diff --git a/.gitattributes b/.gitattributes index b0044cf272fec9..2bc4defac6720c 100644 --- a/.gitattributes +++ b/.gitattributes @@ -4,6 +4,7 @@ *.perl eol=lf diff=perl *.pl eof=lf diff=perl *.pm eol=lf diff=perl +*.png binary *.py eol=lf diff=python *.bat eol=crlf CODE_OF_CONDUCT.md -whitespace diff --git a/.github/ISSUE_TEMPLATE.md b/.github/ISSUE_TEMPLATE.md new file mode 100644 index 00000000000000..4017ed82ca4341 --- /dev/null +++ b/.github/ISSUE_TEMPLATE.md @@ -0,0 +1,64 @@ + - [ ] I was not able to find an [open](https://github.com/git-for-windows/git/issues?q=is%3Aopen) or [closed](https://github.com/git-for-windows/git/issues?q=is%3Aclosed) issue matching what I'm seeing + +### Setup + + - Which version of Git for Windows are you using? Is it 32-bit or 64-bit? + +``` +$ git --version --build-options + +** insert your machine's response here ** +``` + + - Which version of Windows are you running? Vista, 7, 8, 10? Is it 32-bit or 64-bit? + +``` +$ cmd.exe /c ver + +** insert your machine's response here ** +``` + + - What options did you set as part of the installation? Or did you choose the + defaults? + +``` +# One of the following: +> type "C:\Program Files\Git\etc\install-options.txt" +> type "C:\Program Files (x86)\Git\etc\install-options.txt" +> type "%USERPROFILE%\AppData\Local\Programs\Git\etc\install-options.txt" +> type "$env:USERPROFILE\AppData\Local\Programs\Git\etc\install-options.txt" +$ cat /etc/install-options.txt + +** insert your machine's response here ** +``` + + - Any other interesting things about your environment that might be related + to the issue you're seeing? + +** insert your response here ** + +### Details + + - Which terminal/shell are you running Git from? e.g Bash/CMD/PowerShell/other + +** insert your response here ** + + - What commands did you run to trigger this issue? If you can provide a + [Minimal, Complete, and Verifiable example](http://stackoverflow.com/help/mcve) + this will help us understand the issue. + +``` +** insert your commands here ** +``` + - What did you expect to occur after running these commands? + +** insert here ** + + - What actually happened instead? + +** insert here ** + + - If the problem was occurring with a specific repository, can you provide the + URL to that repository to help us with testing? + +** insert URL here ** diff --git a/.github/PULL_REQUEST_TEMPLATE.md b/.github/PULL_REQUEST_TEMPLATE.md index 952c7c3a2aa11e..831ef6f19f1d11 100644 --- a/.github/PULL_REQUEST_TEMPLATE.md +++ b/.github/PULL_REQUEST_TEMPLATE.md @@ -1,7 +1,19 @@ -Thanks for taking the time to contribute to Git! Please be advised that the -Git community does not use github.com for their contributions. Instead, we use -a mailing list (git@vger.kernel.org) for code submissions, code reviews, and -bug reports. Nevertheless, you can use GitGitGadget (https://gitgitgadget.github.io/) +Thanks for taking the time to contribute to Git! + +Those seeking to contribute to the Git for Windows fork should see +http://gitforwindows.org/#contribute on how to contribute Windows specific +enhancements. + +If your contribution is for the core Git functions and documentation +please be aware that the Git community does not use the github.com issues +or pull request mechanism for their contributions. + +Instead, we use the Git mailing list (git@vger.kernel.org) for code and +documentation submissions, code reviews, and bug reports. The +mailing list is plain text only (anything with HTML is sent directly +to the spam folder). + +Nevertheless, you can use GitGitGadget (https://gitgitgadget.github.io/) to conveniently send your Pull Requests commits to our mailing list. Please read the "guidelines for contributing" linked above! diff --git a/.github/config.yml b/.github/config.yml new file mode 100644 index 00000000000000..45edb7ba37ce02 --- /dev/null +++ b/.github/config.yml @@ -0,0 +1,10 @@ +# Configuration for sentiment-bot - https://github.com/behaviorbot/sentiment-bot + +# *Required* toxicity threshold between 0 and .99 with the higher numbers being +# the most toxic. Anything higher than this threshold will be marked as toxic +# and commented on +sentimentBotToxicityThreshold: .7 + +# *Required* Comment to reply with +sentimentBotReplyComment: > + Please be sure to review the code of conduct and be respectful of other users. cc/ @git-for-windows/trusted-git-for-windows-developers diff --git a/.github/workflows/clangarm64-build.yml b/.github/workflows/clangarm64-build.yml new file mode 100644 index 00000000000000..a826dc5b312ac8 --- /dev/null +++ b/.github/workflows/clangarm64-build.yml @@ -0,0 +1,25 @@ +name: CLANG build ARM64 + +on: + workflow_dispatch: + +defaults: + run: + shell: bash + +jobs: + clang-build: + runs-on: [Windows, ARM64] + env: + NO_PERL: 1 + steps: + - uses: actions/checkout@v3 + - uses: git-for-windows/setup-git-for-windows-sdk@v1 + with: + flavor: makepkg-git + architecture: aarch64 + # This assumes that the job is running on a self-hosted runner, + # in which case we need to cleanup SDK files. + cleanup: true + - name: Build Git CLANGARM64 + run: make -j`nproc` diff --git a/.github/workflows/git-artifacts.yml b/.github/workflows/git-artifacts.yml new file mode 100644 index 00000000000000..9b4d1c3aaa3580 --- /dev/null +++ b/.github/workflows/git-artifacts.yml @@ -0,0 +1,514 @@ +name: git-artifacts + +on: + # This workflow can be triggered manually in the Actions tab, see + # https://github.blog/changelog/2020-07-06-github-actions-manual-triggers-with-workflow_dispatch/ + workflow_dispatch: + inputs: + build_only: + description: 'Optionally restrict what artifacts to build' + required: false + ref: + description: 'Optionally override which branch to build' + required: false + repository: + description: 'Optionally override from where to fetch the specified ref' + required: false + +env: + GPG_OPTIONS: "--batch --yes --no-tty --list-options no-show-photos --verify-options no-show-photos --pinentry-mode loopback" + HOME: "${{github.workspace}}\\home" + MSYSTEM: MINGW64 + USERPROFILE: "${{github.workspace}}\\home" + BUILD_ONLY: "${{github.event.inputs.build_only}}" + REPOSITORY: "${{github.event.inputs.repository}}" + REF: "${{github.event.inputs.ref}}" + +jobs: + bundle-artifacts: + runs-on: windows-latest + steps: + - name: Configure user + shell: bash + run: + USER_NAME="${{github.actor}}" && + USER_EMAIL="${{github.actor}}@users.noreply.github.com" && + mkdir "$HOME" && + git config --global user.name "$USER_NAME" && + git config --global user.email "$USER_EMAIL" && + echo "PACKAGER=$USER_NAME <$USER_EMAIL>" >>$GITHUB_ENV + - uses: git-for-windows/setup-git-for-windows-sdk@v1 + with: + flavor: build-installers + - name: Clone build-extra + shell: bash + run: | + d=/usr/src/build-extra && + if test ! -d $d/.git + then + git clone --single-branch -b main https://github.com/git-for-windows/build-extra $d + else + git -C $d fetch https://github.com/git-for-windows/build-extra main && + git -C $d switch -C main FETCH_HEAD + fi + - name: Prepare home directory for GPG signing + if: env.GPGKEY != '' + shell: bash + run: | + echo '${{secrets.PRIVGPGKEY}}' | tr % '\n' | gpg $GPG_OPTIONS --import && + mkdir -p home && + git config --global gpg.program "/usr/src/build-extra/gnupg-with-gpgkey.sh" && + info="$(gpg --list-keys --with-colons "${GPGKEY%% *}" | cut -d : -f 1,10 | sed -n '/^uid/{s|uid:||p;q}')" && + git config --global user.name "${info% <*}" && + git config --global user.email "<${info#*<}" + env: + GPGKEY: ${{secrets.GPGKEY}} + - name: Generate bundle artifacts + env: + GPGKEY: ${{secrets.GPGKEY}} + shell: bash + run: | + printf '#!/bin/sh\n\nexec /mingw64/bin/git.exe "$@"\n' >/usr/bin/git && + mkdir -p bundle-artifacts && + + { test -n "$REPOSITORY" || REPOSITORY='${{github.repository}}'; } && + { test -n "$REF" || REF='${{github.ref}}'; } && + git -c init.defaultBranch=main init --bare && + git remote add -f origin https://github.com/git-for-windows/git && + git fetch "https://github.com/$REPOSITORY" "$REF:$REF" && + + tag_name="$(git describe --match 'v[0-9]*' FETCH_HEAD)-$(date +%Y%m%d%H%M%S)" && + echo "prerelease-${tag_name#v}" >bundle-artifacts/ver && + echo "${tag_name#v}" >bundle-artifacts/display_version && + echo "$tag_name" >bundle-artifacts/next_version && + git tag $(test -z "$GPGKEY" || echo " -s") -m "Snapshot build" "$tag_name" FETCH_HEAD && + git bundle create bundle-artifacts/git.bundle origin/main.."$tag_name" && + + sh -x /usr/src/build-extra/please.sh mention feature "Snapshot of $(git show -s --pretty='tformat:%h (%s, %ad)' --date=short FETCH_HEAD)" && + git -C /usr/src/build-extra bundle create "$PWD/bundle-artifacts/build-extra.bundle" origin/main..main + - name: Clean up temporary files + if: always() + shell: bash + run: rm -rf home + - name: 'Publish Pipeline Artifact: bundle-artifacts' + uses: actions/upload-artifact@v1 + with: + name: bundle-artifacts + path: bundle-artifacts + pkg: + runs-on: windows-latest + needs: bundle-artifacts + strategy: + matrix: + arch: + - name: x86_64 + bitness: 64 + bin: /amd64 + - name: i686 + bitness: 32 + bin: '' + steps: + - name: Determine whether this job should be skipped + shell: bash + run: | + for e in ${BUILD_ONLY:-pkg} + do + case $e in + *-${{matrix.arch.name}}) exit 0;; # build this artifact + *-arm64) test i686 != ${{matrix.arch.name}} || exit 0;; # pkg-i686 is required for the ARM64 version + *-*) ;; # not this build artifact, keep looking + *) exit 0;; # build this artifact + esac + done + echo "SKIP=true" >>$GITHUB_ENV + - name: Configure user + if: env.SKIP != 'true' + shell: bash + run: + USER_NAME="${{github.actor}}" && + USER_EMAIL="${{github.actor}}@users.noreply.github.com" && + mkdir "$HOME" && + git config --global user.name "$USER_NAME" && + git config --global user.email "$USER_EMAIL" && + echo "PACKAGER=$USER_NAME <$USER_EMAIL>" >>$GITHUB_ENV + - uses: git-for-windows/setup-git-for-windows-sdk@v1 + if: env.SKIP != 'true' + with: + flavor: build-installers + - name: Download bundle-artifacts + if: env.SKIP != 'true' + uses: actions/download-artifact@v1 + with: + name: bundle-artifacts + path: bundle-artifacts + - name: Clone and update build-extra + if: env.SKIP != 'true' + shell: bash + run: | + d=/usr/src/build-extra && + if test ! -d $d/.git + then + git clone --single-branch -b main https://github.com/git-for-windows/build-extra $d + else + git -C $d fetch https://github.com/git-for-windows/build-extra main && + git -C $d switch -C main FETCH_HEAD + fi && + git -C $d pull "$PWD"/bundle-artifacts/build-extra.bundle main + - name: Check out git/git + if: env.SKIP != 'true' + shell: bash + run: | + git -c init.defaultBranch=main init && + git remote add -f origin https://github.com/git-for-windows/git && + git fetch --tags bundle-artifacts/git.bundle $(cat bundle-artifacts/next_version) && + git reset --hard $(cat bundle-artifacts/next_version) + - name: Prepare home directory for code-signing + env: + CODESIGN_P12: ${{secrets.CODESIGN_P12}} + CODESIGN_PASS: ${{secrets.CODESIGN_PASS}} + if: env.SKIP != 'true' && env.CODESIGN_P12 != '' && env.CODESIGN_PASS != '' + shell: bash + run: | + cd home && + mkdir -p .sig && + echo -n "$CODESIGN_P12" | tr % '\n' | base64 -d >.sig/codesign.p12 && + echo -n "$CODESIGN_PASS" >.sig/codesign.pass + git config --global alias.signtool '!sh "/usr/src/build-extra/signtool.sh"' + - name: Prepare home directory for GPG signing + if: env.SKIP != 'true' && env.GPGKEY != '' + shell: bash + run: | + echo '${{secrets.PRIVGPGKEY}}' | tr % '\n' | gpg $GPG_OPTIONS --import && + info="$(gpg --list-keys --with-colons "${GPGKEY%% *}" | cut -d : -f 1,10 | sed -n '/^uid/{s|uid:||p;q}')" && + git config --global user.name "${info% <*}" && + git config --global user.email "<${info#*<}" + env: + GPGKEY: ${{secrets.GPGKEY}} + - name: Build mingw-w64-${{matrix.arch.name}}-git + if: env.SKIP != 'true' + env: + GPGKEY: "${{secrets.GPGKEY}}" + shell: bash + run: | + set -x + + # Make sure that there is a `/usr/bin/git` that can be used by `makepkg-mingw` + printf '#!/bin/sh\n\nexec /mingw64/bin/git.exe "$@"\n' >/usr/bin/git && + + # Restrict `PATH` to MSYS2 and to Visual Studio (to let `cv2pdb` find the relevant DLLs) + PATH="/mingw64/bin:/usr/bin:/c/Program Files/Microsoft Visual Studio/2022/Enterprise/Common7/IDE/:/C/Program Files (x86)/Microsoft Visual Studio 14.0/VC/bin${{matrix.arch.bin}}:/C/Windows/system32" + + type -p mspdb140.dll || exit 1 + sh -x /usr/src/build-extra/please.sh build-mingw-w64-git --only-${{matrix.arch.bitness}}-bit --build-src-pkg -o artifacts HEAD && + cp bundle-artifacts/ver artifacts/ && + if test -n "$GPGKEY" + then + for tar in artifacts/*.tar* + do + /usr/src/build-extra/gnupg-with-gpgkey.sh --detach-sign --no-armor $tar + done + fi && + + b=$PWD/artifacts && + version=$(cat bundle-artifacts/next_version) && + (cd /usr/src/MINGW-packages/mingw-w64-git && + cp PKGBUILD.$version PKGBUILD && + git commit -s -m "mingw-w64-git: new version ($version)" PKGBUILD && + git bundle create "$b"/MINGW-packages.bundle origin/main..main) + - name: Clean up temporary files + if: always() && env.SKIP != 'true' + shell: bash + run: rm -rf home + - name: Publish mingw-w64-${{matrix.arch.name}}-git + if: env.SKIP != 'true' + uses: actions/upload-artifact@v1 + with: + name: pkg-${{matrix.arch.name}} + path: artifacts + build-arm64: + needs: bundle-artifacts + runs-on: windows-latest + steps: + - name: Determine whether this job should be skipped + shell: bash + run: | + for e in ${BUILD_ONLY:-pkg} + do + case $e in + *-arm64) exit 0;; # build this artifact + *-*) ;; # not this build artifact, keep looking + *) exit 0;; # build this artifact + esac + done + echo "SKIP=true" >>$GITHUB_ENV + - name: Configure user + if: env.SKIP != 'true' + shell: bash + run: + USER_NAME="${{github.actor}}" && + USER_EMAIL="${{github.actor}}@users.noreply.github.com" && + mkdir -p "$HOME" && + git config --global user.name "$USER_NAME" && + git config --global user.email "$USER_EMAIL" + - name: Download bundle-artifacts + if: env.SKIP != 'true' + uses: actions/download-artifact@v1 + with: + name: bundle-artifacts + path: bundle-artifacts + - name: Check out git/git + if: env.SKIP != 'true' + shell: bash + run: | + git -c init.defaultBranch=main init && + git remote add -f origin https://github.com/git-for-windows/git && + git fetch --tags bundle-artifacts/git.bundle $(cat bundle-artifacts/next_version) && + git reset --hard $(cat bundle-artifacts/next_version) + - name: initialize vcpkg + if: env.SKIP != 'true' + uses: actions/checkout@v2 + with: + repository: 'microsoft/vcpkg' + path: 'compat/vcbuild/vcpkg' + - name: download vcpkg artifacts + if: env.SKIP != 'true' + uses: git-for-windows/get-azure-pipelines-artifact@v0 + with: + repository: git/git + definitionId: 9 + - name: add msbuild to PATH + if: env.SKIP != 'true' + uses: microsoft/setup-msbuild@v1 + - name: copy dlls to root + if: env.SKIP != 'true' + shell: powershell + run: | + & compat\vcbuild\vcpkg_copy_dlls.bat release arm64-windows + if (!$?) { exit(1) } + - name: generate Visual Studio solution + if: env.SKIP != 'true' + shell: bash + run: | + cmake `pwd`/contrib/buildsystems/ -DCMAKE_PREFIX_PATH=`pwd`/compat/vcbuild/vcpkg/installed/arm64-windows \ + -DNO_GETTEXT=YesPlease -DPERL_TESTS=OFF -DPYTHON_TESTS=OFF -DCURL_NO_CURL_CMAKE=ON -DCMAKE_GENERATOR_PLATFORM=arm64 -DVCPKG_ARCH=arm64-windows \ + -DCMAKE_INSTALL_PREFIX="`pwd`/git-arm64" -DSKIP_DASHED_BUILT_INS=ON -DHOST_CPU=arm64 + - name: MSBuild + if: env.SKIP != 'true' + run: msbuild git.sln -property:Configuration=Release + - name: Link the Git executables + if: env.SKIP != 'true' + run: msbuild INSTALL.vcxproj -property:Configuration=Release + - name: upload build artifacts + if: env.SKIP != 'true' + uses: actions/upload-artifact@v1 + with: + name: arm64-artifacts + path: ./git-arm64 + artifacts: + runs-on: windows-latest + needs: [pkg, build-arm64] + strategy: + matrix: + artifact: + - name: installer + fileprefix: Git + fileextension: exe + - name: portable + fileprefix: PortableGit + fileextension: exe + - name: archive + fileprefix: Git + fileextension: tar.bz2 + - name: mingit + fileprefix: MinGit + fileextension: zip + - name: mingit-busybox + fileprefix: MinGit + fileextension: zip + arch: + - name: x86_64 + bitness: 64 + arm64: false + - name: i686 + bitness: 32 + arm64: false + - name: i686 + bitness: 32 + arm64: true + fail-fast: false + env: + MSYSTEM: MINGW${{matrix.arch.bitness}} + steps: + - name: Determine whether this job should be skipped + shell: bash + run: | + suffix=${{matrix.arch.name}} + if test true = ${{matrix.arch.arm64}} + then + suffix=arm64 + fi + case " $BUILD_ONLY " in + ' ') ;; # not set; build all + *" ${{matrix.artifact.name}} "*|*" ${{matrix.artifact.name}}-$suffix "*) ;; # build this artifact + *) echo "SKIP=true" >>$GITHUB_ENV;; + esac + - name: Download pkg-${{matrix.arch.name}} + if: env.SKIP != 'true' + uses: actions/download-artifact@v1 + with: + name: pkg-${{matrix.arch.name}} + path: pkg-${{matrix.arch.name}} + - name: Download bundle-artifacts + if: env.SKIP != 'true' + uses: actions/download-artifact@v1 + with: + name: bundle-artifacts + path: bundle-artifacts + - uses: git-for-windows/setup-git-for-windows-sdk@v1 + if: env.SKIP != 'true' && matrix.arch.bitness == '64' + with: + flavor: build-installers + - uses: git-for-windows/setup-git-for-windows-sdk@v1 + if: env.SKIP != 'true' && matrix.arch.bitness == '32' + with: + flavor: build-installers + architecture: i686 + - name: Download arm64 artifact + if: env.SKIP != 'true' && matrix.arch.arm64 == true + uses: actions/download-artifact@v1 + with: + name: arm64-artifacts + path: ${{github.workspace}}/arm64 + # Workaround for Git Credential Manager Core on ARM64: https://github.com/git-for-windows/git/issues/3015 + - name: Create git-credential-manager-core wrapper for ARM64 + if: env.SKIP != 'true' && matrix.arch.arm64 == true + shell: bash + run: | + printf '%s\n' '#!/bin/sh' 'exec /mingw32/libexec/git-core/git-credential-manager-core.exe "$@"' > arm64/libexec/git-core/git-credential-manager-core + chmod +x arm64/libexec/git-core/git-credential-manager-core + - name: Clone and update build-extra + if: env.SKIP != 'true' + shell: bash + run: | + d=/usr/src/build-extra && + if test ! -d $d/.git + then + git clone --single-branch -b main https://github.com/git-for-windows/build-extra $d + else + git -C $d fetch https://github.com/git-for-windows/build-extra main && + git -C $d switch -C main FETCH_HEAD + fi && + git -C $d pull "$PWD"/bundle-artifacts/build-extra.bundle main + - name: Prepare home directory for code-signing + env: + CODESIGN_P12: ${{secrets.CODESIGN_P12}} + CODESIGN_PASS: ${{secrets.CODESIGN_PASS}} + if: env.SKIP != 'true' && (matrix.artifact.name == 'installer' || matrix.artifact.name == 'portable') && env.CODESIGN_P12 != '' && env.CODESIGN_PASS != '' + shell: bash + run: | + mkdir -p home/.sig && + echo -n "$CODESIGN_P12" | tr % '\n' | base64 -d >home/.sig/codesign.p12 && + echo -n "$CODESIGN_PASS" >home/.sig/codesign.pass && + git config --global alias.signtool '!sh "/usr/src/build-extra/signtool.sh"' + - name: Build ${{matrix.arch.bitness}}-bit ${{matrix.artifact.name}} + if: env.SKIP != 'true' + shell: bash + run: | + set -x + if test "${{matrix.arch.arm64}}" = true + then + ARM64="--include-arm64-artifacts=\"$PWD/arm64\"" + else + ARM64= + fi + + eval /usr/src/build-extra/please.sh make_installers_from_mingw_w64_git $ARM64 --version=$(cat pkg-${{matrix.arch.name}}/ver) -o artifacts --${{matrix.artifact.name}} --pkg=pkg-${{matrix.arch.name}}/mingw-w64-${{matrix.arch.name}}-git-[0-9]*.tar.xz --pkg=pkg-${{matrix.arch.name}}/mingw-w64-${{matrix.arch.name}}-git-doc-html-[0-9]*.tar.xz && + if test portable = '${{matrix.artifact.name}}' && test -n "$(git config alias.signtool)" + then + git signtool artifacts/PortableGit-*.exe + fi && + openssl dgst -sha256 artifacts/${{matrix.artifact.fileprefix}}-*.${{matrix.artifact.fileextension}} | sed "s/.* //" >artifacts/sha-256.txt + - name: Copy package-versions and pdbs + if: env.SKIP != 'true' && matrix.artifact.name == 'installer' + shell: bash + run: | + cp /usr/src/build-extra/installer/package-versions.txt artifacts/ && + + a=$PWD/artifacts && + p=$PWD/pkg-${{matrix.arch.name}} && + (cd /usr/src/build-extra && + mkdir -p cached-source-packages && + cp "$p"/*-pdb* cached-source-packages/ && + GIT_CONFIG_PARAMETERS="'windows.sdk${{matrix.arch.bitness}}.path='" ./please.sh bundle_pdbs --arch=${{matrix.arch.name}} --directory="$a" installer/package-versions.txt) + - name: Clean up temporary files + if: always() && env.SKIP != 'true' + shell: bash + run: rm -rf home + - name: Publish ${{matrix.artifact.name}}-${{matrix.arch.name}} + if: env.SKIP != 'true' && matrix.arch.arm64 != true + uses: actions/upload-artifact@v1 + with: + name: ${{matrix.artifact.name}}-${{matrix.arch.name}} + path: artifacts + - name: Publish ${{matrix.artifact.name}}-arm64 + if: env.SKIP != 'true' && matrix.arch.arm64 == true + uses: actions/upload-artifact@v1 + with: + name: ${{matrix.artifact.name}}-arm64 + path: artifacts + nuget: + runs-on: windows-latest + needs: pkg + steps: + - name: Determine whether this job should be skipped + shell: bash + run: | + case " $BUILD_ONLY " in + ' ') ;; # not set; build all + *" nuget "*) ;; # build this artifact + *) echo "SKIP=true" >>$GITHUB_ENV;; + esac + - name: Download pkg-x86_64 + if: env.SKIP != 'true' + uses: actions/download-artifact@v1 + with: + name: pkg-x86_64 + path: pkg-x86_64 + - name: Download bundle-artifacts + if: env.SKIP != 'true' + uses: actions/download-artifact@v1 + with: + name: bundle-artifacts + path: bundle-artifacts + - uses: git-for-windows/setup-git-for-windows-sdk@v1 + if: env.SKIP != 'true' + with: + flavor: build-installers + - name: Clone and update build-extra + if: env.SKIP != 'true' + shell: bash + run: | + d=/usr/src/build-extra && + if test ! -d $d/.git + then + git clone --single-branch -b main https://github.com/git-for-windows/build-extra $d + else + git -C $d fetch https://github.com/git-for-windows/build-extra main && + git -C $d switch -C main FETCH_HEAD + fi && + git -C $d pull "$PWD"/bundle-artifacts/build-extra.bundle main + - uses: nuget/setup-nuget@v1 + if: env.SKIP != 'true' + - name: Build 64-bit NuGet packages + if: env.SKIP != 'true' + shell: bash + run: | + /usr/src/build-extra/please.sh make_installers_from_mingw_w64_git --version=$(cat pkg-x86_64/ver) -o artifacts --nuget --pkg=pkg-x86_64/mingw-w64-x86_64-git-[0-9]*.tar.xz --pkg=pkg-x86_64/mingw-w64-x86_64-git-doc-html-[0-9]*.tar.xz && + /usr/src/build-extra/please.sh make_installers_from_mingw_w64_git --version=$(cat pkg-x86_64/ver) -o artifacts --nuget-mingit && + openssl dgst -sha256 artifacts/Git*.nupkg | sed "s/.* //" >artifacts/sha-256.txt + - name: Publish nuget-x86_64 + if: env.SKIP != 'true' + uses: actions/upload-artifact@v1 + with: + name: nuget-x86_64 + path: artifacts diff --git a/.github/workflows/main.yml b/.github/workflows/main.yml index 831f4df56c51dc..92063286fffbaf 100644 --- a/.github/workflows/main.yml +++ b/.github/workflows/main.yml @@ -137,6 +137,9 @@ jobs: NO_PERL: 1 GIT_CONFIG_PARAMETERS: "'user.name=CI' 'user.email=ci@git'" runs-on: windows-latest + strategy: + matrix: + arch: [x64, arm64] steps: - uses: actions/checkout@v2 - uses: git-for-windows/setup-git-for-windows-sdk@v1 @@ -146,26 +149,22 @@ jobs: repository: 'microsoft/vcpkg' path: 'compat/vcbuild/vcpkg' - name: download vcpkg artifacts - shell: powershell - run: | - $urlbase = "https://dev.azure.com/git/git/_apis/build/builds" - $id = ((Invoke-WebRequest -UseBasicParsing "${urlbase}?definitions=9&statusFilter=completed&resultFilter=succeeded&`$top=1").content | ConvertFrom-JSON).value[0].id - $downloadUrl = ((Invoke-WebRequest -UseBasicParsing "${urlbase}/$id/artifacts").content | ConvertFrom-JSON).value[0].resource.downloadUrl - (New-Object Net.WebClient).DownloadFile($downloadUrl, "compat.zip") - Expand-Archive compat.zip -DestinationPath . -Force - Remove-Item compat.zip + uses: git-for-windows/get-azure-pipelines-artifact@v0 + with: + repository: git/git + definitionId: 9 - name: add msbuild to PATH uses: microsoft/setup-msbuild@v1 - name: copy dlls to root shell: cmd - run: compat\vcbuild\vcpkg_copy_dlls.bat release + run: compat\vcbuild\vcpkg_copy_dlls.bat release ${{ matrix.arch }}-windows - name: generate Visual Studio solution shell: bash run: | - cmake `pwd`/contrib/buildsystems/ -DCMAKE_PREFIX_PATH=`pwd`/compat/vcbuild/vcpkg/installed/x64-windows \ - -DNO_GETTEXT=YesPlease -DPERL_TESTS=OFF -DPYTHON_TESTS=OFF -DCURL_NO_CURL_CMAKE=ON + cmake `pwd`/contrib/buildsystems/ -DCMAKE_PREFIX_PATH=`pwd`/compat/vcbuild/vcpkg/installed/${{ matrix.arch }}-windows \ + -DNO_GETTEXT=YesPlease -DPERL_TESTS=OFF -DPYTHON_TESTS=OFF -DCURL_NO_CURL_CMAKE=ON -DCMAKE_GENERATOR_PLATFORM=${{ matrix.arch }} -DVCPKG_ARCH=${{ matrix.arch }}-windows -DHOST_CPU=${{ matrix.arch }} - name: MSBuild - run: msbuild git.sln -property:Configuration=Release -property:Platform=x64 -maxCpuCount:4 -property:PlatformToolset=v142 + run: msbuild git.sln -property:Configuration=Release -property:Platform=${{ matrix.arch }} -maxCpuCount:4 -property:PlatformToolset=v142 - name: bundle artifact tar shell: bash env: @@ -179,7 +178,7 @@ jobs: - name: upload tracked files and build artifacts uses: actions/upload-artifact@v2 with: - name: vs-artifacts + name: vs-artifacts-${{ matrix.arch }} path: artifacts vs-test: name: win+VS test @@ -194,7 +193,7 @@ jobs: - name: download tracked files and build artifacts uses: actions/download-artifact@v2 with: - name: vs-artifacts + name: vs-artifacts-x64 path: ${{github.workspace}} - name: extract tracked files and build artifacts shell: bash diff --git a/.github/workflows/monitor-components.yml b/.github/workflows/monitor-components.yml new file mode 100644 index 00000000000000..18b2053dbb9518 --- /dev/null +++ b/.github/workflows/monitor-components.yml @@ -0,0 +1,91 @@ +name: Monitor component updates + +# Git for Windows is a slightly modified subset of MSYS2. Some of its +# components are maintained by Git for Windows, others by MSYS2. To help +# keeping the former up to date, this workflow monitors the Atom/RSS feeds +# and opens new tickets for each new component version. + +on: + schedule: + - cron: "23 8,11,14,17 * * *" + workflow_dispatch: + +env: + CHARACTER_LIMIT: 5000 + MAX_AGE: 7d + +jobs: + job: + runs-on: ubuntu-latest + permissions: + issues: write + strategy: + matrix: + component: + - label: git + feed: https://github.com/git/git/tags.atom + - label: git-lfs + feed: https://github.com/git-lfs/git-lfs/tags.atom + - label: gcm-core + feed: https://github.com/microsoft/git-credential-manager-core/tags.atom + - label: tig + feed: https://github.com/jonas/tig/tags.atom + - label: cygwin + feed: https://github.com/cygwin/cygwin/releases.atom + title-pattern: ^(?!.*newlib) + - label: msys2-runtime-package + feed: https://github.com/msys2/MSYS2-packages/commits/master/msys2-runtime.atom + - label: msys2-runtime + feed: https://github.com/msys2/msys2-runtime/commits/HEAD.atom + aggregate: true + - label: openssh + feed: https://github.com/openssh/openssh-portable/tags.atom + - label: libfido2 + feed: https://github.com/Yubico/libfido2/tags.atom + - label: libcbor + feed: https://github.com/PJK/libcbor/tags.atom + - label: openssl + feed: https://github.com/openssl/openssl/tags.atom + title-pattern: ^(?!.*alpha) + - label: gnutls + feed: https://gnutls.org/news.atom + - label: heimdal + feed: https://github.com/heimdal/heimdal/tags.atom + - label: git-sizer + feed: https://github.com/github/git-sizer/tags.atom + - label: gitflow + feed: https://github.com/petervanderdoes/gitflow-avh/tags.atom + - label: curl + feed: https://github.com/curl/curl/tags.atom + - label: libgpg-error + feed: https://github.com/gpg/libgpg-error/releases.atom + title-pattern: ^libgpg-error-[0-9\.]*$ + - label: libgcrypt + feed: https://github.com/gpg/libgcrypt/releases.atom + title-pattern: ^libgcrypt-[0-9\.]*$ + - label: gpg + feed: https://github.com/gpg/gnupg/releases.atom + - label: mintty + feed: https://github.com/mintty/mintty/releases.atom + - label: p7zip + feed: https://sourceforge.net/projects/p7zip/rss?path=/p7zip + - label: bash + feed: https://git.savannah.gnu.org/cgit/bash.git/atom/?h=master + aggregate: true + - label: perl + feed: https://github.com/Perl/perl5/tags.atom + title-pattern: ^(?!.*(5\.[0-9]+[13579]|RC)) + - label: pcre2 + feed: https://github.com/PCRE2Project/pcre2/tags.atom + fail-fast: false + steps: + - uses: git-for-windows/rss-to-issues@v0 + with: + feed: ${{matrix.component.feed}} + prefix: "[New ${{matrix.component.label}} version]" + labels: component-update + github-token: ${{ secrets.GITHUB_TOKEN }} + character-limit: ${{ env.CHARACTER_LIMIT }} + max-age: ${{ env.MAX_AGE }} + aggregate: ${{matrix.component.aggregate}} + title-pattern: ${{matrix.component.title-pattern}} diff --git a/.gitignore b/.gitignore index b3dcafcb3310e9..9e19c20024c2a1 100644 --- a/.gitignore +++ b/.gitignore @@ -2,6 +2,7 @@ /fuzz_corpora /fuzz-pack-headers /fuzz-pack-idx +/GIT-BUILD-DIR /GIT-BUILD-OPTIONS /GIT-CFLAGS /GIT-LDFLAGS @@ -249,3 +250,4 @@ Release/ /git.VC.db *.dSYM /contrib/buildsystems/out +CMakeSettings.json diff --git a/CODE_OF_CONDUCT.md b/CODE_OF_CONDUCT.md index 0215b1fd4c05e6..8b373cf5518a5d 100644 --- a/CODE_OF_CONDUCT.md +++ b/CODE_OF_CONDUCT.md @@ -1,9 +1,9 @@ -# Git Code of Conduct +# Git for Windows Code of Conduct This code of conduct outlines our expectations for participants within -the Git community, as well as steps for reporting unacceptable behavior. -We are committed to providing a welcoming and inspiring community for -all and expect our code of conduct to be honored. Anyone who violates +the **Git for Windows** community, as well as steps for reporting unacceptable +behavior. We are committed to providing a welcoming and inspiring community +for all and expect our code of conduct to be honored. Anyone who violates this code of conduct may be banned from the community. ## Our Pledge @@ -65,8 +65,8 @@ representative at an online or offline event. ## Enforcement Instances of abusive, harassing, or otherwise unacceptable behavior may be -reported to the community leaders responsible for enforcement at -git@sfconservancy.org, or individually: +reported by contacting the Git for Windows maintainer or the community leaders +responsible for enforcement at git@sfconservancy.org, or individually: - Ævar Arnfjörð Bjarmason - Christian Couder diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md new file mode 100644 index 00000000000000..ca9770df8c4808 --- /dev/null +++ b/CONTRIBUTING.md @@ -0,0 +1,417 @@ +How to Contribute to Git for Windows +==================================== + +Git was originally designed for Unix systems and still today, all the build tools for the Git +codebase assume you have standard Unix tools available in your path. If you have an open-source +mindset and want to start contributing to Git, but primarily use a Windows machine, then you may +have trouble getting started. This guide is for you. + +Get the Source +-------------- + +Clone the [GitForWindows repository on GitHub](https://github.com/git-for-windows/git). +It is helpful to create your own fork for storing your development branches. + +Windows uses different line endings than Unix systems. See +[this GitHub article on working with line endings](https://help.github.com/articles/dealing-with-line-endings/#refreshing-a-repository-after-changing-line-endings) +if you have trouble with line endings. + +Build the Source +---------------- + +First, download and install the latest [Git for Windows SDK (64-bit)](https://github.com/git-for-windows/build-extra/releases/latest). +When complete, you can run the Git SDK, which creates a new Git Bash terminal window with +the additional development commands, such as `make`. + + As of time of writing, the SDK uses a different credential manager, so you may still want to use normal Git + Bash for interacting with your remotes. Alternatively, use SSH rather than HTTPS and + avoid credential manager problems. + +You should now be ready to type `make` from the root of your `git` source directory. +Here are some helpful variations: + +* `make -j[N] DEVELOPER=1`: Compile new sources using up to N concurrent processes. + The `DEVELOPER` flag turns on all warnings; code failing these warnings will not be + accepted upstream ("upstream" = "the core Git project"). +* `make clean`: Delete all compiled files. + +When running `make`, you can use `-j$(nproc)` to automatically use the number of processors +on your machine as the number of concurrent build processes. + +You can go deeper on the Windows-specific build process by reading the +[technical overview](https://github.com/git-for-windows/git/wiki/Technical-overview) or the +[guide to compiling Git with Visual Studio](https://github.com/git-for-windows/git/wiki/Compiling-Git-with-Visual-Studio). + +## Building `git` on Windows with Visual Studio + +The typical approach to building `git` is to use the standard `Makefile` with GCC, as +above. Developers working in a Windows environment may want to instead build with the +[Microsoft Visual C++ compiler and libraries toolset (MSVC)](https://blogs.msdn.microsoft.com/vcblog/2017/03/07/msvc-the-best-choice-for-windows/). +There are a few benefits to using MSVC over GCC during your development, including creating +symbols for debugging and [performance tracing](https://github.com/Microsoft/perfview#perfview-overview). + +There are two ways to build Git for Windows using MSVC. Each have their own merits. + +### Using SDK Command Line + +Use one of the following commands from the SDK Bash window to build Git for Windows: + +``` + make MSVC=1 -j12 + make MSVC=1 DEBUG=1 -j12 +``` + +The first form produces release-mode binaries; the second produces debug-mode binaries. +Both forms produce PDB files and can be debugged. However, the first is best for perf +tracing and the second is best for single-stepping. + +You can then open Visual Studio and select File -> Open -> Project/Solution and select +the compiled `git.exe` file. This creates a basic solution and you can use the debugging +and performance tracing tools in Visual Studio to monitor a Git process. Use the Debug +Properties page to set the working directory and command line arguments. + +Be sure to clean up before switching back to GCC (or to switch between debug and +release MSVC builds): + +``` + make MSVC=1 -j12 clean + make MSVC=1 DEBUG=1 -j12 clean +``` + +### Using the IDE + +If you prefer working in Visual Studio with a solution full of projects, then you can use +CMake, either by letting Visual Studio configure it automatically (simply open Git's +top-level directory via `File>Open>Folder...`) or by (downloading and) running +[CMake](https://cmake.org) manually. + +What to Change? +--------------- + +Many new contributors ask: What should I start working on? + +One way to win big with the open-source community is to look at the +[issues page](https://github.com/git-for-windows/git/issues) and see if there are any issues that +you can fix quickly, or if anything catches your eye. + +You can also look at [the unofficial Chromium issues page](https://crbug.com/git) for +multi-platform issues. You can look at recent user questions on +[the Git mailing list](https://public-inbox.org/git). + +Or you can "scratch your own itch", i.e. address an issue you have with Git. The team at Microsoft where the Git for Windows maintainer works, for example, is focused almost entirely on [improving performance](https://blogs.msdn.microsoft.com/devops/2018/01/11/microsofts-performance-contributions-to-git-in-2017/). +We approach our work by finding something that is slow and try to speed it up. We start our +investigation by reliably reproducing the slow behavior, then running that example using +the MSVC build and tracing the results in PerfView. + +You could also think of something you wish Git could do, and make it do that thing! The +only concern I would have with this approach is whether or not that feature is something +the community also wants. If this excites you though, go for it! Don't be afraid to +[get involved in the mailing list](http://vger.kernel.org/vger-lists.html#git) early for +feedback on the idea. + +Test Your Changes +----------------- + +After you make your changes, it is important that you test your changes. Manual testing is +important, but checking and extending the existing test suite is even more important. You +want to run the functional tests to see if you broke something else during your change, and +you want to extend the functional tests to be sure no one breaks your feature in the future. + +### Functional Tests + +Navigate to the `t/` directory and type `make` to run all tests or use `prove` as +[described in the Git for Windows wiki](https://github.com/git-for-windows/git/wiki/Building-Git): + +``` +prove -j12 --state=failed,save ./t[0-9]*.sh +``` + +You can also run each test directly by running the corresponding shell script with a name +like `tNNNN-descriptor.sh`. + +If you are adding new functionality, you may need to create unit tests by creating +helper commands that test a very limited action. These commands are stored in `t/helpers`. +When adding a helper, be sure to add a line to `t/Makefile` and to the `.gitignore` for the +binary file you add. The Git community prefers functional tests using the full `git` +executable, so try to exercise your new code using `git` commands before creating a test +helper. + +To find out why a test failed, repeat the test with the `-x -v -d -i` options and then +navigate to the appropriate "trash" directory to see the data shape that was used for the +test failed step. + +Read [`t/README`](t/README) for more details. + +### Performance Tests + +If you are working on improving performance, you will need to be acquainted with the +performance tests in `t/perf`. There are not too many performance tests yet, but adding one +as your first commit in a patch series helps to communicate the boost your change provides. + +To check the change in performance across multiple versions of `git`, you can use the +`t/perf/run` script. For example, to compare the performance of `git rev-list` across the +`core/master` and `core/next` branches compared to a `topic` branch, you can run + +``` +cd t/perf +./run core/master core/next topic -- p0001-rev-list.sh +``` + +You can also set certain environment variables to help test the performance on different +repositories or with more repetitions. The full list is available in +[the `t/perf/README` file](t/perf/README), +but here are a few important ones: + +``` +GIT_PERF_REPO=/path/to/repo +GIT_PERF_LARGE_REPO=/path/to/large/repo +GIT_PERF_REPEAT_COUNT=10 +``` + +When running the performance tests on Linux, you may see a message "Can't locate JSON.pm in +@INC" and that means you need to run `sudo cpanm install JSON` to get the JSON perl package. + +For running performance tests, it can be helpful to set up a few repositories with strange +data shapes, such as: + +**Many objects:** Clone repos such as [Kotlin](https://github.com/jetbrains/kotlin), [Linux](https://github.com/torvalds/linux), or [Android](https://source.android.com/setup/downloading). + +**Many pack-files:** You can split a fresh clone into multiple pack-files of size at most +16MB by running `git repack -adfF --max-pack-size=16m`. See the +[`git repack` documentation](https://git-scm.com/docs/git-repack) for more information. +You can count the number of pack-files using `ls .git/objects/pack/*.pack | wc -l`. + +**Many loose objects:** If you already split your repository into multiple pack-files, then +you can pick one to split into loose objects using `cat .git/objects/pack/[id].pack | git unpack-objects`; +delete the `[id].pack` and `[id].idx` files after this. You can count the number of loose +bjects using `ls .git/objects/??/* | wc -l`. + +**Deep history:** Usually large repositories also have deep histories, but you can use the +[test-many-commits-1m repo](https://github.com/cirosantilli/test-many-commits-1m/) to +target deep histories without the overhead of many objects. One issue with this repository: +there are no merge commits, so you will need to use a different repository to test a "wide" +commit history. + +**Large Index:** You can generate a large index and repo by using the scripts in +`t/perf/repos`. There are two scripts. `many-files.sh` which will generate a repo with +same tree and blobs but different paths. Using `many-files.sh -d 5 -w 10 -f 9` will create +a repo with ~1 million entries in the index. `inflate-repo.sh` will use an existing repo +and copy the current work tree until it is a specified size. + +Test Your Changes on Linux +-------------------------- + +It can be important to work directly on the [core Git codebase](https://github.com/git/git), +such as a recent commit into the `master` or `next` branch that has not been incorporated +into Git for Windows. Also, it can help to run functional and performance tests on your +code in Linux before submitting patches to the mailing list, which focuses on many platforms. +The differences between Windows and Linux are usually enough to catch most cross-platform +issues. + +### Using the Windows Subsystem for Linux + +The [Windows Subsystem for Linux (WSL)](https://docs.microsoft.com/en-us/windows/wsl/install-win10) +allows you to [install Ubuntu Linux as an app](https://www.microsoft.com/en-us/store/p/ubuntu/9nblggh4msv6) +that can run Linux executables on top of the Windows kernel. Internally, +Linux syscalls are interpreted by the WSL, everything else is plain Ubuntu. + +First, open WSL (either type "Bash" in Cortana, or execute "bash.exe" in a CMD window). +Then install the prerequisites, and `git` for the initial clone: + +``` +sudo apt-get update +sudo apt-get install git gcc make libssl-dev libcurl4-openssl-dev \ + libexpat-dev tcl tk gettext git-email zlib1g-dev +``` + +Then, clone and build: + +``` +git clone https://github.com/git-for-windows/git +cd git +git remote add -f upstream https://github.com/git/git +make +``` + +Be sure to clone into `/home/[user]/` and not into any folder under `/mnt/?/` or your build +will fail due to colons in file names. + +### Using a Linux Virtual Machine with Hyper-V + +If you prefer, you can use a virtual machine (VM) to run Linux and test your changes in the +full environment. The test suite runs a lot faster on Linux than on Windows or with the WSL. +You can connect to the VM using an SSH terminal like +[PuTTY](https://www.chiark.greenend.org.uk/~sgtatham/putty/). + +The following instructions are for using Hyper-V, which is available in some versions of Windows. +There are many virtual machine alternatives available, if you do not have such a version installed. + +* [Download an Ubuntu Server ISO](https://www.ubuntu.com/download/server). +* Open [Hyper-V Manager](https://docs.microsoft.com/en-us/virtualization/hyper-v-on-windows/quick-start/enable-hyper-v). +* [Set up a virtual switch](https://docs.microsoft.com/en-us/virtualization/hyper-v-on-windows/quick-start/connect-to-network) + so your VM can reach the network. +* Select "Quick Create", name your machine, select the ISO as installation source, and un-check + "This virtual machine will run Windows." +* Go through the Ubuntu install process, being sure to select to install OpenSSH Server. +* When install is complete, log in and check the SSH server status with `sudo service ssh status`. + * If the service is not found, install with `sudo apt-get install openssh-server`. + * If the service is not running, then use `sudo service ssh start`. +* Use `shutdown -h now` to shutdown the VM, go to the Hyper-V settings for the VM, expand Network Adapter + to select "Advanced Features", and set the MAC address to be static (this can save your VM from losing + network if shut down incorrectly). +* Provide as many cores to your VM as you can (for parallel builds). +* Restart your VM, but do not connect. +* Use `ssh` in Git Bash, download [PuTTY](http://www.putty.org/), or use your favorite SSH client to connect to the VM through SSH. + +In order to build and use `git`, you will need the following libraries via `apt-get`: + +``` +sudo apt-get update +sudo apt-get install git gcc make libssl-dev libcurl4-openssl-dev \ + libexpat-dev tcl tk gettext git-email zlib1g-dev +``` + +To get your code from your Windows machine to the Linux VM, it is easiest to push the branch to your fork of Git and clone your fork in the Linux VM. + +Don't forget to set your `git` config with your preferred name, email, and editor. + +Polish Your Commits +------------------- + +Before submitting your patch, be sure to read the [coding guidelines](https://github.com/git/git/blob/master/Documentation/CodingGuidelines) +and check your code to match as best you can. This can be a lot of effort, but it saves +time during review to avoid style issues. + +The other possibly major difference between the mailing list submissions and GitHub PR workflows +is that each commit will be reviewed independently. Even if you are submitting a +patch series with multiple commits, each commit must stand on it's own and be reviewable +by itself. Make sure the commit message clearly explain the why of the commit not the how. +Describe what is wrong with the current code and how your changes have made the code better. + +When preparing your patch, it is important to put yourself in the shoes of the Git community. +Accepting a patch requires more justification than approving a pull request from someone on +your team. The community has a stable product and is responsible for keeping it stable. If +you introduce a bug, then they cannot count on you being around to fix it. When you decided +to start work on a new feature, they were not part of the design discussion and may not +even believe the feature is worth introducing. + +Questions to answer in your patch message (and commit messages) may include: +* Why is this patch necessary? +* How does the current behavior cause pain for users? +* What kinds of repositories are necessary for noticing a difference? +* What design options did you consider before writing this version? Do you have links to + code for those alternate designs? +* Is this a performance fix? Provide clear performance numbers for various well-known repos. + +Here are some other tips that we use when cleaning up our commits: + +* Commit messages should be wrapped at 76 columns per line (or less; 72 is also a + common choice). +* Make sure the commits are signed off using `git commit (-s|--signoff)`. See + [SubmittingPatches](https://github.com/git/git/blob/v2.8.1/Documentation/SubmittingPatches#L234-L286) + for more details about what this sign-off means. +* Check for whitespace errors using `git diff --check [base]...HEAD` or `git log --check`. +* Run `git rebase --whitespace=fix` to correct upstream issues with whitespace. +* Become familiar with interactive rebase (`git rebase -i`) because you will be reordering, + squashing, and editing commits as your patch or series of patches is reviewed. +* Make sure any shell scripts that you add have the executable bit set on them. This is + usually for test files that you add in the `/t` directory. You can use + `git add --chmod=+x [file]` to update it. You can test whether a file is marked as executable + using `git ls-files --stage \*.sh`; the first number is 100755 for executable files. +* Your commit titles should match the "area: change description" format. Rules of thumb: + * Choose ": " prefix appropriately. + * Keep the description short and to the point. + * The word that follows the ": " prefix is not capitalized. + * Do not include a full-stop at the end of the title. + * Read a few commit messages -- using `git log origin/master`, for instance -- to + become acquainted with the preferred commit message style. +* Build source using `make DEVELOPER=1` for extra-strict compiler warnings. + +Submit Your Patch +----------------- + +Git for Windows [accepts pull requests on GitHub](https://github.com/git-for-windows/git/pulls), but +these are reserved for Windows-specific improvements. For core Git, submissions are accepted on +[the Git mailing list](https://public-inbox.org/git). + +### Configure Git to Send Emails + +There are a bunch of options for configuring the `git send-email` command. These options can +be found in the documentation for +[`git config`](https://git-scm.com/docs/git-config) and +[`git send-email`](https://git-scm.com/docs/git-send-email). + +``` +git config --global sendemail.smtpserver +git config --global sendemail.smtpserverport 587 +git config --global sendemail.smtpencryption tls +git config --global sendemail.smtpuser +``` + +To avoid storing your password in the config file, store it in the Git credential manager: + +``` +$ git credential fill +protocol=smtp +host= +username= +password=password +``` + +Before submitting a patch, read the [Git documentation on submitting patches](https://github.com/git/git/blob/master/Documentation/SubmittingPatches). + +To construct a patch set, use the `git format-patch` command. There are three important options: + +* `--cover-letter`: If specified, create a `[v#-]0000-cover-letter.patch` file that can be + edited to describe the patch as a whole. If you previously added a branch description using + `git branch --edit-description`, you will end up with a 0/N mail with that description and + a nice overall diffstat. +* `--in-reply-to=[Message-ID]`: This will mark your cover letter as replying to the given + message (which should correspond to your previous iteration). To determine the correct Message-ID, + find the message you are replying to on [public-inbox.org/git](https://public-inbox.org/git) and take + the ID from between the angle brackets. + +* `--subject-prefix=[prefix]`: This defaults to [PATCH]. For subsequent iterations, you will want to + override it like `--subject-prefix="[PATCH v2]"`. You can also use the `-v` option to have it + automatically generate the version number in the patches. + +If you have multiple commits and use the `--cover-letter` option be sure to open the +`0000-cover-letter.patch` file to update the subject and add some details about the overall purpose +of the patch series. + +### Examples + +To generate a single commit patch file: +``` +git format-patch -s -o [dir] -1 +``` +To generate four patch files from the last three commits with a cover letter: +``` +git format-patch --cover-letter -s -o [dir] HEAD~4 +``` +To generate version 3 with four patch files from the last four commits with a cover letter: +``` +git format-patch --cover-letter -s -o [dir] -v 3 HEAD~4 +``` + +### Submit the Patch + +Run [`git send-email`](https://git-scm.com/docs/git-send-email), starting with a test email: + +``` +git send-email --to=yourself@address.com [dir with patches]/*.patch +``` + +After checking the receipt of your test email, you can send to the list and to any +potentially interested reviewers. + +``` +git send-email --to=git@vger.kernel.org --cc= --cc= [dir with patches]/*.patch +``` + +To submit a nth version patch (say version 3): + +``` +git send-email --to=git@vger.kernel.org --cc= --cc= \ + --in-reply-to= [dir with patches]/*.patch +``` diff --git a/Documentation/config.txt b/Documentation/config.txt index 5b5b9765699933..98f697b52d47ca 100644 --- a/Documentation/config.txt +++ b/Documentation/config.txt @@ -503,6 +503,8 @@ include::config/safe.txt[] include::config/sendemail.txt[] +include::config/sendpack.txt[] + include::config/sequencer.txt[] include::config/showbranch.txt[] @@ -539,4 +541,6 @@ include::config/versionsort.txt[] include::config/web.txt[] +include::config/windows.txt[] + include::config/worktree.txt[] diff --git a/Documentation/config/advice.txt b/Documentation/config/advice.txt index a00d0100a82ba7..72776b4cf36b68 100644 --- a/Documentation/config/advice.txt +++ b/Documentation/config/advice.txt @@ -118,6 +118,9 @@ advice.*:: waitingForEditor:: Print a message to the terminal whenever Git is waiting for editor input from the user. + nameTooLong:: + Advice shown if a filepath operation is attempted where the + path was too long. nestedTag:: Advice shown if a user attempts to recursively tag a tag object. submoduleAlternateErrorStrategyDie:: @@ -136,4 +139,7 @@ advice.*:: Advice shown when either linkgit:git-add[1] or linkgit:git-rm[1] is asked to update index entries outside the current sparse checkout. + useCoreFSMonitorConfig:: + Advice shown if the deprecated 'core.useBuiltinFSMonitor' config + setting is in use. -- diff --git a/Documentation/config/core.txt b/Documentation/config/core.txt index 37afbaf5a419d1..d52fc67af13efc 100644 --- a/Documentation/config/core.txt +++ b/Documentation/config/core.txt @@ -670,6 +670,19 @@ relatively high IO latencies. When enabled, Git will do the index comparison to the filesystem data in parallel, allowing overlapping IO's. Defaults to true. +core.fscache:: + Enable additional caching of file system data for some operations. ++ +Git for Windows uses this to bulk-read and cache lstat data of entire +directories (instead of doing lstat file by file). + +core.longpaths:: + Enable long path (> 260) support for builtin commands in Git for + Windows. This is disabled by default, as long paths are not supported + by Windows Explorer, cmd.exe and the Git for Windows tool chain + (msys, bash, tcl, perl...). Only enable this if you know what you're + doing and are prepared to live with a few quirks. + core.unsetenvvars:: Windows-only: comma-separated list of environment variables' names that need to be unset before spawning any other process. diff --git a/Documentation/config/http.txt b/Documentation/config/http.txt index afeeccfbfa7a70..c9aa1428556ed2 100644 --- a/Documentation/config/http.txt +++ b/Documentation/config/http.txt @@ -189,11 +189,13 @@ http.sslBackend:: http.schannelCheckRevoke:: Used to enforce or disable certificate revocation checks in cURL - when http.sslBackend is set to "schannel". Defaults to `true` if - unset. Only necessary to disable this if Git consistently errors - and the message is about checking the revocation status of a - certificate. This option is ignored if cURL lacks support for - setting the relevant SSL option at runtime. + when http.sslBackend is set to "schannel" via "true" and "false", + respectively. Another accepted value is "best-effort" (the default) + in which case revocation checks are performed, but errors due to + revocation list distribution points that are offline are silently + ignored, as well as errors due to certificates missing revocation + list distribution points. This option is ignored if cURL lacks + support for setting the relevant SSL option at runtime. http.schannelUseSSLCAInfo:: As of cURL v7.60.0, the Secure Channel backend can use the @@ -203,6 +205,11 @@ http.schannelUseSSLCAInfo:: when the `schannel` backend was configured via `http.sslBackend`, unless `http.schannelUseSSLCAInfo` overrides this behavior. +http.sslAutoClientCert:: + As of cURL v7.77.0, the Secure Channel backend won't automatically + send client certificates from the Windows Certificate Store anymore. + To opt in to the old behavior, http.sslAutoClientCert can be set. + http.pinnedPubkey:: Public key of the https service. It may either be the filename of a PEM or DER encoded public key file or a string starting with diff --git a/Documentation/config/sendpack.txt b/Documentation/config/sendpack.txt new file mode 100644 index 00000000000000..e306f657fba7dd --- /dev/null +++ b/Documentation/config/sendpack.txt @@ -0,0 +1,5 @@ +sendpack.sideband:: + Allows to disable the side-band-64k capability for send-pack even + when it is advertised by the server. Makes it possible to work + around a limitation in the git for windows implementation together + with the dump git protocol. Defaults to true. diff --git a/Documentation/config/windows.txt b/Documentation/config/windows.txt new file mode 100644 index 00000000000000..fdaaf1c65504f3 --- /dev/null +++ b/Documentation/config/windows.txt @@ -0,0 +1,4 @@ +windows.appendAtomically:: + By default, append atomic API is used on windows. But it works only with + local disk files, if you're working on a network file system, you should + set it false to turn it off. diff --git a/Documentation/git-reset.txt b/Documentation/git-reset.txt index 01cb4c9b9c56a7..6a44de12788439 100644 --- a/Documentation/git-reset.txt +++ b/Documentation/git-reset.txt @@ -12,6 +12,7 @@ SYNOPSIS 'git reset' [-q] [--pathspec-from-file= [--pathspec-file-nul]] [] 'git reset' (--patch | -p) [] [--] [...] 'git reset' [--soft | --mixed [-N] | --hard | --merge | --keep] [-q] [] +DEPRECATED: 'git reset' [-q] [--stdin [-z]] [] DESCRIPTION ----------- @@ -132,6 +133,16 @@ OPTIONS + For more details, see the 'pathspec' entry in linkgit:gitglossary[7]. +--stdin:: + DEPRECATED (use `--pathspec-from-file=-` instead): Instead of taking + list of paths from the command line, read list of paths from the + standard input. Paths are separated by LF (i.e. one path per line) by + default. + +-z:: + DEPRECATED (use `--pathspec-file-nul` instead): Only meaningful with + `--stdin`; paths are separated with NUL character instead of LF. + EXAMPLES -------- diff --git a/Documentation/git.txt b/Documentation/git.txt index 0c15ef3a8e6a51..455eaf0489a226 100644 --- a/Documentation/git.txt +++ b/Documentation/git.txt @@ -460,6 +460,14 @@ Environment Variables --------------------- Various Git commands use the following environment variables: +System +~~~~~~ +`HOME`:: + Specifies the path to the user's home directory. On Windows, if + unset, Git will set a process environment variable equal to: + `$HOMEDRIVE$HOMEPATH` if both `$HOMEDRIVE` and `$HOMEPATH` exist; + otherwise `$USERPROFILE` if `$USERPROFILE` exists. + The Git Repository ~~~~~~~~~~~~~~~~~~ These environment variables apply to 'all' core Git commands. Nb: it diff --git a/Documentation/gitattributes.txt b/Documentation/gitattributes.txt index 4b36d51beb66f0..85faf9c1429485 100644 --- a/Documentation/gitattributes.txt +++ b/Documentation/gitattributes.txt @@ -385,6 +385,36 @@ sign `$` upon checkout. Any byte sequence that begins with with `$Id$` upon check-in. +`symlink` +^^^^^^^^^ + +On Windows, symbolic links have a type: a "file symlink" must point at +a file, and a "directory symlink" must point at a directory. If the +type of symlink does not match its target, it doesn't work. + +Git does not record the type of symlink in the index or in a tree. On +checkout it'll guess the type, which only works if the target exists +at the time the symlink is created. This may often not be the case, +for example when the link points at a directory inside a submodule. + +The `symlink` attribute allows you to explicitly set the type of symlink +to `file` or `dir`, so Git doesn't have to guess. If you have a set of +symlinks that point at other files, you can do: + +------------------------ +*.gif symlink=file +------------------------ + +To tell Git that a symlink points at a directory, use: + +------------------------ +tools_folder symlink=dir +------------------------ + +The `symlink` attribute is ignored on platforms other than Windows, +since they don't distinguish between different types of symlinks. + + `filter` ^^^^^^^^ diff --git a/Makefile b/Makefile index cac3452edb90b4..746d0dabcf3687 100644 --- a/Makefile +++ b/Makefile @@ -722,6 +722,7 @@ TEST_BUILTINS_OBJS += test-advise.o TEST_BUILTINS_OBJS += test-bitmap.o TEST_BUILTINS_OBJS += test-bloom.o TEST_BUILTINS_OBJS += test-chmtime.o +TEST_BUILTINS_OBJS += test-cmp.o TEST_BUILTINS_OBJS += test-config.o TEST_BUILTINS_OBJS += test-crontab.o TEST_BUILTINS_OBJS += test-csprng.o @@ -744,6 +745,7 @@ TEST_BUILTINS_OBJS += test-hash-speed.o TEST_BUILTINS_OBJS += test-hash.o TEST_BUILTINS_OBJS += test-hashmap.o TEST_BUILTINS_OBJS += test-hexdump.o +TEST_BUILTINS_OBJS += test-iconv.o TEST_BUILTINS_OBJS += test-index-version.o TEST_BUILTINS_OBJS += test-json-writer.o TEST_BUILTINS_OBJS += test-lazy-init-name-hash.o @@ -1259,6 +1261,7 @@ BUILTIN_OBJS += builtin/write-tree.o # upstream unnecessarily (making merging in future changes easier). THIRD_PARTY_SOURCES += compat/inet_ntop.c THIRD_PARTY_SOURCES += compat/inet_pton.c +THIRD_PARTY_SOURCES += compat/mimalloc/% THIRD_PARTY_SOURCES += compat/nedmalloc/% THIRD_PARTY_SOURCES += compat/obstack.% THIRD_PARTY_SOURCES += compat/poll/% @@ -1338,6 +1341,7 @@ BASIC_CFLAGS += -DSHA1DC_FORCE_ALIGNED_ACCESS endif ifneq ($(filter leak,$(SANITIZERS)),) BASIC_CFLAGS += -DSUPPRESS_ANNOTATED_LEAKS +BASIC_CFLAGS += -O0 SANITIZE_LEAK = YesCompiledWithIt endif ifneq ($(filter address,$(SANITIZERS)),) @@ -1941,6 +1945,39 @@ ifdef USE_NED_ALLOCATOR OVERRIDE_STRDUP = YesPlease endif +ifdef USE_MIMALLOC + MIMALLOC_OBJS = \ + compat/mimalloc/alloc-aligned.o \ + compat/mimalloc/alloc.o \ + compat/mimalloc/arena.o \ + compat/mimalloc/bitmap.o \ + compat/mimalloc/heap.o \ + compat/mimalloc/init.o \ + compat/mimalloc/options.o \ + compat/mimalloc/os.o \ + compat/mimalloc/page.o \ + compat/mimalloc/random.o \ + compat/mimalloc/segment.o \ + compat/mimalloc/segment-cache.o \ + compat/mimalloc/stats.o + + COMPAT_CFLAGS += -Icompat/mimalloc -DMI_DEBUG=0 -DUSE_MIMALLOC --std=gnu11 + COMPAT_OBJS += $(MIMALLOC_OBJS) + +$(MIMALLOC_OBJS): COMPAT_CFLAGS += -DBANNED_H + +ifdef DEVELOPER +$(MIMALLOC_OBJS): COMPAT_CFLAGS += \ + -Wno-attributes \ + -Wno-pedantic \ + -Wno-unknown-pragmas \ + -Wno-declaration-after-statement \ + -Wno-old-style-definition \ + -Wno-missing-prototypes \ + -Wno-array-bounds +endif +endif + ifdef OVERRIDE_STRDUP COMPAT_CFLAGS += -DOVERRIDE_STRDUP COMPAT_OBJS += compat/strdup.o @@ -2665,6 +2702,13 @@ compat/nedmalloc/nedmalloc.sp compat/nedmalloc/nedmalloc.o: EXTRA_CPPFLAGS = \ compat/nedmalloc/nedmalloc.sp: SP_EXTRA_FLAGS += -Wno-non-pointer-null endif +headless-git.o: compat/win32/headless.c GIT-CFLAGS + $(QUIET_CC)$(CC) $(ALL_CFLAGS) $(COMPAT_CFLAGS) \ + -fno-stack-protector -o $@ -c -Wall -Wwrite-strings $< + +headless-git$X: headless-git.o git.res GIT-LDFLAGS + $(QUIET_LINK)$(CC) $(ALL_CFLAGS) $(ALL_LDFLAGS) -mwindows -o $@ $< git.res + git-%$X: %.o GIT-LDFLAGS $(GITLIBS) $(QUIET_LINK)$(CC) $(ALL_CFLAGS) -o $@ $(ALL_LDFLAGS) $(filter %.o,$^) $(LIBS) @@ -3038,6 +3082,7 @@ else @echo RUNTIME_PREFIX=\'false\' >>$@+ endif @if cmp $@+ $@ >/dev/null 2>&1; then $(RM) $@+; else mv $@+ $@; fi + @if test -f GIT-BUILD-DIR; then rm GIT-BUILD-DIR; fi ### Detect Python interpreter path changes ifndef NO_PYTHON @@ -3454,6 +3499,7 @@ clean: profile-clean coverage-clean cocciclean $(RM) po/git.pot po/git-core.pot $(RM) git.res $(RM) $(OBJECTS) + $(RM) headless-git.o $(RM) $(LIB_FILE) $(XDIFF_LIB) $(REFTABLE_LIB) $(REFTABLE_TEST_LIB) $(RM) $(ALL_PROGRAMS) $(SCRIPT_LIB) $(BUILT_INS) $(OTHER_PROGRAMS) $(RM) $(TEST_PROGRAMS) @@ -3482,13 +3528,17 @@ endif $(RM) GIT-SCRIPT-DEFINES GIT-PERL-DEFINES GIT-PERL-HEADER GIT-PYTHON-VARS ifdef MSVC $(RM) $(patsubst %.o,%.o.pdb,$(OBJECTS)) + $(RM) headless-git.o.pdb $(RM) $(patsubst %.exe,%.pdb,$(OTHER_PROGRAMS)) + $(RM) $(patsubst %.exe,%.ilk,$(OTHER_PROGRAMS)) $(RM) $(patsubst %.exe,%.iobj,$(OTHER_PROGRAMS)) $(RM) $(patsubst %.exe,%.ipdb,$(OTHER_PROGRAMS)) $(RM) $(patsubst %.exe,%.pdb,$(PROGRAMS)) + $(RM) $(patsubst %.exe,%.ilk,$(PROGRAMS)) $(RM) $(patsubst %.exe,%.iobj,$(PROGRAMS)) $(RM) $(patsubst %.exe,%.ipdb,$(PROGRAMS)) $(RM) $(patsubst %.exe,%.pdb,$(TEST_PROGRAMS)) + $(RM) $(patsubst %.exe,%.ilk,$(TEST_PROGRAMS)) $(RM) $(patsubst %.exe,%.iobj,$(TEST_PROGRAMS)) $(RM) $(patsubst %.exe,%.ipdb,$(TEST_PROGRAMS)) $(RM) compat/vcbuild/MSVC-DEFS-GEN diff --git a/README.md b/README.md index 7ce4f05bae8120..e2314c5a313c08 100644 --- a/README.md +++ b/README.md @@ -1,4 +1,76 @@ -[![Build status](https://github.com/git/git/workflows/CI/badge.svg)](https://github.com/git/git/actions?query=branch%3Amaster+event%3Apush) +Git for Windows +=============== + +[![Open in Visual Studio Code](https://img.shields.io/static/v1?logo=visualstudiocode&label=&message=Open%20in%20Visual%20Studio%20Code&labelColor=2c2c32&color=007acc&logoColor=007acc)](https://open.vscode.dev/git-for-windows/git) +[![Build status](https://github.com/git-for-windows/git/workflows/CI/badge.svg)](https://github.com/git-for-windows/git/actions?query=branch%3Amain+event%3Apush) +[![Join the chat at https://gitter.im/git-for-windows/git](https://badges.gitter.im/Join%20Chat.svg)](https://gitter.im/git-for-windows/git?utm_source=badge&utm_medium=badge&utm_campaign=pr-badge&utm_content=badge) + +This is [Git for Windows](http://git-for-windows.github.io/), the Windows port +of [Git](http://git-scm.com/). + +The Git for Windows project is run using a [governance +model](http://git-for-windows.github.io/governance-model.html). If you +encounter problems, you can report them as [GitHub +issues](https://github.com/git-for-windows/git/issues), discuss them on Git +for Windows' [Google Group](http://groups.google.com/group/git-for-windows), +and [contribute bug +fixes](https://github.com/git-for-windows/git/wiki/How-to-participate). + +To build Git for Windows, please either install [Git for Windows' +SDK](https://gitforwindows.org/#download-sdk), start its `git-bash.exe`, `cd` +to your Git worktree and run `make`, or open the Git worktree as a folder in +Visual Studio. + +To verify that your build works, use one of the following methods: + +- If you want to test the built executables within Git for Windows' SDK, + prepend `/bin-wrappers` to the `PATH`. +- Alternatively, run `make install` in the Git worktree. +- If you need to test this in a full installer, run `sdk build + git-and-installer`. +- You can also "install" Git into an existing portable Git via `make install + DESTDIR=` where `` refers to the top-level directory of the + portable Git. In this instance, you will want to prepend that portable Git's + `/cmd` directory to the `PATH`, or test by running that portable Git's + `git-bash.exe` or `git-cmd.exe`. +- If you built using a recent Visual Studio, you can use the menu item + `Build>Install git` (you will want to click on `Project>CMake Settings for + Git` first, then click on `Edit JSON` and then point `installRoot` to the + `mingw64` directory of an already-unpacked portable Git). + + As in the previous bullet point, you will then prepend `/cmd` to the `PATH` + or run using the portable Git's `git-bash.exe` or `git-cmd.exe`. +- If you want to run the built executables in-place, but in a CMD instead of + inside a Bash, you can run a snippet like this in the `git-bash.exe` window + where Git was built (ensure that the `EOF` line has no leading spaces), and + then paste into the CMD window what was put in the clipboard: + + ```sh + clip.exe <, and other archival sites. +The core git mailing list is plain text (no HTML!). Issues which are security relevant should be disclosed privately to the Git Security mailing list . diff --git a/SECURITY.md b/SECURITY.md index c720c2ae7f9580..328351684298a4 100644 --- a/SECURITY.md +++ b/SECURITY.md @@ -28,24 +28,38 @@ Examples for details to include: ## Supported Versions -There are no official "Long Term Support" versions in Git. -Instead, the maintenance track (i.e. the versions based on the -most recently published feature release, also known as ".0" -version) sees occasional updates with bug fixes. - -Fixes to vulnerabilities are made for the maintenance track for -the latest feature release and merged up to the in-development -branches. The Git project makes no formal guarantee for any -older maintenance tracks to receive updates. In practice, -though, critical vulnerability fixes are applied not only to the -most recent track, but to at least a couple more maintenance -tracks. - -This is typically done by making the fix on the oldest and still -relevant maintenance track, and merging it upwards to newer and -newer maintenance tracks. - -For example, v2.24.1 was released to address a couple of -[CVEs](https://cve.mitre.org/), and at the same time v2.14.6, -v2.15.4, v2.16.6, v2.17.3, v2.18.2, v2.19.3, v2.20.2, v2.21.1, -v2.22.2 and v2.23.1 were released. +Git for Windows is a "friendly fork" of [Git](https://git-scm.com/), i.e. changes in Git for Windows are frequently contributed back, and Git for Windows' release cycle closely following Git's. + +While Git maintains several release trains (when v2.19.1 was released, there were updates to v2.14.x-v2.18.x, too, for example), Git for Windows follows only the latest Git release. For example, there is no Git for Windows release corresponding to Git v2.16.5 (which was released after v2.19.0). + +One exception is [MinGit for Windows](https://github.com/git-for-windows/git/wiki/MinGit) (a minimal subset of Git for Windows, intended for bundling with third-party applications that do not need any interactive commands nor support for `git svn`): critical security fixes are backported to the v2.11.x, v2.14.x, v2.19.x, v2.21.x and v2.23.x release trains. + +## Version number scheme + +The Git for Windows versions reflect the Git version on which they are based. For example, Git for Windows v2.21.0 is based on Git v2.21.0. + +As Git for Windows bundles more than just Git (such as Bash, OpenSSL, OpenSSH, GNU Privacy Guard), sometimes there are interim releases without corresponding Git releases. In these cases, Git for Windows appends a number in parentheses, starting with the number 2, then 3, etc. For example, both Git for Windows v2.17.1 and v2.17.1(2) were based on Git v2.17.1, but the latter included updates for Git Credential Manager and Git LFS, fixing critical regressions. + +## Tag naming scheme + +Every Git for Windows version is tagged using a name that starts with the Git version on which it is based, with the suffix `.windows.` appended. For example, Git for Windows v2.17.1' source code is tagged as [`v2.17.1.windows.1`](https://github.com/git-for-windows/git/releases/tag/v2.17.1.windows.1) (the patch level is always at least 1, given that Git for Windows always has patches on top of Git). Likewise, Git for Windows v2.17.1(2)' source code is tagged as [`v2.17.1.windows.2`](https://github.com/git-for-windows/git/releases/tag/v2.17.1.windows.2). + +## Release Candidate (rc) versions + +As a friendly fork of Git (the "upstream" project), Git for Windows is closely corelated to that project. + +Consequently, Git for Windows publishes versions based on Git's release candidates (for upcoming "`.0`" versions, see [Git's release schedule](https://tinyurl.com/gitCal)). These versions end in `-rc`, starting with `-rc0` for a very early preview of what is to come, and as with regular versions, Git for Windows tries to follow Git's releases as quickly as possible. + +Note: there is currently a bug in the "Check daily for updates" code, where it mistakes the final version as a downgrade from release candidates. Example: if you installed Git for Windows v2.23.0-rc3 and enabled the auto-updater, it would ask you whether you want to "downgrade" to v2.23.0 when that version was available. + +[All releases](https://github.com/git-for-windows/git/releases/), including release candidates, are listed via a link at the footer of the [Git for Windows](https://gitforwindows.org/) home page. + +## Snapshot versions ('nightly builds') + +Git for Windows also provides snapshots (these are not releases) of the the current development as per git-for-Windows/git's `master` branch at the [Snapshots](https://wingit.blob.core.windows.net/files/index.html) page. This link is also listed in the footer of the [Git for Windows](https://gitforwindows.org/) home page. + +Note: even if those builds are not exactly "nightly", they are sometimes referred to as "nightly builds" to keep with other projects' nomenclature. + +## Following upstream's developments + +The [gitforwindows/git repository](https://github.com/git-for-windows/git) also provides the `shears/*` branches. The `shears/*` branches reflect Git for Windows' patches, rebased onto the upstream integration branches, [updated (mostly) via automated CI builds](https://dev.azure.com/git-for-windows/git/_build?definitionId=25). diff --git a/abspath.c b/abspath.c index 39e06b58486e3e..7a5f114382f182 100644 --- a/abspath.c +++ b/abspath.c @@ -91,6 +91,9 @@ static char *strbuf_realpath_1(struct strbuf *resolved, const char *path, goto error_out; } + if (platform_strbuf_realpath(resolved, path)) + return resolved->buf; + strbuf_addstr(&remaining, path); get_root_part(resolved, &remaining); diff --git a/advice.c b/advice.c index fd189689437c75..76c553a71fb419 100644 --- a/advice.c +++ b/advice.c @@ -48,6 +48,7 @@ static struct { [ADVICE_GRAFT_FILE_DEPRECATED] = { "graftFileDeprecated", 1 }, [ADVICE_IGNORED_HOOK] = { "ignoredHook", 1 }, [ADVICE_IMPLICIT_IDENTITY] = { "implicitIdentity", 1 }, + [ADVICE_NAME_TOO_LONG] = { "nameTooLong", 1 }, [ADVICE_NESTED_TAG] = { "nestedTag", 1 }, [ADVICE_OBJECT_NAME_WARNING] = { "objectNameWarning", 1 }, [ADVICE_PUSH_ALREADY_EXISTS] = { "pushAlreadyExists", 1 }, @@ -74,6 +75,7 @@ static struct { [ADVICE_SUBMODULE_ALTERNATE_ERROR_STRATEGY_DIE] = { "submoduleAlternateErrorStrategyDie", 1 }, [ADVICE_SUBMODULES_NOT_UPDATED] = { "submodulesNotUpdated", 1 }, [ADVICE_UPDATE_SPARSE_PATH] = { "updateSparsePath", 1 }, + [ADVICE_USE_CORE_FSMONITOR_CONFIG] = { "useCoreFSMonitorConfig", 1 }, [ADVICE_WAITING_FOR_EDITOR] = { "waitingForEditor", 1 }, }; diff --git a/advice.h b/advice.h index 07e0f76833e780..bd0a2add268428 100644 --- a/advice.h +++ b/advice.h @@ -26,6 +26,7 @@ struct string_list; ADVICE_GRAFT_FILE_DEPRECATED, ADVICE_IGNORED_HOOK, ADVICE_IMPLICIT_IDENTITY, + ADVICE_NAME_TOO_LONG, ADVICE_NESTED_TAG, ADVICE_OBJECT_NAME_WARNING, ADVICE_PUSH_ALREADY_EXISTS, @@ -48,6 +49,7 @@ struct string_list; ADVICE_SUBMODULE_ALTERNATE_ERROR_STRATEGY_DIE, ADVICE_SUBMODULES_NOT_UPDATED, ADVICE_UPDATE_SPARSE_PATH, + ADVICE_USE_CORE_FSMONITOR_CONFIG, ADVICE_WAITING_FOR_EDITOR, ADVICE_SKIPPED_CHERRY_PICKS, }; diff --git a/apply.c b/apply.c index 2b7cd930efa3bd..f86a01b26080dd 100644 --- a/apply.c +++ b/apply.c @@ -4366,7 +4366,7 @@ static int try_create_file(struct apply_state *state, const char *path, /* Although buf:size is counted string, it also is NUL * terminated. */ - return !!symlink(buf, path); + return !!create_symlink(state && state->repo ? state->repo->index : NULL, buf, path); fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666); if (fd < 0) diff --git a/azure-pipelines.yml b/azure-pipelines.yml new file mode 100644 index 00000000000000..fed26341d49a4a --- /dev/null +++ b/azure-pipelines.yml @@ -0,0 +1,393 @@ +variables: + Agent.Source.Git.ShallowFetchDepth: 1 + GIT_CONFIG_PARAMETERS: "'checkout.workers=56' 'user.name=CI' 'user.email=ci@git'" + +jobs: +- job: windows_build + displayName: Windows Build + condition: succeeded() + pool: + vmImage: windows-latest + timeoutInMinutes: 240 + steps: + - bash: git clone --bare --depth=1 --filter=blob:none --single-branch -b main https://github.com/git-for-windows/git-sdk-64 + displayName: 'clone git-sdk-64' + - bash: git clone --depth=1 --single-branch -b main https://github.com/git-for-windows/build-extra + displayName: 'clone build-extra' + - bash: sh -x ./build-extra/please.sh create-sdk-artifact --sdk=git-sdk-64.git --out=git-sdk-64-minimal minimal-sdk + displayName: 'build git-sdk-64-minimal-sdk' + - bash: | + # Let Git ignore the SDK and the test-cache + printf "%s\n" /git-sdk-64.git/ /build-extra/ /git-sdk-64-minimal/ /test-cache/ >>'.git/info/exclude' + displayName: 'Ignore untracked directories' + - bash: ci/make-test-artifacts.sh artifacts + displayName: Build + env: + HOME: $(Build.SourcesDirectory) + MSYSTEM: MINGW64 + DEVELOPER: 1 + NO_PERL: 1 + PATH: "$(Build.SourcesDirectory)\\git-sdk-64-minimal\\mingw64\\bin;$(Build.SourcesDirectory)\\git-sdk-64-minimal\\usr\\bin;C:\\Windows\\system32;C:\\Windows;C:\\Windows\\system32\\wbem" + - task: PublishPipelineArtifact@0 + displayName: 'Publish Pipeline Artifact: test artifacts' + inputs: + artifactName: 'windows-artifacts' + targetPath: '$(Build.SourcesDirectory)\artifacts' + - task: PublishPipelineArtifact@0 + displayName: 'Publish Pipeline Artifact: git-sdk-64-minimal' + inputs: + artifactName: 'git-sdk-64-minimal' + targetPath: '$(Build.SourcesDirectory)\git-sdk-64-minimal' + +- job: windows_test + displayName: Windows Test + dependsOn: windows_build + condition: succeeded() + pool: + vmImage: windows-latest + timeoutInMinutes: 240 + strategy: + parallel: 10 + steps: + - task: DownloadPipelineArtifact@0 + displayName: 'Download Pipeline Artifact: test artifacts' + inputs: + artifactName: 'windows-artifacts' + targetPath: '$(Build.SourcesDirectory)' + - task: DownloadPipelineArtifact@0 + displayName: 'Download Pipeline Artifact: git-sdk-64-minimal' + inputs: + artifactName: 'git-sdk-64-minimal' + targetPath: '$(Build.SourcesDirectory)\git-sdk-64-minimal' + - bash: | + test -f artifacts.tar.gz || { + echo No test artifacts found\; skipping >&2 + exit 0 + } + tar xf artifacts.tar.gz || exit 1 + + # Let Git ignore the SDK and the test-cache + printf '%s\n' /git-sdk-64.git/ /build-extra/ /git-sdk-64-minimal/ /test-cache/ >>.git/info/exclude + + ci/run-test-slice.sh $SYSTEM_JOBPOSITIONINPHASE $SYSTEM_TOTALJOBSINPHASE || { + ci/print-test-failures.sh + exit 1 + } + displayName: 'Test (parallel)' + env: + HOME: $(Build.SourcesDirectory) + MSYSTEM: MINGW64 + NO_SVN_TESTS: 1 + GIT_TEST_SKIP_REBASE_P: 1 + PATH: "$(Build.SourcesDirectory)\\git-sdk-64-minimal\\mingw64\\bin;$(Build.SourcesDirectory)\\git-sdk-64-minimal\\usr\\bin\\core_perl;$(Build.SourcesDirectory)\\git-sdk-64-minimal\\usr\\bin;C:\\Windows\\system32;C:\\Windows;C:\\Windows\\system32\\wbem" + - task: PublishTestResults@2 + displayName: 'Publish Test Results **/TEST-*.xml' + inputs: + mergeTestResults: true + testRunTitle: 'windows' + platform: Windows + publishRunAttachments: false + condition: succeededOrFailed() + - task: PublishBuildArtifacts@1 + displayName: 'Publish trash directories of failed tests' + condition: failed() + inputs: + PathtoPublish: t/failed-test-artifacts + ArtifactName: failed-test-artifacts + +- job: vs_build + displayName: Visual Studio Build + condition: succeeded() + pool: + vmImage: windows-latest + timeoutInMinutes: 240 + steps: + - bash: git clone --bare --depth=1 --filter=blob:none --single-branch -b main https://github.com/git-for-windows/git-sdk-64 + displayName: 'clone git-sdk-64' + - bash: git clone --depth=1 --single-branch -b main https://github.com/git-for-windows/build-extra + displayName: 'clone build-extra' + - bash: sh -x ./build-extra/please.sh create-sdk-artifact --sdk=git-sdk-64.git --out=git-sdk-64-minimal minimal-sdk + displayName: 'build git-sdk-64-minimal-sdk' + - bash: | + # Let Git ignore the SDK and the test-cache + printf "%s\n" /git-sdk-64-minimal/ /test-cache/ >>'.git/info/exclude' + displayName: 'Ignore untracked directories' + - bash: make NDEBUG=1 DEVELOPER=1 vcxproj + displayName: Generate Visual Studio Solution + env: + HOME: $(Build.SourcesDirectory) + MSYSTEM: MINGW64 + DEVELOPER: 1 + NO_PERL: 1 + PATH: "$(Build.SourcesDirectory)\\git-sdk-64-minimal\\mingw64\\bin;$(Build.SourcesDirectory)\\git-sdk-64-minimal\\usr\\bin;C:\\Windows\\system32;C:\\Windows;C:\\Windows\\system32\\wbem" + - powershell: | + $urlbase = "https://dev.azure.com/git/git/_apis/build/builds" + $id = ((Invoke-WebRequest -UseBasicParsing "${urlbase}?definitions=9&statusFilter=completed&resultFilter=succeeded&`$top=1").content | ConvertFrom-JSON).value[0].id + $downloadUrl = ((Invoke-WebRequest -UseBasicParsing "${urlbase}/$id/artifacts").content | ConvertFrom-JSON).value[0].resource.downloadUrl + (New-Object Net.WebClient).DownloadFile($downloadUrl, "compat.zip") + Expand-Archive compat.zip -DestinationPath . -Force + Remove-Item compat.zip + displayName: 'Download vcpkg artifacts' + - task: MSBuild@1 + inputs: + solution: git.sln + platform: x64 + configuration: Release + maximumCpuCount: 4 + msbuildArguments: /p:PlatformToolset=v142 + - bash: | + ./compat/vcbuild/vcpkg_copy_dlls.bat release && + mkdir -p artifacts && + eval "$(make -n artifacts-tar INCLUDE_DLLS_IN_ARTIFACTS=YesPlease ARTIFACTS_DIRECTORY=artifacts | grep ^tar)" + displayName: Bundle artifact tar + env: + HOME: $(Build.SourcesDirectory) + MSYSTEM: MINGW64 + DEVELOPER: 1 + NO_PERL: 1 + MSVC: 1 + VCPKG_ROOT: $(Build.SourcesDirectory)\compat\vcbuild\vcpkg + PATH: "$(Build.SourcesDirectory)\\git-sdk-64-minimal\\mingw64\\bin;$(Build.SourcesDirectory)\\git-sdk-64-minimal\\usr\\bin;C:\\Windows\\system32;C:\\Windows;C:\\Windows\\system32\\wbem" + - powershell: | + $tag = (Invoke-WebRequest -UseBasicParsing "https://gitforwindows.org/latest-tag.txt").content + $version = (Invoke-WebRequest -UseBasicParsing "https://gitforwindows.org/latest-version.txt").content + $url = "https://github.com/git-for-windows/git/releases/download/${tag}/PortableGit-${version}-64-bit.7z.exe" + (New-Object Net.WebClient).DownloadFile($url,"PortableGit.exe") + & .\PortableGit.exe -y -oartifacts\PortableGit + # Wait until it is unpacked + while (-not @(Remove-Item -ErrorAction SilentlyContinue PortableGit.exe; $?)) { sleep 1 } + displayName: Download & extract portable Git + - task: PublishPipelineArtifact@0 + displayName: 'Publish Pipeline Artifact: MSVC test artifacts' + inputs: + artifactName: 'vs-artifacts' + targetPath: '$(Build.SourcesDirectory)\artifacts' + +- job: vs_test + displayName: Visual Studio Test + dependsOn: vs_build + condition: succeeded() + pool: + vmImage: windows-latest + timeoutInMinutes: 240 + strategy: + parallel: 10 + steps: + - task: DownloadPipelineArtifact@0 + displayName: 'Download Pipeline Artifact: VS test artifacts' + inputs: + artifactName: 'vs-artifacts' + targetPath: '$(Build.SourcesDirectory)' + - bash: | + test -f artifacts.tar.gz || { + echo No test artifacts found\; skipping >&2 + exit 0 + } + tar xf artifacts.tar.gz || exit 1 + + # Let Git ignore the SDK and the test-cache + printf '%s\n' /PortableGit/ /test-cache/ >>.git/info/exclude + + cd t && + PATH="$PWD/helper:$PATH" && + test-tool.exe run-command testsuite --jobs=10 -V -x --write-junit-xml \ + $(test-tool.exe path-utils slice-tests \ + $SYSTEM_JOBPOSITIONINPHASE $SYSTEM_TOTALJOBSINPHASE t[0-9]*.sh) + displayName: 'Test (parallel)' + env: + HOME: $(Build.SourcesDirectory) + MSYSTEM: MINGW64 + NO_SVN_TESTS: 1 + GIT_TEST_SKIP_REBASE_P: 1 + PATH: "$(Build.SourcesDirectory)\\PortableGit\\mingw64\\bin;$(Build.SourcesDirectory)\\PortableGit\\usr\\bin;C:\\Windows\\system32;C:\\Windows;C:\\Windows\\system32\\wbem" + - task: PublishTestResults@2 + displayName: 'Publish Test Results **/TEST-*.xml' + inputs: + mergeTestResults: true + testRunTitle: 'vs' + platform: Windows + publishRunAttachments: false + condition: succeededOrFailed() + - task: PublishBuildArtifacts@1 + displayName: 'Publish trash directories of failed tests' + condition: failed() + inputs: + PathtoPublish: t/failed-test-artifacts + ArtifactName: failed-vs-test-artifacts + +- job: linux_clang + displayName: linux-clang + condition: succeeded() + pool: + vmImage: ubuntu-latest + steps: + - bash: | + export CC=clang || exit 1 + + ci/install-dependencies.sh || exit 1 + ci/run-build-and-tests.sh || { + ci/print-test-failures.sh + exit 1 + } + displayName: 'ci/run-build-and-tests.sh' + - task: PublishTestResults@2 + displayName: 'Publish Test Results **/TEST-*.xml' + inputs: + mergeTestResults: true + testRunTitle: 'linux-clang' + platform: Linux + publishRunAttachments: false + condition: succeededOrFailed() + - task: PublishBuildArtifacts@1 + displayName: 'Publish trash directories of failed tests' + condition: failed() + inputs: + PathtoPublish: t/failed-test-artifacts + ArtifactName: failed-test-artifacts + +- job: linux_gcc + displayName: linux-gcc + condition: succeeded() + pool: + vmImage: ubuntu-latest + steps: + - bash: | + ci/install-dependencies.sh || exit 1 + ci/run-build-and-tests.sh || { + ci/print-test-failures.sh + exit 1 + } + displayName: 'ci/run-build-and-tests.sh' + - task: PublishTestResults@2 + displayName: 'Publish Test Results **/TEST-*.xml' + inputs: + mergeTestResults: true + testRunTitle: 'linux-gcc' + platform: Linux + publishRunAttachments: false + condition: succeededOrFailed() + - task: PublishBuildArtifacts@1 + displayName: 'Publish trash directories of failed tests' + condition: failed() + inputs: + PathtoPublish: t/failed-test-artifacts + ArtifactName: failed-test-artifacts + +- job: osx_clang + displayName: osx-clang + condition: succeeded() + pool: + vmImage: macOS-latest + steps: + - bash: | + export CC=clang + + ci/install-dependencies.sh || exit 1 + ci/run-build-and-tests.sh || { + ci/print-test-failures.sh + exit 1 + } + displayName: 'ci/run-build-and-tests.sh' + - task: PublishTestResults@2 + displayName: 'Publish Test Results **/TEST-*.xml' + inputs: + mergeTestResults: true + testRunTitle: 'osx-clang' + platform: macOS + publishRunAttachments: false + condition: succeededOrFailed() + - task: PublishBuildArtifacts@1 + displayName: 'Publish trash directories of failed tests' + condition: failed() + inputs: + PathtoPublish: t/failed-test-artifacts + ArtifactName: failed-test-artifacts + +- job: osx_gcc + displayName: osx-gcc + condition: succeeded() + pool: + vmImage: macOS-latest + steps: + - bash: | + ci/install-dependencies.sh || exit 1 + ci/run-build-and-tests.sh || { + ci/print-test-failures.sh + exit 1 + } + displayName: 'ci/run-build-and-tests.sh' + - task: PublishTestResults@2 + displayName: 'Publish Test Results **/TEST-*.xml' + inputs: + mergeTestResults: true + testRunTitle: 'osx-gcc' + platform: macOS + publishRunAttachments: false + condition: succeededOrFailed() + - task: PublishBuildArtifacts@1 + displayName: 'Publish trash directories of failed tests' + condition: failed() + inputs: + PathtoPublish: t/failed-test-artifacts + ArtifactName: failed-test-artifacts + +- job: linux32 + displayName: Linux32 + condition: succeeded() + pool: + vmImage: ubuntu-latest + steps: + - bash: | + res=0 + sudo AGENT_OS="$AGENT_OS" BUILD_BUILDNUMBER="$BUILD_BUILDNUMBER" BUILD_REPOSITORY_URI="$BUILD_REPOSITORY_URI" BUILD_SOURCEBRANCH="$BUILD_SOURCEBRANCH" BUILD_SOURCEVERSION="$BUILD_SOURCEVERSION" SYSTEM_PHASENAME="$SYSTEM_PHASENAME" SYSTEM_TASKDEFINITIONSURI="$SYSTEM_TASKDEFINITIONSURI" SYSTEM_TEAMPROJECT="$SYSTEM_TEAMPROJECT" CC=$CC MAKEFLAGS="$MAKEFLAGS" jobname=linux32 bash -lxc ci/run-docker.sh || res=1 + + sudo chmod a+r t/out/TEST-*.xml + test ! -d t/failed-test-artifacts || sudo chmod a+r t/failed-test-artifacts + + exit $res + displayName: 'jobname=linux32 ci/run-docker.sh' + - task: PublishTestResults@2 + displayName: 'Publish Test Results **/TEST-*.xml' + inputs: + mergeTestResults: true + testRunTitle: 'linux32' + platform: Linux + publishRunAttachments: false + condition: succeededOrFailed() + - task: PublishBuildArtifacts@1 + displayName: 'Publish trash directories of failed tests' + condition: failed() + inputs: + PathtoPublish: t/failed-test-artifacts + ArtifactName: failed-test-artifacts + +- job: static_analysis + displayName: StaticAnalysis + condition: succeeded() + pool: + vmImage: ubuntu-22.04 + steps: + - bash: | + sudo apt-get update && + sudo apt-get install -y coccinelle libcurl4-openssl-dev libssl-dev libexpat-dev gettext && + + export jobname=StaticAnalysis && + + ci/run-static-analysis.sh || exit 1 + displayName: 'ci/run-static-analysis.sh' + +- job: documentation + displayName: Documentation + condition: succeeded() + pool: + vmImage: ubuntu-latest + steps: + - bash: | + sudo apt-get update && + sudo apt-get install -y asciidoc xmlto asciidoctor docbook-xsl-ns && + + export ALREADY_HAVE_ASCIIDOCTOR=yes. && + export jobname=Documentation && + + ci/test-documentation.sh || exit 1 + displayName: 'ci/test-documentation.sh' diff --git a/builtin/add.c b/builtin/add.c index f84372964c8c48..76d5ad1f5da190 100644 --- a/builtin/add.c +++ b/builtin/add.c @@ -593,6 +593,10 @@ int cmd_add(int argc, const char **argv, const char *prefix) die_in_unpopulated_submodule(&the_index, prefix); die_path_inside_submodule(&the_index, &pathspec); + enable_fscache(0); + /* We do not really re-read the index but update the up-to-date flags */ + preload_index(&the_index, &pathspec, 0); + if (add_new_files) { int baselen; @@ -695,6 +699,7 @@ int cmd_add(int argc, const char **argv, const char *prefix) die(_("Unable to write new index file")); dir_clear(&dir); + enable_fscache(0); UNLEAK(pathspec); return exit_status; } diff --git a/builtin/checkout.c b/builtin/checkout.c index 2a132392fbe747..0490a19fdab2b1 100644 --- a/builtin/checkout.c +++ b/builtin/checkout.c @@ -391,6 +391,7 @@ static int checkout_worktree(const struct checkout_opts *opts, if (pc_workers > 1) init_parallel_checkout(); + enable_fscache(active_nr); for (pos = 0; pos < active_nr; pos++) { struct cache_entry *ce = active_cache[pos]; if (ce->ce_flags & CE_MATCHED) { @@ -415,6 +416,7 @@ static int checkout_worktree(const struct checkout_opts *opts, errs |= run_parallel_checkout(&state, pc_workers, pc_threshold, NULL, NULL); mem_pool_discard(&ce_mem_pool, should_validate_cache_entries()); + disable_fscache(); remove_marked_cache_entries(&the_index, 1); remove_scheduled_dirs(); errs |= finish_delayed_checkout(&state, opts->show_progress); diff --git a/builtin/clean.c b/builtin/clean.c index 5466636e66604e..1852e33ec993b0 100644 --- a/builtin/clean.c +++ b/builtin/clean.c @@ -34,6 +34,10 @@ static const char *msg_remove = N_("Removing %s\n"); static const char *msg_would_remove = N_("Would remove %s\n"); static const char *msg_skip_git_dir = N_("Skipping repository %s\n"); static const char *msg_would_skip_git_dir = N_("Would skip repository %s\n"); +#ifndef CAN_UNLINK_MOUNT_POINTS +static const char *msg_skip_mount_point = N_("Skipping mount point %s\n"); +static const char *msg_would_skip_mount_point = N_("Would skip mount point %s\n"); +#endif static const char *msg_warn_remove_failed = N_("failed to remove %s"); static const char *msg_warn_lstat_failed = N_("could not lstat %s\n"); static const char *msg_skip_cwd = N_("Refusing to remove current working directory\n"); @@ -175,6 +179,29 @@ static int remove_dirs(struct strbuf *path, const char *prefix, int force_flag, goto out; } + if (is_mount_point(path)) { +#ifndef CAN_UNLINK_MOUNT_POINTS + if (!quiet) { + quote_path(path->buf, prefix, "ed, 0); + printf(dry_run ? + _(msg_would_skip_mount_point) : + _(msg_skip_mount_point), quoted.buf); + } + *dir_gone = 0; +#else + if (!dry_run && unlink(path->buf)) { + int saved_errno = errno; + quote_path(path->buf, prefix, "ed, 0); + errno = saved_errno; + warning_errno(_(msg_warn_remove_failed), quoted.buf); + *dir_gone = 0; + ret = -1; + } +#endif + + goto out; + } + dir = opendir(path->buf); if (!dir) { /* an empty dir could be removed even if it is unreadble */ @@ -184,6 +211,9 @@ static int remove_dirs(struct strbuf *path, const char *prefix, int force_flag, quote_path(path->buf, prefix, "ed, 0); errno = saved_errno; warning_errno(_(msg_warn_remove_failed), quoted.buf); + if (saved_errno == ENAMETOOLONG) { + advise_if_enabled(ADVICE_NAME_TOO_LONG, _("Setting `core.longPaths` may allow the deletion to succeed.")); + } *dir_gone = 0; } ret = res; @@ -219,6 +249,9 @@ static int remove_dirs(struct strbuf *path, const char *prefix, int force_flag, quote_path(path->buf, prefix, "ed, 0); errno = saved_errno; warning_errno(_(msg_warn_remove_failed), quoted.buf); + if (saved_errno == ENAMETOOLONG) { + advise_if_enabled(ADVICE_NAME_TOO_LONG, _("Setting `core.longPaths` may allow the deletion to succeed.")); + } *dir_gone = 0; ret = 1; } @@ -262,6 +295,9 @@ static int remove_dirs(struct strbuf *path, const char *prefix, int force_flag, quote_path(path->buf, prefix, "ed, 0); errno = saved_errno; warning_errno(_(msg_warn_remove_failed), quoted.buf); + if (saved_errno == ENAMETOOLONG) { + advise_if_enabled(ADVICE_NAME_TOO_LONG, _("Setting `core.longPaths` may allow the deletion to succeed.")); + } *dir_gone = 0; ret = 1; } @@ -1014,6 +1050,7 @@ int cmd_clean(int argc, const char **argv, const char *prefix) if (read_cache() < 0) die(_("index file corrupt")); + enable_fscache(active_nr); pl = add_pattern_list(&dir, EXC_CMDL, "--exclude option"); for (i = 0; i < exclude_list.nr; i++) @@ -1080,6 +1117,9 @@ int cmd_clean(int argc, const char **argv, const char *prefix) qname = quote_path(item->string, NULL, &buf, 0); errno = saved_errno; warning_errno(_(msg_warn_remove_failed), qname); + if (saved_errno == ENAMETOOLONG) { + advise_if_enabled(ADVICE_NAME_TOO_LONG, _("Setting `core.longPaths` may allow the deletion to succeed.")); + } errors++; } else if (!quiet) { qname = quote_path(item->string, NULL, &buf, 0); @@ -1088,6 +1128,7 @@ int cmd_clean(int argc, const char **argv, const char *prefix) } } + disable_fscache(); strbuf_release(&abs_path); strbuf_release(&buf); string_list_clear(&del_list, 0); diff --git a/builtin/commit.c b/builtin/commit.c index fcf9c85947e6a1..cf0e24d3abf25c 100644 --- a/builtin/commit.c +++ b/builtin/commit.c @@ -1540,6 +1540,7 @@ int cmd_status(int argc, const char **argv, const char *prefix) PATHSPEC_PREFER_FULL, prefix, argv); + enable_fscache(0); if (status_format != STATUS_FORMAT_PORCELAIN && status_format != STATUS_FORMAT_PORCELAIN_V2) progress_flag = REFRESH_PROGRESS; @@ -1580,6 +1581,7 @@ int cmd_status(int argc, const char **argv, const char *prefix) wt_status_print(&s); wt_status_collect_free_buffers(&s); + disable_fscache(); return 0; } diff --git a/builtin/difftool.c b/builtin/difftool.c index 4b10ad1a36908f..a133424fe3899f 100644 --- a/builtin/difftool.c +++ b/builtin/difftool.c @@ -513,7 +513,7 @@ static int run_dir_diff(const char *extcmd, int symlinks, const char *prefix, } add_path(&wtdir, wtdir_len, dst_path); if (symlinks) { - if (symlink(wtdir.buf, rdir.buf)) { + if (create_symlink(lstate.istate, wtdir.buf, rdir.buf)) { ret = error_errno("could not symlink '%s' to '%s'", wtdir.buf, rdir.buf); goto finish; } diff --git a/builtin/gc.c b/builtin/gc.c index 2753bd15a5e85f..effe7ac15661ce 100644 --- a/builtin/gc.c +++ b/builtin/gc.c @@ -1998,7 +1998,7 @@ static int schtasks_schedule_task(const char *exec_path, enum schedule_priority "\n" "\n" "\n" - "\"%s\\git.exe\"\n" + "\"%s\\headless-git.exe\"\n" "--exec-path=\"%s\" for-each-repo --config=maintenance.repo maintenance run --schedule=%s\n" "\n" "\n" diff --git a/builtin/init-db.c b/builtin/init-db.c index 546f9c595e7d8c..665d2da51bce8f 100644 --- a/builtin/init-db.c +++ b/builtin/init-db.c @@ -78,7 +78,7 @@ static void copy_templates_1(struct strbuf *path, struct strbuf *template_path, if (strbuf_readlink(&lnk, template_path->buf, st_template.st_size) < 0) die_errno(_("cannot readlink '%s'"), template_path->buf); - if (symlink(lnk.buf, path->buf)) + if (create_symlink(NULL, lnk.buf, path->buf)) die_errno(_("cannot symlink '%s' '%s'"), lnk.buf, path->buf); strbuf_release(&lnk); @@ -300,7 +300,7 @@ static int create_default_files(const char *template_path, path = git_path_buf(&buf, "tXXXXXX"); if (!close(xmkstemp(path)) && !unlink(path) && - !symlink("testing", path) && + !create_symlink(NULL, "testing", path) && !lstat(path, &st1) && S_ISLNK(st1.st_mode)) unlink(path); /* good */ @@ -410,7 +410,7 @@ int init_db(const char *git_dir, const char *real_git_dir, startup_info->have_repository = 1; /* Ensure `core.hidedotfiles` is processed */ - git_config(platform_core_config, NULL); + git_config(git_default_core_config, NULL); safe_create_dir(git_dir, 0); diff --git a/builtin/reset.c b/builtin/reset.c index fdce6f8c85670c..0d471411ffe398 100644 --- a/builtin/reset.c +++ b/builtin/reset.c @@ -26,6 +26,8 @@ #include "submodule.h" #include "submodule-config.h" #include "dir.h" +#include "strbuf.h" +#include "quote.h" #define REFRESH_INDEX_DELAY_WARNING_IN_MS (2 * 1000) @@ -34,6 +36,7 @@ static const char * const git_reset_usage[] = { N_("git reset [-q] [] [--] ..."), N_("git reset [-q] [--pathspec-from-file [--pathspec-file-nul]] []"), N_("git reset --patch [] [--] [...]"), + N_("DEPRECATED: git reset [-q] [--stdin [-z]] []"), NULL }; @@ -316,6 +319,7 @@ int cmd_reset(int argc, const char **argv, const char *prefix) struct object_id oid; struct pathspec pathspec; int intent_to_add = 0; + int nul_term_line = 0, read_from_stdin = 0; const struct option options[] = { OPT__QUIET(&quiet, N_("be quiet, only report errors")), OPT_BOOL(0, "no-refresh", &no_refresh, @@ -337,6 +341,10 @@ int cmd_reset(int argc, const char **argv, const char *prefix) N_("record only the fact that removed paths will be added later")), OPT_PATHSPEC_FROM_FILE(&pathspec_from_file), OPT_PATHSPEC_FILE_NUL(&pathspec_file_nul), + OPT_BOOL('z', NULL, &nul_term_line, + N_("DEPRECATED (use --pathspec-file-nul instead): paths are separated with NUL character")), + OPT_BOOL(0, "stdin", &read_from_stdin, + N_("DEPRECATED (use --pathspec-from-file=- instead): read paths from ")), OPT_END() }; @@ -346,6 +354,12 @@ int cmd_reset(int argc, const char **argv, const char *prefix) PARSE_OPT_KEEP_DASHDASH); parse_args(&pathspec, argv, prefix, patch_mode, &rev); + if (read_from_stdin) { + pathspec_from_file = "-"; + if (nul_term_line) + pathspec_file_nul = 1; + } + if (pathspec_from_file) { if (patch_mode) die(_("options '%s' and '%s' cannot be used together"), "--pathspec-from-file", "--patch"); diff --git a/cache.h b/cache.h index 26ed03bd6de626..e7344004f5c7e0 100644 --- a/cache.h +++ b/cache.h @@ -1313,6 +1313,7 @@ int normalize_path_copy_len(char *dst, const char *src, int *prefix_len); int normalize_path_copy(char *dst, const char *src); int longest_ancestor_length(const char *path, struct string_list *prefixes); char *strip_path_suffix(const char *path, const char *suffix); +int is_mount_point_via_stat(struct strbuf *path); int daemon_avoid_alias(const char *path); /* diff --git a/ci/lib.sh b/ci/lib.sh index 1b0cc2b57db866..b40719b0df3769 100755 --- a/ci/lib.sh +++ b/ci/lib.sh @@ -161,6 +161,12 @@ then MAKEFLAGS="$MAKEFLAGS --jobs=10" test windows_nt != "$CI_OS_NAME" || GIT_TEST_OPTS="--no-chain-lint --no-bin-wrappers $GIT_TEST_OPTS" + case "$CI_OS_NAME" in + linux) runs_on_pool=ubuntu-latest;; + macos|osx) runs_on_pool=macos-latest;; + windows_nt) runs_on_pool=windows-latest;; + *) echo "Unhandled OS: $CI_OS_NAME" >&2; exit 1;; + esac elif test true = "$GITHUB_ACTIONS" then CI_TYPE=github-actions diff --git a/ci/run-build-and-tests.sh b/ci/run-build-and-tests.sh index 8ebff4259676e3..f3195eb265f37b 100755 --- a/ci/run-build-and-tests.sh +++ b/ci/run-build-and-tests.sh @@ -5,11 +5,6 @@ . ${0%/*}/lib.sh -case "$CI_OS_NAME" in -windows*) cmd //c mklink //j t\\.prove "$(cygpath -aw "$cache_dir/.prove")";; -*) ln -s "$cache_dir/.prove" t/.prove;; -esac - run_tests=t case "$jobname" in @@ -53,4 +48,8 @@ then fi check_unignored_build_artifacts +case " $MAKE_TARGETS " in +*" all "*) make -C contrib/subtree test;; +esac + save_good_tree diff --git a/ci/run-test-slice.sh b/ci/run-test-slice.sh index a3c67956a8df8f..279aff4da9e4db 100755 --- a/ci/run-test-slice.sh +++ b/ci/run-test-slice.sh @@ -5,14 +5,12 @@ . ${0%/*}/lib.sh -case "$CI_OS_NAME" in -windows*) cmd //c mklink //j t\\.prove "$(cygpath -aw "$cache_dir/.prove")";; -*) ln -s "$cache_dir/.prove" t/.prove;; -esac - group "Run tests" make --quiet -C t T="$(cd t && ./helper/test-tool path-utils slice-tests "$1" "$2" t[0-9]*.sh | tr '\n' ' ')" || handle_failed_tests +# Run the git subtree tests only if main tests succeeded +test 0 != "$1" || make -C contrib/subtree test + check_unignored_build_artifacts diff --git a/compat/bswap.h b/compat/bswap.h index 512f6f4b9937c8..a443e99eef2f1c 100644 --- a/compat/bswap.h +++ b/compat/bswap.h @@ -35,7 +35,19 @@ static inline uint64_t default_bswap64(uint64_t val) #undef bswap32 #undef bswap64 -#if defined(__GNUC__) && (defined(__i386__) || defined(__x86_64__)) +/** + * __has_builtin is available since Clang 10 and GCC 10. + * Below is a fallback for older compilers. + */ +#ifndef __has_builtin + #define __has_builtin(x) 0 +#endif + +#if __has_builtin(__builtin_bswap32) && __has_builtin(__builtin_bswap64) +#define bswap32(x) __builtin_bswap32((x)) +#define bswap64(x) __builtin_bswap64((x)) + +#elif defined(__GNUC__) && (defined(__i386__) || defined(__x86_64__)) #define bswap32 git_bswap32 static inline uint32_t git_bswap32(uint32_t x) diff --git a/compat/fsmonitor/fsm-health-win32.c b/compat/fsmonitor/fsm-health-win32.c index 2ea08c1d4e81ec..5551d6d97dee87 100644 --- a/compat/fsmonitor/fsm-health-win32.c +++ b/compat/fsmonitor/fsm-health-win32.c @@ -32,7 +32,7 @@ struct fsm_health_data struct wt_moved { - wchar_t wpath[MAX_PATH + 1]; + wchar_t wpath[MAX_LONG_PATH + 1]; BY_HANDLE_FILE_INFORMATION bhfi; } wt_moved; }; @@ -141,8 +141,8 @@ static int has_worktree_moved(struct fsmonitor_daemon_state *state, return 0; case CTX_INIT: - if (xutftowcs_path(data->wt_moved.wpath, - state->path_worktree_watch.buf) < 0) { + if (xutftowcs_long_path(data->wt_moved.wpath, + state->path_worktree_watch.buf) < 0) { error(_("could not convert to wide characters: '%s'"), state->path_worktree_watch.buf); return -1; diff --git a/compat/fsmonitor/fsm-listen-win32.c b/compat/fsmonitor/fsm-listen-win32.c index 03df8d951b871f..df8704aa5b2ca4 100644 --- a/compat/fsmonitor/fsm-listen-win32.c +++ b/compat/fsmonitor/fsm-listen-win32.c @@ -25,7 +25,7 @@ struct one_watch DWORD count; struct strbuf path; - wchar_t wpath_longname[MAX_PATH + 1]; + wchar_t wpath_longname[MAX_LONG_PATH + 1]; DWORD wpath_longname_len; HANDLE hDir; @@ -128,8 +128,8 @@ static int normalize_path_in_utf8(wchar_t *wpath, DWORD wpath_len, */ static void check_for_shortnames(struct one_watch *watch) { - wchar_t buf_in[MAX_PATH + 1]; - wchar_t buf_out[MAX_PATH + 1]; + wchar_t buf_in[MAX_LONG_PATH + 1]; + wchar_t buf_out[MAX_LONG_PATH + 1]; wchar_t *last; wchar_t *p; @@ -194,8 +194,8 @@ static enum get_relative_result get_relative_longname( const wchar_t *wpath, DWORD wpath_len, wchar_t *wpath_longname, size_t bufsize_wpath_longname) { - wchar_t buf_in[2 * MAX_PATH + 1]; - wchar_t buf_out[MAX_PATH + 1]; + wchar_t buf_in[2 * MAX_LONG_PATH + 1]; + wchar_t buf_out[MAX_LONG_PATH + 1]; DWORD root_len; DWORD out_len; @@ -296,10 +296,10 @@ static struct one_watch *create_watch(struct fsmonitor_daemon_state *state, FILE_SHARE_WRITE | FILE_SHARE_READ | FILE_SHARE_DELETE; HANDLE hDir; DWORD len_longname; - wchar_t wpath[MAX_PATH + 1]; - wchar_t wpath_longname[MAX_PATH + 1]; + wchar_t wpath[MAX_LONG_PATH + 1]; + wchar_t wpath_longname[MAX_LONG_PATH + 1]; - if (xutftowcs_path(wpath, path) < 0) { + if (xutftowcs_long_path(wpath, path) < 0) { error(_("could not convert to wide characters: '%s'"), path); return NULL; } @@ -544,7 +544,7 @@ static int process_worktree_events(struct fsmonitor_daemon_state *state) struct string_list cookie_list = STRING_LIST_INIT_DUP; struct fsmonitor_batch *batch = NULL; const char *p = watch->buffer; - wchar_t wpath_longname[MAX_PATH + 1]; + wchar_t wpath_longname[MAX_LONG_PATH + 1]; /* * If the kernel gets more events than will fit in the kernel diff --git a/compat/fsmonitor/fsm-settings-win32.c b/compat/fsmonitor/fsm-settings-win32.c index e5ec5b0a9f73bc..28e8f3fec92ead 100644 --- a/compat/fsmonitor/fsm-settings-win32.c +++ b/compat/fsmonitor/fsm-settings-win32.c @@ -130,8 +130,8 @@ static int check_remote_protocol(wchar_t *wpath) static enum fsmonitor_reason check_remote(struct repository *r) { int ret; - wchar_t wpath[MAX_PATH]; - wchar_t wfullpath[MAX_PATH]; + wchar_t wpath[MAX_LONG_PATH]; + wchar_t wfullpath[MAX_LONG_PATH]; size_t wlen; UINT driveType; @@ -139,7 +139,7 @@ static enum fsmonitor_reason check_remote(struct repository *r) * Do everything in wide chars because the drive letter might be * a multi-byte sequence. See win32_has_dos_drive_prefix(). */ - if (xutftowcs_path(wpath, r->worktree) < 0) + if (xutftowcs_long_path(wpath, r->worktree) < 0) return FSMONITOR_REASON_ERROR; /* @@ -157,7 +157,7 @@ static enum fsmonitor_reason check_remote(struct repository *r) * slashes to backslashes. This is essential to get GetDriveTypeW() * correctly handle some UNC "\\server\share\..." paths. */ - if (!GetFullPathNameW(wpath, MAX_PATH, wfullpath, NULL)) + if (!GetFullPathNameW(wpath, MAX_LONG_PATH, wfullpath, NULL)) return FSMONITOR_REASON_ERROR; driveType = GetDriveTypeW(wfullpath); diff --git a/compat/mimalloc/LICENSE b/compat/mimalloc/LICENSE new file mode 100644 index 00000000000000..670b668a0c928e --- /dev/null +++ b/compat/mimalloc/LICENSE @@ -0,0 +1,21 @@ +MIT License + +Copyright (c) 2018-2021 Microsoft Corporation, Daan Leijen + +Permission is hereby granted, free of charge, to any person obtaining a copy +of this software and associated documentation files (the "Software"), to deal +in the Software without restriction, including without limitation the rights +to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +copies of the Software, and to permit persons to whom the Software is +furnished to do so, subject to the following conditions: + +The above copyright notice and this permission notice shall be included in all +copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +SOFTWARE. diff --git a/compat/mimalloc/alloc-aligned.c b/compat/mimalloc/alloc-aligned.c new file mode 100644 index 00000000000000..5594b6d38387d2 --- /dev/null +++ b/compat/mimalloc/alloc-aligned.c @@ -0,0 +1,260 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2021, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ + +#include "mimalloc.h" +#include "mimalloc-internal.h" + +#include // memset + +// ------------------------------------------------------ +// Aligned Allocation +// ------------------------------------------------------ + +// Fallback primitive aligned allocation -- split out for better codegen +static mi_decl_noinline void* mi_heap_malloc_zero_aligned_at_fallback(mi_heap_t* const heap, const size_t size, const size_t alignment, const size_t offset, const bool zero) mi_attr_noexcept +{ + mi_assert_internal(size <= PTRDIFF_MAX); + mi_assert_internal(alignment!=0 && _mi_is_power_of_two(alignment) && alignment <= MI_ALIGNMENT_MAX); + + const uintptr_t align_mask = alignment-1; // for any x, `(x & align_mask) == (x % alignment)` + const size_t padsize = size + MI_PADDING_SIZE; + + // use regular allocation if it is guaranteed to fit the alignment constraints + if (offset==0 && alignment<=padsize && padsize<=MI_MAX_ALIGN_GUARANTEE && (padsize&align_mask)==0) { + void* p = _mi_heap_malloc_zero(heap, size, zero); + mi_assert_internal(p == NULL || ((uintptr_t)p % alignment) == 0); + return p; + } + + // otherwise over-allocate + void* p = _mi_heap_malloc_zero(heap, size + alignment - 1, zero); + if (p == NULL) return NULL; + + // .. and align within the allocation + uintptr_t adjust = alignment - (((uintptr_t)p + offset) & align_mask); + mi_assert_internal(adjust <= alignment); + void* aligned_p = (adjust == alignment ? p : (void*)((uintptr_t)p + adjust)); + if (aligned_p != p) mi_page_set_has_aligned(_mi_ptr_page(p), true); + mi_assert_internal(((uintptr_t)aligned_p + offset) % alignment == 0); + mi_assert_internal(p == _mi_page_ptr_unalign(_mi_ptr_segment(aligned_p), _mi_ptr_page(aligned_p), aligned_p)); + return aligned_p; +} + +// Primitive aligned allocation +static void* mi_heap_malloc_zero_aligned_at(mi_heap_t* const heap, const size_t size, const size_t alignment, const size_t offset, const bool zero) mi_attr_noexcept +{ + // note: we don't require `size > offset`, we just guarantee that the address at offset is aligned regardless of the allocated size. + mi_assert(alignment > 0); + if (mi_unlikely(alignment==0 || !_mi_is_power_of_two(alignment))) { // require power-of-two (see ) + #if MI_DEBUG > 0 + _mi_error_message(EOVERFLOW, "aligned allocation requires the alignment to be a power-of-two (size %zu, alignment %zu)\n", size, alignment); + #endif + return NULL; + } + if (mi_unlikely(alignment > MI_ALIGNMENT_MAX)) { // we cannot align at a boundary larger than this (or otherwise we cannot find segment headers) + #if MI_DEBUG > 0 + _mi_error_message(EOVERFLOW, "aligned allocation has a maximum alignment of %zu (size %zu, alignment %zu)\n", MI_ALIGNMENT_MAX, size, alignment); + #endif + return NULL; + } + if (mi_unlikely(size > PTRDIFF_MAX)) { // we don't allocate more than PTRDIFF_MAX (see ) + #if MI_DEBUG > 0 + _mi_error_message(EOVERFLOW, "aligned allocation request is too large (size %zu, alignment %zu)\n", size, alignment); + #endif + return NULL; + } + const uintptr_t align_mask = alignment-1; // for any x, `(x & align_mask) == (x % alignment)` + const size_t padsize = size + MI_PADDING_SIZE; // note: cannot overflow due to earlier size > PTRDIFF_MAX check + + // try first if there happens to be a small block available with just the right alignment + if (mi_likely(padsize <= MI_SMALL_SIZE_MAX)) { + mi_page_t* page = _mi_heap_get_free_small_page(heap, padsize); + const bool is_aligned = (((uintptr_t)page->free+offset) & align_mask)==0; + if (mi_likely(page->free != NULL && is_aligned)) + { + #if MI_STAT>1 + mi_heap_stat_increase(heap, malloc, size); + #endif + void* p = _mi_page_malloc(heap, page, padsize); // TODO: inline _mi_page_malloc + mi_assert_internal(p != NULL); + mi_assert_internal(((uintptr_t)p + offset) % alignment == 0); + if (zero) { _mi_block_zero_init(page, p, size); } + return p; + } + } + // fallback + return mi_heap_malloc_zero_aligned_at_fallback(heap, size, alignment, offset, zero); +} + + +// ------------------------------------------------------ +// Optimized mi_heap_malloc_aligned / mi_malloc_aligned +// ------------------------------------------------------ + +mi_decl_restrict void* mi_heap_malloc_aligned_at(mi_heap_t* heap, size_t size, size_t alignment, size_t offset) mi_attr_noexcept { + return mi_heap_malloc_zero_aligned_at(heap, size, alignment, offset, false); +} + +mi_decl_restrict void* mi_heap_malloc_aligned(mi_heap_t* heap, size_t size, size_t alignment) mi_attr_noexcept { + #if !MI_PADDING + // without padding, any small sized allocation is naturally aligned (see also `_mi_segment_page_start`) + if (!_mi_is_power_of_two(alignment)) return NULL; + if (mi_likely(_mi_is_power_of_two(size) && size >= alignment && size <= MI_SMALL_SIZE_MAX)) + #else + // with padding, we can only guarantee this for fixed alignments + if (mi_likely((alignment == sizeof(void*) || (alignment == MI_MAX_ALIGN_SIZE && size > (MI_MAX_ALIGN_SIZE/2))) + && size <= MI_SMALL_SIZE_MAX)) + #endif + { + // fast path for common alignment and size + return mi_heap_malloc_small(heap, size); + } + else { + return mi_heap_malloc_aligned_at(heap, size, alignment, 0); + } +} + +// ------------------------------------------------------ +// Aligned Allocation +// ------------------------------------------------------ + +mi_decl_restrict void* mi_heap_zalloc_aligned_at(mi_heap_t* heap, size_t size, size_t alignment, size_t offset) mi_attr_noexcept { + return mi_heap_malloc_zero_aligned_at(heap, size, alignment, offset, true); +} + +mi_decl_restrict void* mi_heap_zalloc_aligned(mi_heap_t* heap, size_t size, size_t alignment) mi_attr_noexcept { + return mi_heap_zalloc_aligned_at(heap, size, alignment, 0); +} + +mi_decl_restrict void* mi_heap_calloc_aligned_at(mi_heap_t* heap, size_t count, size_t size, size_t alignment, size_t offset) mi_attr_noexcept { + size_t total; + if (mi_count_size_overflow(count, size, &total)) return NULL; + return mi_heap_zalloc_aligned_at(heap, total, alignment, offset); +} + +mi_decl_restrict void* mi_heap_calloc_aligned(mi_heap_t* heap, size_t count, size_t size, size_t alignment) mi_attr_noexcept { + return mi_heap_calloc_aligned_at(heap,count,size,alignment,0); +} + +mi_decl_restrict void* mi_malloc_aligned_at(size_t size, size_t alignment, size_t offset) mi_attr_noexcept { + return mi_heap_malloc_aligned_at(mi_get_default_heap(), size, alignment, offset); +} + +mi_decl_restrict void* mi_malloc_aligned(size_t size, size_t alignment) mi_attr_noexcept { + return mi_heap_malloc_aligned(mi_get_default_heap(), size, alignment); +} + +mi_decl_restrict void* mi_zalloc_aligned_at(size_t size, size_t alignment, size_t offset) mi_attr_noexcept { + return mi_heap_zalloc_aligned_at(mi_get_default_heap(), size, alignment, offset); +} + +mi_decl_restrict void* mi_zalloc_aligned(size_t size, size_t alignment) mi_attr_noexcept { + return mi_heap_zalloc_aligned(mi_get_default_heap(), size, alignment); +} + +mi_decl_restrict void* mi_calloc_aligned_at(size_t count, size_t size, size_t alignment, size_t offset) mi_attr_noexcept { + return mi_heap_calloc_aligned_at(mi_get_default_heap(), count, size, alignment, offset); +} + +mi_decl_restrict void* mi_calloc_aligned(size_t count, size_t size, size_t alignment) mi_attr_noexcept { + return mi_heap_calloc_aligned(mi_get_default_heap(), count, size, alignment); +} + + +// ------------------------------------------------------ +// Aligned re-allocation +// ------------------------------------------------------ + +static void* mi_heap_realloc_zero_aligned_at(mi_heap_t* heap, void* p, size_t newsize, size_t alignment, size_t offset, bool zero) mi_attr_noexcept { + mi_assert(alignment > 0); + if (alignment <= sizeof(uintptr_t)) return _mi_heap_realloc_zero(heap,p,newsize,zero); + if (p == NULL) return mi_heap_malloc_zero_aligned_at(heap,newsize,alignment,offset,zero); + size_t size = mi_usable_size(p); + if (newsize <= size && newsize >= (size - (size / 2)) + && (((uintptr_t)p + offset) % alignment) == 0) { + return p; // reallocation still fits, is aligned and not more than 50% waste + } + else { + void* newp = mi_heap_malloc_aligned_at(heap,newsize,alignment,offset); + if (newp != NULL) { + if (zero && newsize > size) { + const mi_page_t* page = _mi_ptr_page(newp); + if (page->is_zero) { + // already zero initialized + mi_assert_expensive(mi_mem_is_zero(newp,newsize)); + } + else { + // also set last word in the previous allocation to zero to ensure any padding is zero-initialized + size_t start = (size >= sizeof(intptr_t) ? size - sizeof(intptr_t) : 0); + memset((uint8_t*)newp + start, 0, newsize - start); + } + } + _mi_memcpy_aligned(newp, p, (newsize > size ? size : newsize)); + mi_free(p); // only free if successful + } + return newp; + } +} + +static void* mi_heap_realloc_zero_aligned(mi_heap_t* heap, void* p, size_t newsize, size_t alignment, bool zero) mi_attr_noexcept { + mi_assert(alignment > 0); + if (alignment <= sizeof(uintptr_t)) return _mi_heap_realloc_zero(heap,p,newsize,zero); + size_t offset = ((uintptr_t)p % alignment); // use offset of previous allocation (p can be NULL) + return mi_heap_realloc_zero_aligned_at(heap,p,newsize,alignment,offset,zero); +} + +void* mi_heap_realloc_aligned_at(mi_heap_t* heap, void* p, size_t newsize, size_t alignment, size_t offset) mi_attr_noexcept { + return mi_heap_realloc_zero_aligned_at(heap,p,newsize,alignment,offset,false); +} + +void* mi_heap_realloc_aligned(mi_heap_t* heap, void* p, size_t newsize, size_t alignment) mi_attr_noexcept { + return mi_heap_realloc_zero_aligned(heap,p,newsize,alignment,false); +} + +void* mi_heap_rezalloc_aligned_at(mi_heap_t* heap, void* p, size_t newsize, size_t alignment, size_t offset) mi_attr_noexcept { + return mi_heap_realloc_zero_aligned_at(heap, p, newsize, alignment, offset, true); +} + +void* mi_heap_rezalloc_aligned(mi_heap_t* heap, void* p, size_t newsize, size_t alignment) mi_attr_noexcept { + return mi_heap_realloc_zero_aligned(heap, p, newsize, alignment, true); +} + +void* mi_heap_recalloc_aligned_at(mi_heap_t* heap, void* p, size_t newcount, size_t size, size_t alignment, size_t offset) mi_attr_noexcept { + size_t total; + if (mi_count_size_overflow(newcount, size, &total)) return NULL; + return mi_heap_rezalloc_aligned_at(heap, p, total, alignment, offset); +} + +void* mi_heap_recalloc_aligned(mi_heap_t* heap, void* p, size_t newcount, size_t size, size_t alignment) mi_attr_noexcept { + size_t total; + if (mi_count_size_overflow(newcount, size, &total)) return NULL; + return mi_heap_rezalloc_aligned(heap, p, total, alignment); +} + +void* mi_realloc_aligned_at(void* p, size_t newsize, size_t alignment, size_t offset) mi_attr_noexcept { + return mi_heap_realloc_aligned_at(mi_get_default_heap(), p, newsize, alignment, offset); +} + +void* mi_realloc_aligned(void* p, size_t newsize, size_t alignment) mi_attr_noexcept { + return mi_heap_realloc_aligned(mi_get_default_heap(), p, newsize, alignment); +} + +void* mi_rezalloc_aligned_at(void* p, size_t newsize, size_t alignment, size_t offset) mi_attr_noexcept { + return mi_heap_rezalloc_aligned_at(mi_get_default_heap(), p, newsize, alignment, offset); +} + +void* mi_rezalloc_aligned(void* p, size_t newsize, size_t alignment) mi_attr_noexcept { + return mi_heap_rezalloc_aligned(mi_get_default_heap(), p, newsize, alignment); +} + +void* mi_recalloc_aligned_at(void* p, size_t newcount, size_t size, size_t alignment, size_t offset) mi_attr_noexcept { + return mi_heap_recalloc_aligned_at(mi_get_default_heap(), p, newcount, size, alignment, offset); +} + +void* mi_recalloc_aligned(void* p, size_t newcount, size_t size, size_t alignment) mi_attr_noexcept { + return mi_heap_recalloc_aligned(mi_get_default_heap(), p, newcount, size, alignment); +} diff --git a/compat/mimalloc/alloc.c b/compat/mimalloc/alloc.c new file mode 100644 index 00000000000000..8f084d3ad35170 --- /dev/null +++ b/compat/mimalloc/alloc.c @@ -0,0 +1,930 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2022, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ +#ifndef _DEFAULT_SOURCE +#define _DEFAULT_SOURCE // for realpath() on Linux +#endif + +#include "mimalloc.h" +#include "mimalloc-internal.h" +#include "mimalloc-atomic.h" + +#include // memset, strlen +#include // malloc, exit + +// ------------------------------------------------------ +// Allocation +// ------------------------------------------------------ + +// Fast allocation in a page: just pop from the free list. +// Fall back to generic allocation only if the list is empty. +extern inline void* _mi_page_malloc(mi_heap_t* heap, mi_page_t* page, size_t size) mi_attr_noexcept { + mi_assert_internal(page->xblock_size==0||mi_page_block_size(page) >= size); + mi_block_t* const block = page->free; + if (mi_unlikely(block == NULL)) { + return _mi_malloc_generic(heap, size); + } + mi_assert_internal(block != NULL && _mi_ptr_page(block) == page); + // pop from the free list + page->used++; + page->free = mi_block_next(page, block); + mi_assert_internal(page->free == NULL || _mi_ptr_page(page->free) == page); + +#if (MI_DEBUG>0) + if (!page->is_zero) { memset(block, MI_DEBUG_UNINIT, size); } +#elif (MI_SECURE!=0) + block->next = 0; // don't leak internal data +#endif + +#if (MI_STAT>0) + const size_t bsize = mi_page_usable_block_size(page); + if (bsize <= MI_MEDIUM_OBJ_SIZE_MAX) { + mi_heap_stat_increase(heap, normal, bsize); + mi_heap_stat_counter_increase(heap, normal_count, 1); +#if (MI_STAT>1) + const size_t bin = _mi_bin(bsize); + mi_heap_stat_increase(heap, normal_bins[bin], 1); +#endif + } +#endif + +#if (MI_PADDING > 0) && defined(MI_ENCODE_FREELIST) + mi_padding_t* const padding = (mi_padding_t*)((uint8_t*)block + mi_page_usable_block_size(page)); + ptrdiff_t delta = ((uint8_t*)padding - (uint8_t*)block - (size - MI_PADDING_SIZE)); + mi_assert_internal(delta >= 0 && mi_page_usable_block_size(page) >= (size - MI_PADDING_SIZE + delta)); + padding->canary = (uint32_t)(mi_ptr_encode(page,block,page->keys)); + padding->delta = (uint32_t)(delta); + uint8_t* fill = (uint8_t*)padding - delta; + const size_t maxpad = (delta > MI_MAX_ALIGN_SIZE ? MI_MAX_ALIGN_SIZE : delta); // set at most N initial padding bytes + for (size_t i = 0; i < maxpad; i++) { fill[i] = MI_DEBUG_PADDING; } +#endif + + return block; +} + +// allocate a small block +extern inline mi_decl_restrict void* mi_heap_malloc_small(mi_heap_t* heap, size_t size) mi_attr_noexcept { + mi_assert(heap!=NULL); + mi_assert(heap->thread_id == 0 || heap->thread_id == _mi_thread_id()); // heaps are thread local + mi_assert(size <= MI_SMALL_SIZE_MAX); + #if (MI_PADDING) + if (size == 0) { + size = sizeof(void*); + } + #endif + mi_page_t* page = _mi_heap_get_free_small_page(heap,size + MI_PADDING_SIZE); + void* p = _mi_page_malloc(heap, page, size + MI_PADDING_SIZE); + mi_assert_internal(p==NULL || mi_usable_size(p) >= size); + #if MI_STAT>1 + if (p != NULL) { + if (!mi_heap_is_initialized(heap)) { heap = mi_get_default_heap(); } + mi_heap_stat_increase(heap, malloc, mi_usable_size(p)); + } + #endif + return p; +} + +extern inline mi_decl_restrict void* mi_malloc_small(size_t size) mi_attr_noexcept { + return mi_heap_malloc_small(mi_get_default_heap(), size); +} + +// The main allocation function +extern inline mi_decl_restrict void* mi_heap_malloc(mi_heap_t* heap, size_t size) mi_attr_noexcept { + if (mi_likely(size <= MI_SMALL_SIZE_MAX)) { + return mi_heap_malloc_small(heap, size); + } + else { + mi_assert(heap!=NULL); + mi_assert(heap->thread_id == 0 || heap->thread_id == _mi_thread_id()); // heaps are thread local + void* const p = _mi_malloc_generic(heap, size + MI_PADDING_SIZE); // note: size can overflow but it is detected in malloc_generic + mi_assert_internal(p == NULL || mi_usable_size(p) >= size); + #if MI_STAT>1 + if (p != NULL) { + if (!mi_heap_is_initialized(heap)) { heap = mi_get_default_heap(); } + mi_heap_stat_increase(heap, malloc, mi_usable_size(p)); + } + #endif + return p; + } +} + +extern inline mi_decl_restrict void* mi_malloc(size_t size) mi_attr_noexcept { + return mi_heap_malloc(mi_get_default_heap(), size); +} + + +void _mi_block_zero_init(const mi_page_t* page, void* p, size_t size) { + // note: we need to initialize the whole usable block size to zero, not just the requested size, + // or the recalloc/rezalloc functions cannot safely expand in place (see issue #63) + MI_UNUSED(size); + mi_assert_internal(p != NULL); + mi_assert_internal(mi_usable_size(p) >= size); // size can be zero + mi_assert_internal(_mi_ptr_page(p)==page); + if (page->is_zero && size > sizeof(mi_block_t)) { + // already zero initialized memory + ((mi_block_t*)p)->next = 0; // clear the free list pointer + mi_assert_expensive(mi_mem_is_zero(p, mi_usable_size(p))); + } + else { + // otherwise memset + memset(p, 0, mi_usable_size(p)); + } +} + +// zero initialized small block +mi_decl_restrict void* mi_zalloc_small(size_t size) mi_attr_noexcept { + void* p = mi_malloc_small(size); + if (p != NULL) { + _mi_block_zero_init(_mi_ptr_page(p), p, size); // todo: can we avoid getting the page again? + } + return p; +} + +void* _mi_heap_malloc_zero(mi_heap_t* heap, size_t size, bool zero) mi_attr_noexcept { + void* p = mi_heap_malloc(heap,size); + if (zero && p != NULL) { + _mi_block_zero_init(_mi_ptr_page(p),p,size); // todo: can we avoid getting the page again? + } + return p; +} + +extern inline mi_decl_restrict void* mi_heap_zalloc(mi_heap_t* heap, size_t size) mi_attr_noexcept { + return _mi_heap_malloc_zero(heap, size, true); +} + +mi_decl_restrict void* mi_zalloc(size_t size) mi_attr_noexcept { + return mi_heap_zalloc(mi_get_default_heap(),size); +} + + +// ------------------------------------------------------ +// Check for double free in secure and debug mode +// This is somewhat expensive so only enabled for secure mode 4 +// ------------------------------------------------------ + +#if (MI_ENCODE_FREELIST && (MI_SECURE>=4 || MI_DEBUG!=0)) +// linear check if the free list contains a specific element +static bool mi_list_contains(const mi_page_t* page, const mi_block_t* list, const mi_block_t* elem) { + while (list != NULL) { + if (elem==list) return true; + list = mi_block_next(page, list); + } + return false; +} + +static mi_decl_noinline bool mi_check_is_double_freex(const mi_page_t* page, const mi_block_t* block) { + // The decoded value is in the same page (or NULL). + // Walk the free lists to verify positively if it is already freed + if (mi_list_contains(page, page->free, block) || + mi_list_contains(page, page->local_free, block) || + mi_list_contains(page, mi_page_thread_free(page), block)) + { + _mi_error_message(EAGAIN, "double free detected of block %p with size %zu\n", block, mi_page_block_size(page)); + return true; + } + return false; +} + +static inline bool mi_check_is_double_free(const mi_page_t* page, const mi_block_t* block) { + mi_block_t* n = mi_block_nextx(page, block, page->keys); // pretend it is freed, and get the decoded first field + if (((uintptr_t)n & (MI_INTPTR_SIZE-1))==0 && // quick check: aligned pointer? + (n==NULL || mi_is_in_same_page(block, n))) // quick check: in same page or NULL? + { + // Suspicous: decoded value a in block is in the same page (or NULL) -- maybe a double free? + // (continue in separate function to improve code generation) + return mi_check_is_double_freex(page, block); + } + return false; +} +#else +static inline bool mi_check_is_double_free(const mi_page_t* page, const mi_block_t* block) { + MI_UNUSED(page); + MI_UNUSED(block); + return false; +} +#endif + +// --------------------------------------------------------------------------- +// Check for heap block overflow by setting up padding at the end of the block +// --------------------------------------------------------------------------- + +#if (MI_PADDING>0) && defined(MI_ENCODE_FREELIST) +static bool mi_page_decode_padding(const mi_page_t* page, const mi_block_t* block, size_t* delta, size_t* bsize) { + *bsize = mi_page_usable_block_size(page); + const mi_padding_t* const padding = (mi_padding_t*)((uint8_t*)block + *bsize); + *delta = padding->delta; + return ((uint32_t)mi_ptr_encode(page,block,page->keys) == padding->canary && *delta <= *bsize); +} + +// Return the exact usable size of a block. +static size_t mi_page_usable_size_of(const mi_page_t* page, const mi_block_t* block) { + size_t bsize; + size_t delta; + bool ok = mi_page_decode_padding(page, block, &delta, &bsize); + mi_assert_internal(ok); mi_assert_internal(delta <= bsize); + return (ok ? bsize - delta : 0); +} + +static bool mi_verify_padding(const mi_page_t* page, const mi_block_t* block, size_t* size, size_t* wrong) { + size_t bsize; + size_t delta; + bool ok = mi_page_decode_padding(page, block, &delta, &bsize); + *size = *wrong = bsize; + if (!ok) return false; + mi_assert_internal(bsize >= delta); + *size = bsize - delta; + uint8_t* fill = (uint8_t*)block + bsize - delta; + const size_t maxpad = (delta > MI_MAX_ALIGN_SIZE ? MI_MAX_ALIGN_SIZE : delta); // check at most the first N padding bytes + for (size_t i = 0; i < maxpad; i++) { + if (fill[i] != MI_DEBUG_PADDING) { + *wrong = bsize - delta + i; + return false; + } + } + return true; +} + +static void mi_check_padding(const mi_page_t* page, const mi_block_t* block) { + size_t size; + size_t wrong; + if (!mi_verify_padding(page,block,&size,&wrong)) { + _mi_error_message(EFAULT, "buffer overflow in heap block %p of size %zu: write after %zu bytes\n", block, size, wrong ); + } +} + +// When a non-thread-local block is freed, it becomes part of the thread delayed free +// list that is freed later by the owning heap. If the exact usable size is too small to +// contain the pointer for the delayed list, then shrink the padding (by decreasing delta) +// so it will later not trigger an overflow error in `mi_free_block`. +static void mi_padding_shrink(const mi_page_t* page, const mi_block_t* block, const size_t min_size) { + size_t bsize; + size_t delta; + bool ok = mi_page_decode_padding(page, block, &delta, &bsize); + mi_assert_internal(ok); + if (!ok || (bsize - delta) >= min_size) return; // usually already enough space + mi_assert_internal(bsize >= min_size); + if (bsize < min_size) return; // should never happen + size_t new_delta = (bsize - min_size); + mi_assert_internal(new_delta < bsize); + mi_padding_t* padding = (mi_padding_t*)((uint8_t*)block + bsize); + padding->delta = (uint32_t)new_delta; +} +#else +static void mi_check_padding(const mi_page_t* page, const mi_block_t* block) { + MI_UNUSED(page); + MI_UNUSED(block); +} + +static size_t mi_page_usable_size_of(const mi_page_t* page, const mi_block_t* block) { + MI_UNUSED(block); + return mi_page_usable_block_size(page); +} + +static void mi_padding_shrink(const mi_page_t* page, const mi_block_t* block, const size_t min_size) { + MI_UNUSED(page); + MI_UNUSED(block); + MI_UNUSED(min_size); +} +#endif + +// only maintain stats for smaller objects if requested +#if (MI_STAT>0) +static void mi_stat_free(const mi_page_t* page, const mi_block_t* block) { + #if (MI_STAT < 2) + MI_UNUSED(block); + #endif + mi_heap_t* const heap = mi_heap_get_default(); + const size_t bsize = mi_page_usable_block_size(page); + #if (MI_STAT>1) + const size_t usize = mi_page_usable_size_of(page, block); + mi_heap_stat_decrease(heap, malloc, usize); + #endif + if (bsize <= MI_MEDIUM_OBJ_SIZE_MAX) { + mi_heap_stat_decrease(heap, normal, bsize); + #if (MI_STAT > 1) + mi_heap_stat_decrease(heap, normal_bins[_mi_bin(bsize)], 1); + #endif + } + else if (bsize <= MI_LARGE_OBJ_SIZE_MAX) { + mi_heap_stat_decrease(heap, large, bsize); + } + else { + mi_heap_stat_decrease(heap, huge, bsize); + } +} +#else +static void mi_stat_free(const mi_page_t* page, const mi_block_t* block) { + MI_UNUSED(page); MI_UNUSED(block); +} +#endif + +#if (MI_STAT>0) +// maintain stats for huge objects +static void mi_stat_huge_free(const mi_page_t* page) { + mi_heap_t* const heap = mi_heap_get_default(); + const size_t bsize = mi_page_block_size(page); // to match stats in `page.c:mi_page_huge_alloc` + if (bsize <= MI_LARGE_OBJ_SIZE_MAX) { + mi_heap_stat_decrease(heap, large, bsize); + } + else { + mi_heap_stat_decrease(heap, huge, bsize); + } +} +#else +static void mi_stat_huge_free(const mi_page_t* page) { + MI_UNUSED(page); +} +#endif + +// ------------------------------------------------------ +// Free +// ------------------------------------------------------ + +// multi-threaded free +static mi_decl_noinline void _mi_free_block_mt(mi_page_t* page, mi_block_t* block) +{ + // The padding check may access the non-thread-owned page for the key values. + // that is safe as these are constant and the page won't be freed (as the block is not freed yet). + mi_check_padding(page, block); + mi_padding_shrink(page, block, sizeof(mi_block_t)); // for small size, ensure we can fit the delayed thread pointers without triggering overflow detection + #if (MI_DEBUG!=0) + memset(block, MI_DEBUG_FREED, mi_usable_size(block)); + #endif + + // huge page segments are always abandoned and can be freed immediately + mi_segment_t* segment = _mi_page_segment(page); + if (segment->kind==MI_SEGMENT_HUGE) { + mi_stat_huge_free(page); + _mi_segment_huge_page_free(segment, page, block); + return; + } + + // Try to put the block on either the page-local thread free list, or the heap delayed free list. + mi_thread_free_t tfreex; + bool use_delayed; + mi_thread_free_t tfree = mi_atomic_load_relaxed(&page->xthread_free); + do { + use_delayed = (mi_tf_delayed(tfree) == MI_USE_DELAYED_FREE); + if (mi_unlikely(use_delayed)) { + // unlikely: this only happens on the first concurrent free in a page that is in the full list + tfreex = mi_tf_set_delayed(tfree,MI_DELAYED_FREEING); + } + else { + // usual: directly add to page thread_free list + mi_block_set_next(page, block, mi_tf_block(tfree)); + tfreex = mi_tf_set_block(tfree,block); + } + } while (!mi_atomic_cas_weak_release(&page->xthread_free, &tfree, tfreex)); + + if (mi_unlikely(use_delayed)) { + // racy read on `heap`, but ok because MI_DELAYED_FREEING is set (see `mi_heap_delete` and `mi_heap_collect_abandon`) + mi_heap_t* const heap = (mi_heap_t*)(mi_atomic_load_acquire(&page->xheap)); //mi_page_heap(page); + mi_assert_internal(heap != NULL); + if (heap != NULL) { + // add to the delayed free list of this heap. (do this atomically as the lock only protects heap memory validity) + mi_block_t* dfree = mi_atomic_load_ptr_relaxed(mi_block_t, &heap->thread_delayed_free); + do { + mi_block_set_nextx(heap,block,dfree, heap->keys); + } while (!mi_atomic_cas_ptr_weak_release(mi_block_t,&heap->thread_delayed_free, &dfree, block)); + } + + // and reset the MI_DELAYED_FREEING flag + tfree = mi_atomic_load_relaxed(&page->xthread_free); + do { + tfreex = tfree; + mi_assert_internal(mi_tf_delayed(tfree) == MI_DELAYED_FREEING); + tfreex = mi_tf_set_delayed(tfree,MI_NO_DELAYED_FREE); + } while (!mi_atomic_cas_weak_release(&page->xthread_free, &tfree, tfreex)); + } +} + +// regular free +static inline void _mi_free_block(mi_page_t* page, bool local, mi_block_t* block) +{ + // and push it on the free list + if (mi_likely(local)) { + // owning thread can free a block directly + if (mi_unlikely(mi_check_is_double_free(page, block))) return; + mi_check_padding(page, block); + #if (MI_DEBUG!=0) + memset(block, MI_DEBUG_FREED, mi_page_block_size(page)); + #endif + mi_block_set_next(page, block, page->local_free); + page->local_free = block; + page->used--; + if (mi_unlikely(mi_page_all_free(page))) { + _mi_page_retire(page); + } + else if (mi_unlikely(mi_page_is_in_full(page))) { + _mi_page_unfull(page); + } + } + else { + _mi_free_block_mt(page,block); + } +} + + +// Adjust a block that was allocated aligned, to the actual start of the block in the page. +mi_block_t* _mi_page_ptr_unalign(const mi_segment_t* segment, const mi_page_t* page, const void* p) { + mi_assert_internal(page!=NULL && p!=NULL); + const size_t diff = (uint8_t*)p - _mi_page_start(segment, page, NULL); + const size_t adjust = (diff % mi_page_block_size(page)); + return (mi_block_t*)((uintptr_t)p - adjust); +} + + +static void mi_decl_noinline mi_free_generic(const mi_segment_t* segment, bool local, void* p) mi_attr_noexcept { + mi_page_t* const page = _mi_segment_page_of(segment, p); + mi_block_t* const block = (mi_page_has_aligned(page) ? _mi_page_ptr_unalign(segment, page, p) : (mi_block_t*)p); + mi_stat_free(page, block); + _mi_free_block(page, local, block); +} + +// Get the segment data belonging to a pointer +// This is just a single `and` in assembly but does further checks in debug mode +// (and secure mode) if this was a valid pointer. +static inline mi_segment_t* mi_checked_ptr_segment(const void* p, const char* msg) +{ + MI_UNUSED(msg); +#if (MI_DEBUG>0) + if (mi_unlikely(((uintptr_t)p & (MI_INTPTR_SIZE - 1)) != 0)) { + _mi_error_message(EINVAL, "%s: invalid (unaligned) pointer: %p\n", msg, p); + return NULL; + } +#endif + + mi_segment_t* const segment = _mi_ptr_segment(p); + if (mi_unlikely(segment == NULL)) return NULL; // checks also for (p==NULL) + +#if (MI_DEBUG>0) + if (mi_unlikely(!mi_is_in_heap_region(p))) { + _mi_warning_message("%s: pointer might not point to a valid heap region: %p\n" + "(this may still be a valid very large allocation (over 64MiB))\n", msg, p); + if (mi_likely(_mi_ptr_cookie(segment) == segment->cookie)) { + _mi_warning_message("(yes, the previous pointer %p was valid after all)\n", p); + } + } +#endif +#if (MI_DEBUG>0 || MI_SECURE>=4) + if (mi_unlikely(_mi_ptr_cookie(segment) != segment->cookie)) { + _mi_error_message(EINVAL, "%s: pointer does not point to a valid heap space: %p\n", msg, p); + return NULL; + } +#endif + return segment; +} + +// Free a block +void mi_free(void* p) mi_attr_noexcept +{ + mi_segment_t* const segment = mi_checked_ptr_segment(p,"mi_free"); + if (mi_unlikely(segment == NULL)) return; + + mi_threadid_t tid = _mi_thread_id(); + mi_page_t* const page = _mi_segment_page_of(segment, p); + + if (mi_likely(tid == mi_atomic_load_relaxed(&segment->thread_id) && page->flags.full_aligned == 0)) { // the thread id matches and it is not a full page, nor has aligned blocks + // local, and not full or aligned + mi_block_t* block = (mi_block_t*)(p); + if (mi_unlikely(mi_check_is_double_free(page,block))) return; + mi_check_padding(page, block); + mi_stat_free(page, block); + #if (MI_DEBUG!=0) + memset(block, MI_DEBUG_FREED, mi_page_block_size(page)); + #endif + mi_block_set_next(page, block, page->local_free); + page->local_free = block; + if (mi_unlikely(--page->used == 0)) { // using this expression generates better code than: page->used--; if (mi_page_all_free(page)) + _mi_page_retire(page); + } + } + else { + // non-local, aligned blocks, or a full page; use the more generic path + // note: recalc page in generic to improve code generation + mi_free_generic(segment, tid == segment->thread_id, p); + } +} + +bool _mi_free_delayed_block(mi_block_t* block) { + // get segment and page + const mi_segment_t* const segment = _mi_ptr_segment(block); + mi_assert_internal(_mi_ptr_cookie(segment) == segment->cookie); + mi_assert_internal(_mi_thread_id() == segment->thread_id); + mi_page_t* const page = _mi_segment_page_of(segment, block); + + // Clear the no-delayed flag so delayed freeing is used again for this page. + // This must be done before collecting the free lists on this page -- otherwise + // some blocks may end up in the page `thread_free` list with no blocks in the + // heap `thread_delayed_free` list which may cause the page to be never freed! + // (it would only be freed if we happen to scan it in `mi_page_queue_find_free_ex`) + _mi_page_use_delayed_free(page, MI_USE_DELAYED_FREE, false /* dont overwrite never delayed */); + + // collect all other non-local frees to ensure up-to-date `used` count + _mi_page_free_collect(page, false); + + // and free the block (possibly freeing the page as well since used is updated) + _mi_free_block(page, true, block); + return true; +} + +// Bytes available in a block +mi_decl_noinline static size_t mi_page_usable_aligned_size_of(const mi_segment_t* segment, const mi_page_t* page, const void* p) mi_attr_noexcept { + const mi_block_t* block = _mi_page_ptr_unalign(segment, page, p); + const size_t size = mi_page_usable_size_of(page, block); + const ptrdiff_t adjust = (uint8_t*)p - (uint8_t*)block; + mi_assert_internal(adjust >= 0 && (size_t)adjust <= size); + return (size - adjust); +} + +static inline size_t _mi_usable_size(const void* p, const char* msg) mi_attr_noexcept { + const mi_segment_t* const segment = mi_checked_ptr_segment(p, msg); + if (segment==NULL) return 0; // also returns 0 if `p == NULL` + const mi_page_t* const page = _mi_segment_page_of(segment, p); + if (mi_likely(!mi_page_has_aligned(page))) { + const mi_block_t* block = (const mi_block_t*)p; + return mi_page_usable_size_of(page, block); + } + else { + // split out to separate routine for improved code generation + return mi_page_usable_aligned_size_of(segment, page, p); + } +} + +size_t mi_usable_size(const void* p) mi_attr_noexcept { + return _mi_usable_size(p, "mi_usable_size"); +} + + +// ------------------------------------------------------ +// ensure explicit external inline definitions are emitted! +// ------------------------------------------------------ + +#ifdef __cplusplus +void* _mi_externs[] = { + (void*)&_mi_page_malloc, + (void*)&mi_malloc, + (void*)&mi_malloc_small, + (void*)&mi_zalloc_small, + (void*)&mi_heap_malloc, + (void*)&mi_heap_zalloc, + (void*)&mi_heap_malloc_small +}; +#endif + + +// ------------------------------------------------------ +// Allocation extensions +// ------------------------------------------------------ + +void mi_free_size(void* p, size_t size) mi_attr_noexcept { + MI_UNUSED_RELEASE(size); + mi_assert(p == NULL || size <= _mi_usable_size(p,"mi_free_size")); + mi_free(p); +} + +void mi_free_size_aligned(void* p, size_t size, size_t alignment) mi_attr_noexcept { + MI_UNUSED_RELEASE(alignment); + mi_assert(((uintptr_t)p % alignment) == 0); + mi_free_size(p,size); +} + +void mi_free_aligned(void* p, size_t alignment) mi_attr_noexcept { + MI_UNUSED_RELEASE(alignment); + mi_assert(((uintptr_t)p % alignment) == 0); + mi_free(p); +} + +extern inline mi_decl_restrict void* mi_heap_calloc(mi_heap_t* heap, size_t count, size_t size) mi_attr_noexcept { + size_t total; + if (mi_count_size_overflow(count,size,&total)) return NULL; + return mi_heap_zalloc(heap,total); +} + +mi_decl_restrict void* mi_calloc(size_t count, size_t size) mi_attr_noexcept { + return mi_heap_calloc(mi_get_default_heap(),count,size); +} + +// Uninitialized `calloc` +extern mi_decl_restrict void* mi_heap_mallocn(mi_heap_t* heap, size_t count, size_t size) mi_attr_noexcept { + size_t total; + if (mi_count_size_overflow(count, size, &total)) return NULL; + return mi_heap_malloc(heap, total); +} + +mi_decl_restrict void* mi_mallocn(size_t count, size_t size) mi_attr_noexcept { + return mi_heap_mallocn(mi_get_default_heap(),count,size); +} + +// Expand (or shrink) in place (or fail) +void* mi_expand(void* p, size_t newsize) mi_attr_noexcept { + #if MI_PADDING + // we do not shrink/expand with padding enabled + MI_UNUSED(p); MI_UNUSED(newsize); + return NULL; + #else + if (p == NULL) return NULL; + const size_t size = _mi_usable_size(p,"mi_expand"); + if (newsize > size) return NULL; + return p; // it fits + #endif +} + +void* _mi_heap_realloc_zero(mi_heap_t* heap, void* p, size_t newsize, bool zero) mi_attr_noexcept { + const size_t size = _mi_usable_size(p,"mi_realloc"); // also works if p == NULL + if (mi_unlikely(newsize <= size && newsize >= (size / 2))) { + // todo: adjust potential padding to reflect the new size? + return p; // reallocation still fits and not more than 50% waste + } + void* newp = mi_heap_malloc(heap,newsize); + if (mi_likely(newp != NULL)) { + if (zero && newsize > size) { + // also set last word in the previous allocation to zero to ensure any padding is zero-initialized + const size_t start = (size >= sizeof(intptr_t) ? size - sizeof(intptr_t) : 0); + memset((uint8_t*)newp + start, 0, newsize - start); + } + if (mi_likely(p != NULL)) { + _mi_memcpy_aligned(newp, p, (newsize > size ? size : newsize)); + mi_free(p); // only free the original pointer if successful + } + } + return newp; +} + +void* mi_heap_realloc(mi_heap_t* heap, void* p, size_t newsize) mi_attr_noexcept { + return _mi_heap_realloc_zero(heap, p, newsize, false); +} + +void* mi_heap_reallocn(mi_heap_t* heap, void* p, size_t count, size_t size) mi_attr_noexcept { + size_t total; + if (mi_count_size_overflow(count, size, &total)) return NULL; + return mi_heap_realloc(heap, p, total); +} + + +// Reallocate but free `p` on errors +void* mi_heap_reallocf(mi_heap_t* heap, void* p, size_t newsize) mi_attr_noexcept { + void* newp = mi_heap_realloc(heap, p, newsize); + if (newp==NULL && p!=NULL) mi_free(p); + return newp; +} + +void* mi_heap_rezalloc(mi_heap_t* heap, void* p, size_t newsize) mi_attr_noexcept { + return _mi_heap_realloc_zero(heap, p, newsize, true); +} + +void* mi_heap_recalloc(mi_heap_t* heap, void* p, size_t count, size_t size) mi_attr_noexcept { + size_t total; + if (mi_count_size_overflow(count, size, &total)) return NULL; + return mi_heap_rezalloc(heap, p, total); +} + + +void* mi_realloc(void* p, size_t newsize) mi_attr_noexcept { + return mi_heap_realloc(mi_get_default_heap(),p,newsize); +} + +void* mi_reallocn(void* p, size_t count, size_t size) mi_attr_noexcept { + return mi_heap_reallocn(mi_get_default_heap(),p,count,size); +} + +// Reallocate but free `p` on errors +void* mi_reallocf(void* p, size_t newsize) mi_attr_noexcept { + return mi_heap_reallocf(mi_get_default_heap(),p,newsize); +} + +void* mi_rezalloc(void* p, size_t newsize) mi_attr_noexcept { + return mi_heap_rezalloc(mi_get_default_heap(), p, newsize); +} + +void* mi_recalloc(void* p, size_t count, size_t size) mi_attr_noexcept { + return mi_heap_recalloc(mi_get_default_heap(), p, count, size); +} + + + +// ------------------------------------------------------ +// strdup, strndup, and realpath +// ------------------------------------------------------ + +// `strdup` using mi_malloc +mi_decl_restrict char* mi_heap_strdup(mi_heap_t* heap, const char* s) mi_attr_noexcept { + if (s == NULL) return NULL; + size_t n = strlen(s); + char* t = (char*)mi_heap_malloc(heap,n+1); + if (t != NULL) _mi_memcpy(t, s, n + 1); + return t; +} + +mi_decl_restrict char* mi_strdup(const char* s) mi_attr_noexcept { + return mi_heap_strdup(mi_get_default_heap(), s); +} + +// `strndup` using mi_malloc +mi_decl_restrict char* mi_heap_strndup(mi_heap_t* heap, const char* s, size_t n) mi_attr_noexcept { + if (s == NULL) return NULL; + const char* end = (const char*)memchr(s, 0, n); // find end of string in the first `n` characters (returns NULL if not found) + const size_t m = (end != NULL ? (size_t)(end - s) : n); // `m` is the minimum of `n` or the end-of-string + mi_assert_internal(m <= n); + char* t = (char*)mi_heap_malloc(heap, m+1); + if (t == NULL) return NULL; + _mi_memcpy(t, s, m); + t[m] = 0; + return t; +} + +mi_decl_restrict char* mi_strndup(const char* s, size_t n) mi_attr_noexcept { + return mi_heap_strndup(mi_get_default_heap(),s,n); +} + +#ifndef __wasi__ +// `realpath` using mi_malloc +#ifdef _WIN32 +#ifndef PATH_MAX +#define PATH_MAX MAX_PATH +#endif +#include +mi_decl_restrict char* mi_heap_realpath(mi_heap_t* heap, const char* fname, char* resolved_name) mi_attr_noexcept { + // todo: use GetFullPathNameW to allow longer file names + char buf[PATH_MAX]; + DWORD res = GetFullPathNameA(fname, PATH_MAX, (resolved_name == NULL ? buf : resolved_name), NULL); + if (res == 0) { + errno = GetLastError(); return NULL; + } + else if (res > PATH_MAX) { + errno = EINVAL; return NULL; + } + else if (resolved_name != NULL) { + return resolved_name; + } + else { + return mi_heap_strndup(heap, buf, PATH_MAX); + } +} +#else +#include // pathconf +static size_t mi_path_max(void) { + static size_t path_max = 0; + if (path_max <= 0) { + long m = pathconf("/",_PC_PATH_MAX); + if (m <= 0) path_max = 4096; // guess + else if (m < 256) path_max = 256; // at least 256 + else path_max = m; + } + return path_max; +} + +char* mi_heap_realpath(mi_heap_t* heap, const char* fname, char* resolved_name) mi_attr_noexcept { + if (resolved_name != NULL) { + return realpath(fname,resolved_name); + } + else { + size_t n = mi_path_max(); + char* buf = (char*)mi_malloc(n+1); + if (buf==NULL) return NULL; + char* rname = realpath(fname,buf); + char* result = mi_heap_strndup(heap,rname,n); // ok if `rname==NULL` + mi_free(buf); + return result; + } +} +#endif + +mi_decl_restrict char* mi_realpath(const char* fname, char* resolved_name) mi_attr_noexcept { + return mi_heap_realpath(mi_get_default_heap(),fname,resolved_name); +} +#endif + +/*------------------------------------------------------- +C++ new and new_aligned +The standard requires calling into `get_new_handler` and +throwing the bad_alloc exception on failure. If we compile +with a C++ compiler we can implement this precisely. If we +use a C compiler we cannot throw a `bad_alloc` exception +but we call `exit` instead (i.e. not returning). +-------------------------------------------------------*/ + +#ifdef __cplusplus +#include +static bool mi_try_new_handler(bool nothrow) { + #if defined(_MSC_VER) || (__cplusplus >= 201103L) + std::new_handler h = std::get_new_handler(); + #else + std::new_handler h = std::set_new_handler(); + std::set_new_handler(h); + #endif + if (h==NULL) { + _mi_error_message(ENOMEM, "out of memory in 'new'"); + if (!nothrow) { + throw std::bad_alloc(); + } + return false; + } + else { + h(); + return true; + } +} +#else +typedef void (*std_new_handler_t)(void); + +#if (defined(__GNUC__) || defined(__clang__)) +std_new_handler_t __attribute((weak)) _ZSt15get_new_handlerv(void) { + return NULL; +} +static std_new_handler_t mi_get_new_handler(void) { + return _ZSt15get_new_handlerv(); +} +#else +// note: on windows we could dynamically link to `?get_new_handler@std@@YAP6AXXZXZ`. +static std_new_handler_t mi_get_new_handler() { + return NULL; +} +#endif + +static bool mi_try_new_handler(bool nothrow) { + std_new_handler_t h = mi_get_new_handler(); + if (h==NULL) { + _mi_error_message(ENOMEM, "out of memory in 'new'"); + if (!nothrow) { + abort(); // cannot throw in plain C, use abort + } + return false; + } + else { + h(); + return true; + } +} +#endif + +static mi_decl_noinline void* mi_try_new(size_t size, bool nothrow ) { + void* p = NULL; + while(p == NULL && mi_try_new_handler(nothrow)) { + p = mi_malloc(size); + } + return p; +} + +mi_decl_restrict void* mi_new(size_t size) { + void* p = mi_malloc(size); + if (mi_unlikely(p == NULL)) return mi_try_new(size,false); + return p; +} + +mi_decl_restrict void* mi_new_nothrow(size_t size) mi_attr_noexcept { + void* p = mi_malloc(size); + if (mi_unlikely(p == NULL)) return mi_try_new(size, true); + return p; +} + +mi_decl_restrict void* mi_new_aligned(size_t size, size_t alignment) { + void* p; + do { + p = mi_malloc_aligned(size, alignment); + } + while(p == NULL && mi_try_new_handler(false)); + return p; +} + +mi_decl_restrict void* mi_new_aligned_nothrow(size_t size, size_t alignment) mi_attr_noexcept { + void* p; + do { + p = mi_malloc_aligned(size, alignment); + } + while(p == NULL && mi_try_new_handler(true)); + return p; +} + +mi_decl_restrict void* mi_new_n(size_t count, size_t size) { + size_t total; + if (mi_unlikely(mi_count_size_overflow(count, size, &total))) { + mi_try_new_handler(false); // on overflow we invoke the try_new_handler once to potentially throw std::bad_alloc + return NULL; + } + else { + return mi_new(total); + } +} + +void* mi_new_realloc(void* p, size_t newsize) { + void* q; + do { + q = mi_realloc(p, newsize); + } while (q == NULL && mi_try_new_handler(false)); + return q; +} + +void* mi_new_reallocn(void* p, size_t newcount, size_t size) { + size_t total; + if (mi_unlikely(mi_count_size_overflow(newcount, size, &total))) { + mi_try_new_handler(false); // on overflow we invoke the try_new_handler once to potentially throw std::bad_alloc + return NULL; + } + else { + return mi_new_realloc(p, total); + } +} diff --git a/compat/mimalloc/arena.c b/compat/mimalloc/arena.c new file mode 100644 index 00000000000000..567c8a93ac30c8 --- /dev/null +++ b/compat/mimalloc/arena.c @@ -0,0 +1,446 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2019-2021, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ + +/* ---------------------------------------------------------------------------- +"Arenas" are fixed area's of OS memory from which we can allocate +large blocks (>= MI_ARENA_MIN_BLOCK_SIZE, 4MiB). +In contrast to the rest of mimalloc, the arenas are shared between +threads and need to be accessed using atomic operations. + +Currently arenas are only used to for huge OS page (1GiB) reservations, +or direct OS memory reservations -- otherwise it delegates to direct allocation from the OS. +In the future, we can expose an API to manually add more kinds of arenas +which is sometimes needed for embedded devices or shared memory for example. +(We can also employ this with WASI or `sbrk` systems to reserve large arenas + on demand and be able to reuse them efficiently). + +The arena allocation needs to be thread safe and we use an atomic bitmap to allocate. +-----------------------------------------------------------------------------*/ +#include "mimalloc.h" +#include "mimalloc-internal.h" +#include "mimalloc-atomic.h" + +#include // memset +#include // ENOMEM + +#include "bitmap.h" // atomic bitmap + + +// os.c +void* _mi_os_alloc_aligned(size_t size, size_t alignment, bool commit, bool* large, mi_stats_t* stats); +void _mi_os_free_ex(void* p, size_t size, bool was_committed, mi_stats_t* stats); + +void* _mi_os_alloc_huge_os_pages(size_t pages, int numa_node, mi_msecs_t max_secs, size_t* pages_reserved, size_t* psize); +void _mi_os_free_huge_pages(void* p, size_t size, mi_stats_t* stats); + +bool _mi_os_commit(void* p, size_t size, bool* is_zero, mi_stats_t* stats); +bool _mi_os_decommit(void* addr, size_t size, mi_stats_t* stats); + + +/* ----------------------------------------------------------- + Arena allocation +----------------------------------------------------------- */ + + +// Block info: bit 0 contains the `in_use` bit, the upper bits the +// size in count of arena blocks. +typedef uintptr_t mi_block_info_t; +#define MI_ARENA_BLOCK_SIZE (MI_SEGMENT_SIZE) // 8MiB (must be at least MI_SEGMENT_ALIGN) +#define MI_ARENA_MIN_OBJ_SIZE (MI_ARENA_BLOCK_SIZE/2) // 4MiB +#define MI_MAX_ARENAS (64) // not more than 256 (since we use 8 bits in the memid) + +// A memory arena descriptor +typedef struct mi_arena_s { + _Atomic(uint8_t*) start; // the start of the memory area + size_t block_count; // size of the area in arena blocks (of `MI_ARENA_BLOCK_SIZE`) + size_t field_count; // number of bitmap fields (where `field_count * MI_BITMAP_FIELD_BITS >= block_count`) + int numa_node; // associated NUMA node + bool is_zero_init; // is the arena zero initialized? + bool allow_decommit; // is decommit allowed? if true, is_large should be false and blocks_committed != NULL + bool is_large; // large- or huge OS pages (always committed) + _Atomic(size_t) search_idx; // optimization to start the search for free blocks + mi_bitmap_field_t* blocks_dirty; // are the blocks potentially non-zero? + mi_bitmap_field_t* blocks_committed; // are the blocks committed? (can be NULL for memory that cannot be decommitted) + mi_bitmap_field_t blocks_inuse[1]; // in-place bitmap of in-use blocks (of size `field_count`) +} mi_arena_t; + + +// The available arenas +static mi_decl_cache_align _Atomic(mi_arena_t*) mi_arenas[MI_MAX_ARENAS]; +static mi_decl_cache_align _Atomic(size_t) mi_arena_count; // = 0 + + +/* ----------------------------------------------------------- + Arena allocations get a memory id where the lower 8 bits are + the arena index +1, and the upper bits the block index. +----------------------------------------------------------- */ + +// Use `0` as a special id for direct OS allocated memory. +#define MI_MEMID_OS 0 + +static size_t mi_arena_id_create(size_t arena_index, mi_bitmap_index_t bitmap_index) { + mi_assert_internal(arena_index < 0xFE); + mi_assert_internal(((bitmap_index << 8) >> 8) == bitmap_index); // no overflow? + return ((bitmap_index << 8) | ((arena_index+1) & 0xFF)); +} + +static void mi_arena_id_indices(size_t memid, size_t* arena_index, mi_bitmap_index_t* bitmap_index) { + mi_assert_internal(memid != MI_MEMID_OS); + *arena_index = (memid & 0xFF) - 1; + *bitmap_index = (memid >> 8); +} + +static size_t mi_block_count_of_size(size_t size) { + return _mi_divide_up(size, MI_ARENA_BLOCK_SIZE); +} + +/* ----------------------------------------------------------- + Thread safe allocation in an arena +----------------------------------------------------------- */ +static bool mi_arena_alloc(mi_arena_t* arena, size_t blocks, mi_bitmap_index_t* bitmap_idx) +{ + size_t idx = 0; // mi_atomic_load_relaxed(&arena->search_idx); // start from last search; ok to be relaxed as the exact start does not matter + if (_mi_bitmap_try_find_from_claim_across(arena->blocks_inuse, arena->field_count, idx, blocks, bitmap_idx)) { + mi_atomic_store_relaxed(&arena->search_idx, mi_bitmap_index_field(*bitmap_idx)); // start search from found location next time around + return true; + }; + return false; +} + + +/* ----------------------------------------------------------- + Arena Allocation +----------------------------------------------------------- */ + +static mi_decl_noinline void* mi_arena_alloc_from(mi_arena_t* arena, size_t arena_index, size_t needed_bcount, + bool* commit, bool* large, bool* is_pinned, bool* is_zero, size_t* memid, mi_os_tld_t* tld) +{ + mi_bitmap_index_t bitmap_index; + if (!mi_arena_alloc(arena, needed_bcount, &bitmap_index)) return NULL; + + // claimed it! set the dirty bits (todo: no need for an atomic op here?) + void* p = arena->start + (mi_bitmap_index_bit(bitmap_index)*MI_ARENA_BLOCK_SIZE); + *memid = mi_arena_id_create(arena_index, bitmap_index); + *is_zero = _mi_bitmap_claim_across(arena->blocks_dirty, arena->field_count, needed_bcount, bitmap_index, NULL); + *large = arena->is_large; + *is_pinned = (arena->is_large || !arena->allow_decommit); + if (arena->blocks_committed == NULL) { + // always committed + *commit = true; + } + else if (*commit) { + // arena not committed as a whole, but commit requested: ensure commit now + bool any_uncommitted; + _mi_bitmap_claim_across(arena->blocks_committed, arena->field_count, needed_bcount, bitmap_index, &any_uncommitted); + if (any_uncommitted) { + bool commit_zero; + _mi_os_commit(p, needed_bcount * MI_ARENA_BLOCK_SIZE, &commit_zero, tld->stats); + if (commit_zero) *is_zero = true; + } + } + else { + // no need to commit, but check if already fully committed + *commit = _mi_bitmap_is_claimed_across(arena->blocks_committed, arena->field_count, needed_bcount, bitmap_index); + } + return p; +} + +static mi_decl_noinline void* mi_arena_allocate(int numa_node, size_t size, size_t alignment, bool* commit, bool* large, bool* is_pinned, bool* is_zero, size_t* memid, mi_os_tld_t* tld) +{ + MI_UNUSED_RELEASE(alignment); + mi_assert_internal(alignment <= MI_SEGMENT_ALIGN); + const size_t max_arena = mi_atomic_load_relaxed(&mi_arena_count); + const size_t bcount = mi_block_count_of_size(size); + if (mi_likely(max_arena == 0)) return NULL; + mi_assert_internal(size <= bcount*MI_ARENA_BLOCK_SIZE); + + // try numa affine allocation + for (size_t i = 0; i < max_arena; i++) { + mi_arena_t* arena = mi_atomic_load_ptr_relaxed(mi_arena_t, &mi_arenas[i]); + if (arena==NULL) break; // end reached + if ((arena->numa_node<0 || arena->numa_node==numa_node) && // numa local? + (*large || !arena->is_large)) // large OS pages allowed, or arena is not large OS pages + { + void* p = mi_arena_alloc_from(arena, i, bcount, commit, large, is_pinned, is_zero, memid, tld); + mi_assert_internal((uintptr_t)p % alignment == 0); + if (p != NULL) { + return p; + } + } + } + + // try from another numa node instead.. + for (size_t i = 0; i < max_arena; i++) { + mi_arena_t* arena = mi_atomic_load_ptr_relaxed(mi_arena_t, &mi_arenas[i]); + if (arena==NULL) break; // end reached + if ((arena->numa_node>=0 && arena->numa_node!=numa_node) && // not numa local! + (*large || !arena->is_large)) // large OS pages allowed, or arena is not large OS pages + { + void* p = mi_arena_alloc_from(arena, i, bcount, commit, large, is_pinned, is_zero, memid, tld); + mi_assert_internal((uintptr_t)p % alignment == 0); + if (p != NULL) { + return p; + } + } + } + return NULL; +} + + +void* _mi_arena_alloc_aligned(size_t size, size_t alignment, bool* commit, bool* large, bool* is_pinned, bool* is_zero, + size_t* memid, mi_os_tld_t* tld) +{ + mi_assert_internal(commit != NULL && is_pinned != NULL && is_zero != NULL && memid != NULL && tld != NULL); + mi_assert_internal(size > 0); + *memid = MI_MEMID_OS; + *is_zero = false; + *is_pinned = false; + + bool default_large = false; + if (large==NULL) large = &default_large; // ensure `large != NULL` + const int numa_node = _mi_os_numa_node(tld); // current numa node + + // try to allocate in an arena if the alignment is small enough and the object is not too small (as for heap meta data) + if (size >= MI_ARENA_MIN_OBJ_SIZE && alignment <= MI_SEGMENT_ALIGN) { + void* p = mi_arena_allocate(numa_node, size, alignment, commit, large, is_pinned, is_zero, memid, tld); + if (p != NULL) return p; + } + + // finally, fall back to the OS + if (mi_option_is_enabled(mi_option_limit_os_alloc)) { + errno = ENOMEM; + return NULL; + } + *is_zero = true; + *memid = MI_MEMID_OS; + void* p = _mi_os_alloc_aligned(size, alignment, *commit, large, tld->stats); + if (p != NULL) *is_pinned = *large; + return p; +} + +void* _mi_arena_alloc(size_t size, bool* commit, bool* large, bool* is_pinned, bool* is_zero, size_t* memid, mi_os_tld_t* tld) +{ + return _mi_arena_alloc_aligned(size, MI_ARENA_BLOCK_SIZE, commit, large, is_pinned, is_zero, memid, tld); +} + +/* ----------------------------------------------------------- + Arena free +----------------------------------------------------------- */ + +void _mi_arena_free(void* p, size_t size, size_t memid, bool all_committed, mi_os_tld_t* tld) { + mi_assert_internal(size > 0 && tld->stats != NULL); + if (p==NULL) return; + if (size==0) return; + + if (memid == MI_MEMID_OS) { + // was a direct OS allocation, pass through + _mi_os_free_ex(p, size, all_committed, tld->stats); + } + else { + // allocated in an arena + size_t arena_idx; + size_t bitmap_idx; + mi_arena_id_indices(memid, &arena_idx, &bitmap_idx); + mi_assert_internal(arena_idx < MI_MAX_ARENAS); + mi_arena_t* arena = mi_atomic_load_ptr_relaxed(mi_arena_t,&mi_arenas[arena_idx]); + mi_assert_internal(arena != NULL); + const size_t blocks = mi_block_count_of_size(size); + // checks + if (arena == NULL) { + _mi_error_message(EINVAL, "trying to free from non-existent arena: %p, size %zu, memid: 0x%zx\n", p, size, memid); + return; + } + mi_assert_internal(arena->field_count > mi_bitmap_index_field(bitmap_idx)); + if (arena->field_count <= mi_bitmap_index_field(bitmap_idx)) { + _mi_error_message(EINVAL, "trying to free from non-existent arena block: %p, size %zu, memid: 0x%zx\n", p, size, memid); + return; + } + // potentially decommit + if (!arena->allow_decommit || arena->blocks_committed == NULL) { + mi_assert_internal(all_committed); // note: may be not true as we may "pretend" to be not committed (in segment.c) + } + else { + mi_assert_internal(arena->blocks_committed != NULL); + _mi_os_decommit(p, blocks * MI_ARENA_BLOCK_SIZE, tld->stats); // ok if this fails + _mi_bitmap_unclaim_across(arena->blocks_committed, arena->field_count, blocks, bitmap_idx); + } + // and make it available to others again + bool all_inuse = _mi_bitmap_unclaim_across(arena->blocks_inuse, arena->field_count, blocks, bitmap_idx); + if (!all_inuse) { + _mi_error_message(EAGAIN, "trying to free an already freed block: %p, size %zu\n", p, size); + return; + }; + } +} + +/* ----------------------------------------------------------- + Add an arena. +----------------------------------------------------------- */ + +static bool mi_arena_add(mi_arena_t* arena) { + mi_assert_internal(arena != NULL); + mi_assert_internal((uintptr_t)mi_atomic_load_ptr_relaxed(uint8_t,&arena->start) % MI_SEGMENT_ALIGN == 0); + mi_assert_internal(arena->block_count > 0); + + size_t i = mi_atomic_increment_acq_rel(&mi_arena_count); + if (i >= MI_MAX_ARENAS) { + mi_atomic_decrement_acq_rel(&mi_arena_count); + return false; + } + mi_atomic_store_ptr_release(mi_arena_t,&mi_arenas[i], arena); + return true; +} + +bool mi_manage_os_memory(void* start, size_t size, bool is_committed, bool is_large, bool is_zero, int numa_node) mi_attr_noexcept +{ + if (size < MI_ARENA_BLOCK_SIZE) return false; + + if (is_large) { + mi_assert_internal(is_committed); + is_committed = true; + } + + const size_t bcount = size / MI_ARENA_BLOCK_SIZE; + const size_t fields = _mi_divide_up(bcount, MI_BITMAP_FIELD_BITS); + const size_t bitmaps = (is_committed ? 2 : 3); + const size_t asize = sizeof(mi_arena_t) + (bitmaps*fields*sizeof(mi_bitmap_field_t)); + mi_arena_t* arena = (mi_arena_t*)_mi_os_alloc(asize, &_mi_stats_main); // TODO: can we avoid allocating from the OS? + if (arena == NULL) return false; + + arena->block_count = bcount; + arena->field_count = fields; + arena->start = (uint8_t*)start; + arena->numa_node = numa_node; // TODO: or get the current numa node if -1? (now it allows anyone to allocate on -1) + arena->is_large = is_large; + arena->is_zero_init = is_zero; + arena->allow_decommit = !is_large && !is_committed; // only allow decommit for initially uncommitted memory + arena->search_idx = 0; + arena->blocks_dirty = &arena->blocks_inuse[fields]; // just after inuse bitmap + arena->blocks_committed = (!arena->allow_decommit ? NULL : &arena->blocks_inuse[2*fields]); // just after dirty bitmap + // the bitmaps are already zero initialized due to os_alloc + // initialize committed bitmap? + if (arena->blocks_committed != NULL && is_committed) { + memset((void*)arena->blocks_committed, 0xFF, fields*sizeof(mi_bitmap_field_t)); // cast to void* to avoid atomic warning + } + // and claim leftover blocks if needed (so we never allocate there) + ptrdiff_t post = (fields * MI_BITMAP_FIELD_BITS) - bcount; + mi_assert_internal(post >= 0); + if (post > 0) { + // don't use leftover bits at the end + mi_bitmap_index_t postidx = mi_bitmap_index_create(fields - 1, MI_BITMAP_FIELD_BITS - post); + _mi_bitmap_claim(arena->blocks_inuse, fields, post, postidx, NULL); + } + + mi_arena_add(arena); + return true; +} + +// Reserve a range of regular OS memory +int mi_reserve_os_memory(size_t size, bool commit, bool allow_large) mi_attr_noexcept +{ + size = _mi_align_up(size, MI_ARENA_BLOCK_SIZE); // at least one block + bool large = allow_large; + void* start = _mi_os_alloc_aligned(size, MI_SEGMENT_ALIGN, commit, &large, &_mi_stats_main); + if (start==NULL) return ENOMEM; + if (!mi_manage_os_memory(start, size, (large || commit), large, true, -1)) { + _mi_os_free_ex(start, size, commit, &_mi_stats_main); + _mi_verbose_message("failed to reserve %zu k memory\n", _mi_divide_up(size,1024)); + return ENOMEM; + } + _mi_verbose_message("reserved %zu KiB memory%s\n", _mi_divide_up(size,1024), large ? " (in large os pages)" : ""); + return 0; +} + +static size_t mi_debug_show_bitmap(const char* prefix, mi_bitmap_field_t* fields, size_t field_count ) { + size_t inuse_count = 0; + for (size_t i = 0; i < field_count; i++) { + char buf[MI_BITMAP_FIELD_BITS + 1]; + uintptr_t field = mi_atomic_load_relaxed(&fields[i]); + for (size_t bit = 0; bit < MI_BITMAP_FIELD_BITS; bit++) { + bool inuse = ((((uintptr_t)1 << bit) & field) != 0); + if (inuse) inuse_count++; + buf[MI_BITMAP_FIELD_BITS - 1 - bit] = (inuse ? 'x' : '.'); + } + buf[MI_BITMAP_FIELD_BITS] = 0; + _mi_verbose_message("%s%s\n", prefix, buf); + } + return inuse_count; +} + +void mi_debug_show_arenas(void) mi_attr_noexcept { + size_t max_arenas = mi_atomic_load_relaxed(&mi_arena_count); + for (size_t i = 0; i < max_arenas; i++) { + mi_arena_t* arena = mi_atomic_load_ptr_relaxed(mi_arena_t, &mi_arenas[i]); + if (arena == NULL) break; + size_t inuse_count = 0; + _mi_verbose_message("arena %zu: %zu blocks with %zu fields\n", i, arena->block_count, arena->field_count); + inuse_count += mi_debug_show_bitmap(" ", arena->blocks_inuse, arena->field_count); + _mi_verbose_message(" blocks in use ('x'): %zu\n", inuse_count); + } +} + +/* ----------------------------------------------------------- + Reserve a huge page arena. +----------------------------------------------------------- */ +// reserve at a specific numa node +int mi_reserve_huge_os_pages_at(size_t pages, int numa_node, size_t timeout_msecs) mi_attr_noexcept { + if (pages==0) return 0; + if (numa_node < -1) numa_node = -1; + if (numa_node >= 0) numa_node = numa_node % _mi_os_numa_node_count(); + size_t hsize = 0; + size_t pages_reserved = 0; + void* p = _mi_os_alloc_huge_os_pages(pages, numa_node, timeout_msecs, &pages_reserved, &hsize); + if (p==NULL || pages_reserved==0) { + _mi_warning_message("failed to reserve %zu GiB huge pages\n", pages); + return ENOMEM; + } + _mi_verbose_message("numa node %i: reserved %zu GiB huge pages (of the %zu GiB requested)\n", numa_node, pages_reserved, pages); + + if (!mi_manage_os_memory(p, hsize, true, true, true, numa_node)) { + _mi_os_free_huge_pages(p, hsize, &_mi_stats_main); + return ENOMEM; + } + return 0; +} + + +// reserve huge pages evenly among the given number of numa nodes (or use the available ones as detected) +int mi_reserve_huge_os_pages_interleave(size_t pages, size_t numa_nodes, size_t timeout_msecs) mi_attr_noexcept { + if (pages == 0) return 0; + + // pages per numa node + size_t numa_count = (numa_nodes > 0 ? numa_nodes : _mi_os_numa_node_count()); + if (numa_count <= 0) numa_count = 1; + const size_t pages_per = pages / numa_count; + const size_t pages_mod = pages % numa_count; + const size_t timeout_per = (timeout_msecs==0 ? 0 : (timeout_msecs / numa_count) + 50); + + // reserve evenly among numa nodes + for (size_t numa_node = 0; numa_node < numa_count && pages > 0; numa_node++) { + size_t node_pages = pages_per; // can be 0 + if (numa_node < pages_mod) node_pages++; + int err = mi_reserve_huge_os_pages_at(node_pages, (int)numa_node, timeout_per); + if (err) return err; + if (pages < node_pages) { + pages = 0; + } + else { + pages -= node_pages; + } + } + + return 0; +} + +int mi_reserve_huge_os_pages(size_t pages, double max_secs, size_t* pages_reserved) mi_attr_noexcept { + MI_UNUSED(max_secs); + _mi_warning_message("mi_reserve_huge_os_pages is deprecated: use mi_reserve_huge_os_pages_interleave/at instead\n"); + if (pages_reserved != NULL) *pages_reserved = 0; + int err = mi_reserve_huge_os_pages_interleave(pages, 0, (size_t)(max_secs * 1000.0)); + if (err==0 && pages_reserved!=NULL) *pages_reserved = pages; + return err; +} diff --git a/compat/mimalloc/bitmap.c b/compat/mimalloc/bitmap.c new file mode 100644 index 00000000000000..8634b32ab13fa9 --- /dev/null +++ b/compat/mimalloc/bitmap.c @@ -0,0 +1,395 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2019-2021 Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ + +/* ---------------------------------------------------------------------------- +Concurrent bitmap that can set/reset sequences of bits atomically, +represeted as an array of fields where each field is a machine word (`size_t`) + +There are two api's; the standard one cannot have sequences that cross +between the bitmap fields (and a sequence must be <= MI_BITMAP_FIELD_BITS). +(this is used in region allocation) + +The `_across` postfixed functions do allow sequences that can cross over +between the fields. (This is used in arena allocation) +---------------------------------------------------------------------------- */ + +#include "mimalloc.h" +#include "mimalloc-internal.h" +#include "bitmap.h" + +/* ----------------------------------------------------------- + Bitmap definition +----------------------------------------------------------- */ + +// The bit mask for a given number of blocks at a specified bit index. +static inline size_t mi_bitmap_mask_(size_t count, size_t bitidx) { + mi_assert_internal(count + bitidx <= MI_BITMAP_FIELD_BITS); + mi_assert_internal(count > 0); + if (count >= MI_BITMAP_FIELD_BITS) return MI_BITMAP_FIELD_FULL; + if (count == 0) return 0; + return ((((size_t)1 << count) - 1) << bitidx); +} + + +/* ----------------------------------------------------------- + Claim a bit sequence atomically +----------------------------------------------------------- */ + +// Try to atomically claim a sequence of `count` bits in a single +// field at `idx` in `bitmap`. Returns `true` on success. +inline bool _mi_bitmap_try_find_claim_field(mi_bitmap_t bitmap, size_t idx, const size_t count, mi_bitmap_index_t* bitmap_idx) +{ + mi_assert_internal(bitmap_idx != NULL); + mi_assert_internal(count <= MI_BITMAP_FIELD_BITS); + mi_assert_internal(count > 0); + mi_bitmap_field_t* field = &bitmap[idx]; + size_t map = mi_atomic_load_relaxed(field); + if (map==MI_BITMAP_FIELD_FULL) return false; // short cut + + // search for 0-bit sequence of length count + const size_t mask = mi_bitmap_mask_(count, 0); + const size_t bitidx_max = MI_BITMAP_FIELD_BITS - count; + +#ifdef MI_HAVE_FAST_BITSCAN + size_t bitidx = mi_ctz(~map); // quickly find the first zero bit if possible +#else + size_t bitidx = 0; // otherwise start at 0 +#endif + size_t m = (mask << bitidx); // invariant: m == mask shifted by bitidx + + // scan linearly for a free range of zero bits + while (bitidx <= bitidx_max) { + const size_t mapm = map & m; + if (mapm == 0) { // are the mask bits free at bitidx? + mi_assert_internal((m >> bitidx) == mask); // no overflow? + const size_t newmap = map | m; + mi_assert_internal((newmap^map) >> bitidx == mask); + if (!mi_atomic_cas_weak_acq_rel(field, &map, newmap)) { // TODO: use strong cas here? + // no success, another thread claimed concurrently.. keep going (with updated `map`) + continue; + } + else { + // success, we claimed the bits! + *bitmap_idx = mi_bitmap_index_create(idx, bitidx); + return true; + } + } + else { + // on to the next bit range +#ifdef MI_HAVE_FAST_BITSCAN + const size_t shift = (count == 1 ? 1 : mi_bsr(mapm) - bitidx + 1); + mi_assert_internal(shift > 0 && shift <= count); +#else + const size_t shift = 1; +#endif + bitidx += shift; + m <<= shift; + } + } + // no bits found + return false; +} + +// Find `count` bits of 0 and set them to 1 atomically; returns `true` on success. +// Starts at idx, and wraps around to search in all `bitmap_fields` fields. +// `count` can be at most MI_BITMAP_FIELD_BITS and will never cross fields. +bool _mi_bitmap_try_find_from_claim(mi_bitmap_t bitmap, const size_t bitmap_fields, const size_t start_field_idx, const size_t count, mi_bitmap_index_t* bitmap_idx) { + size_t idx = start_field_idx; + for (size_t visited = 0; visited < bitmap_fields; visited++, idx++) { + if (idx >= bitmap_fields) idx = 0; // wrap + if (_mi_bitmap_try_find_claim_field(bitmap, idx, count, bitmap_idx)) { + return true; + } + } + return false; +} + +/* +// Find `count` bits of 0 and set them to 1 atomically; returns `true` on success. +// For now, `count` can be at most MI_BITMAP_FIELD_BITS and will never span fields. +bool _mi_bitmap_try_find_claim(mi_bitmap_t bitmap, const size_t bitmap_fields, const size_t count, mi_bitmap_index_t* bitmap_idx) { + return _mi_bitmap_try_find_from_claim(bitmap, bitmap_fields, 0, count, bitmap_idx); +} +*/ + +// Set `count` bits at `bitmap_idx` to 0 atomically +// Returns `true` if all `count` bits were 1 previously. +bool _mi_bitmap_unclaim(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx) { + const size_t idx = mi_bitmap_index_field(bitmap_idx); + const size_t bitidx = mi_bitmap_index_bit_in_field(bitmap_idx); + const size_t mask = mi_bitmap_mask_(count, bitidx); + mi_assert_internal(bitmap_fields > idx); MI_UNUSED(bitmap_fields); + // mi_assert_internal((bitmap[idx] & mask) == mask); + size_t prev = mi_atomic_and_acq_rel(&bitmap[idx], ~mask); + return ((prev & mask) == mask); +} + + +// Set `count` bits at `bitmap_idx` to 1 atomically +// Returns `true` if all `count` bits were 0 previously. `any_zero` is `true` if there was at least one zero bit. +bool _mi_bitmap_claim(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, bool* any_zero) { + const size_t idx = mi_bitmap_index_field(bitmap_idx); + const size_t bitidx = mi_bitmap_index_bit_in_field(bitmap_idx); + const size_t mask = mi_bitmap_mask_(count, bitidx); + mi_assert_internal(bitmap_fields > idx); MI_UNUSED(bitmap_fields); + //mi_assert_internal(any_zero != NULL || (bitmap[idx] & mask) == 0); + size_t prev = mi_atomic_or_acq_rel(&bitmap[idx], mask); + if (any_zero != NULL) *any_zero = ((prev & mask) != mask); + return ((prev & mask) == 0); +} + +// Returns `true` if all `count` bits were 1. `any_ones` is `true` if there was at least one bit set to one. +static bool mi_bitmap_is_claimedx(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, bool* any_ones) { + const size_t idx = mi_bitmap_index_field(bitmap_idx); + const size_t bitidx = mi_bitmap_index_bit_in_field(bitmap_idx); + const size_t mask = mi_bitmap_mask_(count, bitidx); + mi_assert_internal(bitmap_fields > idx); MI_UNUSED(bitmap_fields); + size_t field = mi_atomic_load_relaxed(&bitmap[idx]); + if (any_ones != NULL) *any_ones = ((field & mask) != 0); + return ((field & mask) == mask); +} + +bool _mi_bitmap_is_claimed(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx) { + return mi_bitmap_is_claimedx(bitmap, bitmap_fields, count, bitmap_idx, NULL); +} + +bool _mi_bitmap_is_any_claimed(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx) { + bool any_ones; + mi_bitmap_is_claimedx(bitmap, bitmap_fields, count, bitmap_idx, &any_ones); + return any_ones; +} + + +//-------------------------------------------------------------------------- +// the `_across` functions work on bitmaps where sequences can cross over +// between the fields. This is used in arena allocation +//-------------------------------------------------------------------------- + +// Try to atomically claim a sequence of `count` bits starting from the field +// at `idx` in `bitmap` and crossing into subsequent fields. Returns `true` on success. +static bool mi_bitmap_try_find_claim_field_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t idx, const size_t count, const size_t retries, mi_bitmap_index_t* bitmap_idx) +{ + mi_assert_internal(bitmap_idx != NULL); + + // check initial trailing zeros + mi_bitmap_field_t* field = &bitmap[idx]; + size_t map = mi_atomic_load_relaxed(field); + const size_t initial = mi_clz(map); // count of initial zeros starting at idx + mi_assert_internal(initial <= MI_BITMAP_FIELD_BITS); + if (initial == 0) return false; + if (initial >= count) return _mi_bitmap_try_find_claim_field(bitmap, idx, count, bitmap_idx); // no need to cross fields + if (_mi_divide_up(count - initial, MI_BITMAP_FIELD_BITS) >= (bitmap_fields - idx)) return false; // not enough entries + + // scan ahead + size_t found = initial; + size_t mask = 0; // mask bits for the final field + while(found < count) { + field++; + map = mi_atomic_load_relaxed(field); + const size_t mask_bits = (found + MI_BITMAP_FIELD_BITS <= count ? MI_BITMAP_FIELD_BITS : (count - found)); + mask = mi_bitmap_mask_(mask_bits, 0); + if ((map & mask) != 0) return false; + found += mask_bits; + } + mi_assert_internal(field < &bitmap[bitmap_fields]); + + // found range of zeros up to the final field; mask contains mask in the final field + // now claim it atomically + mi_bitmap_field_t* const final_field = field; + const size_t final_mask = mask; + mi_bitmap_field_t* const initial_field = &bitmap[idx]; + const size_t initial_mask = mi_bitmap_mask_(initial, MI_BITMAP_FIELD_BITS - initial); + + // initial field + size_t newmap; + field = initial_field; + map = mi_atomic_load_relaxed(field); + do { + newmap = map | initial_mask; + if ((map & initial_mask) != 0) { goto rollback; }; + } while (!mi_atomic_cas_strong_acq_rel(field, &map, newmap)); + + // intermediate fields + while (++field < final_field) { + newmap = MI_BITMAP_FIELD_FULL; + map = 0; + if (!mi_atomic_cas_strong_acq_rel(field, &map, newmap)) { goto rollback; } + } + + // final field + mi_assert_internal(field == final_field); + map = mi_atomic_load_relaxed(field); + do { + newmap = map | final_mask; + if ((map & final_mask) != 0) { goto rollback; } + } while (!mi_atomic_cas_strong_acq_rel(field, &map, newmap)); + + // claimed! + *bitmap_idx = mi_bitmap_index_create(idx, MI_BITMAP_FIELD_BITS - initial); + return true; + +rollback: + // roll back intermediate fields + while (--field > initial_field) { + newmap = 0; + map = MI_BITMAP_FIELD_FULL; + mi_assert_internal(mi_atomic_load_relaxed(field) == map); + mi_atomic_store_release(field, newmap); + } + if (field == initial_field) { + map = mi_atomic_load_relaxed(field); + do { + mi_assert_internal((map & initial_mask) == initial_mask); + newmap = map & ~initial_mask; + } while (!mi_atomic_cas_strong_acq_rel(field, &map, newmap)); + } + // retry? (we make a recursive call instead of goto to be able to use const declarations) + if (retries < 4) { + return mi_bitmap_try_find_claim_field_across(bitmap, bitmap_fields, idx, count, retries+1, bitmap_idx); + } + else { + return false; + } +} + + +// Find `count` bits of zeros and set them to 1 atomically; returns `true` on success. +// Starts at idx, and wraps around to search in all `bitmap_fields` fields. +bool _mi_bitmap_try_find_from_claim_across(mi_bitmap_t bitmap, const size_t bitmap_fields, const size_t start_field_idx, const size_t count, mi_bitmap_index_t* bitmap_idx) { + mi_assert_internal(count > 0); + if (count==1) return _mi_bitmap_try_find_from_claim(bitmap, bitmap_fields, start_field_idx, count, bitmap_idx); + size_t idx = start_field_idx; + for (size_t visited = 0; visited < bitmap_fields; visited++, idx++) { + if (idx >= bitmap_fields) idx = 0; // wrap + // try to claim inside the field + if (count <= MI_BITMAP_FIELD_BITS) { + if (_mi_bitmap_try_find_claim_field(bitmap, idx, count, bitmap_idx)) { + return true; + } + } + // try to claim across fields + if (mi_bitmap_try_find_claim_field_across(bitmap, bitmap_fields, idx, count, 0, bitmap_idx)) { + return true; + } + } + return false; +} + +// Helper for masks across fields; returns the mid count, post_mask may be 0 +static size_t mi_bitmap_mask_across(mi_bitmap_index_t bitmap_idx, size_t bitmap_fields, size_t count, size_t* pre_mask, size_t* mid_mask, size_t* post_mask) { + MI_UNUSED_RELEASE(bitmap_fields); + const size_t bitidx = mi_bitmap_index_bit_in_field(bitmap_idx); + if (mi_likely(bitidx + count <= MI_BITMAP_FIELD_BITS)) { + *pre_mask = mi_bitmap_mask_(count, bitidx); + *mid_mask = 0; + *post_mask = 0; + mi_assert_internal(mi_bitmap_index_field(bitmap_idx) < bitmap_fields); + return 0; + } + else { + const size_t pre_bits = MI_BITMAP_FIELD_BITS - bitidx; + mi_assert_internal(pre_bits < count); + *pre_mask = mi_bitmap_mask_(pre_bits, bitidx); + count -= pre_bits; + const size_t mid_count = (count / MI_BITMAP_FIELD_BITS); + *mid_mask = MI_BITMAP_FIELD_FULL; + count %= MI_BITMAP_FIELD_BITS; + *post_mask = (count==0 ? 0 : mi_bitmap_mask_(count, 0)); + mi_assert_internal(mi_bitmap_index_field(bitmap_idx) + mid_count + (count==0 ? 0 : 1) < bitmap_fields); + return mid_count; + } +} + +// Set `count` bits at `bitmap_idx` to 0 atomically +// Returns `true` if all `count` bits were 1 previously. +bool _mi_bitmap_unclaim_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx) { + size_t idx = mi_bitmap_index_field(bitmap_idx); + size_t pre_mask; + size_t mid_mask; + size_t post_mask; + size_t mid_count = mi_bitmap_mask_across(bitmap_idx, bitmap_fields, count, &pre_mask, &mid_mask, &post_mask); + bool all_one = true; + mi_bitmap_field_t* field = &bitmap[idx]; + size_t prev = mi_atomic_and_acq_rel(field++, ~pre_mask); + if ((prev & pre_mask) != pre_mask) all_one = false; + while(mid_count-- > 0) { + prev = mi_atomic_and_acq_rel(field++, ~mid_mask); + if ((prev & mid_mask) != mid_mask) all_one = false; + } + if (post_mask!=0) { + prev = mi_atomic_and_acq_rel(field, ~post_mask); + if ((prev & post_mask) != post_mask) all_one = false; + } + return all_one; +} + +// Set `count` bits at `bitmap_idx` to 1 atomically +// Returns `true` if all `count` bits were 0 previously. `any_zero` is `true` if there was at least one zero bit. +bool _mi_bitmap_claim_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, bool* pany_zero) { + size_t idx = mi_bitmap_index_field(bitmap_idx); + size_t pre_mask; + size_t mid_mask; + size_t post_mask; + size_t mid_count = mi_bitmap_mask_across(bitmap_idx, bitmap_fields, count, &pre_mask, &mid_mask, &post_mask); + bool all_zero = true; + bool any_zero = false; + _Atomic(size_t)*field = &bitmap[idx]; + size_t prev = mi_atomic_or_acq_rel(field++, pre_mask); + if ((prev & pre_mask) != 0) all_zero = false; + if ((prev & pre_mask) != pre_mask) any_zero = true; + while (mid_count-- > 0) { + prev = mi_atomic_or_acq_rel(field++, mid_mask); + if ((prev & mid_mask) != 0) all_zero = false; + if ((prev & mid_mask) != mid_mask) any_zero = true; + } + if (post_mask!=0) { + prev = mi_atomic_or_acq_rel(field, post_mask); + if ((prev & post_mask) != 0) all_zero = false; + if ((prev & post_mask) != post_mask) any_zero = true; + } + if (pany_zero != NULL) *pany_zero = any_zero; + return all_zero; +} + + +// Returns `true` if all `count` bits were 1. +// `any_ones` is `true` if there was at least one bit set to one. +static bool mi_bitmap_is_claimedx_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, bool* pany_ones) { + size_t idx = mi_bitmap_index_field(bitmap_idx); + size_t pre_mask; + size_t mid_mask; + size_t post_mask; + size_t mid_count = mi_bitmap_mask_across(bitmap_idx, bitmap_fields, count, &pre_mask, &mid_mask, &post_mask); + bool all_ones = true; + bool any_ones = false; + mi_bitmap_field_t* field = &bitmap[idx]; + size_t prev = mi_atomic_load_relaxed(field++); + if ((prev & pre_mask) != pre_mask) all_ones = false; + if ((prev & pre_mask) != 0) any_ones = true; + while (mid_count-- > 0) { + prev = mi_atomic_load_relaxed(field++); + if ((prev & mid_mask) != mid_mask) all_ones = false; + if ((prev & mid_mask) != 0) any_ones = true; + } + if (post_mask!=0) { + prev = mi_atomic_load_relaxed(field); + if ((prev & post_mask) != post_mask) all_ones = false; + if ((prev & post_mask) != 0) any_ones = true; + } + if (pany_ones != NULL) *pany_ones = any_ones; + return all_ones; +} + +bool _mi_bitmap_is_claimed_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx) { + return mi_bitmap_is_claimedx_across(bitmap, bitmap_fields, count, bitmap_idx, NULL); +} + +bool _mi_bitmap_is_any_claimed_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx) { + bool any_ones; + mi_bitmap_is_claimedx_across(bitmap, bitmap_fields, count, bitmap_idx, &any_ones); + return any_ones; +} diff --git a/compat/mimalloc/bitmap.h b/compat/mimalloc/bitmap.h new file mode 100644 index 00000000000000..e3375033a9326e --- /dev/null +++ b/compat/mimalloc/bitmap.h @@ -0,0 +1,107 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2019-2020 Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ + +/* ---------------------------------------------------------------------------- +Concurrent bitmap that can set/reset sequences of bits atomically, +represeted as an array of fields where each field is a machine word (`size_t`) + +There are two api's; the standard one cannot have sequences that cross +between the bitmap fields (and a sequence must be <= MI_BITMAP_FIELD_BITS). +(this is used in region allocation) + +The `_across` postfixed functions do allow sequences that can cross over +between the fields. (This is used in arena allocation) +---------------------------------------------------------------------------- */ +#pragma once +#ifndef MI_BITMAP_H +#define MI_BITMAP_H + +/* ----------------------------------------------------------- + Bitmap definition +----------------------------------------------------------- */ + +#define MI_BITMAP_FIELD_BITS (8*MI_SIZE_SIZE) +#define MI_BITMAP_FIELD_FULL (~((size_t)0)) // all bits set + +// An atomic bitmap of `size_t` fields +typedef _Atomic(size_t) mi_bitmap_field_t; +typedef mi_bitmap_field_t* mi_bitmap_t; + +// A bitmap index is the index of the bit in a bitmap. +typedef size_t mi_bitmap_index_t; + +// Create a bit index. +static inline mi_bitmap_index_t mi_bitmap_index_create(size_t idx, size_t bitidx) { + mi_assert_internal(bitidx < MI_BITMAP_FIELD_BITS); + return (idx*MI_BITMAP_FIELD_BITS) + bitidx; +} + +// Create a bit index. +static inline mi_bitmap_index_t mi_bitmap_index_create_from_bit(size_t full_bitidx) { + return mi_bitmap_index_create(full_bitidx / MI_BITMAP_FIELD_BITS, full_bitidx % MI_BITMAP_FIELD_BITS); +} + +// Get the field index from a bit index. +static inline size_t mi_bitmap_index_field(mi_bitmap_index_t bitmap_idx) { + return (bitmap_idx / MI_BITMAP_FIELD_BITS); +} + +// Get the bit index in a bitmap field +static inline size_t mi_bitmap_index_bit_in_field(mi_bitmap_index_t bitmap_idx) { + return (bitmap_idx % MI_BITMAP_FIELD_BITS); +} + +// Get the full bit index +static inline size_t mi_bitmap_index_bit(mi_bitmap_index_t bitmap_idx) { + return bitmap_idx; +} + +/* ----------------------------------------------------------- + Claim a bit sequence atomically +----------------------------------------------------------- */ + +// Try to atomically claim a sequence of `count` bits in a single +// field at `idx` in `bitmap`. Returns `true` on success. +bool _mi_bitmap_try_find_claim_field(mi_bitmap_t bitmap, size_t idx, const size_t count, mi_bitmap_index_t* bitmap_idx); + +// Starts at idx, and wraps around to search in all `bitmap_fields` fields. +// For now, `count` can be at most MI_BITMAP_FIELD_BITS and will never cross fields. +bool _mi_bitmap_try_find_from_claim(mi_bitmap_t bitmap, const size_t bitmap_fields, const size_t start_field_idx, const size_t count, mi_bitmap_index_t* bitmap_idx); + +// Set `count` bits at `bitmap_idx` to 0 atomically +// Returns `true` if all `count` bits were 1 previously. +bool _mi_bitmap_unclaim(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx); + +// Set `count` bits at `bitmap_idx` to 1 atomically +// Returns `true` if all `count` bits were 0 previously. `any_zero` is `true` if there was at least one zero bit. +bool _mi_bitmap_claim(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, bool* any_zero); + +bool _mi_bitmap_is_claimed(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx); +bool _mi_bitmap_is_any_claimed(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx); + + +//-------------------------------------------------------------------------- +// the `_across` functions work on bitmaps where sequences can cross over +// between the fields. This is used in arena allocation +//-------------------------------------------------------------------------- + +// Find `count` bits of zeros and set them to 1 atomically; returns `true` on success. +// Starts at idx, and wraps around to search in all `bitmap_fields` fields. +bool _mi_bitmap_try_find_from_claim_across(mi_bitmap_t bitmap, const size_t bitmap_fields, const size_t start_field_idx, const size_t count, mi_bitmap_index_t* bitmap_idx); + +// Set `count` bits at `bitmap_idx` to 0 atomically +// Returns `true` if all `count` bits were 1 previously. +bool _mi_bitmap_unclaim_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx); + +// Set `count` bits at `bitmap_idx` to 1 atomically +// Returns `true` if all `count` bits were 0 previously. `any_zero` is `true` if there was at least one zero bit. +bool _mi_bitmap_claim_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, bool* pany_zero); + +bool _mi_bitmap_is_claimed_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx); +bool _mi_bitmap_is_any_claimed_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx); + +#endif diff --git a/compat/mimalloc/heap.c b/compat/mimalloc/heap.c new file mode 100644 index 00000000000000..42c6cfd63699db --- /dev/null +++ b/compat/mimalloc/heap.c @@ -0,0 +1,580 @@ +/*---------------------------------------------------------------------------- +Copyright (c) 2018-2021, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ + +#include "mimalloc.h" +#include "mimalloc-internal.h" +#include "mimalloc-atomic.h" + +#include // memset, memcpy + +#if defined(_MSC_VER) && (_MSC_VER < 1920) +#pragma warning(disable:4204) // non-constant aggregate initializer +#endif + +/* ----------------------------------------------------------- + Helpers +----------------------------------------------------------- */ + +// return `true` if ok, `false` to break +typedef bool (heap_page_visitor_fun)(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_t* page, void* arg1, void* arg2); + +// Visit all pages in a heap; returns `false` if break was called. +static bool mi_heap_visit_pages(mi_heap_t* heap, heap_page_visitor_fun* fn, void* arg1, void* arg2) +{ + if (heap==NULL || heap->page_count==0) return 0; + + // visit all pages + #if MI_DEBUG>1 + size_t total = heap->page_count; + #endif + size_t count = 0; + for (size_t i = 0; i <= MI_BIN_FULL; i++) { + mi_page_queue_t* pq = &heap->pages[i]; + mi_page_t* page = pq->first; + while(page != NULL) { + mi_page_t* next = page->next; // save next in case the page gets removed from the queue + mi_assert_internal(mi_page_heap(page) == heap); + count++; + if (!fn(heap, pq, page, arg1, arg2)) return false; + page = next; // and continue + } + } + mi_assert_internal(count == total); + return true; +} + + +#if MI_DEBUG>=2 +static bool mi_heap_page_is_valid(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_t* page, void* arg1, void* arg2) { + MI_UNUSED(arg1); + MI_UNUSED(arg2); + MI_UNUSED(pq); + mi_assert_internal(mi_page_heap(page) == heap); + mi_segment_t* segment = _mi_page_segment(page); + mi_assert_internal(segment->thread_id == heap->thread_id); + mi_assert_expensive(_mi_page_is_valid(page)); + return true; +} +#endif +#if MI_DEBUG>=3 +static bool mi_heap_is_valid(mi_heap_t* heap) { + mi_assert_internal(heap!=NULL); + mi_heap_visit_pages(heap, &mi_heap_page_is_valid, NULL, NULL); + return true; +} +#endif + + + + +/* ----------------------------------------------------------- + "Collect" pages by migrating `local_free` and `thread_free` + lists and freeing empty pages. This is done when a thread + stops (and in that case abandons pages if there are still + blocks alive) +----------------------------------------------------------- */ + +typedef enum mi_collect_e { + MI_NORMAL, + MI_FORCE, + MI_ABANDON +} mi_collect_t; + + +static bool mi_heap_page_collect(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_t* page, void* arg_collect, void* arg2 ) { + MI_UNUSED(arg2); + MI_UNUSED(heap); + mi_assert_internal(mi_heap_page_is_valid(heap, pq, page, NULL, NULL)); + mi_collect_t collect = *((mi_collect_t*)arg_collect); + _mi_page_free_collect(page, collect >= MI_FORCE); + if (mi_page_all_free(page)) { + // no more used blocks, free the page. + // note: this will free retired pages as well. + _mi_page_free(page, pq, collect >= MI_FORCE); + } + else if (collect == MI_ABANDON) { + // still used blocks but the thread is done; abandon the page + _mi_page_abandon(page, pq); + } + return true; // don't break +} + +static bool mi_heap_page_never_delayed_free(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_t* page, void* arg1, void* arg2) { + MI_UNUSED(arg1); + MI_UNUSED(arg2); + MI_UNUSED(heap); + MI_UNUSED(pq); + _mi_page_use_delayed_free(page, MI_NEVER_DELAYED_FREE, false); + return true; // don't break +} + +static void mi_heap_collect_ex(mi_heap_t* heap, mi_collect_t collect) +{ + if (heap==NULL || !mi_heap_is_initialized(heap)) return; + + const bool force = collect >= MI_FORCE; + _mi_deferred_free(heap, force); + + // note: never reclaim on collect but leave it to threads that need storage to reclaim + const bool force_main = + #ifdef NDEBUG + collect == MI_FORCE + #else + collect >= MI_FORCE + #endif + && _mi_is_main_thread() && mi_heap_is_backing(heap) && !heap->no_reclaim; + + if (force_main) { + // the main thread is abandoned (end-of-program), try to reclaim all abandoned segments. + // if all memory is freed by now, all segments should be freed. + _mi_abandoned_reclaim_all(heap, &heap->tld->segments); + } + + // if abandoning, mark all pages to no longer add to delayed_free + if (collect == MI_ABANDON) { + mi_heap_visit_pages(heap, &mi_heap_page_never_delayed_free, NULL, NULL); + } + + // free thread delayed blocks. + // (if abandoning, after this there are no more thread-delayed references into the pages.) + _mi_heap_delayed_free(heap); + + // collect retired pages + _mi_heap_collect_retired(heap, force); + + // collect all pages owned by this thread + mi_heap_visit_pages(heap, &mi_heap_page_collect, &collect, NULL); + mi_assert_internal( collect != MI_ABANDON || mi_atomic_load_ptr_acquire(mi_block_t,&heap->thread_delayed_free) == NULL ); + + // collect abandoned segments (in particular, decommit expired parts of segments in the abandoned segment list) + // note: forced decommit can be quite expensive if many threads are created/destroyed so we do not force on abandonment + _mi_abandoned_collect(heap, collect == MI_FORCE /* force? */, &heap->tld->segments); + + // collect segment local caches + if (force) { + _mi_segment_thread_collect(&heap->tld->segments); + } + + // decommit in global segment caches + // note: forced decommit can be quite expensive if many threads are created/destroyed so we do not force on abandonment + _mi_segment_cache_collect( collect == MI_FORCE, &heap->tld->os); + + // collect regions on program-exit (or shared library unload) + if (force && _mi_is_main_thread() && mi_heap_is_backing(heap)) { + //_mi_mem_collect(&heap->tld->os); + } +} + +void _mi_heap_collect_abandon(mi_heap_t* heap) { + mi_heap_collect_ex(heap, MI_ABANDON); +} + +void mi_heap_collect(mi_heap_t* heap, bool force) mi_attr_noexcept { + mi_heap_collect_ex(heap, (force ? MI_FORCE : MI_NORMAL)); +} + +void mi_collect(bool force) mi_attr_noexcept { + mi_heap_collect(mi_get_default_heap(), force); +} + + +/* ----------------------------------------------------------- + Heap new +----------------------------------------------------------- */ + +mi_heap_t* mi_heap_get_default(void) { + mi_thread_init(); + return mi_get_default_heap(); +} + +mi_heap_t* mi_heap_get_backing(void) { + mi_heap_t* heap = mi_heap_get_default(); + mi_assert_internal(heap!=NULL); + mi_heap_t* bheap = heap->tld->heap_backing; + mi_assert_internal(bheap!=NULL); + mi_assert_internal(bheap->thread_id == _mi_thread_id()); + return bheap; +} + +mi_heap_t* mi_heap_new(void) { + mi_heap_t* bheap = mi_heap_get_backing(); + mi_heap_t* heap = mi_heap_malloc_tp(bheap, mi_heap_t); // todo: OS allocate in secure mode? + if (heap==NULL) return NULL; + _mi_memcpy_aligned(heap, &_mi_heap_empty, sizeof(mi_heap_t)); + heap->tld = bheap->tld; + heap->thread_id = _mi_thread_id(); + _mi_random_split(&bheap->random, &heap->random); + heap->cookie = _mi_heap_random_next(heap) | 1; + heap->keys[0] = _mi_heap_random_next(heap); + heap->keys[1] = _mi_heap_random_next(heap); + heap->no_reclaim = true; // don't reclaim abandoned pages or otherwise destroy is unsafe + // push on the thread local heaps list + heap->next = heap->tld->heaps; + heap->tld->heaps = heap; + return heap; +} + +uintptr_t _mi_heap_random_next(mi_heap_t* heap) { + return _mi_random_next(&heap->random); +} + +// zero out the page queues +static void mi_heap_reset_pages(mi_heap_t* heap) { + mi_assert_internal(heap != NULL); + mi_assert_internal(mi_heap_is_initialized(heap)); + // TODO: copy full empty heap instead? + memset(&heap->pages_free_direct, 0, sizeof(heap->pages_free_direct)); +#ifdef MI_MEDIUM_DIRECT + memset(&heap->pages_free_medium, 0, sizeof(heap->pages_free_medium)); +#endif + _mi_memcpy_aligned(&heap->pages, &_mi_heap_empty.pages, sizeof(heap->pages)); + heap->thread_delayed_free = NULL; + heap->page_count = 0; +} + +// called from `mi_heap_destroy` and `mi_heap_delete` to free the internal heap resources. +static void mi_heap_free(mi_heap_t* heap) { + mi_assert(heap != NULL); + mi_assert_internal(mi_heap_is_initialized(heap)); + if (heap==NULL || !mi_heap_is_initialized(heap)) return; + if (mi_heap_is_backing(heap)) return; // dont free the backing heap + + // reset default + if (mi_heap_is_default(heap)) { + _mi_heap_set_default_direct(heap->tld->heap_backing); + } + + // remove ourselves from the thread local heaps list + // linear search but we expect the number of heaps to be relatively small + mi_heap_t* prev = NULL; + mi_heap_t* curr = heap->tld->heaps; + while (curr != heap && curr != NULL) { + prev = curr; + curr = curr->next; + } + mi_assert_internal(curr == heap); + if (curr == heap) { + if (prev != NULL) { prev->next = heap->next; } + else { heap->tld->heaps = heap->next; } + } + mi_assert_internal(heap->tld->heaps != NULL); + + // and free the used memory + mi_free(heap); +} + + +/* ----------------------------------------------------------- + Heap destroy +----------------------------------------------------------- */ + +static bool _mi_heap_page_destroy(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_t* page, void* arg1, void* arg2) { + MI_UNUSED(arg1); + MI_UNUSED(arg2); + MI_UNUSED(heap); + MI_UNUSED(pq); + + // ensure no more thread_delayed_free will be added + _mi_page_use_delayed_free(page, MI_NEVER_DELAYED_FREE, false); + + // stats + const size_t bsize = mi_page_block_size(page); + if (bsize > MI_MEDIUM_OBJ_SIZE_MAX) { + if (bsize <= MI_LARGE_OBJ_SIZE_MAX) { + mi_heap_stat_decrease(heap, large, bsize); + } + else { + mi_heap_stat_decrease(heap, huge, bsize); + } + } +#if (MI_STAT) + _mi_page_free_collect(page, false); // update used count + const size_t inuse = page->used; + if (bsize <= MI_LARGE_OBJ_SIZE_MAX) { + mi_heap_stat_decrease(heap, normal, bsize * inuse); +#if (MI_STAT>1) + mi_heap_stat_decrease(heap, normal_bins[_mi_bin(bsize)], inuse); +#endif + } + mi_heap_stat_decrease(heap, malloc, bsize * inuse); // todo: off for aligned blocks... +#endif + + /// pretend it is all free now + mi_assert_internal(mi_page_thread_free(page) == NULL); + page->used = 0; + + // and free the page + // mi_page_free(page,false); + page->next = NULL; + page->prev = NULL; + _mi_segment_page_free(page,false /* no force? */, &heap->tld->segments); + + return true; // keep going +} + +void _mi_heap_destroy_pages(mi_heap_t* heap) { + mi_heap_visit_pages(heap, &_mi_heap_page_destroy, NULL, NULL); + mi_heap_reset_pages(heap); +} + +void mi_heap_destroy(mi_heap_t* heap) { + mi_assert(heap != NULL); + mi_assert(mi_heap_is_initialized(heap)); + mi_assert(heap->no_reclaim); + mi_assert_expensive(mi_heap_is_valid(heap)); + if (heap==NULL || !mi_heap_is_initialized(heap)) return; + if (!heap->no_reclaim) { + // don't free in case it may contain reclaimed pages + mi_heap_delete(heap); + } + else { + // free all pages + _mi_heap_destroy_pages(heap); + mi_heap_free(heap); + } +} + + + +/* ----------------------------------------------------------- + Safe Heap delete +----------------------------------------------------------- */ + +// Transfer the pages from one heap to the other +static void mi_heap_absorb(mi_heap_t* heap, mi_heap_t* from) { + mi_assert_internal(heap!=NULL); + if (from==NULL || from->page_count == 0) return; + + // reduce the size of the delayed frees + _mi_heap_delayed_free(from); + + // transfer all pages by appending the queues; this will set a new heap field + // so threads may do delayed frees in either heap for a while. + // note: appending waits for each page to not be in the `MI_DELAYED_FREEING` state + // so after this only the new heap will get delayed frees + for (size_t i = 0; i <= MI_BIN_FULL; i++) { + mi_page_queue_t* pq = &heap->pages[i]; + mi_page_queue_t* append = &from->pages[i]; + size_t pcount = _mi_page_queue_append(heap, pq, append); + heap->page_count += pcount; + from->page_count -= pcount; + } + mi_assert_internal(from->page_count == 0); + + // and do outstanding delayed frees in the `from` heap + // note: be careful here as the `heap` field in all those pages no longer point to `from`, + // turns out to be ok as `_mi_heap_delayed_free` only visits the list and calls a + // the regular `_mi_free_delayed_block` which is safe. + _mi_heap_delayed_free(from); + #if !defined(_MSC_VER) || (_MSC_VER > 1900) // somehow the following line gives an error in VS2015, issue #353 + mi_assert_internal(mi_atomic_load_ptr_relaxed(mi_block_t,&from->thread_delayed_free) == NULL); + #endif + + // and reset the `from` heap + mi_heap_reset_pages(from); +} + +// Safe delete a heap without freeing any still allocated blocks in that heap. +void mi_heap_delete(mi_heap_t* heap) +{ + mi_assert(heap != NULL); + mi_assert(mi_heap_is_initialized(heap)); + mi_assert_expensive(mi_heap_is_valid(heap)); + if (heap==NULL || !mi_heap_is_initialized(heap)) return; + + if (!mi_heap_is_backing(heap)) { + // tranfer still used pages to the backing heap + mi_heap_absorb(heap->tld->heap_backing, heap); + } + else { + // the backing heap abandons its pages + _mi_heap_collect_abandon(heap); + } + mi_assert_internal(heap->page_count==0); + mi_heap_free(heap); +} + +mi_heap_t* mi_heap_set_default(mi_heap_t* heap) { + mi_assert(heap != NULL); + mi_assert(mi_heap_is_initialized(heap)); + if (heap==NULL || !mi_heap_is_initialized(heap)) return NULL; + mi_assert_expensive(mi_heap_is_valid(heap)); + mi_heap_t* old = mi_get_default_heap(); + _mi_heap_set_default_direct(heap); + return old; +} + + + + +/* ----------------------------------------------------------- + Analysis +----------------------------------------------------------- */ + +// static since it is not thread safe to access heaps from other threads. +static mi_heap_t* mi_heap_of_block(const void* p) { + if (p == NULL) return NULL; + mi_segment_t* segment = _mi_ptr_segment(p); + bool valid = (_mi_ptr_cookie(segment) == segment->cookie); + mi_assert_internal(valid); + if (mi_unlikely(!valid)) return NULL; + return mi_page_heap(_mi_segment_page_of(segment,p)); +} + +bool mi_heap_contains_block(mi_heap_t* heap, const void* p) { + mi_assert(heap != NULL); + if (heap==NULL || !mi_heap_is_initialized(heap)) return false; + return (heap == mi_heap_of_block(p)); +} + + +static bool mi_heap_page_check_owned(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_t* page, void* p, void* vfound) { + MI_UNUSED(heap); + MI_UNUSED(pq); + bool* found = (bool*)vfound; + mi_segment_t* segment = _mi_page_segment(page); + void* start = _mi_page_start(segment, page, NULL); + void* end = (uint8_t*)start + (page->capacity * mi_page_block_size(page)); + *found = (p >= start && p < end); + return (!*found); // continue if not found +} + +bool mi_heap_check_owned(mi_heap_t* heap, const void* p) { + mi_assert(heap != NULL); + if (heap==NULL || !mi_heap_is_initialized(heap)) return false; + if (((uintptr_t)p & (MI_INTPTR_SIZE - 1)) != 0) return false; // only aligned pointers + bool found = false; + mi_heap_visit_pages(heap, &mi_heap_page_check_owned, (void*)p, &found); + return found; +} + +bool mi_check_owned(const void* p) { + return mi_heap_check_owned(mi_get_default_heap(), p); +} + +/* ----------------------------------------------------------- + Visit all heap blocks and areas + Todo: enable visiting abandoned pages, and + enable visiting all blocks of all heaps across threads +----------------------------------------------------------- */ + +// Separate struct to keep `mi_page_t` out of the public interface +typedef struct mi_heap_area_ex_s { + mi_heap_area_t area; + mi_page_t* page; +} mi_heap_area_ex_t; + +static bool mi_heap_area_visit_blocks(const mi_heap_area_ex_t* xarea, mi_block_visit_fun* visitor, void* arg) { + mi_assert(xarea != NULL); + if (xarea==NULL) return true; + const mi_heap_area_t* area = &xarea->area; + mi_page_t* page = xarea->page; + mi_assert(page != NULL); + if (page == NULL) return true; + + _mi_page_free_collect(page,true); + mi_assert_internal(page->local_free == NULL); + if (page->used == 0) return true; + + const size_t bsize = mi_page_block_size(page); + const size_t ubsize = mi_page_usable_block_size(page); // without padding + size_t psize; + uint8_t* pstart = _mi_page_start(_mi_page_segment(page), page, &psize); + + if (page->capacity == 1) { + // optimize page with one block + mi_assert_internal(page->used == 1 && page->free == NULL); + return visitor(mi_page_heap(page), area, pstart, ubsize, arg); + } + + // create a bitmap of free blocks. + #define MI_MAX_BLOCKS (MI_SMALL_PAGE_SIZE / sizeof(void*)) + uintptr_t free_map[MI_MAX_BLOCKS / sizeof(uintptr_t)]; + memset(free_map, 0, sizeof(free_map)); + + size_t free_count = 0; + for (mi_block_t* block = page->free; block != NULL; block = mi_block_next(page,block)) { + free_count++; + mi_assert_internal((uint8_t*)block >= pstart && (uint8_t*)block < (pstart + psize)); + size_t offset = (uint8_t*)block - pstart; + mi_assert_internal(offset % bsize == 0); + size_t blockidx = offset / bsize; // Todo: avoid division? + mi_assert_internal( blockidx < MI_MAX_BLOCKS); + size_t bitidx = (blockidx / sizeof(uintptr_t)); + size_t bit = blockidx - (bitidx * sizeof(uintptr_t)); + free_map[bitidx] |= ((uintptr_t)1 << bit); + } + mi_assert_internal(page->capacity == (free_count + page->used)); + + // walk through all blocks skipping the free ones + size_t used_count = 0; + for (size_t i = 0; i < page->capacity; i++) { + size_t bitidx = (i / sizeof(uintptr_t)); + size_t bit = i - (bitidx * sizeof(uintptr_t)); + uintptr_t m = free_map[bitidx]; + if (bit == 0 && m == UINTPTR_MAX) { + i += (sizeof(uintptr_t) - 1); // skip a run of free blocks + } + else if ((m & ((uintptr_t)1 << bit)) == 0) { + used_count++; + uint8_t* block = pstart + (i * bsize); + if (!visitor(mi_page_heap(page), area, block, ubsize, arg)) return false; + } + } + mi_assert_internal(page->used == used_count); + return true; +} + +typedef bool (mi_heap_area_visit_fun)(const mi_heap_t* heap, const mi_heap_area_ex_t* area, void* arg); + + +static bool mi_heap_visit_areas_page(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_t* page, void* vfun, void* arg) { + MI_UNUSED(heap); + MI_UNUSED(pq); + mi_heap_area_visit_fun* fun = (mi_heap_area_visit_fun*)vfun; + mi_heap_area_ex_t xarea; + const size_t bsize = mi_page_block_size(page); + const size_t ubsize = mi_page_usable_block_size(page); + xarea.page = page; + xarea.area.reserved = page->reserved * bsize; + xarea.area.committed = page->capacity * bsize; + xarea.area.blocks = _mi_page_start(_mi_page_segment(page), page, NULL); + xarea.area.used = page->used * bsize; + xarea.area.block_size = ubsize; + xarea.area.full_block_size = bsize; + return fun(heap, &xarea, arg); +} + +// Visit all heap pages as areas +static bool mi_heap_visit_areas(const mi_heap_t* heap, mi_heap_area_visit_fun* visitor, void* arg) { + if (visitor == NULL) return false; + return mi_heap_visit_pages((mi_heap_t*)heap, &mi_heap_visit_areas_page, (void*)(visitor), arg); // note: function pointer to void* :-{ +} + +// Just to pass arguments +typedef struct mi_visit_blocks_args_s { + bool visit_blocks; + mi_block_visit_fun* visitor; + void* arg; +} mi_visit_blocks_args_t; + +static bool mi_heap_area_visitor(const mi_heap_t* heap, const mi_heap_area_ex_t* xarea, void* arg) { + mi_visit_blocks_args_t* args = (mi_visit_blocks_args_t*)arg; + if (!args->visitor(heap, &xarea->area, NULL, xarea->area.block_size, args->arg)) return false; + if (args->visit_blocks) { + return mi_heap_area_visit_blocks(xarea, args->visitor, args->arg); + } + else { + return true; + } +} + +// Visit all blocks in a heap +bool mi_heap_visit_blocks(const mi_heap_t* heap, bool visit_blocks, mi_block_visit_fun* visitor, void* arg) { + mi_visit_blocks_args_t args = { visit_blocks, visitor, arg }; + return mi_heap_visit_areas(heap, &mi_heap_area_visitor, &args); +} diff --git a/compat/mimalloc/init.c b/compat/mimalloc/init.c new file mode 100644 index 00000000000000..6b2a99e47a6180 --- /dev/null +++ b/compat/mimalloc/init.c @@ -0,0 +1,693 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2022, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ +#include "mimalloc.h" +#include "mimalloc-internal.h" + +#include // memcpy, memset +#include // atexit + +// Empty page used to initialize the small free pages array +const mi_page_t _mi_page_empty = { + 0, false, false, false, false, + 0, // capacity + 0, // reserved capacity + { 0 }, // flags + false, // is_zero + 0, // retire_expire + NULL, // free + #if MI_ENCODE_FREELIST + { 0, 0 }, + #endif + 0, // used + 0, // xblock_size + NULL, // local_free + MI_ATOMIC_VAR_INIT(0), // xthread_free + MI_ATOMIC_VAR_INIT(0), // xheap + NULL, NULL + #if MI_INTPTR_SIZE==8 + , { 0 } // padding + #endif +}; + +#define MI_PAGE_EMPTY() ((mi_page_t*)&_mi_page_empty) + +#if (MI_PADDING>0) && (MI_INTPTR_SIZE >= 8) +#define MI_SMALL_PAGES_EMPTY { MI_INIT128(MI_PAGE_EMPTY), MI_PAGE_EMPTY(), MI_PAGE_EMPTY() } +#elif (MI_PADDING>0) +#define MI_SMALL_PAGES_EMPTY { MI_INIT128(MI_PAGE_EMPTY), MI_PAGE_EMPTY(), MI_PAGE_EMPTY(), MI_PAGE_EMPTY() } +#else +#define MI_SMALL_PAGES_EMPTY { MI_INIT128(MI_PAGE_EMPTY), MI_PAGE_EMPTY() } +#endif + + +// Empty page queues for every bin +#define QNULL(sz) { NULL, NULL, (sz)*sizeof(uintptr_t) } +#define MI_PAGE_QUEUES_EMPTY \ + { QNULL(1), \ + QNULL( 1), QNULL( 2), QNULL( 3), QNULL( 4), QNULL( 5), QNULL( 6), QNULL( 7), QNULL( 8), /* 8 */ \ + QNULL( 10), QNULL( 12), QNULL( 14), QNULL( 16), QNULL( 20), QNULL( 24), QNULL( 28), QNULL( 32), /* 16 */ \ + QNULL( 40), QNULL( 48), QNULL( 56), QNULL( 64), QNULL( 80), QNULL( 96), QNULL( 112), QNULL( 128), /* 24 */ \ + QNULL( 160), QNULL( 192), QNULL( 224), QNULL( 256), QNULL( 320), QNULL( 384), QNULL( 448), QNULL( 512), /* 32 */ \ + QNULL( 640), QNULL( 768), QNULL( 896), QNULL( 1024), QNULL( 1280), QNULL( 1536), QNULL( 1792), QNULL( 2048), /* 40 */ \ + QNULL( 2560), QNULL( 3072), QNULL( 3584), QNULL( 4096), QNULL( 5120), QNULL( 6144), QNULL( 7168), QNULL( 8192), /* 48 */ \ + QNULL( 10240), QNULL( 12288), QNULL( 14336), QNULL( 16384), QNULL( 20480), QNULL( 24576), QNULL( 28672), QNULL( 32768), /* 56 */ \ + QNULL( 40960), QNULL( 49152), QNULL( 57344), QNULL( 65536), QNULL( 81920), QNULL( 98304), QNULL(114688), QNULL(131072), /* 64 */ \ + QNULL(163840), QNULL(196608), QNULL(229376), QNULL(262144), QNULL(327680), QNULL(393216), QNULL(458752), QNULL(524288), /* 72 */ \ + QNULL(MI_MEDIUM_OBJ_WSIZE_MAX + 1 /* 655360, Huge queue */), \ + QNULL(MI_MEDIUM_OBJ_WSIZE_MAX + 2) /* Full queue */ } + +#define MI_STAT_COUNT_NULL() {0,0,0,0} + +// Empty statistics +#if MI_STAT>1 +#define MI_STAT_COUNT_END_NULL() , { MI_STAT_COUNT_NULL(), MI_INIT32(MI_STAT_COUNT_NULL) } +#else +#define MI_STAT_COUNT_END_NULL() +#endif + +#define MI_STATS_NULL \ + MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), \ + MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), \ + MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), \ + MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), \ + MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), \ + MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), \ + MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), \ + { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, \ + { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 } \ + MI_STAT_COUNT_END_NULL() + + +// Empty slice span queues for every bin +#define SQNULL(sz) { NULL, NULL, sz } +#define MI_SEGMENT_SPAN_QUEUES_EMPTY \ + { SQNULL(1), \ + SQNULL( 1), SQNULL( 2), SQNULL( 3), SQNULL( 4), SQNULL( 5), SQNULL( 6), SQNULL( 7), SQNULL( 10), /* 8 */ \ + SQNULL( 12), SQNULL( 14), SQNULL( 16), SQNULL( 20), SQNULL( 24), SQNULL( 28), SQNULL( 32), SQNULL( 40), /* 16 */ \ + SQNULL( 48), SQNULL( 56), SQNULL( 64), SQNULL( 80), SQNULL( 96), SQNULL( 112), SQNULL( 128), SQNULL( 160), /* 24 */ \ + SQNULL( 192), SQNULL( 224), SQNULL( 256), SQNULL( 320), SQNULL( 384), SQNULL( 448), SQNULL( 512), SQNULL( 640), /* 32 */ \ + SQNULL( 768), SQNULL( 896), SQNULL( 1024) /* 35 */ } + + +// -------------------------------------------------------- +// Statically allocate an empty heap as the initial +// thread local value for the default heap, +// and statically allocate the backing heap for the main +// thread so it can function without doing any allocation +// itself (as accessing a thread local for the first time +// may lead to allocation itself on some platforms) +// -------------------------------------------------------- + +mi_decl_cache_align const mi_heap_t _mi_heap_empty = { + NULL, + MI_SMALL_PAGES_EMPTY, + MI_PAGE_QUEUES_EMPTY, + MI_ATOMIC_VAR_INIT(NULL), + 0, // tid + 0, // cookie + { 0, 0 }, // keys + { {0}, {0}, 0 }, + 0, // page count + MI_BIN_FULL, 0, // page retired min/max + NULL, // next + false +}; + +#define tld_empty_stats ((mi_stats_t*)((uint8_t*)&tld_empty + offsetof(mi_tld_t,stats))) +#define tld_empty_os ((mi_os_tld_t*)((uint8_t*)&tld_empty + offsetof(mi_tld_t,os))) + +mi_decl_cache_align static const mi_tld_t tld_empty = { + 0, + false, + NULL, NULL, + { MI_SEGMENT_SPAN_QUEUES_EMPTY, 0, 0, 0, 0, tld_empty_stats, tld_empty_os }, // segments + { 0, tld_empty_stats }, // os + { MI_STATS_NULL } // stats +}; + +// the thread-local default heap for allocation +mi_decl_thread mi_heap_t* _mi_heap_default = (mi_heap_t*)&_mi_heap_empty; + +extern mi_heap_t _mi_heap_main; + +static mi_tld_t tld_main = { + 0, false, + &_mi_heap_main, & _mi_heap_main, + { MI_SEGMENT_SPAN_QUEUES_EMPTY, 0, 0, 0, 0, &tld_main.stats, &tld_main.os }, // segments + { 0, &tld_main.stats }, // os + { MI_STATS_NULL } // stats +}; + +mi_heap_t _mi_heap_main = { + &tld_main, + MI_SMALL_PAGES_EMPTY, + MI_PAGE_QUEUES_EMPTY, + MI_ATOMIC_VAR_INIT(NULL), + 0, // thread id + 0, // initial cookie + { 0, 0 }, // the key of the main heap can be fixed (unlike page keys that need to be secure!) + { {0x846ca68b}, {0}, 0 }, // random + 0, // page count + MI_BIN_FULL, 0, // page retired min/max + NULL, // next heap + false // can reclaim +}; + +bool _mi_process_is_initialized = false; // set to `true` in `mi_process_init`. + +mi_stats_t _mi_stats_main = { MI_STATS_NULL }; + + +static void mi_heap_main_init(void) { + if (_mi_heap_main.cookie == 0) { + _mi_heap_main.thread_id = _mi_thread_id(); + _mi_heap_main.cookie = _mi_os_random_weak((uintptr_t)&mi_heap_main_init); + _mi_random_init(&_mi_heap_main.random); + _mi_heap_main.keys[0] = _mi_heap_random_next(&_mi_heap_main); + _mi_heap_main.keys[1] = _mi_heap_random_next(&_mi_heap_main); + } +} + +mi_heap_t* _mi_heap_main_get(void) { + mi_heap_main_init(); + return &_mi_heap_main; +} + + +/* ----------------------------------------------------------- + Initialization and freeing of the thread local heaps +----------------------------------------------------------- */ + +// note: in x64 in release build `sizeof(mi_thread_data_t)` is under 4KiB (= OS page size). +typedef struct mi_thread_data_s { + mi_heap_t heap; // must come first due to cast in `_mi_heap_done` + mi_tld_t tld; +} mi_thread_data_t; + + +// Thread meta-data is allocated directly from the OS. For +// some programs that do not use thread pools and allocate and +// destroy many OS threads, this may causes too much overhead +// per thread so we maintain a small cache of recently freed metadata. + +#define TD_CACHE_SIZE (8) +static _Atomic(mi_thread_data_t*) td_cache[TD_CACHE_SIZE]; + +static mi_thread_data_t* mi_thread_data_alloc(void) { + // try to find thread metadata in the cache + mi_thread_data_t* td; + for (int i = 0; i < TD_CACHE_SIZE; i++) { + td = mi_atomic_load_ptr_relaxed(mi_thread_data_t, &td_cache[i]); + if (td != NULL) { + td = mi_atomic_exchange_ptr_acq_rel(mi_thread_data_t, &td_cache[i], NULL); + if (td != NULL) { + return td; + } + } + } + // if that fails, allocate directly from the OS + td = (mi_thread_data_t*)_mi_os_alloc(sizeof(mi_thread_data_t), &_mi_stats_main); + if (td == NULL) { + // if this fails, try once more. (issue #257) + td = (mi_thread_data_t*)_mi_os_alloc(sizeof(mi_thread_data_t), &_mi_stats_main); + if (td == NULL) { + // really out of memory + _mi_error_message(ENOMEM, "unable to allocate thread local heap metadata (%zu bytes)\n", sizeof(mi_thread_data_t)); + } + } + return td; +} + +static void mi_thread_data_free( mi_thread_data_t* tdfree ) { + // try to add the thread metadata to the cache + for (int i = 0; i < TD_CACHE_SIZE; i++) { + mi_thread_data_t* td = mi_atomic_load_ptr_relaxed(mi_thread_data_t, &td_cache[i]); + if (td == NULL) { + mi_thread_data_t* expected = NULL; + if (mi_atomic_cas_ptr_weak_acq_rel(mi_thread_data_t, &td_cache[i], &expected, tdfree)) { + return; + } + } + } + // if that fails, just free it directly + _mi_os_free(tdfree, sizeof(mi_thread_data_t), &_mi_stats_main); +} + +static void mi_thread_data_collect(void) { + // free all thread metadata from the cache + for (int i = 0; i < TD_CACHE_SIZE; i++) { + mi_thread_data_t* td = mi_atomic_load_ptr_relaxed(mi_thread_data_t, &td_cache[i]); + if (td != NULL) { + td = mi_atomic_exchange_ptr_acq_rel(mi_thread_data_t, &td_cache[i], NULL); + if (td != NULL) { + _mi_os_free( td, sizeof(mi_thread_data_t), &_mi_stats_main ); + } + } + } +} + +// Initialize the thread local default heap, called from `mi_thread_init` +static bool _mi_heap_init(void) { + if (mi_heap_is_initialized(mi_get_default_heap())) return true; + if (_mi_is_main_thread()) { + // mi_assert_internal(_mi_heap_main.thread_id != 0); // can happen on freeBSD where alloc is called before any initialization + // the main heap is statically allocated + mi_heap_main_init(); + _mi_heap_set_default_direct(&_mi_heap_main); + //mi_assert_internal(_mi_heap_default->tld->heap_backing == mi_get_default_heap()); + } + else { + // use `_mi_os_alloc` to allocate directly from the OS + mi_thread_data_t* td = mi_thread_data_alloc(); + if (td == NULL) return false; + + // OS allocated so already zero initialized + mi_tld_t* tld = &td->tld; + mi_heap_t* heap = &td->heap; + _mi_memcpy_aligned(tld, &tld_empty, sizeof(*tld)); + _mi_memcpy_aligned(heap, &_mi_heap_empty, sizeof(*heap)); + heap->thread_id = _mi_thread_id(); + _mi_random_init(&heap->random); + heap->cookie = _mi_heap_random_next(heap) | 1; + heap->keys[0] = _mi_heap_random_next(heap); + heap->keys[1] = _mi_heap_random_next(heap); + heap->tld = tld; + tld->heap_backing = heap; + tld->heaps = heap; + tld->segments.stats = &tld->stats; + tld->segments.os = &tld->os; + tld->os.stats = &tld->stats; + _mi_heap_set_default_direct(heap); + } + return false; +} + +// Free the thread local default heap (called from `mi_thread_done`) +static bool _mi_heap_done(mi_heap_t* heap) { + if (!mi_heap_is_initialized(heap)) return true; + + // reset default heap + _mi_heap_set_default_direct(_mi_is_main_thread() ? &_mi_heap_main : (mi_heap_t*)&_mi_heap_empty); + + // switch to backing heap + heap = heap->tld->heap_backing; + if (!mi_heap_is_initialized(heap)) return false; + + // delete all non-backing heaps in this thread + mi_heap_t* curr = heap->tld->heaps; + while (curr != NULL) { + mi_heap_t* next = curr->next; // save `next` as `curr` will be freed + if (curr != heap) { + mi_assert_internal(!mi_heap_is_backing(curr)); + mi_heap_delete(curr); + } + curr = next; + } + mi_assert_internal(heap->tld->heaps == heap && heap->next == NULL); + mi_assert_internal(mi_heap_is_backing(heap)); + + // collect if not the main thread + if (heap != &_mi_heap_main) { + _mi_heap_collect_abandon(heap); + } + + // merge stats + _mi_stats_done(&heap->tld->stats); + + // free if not the main thread + if (heap != &_mi_heap_main) { + // the following assertion does not always hold for huge segments as those are always treated + // as abondened: one may allocate it in one thread, but deallocate in another in which case + // the count can be too large or negative. todo: perhaps not count huge segments? see issue #363 + // mi_assert_internal(heap->tld->segments.count == 0 || heap->thread_id != _mi_thread_id()); + mi_thread_data_free((mi_thread_data_t*)heap); + } + else { + mi_thread_data_collect(); // free cached thread metadata + #if 0 + // never free the main thread even in debug mode; if a dll is linked statically with mimalloc, + // there may still be delete/free calls after the mi_fls_done is called. Issue #207 + _mi_heap_destroy_pages(heap); + mi_assert_internal(heap->tld->heap_backing == &_mi_heap_main); + #endif + } + return false; +} + + + +// -------------------------------------------------------- +// Try to run `mi_thread_done()` automatically so any memory +// owned by the thread but not yet released can be abandoned +// and re-owned by another thread. +// +// 1. windows dynamic library: +// call from DllMain on DLL_THREAD_DETACH +// 2. windows static library: +// use `FlsAlloc` to call a destructor when the thread is done +// 3. unix, pthreads: +// use a pthread key to call a destructor when a pthread is done +// +// In the last two cases we also need to call `mi_process_init` +// to set up the thread local keys. +// -------------------------------------------------------- + +static void _mi_thread_done(mi_heap_t* default_heap); + +#if defined(_WIN32) && defined(MI_SHARED_LIB) + // nothing to do as it is done in DllMain +#elif defined(_WIN32) && !defined(MI_SHARED_LIB) + // use thread local storage keys to detect thread ending + #include + #include + #if (_WIN32_WINNT < 0x600) // before Windows Vista + WINBASEAPI DWORD WINAPI FlsAlloc( _In_opt_ PFLS_CALLBACK_FUNCTION lpCallback ); + WINBASEAPI PVOID WINAPI FlsGetValue( _In_ DWORD dwFlsIndex ); + WINBASEAPI BOOL WINAPI FlsSetValue( _In_ DWORD dwFlsIndex, _In_opt_ PVOID lpFlsData ); + WINBASEAPI BOOL WINAPI FlsFree(_In_ DWORD dwFlsIndex); + #endif + static DWORD mi_fls_key = (DWORD)(-1); + static void NTAPI mi_fls_done(PVOID value) { + if (value!=NULL) _mi_thread_done((mi_heap_t*)value); + } +#elif defined(MI_USE_PTHREADS) + // use pthread local storage keys to detect thread ending + // (and used with MI_TLS_PTHREADS for the default heap) + pthread_key_t _mi_heap_default_key = (pthread_key_t)(-1); + static void mi_pthread_done(void* value) { + if (value!=NULL) _mi_thread_done((mi_heap_t*)value); + } +#elif defined(__wasi__) +// no pthreads in the WebAssembly Standard Interface +#else + #pragma message("define a way to call mi_thread_done when a thread is done") +#endif + +// Set up handlers so `mi_thread_done` is called automatically +static void mi_process_setup_auto_thread_done(void) { + static bool tls_initialized = false; // fine if it races + if (tls_initialized) return; + tls_initialized = true; + #if defined(_WIN32) && defined(MI_SHARED_LIB) + // nothing to do as it is done in DllMain + #elif defined(_WIN32) && !defined(MI_SHARED_LIB) + mi_fls_key = FlsAlloc(&mi_fls_done); + #elif defined(MI_USE_PTHREADS) + mi_assert_internal(_mi_heap_default_key == (pthread_key_t)(-1)); + pthread_key_create(&_mi_heap_default_key, &mi_pthread_done); + #endif + _mi_heap_set_default_direct(&_mi_heap_main); +} + + +bool _mi_is_main_thread(void) { + return (_mi_heap_main.thread_id==0 || _mi_heap_main.thread_id == _mi_thread_id()); +} + +static _Atomic(size_t) thread_count = MI_ATOMIC_VAR_INIT(1); + +size_t _mi_current_thread_count(void) { + return mi_atomic_load_relaxed(&thread_count); +} + +// This is called from the `mi_malloc_generic` +void mi_thread_init(void) mi_attr_noexcept +{ + // ensure our process has started already + mi_process_init(); + + // initialize the thread local default heap + // (this will call `_mi_heap_set_default_direct` and thus set the + // fiber/pthread key to a non-zero value, ensuring `_mi_thread_done` is called) + if (_mi_heap_init()) return; // returns true if already initialized + + _mi_stat_increase(&_mi_stats_main.threads, 1); + mi_atomic_increment_relaxed(&thread_count); + //_mi_verbose_message("thread init: 0x%zx\n", _mi_thread_id()); +} + +void mi_thread_done(void) mi_attr_noexcept { + _mi_thread_done(mi_get_default_heap()); +} + +static void _mi_thread_done(mi_heap_t* heap) { + mi_atomic_decrement_relaxed(&thread_count); + _mi_stat_decrease(&_mi_stats_main.threads, 1); + + // check thread-id as on Windows shutdown with FLS the main (exit) thread may call this on thread-local heaps... + if (heap->thread_id != _mi_thread_id()) return; + + // abandon the thread local heap + if (_mi_heap_done(heap)) return; // returns true if already ran +} + +void _mi_heap_set_default_direct(mi_heap_t* heap) { + mi_assert_internal(heap != NULL); + #if defined(MI_TLS_SLOT) + mi_tls_slot_set(MI_TLS_SLOT,heap); + #elif defined(MI_TLS_PTHREAD_SLOT_OFS) + *mi_tls_pthread_heap_slot() = heap; + #elif defined(MI_TLS_PTHREAD) + // we use _mi_heap_default_key + #else + _mi_heap_default = heap; + #endif + + // ensure the default heap is passed to `_mi_thread_done` + // setting to a non-NULL value also ensures `mi_thread_done` is called. + #if defined(_WIN32) && defined(MI_SHARED_LIB) + // nothing to do as it is done in DllMain + #elif defined(_WIN32) && !defined(MI_SHARED_LIB) + mi_assert_internal(mi_fls_key != 0); + FlsSetValue(mi_fls_key, heap); + #elif defined(MI_USE_PTHREADS) + if (_mi_heap_default_key != (pthread_key_t)(-1)) { // can happen during recursive invocation on freeBSD + pthread_setspecific(_mi_heap_default_key, heap); + } + #endif +} + + +// -------------------------------------------------------- +// Run functions on process init/done, and thread init/done +// -------------------------------------------------------- +static void mi_process_done(void); + +static bool os_preloading = true; // true until this module is initialized +static bool mi_redirected = false; // true if malloc redirects to mi_malloc + +// Returns true if this module has not been initialized; Don't use C runtime routines until it returns false. +bool _mi_preloading(void) { + return os_preloading; +} + +mi_decl_nodiscard bool mi_is_redirected(void) mi_attr_noexcept { + return mi_redirected; +} + +// Communicate with the redirection module on Windows +#if defined(_WIN32) && defined(MI_SHARED_LIB) +#ifdef __cplusplus +extern "C" { +#endif +mi_decl_export void _mi_redirect_entry(DWORD reason) { + // called on redirection; careful as this may be called before DllMain + if (reason == DLL_PROCESS_ATTACH) { + mi_redirected = true; + } + else if (reason == DLL_PROCESS_DETACH) { + mi_redirected = false; + } + else if (reason == DLL_THREAD_DETACH) { + mi_thread_done(); + } +} +__declspec(dllimport) bool mi_allocator_init(const char** message); +__declspec(dllimport) void mi_allocator_done(void); +#ifdef __cplusplus +} +#endif +#else +static bool mi_allocator_init(const char** message) { + if (message != NULL) *message = NULL; + return true; +} +static void mi_allocator_done(void) { + // nothing to do +} +#endif + +// Called once by the process loader +static void mi_process_load(void) { + mi_heap_main_init(); + #if defined(MI_TLS_RECURSE_GUARD) + volatile mi_heap_t* dummy = _mi_heap_default; // access TLS to allocate it before setting tls_initialized to true; + MI_UNUSED(dummy); + #endif + os_preloading = false; + #if !(defined(_WIN32) && defined(MI_SHARED_LIB)) // use Dll process detach (see below) instead of atexit (issue #521) + atexit(&mi_process_done); + #endif + _mi_options_init(); + mi_process_init(); + //mi_stats_reset();- + if (mi_redirected) _mi_verbose_message("malloc is redirected.\n"); + + // show message from the redirector (if present) + const char* msg = NULL; + mi_allocator_init(&msg); + if (msg != NULL && (mi_option_is_enabled(mi_option_verbose) || mi_option_is_enabled(mi_option_show_errors))) { + _mi_fputs(NULL,NULL,NULL,msg); + } +} + +#if defined(_WIN32) && (defined(_M_IX86) || defined(_M_X64)) +#include +mi_decl_cache_align bool _mi_cpu_has_fsrm = false; + +static void mi_detect_cpu_features(void) { + // FSRM for fast rep movsb support (AMD Zen3+ (~2020) or Intel Ice Lake+ (~2017)) + int32_t cpu_info[4]; + __cpuid(cpu_info, 7); + _mi_cpu_has_fsrm = ((cpu_info[3] & (1 << 4)) != 0); // bit 4 of EDX : see +} +#else +static void mi_detect_cpu_features(void) { + // nothing +} +#endif + +// Initialize the process; called by thread_init or the process loader +void mi_process_init(void) mi_attr_noexcept { + // ensure we are called once + if (_mi_process_is_initialized) return; + _mi_verbose_message("process init: 0x%zx\n", _mi_thread_id()); + _mi_process_is_initialized = true; + mi_process_setup_auto_thread_done(); + + + mi_detect_cpu_features(); + _mi_os_init(); + mi_heap_main_init(); + #if (MI_DEBUG) + _mi_verbose_message("debug level : %d\n", MI_DEBUG); + #endif + _mi_verbose_message("secure level: %d\n", MI_SECURE); + mi_thread_init(); + + #if defined(_WIN32) && !defined(MI_SHARED_LIB) + // When building as a static lib the FLS cleanup happens to early for the main thread. + // To avoid this, set the FLS value for the main thread to NULL so the fls cleanup + // will not call _mi_thread_done on the (still executing) main thread. See issue #508. + FlsSetValue(mi_fls_key, NULL); + #endif + + mi_stats_reset(); // only call stat reset *after* thread init (or the heap tld == NULL) + + if (mi_option_is_enabled(mi_option_reserve_huge_os_pages)) { + size_t pages = mi_option_get_clamp(mi_option_reserve_huge_os_pages, 0, 128*1024); + long reserve_at = mi_option_get(mi_option_reserve_huge_os_pages_at); + if (reserve_at != -1) { + mi_reserve_huge_os_pages_at(pages, reserve_at, pages*500); + } else { + mi_reserve_huge_os_pages_interleave(pages, 0, pages*500); + } + } + if (mi_option_is_enabled(mi_option_reserve_os_memory)) { + long ksize = mi_option_get(mi_option_reserve_os_memory); + if (ksize > 0) { + mi_reserve_os_memory((size_t)ksize*MI_KiB, true /* commit? */, true /* allow large pages? */); + } + } +} + +// Called when the process is done (through `at_exit`) +static void mi_process_done(void) { + // only shutdown if we were initialized + if (!_mi_process_is_initialized) return; + // ensure we are called once + static bool process_done = false; + if (process_done) return; + process_done = true; + + #if defined(_WIN32) && !defined(MI_SHARED_LIB) + FlsFree(mi_fls_key); // call thread-done on all threads (except the main thread) to prevent dangling callback pointer if statically linked with a DLL; Issue #208 + #endif + + #ifndef MI_SKIP_COLLECT_ON_EXIT + #if (MI_DEBUG != 0) || !defined(MI_SHARED_LIB) + // free all memory if possible on process exit. This is not needed for a stand-alone process + // but should be done if mimalloc is statically linked into another shared library which + // is repeatedly loaded/unloaded, see issue #281. + mi_collect(true /* force */ ); + #endif + #endif + + if (mi_option_is_enabled(mi_option_show_stats) || mi_option_is_enabled(mi_option_verbose)) { + mi_stats_print(NULL); + } + mi_allocator_done(); + _mi_verbose_message("process done: 0x%zx\n", _mi_heap_main.thread_id); + os_preloading = true; // don't call the C runtime anymore +} + + + +#if defined(_WIN32) && defined(MI_SHARED_LIB) + // Windows DLL: easy to hook into process_init and thread_done + __declspec(dllexport) BOOL WINAPI DllMain(HINSTANCE inst, DWORD reason, LPVOID reserved) { + MI_UNUSED(reserved); + MI_UNUSED(inst); + if (reason==DLL_PROCESS_ATTACH) { + mi_process_load(); + } + else if (reason==DLL_PROCESS_DETACH) { + mi_process_done(); + } + else if (reason==DLL_THREAD_DETACH) { + if (!mi_is_redirected()) { + mi_thread_done(); + } + } + return TRUE; + } + +#elif defined(_MSC_VER) + // MSVC: use data section magic for static libraries + // See + static int _mi_process_init(void) { + mi_process_load(); + return 0; + } + typedef int(*_mi_crt_callback_t)(void); + #if defined(_M_X64) || defined(_M_ARM64) + __pragma(comment(linker, "/include:" "_mi_msvc_initu")) + #pragma section(".CRT$XIU", long, read) + #else + __pragma(comment(linker, "/include:" "__mi_msvc_initu")) + #endif + #pragma data_seg(".CRT$XIU") + mi_decl_externc _mi_crt_callback_t _mi_msvc_initu[] = { &_mi_process_init }; + #pragma data_seg() + +#elif defined(__cplusplus) + // C++: use static initialization to detect process start + static bool _mi_process_init(void) { + mi_process_load(); + return (_mi_heap_main.thread_id != 0); + } + static bool mi_initialized = _mi_process_init(); + +#elif defined(__GNUC__) || defined(__clang__) + // GCC,Clang: use the constructor attribute + static void __attribute__((constructor)) _mi_process_init(void) { + mi_process_load(); + } + +#else +#pragma message("define a way to call mi_process_load on your platform") +#endif diff --git a/compat/mimalloc/mimalloc-atomic.h b/compat/mimalloc/mimalloc-atomic.h new file mode 100644 index 00000000000000..c66f80493321ee --- /dev/null +++ b/compat/mimalloc/mimalloc-atomic.h @@ -0,0 +1,338 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2021 Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ +#pragma once +#ifndef MIMALLOC_ATOMIC_H +#define MIMALLOC_ATOMIC_H + +// -------------------------------------------------------------------------------------------- +// Atomics +// We need to be portable between C, C++, and MSVC. +// We base the primitives on the C/C++ atomics and create a mimimal wrapper for MSVC in C compilation mode. +// This is why we try to use only `uintptr_t` and `*` as atomic types. +// To gain better insight in the range of used atomics, we use explicitly named memory order operations +// instead of passing the memory order as a parameter. +// ----------------------------------------------------------------------------------------------- + +#if defined(__cplusplus) +// Use C++ atomics +#include +#define _Atomic(tp) std::atomic +#define mi_atomic(name) std::atomic_##name +#define mi_memory_order(name) std::memory_order_##name +#if !defined(ATOMIC_VAR_INIT) || (__cplusplus >= 202002L) // c++20, see issue #571 + #define MI_ATOMIC_VAR_INIT(x) x +#else + #define MI_ATOMIC_VAR_INIT(x) ATOMIC_VAR_INIT(x) +#endif +#elif defined(_MSC_VER) +// Use MSVC C wrapper for C11 atomics +#define _Atomic(tp) tp +#define MI_ATOMIC_VAR_INIT(x) x +#define mi_atomic(name) mi_atomic_##name +#define mi_memory_order(name) mi_memory_order_##name +#else +// Use C11 atomics +#include +#define mi_atomic(name) atomic_##name +#define mi_memory_order(name) memory_order_##name +#define MI_ATOMIC_VAR_INIT(x) ATOMIC_VAR_INIT(x) +#endif + +// Various defines for all used memory orders in mimalloc +#define mi_atomic_cas_weak(p,expected,desired,mem_success,mem_fail) \ + mi_atomic(compare_exchange_weak_explicit)(p,expected,desired,mem_success,mem_fail) + +#define mi_atomic_cas_strong(p,expected,desired,mem_success,mem_fail) \ + mi_atomic(compare_exchange_strong_explicit)(p,expected,desired,mem_success,mem_fail) + +#define mi_atomic_load_acquire(p) mi_atomic(load_explicit)(p,mi_memory_order(acquire)) +#define mi_atomic_load_relaxed(p) mi_atomic(load_explicit)(p,mi_memory_order(relaxed)) +#define mi_atomic_store_release(p,x) mi_atomic(store_explicit)(p,x,mi_memory_order(release)) +#define mi_atomic_store_relaxed(p,x) mi_atomic(store_explicit)(p,x,mi_memory_order(relaxed)) +#define mi_atomic_exchange_release(p,x) mi_atomic(exchange_explicit)(p,x,mi_memory_order(release)) +#define mi_atomic_exchange_acq_rel(p,x) mi_atomic(exchange_explicit)(p,x,mi_memory_order(acq_rel)) +#define mi_atomic_cas_weak_release(p,exp,des) mi_atomic_cas_weak(p,exp,des,mi_memory_order(release),mi_memory_order(relaxed)) +#define mi_atomic_cas_weak_acq_rel(p,exp,des) mi_atomic_cas_weak(p,exp,des,mi_memory_order(acq_rel),mi_memory_order(acquire)) +#define mi_atomic_cas_strong_release(p,exp,des) mi_atomic_cas_strong(p,exp,des,mi_memory_order(release),mi_memory_order(relaxed)) +#define mi_atomic_cas_strong_acq_rel(p,exp,des) mi_atomic_cas_strong(p,exp,des,mi_memory_order(acq_rel),mi_memory_order(acquire)) + +#define mi_atomic_add_relaxed(p,x) mi_atomic(fetch_add_explicit)(p,x,mi_memory_order(relaxed)) +#define mi_atomic_sub_relaxed(p,x) mi_atomic(fetch_sub_explicit)(p,x,mi_memory_order(relaxed)) +#define mi_atomic_add_acq_rel(p,x) mi_atomic(fetch_add_explicit)(p,x,mi_memory_order(acq_rel)) +#define mi_atomic_sub_acq_rel(p,x) mi_atomic(fetch_sub_explicit)(p,x,mi_memory_order(acq_rel)) +#define mi_atomic_and_acq_rel(p,x) mi_atomic(fetch_and_explicit)(p,x,mi_memory_order(acq_rel)) +#define mi_atomic_or_acq_rel(p,x) mi_atomic(fetch_or_explicit)(p,x,mi_memory_order(acq_rel)) + +#define mi_atomic_increment_relaxed(p) mi_atomic_add_relaxed(p,(uintptr_t)1) +#define mi_atomic_decrement_relaxed(p) mi_atomic_sub_relaxed(p,(uintptr_t)1) +#define mi_atomic_increment_acq_rel(p) mi_atomic_add_acq_rel(p,(uintptr_t)1) +#define mi_atomic_decrement_acq_rel(p) mi_atomic_sub_acq_rel(p,(uintptr_t)1) + +static inline void mi_atomic_yield(void); +static inline intptr_t mi_atomic_addi(_Atomic(intptr_t)*p, intptr_t add); +static inline intptr_t mi_atomic_subi(_Atomic(intptr_t)*p, intptr_t sub); + + +#if defined(__cplusplus) || !defined(_MSC_VER) + +// In C++/C11 atomics we have polymorphic atomics so can use the typed `ptr` variants (where `tp` is the type of atomic value) +// We use these macros so we can provide a typed wrapper in MSVC in C compilation mode as well +#define mi_atomic_load_ptr_acquire(tp,p) mi_atomic_load_acquire(p) +#define mi_atomic_load_ptr_relaxed(tp,p) mi_atomic_load_relaxed(p) + +// In C++ we need to add casts to help resolve templates if NULL is passed +#if defined(__cplusplus) +#define mi_atomic_store_ptr_release(tp,p,x) mi_atomic_store_release(p,(tp*)x) +#define mi_atomic_store_ptr_relaxed(tp,p,x) mi_atomic_store_relaxed(p,(tp*)x) +#define mi_atomic_cas_ptr_weak_release(tp,p,exp,des) mi_atomic_cas_weak_release(p,exp,(tp*)des) +#define mi_atomic_cas_ptr_weak_acq_rel(tp,p,exp,des) mi_atomic_cas_weak_acq_rel(p,exp,(tp*)des) +#define mi_atomic_cas_ptr_strong_release(tp,p,exp,des) mi_atomic_cas_strong_release(p,exp,(tp*)des) +#define mi_atomic_exchange_ptr_release(tp,p,x) mi_atomic_exchange_release(p,(tp*)x) +#define mi_atomic_exchange_ptr_acq_rel(tp,p,x) mi_atomic_exchange_acq_rel(p,(tp*)x) +#else +#define mi_atomic_store_ptr_release(tp,p,x) mi_atomic_store_release(p,x) +#define mi_atomic_store_ptr_relaxed(tp,p,x) mi_atomic_store_relaxed(p,x) +#define mi_atomic_cas_ptr_weak_release(tp,p,exp,des) mi_atomic_cas_weak_release(p,exp,des) +#define mi_atomic_cas_ptr_weak_acq_rel(tp,p,exp,des) mi_atomic_cas_weak_acq_rel(p,exp,des) +#define mi_atomic_cas_ptr_strong_release(tp,p,exp,des) mi_atomic_cas_strong_release(p,exp,des) +#define mi_atomic_exchange_ptr_release(tp,p,x) mi_atomic_exchange_release(p,x) +#define mi_atomic_exchange_ptr_acq_rel(tp,p,x) mi_atomic_exchange_acq_rel(p,x) +#endif + +// These are used by the statistics +static inline int64_t mi_atomic_addi64_relaxed(volatile int64_t* p, int64_t add) { + return mi_atomic(fetch_add_explicit)((_Atomic(int64_t)*)p, add, mi_memory_order(relaxed)); +} +static inline void mi_atomic_maxi64_relaxed(volatile int64_t* p, int64_t x) { + int64_t current = mi_atomic_load_relaxed((_Atomic(int64_t)*)p); + while (current < x && !mi_atomic_cas_weak_release((_Atomic(int64_t)*)p, ¤t, x)) { /* nothing */ }; +} + +// Used by timers +#define mi_atomic_loadi64_acquire(p) mi_atomic(load_explicit)(p,mi_memory_order(acquire)) +#define mi_atomic_loadi64_relaxed(p) mi_atomic(load_explicit)(p,mi_memory_order(relaxed)) +#define mi_atomic_storei64_release(p,x) mi_atomic(store_explicit)(p,x,mi_memory_order(release)) +#define mi_atomic_storei64_relaxed(p,x) mi_atomic(store_explicit)(p,x,mi_memory_order(relaxed)) + + + +#elif defined(_MSC_VER) + +// MSVC C compilation wrapper that uses Interlocked operations to model C11 atomics. +#define WIN32_LEAN_AND_MEAN +#include +#include +#ifdef _WIN64 +typedef LONG64 msc_intptr_t; +#define MI_64(f) f##64 +#else +typedef LONG msc_intptr_t; +#define MI_64(f) f +#endif + +typedef enum mi_memory_order_e { + mi_memory_order_relaxed, + mi_memory_order_consume, + mi_memory_order_acquire, + mi_memory_order_release, + mi_memory_order_acq_rel, + mi_memory_order_seq_cst +} mi_memory_order; + +static inline uintptr_t mi_atomic_fetch_add_explicit(_Atomic(uintptr_t)*p, uintptr_t add, mi_memory_order mo) { + (void)(mo); + return (uintptr_t)MI_64(_InterlockedExchangeAdd)((volatile msc_intptr_t*)p, (msc_intptr_t)add); +} +static inline uintptr_t mi_atomic_fetch_sub_explicit(_Atomic(uintptr_t)*p, uintptr_t sub, mi_memory_order mo) { + (void)(mo); + return (uintptr_t)MI_64(_InterlockedExchangeAdd)((volatile msc_intptr_t*)p, -((msc_intptr_t)sub)); +} +static inline uintptr_t mi_atomic_fetch_and_explicit(_Atomic(uintptr_t)*p, uintptr_t x, mi_memory_order mo) { + (void)(mo); + return (uintptr_t)MI_64(_InterlockedAnd)((volatile msc_intptr_t*)p, (msc_intptr_t)x); +} +static inline uintptr_t mi_atomic_fetch_or_explicit(_Atomic(uintptr_t)*p, uintptr_t x, mi_memory_order mo) { + (void)(mo); + return (uintptr_t)MI_64(_InterlockedOr)((volatile msc_intptr_t*)p, (msc_intptr_t)x); +} +static inline bool mi_atomic_compare_exchange_strong_explicit(_Atomic(uintptr_t)*p, uintptr_t* expected, uintptr_t desired, mi_memory_order mo1, mi_memory_order mo2) { + (void)(mo1); (void)(mo2); + uintptr_t read = (uintptr_t)MI_64(_InterlockedCompareExchange)((volatile msc_intptr_t*)p, (msc_intptr_t)desired, (msc_intptr_t)(*expected)); + if (read == *expected) { + return true; + } + else { + *expected = read; + return false; + } +} +static inline bool mi_atomic_compare_exchange_weak_explicit(_Atomic(uintptr_t)*p, uintptr_t* expected, uintptr_t desired, mi_memory_order mo1, mi_memory_order mo2) { + return mi_atomic_compare_exchange_strong_explicit(p, expected, desired, mo1, mo2); +} +static inline uintptr_t mi_atomic_exchange_explicit(_Atomic(uintptr_t)*p, uintptr_t exchange, mi_memory_order mo) { + (void)(mo); + return (uintptr_t)MI_64(_InterlockedExchange)((volatile msc_intptr_t*)p, (msc_intptr_t)exchange); +} +static inline void mi_atomic_thread_fence(mi_memory_order mo) { + (void)(mo); + _Atomic(uintptr_t) x = 0; + mi_atomic_exchange_explicit(&x, 1, mo); +} +static inline uintptr_t mi_atomic_load_explicit(_Atomic(uintptr_t) const* p, mi_memory_order mo) { + (void)(mo); +#if defined(_M_IX86) || defined(_M_X64) + return *p; +#else + uintptr_t x = *p; + if (mo > mi_memory_order_relaxed) { + while (!mi_atomic_compare_exchange_weak_explicit(p, &x, x, mo, mi_memory_order_relaxed)) { /* nothing */ }; + } + return x; +#endif +} +static inline void mi_atomic_store_explicit(_Atomic(uintptr_t)*p, uintptr_t x, mi_memory_order mo) { + (void)(mo); +#if defined(_M_IX86) || defined(_M_X64) + *p = x; +#else + mi_atomic_exchange_explicit(p, x, mo); +#endif +} +static inline int64_t mi_atomic_loadi64_explicit(_Atomic(int64_t)*p, mi_memory_order mo) { + (void)(mo); +#if defined(_M_X64) + return *p; +#else + int64_t old = *p; + int64_t x = old; + while ((old = InterlockedCompareExchange64(p, x, old)) != x) { + x = old; + } + return x; +#endif +} +static inline void mi_atomic_storei64_explicit(_Atomic(int64_t)*p, int64_t x, mi_memory_order mo) { + (void)(mo); +#if defined(x_M_IX86) || defined(_M_X64) + *p = x; +#else + InterlockedExchange64(p, x); +#endif +} + +// These are used by the statistics +static inline int64_t mi_atomic_addi64_relaxed(volatile _Atomic(int64_t)*p, int64_t add) { +#ifdef _WIN64 + return (int64_t)mi_atomic_addi((int64_t*)p, add); +#else + int64_t current; + int64_t sum; + do { + current = *p; + sum = current + add; + } while (_InterlockedCompareExchange64(p, sum, current) != current); + return current; +#endif +} +static inline void mi_atomic_maxi64_relaxed(volatile _Atomic(int64_t)*p, int64_t x) { + int64_t current; + do { + current = *p; + } while (current < x && _InterlockedCompareExchange64(p, x, current) != current); +} + +// The pointer macros cast to `uintptr_t`. +#define mi_atomic_load_ptr_acquire(tp,p) (tp*)mi_atomic_load_acquire((_Atomic(uintptr_t)*)(p)) +#define mi_atomic_load_ptr_relaxed(tp,p) (tp*)mi_atomic_load_relaxed((_Atomic(uintptr_t)*)(p)) +#define mi_atomic_store_ptr_release(tp,p,x) mi_atomic_store_release((_Atomic(uintptr_t)*)(p),(uintptr_t)(x)) +#define mi_atomic_store_ptr_relaxed(tp,p,x) mi_atomic_store_relaxed((_Atomic(uintptr_t)*)(p),(uintptr_t)(x)) +#define mi_atomic_cas_ptr_weak_release(tp,p,exp,des) mi_atomic_cas_weak_release((_Atomic(uintptr_t)*)(p),(uintptr_t*)exp,(uintptr_t)des) +#define mi_atomic_cas_ptr_weak_acq_rel(tp,p,exp,des) mi_atomic_cas_weak_acq_rel((_Atomic(uintptr_t)*)(p),(uintptr_t*)exp,(uintptr_t)des) +#define mi_atomic_cas_ptr_strong_release(tp,p,exp,des) mi_atomic_cas_strong_release((_Atomic(uintptr_t)*)(p),(uintptr_t*)exp,(uintptr_t)des) +#define mi_atomic_exchange_ptr_release(tp,p,x) (tp*)mi_atomic_exchange_release((_Atomic(uintptr_t)*)(p),(uintptr_t)x) +#define mi_atomic_exchange_ptr_acq_rel(tp,p,x) (tp*)mi_atomic_exchange_acq_rel((_Atomic(uintptr_t)*)(p),(uintptr_t)x) + +#define mi_atomic_loadi64_acquire(p) mi_atomic(loadi64_explicit)(p,mi_memory_order(acquire)) +#define mi_atomic_loadi64_relaxed(p) mi_atomic(loadi64_explicit)(p,mi_memory_order(relaxed)) +#define mi_atomic_storei64_release(p,x) mi_atomic(storei64_explicit)(p,x,mi_memory_order(release)) +#define mi_atomic_storei64_relaxed(p,x) mi_atomic(storei64_explicit)(p,x,mi_memory_order(relaxed)) + + +#endif + + +// Atomically add a signed value; returns the previous value. +static inline intptr_t mi_atomic_addi(_Atomic(intptr_t)*p, intptr_t add) { + return (intptr_t)mi_atomic_add_acq_rel((_Atomic(uintptr_t)*)p, (uintptr_t)add); +} + +// Atomically subtract a signed value; returns the previous value. +static inline intptr_t mi_atomic_subi(_Atomic(intptr_t)*p, intptr_t sub) { + return (intptr_t)mi_atomic_addi(p, -sub); +} + +// Yield +#if defined(__cplusplus) +#include +static inline void mi_atomic_yield(void) { + std::this_thread::yield(); +} +#elif defined(_WIN32) +#define WIN32_LEAN_AND_MEAN +#include +static inline void mi_atomic_yield(void) { + YieldProcessor(); +} +#elif defined(__SSE2__) +#include +static inline void mi_atomic_yield(void) { + _mm_pause(); +} +#elif (defined(__GNUC__) || defined(__clang__)) && \ + (defined(__x86_64__) || defined(__i386__) || defined(__arm__) || defined(__armel__) || defined(__ARMEL__) || \ + defined(__aarch64__) || defined(__powerpc__) || defined(__ppc__) || defined(__PPC__)) +#if defined(__x86_64__) || defined(__i386__) +static inline void mi_atomic_yield(void) { + __asm__ volatile ("pause" ::: "memory"); +} +#elif defined(__aarch64__) +static inline void mi_atomic_yield(void) { + __asm__ volatile("wfe"); +} +#elif (defined(__arm__) && __ARM_ARCH__ >= 7) +static inline void mi_atomic_yield(void) { + __asm__ volatile("yield" ::: "memory"); +} +#elif defined(__powerpc__) || defined(__ppc__) || defined(__PPC__) +static inline void mi_atomic_yield(void) { + __asm__ __volatile__ ("or 27,27,27" ::: "memory"); +} +#elif defined(__armel__) || defined(__ARMEL__) +static inline void mi_atomic_yield(void) { + __asm__ volatile ("nop" ::: "memory"); +} +#endif +#elif defined(__sun) +// Fallback for other archs +#include +static inline void mi_atomic_yield(void) { + smt_pause(); +} +#elif defined(__wasi__) +#include +static inline void mi_atomic_yield(void) { + sched_yield(); +} +#else +#include +static inline void mi_atomic_yield(void) { + sleep(0); +} +#endif + + +#endif // __MIMALLOC_ATOMIC_H diff --git a/compat/mimalloc/mimalloc-internal.h b/compat/mimalloc/mimalloc-internal.h new file mode 100644 index 00000000000000..de5c53b1e52f93 --- /dev/null +++ b/compat/mimalloc/mimalloc-internal.h @@ -0,0 +1,1049 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2022, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ +#pragma once +#ifndef MIMALLOC_INTERNAL_H +#define MIMALLOC_INTERNAL_H + +#include "mimalloc-types.h" + +#if (MI_DEBUG>0) +#define mi_trace_message(...) _mi_trace_message(__VA_ARGS__) +#else +#define mi_trace_message(...) +#endif + +#define MI_CACHE_LINE 64 +#if defined(_MSC_VER) +#pragma warning(disable:4127) // suppress constant conditional warning (due to MI_SECURE paths) +#pragma warning(disable:26812) // unscoped enum warning +#define mi_decl_noinline __declspec(noinline) +#define mi_decl_thread __declspec(thread) +#define mi_decl_cache_align __declspec(align(MI_CACHE_LINE)) +#elif (defined(__GNUC__) && (__GNUC__ >= 3)) || defined(__clang__) // includes clang and icc +#define mi_decl_noinline __attribute__((noinline)) +#define mi_decl_thread __thread +#define mi_decl_cache_align __attribute__((aligned(MI_CACHE_LINE))) +#else +#define mi_decl_noinline +#define mi_decl_thread __thread // hope for the best :-) +#define mi_decl_cache_align +#endif + +#if defined(__EMSCRIPTEN__) && !defined(__wasi__) +#define __wasi__ +#endif + +#if defined(__cplusplus) +#define mi_decl_externc extern "C" +#else +#define mi_decl_externc +#endif + +#if !defined(_WIN32) && !defined(__wasi__) +#define MI_USE_PTHREADS +#include +#endif + +// "options.c" +void _mi_fputs(mi_output_fun* out, void* arg, const char* prefix, const char* message); +void _mi_fprintf(mi_output_fun* out, void* arg, const char* fmt, ...); +void _mi_warning_message(const char* fmt, ...); +void _mi_verbose_message(const char* fmt, ...); +void _mi_trace_message(const char* fmt, ...); +void _mi_options_init(void); +void _mi_error_message(int err, const char* fmt, ...); + +// random.c +void _mi_random_init(mi_random_ctx_t* ctx); +void _mi_random_split(mi_random_ctx_t* ctx, mi_random_ctx_t* new_ctx); +uintptr_t _mi_random_next(mi_random_ctx_t* ctx); +uintptr_t _mi_heap_random_next(mi_heap_t* heap); +uintptr_t _mi_os_random_weak(uintptr_t extra_seed); +static inline uintptr_t _mi_random_shuffle(uintptr_t x); + +// init.c +extern mi_decl_cache_align mi_stats_t _mi_stats_main; +extern mi_decl_cache_align const mi_page_t _mi_page_empty; +bool _mi_is_main_thread(void); +size_t _mi_current_thread_count(void); +bool _mi_preloading(void); // true while the C runtime is not ready + +// os.c +size_t _mi_os_page_size(void); +void _mi_os_init(void); // called from process init +void* _mi_os_alloc(size_t size, mi_stats_t* stats); // to allocate thread local data +void _mi_os_free(void* p, size_t size, mi_stats_t* stats); // to free thread local data + +bool _mi_os_protect(void* addr, size_t size); +bool _mi_os_unprotect(void* addr, size_t size); +bool _mi_os_commit(void* addr, size_t size, bool* is_zero, mi_stats_t* stats); +bool _mi_os_decommit(void* p, size_t size, mi_stats_t* stats); +bool _mi_os_reset(void* p, size_t size, mi_stats_t* stats); +// bool _mi_os_unreset(void* p, size_t size, bool* is_zero, mi_stats_t* stats); +size_t _mi_os_good_alloc_size(size_t size); +bool _mi_os_has_overcommit(void); + +// arena.c +void* _mi_arena_alloc_aligned(size_t size, size_t alignment, bool* commit, bool* large, bool* is_pinned, bool* is_zero, size_t* memid, mi_os_tld_t* tld); +void* _mi_arena_alloc(size_t size, bool* commit, bool* large, bool* is_pinned, bool* is_zero, size_t* memid, mi_os_tld_t* tld); +void _mi_arena_free(void* p, size_t size, size_t memid, bool is_committed, mi_os_tld_t* tld); + +// "segment-cache.c" +void* _mi_segment_cache_pop(size_t size, mi_commit_mask_t* commit_mask, mi_commit_mask_t* decommit_mask, bool* large, bool* is_pinned, bool* is_zero, size_t* memid, mi_os_tld_t* tld); +bool _mi_segment_cache_push(void* start, size_t size, size_t memid, const mi_commit_mask_t* commit_mask, const mi_commit_mask_t* decommit_mask, bool is_large, bool is_pinned, mi_os_tld_t* tld); +void _mi_segment_cache_collect(bool force, mi_os_tld_t* tld); +void _mi_segment_map_allocated_at(const mi_segment_t* segment); +void _mi_segment_map_freed_at(const mi_segment_t* segment); + +// "segment.c" +mi_page_t* _mi_segment_page_alloc(mi_heap_t* heap, size_t block_wsize, mi_segments_tld_t* tld, mi_os_tld_t* os_tld); +void _mi_segment_page_free(mi_page_t* page, bool force, mi_segments_tld_t* tld); +void _mi_segment_page_abandon(mi_page_t* page, mi_segments_tld_t* tld); +bool _mi_segment_try_reclaim_abandoned( mi_heap_t* heap, bool try_all, mi_segments_tld_t* tld); +void _mi_segment_thread_collect(mi_segments_tld_t* tld); +void _mi_segment_huge_page_free(mi_segment_t* segment, mi_page_t* page, mi_block_t* block); + +uint8_t* _mi_segment_page_start(const mi_segment_t* segment, const mi_page_t* page, size_t* page_size); // page start for any page +void _mi_abandoned_reclaim_all(mi_heap_t* heap, mi_segments_tld_t* tld); +void _mi_abandoned_await_readers(void); +void _mi_abandoned_collect(mi_heap_t* heap, bool force, mi_segments_tld_t* tld); + + + +// "page.c" +void* _mi_malloc_generic(mi_heap_t* heap, size_t size) mi_attr_noexcept mi_attr_malloc; + +void _mi_page_retire(mi_page_t* page) mi_attr_noexcept; // free the page if there are no other pages with many free blocks +void _mi_page_unfull(mi_page_t* page); +void _mi_page_free(mi_page_t* page, mi_page_queue_t* pq, bool force); // free the page +void _mi_page_abandon(mi_page_t* page, mi_page_queue_t* pq); // abandon the page, to be picked up by another thread... +void _mi_heap_delayed_free(mi_heap_t* heap); +void _mi_heap_collect_retired(mi_heap_t* heap, bool force); + +void _mi_page_use_delayed_free(mi_page_t* page, mi_delayed_t delay, bool override_never); +size_t _mi_page_queue_append(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_queue_t* append); +void _mi_deferred_free(mi_heap_t* heap, bool force); + +void _mi_page_free_collect(mi_page_t* page,bool force); +void _mi_page_reclaim(mi_heap_t* heap, mi_page_t* page); // callback from segments + +size_t _mi_bin_size(uint8_t bin); // for stats +uint8_t _mi_bin(size_t size); // for stats + +// "heap.c" +void _mi_heap_destroy_pages(mi_heap_t* heap); +void _mi_heap_collect_abandon(mi_heap_t* heap); +void _mi_heap_set_default_direct(mi_heap_t* heap); + +// "stats.c" +void _mi_stats_done(mi_stats_t* stats); + +mi_msecs_t _mi_clock_now(void); +mi_msecs_t _mi_clock_end(mi_msecs_t start); +mi_msecs_t _mi_clock_start(void); + +// "alloc.c" +void* _mi_page_malloc(mi_heap_t* heap, mi_page_t* page, size_t size) mi_attr_noexcept; // called from `_mi_malloc_generic` +void* _mi_heap_malloc_zero(mi_heap_t* heap, size_t size, bool zero) mi_attr_noexcept; +void* _mi_heap_realloc_zero(mi_heap_t* heap, void* p, size_t newsize, bool zero) mi_attr_noexcept; +mi_block_t* _mi_page_ptr_unalign(const mi_segment_t* segment, const mi_page_t* page, const void* p); +bool _mi_free_delayed_block(mi_block_t* block); +void _mi_block_zero_init(const mi_page_t* page, void* p, size_t size); + +#if MI_DEBUG>1 +bool _mi_page_is_valid(mi_page_t* page); +#endif + + +// ------------------------------------------------------ +// Branches +// ------------------------------------------------------ + +#if defined(__GNUC__) || defined(__clang__) +#define mi_unlikely(x) __builtin_expect(!!(x),false) +#define mi_likely(x) __builtin_expect(!!(x),true) +#else +#define mi_unlikely(x) (x) +#define mi_likely(x) (x) +#endif + +#ifndef __has_builtin +#define __has_builtin(x) 0 +#endif + + +/* ----------------------------------------------------------- + Error codes passed to `_mi_fatal_error` + All are recoverable but EFAULT is a serious error and aborts by default in secure mode. + For portability define undefined error codes using common Unix codes: + +----------------------------------------------------------- */ +#include +#ifndef EAGAIN // double free +#define EAGAIN (11) +#endif +#ifndef ENOMEM // out of memory +#define ENOMEM (12) +#endif +#ifndef EFAULT // corrupted free-list or meta-data +#define EFAULT (14) +#endif +#ifndef EINVAL // trying to free an invalid pointer +#define EINVAL (22) +#endif +#ifndef EOVERFLOW // count*size overflow +#define EOVERFLOW (75) +#endif + + +/* ----------------------------------------------------------- + Inlined definitions +----------------------------------------------------------- */ +#define MI_UNUSED(x) (void)(x) +#if (MI_DEBUG>0) +#define MI_UNUSED_RELEASE(x) +#else +#define MI_UNUSED_RELEASE(x) MI_UNUSED(x) +#endif + +#define MI_INIT4(x) x(),x(),x(),x() +#define MI_INIT8(x) MI_INIT4(x),MI_INIT4(x) +#define MI_INIT16(x) MI_INIT8(x),MI_INIT8(x) +#define MI_INIT32(x) MI_INIT16(x),MI_INIT16(x) +#define MI_INIT64(x) MI_INIT32(x),MI_INIT32(x) +#define MI_INIT128(x) MI_INIT64(x),MI_INIT64(x) +#define MI_INIT256(x) MI_INIT128(x),MI_INIT128(x) + + +// Is `x` a power of two? (0 is considered a power of two) +static inline bool _mi_is_power_of_two(uintptr_t x) { + return ((x & (x - 1)) == 0); +} + +// Align upwards +static inline uintptr_t _mi_align_up(uintptr_t sz, size_t alignment) { + mi_assert_internal(alignment != 0); + uintptr_t mask = alignment - 1; + if ((alignment & mask) == 0) { // power of two? + return ((sz + mask) & ~mask); + } + else { + return (((sz + mask)/alignment)*alignment); + } +} + +// Align downwards +static inline uintptr_t _mi_align_down(uintptr_t sz, size_t alignment) { + mi_assert_internal(alignment != 0); + uintptr_t mask = alignment - 1; + if ((alignment & mask) == 0) { // power of two? + return (sz & ~mask); + } + else { + return ((sz / alignment) * alignment); + } +} + +// Divide upwards: `s <= _mi_divide_up(s,d)*d < s+d`. +static inline uintptr_t _mi_divide_up(uintptr_t size, size_t divider) { + mi_assert_internal(divider != 0); + return (divider == 0 ? size : ((size + divider - 1) / divider)); +} + +// Is memory zero initialized? +static inline bool mi_mem_is_zero(void* p, size_t size) { + for (size_t i = 0; i < size; i++) { + if (((uint8_t*)p)[i] != 0) return false; + } + return true; +} + + +// Align a byte size to a size in _machine words_, +// i.e. byte size == `wsize*sizeof(void*)`. +static inline size_t _mi_wsize_from_size(size_t size) { + mi_assert_internal(size <= SIZE_MAX - sizeof(uintptr_t)); + return (size + sizeof(uintptr_t) - 1) / sizeof(uintptr_t); +} + +// Overflow detecting multiply +#if __has_builtin(__builtin_umul_overflow) || (defined(__GNUC__) && (__GNUC__ >= 5)) +#include // UINT_MAX, ULONG_MAX +#if defined(_CLOCK_T) // for Illumos +#undef _CLOCK_T +#endif +static inline bool mi_mul_overflow(size_t count, size_t size, size_t* total) { + #if (SIZE_MAX == ULONG_MAX) + return __builtin_umull_overflow(count, size, (unsigned long *)total); + #elif (SIZE_MAX == UINT_MAX) + return __builtin_umul_overflow(count, size, (unsigned int *)total); + #else + return __builtin_umulll_overflow(count, size, (unsigned long long *)total); + #endif +} +#else /* __builtin_umul_overflow is unavailable */ +static inline bool mi_mul_overflow(size_t count, size_t size, size_t* total) { + #define MI_MUL_NO_OVERFLOW ((size_t)1 << (4*sizeof(size_t))) // sqrt(SIZE_MAX) + *total = count * size; + return ((size >= MI_MUL_NO_OVERFLOW || count >= MI_MUL_NO_OVERFLOW) + && size > 0 && (SIZE_MAX / size) < count); +} +#endif + +// Safe multiply `count*size` into `total`; return `true` on overflow. +static inline bool mi_count_size_overflow(size_t count, size_t size, size_t* total) { + if (count==1) { // quick check for the case where count is one (common for C++ allocators) + *total = size; + return false; + } + else if (mi_unlikely(mi_mul_overflow(count, size, total))) { + _mi_error_message(EOVERFLOW, "allocation request is too large (%zu * %zu bytes)\n", count, size); + *total = SIZE_MAX; + return true; + } + else return false; +} + + +/* ---------------------------------------------------------------------------------------- +The thread local default heap: `_mi_get_default_heap` returns the thread local heap. +On most platforms (Windows, Linux, FreeBSD, NetBSD, etc), this just returns a +__thread local variable (`_mi_heap_default`). With the initial-exec TLS model this ensures +that the storage will always be available (allocated on the thread stacks). +On some platforms though we cannot use that when overriding `malloc` since the underlying +TLS implementation (or the loader) will call itself `malloc` on a first access and recurse. +We try to circumvent this in an efficient way: +- macOSX : we use an unused TLS slot from the OS allocated slots (MI_TLS_SLOT). On OSX, the + loader itself calls `malloc` even before the modules are initialized. +- OpenBSD: we use an unused slot from the pthread block (MI_TLS_PTHREAD_SLOT_OFS). +- DragonFly: defaults are working but seem slow compared to freeBSD (see PR #323) +------------------------------------------------------------------------------------------- */ + +extern const mi_heap_t _mi_heap_empty; // read-only empty heap, initial value of the thread local default heap +extern bool _mi_process_is_initialized; +mi_heap_t* _mi_heap_main_get(void); // statically allocated main backing heap + +#if defined(MI_MALLOC_OVERRIDE) +#if defined(__APPLE__) // macOS +#define MI_TLS_SLOT 89 // seems unused? +// #define MI_TLS_RECURSE_GUARD 1 +// other possible unused ones are 9, 29, __PTK_FRAMEWORK_JAVASCRIPTCORE_KEY4 (94), __PTK_FRAMEWORK_GC_KEY9 (112) and __PTK_FRAMEWORK_OLDGC_KEY9 (89) +// see +#elif defined(__OpenBSD__) +// use end bytes of a name; goes wrong if anyone uses names > 23 characters (ptrhread specifies 16) +// see +#define MI_TLS_PTHREAD_SLOT_OFS (6*sizeof(int) + 4*sizeof(void*) + 24) +// #elif defined(__DragonFly__) +// #warning "mimalloc is not working correctly on DragonFly yet." +// #define MI_TLS_PTHREAD_SLOT_OFS (4 + 1*sizeof(void*)) // offset `uniqueid` (also used by gdb?) +#elif defined(__ANDROID__) +// See issue #381 +#define MI_TLS_PTHREAD +#endif +#endif + +#if defined(MI_TLS_SLOT) +static inline void* mi_tls_slot(size_t slot) mi_attr_noexcept; // forward declaration +#elif defined(MI_TLS_PTHREAD_SLOT_OFS) +static inline mi_heap_t** mi_tls_pthread_heap_slot(void) { + pthread_t self = pthread_self(); + #if defined(__DragonFly__) + if (self==NULL) { + mi_heap_t* pheap_main = _mi_heap_main_get(); + return &pheap_main; + } + #endif + return (mi_heap_t**)((uint8_t*)self + MI_TLS_PTHREAD_SLOT_OFS); +} +#elif defined(MI_TLS_PTHREAD) +extern pthread_key_t _mi_heap_default_key; +#endif + +// Default heap to allocate from (if not using TLS- or pthread slots). +// Do not use this directly but use through `mi_heap_get_default()` (or the unchecked `mi_get_default_heap`). +// This thread local variable is only used when neither MI_TLS_SLOT, MI_TLS_PTHREAD, or MI_TLS_PTHREAD_SLOT_OFS are defined. +// However, on the Apple M1 we do use the address of this variable as the unique thread-id (issue #356). +extern mi_decl_thread mi_heap_t* _mi_heap_default; // default heap to allocate from + +static inline mi_heap_t* mi_get_default_heap(void) { +#if defined(MI_TLS_SLOT) + mi_heap_t* heap = (mi_heap_t*)mi_tls_slot(MI_TLS_SLOT); + if (mi_unlikely(heap == NULL)) { + #ifdef __GNUC__ + __asm(""); // prevent conditional load of the address of _mi_heap_empty + #endif + heap = (mi_heap_t*)&_mi_heap_empty; + } + return heap; +#elif defined(MI_TLS_PTHREAD_SLOT_OFS) + mi_heap_t* heap = *mi_tls_pthread_heap_slot(); + return (mi_unlikely(heap == NULL) ? (mi_heap_t*)&_mi_heap_empty : heap); +#elif defined(MI_TLS_PTHREAD) + mi_heap_t* heap = (mi_unlikely(_mi_heap_default_key == (pthread_key_t)(-1)) ? _mi_heap_main_get() : (mi_heap_t*)pthread_getspecific(_mi_heap_default_key)); + return (mi_unlikely(heap == NULL) ? (mi_heap_t*)&_mi_heap_empty : heap); +#else + #if defined(MI_TLS_RECURSE_GUARD) + if (mi_unlikely(!_mi_process_is_initialized)) return _mi_heap_main_get(); + #endif + return _mi_heap_default; +#endif +} + +static inline bool mi_heap_is_default(const mi_heap_t* heap) { + return (heap == mi_get_default_heap()); +} + +static inline bool mi_heap_is_backing(const mi_heap_t* heap) { + return (heap->tld->heap_backing == heap); +} + +static inline bool mi_heap_is_initialized(mi_heap_t* heap) { + mi_assert_internal(heap != NULL); + return (heap != &_mi_heap_empty); +} + +static inline uintptr_t _mi_ptr_cookie(const void* p) { + extern mi_heap_t _mi_heap_main; + mi_assert_internal(_mi_heap_main.cookie != 0); + return ((uintptr_t)p ^ _mi_heap_main.cookie); +} + +/* ----------------------------------------------------------- + Pages +----------------------------------------------------------- */ + +static inline mi_page_t* _mi_heap_get_free_small_page(mi_heap_t* heap, size_t size) { + mi_assert_internal(size <= (MI_SMALL_SIZE_MAX + MI_PADDING_SIZE)); + const size_t idx = _mi_wsize_from_size(size); + mi_assert_internal(idx < MI_PAGES_DIRECT); + return heap->pages_free_direct[idx]; +} + +// Get the page belonging to a certain size class +static inline mi_page_t* _mi_get_free_small_page(size_t size) { + return _mi_heap_get_free_small_page(mi_get_default_heap(), size); +} + +// Segment that contains the pointer +static inline mi_segment_t* _mi_ptr_segment(const void* p) { + // mi_assert_internal(p != NULL); + return (mi_segment_t*)((uintptr_t)p & ~MI_SEGMENT_MASK); +} + +static inline mi_page_t* mi_slice_to_page(mi_slice_t* s) { + mi_assert_internal(s->slice_offset== 0 && s->slice_count > 0); + return (mi_page_t*)(s); +} + +static inline mi_slice_t* mi_page_to_slice(mi_page_t* p) { + mi_assert_internal(p->slice_offset== 0 && p->slice_count > 0); + return (mi_slice_t*)(p); +} + +// Segment belonging to a page +static inline mi_segment_t* _mi_page_segment(const mi_page_t* page) { + mi_segment_t* segment = _mi_ptr_segment(page); + mi_assert_internal(segment == NULL || ((mi_slice_t*)page >= segment->slices && (mi_slice_t*)page < segment->slices + segment->slice_entries)); + return segment; +} + +static inline mi_slice_t* mi_slice_first(const mi_slice_t* slice) { + mi_slice_t* start = (mi_slice_t*)((uint8_t*)slice - slice->slice_offset); + mi_assert_internal(start >= _mi_ptr_segment(slice)->slices); + mi_assert_internal(start->slice_offset == 0); + mi_assert_internal(start + start->slice_count > slice); + return start; +} + +// Get the page containing the pointer +static inline mi_page_t* _mi_segment_page_of(const mi_segment_t* segment, const void* p) { + ptrdiff_t diff = (uint8_t*)p - (uint8_t*)segment; + mi_assert_internal(diff >= 0 && diff < (ptrdiff_t)MI_SEGMENT_SIZE); + size_t idx = (size_t)diff >> MI_SEGMENT_SLICE_SHIFT; + mi_assert_internal(idx < segment->slice_entries); + mi_slice_t* slice0 = (mi_slice_t*)&segment->slices[idx]; + mi_slice_t* slice = mi_slice_first(slice0); // adjust to the block that holds the page data + mi_assert_internal(slice->slice_offset == 0); + mi_assert_internal(slice >= segment->slices && slice < segment->slices + segment->slice_entries); + return mi_slice_to_page(slice); +} + +// Quick page start for initialized pages +static inline uint8_t* _mi_page_start(const mi_segment_t* segment, const mi_page_t* page, size_t* page_size) { + return _mi_segment_page_start(segment, page, page_size); +} + +// Get the page containing the pointer +static inline mi_page_t* _mi_ptr_page(void* p) { + return _mi_segment_page_of(_mi_ptr_segment(p), p); +} + +// Get the block size of a page (special case for huge objects) +static inline size_t mi_page_block_size(const mi_page_t* page) { + const size_t bsize = page->xblock_size; + mi_assert_internal(bsize > 0); + if (mi_likely(bsize < MI_HUGE_BLOCK_SIZE)) { + return bsize; + } + else { + size_t psize; + _mi_segment_page_start(_mi_page_segment(page), page, &psize); + return psize; + } +} + +// Get the usable block size of a page without fixed padding. +// This may still include internal padding due to alignment and rounding up size classes. +static inline size_t mi_page_usable_block_size(const mi_page_t* page) { + return mi_page_block_size(page) - MI_PADDING_SIZE; +} + +// size of a segment +static inline size_t mi_segment_size(mi_segment_t* segment) { + return segment->segment_slices * MI_SEGMENT_SLICE_SIZE; +} + +static inline uint8_t* mi_segment_end(mi_segment_t* segment) { + return (uint8_t*)segment + mi_segment_size(segment); +} + +// Thread free access +static inline mi_block_t* mi_page_thread_free(const mi_page_t* page) { + return (mi_block_t*)(mi_atomic_load_relaxed(&((mi_page_t*)page)->xthread_free) & ~3); +} + +static inline mi_delayed_t mi_page_thread_free_flag(const mi_page_t* page) { + return (mi_delayed_t)(mi_atomic_load_relaxed(&((mi_page_t*)page)->xthread_free) & 3); +} + +// Heap access +static inline mi_heap_t* mi_page_heap(const mi_page_t* page) { + return (mi_heap_t*)(mi_atomic_load_relaxed(&((mi_page_t*)page)->xheap)); +} + +static inline void mi_page_set_heap(mi_page_t* page, mi_heap_t* heap) { + mi_assert_internal(mi_page_thread_free_flag(page) != MI_DELAYED_FREEING); + mi_atomic_store_release(&page->xheap,(uintptr_t)heap); +} + +// Thread free flag helpers +static inline mi_block_t* mi_tf_block(mi_thread_free_t tf) { + return (mi_block_t*)(tf & ~0x03); +} +static inline mi_delayed_t mi_tf_delayed(mi_thread_free_t tf) { + return (mi_delayed_t)(tf & 0x03); +} +static inline mi_thread_free_t mi_tf_make(mi_block_t* block, mi_delayed_t delayed) { + return (mi_thread_free_t)((uintptr_t)block | (uintptr_t)delayed); +} +static inline mi_thread_free_t mi_tf_set_delayed(mi_thread_free_t tf, mi_delayed_t delayed) { + return mi_tf_make(mi_tf_block(tf),delayed); +} +static inline mi_thread_free_t mi_tf_set_block(mi_thread_free_t tf, mi_block_t* block) { + return mi_tf_make(block, mi_tf_delayed(tf)); +} + +// are all blocks in a page freed? +// note: needs up-to-date used count, (as the `xthread_free` list may not be empty). see `_mi_page_collect_free`. +static inline bool mi_page_all_free(const mi_page_t* page) { + mi_assert_internal(page != NULL); + return (page->used == 0); +} + +// are there any available blocks? +static inline bool mi_page_has_any_available(const mi_page_t* page) { + mi_assert_internal(page != NULL && page->reserved > 0); + return (page->used < page->reserved || (mi_page_thread_free(page) != NULL)); +} + +// are there immediately available blocks, i.e. blocks available on the free list. +static inline bool mi_page_immediate_available(const mi_page_t* page) { + mi_assert_internal(page != NULL); + return (page->free != NULL); +} + +// is more than 7/8th of a page in use? +static inline bool mi_page_mostly_used(const mi_page_t* page) { + if (page==NULL) return true; + uint16_t frac = page->reserved / 8U; + return (page->reserved - page->used <= frac); +} + +static inline mi_page_queue_t* mi_page_queue(const mi_heap_t* heap, size_t size) { + return &((mi_heap_t*)heap)->pages[_mi_bin(size)]; +} + + + +//----------------------------------------------------------- +// Page flags +//----------------------------------------------------------- +static inline bool mi_page_is_in_full(const mi_page_t* page) { + return page->flags.x.in_full; +} + +static inline void mi_page_set_in_full(mi_page_t* page, bool in_full) { + page->flags.x.in_full = in_full; +} + +static inline bool mi_page_has_aligned(const mi_page_t* page) { + return page->flags.x.has_aligned; +} + +static inline void mi_page_set_has_aligned(mi_page_t* page, bool has_aligned) { + page->flags.x.has_aligned = has_aligned; +} + + +/* ------------------------------------------------------------------- +Encoding/Decoding the free list next pointers + +This is to protect against buffer overflow exploits where the +free list is mutated. Many hardened allocators xor the next pointer `p` +with a secret key `k1`, as `p^k1`. This prevents overwriting with known +values but might be still too weak: if the attacker can guess +the pointer `p` this can reveal `k1` (since `p^k1^p == k1`). +Moreover, if multiple blocks can be read as well, the attacker can +xor both as `(p1^k1) ^ (p2^k1) == p1^p2` which may reveal a lot +about the pointers (and subsequently `k1`). + +Instead mimalloc uses an extra key `k2` and encodes as `((p^k2)<<> (MI_INTPTR_BITS - shift)))); +} +static inline uintptr_t mi_rotr(uintptr_t x, uintptr_t shift) { + shift %= MI_INTPTR_BITS; + return (shift==0 ? x : ((x >> shift) | (x << (MI_INTPTR_BITS - shift)))); +} + +static inline void* mi_ptr_decode(const void* null, const mi_encoded_t x, const uintptr_t* keys) { + void* p = (void*)(mi_rotr(x - keys[0], keys[0]) ^ keys[1]); + return (mi_unlikely(p==null) ? NULL : p); +} + +static inline mi_encoded_t mi_ptr_encode(const void* null, const void* p, const uintptr_t* keys) { + uintptr_t x = (uintptr_t)(mi_unlikely(p==NULL) ? null : p); + return mi_rotl(x ^ keys[1], keys[0]) + keys[0]; +} + +static inline mi_block_t* mi_block_nextx( const void* null, const mi_block_t* block, const uintptr_t* keys ) { + #ifdef MI_ENCODE_FREELIST + return (mi_block_t*)mi_ptr_decode(null, block->next, keys); + #else + MI_UNUSED(keys); MI_UNUSED(null); + return (mi_block_t*)block->next; + #endif +} + +static inline void mi_block_set_nextx(const void* null, mi_block_t* block, const mi_block_t* next, const uintptr_t* keys) { + #ifdef MI_ENCODE_FREELIST + block->next = mi_ptr_encode(null, next, keys); + #else + MI_UNUSED(keys); MI_UNUSED(null); + block->next = (mi_encoded_t)next; + #endif +} + +static inline mi_block_t* mi_block_next(const mi_page_t* page, const mi_block_t* block) { + #ifdef MI_ENCODE_FREELIST + mi_block_t* next = mi_block_nextx(page,block,page->keys); + // check for free list corruption: is `next` at least in the same page? + // TODO: check if `next` is `page->block_size` aligned? + if (mi_unlikely(next!=NULL && !mi_is_in_same_page(block, next))) { + _mi_error_message(EFAULT, "corrupted free list entry of size %zub at %p: value 0x%zx\n", mi_page_block_size(page), block, (uintptr_t)next); + next = NULL; + } + return next; + #else + MI_UNUSED(page); + return mi_block_nextx(page,block,NULL); + #endif +} + +static inline void mi_block_set_next(const mi_page_t* page, mi_block_t* block, const mi_block_t* next) { + #ifdef MI_ENCODE_FREELIST + mi_block_set_nextx(page,block,next, page->keys); + #else + MI_UNUSED(page); + mi_block_set_nextx(page,block,next,NULL); + #endif +} + + +// ------------------------------------------------------------------- +// commit mask +// ------------------------------------------------------------------- + +static inline void mi_commit_mask_create_empty(mi_commit_mask_t* cm) { + for (size_t i = 0; i < MI_COMMIT_MASK_FIELD_COUNT; i++) { + cm->mask[i] = 0; + } +} + +static inline void mi_commit_mask_create_full(mi_commit_mask_t* cm) { + for (size_t i = 0; i < MI_COMMIT_MASK_FIELD_COUNT; i++) { + cm->mask[i] = ~((size_t)0); + } +} + +static inline bool mi_commit_mask_is_empty(const mi_commit_mask_t* cm) { + for (size_t i = 0; i < MI_COMMIT_MASK_FIELD_COUNT; i++) { + if (cm->mask[i] != 0) return false; + } + return true; +} + +static inline bool mi_commit_mask_is_full(const mi_commit_mask_t* cm) { + for (size_t i = 0; i < MI_COMMIT_MASK_FIELD_COUNT; i++) { + if (cm->mask[i] != ~((size_t)0)) return false; + } + return true; +} + +// defined in `segment.c`: +size_t _mi_commit_mask_committed_size(const mi_commit_mask_t* cm, size_t total); +size_t _mi_commit_mask_next_run(const mi_commit_mask_t* cm, size_t* idx); + +#define mi_commit_mask_foreach(cm,idx,count) \ + idx = 0; \ + while ((count = _mi_commit_mask_next_run(cm,&idx)) > 0) { + +#define mi_commit_mask_foreach_end() \ + idx += count; \ + } + + + + +// ------------------------------------------------------------------- +// Fast "random" shuffle +// ------------------------------------------------------------------- + +static inline uintptr_t _mi_random_shuffle(uintptr_t x) { + if (x==0) { x = 17; } // ensure we don't get stuck in generating zeros +#if (MI_INTPTR_SIZE==8) + // by Sebastiano Vigna, see: + x ^= x >> 30; + x *= 0xbf58476d1ce4e5b9UL; + x ^= x >> 27; + x *= 0x94d049bb133111ebUL; + x ^= x >> 31; +#elif (MI_INTPTR_SIZE==4) + // by Chris Wellons, see: + x ^= x >> 16; + x *= 0x7feb352dUL; + x ^= x >> 15; + x *= 0x846ca68bUL; + x ^= x >> 16; +#endif + return x; +} + +// ------------------------------------------------------------------- +// Optimize numa node access for the common case (= one node) +// ------------------------------------------------------------------- + +int _mi_os_numa_node_get(mi_os_tld_t* tld); +size_t _mi_os_numa_node_count_get(void); + +extern _Atomic(size_t) _mi_numa_node_count; +static inline int _mi_os_numa_node(mi_os_tld_t* tld) { + if (mi_likely(mi_atomic_load_relaxed(&_mi_numa_node_count) == 1)) return 0; + else return _mi_os_numa_node_get(tld); +} +static inline size_t _mi_os_numa_node_count(void) { + const size_t count = mi_atomic_load_relaxed(&_mi_numa_node_count); + if (mi_likely(count>0)) return count; + else return _mi_os_numa_node_count_get(); +} + + +// ------------------------------------------------------------------- +// Getting the thread id should be performant as it is called in the +// fast path of `_mi_free` and we specialize for various platforms. +// We only require _mi_threadid() to return a unique id for each thread. +// ------------------------------------------------------------------- +#if defined(_WIN32) + +#define WIN32_LEAN_AND_MEAN +#include +static inline mi_threadid_t _mi_thread_id(void) mi_attr_noexcept { + // Windows: works on Intel and ARM in both 32- and 64-bit + return (uintptr_t)NtCurrentTeb(); +} + +// We use assembly for a fast thread id on the main platforms. The TLS layout depends on +// both the OS and libc implementation so we use specific tests for each main platform. +// If you test on another platform and it works please send a PR :-) +// see also https://akkadia.org/drepper/tls.pdf for more info on the TLS register. +#elif defined(__GNUC__) && ( \ + (defined(__GLIBC__) && (defined(__x86_64__) || defined(__i386__) || defined(__arm__) || defined(__aarch64__))) \ + || (defined(__APPLE__) && (defined(__x86_64__) || defined(__aarch64__))) \ + || (defined(__BIONIC__) && (defined(__x86_64__) || defined(__i386__) || defined(__arm__) || defined(__aarch64__))) \ + || (defined(__FreeBSD__) && (defined(__x86_64__) || defined(__i386__) || defined(__aarch64__))) \ + || (defined(__OpenBSD__) && (defined(__x86_64__) || defined(__i386__) || defined(__aarch64__))) \ + ) + +static inline void* mi_tls_slot(size_t slot) mi_attr_noexcept { + void* res; + const size_t ofs = (slot*sizeof(void*)); + #if defined(__i386__) + __asm__("movl %%gs:%1, %0" : "=r" (res) : "m" (*((void**)ofs)) : ); // x86 32-bit always uses GS + #elif defined(__APPLE__) && defined(__x86_64__) + __asm__("movq %%gs:%1, %0" : "=r" (res) : "m" (*((void**)ofs)) : ); // x86_64 macOSX uses GS + #elif defined(__x86_64__) && (MI_INTPTR_SIZE==4) + __asm__("movl %%fs:%1, %0" : "=r" (res) : "m" (*((void**)ofs)) : ); // x32 ABI + #elif defined(__x86_64__) + __asm__("movq %%fs:%1, %0" : "=r" (res) : "m" (*((void**)ofs)) : ); // x86_64 Linux, BSD uses FS + #elif defined(__arm__) + void** tcb; MI_UNUSED(ofs); + __asm__ volatile ("mrc p15, 0, %0, c13, c0, 3\nbic %0, %0, #3" : "=r" (tcb)); + res = tcb[slot]; + #elif defined(__aarch64__) + void** tcb; MI_UNUSED(ofs); + #if defined(__APPLE__) // M1, issue #343 + __asm__ volatile ("mrs %0, tpidrro_el0\nbic %0, %0, #7" : "=r" (tcb)); + #else + __asm__ volatile ("mrs %0, tpidr_el0" : "=r" (tcb)); + #endif + res = tcb[slot]; + #endif + return res; +} + +// setting a tls slot is only used on macOS for now +static inline void mi_tls_slot_set(size_t slot, void* value) mi_attr_noexcept { + const size_t ofs = (slot*sizeof(void*)); + #if defined(__i386__) + __asm__("movl %1,%%gs:%0" : "=m" (*((void**)ofs)) : "rn" (value) : ); // 32-bit always uses GS + #elif defined(__APPLE__) && defined(__x86_64__) + __asm__("movq %1,%%gs:%0" : "=m" (*((void**)ofs)) : "rn" (value) : ); // x86_64 macOS uses GS + #elif defined(__x86_64__) && (MI_INTPTR_SIZE==4) + __asm__("movl %1,%%fs:%0" : "=m" (*((void**)ofs)) : "rn" (value) : ); // x32 ABI + #elif defined(__x86_64__) + __asm__("movq %1,%%fs:%0" : "=m" (*((void**)ofs)) : "rn" (value) : ); // x86_64 Linux, BSD uses FS + #elif defined(__arm__) + void** tcb; MI_UNUSED(ofs); + __asm__ volatile ("mrc p15, 0, %0, c13, c0, 3\nbic %0, %0, #3" : "=r" (tcb)); + tcb[slot] = value; + #elif defined(__aarch64__) + void** tcb; MI_UNUSED(ofs); + #if defined(__APPLE__) // M1, issue #343 + __asm__ volatile ("mrs %0, tpidrro_el0\nbic %0, %0, #7" : "=r" (tcb)); + #else + __asm__ volatile ("mrs %0, tpidr_el0" : "=r" (tcb)); + #endif + tcb[slot] = value; + #endif +} + +static inline mi_threadid_t _mi_thread_id(void) mi_attr_noexcept { + #if defined(__BIONIC__) + // issue #384, #495: on the Bionic libc (Android), slot 1 is the thread id + // see: https://github.com/aosp-mirror/platform_bionic/blob/c44b1d0676ded732df4b3b21c5f798eacae93228/libc/platform/bionic/tls_defines.h#L86 + return (uintptr_t)mi_tls_slot(1); + #else + // in all our other targets, slot 0 is the thread id + // glibc: https://sourceware.org/git/?p=glibc.git;a=blob_plain;f=sysdeps/x86_64/nptl/tls.h + // apple: https://github.com/apple/darwin-xnu/blob/main/libsyscall/os/tsd.h#L36 + return (uintptr_t)mi_tls_slot(0); + #endif +} + +#else + +// otherwise use portable C, taking the address of a thread local variable (this is still very fast on most platforms). +static inline mi_threadid_t _mi_thread_id(void) mi_attr_noexcept { + return (uintptr_t)&_mi_heap_default; +} + +#endif + + +// ----------------------------------------------------------------------- +// Count bits: trailing or leading zeros (with MI_INTPTR_BITS on all zero) +// ----------------------------------------------------------------------- + +#if defined(__GNUC__) + +#include // LONG_MAX +#define MI_HAVE_FAST_BITSCAN +static inline size_t mi_clz(uintptr_t x) { + if (x==0) return MI_INTPTR_BITS; +#if (INTPTR_MAX == LONG_MAX) + return __builtin_clzl(x); +#else + return __builtin_clzll(x); +#endif +} +static inline size_t mi_ctz(uintptr_t x) { + if (x==0) return MI_INTPTR_BITS; +#if (INTPTR_MAX == LONG_MAX) + return __builtin_ctzl(x); +#else + return __builtin_ctzll(x); +#endif +} + +#elif defined(_MSC_VER) + +#include // LONG_MAX +#define MI_HAVE_FAST_BITSCAN +static inline size_t mi_clz(uintptr_t x) { + if (x==0) return MI_INTPTR_BITS; + unsigned long idx; +#if (INTPTR_MAX == LONG_MAX) + _BitScanReverse(&idx, x); +#else + _BitScanReverse64(&idx, x); +#endif + return ((MI_INTPTR_BITS - 1) - idx); +} +static inline size_t mi_ctz(uintptr_t x) { + if (x==0) return MI_INTPTR_BITS; + unsigned long idx; +#if (INTPTR_MAX == LONG_MAX) + _BitScanForward(&idx, x); +#else + _BitScanForward64(&idx, x); +#endif + return idx; +} + +#else +static inline size_t mi_ctz32(uint32_t x) { + // de Bruijn multiplication, see + static const unsigned char debruijn[32] = { + 0, 1, 28, 2, 29, 14, 24, 3, 30, 22, 20, 15, 25, 17, 4, 8, + 31, 27, 13, 23, 21, 19, 16, 7, 26, 12, 18, 6, 11, 5, 10, 9 + }; + if (x==0) return 32; + return debruijn[((x & -(int32_t)x) * 0x077CB531UL) >> 27]; +} +static inline size_t mi_clz32(uint32_t x) { + // de Bruijn multiplication, see + static const uint8_t debruijn[32] = { + 31, 22, 30, 21, 18, 10, 29, 2, 20, 17, 15, 13, 9, 6, 28, 1, + 23, 19, 11, 3, 16, 14, 7, 24, 12, 4, 8, 25, 5, 26, 27, 0 + }; + if (x==0) return 32; + x |= x >> 1; + x |= x >> 2; + x |= x >> 4; + x |= x >> 8; + x |= x >> 16; + return debruijn[(uint32_t)(x * 0x07C4ACDDUL) >> 27]; +} + +static inline size_t mi_clz(uintptr_t x) { + if (x==0) return MI_INTPTR_BITS; +#if (MI_INTPTR_BITS <= 32) + return mi_clz32((uint32_t)x); +#else + size_t count = mi_clz32((uint32_t)(x >> 32)); + if (count < 32) return count; + return (32 + mi_clz32((uint32_t)x)); +#endif +} +static inline size_t mi_ctz(uintptr_t x) { + if (x==0) return MI_INTPTR_BITS; +#if (MI_INTPTR_BITS <= 32) + return mi_ctz32((uint32_t)x); +#else + size_t count = mi_ctz32((uint32_t)x); + if (count < 32) return count; + return (32 + mi_ctz32((uint32_t)(x>>32))); +#endif +} + +#endif + +// "bit scan reverse": Return index of the highest bit (or MI_INTPTR_BITS if `x` is zero) +static inline size_t mi_bsr(uintptr_t x) { + return (x==0 ? MI_INTPTR_BITS : MI_INTPTR_BITS - 1 - mi_clz(x)); +} + + +// --------------------------------------------------------------------------------- +// Provide our own `_mi_memcpy` for potential performance optimizations. +// +// For now, only on Windows with msvc/clang-cl we optimize to `rep movsb` if +// we happen to run on x86/x64 cpu's that have "fast short rep movsb" (FSRM) support +// (AMD Zen3+ (~2020) or Intel Ice Lake+ (~2017). See also issue #201 and pr #253. +// --------------------------------------------------------------------------------- + +#if defined(_WIN32) && (defined(_M_IX86) || defined(_M_X64)) +#include +#include +extern bool _mi_cpu_has_fsrm; +static inline void _mi_memcpy(void* dst, const void* src, size_t n) { + if (_mi_cpu_has_fsrm) { + __movsb((unsigned char*)dst, (const unsigned char*)src, n); + } + else { + memcpy(dst, src, n); // todo: use noinline? + } +} +#else +#include +static inline void _mi_memcpy(void* dst, const void* src, size_t n) { + memcpy(dst, src, n); +} +#endif + + +// ------------------------------------------------------------------------------- +// The `_mi_memcpy_aligned` can be used if the pointers are machine-word aligned +// This is used for example in `mi_realloc`. +// ------------------------------------------------------------------------------- + +#if (defined(__GNUC__) && (__GNUC__ >= 4)) || defined(__clang__) +// On GCC/CLang we provide a hint that the pointers are word aligned. +#include +static inline void _mi_memcpy_aligned(void* dst, const void* src, size_t n) { + mi_assert_internal(((uintptr_t)dst % MI_INTPTR_SIZE == 0) && ((uintptr_t)src % MI_INTPTR_SIZE == 0)); + void* adst = __builtin_assume_aligned(dst, MI_INTPTR_SIZE); + const void* asrc = __builtin_assume_aligned(src, MI_INTPTR_SIZE); + _mi_memcpy(adst, asrc, n); +} +#else +// Default fallback on `_mi_memcpy` +static inline void _mi_memcpy_aligned(void* dst, const void* src, size_t n) { + mi_assert_internal(((uintptr_t)dst % MI_INTPTR_SIZE == 0) && ((uintptr_t)src % MI_INTPTR_SIZE == 0)); + _mi_memcpy(dst, src, n); +} +#endif + + +#endif diff --git a/compat/mimalloc/mimalloc-types.h b/compat/mimalloc/mimalloc-types.h new file mode 100644 index 00000000000000..a07858e22bccc0 --- /dev/null +++ b/compat/mimalloc/mimalloc-types.h @@ -0,0 +1,598 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2021, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ +#pragma once +#ifndef MIMALLOC_TYPES_H +#define MIMALLOC_TYPES_H + +#include // ptrdiff_t +#include // uintptr_t, uint16_t, etc +#include "mimalloc-atomic.h" // _Atomic + +#ifdef _MSC_VER +#pragma warning(disable:4214) // bitfield is not int +#endif + +// Minimal alignment necessary. On most platforms 16 bytes are needed +// due to SSE registers for example. This must be at least `sizeof(void*)` +#ifndef MI_MAX_ALIGN_SIZE +#define MI_MAX_ALIGN_SIZE 16 // sizeof(max_align_t) +#endif + +// ------------------------------------------------------ +// Variants +// ------------------------------------------------------ + +// Define NDEBUG in the release version to disable assertions. +// #define NDEBUG + +// Define MI_STAT as 1 to maintain statistics; set it to 2 to have detailed statistics (but costs some performance). +// #define MI_STAT 1 + +// Define MI_SECURE to enable security mitigations +// #define MI_SECURE 1 // guard page around metadata +// #define MI_SECURE 2 // guard page around each mimalloc page +// #define MI_SECURE 3 // encode free lists (detect corrupted free list (buffer overflow), and invalid pointer free) +// #define MI_SECURE 4 // checks for double free. (may be more expensive) + +#if !defined(MI_SECURE) +#define MI_SECURE 0 +#endif + +// Define MI_DEBUG for debug mode +// #define MI_DEBUG 1 // basic assertion checks and statistics, check double free, corrupted free list, and invalid pointer free. +// #define MI_DEBUG 2 // + internal assertion checks +// #define MI_DEBUG 3 // + extensive internal invariant checking (cmake -DMI_DEBUG_FULL=ON) +#if !defined(MI_DEBUG) +#if !defined(NDEBUG) || defined(_DEBUG) +#define MI_DEBUG 2 +#else +#define MI_DEBUG 0 +#endif +#endif + +// Reserve extra padding at the end of each block to be more resilient against heap block overflows. +// The padding can detect byte-precise buffer overflow on free. +#if !defined(MI_PADDING) && (MI_DEBUG>=1) +#define MI_PADDING 1 +#endif + + +// Encoded free lists allow detection of corrupted free lists +// and can detect buffer overflows, modify after free, and double `free`s. +#if (MI_SECURE>=3 || MI_DEBUG>=1 || MI_PADDING > 0) +#define MI_ENCODE_FREELIST 1 +#endif + + +// ------------------------------------------------------ +// Platform specific values +// ------------------------------------------------------ + +// ------------------------------------------------------ +// Size of a pointer. +// We assume that `sizeof(void*)==sizeof(intptr_t)` +// and it holds for all platforms we know of. +// +// However, the C standard only requires that: +// p == (void*)((intptr_t)p)) +// but we also need: +// i == (intptr_t)((void*)i) +// or otherwise one might define an intptr_t type that is larger than a pointer... +// ------------------------------------------------------ + +#if INTPTR_MAX > INT64_MAX +# define MI_INTPTR_SHIFT (4) // assume 128-bit (as on arm CHERI for example) +#elif INTPTR_MAX == INT64_MAX +# define MI_INTPTR_SHIFT (3) +#elif INTPTR_MAX == INT32_MAX +# define MI_INTPTR_SHIFT (2) +#else +#error platform pointers must be 32, 64, or 128 bits +#endif + +#if SIZE_MAX == UINT64_MAX +# define MI_SIZE_SHIFT (3) +typedef int64_t mi_ssize_t; +#elif SIZE_MAX == UINT32_MAX +# define MI_SIZE_SHIFT (2) +typedef int32_t mi_ssize_t; +#else +#error platform objects must be 32 or 64 bits +#endif + +#if (SIZE_MAX/2) > LONG_MAX +# define MI_ZU(x) x##ULL +# define MI_ZI(x) x##LL +#else +# define MI_ZU(x) x##UL +# define MI_ZI(x) x##L +#endif + +#define MI_INTPTR_SIZE (1< 4 +#define MI_SEGMENT_SHIFT (10 + MI_SEGMENT_SLICE_SHIFT) // 64MiB +#else +#define MI_SEGMENT_SHIFT ( 7 + MI_SEGMENT_SLICE_SHIFT) // 4MiB on 32-bit +#endif + +#define MI_SMALL_PAGE_SHIFT (MI_SEGMENT_SLICE_SHIFT) // 64KiB +#define MI_MEDIUM_PAGE_SHIFT ( 3 + MI_SMALL_PAGE_SHIFT) // 512KiB + + +// Derived constants +#define MI_SEGMENT_SIZE (MI_ZU(1)<= 655360) +#error "mimalloc internal: define more bins" +#endif +#if (MI_ALIGNMENT_MAX > MI_SEGMENT_SIZE/2) +#error "mimalloc internal: the max aligned boundary is too large for the segment size" +#endif +#if (MI_ALIGNED_MAX % MI_SEGMENT_SLICE_SIZE != 0) +#error "mimalloc internal: the max aligned boundary must be an integral multiple of the segment slice size" +#endif + +// Maximum slice offset (15) +#define MI_MAX_SLICE_OFFSET ((MI_ALIGNMENT_MAX / MI_SEGMENT_SLICE_SIZE) - 1) + +// Used as a special value to encode block sizes in 32 bits. +#define MI_HUGE_BLOCK_SIZE ((uint32_t)(2*MI_GiB)) + +// blocks up to this size are always allocated aligned +#define MI_MAX_ALIGN_GUARANTEE (8*MI_MAX_ALIGN_SIZE) + + + + +// ------------------------------------------------------ +// Mimalloc pages contain allocated blocks +// ------------------------------------------------------ + +// The free lists use encoded next fields +// (Only actually encodes when MI_ENCODED_FREELIST is defined.) +typedef uintptr_t mi_encoded_t; + +// thread id's +typedef size_t mi_threadid_t; + +// free lists contain blocks +typedef struct mi_block_s { + mi_encoded_t next; +} mi_block_t; + + +// The delayed flags are used for efficient multi-threaded free-ing +typedef enum mi_delayed_e { + MI_USE_DELAYED_FREE = 0, // push on the owning heap thread delayed list + MI_DELAYED_FREEING = 1, // temporary: another thread is accessing the owning heap + MI_NO_DELAYED_FREE = 2, // optimize: push on page local thread free queue if another block is already in the heap thread delayed free list + MI_NEVER_DELAYED_FREE = 3 // sticky, only resets on page reclaim +} mi_delayed_t; + + +// The `in_full` and `has_aligned` page flags are put in a union to efficiently +// test if both are false (`full_aligned == 0`) in the `mi_free` routine. +#if !MI_TSAN +typedef union mi_page_flags_s { + uint8_t full_aligned; + struct { + uint8_t in_full : 1; + uint8_t has_aligned : 1; + } x; +} mi_page_flags_t; +#else +// under thread sanitizer, use a byte for each flag to suppress warning, issue #130 +typedef union mi_page_flags_s { + uint16_t full_aligned; + struct { + uint8_t in_full; + uint8_t has_aligned; + } x; +} mi_page_flags_t; +#endif + +// Thread free list. +// We use the bottom 2 bits of the pointer for mi_delayed_t flags +typedef uintptr_t mi_thread_free_t; + +// A page contains blocks of one specific size (`block_size`). +// Each page has three list of free blocks: +// `free` for blocks that can be allocated, +// `local_free` for freed blocks that are not yet available to `mi_malloc` +// `thread_free` for freed blocks by other threads +// The `local_free` and `thread_free` lists are migrated to the `free` list +// when it is exhausted. The separate `local_free` list is necessary to +// implement a monotonic heartbeat. The `thread_free` list is needed for +// avoiding atomic operations in the common case. +// +// +// `used - |thread_free|` == actual blocks that are in use (alive) +// `used - |thread_free| + |free| + |local_free| == capacity` +// +// We don't count `freed` (as |free|) but use `used` to reduce +// the number of memory accesses in the `mi_page_all_free` function(s). +// +// Notes: +// - Access is optimized for `mi_free` and `mi_page_alloc` (in `alloc.c`) +// - Using `uint16_t` does not seem to slow things down +// - The size is 8 words on 64-bit which helps the page index calculations +// (and 10 words on 32-bit, and encoded free lists add 2 words. Sizes 10 +// and 12 are still good for address calculation) +// - To limit the structure size, the `xblock_size` is 32-bits only; for +// blocks > MI_HUGE_BLOCK_SIZE the size is determined from the segment page size +// - `thread_free` uses the bottom bits as a delayed-free flags to optimize +// concurrent frees where only the first concurrent free adds to the owning +// heap `thread_delayed_free` list (see `alloc.c:mi_free_block_mt`). +// The invariant is that no-delayed-free is only set if there is +// at least one block that will be added, or as already been added, to +// the owning heap `thread_delayed_free` list. This guarantees that pages +// will be freed correctly even if only other threads free blocks. +typedef struct mi_page_s { + // "owned" by the segment + uint32_t slice_count; // slices in this page (0 if not a page) + uint32_t slice_offset; // distance from the actual page data slice (0 if a page) + uint8_t is_reset : 1; // `true` if the page memory was reset + uint8_t is_committed : 1; // `true` if the page virtual memory is committed + uint8_t is_zero_init : 1; // `true` if the page was zero initialized + + // layout like this to optimize access in `mi_malloc` and `mi_free` + uint16_t capacity; // number of blocks committed, must be the first field, see `segment.c:page_clear` + uint16_t reserved; // number of blocks reserved in memory + mi_page_flags_t flags; // `in_full` and `has_aligned` flags (8 bits) + uint8_t is_zero : 1; // `true` if the blocks in the free list are zero initialized + uint8_t retire_expire : 7; // expiration count for retired blocks + + mi_block_t* free; // list of available free blocks (`malloc` allocates from this list) + #ifdef MI_ENCODE_FREELIST + uintptr_t keys[2]; // two random keys to encode the free lists (see `_mi_block_next`) + #endif + uint32_t used; // number of blocks in use (including blocks in `local_free` and `thread_free`) + uint32_t xblock_size; // size available in each block (always `>0`) + + mi_block_t* local_free; // list of deferred free blocks by this thread (migrates to `free`) + _Atomic(mi_thread_free_t) xthread_free; // list of deferred free blocks freed by other threads + _Atomic(uintptr_t) xheap; + + struct mi_page_s* next; // next page owned by this thread with the same `block_size` + struct mi_page_s* prev; // previous page owned by this thread with the same `block_size` + + // 64-bit 9 words, 32-bit 12 words, (+2 for secure) + #if MI_INTPTR_SIZE==8 + uintptr_t padding[1]; + #endif +} mi_page_t; + + + +typedef enum mi_page_kind_e { + MI_PAGE_SMALL, // small blocks go into 64KiB pages inside a segment + MI_PAGE_MEDIUM, // medium blocks go into medium pages inside a segment + MI_PAGE_LARGE, // larger blocks go into a page of just one block + MI_PAGE_HUGE, // huge blocks (> 16 MiB) are put into a single page in a single segment. +} mi_page_kind_t; + +typedef enum mi_segment_kind_e { + MI_SEGMENT_NORMAL, // MI_SEGMENT_SIZE size with pages inside. + MI_SEGMENT_HUGE, // > MI_LARGE_SIZE_MAX segment with just one huge page inside. +} mi_segment_kind_t; + +// ------------------------------------------------------ +// A segment holds a commit mask where a bit is set if +// the corresponding MI_COMMIT_SIZE area is committed. +// The MI_COMMIT_SIZE must be a multiple of the slice +// size. If it is equal we have the most fine grained +// decommit (but setting it higher can be more efficient). +// The MI_MINIMAL_COMMIT_SIZE is the minimal amount that will +// be committed in one go which can be set higher than +// MI_COMMIT_SIZE for efficiency (while the decommit mask +// is still tracked in fine-grained MI_COMMIT_SIZE chunks) +// ------------------------------------------------------ + +#define MI_MINIMAL_COMMIT_SIZE (2*MI_MiB) +#define MI_COMMIT_SIZE (MI_SEGMENT_SLICE_SIZE) // 64KiB +#define MI_COMMIT_MASK_BITS (MI_SEGMENT_SIZE / MI_COMMIT_SIZE) +#define MI_COMMIT_MASK_FIELD_BITS MI_SIZE_BITS +#define MI_COMMIT_MASK_FIELD_COUNT (MI_COMMIT_MASK_BITS / MI_COMMIT_MASK_FIELD_BITS) + +#if (MI_COMMIT_MASK_BITS != (MI_COMMIT_MASK_FIELD_COUNT * MI_COMMIT_MASK_FIELD_BITS)) +#error "the segment size must be exactly divisible by the (commit size * size_t bits)" +#endif + +typedef struct mi_commit_mask_s { + size_t mask[MI_COMMIT_MASK_FIELD_COUNT]; +} mi_commit_mask_t; + +typedef mi_page_t mi_slice_t; +typedef int64_t mi_msecs_t; + + +// Segments are large allocated memory blocks (8mb on 64 bit) from +// the OS. Inside segments we allocated fixed size _pages_ that +// contain blocks. +typedef struct mi_segment_s { + size_t memid; // memory id for arena allocation + bool mem_is_pinned; // `true` if we cannot decommit/reset/protect in this memory (i.e. when allocated using large OS pages) + bool mem_is_large; // in large/huge os pages? + bool mem_is_committed; // `true` if the whole segment is eagerly committed + + bool allow_decommit; + mi_msecs_t decommit_expire; + mi_commit_mask_t decommit_mask; + mi_commit_mask_t commit_mask; + + _Atomic(struct mi_segment_s*) abandoned_next; + + // from here is zero initialized + struct mi_segment_s* next; // the list of freed segments in the cache (must be first field, see `segment.c:mi_segment_init`) + + size_t abandoned; // abandoned pages (i.e. the original owning thread stopped) (`abandoned <= used`) + size_t abandoned_visits; // count how often this segment is visited in the abandoned list (to force reclaim it it is too long) + size_t used; // count of pages in use + uintptr_t cookie; // verify addresses in debug mode: `mi_ptr_cookie(segment) == segment->cookie` + + size_t segment_slices; // for huge segments this may be different from `MI_SLICES_PER_SEGMENT` + size_t segment_info_slices; // initial slices we are using segment info and possible guard pages. + + // layout like this to optimize access in `mi_free` + mi_segment_kind_t kind; + _Atomic(mi_threadid_t) thread_id; // unique id of the thread owning this segment + size_t slice_entries; // entries in the `slices` array, at most `MI_SLICES_PER_SEGMENT` + mi_slice_t slices[MI_SLICES_PER_SEGMENT]; +} mi_segment_t; + + +// ------------------------------------------------------ +// Heaps +// Provide first-class heaps to allocate from. +// A heap just owns a set of pages for allocation and +// can only be allocate/reallocate from the thread that created it. +// Freeing blocks can be done from any thread though. +// Per thread, the segments are shared among its heaps. +// Per thread, there is always a default heap that is +// used for allocation; it is initialized to statically +// point to an empty heap to avoid initialization checks +// in the fast path. +// ------------------------------------------------------ + +// Thread local data +typedef struct mi_tld_s mi_tld_t; + +// Pages of a certain block size are held in a queue. +typedef struct mi_page_queue_s { + mi_page_t* first; + mi_page_t* last; + size_t block_size; +} mi_page_queue_t; + +#define MI_BIN_FULL (MI_BIN_HUGE+1) + +// Random context +typedef struct mi_random_cxt_s { + uint32_t input[16]; + uint32_t output[16]; + int output_available; +} mi_random_ctx_t; + + +// In debug mode there is a padding structure at the end of the blocks to check for buffer overflows +#if (MI_PADDING) +typedef struct mi_padding_s { + uint32_t canary; // encoded block value to check validity of the padding (in case of overflow) + uint32_t delta; // padding bytes before the block. (mi_usable_size(p) - delta == exact allocated bytes) +} mi_padding_t; +#define MI_PADDING_SIZE (sizeof(mi_padding_t)) +#define MI_PADDING_WSIZE ((MI_PADDING_SIZE + MI_INTPTR_SIZE - 1) / MI_INTPTR_SIZE) +#else +#define MI_PADDING_SIZE 0 +#define MI_PADDING_WSIZE 0 +#endif + +#define MI_PAGES_DIRECT (MI_SMALL_WSIZE_MAX + MI_PADDING_WSIZE + 1) + + +// A heap owns a set of pages. +struct mi_heap_s { + mi_tld_t* tld; + mi_page_t* pages_free_direct[MI_PAGES_DIRECT]; // optimize: array where every entry points a page with possibly free blocks in the corresponding queue for that size. + mi_page_queue_t pages[MI_BIN_FULL + 1]; // queue of pages for each size class (or "bin") + _Atomic(mi_block_t*) thread_delayed_free; + mi_threadid_t thread_id; // thread this heap belongs too + uintptr_t cookie; // random cookie to verify pointers (see `_mi_ptr_cookie`) + uintptr_t keys[2]; // two random keys used to encode the `thread_delayed_free` list + mi_random_ctx_t random; // random number context used for secure allocation + size_t page_count; // total number of pages in the `pages` queues. + size_t page_retired_min; // smallest retired index (retired pages are fully free, but still in the page queues) + size_t page_retired_max; // largest retired index into the `pages` array. + mi_heap_t* next; // list of heaps per thread + bool no_reclaim; // `true` if this heap should not reclaim abandoned pages +}; + + + +// ------------------------------------------------------ +// Debug +// ------------------------------------------------------ + +#if !defined(MI_DEBUG_UNINIT) +#define MI_DEBUG_UNINIT (0xD0) +#endif +#if !defined(MI_DEBUG_FREED) +#define MI_DEBUG_FREED (0xDF) +#endif +#if !defined(MI_DEBUG_PADDING) +#define MI_DEBUG_PADDING (0xDE) +#endif + +#if (MI_DEBUG) +// use our own assertion to print without memory allocation +void _mi_assert_fail(const char* assertion, const char* fname, unsigned int line, const char* func ); +#define mi_assert(expr) ((expr) ? (void)0 : _mi_assert_fail(#expr,__FILE__,__LINE__,__func__)) +#else +#define mi_assert(x) +#endif + +#if (MI_DEBUG>1) +#define mi_assert_internal mi_assert +#else +#define mi_assert_internal(x) +#endif + +#if (MI_DEBUG>2) +#define mi_assert_expensive mi_assert +#else +#define mi_assert_expensive(x) +#endif + +// ------------------------------------------------------ +// Statistics +// ------------------------------------------------------ + +#ifndef MI_STAT +#if (MI_DEBUG>0) +#define MI_STAT 2 +#else +#define MI_STAT 0 +#endif +#endif + +typedef struct mi_stat_count_s { + int64_t allocated; + int64_t freed; + int64_t peak; + int64_t current; +} mi_stat_count_t; + +typedef struct mi_stat_counter_s { + int64_t total; + int64_t count; +} mi_stat_counter_t; + +typedef struct mi_stats_s { + mi_stat_count_t segments; + mi_stat_count_t pages; + mi_stat_count_t reserved; + mi_stat_count_t committed; + mi_stat_count_t reset; + mi_stat_count_t page_committed; + mi_stat_count_t segments_abandoned; + mi_stat_count_t pages_abandoned; + mi_stat_count_t threads; + mi_stat_count_t normal; + mi_stat_count_t huge; + mi_stat_count_t large; + mi_stat_count_t malloc; + mi_stat_count_t segments_cache; + mi_stat_counter_t pages_extended; + mi_stat_counter_t mmap_calls; + mi_stat_counter_t commit_calls; + mi_stat_counter_t page_no_retire; + mi_stat_counter_t searches; + mi_stat_counter_t normal_count; + mi_stat_counter_t huge_count; + mi_stat_counter_t large_count; +#if MI_STAT>1 + mi_stat_count_t normal_bins[MI_BIN_HUGE+1]; +#endif +} mi_stats_t; + + +void _mi_stat_increase(mi_stat_count_t* stat, size_t amount); +void _mi_stat_decrease(mi_stat_count_t* stat, size_t amount); +void _mi_stat_counter_increase(mi_stat_counter_t* stat, size_t amount); + +#if (MI_STAT) +#define mi_stat_increase(stat,amount) _mi_stat_increase( &(stat), amount) +#define mi_stat_decrease(stat,amount) _mi_stat_decrease( &(stat), amount) +#define mi_stat_counter_increase(stat,amount) _mi_stat_counter_increase( &(stat), amount) +#else +#define mi_stat_increase(stat,amount) (void)0 +#define mi_stat_decrease(stat,amount) (void)0 +#define mi_stat_counter_increase(stat,amount) (void)0 +#endif + +#define mi_heap_stat_counter_increase(heap,stat,amount) mi_stat_counter_increase( (heap)->tld->stats.stat, amount) +#define mi_heap_stat_increase(heap,stat,amount) mi_stat_increase( (heap)->tld->stats.stat, amount) +#define mi_heap_stat_decrease(heap,stat,amount) mi_stat_decrease( (heap)->tld->stats.stat, amount) + +// ------------------------------------------------------ +// Thread Local data +// ------------------------------------------------------ + +// A "span" is is an available range of slices. The span queues keep +// track of slice spans of at most the given `slice_count` (but more than the previous size class). +typedef struct mi_span_queue_s { + mi_slice_t* first; + mi_slice_t* last; + size_t slice_count; +} mi_span_queue_t; + +#define MI_SEGMENT_BIN_MAX (35) // 35 == mi_segment_bin(MI_SLICES_PER_SEGMENT) + +// OS thread local data +typedef struct mi_os_tld_s { + size_t region_idx; // start point for next allocation + mi_stats_t* stats; // points to tld stats +} mi_os_tld_t; + + +// Segments thread local data +typedef struct mi_segments_tld_s { + mi_span_queue_t spans[MI_SEGMENT_BIN_MAX+1]; // free slice spans inside segments + size_t count; // current number of segments; + size_t peak_count; // peak number of segments + size_t current_size; // current size of all segments + size_t peak_size; // peak size of all segments + mi_stats_t* stats; // points to tld stats + mi_os_tld_t* os; // points to os stats +} mi_segments_tld_t; + +// Thread local data +struct mi_tld_s { + unsigned long long heartbeat; // monotonic heartbeat count + bool recurse; // true if deferred was called; used to prevent infinite recursion. + mi_heap_t* heap_backing; // backing heap of this thread (cannot be deleted) + mi_heap_t* heaps; // list of heaps in this thread (so we can abandon all when the thread terminates) + mi_segments_tld_t segments; // segment tld + mi_os_tld_t os; // os tld + mi_stats_t stats; // statistics +}; + +#endif diff --git a/compat/mimalloc/mimalloc.h b/compat/mimalloc/mimalloc.h new file mode 100644 index 00000000000000..41ccc62d73d182 --- /dev/null +++ b/compat/mimalloc/mimalloc.h @@ -0,0 +1,454 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2022, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ +#pragma once +#ifndef MIMALLOC_H +#define MIMALLOC_H + +#define MI_MALLOC_VERSION 206 // major + 2 digits minor + +// ------------------------------------------------------ +// Compiler specific attributes +// ------------------------------------------------------ + +#ifdef __cplusplus + #if (__cplusplus >= 201103L) || (_MSC_VER > 1900) // C++11 + #define mi_attr_noexcept noexcept + #else + #define mi_attr_noexcept throw() + #endif +#else + #define mi_attr_noexcept +#endif + +#if defined(__cplusplus) && (__cplusplus >= 201703) + #define mi_decl_nodiscard [[nodiscard]] +#elif (defined(__GNUC__) && (__GNUC__ >= 4)) || defined(__clang__) // includes clang, icc, and clang-cl + #define mi_decl_nodiscard __attribute__((warn_unused_result)) +#elif (_MSC_VER >= 1700) + #define mi_decl_nodiscard _Check_return_ +#else + #define mi_decl_nodiscard +#endif + +#if defined(_MSC_VER) || defined(__MINGW32__) + #if !defined(MI_SHARED_LIB) + #define mi_decl_export + #elif defined(MI_SHARED_LIB_EXPORT) + #define mi_decl_export __declspec(dllexport) + #else + #define mi_decl_export __declspec(dllimport) + #endif + #if defined(__MINGW32__) + #define mi_decl_restrict + #define mi_attr_malloc __attribute__((malloc)) + #else + #if (_MSC_VER >= 1900) && !defined(__EDG__) + #define mi_decl_restrict __declspec(allocator) __declspec(restrict) + #else + #define mi_decl_restrict __declspec(restrict) + #endif + #define mi_attr_malloc + #endif + #define mi_cdecl __cdecl + #define mi_attr_alloc_size(s) + #define mi_attr_alloc_size2(s1,s2) + #define mi_attr_alloc_align(p) +#elif defined(__GNUC__) // includes clang and icc + #if defined(MI_SHARED_LIB) && defined(MI_SHARED_LIB_EXPORT) + #define mi_decl_export __attribute__((visibility("default"))) + #else + #define mi_decl_export + #endif + #define mi_cdecl // leads to warnings... __attribute__((cdecl)) + #define mi_decl_restrict + #define mi_attr_malloc __attribute__((malloc)) + #if (defined(__clang_major__) && (__clang_major__ < 4)) || (__GNUC__ < 5) + #define mi_attr_alloc_size(s) + #define mi_attr_alloc_size2(s1,s2) + #define mi_attr_alloc_align(p) + #elif defined(__INTEL_COMPILER) + #define mi_attr_alloc_size(s) __attribute__((alloc_size(s))) + #define mi_attr_alloc_size2(s1,s2) __attribute__((alloc_size(s1,s2))) + #define mi_attr_alloc_align(p) + #else + #define mi_attr_alloc_size(s) __attribute__((alloc_size(s))) + #define mi_attr_alloc_size2(s1,s2) __attribute__((alloc_size(s1,s2))) + #define mi_attr_alloc_align(p) __attribute__((alloc_align(p))) + #endif +#else + #define mi_cdecl + #define mi_decl_export + #define mi_decl_restrict + #define mi_attr_malloc + #define mi_attr_alloc_size(s) + #define mi_attr_alloc_size2(s1,s2) + #define mi_attr_alloc_align(p) +#endif + +// ------------------------------------------------------ +// Includes +// ------------------------------------------------------ + +#include "git-compat-util.h" + +#include // bool + +#ifdef __cplusplus +extern "C" { +#endif + +// ------------------------------------------------------ +// Standard malloc interface +// ------------------------------------------------------ + +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_malloc(size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_calloc(size_t count, size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size2(1,2); +mi_decl_nodiscard mi_decl_export void* mi_realloc(void* p, size_t newsize) mi_attr_noexcept mi_attr_alloc_size(2); +mi_decl_export void* mi_expand(void* p, size_t newsize) mi_attr_noexcept mi_attr_alloc_size(2); + +mi_decl_export void mi_free(void* p) mi_attr_noexcept; +mi_decl_nodiscard mi_decl_export mi_decl_restrict char* mi_strdup(const char* s) mi_attr_noexcept mi_attr_malloc; +mi_decl_nodiscard mi_decl_export mi_decl_restrict char* mi_strndup(const char* s, size_t n) mi_attr_noexcept mi_attr_malloc; +mi_decl_nodiscard mi_decl_export mi_decl_restrict char* mi_realpath(const char* fname, char* resolved_name) mi_attr_noexcept mi_attr_malloc; + +// ------------------------------------------------------ +// Extended functionality +// ------------------------------------------------------ +#define MI_SMALL_WSIZE_MAX (128) +#define MI_SMALL_SIZE_MAX (MI_SMALL_WSIZE_MAX*sizeof(void*)) + +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_malloc_small(size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_zalloc_small(size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_zalloc(size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1); + +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_mallocn(size_t count, size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size2(1,2); +mi_decl_nodiscard mi_decl_export void* mi_reallocn(void* p, size_t count, size_t size) mi_attr_noexcept mi_attr_alloc_size2(2,3); +mi_decl_nodiscard mi_decl_export void* mi_reallocf(void* p, size_t newsize) mi_attr_noexcept mi_attr_alloc_size(2); + +mi_decl_nodiscard mi_decl_export size_t mi_usable_size(const void* p) mi_attr_noexcept; +mi_decl_nodiscard mi_decl_export size_t mi_good_size(size_t size) mi_attr_noexcept; + + +// ------------------------------------------------------ +// Internals +// ------------------------------------------------------ + +typedef void (mi_cdecl mi_deferred_free_fun)(bool force, unsigned long long heartbeat, void* arg); +mi_decl_export void mi_register_deferred_free(mi_deferred_free_fun* deferred_free, void* arg) mi_attr_noexcept; + +typedef void (mi_cdecl mi_output_fun)(const char* msg, void* arg); +mi_decl_export void mi_register_output(mi_output_fun* out, void* arg) mi_attr_noexcept; + +typedef void (mi_cdecl mi_error_fun)(int err, void* arg); +mi_decl_export void mi_register_error(mi_error_fun* fun, void* arg); + +mi_decl_export void mi_collect(bool force) mi_attr_noexcept; +mi_decl_export int mi_version(void) mi_attr_noexcept; +mi_decl_export void mi_stats_reset(void) mi_attr_noexcept; +mi_decl_export void mi_stats_merge(void) mi_attr_noexcept; +mi_decl_export void mi_stats_print(void* out) mi_attr_noexcept; // backward compatibility: `out` is ignored and should be NULL +mi_decl_export void mi_stats_print_out(mi_output_fun* out, void* arg) mi_attr_noexcept; + +mi_decl_export void mi_process_init(void) mi_attr_noexcept; +mi_decl_export void mi_thread_init(void) mi_attr_noexcept; +mi_decl_export void mi_thread_done(void) mi_attr_noexcept; +mi_decl_export void mi_thread_stats_print_out(mi_output_fun* out, void* arg) mi_attr_noexcept; + +mi_decl_export void mi_process_info(size_t* elapsed_msecs, size_t* user_msecs, size_t* system_msecs, + size_t* current_rss, size_t* peak_rss, + size_t* current_commit, size_t* peak_commit, size_t* page_faults) mi_attr_noexcept; + +// ------------------------------------------------------------------------------------- +// Aligned allocation +// Note that `alignment` always follows `size` for consistency with unaligned +// allocation, but unfortunately this differs from `posix_memalign` and `aligned_alloc`. +// ------------------------------------------------------------------------------------- +#define MI_ALIGNMENT_MAX (1024*1024UL) // maximum supported alignment is 1MiB + +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_malloc_aligned(size_t size, size_t alignment) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1) mi_attr_alloc_align(2); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_malloc_aligned_at(size_t size, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_zalloc_aligned(size_t size, size_t alignment) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1) mi_attr_alloc_align(2); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_zalloc_aligned_at(size_t size, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_calloc_aligned(size_t count, size_t size, size_t alignment) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size2(1,2) mi_attr_alloc_align(3); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_calloc_aligned_at(size_t count, size_t size, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size2(1,2); +mi_decl_nodiscard mi_decl_export void* mi_realloc_aligned(void* p, size_t newsize, size_t alignment) mi_attr_noexcept mi_attr_alloc_size(2) mi_attr_alloc_align(3); +mi_decl_nodiscard mi_decl_export void* mi_realloc_aligned_at(void* p, size_t newsize, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_alloc_size(2); + + +// ------------------------------------------------------------------------------------- +// Heaps: first-class, but can only allocate from the same thread that created it. +// ------------------------------------------------------------------------------------- + +struct mi_heap_s; +typedef struct mi_heap_s mi_heap_t; + +mi_decl_nodiscard mi_decl_export mi_heap_t* mi_heap_new(void); +mi_decl_export void mi_heap_delete(mi_heap_t* heap); +mi_decl_export void mi_heap_destroy(mi_heap_t* heap); +mi_decl_export mi_heap_t* mi_heap_set_default(mi_heap_t* heap); +mi_decl_export mi_heap_t* mi_heap_get_default(void); +mi_decl_export mi_heap_t* mi_heap_get_backing(void); +mi_decl_export void mi_heap_collect(mi_heap_t* heap, bool force) mi_attr_noexcept; + +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_malloc(mi_heap_t* heap, size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_zalloc(mi_heap_t* heap, size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_calloc(mi_heap_t* heap, size_t count, size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size2(2, 3); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_mallocn(mi_heap_t* heap, size_t count, size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size2(2, 3); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_malloc_small(mi_heap_t* heap, size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2); + +mi_decl_nodiscard mi_decl_export void* mi_heap_realloc(mi_heap_t* heap, void* p, size_t newsize) mi_attr_noexcept mi_attr_alloc_size(3); +mi_decl_nodiscard mi_decl_export void* mi_heap_reallocn(mi_heap_t* heap, void* p, size_t count, size_t size) mi_attr_noexcept mi_attr_alloc_size2(3,4); +mi_decl_nodiscard mi_decl_export void* mi_heap_reallocf(mi_heap_t* heap, void* p, size_t newsize) mi_attr_noexcept mi_attr_alloc_size(3); + +mi_decl_nodiscard mi_decl_export mi_decl_restrict char* mi_heap_strdup(mi_heap_t* heap, const char* s) mi_attr_noexcept mi_attr_malloc; +mi_decl_nodiscard mi_decl_export mi_decl_restrict char* mi_heap_strndup(mi_heap_t* heap, const char* s, size_t n) mi_attr_noexcept mi_attr_malloc; +mi_decl_nodiscard mi_decl_export mi_decl_restrict char* mi_heap_realpath(mi_heap_t* heap, const char* fname, char* resolved_name) mi_attr_noexcept mi_attr_malloc; + +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_malloc_aligned(mi_heap_t* heap, size_t size, size_t alignment) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2) mi_attr_alloc_align(3); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_malloc_aligned_at(mi_heap_t* heap, size_t size, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_zalloc_aligned(mi_heap_t* heap, size_t size, size_t alignment) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2) mi_attr_alloc_align(3); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_zalloc_aligned_at(mi_heap_t* heap, size_t size, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_calloc_aligned(mi_heap_t* heap, size_t count, size_t size, size_t alignment) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size2(2, 3) mi_attr_alloc_align(4); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_calloc_aligned_at(mi_heap_t* heap, size_t count, size_t size, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size2(2, 3); +mi_decl_nodiscard mi_decl_export void* mi_heap_realloc_aligned(mi_heap_t* heap, void* p, size_t newsize, size_t alignment) mi_attr_noexcept mi_attr_alloc_size(3) mi_attr_alloc_align(4); +mi_decl_nodiscard mi_decl_export void* mi_heap_realloc_aligned_at(mi_heap_t* heap, void* p, size_t newsize, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_alloc_size(3); + + +// -------------------------------------------------------------------------------- +// Zero initialized re-allocation. +// Only valid on memory that was originally allocated with zero initialization too. +// e.g. `mi_calloc`, `mi_zalloc`, `mi_zalloc_aligned` etc. +// see +// -------------------------------------------------------------------------------- + +mi_decl_nodiscard mi_decl_export void* mi_rezalloc(void* p, size_t newsize) mi_attr_noexcept mi_attr_alloc_size(2); +mi_decl_nodiscard mi_decl_export void* mi_recalloc(void* p, size_t newcount, size_t size) mi_attr_noexcept mi_attr_alloc_size2(2,3); + +mi_decl_nodiscard mi_decl_export void* mi_rezalloc_aligned(void* p, size_t newsize, size_t alignment) mi_attr_noexcept mi_attr_alloc_size(2) mi_attr_alloc_align(3); +mi_decl_nodiscard mi_decl_export void* mi_rezalloc_aligned_at(void* p, size_t newsize, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_alloc_size(2); +mi_decl_nodiscard mi_decl_export void* mi_recalloc_aligned(void* p, size_t newcount, size_t size, size_t alignment) mi_attr_noexcept mi_attr_alloc_size2(2,3) mi_attr_alloc_align(4); +mi_decl_nodiscard mi_decl_export void* mi_recalloc_aligned_at(void* p, size_t newcount, size_t size, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_alloc_size2(2,3); + +mi_decl_nodiscard mi_decl_export void* mi_heap_rezalloc(mi_heap_t* heap, void* p, size_t newsize) mi_attr_noexcept mi_attr_alloc_size(3); +mi_decl_nodiscard mi_decl_export void* mi_heap_recalloc(mi_heap_t* heap, void* p, size_t newcount, size_t size) mi_attr_noexcept mi_attr_alloc_size2(3,4); + +mi_decl_nodiscard mi_decl_export void* mi_heap_rezalloc_aligned(mi_heap_t* heap, void* p, size_t newsize, size_t alignment) mi_attr_noexcept mi_attr_alloc_size(3) mi_attr_alloc_align(4); +mi_decl_nodiscard mi_decl_export void* mi_heap_rezalloc_aligned_at(mi_heap_t* heap, void* p, size_t newsize, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_alloc_size(3); +mi_decl_nodiscard mi_decl_export void* mi_heap_recalloc_aligned(mi_heap_t* heap, void* p, size_t newcount, size_t size, size_t alignment) mi_attr_noexcept mi_attr_alloc_size2(3,4) mi_attr_alloc_align(5); +mi_decl_nodiscard mi_decl_export void* mi_heap_recalloc_aligned_at(mi_heap_t* heap, void* p, size_t newcount, size_t size, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_alloc_size2(3,4); + + +// ------------------------------------------------------ +// Analysis +// ------------------------------------------------------ + +mi_decl_export bool mi_heap_contains_block(mi_heap_t* heap, const void* p); +mi_decl_export bool mi_heap_check_owned(mi_heap_t* heap, const void* p); +mi_decl_export bool mi_check_owned(const void* p); + +// An area of heap space contains blocks of a single size. +typedef struct mi_heap_area_s { + void* blocks; // start of the area containing heap blocks + size_t reserved; // bytes reserved for this area (virtual) + size_t committed; // current available bytes for this area + size_t used; // number of allocated blocks + size_t block_size; // size in bytes of each block + size_t full_block_size; // size in bytes of a full block including padding and metadata. +} mi_heap_area_t; + +typedef bool (mi_cdecl mi_block_visit_fun)(const mi_heap_t* heap, const mi_heap_area_t* area, void* block, size_t block_size, void* arg); + +mi_decl_export bool mi_heap_visit_blocks(const mi_heap_t* heap, bool visit_all_blocks, mi_block_visit_fun* visitor, void* arg); + +// Experimental +mi_decl_nodiscard mi_decl_export bool mi_is_in_heap_region(const void* p) mi_attr_noexcept; +mi_decl_nodiscard mi_decl_export bool mi_is_redirected(void) mi_attr_noexcept; + +mi_decl_export int mi_reserve_huge_os_pages_interleave(size_t pages, size_t numa_nodes, size_t timeout_msecs) mi_attr_noexcept; +mi_decl_export int mi_reserve_huge_os_pages_at(size_t pages, int numa_node, size_t timeout_msecs) mi_attr_noexcept; + +mi_decl_export int mi_reserve_os_memory(size_t size, bool commit, bool allow_large) mi_attr_noexcept; +mi_decl_export bool mi_manage_os_memory(void* start, size_t size, bool is_committed, bool is_large, bool is_zero, int numa_node) mi_attr_noexcept; + +mi_decl_export void mi_debug_show_arenas(void) mi_attr_noexcept; + +// deprecated +mi_decl_export int mi_reserve_huge_os_pages(size_t pages, double max_secs, size_t* pages_reserved) mi_attr_noexcept; + + +// ------------------------------------------------------ +// Convenience +// ------------------------------------------------------ + +#define mi_malloc_tp(tp) ((tp*)mi_malloc(sizeof(tp))) +#define mi_zalloc_tp(tp) ((tp*)mi_zalloc(sizeof(tp))) +#define mi_calloc_tp(tp,n) ((tp*)mi_calloc(n,sizeof(tp))) +#define mi_mallocn_tp(tp,n) ((tp*)mi_mallocn(n,sizeof(tp))) +#define mi_reallocn_tp(p,tp,n) ((tp*)mi_reallocn(p,n,sizeof(tp))) +#define mi_recalloc_tp(p,tp,n) ((tp*)mi_recalloc(p,n,sizeof(tp))) + +#define mi_heap_malloc_tp(hp,tp) ((tp*)mi_heap_malloc(hp,sizeof(tp))) +#define mi_heap_zalloc_tp(hp,tp) ((tp*)mi_heap_zalloc(hp,sizeof(tp))) +#define mi_heap_calloc_tp(hp,tp,n) ((tp*)mi_heap_calloc(hp,n,sizeof(tp))) +#define mi_heap_mallocn_tp(hp,tp,n) ((tp*)mi_heap_mallocn(hp,n,sizeof(tp))) +#define mi_heap_reallocn_tp(hp,p,tp,n) ((tp*)mi_heap_reallocn(hp,p,n,sizeof(tp))) +#define mi_heap_recalloc_tp(hp,p,tp,n) ((tp*)mi_heap_recalloc(hp,p,n,sizeof(tp))) + + +// ------------------------------------------------------ +// Options +// ------------------------------------------------------ + +typedef enum mi_option_e { + // stable options + mi_option_show_errors, + mi_option_show_stats, + mi_option_verbose, + // some of the following options are experimental + // (deprecated options are kept for binary backward compatibility with v1.x versions) + mi_option_eager_commit, + mi_option_deprecated_eager_region_commit, + mi_option_deprecated_reset_decommits, + mi_option_large_os_pages, // use large (2MiB) OS pages, implies eager commit + mi_option_reserve_huge_os_pages, // reserve N huge OS pages (1GiB) at startup + mi_option_reserve_huge_os_pages_at, // reserve huge OS pages at a specific NUMA node + mi_option_reserve_os_memory, // reserve specified amount of OS memory at startup + mi_option_deprecated_segment_cache, + mi_option_page_reset, + mi_option_abandoned_page_decommit, + mi_option_deprecated_segment_reset, + mi_option_eager_commit_delay, + mi_option_decommit_delay, + mi_option_use_numa_nodes, // 0 = use available numa nodes, otherwise use at most N nodes. + mi_option_limit_os_alloc, // 1 = do not use OS memory for allocation (but only reserved arenas) + mi_option_os_tag, + mi_option_max_errors, + mi_option_max_warnings, + mi_option_max_segment_reclaim, + mi_option_allow_decommit, + mi_option_segment_decommit_delay, + mi_option_decommit_extend_delay, + _mi_option_last +} mi_option_t; + + +mi_decl_nodiscard mi_decl_export bool mi_option_is_enabled(mi_option_t option); +mi_decl_export void mi_option_enable(mi_option_t option); +mi_decl_export void mi_option_disable(mi_option_t option); +mi_decl_export void mi_option_set_enabled(mi_option_t option, bool enable); +mi_decl_export void mi_option_set_enabled_default(mi_option_t option, bool enable); + +mi_decl_nodiscard mi_decl_export long mi_option_get(mi_option_t option); +mi_decl_nodiscard mi_decl_export long mi_option_get_clamp(mi_option_t option, long min, long max); +mi_decl_export void mi_option_set(mi_option_t option, long value); +mi_decl_export void mi_option_set_default(mi_option_t option, long value); + + +// ------------------------------------------------------------------------------------------------------- +// "mi" prefixed implementations of various posix, Unix, Windows, and C++ allocation functions. +// (This can be convenient when providing overrides of these functions as done in `mimalloc-override.h`.) +// note: we use `mi_cfree` as "checked free" and it checks if the pointer is in our heap before free-ing. +// ------------------------------------------------------------------------------------------------------- + +mi_decl_export void mi_cfree(void* p) mi_attr_noexcept; +mi_decl_export void* mi__expand(void* p, size_t newsize) mi_attr_noexcept; +mi_decl_nodiscard mi_decl_export size_t mi_malloc_size(const void* p) mi_attr_noexcept; +mi_decl_nodiscard mi_decl_export size_t mi_malloc_good_size(size_t size) mi_attr_noexcept; +mi_decl_nodiscard mi_decl_export size_t mi_malloc_usable_size(const void *p) mi_attr_noexcept; + +mi_decl_export int mi_posix_memalign(void** p, size_t alignment, size_t size) mi_attr_noexcept; +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_memalign(size_t alignment, size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2) mi_attr_alloc_align(1); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_valloc(size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_pvalloc(size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_aligned_alloc(size_t alignment, size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2) mi_attr_alloc_align(1); + +mi_decl_nodiscard mi_decl_export void* mi_reallocarray(void* p, size_t count, size_t size) mi_attr_noexcept mi_attr_alloc_size2(2,3); +mi_decl_nodiscard mi_decl_export int mi_reallocarr(void* p, size_t count, size_t size) mi_attr_noexcept; +mi_decl_nodiscard mi_decl_export void* mi_aligned_recalloc(void* p, size_t newcount, size_t size, size_t alignment) mi_attr_noexcept; +mi_decl_nodiscard mi_decl_export void* mi_aligned_offset_recalloc(void* p, size_t newcount, size_t size, size_t alignment, size_t offset) mi_attr_noexcept; + +mi_decl_nodiscard mi_decl_export mi_decl_restrict unsigned short* mi_wcsdup(const unsigned short* s) mi_attr_noexcept mi_attr_malloc; +mi_decl_nodiscard mi_decl_export mi_decl_restrict unsigned char* mi_mbsdup(const unsigned char* s) mi_attr_noexcept mi_attr_malloc; +mi_decl_export int mi_dupenv_s(char** buf, size_t* size, const char* name) mi_attr_noexcept; +mi_decl_export int mi_wdupenv_s(unsigned short** buf, size_t* size, const unsigned short* name) mi_attr_noexcept; + +mi_decl_export void mi_free_size(void* p, size_t size) mi_attr_noexcept; +mi_decl_export void mi_free_size_aligned(void* p, size_t size, size_t alignment) mi_attr_noexcept; +mi_decl_export void mi_free_aligned(void* p, size_t alignment) mi_attr_noexcept; + +// The `mi_new` wrappers implement C++ semantics on out-of-memory instead of directly returning `NULL`. +// (and call `std::get_new_handler` and potentially raise a `std::bad_alloc` exception). +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_new(size_t size) mi_attr_malloc mi_attr_alloc_size(1); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_new_aligned(size_t size, size_t alignment) mi_attr_malloc mi_attr_alloc_size(1) mi_attr_alloc_align(2); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_new_nothrow(size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_new_aligned_nothrow(size_t size, size_t alignment) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1) mi_attr_alloc_align(2); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_new_n(size_t count, size_t size) mi_attr_malloc mi_attr_alloc_size2(1, 2); +mi_decl_nodiscard mi_decl_export void* mi_new_realloc(void* p, size_t newsize) mi_attr_alloc_size(2); +mi_decl_nodiscard mi_decl_export void* mi_new_reallocn(void* p, size_t newcount, size_t size) mi_attr_alloc_size2(2, 3); + +#ifdef __cplusplus +} +#endif + +// --------------------------------------------------------------------------------------------- +// Implement the C++ std::allocator interface for use in STL containers. +// (note: see `mimalloc-new-delete.h` for overriding the new/delete operators globally) +// --------------------------------------------------------------------------------------------- +#ifdef __cplusplus + +#include // std::size_t +#include // PTRDIFF_MAX +#if (__cplusplus >= 201103L) || (_MSC_VER > 1900) // C++11 +#include // std::true_type +#include // std::forward +#endif + +template struct mi_stl_allocator { + typedef T value_type; + typedef std::size_t size_type; + typedef std::ptrdiff_t difference_type; + typedef value_type& reference; + typedef value_type const& const_reference; + typedef value_type* pointer; + typedef value_type const* const_pointer; + template struct rebind { typedef mi_stl_allocator other; }; + + mi_stl_allocator() mi_attr_noexcept = default; + mi_stl_allocator(const mi_stl_allocator&) mi_attr_noexcept = default; + template mi_stl_allocator(const mi_stl_allocator&) mi_attr_noexcept { } + mi_stl_allocator select_on_container_copy_construction() const { return *this; } + void deallocate(T* p, size_type) { mi_free(p); } + + #if (__cplusplus >= 201703L) // C++17 + mi_decl_nodiscard T* allocate(size_type count) { return static_cast(mi_new_n(count, sizeof(T))); } + mi_decl_nodiscard T* allocate(size_type count, const void*) { return allocate(count); } + #else + mi_decl_nodiscard pointer allocate(size_type count, const void* = 0) { return static_cast(mi_new_n(count, sizeof(value_type))); } + #endif + + #if ((__cplusplus >= 201103L) || (_MSC_VER > 1900)) // C++11 + using propagate_on_container_copy_assignment = std::true_type; + using propagate_on_container_move_assignment = std::true_type; + using propagate_on_container_swap = std::true_type; + using is_always_equal = std::true_type; + template void construct(U* p, Args&& ...args) { ::new(p) U(std::forward(args)...); } + template void destroy(U* p) mi_attr_noexcept { p->~U(); } + #else + void construct(pointer p, value_type const& val) { ::new(p) value_type(val); } + void destroy(pointer p) { p->~value_type(); } + #endif + + size_type max_size() const mi_attr_noexcept { return (PTRDIFF_MAX/sizeof(value_type)); } + pointer address(reference x) const { return &x; } + const_pointer address(const_reference x) const { return &x; } +}; + +template bool operator==(const mi_stl_allocator& , const mi_stl_allocator& ) mi_attr_noexcept { return true; } +template bool operator!=(const mi_stl_allocator& , const mi_stl_allocator& ) mi_attr_noexcept { return false; } +#endif // __cplusplus + +#endif diff --git a/compat/mimalloc/options.c b/compat/mimalloc/options.c new file mode 100644 index 00000000000000..3c68bff00d4d16 --- /dev/null +++ b/compat/mimalloc/options.c @@ -0,0 +1,634 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2021, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ +#include "mimalloc.h" +#include "mimalloc-internal.h" +#include "mimalloc-atomic.h" + +#include +#include // strtol +#include // strncpy, strncat, strlen, strstr +#include // toupper +#include + +#ifdef _MSC_VER +#pragma warning(disable:4996) // strncpy, strncat +#endif + + +static long mi_max_error_count = 16; // stop outputting errors after this (use < 0 for no limit) +static long mi_max_warning_count = 16; // stop outputting warnings after this (use < 0 for no limit) + +static void mi_add_stderr_output(void); + +int mi_version(void) mi_attr_noexcept { + return MI_MALLOC_VERSION; +} + +#ifdef _WIN32 +#include +#endif + +// -------------------------------------------------------- +// Options +// These can be accessed by multiple threads and may be +// concurrently initialized, but an initializing data race +// is ok since they resolve to the same value. +// -------------------------------------------------------- +typedef enum mi_init_e { + UNINIT, // not yet initialized + DEFAULTED, // not found in the environment, use default value + INITIALIZED // found in environment or set explicitly +} mi_init_t; + +typedef struct mi_option_desc_s { + long value; // the value + mi_init_t init; // is it initialized yet? (from the environment) + mi_option_t option; // for debugging: the option index should match the option + const char* name; // option name without `mimalloc_` prefix + const char* legacy_name; // potential legacy v1.x option name +} mi_option_desc_t; + +#define MI_OPTION(opt) mi_option_##opt, #opt, NULL +#define MI_OPTION_LEGACY(opt,legacy) mi_option_##opt, #opt, #legacy + +static mi_option_desc_t options[_mi_option_last] = +{ + // stable options + #if MI_DEBUG || defined(MI_SHOW_ERRORS) + { 1, UNINIT, MI_OPTION(show_errors) }, + #else + { 0, UNINIT, MI_OPTION(show_errors) }, + #endif + { 0, UNINIT, MI_OPTION(show_stats) }, + { 0, UNINIT, MI_OPTION(verbose) }, + + // Some of the following options are experimental and not all combinations are valid. Use with care. + { 1, UNINIT, MI_OPTION(eager_commit) }, // commit per segment directly (8MiB) (but see also `eager_commit_delay`) + { 0, UNINIT, MI_OPTION(deprecated_eager_region_commit) }, + { 0, UNINIT, MI_OPTION(deprecated_reset_decommits) }, + { 0, UNINIT, MI_OPTION(large_os_pages) }, // use large OS pages, use only with eager commit to prevent fragmentation of VMA's + { 0, UNINIT, MI_OPTION(reserve_huge_os_pages) }, // per 1GiB huge pages + { -1, UNINIT, MI_OPTION(reserve_huge_os_pages_at) }, // reserve huge pages at node N + { 0, UNINIT, MI_OPTION(reserve_os_memory) }, + { 0, UNINIT, MI_OPTION(deprecated_segment_cache) }, // cache N segments per thread + { 0, UNINIT, MI_OPTION(page_reset) }, // reset page memory on free + { 0, UNINIT, MI_OPTION_LEGACY(abandoned_page_decommit, abandoned_page_reset) },// decommit free page memory when a thread terminates + { 0, UNINIT, MI_OPTION(deprecated_segment_reset) }, + #if defined(__NetBSD__) + { 0, UNINIT, MI_OPTION(eager_commit_delay) }, // the first N segments per thread are not eagerly committed + #elif defined(_WIN32) + { 4, UNINIT, MI_OPTION(eager_commit_delay) }, // the first N segments per thread are not eagerly committed (but per page in the segment on demand) + #else + { 1, UNINIT, MI_OPTION(eager_commit_delay) }, // the first N segments per thread are not eagerly committed (but per page in the segment on demand) + #endif + { 25, UNINIT, MI_OPTION_LEGACY(decommit_delay, reset_delay) }, // page decommit delay in milli-seconds + { 0, UNINIT, MI_OPTION(use_numa_nodes) }, // 0 = use available numa nodes, otherwise use at most N nodes. + { 0, UNINIT, MI_OPTION(limit_os_alloc) }, // 1 = do not use OS memory for allocation (but only reserved arenas) + { 100, UNINIT, MI_OPTION(os_tag) }, // only apple specific for now but might serve more or less related purpose + { 16, UNINIT, MI_OPTION(max_errors) }, // maximum errors that are output + { 16, UNINIT, MI_OPTION(max_warnings) }, // maximum warnings that are output + { 8, UNINIT, MI_OPTION(max_segment_reclaim)},// max. number of segment reclaims from the abandoned segments per try. + { 1, UNINIT, MI_OPTION(allow_decommit) }, // decommit slices when no longer used (after decommit_delay milli-seconds) + { 500, UNINIT, MI_OPTION(segment_decommit_delay) }, // decommit delay in milli-seconds for freed segments + { 2, UNINIT, MI_OPTION(decommit_extend_delay) } +}; + +static void mi_option_init(mi_option_desc_t* desc); + +void _mi_options_init(void) { + // called on process load; should not be called before the CRT is initialized! + // (e.g. do not call this from process_init as that may run before CRT initialization) + mi_add_stderr_output(); // now it safe to use stderr for output + for(int i = 0; i < _mi_option_last; i++ ) { + mi_option_t option = (mi_option_t)i; + long l = mi_option_get(option); MI_UNUSED(l); // initialize + if (option != mi_option_verbose) { + mi_option_desc_t* desc = &options[option]; + _mi_verbose_message("option '%s': %ld\n", desc->name, desc->value); + } + } + mi_max_error_count = mi_option_get(mi_option_max_errors); + mi_max_warning_count = mi_option_get(mi_option_max_warnings); +} + +mi_decl_nodiscard long mi_option_get(mi_option_t option) { + mi_assert(option >= 0 && option < _mi_option_last); + if (option < 0 || option >= _mi_option_last) return 0; + mi_option_desc_t* desc = &options[option]; + mi_assert(desc->option == option); // index should match the option + if (mi_unlikely(desc->init == UNINIT)) { + mi_option_init(desc); + } + return desc->value; +} + +mi_decl_nodiscard long mi_option_get_clamp(mi_option_t option, long min, long max) { + long x = mi_option_get(option); + return (x < min ? min : (x > max ? max : x)); +} + +void mi_option_set(mi_option_t option, long value) { + mi_assert(option >= 0 && option < _mi_option_last); + if (option < 0 || option >= _mi_option_last) return; + mi_option_desc_t* desc = &options[option]; + mi_assert(desc->option == option); // index should match the option + desc->value = value; + desc->init = INITIALIZED; +} + +void mi_option_set_default(mi_option_t option, long value) { + mi_assert(option >= 0 && option < _mi_option_last); + if (option < 0 || option >= _mi_option_last) return; + mi_option_desc_t* desc = &options[option]; + if (desc->init != INITIALIZED) { + desc->value = value; + } +} + +mi_decl_nodiscard bool mi_option_is_enabled(mi_option_t option) { + return (mi_option_get(option) != 0); +} + +void mi_option_set_enabled(mi_option_t option, bool enable) { + mi_option_set(option, (enable ? 1 : 0)); +} + +void mi_option_set_enabled_default(mi_option_t option, bool enable) { + mi_option_set_default(option, (enable ? 1 : 0)); +} + +void mi_option_enable(mi_option_t option) { + mi_option_set_enabled(option,true); +} + +void mi_option_disable(mi_option_t option) { + mi_option_set_enabled(option,false); +} + + +static void mi_out_stderr(const char* msg, void* arg) { + MI_UNUSED(arg); + if (msg == NULL) return; + #ifdef _WIN32 + // on windows with redirection, the C runtime cannot handle locale dependent output + // after the main thread closes so we use direct console output. + if (!_mi_preloading()) { + // _cputs(msg); // _cputs cannot be used at is aborts if it fails to lock the console + static HANDLE hcon = INVALID_HANDLE_VALUE; + static int write_to_console; + if (hcon == INVALID_HANDLE_VALUE) { + CONSOLE_SCREEN_BUFFER_INFO sbi; + hcon = GetStdHandle(STD_ERROR_HANDLE); + write_to_console = GetConsoleScreenBufferInfo(hcon, &sbi) ? 1 : 0; + } + if (!write_to_console) { + fputs(msg, stderr); + return; + } + const size_t len = strlen(msg); + if (hcon != INVALID_HANDLE_VALUE && len > 0 && len < UINT32_MAX) { + DWORD written = 0; + WriteConsoleA(hcon, msg, (DWORD)len, &written, NULL); + } + } + #else + fputs(msg, stderr); + #endif +} + +// Since an output function can be registered earliest in the `main` +// function we also buffer output that happens earlier. When +// an output function is registered it is called immediately with +// the output up to that point. +#ifndef MI_MAX_DELAY_OUTPUT +#define MI_MAX_DELAY_OUTPUT ((size_t)(32*1024)) +#endif +static char out_buf[MI_MAX_DELAY_OUTPUT+1]; +static _Atomic(size_t) out_len; + +static void mi_out_buf(const char* msg, void* arg) { + MI_UNUSED(arg); + if (msg==NULL) return; + if (mi_atomic_load_relaxed(&out_len)>=MI_MAX_DELAY_OUTPUT) return; + size_t n = strlen(msg); + if (n==0) return; + // claim space + size_t start = mi_atomic_add_acq_rel(&out_len, n); + if (start >= MI_MAX_DELAY_OUTPUT) return; + // check bound + if (start+n >= MI_MAX_DELAY_OUTPUT) { + n = MI_MAX_DELAY_OUTPUT-start-1; + } + _mi_memcpy(&out_buf[start], msg, n); +} + +static void mi_out_buf_flush(mi_output_fun* out, bool no_more_buf, void* arg) { + if (out==NULL) return; + // claim (if `no_more_buf == true`, no more output will be added after this point) + size_t count = mi_atomic_add_acq_rel(&out_len, (no_more_buf ? MI_MAX_DELAY_OUTPUT : 1)); + // and output the current contents + if (count>MI_MAX_DELAY_OUTPUT) count = MI_MAX_DELAY_OUTPUT; + out_buf[count] = 0; + out(out_buf,arg); + if (!no_more_buf) { + out_buf[count] = '\n'; // if continue with the buffer, insert a newline + } +} + + +// Once this module is loaded, switch to this routine +// which outputs to stderr and the delayed output buffer. +static void mi_out_buf_stderr(const char* msg, void* arg) { + mi_out_stderr(msg,arg); + mi_out_buf(msg,arg); +} + + + +// -------------------------------------------------------- +// Default output handler +// -------------------------------------------------------- + +// Should be atomic but gives errors on many platforms as generally we cannot cast a function pointer to a uintptr_t. +// For now, don't register output from multiple threads. +static mi_output_fun* volatile mi_out_default; // = NULL +static _Atomic(void*) mi_out_arg; // = NULL + +static mi_output_fun* mi_out_get_default(void** parg) { + if (parg != NULL) { *parg = mi_atomic_load_ptr_acquire(void,&mi_out_arg); } + mi_output_fun* out = mi_out_default; + return (out == NULL ? &mi_out_buf : out); +} + +void mi_register_output(mi_output_fun* out, void* arg) mi_attr_noexcept { + mi_out_default = (out == NULL ? &mi_out_stderr : out); // stop using the delayed output buffer + mi_atomic_store_ptr_release(void,&mi_out_arg, arg); + if (out!=NULL) mi_out_buf_flush(out,true,arg); // output all the delayed output now +} + +// add stderr to the delayed output after the module is loaded +static void mi_add_stderr_output() { + mi_assert_internal(mi_out_default == NULL); + mi_out_buf_flush(&mi_out_stderr, false, NULL); // flush current contents to stderr + mi_out_default = &mi_out_buf_stderr; // and add stderr to the delayed output +} + +// -------------------------------------------------------- +// Messages, all end up calling `_mi_fputs`. +// -------------------------------------------------------- +static _Atomic(size_t) error_count; // = 0; // when >= max_error_count stop emitting errors +static _Atomic(size_t) warning_count; // = 0; // when >= max_warning_count stop emitting warnings + +// When overriding malloc, we may recurse into mi_vfprintf if an allocation +// inside the C runtime causes another message. +// In some cases (like on macOS) the loader already allocates which +// calls into mimalloc; if we then access thread locals (like `recurse`) +// this may crash as the access may call _tlv_bootstrap that tries to +// (recursively) invoke malloc again to allocate space for the thread local +// variables on demand. This is why we use a _mi_preloading test on such +// platforms. However, C code generator may move the initial thread local address +// load before the `if` and we therefore split it out in a separate funcion. +static mi_decl_thread bool recurse = false; + +static mi_decl_noinline bool mi_recurse_enter_prim(void) { + if (recurse) return false; + recurse = true; + return true; +} + +static mi_decl_noinline void mi_recurse_exit_prim(void) { + recurse = false; +} + +static bool mi_recurse_enter(void) { + #if defined(__APPLE__) || defined(MI_TLS_RECURSE_GUARD) + if (_mi_preloading()) return true; + #endif + return mi_recurse_enter_prim(); +} + +static void mi_recurse_exit(void) { + #if defined(__APPLE__) || defined(MI_TLS_RECURSE_GUARD) + if (_mi_preloading()) return; + #endif + mi_recurse_exit_prim(); +} + +void _mi_fputs(mi_output_fun* out, void* arg, const char* prefix, const char* message) { + if (out==NULL || (FILE*)out==stdout || (FILE*)out==stderr) { // TODO: use mi_out_stderr for stderr? + if (!mi_recurse_enter()) return; + out = mi_out_get_default(&arg); + if (prefix != NULL) out(prefix, arg); + out(message, arg); + mi_recurse_exit(); + } + else { + if (prefix != NULL) out(prefix, arg); + out(message, arg); + } +} + +// Define our own limited `fprintf` that avoids memory allocation. +// We do this using `snprintf` with a limited buffer. +static void mi_vfprintf( mi_output_fun* out, void* arg, const char* prefix, const char* fmt, va_list args ) { + char buf[512]; + if (fmt==NULL) return; + if (!mi_recurse_enter()) return; + vsnprintf(buf,sizeof(buf)-1,fmt,args); + mi_recurse_exit(); + _mi_fputs(out,arg,prefix,buf); +} + +void _mi_fprintf( mi_output_fun* out, void* arg, const char* fmt, ... ) { + va_list args; + va_start(args,fmt); + mi_vfprintf(out,arg,NULL,fmt,args); + va_end(args); +} + +static void mi_vfprintf_thread(mi_output_fun* out, void* arg, const char* prefix, const char* fmt, va_list args) { + if (prefix != NULL && strlen(prefix) <= 32 && !_mi_is_main_thread()) { + char tprefix[64]; + snprintf(tprefix, sizeof(tprefix), "%sthread 0x%llx: ", prefix, (unsigned long long)_mi_thread_id()); + mi_vfprintf(out, arg, tprefix, fmt, args); + } + else { + mi_vfprintf(out, arg, prefix, fmt, args); + } +} + +void _mi_trace_message(const char* fmt, ...) { + if (mi_option_get(mi_option_verbose) <= 1) return; // only with verbose level 2 or higher + va_list args; + va_start(args, fmt); + mi_vfprintf_thread(NULL, NULL, "mimalloc: ", fmt, args); + va_end(args); +} + +void _mi_verbose_message(const char* fmt, ...) { + if (!mi_option_is_enabled(mi_option_verbose)) return; + va_list args; + va_start(args,fmt); + mi_vfprintf(NULL, NULL, "mimalloc: ", fmt, args); + va_end(args); +} + +static void mi_show_error_message(const char* fmt, va_list args) { + if (!mi_option_is_enabled(mi_option_verbose)) { + if (!mi_option_is_enabled(mi_option_show_errors)) return; + if (mi_max_error_count >= 0 && (long)mi_atomic_increment_acq_rel(&error_count) > mi_max_error_count) return; + } + mi_vfprintf_thread(NULL, NULL, "mimalloc: error: ", fmt, args); +} + +void _mi_warning_message(const char* fmt, ...) { + if (!mi_option_is_enabled(mi_option_verbose)) { + if (!mi_option_is_enabled(mi_option_show_errors)) return; + if (mi_max_warning_count >= 0 && (long)mi_atomic_increment_acq_rel(&warning_count) > mi_max_warning_count) return; + } + va_list args; + va_start(args,fmt); + mi_vfprintf_thread(NULL, NULL, "mimalloc: warning: ", fmt, args); + va_end(args); +} + + +#if MI_DEBUG +void _mi_assert_fail(const char* assertion, const char* fname, unsigned line, const char* func ) { + _mi_fprintf(NULL, NULL, "mimalloc: assertion failed: at \"%s\":%u, %s\n assertion: \"%s\"\n", fname, line, (func==NULL?"":func), assertion); + abort(); +} +#endif + +// -------------------------------------------------------- +// Errors +// -------------------------------------------------------- + +static mi_error_fun* volatile mi_error_handler; // = NULL +static _Atomic(void*) mi_error_arg; // = NULL + +static void mi_error_default(int err) { + MI_UNUSED(err); +#if (MI_DEBUG>0) + if (err==EFAULT) { + #ifdef _MSC_VER + __debugbreak(); + #endif + abort(); + } +#endif +#if (MI_SECURE>0) + if (err==EFAULT) { // abort on serious errors in secure mode (corrupted meta-data) + abort(); + } +#endif +#if defined(MI_XMALLOC) + if (err==ENOMEM || err==EOVERFLOW) { // abort on memory allocation fails in xmalloc mode + abort(); + } +#endif +} + +void mi_register_error(mi_error_fun* fun, void* arg) { + mi_error_handler = fun; // can be NULL + mi_atomic_store_ptr_release(void,&mi_error_arg, arg); +} + +void _mi_error_message(int err, const char* fmt, ...) { + // show detailed error message + va_list args; + va_start(args, fmt); + mi_show_error_message(fmt, args); + va_end(args); + // and call the error handler which may abort (or return normally) + if (mi_error_handler != NULL) { + mi_error_handler(err, mi_atomic_load_ptr_acquire(void,&mi_error_arg)); + } + else { + mi_error_default(err); + } +} + +// -------------------------------------------------------- +// Initialize options by checking the environment +// -------------------------------------------------------- + +static void mi_strlcpy(char* dest, const char* src, size_t dest_size) { + if (dest==NULL || src==NULL || dest_size == 0) return; + // copy until end of src, or when dest is (almost) full + while (*src != 0 && dest_size > 1) { + *dest++ = *src++; + dest_size--; + } + // always zero terminate + *dest = 0; +} + +static void mi_strlcat(char* dest, const char* src, size_t dest_size) { + if (dest==NULL || src==NULL || dest_size == 0) return; + // find end of string in the dest buffer + while (*dest != 0 && dest_size > 1) { + dest++; + dest_size--; + } + // and catenate + mi_strlcpy(dest, src, dest_size); +} + +#ifdef MI_NO_GETENV +static bool mi_getenv(const char* name, char* result, size_t result_size) { + MI_UNUSED(name); + MI_UNUSED(result); + MI_UNUSED(result_size); + return false; +} +#else +static inline int mi_strnicmp(const char* s, const char* t, size_t n) { + if (n==0) return 0; + for (; *s != 0 && *t != 0 && n > 0; s++, t++, n--) { + if (toupper(*s) != toupper(*t)) break; + } + return (n==0 ? 0 : *s - *t); +} +#if defined _WIN32 +// On Windows use GetEnvironmentVariable instead of getenv to work +// reliably even when this is invoked before the C runtime is initialized. +// i.e. when `_mi_preloading() == true`. +// Note: on windows, environment names are not case sensitive. +#include +static bool mi_getenv(const char* name, char* result, size_t result_size) { + result[0] = 0; + size_t len = GetEnvironmentVariableA(name, result, (DWORD)result_size); + return (len > 0 && len < result_size); +} +#elif !defined(MI_USE_ENVIRON) || (MI_USE_ENVIRON!=0) +// On Posix systemsr use `environ` to acces environment variables +// even before the C runtime is initialized. +#if defined(__APPLE__) && defined(__has_include) && __has_include() +#include +static char** mi_get_environ(void) { + return (*_NSGetEnviron()); +} +#else +extern char** environ; +static char** mi_get_environ(void) { + return environ; +} +#endif +static bool mi_getenv(const char* name, char* result, size_t result_size) { + if (name==NULL) return false; + const size_t len = strlen(name); + if (len == 0) return false; + char** env = mi_get_environ(); + if (env == NULL) return false; + // compare up to 256 entries + for (int i = 0; i < 256 && env[i] != NULL; i++) { + const char* s = env[i]; + if (mi_strnicmp(name, s, len) == 0 && s[len] == '=') { // case insensitive + // found it + mi_strlcpy(result, s + len + 1, result_size); + return true; + } + } + return false; +} +#else +// fallback: use standard C `getenv` but this cannot be used while initializing the C runtime +static bool mi_getenv(const char* name, char* result, size_t result_size) { + // cannot call getenv() when still initializing the C runtime. + if (_mi_preloading()) return false; + const char* s = getenv(name); + if (s == NULL) { + // we check the upper case name too. + char buf[64+1]; + size_t len = strlen(name); + if (len >= sizeof(buf)) len = sizeof(buf) - 1; + for (size_t i = 0; i < len; i++) { + buf[i] = toupper(name[i]); + } + buf[len] = 0; + s = getenv(buf); + } + if (s != NULL && strlen(s) < result_size) { + mi_strlcpy(result, s, result_size); + return true; + } + else { + return false; + } +} +#endif // !MI_USE_ENVIRON +#endif // !MI_NO_GETENV + +static void mi_option_init(mi_option_desc_t* desc) { + // Read option value from the environment + char s[64+1]; + char buf[64+1]; + mi_strlcpy(buf, "mimalloc_", sizeof(buf)); + mi_strlcat(buf, desc->name, sizeof(buf)); + bool found = mi_getenv(buf,s,sizeof(s)); + if (!found && desc->legacy_name != NULL) { + mi_strlcpy(buf, "mimalloc_", sizeof(buf)); + mi_strlcat(buf, desc->legacy_name, sizeof(buf)); + found = mi_getenv(buf,s,sizeof(s)); + if (found) { + _mi_warning_message("environment option \"mimalloc_%s\" is deprecated -- use \"mimalloc_%s\" instead.\n", desc->legacy_name, desc->name ); + } + } + + if (found) { + size_t len = strlen(s); + if (len >= sizeof(buf)) len = sizeof(buf) - 1; + for (size_t i = 0; i < len; i++) { + buf[i] = (char)toupper(s[i]); + } + buf[len] = 0; + if (buf[0]==0 || strstr("1;TRUE;YES;ON", buf) != NULL) { + desc->value = 1; + desc->init = INITIALIZED; + } + else if (strstr("0;FALSE;NO;OFF", buf) != NULL) { + desc->value = 0; + desc->init = INITIALIZED; + } + else { + char* end = buf; + long value = strtol(buf, &end, 10); + if (desc->option == mi_option_reserve_os_memory) { + // this option is interpreted in KiB to prevent overflow of `long` + if (*end == 'K') { end++; } + else if (*end == 'M') { value *= MI_KiB; end++; } + else if (*end == 'G') { value *= MI_MiB; end++; } + else { value = (value + MI_KiB - 1) / MI_KiB; } + if (end[0] == 'I' && end[1] == 'B') { end += 2; } + else if (*end == 'B') { end++; } + } + if (*end == 0) { + desc->value = value; + desc->init = INITIALIZED; + } + else { + // set `init` first to avoid recursion through _mi_warning_message on mimalloc_verbose. + desc->init = DEFAULTED; + if (desc->option == mi_option_verbose && desc->value == 0) { + // if the 'mimalloc_verbose' env var has a bogus value we'd never know + // (since the value defaults to 'off') so in that case briefly enable verbose + desc->value = 1; + _mi_warning_message("environment option mimalloc_%s has an invalid value.\n", desc->name ); + desc->value = 0; + } + else { + _mi_warning_message("environment option mimalloc_%s has an invalid value.\n", desc->name ); + } + } + } + mi_assert_internal(desc->init != UNINIT); + } + else if (!_mi_preloading()) { + desc->init = DEFAULTED; + } +} diff --git a/compat/mimalloc/os.c b/compat/mimalloc/os.c new file mode 100644 index 00000000000000..e3e5f84eeb6871 --- /dev/null +++ b/compat/mimalloc/os.c @@ -0,0 +1,1446 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2021, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ +#ifndef _DEFAULT_SOURCE +#define _DEFAULT_SOURCE // ensure mmap flags are defined +#endif + +#if defined(__sun) +// illumos provides new mman.h api when any of these are defined +// otherwise the old api based on caddr_t which predates the void pointers one. +// stock solaris provides only the former, chose to atomically to discard those +// flags only here rather than project wide tough. +#undef _XOPEN_SOURCE +#undef _POSIX_C_SOURCE +#endif +#include "mimalloc.h" +#include "mimalloc-internal.h" +#include "mimalloc-atomic.h" + +#include // strerror + +#ifdef _MSC_VER +#pragma warning(disable:4996) // strerror +#endif + +#if defined(__wasi__) +#define MI_USE_SBRK +#endif + +#if defined(_WIN32) +#include +#elif defined(__wasi__) +#include // sbrk +#else +#include // mmap +#include // sysconf +#if defined(__linux__) +#include +#include +#if defined(__GLIBC__) +#include // linux mmap flags +#else +#include +#endif +#endif +#if defined(__APPLE__) +#include +#if !TARGET_IOS_IPHONE && !TARGET_IOS_SIMULATOR +#include +#endif +#endif +#if defined(__FreeBSD__) || defined(__DragonFly__) +#include +#if __FreeBSD_version >= 1200000 +#include +#include +#endif +#include +#endif +#endif + +/* ----------------------------------------------------------- + Initialization. + On windows initializes support for aligned allocation and + large OS pages (if MIMALLOC_LARGE_OS_PAGES is true). +----------------------------------------------------------- */ +bool _mi_os_decommit(void* addr, size_t size, mi_stats_t* stats); +bool _mi_os_commit(void* addr, size_t size, bool* is_zero, mi_stats_t* tld_stats); + +static void* mi_align_up_ptr(void* p, size_t alignment) { + return (void*)_mi_align_up((uintptr_t)p, alignment); +} + +static void* mi_align_down_ptr(void* p, size_t alignment) { + return (void*)_mi_align_down((uintptr_t)p, alignment); +} + + +// page size (initialized properly in `os_init`) +static size_t os_page_size = 4096; + +// minimal allocation granularity +static size_t os_alloc_granularity = 4096; + +// if non-zero, use large page allocation +static size_t large_os_page_size = 0; + +// is memory overcommit allowed? +// set dynamically in _mi_os_init (and if true we use MAP_NORESERVE) +static bool os_overcommit = true; + +bool _mi_os_has_overcommit(void) { + return os_overcommit; +} + +// OS (small) page size +size_t _mi_os_page_size(void) { + return os_page_size; +} + +// if large OS pages are supported (2 or 4MiB), then return the size, otherwise return the small page size (4KiB) +size_t _mi_os_large_page_size(void) { + return (large_os_page_size != 0 ? large_os_page_size : _mi_os_page_size()); +} + +#if !defined(MI_USE_SBRK) && !defined(__wasi__) +static bool use_large_os_page(size_t size, size_t alignment) { + // if we have access, check the size and alignment requirements + if (large_os_page_size == 0 || !mi_option_is_enabled(mi_option_large_os_pages)) return false; + return ((size % large_os_page_size) == 0 && (alignment % large_os_page_size) == 0); +} +#endif + +// round to a good OS allocation size (bounded by max 12.5% waste) +size_t _mi_os_good_alloc_size(size_t size) { + size_t align_size; + if (size < 512*MI_KiB) align_size = _mi_os_page_size(); + else if (size < 2*MI_MiB) align_size = 64*MI_KiB; + else if (size < 8*MI_MiB) align_size = 256*MI_KiB; + else if (size < 32*MI_MiB) align_size = 1*MI_MiB; + else align_size = 4*MI_MiB; + if (mi_unlikely(size >= (SIZE_MAX - align_size))) return size; // possible overflow? + return _mi_align_up(size, align_size); +} + +#if defined(_WIN32) +// We use VirtualAlloc2 for aligned allocation, but it is only supported on Windows 10 and Windows Server 2016. +// So, we need to look it up dynamically to run on older systems. (use __stdcall for 32-bit compatibility) +// NtAllocateVirtualAllocEx is used for huge OS page allocation (1GiB) +// We define a minimal MEM_EXTENDED_PARAMETER ourselves in order to be able to compile with older SDK's. +typedef enum MI_MEM_EXTENDED_PARAMETER_TYPE_E { + MiMemExtendedParameterInvalidType = 0, + MiMemExtendedParameterAddressRequirements, + MiMemExtendedParameterNumaNode, + MiMemExtendedParameterPartitionHandle, + MiMemExtendedParameterUserPhysicalHandle, + MiMemExtendedParameterAttributeFlags, + MiMemExtendedParameterMax +} MI_MEM_EXTENDED_PARAMETER_TYPE; + +typedef struct DECLSPEC_ALIGN(8) MI_MEM_EXTENDED_PARAMETER_S { + struct { DWORD64 Type : 8; DWORD64 Reserved : 56; } Type; + union { DWORD64 ULong64; PVOID Pointer; SIZE_T Size; HANDLE Handle; DWORD ULong; } Arg; +} MI_MEM_EXTENDED_PARAMETER; + +typedef struct MI_MEM_ADDRESS_REQUIREMENTS_S { + PVOID LowestStartingAddress; + PVOID HighestEndingAddress; + SIZE_T Alignment; +} MI_MEM_ADDRESS_REQUIREMENTS; + +#define MI_MEM_EXTENDED_PARAMETER_NONPAGED_HUGE 0x00000010 + +#include +typedef PVOID (__stdcall *PVirtualAlloc2)(HANDLE, PVOID, SIZE_T, ULONG, ULONG, MI_MEM_EXTENDED_PARAMETER*, ULONG); +typedef NTSTATUS (__stdcall *PNtAllocateVirtualMemoryEx)(HANDLE, PVOID*, SIZE_T*, ULONG, ULONG, MI_MEM_EXTENDED_PARAMETER*, ULONG); +static PVirtualAlloc2 pVirtualAlloc2 = NULL; +static PNtAllocateVirtualMemoryEx pNtAllocateVirtualMemoryEx = NULL; + +// Similarly, GetNumaProcesorNodeEx is only supported since Windows 7 +typedef struct MI_PROCESSOR_NUMBER_S { WORD Group; BYTE Number; BYTE Reserved; } MI_PROCESSOR_NUMBER; + +typedef VOID (__stdcall *PGetCurrentProcessorNumberEx)(MI_PROCESSOR_NUMBER* ProcNumber); +typedef BOOL (__stdcall *PGetNumaProcessorNodeEx)(MI_PROCESSOR_NUMBER* Processor, PUSHORT NodeNumber); +typedef BOOL (__stdcall* PGetNumaNodeProcessorMaskEx)(USHORT Node, PGROUP_AFFINITY ProcessorMask); +typedef BOOL (__stdcall *PGetNumaProcessorNode)(UCHAR Processor, PUCHAR NodeNumber); +static PGetCurrentProcessorNumberEx pGetCurrentProcessorNumberEx = NULL; +static PGetNumaProcessorNodeEx pGetNumaProcessorNodeEx = NULL; +static PGetNumaNodeProcessorMaskEx pGetNumaNodeProcessorMaskEx = NULL; +static PGetNumaProcessorNode pGetNumaProcessorNode = NULL; + +static bool mi_win_enable_large_os_pages(void) +{ + if (large_os_page_size > 0) return true; + + // Try to see if large OS pages are supported + // To use large pages on Windows, we first need access permission + // Set "Lock pages in memory" permission in the group policy editor + // + unsigned long err = 0; + HANDLE token = NULL; + BOOL ok = OpenProcessToken(GetCurrentProcess(), TOKEN_ADJUST_PRIVILEGES | TOKEN_QUERY, &token); + if (ok) { + TOKEN_PRIVILEGES tp; + ok = LookupPrivilegeValue(NULL, TEXT("SeLockMemoryPrivilege"), &tp.Privileges[0].Luid); + if (ok) { + tp.PrivilegeCount = 1; + tp.Privileges[0].Attributes = SE_PRIVILEGE_ENABLED; + ok = AdjustTokenPrivileges(token, FALSE, &tp, 0, (PTOKEN_PRIVILEGES)NULL, 0); + if (ok) { + err = GetLastError(); + ok = (err == ERROR_SUCCESS); + if (ok) { + large_os_page_size = GetLargePageMinimum(); + } + } + } + CloseHandle(token); + } + if (!ok) { + if (err == 0) err = GetLastError(); + _mi_warning_message("cannot enable large OS page support, error %lu\n", err); + } + return (ok!=0); +} + +void _mi_os_init(void) +{ + os_overcommit = false; + // get the page size + SYSTEM_INFO si; + GetSystemInfo(&si); + if (si.dwPageSize > 0) os_page_size = si.dwPageSize; + if (si.dwAllocationGranularity > 0) os_alloc_granularity = si.dwAllocationGranularity; + // get the VirtualAlloc2 function + HINSTANCE hDll; + hDll = LoadLibrary(TEXT("kernelbase.dll")); + if (hDll != NULL) { + // use VirtualAlloc2FromApp if possible as it is available to Windows store apps + pVirtualAlloc2 = (PVirtualAlloc2)(void (*)(void))GetProcAddress(hDll, "VirtualAlloc2FromApp"); + if (pVirtualAlloc2==NULL) pVirtualAlloc2 = (PVirtualAlloc2)(void (*)(void))GetProcAddress(hDll, "VirtualAlloc2"); + FreeLibrary(hDll); + } + // NtAllocateVirtualMemoryEx is used for huge page allocation + hDll = LoadLibrary(TEXT("ntdll.dll")); + if (hDll != NULL) { + pNtAllocateVirtualMemoryEx = (PNtAllocateVirtualMemoryEx)(void (*)(void))GetProcAddress(hDll, "NtAllocateVirtualMemoryEx"); + FreeLibrary(hDll); + } + // Try to use Win7+ numa API + hDll = LoadLibrary(TEXT("kernel32.dll")); + if (hDll != NULL) { + pGetCurrentProcessorNumberEx = (PGetCurrentProcessorNumberEx)(void (*)(void))GetProcAddress(hDll, "GetCurrentProcessorNumberEx"); + pGetNumaProcessorNodeEx = (PGetNumaProcessorNodeEx)(void (*)(void))GetProcAddress(hDll, "GetNumaProcessorNodeEx"); + pGetNumaNodeProcessorMaskEx = (PGetNumaNodeProcessorMaskEx)(void (*)(void))GetProcAddress(hDll, "GetNumaNodeProcessorMaskEx"); + pGetNumaProcessorNode = (PGetNumaProcessorNode)(void (*)(void))GetProcAddress(hDll, "GetNumaProcessorNode"); + FreeLibrary(hDll); + } + if (mi_option_is_enabled(mi_option_large_os_pages) || mi_option_is_enabled(mi_option_reserve_huge_os_pages)) { + mi_win_enable_large_os_pages(); + } +} +#elif defined(__wasi__) +void _mi_os_init(void) { + os_overcommit = false; + os_page_size = 64*MI_KiB; // WebAssembly has a fixed page size: 64KiB + os_alloc_granularity = 16; +} + +#else // generic unix + +static void os_detect_overcommit(void) { +#if defined(__linux__) + int fd = open("/proc/sys/vm/overcommit_memory", O_RDONLY); + if (fd < 0) return; + char buf[32]; + ssize_t nread = read(fd, &buf, sizeof(buf)); + close(fd); + // + // 0: heuristic overcommit, 1: always overcommit, 2: never overcommit (ignore NORESERVE) + if (nread >= 1) { + os_overcommit = (buf[0] == '0' || buf[0] == '1'); + } +#elif defined(__FreeBSD__) + int val = 0; + size_t olen = sizeof(val); + if (sysctlbyname("vm.overcommit", &val, &olen, NULL, 0) == 0) { + os_overcommit = (val != 0); + } +#else + // default: overcommit is true +#endif +} + +void _mi_os_init(void) { + // get the page size + long result = sysconf(_SC_PAGESIZE); + if (result > 0) { + os_page_size = (size_t)result; + os_alloc_granularity = os_page_size; + } + large_os_page_size = 2*MI_MiB; // TODO: can we query the OS for this? + os_detect_overcommit(); +} +#endif + + +#if defined(MADV_NORMAL) +static int mi_madvise(void* addr, size_t length, int advice) { + #if defined(__sun) + return madvise((caddr_t)addr, length, advice); // Solaris needs cast (issue #520) + #else + return madvise(addr, length, advice); + #endif +} +#endif + + +/* ----------------------------------------------------------- + aligned hinting +-------------------------------------------------------------- */ + +// On 64-bit systems, we can do efficient aligned allocation by using +// the 2TiB to 30TiB area to allocate those. +#if (MI_INTPTR_SIZE >= 8) +static mi_decl_cache_align _Atomic(uintptr_t)aligned_base; + +// Return a MI_SEGMENT_SIZE aligned address that is probably available. +// If this returns NULL, the OS will determine the address but on some OS's that may not be +// properly aligned which can be more costly as it needs to be adjusted afterwards. +// For a size > 1GiB this always returns NULL in order to guarantee good ASLR randomization; +// (otherwise an initial large allocation of say 2TiB has a 50% chance to include (known) addresses +// in the middle of the 2TiB - 6TiB address range (see issue #372)) + +#define MI_HINT_BASE ((uintptr_t)2 << 40) // 2TiB start +#define MI_HINT_AREA ((uintptr_t)4 << 40) // upto 6TiB (since before win8 there is "only" 8TiB available to processes) +#define MI_HINT_MAX ((uintptr_t)30 << 40) // wrap after 30TiB (area after 32TiB is used for huge OS pages) + +static void* mi_os_get_aligned_hint(size_t try_alignment, size_t size) +{ + if (try_alignment <= 1 || try_alignment > MI_SEGMENT_SIZE) return NULL; + size = _mi_align_up(size, MI_SEGMENT_SIZE); + if (size > 1*MI_GiB) return NULL; // guarantee the chance of fixed valid address is at most 1/(MI_HINT_AREA / 1<<30) = 1/4096. + #if (MI_SECURE>0) + size += MI_SEGMENT_SIZE; // put in `MI_SEGMENT_SIZE` virtual gaps between hinted blocks; this splits VLA's but increases guarded areas. + #endif + + uintptr_t hint = mi_atomic_add_acq_rel(&aligned_base, size); + if (hint == 0 || hint > MI_HINT_MAX) { // wrap or initialize + uintptr_t init = MI_HINT_BASE; + #if (MI_SECURE>0 || MI_DEBUG==0) // security: randomize start of aligned allocations unless in debug mode + uintptr_t r = _mi_heap_random_next(mi_get_default_heap()); + init = init + ((MI_SEGMENT_SIZE * ((r>>17) & 0xFFFFF)) % MI_HINT_AREA); // (randomly 20 bits)*4MiB == 0 to 4TiB + #endif + uintptr_t expected = hint + size; + mi_atomic_cas_strong_acq_rel(&aligned_base, &expected, init); + hint = mi_atomic_add_acq_rel(&aligned_base, size); // this may still give 0 or > MI_HINT_MAX but that is ok, it is a hint after all + } + if (hint%try_alignment != 0) return NULL; + return (void*)hint; +} +#else +static void* mi_os_get_aligned_hint(size_t try_alignment, size_t size) { + MI_UNUSED(try_alignment); MI_UNUSED(size); + return NULL; +} +#endif + +/* ----------------------------------------------------------- + Free memory +-------------------------------------------------------------- */ + +static bool mi_os_mem_free(void* addr, size_t size, bool was_committed, mi_stats_t* stats) +{ + if (addr == NULL || size == 0) return true; // || _mi_os_is_huge_reserved(addr) + bool err = false; +#if defined(_WIN32) + DWORD errcode = 0; + err = (VirtualFree(addr, 0, MEM_RELEASE) == 0); + if (err) { errcode = GetLastError(); } + if (errcode == ERROR_INVALID_ADDRESS) { + // In mi_os_mem_alloc_aligned the fallback path may have returned a pointer inside + // the memory region returned by VirtualAlloc; in that case we need to free using + // the start of the region. + MEMORY_BASIC_INFORMATION info = { 0, 0 }; + VirtualQuery(addr, &info, sizeof(info)); + if (info.AllocationBase < addr && ((uint8_t*)addr - (uint8_t*)info.AllocationBase) < MI_SEGMENT_SIZE) { + errcode = 0; + err = (VirtualFree(info.AllocationBase, 0, MEM_RELEASE) == 0); + if (err) { errcode = GetLastError(); } + } + } + if (errcode != 0) { + _mi_warning_message("unable to release OS memory: error code 0x%x, addr: %p, size: %zu\n", errcode, addr, size); + } +#elif defined(MI_USE_SBRK) || defined(__wasi__) + err = false; // sbrk heap cannot be shrunk +#else + err = (munmap(addr, size) == -1); + if (err) { + _mi_warning_message("unable to release OS memory: %s, addr: %p, size: %zu\n", strerror(errno), addr, size); + } +#endif + if (was_committed) { _mi_stat_decrease(&stats->committed, size); } + _mi_stat_decrease(&stats->reserved, size); + return !err; +} + + +/* ----------------------------------------------------------- + Raw allocation on Windows (VirtualAlloc) +-------------------------------------------------------------- */ + +#ifdef _WIN32 + +#define MEM_COMMIT_RESERVE (MEM_COMMIT|MEM_RESERVE) + +static void* mi_win_virtual_allocx(void* addr, size_t size, size_t try_alignment, DWORD flags) { +#if (MI_INTPTR_SIZE >= 8) + // on 64-bit systems, try to use the virtual address area after 2TiB for 4MiB aligned allocations + if (addr == NULL) { + void* hint = mi_os_get_aligned_hint(try_alignment,size); + if (hint != NULL) { + void* p = VirtualAlloc(hint, size, flags, PAGE_READWRITE); + if (p != NULL) return p; + _mi_verbose_message("warning: unable to allocate hinted aligned OS memory (%zu bytes, error code: 0x%x, address: %p, alignment: %zu, flags: 0x%x)\n", size, GetLastError(), hint, try_alignment, flags); + // fall through on error + } + } +#endif + // on modern Windows try use VirtualAlloc2 for aligned allocation + if (try_alignment > 1 && (try_alignment % _mi_os_page_size()) == 0 && pVirtualAlloc2 != NULL) { + MI_MEM_ADDRESS_REQUIREMENTS reqs = { 0, 0, 0 }; + reqs.Alignment = try_alignment; + MI_MEM_EXTENDED_PARAMETER param = { {0, 0}, {0} }; + param.Type.Type = MiMemExtendedParameterAddressRequirements; + param.Arg.Pointer = &reqs; + void* p = (*pVirtualAlloc2)(GetCurrentProcess(), addr, size, flags, PAGE_READWRITE, ¶m, 1); + if (p != NULL) return p; + _mi_warning_message("unable to allocate aligned OS memory (%zu bytes, error code: 0x%x, address: %p, alignment: %zu, flags: 0x%x)\n", size, GetLastError(), addr, try_alignment, flags); + // fall through on error + } + // last resort + return VirtualAlloc(addr, size, flags, PAGE_READWRITE); +} + +static void* mi_win_virtual_alloc(void* addr, size_t size, size_t try_alignment, DWORD flags, bool large_only, bool allow_large, bool* is_large) { + mi_assert_internal(!(large_only && !allow_large)); + static _Atomic(size_t) large_page_try_ok; // = 0; + void* p = NULL; + // Try to allocate large OS pages (2MiB) if allowed or required. + if ((large_only || use_large_os_page(size, try_alignment)) + && allow_large && (flags&MEM_COMMIT)!=0 && (flags&MEM_RESERVE)!=0) { + size_t try_ok = mi_atomic_load_acquire(&large_page_try_ok); + if (!large_only && try_ok > 0) { + // if a large page allocation fails, it seems the calls to VirtualAlloc get very expensive. + // therefore, once a large page allocation failed, we don't try again for `large_page_try_ok` times. + mi_atomic_cas_strong_acq_rel(&large_page_try_ok, &try_ok, try_ok - 1); + } + else { + // large OS pages must always reserve and commit. + *is_large = true; + p = mi_win_virtual_allocx(addr, size, try_alignment, flags | MEM_LARGE_PAGES); + if (large_only) return p; + // fall back to non-large page allocation on error (`p == NULL`). + if (p == NULL) { + mi_atomic_store_release(&large_page_try_ok,10UL); // on error, don't try again for the next N allocations + } + } + } + // Fall back to regular page allocation + if (p == NULL) { + *is_large = ((flags&MEM_LARGE_PAGES) != 0); + p = mi_win_virtual_allocx(addr, size, try_alignment, flags); + } + if (p == NULL) { + _mi_warning_message("unable to allocate OS memory (%zu bytes, error code: 0x%x, address: %p, alignment: %zu, flags: 0x%x, large only: %d, allow large: %d)\n", size, GetLastError(), addr, try_alignment, flags, large_only, allow_large); + } + return p; +} + +/* ----------------------------------------------------------- + Raw allocation using `sbrk` or `wasm_memory_grow` +-------------------------------------------------------------- */ + +#elif defined(MI_USE_SBRK) || defined(__wasi__) +#if defined(MI_USE_SBRK) + static void* mi_memory_grow( size_t size ) { + void* p = sbrk(size); + if (p == (void*)(-1)) return NULL; + #if !defined(__wasi__) // on wasi this is always zero initialized already (?) + memset(p,0,size); + #endif + return p; + } +#elif defined(__wasi__) + static void* mi_memory_grow( size_t size ) { + size_t base = (size > 0 ? __builtin_wasm_memory_grow(0,_mi_divide_up(size, _mi_os_page_size())) + : __builtin_wasm_memory_size(0)); + if (base == SIZE_MAX) return NULL; + return (void*)(base * _mi_os_page_size()); + } +#endif + +#if defined(MI_USE_PTHREADS) +static pthread_mutex_t mi_heap_grow_mutex = PTHREAD_MUTEX_INITIALIZER; +#endif + +static void* mi_heap_grow(size_t size, size_t try_alignment) { + void* p = NULL; + if (try_alignment <= 1) { + // `sbrk` is not thread safe in general so try to protect it (we could skip this on WASM but leave it in for now) + #if defined(MI_USE_PTHREADS) + pthread_mutex_lock(&mi_heap_grow_mutex); + #endif + p = mi_memory_grow(size); + #if defined(MI_USE_PTHREADS) + pthread_mutex_unlock(&mi_heap_grow_mutex); + #endif + } + else { + void* base = NULL; + size_t alloc_size = 0; + // to allocate aligned use a lock to try to avoid thread interaction + // between getting the current size and actual allocation + // (also, `sbrk` is not thread safe in general) + #if defined(MI_USE_PTHREADS) + pthread_mutex_lock(&mi_heap_grow_mutex); + #endif + { + void* current = mi_memory_grow(0); // get current size + if (current != NULL) { + void* aligned_current = mi_align_up_ptr(current, try_alignment); // and align from there to minimize wasted space + alloc_size = _mi_align_up( ((uint8_t*)aligned_current - (uint8_t*)current) + size, _mi_os_page_size()); + base = mi_memory_grow(alloc_size); + } + } + #if defined(MI_USE_PTHREADS) + pthread_mutex_unlock(&mi_heap_grow_mutex); + #endif + if (base != NULL) { + p = mi_align_up_ptr(base, try_alignment); + if ((uint8_t*)p + size > (uint8_t*)base + alloc_size) { + // another thread used wasm_memory_grow/sbrk in-between and we do not have enough + // space after alignment. Give up (and waste the space as we cannot shrink :-( ) + // (in `mi_os_mem_alloc_aligned` this will fall back to overallocation to align) + p = NULL; + } + } + } + if (p == NULL) { + _mi_warning_message("unable to allocate sbrk/wasm_memory_grow OS memory (%zu bytes, %zu alignment)\n", size, try_alignment); + errno = ENOMEM; + return NULL; + } + mi_assert_internal( try_alignment == 0 || (uintptr_t)p % try_alignment == 0 ); + return p; +} + +/* ----------------------------------------------------------- + Raw allocation on Unix's (mmap) +-------------------------------------------------------------- */ +#else +#define MI_OS_USE_MMAP +static void* mi_unix_mmapx(void* addr, size_t size, size_t try_alignment, int protect_flags, int flags, int fd) { + MI_UNUSED(try_alignment); + #if defined(MAP_ALIGNED) // BSD + if (addr == NULL && try_alignment > 1 && (try_alignment % _mi_os_page_size()) == 0) { + size_t n = mi_bsr(try_alignment); + if (((size_t)1 << n) == try_alignment && n >= 12 && n <= 30) { // alignment is a power of 2 and 4096 <= alignment <= 1GiB + flags |= MAP_ALIGNED(n); + void* p = mmap(addr, size, protect_flags, flags | MAP_ALIGNED(n), fd, 0); + if (p!=MAP_FAILED) return p; + // fall back to regular mmap + } + } + #elif defined(MAP_ALIGN) // Solaris + if (addr == NULL && try_alignment > 1 && (try_alignment % _mi_os_page_size()) == 0) { + void* p = mmap((void*)try_alignment, size, protect_flags, flags | MAP_ALIGN, fd, 0); // addr parameter is the required alignment + if (p!=MAP_FAILED) return p; + // fall back to regular mmap + } + #endif + #if (MI_INTPTR_SIZE >= 8) && !defined(MAP_ALIGNED) + // on 64-bit systems, use the virtual address area after 2TiB for 4MiB aligned allocations + if (addr == NULL) { + void* hint = mi_os_get_aligned_hint(try_alignment, size); + if (hint != NULL) { + void* p = mmap(hint, size, protect_flags, flags, fd, 0); + if (p!=MAP_FAILED) return p; + // fall back to regular mmap + } + } + #endif + // regular mmap + void* p = mmap(addr, size, protect_flags, flags, fd, 0); + if (p!=MAP_FAILED) return p; + // failed to allocate + return NULL; +} + +static int mi_unix_mmap_fd(void) { +#if defined(VM_MAKE_TAG) + // macOS: tracking anonymous page with a specific ID. (All up to 98 are taken officially but LLVM sanitizers had taken 99) + int os_tag = (int)mi_option_get(mi_option_os_tag); + if (os_tag < 100 || os_tag > 255) os_tag = 100; + return VM_MAKE_TAG(os_tag); +#else + return -1; +#endif +} + +static void* mi_unix_mmap(void* addr, size_t size, size_t try_alignment, int protect_flags, bool large_only, bool allow_large, bool* is_large) { + void* p = NULL; + #if !defined(MAP_ANONYMOUS) + #define MAP_ANONYMOUS MAP_ANON + #endif + #if !defined(MAP_NORESERVE) + #define MAP_NORESERVE 0 + #endif + const int fd = mi_unix_mmap_fd(); + int flags = MAP_PRIVATE | MAP_ANONYMOUS; + if (_mi_os_has_overcommit()) { + flags |= MAP_NORESERVE; + } + #if defined(PROT_MAX) + protect_flags |= PROT_MAX(PROT_READ | PROT_WRITE); // BSD + #endif + // huge page allocation + if ((large_only || use_large_os_page(size, try_alignment)) && allow_large) { + static _Atomic(size_t) large_page_try_ok; // = 0; + size_t try_ok = mi_atomic_load_acquire(&large_page_try_ok); + if (!large_only && try_ok > 0) { + // If the OS is not configured for large OS pages, or the user does not have + // enough permission, the `mmap` will always fail (but it might also fail for other reasons). + // Therefore, once a large page allocation failed, we don't try again for `large_page_try_ok` times + // to avoid too many failing calls to mmap. + mi_atomic_cas_strong_acq_rel(&large_page_try_ok, &try_ok, try_ok - 1); + } + else { + int lflags = flags & ~MAP_NORESERVE; // using NORESERVE on huge pages seems to fail on Linux + int lfd = fd; + #ifdef MAP_ALIGNED_SUPER + lflags |= MAP_ALIGNED_SUPER; + #endif + #ifdef MAP_HUGETLB + lflags |= MAP_HUGETLB; + #endif + #ifdef MAP_HUGE_1GB + static bool mi_huge_pages_available = true; + if ((size % MI_GiB) == 0 && mi_huge_pages_available) { + lflags |= MAP_HUGE_1GB; + } + else + #endif + { + #ifdef MAP_HUGE_2MB + lflags |= MAP_HUGE_2MB; + #endif + } + #ifdef VM_FLAGS_SUPERPAGE_SIZE_2MB + lfd |= VM_FLAGS_SUPERPAGE_SIZE_2MB; + #endif + if (large_only || lflags != flags) { + // try large OS page allocation + *is_large = true; + p = mi_unix_mmapx(addr, size, try_alignment, protect_flags, lflags, lfd); + #ifdef MAP_HUGE_1GB + if (p == NULL && (lflags & MAP_HUGE_1GB) != 0) { + mi_huge_pages_available = false; // don't try huge 1GiB pages again + _mi_warning_message("unable to allocate huge (1GiB) page, trying large (2MiB) pages instead (error %i)\n", errno); + lflags = ((lflags & ~MAP_HUGE_1GB) | MAP_HUGE_2MB); + p = mi_unix_mmapx(addr, size, try_alignment, protect_flags, lflags, lfd); + } + #endif + if (large_only) return p; + if (p == NULL) { + mi_atomic_store_release(&large_page_try_ok, (size_t)8); // on error, don't try again for the next N allocations + } + } + } + } + // regular allocation + if (p == NULL) { + *is_large = false; + p = mi_unix_mmapx(addr, size, try_alignment, protect_flags, flags, fd); + if (p != NULL) { + #if defined(MADV_HUGEPAGE) + // Many Linux systems don't allow MAP_HUGETLB but they support instead + // transparent huge pages (THP). Generally, it is not required to call `madvise` with MADV_HUGE + // though since properly aligned allocations will already use large pages if available + // in that case -- in particular for our large regions (in `memory.c`). + // However, some systems only allow THP if called with explicit `madvise`, so + // when large OS pages are enabled for mimalloc, we call `madvise` anyways. + if (allow_large && use_large_os_page(size, try_alignment)) { + if (mi_madvise(p, size, MADV_HUGEPAGE) == 0) { + *is_large = true; // possibly + }; + } + #elif defined(__sun) + if (allow_large && use_large_os_page(size, try_alignment)) { + struct memcntl_mha cmd = {0}; + cmd.mha_pagesize = large_os_page_size; + cmd.mha_cmd = MHA_MAPSIZE_VA; + if (memcntl((caddr_t)p, size, MC_HAT_ADVISE, (caddr_t)&cmd, 0, 0) == 0) { + *is_large = true; + } + } + #endif + } + } + if (p == NULL) { + _mi_warning_message("unable to allocate OS memory (%zu bytes, error code: %i, address: %p, large only: %d, allow large: %d)\n", size, errno, addr, large_only, allow_large); + } + return p; +} +#endif + + +/* ----------------------------------------------------------- + Primitive allocation from the OS. +-------------------------------------------------------------- */ + +// Note: the `try_alignment` is just a hint and the returned pointer is not guaranteed to be aligned. +static void* mi_os_mem_alloc(size_t size, size_t try_alignment, bool commit, bool allow_large, bool* is_large, mi_stats_t* stats) { + mi_assert_internal(size > 0 && (size % _mi_os_page_size()) == 0); + if (size == 0) return NULL; + if (!commit) allow_large = false; + if (try_alignment == 0) try_alignment = 1; // avoid 0 to ensure there will be no divide by zero when aligning + + void* p = NULL; + /* + if (commit && allow_large) { + p = _mi_os_try_alloc_from_huge_reserved(size, try_alignment); + if (p != NULL) { + *is_large = true; + return p; + } + } + */ + + #if defined(_WIN32) + int flags = MEM_RESERVE; + if (commit) { flags |= MEM_COMMIT; } + p = mi_win_virtual_alloc(NULL, size, try_alignment, flags, false, allow_large, is_large); + #elif defined(MI_USE_SBRK) || defined(__wasi__) + MI_UNUSED(allow_large); + *is_large = false; + p = mi_heap_grow(size, try_alignment); + #else + int protect_flags = (commit ? (PROT_WRITE | PROT_READ) : PROT_NONE); + p = mi_unix_mmap(NULL, size, try_alignment, protect_flags, false, allow_large, is_large); + #endif + mi_stat_counter_increase(stats->mmap_calls, 1); + if (p != NULL) { + _mi_stat_increase(&stats->reserved, size); + if (commit) { _mi_stat_increase(&stats->committed, size); } + } + return p; +} + + +// Primitive aligned allocation from the OS. +// This function guarantees the allocated memory is aligned. +static void* mi_os_mem_alloc_aligned(size_t size, size_t alignment, bool commit, bool allow_large, bool* is_large, mi_stats_t* stats) { + mi_assert_internal(alignment >= _mi_os_page_size() && ((alignment & (alignment - 1)) == 0)); + mi_assert_internal(size > 0 && (size % _mi_os_page_size()) == 0); + mi_assert_internal(is_large != NULL); + if (!commit) allow_large = false; + if (!(alignment >= _mi_os_page_size() && ((alignment & (alignment - 1)) == 0))) return NULL; + size = _mi_align_up(size, _mi_os_page_size()); + + // try first with a hint (this will be aligned directly on Win 10+ or BSD) + void* p = mi_os_mem_alloc(size, alignment, commit, allow_large, is_large, stats); + if (p == NULL) return NULL; + + // if not aligned, free it, overallocate, and unmap around it + if (((uintptr_t)p % alignment != 0)) { + mi_os_mem_free(p, size, commit, stats); + _mi_warning_message("unable to allocate aligned OS memory directly, fall back to over-allocation (%zu bytes, address: %p, alignment: %zu, commit: %d)\n", size, p, alignment, commit); + if (size >= (SIZE_MAX - alignment)) return NULL; // overflow + const size_t over_size = size + alignment; + +#if _WIN32 + // over-allocate uncommitted (virtual) memory + p = mi_os_mem_alloc(over_size, 0 /*alignment*/, false /* commit? */, false /* allow_large */, is_large, stats); + if (p == NULL) return NULL; + + // set p to the aligned part in the full region + // note: this is dangerous on Windows as VirtualFree needs the actual region pointer + // but in mi_os_mem_free we handle this (hopefully exceptional) situation. + p = mi_align_up_ptr(p, alignment); + + // explicitly commit only the aligned part + if (commit) { + _mi_os_commit(p, size, NULL, stats); + } +#else + // overallocate... + p = mi_os_mem_alloc(over_size, 1, commit, false, is_large, stats); + if (p == NULL) return NULL; + // and selectively unmap parts around the over-allocated area. (noop on sbrk) + void* aligned_p = mi_align_up_ptr(p, alignment); + size_t pre_size = (uint8_t*)aligned_p - (uint8_t*)p; + size_t mid_size = _mi_align_up(size, _mi_os_page_size()); + size_t post_size = over_size - pre_size - mid_size; + mi_assert_internal(pre_size < over_size && post_size < over_size && mid_size >= size); + if (pre_size > 0) mi_os_mem_free(p, pre_size, commit, stats); + if (post_size > 0) mi_os_mem_free((uint8_t*)aligned_p + mid_size, post_size, commit, stats); + // we can return the aligned pointer on `mmap` (and sbrk) systems + p = aligned_p; +#endif + } + + mi_assert_internal(p == NULL || (p != NULL && ((uintptr_t)p % alignment) == 0)); + return p; +} + + +/* ----------------------------------------------------------- + OS API: alloc, free, alloc_aligned +----------------------------------------------------------- */ + +void* _mi_os_alloc(size_t size, mi_stats_t* tld_stats) { + MI_UNUSED(tld_stats); + mi_stats_t* stats = &_mi_stats_main; + if (size == 0) return NULL; + size = _mi_os_good_alloc_size(size); + bool is_large = false; + return mi_os_mem_alloc(size, 0, true, false, &is_large, stats); +} + +void _mi_os_free_ex(void* p, size_t size, bool was_committed, mi_stats_t* tld_stats) { + MI_UNUSED(tld_stats); + mi_stats_t* stats = &_mi_stats_main; + if (size == 0 || p == NULL) return; + size = _mi_os_good_alloc_size(size); + mi_os_mem_free(p, size, was_committed, stats); +} + +void _mi_os_free(void* p, size_t size, mi_stats_t* stats) { + _mi_os_free_ex(p, size, true, stats); +} + +void* _mi_os_alloc_aligned(size_t size, size_t alignment, bool commit, bool* large, mi_stats_t* tld_stats) +{ + MI_UNUSED(&mi_os_get_aligned_hint); // suppress unused warnings + MI_UNUSED(tld_stats); + if (size == 0) return NULL; + size = _mi_os_good_alloc_size(size); + alignment = _mi_align_up(alignment, _mi_os_page_size()); + bool allow_large = false; + if (large != NULL) { + allow_large = *large; + *large = false; + } + return mi_os_mem_alloc_aligned(size, alignment, commit, allow_large, (large!=NULL?large:&allow_large), &_mi_stats_main /*tld->stats*/ ); +} + + + +/* ----------------------------------------------------------- + OS memory API: reset, commit, decommit, protect, unprotect. +----------------------------------------------------------- */ + + +// OS page align within a given area, either conservative (pages inside the area only), +// or not (straddling pages outside the area is possible) +static void* mi_os_page_align_areax(bool conservative, void* addr, size_t size, size_t* newsize) { + mi_assert(addr != NULL && size > 0); + if (newsize != NULL) *newsize = 0; + if (size == 0 || addr == NULL) return NULL; + + // page align conservatively within the range + void* start = (conservative ? mi_align_up_ptr(addr, _mi_os_page_size()) + : mi_align_down_ptr(addr, _mi_os_page_size())); + void* end = (conservative ? mi_align_down_ptr((uint8_t*)addr + size, _mi_os_page_size()) + : mi_align_up_ptr((uint8_t*)addr + size, _mi_os_page_size())); + ptrdiff_t diff = (uint8_t*)end - (uint8_t*)start; + if (diff <= 0) return NULL; + + mi_assert_internal((conservative && (size_t)diff <= size) || (!conservative && (size_t)diff >= size)); + if (newsize != NULL) *newsize = (size_t)diff; + return start; +} + +static void* mi_os_page_align_area_conservative(void* addr, size_t size, size_t* newsize) { + return mi_os_page_align_areax(true, addr, size, newsize); +} + +static void mi_mprotect_hint(int err) { +#if defined(MI_OS_USE_MMAP) && (MI_SECURE>=2) // guard page around every mimalloc page + if (err == ENOMEM) { + _mi_warning_message("the previous warning may have been caused by a low memory map limit.\n" + " On Linux this is controlled by the vm.max_map_count. For example:\n" + " > sudo sysctl -w vm.max_map_count=262144\n"); + } +#else + MI_UNUSED(err); +#endif +} + +// Commit/Decommit memory. +// Usually commit is aligned liberal, while decommit is aligned conservative. +// (but not for the reset version where we want commit to be conservative as well) +static bool mi_os_commitx(void* addr, size_t size, bool commit, bool conservative, bool* is_zero, mi_stats_t* stats) { + // page align in the range, commit liberally, decommit conservative + if (is_zero != NULL) { *is_zero = false; } + size_t csize; + void* start = mi_os_page_align_areax(conservative, addr, size, &csize); + if (csize == 0) return true; // || _mi_os_is_huge_reserved(addr)) + int err = 0; + if (commit) { + _mi_stat_increase(&stats->committed, size); // use size for precise commit vs. decommit + _mi_stat_counter_increase(&stats->commit_calls, 1); + } + else { + _mi_stat_decrease(&stats->committed, size); + } + + #if defined(_WIN32) + if (commit) { + // *is_zero = true; // note: if the memory was already committed, the call succeeds but the memory is not zero'd + void* p = VirtualAlloc(start, csize, MEM_COMMIT, PAGE_READWRITE); + err = (p == start ? 0 : GetLastError()); + } + else { + BOOL ok = VirtualFree(start, csize, MEM_DECOMMIT); + err = (ok ? 0 : GetLastError()); + } + #elif defined(__wasi__) + // WebAssembly guests can't control memory protection + #elif 0 && defined(MAP_FIXED) && !defined(__APPLE__) + // Linux: disabled for now as mmap fixed seems much more expensive than MADV_DONTNEED (and splits VMA's?) + if (commit) { + // commit: just change the protection + err = mprotect(start, csize, (PROT_READ | PROT_WRITE)); + if (err != 0) { err = errno; } + } + else { + // decommit: use mmap with MAP_FIXED to discard the existing memory (and reduce rss) + const int fd = mi_unix_mmap_fd(); + void* p = mmap(start, csize, PROT_NONE, (MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS | MAP_NORESERVE), fd, 0); + if (p != start) { err = errno; } + } + #else + // Linux, macOSX and others. + if (commit) { + // commit: ensure we can access the area + err = mprotect(start, csize, (PROT_READ | PROT_WRITE)); + if (err != 0) { err = errno; } + } + else { + #if defined(MADV_DONTNEED) && MI_DEBUG == 0 && MI_SECURE == 0 + // decommit: use MADV_DONTNEED as it decreases rss immediately (unlike MADV_FREE) + // (on the other hand, MADV_FREE would be good enough.. it is just not reflected in the stats :-( ) + err = madvise(start, csize, MADV_DONTNEED); + #else + // decommit: just disable access (also used in debug and secure mode to trap on illegal access) + err = mprotect(start, csize, PROT_NONE); + if (err != 0) { err = errno; } + #endif + //#if defined(MADV_FREE_REUSE) + // while ((err = mi_madvise(start, csize, MADV_FREE_REUSE)) != 0 && errno == EAGAIN) { errno = 0; } + //#endif + } + #endif + if (err != 0) { + _mi_warning_message("%s error: start: %p, csize: 0x%zx, err: %i\n", commit ? "commit" : "decommit", start, csize, err); + mi_mprotect_hint(err); + } + mi_assert_internal(err == 0); + return (err == 0); +} + +bool _mi_os_commit(void* addr, size_t size, bool* is_zero, mi_stats_t* tld_stats) { + MI_UNUSED(tld_stats); + mi_stats_t* stats = &_mi_stats_main; + return mi_os_commitx(addr, size, true, false /* liberal */, is_zero, stats); +} + +bool _mi_os_decommit(void* addr, size_t size, mi_stats_t* tld_stats) { + MI_UNUSED(tld_stats); + mi_stats_t* stats = &_mi_stats_main; + bool is_zero; + return mi_os_commitx(addr, size, false, true /* conservative */, &is_zero, stats); +} + +/* +static bool mi_os_commit_unreset(void* addr, size_t size, bool* is_zero, mi_stats_t* stats) { + return mi_os_commitx(addr, size, true, true // conservative + , is_zero, stats); +} +*/ + +// Signal to the OS that the address range is no longer in use +// but may be used later again. This will release physical memory +// pages and reduce swapping while keeping the memory committed. +// We page align to a conservative area inside the range to reset. +static bool mi_os_resetx(void* addr, size_t size, bool reset, mi_stats_t* stats) { + // page align conservatively within the range + size_t csize; + void* start = mi_os_page_align_area_conservative(addr, size, &csize); + if (csize == 0) return true; // || _mi_os_is_huge_reserved(addr) + if (reset) _mi_stat_increase(&stats->reset, csize); + else _mi_stat_decrease(&stats->reset, csize); + if (!reset) return true; // nothing to do on unreset! + + #if (MI_DEBUG>1) + if (MI_SECURE==0) { + memset(start, 0, csize); // pretend it is eagerly reset + } + #endif + +#if defined(_WIN32) + // Testing shows that for us (on `malloc-large`) MEM_RESET is 2x faster than DiscardVirtualMemory + void* p = VirtualAlloc(start, csize, MEM_RESET, PAGE_READWRITE); + mi_assert_internal(p == start); + #if 1 + if (p == start && start != NULL) { + VirtualUnlock(start,csize); // VirtualUnlock after MEM_RESET removes the memory from the working set + } + #endif + if (p != start) return false; +#else +#if defined(MADV_FREE) + static _Atomic(size_t) advice = MI_ATOMIC_VAR_INIT(MADV_FREE); + int oadvice = (int)mi_atomic_load_relaxed(&advice); + int err; + while ((err = mi_madvise(start, csize, oadvice)) != 0 && errno == EAGAIN) { errno = 0; }; + if (err != 0 && errno == EINVAL && oadvice == MADV_FREE) { + // if MADV_FREE is not supported, fall back to MADV_DONTNEED from now on + mi_atomic_store_release(&advice, (size_t)MADV_DONTNEED); + err = mi_madvise(start, csize, MADV_DONTNEED); + } +#elif defined(__wasi__) + int err = 0; +#else + int err = mi_madvise(start, csize, MADV_DONTNEED); +#endif + if (err != 0) { + _mi_warning_message("madvise reset error: start: %p, csize: 0x%zx, errno: %i\n", start, csize, errno); + } + //mi_assert(err == 0); + if (err != 0) return false; +#endif + return true; +} + +// Signal to the OS that the address range is no longer in use +// but may be used later again. This will release physical memory +// pages and reduce swapping while keeping the memory committed. +// We page align to a conservative area inside the range to reset. +bool _mi_os_reset(void* addr, size_t size, mi_stats_t* tld_stats) { + MI_UNUSED(tld_stats); + mi_stats_t* stats = &_mi_stats_main; + return mi_os_resetx(addr, size, true, stats); +} + +/* +bool _mi_os_unreset(void* addr, size_t size, bool* is_zero, mi_stats_t* tld_stats) { + MI_UNUSED(tld_stats); + mi_stats_t* stats = &_mi_stats_main; + if (mi_option_is_enabled(mi_option_reset_decommits)) { + return mi_os_commit_unreset(addr, size, is_zero, stats); // re-commit it (conservatively!) + } + else { + *is_zero = false; + return mi_os_resetx(addr, size, false, stats); + } +} +*/ + +// Protect a region in memory to be not accessible. +static bool mi_os_protectx(void* addr, size_t size, bool protect) { + // page align conservatively within the range + size_t csize = 0; + void* start = mi_os_page_align_area_conservative(addr, size, &csize); + if (csize == 0) return false; + /* + if (_mi_os_is_huge_reserved(addr)) { + _mi_warning_message("cannot mprotect memory allocated in huge OS pages\n"); + } + */ + int err = 0; +#ifdef _WIN32 + DWORD oldprotect = 0; + BOOL ok = VirtualProtect(start, csize, protect ? PAGE_NOACCESS : PAGE_READWRITE, &oldprotect); + err = (ok ? 0 : GetLastError()); +#elif defined(__wasi__) + err = 0; +#else + err = mprotect(start, csize, protect ? PROT_NONE : (PROT_READ | PROT_WRITE)); + if (err != 0) { err = errno; } +#endif + if (err != 0) { + _mi_warning_message("mprotect error: start: %p, csize: 0x%zx, err: %i\n", start, csize, err); + mi_mprotect_hint(err); + } + return (err == 0); +} + +bool _mi_os_protect(void* addr, size_t size) { + return mi_os_protectx(addr, size, true); +} + +bool _mi_os_unprotect(void* addr, size_t size) { + return mi_os_protectx(addr, size, false); +} + + + +bool _mi_os_shrink(void* p, size_t oldsize, size_t newsize, mi_stats_t* stats) { + // page align conservatively within the range + mi_assert_internal(oldsize > newsize && p != NULL); + if (oldsize < newsize || p == NULL) return false; + if (oldsize == newsize) return true; + + // oldsize and newsize should be page aligned or we cannot shrink precisely + void* addr = (uint8_t*)p + newsize; + size_t size = 0; + void* start = mi_os_page_align_area_conservative(addr, oldsize - newsize, &size); + if (size == 0 || start != addr) return false; + +#ifdef _WIN32 + // we cannot shrink on windows, but we can decommit + return _mi_os_decommit(start, size, stats); +#else + return mi_os_mem_free(start, size, true, stats); +#endif +} + + +/* ---------------------------------------------------------------------------- +Support for allocating huge OS pages (1Gib) that are reserved up-front +and possibly associated with a specific NUMA node. (use `numa_node>=0`) +-----------------------------------------------------------------------------*/ +#define MI_HUGE_OS_PAGE_SIZE (MI_GiB) + +#if defined(_WIN32) && (MI_INTPTR_SIZE >= 8) +static void* mi_os_alloc_huge_os_pagesx(void* addr, size_t size, int numa_node) +{ + mi_assert_internal(size%MI_GiB == 0); + mi_assert_internal(addr != NULL); + const DWORD flags = MEM_LARGE_PAGES | MEM_COMMIT | MEM_RESERVE; + + mi_win_enable_large_os_pages(); + + MI_MEM_EXTENDED_PARAMETER params[3] = { {{0,0},{0}},{{0,0},{0}},{{0,0},{0}} }; + // on modern Windows try use NtAllocateVirtualMemoryEx for 1GiB huge pages + static bool mi_huge_pages_available = true; + if (pNtAllocateVirtualMemoryEx != NULL && mi_huge_pages_available) { + params[0].Type.Type = MiMemExtendedParameterAttributeFlags; + params[0].Arg.ULong64 = MI_MEM_EXTENDED_PARAMETER_NONPAGED_HUGE; + ULONG param_count = 1; + if (numa_node >= 0) { + param_count++; + params[1].Type.Type = MiMemExtendedParameterNumaNode; + params[1].Arg.ULong = (unsigned)numa_node; + } + SIZE_T psize = size; + void* base = addr; + NTSTATUS err = (*pNtAllocateVirtualMemoryEx)(GetCurrentProcess(), &base, &psize, flags, PAGE_READWRITE, params, param_count); + if (err == 0 && base != NULL) { + return base; + } + else { + // fall back to regular large pages + mi_huge_pages_available = false; // don't try further huge pages + _mi_warning_message("unable to allocate using huge (1GiB) pages, trying large (2MiB) pages instead (status 0x%lx)\n", err); + } + } + // on modern Windows try use VirtualAlloc2 for numa aware large OS page allocation + if (pVirtualAlloc2 != NULL && numa_node >= 0) { + params[0].Type.Type = MiMemExtendedParameterNumaNode; + params[0].Arg.ULong = (unsigned)numa_node; + return (*pVirtualAlloc2)(GetCurrentProcess(), addr, size, flags, PAGE_READWRITE, params, 1); + } + + // otherwise use regular virtual alloc on older windows + return VirtualAlloc(addr, size, flags, PAGE_READWRITE); +} + +#elif defined(MI_OS_USE_MMAP) && (MI_INTPTR_SIZE >= 8) && !defined(__HAIKU__) +#include +#ifndef MPOL_PREFERRED +#define MPOL_PREFERRED 1 +#endif +#if defined(SYS_mbind) +static long mi_os_mbind(void* start, unsigned long len, unsigned long mode, const unsigned long* nmask, unsigned long maxnode, unsigned flags) { + return syscall(SYS_mbind, start, len, mode, nmask, maxnode, flags); +} +#else +static long mi_os_mbind(void* start, unsigned long len, unsigned long mode, const unsigned long* nmask, unsigned long maxnode, unsigned flags) { + MI_UNUSED(start); MI_UNUSED(len); MI_UNUSED(mode); MI_UNUSED(nmask); MI_UNUSED(maxnode); MI_UNUSED(flags); + return 0; +} +#endif +static void* mi_os_alloc_huge_os_pagesx(void* addr, size_t size, int numa_node) { + mi_assert_internal(size%MI_GiB == 0); + bool is_large = true; + void* p = mi_unix_mmap(addr, size, MI_SEGMENT_SIZE, PROT_READ | PROT_WRITE, true, true, &is_large); + if (p == NULL) return NULL; + if (numa_node >= 0 && numa_node < 8*MI_INTPTR_SIZE) { // at most 64 nodes + unsigned long numa_mask = (1UL << numa_node); + // TODO: does `mbind` work correctly for huge OS pages? should we + // use `set_mempolicy` before calling mmap instead? + // see: + long err = mi_os_mbind(p, size, MPOL_PREFERRED, &numa_mask, 8*MI_INTPTR_SIZE, 0); + if (err != 0) { + _mi_warning_message("failed to bind huge (1GiB) pages to numa node %d: %s\n", numa_node, strerror(errno)); + } + } + return p; +} +#else +static void* mi_os_alloc_huge_os_pagesx(void* addr, size_t size, int numa_node) { + MI_UNUSED(addr); MI_UNUSED(size); MI_UNUSED(numa_node); + return NULL; +} +#endif + +#if (MI_INTPTR_SIZE >= 8) +// To ensure proper alignment, use our own area for huge OS pages +static mi_decl_cache_align _Atomic(uintptr_t) mi_huge_start; // = 0 + +// Claim an aligned address range for huge pages +static uint8_t* mi_os_claim_huge_pages(size_t pages, size_t* total_size) { + if (total_size != NULL) *total_size = 0; + const size_t size = pages * MI_HUGE_OS_PAGE_SIZE; + + uintptr_t start = 0; + uintptr_t end = 0; + uintptr_t huge_start = mi_atomic_load_relaxed(&mi_huge_start); + do { + start = huge_start; + if (start == 0) { + // Initialize the start address after the 32TiB area + start = ((uintptr_t)32 << 40); // 32TiB virtual start address +#if (MI_SECURE>0 || MI_DEBUG==0) // security: randomize start of huge pages unless in debug mode + uintptr_t r = _mi_heap_random_next(mi_get_default_heap()); + start = start + ((uintptr_t)MI_HUGE_OS_PAGE_SIZE * ((r>>17) & 0x0FFF)); // (randomly 12bits)*1GiB == between 0 to 4TiB +#endif + } + end = start + size; + mi_assert_internal(end % MI_SEGMENT_SIZE == 0); + } while (!mi_atomic_cas_strong_acq_rel(&mi_huge_start, &huge_start, end)); + + if (total_size != NULL) *total_size = size; + return (uint8_t*)start; +} +#else +static uint8_t* mi_os_claim_huge_pages(size_t pages, size_t* total_size) { + MI_UNUSED(pages); + if (total_size != NULL) *total_size = 0; + return NULL; +} +#endif + +// Allocate MI_SEGMENT_SIZE aligned huge pages +void* _mi_os_alloc_huge_os_pages(size_t pages, int numa_node, mi_msecs_t max_msecs, size_t* pages_reserved, size_t* psize) { + if (psize != NULL) *psize = 0; + if (pages_reserved != NULL) *pages_reserved = 0; + size_t size = 0; + uint8_t* start = mi_os_claim_huge_pages(pages, &size); + if (start == NULL) return NULL; // or 32-bit systems + + // Allocate one page at the time but try to place them contiguously + // We allocate one page at the time to be able to abort if it takes too long + // or to at least allocate as many as available on the system. + mi_msecs_t start_t = _mi_clock_start(); + size_t page; + for (page = 0; page < pages; page++) { + // allocate a page + void* addr = start + (page * MI_HUGE_OS_PAGE_SIZE); + void* p = mi_os_alloc_huge_os_pagesx(addr, MI_HUGE_OS_PAGE_SIZE, numa_node); + + // Did we succeed at a contiguous address? + if (p != addr) { + // no success, issue a warning and break + if (p != NULL) { + _mi_warning_message("could not allocate contiguous huge page %zu at %p\n", page, addr); + _mi_os_free(p, MI_HUGE_OS_PAGE_SIZE, &_mi_stats_main); + } + break; + } + + // success, record it + _mi_stat_increase(&_mi_stats_main.committed, MI_HUGE_OS_PAGE_SIZE); + _mi_stat_increase(&_mi_stats_main.reserved, MI_HUGE_OS_PAGE_SIZE); + + // check for timeout + if (max_msecs > 0) { + mi_msecs_t elapsed = _mi_clock_end(start_t); + if (page >= 1) { + mi_msecs_t estimate = ((elapsed / (page+1)) * pages); + if (estimate > 2*max_msecs) { // seems like we are going to timeout, break + elapsed = max_msecs + 1; + } + } + if (elapsed > max_msecs) { + _mi_warning_message("huge page allocation timed out\n"); + break; + } + } + } + mi_assert_internal(page*MI_HUGE_OS_PAGE_SIZE <= size); + if (pages_reserved != NULL) { *pages_reserved = page; } + if (psize != NULL) { *psize = page * MI_HUGE_OS_PAGE_SIZE; } + return (page == 0 ? NULL : start); +} + +// free every huge page in a range individually (as we allocated per page) +// note: needed with VirtualAlloc but could potentially be done in one go on mmap'd systems. +void _mi_os_free_huge_pages(void* p, size_t size, mi_stats_t* stats) { + if (p==NULL || size==0) return; + uint8_t* base = (uint8_t*)p; + while (size >= MI_HUGE_OS_PAGE_SIZE) { + _mi_os_free(base, MI_HUGE_OS_PAGE_SIZE, stats); + size -= MI_HUGE_OS_PAGE_SIZE; + base += MI_HUGE_OS_PAGE_SIZE; + } +} + +/* ---------------------------------------------------------------------------- +Support NUMA aware allocation +-----------------------------------------------------------------------------*/ +#ifdef _WIN32 +static size_t mi_os_numa_nodex(void) { + USHORT numa_node = 0; + if (pGetCurrentProcessorNumberEx != NULL && pGetNumaProcessorNodeEx != NULL) { + // Extended API is supported + MI_PROCESSOR_NUMBER pnum; + (*pGetCurrentProcessorNumberEx)(&pnum); + USHORT nnode = 0; + BOOL ok = (*pGetNumaProcessorNodeEx)(&pnum, &nnode); + if (ok) numa_node = nnode; + } + else if (pGetNumaProcessorNode != NULL) { + // Vista or earlier, use older API that is limited to 64 processors. Issue #277 + DWORD pnum = GetCurrentProcessorNumber(); + UCHAR nnode = 0; + BOOL ok = pGetNumaProcessorNode((UCHAR)pnum, &nnode); + if (ok) numa_node = nnode; + } + return numa_node; +} + +static size_t mi_os_numa_node_countx(void) { + ULONG numa_max = 0; + GetNumaHighestNodeNumber(&numa_max); + // find the highest node number that has actual processors assigned to it. Issue #282 + while(numa_max > 0) { + if (pGetNumaNodeProcessorMaskEx != NULL) { + // Extended API is supported + GROUP_AFFINITY affinity; + if ((*pGetNumaNodeProcessorMaskEx)((USHORT)numa_max, &affinity)) { + if (affinity.Mask != 0) break; // found the maximum non-empty node + } + } + else { + // Vista or earlier, use older API that is limited to 64 processors. + ULONGLONG mask; + if (GetNumaNodeProcessorMask((UCHAR)numa_max, &mask)) { + if (mask != 0) break; // found the maximum non-empty node + }; + } + // max node was invalid or had no processor assigned, try again + numa_max--; + } + return ((size_t)numa_max + 1); +} +#elif defined(__linux__) +#include // getcpu +#include // access + +static size_t mi_os_numa_nodex(void) { +#ifdef SYS_getcpu + unsigned long node = 0; + unsigned long ncpu = 0; + long err = syscall(SYS_getcpu, &ncpu, &node, NULL); + if (err != 0) return 0; + return node; +#else + return 0; +#endif +} +static size_t mi_os_numa_node_countx(void) { + char buf[128]; + unsigned node = 0; + for(node = 0; node < 256; node++) { + // enumerate node entries -- todo: it there a more efficient way to do this? (but ensure there is no allocation) + snprintf(buf, 127, "/sys/devices/system/node/node%u", node + 1); + if (access(buf,R_OK) != 0) break; + } + return (node+1); +} +#elif defined(__FreeBSD__) && __FreeBSD_version >= 1200000 +static size_t mi_os_numa_nodex(void) { + domainset_t dom; + size_t node; + int policy; + if (cpuset_getdomain(CPU_LEVEL_CPUSET, CPU_WHICH_PID, -1, sizeof(dom), &dom, &policy) == -1) return 0ul; + for (node = 0; node < MAXMEMDOM; node++) { + if (DOMAINSET_ISSET(node, &dom)) return node; + } + return 0ul; +} +static size_t mi_os_numa_node_countx(void) { + size_t ndomains = 0; + size_t len = sizeof(ndomains); + if (sysctlbyname("vm.ndomains", &ndomains, &len, NULL, 0) == -1) return 0ul; + return ndomains; +} +#elif defined(__DragonFly__) +static size_t mi_os_numa_nodex(void) { + // TODO: DragonFly does not seem to provide any userland means to get this information. + return 0ul; +} +static size_t mi_os_numa_node_countx(void) { + size_t ncpus = 0, nvirtcoresperphys = 0; + size_t len = sizeof(size_t); + if (sysctlbyname("hw.ncpu", &ncpus, &len, NULL, 0) == -1) return 0ul; + if (sysctlbyname("hw.cpu_topology_ht_ids", &nvirtcoresperphys, &len, NULL, 0) == -1) return 0ul; + return nvirtcoresperphys * ncpus; +} +#else +static size_t mi_os_numa_nodex(void) { + return 0; +} +static size_t mi_os_numa_node_countx(void) { + return 1; +} +#endif + +_Atomic(size_t) _mi_numa_node_count; // = 0 // cache the node count + +size_t _mi_os_numa_node_count_get(void) { + size_t count = mi_atomic_load_acquire(&_mi_numa_node_count); + if (count <= 0) { + long ncount = mi_option_get(mi_option_use_numa_nodes); // given explicitly? + if (ncount > 0) { + count = (size_t)ncount; + } + else { + count = mi_os_numa_node_countx(); // or detect dynamically + if (count == 0) count = 1; + } + mi_atomic_store_release(&_mi_numa_node_count, count); // save it + _mi_verbose_message("using %zd numa regions\n", count); + } + return count; +} + +int _mi_os_numa_node_get(mi_os_tld_t* tld) { + MI_UNUSED(tld); + size_t numa_count = _mi_os_numa_node_count(); + if (numa_count<=1) return 0; // optimize on single numa node systems: always node 0 + // never more than the node count and >= 0 + size_t numa_node = mi_os_numa_nodex(); + if (numa_node >= numa_count) { numa_node = numa_node % numa_count; } + return (int)numa_node; +} diff --git a/compat/mimalloc/page-queue.c b/compat/mimalloc/page-queue.c new file mode 100644 index 00000000000000..e1a8a6a6592b86 --- /dev/null +++ b/compat/mimalloc/page-queue.c @@ -0,0 +1,331 @@ +/*---------------------------------------------------------------------------- +Copyright (c) 2018-2020, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ + +/* ----------------------------------------------------------- + Definition of page queues for each block size +----------------------------------------------------------- */ + +#ifndef MI_IN_PAGE_C +#error "this file should be included from 'page.c'" +#endif + +/* ----------------------------------------------------------- + Minimal alignment in machine words (i.e. `sizeof(void*)`) +----------------------------------------------------------- */ + +#if (MI_MAX_ALIGN_SIZE > 4*MI_INTPTR_SIZE) + #error "define alignment for more than 4x word size for this platform" +#elif (MI_MAX_ALIGN_SIZE > 2*MI_INTPTR_SIZE) + #define MI_ALIGN4W // 4 machine words minimal alignment +#elif (MI_MAX_ALIGN_SIZE > MI_INTPTR_SIZE) + #define MI_ALIGN2W // 2 machine words minimal alignment +#else + // ok, default alignment is 1 word +#endif + + +/* ----------------------------------------------------------- + Queue query +----------------------------------------------------------- */ + + +static inline bool mi_page_queue_is_huge(const mi_page_queue_t* pq) { + return (pq->block_size == (MI_MEDIUM_OBJ_SIZE_MAX+sizeof(uintptr_t))); +} + +static inline bool mi_page_queue_is_full(const mi_page_queue_t* pq) { + return (pq->block_size == (MI_MEDIUM_OBJ_SIZE_MAX+(2*sizeof(uintptr_t)))); +} + +static inline bool mi_page_queue_is_special(const mi_page_queue_t* pq) { + return (pq->block_size > MI_MEDIUM_OBJ_SIZE_MAX); +} + +/* ----------------------------------------------------------- + Bins +----------------------------------------------------------- */ + +// Return the bin for a given field size. +// Returns MI_BIN_HUGE if the size is too large. +// We use `wsize` for the size in "machine word sizes", +// i.e. byte size == `wsize*sizeof(void*)`. +static inline uint8_t mi_bin(size_t size) { + size_t wsize = _mi_wsize_from_size(size); + uint8_t bin; + if (wsize <= 1) { + bin = 1; + } + #if defined(MI_ALIGN4W) + else if (wsize <= 4) { + bin = (uint8_t)((wsize+1)&~1); // round to double word sizes + } + #elif defined(MI_ALIGN2W) + else if (wsize <= 8) { + bin = (uint8_t)((wsize+1)&~1); // round to double word sizes + } + #else + else if (wsize <= 8) { + bin = (uint8_t)wsize; + } + #endif + else if (wsize > MI_MEDIUM_OBJ_WSIZE_MAX) { + bin = MI_BIN_HUGE; + } + else { + #if defined(MI_ALIGN4W) + if (wsize <= 16) { wsize = (wsize+3)&~3; } // round to 4x word sizes + #endif + wsize--; + // find the highest bit + uint8_t b = (uint8_t)mi_bsr(wsize); // note: wsize != 0 + // and use the top 3 bits to determine the bin (~12.5% worst internal fragmentation). + // - adjust with 3 because we use do not round the first 8 sizes + // which each get an exact bin + bin = ((b << 2) + (uint8_t)((wsize >> (b - 2)) & 0x03)) - 3; + mi_assert_internal(bin < MI_BIN_HUGE); + } + mi_assert_internal(bin > 0 && bin <= MI_BIN_HUGE); + return bin; +} + + + +/* ----------------------------------------------------------- + Queue of pages with free blocks +----------------------------------------------------------- */ + +uint8_t _mi_bin(size_t size) { + return mi_bin(size); +} + +size_t _mi_bin_size(uint8_t bin) { + return _mi_heap_empty.pages[bin].block_size; +} + +// Good size for allocation +size_t mi_good_size(size_t size) mi_attr_noexcept { + if (size <= MI_MEDIUM_OBJ_SIZE_MAX) { + return _mi_bin_size(mi_bin(size)); + } + else { + return _mi_align_up(size,_mi_os_page_size()); + } +} + +#if (MI_DEBUG>1) +static bool mi_page_queue_contains(mi_page_queue_t* queue, const mi_page_t* page) { + mi_assert_internal(page != NULL); + mi_page_t* list = queue->first; + while (list != NULL) { + mi_assert_internal(list->next == NULL || list->next->prev == list); + mi_assert_internal(list->prev == NULL || list->prev->next == list); + if (list == page) break; + list = list->next; + } + return (list == page); +} + +#endif + +#if (MI_DEBUG>1) +static bool mi_heap_contains_queue(const mi_heap_t* heap, const mi_page_queue_t* pq) { + return (pq >= &heap->pages[0] && pq <= &heap->pages[MI_BIN_FULL]); +} +#endif + +static mi_page_queue_t* mi_page_queue_of(const mi_page_t* page) { + uint8_t bin = (mi_page_is_in_full(page) ? MI_BIN_FULL : mi_bin(page->xblock_size)); + mi_heap_t* heap = mi_page_heap(page); + mi_assert_internal(heap != NULL && bin <= MI_BIN_FULL); + mi_page_queue_t* pq = &heap->pages[bin]; + mi_assert_internal(bin >= MI_BIN_HUGE || page->xblock_size == pq->block_size); + mi_assert_expensive(mi_page_queue_contains(pq, page)); + return pq; +} + +static mi_page_queue_t* mi_heap_page_queue_of(mi_heap_t* heap, const mi_page_t* page) { + uint8_t bin = (mi_page_is_in_full(page) ? MI_BIN_FULL : mi_bin(page->xblock_size)); + mi_assert_internal(bin <= MI_BIN_FULL); + mi_page_queue_t* pq = &heap->pages[bin]; + mi_assert_internal(mi_page_is_in_full(page) || page->xblock_size == pq->block_size); + return pq; +} + +// The current small page array is for efficiency and for each +// small size (up to 256) it points directly to the page for that +// size without having to compute the bin. This means when the +// current free page queue is updated for a small bin, we need to update a +// range of entries in `_mi_page_small_free`. +static inline void mi_heap_queue_first_update(mi_heap_t* heap, const mi_page_queue_t* pq) { + mi_assert_internal(mi_heap_contains_queue(heap,pq)); + size_t size = pq->block_size; + if (size > MI_SMALL_SIZE_MAX) return; + + mi_page_t* page = pq->first; + if (pq->first == NULL) page = (mi_page_t*)&_mi_page_empty; + + // find index in the right direct page array + size_t start; + size_t idx = _mi_wsize_from_size(size); + mi_page_t** pages_free = heap->pages_free_direct; + + if (pages_free[idx] == page) return; // already set + + // find start slot + if (idx<=1) { + start = 0; + } + else { + // find previous size; due to minimal alignment upto 3 previous bins may need to be skipped + uint8_t bin = mi_bin(size); + const mi_page_queue_t* prev = pq - 1; + while( bin == mi_bin(prev->block_size) && prev > &heap->pages[0]) { + prev--; + } + start = 1 + _mi_wsize_from_size(prev->block_size); + if (start > idx) start = idx; + } + + // set size range to the right page + mi_assert(start <= idx); + for (size_t sz = start; sz <= idx; sz++) { + pages_free[sz] = page; + } +} + +/* +static bool mi_page_queue_is_empty(mi_page_queue_t* queue) { + return (queue->first == NULL); +} +*/ + +static void mi_page_queue_remove(mi_page_queue_t* queue, mi_page_t* page) { + mi_assert_internal(page != NULL); + mi_assert_expensive(mi_page_queue_contains(queue, page)); + mi_assert_internal(page->xblock_size == queue->block_size || (page->xblock_size > MI_MEDIUM_OBJ_SIZE_MAX && mi_page_queue_is_huge(queue)) || (mi_page_is_in_full(page) && mi_page_queue_is_full(queue))); + mi_heap_t* heap = mi_page_heap(page); + + if (page->prev != NULL) page->prev->next = page->next; + if (page->next != NULL) page->next->prev = page->prev; + if (page == queue->last) queue->last = page->prev; + if (page == queue->first) { + queue->first = page->next; + // update first + mi_assert_internal(mi_heap_contains_queue(heap, queue)); + mi_heap_queue_first_update(heap,queue); + } + heap->page_count--; + page->next = NULL; + page->prev = NULL; + // mi_atomic_store_ptr_release(mi_atomic_cast(void*, &page->heap), NULL); + mi_page_set_in_full(page,false); +} + + +static void mi_page_queue_push(mi_heap_t* heap, mi_page_queue_t* queue, mi_page_t* page) { + mi_assert_internal(mi_page_heap(page) == heap); + mi_assert_internal(!mi_page_queue_contains(queue, page)); + + mi_assert_internal(_mi_page_segment(page)->kind != MI_SEGMENT_HUGE); + mi_assert_internal(page->xblock_size == queue->block_size || + (page->xblock_size > MI_MEDIUM_OBJ_SIZE_MAX) || + (mi_page_is_in_full(page) && mi_page_queue_is_full(queue))); + + mi_page_set_in_full(page, mi_page_queue_is_full(queue)); + // mi_atomic_store_ptr_release(mi_atomic_cast(void*, &page->heap), heap); + page->next = queue->first; + page->prev = NULL; + if (queue->first != NULL) { + mi_assert_internal(queue->first->prev == NULL); + queue->first->prev = page; + queue->first = page; + } + else { + queue->first = queue->last = page; + } + + // update direct + mi_heap_queue_first_update(heap, queue); + heap->page_count++; +} + + +static void mi_page_queue_enqueue_from(mi_page_queue_t* to, mi_page_queue_t* from, mi_page_t* page) { + mi_assert_internal(page != NULL); + mi_assert_expensive(mi_page_queue_contains(from, page)); + mi_assert_expensive(!mi_page_queue_contains(to, page)); + + mi_assert_internal((page->xblock_size == to->block_size && page->xblock_size == from->block_size) || + (page->xblock_size == to->block_size && mi_page_queue_is_full(from)) || + (page->xblock_size == from->block_size && mi_page_queue_is_full(to)) || + (page->xblock_size > MI_LARGE_OBJ_SIZE_MAX && mi_page_queue_is_huge(to)) || + (page->xblock_size > MI_LARGE_OBJ_SIZE_MAX && mi_page_queue_is_full(to))); + + mi_heap_t* heap = mi_page_heap(page); + if (page->prev != NULL) page->prev->next = page->next; + if (page->next != NULL) page->next->prev = page->prev; + if (page == from->last) from->last = page->prev; + if (page == from->first) { + from->first = page->next; + // update first + mi_assert_internal(mi_heap_contains_queue(heap, from)); + mi_heap_queue_first_update(heap, from); + } + + page->prev = to->last; + page->next = NULL; + if (to->last != NULL) { + mi_assert_internal(heap == mi_page_heap(to->last)); + to->last->next = page; + to->last = page; + } + else { + to->first = page; + to->last = page; + mi_heap_queue_first_update(heap, to); + } + + mi_page_set_in_full(page, mi_page_queue_is_full(to)); +} + +// Only called from `mi_heap_absorb`. +size_t _mi_page_queue_append(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_queue_t* append) { + mi_assert_internal(mi_heap_contains_queue(heap,pq)); + mi_assert_internal(pq->block_size == append->block_size); + + if (append->first==NULL) return 0; + + // set append pages to new heap and count + size_t count = 0; + for (mi_page_t* page = append->first; page != NULL; page = page->next) { + // inline `mi_page_set_heap` to avoid wrong assertion during absorption; + // in this case it is ok to be delayed freeing since both "to" and "from" heap are still alive. + mi_atomic_store_release(&page->xheap, (uintptr_t)heap); + // set the flag to delayed free (not overriding NEVER_DELAYED_FREE) which has as a + // side effect that it spins until any DELAYED_FREEING is finished. This ensures + // that after appending only the new heap will be used for delayed free operations. + _mi_page_use_delayed_free(page, MI_USE_DELAYED_FREE, false); + count++; + } + + if (pq->last==NULL) { + // take over afresh + mi_assert_internal(pq->first==NULL); + pq->first = append->first; + pq->last = append->last; + mi_heap_queue_first_update(heap, pq); + } + else { + // append to end + mi_assert_internal(pq->last!=NULL); + mi_assert_internal(append->first!=NULL); + pq->last->next = append->first; + append->first->prev = pq->last; + pq->last = append->last; + } + return count; +} diff --git a/compat/mimalloc/page.c b/compat/mimalloc/page.c new file mode 100644 index 00000000000000..121683015de6b2 --- /dev/null +++ b/compat/mimalloc/page.c @@ -0,0 +1,869 @@ +/*---------------------------------------------------------------------------- +Copyright (c) 2018-2020, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ + +/* ----------------------------------------------------------- + The core of the allocator. Every segment contains + pages of a certain block size. The main function + exported is `mi_malloc_generic`. +----------------------------------------------------------- */ + +#include "mimalloc.h" +#include "mimalloc-internal.h" +#include "mimalloc-atomic.h" + +/* ----------------------------------------------------------- + Definition of page queues for each block size +----------------------------------------------------------- */ + +#define MI_IN_PAGE_C +#include "page-queue.c" +#undef MI_IN_PAGE_C + + +/* ----------------------------------------------------------- + Page helpers +----------------------------------------------------------- */ + +// Index a block in a page +static inline mi_block_t* mi_page_block_at(const mi_page_t* page, void* page_start, size_t block_size, size_t i) { + MI_UNUSED(page); + mi_assert_internal(page != NULL); + mi_assert_internal(i <= page->reserved); + return (mi_block_t*)((uint8_t*)page_start + (i * block_size)); +} + +static void mi_page_init(mi_heap_t* heap, mi_page_t* page, size_t size, mi_tld_t* tld); +static void mi_page_extend_free(mi_heap_t* heap, mi_page_t* page, mi_tld_t* tld); + +#if (MI_DEBUG>=3) +static size_t mi_page_list_count(mi_page_t* page, mi_block_t* head) { + size_t count = 0; + while (head != NULL) { + mi_assert_internal(page == _mi_ptr_page(head)); + count++; + head = mi_block_next(page, head); + } + return count; +} + +/* +// Start of the page available memory +static inline uint8_t* mi_page_area(const mi_page_t* page) { + return _mi_page_start(_mi_page_segment(page), page, NULL); +} +*/ + +static bool mi_page_list_is_valid(mi_page_t* page, mi_block_t* p) { + size_t psize; + uint8_t* page_area = _mi_page_start(_mi_page_segment(page), page, &psize); + mi_block_t* start = (mi_block_t*)page_area; + mi_block_t* end = (mi_block_t*)(page_area + psize); + while(p != NULL) { + if (p < start || p >= end) return false; + p = mi_block_next(page, p); + } + return true; +} + +static bool mi_page_is_valid_init(mi_page_t* page) { + mi_assert_internal(page->xblock_size > 0); + mi_assert_internal(page->used <= page->capacity); + mi_assert_internal(page->capacity <= page->reserved); + + mi_segment_t* segment = _mi_page_segment(page); + uint8_t* start = _mi_page_start(segment,page,NULL); + mi_assert_internal(start == _mi_segment_page_start(segment,page,NULL)); + //const size_t bsize = mi_page_block_size(page); + //mi_assert_internal(start + page->capacity*page->block_size == page->top); + + mi_assert_internal(mi_page_list_is_valid(page,page->free)); + mi_assert_internal(mi_page_list_is_valid(page,page->local_free)); + + #if MI_DEBUG>3 // generally too expensive to check this + if (page->is_zero) { + const size_t ubsize = mi_page_usable_block_size(page); + for(mi_block_t* block = page->free; block != NULL; block = mi_block_next(page,block)) { + mi_assert_expensive(mi_mem_is_zero(block + 1, ubsize - sizeof(mi_block_t))); + } + } + #endif + + mi_block_t* tfree = mi_page_thread_free(page); + mi_assert_internal(mi_page_list_is_valid(page, tfree)); + //size_t tfree_count = mi_page_list_count(page, tfree); + //mi_assert_internal(tfree_count <= page->thread_freed + 1); + + size_t free_count = mi_page_list_count(page, page->free) + mi_page_list_count(page, page->local_free); + mi_assert_internal(page->used + free_count == page->capacity); + + return true; +} + +bool _mi_page_is_valid(mi_page_t* page) { + mi_assert_internal(mi_page_is_valid_init(page)); + #if MI_SECURE + mi_assert_internal(page->keys[0] != 0); + #endif + if (mi_page_heap(page)!=NULL) { + mi_segment_t* segment = _mi_page_segment(page); + + mi_assert_internal(!_mi_process_is_initialized || segment->thread_id==0 || segment->thread_id == mi_page_heap(page)->thread_id); + if (segment->kind != MI_SEGMENT_HUGE) { + mi_page_queue_t* pq = mi_page_queue_of(page); + mi_assert_internal(mi_page_queue_contains(pq, page)); + mi_assert_internal(pq->block_size==mi_page_block_size(page) || mi_page_block_size(page) > MI_MEDIUM_OBJ_SIZE_MAX || mi_page_is_in_full(page)); + mi_assert_internal(mi_heap_contains_queue(mi_page_heap(page),pq)); + } + } + return true; +} +#endif + +void _mi_page_use_delayed_free(mi_page_t* page, mi_delayed_t delay, bool override_never) { + mi_thread_free_t tfreex; + mi_delayed_t old_delay; + mi_thread_free_t tfree; + do { + tfree = mi_atomic_load_acquire(&page->xthread_free); // note: must acquire as we can break/repeat this loop and not do a CAS; + tfreex = mi_tf_set_delayed(tfree, delay); + old_delay = mi_tf_delayed(tfree); + if (mi_unlikely(old_delay == MI_DELAYED_FREEING)) { + mi_atomic_yield(); // delay until outstanding MI_DELAYED_FREEING are done. + // tfree = mi_tf_set_delayed(tfree, MI_NO_DELAYED_FREE); // will cause CAS to busy fail + } + else if (delay == old_delay) { + break; // avoid atomic operation if already equal + } + else if (!override_never && old_delay == MI_NEVER_DELAYED_FREE) { + break; // leave never-delayed flag set + } + } while ((old_delay == MI_DELAYED_FREEING) || + !mi_atomic_cas_weak_release(&page->xthread_free, &tfree, tfreex)); +} + +/* ----------------------------------------------------------- + Page collect the `local_free` and `thread_free` lists +----------------------------------------------------------- */ + +// Collect the local `thread_free` list using an atomic exchange. +// Note: The exchange must be done atomically as this is used right after +// moving to the full list in `mi_page_collect_ex` and we need to +// ensure that there was no race where the page became unfull just before the move. +static void _mi_page_thread_free_collect(mi_page_t* page) +{ + mi_block_t* head; + mi_thread_free_t tfreex; + mi_thread_free_t tfree = mi_atomic_load_relaxed(&page->xthread_free); + do { + head = mi_tf_block(tfree); + tfreex = mi_tf_set_block(tfree,NULL); + } while (!mi_atomic_cas_weak_acq_rel(&page->xthread_free, &tfree, tfreex)); + + // return if the list is empty + if (head == NULL) return; + + // find the tail -- also to get a proper count (without data races) + uint32_t max_count = page->capacity; // cannot collect more than capacity + uint32_t count = 1; + mi_block_t* tail = head; + mi_block_t* next; + while ((next = mi_block_next(page,tail)) != NULL && count <= max_count) { + count++; + tail = next; + } + // if `count > max_count` there was a memory corruption (possibly infinite list due to double multi-threaded free) + if (count > max_count) { + _mi_error_message(EFAULT, "corrupted thread-free list\n"); + return; // the thread-free items cannot be freed + } + + // and append the current local free list + mi_block_set_next(page,tail, page->local_free); + page->local_free = head; + + // update counts now + page->used -= count; +} + +void _mi_page_free_collect(mi_page_t* page, bool force) { + mi_assert_internal(page!=NULL); + + // collect the thread free list + if (force || mi_page_thread_free(page) != NULL) { // quick test to avoid an atomic operation + _mi_page_thread_free_collect(page); + } + + // and the local free list + if (page->local_free != NULL) { + if (mi_likely(page->free == NULL)) { + // usual case + page->free = page->local_free; + page->local_free = NULL; + page->is_zero = false; + } + else if (force) { + // append -- only on shutdown (force) as this is a linear operation + mi_block_t* tail = page->local_free; + mi_block_t* next; + while ((next = mi_block_next(page, tail)) != NULL) { + tail = next; + } + mi_block_set_next(page, tail, page->free); + page->free = page->local_free; + page->local_free = NULL; + page->is_zero = false; + } + } + + mi_assert_internal(!force || page->local_free == NULL); +} + + + +/* ----------------------------------------------------------- + Page fresh and retire +----------------------------------------------------------- */ + +// called from segments when reclaiming abandoned pages +void _mi_page_reclaim(mi_heap_t* heap, mi_page_t* page) { + mi_assert_expensive(mi_page_is_valid_init(page)); + + mi_assert_internal(mi_page_heap(page) == heap); + mi_assert_internal(mi_page_thread_free_flag(page) != MI_NEVER_DELAYED_FREE); + mi_assert_internal(_mi_page_segment(page)->kind != MI_SEGMENT_HUGE); + mi_assert_internal(!page->is_reset); + // TODO: push on full queue immediately if it is full? + mi_page_queue_t* pq = mi_page_queue(heap, mi_page_block_size(page)); + mi_page_queue_push(heap, pq, page); + mi_assert_expensive(_mi_page_is_valid(page)); +} + +// allocate a fresh page from a segment +static mi_page_t* mi_page_fresh_alloc(mi_heap_t* heap, mi_page_queue_t* pq, size_t block_size) { + mi_assert_internal(pq==NULL||mi_heap_contains_queue(heap, pq)); + mi_page_t* page = _mi_segment_page_alloc(heap, block_size, &heap->tld->segments, &heap->tld->os); + if (page == NULL) { + // this may be out-of-memory, or an abandoned page was reclaimed (and in our queue) + return NULL; + } + mi_assert_internal(pq==NULL || _mi_page_segment(page)->kind != MI_SEGMENT_HUGE); + mi_page_init(heap, page, block_size, heap->tld); + mi_heap_stat_increase(heap, pages, 1); + if (pq!=NULL) mi_page_queue_push(heap, pq, page); // huge pages use pq==NULL + mi_assert_expensive(_mi_page_is_valid(page)); + return page; +} + +// Get a fresh page to use +static mi_page_t* mi_page_fresh(mi_heap_t* heap, mi_page_queue_t* pq) { + mi_assert_internal(mi_heap_contains_queue(heap, pq)); + mi_page_t* page = mi_page_fresh_alloc(heap, pq, pq->block_size); + if (page==NULL) return NULL; + mi_assert_internal(pq->block_size==mi_page_block_size(page)); + mi_assert_internal(pq==mi_page_queue(heap, mi_page_block_size(page))); + return page; +} + +/* ----------------------------------------------------------- + Do any delayed frees + (put there by other threads if they deallocated in a full page) +----------------------------------------------------------- */ +void _mi_heap_delayed_free(mi_heap_t* heap) { + // take over the list (note: no atomic exchange since it is often NULL) + mi_block_t* block = mi_atomic_load_ptr_relaxed(mi_block_t, &heap->thread_delayed_free); + while (block != NULL && !mi_atomic_cas_ptr_weak_acq_rel(mi_block_t, &heap->thread_delayed_free, &block, NULL)) { /* nothing */ }; + + // and free them all + while(block != NULL) { + mi_block_t* next = mi_block_nextx(heap,block, heap->keys); + // use internal free instead of regular one to keep stats etc correct + if (!_mi_free_delayed_block(block)) { + // we might already start delayed freeing while another thread has not yet + // reset the delayed_freeing flag; in that case delay it further by reinserting. + mi_block_t* dfree = mi_atomic_load_ptr_relaxed(mi_block_t, &heap->thread_delayed_free); + do { + mi_block_set_nextx(heap, block, dfree, heap->keys); + } while (!mi_atomic_cas_ptr_weak_release(mi_block_t,&heap->thread_delayed_free, &dfree, block)); + } + block = next; + } +} + +/* ----------------------------------------------------------- + Unfull, abandon, free and retire +----------------------------------------------------------- */ + +// Move a page from the full list back to a regular list +void _mi_page_unfull(mi_page_t* page) { + mi_assert_internal(page != NULL); + mi_assert_expensive(_mi_page_is_valid(page)); + mi_assert_internal(mi_page_is_in_full(page)); + if (!mi_page_is_in_full(page)) return; + + mi_heap_t* heap = mi_page_heap(page); + mi_page_queue_t* pqfull = &heap->pages[MI_BIN_FULL]; + mi_page_set_in_full(page, false); // to get the right queue + mi_page_queue_t* pq = mi_heap_page_queue_of(heap, page); + mi_page_set_in_full(page, true); + mi_page_queue_enqueue_from(pq, pqfull, page); +} + +static void mi_page_to_full(mi_page_t* page, mi_page_queue_t* pq) { + mi_assert_internal(pq == mi_page_queue_of(page)); + mi_assert_internal(!mi_page_immediate_available(page)); + mi_assert_internal(!mi_page_is_in_full(page)); + + if (mi_page_is_in_full(page)) return; + mi_page_queue_enqueue_from(&mi_page_heap(page)->pages[MI_BIN_FULL], pq, page); + _mi_page_free_collect(page,false); // try to collect right away in case another thread freed just before MI_USE_DELAYED_FREE was set +} + + +// Abandon a page with used blocks at the end of a thread. +// Note: only call if it is ensured that no references exist from +// the `page->heap->thread_delayed_free` into this page. +// Currently only called through `mi_heap_collect_ex` which ensures this. +void _mi_page_abandon(mi_page_t* page, mi_page_queue_t* pq) { + mi_assert_internal(page != NULL); + mi_assert_expensive(_mi_page_is_valid(page)); + mi_assert_internal(pq == mi_page_queue_of(page)); + mi_assert_internal(mi_page_heap(page) != NULL); + + mi_heap_t* pheap = mi_page_heap(page); + + // remove from our page list + mi_segments_tld_t* segments_tld = &pheap->tld->segments; + mi_page_queue_remove(pq, page); + + // page is no longer associated with our heap + mi_assert_internal(mi_page_thread_free_flag(page)==MI_NEVER_DELAYED_FREE); + mi_page_set_heap(page, NULL); + +#if MI_DEBUG>1 + // check there are no references left.. + for (mi_block_t* block = (mi_block_t*)pheap->thread_delayed_free; block != NULL; block = mi_block_nextx(pheap, block, pheap->keys)) { + mi_assert_internal(_mi_ptr_page(block) != page); + } +#endif + + // and abandon it + mi_assert_internal(mi_page_heap(page) == NULL); + _mi_segment_page_abandon(page,segments_tld); +} + + +// Free a page with no more free blocks +void _mi_page_free(mi_page_t* page, mi_page_queue_t* pq, bool force) { + mi_assert_internal(page != NULL); + mi_assert_expensive(_mi_page_is_valid(page)); + mi_assert_internal(pq == mi_page_queue_of(page)); + mi_assert_internal(mi_page_all_free(page)); + mi_assert_internal(mi_page_thread_free_flag(page)!=MI_DELAYED_FREEING); + + // no more aligned blocks in here + mi_page_set_has_aligned(page, false); + + mi_heap_t* heap = mi_page_heap(page); + + // remove from the page list + // (no need to do _mi_heap_delayed_free first as all blocks are already free) + mi_segments_tld_t* segments_tld = &heap->tld->segments; + mi_page_queue_remove(pq, page); + + // and free it + mi_page_set_heap(page,NULL); + _mi_segment_page_free(page, force, segments_tld); +} + +// Retire parameters +#define MI_MAX_RETIRE_SIZE MI_MEDIUM_OBJ_SIZE_MAX +#define MI_RETIRE_CYCLES (8) + +// Retire a page with no more used blocks +// Important to not retire too quickly though as new +// allocations might coming. +// Note: called from `mi_free` and benchmarks often +// trigger this due to freeing everything and then +// allocating again so careful when changing this. +void _mi_page_retire(mi_page_t* page) mi_attr_noexcept { + mi_assert_internal(page != NULL); + mi_assert_expensive(_mi_page_is_valid(page)); + mi_assert_internal(mi_page_all_free(page)); + + mi_page_set_has_aligned(page, false); + + // don't retire too often.. + // (or we end up retiring and re-allocating most of the time) + // NOTE: refine this more: we should not retire if this + // is the only page left with free blocks. It is not clear + // how to check this efficiently though... + // for now, we don't retire if it is the only page left of this size class. + mi_page_queue_t* pq = mi_page_queue_of(page); + if (mi_likely(page->xblock_size <= MI_MAX_RETIRE_SIZE && !mi_page_is_in_full(page))) { + if (pq->last==page && pq->first==page) { // the only page in the queue? + mi_stat_counter_increase(_mi_stats_main.page_no_retire,1); + page->retire_expire = 1 + (page->xblock_size <= MI_SMALL_OBJ_SIZE_MAX ? MI_RETIRE_CYCLES : MI_RETIRE_CYCLES/4); + mi_heap_t* heap = mi_page_heap(page); + mi_assert_internal(pq >= heap->pages); + const size_t index = pq - heap->pages; + mi_assert_internal(index < MI_BIN_FULL && index < MI_BIN_HUGE); + if (index < heap->page_retired_min) heap->page_retired_min = index; + if (index > heap->page_retired_max) heap->page_retired_max = index; + mi_assert_internal(mi_page_all_free(page)); + return; // dont't free after all + } + } + _mi_page_free(page, pq, false); +} + +// free retired pages: we don't need to look at the entire queues +// since we only retire pages that are at the head position in a queue. +void _mi_heap_collect_retired(mi_heap_t* heap, bool force) { + size_t min = MI_BIN_FULL; + size_t max = 0; + for(size_t bin = heap->page_retired_min; bin <= heap->page_retired_max; bin++) { + mi_page_queue_t* pq = &heap->pages[bin]; + mi_page_t* page = pq->first; + if (page != NULL && page->retire_expire != 0) { + if (mi_page_all_free(page)) { + page->retire_expire--; + if (force || page->retire_expire == 0) { + _mi_page_free(pq->first, pq, force); + } + else { + // keep retired, update min/max + if (bin < min) min = bin; + if (bin > max) max = bin; + } + } + else { + page->retire_expire = 0; + } + } + } + heap->page_retired_min = min; + heap->page_retired_max = max; +} + + +/* ----------------------------------------------------------- + Initialize the initial free list in a page. + In secure mode we initialize a randomized list by + alternating between slices. +----------------------------------------------------------- */ + +#define MI_MAX_SLICE_SHIFT (6) // at most 64 slices +#define MI_MAX_SLICES (1UL << MI_MAX_SLICE_SHIFT) +#define MI_MIN_SLICES (2) + +static void mi_page_free_list_extend_secure(mi_heap_t* const heap, mi_page_t* const page, const size_t bsize, const size_t extend, mi_stats_t* const stats) { + MI_UNUSED(stats); + #if (MI_SECURE<=2) + mi_assert_internal(page->free == NULL); + mi_assert_internal(page->local_free == NULL); + #endif + mi_assert_internal(page->capacity + extend <= page->reserved); + mi_assert_internal(bsize == mi_page_block_size(page)); + void* const page_area = _mi_page_start(_mi_page_segment(page), page, NULL); + + // initialize a randomized free list + // set up `slice_count` slices to alternate between + size_t shift = MI_MAX_SLICE_SHIFT; + while ((extend >> shift) == 0) { + shift--; + } + const size_t slice_count = (size_t)1U << shift; + const size_t slice_extend = extend / slice_count; + mi_assert_internal(slice_extend >= 1); + mi_block_t* blocks[MI_MAX_SLICES]; // current start of the slice + size_t counts[MI_MAX_SLICES]; // available objects in the slice + for (size_t i = 0; i < slice_count; i++) { + blocks[i] = mi_page_block_at(page, page_area, bsize, page->capacity + i*slice_extend); + counts[i] = slice_extend; + } + counts[slice_count-1] += (extend % slice_count); // final slice holds the modulus too (todo: distribute evenly?) + + // and initialize the free list by randomly threading through them + // set up first element + const uintptr_t r = _mi_heap_random_next(heap); + size_t current = r % slice_count; + counts[current]--; + mi_block_t* const free_start = blocks[current]; + // and iterate through the rest; use `random_shuffle` for performance + uintptr_t rnd = _mi_random_shuffle(r|1); // ensure not 0 + for (size_t i = 1; i < extend; i++) { + // call random_shuffle only every INTPTR_SIZE rounds + const size_t round = i%MI_INTPTR_SIZE; + if (round == 0) rnd = _mi_random_shuffle(rnd); + // select a random next slice index + size_t next = ((rnd >> 8*round) & (slice_count-1)); + while (counts[next]==0) { // ensure it still has space + next++; + if (next==slice_count) next = 0; + } + // and link the current block to it + counts[next]--; + mi_block_t* const block = blocks[current]; + blocks[current] = (mi_block_t*)((uint8_t*)block + bsize); // bump to the following block + mi_block_set_next(page, block, blocks[next]); // and set next; note: we may have `current == next` + current = next; + } + // prepend to the free list (usually NULL) + mi_block_set_next(page, blocks[current], page->free); // end of the list + page->free = free_start; +} + +static mi_decl_noinline void mi_page_free_list_extend( mi_page_t* const page, const size_t bsize, const size_t extend, mi_stats_t* const stats) +{ + MI_UNUSED(stats); + #if (MI_SECURE <= 2) + mi_assert_internal(page->free == NULL); + mi_assert_internal(page->local_free == NULL); + #endif + mi_assert_internal(page->capacity + extend <= page->reserved); + mi_assert_internal(bsize == mi_page_block_size(page)); + void* const page_area = _mi_page_start(_mi_page_segment(page), page, NULL ); + + mi_block_t* const start = mi_page_block_at(page, page_area, bsize, page->capacity); + + // initialize a sequential free list + mi_block_t* const last = mi_page_block_at(page, page_area, bsize, page->capacity + extend - 1); + mi_block_t* block = start; + while(block <= last) { + mi_block_t* next = (mi_block_t*)((uint8_t*)block + bsize); + mi_block_set_next(page,block,next); + block = next; + } + // prepend to free list (usually `NULL`) + mi_block_set_next(page, last, page->free); + page->free = start; +} + +/* ----------------------------------------------------------- + Page initialize and extend the capacity +----------------------------------------------------------- */ + +#define MI_MAX_EXTEND_SIZE (4*1024) // heuristic, one OS page seems to work well. +#if (MI_SECURE>0) +#define MI_MIN_EXTEND (8*MI_SECURE) // extend at least by this many +#else +#define MI_MIN_EXTEND (1) +#endif + +// Extend the capacity (up to reserved) by initializing a free list +// We do at most `MI_MAX_EXTEND` to avoid touching too much memory +// Note: we also experimented with "bump" allocation on the first +// allocations but this did not speed up any benchmark (due to an +// extra test in malloc? or cache effects?) +static void mi_page_extend_free(mi_heap_t* heap, mi_page_t* page, mi_tld_t* tld) { + MI_UNUSED(tld); + mi_assert_expensive(mi_page_is_valid_init(page)); + #if (MI_SECURE<=2) + mi_assert(page->free == NULL); + mi_assert(page->local_free == NULL); + if (page->free != NULL) return; + #endif + if (page->capacity >= page->reserved) return; + + size_t page_size; + _mi_page_start(_mi_page_segment(page), page, &page_size); + mi_stat_counter_increase(tld->stats.pages_extended, 1); + + // calculate the extend count + const size_t bsize = (page->xblock_size < MI_HUGE_BLOCK_SIZE ? page->xblock_size : page_size); + size_t extend = page->reserved - page->capacity; + mi_assert_internal(extend > 0); + + size_t max_extend = (bsize >= MI_MAX_EXTEND_SIZE ? MI_MIN_EXTEND : MI_MAX_EXTEND_SIZE/(uint32_t)bsize); + if (max_extend < MI_MIN_EXTEND) { max_extend = MI_MIN_EXTEND; } + mi_assert_internal(max_extend > 0); + + if (extend > max_extend) { + // ensure we don't touch memory beyond the page to reduce page commit. + // the `lean` benchmark tests this. Going from 1 to 8 increases rss by 50%. + extend = max_extend; + } + + mi_assert_internal(extend > 0 && extend + page->capacity <= page->reserved); + mi_assert_internal(extend < (1UL<<16)); + + // and append the extend the free list + if (extend < MI_MIN_SLICES || MI_SECURE==0) { //!mi_option_is_enabled(mi_option_secure)) { + mi_page_free_list_extend(page, bsize, extend, &tld->stats ); + } + else { + mi_page_free_list_extend_secure(heap, page, bsize, extend, &tld->stats); + } + // enable the new free list + page->capacity += (uint16_t)extend; + mi_stat_increase(tld->stats.page_committed, extend * bsize); + + // extension into zero initialized memory preserves the zero'd free list + if (!page->is_zero_init) { + page->is_zero = false; + } + mi_assert_expensive(mi_page_is_valid_init(page)); +} + +// Initialize a fresh page +static void mi_page_init(mi_heap_t* heap, mi_page_t* page, size_t block_size, mi_tld_t* tld) { + mi_assert(page != NULL); + mi_segment_t* segment = _mi_page_segment(page); + mi_assert(segment != NULL); + mi_assert_internal(block_size > 0); + // set fields + mi_page_set_heap(page, heap); + page->xblock_size = (block_size < MI_HUGE_BLOCK_SIZE ? (uint32_t)block_size : MI_HUGE_BLOCK_SIZE); // initialize before _mi_segment_page_start + size_t page_size; + _mi_segment_page_start(segment, page, &page_size); + mi_assert_internal(mi_page_block_size(page) <= page_size); + mi_assert_internal(page_size <= page->slice_count*MI_SEGMENT_SLICE_SIZE); + mi_assert_internal(page_size / block_size < (1L<<16)); + page->reserved = (uint16_t)(page_size / block_size); + #ifdef MI_ENCODE_FREELIST + page->keys[0] = _mi_heap_random_next(heap); + page->keys[1] = _mi_heap_random_next(heap); + #endif + #if MI_DEBUG > 0 + page->is_zero = false; // ensure in debug mode we initialize with MI_DEBUG_UNINIT, see issue #501 + #else + page->is_zero = page->is_zero_init; + #endif + + mi_assert_internal(page->is_committed); + mi_assert_internal(!page->is_reset); + mi_assert_internal(page->capacity == 0); + mi_assert_internal(page->free == NULL); + mi_assert_internal(page->used == 0); + mi_assert_internal(page->xthread_free == 0); + mi_assert_internal(page->next == NULL); + mi_assert_internal(page->prev == NULL); + mi_assert_internal(page->retire_expire == 0); + mi_assert_internal(!mi_page_has_aligned(page)); + #if (MI_ENCODE_FREELIST) + mi_assert_internal(page->keys[0] != 0); + mi_assert_internal(page->keys[1] != 0); + #endif + mi_assert_expensive(mi_page_is_valid_init(page)); + + // initialize an initial free list + mi_page_extend_free(heap,page,tld); + mi_assert(mi_page_immediate_available(page)); +} + + +/* ----------------------------------------------------------- + Find pages with free blocks +-------------------------------------------------------------*/ + +// Find a page with free blocks of `page->block_size`. +static mi_page_t* mi_page_queue_find_free_ex(mi_heap_t* heap, mi_page_queue_t* pq, bool first_try) +{ + // search through the pages in "next fit" order + size_t count = 0; + mi_page_t* page = pq->first; + while (page != NULL) + { + mi_page_t* next = page->next; // remember next + count++; + + // 0. collect freed blocks by us and other threads + _mi_page_free_collect(page, false); + + // 1. if the page contains free blocks, we are done + if (mi_page_immediate_available(page)) { + break; // pick this one + } + + // 2. Try to extend + if (page->capacity < page->reserved) { + mi_page_extend_free(heap, page, heap->tld); + mi_assert_internal(mi_page_immediate_available(page)); + break; + } + + // 3. If the page is completely full, move it to the `mi_pages_full` + // queue so we don't visit long-lived pages too often. + mi_assert_internal(!mi_page_is_in_full(page) && !mi_page_immediate_available(page)); + mi_page_to_full(page, pq); + + page = next; + } // for each page + + mi_heap_stat_counter_increase(heap, searches, count); + + if (page == NULL) { + _mi_heap_collect_retired(heap, false); // perhaps make a page available? + page = mi_page_fresh(heap, pq); + if (page == NULL && first_try) { + // out-of-memory _or_ an abandoned page with free blocks was reclaimed, try once again + page = mi_page_queue_find_free_ex(heap, pq, false); + } + } + else { + mi_assert(pq->first == page); + page->retire_expire = 0; + } + mi_assert_internal(page == NULL || mi_page_immediate_available(page)); + return page; +} + + + +// Find a page with free blocks of `size`. +static inline mi_page_t* mi_find_free_page(mi_heap_t* heap, size_t size) { + mi_page_queue_t* pq = mi_page_queue(heap,size); + mi_page_t* page = pq->first; + if (page != NULL) { + #if (MI_SECURE>=3) // in secure mode, we extend half the time to increase randomness + if (page->capacity < page->reserved && ((_mi_heap_random_next(heap) & 1) == 1)) { + mi_page_extend_free(heap, page, heap->tld); + mi_assert_internal(mi_page_immediate_available(page)); + } + else + #endif + { + _mi_page_free_collect(page,false); + } + + if (mi_page_immediate_available(page)) { + page->retire_expire = 0; + return page; // fast path + } + } + return mi_page_queue_find_free_ex(heap, pq, true); +} + + +/* ----------------------------------------------------------- + Users can register a deferred free function called + when the `free` list is empty. Since the `local_free` + is separate this is deterministically called after + a certain number of allocations. +----------------------------------------------------------- */ + +static mi_deferred_free_fun* volatile deferred_free = NULL; +static _Atomic(void*) deferred_arg; // = NULL + +void _mi_deferred_free(mi_heap_t* heap, bool force) { + heap->tld->heartbeat++; + if (deferred_free != NULL && !heap->tld->recurse) { + heap->tld->recurse = true; + deferred_free(force, heap->tld->heartbeat, mi_atomic_load_ptr_relaxed(void,&deferred_arg)); + heap->tld->recurse = false; + } +} + +void mi_register_deferred_free(mi_deferred_free_fun* fn, void* arg) mi_attr_noexcept { + deferred_free = fn; + mi_atomic_store_ptr_release(void,&deferred_arg, arg); +} + + +/* ----------------------------------------------------------- + General allocation +----------------------------------------------------------- */ + +// Large and huge page allocation. +// Huge pages are allocated directly without being in a queue. +// Because huge pages contain just one block, and the segment contains +// just that page, we always treat them as abandoned and any thread +// that frees the block can free the whole page and segment directly. +static mi_page_t* mi_large_huge_page_alloc(mi_heap_t* heap, size_t size) { + size_t block_size = _mi_os_good_alloc_size(size); + mi_assert_internal(mi_bin(block_size) == MI_BIN_HUGE); + bool is_huge = (block_size > MI_LARGE_OBJ_SIZE_MAX); + mi_page_queue_t* pq = (is_huge ? NULL : mi_page_queue(heap, block_size)); + mi_page_t* page = mi_page_fresh_alloc(heap, pq, block_size); + if (page != NULL) { + mi_assert_internal(mi_page_immediate_available(page)); + + if (pq == NULL) { + // huge pages are directly abandoned + mi_assert_internal(_mi_page_segment(page)->kind == MI_SEGMENT_HUGE); + mi_assert_internal(_mi_page_segment(page)->used==1); + mi_assert_internal(_mi_page_segment(page)->thread_id==0); // abandoned, not in the huge queue + mi_page_set_heap(page, NULL); + } + else { + mi_assert_internal(_mi_page_segment(page)->kind != MI_SEGMENT_HUGE); + } + + const size_t bsize = mi_page_usable_block_size(page); // note: not `mi_page_block_size` to account for padding + if (bsize <= MI_LARGE_OBJ_SIZE_MAX) { + mi_heap_stat_increase(heap, large, bsize); + mi_heap_stat_counter_increase(heap, large_count, 1); + } + else { + mi_heap_stat_increase(heap, huge, bsize); + mi_heap_stat_counter_increase(heap, huge_count, 1); + } + } + return page; +} + + +// Allocate a page +// Note: in debug mode the size includes MI_PADDING_SIZE and might have overflowed. +static mi_page_t* mi_find_page(mi_heap_t* heap, size_t size) mi_attr_noexcept { + // huge allocation? + const size_t req_size = size - MI_PADDING_SIZE; // correct for padding_size in case of an overflow on `size` + if (mi_unlikely(req_size > (MI_MEDIUM_OBJ_SIZE_MAX - MI_PADDING_SIZE) )) { + if (mi_unlikely(req_size > PTRDIFF_MAX)) { // we don't allocate more than PTRDIFF_MAX (see ) + _mi_error_message(EOVERFLOW, "allocation request is too large (%zu bytes)\n", req_size); + return NULL; + } + else { + return mi_large_huge_page_alloc(heap,size); + } + } + else { + // otherwise find a page with free blocks in our size segregated queues + mi_assert_internal(size >= MI_PADDING_SIZE); + return mi_find_free_page(heap, size); + } +} + +// Generic allocation routine if the fast path (`alloc.c:mi_page_malloc`) does not succeed. +// Note: in debug mode the size includes MI_PADDING_SIZE and might have overflowed. +void* _mi_malloc_generic(mi_heap_t* heap, size_t size) mi_attr_noexcept +{ + mi_assert_internal(heap != NULL); + + // initialize if necessary + if (mi_unlikely(!mi_heap_is_initialized(heap))) { + mi_thread_init(); // calls `_mi_heap_init` in turn + heap = mi_get_default_heap(); + if (mi_unlikely(!mi_heap_is_initialized(heap))) { return NULL; } + } + mi_assert_internal(mi_heap_is_initialized(heap)); + + // call potential deferred free routines + _mi_deferred_free(heap, false); + + // free delayed frees from other threads + _mi_heap_delayed_free(heap); + + // find (or allocate) a page of the right size + mi_page_t* page = mi_find_page(heap, size); + if (mi_unlikely(page == NULL)) { // first time out of memory, try to collect and retry the allocation once more + mi_heap_collect(heap, true /* force */); + page = mi_find_page(heap, size); + } + + if (mi_unlikely(page == NULL)) { // out of memory + const size_t req_size = size - MI_PADDING_SIZE; // correct for padding_size in case of an overflow on `size` + _mi_error_message(ENOMEM, "unable to allocate memory (%zu bytes)\n", req_size); + return NULL; + } + + mi_assert_internal(mi_page_immediate_available(page)); + mi_assert_internal(mi_page_block_size(page) >= size); + + // and try again, this time succeeding! (i.e. this should never recurse) + return _mi_page_malloc(heap, page, size); +} diff --git a/compat/mimalloc/random.c b/compat/mimalloc/random.c new file mode 100644 index 00000000000000..4e334d3dd76ebd --- /dev/null +++ b/compat/mimalloc/random.c @@ -0,0 +1,373 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2019-2021, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ +#ifndef _DEFAULT_SOURCE +#define _DEFAULT_SOURCE // for syscall() on Linux +#endif + +#include "mimalloc.h" +#include "mimalloc-internal.h" + +#include // memset + +/* ---------------------------------------------------------------------------- +We use our own PRNG to keep predictable performance of random number generation +and to avoid implementations that use a lock. We only use the OS provided +random source to initialize the initial seeds. Since we do not need ultimate +performance but we do rely on the security (for secret cookies in secure mode) +we use a cryptographically secure generator (chacha20). +-----------------------------------------------------------------------------*/ + +#define MI_CHACHA_ROUNDS (20) // perhaps use 12 for better performance? + + +/* ---------------------------------------------------------------------------- +Chacha20 implementation as the original algorithm with a 64-bit nonce +and counter: https://en.wikipedia.org/wiki/Salsa20 +The input matrix has sixteen 32-bit values: +Position 0 to 3: constant key +Position 4 to 11: the key +Position 12 to 13: the counter. +Position 14 to 15: the nonce. + +The implementation uses regular C code which compiles very well on modern compilers. +(gcc x64 has no register spills, and clang 6+ uses SSE instructions) +-----------------------------------------------------------------------------*/ + +static inline uint32_t rotl(uint32_t x, uint32_t shift) { + return (x << shift) | (x >> (32 - shift)); +} + +static inline void qround(uint32_t x[16], size_t a, size_t b, size_t c, size_t d) { + x[a] += x[b]; x[d] = rotl(x[d] ^ x[a], 16); + x[c] += x[d]; x[b] = rotl(x[b] ^ x[c], 12); + x[a] += x[b]; x[d] = rotl(x[d] ^ x[a], 8); + x[c] += x[d]; x[b] = rotl(x[b] ^ x[c], 7); +} + +static void chacha_block(mi_random_ctx_t* ctx) +{ + // scramble into `x` + uint32_t x[16]; + for (size_t i = 0; i < 16; i++) { + x[i] = ctx->input[i]; + } + for (size_t i = 0; i < MI_CHACHA_ROUNDS; i += 2) { + qround(x, 0, 4, 8, 12); + qround(x, 1, 5, 9, 13); + qround(x, 2, 6, 10, 14); + qround(x, 3, 7, 11, 15); + qround(x, 0, 5, 10, 15); + qround(x, 1, 6, 11, 12); + qround(x, 2, 7, 8, 13); + qround(x, 3, 4, 9, 14); + } + + // add scrambled data to the initial state + for (size_t i = 0; i < 16; i++) { + ctx->output[i] = x[i] + ctx->input[i]; + } + ctx->output_available = 16; + + // increment the counter for the next round + ctx->input[12] += 1; + if (ctx->input[12] == 0) { + ctx->input[13] += 1; + if (ctx->input[13] == 0) { // and keep increasing into the nonce + ctx->input[14] += 1; + } + } +} + +static uint32_t chacha_next32(mi_random_ctx_t* ctx) { + if (ctx->output_available <= 0) { + chacha_block(ctx); + ctx->output_available = 16; // (assign again to suppress static analysis warning) + } + const uint32_t x = ctx->output[16 - ctx->output_available]; + ctx->output[16 - ctx->output_available] = 0; // reset once the data is handed out + ctx->output_available--; + return x; +} + +static inline uint32_t read32(const uint8_t* p, size_t idx32) { + const size_t i = 4*idx32; + return ((uint32_t)p[i+0] | (uint32_t)p[i+1] << 8 | (uint32_t)p[i+2] << 16 | (uint32_t)p[i+3] << 24); +} + +static void chacha_init(mi_random_ctx_t* ctx, const uint8_t key[32], uint64_t nonce) +{ + // since we only use chacha for randomness (and not encryption) we + // do not _need_ to read 32-bit values as little endian but we do anyways + // just for being compatible :-) + memset(ctx, 0, sizeof(*ctx)); + for (size_t i = 0; i < 4; i++) { + const uint8_t* sigma = (uint8_t*)"expand 32-byte k"; + ctx->input[i] = read32(sigma,i); + } + for (size_t i = 0; i < 8; i++) { + ctx->input[i + 4] = read32(key,i); + } + ctx->input[12] = 0; + ctx->input[13] = 0; + ctx->input[14] = (uint32_t)nonce; + ctx->input[15] = (uint32_t)(nonce >> 32); +} + +static void chacha_split(mi_random_ctx_t* ctx, uint64_t nonce, mi_random_ctx_t* ctx_new) { + memset(ctx_new, 0, sizeof(*ctx_new)); + _mi_memcpy(ctx_new->input, ctx->input, sizeof(ctx_new->input)); + ctx_new->input[12] = 0; + ctx_new->input[13] = 0; + ctx_new->input[14] = (uint32_t)nonce; + ctx_new->input[15] = (uint32_t)(nonce >> 32); + mi_assert_internal(ctx->input[14] != ctx_new->input[14] || ctx->input[15] != ctx_new->input[15]); // do not reuse nonces! + chacha_block(ctx_new); +} + + +/* ---------------------------------------------------------------------------- +Random interface +-----------------------------------------------------------------------------*/ + +#if MI_DEBUG>1 +static bool mi_random_is_initialized(mi_random_ctx_t* ctx) { + return (ctx != NULL && ctx->input[0] != 0); +} +#endif + +void _mi_random_split(mi_random_ctx_t* ctx, mi_random_ctx_t* ctx_new) { + mi_assert_internal(mi_random_is_initialized(ctx)); + mi_assert_internal(ctx != ctx_new); + chacha_split(ctx, (uintptr_t)ctx_new /*nonce*/, ctx_new); +} + +uintptr_t _mi_random_next(mi_random_ctx_t* ctx) { + mi_assert_internal(mi_random_is_initialized(ctx)); + #if MI_INTPTR_SIZE <= 4 + return chacha_next32(ctx); + #elif MI_INTPTR_SIZE == 8 + return (((uintptr_t)chacha_next32(ctx) << 32) | chacha_next32(ctx)); + #else + # error "define mi_random_next for this platform" + #endif +} + + +/* ---------------------------------------------------------------------------- +To initialize a fresh random context we rely on the OS: +- Windows : BCryptGenRandom (or RtlGenRandom) +- macOS : CCRandomGenerateBytes, arc4random_buf +- bsd,wasi : arc4random_buf +- Linux : getrandom,/dev/urandom +If we cannot get good randomness, we fall back to weak randomness based on a timer and ASLR. +-----------------------------------------------------------------------------*/ + +#if defined(_WIN32) + +#if defined(MI_USE_RTLGENRANDOM) || defined(__cplusplus) +// We prefer to use BCryptGenRandom instead of (the unofficial) RtlGenRandom but when using +// dynamic overriding, we observed it can raise an exception when compiled with C++, and +// sometimes deadlocks when also running under the VS debugger. +#pragma comment (lib,"advapi32.lib") +#define RtlGenRandom SystemFunction036 +#ifdef __cplusplus +extern "C" { +#endif +BOOLEAN NTAPI RtlGenRandom(PVOID RandomBuffer, ULONG RandomBufferLength); +#ifdef __cplusplus +} +#endif +static bool os_random_buf(void* buf, size_t buf_len) { + return (RtlGenRandom(buf, (ULONG)buf_len) != 0); +} +#else +#include "compat/win32/lazyload.h" +#ifndef BCRYPT_USE_SYSTEM_PREFERRED_RNG +#define BCRYPT_USE_SYSTEM_PREFERRED_RNG 0x00000002 +#endif + +static bool os_random_buf(void* buf, size_t buf_len) { + DECLARE_PROC_ADDR(bcrypt, LONG, NTAPI, BCryptGenRandom, HANDLE, PUCHAR, ULONG, ULONG); + if (!INIT_PROC_ADDR(BCryptGenRandom)) + return 0; + return (BCryptGenRandom(NULL, (PUCHAR)buf, (ULONG)buf_len, BCRYPT_USE_SYSTEM_PREFERRED_RNG) >= 0); +} +#endif + +#elif defined(__APPLE__) +#include +#if defined(MAC_OS_X_VERSION_10_10) && MAC_OS_X_VERSION_MAX_ALLOWED >= MAC_OS_X_VERSION_10_10 +#include +#include +#endif +static bool os_random_buf(void* buf, size_t buf_len) { + #if defined(MAC_OS_X_VERSION_10_15) && MAC_OS_X_VERSION_MAX_ALLOWED >= MAC_OS_X_VERSION_10_15 + // We prefere CCRandomGenerateBytes as it returns an error code while arc4random_buf + // may fail silently on macOS. See PR #390, and + return (CCRandomGenerateBytes(buf, buf_len) == kCCSuccess); + #else + // fall back on older macOS + arc4random_buf(buf, buf_len); + return true; + #endif +} + +#elif defined(__ANDROID__) || defined(__DragonFly__) || \ + defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) || \ + defined(__sun) // todo: what to use with __wasi__? +#include +static bool os_random_buf(void* buf, size_t buf_len) { + arc4random_buf(buf, buf_len); + return true; +} +#elif defined(__linux__) || defined(__HAIKU__) +#if defined(__linux__) +#include +#endif +#include +#include +#include +#include +#include +static bool os_random_buf(void* buf, size_t buf_len) { + // Modern Linux provides `getrandom` but different distributions either use `sys/random.h` or `linux/random.h` + // and for the latter the actual `getrandom` call is not always defined. + // (see ) + // We therefore use a syscall directly and fall back dynamically to /dev/urandom when needed. +#ifdef SYS_getrandom + #ifndef GRND_NONBLOCK + #define GRND_NONBLOCK (1) + #endif + static _Atomic(uintptr_t) no_getrandom; // = 0 + if (mi_atomic_load_acquire(&no_getrandom)==0) { + ssize_t ret = syscall(SYS_getrandom, buf, buf_len, GRND_NONBLOCK); + if (ret >= 0) return (buf_len == (size_t)ret); + if (errno != ENOSYS) return false; + mi_atomic_store_release(&no_getrandom, 1UL); // don't call again, and fall back to /dev/urandom + } +#endif + int flags = O_RDONLY; + #if defined(O_CLOEXEC) + flags |= O_CLOEXEC; + #endif + int fd = open("/dev/urandom", flags, 0); + if (fd < 0) return false; + size_t count = 0; + while(count < buf_len) { + ssize_t ret = read(fd, (char*)buf + count, buf_len - count); + if (ret<=0) { + if (errno!=EAGAIN && errno!=EINTR) break; + } + else { + count += ret; + } + } + close(fd); + return (count==buf_len); +} +#else +static bool os_random_buf(void* buf, size_t buf_len) { + return false; +} +#endif + +#if defined(_WIN32) +#include +#elif defined(__APPLE__) +#include +#else +#include +#endif + +uintptr_t _mi_os_random_weak(uintptr_t extra_seed) { + uintptr_t x = (uintptr_t)&_mi_os_random_weak ^ extra_seed; // ASLR makes the address random + + #if defined(_WIN32) + LARGE_INTEGER pcount; + QueryPerformanceCounter(&pcount); + x ^= (uintptr_t)(pcount.QuadPart); + #elif defined(__APPLE__) + x ^= (uintptr_t)mach_absolute_time(); + #else + struct timespec time; + clock_gettime(CLOCK_MONOTONIC, &time); + x ^= (uintptr_t)time.tv_sec; + x ^= (uintptr_t)time.tv_nsec; + #endif + // and do a few randomization steps + uintptr_t max = ((x ^ (x >> 17)) & 0x0F) + 1; + for (uintptr_t i = 0; i < max; i++) { + x = _mi_random_shuffle(x); + } + mi_assert_internal(x != 0); + return x; +} + +void _mi_random_init(mi_random_ctx_t* ctx) { + uint8_t key[32]; + if (!os_random_buf(key, sizeof(key))) { + // if we fail to get random data from the OS, we fall back to a + // weak random source based on the current time + #if !defined(__wasi__) + _mi_warning_message("unable to use secure randomness\n"); + #endif + uintptr_t x = _mi_os_random_weak(0); + for (size_t i = 0; i < 8; i++) { // key is eight 32-bit words. + x = _mi_random_shuffle(x); + ((uint32_t*)key)[i] = (uint32_t)x; + } + } + chacha_init(ctx, key, (uintptr_t)ctx /*nonce*/ ); +} + +/* -------------------------------------------------------- +test vectors from +----------------------------------------------------------- */ +/* +static bool array_equals(uint32_t* x, uint32_t* y, size_t n) { + for (size_t i = 0; i < n; i++) { + if (x[i] != y[i]) return false; + } + return true; +} +static void chacha_test(void) +{ + uint32_t x[4] = { 0x11111111, 0x01020304, 0x9b8d6f43, 0x01234567 }; + uint32_t x_out[4] = { 0xea2a92f4, 0xcb1cf8ce, 0x4581472e, 0x5881c4bb }; + qround(x, 0, 1, 2, 3); + mi_assert_internal(array_equals(x, x_out, 4)); + + uint32_t y[16] = { + 0x879531e0, 0xc5ecf37d, 0x516461b1, 0xc9a62f8a, + 0x44c20ef3, 0x3390af7f, 0xd9fc690b, 0x2a5f714c, + 0x53372767, 0xb00a5631, 0x974c541a, 0x359e9963, + 0x5c971061, 0x3d631689, 0x2098d9d6, 0x91dbd320 }; + uint32_t y_out[16] = { + 0x879531e0, 0xc5ecf37d, 0xbdb886dc, 0xc9a62f8a, + 0x44c20ef3, 0x3390af7f, 0xd9fc690b, 0xcfacafd2, + 0xe46bea80, 0xb00a5631, 0x974c541a, 0x359e9963, + 0x5c971061, 0xccc07c79, 0x2098d9d6, 0x91dbd320 }; + qround(y, 2, 7, 8, 13); + mi_assert_internal(array_equals(y, y_out, 16)); + + mi_random_ctx_t r = { + { 0x61707865, 0x3320646e, 0x79622d32, 0x6b206574, + 0x03020100, 0x07060504, 0x0b0a0908, 0x0f0e0d0c, + 0x13121110, 0x17161514, 0x1b1a1918, 0x1f1e1d1c, + 0x00000001, 0x09000000, 0x4a000000, 0x00000000 }, + {0}, + 0 + }; + uint32_t r_out[16] = { + 0xe4e7f110, 0x15593bd1, 0x1fdd0f50, 0xc47120a3, + 0xc7f4d1c7, 0x0368c033, 0x9aaa2204, 0x4e6cd4c3, + 0x466482d2, 0x09aa9f07, 0x05d7c214, 0xa2028bd9, + 0xd19c12b5, 0xb94e16de, 0xe883d0cb, 0x4e3c50a2 }; + chacha_block(&r); + mi_assert_internal(array_equals(r.output, r_out, 16)); +} +*/ diff --git a/compat/mimalloc/readme.md b/compat/mimalloc/readme.md new file mode 100644 index 00000000000000..0db3ff6f112ca0 --- /dev/null +++ b/compat/mimalloc/readme.md @@ -0,0 +1,715 @@ + + + +[](https://dev.azure.com/Daan0324/mimalloc/_build?definitionId=1&_a=summary) + +# mimalloc + +  + +mimalloc (pronounced "me-malloc") +is a general purpose allocator with excellent [performance](#performance) characteristics. +Initially developed by Daan Leijen for the run-time systems of the +[Koka](https://koka-lang.github.io) and [Lean](https://github.com/leanprover/lean) languages. + +Latest release tag: `v2.0.6` (2022-04-14). +Latest stable tag: `v1.7.6` (2022-02-14). + +mimalloc is a drop-in replacement for `malloc` and can be used in other programs +without code changes, for example, on dynamically linked ELF-based systems (Linux, BSD, etc.) you can use it as: +``` +> LD_PRELOAD=/usr/lib/libmimalloc.so myprogram +``` +It also has an easy way to override the default allocator in [Windows](#override_on_windows). Notable aspects of the design include: + +- __small and consistent__: the library is about 8k LOC using simple and + consistent data structures. This makes it very suitable + to integrate and adapt in other projects. For runtime systems it + provides hooks for a monotonic _heartbeat_ and deferred freeing (for + bounded worst-case times with reference counting). +- __free list sharding__: instead of one big free list (per size class) we have + many smaller lists per "mimalloc page" which reduces fragmentation and + increases locality -- + things that are allocated close in time get allocated close in memory. + (A mimalloc page contains blocks of one size class and is usually 64KiB on a 64-bit system). +- __free list multi-sharding__: the big idea! Not only do we shard the free list + per mimalloc page, but for each page we have multiple free lists. In particular, there + is one list for thread-local `free` operations, and another one for concurrent `free` + operations. Free-ing from another thread can now be a single CAS without needing + sophisticated coordination between threads. Since there will be + thousands of separate free lists, contention is naturally distributed over the heap, + and the chance of contending on a single location will be low -- this is quite + similar to randomized algorithms like skip lists where adding + a random oracle removes the need for a more complex algorithm. +- __eager page reset__: when a "page" becomes empty (with increased chance + due to free list sharding) the memory is marked to the OS as unused ("reset" or "purged") + reducing (real) memory pressure and fragmentation, especially in long running + programs. +- __secure__: _mimalloc_ can be built in secure mode, adding guard pages, + randomized allocation, encrypted free lists, etc. to protect against various + heap vulnerabilities. The performance penalty is usually around 10% on average + over our benchmarks. +- __first-class heaps__: efficiently create and use multiple heaps to allocate across different regions. + A heap can be destroyed at once instead of deallocating each object separately. +- __bounded__: it does not suffer from _blowup_ \[1\], has bounded worst-case allocation + times (_wcat_), bounded space overhead (~0.2% meta-data, with low internal fragmentation), + and has no internal points of contention using only atomic operations. +- __fast__: In our benchmarks (see [below](#performance)), + _mimalloc_ outperforms other leading allocators (_jemalloc_, _tcmalloc_, _Hoard_, etc), + and often uses less memory. A nice property + is that it does consistently well over a wide range of benchmarks. There is also good huge OS page + support for larger server programs. + +The [documentation](https://microsoft.github.io/mimalloc) gives a full overview of the API. +You can read more on the design of _mimalloc_ in the [technical report](https://www.microsoft.com/en-us/research/publication/mimalloc-free-list-sharding-in-action) which also has detailed benchmark results. + +Enjoy! + +### Branches + +* `master`: latest stable release (based on `dev-slice`). +* `dev`: development branch for mimalloc v1. Use this branch for submitting PR's. +* `dev-slice`: development branch for mimalloc v2. This branch is downstream of `dev`. + +### Releases + +Note: the `v2.x` version has a new algorithm for managing internal mimalloc pages that tends to use reduce memory usage + and fragmentation compared to mimalloc `v1.x` (especially for large workloads). Should otherwise have similar performance + (see [below](#performance)); please report if you observe any significant performance regression. + +* 2022-04-14, `v1.7.6`, `v2.0.6`: fix fallback path for aligned OS allocation on Windows, improve Windows aligned allocation + even when compiling with older SDK's, fix dynamic overriding on macOS Monterey, fix MSVC C++ dynamic overriding, fix + warnings under Clang 14, improve performance if many OS threads are created and destroyed, fix statistics for large object + allocations, using MIMALLOC_VERBOSE=1 has no maximum on the number of error messages, various small fixes. + +* 2022-02-14, `v1.7.5`, `v2.0.5` (alpha): fix malloc override on + Windows 11, fix compilation with musl, potentially reduced + committed memory, add `bin/minject` for Windows, + improved wasm support, faster aligned allocation, + various small fixes. + +* 2021-11-14, `v1.7.3`, `v2.0.3` (beta): improved WASM support, improved macOS support and performance (including + M1), improved performance for v2 for large objects, Python integration improvements, more standard + installation directories, various small fixes. + +* 2021-06-17, `v1.7.2`, `v2.0.2` (beta): support M1, better installation layout on Linux, fix + thread_id on Android, prefer 2-6TiB area for aligned allocation to work better on pre-windows 8, various small fixes. + +* 2021-04-06, `v1.7.1`, `v2.0.1` (beta): fix bug in arena allocation for huge pages, improved aslr on large allocations, initial M1 support (still experimental). + +* 2021-01-31, `v2.0.0`: beta release 2.0: new slice algorithm for managing internal mimalloc pages. + +* 2021-01-31, `v1.7.0`: stable release 1.7: support explicit user provided memory regions, more precise statistics, + improve macOS overriding, initial support for Apple M1, improved DragonFly support, faster memcpy on Windows, various small fixes. + +* [Older release notes](#older-release-notes) + +Special thanks to: + +* [David Carlier](https://devnexen.blogspot.com/) (@devnexen) for his many contributions, and making + mimalloc work better on many less common operating systems, like Haiku, Dragonfly, etc. +* Mary Feofanova (@mary3000), Evgeniy Moiseenko, and Manuel Pöter (@mpoeter) for making mimalloc TSAN checkable, and finding + memory model bugs using the [genMC] model checker. +* Weipeng Liu (@pongba), Zhuowei Li, Junhua Wang, and Jakub Szymanski, for their early support of mimalloc and deployment + at large scale services, leading to many improvements in the mimalloc algorithms for large workloads. +* Jason Gibson (@jasongibson) for exhaustive testing on large scale workloads and server environments, and finding complex bugs + in (early versions of) `mimalloc`. +* Manuel Pöter (@mpoeter) and Sam Gross(@colesbury) for finding an ABA concurrency issue in abandoned segment reclamation. Sam also created the [no GIL](https://github.com/colesbury/nogil) Python fork which + uses mimalloc internally. + + +[genMC]: https://plv.mpi-sws.org/genmc/ + +### Usage + +mimalloc is used in various large scale low-latency services and programs, for example: + + + + + + + + +# Building + +## Windows + +Open `ide/vs2019/mimalloc.sln` in Visual Studio 2019 and build. +The `mimalloc` project builds a static library (in `out/msvc-x64`), while the +`mimalloc-override` project builds a DLL for overriding malloc +in the entire program. + +## macOS, Linux, BSD, etc. + +We use [`cmake`](https://cmake.org)1 as the build system: + +``` +> mkdir -p out/release +> cd out/release +> cmake ../.. +> make +``` +This builds the library as a shared (dynamic) +library (`.so` or `.dylib`), a static library (`.a`), and +as a single object file (`.o`). + +`> sudo make install` (install the library and header files in `/usr/local/lib` and `/usr/local/include`) + +You can build the debug version which does many internal checks and +maintains detailed statistics as: + +``` +> mkdir -p out/debug +> cd out/debug +> cmake -DCMAKE_BUILD_TYPE=Debug ../.. +> make +``` +This will name the shared library as `libmimalloc-debug.so`. + +Finally, you can build a _secure_ version that uses guard pages, encrypted +free lists, etc., as: +``` +> mkdir -p out/secure +> cd out/secure +> cmake -DMI_SECURE=ON ../.. +> make +``` +This will name the shared library as `libmimalloc-secure.so`. +Use `ccmake`2 instead of `cmake` +to see and customize all the available build options. + +Notes: +1. Install CMake: `sudo apt-get install cmake` +2. Install CCMake: `sudo apt-get install cmake-curses-gui` + + +## Single source + +You can also directly build the single `src/static.c` file as part of your project without +needing `cmake` at all. Make sure to also add the mimalloc `include` directory to the include path. + + +# Using the library + +The preferred usage is including ``, linking with +the shared- or static library, and using the `mi_malloc` API exclusively for allocation. For example, +``` +> gcc -o myprogram -lmimalloc myfile.c +``` + +mimalloc uses only safe OS calls (`mmap` and `VirtualAlloc`) and can co-exist +with other allocators linked to the same program. +If you use `cmake`, you can simply use: +``` +find_package(mimalloc 1.4 REQUIRED) +``` +in your `CMakeLists.txt` to find a locally installed mimalloc. Then use either: +``` +target_link_libraries(myapp PUBLIC mimalloc) +``` +to link with the shared (dynamic) library, or: +``` +target_link_libraries(myapp PUBLIC mimalloc-static) +``` +to link with the static library. See `test\CMakeLists.txt` for an example. + +For best performance in C++ programs, it is also recommended to override the +global `new` and `delete` operators. For convience, mimalloc provides +[`mimalloc-new-delete.h`](https://github.com/microsoft/mimalloc/blob/master/include/mimalloc-new-delete.h) which does this for you -- just include it in a single(!) source file in your project. +In C++, mimalloc also provides the `mi_stl_allocator` struct which implements the `std::allocator` +interface. + +You can pass environment variables to print verbose messages (`MIMALLOC_VERBOSE=1`) +and statistics (`MIMALLOC_SHOW_STATS=1`) (in the debug version): +``` +> env MIMALLOC_SHOW_STATS=1 ./cfrac 175451865205073170563711388363 + +175451865205073170563711388363 = 374456281610909315237213 * 468551 + +heap stats: peak total freed unit +normal 2: 16.4 kb 17.5 mb 17.5 mb 16 b ok +normal 3: 16.3 kb 15.2 mb 15.2 mb 24 b ok +normal 4: 64 b 4.6 kb 4.6 kb 32 b ok +normal 5: 80 b 118.4 kb 118.4 kb 40 b ok +normal 6: 48 b 48 b 48 b 48 b ok +normal 17: 960 b 960 b 960 b 320 b ok + +heap stats: peak total freed unit + normal: 33.9 kb 32.8 mb 32.8 mb 1 b ok + huge: 0 b 0 b 0 b 1 b ok + total: 33.9 kb 32.8 mb 32.8 mb 1 b ok +malloc requested: 32.8 mb + + committed: 58.2 kb 58.2 kb 58.2 kb 1 b ok + reserved: 2.0 mb 2.0 mb 2.0 mb 1 b ok + reset: 0 b 0 b 0 b 1 b ok + segments: 1 1 1 +-abandoned: 0 + pages: 6 6 6 +-abandoned: 0 + mmaps: 3 + mmap fast: 0 + mmap slow: 1 + threads: 0 + elapsed: 2.022s + process: user: 1.781s, system: 0.016s, faults: 756, reclaims: 0, rss: 2.7 mb +``` + +The above model of using the `mi_` prefixed API is not always possible +though in existing programs that already use the standard malloc interface, +and another option is to override the standard malloc interface +completely and redirect all calls to the _mimalloc_ library instead . + +## Environment Options + +You can set further options either programmatically (using [`mi_option_set`](https://microsoft.github.io/mimalloc/group__options.html)), +or via environment variables: + +- `MIMALLOC_SHOW_STATS=1`: show statistics when the program terminates. +- `MIMALLOC_VERBOSE=1`: show verbose messages. +- `MIMALLOC_SHOW_ERRORS=1`: show error and warning messages. +- `MIMALLOC_PAGE_RESET=0`: by default, mimalloc will reset (or purge) OS pages that are not in use, to signal to the OS + that the underlying physical memory can be reused. This can reduce memory fragmentation in long running (server) + programs. By setting it to `0` this will no longer be done which can improve performance for batch-like programs. + As an alternative, the `MIMALLOC_RESET_DELAY=` can be set higher (100ms by default) to make the page + reset occur less frequently instead of turning it off completely. +- `MIMALLOC_USE_NUMA_NODES=N`: pretend there are at most `N` NUMA nodes. If not set, the actual NUMA nodes are detected + at runtime. Setting `N` to 1 may avoid problems in some virtual environments. Also, setting it to a lower number than + the actual NUMA nodes is fine and will only cause threads to potentially allocate more memory across actual NUMA + nodes (but this can happen in any case as NUMA local allocation is always a best effort but not guaranteed). +- `MIMALLOC_LARGE_OS_PAGES=1`: use large OS pages (2MiB) when available; for some workloads this can significantly + improve performance. Use `MIMALLOC_VERBOSE` to check if the large OS pages are enabled -- usually one needs + to explicitly allow large OS pages (as on [Windows][windows-huge] and [Linux][linux-huge]). However, sometimes + the OS is very slow to reserve contiguous physical memory for large OS pages so use with care on systems that + can have fragmented memory (for that reason, we generally recommend to use `MIMALLOC_RESERVE_HUGE_OS_PAGES` instead whenever possible). + +- `MIMALLOC_RESERVE_HUGE_OS_PAGES=N`: where N is the number of 1GiB _huge_ OS pages. This reserves the huge pages at + startup and sometimes this can give a large (latency) performance improvement on big workloads. + Usually it is better to not use + `MIMALLOC_LARGE_OS_PAGES` in combination with this setting. Just like large OS pages, use with care as reserving + contiguous physical memory can take a long time when memory is fragmented (but reserving the huge pages is done at + startup only once). + Note that we usually need to explicitly enable huge OS pages (as on [Windows][windows-huge] and [Linux][linux-huge])). + With huge OS pages, it may be beneficial to set the setting + `MIMALLOC_EAGER_COMMIT_DELAY=N` (`N` is 1 by default) to delay the initial `N` segments (of 4MiB) + of a thread to not allocate in the huge OS pages; this prevents threads that are short lived + and allocate just a little to take up space in the huge OS page area (which cannot be reset). + The huge pages are usually allocated evenly among NUMA nodes. + We can use `MIMALLOC_RESERVE_HUGE_OS_PAGES_AT=N` where `N` is the numa node (starting at 0) to allocate all + the huge pages at a specific numa node instead. + +Use caution when using `fork` in combination with either large or huge OS pages: on a fork, the OS uses copy-on-write +for all pages in the original process including the huge OS pages. When any memory is now written in that area, the +OS will copy the entire 1GiB huge page (or 2MiB large page) which can cause the memory usage to grow in large increments. + +[linux-huge]: https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/5/html/tuning_and_optimizing_red_hat_enterprise_linux_for_oracle_9i_and_10g_databases/sect-oracle_9i_and_10g_tuning_guide-large_memory_optimization_big_pages_and_huge_pages-configuring_huge_pages_in_red_hat_enterprise_linux_4_or_5 +[windows-huge]: https://docs.microsoft.com/en-us/sql/database-engine/configure-windows/enable-the-lock-pages-in-memory-option-windows?view=sql-server-2017 + +## Secure Mode + +_mimalloc_ can be build in secure mode by using the `-DMI_SECURE=ON` flags in `cmake`. This build enables various mitigations +to make mimalloc more robust against exploits. In particular: + +- All internal mimalloc pages are surrounded by guard pages and the heap metadata is behind a guard page as well (so a buffer overflow + exploit cannot reach into the metadata). +- All free list pointers are + [encoded](https://github.com/microsoft/mimalloc/blob/783e3377f79ee82af43a0793910a9f2d01ac7863/include/mimalloc-internal.h#L396) + with per-page keys which is used both to prevent overwrites with a known pointer, as well as to detect heap corruption. +- Double free's are detected (and ignored). +- The free lists are initialized in a random order and allocation randomly chooses between extension and reuse within a page to + mitigate against attacks that rely on a predicable allocation order. Similarly, the larger heap blocks allocated by mimalloc + from the OS are also address randomized. + +As always, evaluate with care as part of an overall security strategy as all of the above are mitigations but not guarantees. + +## Debug Mode + +When _mimalloc_ is built using debug mode, various checks are done at runtime to catch development errors. + +- Statistics are maintained in detail for each object size. They can be shown using `MIMALLOC_SHOW_STATS=1` at runtime. +- All objects have padding at the end to detect (byte precise) heap block overflows. +- Double free's, and freeing invalid heap pointers are detected. +- Corrupted free-lists and some forms of use-after-free are detected. + + +# Overriding Standard Malloc + +Overriding the standard `malloc` (and `new`) can be done either _dynamically_ or _statically_. + +## Dynamic override + +This is the recommended way to override the standard malloc interface. + +### Override on Linux, BSD + +On these ELF-based systems we preload the mimalloc shared +library so all calls to the standard `malloc` interface are +resolved to the _mimalloc_ library. +``` +> env LD_PRELOAD=/usr/lib/libmimalloc.so myprogram +``` + +You can set extra environment variables to check that mimalloc is running, +like: +``` +> env MIMALLOC_VERBOSE=1 LD_PRELOAD=/usr/lib/libmimalloc.so myprogram +``` +or run with the debug version to get detailed statistics: +``` +> env MIMALLOC_SHOW_STATS=1 LD_PRELOAD=/usr/lib/libmimalloc-debug.so myprogram +``` + +### Override on MacOS + +On macOS we can also preload the mimalloc shared +library so all calls to the standard `malloc` interface are +resolved to the _mimalloc_ library. +``` +> env DYLD_INSERT_LIBRARIES=/usr/lib/libmimalloc.dylib myprogram +``` + +Note that certain security restrictions may apply when doing this from +the [shell](https://stackoverflow.com/questions/43941322/dyld-insert-libraries-ignored-when-calling-application-through-bash). + + +### Override on Windows + +Overriding on Windows is robust and has the +particular advantage to be able to redirect all malloc/free calls that go through +the (dynamic) C runtime allocator, including those from other DLL's or libraries. + +The overriding on Windows requires that you link your program explicitly with +the mimalloc DLL and use the C-runtime library as a DLL (using the `/MD` or `/MDd` switch). +Also, the `mimalloc-redirect.dll` (or `mimalloc-redirect32.dll`) must be put +in the same folder as the main `mimalloc-override.dll` at runtime (as it is a dependency). +The redirection DLL ensures that all calls to the C runtime malloc API get redirected to +mimalloc (in `mimalloc-override.dll`). + +To ensure the mimalloc DLL is loaded at run-time it is easiest to insert some +call to the mimalloc API in the `main` function, like `mi_version()` +(or use the `/INCLUDE:mi_version` switch on the linker). See the `mimalloc-override-test` project +for an example on how to use this. For best performance on Windows with C++, it +is also recommended to also override the `new`/`delete` operations (by including +[`mimalloc-new-delete.h`](https://github.com/microsoft/mimalloc/blob/master/include/mimalloc-new-delete.h) a single(!) source file in your project). + +The environment variable `MIMALLOC_DISABLE_REDIRECT=1` can be used to disable dynamic +overriding at run-time. Use `MIMALLOC_VERBOSE=1` to check if mimalloc was successfully redirected. + +(Note: in principle, it is possible to even patch existing executables without any recompilation +if they are linked with the dynamic C runtime (`ucrtbase.dll`) -- just put the `mimalloc-override.dll` +into the import table (and put `mimalloc-redirect.dll` in the same folder) +Such patching can be done for example with [CFF Explorer](https://ntcore.com/?page_id=388)). + + +## Static override + +On Unix-like systems, you can also statically link with _mimalloc_ to override the standard +malloc interface. The recommended way is to link the final program with the +_mimalloc_ single object file (`mimalloc-override.o`). We use +an object file instead of a library file as linkers give preference to +that over archives to resolve symbols. To ensure that the standard +malloc interface resolves to the _mimalloc_ library, link it as the first +object file. For example: +``` +> gcc -o myprogram mimalloc-override.o myfile1.c ... +``` + +Another way to override statically that works on all platforms, is to +link statically to mimalloc (as shown in the introduction) and include a +header file in each source file that re-defines `malloc` etc. to `mi_malloc`. +This is provided by [`mimalloc-override.h`](https://github.com/microsoft/mimalloc/blob/master/include/mimalloc-override.h). This only works reliably though if all sources are +under your control or otherwise mixing of pointers from different heaps may occur! + + +# Performance + +Last update: 2021-01-30 + +We tested _mimalloc_ against many other top allocators over a wide +range of benchmarks, ranging from various real world programs to +synthetic benchmarks that see how the allocator behaves under more +extreme circumstances. In our benchmark suite, _mimalloc_ outperforms other leading +allocators (_jemalloc_, _tcmalloc_, _Hoard_, etc), and has a similar memory footprint. A nice property is that it +does consistently well over the wide range of benchmarks. + +General memory allocators are interesting as there exists no algorithm that is +optimal -- for a given allocator one can usually construct a workload +where it does not do so well. The goal is thus to find an allocation +strategy that performs well over a wide range of benchmarks without +suffering from (too much) underperformance in less common situations. + +As always, interpret these results with care since some benchmarks test synthetic +or uncommon situations that may never apply to your workloads. For example, most +allocators do not do well on `xmalloc-testN` but that includes even the best +industrial allocators like _jemalloc_ and _tcmalloc_ that are used in some of +the world's largest systems (like Chrome or FreeBSD). + +Also, the benchmarks here do not measure the behaviour on very large and long-running server workloads, +or worst-case latencies of allocation. Much work has gone into `mimalloc` to work well on such +workloads (for example, to reduce virtual memory fragmentation on long-running services) +but such optimizations are not always reflected in the current benchmark suite. + +We show here only an overview -- for +more specific details and further benchmarks we refer to the +[technical report](https://www.microsoft.com/en-us/research/publication/mimalloc-free-list-sharding-in-action). +The benchmark suite is automated and available separately +as [mimalloc-bench](https://github.com/daanx/mimalloc-bench). + + +## Benchmark Results on a 16-core AMD 5950x (Zen3) + +Testing on the 16-core AMD 5950x processor at 3.4Ghz (4.9Ghz boost), with +with 32GiB memory at 3600Mhz, running Ubuntu 20.04 with glibc 2.31 and GCC 9.3.0. + +We measure three versions of _mimalloc_: the main version `mi` (tag:v1.7.0), +the new v2.0 beta version as `xmi` (tag:v2.0.0), and the main version in secure mode as `smi` (tag:v1.7.0). + +The other allocators are +Google's [_tcmalloc_](https://github.com/gperftools/gperftools) (`tc`, tag:gperftools-2.8.1) used in Chrome, +Facebook's [_jemalloc_](https://github.com/jemalloc/jemalloc) (`je`, tag:5.2.1) by Jason Evans used in Firefox and FreeBSD, +the Intel thread building blocks [allocator](https://github.com/intel/tbb) (`tbb`, tag:v2020.3), +[rpmalloc](https://github.com/mjansson/rpmalloc) (`rp`,tag:1.4.1) by Mattias Jansson, +the original scalable [_Hoard_](https://github.com/emeryberger/Hoard) (git:d880f72) allocator by Emery Berger \[1], +the memory compacting [_Mesh_](https://github.com/plasma-umass/Mesh) (git:67ff31a) allocator by +Bobby Powers _et al_ \[8], +and finally the default system allocator (`glibc`, 2.31) (based on _PtMalloc2_). + + + + +Any benchmarks ending in `N` run on all 32 logical cores in parallel. +Results are averaged over 10 runs and reported relative +to mimalloc (where 1.2 means it took 1.2× longer to run). +The legend also contains the _overall relative score_ between the +allocators where 100 points is the maximum if an allocator is fastest on +all benchmarks. + +The single threaded _cfrac_ benchmark by Dave Barrett is an implementation of +continued fraction factorization which uses many small short-lived allocations. +All allocators do well on such common usage, where _mimalloc_ is just a tad +faster than _tcmalloc_ and +_jemalloc_. + +The _leanN_ program is interesting as a large realistic and +concurrent workload of the [Lean](https://github.com/leanprover/lean) +theorem prover compiling its own standard library, and there is a 13% +speedup over _tcmalloc_. This is +quite significant: if Lean spends 20% of its time in the +allocator that means that _mimalloc_ is 1.6× faster than _tcmalloc_ +here. (This is surprising as that is not measured in a pure +allocation benchmark like _alloc-test_. We conjecture that we see this +outsized improvement here because _mimalloc_ has better locality in +the allocation which improves performance for the *other* computations +in a program as well). + +The single threaded _redis_ benchmark again show that most allocators do well on such workloads. + +The _larsonN_ server benchmark by Larson and Krishnan \[2] allocates and frees between threads. They observed this +behavior (which they call _bleeding_) in actual server applications, and the benchmark simulates this. +Here, _mimalloc_ is quite a bit faster than _tcmalloc_ and _jemalloc_ probably due to the object migration between different threads. + +The _mstressN_ workload performs many allocations and re-allocations, +and migrates objects between threads (as in _larsonN_). However, it also +creates and destroys the _N_ worker threads a few times keeping some objects +alive beyond the life time of the allocating thread. We observed this +behavior in many larger server applications. + +The [_rptestN_](https://github.com/mjansson/rpmalloc-benchmark) benchmark +by Mattias Jansson is a allocator test originally designed +for _rpmalloc_, and tries to simulate realistic allocation patterns over +multiple threads. Here the differences between allocators become more apparent. + +The second benchmark set tests specific aspects of the allocators and +shows even more extreme differences between them. + +The _alloc-test_, by +[OLogN Technologies AG](http://ithare.com/testing-memory-allocators-ptmalloc2-tcmalloc-hoard-jemalloc-while-trying-to-simulate-real-world-loads/), is a very allocation intensive benchmark doing millions of +allocations in various size classes. The test is scaled such that when an +allocator performs almost identically on _alloc-test1_ as _alloc-testN_ it +means that it scales linearly. + +The _sh6bench_ and _sh8bench_ benchmarks are +developed by [MicroQuill](http://www.microquill.com/) as part of SmartHeap. +In _sh6bench_ _mimalloc_ does much +better than the others (more than 2.5× faster than _jemalloc_). +We cannot explain this well but believe it is +caused in part by the "reverse" free-ing pattern in _sh6bench_. +The _sh8bench_ is a variation with object migration +between threads; whereas _tcmalloc_ did well on _sh6bench_, the addition of object migration causes it to be 10× slower than before. + +The _xmalloc-testN_ benchmark by Lever and Boreham \[5] and Christian Eder, simulates an asymmetric workload where +some threads only allocate, and others only free -- they observed this pattern in +larger server applications. Here we see that +the _mimalloc_ technique of having non-contended sharded thread free +lists pays off as it outperforms others by a very large margin. Only _rpmalloc_, _tbb_, and _glibc_ also scale well on this benchmark. + +The _cache-scratch_ benchmark by Emery Berger \[1], and introduced with +the Hoard allocator to test for _passive-false_ sharing of cache lines. +With a single thread they all +perform the same, but when running with multiple threads the potential allocator +induced false sharing of the cache lines can cause large run-time differences. +Crundal \[6] describes in detail why the false cache line sharing occurs in the _tcmalloc_ design, and also discusses how this +can be avoided with some small implementation changes. +Only the _tbb_, _rpmalloc_ and _mesh_ allocators also avoid the +cache line sharing completely, while _Hoard_ and _glibc_ seem to mitigate +the effects. Kukanov and Voss \[7] describe in detail +how the design of _tbb_ avoids the false cache line sharing. + + +## On a 36-core Intel Xeon + +For completeness, here are the results on a big Amazon +[c5.18xlarge](https://aws.amazon.com/ec2/instance-types/#Compute_Optimized) instance +consisting of a 2×18-core Intel Xeon (Cascade Lake) at 3.4GHz (boost 3.5GHz) +with 144GiB ECC memory, running Ubuntu 20.04 with glibc 2.31, GCC 9.3.0, and +Clang 10.0.0. This time, the mimalloc allocators (mi, xmi, and smi) were +compiled with the Clang compiler instead of GCC. +The results are similar to the AMD results but it is interesting to +see the differences in the _larsonN_, _mstressN_, and _xmalloc-testN_ benchmarks. + + + + + +## Peak Working Set + +The following figure shows the peak working set (rss) of the allocators +on the benchmarks (on the c5.18xlarge instance). + + + + +Note that the _xmalloc-testN_ memory usage should be disregarded as it +allocates more the faster the program runs. Similarly, memory usage of +_larsonN_, _mstressN_, _rptestN_ and _sh8bench_ can vary depending on scheduling and +speed. Nevertheless, we hope to improve the memory usage on _mstressN_ +and _rptestN_ (just as _cfrac_, _larsonN_ and _sh8bench_ have a small working set which skews the results). + + + + +# References + +- \[1] Emery D. Berger, Kathryn S. McKinley, Robert D. Blumofe, and Paul R. Wilson. + _Hoard: A Scalable Memory Allocator for Multithreaded Applications_ + the Ninth International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS-IX). Cambridge, MA, November 2000. + [pdf](http://www.cs.utexas.edu/users/mckinley/papers/asplos-2000.pdf) + +- \[2] P. Larson and M. Krishnan. _Memory allocation for long-running server applications_. + In ISMM, Vancouver, B.C., Canada, 1998. [pdf](http://citeseer.ist.psu.edu/viewdoc/download?doi=10.1.1.45.1947&rep=rep1&type=pdf) + +- \[3] D. Grunwald, B. Zorn, and R. Henderson. + _Improving the cache locality of memory allocation_. In R. Cartwright, editor, + Proceedings of the Conference on Programming Language Design and Implementation, pages 177–186, New York, NY, USA, June 1993. [pdf](http://citeseer.ist.psu.edu/viewdoc/download?doi=10.1.1.43.6621&rep=rep1&type=pdf) + +- \[4] J. Barnes and P. Hut. _A hierarchical O(n*log(n)) force-calculation algorithm_. Nature, 324:446-449, 1986. + +- \[5] C. Lever, and D. Boreham. _Malloc() Performance in a Multithreaded Linux Environment._ + In USENIX Annual Technical Conference, Freenix Session. San Diego, CA. Jun. 2000. + Available at + +- \[6] Timothy Crundal. _Reducing Active-False Sharing in TCMalloc_. 2016. CS16S1 project at the Australian National University. [pdf](http://courses.cecs.anu.edu.au/courses/CSPROJECTS/16S1/Reports/Timothy_Crundal_Report.pdf) + +- \[7] Alexey Kukanov, and Michael J Voss. + _The Foundations for Scalable Multi-Core Software in Intel Threading Building Blocks._ + Intel Technology Journal 11 (4). 2007 + +- \[8] Bobby Powers, David Tench, Emery D. Berger, and Andrew McGregor. + _Mesh: Compacting Memory Management for C/C++_ + In Proceedings of the 40th ACM SIGPLAN Conference on Programming Language Design and Implementation (PLDI'19), June 2019, pages 333-–346. + + + +# Contributing + +This project welcomes contributions and suggestions. Most contributions require you to agree to a +Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us +the rights to use your contribution. For details, visit https://cla.microsoft.com. + +When you submit a pull request, a CLA-bot will automatically determine whether you need to provide +a CLA and decorate the PR appropriately (e.g., label, comment). Simply follow the instructions +provided by the bot. You will only need to do this once across all repos using our CLA. + + +# Older Release Notes + +* 2020-09-24, `v1.6.7`: stable release 1.6: using standard C atomics, passing tsan testing, improved + handling of failing to commit on Windows, add [`mi_process_info`](https://github.com/microsoft/mimalloc/blob/master/include/mimalloc.h#L156) api call. +* 2020-08-06, `v1.6.4`: stable release 1.6: improved error recovery in low-memory situations, + support for IllumOS and Haiku, NUMA support for Vista/XP, improved NUMA detection for AMD Ryzen, ubsan support. +* 2020-05-05, `v1.6.3`: stable release 1.6: improved behavior in out-of-memory situations, improved malloc zones on macOS, + build PIC static libraries by default, add option to abort on out-of-memory, line buffered statistics. +* 2020-04-20, `v1.6.2`: stable release 1.6: fix compilation on Android, MingW, Raspberry, and Conda, + stability fix for Windows 7, fix multiple mimalloc instances in one executable, fix `strnlen` overload, + fix aligned debug padding. +* 2020-02-17, `v1.6.1`: stable release 1.6: minor updates (build with clang-cl, fix alignment issue for small objects). +* 2020-02-09, `v1.6.0`: stable release 1.6: fixed potential memory leak, improved overriding + and thread local support on FreeBSD, NetBSD, DragonFly, and macOSX. New byte-precise + heap block overflow detection in debug mode (besides the double-free detection and free-list + corruption detection). Add `nodiscard` attribute to most allocation functions. + Enable `MIMALLOC_PAGE_RESET` by default. New reclamation strategy for abandoned heap pages + for better memory footprint. +* 2020-02-09, `v1.5.0`: stable release 1.5: improved free performance, small bug fixes. +* 2020-01-22, `v1.4.0`: stable release 1.4: improved performance for delayed OS page reset, +more eager concurrent free, addition of STL allocator, fixed potential memory leak. +* 2020-01-15, `v1.3.0`: stable release 1.3: bug fixes, improved randomness and [stronger +free list encoding](https://github.com/microsoft/mimalloc/blob/783e3377f79ee82af43a0793910a9f2d01ac7863/include/mimalloc-internal.h#L396) in secure mode. +* 2019-12-22, `v1.2.2`: stable release 1.2: minor updates. +* 2019-11-22, `v1.2.0`: stable release 1.2: bug fixes, improved secure mode (free list corruption checks, double free mitigation). Improved dynamic overriding on Windows. +* 2019-10-07, `v1.1.0`: stable release 1.1. +* 2019-09-01, `v1.0.8`: pre-release 8: more robust windows dynamic overriding, initial huge page support. +* 2019-08-10, `v1.0.6`: pre-release 6: various performance improvements. diff --git a/compat/mimalloc/segment-cache.c b/compat/mimalloc/segment-cache.c new file mode 100644 index 00000000000000..c071239ce32bee --- /dev/null +++ b/compat/mimalloc/segment-cache.c @@ -0,0 +1,360 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2020, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ + +/* ---------------------------------------------------------------------------- + Implements a cache of segments to avoid expensive OS calls and to reuse + the commit_mask to optimize the commit/decommit calls. + The full memory map of all segments is also implemented here. +-----------------------------------------------------------------------------*/ +#include "mimalloc.h" +#include "mimalloc-internal.h" +#include "mimalloc-atomic.h" + +#include "bitmap.h" // atomic bitmap + +//#define MI_CACHE_DISABLE 1 // define to completely disable the segment cache + +#define MI_CACHE_FIELDS (16) +#define MI_CACHE_MAX (MI_BITMAP_FIELD_BITS*MI_CACHE_FIELDS) // 1024 on 64-bit + +#define BITS_SET() MI_ATOMIC_VAR_INIT(UINTPTR_MAX) +#define MI_CACHE_BITS_SET MI_INIT16(BITS_SET) // note: update if MI_CACHE_FIELDS changes + +typedef struct mi_cache_slot_s { + void* p; + size_t memid; + bool is_pinned; + mi_commit_mask_t commit_mask; + mi_commit_mask_t decommit_mask; + _Atomic(mi_msecs_t) expire; +} mi_cache_slot_t; + +static mi_decl_cache_align mi_cache_slot_t cache[MI_CACHE_MAX]; // = 0 + +static mi_decl_cache_align mi_bitmap_field_t cache_available[MI_CACHE_FIELDS] = { MI_CACHE_BITS_SET }; // zero bit = available! +static mi_decl_cache_align mi_bitmap_field_t cache_available_large[MI_CACHE_FIELDS] = { MI_CACHE_BITS_SET }; +static mi_decl_cache_align mi_bitmap_field_t cache_inuse[MI_CACHE_FIELDS]; // zero bit = free + + +mi_decl_noinline void* _mi_segment_cache_pop(size_t size, mi_commit_mask_t* commit_mask, mi_commit_mask_t* decommit_mask, bool* large, bool* is_pinned, bool* is_zero, size_t* memid, mi_os_tld_t* tld) +{ +#ifdef MI_CACHE_DISABLE + return NULL; +#else + + // only segment blocks + if (size != MI_SEGMENT_SIZE) return NULL; + + // numa node determines start field + const int numa_node = _mi_os_numa_node(tld); + size_t start_field = 0; + if (numa_node > 0) { + start_field = (MI_CACHE_FIELDS / _mi_os_numa_node_count())*numa_node; + if (start_field >= MI_CACHE_FIELDS) start_field = 0; + } + + // find an available slot + mi_bitmap_index_t bitidx = 0; + bool claimed = false; + if (*large) { // large allowed? + claimed = _mi_bitmap_try_find_from_claim(cache_available_large, MI_CACHE_FIELDS, start_field, 1, &bitidx); + if (claimed) *large = true; + } + if (!claimed) { + claimed = _mi_bitmap_try_find_from_claim(cache_available, MI_CACHE_FIELDS, start_field, 1, &bitidx); + if (claimed) *large = false; + } + + if (!claimed) return NULL; + + // found a slot + mi_cache_slot_t* slot = &cache[mi_bitmap_index_bit(bitidx)]; + void* p = slot->p; + *memid = slot->memid; + *is_pinned = slot->is_pinned; + *is_zero = false; + *commit_mask = slot->commit_mask; + *decommit_mask = slot->decommit_mask; + slot->p = NULL; + mi_atomic_storei64_release(&slot->expire,(mi_msecs_t)0); + + // mark the slot as free again + mi_assert_internal(_mi_bitmap_is_claimed(cache_inuse, MI_CACHE_FIELDS, 1, bitidx)); + _mi_bitmap_unclaim(cache_inuse, MI_CACHE_FIELDS, 1, bitidx); + return p; +#endif +} + +static mi_decl_noinline void mi_commit_mask_decommit(mi_commit_mask_t* cmask, void* p, size_t total, mi_stats_t* stats) +{ + if (mi_commit_mask_is_empty(cmask)) { + // nothing + } + else if (mi_commit_mask_is_full(cmask)) { + _mi_os_decommit(p, total, stats); + } + else { + // todo: one call to decommit the whole at once? + mi_assert_internal((total%MI_COMMIT_MASK_BITS)==0); + size_t part = total/MI_COMMIT_MASK_BITS; + size_t idx; + size_t count; + mi_commit_mask_foreach(cmask, idx, count) { + void* start = (uint8_t*)p + (idx*part); + size_t size = count*part; + _mi_os_decommit(start, size, stats); + } + mi_commit_mask_foreach_end() + } + mi_commit_mask_create_empty(cmask); +} + +#define MI_MAX_PURGE_PER_PUSH (4) + +static mi_decl_noinline void mi_segment_cache_purge(bool force, mi_os_tld_t* tld) +{ + MI_UNUSED(tld); + if (!mi_option_is_enabled(mi_option_allow_decommit)) return; + mi_msecs_t now = _mi_clock_now(); + size_t purged = 0; + const size_t max_visits = (force ? MI_CACHE_MAX /* visit all */ : MI_CACHE_FIELDS /* probe at most N (=16) slots */); + size_t idx = (force ? 0 : _mi_random_shuffle((uintptr_t)now) % MI_CACHE_MAX /* random start */ ); + for (size_t visited = 0; visited < max_visits; visited++,idx++) { // visit N slots + if (idx >= MI_CACHE_MAX) idx = 0; // wrap + mi_cache_slot_t* slot = &cache[idx]; + mi_msecs_t expire = mi_atomic_loadi64_relaxed(&slot->expire); + if (expire != 0 && (force || now >= expire)) { // racy read + // seems expired, first claim it from available + purged++; + mi_bitmap_index_t bitidx = mi_bitmap_index_create_from_bit(idx); + if (_mi_bitmap_claim(cache_available, MI_CACHE_FIELDS, 1, bitidx, NULL)) { + // was available, we claimed it + expire = mi_atomic_loadi64_acquire(&slot->expire); + if (expire != 0 && (force || now >= expire)) { // safe read + // still expired, decommit it + mi_atomic_storei64_relaxed(&slot->expire,(mi_msecs_t)0); + mi_assert_internal(!mi_commit_mask_is_empty(&slot->commit_mask) && _mi_bitmap_is_claimed(cache_available_large, MI_CACHE_FIELDS, 1, bitidx)); + _mi_abandoned_await_readers(); // wait until safe to decommit + // decommit committed parts + // TODO: instead of decommit, we could also free to the OS? + mi_commit_mask_decommit(&slot->commit_mask, slot->p, MI_SEGMENT_SIZE, tld->stats); + mi_commit_mask_create_empty(&slot->decommit_mask); + } + _mi_bitmap_unclaim(cache_available, MI_CACHE_FIELDS, 1, bitidx); // make it available again for a pop + } + if (!force && purged > MI_MAX_PURGE_PER_PUSH) break; // bound to no more than N purge tries per push + } + } +} + +void _mi_segment_cache_collect(bool force, mi_os_tld_t* tld) { + mi_segment_cache_purge(force, tld ); +} + +mi_decl_noinline bool _mi_segment_cache_push(void* start, size_t size, size_t memid, const mi_commit_mask_t* commit_mask, const mi_commit_mask_t* decommit_mask, bool is_large, bool is_pinned, mi_os_tld_t* tld) +{ +#ifdef MI_CACHE_DISABLE + return false; +#else + + // only for normal segment blocks + if (size != MI_SEGMENT_SIZE || ((uintptr_t)start % MI_SEGMENT_ALIGN) != 0) return false; + + // numa node determines start field + int numa_node = _mi_os_numa_node(NULL); + size_t start_field = 0; + if (numa_node > 0) { + start_field = (MI_CACHE_FIELDS / _mi_os_numa_node_count())*numa_node; + if (start_field >= MI_CACHE_FIELDS) start_field = 0; + } + + // purge expired entries + mi_segment_cache_purge(false /* force? */, tld); + + // find an available slot + mi_bitmap_index_t bitidx; + bool claimed = _mi_bitmap_try_find_from_claim(cache_inuse, MI_CACHE_FIELDS, start_field, 1, &bitidx); + if (!claimed) return false; + + mi_assert_internal(_mi_bitmap_is_claimed(cache_available, MI_CACHE_FIELDS, 1, bitidx)); + mi_assert_internal(_mi_bitmap_is_claimed(cache_available_large, MI_CACHE_FIELDS, 1, bitidx)); +#if MI_DEBUG>1 + if (is_pinned || is_large) { + mi_assert_internal(mi_commit_mask_is_full(commit_mask)); + } +#endif + + // set the slot + mi_cache_slot_t* slot = &cache[mi_bitmap_index_bit(bitidx)]; + slot->p = start; + slot->memid = memid; + slot->is_pinned = is_pinned; + mi_atomic_storei64_relaxed(&slot->expire,(mi_msecs_t)0); + slot->commit_mask = *commit_mask; + slot->decommit_mask = *decommit_mask; + if (!mi_commit_mask_is_empty(commit_mask) && !is_large && !is_pinned && mi_option_is_enabled(mi_option_allow_decommit)) { + long delay = mi_option_get(mi_option_segment_decommit_delay); + if (delay == 0) { + _mi_abandoned_await_readers(); // wait until safe to decommit + mi_commit_mask_decommit(&slot->commit_mask, start, MI_SEGMENT_SIZE, tld->stats); + mi_commit_mask_create_empty(&slot->decommit_mask); + } + else { + mi_atomic_storei64_release(&slot->expire, _mi_clock_now() + delay); + } + } + + // make it available + _mi_bitmap_unclaim((is_large ? cache_available_large : cache_available), MI_CACHE_FIELDS, 1, bitidx); + return true; +#endif +} + + +/* ----------------------------------------------------------- + The following functions are to reliably find the segment or + block that encompasses any pointer p (or NULL if it is not + in any of our segments). + We maintain a bitmap of all memory with 1 bit per MI_SEGMENT_SIZE (64MiB) + set to 1 if it contains the segment meta data. +----------------------------------------------------------- */ + + +#if (MI_INTPTR_SIZE==8) +#define MI_MAX_ADDRESS ((size_t)20 << 40) // 20TB +#else +#define MI_MAX_ADDRESS ((size_t)2 << 30) // 2Gb +#endif + +#define MI_SEGMENT_MAP_BITS (MI_MAX_ADDRESS / MI_SEGMENT_SIZE) +#define MI_SEGMENT_MAP_SIZE (MI_SEGMENT_MAP_BITS / 8) +#define MI_SEGMENT_MAP_WSIZE (MI_SEGMENT_MAP_SIZE / MI_INTPTR_SIZE) + +static _Atomic(uintptr_t) mi_segment_map[MI_SEGMENT_MAP_WSIZE + 1]; // 2KiB per TB with 64MiB segments + +static size_t mi_segment_map_index_of(const mi_segment_t* segment, size_t* bitidx) { + mi_assert_internal(_mi_ptr_segment(segment) == segment); // is it aligned on MI_SEGMENT_SIZE? + if ((uintptr_t)segment >= MI_MAX_ADDRESS) { + *bitidx = 0; + return MI_SEGMENT_MAP_WSIZE; + } + else { + const uintptr_t segindex = ((uintptr_t)segment) / MI_SEGMENT_SIZE; + *bitidx = segindex % MI_INTPTR_BITS; + const size_t mapindex = segindex / MI_INTPTR_BITS; + mi_assert_internal(mapindex < MI_SEGMENT_MAP_WSIZE); + return mapindex; + } +} + +void _mi_segment_map_allocated_at(const mi_segment_t* segment) { + size_t bitidx; + size_t index = mi_segment_map_index_of(segment, &bitidx); + mi_assert_internal(index <= MI_SEGMENT_MAP_WSIZE); + if (index==MI_SEGMENT_MAP_WSIZE) return; + uintptr_t mask = mi_atomic_load_relaxed(&mi_segment_map[index]); + uintptr_t newmask; + do { + newmask = (mask | ((uintptr_t)1 << bitidx)); + } while (!mi_atomic_cas_weak_release(&mi_segment_map[index], &mask, newmask)); +} + +void _mi_segment_map_freed_at(const mi_segment_t* segment) { + size_t bitidx; + size_t index = mi_segment_map_index_of(segment, &bitidx); + mi_assert_internal(index <= MI_SEGMENT_MAP_WSIZE); + if (index == MI_SEGMENT_MAP_WSIZE) return; + uintptr_t mask = mi_atomic_load_relaxed(&mi_segment_map[index]); + uintptr_t newmask; + do { + newmask = (mask & ~((uintptr_t)1 << bitidx)); + } while (!mi_atomic_cas_weak_release(&mi_segment_map[index], &mask, newmask)); +} + +// Determine the segment belonging to a pointer or NULL if it is not in a valid segment. +static mi_segment_t* _mi_segment_of(const void* p) { + mi_segment_t* segment = _mi_ptr_segment(p); + if (segment == NULL) return NULL; + size_t bitidx; + size_t index = mi_segment_map_index_of(segment, &bitidx); + // fast path: for any pointer to valid small/medium/large object or first MI_SEGMENT_SIZE in huge + const uintptr_t mask = mi_atomic_load_relaxed(&mi_segment_map[index]); + if (mi_likely((mask & ((uintptr_t)1 << bitidx)) != 0)) { + return segment; // yes, allocated by us + } + if (index==MI_SEGMENT_MAP_WSIZE) return NULL; + + // TODO: maintain max/min allocated range for efficiency for more efficient rejection of invalid pointers? + + // search downwards for the first segment in case it is an interior pointer + // could be slow but searches in MI_INTPTR_SIZE * MI_SEGMENT_SIZE (512MiB) steps trough + // valid huge objects + // note: we could maintain a lowest index to speed up the path for invalid pointers? + size_t lobitidx; + size_t loindex; + uintptr_t lobits = mask & (((uintptr_t)1 << bitidx) - 1); + if (lobits != 0) { + loindex = index; + lobitidx = mi_bsr(lobits); // lobits != 0 + } + else if (index == 0) { + return NULL; + } + else { + mi_assert_internal(index > 0); + uintptr_t lomask = mask; + loindex = index; + do { + loindex--; + lomask = mi_atomic_load_relaxed(&mi_segment_map[loindex]); + } while (lomask != 0 && loindex > 0); + if (lomask == 0) return NULL; + lobitidx = mi_bsr(lomask); // lomask != 0 + } + mi_assert_internal(loindex < MI_SEGMENT_MAP_WSIZE); + // take difference as the addresses could be larger than the MAX_ADDRESS space. + size_t diff = (((index - loindex) * (8*MI_INTPTR_SIZE)) + bitidx - lobitidx) * MI_SEGMENT_SIZE; + segment = (mi_segment_t*)((uint8_t*)segment - diff); + + if (segment == NULL) return NULL; + mi_assert_internal((void*)segment < p); + bool cookie_ok = (_mi_ptr_cookie(segment) == segment->cookie); + mi_assert_internal(cookie_ok); + if (mi_unlikely(!cookie_ok)) return NULL; + if (((uint8_t*)segment + mi_segment_size(segment)) <= (uint8_t*)p) return NULL; // outside the range + mi_assert_internal(p >= (void*)segment && (uint8_t*)p < (uint8_t*)segment + mi_segment_size(segment)); + return segment; +} + +// Is this a valid pointer in our heap? +static bool mi_is_valid_pointer(const void* p) { + return (_mi_segment_of(p) != NULL); +} + +mi_decl_nodiscard mi_decl_export bool mi_is_in_heap_region(const void* p) mi_attr_noexcept { + return mi_is_valid_pointer(p); +} + +/* +// Return the full segment range belonging to a pointer +static void* mi_segment_range_of(const void* p, size_t* size) { + mi_segment_t* segment = _mi_segment_of(p); + if (segment == NULL) { + if (size != NULL) *size = 0; + return NULL; + } + else { + if (size != NULL) *size = segment->segment_size; + return segment; + } + mi_assert_expensive(page == NULL || mi_segment_is_valid(_mi_page_segment(page),tld)); + mi_assert_internal(page == NULL || (mi_segment_page_size(_mi_page_segment(page)) - (MI_SECURE == 0 ? 0 : _mi_os_page_size())) >= block_size); + mi_reset_delayed(tld); + mi_assert_internal(page == NULL || mi_page_not_in_queue(page, tld)); + return page; +} +*/ diff --git a/compat/mimalloc/segment.c b/compat/mimalloc/segment.c new file mode 100644 index 00000000000000..d772440d69f032 --- /dev/null +++ b/compat/mimalloc/segment.c @@ -0,0 +1,1542 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2020, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ +#include "mimalloc.h" +#include "mimalloc-internal.h" +#include "mimalloc-atomic.h" + +#include // memset +#include + +#define MI_PAGE_HUGE_ALIGN (256*1024) + +static void mi_segment_delayed_decommit(mi_segment_t* segment, bool force, mi_stats_t* stats); + + +// ------------------------------------------------------------------- +// commit mask +// ------------------------------------------------------------------- + +static bool mi_commit_mask_all_set(const mi_commit_mask_t* commit, const mi_commit_mask_t* cm) { + for (size_t i = 0; i < MI_COMMIT_MASK_FIELD_COUNT; i++) { + if ((commit->mask[i] & cm->mask[i]) != cm->mask[i]) return false; + } + return true; +} + +static bool mi_commit_mask_any_set(const mi_commit_mask_t* commit, const mi_commit_mask_t* cm) { + for (size_t i = 0; i < MI_COMMIT_MASK_FIELD_COUNT; i++) { + if ((commit->mask[i] & cm->mask[i]) != 0) return true; + } + return false; +} + +static void mi_commit_mask_create_intersect(const mi_commit_mask_t* commit, const mi_commit_mask_t* cm, mi_commit_mask_t* res) { + for (size_t i = 0; i < MI_COMMIT_MASK_FIELD_COUNT; i++) { + res->mask[i] = (commit->mask[i] & cm->mask[i]); + } +} + +static void mi_commit_mask_clear(mi_commit_mask_t* res, const mi_commit_mask_t* cm) { + for (size_t i = 0; i < MI_COMMIT_MASK_FIELD_COUNT; i++) { + res->mask[i] &= ~(cm->mask[i]); + } +} + +static void mi_commit_mask_set(mi_commit_mask_t* res, const mi_commit_mask_t* cm) { + for (size_t i = 0; i < MI_COMMIT_MASK_FIELD_COUNT; i++) { + res->mask[i] |= cm->mask[i]; + } +} + +static void mi_commit_mask_create(size_t bitidx, size_t bitcount, mi_commit_mask_t* cm) { + mi_assert_internal(bitidx < MI_COMMIT_MASK_BITS); + mi_assert_internal((bitidx + bitcount) <= MI_COMMIT_MASK_BITS); + if (bitcount == MI_COMMIT_MASK_BITS) { + mi_assert_internal(bitidx==0); + mi_commit_mask_create_full(cm); + } + else if (bitcount == 0) { + mi_commit_mask_create_empty(cm); + } + else { + mi_commit_mask_create_empty(cm); + size_t i = bitidx / MI_COMMIT_MASK_FIELD_BITS; + size_t ofs = bitidx % MI_COMMIT_MASK_FIELD_BITS; + while (bitcount > 0) { + mi_assert_internal(i < MI_COMMIT_MASK_FIELD_COUNT); + size_t avail = MI_COMMIT_MASK_FIELD_BITS - ofs; + size_t count = (bitcount > avail ? avail : bitcount); + size_t mask = (count >= MI_COMMIT_MASK_FIELD_BITS ? ~((size_t)0) : (((size_t)1 << count) - 1) << ofs); + cm->mask[i] = mask; + bitcount -= count; + ofs = 0; + i++; + } + } +} + +size_t _mi_commit_mask_committed_size(const mi_commit_mask_t* cm, size_t total) { + mi_assert_internal((total%MI_COMMIT_MASK_BITS)==0); + size_t count = 0; + for (size_t i = 0; i < MI_COMMIT_MASK_FIELD_COUNT; i++) { + size_t mask = cm->mask[i]; + if (~mask == 0) { + count += MI_COMMIT_MASK_FIELD_BITS; + } + else { + for (; mask != 0; mask >>= 1) { // todo: use popcount + if ((mask&1)!=0) count++; + } + } + } + // we use total since for huge segments each commit bit may represent a larger size + return ((total / MI_COMMIT_MASK_BITS) * count); +} + + +size_t _mi_commit_mask_next_run(const mi_commit_mask_t* cm, size_t* idx) { + size_t i = (*idx) / MI_COMMIT_MASK_FIELD_BITS; + size_t ofs = (*idx) % MI_COMMIT_MASK_FIELD_BITS; + size_t mask = 0; + // find first ones + while (i < MI_COMMIT_MASK_FIELD_COUNT) { + mask = cm->mask[i]; + mask >>= ofs; + if (mask != 0) { + while ((mask&1) == 0) { + mask >>= 1; + ofs++; + } + break; + } + i++; + ofs = 0; + } + if (i >= MI_COMMIT_MASK_FIELD_COUNT) { + // not found + *idx = MI_COMMIT_MASK_BITS; + return 0; + } + else { + // found, count ones + size_t count = 0; + *idx = (i*MI_COMMIT_MASK_FIELD_BITS) + ofs; + do { + mi_assert_internal(ofs < MI_COMMIT_MASK_FIELD_BITS && (mask&1) == 1); + do { + count++; + mask >>= 1; + } while ((mask&1) == 1); + if ((((*idx + count) % MI_COMMIT_MASK_FIELD_BITS) == 0)) { + i++; + if (i >= MI_COMMIT_MASK_FIELD_COUNT) break; + mask = cm->mask[i]; + ofs = 0; + } + } while ((mask&1) == 1); + mi_assert_internal(count > 0); + return count; + } +} + + +/* -------------------------------------------------------------------------------- + Segment allocation + + If a thread ends, it "abandons" pages with used blocks + and there is an abandoned segment list whose segments can + be reclaimed by still running threads, much like work-stealing. +-------------------------------------------------------------------------------- */ + + +/* ----------------------------------------------------------- + Slices +----------------------------------------------------------- */ + + +static const mi_slice_t* mi_segment_slices_end(const mi_segment_t* segment) { + return &segment->slices[segment->slice_entries]; +} + +static uint8_t* mi_slice_start(const mi_slice_t* slice) { + mi_segment_t* segment = _mi_ptr_segment(slice); + mi_assert_internal(slice >= segment->slices && slice < mi_segment_slices_end(segment)); + return ((uint8_t*)segment + ((slice - segment->slices)*MI_SEGMENT_SLICE_SIZE)); +} + + +/* ----------------------------------------------------------- + Bins +----------------------------------------------------------- */ +// Use bit scan forward to quickly find the first zero bit if it is available + +static inline size_t mi_slice_bin8(size_t slice_count) { + if (slice_count<=1) return slice_count; + mi_assert_internal(slice_count <= MI_SLICES_PER_SEGMENT); + slice_count--; + size_t s = mi_bsr(slice_count); // slice_count > 1 + if (s <= 2) return slice_count + 1; + size_t bin = ((s << 2) | ((slice_count >> (s - 2))&0x03)) - 4; + return bin; +} + +static inline size_t mi_slice_bin(size_t slice_count) { + mi_assert_internal(slice_count*MI_SEGMENT_SLICE_SIZE <= MI_SEGMENT_SIZE); + mi_assert_internal(mi_slice_bin8(MI_SLICES_PER_SEGMENT) <= MI_SEGMENT_BIN_MAX); + size_t bin = mi_slice_bin8(slice_count); + mi_assert_internal(bin <= MI_SEGMENT_BIN_MAX); + return bin; +} + +static inline size_t mi_slice_index(const mi_slice_t* slice) { + mi_segment_t* segment = _mi_ptr_segment(slice); + ptrdiff_t index = slice - segment->slices; + mi_assert_internal(index >= 0 && index < (ptrdiff_t)segment->slice_entries); + return index; +} + + +/* ----------------------------------------------------------- + Slice span queues +----------------------------------------------------------- */ + +static void mi_span_queue_push(mi_span_queue_t* sq, mi_slice_t* slice) { + // todo: or push to the end? + mi_assert_internal(slice->prev == NULL && slice->next==NULL); + slice->prev = NULL; // paranoia + slice->next = sq->first; + sq->first = slice; + if (slice->next != NULL) slice->next->prev = slice; + else sq->last = slice; + slice->xblock_size = 0; // free +} + +static mi_span_queue_t* mi_span_queue_for(size_t slice_count, mi_segments_tld_t* tld) { + size_t bin = mi_slice_bin(slice_count); + mi_span_queue_t* sq = &tld->spans[bin]; + mi_assert_internal(sq->slice_count >= slice_count); + return sq; +} + +static void mi_span_queue_delete(mi_span_queue_t* sq, mi_slice_t* slice) { + mi_assert_internal(slice->xblock_size==0 && slice->slice_count>0 && slice->slice_offset==0); + // should work too if the queue does not contain slice (which can happen during reclaim) + if (slice->prev != NULL) slice->prev->next = slice->next; + if (slice == sq->first) sq->first = slice->next; + if (slice->next != NULL) slice->next->prev = slice->prev; + if (slice == sq->last) sq->last = slice->prev; + slice->prev = NULL; + slice->next = NULL; + slice->xblock_size = 1; // no more free +} + + +/* ----------------------------------------------------------- + Invariant checking +----------------------------------------------------------- */ + +static bool mi_slice_is_used(const mi_slice_t* slice) { + return (slice->xblock_size > 0); +} + + +#if (MI_DEBUG>=3) +static bool mi_span_queue_contains(mi_span_queue_t* sq, mi_slice_t* slice) { + for (mi_slice_t* s = sq->first; s != NULL; s = s->next) { + if (s==slice) return true; + } + return false; +} + +static bool mi_segment_is_valid(mi_segment_t* segment, mi_segments_tld_t* tld) { + mi_assert_internal(segment != NULL); + mi_assert_internal(_mi_ptr_cookie(segment) == segment->cookie); + mi_assert_internal(segment->abandoned <= segment->used); + mi_assert_internal(segment->thread_id == 0 || segment->thread_id == _mi_thread_id()); + mi_assert_internal(mi_commit_mask_all_set(&segment->commit_mask, &segment->decommit_mask)); // can only decommit committed blocks + //mi_assert_internal(segment->segment_info_size % MI_SEGMENT_SLICE_SIZE == 0); + mi_slice_t* slice = &segment->slices[0]; + const mi_slice_t* end = mi_segment_slices_end(segment); + size_t used_count = 0; + mi_span_queue_t* sq; + while(slice < end) { + mi_assert_internal(slice->slice_count > 0); + mi_assert_internal(slice->slice_offset == 0); + size_t index = mi_slice_index(slice); + size_t maxindex = (index + slice->slice_count >= segment->slice_entries ? segment->slice_entries : index + slice->slice_count) - 1; + if (mi_slice_is_used(slice)) { // a page in use, we need at least MAX_SLICE_OFFSET valid back offsets + used_count++; + for (size_t i = 0; i <= MI_MAX_SLICE_OFFSET && index + i <= maxindex; i++) { + mi_assert_internal(segment->slices[index + i].slice_offset == i*sizeof(mi_slice_t)); + mi_assert_internal(i==0 || segment->slices[index + i].slice_count == 0); + mi_assert_internal(i==0 || segment->slices[index + i].xblock_size == 1); + } + // and the last entry as well (for coalescing) + const mi_slice_t* last = slice + slice->slice_count - 1; + if (last > slice && last < mi_segment_slices_end(segment)) { + mi_assert_internal(last->slice_offset == (slice->slice_count-1)*sizeof(mi_slice_t)); + mi_assert_internal(last->slice_count == 0); + mi_assert_internal(last->xblock_size == 1); + } + } + else { // free range of slices; only last slice needs a valid back offset + mi_slice_t* last = &segment->slices[maxindex]; + if (segment->kind != MI_SEGMENT_HUGE || slice->slice_count <= (segment->slice_entries - segment->segment_info_slices)) { + mi_assert_internal((uint8_t*)slice == (uint8_t*)last - last->slice_offset); + } + mi_assert_internal(slice == last || last->slice_count == 0 ); + mi_assert_internal(last->xblock_size == 0 || (segment->kind==MI_SEGMENT_HUGE && last->xblock_size==1)); + if (segment->kind != MI_SEGMENT_HUGE && segment->thread_id != 0) { // segment is not huge or abandoned + sq = mi_span_queue_for(slice->slice_count,tld); + mi_assert_internal(mi_span_queue_contains(sq,slice)); + } + } + slice = &segment->slices[maxindex+1]; + } + mi_assert_internal(slice == end); + mi_assert_internal(used_count == segment->used + 1); + return true; +} +#endif + +/* ----------------------------------------------------------- + Segment size calculations +----------------------------------------------------------- */ + +static size_t mi_segment_info_size(mi_segment_t* segment) { + return segment->segment_info_slices * MI_SEGMENT_SLICE_SIZE; +} + +static uint8_t* _mi_segment_page_start_from_slice(const mi_segment_t* segment, const mi_slice_t* slice, size_t xblock_size, size_t* page_size) +{ + ptrdiff_t idx = slice - segment->slices; + size_t psize = (size_t)slice->slice_count * MI_SEGMENT_SLICE_SIZE; + // make the start not OS page aligned for smaller blocks to avoid page/cache effects + size_t start_offset = (xblock_size >= MI_INTPTR_SIZE && xblock_size <= 1024 ? MI_MAX_ALIGN_GUARANTEE : 0); + if (page_size != NULL) { *page_size = psize - start_offset; } + return (uint8_t*)segment + ((idx*MI_SEGMENT_SLICE_SIZE) + start_offset); +} + +// Start of the page available memory; can be used on uninitialized pages +uint8_t* _mi_segment_page_start(const mi_segment_t* segment, const mi_page_t* page, size_t* page_size) +{ + const mi_slice_t* slice = mi_page_to_slice((mi_page_t*)page); + uint8_t* p = _mi_segment_page_start_from_slice(segment, slice, page->xblock_size, page_size); + mi_assert_internal(page->xblock_size > 0 || _mi_ptr_page(p) == page); + mi_assert_internal(_mi_ptr_segment(p) == segment); + return p; +} + + +static size_t mi_segment_calculate_slices(size_t required, size_t* pre_size, size_t* info_slices) { + size_t page_size = _mi_os_page_size(); + size_t isize = _mi_align_up(sizeof(mi_segment_t), page_size); + size_t guardsize = 0; + + if (MI_SECURE>0) { + // in secure mode, we set up a protected page in between the segment info + // and the page data (and one at the end of the segment) + guardsize = page_size; + required = _mi_align_up(required, page_size); + } + + if (pre_size != NULL) *pre_size = isize; + isize = _mi_align_up(isize + guardsize, MI_SEGMENT_SLICE_SIZE); + if (info_slices != NULL) *info_slices = isize / MI_SEGMENT_SLICE_SIZE; + size_t segment_size = (required==0 ? MI_SEGMENT_SIZE : _mi_align_up( required + isize + guardsize, MI_SEGMENT_SLICE_SIZE) ); + mi_assert_internal(segment_size % MI_SEGMENT_SLICE_SIZE == 0); + return (segment_size / MI_SEGMENT_SLICE_SIZE); +} + + +/* ---------------------------------------------------------------------------- +Segment caches +We keep a small segment cache per thread to increase local +reuse and avoid setting/clearing guard pages in secure mode. +------------------------------------------------------------------------------- */ + +static void mi_segments_track_size(long segment_size, mi_segments_tld_t* tld) { + if (segment_size>=0) _mi_stat_increase(&tld->stats->segments,1); + else _mi_stat_decrease(&tld->stats->segments,1); + tld->count += (segment_size >= 0 ? 1 : -1); + if (tld->count > tld->peak_count) tld->peak_count = tld->count; + tld->current_size += segment_size; + if (tld->current_size > tld->peak_size) tld->peak_size = tld->current_size; +} + +static void mi_segment_os_free(mi_segment_t* segment, mi_segments_tld_t* tld) { + segment->thread_id = 0; + _mi_segment_map_freed_at(segment); + mi_segments_track_size(-((long)mi_segment_size(segment)),tld); + if (MI_SECURE>0) { + // _mi_os_unprotect(segment, mi_segment_size(segment)); // ensure no more guard pages are set + // unprotect the guard pages; we cannot just unprotect the whole segment size as part may be decommitted + size_t os_pagesize = _mi_os_page_size(); + _mi_os_unprotect((uint8_t*)segment + mi_segment_info_size(segment) - os_pagesize, os_pagesize); + uint8_t* end = (uint8_t*)segment + mi_segment_size(segment) - os_pagesize; + _mi_os_unprotect(end, os_pagesize); + } + + // purge delayed decommits now? (no, leave it to the cache) + // mi_segment_delayed_decommit(segment,true,tld->stats); + + // _mi_os_free(segment, mi_segment_size(segment), /*segment->memid,*/ tld->stats); + const size_t size = mi_segment_size(segment); + if (size != MI_SEGMENT_SIZE || !_mi_segment_cache_push(segment, size, segment->memid, &segment->commit_mask, &segment->decommit_mask, segment->mem_is_large, segment->mem_is_pinned, tld->os)) { + const size_t csize = _mi_commit_mask_committed_size(&segment->commit_mask, size); + if (csize > 0 && !segment->mem_is_pinned) _mi_stat_decrease(&_mi_stats_main.committed, csize); + _mi_abandoned_await_readers(); // wait until safe to free + _mi_arena_free(segment, mi_segment_size(segment), segment->memid, segment->mem_is_pinned /* pretend not committed to not double count decommits */, tld->os); + } +} + +// called by threads that are terminating +void _mi_segment_thread_collect(mi_segments_tld_t* tld) { + MI_UNUSED(tld); + // nothing to do +} + + +/* ----------------------------------------------------------- + Span management +----------------------------------------------------------- */ + +static void mi_segment_commit_mask(mi_segment_t* segment, bool conservative, uint8_t* p, size_t size, uint8_t** start_p, size_t* full_size, mi_commit_mask_t* cm) { + mi_assert_internal(_mi_ptr_segment(p) == segment); + mi_assert_internal(segment->kind != MI_SEGMENT_HUGE); + mi_commit_mask_create_empty(cm); + if (size == 0 || size > MI_SEGMENT_SIZE || segment->kind == MI_SEGMENT_HUGE) return; + const size_t segstart = mi_segment_info_size(segment); + const size_t segsize = mi_segment_size(segment); + if (p >= (uint8_t*)segment + segsize) return; + + size_t pstart = (p - (uint8_t*)segment); + mi_assert_internal(pstart + size <= segsize); + + size_t start; + size_t end; + if (conservative) { + // decommit conservative + start = _mi_align_up(pstart, MI_COMMIT_SIZE); + end = _mi_align_down(pstart + size, MI_COMMIT_SIZE); + mi_assert_internal(start >= segstart); + mi_assert_internal(end <= segsize); + } + else { + // commit liberal + start = _mi_align_down(pstart, MI_MINIMAL_COMMIT_SIZE); + end = _mi_align_up(pstart + size, MI_MINIMAL_COMMIT_SIZE); + } + if (pstart >= segstart && start < segstart) { // note: the mask is also calculated for an initial commit of the info area + start = segstart; + } + if (end > segsize) { + end = segsize; + } + + mi_assert_internal(start <= pstart && (pstart + size) <= end); + mi_assert_internal(start % MI_COMMIT_SIZE==0 && end % MI_COMMIT_SIZE == 0); + *start_p = (uint8_t*)segment + start; + *full_size = (end > start ? end - start : 0); + if (*full_size == 0) return; + + size_t bitidx = start / MI_COMMIT_SIZE; + mi_assert_internal(bitidx < MI_COMMIT_MASK_BITS); + + size_t bitcount = *full_size / MI_COMMIT_SIZE; // can be 0 + if (bitidx + bitcount > MI_COMMIT_MASK_BITS) { + _mi_warning_message("commit mask overflow: idx=%zu count=%zu start=%zx end=%zx p=0x%p size=%zu fullsize=%zu\n", bitidx, bitcount, start, end, p, size, *full_size); + } + mi_assert_internal((bitidx + bitcount) <= MI_COMMIT_MASK_BITS); + mi_commit_mask_create(bitidx, bitcount, cm); +} + + +static bool mi_segment_commitx(mi_segment_t* segment, bool commit, uint8_t* p, size_t size, mi_stats_t* stats) { + mi_assert_internal(mi_commit_mask_all_set(&segment->commit_mask, &segment->decommit_mask)); + + // try to commit in at least MI_MINIMAL_COMMIT_SIZE sizes. + /* + if (commit && size > 0) { + const size_t csize = _mi_align_up(size, MI_MINIMAL_COMMIT_SIZE); + if (p + csize <= mi_segment_end(segment)) { + size = csize; + } + } + */ + // commit liberal, but decommit conservative + uint8_t* start = NULL; + size_t full_size = 0; + mi_commit_mask_t mask; + mi_segment_commit_mask(segment, !commit/*conservative*/, p, size, &start, &full_size, &mask); + if (mi_commit_mask_is_empty(&mask) || full_size==0) return true; + + if (commit && !mi_commit_mask_all_set(&segment->commit_mask, &mask)) { + bool is_zero = false; + mi_commit_mask_t cmask; + mi_commit_mask_create_intersect(&segment->commit_mask, &mask, &cmask); + _mi_stat_decrease(&_mi_stats_main.committed, _mi_commit_mask_committed_size(&cmask, MI_SEGMENT_SIZE)); // adjust for overlap + if (!_mi_os_commit(start,full_size,&is_zero,stats)) return false; + mi_commit_mask_set(&segment->commit_mask, &mask); + } + else if (!commit && mi_commit_mask_any_set(&segment->commit_mask, &mask)) { + mi_assert_internal((void*)start != (void*)segment); + //mi_assert_internal(mi_commit_mask_all_set(&segment->commit_mask, &mask)); + + mi_commit_mask_t cmask; + mi_commit_mask_create_intersect(&segment->commit_mask, &mask, &cmask); + _mi_stat_increase(&_mi_stats_main.committed, full_size - _mi_commit_mask_committed_size(&cmask, MI_SEGMENT_SIZE)); // adjust for overlap + if (segment->allow_decommit) { + _mi_os_decommit(start, full_size, stats); // ok if this fails + } + mi_commit_mask_clear(&segment->commit_mask, &mask); + } + // increase expiration of reusing part of the delayed decommit + if (commit && mi_commit_mask_any_set(&segment->decommit_mask, &mask)) { + segment->decommit_expire = _mi_clock_now() + mi_option_get(mi_option_decommit_delay); + } + // always undo delayed decommits + mi_commit_mask_clear(&segment->decommit_mask, &mask); + return true; +} + +static bool mi_segment_ensure_committed(mi_segment_t* segment, uint8_t* p, size_t size, mi_stats_t* stats) { + mi_assert_internal(mi_commit_mask_all_set(&segment->commit_mask, &segment->decommit_mask)); + // note: assumes commit_mask is always full for huge segments as otherwise the commit mask bits can overflow + if (mi_commit_mask_is_full(&segment->commit_mask) && mi_commit_mask_is_empty(&segment->decommit_mask)) return true; // fully committed + return mi_segment_commitx(segment,true,p,size,stats); +} + +static void mi_segment_perhaps_decommit(mi_segment_t* segment, uint8_t* p, size_t size, mi_stats_t* stats) { + if (!segment->allow_decommit) return; + if (mi_option_get(mi_option_decommit_delay) == 0) { + mi_segment_commitx(segment, false, p, size, stats); + } + else { + // register for future decommit in the decommit mask + uint8_t* start = NULL; + size_t full_size = 0; + mi_commit_mask_t mask; + mi_segment_commit_mask(segment, true /*conservative*/, p, size, &start, &full_size, &mask); + if (mi_commit_mask_is_empty(&mask) || full_size==0) return; + + // update delayed commit + mi_assert_internal(segment->decommit_expire > 0 || mi_commit_mask_is_empty(&segment->decommit_mask)); + mi_commit_mask_t cmask; + mi_commit_mask_create_intersect(&segment->commit_mask, &mask, &cmask); // only decommit what is committed; span_free may try to decommit more + mi_commit_mask_set(&segment->decommit_mask, &cmask); + mi_msecs_t now = _mi_clock_now(); + if (segment->decommit_expire == 0) { + // no previous decommits, initialize now + segment->decommit_expire = now + mi_option_get(mi_option_decommit_delay); + } + else if (segment->decommit_expire <= now) { + // previous decommit mask already expired + // mi_segment_delayed_decommit(segment, true, stats); + segment->decommit_expire = now + mi_option_get(mi_option_decommit_extend_delay); // (mi_option_get(mi_option_decommit_delay) / 8); // wait a tiny bit longer in case there is a series of free's + } + else { + // previous decommit mask is not yet expired, increase the expiration by a bit. + segment->decommit_expire += mi_option_get(mi_option_decommit_extend_delay); + } + } +} + +static void mi_segment_delayed_decommit(mi_segment_t* segment, bool force, mi_stats_t* stats) { + if (!segment->allow_decommit || mi_commit_mask_is_empty(&segment->decommit_mask)) return; + mi_msecs_t now = _mi_clock_now(); + if (!force && now < segment->decommit_expire) return; + + mi_commit_mask_t mask = segment->decommit_mask; + segment->decommit_expire = 0; + mi_commit_mask_create_empty(&segment->decommit_mask); + + size_t idx; + size_t count; + mi_commit_mask_foreach(&mask, idx, count) { + // if found, decommit that sequence + if (count > 0) { + uint8_t* p = (uint8_t*)segment + (idx*MI_COMMIT_SIZE); + size_t size = count * MI_COMMIT_SIZE; + mi_segment_commitx(segment, false, p, size, stats); + } + } + mi_commit_mask_foreach_end() + mi_assert_internal(mi_commit_mask_is_empty(&segment->decommit_mask)); +} + + +static bool mi_segment_is_abandoned(mi_segment_t* segment) { + return (segment->thread_id == 0); +} + +// note: can be called on abandoned segments +static void mi_segment_span_free(mi_segment_t* segment, size_t slice_index, size_t slice_count, mi_segments_tld_t* tld) { + mi_assert_internal(slice_index < segment->slice_entries); + mi_span_queue_t* sq = (segment->kind == MI_SEGMENT_HUGE || mi_segment_is_abandoned(segment) + ? NULL : mi_span_queue_for(slice_count,tld)); + if (slice_count==0) slice_count = 1; + mi_assert_internal(slice_index + slice_count - 1 < segment->slice_entries); + + // set first and last slice (the intermediates can be undetermined) + mi_slice_t* slice = &segment->slices[slice_index]; + slice->slice_count = (uint32_t)slice_count; + mi_assert_internal(slice->slice_count == slice_count); // no overflow? + slice->slice_offset = 0; + if (slice_count > 1) { + mi_slice_t* last = &segment->slices[slice_index + slice_count - 1]; + last->slice_count = 0; + last->slice_offset = (uint32_t)(sizeof(mi_page_t)*(slice_count - 1)); + last->xblock_size = 0; + } + + // perhaps decommit + mi_segment_perhaps_decommit(segment,mi_slice_start(slice),slice_count*MI_SEGMENT_SLICE_SIZE,tld->stats); + + // and push it on the free page queue (if it was not a huge page) + if (sq != NULL) mi_span_queue_push( sq, slice ); + else slice->xblock_size = 0; // mark huge page as free anyways +} + +/* +// called from reclaim to add existing free spans +static void mi_segment_span_add_free(mi_slice_t* slice, mi_segments_tld_t* tld) { + mi_segment_t* segment = _mi_ptr_segment(slice); + mi_assert_internal(slice->xblock_size==0 && slice->slice_count>0 && slice->slice_offset==0); + size_t slice_index = mi_slice_index(slice); + mi_segment_span_free(segment,slice_index,slice->slice_count,tld); +} +*/ + +static void mi_segment_span_remove_from_queue(mi_slice_t* slice, mi_segments_tld_t* tld) { + mi_assert_internal(slice->slice_count > 0 && slice->slice_offset==0 && slice->xblock_size==0); + mi_assert_internal(_mi_ptr_segment(slice)->kind != MI_SEGMENT_HUGE); + mi_span_queue_t* sq = mi_span_queue_for(slice->slice_count, tld); + mi_span_queue_delete(sq, slice); +} + +// note: can be called on abandoned segments +static mi_slice_t* mi_segment_span_free_coalesce(mi_slice_t* slice, mi_segments_tld_t* tld) { + mi_assert_internal(slice != NULL && slice->slice_count > 0 && slice->slice_offset == 0); + mi_segment_t* segment = _mi_ptr_segment(slice); + bool is_abandoned = mi_segment_is_abandoned(segment); + + // for huge pages, just mark as free but don't add to the queues + if (segment->kind == MI_SEGMENT_HUGE) { + mi_assert_internal(segment->used == 1); // decreased right after this call in `mi_segment_page_clear` + slice->xblock_size = 0; // mark as free anyways + // we should mark the last slice `xblock_size=0` now to maintain invariants but we skip it to + // avoid a possible cache miss (and the segment is about to be freed) + return slice; + } + + // otherwise coalesce the span and add to the free span queues + size_t slice_count = slice->slice_count; + mi_slice_t* next = slice + slice->slice_count; + mi_assert_internal(next <= mi_segment_slices_end(segment)); + if (next < mi_segment_slices_end(segment) && next->xblock_size==0) { + // free next block -- remove it from free and merge + mi_assert_internal(next->slice_count > 0 && next->slice_offset==0); + slice_count += next->slice_count; // extend + if (!is_abandoned) { mi_segment_span_remove_from_queue(next, tld); } + } + if (slice > segment->slices) { + mi_slice_t* prev = mi_slice_first(slice - 1); + mi_assert_internal(prev >= segment->slices); + if (prev->xblock_size==0) { + // free previous slice -- remove it from free and merge + mi_assert_internal(prev->slice_count > 0 && prev->slice_offset==0); + slice_count += prev->slice_count; + if (!is_abandoned) { mi_segment_span_remove_from_queue(prev, tld); } + slice = prev; + } + } + + // and add the new free page + mi_segment_span_free(segment, mi_slice_index(slice), slice_count, tld); + return slice; +} + + +static void mi_segment_slice_split(mi_segment_t* segment, mi_slice_t* slice, size_t slice_count, mi_segments_tld_t* tld) { + mi_assert_internal(_mi_ptr_segment(slice)==segment); + mi_assert_internal(slice->slice_count >= slice_count); + mi_assert_internal(slice->xblock_size > 0); // no more in free queue + if (slice->slice_count <= slice_count) return; + mi_assert_internal(segment->kind != MI_SEGMENT_HUGE); + size_t next_index = mi_slice_index(slice) + slice_count; + size_t next_count = slice->slice_count - slice_count; + mi_segment_span_free(segment, next_index, next_count, tld); + slice->slice_count = (uint32_t)slice_count; +} + +// Note: may still return NULL if committing the memory failed +static mi_page_t* mi_segment_span_allocate(mi_segment_t* segment, size_t slice_index, size_t slice_count, mi_segments_tld_t* tld) { + mi_assert_internal(slice_index < segment->slice_entries); + mi_slice_t* slice = &segment->slices[slice_index]; + mi_assert_internal(slice->xblock_size==0 || slice->xblock_size==1); + + // commit before changing the slice data + if (!mi_segment_ensure_committed(segment, _mi_segment_page_start_from_slice(segment, slice, 0, NULL), slice_count * MI_SEGMENT_SLICE_SIZE, tld->stats)) { + return NULL; // commit failed! + } + + // convert the slices to a page + slice->slice_offset = 0; + slice->slice_count = (uint32_t)slice_count; + mi_assert_internal(slice->slice_count == slice_count); + const size_t bsize = slice_count * MI_SEGMENT_SLICE_SIZE; + slice->xblock_size = (uint32_t)(bsize >= MI_HUGE_BLOCK_SIZE ? MI_HUGE_BLOCK_SIZE : bsize); + mi_page_t* page = mi_slice_to_page(slice); + mi_assert_internal(mi_page_block_size(page) == bsize); + + // set slice back pointers for the first MI_MAX_SLICE_OFFSET entries + size_t extra = slice_count-1; + if (extra > MI_MAX_SLICE_OFFSET) extra = MI_MAX_SLICE_OFFSET; + if (slice_index + extra >= segment->slice_entries) extra = segment->slice_entries - slice_index - 1; // huge objects may have more slices than avaiable entries in the segment->slices + slice++; + for (size_t i = 1; i <= extra; i++, slice++) { + slice->slice_offset = (uint32_t)(sizeof(mi_slice_t)*i); + slice->slice_count = 0; + slice->xblock_size = 1; + } + + // and also for the last one (if not set already) (the last one is needed for coalescing) + // note: the cast is needed for ubsan since the index can be larger than MI_SLICES_PER_SEGMENT for huge allocations (see #543) + mi_slice_t* last = &((mi_slice_t*)segment->slices)[slice_index + slice_count - 1]; + if (last < mi_segment_slices_end(segment) && last >= slice) { + last->slice_offset = (uint32_t)(sizeof(mi_slice_t)*(slice_count-1)); + last->slice_count = 0; + last->xblock_size = 1; + } + + // and initialize the page + page->is_reset = false; + page->is_committed = true; + segment->used++; + return page; +} + +static mi_page_t* mi_segments_page_find_and_allocate(size_t slice_count, mi_segments_tld_t* tld) { + mi_assert_internal(slice_count*MI_SEGMENT_SLICE_SIZE <= MI_LARGE_OBJ_SIZE_MAX); + // search from best fit up + mi_span_queue_t* sq = mi_span_queue_for(slice_count, tld); + if (slice_count == 0) slice_count = 1; + while (sq <= &tld->spans[MI_SEGMENT_BIN_MAX]) { + for (mi_slice_t* slice = sq->first; slice != NULL; slice = slice->next) { + if (slice->slice_count >= slice_count) { + // found one + mi_span_queue_delete(sq, slice); + mi_segment_t* segment = _mi_ptr_segment(slice); + if (slice->slice_count > slice_count) { + mi_segment_slice_split(segment, slice, slice_count, tld); + } + mi_assert_internal(slice != NULL && slice->slice_count == slice_count && slice->xblock_size > 0); + mi_page_t* page = mi_segment_span_allocate(segment, mi_slice_index(slice), slice->slice_count, tld); + if (page == NULL) { + // commit failed; return NULL but first restore the slice + mi_segment_span_free_coalesce(slice, tld); + return NULL; + } + return page; + } + } + sq++; + } + // could not find a page.. + return NULL; +} + + +/* ----------------------------------------------------------- + Segment allocation +----------------------------------------------------------- */ + +// Allocate a segment from the OS aligned to `MI_SEGMENT_SIZE` . +static mi_segment_t* mi_segment_init(mi_segment_t* segment, size_t required, mi_segments_tld_t* tld, mi_os_tld_t* os_tld, mi_page_t** huge_page) +{ + mi_assert_internal((required==0 && huge_page==NULL) || (required>0 && huge_page != NULL)); + mi_assert_internal((segment==NULL) || (segment!=NULL && required==0)); + // calculate needed sizes first + size_t info_slices; + size_t pre_size; + const size_t segment_slices = mi_segment_calculate_slices(required, &pre_size, &info_slices); + const size_t slice_entries = (segment_slices > MI_SLICES_PER_SEGMENT ? MI_SLICES_PER_SEGMENT : segment_slices); + const size_t segment_size = segment_slices * MI_SEGMENT_SLICE_SIZE; + + // Commit eagerly only if not the first N lazy segments (to reduce impact of many threads that allocate just a little) + const bool eager_delay = (// !_mi_os_has_overcommit() && // never delay on overcommit systems + _mi_current_thread_count() > 1 && // do not delay for the first N threads + tld->count < (size_t)mi_option_get(mi_option_eager_commit_delay)); + const bool eager = !eager_delay && mi_option_is_enabled(mi_option_eager_commit); + bool commit = eager || (required > 0); + + // Try to get from our cache first + bool is_zero = false; + const bool commit_info_still_good = (segment != NULL); + mi_commit_mask_t commit_mask; + mi_commit_mask_t decommit_mask; + if (segment != NULL) { + commit_mask = segment->commit_mask; + decommit_mask = segment->decommit_mask; + } + else { + mi_commit_mask_create_empty(&commit_mask); + mi_commit_mask_create_empty(&decommit_mask); + } + if (segment==NULL) { + // Allocate the segment from the OS + bool mem_large = (!eager_delay && (MI_SECURE==0)); // only allow large OS pages once we are no longer lazy + bool is_pinned = false; + size_t memid = 0; + segment = (mi_segment_t*)_mi_segment_cache_pop(segment_size, &commit_mask, &decommit_mask, &mem_large, &is_pinned, &is_zero, &memid, os_tld); + if (segment==NULL) { + segment = (mi_segment_t*)_mi_arena_alloc_aligned(segment_size, MI_SEGMENT_SIZE, &commit, &mem_large, &is_pinned, &is_zero, &memid, os_tld); + if (segment == NULL) return NULL; // failed to allocate + if (commit) { + mi_commit_mask_create_full(&commit_mask); + } + else { + mi_commit_mask_create_empty(&commit_mask); + } + } + mi_assert_internal(segment != NULL && (uintptr_t)segment % MI_SEGMENT_SIZE == 0); + + const size_t commit_needed = _mi_divide_up(info_slices*MI_SEGMENT_SLICE_SIZE, MI_COMMIT_SIZE); + mi_assert_internal(commit_needed>0); + mi_commit_mask_t commit_needed_mask; + mi_commit_mask_create(0, commit_needed, &commit_needed_mask); + if (!mi_commit_mask_all_set(&commit_mask, &commit_needed_mask)) { + // at least commit the info slices + mi_assert_internal(commit_needed*MI_COMMIT_SIZE >= info_slices*MI_SEGMENT_SLICE_SIZE); + bool ok = _mi_os_commit(segment, commit_needed*MI_COMMIT_SIZE, &is_zero, tld->stats); + if (!ok) return NULL; // failed to commit + mi_commit_mask_set(&commit_mask, &commit_needed_mask); + } + segment->memid = memid; + segment->mem_is_pinned = is_pinned; + segment->mem_is_large = mem_large; + segment->mem_is_committed = mi_commit_mask_is_full(&commit_mask); + mi_segments_track_size((long)(segment_size), tld); + _mi_segment_map_allocated_at(segment); + } + + // zero the segment info? -- not always needed as it is zero initialized from the OS + mi_atomic_store_ptr_release(mi_segment_t, &segment->abandoned_next, NULL); // tsan + if (!is_zero) { + ptrdiff_t ofs = offsetof(mi_segment_t, next); + size_t prefix = offsetof(mi_segment_t, slices) - ofs; + memset((uint8_t*)segment+ofs, 0, prefix + sizeof(mi_slice_t)*segment_slices); + } + + if (!commit_info_still_good) { + segment->commit_mask = commit_mask; // on lazy commit, the initial part is always committed + segment->allow_decommit = (mi_option_is_enabled(mi_option_allow_decommit) && !segment->mem_is_pinned && !segment->mem_is_large); + if (segment->allow_decommit) { + segment->decommit_expire = _mi_clock_now() + mi_option_get(mi_option_decommit_delay); + segment->decommit_mask = decommit_mask; + mi_assert_internal(mi_commit_mask_all_set(&segment->commit_mask, &segment->decommit_mask)); + #if MI_DEBUG>2 + const size_t commit_needed = _mi_divide_up(info_slices*MI_SEGMENT_SLICE_SIZE, MI_COMMIT_SIZE); + mi_commit_mask_t commit_needed_mask; + mi_commit_mask_create(0, commit_needed, &commit_needed_mask); + mi_assert_internal(!mi_commit_mask_any_set(&segment->decommit_mask, &commit_needed_mask)); + #endif + } + else { + mi_assert_internal(mi_commit_mask_is_empty(&decommit_mask)); + segment->decommit_expire = 0; + mi_commit_mask_create_empty( &segment->decommit_mask ); + mi_assert_internal(mi_commit_mask_is_empty(&segment->decommit_mask)); + } + } + + + // initialize segment info + segment->segment_slices = segment_slices; + segment->segment_info_slices = info_slices; + segment->thread_id = _mi_thread_id(); + segment->cookie = _mi_ptr_cookie(segment); + segment->slice_entries = slice_entries; + segment->kind = (required == 0 ? MI_SEGMENT_NORMAL : MI_SEGMENT_HUGE); + + // memset(segment->slices, 0, sizeof(mi_slice_t)*(info_slices+1)); + _mi_stat_increase(&tld->stats->page_committed, mi_segment_info_size(segment)); + + // set up guard pages + size_t guard_slices = 0; + if (MI_SECURE>0) { + // in secure mode, we set up a protected page in between the segment info + // and the page data, and at the end of the segment. + size_t os_pagesize = _mi_os_page_size(); + mi_assert_internal(mi_segment_info_size(segment) - os_pagesize >= pre_size); + _mi_os_protect((uint8_t*)segment + mi_segment_info_size(segment) - os_pagesize, os_pagesize); + uint8_t* end = (uint8_t*)segment + mi_segment_size(segment) - os_pagesize; + mi_segment_ensure_committed(segment, end, os_pagesize, tld->stats); + _mi_os_protect(end, os_pagesize); + if (slice_entries == segment_slices) segment->slice_entries--; // don't use the last slice :-( + guard_slices = 1; + } + + // reserve first slices for segment info + mi_page_t* page0 = mi_segment_span_allocate(segment, 0, info_slices, tld); + mi_assert_internal(page0!=NULL); if (page0==NULL) return NULL; // cannot fail as we always commit in advance + mi_assert_internal(segment->used == 1); + segment->used = 0; // don't count our internal slices towards usage + + // initialize initial free pages + if (segment->kind == MI_SEGMENT_NORMAL) { // not a huge page + mi_assert_internal(huge_page==NULL); + mi_segment_span_free(segment, info_slices, segment->slice_entries - info_slices, tld); + } + else { + mi_assert_internal(huge_page!=NULL); + mi_assert_internal(mi_commit_mask_is_empty(&segment->decommit_mask)); + mi_assert_internal(mi_commit_mask_is_full(&segment->commit_mask)); + *huge_page = mi_segment_span_allocate(segment, info_slices, segment_slices - info_slices - guard_slices, tld); + mi_assert_internal(*huge_page != NULL); // cannot fail as we commit in advance + } + + mi_assert_expensive(mi_segment_is_valid(segment,tld)); + return segment; +} + + +// Allocate a segment from the OS aligned to `MI_SEGMENT_SIZE` . +static mi_segment_t* mi_segment_alloc(size_t required, mi_segments_tld_t* tld, mi_os_tld_t* os_tld, mi_page_t** huge_page) { + return mi_segment_init(NULL, required, tld, os_tld, huge_page); +} + + +static void mi_segment_free(mi_segment_t* segment, bool force, mi_segments_tld_t* tld) { + MI_UNUSED(force); + mi_assert_internal(segment != NULL); + mi_assert_internal(segment->next == NULL); + mi_assert_internal(segment->used == 0); + + // Remove the free pages + mi_slice_t* slice = &segment->slices[0]; + const mi_slice_t* end = mi_segment_slices_end(segment); + size_t page_count = 0; + while (slice < end) { + mi_assert_internal(slice->slice_count > 0); + mi_assert_internal(slice->slice_offset == 0); + mi_assert_internal(mi_slice_index(slice)==0 || slice->xblock_size == 0); // no more used pages .. + if (slice->xblock_size == 0 && segment->kind != MI_SEGMENT_HUGE) { + mi_segment_span_remove_from_queue(slice, tld); + } + page_count++; + slice = slice + slice->slice_count; + } + mi_assert_internal(page_count == 2); // first page is allocated by the segment itself + + // stats + _mi_stat_decrease(&tld->stats->page_committed, mi_segment_info_size(segment)); + + // return it to the OS + mi_segment_os_free(segment, tld); +} + + +/* ----------------------------------------------------------- + Page Free +----------------------------------------------------------- */ + +static void mi_segment_abandon(mi_segment_t* segment, mi_segments_tld_t* tld); + +// note: can be called on abandoned pages +static mi_slice_t* mi_segment_page_clear(mi_page_t* page, mi_segments_tld_t* tld) { + mi_assert_internal(page->xblock_size > 0); + mi_assert_internal(mi_page_all_free(page)); + mi_segment_t* segment = _mi_ptr_segment(page); + mi_assert_internal(segment->used > 0); + + size_t inuse = page->capacity * mi_page_block_size(page); + _mi_stat_decrease(&tld->stats->page_committed, inuse); + _mi_stat_decrease(&tld->stats->pages, 1); + + // reset the page memory to reduce memory pressure? + if (!segment->mem_is_pinned && !page->is_reset && mi_option_is_enabled(mi_option_page_reset)) { + size_t psize; + uint8_t* start = _mi_page_start(segment, page, &psize); + page->is_reset = true; + _mi_os_reset(start, psize, tld->stats); + } + + // zero the page data, but not the segment fields + page->is_zero_init = false; + ptrdiff_t ofs = offsetof(mi_page_t, capacity); + memset((uint8_t*)page + ofs, 0, sizeof(*page) - ofs); + page->xblock_size = 1; + + // and free it + mi_slice_t* slice = mi_segment_span_free_coalesce(mi_page_to_slice(page), tld); + segment->used--; + // cannot assert segment valid as it is called during reclaim + // mi_assert_expensive(mi_segment_is_valid(segment, tld)); + return slice; +} + +void _mi_segment_page_free(mi_page_t* page, bool force, mi_segments_tld_t* tld) +{ + mi_assert(page != NULL); + + mi_segment_t* segment = _mi_page_segment(page); + mi_assert_expensive(mi_segment_is_valid(segment,tld)); + + // mark it as free now + mi_segment_page_clear(page, tld); + mi_assert_expensive(mi_segment_is_valid(segment, tld)); + + if (segment->used == 0) { + // no more used pages; remove from the free list and free the segment + mi_segment_free(segment, force, tld); + } + else if (segment->used == segment->abandoned) { + // only abandoned pages; remove from free list and abandon + mi_segment_abandon(segment,tld); + } +} + + +/* ----------------------------------------------------------- +Abandonment + +When threads terminate, they can leave segments with +live blocks (reachable through other threads). Such segments +are "abandoned" and will be reclaimed by other threads to +reuse their pages and/or free them eventually + +We maintain a global list of abandoned segments that are +reclaimed on demand. Since this is shared among threads +the implementation needs to avoid the A-B-A problem on +popping abandoned segments: +We use tagged pointers to avoid accidentially identifying +reused segments, much like stamped references in Java. +Secondly, we maintain a reader counter to avoid resetting +or decommitting segments that have a pending read operation. + +Note: the current implementation is one possible design; +another way might be to keep track of abandoned segments +in the arenas/segment_cache's. This would have the advantage of keeping +all concurrent code in one place and not needing to deal +with ABA issues. The drawback is that it is unclear how to +scan abandoned segments efficiently in that case as they +would be spread among all other segments in the arenas. +----------------------------------------------------------- */ + +// Use the bottom 20-bits (on 64-bit) of the aligned segment pointers +// to put in a tag that increments on update to avoid the A-B-A problem. +#define MI_TAGGED_MASK MI_SEGMENT_MASK +typedef uintptr_t mi_tagged_segment_t; + +static mi_segment_t* mi_tagged_segment_ptr(mi_tagged_segment_t ts) { + return (mi_segment_t*)(ts & ~MI_TAGGED_MASK); +} + +static mi_tagged_segment_t mi_tagged_segment(mi_segment_t* segment, mi_tagged_segment_t ts) { + mi_assert_internal(((uintptr_t)segment & MI_TAGGED_MASK) == 0); + uintptr_t tag = ((ts & MI_TAGGED_MASK) + 1) & MI_TAGGED_MASK; + return ((uintptr_t)segment | tag); +} + +// This is a list of visited abandoned pages that were full at the time. +// this list migrates to `abandoned` when that becomes NULL. The use of +// this list reduces contention and the rate at which segments are visited. +static mi_decl_cache_align _Atomic(mi_segment_t*) abandoned_visited; // = NULL + +// The abandoned page list (tagged as it supports pop) +static mi_decl_cache_align _Atomic(mi_tagged_segment_t) abandoned; // = NULL + +// Maintain these for debug purposes (these counts may be a bit off) +static mi_decl_cache_align _Atomic(size_t) abandoned_count; +static mi_decl_cache_align _Atomic(size_t) abandoned_visited_count; + +// We also maintain a count of current readers of the abandoned list +// in order to prevent resetting/decommitting segment memory if it might +// still be read. +static mi_decl_cache_align _Atomic(size_t) abandoned_readers; // = 0 + +// Push on the visited list +static void mi_abandoned_visited_push(mi_segment_t* segment) { + mi_assert_internal(segment->thread_id == 0); + mi_assert_internal(mi_atomic_load_ptr_relaxed(mi_segment_t,&segment->abandoned_next) == NULL); + mi_assert_internal(segment->next == NULL); + mi_assert_internal(segment->used > 0); + mi_segment_t* anext = mi_atomic_load_ptr_relaxed(mi_segment_t, &abandoned_visited); + do { + mi_atomic_store_ptr_release(mi_segment_t, &segment->abandoned_next, anext); + } while (!mi_atomic_cas_ptr_weak_release(mi_segment_t, &abandoned_visited, &anext, segment)); + mi_atomic_increment_relaxed(&abandoned_visited_count); +} + +// Move the visited list to the abandoned list. +static bool mi_abandoned_visited_revisit(void) +{ + // quick check if the visited list is empty + if (mi_atomic_load_ptr_relaxed(mi_segment_t, &abandoned_visited) == NULL) return false; + + // grab the whole visited list + mi_segment_t* first = mi_atomic_exchange_ptr_acq_rel(mi_segment_t, &abandoned_visited, NULL); + if (first == NULL) return false; + + // first try to swap directly if the abandoned list happens to be NULL + mi_tagged_segment_t afirst; + mi_tagged_segment_t ts = mi_atomic_load_relaxed(&abandoned); + if (mi_tagged_segment_ptr(ts)==NULL) { + size_t count = mi_atomic_load_relaxed(&abandoned_visited_count); + afirst = mi_tagged_segment(first, ts); + if (mi_atomic_cas_strong_acq_rel(&abandoned, &ts, afirst)) { + mi_atomic_add_relaxed(&abandoned_count, count); + mi_atomic_sub_relaxed(&abandoned_visited_count, count); + return true; + } + } + + // find the last element of the visited list: O(n) + mi_segment_t* last = first; + mi_segment_t* next; + while ((next = mi_atomic_load_ptr_relaxed(mi_segment_t, &last->abandoned_next)) != NULL) { + last = next; + } + + // and atomically prepend to the abandoned list + // (no need to increase the readers as we don't access the abandoned segments) + mi_tagged_segment_t anext = mi_atomic_load_relaxed(&abandoned); + size_t count; + do { + count = mi_atomic_load_relaxed(&abandoned_visited_count); + mi_atomic_store_ptr_release(mi_segment_t, &last->abandoned_next, mi_tagged_segment_ptr(anext)); + afirst = mi_tagged_segment(first, anext); + } while (!mi_atomic_cas_weak_release(&abandoned, &anext, afirst)); + mi_atomic_add_relaxed(&abandoned_count, count); + mi_atomic_sub_relaxed(&abandoned_visited_count, count); + return true; +} + +// Push on the abandoned list. +static void mi_abandoned_push(mi_segment_t* segment) { + mi_assert_internal(segment->thread_id == 0); + mi_assert_internal(mi_atomic_load_ptr_relaxed(mi_segment_t, &segment->abandoned_next) == NULL); + mi_assert_internal(segment->next == NULL); + mi_assert_internal(segment->used > 0); + mi_tagged_segment_t next; + mi_tagged_segment_t ts = mi_atomic_load_relaxed(&abandoned); + do { + mi_atomic_store_ptr_release(mi_segment_t, &segment->abandoned_next, mi_tagged_segment_ptr(ts)); + next = mi_tagged_segment(segment, ts); + } while (!mi_atomic_cas_weak_release(&abandoned, &ts, next)); + mi_atomic_increment_relaxed(&abandoned_count); +} + +// Wait until there are no more pending reads on segments that used to be in the abandoned list +// called for example from `arena.c` before decommitting +void _mi_abandoned_await_readers(void) { + size_t n; + do { + n = mi_atomic_load_acquire(&abandoned_readers); + if (n != 0) mi_atomic_yield(); + } while (n != 0); +} + +// Pop from the abandoned list +static mi_segment_t* mi_abandoned_pop(void) { + mi_segment_t* segment; + // Check efficiently if it is empty (or if the visited list needs to be moved) + mi_tagged_segment_t ts = mi_atomic_load_relaxed(&abandoned); + segment = mi_tagged_segment_ptr(ts); + if (mi_likely(segment == NULL)) { + if (mi_likely(!mi_abandoned_visited_revisit())) { // try to swap in the visited list on NULL + return NULL; + } + } + + // Do a pop. We use a reader count to prevent + // a segment to be decommitted while a read is still pending, + // and a tagged pointer to prevent A-B-A link corruption. + // (this is called from `region.c:_mi_mem_free` for example) + mi_atomic_increment_relaxed(&abandoned_readers); // ensure no segment gets decommitted + mi_tagged_segment_t next = 0; + ts = mi_atomic_load_acquire(&abandoned); + do { + segment = mi_tagged_segment_ptr(ts); + if (segment != NULL) { + mi_segment_t* anext = mi_atomic_load_ptr_relaxed(mi_segment_t, &segment->abandoned_next); + next = mi_tagged_segment(anext, ts); // note: reads the segment's `abandoned_next` field so should not be decommitted + } + } while (segment != NULL && !mi_atomic_cas_weak_acq_rel(&abandoned, &ts, next)); + mi_atomic_decrement_relaxed(&abandoned_readers); // release reader lock + if (segment != NULL) { + mi_atomic_store_ptr_release(mi_segment_t, &segment->abandoned_next, NULL); + mi_atomic_decrement_relaxed(&abandoned_count); + } + return segment; +} + +/* ----------------------------------------------------------- + Abandon segment/page +----------------------------------------------------------- */ + +static void mi_segment_abandon(mi_segment_t* segment, mi_segments_tld_t* tld) { + mi_assert_internal(segment->used == segment->abandoned); + mi_assert_internal(segment->used > 0); + mi_assert_internal(mi_atomic_load_ptr_relaxed(mi_segment_t, &segment->abandoned_next) == NULL); + mi_assert_internal(segment->abandoned_visits == 0); + mi_assert_expensive(mi_segment_is_valid(segment,tld)); + + // remove the free pages from the free page queues + mi_slice_t* slice = &segment->slices[0]; + const mi_slice_t* end = mi_segment_slices_end(segment); + while (slice < end) { + mi_assert_internal(slice->slice_count > 0); + mi_assert_internal(slice->slice_offset == 0); + if (slice->xblock_size == 0) { // a free page + mi_segment_span_remove_from_queue(slice,tld); + slice->xblock_size = 0; // but keep it free + } + slice = slice + slice->slice_count; + } + + // perform delayed decommits + mi_segment_delayed_decommit(segment, mi_option_is_enabled(mi_option_abandoned_page_decommit) /* force? */, tld->stats); + + // all pages in the segment are abandoned; add it to the abandoned list + _mi_stat_increase(&tld->stats->segments_abandoned, 1); + mi_segments_track_size(-((long)mi_segment_size(segment)), tld); + segment->thread_id = 0; + mi_atomic_store_ptr_release(mi_segment_t, &segment->abandoned_next, NULL); + segment->abandoned_visits = 1; // from 0 to 1 to signify it is abandoned + mi_abandoned_push(segment); +} + +void _mi_segment_page_abandon(mi_page_t* page, mi_segments_tld_t* tld) { + mi_assert(page != NULL); + mi_assert_internal(mi_page_thread_free_flag(page)==MI_NEVER_DELAYED_FREE); + mi_assert_internal(mi_page_heap(page) == NULL); + mi_segment_t* segment = _mi_page_segment(page); + + mi_assert_expensive(mi_segment_is_valid(segment,tld)); + segment->abandoned++; + + _mi_stat_increase(&tld->stats->pages_abandoned, 1); + mi_assert_internal(segment->abandoned <= segment->used); + if (segment->used == segment->abandoned) { + // all pages are abandoned, abandon the entire segment + mi_segment_abandon(segment, tld); + } +} + +/* ----------------------------------------------------------- + Reclaim abandoned pages +----------------------------------------------------------- */ + +static mi_slice_t* mi_slices_start_iterate(mi_segment_t* segment, const mi_slice_t** end) { + mi_slice_t* slice = &segment->slices[0]; + *end = mi_segment_slices_end(segment); + mi_assert_internal(slice->slice_count>0 && slice->xblock_size>0); // segment allocated page + slice = slice + slice->slice_count; // skip the first segment allocated page + return slice; +} + +// Possibly free pages and check if free space is available +static bool mi_segment_check_free(mi_segment_t* segment, size_t slices_needed, size_t block_size, mi_segments_tld_t* tld) +{ + mi_assert_internal(block_size < MI_HUGE_BLOCK_SIZE); + mi_assert_internal(mi_segment_is_abandoned(segment)); + bool has_page = false; + + // for all slices + const mi_slice_t* end; + mi_slice_t* slice = mi_slices_start_iterate(segment, &end); + while (slice < end) { + mi_assert_internal(slice->slice_count > 0); + mi_assert_internal(slice->slice_offset == 0); + if (mi_slice_is_used(slice)) { // used page + // ensure used count is up to date and collect potential concurrent frees + mi_page_t* const page = mi_slice_to_page(slice); + _mi_page_free_collect(page, false); + if (mi_page_all_free(page)) { + // if this page is all free now, free it without adding to any queues (yet) + mi_assert_internal(page->next == NULL && page->prev==NULL); + _mi_stat_decrease(&tld->stats->pages_abandoned, 1); + segment->abandoned--; + slice = mi_segment_page_clear(page, tld); // re-assign slice due to coalesce! + mi_assert_internal(!mi_slice_is_used(slice)); + if (slice->slice_count >= slices_needed) { + has_page = true; + } + } + else { + if (page->xblock_size == block_size && mi_page_has_any_available(page)) { + // a page has available free blocks of the right size + has_page = true; + } + } + } + else { + // empty span + if (slice->slice_count >= slices_needed) { + has_page = true; + } + } + slice = slice + slice->slice_count; + } + return has_page; +} + +// Reclaim an abandoned segment; returns NULL if the segment was freed +// set `right_page_reclaimed` to `true` if it reclaimed a page of the right `block_size` that was not full. +static mi_segment_t* mi_segment_reclaim(mi_segment_t* segment, mi_heap_t* heap, size_t requested_block_size, bool* right_page_reclaimed, mi_segments_tld_t* tld) { + mi_assert_internal(mi_atomic_load_ptr_relaxed(mi_segment_t, &segment->abandoned_next) == NULL); + mi_assert_expensive(mi_segment_is_valid(segment, tld)); + if (right_page_reclaimed != NULL) { *right_page_reclaimed = false; } + + segment->thread_id = _mi_thread_id(); + segment->abandoned_visits = 0; + mi_segments_track_size((long)mi_segment_size(segment), tld); + mi_assert_internal(segment->next == NULL); + _mi_stat_decrease(&tld->stats->segments_abandoned, 1); + + // for all slices + const mi_slice_t* end; + mi_slice_t* slice = mi_slices_start_iterate(segment, &end); + while (slice < end) { + mi_assert_internal(slice->slice_count > 0); + mi_assert_internal(slice->slice_offset == 0); + if (mi_slice_is_used(slice)) { + // in use: reclaim the page in our heap + mi_page_t* page = mi_slice_to_page(slice); + mi_assert_internal(!page->is_reset); + mi_assert_internal(page->is_committed); + mi_assert_internal(mi_page_thread_free_flag(page)==MI_NEVER_DELAYED_FREE); + mi_assert_internal(mi_page_heap(page) == NULL); + mi_assert_internal(page->next == NULL && page->prev==NULL); + _mi_stat_decrease(&tld->stats->pages_abandoned, 1); + segment->abandoned--; + // set the heap again and allow delayed free again + mi_page_set_heap(page, heap); + _mi_page_use_delayed_free(page, MI_USE_DELAYED_FREE, true); // override never (after heap is set) + _mi_page_free_collect(page, false); // ensure used count is up to date + if (mi_page_all_free(page)) { + // if everything free by now, free the page + slice = mi_segment_page_clear(page, tld); // set slice again due to coalesceing + } + else { + // otherwise reclaim it into the heap + _mi_page_reclaim(heap, page); + if (requested_block_size == page->xblock_size && mi_page_has_any_available(page)) { + if (right_page_reclaimed != NULL) { *right_page_reclaimed = true; } + } + } + } + else { + // the span is free, add it to our page queues + slice = mi_segment_span_free_coalesce(slice, tld); // set slice again due to coalesceing + } + mi_assert_internal(slice->slice_count>0 && slice->slice_offset==0); + slice = slice + slice->slice_count; + } + + mi_assert(segment->abandoned == 0); + if (segment->used == 0) { // due to page_clear + mi_assert_internal(right_page_reclaimed == NULL || !(*right_page_reclaimed)); + mi_segment_free(segment, false, tld); + return NULL; + } + else { + return segment; + } +} + + +void _mi_abandoned_reclaim_all(mi_heap_t* heap, mi_segments_tld_t* tld) { + mi_segment_t* segment; + while ((segment = mi_abandoned_pop()) != NULL) { + mi_segment_reclaim(segment, heap, 0, NULL, tld); + } +} + +static mi_segment_t* mi_segment_try_reclaim(mi_heap_t* heap, size_t needed_slices, size_t block_size, bool* reclaimed, mi_segments_tld_t* tld) +{ + *reclaimed = false; + mi_segment_t* segment; + long max_tries = mi_option_get_clamp(mi_option_max_segment_reclaim, 8, 1024); // limit the work to bound allocation times + while ((max_tries-- > 0) && ((segment = mi_abandoned_pop()) != NULL)) { + segment->abandoned_visits++; + bool has_page = mi_segment_check_free(segment,needed_slices,block_size,tld); // try to free up pages (due to concurrent frees) + if (segment->used == 0) { + // free the segment (by forced reclaim) to make it available to other threads. + // note1: we prefer to free a segment as that might lead to reclaiming another + // segment that is still partially used. + // note2: we could in principle optimize this by skipping reclaim and directly + // freeing but that would violate some invariants temporarily) + mi_segment_reclaim(segment, heap, 0, NULL, tld); + } + else if (has_page) { + // found a large enough free span, or a page of the right block_size with free space + // we return the result of reclaim (which is usually `segment`) as it might free + // the segment due to concurrent frees (in which case `NULL` is returned). + return mi_segment_reclaim(segment, heap, block_size, reclaimed, tld); + } + else if (segment->abandoned_visits > 3) { + // always reclaim on 3rd visit to limit the abandoned queue length. + mi_segment_reclaim(segment, heap, 0, NULL, tld); + } + else { + // otherwise, push on the visited list so it gets not looked at too quickly again + mi_segment_delayed_decommit(segment, true /* force? */, tld->stats); // forced decommit if needed as we may not visit soon again + mi_abandoned_visited_push(segment); + } + } + return NULL; +} + + +void _mi_abandoned_collect(mi_heap_t* heap, bool force, mi_segments_tld_t* tld) +{ + mi_segment_t* segment; + int max_tries = (force ? 16*1024 : 1024); // limit latency + if (force) { + mi_abandoned_visited_revisit(); + } + while ((max_tries-- > 0) && ((segment = mi_abandoned_pop()) != NULL)) { + mi_segment_check_free(segment,0,0,tld); // try to free up pages (due to concurrent frees) + if (segment->used == 0) { + // free the segment (by forced reclaim) to make it available to other threads. + // note: we could in principle optimize this by skipping reclaim and directly + // freeing but that would violate some invariants temporarily) + mi_segment_reclaim(segment, heap, 0, NULL, tld); + } + else { + // otherwise, decommit if needed and push on the visited list + // note: forced decommit can be expensive if many threads are destroyed/created as in mstress. + mi_segment_delayed_decommit(segment, force, tld->stats); + mi_abandoned_visited_push(segment); + } + } +} + +/* ----------------------------------------------------------- + Reclaim or allocate +----------------------------------------------------------- */ + +static mi_segment_t* mi_segment_reclaim_or_alloc(mi_heap_t* heap, size_t needed_slices, size_t block_size, mi_segments_tld_t* tld, mi_os_tld_t* os_tld) +{ + mi_assert_internal(block_size < MI_HUGE_BLOCK_SIZE); + mi_assert_internal(block_size <= MI_LARGE_OBJ_SIZE_MAX); + + // 1. try to reclaim an abandoned segment + bool reclaimed; + mi_segment_t* segment = mi_segment_try_reclaim(heap, needed_slices, block_size, &reclaimed, tld); + if (reclaimed) { + // reclaimed the right page right into the heap + mi_assert_internal(segment != NULL); + return NULL; // pretend out-of-memory as the page will be in the page queue of the heap with available blocks + } + else if (segment != NULL) { + // reclaimed a segment with a large enough empty span in it + return segment; + } + // 2. otherwise allocate a fresh segment + return mi_segment_alloc(0, tld, os_tld, NULL); +} + + +/* ----------------------------------------------------------- + Page allocation +----------------------------------------------------------- */ + +static mi_page_t* mi_segments_page_alloc(mi_heap_t* heap, mi_page_kind_t page_kind, size_t required, size_t block_size, mi_segments_tld_t* tld, mi_os_tld_t* os_tld) +{ + mi_assert_internal(required <= MI_LARGE_OBJ_SIZE_MAX && page_kind <= MI_PAGE_LARGE); + + // find a free page + size_t page_size = _mi_align_up(required, (required > MI_MEDIUM_PAGE_SIZE ? MI_MEDIUM_PAGE_SIZE : MI_SEGMENT_SLICE_SIZE)); + size_t slices_needed = page_size / MI_SEGMENT_SLICE_SIZE; + mi_assert_internal(slices_needed * MI_SEGMENT_SLICE_SIZE == page_size); + mi_page_t* page = mi_segments_page_find_and_allocate(slices_needed, tld); //(required <= MI_SMALL_SIZE_MAX ? 0 : slices_needed), tld); + if (page==NULL) { + // no free page, allocate a new segment and try again + if (mi_segment_reclaim_or_alloc(heap, slices_needed, block_size, tld, os_tld) == NULL) { + // OOM or reclaimed a good page in the heap + return NULL; + } + else { + // otherwise try again + return mi_segments_page_alloc(heap, page_kind, required, block_size, tld, os_tld); + } + } + mi_assert_internal(page != NULL && page->slice_count*MI_SEGMENT_SLICE_SIZE == page_size); + mi_assert_internal(_mi_ptr_segment(page)->thread_id == _mi_thread_id()); + mi_segment_delayed_decommit(_mi_ptr_segment(page), false, tld->stats); + return page; +} + + + +/* ----------------------------------------------------------- + Huge page allocation +----------------------------------------------------------- */ + +static mi_page_t* mi_segment_huge_page_alloc(size_t size, mi_segments_tld_t* tld, mi_os_tld_t* os_tld) +{ + mi_page_t* page = NULL; + mi_segment_t* segment = mi_segment_alloc(size,tld,os_tld,&page); + if (segment == NULL || page==NULL) return NULL; + mi_assert_internal(segment->used==1); + mi_assert_internal(mi_page_block_size(page) >= size); + segment->thread_id = 0; // huge segments are immediately abandoned + return page; +} + +// free huge block from another thread +void _mi_segment_huge_page_free(mi_segment_t* segment, mi_page_t* page, mi_block_t* block) { + // huge page segments are always abandoned and can be freed immediately by any thread + mi_assert_internal(segment->kind==MI_SEGMENT_HUGE); + mi_assert_internal(segment == _mi_page_segment(page)); + mi_assert_internal(mi_atomic_load_relaxed(&segment->thread_id)==0); + + // claim it and free + mi_heap_t* heap = mi_heap_get_default(); // issue #221; don't use the internal get_default_heap as we need to ensure the thread is initialized. + // paranoia: if this it the last reference, the cas should always succeed + size_t expected_tid = 0; + if (mi_atomic_cas_strong_acq_rel(&segment->thread_id, &expected_tid, heap->thread_id)) { + mi_block_set_next(page, block, page->free); + page->free = block; + page->used--; + page->is_zero = false; + mi_assert(page->used == 0); + mi_tld_t* tld = heap->tld; + _mi_segment_page_free(page, true, &tld->segments); + } +#if (MI_DEBUG!=0) + else { + mi_assert_internal(false); + } +#endif +} + +/* ----------------------------------------------------------- + Page allocation and free +----------------------------------------------------------- */ +mi_page_t* _mi_segment_page_alloc(mi_heap_t* heap, size_t block_size, mi_segments_tld_t* tld, mi_os_tld_t* os_tld) { + mi_page_t* page; + if (block_size <= MI_SMALL_OBJ_SIZE_MAX) { + page = mi_segments_page_alloc(heap,MI_PAGE_SMALL,block_size,block_size,tld,os_tld); + } + else if (block_size <= MI_MEDIUM_OBJ_SIZE_MAX) { + page = mi_segments_page_alloc(heap,MI_PAGE_MEDIUM,MI_MEDIUM_PAGE_SIZE,block_size,tld, os_tld); + } + else if (block_size <= MI_LARGE_OBJ_SIZE_MAX) { + page = mi_segments_page_alloc(heap,MI_PAGE_LARGE,block_size,block_size,tld, os_tld); + } + else { + page = mi_segment_huge_page_alloc(block_size,tld,os_tld); + } + mi_assert_expensive(page == NULL || mi_segment_is_valid(_mi_page_segment(page),tld)); + return page; +} diff --git a/compat/mimalloc/stats.c b/compat/mimalloc/stats.c new file mode 100644 index 00000000000000..c722189f7044e7 --- /dev/null +++ b/compat/mimalloc/stats.c @@ -0,0 +1,589 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2021, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ +#include "mimalloc.h" +#include "mimalloc-internal.h" +#include "mimalloc-atomic.h" + +#include // fputs, stderr +#include // memset + +#if defined(_MSC_VER) && (_MSC_VER < 1920) +#pragma warning(disable:4204) // non-constant aggregate initializer +#endif + +/* ----------------------------------------------------------- + Statistics operations +----------------------------------------------------------- */ + +static bool mi_is_in_main(void* stat) { + return ((uint8_t*)stat >= (uint8_t*)&_mi_stats_main + && (uint8_t*)stat < ((uint8_t*)&_mi_stats_main + sizeof(mi_stats_t))); +} + +static void mi_stat_update(mi_stat_count_t* stat, int64_t amount) { + if (amount == 0) return; + if (mi_is_in_main(stat)) + { + // add atomically (for abandoned pages) + int64_t current = mi_atomic_addi64_relaxed(&stat->current, amount); + mi_atomic_maxi64_relaxed(&stat->peak, current + amount); + if (amount > 0) { + mi_atomic_addi64_relaxed(&stat->allocated,amount); + } + else { + mi_atomic_addi64_relaxed(&stat->freed, -amount); + } + } + else { + // add thread local + stat->current += amount; + if (stat->current > stat->peak) stat->peak = stat->current; + if (amount > 0) { + stat->allocated += amount; + } + else { + stat->freed += -amount; + } + } +} + +void _mi_stat_counter_increase(mi_stat_counter_t* stat, size_t amount) { + if (mi_is_in_main(stat)) { + mi_atomic_addi64_relaxed( &stat->count, 1 ); + mi_atomic_addi64_relaxed( &stat->total, (int64_t)amount ); + } + else { + stat->count++; + stat->total += amount; + } +} + +void _mi_stat_increase(mi_stat_count_t* stat, size_t amount) { + mi_stat_update(stat, (int64_t)amount); +} + +void _mi_stat_decrease(mi_stat_count_t* stat, size_t amount) { + mi_stat_update(stat, -((int64_t)amount)); +} + +// must be thread safe as it is called from stats_merge +static void mi_stat_add(mi_stat_count_t* stat, const mi_stat_count_t* src, int64_t unit) { + if (stat==src) return; + if (src->allocated==0 && src->freed==0) return; + mi_atomic_addi64_relaxed( &stat->allocated, src->allocated * unit); + mi_atomic_addi64_relaxed( &stat->current, src->current * unit); + mi_atomic_addi64_relaxed( &stat->freed, src->freed * unit); + // peak scores do not work across threads.. + mi_atomic_addi64_relaxed( &stat->peak, src->peak * unit); +} + +static void mi_stat_counter_add(mi_stat_counter_t* stat, const mi_stat_counter_t* src, int64_t unit) { + if (stat==src) return; + mi_atomic_addi64_relaxed( &stat->total, src->total * unit); + mi_atomic_addi64_relaxed( &stat->count, src->count * unit); +} + +// must be thread safe as it is called from stats_merge +static void mi_stats_add(mi_stats_t* stats, const mi_stats_t* src) { + if (stats==src) return; + mi_stat_add(&stats->segments, &src->segments,1); + mi_stat_add(&stats->pages, &src->pages,1); + mi_stat_add(&stats->reserved, &src->reserved, 1); + mi_stat_add(&stats->committed, &src->committed, 1); + mi_stat_add(&stats->reset, &src->reset, 1); + mi_stat_add(&stats->page_committed, &src->page_committed, 1); + + mi_stat_add(&stats->pages_abandoned, &src->pages_abandoned, 1); + mi_stat_add(&stats->segments_abandoned, &src->segments_abandoned, 1); + mi_stat_add(&stats->threads, &src->threads, 1); + + mi_stat_add(&stats->malloc, &src->malloc, 1); + mi_stat_add(&stats->segments_cache, &src->segments_cache, 1); + mi_stat_add(&stats->normal, &src->normal, 1); + mi_stat_add(&stats->huge, &src->huge, 1); + mi_stat_add(&stats->large, &src->large, 1); + + mi_stat_counter_add(&stats->pages_extended, &src->pages_extended, 1); + mi_stat_counter_add(&stats->mmap_calls, &src->mmap_calls, 1); + mi_stat_counter_add(&stats->commit_calls, &src->commit_calls, 1); + + mi_stat_counter_add(&stats->page_no_retire, &src->page_no_retire, 1); + mi_stat_counter_add(&stats->searches, &src->searches, 1); + mi_stat_counter_add(&stats->normal_count, &src->normal_count, 1); + mi_stat_counter_add(&stats->huge_count, &src->huge_count, 1); + mi_stat_counter_add(&stats->large_count, &src->large_count, 1); +#if MI_STAT>1 + for (size_t i = 0; i <= MI_BIN_HUGE; i++) { + if (src->normal_bins[i].allocated > 0 || src->normal_bins[i].freed > 0) { + mi_stat_add(&stats->normal_bins[i], &src->normal_bins[i], 1); + } + } +#endif +} + +/* ----------------------------------------------------------- + Display statistics +----------------------------------------------------------- */ + +// unit > 0 : size in binary bytes +// unit == 0: count as decimal +// unit < 0 : count in binary +static void mi_printf_amount(int64_t n, int64_t unit, mi_output_fun* out, void* arg, const char* fmt) { + char buf[32]; buf[0] = 0; + int len = 32; + const char* suffix = (unit <= 0 ? " " : "B"); + const int64_t base = (unit == 0 ? 1000 : 1024); + if (unit>0) n *= unit; + + const int64_t pos = (n < 0 ? -n : n); + if (pos < base) { + if (n!=1 || suffix[0] != 'B') { // skip printing 1 B for the unit column + snprintf(buf, len, "%d %-3s", (int)n, (n==0 ? "" : suffix)); + } + } + else { + int64_t divider = base; + const char* magnitude = "K"; + if (pos >= divider*base) { divider *= base; magnitude = "M"; } + if (pos >= divider*base) { divider *= base; magnitude = "G"; } + const int64_t tens = (n / (divider/10)); + const long whole = (long)(tens/10); + const long frac1 = (long)(tens%10); + char unitdesc[8]; + snprintf(unitdesc, 8, "%s%s%s", magnitude, (base==1024 ? "i" : ""), suffix); + snprintf(buf, len, "%ld.%ld %-3s", whole, (frac1 < 0 ? -frac1 : frac1), unitdesc); + } + _mi_fprintf(out, arg, (fmt==NULL ? "%11s" : fmt), buf); +} + + +static void mi_print_amount(int64_t n, int64_t unit, mi_output_fun* out, void* arg) { + mi_printf_amount(n,unit,out,arg,NULL); +} + +static void mi_print_count(int64_t n, int64_t unit, mi_output_fun* out, void* arg) { + if (unit==1) _mi_fprintf(out, arg, "%11s"," "); + else mi_print_amount(n,0,out,arg); +} + +static void mi_stat_print(const mi_stat_count_t* stat, const char* msg, int64_t unit, mi_output_fun* out, void* arg ) { + _mi_fprintf(out, arg,"%10s:", msg); + if (unit>0) { + mi_print_amount(stat->peak, unit, out, arg); + mi_print_amount(stat->allocated, unit, out, arg); + mi_print_amount(stat->freed, unit, out, arg); + mi_print_amount(stat->current, unit, out, arg); + mi_print_amount(unit, 1, out, arg); + mi_print_count(stat->allocated, unit, out, arg); + if (stat->allocated > stat->freed) + _mi_fprintf(out, arg, " not all freed!\n"); + else + _mi_fprintf(out, arg, " ok\n"); + } + else if (unit<0) { + mi_print_amount(stat->peak, -1, out, arg); + mi_print_amount(stat->allocated, -1, out, arg); + mi_print_amount(stat->freed, -1, out, arg); + mi_print_amount(stat->current, -1, out, arg); + if (unit==-1) { + _mi_fprintf(out, arg, "%22s", ""); + } + else { + mi_print_amount(-unit, 1, out, arg); + mi_print_count((stat->allocated / -unit), 0, out, arg); + } + if (stat->allocated > stat->freed) + _mi_fprintf(out, arg, " not all freed!\n"); + else + _mi_fprintf(out, arg, " ok\n"); + } + else { + mi_print_amount(stat->peak, 1, out, arg); + mi_print_amount(stat->allocated, 1, out, arg); + _mi_fprintf(out, arg, "%11s", " "); // no freed + mi_print_amount(stat->current, 1, out, arg); + _mi_fprintf(out, arg, "\n"); + } +} + +static void mi_stat_counter_print(const mi_stat_counter_t* stat, const char* msg, mi_output_fun* out, void* arg ) { + _mi_fprintf(out, arg, "%10s:", msg); + mi_print_amount(stat->total, -1, out, arg); + _mi_fprintf(out, arg, "\n"); +} + +static void mi_stat_counter_print_avg(const mi_stat_counter_t* stat, const char* msg, mi_output_fun* out, void* arg) { + const int64_t avg_tens = (stat->count == 0 ? 0 : (stat->total*10 / stat->count)); + const long avg_whole = (long)(avg_tens/10); + const long avg_frac1 = (long)(avg_tens%10); + _mi_fprintf(out, arg, "%10s: %5ld.%ld avg\n", msg, avg_whole, avg_frac1); +} + + +static void mi_print_header(mi_output_fun* out, void* arg ) { + _mi_fprintf(out, arg, "%10s: %10s %10s %10s %10s %10s %10s\n", "heap stats", "peak ", "total ", "freed ", "current ", "unit ", "count "); +} + +#if MI_STAT>1 +static void mi_stats_print_bins(const mi_stat_count_t* bins, size_t max, const char* fmt, mi_output_fun* out, void* arg) { + bool found = false; + char buf[64]; + for (size_t i = 0; i <= max; i++) { + if (bins[i].allocated > 0) { + found = true; + int64_t unit = _mi_bin_size((uint8_t)i); + snprintf(buf, 64, "%s %3lu", fmt, (long)i); + mi_stat_print(&bins[i], buf, unit, out, arg); + } + } + if (found) { + _mi_fprintf(out, arg, "\n"); + mi_print_header(out, arg); + } +} +#endif + + + +//------------------------------------------------------------ +// Use an output wrapper for line-buffered output +// (which is nice when using loggers etc.) +//------------------------------------------------------------ +typedef struct buffered_s { + mi_output_fun* out; // original output function + void* arg; // and state + char* buf; // local buffer of at least size `count+1` + size_t used; // currently used chars `used <= count` + size_t count; // total chars available for output +} buffered_t; + +static void mi_buffered_flush(buffered_t* buf) { + buf->buf[buf->used] = 0; + _mi_fputs(buf->out, buf->arg, NULL, buf->buf); + buf->used = 0; +} + +static void mi_buffered_out(const char* msg, void* arg) { + buffered_t* buf = (buffered_t*)arg; + if (msg==NULL || buf==NULL) return; + for (const char* src = msg; *src != 0; src++) { + char c = *src; + if (buf->used >= buf->count) mi_buffered_flush(buf); + mi_assert_internal(buf->used < buf->count); + buf->buf[buf->used++] = c; + if (c == '\n') mi_buffered_flush(buf); + } +} + +//------------------------------------------------------------ +// Print statistics +//------------------------------------------------------------ + +static void mi_stat_process_info(mi_msecs_t* elapsed, mi_msecs_t* utime, mi_msecs_t* stime, size_t* current_rss, size_t* peak_rss, size_t* current_commit, size_t* peak_commit, size_t* page_faults); + +static void _mi_stats_print(mi_stats_t* stats, mi_output_fun* out0, void* arg0) mi_attr_noexcept { + // wrap the output function to be line buffered + char buf[256]; + buffered_t buffer = { out0, arg0, NULL, 0, 255 }; + buffer.buf = buf; + mi_output_fun* out = &mi_buffered_out; + void* arg = &buffer; + + // and print using that + mi_print_header(out,arg); + #if MI_STAT>1 + mi_stats_print_bins(stats->normal_bins, MI_BIN_HUGE, "normal",out,arg); + #endif + #if MI_STAT + mi_stat_print(&stats->normal, "normal", (stats->normal_count.count == 0 ? 1 : -(stats->normal.allocated / stats->normal_count.count)), out, arg); + mi_stat_print(&stats->large, "large", (stats->large_count.count == 0 ? 1 : -(stats->large.allocated / stats->large_count.count)), out, arg); + mi_stat_print(&stats->huge, "huge", (stats->huge_count.count == 0 ? 1 : -(stats->huge.allocated / stats->huge_count.count)), out, arg); + mi_stat_count_t total = { 0,0,0,0 }; + mi_stat_add(&total, &stats->normal, 1); + mi_stat_add(&total, &stats->large, 1); + mi_stat_add(&total, &stats->huge, 1); + mi_stat_print(&total, "total", 1, out, arg); + #endif + #if MI_STAT>1 + mi_stat_print(&stats->malloc, "malloc req", 1, out, arg); + _mi_fprintf(out, arg, "\n"); + #endif + mi_stat_print(&stats->reserved, "reserved", 1, out, arg); + mi_stat_print(&stats->committed, "committed", 1, out, arg); + mi_stat_print(&stats->reset, "reset", 1, out, arg); + mi_stat_print(&stats->page_committed, "touched", 1, out, arg); + mi_stat_print(&stats->segments, "segments", -1, out, arg); + mi_stat_print(&stats->segments_abandoned, "-abandoned", -1, out, arg); + mi_stat_print(&stats->segments_cache, "-cached", -1, out, arg); + mi_stat_print(&stats->pages, "pages", -1, out, arg); + mi_stat_print(&stats->pages_abandoned, "-abandoned", -1, out, arg); + mi_stat_counter_print(&stats->pages_extended, "-extended", out, arg); + mi_stat_counter_print(&stats->page_no_retire, "-noretire", out, arg); + mi_stat_counter_print(&stats->mmap_calls, "mmaps", out, arg); + mi_stat_counter_print(&stats->commit_calls, "commits", out, arg); + mi_stat_print(&stats->threads, "threads", -1, out, arg); + mi_stat_counter_print_avg(&stats->searches, "searches", out, arg); + _mi_fprintf(out, arg, "%10s: %7zu\n", "numa nodes", _mi_os_numa_node_count()); + + mi_msecs_t elapsed; + mi_msecs_t user_time; + mi_msecs_t sys_time; + size_t current_rss; + size_t peak_rss; + size_t current_commit; + size_t peak_commit; + size_t page_faults; + mi_stat_process_info(&elapsed, &user_time, &sys_time, ¤t_rss, &peak_rss, ¤t_commit, &peak_commit, &page_faults); + _mi_fprintf(out, arg, "%10s: %7ld.%03ld s\n", "elapsed", elapsed/1000, elapsed%1000); + _mi_fprintf(out, arg, "%10s: user: %ld.%03ld s, system: %ld.%03ld s, faults: %lu, rss: ", "process", + user_time/1000, user_time%1000, sys_time/1000, sys_time%1000, (unsigned long)page_faults ); + mi_printf_amount((int64_t)peak_rss, 1, out, arg, "%s"); + if (peak_commit > 0) { + _mi_fprintf(out, arg, ", commit: "); + mi_printf_amount((int64_t)peak_commit, 1, out, arg, "%s"); + } + _mi_fprintf(out, arg, "\n"); +} + +static mi_msecs_t mi_process_start; // = 0 + +static mi_stats_t* mi_stats_get_default(void) { + mi_heap_t* heap = mi_heap_get_default(); + return &heap->tld->stats; +} + +static void mi_stats_merge_from(mi_stats_t* stats) { + if (stats != &_mi_stats_main) { + mi_stats_add(&_mi_stats_main, stats); + memset(stats, 0, sizeof(mi_stats_t)); + } +} + +void mi_stats_reset(void) mi_attr_noexcept { + mi_stats_t* stats = mi_stats_get_default(); + if (stats != &_mi_stats_main) { memset(stats, 0, sizeof(mi_stats_t)); } + memset(&_mi_stats_main, 0, sizeof(mi_stats_t)); + if (mi_process_start == 0) { mi_process_start = _mi_clock_start(); }; +} + +void mi_stats_merge(void) mi_attr_noexcept { + mi_stats_merge_from( mi_stats_get_default() ); +} + +void _mi_stats_done(mi_stats_t* stats) { // called from `mi_thread_done` + mi_stats_merge_from(stats); +} + +void mi_stats_print_out(mi_output_fun* out, void* arg) mi_attr_noexcept { + mi_stats_merge_from(mi_stats_get_default()); + _mi_stats_print(&_mi_stats_main, out, arg); +} + +void mi_stats_print(void* out) mi_attr_noexcept { + // for compatibility there is an `out` parameter (which can be `stdout` or `stderr`) + mi_stats_print_out((mi_output_fun*)out, NULL); +} + +void mi_thread_stats_print_out(mi_output_fun* out, void* arg) mi_attr_noexcept { + _mi_stats_print(mi_stats_get_default(), out, arg); +} + + +// ---------------------------------------------------------------- +// Basic timer for convenience; use milli-seconds to avoid doubles +// ---------------------------------------------------------------- +#ifdef _WIN32 +#include +static mi_msecs_t mi_to_msecs(LARGE_INTEGER t) { + static LARGE_INTEGER mfreq; // = 0 + if (mfreq.QuadPart == 0LL) { + LARGE_INTEGER f; + QueryPerformanceFrequency(&f); + mfreq.QuadPart = f.QuadPart/1000LL; + if (mfreq.QuadPart == 0) mfreq.QuadPart = 1; + } + return (mi_msecs_t)(t.QuadPart / mfreq.QuadPart); +} + +mi_msecs_t _mi_clock_now(void) { + LARGE_INTEGER t; + QueryPerformanceCounter(&t); + return mi_to_msecs(t); +} +#else +#include +#if defined(CLOCK_REALTIME) || defined(CLOCK_MONOTONIC) +mi_msecs_t _mi_clock_now(void) { + struct timespec t; + #ifdef CLOCK_MONOTONIC + clock_gettime(CLOCK_MONOTONIC, &t); + #else + clock_gettime(CLOCK_REALTIME, &t); + #endif + return ((mi_msecs_t)t.tv_sec * 1000) + ((mi_msecs_t)t.tv_nsec / 1000000); +} +#else +// low resolution timer +mi_msecs_t _mi_clock_now(void) { + return ((mi_msecs_t)clock() / ((mi_msecs_t)CLOCKS_PER_SEC / 1000)); +} +#endif +#endif + + +static mi_msecs_t mi_clock_diff; + +mi_msecs_t _mi_clock_start(void) { + if (mi_clock_diff == 0.0) { + mi_msecs_t t0 = _mi_clock_now(); + mi_clock_diff = _mi_clock_now() - t0; + } + return _mi_clock_now(); +} + +mi_msecs_t _mi_clock_end(mi_msecs_t start) { + mi_msecs_t end = _mi_clock_now(); + return (end - start - mi_clock_diff); +} + + +// -------------------------------------------------------- +// Basic process statistics +// -------------------------------------------------------- + +#if defined(_WIN32) +#include +#include +#pragma comment(lib,"psapi.lib") +#include "compat/win32/lazyload.h" + +static mi_msecs_t filetime_msecs(const FILETIME* ftime) { + ULARGE_INTEGER i; + i.LowPart = ftime->dwLowDateTime; + i.HighPart = ftime->dwHighDateTime; + mi_msecs_t msecs = (i.QuadPart / 10000); // FILETIME is in 100 nano seconds + return msecs; +} + +static void mi_stat_process_info(mi_msecs_t* elapsed, mi_msecs_t* utime, mi_msecs_t* stime, size_t* current_rss, size_t* peak_rss, size_t* current_commit, size_t* peak_commit, size_t* page_faults) +{ + *elapsed = _mi_clock_end(mi_process_start); + FILETIME ct; + FILETIME ut; + FILETIME st; + FILETIME et; + GetProcessTimes(GetCurrentProcess(), &ct, &et, &st, &ut); + *utime = filetime_msecs(&ut); + *stime = filetime_msecs(&st); + PROCESS_MEMORY_COUNTERS info; + DECLARE_PROC_ADDR(psapi, BOOL, WINAPI, GetProcessMemoryInfo, HANDLE, PPROCESS_MEMORY_COUNTERS, DWORD); + if (INIT_PROC_ADDR(GetProcessMemoryInfo)) { + GetProcessMemoryInfo(GetCurrentProcess(), &info, sizeof(info)); + *current_rss = (size_t)info.WorkingSetSize; + *peak_rss = (size_t)info.PeakWorkingSetSize; + *current_commit = (size_t)info.PagefileUsage; + *peak_commit = (size_t)info.PeakPagefileUsage; + *page_faults = (size_t)info.PageFaultCount; + } else { + *current_rss = *peak_rss = *current_commit = *peak_commit = *page_faults = 0; + } +} + +#elif !defined(__wasi__) && (defined(__unix__) || defined(__unix) || defined(unix) || defined(__APPLE__) || defined(__HAIKU__)) +#include +#include +#include + +#if defined(__APPLE__) +#include +#endif + +#if defined(__HAIKU__) +#include +#endif + +static mi_msecs_t timeval_secs(const struct timeval* tv) { + return ((mi_msecs_t)tv->tv_sec * 1000L) + ((mi_msecs_t)tv->tv_usec / 1000L); +} + +static void mi_stat_process_info(mi_msecs_t* elapsed, mi_msecs_t* utime, mi_msecs_t* stime, size_t* current_rss, size_t* peak_rss, size_t* current_commit, size_t* peak_commit, size_t* page_faults) +{ + *elapsed = _mi_clock_end(mi_process_start); + struct rusage rusage; + getrusage(RUSAGE_SELF, &rusage); + *utime = timeval_secs(&rusage.ru_utime); + *stime = timeval_secs(&rusage.ru_stime); +#if !defined(__HAIKU__) + *page_faults = rusage.ru_majflt; +#endif + // estimate commit using our stats + *peak_commit = (size_t)(mi_atomic_loadi64_relaxed((_Atomic(int64_t)*)&_mi_stats_main.committed.peak)); + *current_commit = (size_t)(mi_atomic_loadi64_relaxed((_Atomic(int64_t)*)&_mi_stats_main.committed.current)); + *current_rss = *current_commit; // estimate +#if defined(__HAIKU__) + // Haiku does not have (yet?) a way to + // get these stats per process + thread_info tid; + area_info mem; + ssize_t c; + get_thread_info(find_thread(0), &tid); + while (get_next_area_info(tid.team, &c, &mem) == B_OK) { + *peak_rss += mem.ram_size; + } + *page_faults = 0; +#elif defined(__APPLE__) + *peak_rss = rusage.ru_maxrss; // BSD reports in bytes + struct mach_task_basic_info info; + mach_msg_type_number_t infoCount = MACH_TASK_BASIC_INFO_COUNT; + if (task_info(mach_task_self(), MACH_TASK_BASIC_INFO, (task_info_t)&info, &infoCount) == KERN_SUCCESS) { + *current_rss = (size_t)info.resident_size; + } +#else + *peak_rss = rusage.ru_maxrss * 1024; // Linux reports in KiB +#endif +} + +#else +#ifndef __wasi__ +// WebAssembly instances are not processes +#pragma message("define a way to get process info") +#endif + +static void mi_stat_process_info(mi_msecs_t* elapsed, mi_msecs_t* utime, mi_msecs_t* stime, size_t* current_rss, size_t* peak_rss, size_t* current_commit, size_t* peak_commit, size_t* page_faults) +{ + *elapsed = _mi_clock_end(mi_process_start); + *peak_commit = (size_t)(mi_atomic_loadi64_relaxed((_Atomic(int64_t)*)&_mi_stats_main.committed.peak)); + *current_commit = (size_t)(mi_atomic_loadi64_relaxed((_Atomic(int64_t)*)&_mi_stats_main.committed.current)); + *peak_rss = *peak_commit; + *current_rss = *current_commit; + *page_faults = 0; + *utime = 0; + *stime = 0; +} +#endif + + +mi_decl_export void mi_process_info(size_t* elapsed_msecs, size_t* user_msecs, size_t* system_msecs, size_t* current_rss, size_t* peak_rss, size_t* current_commit, size_t* peak_commit, size_t* page_faults) mi_attr_noexcept +{ + mi_msecs_t elapsed = 0; + mi_msecs_t utime = 0; + mi_msecs_t stime = 0; + size_t current_rss0 = 0; + size_t peak_rss0 = 0; + size_t current_commit0 = 0; + size_t peak_commit0 = 0; + size_t page_faults0 = 0; + mi_stat_process_info(&elapsed,&utime, &stime, ¤t_rss0, &peak_rss0, ¤t_commit0, &peak_commit0, &page_faults0); + if (elapsed_msecs!=NULL) *elapsed_msecs = (elapsed < 0 ? 0 : (elapsed < (mi_msecs_t)PTRDIFF_MAX ? (size_t)elapsed : PTRDIFF_MAX)); + if (user_msecs!=NULL) *user_msecs = (utime < 0 ? 0 : (utime < (mi_msecs_t)PTRDIFF_MAX ? (size_t)utime : PTRDIFF_MAX)); + if (system_msecs!=NULL) *system_msecs = (stime < 0 ? 0 : (stime < (mi_msecs_t)PTRDIFF_MAX ? (size_t)stime : PTRDIFF_MAX)); + if (current_rss!=NULL) *current_rss = current_rss0; + if (peak_rss!=NULL) *peak_rss = peak_rss0; + if (current_commit!=NULL) *current_commit = current_commit0; + if (peak_commit!=NULL) *peak_commit = peak_commit0; + if (page_faults!=NULL) *page_faults = page_faults0; +} diff --git a/compat/mingw.c b/compat/mingw.c index 901375d58415a3..969534681a5c6e 100644 --- a/compat/mingw.c +++ b/compat/mingw.c @@ -4,19 +4,22 @@ #include #include #include +#include #include "../strbuf.h" #include "../run-command.h" #include "../cache.h" +#include "win32/exit-process.h" #include "win32/lazyload.h" #include "../config.h" #include "dir.h" #define SECURITY_WIN32 #include +#include "win32/fscache.h" +#include "../attr.h" +#include "../string-list.h" #define HCAST(type, handle) ((type)(intptr_t)handle) -static const int delay[] = { 0, 1, 10, 20, 40 }; - void open_in_gdb(void) { static struct child_process cp = CHILD_PROCESS_INIT; @@ -92,6 +95,7 @@ int err_win_to_posix(DWORD winerr) case ERROR_INVALID_PARAMETER: error = EINVAL; break; case ERROR_INVALID_PASSWORD: error = EPERM; break; case ERROR_INVALID_PRIMARY_GROUP: error = EINVAL; break; + case ERROR_INVALID_REPARSE_DATA: error = EINVAL; break; case ERROR_INVALID_SIGNAL_NUMBER: error = EINVAL; break; case ERROR_INVALID_TARGET_HANDLE: error = EIO; break; case ERROR_INVALID_WORKSTATION: error = EACCES; break; @@ -106,6 +110,7 @@ int err_win_to_posix(DWORD winerr) case ERROR_NEGATIVE_SEEK: error = ESPIPE; break; case ERROR_NOACCESS: error = EFAULT; break; case ERROR_NONE_MAPPED: error = EINVAL; break; + case ERROR_NOT_A_REPARSE_POINT: error = EINVAL; break; case ERROR_NOT_ENOUGH_MEMORY: error = ENOMEM; break; case ERROR_NOT_READY: error = EAGAIN; break; case ERROR_NOT_SAME_DEVICE: error = EXDEV; break; @@ -126,6 +131,9 @@ int err_win_to_posix(DWORD winerr) case ERROR_PIPE_NOT_CONNECTED: error = EPIPE; break; case ERROR_PRIVILEGE_NOT_HELD: error = EACCES; break; case ERROR_READ_FAULT: error = EIO; break; + case ERROR_REPARSE_ATTRIBUTE_CONFLICT: error = EINVAL; break; + case ERROR_REPARSE_TAG_INVALID: error = EINVAL; break; + case ERROR_REPARSE_TAG_MISMATCH: error = EINVAL; break; case ERROR_SEEK: error = EIO; break; case ERROR_SEEK_ON_DEVICE: error = ESPIPE; break; case ERROR_SHARING_BUFFER_EXCEEDED: error = ENFILE; break; @@ -193,15 +201,12 @@ static int read_yes_no_answer(void) return -1; } -static int ask_yes_no_if_possible(const char *format, ...) +static int ask_yes_no_if_possible(const char *format, va_list args) { char question[4096]; const char *retry_hook[] = { NULL, NULL, NULL }; - va_list args; - va_start(args, format); vsnprintf(question, sizeof(question), format, args); - va_end(args); if ((retry_hook[0] = mingw_getenv("GIT_ASK_YESNO"))) { retry_hook[1] = question; @@ -223,6 +228,31 @@ static int ask_yes_no_if_possible(const char *format, ...) } } +static int retry_ask_yes_no(int *tries, const char *format, ...) +{ + static const int delay[] = { 0, 1, 10, 20, 40 }; + va_list args; + int result, saved_errno = errno; + + if ((*tries) < ARRAY_SIZE(delay)) { + /* + * We assume that some other process had the file open at the wrong + * moment and retry. In order to give the other process a higher + * chance to complete its operation, we give up our time slice now. + * If we have to retry again, we do sleep a bit. + */ + Sleep(delay[*tries]); + (*tries)++; + return 1; + } + + va_start(args, format); + result = ask_yes_no_if_possible(format, args); + va_end(args); + errno = saved_errno; + return result; +} + /* Windows only */ enum hide_dotfiles_type { HIDE_DOTFILES_FALSE = 0, @@ -233,6 +263,28 @@ enum hide_dotfiles_type { static int core_restrict_inherited_handles = -1; static enum hide_dotfiles_type hide_dotfiles = HIDE_DOTFILES_DOTGITONLY; static char *unset_environment_variables; +int core_fscache; + +int are_long_paths_enabled(void) +{ + /* default to `false` during initialization */ + static const int fallback = 0; + + static int enabled = -1; + + if (enabled < 0) { + /* avoid infinite recursion */ + if (!the_repository) + return fallback; + + if (the_repository->config && + the_repository->config->hash_initialized && + git_config_get_bool("core.longpaths", &enabled) < 0) + enabled = 0; + } + + return enabled < 0 ? fallback : enabled; +} int mingw_core_config(const char *var, const char *value, void *cb) { @@ -244,6 +296,11 @@ int mingw_core_config(const char *var, const char *value, void *cb) return 0; } + if (!strcmp(var, "core.fscache")) { + core_fscache = git_config_bool(var, value); + return 0; + } + if (!strcmp(var, "core.unsetenvvars")) { free(unset_environment_variables); unset_environment_variables = xstrdup(value); @@ -262,6 +319,182 @@ int mingw_core_config(const char *var, const char *value, void *cb) return 0; } +static DWORD symlink_file_flags = 0, symlink_directory_flags = 1; + +enum phantom_symlink_result { + PHANTOM_SYMLINK_RETRY, + PHANTOM_SYMLINK_DONE, + PHANTOM_SYMLINK_DIRECTORY +}; + +static inline int is_wdir_sep(wchar_t wchar) +{ + return wchar == L'/' || wchar == L'\\'; +} + +static const wchar_t *make_relative_to(const wchar_t *path, + const wchar_t *relative_to, wchar_t *out, + size_t size) +{ + size_t i = wcslen(relative_to), len; + + /* Is `path` already absolute? */ + if (is_wdir_sep(path[0]) || + (iswalpha(path[0]) && path[1] == L':' && is_wdir_sep(path[2]))) + return path; + + while (i > 0 && !is_wdir_sep(relative_to[i - 1])) + i--; + + /* Is `relative_to` in the current directory? */ + if (!i) + return path; + + len = wcslen(path); + if (i + len + 1 > size) { + error("Could not make '%ls' relative to '%ls' (too large)", + path, relative_to); + return NULL; + } + + memcpy(out, relative_to, i * sizeof(wchar_t)); + wcscpy(out + i, path); + return out; +} + +/* + * Changes a file symlink to a directory symlink if the target exists and is a + * directory. + */ +static enum phantom_symlink_result +process_phantom_symlink(const wchar_t *wtarget, const wchar_t *wlink) +{ + HANDLE hnd; + BY_HANDLE_FILE_INFORMATION fdata; + wchar_t relative[MAX_LONG_PATH]; + const wchar_t *rel; + + /* check that wlink is still a file symlink */ + if ((GetFileAttributesW(wlink) + & (FILE_ATTRIBUTE_REPARSE_POINT | FILE_ATTRIBUTE_DIRECTORY)) + != FILE_ATTRIBUTE_REPARSE_POINT) + return PHANTOM_SYMLINK_DONE; + + /* make it relative, if necessary */ + rel = make_relative_to(wtarget, wlink, relative, ARRAY_SIZE(relative)); + if (!rel) + return PHANTOM_SYMLINK_DONE; + + /* let Windows resolve the link by opening it */ + hnd = CreateFileW(rel, 0, + FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE, NULL, + OPEN_EXISTING, FILE_FLAG_BACKUP_SEMANTICS, NULL); + if (hnd == INVALID_HANDLE_VALUE) { + errno = err_win_to_posix(GetLastError()); + return PHANTOM_SYMLINK_RETRY; + } + + if (!GetFileInformationByHandle(hnd, &fdata)) { + errno = err_win_to_posix(GetLastError()); + CloseHandle(hnd); + return PHANTOM_SYMLINK_RETRY; + } + CloseHandle(hnd); + + /* if target exists and is a file, we're done */ + if (!(fdata.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY)) + return PHANTOM_SYMLINK_DONE; + + /* otherwise recreate the symlink with directory flag */ + if (DeleteFileW(wlink) && + CreateSymbolicLinkW(wlink, wtarget, symlink_directory_flags)) + return PHANTOM_SYMLINK_DIRECTORY; + + errno = err_win_to_posix(GetLastError()); + return PHANTOM_SYMLINK_RETRY; +} + +/* keep track of newly created symlinks to non-existing targets */ +struct phantom_symlink_info { + struct phantom_symlink_info *next; + wchar_t *wlink; + wchar_t *wtarget; +}; + +static struct phantom_symlink_info *phantom_symlinks = NULL; +static CRITICAL_SECTION phantom_symlinks_cs; + +static void process_phantom_symlinks(void) +{ + struct phantom_symlink_info *current, **psi; + EnterCriticalSection(&phantom_symlinks_cs); + /* process phantom symlinks list */ + psi = &phantom_symlinks; + while ((current = *psi)) { + enum phantom_symlink_result result = process_phantom_symlink( + current->wtarget, current->wlink); + if (result == PHANTOM_SYMLINK_RETRY) { + psi = ¤t->next; + } else { + /* symlink was processed, remove from list */ + *psi = current->next; + free(current); + /* if symlink was a directory, start over */ + if (result == PHANTOM_SYMLINK_DIRECTORY) + psi = &phantom_symlinks; + } + } + LeaveCriticalSection(&phantom_symlinks_cs); +} + +static int create_phantom_symlink(wchar_t *wtarget, wchar_t *wlink) +{ + int len; + + /* create file symlink */ + if (!CreateSymbolicLinkW(wlink, wtarget, symlink_file_flags)) { + errno = err_win_to_posix(GetLastError()); + return -1; + } + + /* convert to directory symlink if target exists */ + switch (process_phantom_symlink(wtarget, wlink)) { + case PHANTOM_SYMLINK_RETRY: { + /* if target doesn't exist, add to phantom symlinks list */ + wchar_t wfullpath[MAX_LONG_PATH]; + struct phantom_symlink_info *psi; + + /* convert to absolute path to be independent of cwd */ + len = GetFullPathNameW(wlink, MAX_LONG_PATH, wfullpath, NULL); + if (!len || len >= MAX_LONG_PATH) { + errno = err_win_to_posix(GetLastError()); + return -1; + } + + /* over-allocate and fill phantom_symlink_info structure */ + psi = xmalloc(sizeof(struct phantom_symlink_info) + + sizeof(wchar_t) * (len + wcslen(wtarget) + 2)); + psi->wlink = (wchar_t *)(psi + 1); + wcscpy(psi->wlink, wfullpath); + psi->wtarget = psi->wlink + len + 1; + wcscpy(psi->wtarget, wtarget); + + EnterCriticalSection(&phantom_symlinks_cs); + psi->next = phantom_symlinks; + phantom_symlinks = psi; + LeaveCriticalSection(&phantom_symlinks_cs); + break; + } + case PHANTOM_SYMLINK_DIRECTORY: + /* if we created a dir symlink, process other phantom symlinks */ + process_phantom_symlinks(); + break; + default: + break; + } + return 0; +} + /* Normalizes NT paths as returned by some low-level APIs. */ static wchar_t *normalize_ntpath(wchar_t *wbuf) { @@ -289,41 +522,38 @@ static wchar_t *normalize_ntpath(wchar_t *wbuf) int mingw_unlink(const char *pathname) { - int ret, tries = 0; - wchar_t wpathname[MAX_PATH]; - if (xutftowcs_path(wpathname, pathname) < 0) + int tries = 0; + wchar_t wpathname[MAX_LONG_PATH]; + if (xutftowcs_long_path(wpathname, pathname) < 0) return -1; if (DeleteFileW(wpathname)) return 0; - /* read-only files cannot be removed */ - _wchmod(wpathname, 0666); - while ((ret = _wunlink(wpathname)) == -1 && tries < ARRAY_SIZE(delay)) { + do { + /* read-only files cannot be removed */ + _wchmod(wpathname, 0666); + if (!_wunlink(wpathname)) + return 0; if (!is_file_in_use_error(GetLastError())) break; /* - * We assume that some other process had the source or - * destination file open at the wrong moment and retry. - * In order to give the other process a higher chance to - * complete its operation, we give up our time slice now. - * If we have to retry again, we do sleep a bit. + * _wunlink() / DeleteFileW() for directory symlinks fails with + * ERROR_ACCESS_DENIED (EACCES), so try _wrmdir() as well. This is the + * same error we get if a file is in use (already checked above). */ - Sleep(delay[tries]); - tries++; - } - while (ret == -1 && is_file_in_use_error(GetLastError()) && - ask_yes_no_if_possible("Unlink of file '%s' failed. " - "Should I try again?", pathname)) - ret = _wunlink(wpathname); - return ret; + if (!_wrmdir(wpathname)) + return 0; + } while (retry_ask_yes_no(&tries, "Unlink of file '%s' failed. " + "Should I try again?", pathname)); + return -1; } static int is_dir_empty(const wchar_t *wpath) { WIN32_FIND_DATAW findbuf; HANDLE handle; - wchar_t wbuf[MAX_PATH + 2]; + wchar_t wbuf[MAX_LONG_PATH + 2]; wcscpy(wbuf, wpath); wcscat(wbuf, L"\\*"); handle = FindFirstFileW(wbuf, &findbuf); @@ -343,8 +573,8 @@ static int is_dir_empty(const wchar_t *wpath) int mingw_rmdir(const char *pathname) { - int ret, tries = 0; - wchar_t wpathname[MAX_PATH]; + int tries = 0; + wchar_t wpathname[MAX_LONG_PATH]; struct stat st; /* @@ -366,10 +596,14 @@ int mingw_rmdir(const char *pathname) return -1; } - if (xutftowcs_path(wpathname, pathname) < 0) + if (xutftowcs_long_path(wpathname, pathname) < 0) return -1; - while ((ret = _wrmdir(wpathname)) == -1 && tries < ARRAY_SIZE(delay)) { + do { + if (!_wrmdir(wpathname)) { + invalidate_lstat_cache(); + return 0; + } if (!is_file_in_use_error(GetLastError())) errno = err_win_to_posix(GetLastError()); if (errno != EACCES) @@ -378,23 +612,9 @@ int mingw_rmdir(const char *pathname) errno = ENOTEMPTY; break; } - /* - * We assume that some other process had the source or - * destination file open at the wrong moment and retry. - * In order to give the other process a higher chance to - * complete its operation, we give up our time slice now. - * If we have to retry again, we do sleep a bit. - */ - Sleep(delay[tries]); - tries++; - } - while (ret == -1 && errno == EACCES && is_file_in_use_error(GetLastError()) && - ask_yes_no_if_possible("Deletion of directory '%s' failed. " - "Should I try again?", pathname)) - ret = _wrmdir(wpathname); - if (!ret) - invalidate_lstat_cache(); - return ret; + } while (retry_ask_yes_no(&tries, "Deletion of directory '%s' failed. " + "Should I try again?", pathname)); + return -1; } static inline int needs_hiding(const char *path) @@ -445,16 +665,21 @@ static int set_hidden_flag(const wchar_t *path, int set) int mingw_mkdir(const char *path, int mode) { int ret; - wchar_t wpath[MAX_PATH]; + wchar_t wpath[MAX_LONG_PATH]; if (!is_valid_win32_path(path, 0)) { errno = EINVAL; return -1; } - if (xutftowcs_path(wpath, path) < 0) + /* CreateDirectoryW path limit is 248 (MAX_PATH - 8.3 file name) */ + if (xutftowcs_path_ex(wpath, path, MAX_LONG_PATH, -1, 248, + are_long_paths_enabled()) < 0) return -1; + ret = _wmkdir(wpath); + if (!ret) + process_phantom_symlinks(); if (!ret && needs_hiding(path)) return set_hidden_flag(wpath, 1); return ret; @@ -535,11 +760,12 @@ static int is_local_named_pipe_path(const char *filename) int mingw_open (const char *filename, int oflags, ...) { + static int append_atomically = -1; typedef int (*open_fn_t)(wchar_t const *wfilename, int oflags, ...); va_list args; unsigned mode; int fd, create = (oflags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL); - wchar_t wfilename[MAX_PATH]; + wchar_t wfilename[MAX_LONG_PATH]; open_fn_t open_fn; va_start(args, oflags); @@ -551,14 +777,23 @@ int mingw_open (const char *filename, int oflags, ...) return -1; } - if ((oflags & O_APPEND) && !is_local_named_pipe_path(filename)) + /* + * Only set append_atomically to default value(1) when repo is initialized + * and fail to get config value + */ + if (append_atomically < 0 && the_repository && the_repository->commondir && + git_config_get_bool("windows.appendatomically", &append_atomically)) + append_atomically = 1; + + if (append_atomically && (oflags & O_APPEND) && + !is_local_named_pipe_path(filename)) open_fn = mingw_open_append; else open_fn = _wopen; if (filename && !strcmp(filename, "/dev/null")) wcscpy(wfilename, L"nul"); - else if (xutftowcs_path(wfilename, filename) < 0) + else if (xutftowcs_long_path(wfilename, filename) < 0) return -1; fd = open_fn(wfilename, oflags, mode); @@ -616,14 +851,14 @@ FILE *mingw_fopen (const char *filename, const char *otype) { int hide = needs_hiding(filename); FILE *file; - wchar_t wfilename[MAX_PATH], wotype[4]; + wchar_t wfilename[MAX_LONG_PATH], wotype[4]; if (filename && !strcmp(filename, "/dev/null")) wcscpy(wfilename, L"nul"); else if (!is_valid_win32_path(filename, 1)) { int create = otype && strchr(otype, 'w'); errno = create ? EINVAL : ENOENT; return NULL; - } else if (xutftowcs_path(wfilename, filename) < 0) + } else if (xutftowcs_long_path(wfilename, filename) < 0) return NULL; if (xutftowcs(wotype, otype, ARRAY_SIZE(wotype)) < 0) @@ -645,14 +880,14 @@ FILE *mingw_freopen (const char *filename, const char *otype, FILE *stream) { int hide = needs_hiding(filename); FILE *file; - wchar_t wfilename[MAX_PATH], wotype[4]; + wchar_t wfilename[MAX_LONG_PATH], wotype[4]; if (filename && !strcmp(filename, "/dev/null")) wcscpy(wfilename, L"nul"); else if (!is_valid_win32_path(filename, 1)) { int create = otype && strchr(otype, 'w'); errno = create ? EINVAL : ENOENT; return NULL; - } else if (xutftowcs_path(wfilename, filename) < 0) + } else if (xutftowcs_long_path(wfilename, filename) < 0) return NULL; if (xutftowcs(wotype, otype, ARRAY_SIZE(wotype)) < 0) @@ -700,8 +935,26 @@ ssize_t mingw_write(int fd, const void *buf, size_t len) HANDLE h = (HANDLE) _get_osfhandle(fd); if (GetFileType(h) == FILE_TYPE_PIPE) errno = EPIPE; - else + else { + wchar_t path[MAX_LONG_PATH]; + DWORD ret = GetFinalPathNameByHandleW(h, path, + ARRAY_SIZE(path), 0); + UINT drive_type = ret > 0 && ret < ARRAY_SIZE(path) ? + GetDriveTypeW(path) : DRIVE_UNKNOWN; + + /* + * The default atomic append causes such an error on + * network file systems, in such a case, it should be + * turned off via config. + * + * `drive_type` of UNC path: DRIVE_NO_ROOT_DIR + */ + if (DRIVE_NO_ROOT_DIR == drive_type || DRIVE_REMOTE == drive_type) + warning("invalid write operation detected; you may try:\n" + "\n\tgit config windows.appendAtomically false"); + errno = EINVAL; + } } return result; @@ -709,49 +962,54 @@ ssize_t mingw_write(int fd, const void *buf, size_t len) int mingw_access(const char *filename, int mode) { - wchar_t wfilename[MAX_PATH]; + wchar_t wfilename[MAX_LONG_PATH]; if (!strcmp("nul", filename) || !strcmp("/dev/null", filename)) return 0; - if (xutftowcs_path(wfilename, filename) < 0) + if (xutftowcs_long_path(wfilename, filename) < 0) return -1; /* X_OK is not supported by the MSVCRT version */ return _waccess(wfilename, mode & ~X_OK); } +/* cached length of current directory for handle_long_path */ +static int current_directory_len = 0; + int mingw_chdir(const char *dirname) { - wchar_t wdirname[MAX_PATH]; - if (xutftowcs_path(wdirname, dirname) < 0) + int result; + wchar_t wdirname[MAX_LONG_PATH]; + if (xutftowcs_long_path(wdirname, dirname) < 0) return -1; - return _wchdir(wdirname); + + if (has_symlinks) { + HANDLE hnd = CreateFileW(wdirname, 0, + FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE, NULL, + OPEN_EXISTING, FILE_FLAG_BACKUP_SEMANTICS, NULL); + if (hnd == INVALID_HANDLE_VALUE) { + errno = err_win_to_posix(GetLastError()); + return -1; + } + if (!GetFinalPathNameByHandleW(hnd, wdirname, ARRAY_SIZE(wdirname), 0)) { + errno = err_win_to_posix(GetLastError()); + CloseHandle(hnd); + return -1; + } + CloseHandle(hnd); + } + + result = _wchdir(normalize_ntpath(wdirname)); + current_directory_len = GetCurrentDirectoryW(0, NULL); + return result; } int mingw_chmod(const char *filename, int mode) { - wchar_t wfilename[MAX_PATH]; - if (xutftowcs_path(wfilename, filename) < 0) + wchar_t wfilename[MAX_LONG_PATH]; + if (xutftowcs_long_path(wfilename, filename) < 0) return -1; return _wchmod(wfilename, mode); } -/* - * The unit of FILETIME is 100-nanoseconds since January 1, 1601, UTC. - * Returns the 100-nanoseconds ("hekto nanoseconds") since the epoch. - */ -static inline long long filetime_to_hnsec(const FILETIME *ft) -{ - long long winTime = ((long long)ft->dwHighDateTime << 32) + ft->dwLowDateTime; - /* Windows to Unix Epoch conversion */ - return winTime - 116444736000000000LL; -} - -static inline void filetime_to_timespec(const FILETIME *ft, struct timespec *ts) -{ - long long hnsec = filetime_to_hnsec(ft); - ts->tv_sec = (time_t)(hnsec / 10000000); - ts->tv_nsec = (hnsec % 10000000) * 100; -} - /** * Verifies that safe_create_leading_directories() would succeed. */ @@ -785,53 +1043,51 @@ static int has_valid_directory_prefix(wchar_t *wfilename) return 1; } -/* We keep the do_lstat code in a separate function to avoid recursion. - * When a path ends with a slash, the stat will fail with ENOENT. In - * this case, we strip the trailing slashes and stat again. - * - * If follow is true then act like stat() and report on the link - * target. Otherwise report on the link itself. - */ -static int do_lstat(int follow, const char *file_name, struct stat *buf) +static int readlink_1(const WCHAR *wpath, BOOL fail_on_unknown_tag, + char *tmpbuf, int *plen, DWORD *ptag); + +int mingw_lstat(const char *file_name, struct stat *buf) { WIN32_FILE_ATTRIBUTE_DATA fdata; - wchar_t wfilename[MAX_PATH]; - if (xutftowcs_path(wfilename, file_name) < 0) + DWORD reparse_tag = 0; + int link_len = 0; + wchar_t wfilename[MAX_LONG_PATH]; + int wlen = xutftowcs_long_path(wfilename, file_name); + if (wlen < 0) + return -1; + + /* strip trailing '/', or GetFileAttributes will fail */ + while (wlen && is_dir_sep(wfilename[wlen - 1])) + wfilename[--wlen] = 0; + if (!wlen) { + errno = ENOENT; return -1; + } if (GetFileAttributesExW(wfilename, GetFileExInfoStandard, &fdata)) { + /* for reparse points, get the link tag and length */ + if (fdata.dwFileAttributes & FILE_ATTRIBUTE_REPARSE_POINT) { + char tmpbuf[MAX_LONG_PATH]; + + if (readlink_1(wfilename, FALSE, tmpbuf, &link_len, + &reparse_tag) < 0) + return -1; + } buf->st_ino = 0; buf->st_gid = 0; buf->st_uid = 0; buf->st_nlink = 1; - buf->st_mode = file_attr_to_st_mode(fdata.dwFileAttributes); - buf->st_size = fdata.nFileSizeLow | - (((off_t)fdata.nFileSizeHigh)<<32); + buf->st_mode = file_attr_to_st_mode(fdata.dwFileAttributes, + reparse_tag, file_name); + buf->st_size = S_ISLNK(buf->st_mode) ? link_len : + fdata.nFileSizeLow | (((off_t) fdata.nFileSizeHigh) << 32); buf->st_dev = buf->st_rdev = 0; /* not used by Git */ filetime_to_timespec(&(fdata.ftLastAccessTime), &(buf->st_atim)); filetime_to_timespec(&(fdata.ftLastWriteTime), &(buf->st_mtim)); filetime_to_timespec(&(fdata.ftCreationTime), &(buf->st_ctim)); - if (fdata.dwFileAttributes & FILE_ATTRIBUTE_REPARSE_POINT) { - WIN32_FIND_DATAW findbuf; - HANDLE handle = FindFirstFileW(wfilename, &findbuf); - if (handle != INVALID_HANDLE_VALUE) { - if ((findbuf.dwFileAttributes & FILE_ATTRIBUTE_REPARSE_POINT) && - (findbuf.dwReserved0 == IO_REPARSE_TAG_SYMLINK)) { - if (follow) { - char buffer[MAXIMUM_REPARSE_DATA_BUFFER_SIZE]; - buf->st_size = readlink(file_name, buffer, MAXIMUM_REPARSE_DATA_BUFFER_SIZE); - } else { - buf->st_mode = S_IFLNK; - } - buf->st_mode |= S_IREAD; - if (!(findbuf.dwFileAttributes & FILE_ATTRIBUTE_READONLY)) - buf->st_mode |= S_IWRITE; - } - FindClose(handle); - } - } return 0; } + switch (GetLastError()) { case ERROR_ACCESS_DENIED: case ERROR_SHARING_VIOLATION: @@ -858,38 +1114,7 @@ static int do_lstat(int follow, const char *file_name, struct stat *buf) return -1; } -/* We provide our own lstat/fstat functions, since the provided - * lstat/fstat functions are so slow. These stat functions are - * tailored for Git's usage (read: fast), and are not meant to be - * complete. Note that Git stat()s are redirected to mingw_lstat() - * too, since Windows doesn't really handle symlinks that well. - */ -static int do_stat_internal(int follow, const char *file_name, struct stat *buf) -{ - int namelen; - char alt_name[PATH_MAX]; - - if (!do_lstat(follow, file_name, buf)) - return 0; - - /* if file_name ended in a '/', Windows returned ENOENT; - * try again without trailing slashes - */ - if (errno != ENOENT) - return -1; - - namelen = strlen(file_name); - if (namelen && file_name[namelen-1] != '/') - return -1; - while (namelen && file_name[namelen-1] == '/') - --namelen; - if (!namelen || namelen >= PATH_MAX) - return -1; - - memcpy(alt_name, file_name, namelen); - alt_name[namelen] = 0; - return do_lstat(follow, alt_name, buf); -} +int (*lstat)(const char *file_name, struct stat *buf) = mingw_lstat; static int get_file_info_by_handle(HANDLE hnd, struct stat *buf) { @@ -904,7 +1129,7 @@ static int get_file_info_by_handle(HANDLE hnd, struct stat *buf) buf->st_gid = 0; buf->st_uid = 0; buf->st_nlink = 1; - buf->st_mode = file_attr_to_st_mode(fdata.dwFileAttributes); + buf->st_mode = file_attr_to_st_mode(fdata.dwFileAttributes, 0, NULL); buf->st_size = fdata.nFileSizeLow | (((off_t)fdata.nFileSizeHigh)<<32); buf->st_dev = buf->st_rdev = 0; /* not used by Git */ @@ -914,13 +1139,37 @@ static int get_file_info_by_handle(HANDLE hnd, struct stat *buf) return 0; } -int mingw_lstat(const char *file_name, struct stat *buf) -{ - return do_stat_internal(0, file_name, buf); -} int mingw_stat(const char *file_name, struct stat *buf) { - return do_stat_internal(1, file_name, buf); + wchar_t wfile_name[MAX_LONG_PATH]; + HANDLE hnd; + int result; + + /* open the file and let Windows resolve the links */ + if (xutftowcs_long_path(wfile_name, file_name) < 0) + return -1; + hnd = CreateFileW(wfile_name, 0, + FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE, NULL, + OPEN_EXISTING, FILE_FLAG_BACKUP_SEMANTICS, NULL); + if (hnd == INVALID_HANDLE_VALUE) { + DWORD err = GetLastError(); + + if (err == ERROR_ACCESS_DENIED && + !mingw_lstat(file_name, buf) && + !S_ISLNK(buf->st_mode)) + /* + * POSIX semantics state to still try to fill + * information, even if permission is denied to create + * a file handle. + */ + return 0; + + errno = err_win_to_posix(err); + return -1; + } + result = get_file_info_by_handle(hnd, buf); + CloseHandle(hnd); + return result; } int mingw_fstat(int fd, struct stat *buf) @@ -965,10 +1214,10 @@ int mingw_utime (const char *file_name, const struct utimbuf *times) FILETIME mft, aft; int rc; DWORD attrs; - wchar_t wfilename[MAX_PATH]; + wchar_t wfilename[MAX_LONG_PATH]; HANDLE osfilehandle; - if (xutftowcs_path(wfilename, file_name) < 0) + if (xutftowcs_long_path(wfilename, file_name) < 0) return -1; /* must have write permission */ @@ -1049,11 +1298,23 @@ unsigned int sleep (unsigned int seconds) char *mingw_mktemp(char *template) { wchar_t wtemplate[MAX_PATH]; - if (xutftowcs_path(wtemplate, template) < 0) + int offset = 0; + + /* we need to return the path, thus no long paths here! */ + if (xutftowcsn(wtemplate, template, MAX_PATH, -1) < 0) { + if (errno == ERANGE) + errno = ENAMETOOLONG; return NULL; + } + + if (is_dir_sep(template[0]) && !is_dir_sep(template[1]) && + iswalpha(wtemplate[0]) && wtemplate[1] == L':') { + /* We have an absolute path missing the drive prefix */ + offset = 2; + } if (!_wmktemp(wtemplate)) return NULL; - if (xwcstoutf(template, wtemplate, strlen(template) + 1) < 0) + if (xwcstoutf(template, wtemplate + offset, strlen(template) + 1) < 0) return NULL; return template; } @@ -1115,37 +1376,101 @@ struct tm *localtime_r(const time_t *timep, struct tm *result) } #endif +char *mingw_strbuf_realpath(struct strbuf *resolved, const char *path) +{ + wchar_t wpath[MAX_PATH]; + HANDLE h; + DWORD ret; + int len; + const char *last_component = NULL; + + if (xutftowcs_path(wpath, path) < 0) + return NULL; + + h = CreateFileW(wpath, 0, + FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE, NULL, + OPEN_EXISTING, FILE_FLAG_BACKUP_SEMANTICS, NULL); + + /* + * strbuf_realpath() allows the last path component to not exist. If + * that is the case, now it's time to try without last component. + */ + if (h == INVALID_HANDLE_VALUE && + GetLastError() == ERROR_FILE_NOT_FOUND) { + /* cut last component off of `wpath` */ + wchar_t *p = wpath + wcslen(wpath); + + while (p != wpath) + if (*(--p) == L'/' || *p == L'\\') + break; /* found start of last component */ + + if (p != wpath && (last_component = find_last_dir_sep(path))) { + last_component++; /* skip directory separator */ + *p = L'\0'; + h = CreateFileW(wpath, 0, FILE_SHARE_READ | + FILE_SHARE_WRITE | FILE_SHARE_DELETE, + NULL, OPEN_EXISTING, + FILE_FLAG_BACKUP_SEMANTICS, NULL); + } + } + + if (h == INVALID_HANDLE_VALUE) + return NULL; + + ret = GetFinalPathNameByHandleW(h, wpath, ARRAY_SIZE(wpath), 0); + CloseHandle(h); + if (!ret || ret >= ARRAY_SIZE(wpath)) + return NULL; + + len = wcslen(wpath) * 3; + strbuf_grow(resolved, len); + len = xwcstoutf(resolved->buf, normalize_ntpath(wpath), len); + if (len < 0) + return NULL; + resolved->len = len; + + if (last_component) { + /* Use forward-slash, like `normalize_ntpath()` */ + strbuf_addch(resolved, '/'); + strbuf_addstr(resolved, last_component); + } + + return resolved->buf; + +} + char *mingw_getcwd(char *pointer, int len) { wchar_t cwd[MAX_PATH], wpointer[MAX_PATH]; DWORD ret = GetCurrentDirectoryW(ARRAY_SIZE(cwd), cwd); + HANDLE hnd; if (!ret || ret >= ARRAY_SIZE(cwd)) { errno = ret ? ENAMETOOLONG : err_win_to_posix(GetLastError()); return NULL; } - ret = GetLongPathNameW(cwd, wpointer, ARRAY_SIZE(wpointer)); - if (!ret && GetLastError() == ERROR_ACCESS_DENIED) { - HANDLE hnd = CreateFileW(cwd, 0, - FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE, NULL, - OPEN_EXISTING, FILE_FLAG_BACKUP_SEMANTICS, NULL); - if (hnd == INVALID_HANDLE_VALUE) - return NULL; + hnd = CreateFileW(cwd, 0, + FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE, NULL, + OPEN_EXISTING, FILE_FLAG_BACKUP_SEMANTICS, NULL); + if (hnd != INVALID_HANDLE_VALUE) { ret = GetFinalPathNameByHandleW(hnd, wpointer, ARRAY_SIZE(wpointer), 0); CloseHandle(hnd); - if (!ret || ret >= ARRAY_SIZE(wpointer)) - return NULL; + if (!ret || ret >= ARRAY_SIZE(wpointer)) { + ret = GetLongPathNameW(cwd, wpointer, ARRAY_SIZE(wpointer)); + if (!ret || ret >= ARRAY_SIZE(wpointer)) { + errno = ret ? ENAMETOOLONG : err_win_to_posix(GetLastError()); + return NULL; + } + } if (xwcstoutf(pointer, normalize_ntpath(wpointer), len) < 0) return NULL; return pointer; } - if (!ret || ret >= ARRAY_SIZE(wpointer)) - return NULL; - if (GetFileAttributesW(wpointer) == INVALID_FILE_ATTRIBUTES) { + if (GetFileAttributesW(cwd) == INVALID_FILE_ATTRIBUTES) { errno = ENOENT; return NULL; } - if (xwcstoutf(pointer, wpointer, len) < 0) + if (xwcstoutf(pointer, cwd, len) < 0) return NULL; convert_slashes(pointer); return pointer; @@ -1247,7 +1572,7 @@ static const char *quote_arg_msys2(const char *arg) static const char *parse_interpreter(const char *cmd) { - static char buf[100]; + static char buf[MAX_PATH]; char *p, *opt; int n, fd; @@ -1306,6 +1631,65 @@ static char *lookup_prog(const char *dir, int dirlen, const char *cmd, return NULL; } +static char *path_lookup(const char *cmd, int exe_only); + +static char *is_busybox_applet(const char *cmd) +{ + static struct string_list applets = STRING_LIST_INIT_DUP; + static char *busybox_path; + static int busybox_path_initialized; + + /* Avoid infinite loop */ + if (!strncasecmp(cmd, "busybox", 7) && + (!cmd[7] || !strcasecmp(cmd + 7, ".exe"))) + return NULL; + + if (!busybox_path_initialized) { + busybox_path = path_lookup("busybox.exe", 1); + busybox_path_initialized = 1; + } + + /* Assume that sh is compiled in... */ + if (!busybox_path || !strcasecmp(cmd, "sh")) + return xstrdup_or_null(busybox_path); + + if (!applets.nr) { + struct child_process cp = CHILD_PROCESS_INIT; + struct strbuf buf = STRBUF_INIT; + char *p; + + strvec_pushl(&cp.args, busybox_path, "--help", NULL); + + if (capture_command(&cp, &buf, 2048)) { + string_list_append(&applets, ""); + return NULL; + } + + /* parse output */ + p = strstr(buf.buf, "Currently defined functions:\n"); + if (!p) { + warning("Could not parse output of busybox --help"); + string_list_append(&applets, ""); + return NULL; + } + p = strchrnul(p, '\n'); + for (;;) { + size_t len; + + p += strspn(p, "\n\t ,"); + len = strcspn(p, "\n\t ,"); + if (!len) + break; + p[len] = '\0'; + string_list_insert(&applets, p); + p = p + len + 1; + } + } + + return string_list_has_string(&applets, cmd) ? + xstrdup(busybox_path) : NULL; +} + /* * Determines the absolute path of cmd using the split path in path. * If cmd contains a slash or backslash, no lookup is performed. @@ -1334,6 +1718,9 @@ static char *path_lookup(const char *cmd, int exe_only) path = sep + 1; } + if (!prog && !isexe) + prog = is_busybox_applet(cmd); + return prog; } @@ -1533,8 +1920,8 @@ static int is_msys2_sh(const char *cmd) } static pid_t mingw_spawnve_fd(const char *cmd, const char **argv, char **deltaenv, - const char *dir, - int prepend_cmd, int fhin, int fhout, int fherr) + const char *dir, const char *prepend_cmd, + int fhin, int fhout, int fherr) { static int restrict_handle_inheritance = -1; STARTUPINFOEXW si; @@ -1613,6 +2000,10 @@ static pid_t mingw_spawnve_fd(const char *cmd, const char **argv, char **deltaen if (*argv && !strcmp(cmd, *argv)) wcmd[0] = L'\0'; + /* + * Paths to executables and to the current directory do not support + * long paths, therefore we cannot use xutftowcs_long_path() here. + */ else if (xutftowcs_path(wcmd, cmd) < 0) return -1; if (dir && xutftowcs_path(wdir, dir) < 0) @@ -1621,9 +2012,9 @@ static pid_t mingw_spawnve_fd(const char *cmd, const char **argv, char **deltaen /* concatenate argv, quoting args as we go */ strbuf_init(&args, 0); if (prepend_cmd) { - char *quoted = (char *)quote_arg(cmd); + char *quoted = (char *)quote_arg(prepend_cmd); strbuf_addstr(&args, quoted); - if (quoted != cmd) + if (quoted != prepend_cmd) free(quoted); } for (; *argv; argv++) { @@ -1782,7 +2173,8 @@ static pid_t mingw_spawnve_fd(const char *cmd, const char **argv, char **deltaen return (pid_t)pi.dwProcessId; } -static pid_t mingw_spawnv(const char *cmd, const char **argv, int prepend_cmd) +static pid_t mingw_spawnv(const char *cmd, const char **argv, + const char *prepend_cmd) { return mingw_spawnve_fd(cmd, argv, NULL, NULL, prepend_cmd, 0, 1, 2); } @@ -1810,14 +2202,14 @@ pid_t mingw_spawnvpe(const char *cmd, const char **argv, char **deltaenv, pid = -1; } else { - pid = mingw_spawnve_fd(iprog, argv, deltaenv, dir, 1, + pid = mingw_spawnve_fd(iprog, argv, deltaenv, dir, interpr, fhin, fhout, fherr); free(iprog); } argv[0] = argv0; } else - pid = mingw_spawnve_fd(prog, argv, deltaenv, dir, 0, + pid = mingw_spawnve_fd(prog, argv, deltaenv, dir, NULL, fhin, fhout, fherr); free(prog); } @@ -1845,7 +2237,7 @@ static int try_shell_exec(const char *cmd, char *const *argv) argv2[0] = (char *)cmd; /* full path to the script file */ COPY_ARRAY(&argv2[1], &argv[1], argc); exec_id = trace2_exec(prog, argv2); - pid = mingw_spawnv(prog, argv2, 1); + pid = mingw_spawnv(prog, argv2, interpr); if (pid >= 0) { int status; if (waitpid(pid, &status, 0) < 0) @@ -1869,7 +2261,7 @@ int mingw_execv(const char *cmd, char *const *argv) int exec_id; exec_id = trace2_exec(cmd, (const char **)argv); - pid = mingw_spawnv(cmd, (const char **)argv, 0); + pid = mingw_spawnv(cmd, (const char **)argv, NULL); if (pid < 0) { trace2_exec_result(exec_id, -1); return -1; @@ -1898,16 +2290,28 @@ int mingw_execvp(const char *cmd, char *const *argv) int mingw_kill(pid_t pid, int sig) { if (pid > 0 && sig == SIGTERM) { - HANDLE h = OpenProcess(PROCESS_TERMINATE, FALSE, pid); - - if (TerminateProcess(h, -1)) { + HANDLE h = OpenProcess(PROCESS_CREATE_THREAD | + PROCESS_QUERY_INFORMATION | + PROCESS_VM_OPERATION | PROCESS_VM_WRITE | + PROCESS_VM_READ | PROCESS_TERMINATE, + FALSE, pid); + int ret; + + if (h) + ret = exit_process(h, 128 + sig); + else { + h = OpenProcess(PROCESS_TERMINATE, FALSE, pid); + if (!h) { + errno = err_win_to_posix(GetLastError()); + return -1; + } + ret = terminate_process_tree(h, 128 + sig); + } + if (ret) { + errno = err_win_to_posix(GetLastError()); CloseHandle(h); - return 0; } - - errno = err_win_to_posix(GetLastError()); - CloseHandle(h); - return -1; + return ret; } else if (pid > 0 && sig == 0) { HANDLE h = OpenProcess(PROCESS_QUERY_INFORMATION, FALSE, pid); if (h) { @@ -2018,18 +2422,150 @@ static void ensure_socket_initialization(void) initialized = 1; } +static int winsock_error_to_errno(DWORD err) +{ + switch (err) { + case WSAEINTR: return EINTR; + case WSAEBADF: return EBADF; + case WSAEACCES: return EACCES; + case WSAEFAULT: return EFAULT; + case WSAEINVAL: return EINVAL; + case WSAEMFILE: return EMFILE; + case WSAEWOULDBLOCK: return EWOULDBLOCK; + case WSAEINPROGRESS: return EINPROGRESS; + case WSAEALREADY: return EALREADY; + case WSAENOTSOCK: return ENOTSOCK; + case WSAEDESTADDRREQ: return EDESTADDRREQ; + case WSAEMSGSIZE: return EMSGSIZE; + case WSAEPROTOTYPE: return EPROTOTYPE; + case WSAENOPROTOOPT: return ENOPROTOOPT; + case WSAEPROTONOSUPPORT: return EPROTONOSUPPORT; + case WSAEOPNOTSUPP: return EOPNOTSUPP; + case WSAEAFNOSUPPORT: return EAFNOSUPPORT; + case WSAEADDRINUSE: return EADDRINUSE; + case WSAEADDRNOTAVAIL: return EADDRNOTAVAIL; + case WSAENETDOWN: return ENETDOWN; + case WSAENETUNREACH: return ENETUNREACH; + case WSAENETRESET: return ENETRESET; + case WSAECONNABORTED: return ECONNABORTED; + case WSAECONNRESET: return ECONNRESET; + case WSAENOBUFS: return ENOBUFS; + case WSAEISCONN: return EISCONN; + case WSAENOTCONN: return ENOTCONN; + case WSAETIMEDOUT: return ETIMEDOUT; + case WSAECONNREFUSED: return ECONNREFUSED; + case WSAELOOP: return ELOOP; + case WSAENAMETOOLONG: return ENAMETOOLONG; + case WSAEHOSTUNREACH: return EHOSTUNREACH; + case WSAENOTEMPTY: return ENOTEMPTY; + /* No errno equivalent; default to EIO */ + case WSAESOCKTNOSUPPORT: + case WSAEPFNOSUPPORT: + case WSAESHUTDOWN: + case WSAETOOMANYREFS: + case WSAEHOSTDOWN: + case WSAEPROCLIM: + case WSAEUSERS: + case WSAEDQUOT: + case WSAESTALE: + case WSAEREMOTE: + case WSASYSNOTREADY: + case WSAVERNOTSUPPORTED: + case WSANOTINITIALISED: + case WSAEDISCON: + case WSAENOMORE: + case WSAECANCELLED: + case WSAEINVALIDPROCTABLE: + case WSAEINVALIDPROVIDER: + case WSAEPROVIDERFAILEDINIT: + case WSASYSCALLFAILURE: + case WSASERVICE_NOT_FOUND: + case WSATYPE_NOT_FOUND: + case WSA_E_NO_MORE: + case WSA_E_CANCELLED: + case WSAEREFUSED: + case WSAHOST_NOT_FOUND: + case WSATRY_AGAIN: + case WSANO_RECOVERY: + case WSANO_DATA: + case WSA_QOS_RECEIVERS: + case WSA_QOS_SENDERS: + case WSA_QOS_NO_SENDERS: + case WSA_QOS_NO_RECEIVERS: + case WSA_QOS_REQUEST_CONFIRMED: + case WSA_QOS_ADMISSION_FAILURE: + case WSA_QOS_POLICY_FAILURE: + case WSA_QOS_BAD_STYLE: + case WSA_QOS_BAD_OBJECT: + case WSA_QOS_TRAFFIC_CTRL_ERROR: + case WSA_QOS_GENERIC_ERROR: + case WSA_QOS_ESERVICETYPE: + case WSA_QOS_EFLOWSPEC: + case WSA_QOS_EPROVSPECBUF: + case WSA_QOS_EFILTERSTYLE: + case WSA_QOS_EFILTERTYPE: + case WSA_QOS_EFILTERCOUNT: + case WSA_QOS_EOBJLENGTH: + case WSA_QOS_EFLOWCOUNT: +#ifndef _MSC_VER + case WSA_QOS_EUNKNOWNPSOBJ: +#endif + case WSA_QOS_EPOLICYOBJ: + case WSA_QOS_EFLOWDESC: + case WSA_QOS_EPSFLOWSPEC: + case WSA_QOS_EPSFILTERSPEC: + case WSA_QOS_ESDMODEOBJ: + case WSA_QOS_ESHAPERATEOBJ: + case WSA_QOS_RESERVED_PETYPE: + default: return EIO; + } +} + +/* + * On Windows, `errno` is a global macro to a function call. + * This makes it difficult to debug and single-step our mappings. + */ +static inline void set_wsa_errno(void) +{ + DWORD wsa = WSAGetLastError(); + int e = winsock_error_to_errno(wsa); + errno = e; + +#ifdef DEBUG_WSA_ERRNO + fprintf(stderr, "winsock error: %d -> %d\n", wsa, e); + fflush(stderr); +#endif +} + +static inline int winsock_return(int ret) +{ + if (ret < 0) + set_wsa_errno(); + + return ret; +} + +#define WINSOCK_RETURN(x) do { return winsock_return(x); } while (0) + #undef gethostname int mingw_gethostname(char *name, int namelen) { - ensure_socket_initialization(); - return gethostname(name, namelen); + ensure_socket_initialization(); + WINSOCK_RETURN(gethostname(name, namelen)); } #undef gethostbyname struct hostent *mingw_gethostbyname(const char *host) { + struct hostent *ret; + ensure_socket_initialization(); - return gethostbyname(host); + + ret = gethostbyname(host); + if (!ret) + set_wsa_errno(); + + return ret; } #undef getaddrinfo @@ -2037,7 +2573,7 @@ int mingw_getaddrinfo(const char *node, const char *service, const struct addrinfo *hints, struct addrinfo **res) { ensure_socket_initialization(); - return getaddrinfo(node, service, hints, res); + WINSOCK_RETURN(getaddrinfo(node, service, hints, res)); } int mingw_socket(int domain, int type, int protocol) @@ -2057,7 +2593,7 @@ int mingw_socket(int domain, int type, int protocol) * in errno so that _if_ someone looks up the code somewhere, * then it is at least the number that are usually listed. */ - errno = WSAGetLastError(); + set_wsa_errno(); return -1; } /* convert into a file descriptor */ @@ -2073,35 +2609,35 @@ int mingw_socket(int domain, int type, int protocol) int mingw_connect(int sockfd, struct sockaddr *sa, size_t sz) { SOCKET s = (SOCKET)_get_osfhandle(sockfd); - return connect(s, sa, sz); + WINSOCK_RETURN(connect(s, sa, sz)); } #undef bind int mingw_bind(int sockfd, struct sockaddr *sa, size_t sz) { SOCKET s = (SOCKET)_get_osfhandle(sockfd); - return bind(s, sa, sz); + WINSOCK_RETURN(bind(s, sa, sz)); } #undef setsockopt int mingw_setsockopt(int sockfd, int lvl, int optname, void *optval, int optlen) { SOCKET s = (SOCKET)_get_osfhandle(sockfd); - return setsockopt(s, lvl, optname, (const char*)optval, optlen); + WINSOCK_RETURN(setsockopt(s, lvl, optname, (const char*)optval, optlen)); } #undef shutdown int mingw_shutdown(int sockfd, int how) { SOCKET s = (SOCKET)_get_osfhandle(sockfd); - return shutdown(s, how); + WINSOCK_RETURN(shutdown(s, how)); } #undef listen int mingw_listen(int sockfd, int backlog) { SOCKET s = (SOCKET)_get_osfhandle(sockfd); - return listen(s, backlog); + WINSOCK_RETURN(listen(s, backlog)); } #undef accept @@ -2112,6 +2648,11 @@ int mingw_accept(int sockfd1, struct sockaddr *sa, socklen_t *sz) SOCKET s1 = (SOCKET)_get_osfhandle(sockfd1); SOCKET s2 = accept(s1, sa, sz); + if (s2 == INVALID_SOCKET) { + set_wsa_errno(); + return -1; + } + /* convert into a file descriptor */ if ((sockfd2 = _open_osfhandle(s2, O_RDWR|O_BINARY)) < 0) { int err = errno; @@ -2125,26 +2666,36 @@ int mingw_accept(int sockfd1, struct sockaddr *sa, socklen_t *sz) #undef rename int mingw_rename(const char *pold, const char *pnew) { - DWORD attrs, gle; + DWORD attrs = INVALID_FILE_ATTRIBUTES, gle; int tries = 0; - wchar_t wpold[MAX_PATH], wpnew[MAX_PATH]; - if (xutftowcs_path(wpold, pold) < 0 || xutftowcs_path(wpnew, pnew) < 0) + wchar_t wpold[MAX_LONG_PATH], wpnew[MAX_LONG_PATH]; + if (xutftowcs_long_path(wpold, pold) < 0 || + xutftowcs_long_path(wpnew, pnew) < 0) return -1; - /* - * Try native rename() first to get errno right. - * It is based on MoveFile(), which cannot overwrite existing files. - */ - if (!_wrename(wpold, wpnew)) - return 0; - if (errno != EEXIST) - return -1; repeat: - if (MoveFileExW(wpold, wpnew, MOVEFILE_REPLACE_EXISTING)) + if (MoveFileExW(wpold, wpnew, + MOVEFILE_REPLACE_EXISTING | MOVEFILE_COPY_ALLOWED)) return 0; - /* TODO: translate more errors */ gle = GetLastError(); - if (gle == ERROR_ACCESS_DENIED && + + if (gle == ERROR_ACCESS_DENIED && is_inside_windows_container()) { + /* Fall back to copy to destination & remove source */ + if (CopyFileW(wpold, wpnew, FALSE) && !mingw_unlink(pold)) + return 0; + gle = GetLastError(); + } + + /* revert file attributes on failure */ + if (attrs != INVALID_FILE_ATTRIBUTES) + SetFileAttributesW(wpnew, attrs); + + if (!is_file_in_use_error(gle)) { + errno = err_win_to_posix(gle); + return -1; + } + + if (attrs == INVALID_FILE_ATTRIBUTES && (attrs = GetFileAttributesW(wpnew)) != INVALID_FILE_ATTRIBUTES) { if (attrs & FILE_ATTRIBUTE_DIRECTORY) { DWORD attrsold = GetFileAttributesW(wpold); @@ -2156,28 +2707,10 @@ int mingw_rename(const char *pold, const char *pnew) return -1; } if ((attrs & FILE_ATTRIBUTE_READONLY) && - SetFileAttributesW(wpnew, attrs & ~FILE_ATTRIBUTE_READONLY)) { - if (MoveFileExW(wpold, wpnew, MOVEFILE_REPLACE_EXISTING)) - return 0; - gle = GetLastError(); - /* revert file attributes on failure */ - SetFileAttributesW(wpnew, attrs); - } + SetFileAttributesW(wpnew, attrs & ~FILE_ATTRIBUTE_READONLY)) + goto repeat; } - if (tries < ARRAY_SIZE(delay) && gle == ERROR_ACCESS_DENIED) { - /* - * We assume that some other process had the source or - * destination file open at the wrong moment and retry. - * In order to give the other process a higher chance to - * complete its operation, we give up our time slice now. - * If we have to retry again, we do sleep a bit. - */ - Sleep(delay[tries]); - tries++; - goto repeat; - } - if (gle == ERROR_ACCESS_DENIED && - ask_yes_no_if_possible("Rename from '%s' to '%s' failed. " + if (retry_ask_yes_no(&tries, "Rename from '%s' to '%s' failed. " "Should I try again?", pold, pnew)) goto repeat; @@ -2442,9 +2975,9 @@ int mingw_raise(int sig) int link(const char *oldpath, const char *newpath) { - wchar_t woldpath[MAX_PATH], wnewpath[MAX_PATH]; - if (xutftowcs_path(woldpath, oldpath) < 0 || - xutftowcs_path(wnewpath, newpath) < 0) + wchar_t woldpath[MAX_LONG_PATH], wnewpath[MAX_LONG_PATH]; + if (xutftowcs_long_path(woldpath, oldpath) < 0 || + xutftowcs_long_path(wnewpath, newpath) < 0) return -1; if (!CreateHardLinkW(wnewpath, woldpath, NULL)) { @@ -2454,6 +2987,199 @@ int link(const char *oldpath, const char *newpath) return 0; } +enum symlink_type { + SYMLINK_TYPE_UNSPECIFIED = 0, + SYMLINK_TYPE_FILE, + SYMLINK_TYPE_DIRECTORY, +}; + +static enum symlink_type check_symlink_attr(struct index_state *index, const char *link) +{ + static struct attr_check *check; + const char *value; + + if (!index) + return SYMLINK_TYPE_UNSPECIFIED; + + if (!check) + check = attr_check_initl("symlink", NULL); + + git_check_attr(index, link, check); + + value = check->items[0].value; + if (ATTR_UNSET(value)) + return SYMLINK_TYPE_UNSPECIFIED; + if (!strcmp(value, "file")) + return SYMLINK_TYPE_FILE; + if (!strcmp(value, "dir") || !strcmp(value, "directory")) + return SYMLINK_TYPE_DIRECTORY; + + warning(_("ignoring invalid symlink type '%s' for '%s'"), value, link); + return SYMLINK_TYPE_UNSPECIFIED; +} + +int mingw_create_symlink(struct index_state *index, const char *target, const char *link) +{ + wchar_t wtarget[MAX_LONG_PATH], wlink[MAX_LONG_PATH]; + int len; + + /* fail if symlinks are disabled or API is not supported (WinXP) */ + if (!has_symlinks) { + errno = ENOSYS; + return -1; + } + + if ((len = xutftowcs_long_path(wtarget, target)) < 0 + || xutftowcs_long_path(wlink, link) < 0) + return -1; + + /* convert target dir separators to backslashes */ + while (len--) + if (wtarget[len] == '/') + wtarget[len] = '\\'; + + switch (check_symlink_attr(index, link)) { + case SYMLINK_TYPE_UNSPECIFIED: + /* Create a phantom symlink: it is initially created as a file + * symlink, but may change to a directory symlink later if/when + * the target exists. */ + return create_phantom_symlink(wtarget, wlink); + case SYMLINK_TYPE_FILE: + if (!CreateSymbolicLinkW(wlink, wtarget, symlink_file_flags)) + break; + return 0; + case SYMLINK_TYPE_DIRECTORY: + if (!CreateSymbolicLinkW(wlink, wtarget, + symlink_directory_flags)) + break; + /* There may be dangling phantom symlinks that point at this + * one, which should now morph into directory symlinks. */ + process_phantom_symlinks(); + return 0; + default: + BUG("unhandled symlink type"); + } + + /* CreateSymbolicLinkW failed. */ + errno = err_win_to_posix(GetLastError()); + return -1; +} + +#ifndef _WINNT_H +/* + * The REPARSE_DATA_BUFFER structure is defined in the Windows DDK (in + * ntifs.h) and in MSYS1's winnt.h (which defines _WINNT_H). So define + * it ourselves if we are on MSYS2 (whose winnt.h defines _WINNT_). + */ +typedef struct _REPARSE_DATA_BUFFER { + DWORD ReparseTag; + WORD ReparseDataLength; + WORD Reserved; +#ifndef _MSC_VER + _ANONYMOUS_UNION +#endif + union { + struct { + WORD SubstituteNameOffset; + WORD SubstituteNameLength; + WORD PrintNameOffset; + WORD PrintNameLength; + ULONG Flags; + WCHAR PathBuffer[1]; + } SymbolicLinkReparseBuffer; + struct { + WORD SubstituteNameOffset; + WORD SubstituteNameLength; + WORD PrintNameOffset; + WORD PrintNameLength; + WCHAR PathBuffer[1]; + } MountPointReparseBuffer; + struct { + BYTE DataBuffer[1]; + } GenericReparseBuffer; + } DUMMYUNIONNAME; +} REPARSE_DATA_BUFFER, *PREPARSE_DATA_BUFFER; +#endif + +static int readlink_1(const WCHAR *wpath, BOOL fail_on_unknown_tag, + char *tmpbuf, int *plen, DWORD *ptag) +{ + HANDLE handle; + WCHAR *wbuf; + REPARSE_DATA_BUFFER *b = alloca(MAXIMUM_REPARSE_DATA_BUFFER_SIZE); + DWORD dummy; + + /* read reparse point data */ + handle = CreateFileW(wpath, 0, + FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE, NULL, + OPEN_EXISTING, + FILE_FLAG_BACKUP_SEMANTICS | FILE_FLAG_OPEN_REPARSE_POINT, NULL); + if (handle == INVALID_HANDLE_VALUE) { + errno = err_win_to_posix(GetLastError()); + return -1; + } + if (!DeviceIoControl(handle, FSCTL_GET_REPARSE_POINT, NULL, 0, b, + MAXIMUM_REPARSE_DATA_BUFFER_SIZE, &dummy, NULL)) { + errno = err_win_to_posix(GetLastError()); + CloseHandle(handle); + return -1; + } + CloseHandle(handle); + + /* get target path for symlinks or mount points (aka 'junctions') */ + switch ((*ptag = b->ReparseTag)) { + case IO_REPARSE_TAG_SYMLINK: + wbuf = (WCHAR*) (((char*) b->SymbolicLinkReparseBuffer.PathBuffer) + + b->SymbolicLinkReparseBuffer.SubstituteNameOffset); + *(WCHAR*) (((char*) wbuf) + + b->SymbolicLinkReparseBuffer.SubstituteNameLength) = 0; + break; + case IO_REPARSE_TAG_MOUNT_POINT: + wbuf = (WCHAR*) (((char*) b->MountPointReparseBuffer.PathBuffer) + + b->MountPointReparseBuffer.SubstituteNameOffset); + *(WCHAR*) (((char*) wbuf) + + b->MountPointReparseBuffer.SubstituteNameLength) = 0; + break; + default: + if (fail_on_unknown_tag) { + errno = EINVAL; + return -1; + } else { + *plen = MAX_LONG_PATH; + return 0; + } + } + + if ((*plen = + xwcstoutf(tmpbuf, normalize_ntpath(wbuf), MAX_LONG_PATH)) < 0) + return -1; + return 0; +} + +int readlink(const char *path, char *buf, size_t bufsiz) +{ + WCHAR wpath[MAX_LONG_PATH]; + char tmpbuf[MAX_LONG_PATH]; + int len; + DWORD tag; + + if (xutftowcs_long_path(wpath, path) < 0) + return -1; + + if (readlink_1(wpath, TRUE, tmpbuf, &len, &tag) < 0) + return -1; + + /* + * Adapt to strange readlink() API: Copy up to bufsiz *bytes*, potentially + * cutting off a UTF-8 sequence. Insufficient bufsize is *not* a failure + * condition. There is no conversion function that produces invalid UTF-8, + * so convert to a (hopefully large enough) temporary buffer, then memcpy + * the requested number of bytes (including '\0' for robustness). + */ + memcpy(buf, tmpbuf, min(bufsiz, len + 1)); + return min(bufsiz, len); +} + pid_t waitpid(pid_t pid, int *status, int options) { HANDLE h = OpenProcess(SYNCHRONIZE | PROCESS_QUERY_INFORMATION, @@ -2506,6 +3232,30 @@ pid_t waitpid(pid_t pid, int *status, int options) return -1; } +int (*win32_is_mount_point)(struct strbuf *path) = mingw_is_mount_point; + +int mingw_is_mount_point(struct strbuf *path) +{ + WIN32_FIND_DATAW findbuf = { 0 }; + HANDLE handle; + wchar_t wfilename[MAX_LONG_PATH]; + int wlen = xutftowcs_long_path(wfilename, path->buf); + if (wlen < 0) + die(_("could not get long path for '%s'"), path->buf); + + /* remove trailing slash, if any */ + if (wlen > 0 && wfilename[wlen - 1] == L'/') + wfilename[--wlen] = L'\0'; + + handle = FindFirstFileW(wfilename, &findbuf); + if (handle == INVALID_HANDLE_VALUE) + return 0; + FindClose(handle); + + return (findbuf.dwFileAttributes & FILE_ATTRIBUTE_REPARSE_POINT) && + (findbuf.dwReserved0 == IO_REPARSE_TAG_MOUNT_POINT); +} + int xutftowcsn(wchar_t *wcs, const char *utfs, size_t wcslen, int utflen) { int upos = 0, wpos = 0; @@ -2591,6 +3341,59 @@ int xwcstoutf(char *utf, const wchar_t *wcs, size_t utflen) return -1; } +#ifdef ENSURE_MSYSTEM_IS_SET +static size_t append_system_bin_dirs(char *path, size_t size) +{ +#if !defined(RUNTIME_PREFIX) || !defined(HAVE_WPGMPTR) + return 0; +#else + char prefix[32768]; + const char *slash; + size_t len = xwcstoutf(prefix, _wpgmptr, sizeof(prefix)), off = 0; + + if (len == 0 || len >= sizeof(prefix) || + !(slash = find_last_dir_sep(prefix))) + return 0; + /* strip trailing `git.exe` */ + len = slash - prefix; + + /* strip trailing `cmd` or `mingw64\bin` or `mingw32\bin` or `bin` or `libexec\git-core` */ + if (strip_suffix_mem(prefix, &len, "\\mingw64\\libexec\\git-core") || + strip_suffix_mem(prefix, &len, "\\mingw64\\bin")) + off += xsnprintf(path + off, size - off, + "%.*s\\mingw64\\bin;", (int)len, prefix); + else if (strip_suffix_mem(prefix, &len, "\\mingw32\\libexec\\git-core") || + strip_suffix_mem(prefix, &len, "\\mingw32\\bin")) + off += xsnprintf(path + off, size - off, + "%.*s\\mingw32\\bin;", (int)len, prefix); + else if (strip_suffix_mem(prefix, &len, "\\cmd") || + strip_suffix_mem(prefix, &len, "\\bin") || + strip_suffix_mem(prefix, &len, "\\libexec\\git-core")) + off += xsnprintf(path + off, size - off, + "%.*s\\mingw%d\\bin;", (int)len, prefix, + (int)(sizeof(void *) * 8)); + else + return 0; + + off += xsnprintf(path + off, size - off, + "%.*s\\usr\\bin;", (int)len, prefix); + return off; +#endif +} +#endif + +static int is_system32_path(const char *path) +{ + WCHAR system32[MAX_LONG_PATH], wpath[MAX_LONG_PATH]; + + if (xutftowcs_long_path(wpath, path) < 0 || + !GetSystemDirectoryW(system32, ARRAY_SIZE(system32)) || + _wcsicmp(system32, wpath)) + return 0; + + return 1; +} + static void setup_windows_environment(void) { char *tmp = getenv("TMPDIR"); @@ -2615,9 +3418,20 @@ static void setup_windows_environment(void) convert_slashes(tmp); } - /* simulate TERM to enable auto-color (see color.c) */ - if (!getenv("TERM")) - setenv("TERM", "cygwin", 1); + + /* + * Make sure TERM is set up correctly to enable auto-color + * (see color.c .) Use "cygwin" for older OS releases which + * works correctly with MSYS2 utilities on older consoles. + */ + if (!getenv("TERM")) { + if ((GetVersion() >> 16) < 15063) + setenv("TERM", "cygwin", 0); + else { + setenv("TERM", "xterm-256color", 0); + setenv("COLORTERM", "truecolor", 0); + } + } /* calculate HOME if not set */ if (!getenv("HOME")) { @@ -2631,7 +3445,8 @@ static void setup_windows_environment(void) strbuf_addstr(&buf, tmp); if ((tmp = getenv("HOMEPATH"))) { strbuf_addstr(&buf, tmp); - if (is_directory(buf.buf)) + if (!is_system32_path(buf.buf) && + is_directory(buf.buf)) setenv("HOME", buf.buf, 1); else tmp = NULL; /* use $USERPROFILE */ @@ -2642,6 +3457,46 @@ static void setup_windows_environment(void) if (!tmp && (tmp = getenv("USERPROFILE"))) setenv("HOME", tmp, 1); } + + if (!getenv("PLINK_PROTOCOL")) + setenv("PLINK_PROTOCOL", "ssh", 0); + +#ifdef ENSURE_MSYSTEM_IS_SET + if (!(tmp = getenv("MSYSTEM")) || !tmp[0]) { + const char *home = getenv("HOME"), *path = getenv("PATH"); + char buf[32768]; + size_t off = 0; + + xsnprintf(buf, sizeof(buf), + "MINGW%d", (int)(sizeof(void *) * 8)); + setenv("MSYSTEM", buf, 1); + + if (home) + off += xsnprintf(buf + off, sizeof(buf) - off, + "%s\\bin;", home); + off += append_system_bin_dirs(buf + off, sizeof(buf) - off); + if (path) + off += xsnprintf(buf + off, sizeof(buf) - off, + "%s", path); + else if (off > 0) + buf[off - 1] = '\0'; + else + buf[0] = '\0'; + setenv("PATH", buf, 1); + } +#endif + + if (!getenv("LC_ALL") && !getenv("LC_CTYPE") && !getenv("LANG")) + setenv("LC_CTYPE", "C.UTF-8", 1); + + /* + * Change 'core.symlinks' default to false, unless native symlinks are + * enabled in MSys2 (via 'MSYS=winsymlinks:nativestrict'). Thus we can + * run the test suite (which doesn't obey config files) with or without + * symlink support. + */ + if (!(tmp = getenv("MSYS")) || !strstr(tmp, "winsymlinks:nativestrict")) + has_symlinks = 0; } static PSID get_current_user_sid(void) @@ -2720,9 +3575,7 @@ int is_path_owned_by_current_sid(const char *path, struct strbuf *report) DACL_SECURITY_INFORMATION, &sid, NULL, NULL, NULL, &descriptor); - if (err != ERROR_SUCCESS) - error(_("failed to get owner for '%s' (%ld)"), path, err); - else if (sid && IsValidSid(sid)) { + if (err == ERROR_SUCCESS && sid && IsValidSid(sid)) { /* Now, verify that the SID matches the current user's */ static PSID current_user_sid; BOOL is_member; @@ -2910,6 +3763,68 @@ int is_valid_win32_path(const char *path, int allow_literal_nul) } } +int handle_long_path(wchar_t *path, int len, int max_path, int expand) +{ + int result; + wchar_t buf[MAX_LONG_PATH]; + + /* + * we don't need special handling if path is relative to the current + * directory, and current directory + path don't exceed the desired + * max_path limit. This should cover > 99 % of cases with minimal + * performance impact (git almost always uses relative paths). + */ + if ((len < 2 || (!is_dir_sep(path[0]) && path[1] != ':')) && + (current_directory_len + len < max_path)) + return len; + + /* + * handle everything else: + * - absolute paths: "C:\dir\file" + * - absolute UNC paths: "\\server\share\dir\file" + * - absolute paths on current drive: "\dir\file" + * - relative paths on other drive: "X:file" + * - prefixed paths: "\\?\...", "\\.\..." + */ + + /* convert to absolute path using GetFullPathNameW */ + result = GetFullPathNameW(path, MAX_LONG_PATH, buf, NULL); + if (!result) { + errno = err_win_to_posix(GetLastError()); + return -1; + } + + /* + * return absolute path if it fits within max_path (even if + * "cwd + path" doesn't due to '..' components) + */ + if (result < max_path) { + wcscpy(path, buf); + return result; + } + + /* error out if we shouldn't expand the path or buf is too small */ + if (!expand || result >= MAX_LONG_PATH - 6) { + errno = ENAMETOOLONG; + return -1; + } + + /* prefix full path with "\\?\" or "\\?\UNC\" */ + if (buf[0] == '\\') { + /* ...unless already prefixed */ + if (buf[1] == '\\' && (buf[2] == '?' || buf[2] == '.')) + return len; + + wcscpy(path, L"\\\\?\\UNC\\"); + wcscpy(path + 8, buf + 2); + return result + 6; + } else { + wcscpy(path, L"\\\\?\\"); + wcscpy(path + 4, buf); + return result + 4; + } +} + #if !defined(_MSC_VER) /* * Disable MSVCRT command line wildcard expansion (__getmainargs called from @@ -2994,6 +3909,31 @@ static void maybe_redirect_std_handles(void) GENERIC_WRITE, FILE_FLAG_NO_BUFFERING); } +static void adjust_symlink_flags(void) +{ + /* + * Starting with Windows 10 Build 14972, symbolic links can be created + * using CreateSymbolicLink() without elevation by passing the flag + * SYMBOLIC_LINK_FLAG_ALLOW_UNPRIVILEGED_CREATE (0x02) as last + * parameter, provided the Developer Mode has been enabled. Some + * earlier Windows versions complain about this flag with an + * ERROR_INVALID_PARAMETER, hence we have to test the build number + * specifically. + */ + if (GetVersion() >= 14972 << 16) { + symlink_file_flags |= 2; + symlink_directory_flags |= 2; + } +} + +static BOOL WINAPI handle_ctrl_c(DWORD ctrl_type) +{ + if (ctrl_type != CTRL_C_EVENT) + return FALSE; /* we did not handle this */ + mingw_raise(SIGINT); + return TRUE; /* we did handle this */ +} + #ifdef _MSC_VER #ifdef _DEBUG #include @@ -3027,7 +3967,11 @@ int wmain(int argc, const wchar_t **wargv) #endif #endif + SetConsoleCtrlHandler(handle_ctrl_c, TRUE); + maybe_redirect_std_handles(); + adjust_symlink_flags(); + fsync_object_files = 1; /* determine size of argv and environ conversion buffer */ maxlen = wcslen(wargv[0]); @@ -3057,6 +4001,10 @@ int wmain(int argc, const wchar_t **wargv) /* initialize critical section for waitpid pinfo_t list */ InitializeCriticalSection(&pinfo_cs); + InitializeCriticalSection(&phantom_symlinks_cs); + + /* initialize critical section for fscache */ + InitializeCriticalSection(&fscache_cs); /* set up default file mode and file modes for stdin/out/err */ _fmode = _O_BINARY; @@ -3067,6 +4015,9 @@ int wmain(int argc, const wchar_t **wargv) /* initialize Unicode console */ winansi_init(); + /* init length of current directory for handle_long_path */ + current_directory_len = GetCurrentDirectoryW(0, NULL); + /* invoke the real main() using our utf8 version of argv. */ exit_status = main(argc, argv); @@ -3090,3 +4041,62 @@ int uname(struct utsname *buf) "%u", (v >> 16) & 0x7fff); return 0; } + +/* + * Based on https://stackoverflow.com/questions/43002803 + * + * [HKLM\SYSTEM\CurrentControlSet\Services\cexecsvc] + * "DisplayName"="@%systemroot%\\system32\\cexecsvc.exe,-100" + * "ErrorControl"=dword:00000001 + * "ImagePath"=hex(2):25,00,73,00,79,00,73,00,74,00,65,00,6d,00,72,00,6f,00, + * 6f,00,74,00,25,00,5c,00,73,00,79,00,73,00,74,00,65,00,6d,00,33,00,32,00, + * 5c,00,63,00,65,00,78,00,65,00,63,00,73,00,76,00,63,00,2e,00,65,00,78,00, + * 65,00,00,00 + * "Start"=dword:00000002 + * "Type"=dword:00000010 + * "Description"="@%systemroot%\\system32\\cexecsvc.exe,-101" + * "ObjectName"="LocalSystem" + * "ServiceSidType"=dword:00000001 + */ +int is_inside_windows_container(void) +{ + static int inside_container = -1; /* -1 uninitialized */ + const char *key = "SYSTEM\\CurrentControlSet\\Services\\cexecsvc"; + HKEY handle = NULL; + + if (inside_container != -1) + return inside_container; + + inside_container = ERROR_SUCCESS == + RegOpenKeyExA(HKEY_LOCAL_MACHINE, key, 0, KEY_READ, &handle); + RegCloseKey(handle); + + return inside_container; +} + +int file_attr_to_st_mode (DWORD attr, DWORD tag, const char *path) +{ + int fMode = S_IREAD; + if ((attr & FILE_ATTRIBUTE_REPARSE_POINT) && + tag == IO_REPARSE_TAG_SYMLINK) { + int flag = S_IFLNK; + char buf[MAX_LONG_PATH]; + + /* + * Windows containers' mapped volumes are marked as reparse + * points and look like symbolic links, but they are not. + */ + if (path && is_inside_windows_container() && + readlink(path, buf, sizeof(buf)) > 27 && + starts_with(buf, "/ContainerMappedDirectories/")) + flag = S_IFDIR; + + fMode |= flag; + } else if (attr & FILE_ATTRIBUTE_DIRECTORY) + fMode |= S_IFDIR; + else + fMode |= S_IFREG; + if (!(attr & FILE_ATTRIBUTE_READONLY)) + fMode |= S_IWRITE; + return fMode; +} diff --git a/compat/mingw.h b/compat/mingw.h index 209cf7cebadd17..c9624a932bbc0a 100644 --- a/compat/mingw.h +++ b/compat/mingw.h @@ -11,6 +11,11 @@ typedef _sigset_t sigset_t; #undef _POSIX_THREAD_SAFE_FUNCTIONS #endif +extern int core_fscache; + +struct repository; +int are_long_paths_enabled(void); + int mingw_core_config(const char *var, const char *value, void *cb); #define platform_core_config mingw_core_config @@ -120,10 +125,6 @@ struct utsname { * trivial stubs */ -static inline int readlink(const char *path, char *buf, size_t bufsiz) -{ errno = ENOSYS; return -1; } -static inline int symlink(const char *oldpath, const char *newpath) -{ errno = ENOSYS; return -1; } static inline int fchmod(int fildes, mode_t mode) { errno = ENOSYS; return -1; } #ifndef __MINGW64_VERSION_MAJOR @@ -214,6 +215,10 @@ int setitimer(int type, struct itimerval *in, struct itimerval *out); int sigaction(int sig, struct sigaction *in, struct sigaction *out); int link(const char *oldpath, const char *newpath); int uname(struct utsname *buf); +int readlink(const char *path, char *buf, size_t bufsiz); +struct index_state; +int mingw_create_symlink(struct index_state *index, const char *target, const char *link); +#define create_symlink mingw_create_symlink /* * replacements of existing functions @@ -351,6 +356,17 @@ static inline int getrlimit(int resource, struct rlimit *rlp) return 0; } +/* + * The unit of FILETIME is 100-nanoseconds since January 1, 1601, UTC. + * Returns the 100-nanoseconds ("hekto nanoseconds") since the epoch. + */ +static inline long long filetime_to_hnsec(const FILETIME *ft) +{ + long long winTime = ((long long)ft->dwHighDateTime << 32) + ft->dwLowDateTime; + /* Windows to Unix Epoch conversion */ + return winTime - 116444736000000000LL; +} + /* * Use mingw specific stat()/lstat()/fstat() implementations on Windows, * including our own struct stat with 64 bit st_size and nanosecond-precision @@ -367,6 +383,13 @@ struct timespec { #endif #endif +static inline void filetime_to_timespec(const FILETIME *ft, struct timespec *ts) +{ + long long hnsec = filetime_to_hnsec(ft); + ts->tv_sec = (time_t)(hnsec / 10000000); + ts->tv_nsec = (hnsec % 10000000) * 100; +} + struct mingw_stat { _dev_t st_dev; _ino_t st_ino; @@ -399,7 +422,7 @@ int mingw_fstat(int fd, struct stat *buf); #ifdef lstat #undef lstat #endif -#define lstat mingw_lstat +extern int (*lstat)(const char *file_name, struct stat *buf); int mingw_utime(const char *file_name, const struct utimbuf *times); @@ -449,9 +472,17 @@ static inline void convert_slashes(char *path) if (*path == '\\') *path = '/'; } +struct strbuf; +int mingw_is_mount_point(struct strbuf *path); +extern int (*win32_is_mount_point)(struct strbuf *path); +#define is_mount_point win32_is_mount_point +#define CAN_UNLINK_MOUNT_POINTS 1 #define PATH_SEP ';' char *mingw_query_user_email(void); #define query_user_email mingw_query_user_email +struct strbuf; +char *mingw_strbuf_realpath(struct strbuf *resolved, const char *path); +#define platform_strbuf_realpath mingw_strbuf_realpath #if !defined(__MINGW64_VERSION_MAJOR) && (!defined(_MSC_VER) || _MSC_VER < 1800) #define PRIuMAX "I64u" #define PRId64 "I64d" @@ -487,6 +518,42 @@ int is_path_owned_by_current_sid(const char *path, struct strbuf *report); int is_valid_win32_path(const char *path, int allow_literal_nul); #define is_valid_path(path) is_valid_win32_path(path, 0) +/** + * Max length of long paths (exceeding MAX_PATH). The actual maximum supported + * by NTFS is 32,767 (* sizeof(wchar_t)), but we choose an arbitrary smaller + * value to limit required stack memory. + */ +#define MAX_LONG_PATH 4096 + +/** + * Handles paths that would exceed the MAX_PATH limit of Windows Unicode APIs. + * + * With expand == false, the function checks for over-long paths and fails + * with ENAMETOOLONG. The path parameter is not modified, except if cwd + path + * exceeds max_path, but the resulting absolute path doesn't (e.g. due to + * eliminating '..' components). The path parameter must point to a buffer + * of max_path wide characters. + * + * With expand == true, an over-long path is automatically converted in place + * to an absolute path prefixed with '\\?\', and the new length is returned. + * The path parameter must point to a buffer of MAX_LONG_PATH wide characters. + * + * Parameters: + * path: path to check and / or convert + * len: size of path on input (number of wide chars without \0) + * max_path: max short path length to check (usually MAX_PATH = 260, but just + * 248 for CreateDirectoryW) + * expand: false to only check the length, true to expand the path to a + * '\\?\'-prefixed absolute path + * + * Return: + * length of the resulting path, or -1 on failure + * + * Errors: + * ENAMETOOLONG if path is too long + */ +int handle_long_path(wchar_t *path, int len, int max_path, int expand); + /** * Converts UTF-8 encoded string to UTF-16LE. * @@ -545,18 +612,46 @@ static inline int xutftowcs(wchar_t *wcs, const char *utf, size_t wcslen) } /** - * Simplified file system specific variant of xutftowcsn, assumes output - * buffer size is MAX_PATH wide chars and input string is \0-terminated, - * fails with ENAMETOOLONG if input string is too long. + * Simplified file system specific wrapper of xutftowcsn and handle_long_path. + * Converts ERANGE to ENAMETOOLONG. If expand is true, wcs must be at least + * MAX_LONG_PATH wide chars (see handle_long_path). */ -static inline int xutftowcs_path(wchar_t *wcs, const char *utf) +static inline int xutftowcs_path_ex(wchar_t *wcs, const char *utf, + size_t wcslen, int utflen, int max_path, int expand) { - int result = xutftowcsn(wcs, utf, MAX_PATH, -1); + int result = xutftowcsn(wcs, utf, wcslen, utflen); if (result < 0 && errno == ERANGE) errno = ENAMETOOLONG; + if (result >= 0) + result = handle_long_path(wcs, result, max_path, expand); return result; } +/** + * Simplified file system specific variant of xutftowcsn, assumes output + * buffer size is MAX_PATH wide chars and input string is \0-terminated, + * fails with ENAMETOOLONG if input string is too long. Typically used for + * Windows APIs that don't support long paths, e.g. SetCurrentDirectory, + * LoadLibrary, CreateProcess... + */ +static inline int xutftowcs_path(wchar_t *wcs, const char *utf) +{ + return xutftowcs_path_ex(wcs, utf, MAX_PATH, -1, MAX_PATH, 0); +} + +/** + * Simplified file system specific variant of xutftowcsn for Windows APIs + * that support long paths via '\\?\'-prefix, assumes output buffer size is + * MAX_LONG_PATH wide chars, fails with ENAMETOOLONG if input string is too + * long. The 'core.longpaths' git-config option controls whether the path + * is only checked or expanded to a long path. + */ +static inline int xutftowcs_long_path(wchar_t *wcs, const char *utf) +{ + return xutftowcs_path_ex(wcs, utf, MAX_LONG_PATH, -1, MAX_PATH, + are_long_paths_enabled()); +} + /** * Converts UTF-16LE encoded string to UTF-8. * @@ -626,3 +721,8 @@ void open_in_gdb(void); * Used by Pthread API implementation for Windows */ int err_win_to_posix(DWORD winerr); + +/* + * Check current process is inside Windows Container. + */ +int is_inside_windows_container(void); diff --git a/compat/terminal.c b/compat/terminal.c index ea490a7ced431a..70a18f82f619ae 100644 --- a/compat/terminal.c +++ b/compat/terminal.c @@ -418,6 +418,55 @@ static int getchar_with_timeout(int timeout) return getchar(); } +static char *shell_prompt(const char *prompt, int echo) +{ + const char *read_input[] = { + /* Note: call 'bash' explicitly, as 'read -s' is bash-specific */ + "bash", "-c", echo ? + "cat >/dev/tty && read -r line /dev/tty && read -r -s line /dev/tty", + NULL + }; + struct child_process child = CHILD_PROCESS_INIT; + static struct strbuf buffer = STRBUF_INIT; + int prompt_len = strlen(prompt), len = -1, code; + + strvec_pushv(&child.args, read_input); + child.in = -1; + child.out = -1; + child.silent_exec_failure = 1; + + if (start_command(&child)) + return NULL; + + if (write_in_full(child.in, prompt, prompt_len) != prompt_len) { + error("could not write to prompt script"); + close(child.in); + goto ret; + } + close(child.in); + + strbuf_reset(&buffer); + len = strbuf_read(&buffer, child.out, 1024); + if (len < 0) { + error("could not read from prompt script"); + goto ret; + } + + strbuf_strip_suffix(&buffer, "\n"); + strbuf_strip_suffix(&buffer, "\r"); + +ret: + close(child.out); + code = finish_command(&child); + if (code) { + error("failed to execute prompt script (exit code %d)", code); + return NULL; + } + + return len < 0 ? NULL : buffer.buf; +} + #endif #ifndef FORCE_TEXT @@ -430,6 +479,15 @@ char *git_terminal_prompt(const char *prompt, int echo) int r; FILE *input_fh, *output_fh; +#ifdef GIT_WINDOWS_NATIVE + + /* try shell_prompt first, fall back to CONIN/OUT if bash is missing */ + char *result = shell_prompt(prompt, echo); + if (result) + return result; + +#endif + input_fh = fopen(INPUT_PATH, "r" FORCE_TEXT); if (!input_fh) return NULL; diff --git a/compat/vcbuild/README b/compat/vcbuild/README index 29ec1d0f104b80..9ac9760397f479 100644 --- a/compat/vcbuild/README +++ b/compat/vcbuild/README @@ -6,7 +6,11 @@ The Steps to Build Git with VS2015 or VS2017 from the command line. Prompt or from an SDK bash window: $ cd - $ ./compat/vcbuild/vcpkg_install.bat + $ ./compat/vcbuild/vcpkg_install.bat x64-windows + + or + + $ ./compat/vcbuild/vcpkg_install.bat arm64-windows The vcpkg tools and all of the third-party sources will be installed in this folder: @@ -37,27 +41,17 @@ The Steps to Build Git with VS2015 or VS2017 from the command line. ================================================================ -Alternatively, run `make vcxproj` and then load the generated `git.sln` in -Visual Studio. The initial build will install the vcpkg system and build the +Alternatively, just open Git's top-level directory in Visual Studio, via +`File>Open>Folder...`. This will use CMake internally to generate the +project definitions. It will also install the vcpkg system and build the dependencies automatically. This will take a while. -Instead of generating the `git.sln` file yourself (which requires a full Git -for Windows SDK), you may want to consider fetching the `vs/master` branch of -https://github.com/git-for-windows/git instead (which is updated automatically -via CI running `make vcxproj`). The `vs/master` branch does not require a Git -for Windows to build, but you can run the test scripts in a regular Git Bash. - -Note that `make vcxproj` will automatically add and commit the generated `.sln` -and `.vcxproj` files to the repo. This is necessary to allow building a -fully-testable Git in Visual Studio, where a regular Git Bash can be used to -run the test scripts (as opposed to a full Git for Windows SDK): a number of -build targets, such as Git commands implemented as Unix shell scripts (where -`@@SHELL_PATH@@` and other placeholders are interpolated) require a full-blown -Git for Windows SDK (which is about 10x the size of a regular Git for Windows -installation). - -If your plan is to open a Pull Request with Git for Windows, it is a good idea -to drop this commit before submitting. +You can also generate the Visual Studio solution manually by downloading +and running CMake explicitly rather than letting Visual Studio doing +that implicitly. + +Another, deprecated option is to run `make vcxproj`. This option is +superseded by the CMake-based build, and will be removed at some point. ================================================================ The Steps of Build Git with VS2008 diff --git a/compat/vcbuild/find_vs_env.bat b/compat/vcbuild/find_vs_env.bat index b35d264c0e6bed..379b16296e09c2 100644 --- a/compat/vcbuild/find_vs_env.bat +++ b/compat/vcbuild/find_vs_env.bat @@ -99,6 +99,7 @@ REM ================================================================ SET sdk_dir=%WindowsSdkDir% SET sdk_ver=%WindowsSDKVersion% + SET sdk_ver_bin_dir=%WindowsSdkVerBinPath%%tgt% SET si=%sdk_dir%Include\%sdk_ver% SET sdk_includes=-I"%si%ucrt" -I"%si%um" -I"%si%shared" SET sl=%sdk_dir%lib\%sdk_ver% @@ -130,6 +131,7 @@ REM ================================================================ SET sdk_dir=%WindowsSdkDir% SET sdk_ver=%WindowsSDKVersion% + SET sdk_ver_bin_dir=%WindowsSdkVerBinPath%bin\amd64 SET si=%sdk_dir%Include\%sdk_ver% SET sdk_includes=-I"%si%ucrt" -I"%si%um" -I"%si%shared" -I"%si%winrt" SET sl=%sdk_dir%lib\%sdk_ver% @@ -160,6 +162,11 @@ REM ================================================================ echo msvc_includes=%msvc_includes% echo msvc_libs=%msvc_libs% + echo sdk_ver_bin_dir=%sdk_ver_bin_dir% + SET X1=%sdk_ver_bin_dir:C:=/C% + SET X2=%X1:\=/% + echo sdk_ver_bin_dir_msys=%X2% + echo sdk_includes=%sdk_includes% echo sdk_libs=%sdk_libs% diff --git a/compat/vcbuild/scripts/clink.pl b/compat/vcbuild/scripts/clink.pl index 3bd824154be381..677d44e46f98d6 100755 --- a/compat/vcbuild/scripts/clink.pl +++ b/compat/vcbuild/scripts/clink.pl @@ -15,6 +15,7 @@ my @lflags = (); my $is_linking = 0; my $is_debug = 0; +my $is_gui = 0; while (@ARGV) { my $arg = shift @ARGV; if ("$arg" eq "-DDEBUG") { @@ -56,7 +57,8 @@ # need to use that instead? foreach my $flag (@lflags) { if ($flag =~ /^-LIBPATH:(.*)/) { - foreach my $l ("libcurl_imp.lib", "libcurl.lib") { + my $libcurl = $is_debug ? "libcurl-d.lib" : "libcurl.lib"; + foreach my $l ("libcurl_imp.lib", $libcurl) { if (-f "$1/$l") { $lib = $l; last; @@ -66,7 +68,11 @@ } push(@args, $lib); } elsif ("$arg" eq "-lexpat") { + if ($is_debug) { + push(@args, "libexpatd.lib"); + } else { push(@args, "libexpat.lib"); + } } elsif ("$arg" =~ /^-L/ && "$arg" ne "-LTCG") { $arg =~ s/^-L/-LIBPATH:/; push(@lflags, $arg); @@ -118,11 +124,23 @@ push(@cflags, "-wd4996"); } elsif ("$arg" =~ /^-W[a-z]/) { # let's ignore those + } elsif ("$arg" eq "-fno-stack-protector") { + # eat this + } elsif ("$arg" eq "-mwindows") { + $is_gui = 1; } else { push(@args, $arg); } } if ($is_linking) { + if ($is_gui) { + push(@args, "-ENTRY:wWinMainCRTStartup"); + push(@args, "-SUBSYSTEM:WINDOWS"); + } else { + push(@args, "-ENTRY:wmainCRTStartup"); + push(@args, "-SUBSYSTEM:CONSOLE"); + } + push(@args, @lflags); unshift(@args, "link.exe"); } else { diff --git a/compat/vcbuild/scripts/rc.pl b/compat/vcbuild/scripts/rc.pl new file mode 100644 index 00000000000000..7bca4cd81c6c63 --- /dev/null +++ b/compat/vcbuild/scripts/rc.pl @@ -0,0 +1,46 @@ +#!/usr/bin/perl -w +###################################################################### +# Compile Resources on Windows +# +# This is a wrapper to facilitate the compilation of Git with MSVC +# using GNU Make as the build system. So, instead of manipulating the +# Makefile into something nasty, just to support non-space arguments +# etc, we use this wrapper to fix the command line options +# +###################################################################### +use strict; +my @args = (); +my @input = (); + +while (@ARGV) { + my $arg = shift @ARGV; + if ("$arg" =~ /^-[dD]/) { + # GIT_VERSION gets passed with too many + # layers of dquote escaping. + $arg =~ s/\\"/"/g; + + push(@args, $arg); + + } elsif ("$arg" eq "-i") { + my $arg = shift @ARGV; + # TODO complain if NULL or is dashed ?? + push(@input, $arg); + + } elsif ("$arg" eq "-o") { + my $arg = shift @ARGV; + # TODO complain if NULL or is dashed ?? + push(@args, "-fo$arg"); + + } else { + push(@args, $arg); + } +} + +push(@args, "-nologo"); +push(@args, "-v"); +push(@args, @input); + +unshift(@args, "rc.exe"); +printf("**** @args\n"); + +exit (system(@args) != 0); diff --git a/compat/vcbuild/vcpkg_copy_dlls.bat b/compat/vcbuild/vcpkg_copy_dlls.bat index 13661c14f8705c..8bea0cbf83b6cf 100644 --- a/compat/vcbuild/vcpkg_copy_dlls.bat +++ b/compat/vcbuild/vcpkg_copy_dlls.bat @@ -15,7 +15,12 @@ REM ================================================================ @FOR /F "delims=" %%D IN ("%~dp0") DO @SET cwd=%%~fD cd %cwd% - SET arch=x64-windows + SET arch=%2 + IF NOT DEFINED arch ( + echo defaulting to 'x64-windows`. Invoke %0 with 'x86-windows', 'x64-windows', or 'arm64-windows' + set arch=x64-windows + ) + SET inst=%cwd%vcpkg\installed\%arch% IF [%1]==[release] ( diff --git a/compat/vcbuild/vcpkg_install.bat b/compat/vcbuild/vcpkg_install.bat index ebd0bad242a8ca..575c65c20ba307 100644 --- a/compat/vcbuild/vcpkg_install.bat +++ b/compat/vcbuild/vcpkg_install.bat @@ -31,11 +31,24 @@ REM ================================================================ SETLOCAL EnableDelayedExpansion + SET arch=%1 + IF NOT DEFINED arch ( + echo defaulting to 'x64-windows`. Invoke %0 with 'x86-windows', 'x64-windows', or 'arm64-windows' + set arch=x64-windows + ) + @FOR /F "delims=" %%D IN ("%~dp0") DO @SET cwd=%%~fD cd %cwd% dir vcpkg\vcpkg.exe >nul 2>nul && GOTO :install_libraries + git.exe version 2>nul + IF ERRORLEVEL 1 ( + echo "***" + echo "Git not found. Please adjust your CMD path or Git install option." + echo "***" + EXIT /B 1 ) + echo Fetching vcpkg in %cwd%vcpkg git.exe clone https://github.com/Microsoft/vcpkg vcpkg IF ERRORLEVEL 1 ( EXIT /B 1 ) @@ -48,9 +61,8 @@ REM ================================================================ echo Successfully installed %cwd%vcpkg\vcpkg.exe :install_libraries - SET arch=x64-windows - echo Installing third-party libraries... + echo Installing third-party libraries(%arch%)... FOR %%i IN (zlib expat libiconv openssl libssh2 curl) DO ( cd %cwd%vcpkg IF NOT EXIST "packages\%%i_%arch%" CALL :sub__install_one %%i @@ -73,8 +85,47 @@ REM ================================================================ :sub__install_one echo Installing package %1... - .\vcpkg.exe install %1:%arch% + call :%1_features + + REM vcpkg may not be reliable on slow, intermittent or proxy + REM connections, see e.g. + REM https://social.msdn.microsoft.com/Forums/windowsdesktop/en-US/4a8f7be5-5e15-4213-a7bb-ddf424a954e6/winhttpsendrequest-ends-with-12002-errorhttptimeout-after-21-seconds-no-matter-what-timeout?forum=windowssdk + REM which explains the hidden 21 second timeout + REM (last post by Dave : Microsoft - Windows Networking team) + + .\vcpkg.exe install %1%features%:%arch% IF ERRORLEVEL 1 ( EXIT /B 1 ) echo Finished %1 goto :EOF + +:: +:: features for each vcpkg to install +:: there should be an entry here for each package to install +:: 'set features=' means use the default otherwise +:: 'set features=[comma-delimited-feature-set]' is the syntax +:: + +:zlib_features +set features= +goto :EOF + +:expat_features +set features= +goto :EOF + +:libiconv_features +set features= +goto :EOF + +:openssl_features +set features= +goto :EOF + +:libssh2_features +set features= +goto :EOF + +:curl_features +set features=[core,openssl,schannel] +goto :EOF diff --git a/compat/win32.h b/compat/win32.h index a97e880757b6f1..299f01bdf0f5a4 100644 --- a/compat/win32.h +++ b/compat/win32.h @@ -6,17 +6,7 @@ #include #endif -static inline int file_attr_to_st_mode (DWORD attr) -{ - int fMode = S_IREAD; - if (attr & FILE_ATTRIBUTE_DIRECTORY) - fMode |= S_IFDIR; - else - fMode |= S_IFREG; - if (!(attr & FILE_ATTRIBUTE_READONLY)) - fMode |= S_IWRITE; - return fMode; -} +extern int file_attr_to_st_mode (DWORD attr, DWORD tag, const char *path); static inline int get_file_attr(const char *fname, WIN32_FILE_ATTRIBUTE_DATA *fdata) { diff --git a/compat/win32/dirent.c b/compat/win32/dirent.c index 52420ec7d4dad7..87063101f57202 100644 --- a/compat/win32/dirent.c +++ b/compat/win32/dirent.c @@ -1,58 +1,33 @@ #include "../../git-compat-util.h" -struct DIR { - struct dirent dd_dir; /* includes d_type */ +#pragma GCC diagnostic push +#pragma GCC diagnostic ignored "-Wpedantic" +typedef struct dirent_DIR { + struct DIR base_dir; /* extend base struct DIR */ HANDLE dd_handle; /* FindFirstFile handle */ int dd_stat; /* 0-based index */ -}; + struct dirent dd_dir; /* includes d_type */ +} dirent_DIR; +#pragma GCC diagnostic pop + +DIR *(*opendir)(const char *dirname) = dirent_opendir; static inline void finddata2dirent(struct dirent *ent, WIN32_FIND_DATAW *fdata) { - /* convert UTF-16 name to UTF-8 */ - xwcstoutf(ent->d_name, fdata->cFileName, sizeof(ent->d_name)); + /* convert UTF-16 name to UTF-8 (d_name points to dirent_DIR.dd_name) */ + xwcstoutf(ent->d_name, fdata->cFileName, MAX_PATH * 3); /* Set file type, based on WIN32_FIND_DATA */ - if (fdata->dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY) + if ((fdata->dwFileAttributes & FILE_ATTRIBUTE_REPARSE_POINT) + && fdata->dwReserved0 == IO_REPARSE_TAG_SYMLINK) + ent->d_type = DT_LNK; + else if (fdata->dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY) ent->d_type = DT_DIR; else ent->d_type = DT_REG; } -DIR *opendir(const char *name) -{ - wchar_t pattern[MAX_PATH + 2]; /* + 2 for '/' '*' */ - WIN32_FIND_DATAW fdata; - HANDLE h; - int len; - DIR *dir; - - /* convert name to UTF-16 and check length < MAX_PATH */ - if ((len = xutftowcs_path(pattern, name)) < 0) - return NULL; - - /* append optional '/' and wildcard '*' */ - if (len && !is_dir_sep(pattern[len - 1])) - pattern[len++] = '/'; - pattern[len++] = '*'; - pattern[len] = 0; - - /* open find handle */ - h = FindFirstFileW(pattern, &fdata); - if (h == INVALID_HANDLE_VALUE) { - DWORD err = GetLastError(); - errno = (err == ERROR_DIRECTORY) ? ENOTDIR : err_win_to_posix(err); - return NULL; - } - - /* initialize DIR structure and copy first dir entry */ - dir = xmalloc(sizeof(DIR)); - dir->dd_handle = h; - dir->dd_stat = 0; - finddata2dirent(&dir->dd_dir, &fdata); - return dir; -} - -struct dirent *readdir(DIR *dir) +static struct dirent *dirent_readdir(dirent_DIR *dir) { if (!dir) { errno = EBADF; /* No set_errno for mingw */ @@ -79,7 +54,7 @@ struct dirent *readdir(DIR *dir) return &dir->dd_dir; } -int closedir(DIR *dir) +static int dirent_closedir(dirent_DIR *dir) { if (!dir) { errno = EBADF; @@ -90,3 +65,44 @@ int closedir(DIR *dir) free(dir); return 0; } + +DIR *dirent_opendir(const char *name) +{ + wchar_t pattern[MAX_LONG_PATH + 2]; /* + 2 for "\*" */ + WIN32_FIND_DATAW fdata; + HANDLE h; + int len; + dirent_DIR *dir; + + /* convert name to UTF-16 and check length */ + if ((len = xutftowcs_path_ex(pattern, name, MAX_LONG_PATH, -1, + MAX_PATH - 2, + are_long_paths_enabled())) < 0) + return NULL; + + /* + * append optional '\' and wildcard '*'. Note: we need to use '\' as + * Windows doesn't translate '/' to '\' for "\\?\"-prefixed paths. + */ + if (len && !is_dir_sep(pattern[len - 1])) + pattern[len++] = '\\'; + pattern[len++] = '*'; + pattern[len] = 0; + + /* open find handle */ + h = FindFirstFileW(pattern, &fdata); + if (h == INVALID_HANDLE_VALUE) { + DWORD err = GetLastError(); + errno = (err == ERROR_DIRECTORY) ? ENOTDIR : err_win_to_posix(err); + return NULL; + } + + /* initialize DIR structure and copy first dir entry */ + dir = xmalloc(sizeof(dirent_DIR) + MAX_LONG_PATH); + dir->base_dir.preaddir = (struct dirent *(*)(DIR *dir)) dirent_readdir; + dir->base_dir.pclosedir = (int (*)(DIR *dir)) dirent_closedir; + dir->dd_handle = h; + dir->dd_stat = 0; + finddata2dirent(&dir->dd_dir, &fdata); + return (DIR*) dir; +} diff --git a/compat/win32/dirent.h b/compat/win32/dirent.h index 058207e4bfed62..e0e0e1700f64d1 100644 --- a/compat/win32/dirent.h +++ b/compat/win32/dirent.h @@ -1,20 +1,32 @@ #ifndef DIRENT_H #define DIRENT_H -typedef struct DIR DIR; - #define DT_UNKNOWN 0 #define DT_DIR 1 #define DT_REG 2 #define DT_LNK 3 struct dirent { - unsigned char d_type; /* file type to prevent lstat after readdir */ - char d_name[MAX_PATH * 3]; /* file name (* 3 for UTF-8 conversion) */ + unsigned char d_type; /* file type to prevent lstat after readdir */ + char d_name[FLEX_ARRAY]; /* file name */ }; -DIR *opendir(const char *dirname); -struct dirent *readdir(DIR *dir); -int closedir(DIR *dir); +/* + * Base DIR structure, contains pointers to readdir/closedir implementations so + * that opendir may choose a concrete implementation on a call-by-call basis. + */ +typedef struct DIR { + struct dirent *(*preaddir)(struct DIR *dir); + int (*pclosedir)(struct DIR *dir); +} DIR; + +/* default dirent implementation */ +extern DIR *dirent_opendir(const char *dirname); + +/* current dirent implementation */ +extern DIR *(*opendir)(const char *dirname); + +#define readdir(dir) (dir->preaddir(dir)) +#define closedir(dir) (dir->pclosedir(dir)) #endif /* DIRENT_H */ diff --git a/compat/win32/exit-process.h b/compat/win32/exit-process.h new file mode 100644 index 00000000000000..d53989884cfb0c --- /dev/null +++ b/compat/win32/exit-process.h @@ -0,0 +1,165 @@ +#ifndef EXIT_PROCESS_H +#define EXIT_PROCESS_H + +/* + * This file contains functions to terminate a Win32 process, as gently as + * possible. + * + * At first, we will attempt to inject a thread that calls ExitProcess(). If + * that fails, we will fall back to terminating the entire process tree. + * + * For simplicity, these functions are marked as file-local. + */ + +#include + +/* + * Terminates the process corresponding to the process ID and all of its + * directly and indirectly spawned subprocesses. + * + * This way of terminating the processes is not gentle: the processes get + * no chance of cleaning up after themselves (closing file handles, removing + * .lock files, terminating spawned processes (if any), etc). + */ +static int terminate_process_tree(HANDLE main_process, int exit_status) +{ + HANDLE snapshot = CreateToolhelp32Snapshot(TH32CS_SNAPPROCESS, 0); + PROCESSENTRY32 entry; + DWORD pids[16384]; + int max_len = sizeof(pids) / sizeof(*pids), i, len, ret = 0; + pid_t pid = GetProcessId(main_process); + + pids[0] = (DWORD)pid; + len = 1; + + /* + * Even if Process32First()/Process32Next() seem to traverse the + * processes in topological order (i.e. parent processes before + * child processes), there is nothing in the Win32 API documentation + * suggesting that this is guaranteed. + * + * Therefore, run through them at least twice and stop when no more + * process IDs were added to the list. + */ + for (;;) { + int orig_len = len; + + memset(&entry, 0, sizeof(entry)); + entry.dwSize = sizeof(entry); + + if (!Process32First(snapshot, &entry)) + break; + + do { + for (i = len - 1; i >= 0; i--) { + if (pids[i] == entry.th32ProcessID) + break; + if (pids[i] == entry.th32ParentProcessID) + pids[len++] = entry.th32ProcessID; + } + } while (len < max_len && Process32Next(snapshot, &entry)); + + if (orig_len == len || len >= max_len) + break; + } + + for (i = len - 1; i > 0; i--) { + HANDLE process = OpenProcess(PROCESS_TERMINATE, FALSE, pids[i]); + + if (process) { + if (!TerminateProcess(process, exit_status)) + ret = -1; + CloseHandle(process); + } + } + if (!TerminateProcess(main_process, exit_status)) + ret = -1; + CloseHandle(main_process); + + return ret; +} + +/** + * Determine whether a process runs in the same architecture as the current + * one. That test is required before we assume that GetProcAddress() returns + * a valid address *for the target process*. + */ +static inline int process_architecture_matches_current(HANDLE process) +{ + static BOOL current_is_wow = -1; + BOOL is_wow; + + if (current_is_wow == -1 && + !IsWow64Process (GetCurrentProcess(), ¤t_is_wow)) + current_is_wow = -2; + if (current_is_wow == -2) + return 0; /* could not determine current process' WoW-ness */ + if (!IsWow64Process (process, &is_wow)) + return 0; /* cannot determine */ + return is_wow == current_is_wow; +} + +/** + * Inject a thread into the given process that runs ExitProcess(). + * + * Note: as kernel32.dll is loaded before any process, the other process and + * this process will have ExitProcess() at the same address. + * + * This function expects the process handle to have the access rights for + * CreateRemoteThread(): PROCESS_CREATE_THREAD, PROCESS_QUERY_INFORMATION, + * PROCESS_VM_OPERATION, PROCESS_VM_WRITE, and PROCESS_VM_READ. + * + * The idea comes from the Dr Dobb's article "A Safer Alternative to + * TerminateProcess()" by Andrew Tucker (July 1, 1999), + * http://www.drdobbs.com/a-safer-alternative-to-terminateprocess/184416547 + * + * If this method fails, we fall back to running terminate_process_tree(). + */ +static int exit_process(HANDLE process, int exit_code) +{ + DWORD code; + + if (GetExitCodeProcess(process, &code) && code == STILL_ACTIVE) { + static int initialized; + static LPTHREAD_START_ROUTINE exit_process_address; + PVOID arg = (PVOID)(intptr_t)exit_code; + DWORD thread_id; + HANDLE thread = NULL; + + if (!initialized) { + HINSTANCE kernel32 = GetModuleHandleA("kernel32"); + if (!kernel32) + die("BUG: cannot find kernel32"); + exit_process_address = + (LPTHREAD_START_ROUTINE)(void (*)(void)) + GetProcAddress(kernel32, "ExitProcess"); + initialized = 1; + } + if (!exit_process_address || + !process_architecture_matches_current(process)) + return terminate_process_tree(process, exit_code); + + thread = CreateRemoteThread(process, NULL, 0, + exit_process_address, + arg, 0, &thread_id); + if (thread) { + CloseHandle(thread); + /* + * If the process survives for 10 seconds (a completely + * arbitrary value picked from thin air), fall back to + * killing the process tree via TerminateProcess(). + */ + if (WaitForSingleObject(process, 10000) == + WAIT_OBJECT_0) { + CloseHandle(process); + return 0; + } + } + + return terminate_process_tree(process, exit_code); + } + + return 0; +} + +#endif diff --git a/compat/win32/fscache.c b/compat/win32/fscache.c new file mode 100644 index 00000000000000..5edbb6150a8b0f --- /dev/null +++ b/compat/win32/fscache.c @@ -0,0 +1,787 @@ +#include "../../cache.h" +#include "../../hashmap.h" +#include "../win32.h" +#include "fscache.h" +#include "../../dir.h" +#include "config.h" +#include "../../mem-pool.h" +#include "ntifs.h" + +static volatile long initialized; +static DWORD dwTlsIndex; +CRITICAL_SECTION fscache_cs; + +/* + * Store one fscache per thread to avoid thread contention and locking. + * This is ok because multi-threaded access is 1) uncommon and 2) always + * splitting up the cache entries across multiple threads so there isn't + * any overlap between threads anyway. + */ +struct fscache { + volatile long enabled; + struct hashmap map; + struct mem_pool mem_pool; + unsigned int lstat_requests; + unsigned int opendir_requests; + unsigned int fscache_requests; + unsigned int fscache_misses; + /* + * 32k wide characters translates to 64kB, which is the maximum that + * Windows 8.1 and earlier can handle. On network drives, not only + * the client's Windows version matters, but also the server's, + * therefore we need to keep this to 64kB. + */ + WCHAR buffer[32 * 1024]; +}; +static struct trace_key trace_fscache = TRACE_KEY_INIT(FSCACHE); + +/* + * An entry in the file system cache. Used for both entire directory listings + * and file entries. + */ +#pragma GCC diagnostic push +#pragma GCC diagnostic ignored "-Wpedantic" +struct fsentry { + struct hashmap_entry ent; + mode_t st_mode; + ULONG reparse_tag; + /* Pointer to the directory listing, or NULL for the listing itself. */ + struct fsentry *list; + /* Pointer to the next file entry of the list. */ + struct fsentry *next; + + union { + /* Reference count of the directory listing. */ + volatile long refcnt; + struct { + /* More stat members (only used for file entries). */ + off64_t st_size; + struct timespec st_atim; + struct timespec st_mtim; + struct timespec st_ctim; + } s; + } u; + + /* Length of name. */ + unsigned short len; + /* + * Name of the entry. For directory listings: relative path of the + * directory, without trailing '/' (empty for cwd()). For file entries: + * name of the file. Typically points to the end of the structure if + * the fsentry is allocated on the heap (see fsentry_alloc), or to a + * local variable if on the stack (see fsentry_init). + */ + struct dirent dirent; +}; +#pragma GCC diagnostic pop + +struct heap_fsentry { + union { + struct fsentry ent; + char dummy[sizeof(struct fsentry) + MAX_LONG_PATH]; + } u; +}; + +/* + * Compares the paths of two fsentry structures for equality. + */ +static int fsentry_cmp(void *unused_cmp_data, + const struct fsentry *fse1, const struct fsentry *fse2, + void *unused_keydata) +{ + int res; + if (fse1 == fse2) + return 0; + + /* compare the list parts first */ + if (fse1->list != fse2->list && + (res = fsentry_cmp(NULL, fse1->list ? fse1->list : fse1, + fse2->list ? fse2->list : fse2, NULL))) + return res; + + /* if list parts are equal, compare len and name */ + if (fse1->len != fse2->len) + return fse1->len - fse2->len; + return fspathncmp(fse1->dirent.d_name, fse2->dirent.d_name, fse1->len); +} + +/* + * Calculates the hash code of an fsentry structure's path. + */ +static unsigned int fsentry_hash(const struct fsentry *fse) +{ + unsigned int hash = fse->list ? fse->list->ent.hash : 0; + return hash ^ memihash(fse->dirent.d_name, fse->len); +} + +/* + * Initialize an fsentry structure for use by fsentry_hash and fsentry_cmp. + */ +static void fsentry_init(struct fsentry *fse, struct fsentry *list, + const char *name, size_t len) +{ + fse->list = list; + if (len > MAX_LONG_PATH) + BUG("Trying to allocate fsentry for long path '%.*s'", + (int)len, name); + memcpy(fse->dirent.d_name, name, len); + fse->dirent.d_name[len] = 0; + fse->len = len; + hashmap_entry_init(&fse->ent, fsentry_hash(fse)); +} + +/* + * Allocate an fsentry structure on the heap. + */ +static struct fsentry *fsentry_alloc(struct fscache *cache, struct fsentry *list, const char *name, + size_t len) +{ + /* overallocate fsentry and copy the name to the end */ + struct fsentry *fse = + mem_pool_alloc(&cache->mem_pool, sizeof(*fse) + len + 1); + /* init the rest of the structure */ + fsentry_init(fse, list, name, len); + fse->next = NULL; + fse->u.refcnt = 1; + return fse; +} + +/* + * Add a reference to an fsentry. + */ +inline static void fsentry_addref(struct fsentry *fse) +{ + if (fse->list) + fse = fse->list; + + InterlockedIncrement(&(fse->u.refcnt)); +} + +/* + * Release the reference to an fsentry. + */ +static void fsentry_release(struct fsentry *fse) +{ + if (fse->list) + fse = fse->list; + + InterlockedDecrement(&(fse->u.refcnt)); +} + +static int xwcstoutfn(char *utf, int utflen, const wchar_t *wcs, int wcslen) +{ + if (!wcs || !utf || utflen < 1) { + errno = EINVAL; + return -1; + } + utflen = WideCharToMultiByte(CP_UTF8, 0, wcs, wcslen, utf, utflen, NULL, NULL); + if (utflen) + return utflen; + errno = ERANGE; + return -1; +} + +/* + * Allocate and initialize an fsentry from a FILE_FULL_DIR_INFORMATION structure. + */ +static struct fsentry *fseentry_create_entry(struct fscache *cache, + struct fsentry *list, + PFILE_FULL_DIR_INFORMATION fdata) +{ + char buf[MAX_PATH * 3]; + int len; + struct fsentry *fse; + + len = xwcstoutfn(buf, ARRAY_SIZE(buf), fdata->FileName, fdata->FileNameLength / sizeof(wchar_t)); + + fse = fsentry_alloc(cache, list, buf, len); + + fse->reparse_tag = + fdata->FileAttributes & FILE_ATTRIBUTE_REPARSE_POINT ? + fdata->EaSize : 0; + + /* + * On certain Windows versions, host directories mapped into + * Windows Containers ("Volumes", see https://docs.docker.com/storage/volumes/) + * look like symbolic links, but their targets are paths that + * are valid only in kernel mode. + * + * Let's work around this by detecting that situation and + * telling Git that these are *not* symbolic links. + */ + if (fse->reparse_tag == IO_REPARSE_TAG_SYMLINK && + sizeof(buf) > (list ? list->len + 1 : 0) + fse->len + 1 && + is_inside_windows_container()) { + size_t off = 0; + if (list) { + memcpy(buf, list->dirent.d_name, list->len); + buf[list->len] = '/'; + off = list->len + 1; + } + memcpy(buf + off, fse->dirent.d_name, fse->len); + buf[off + fse->len] = '\0'; + } + + fse->st_mode = file_attr_to_st_mode(fdata->FileAttributes, + fdata->EaSize, buf); + fse->dirent.d_type = S_ISREG(fse->st_mode) ? DT_REG : + S_ISDIR(fse->st_mode) ? DT_DIR : DT_LNK; + fse->u.s.st_size = S_ISLNK(fse->st_mode) ? MAX_LONG_PATH : + fdata->EndOfFile.LowPart | + (((off_t)fdata->EndOfFile.HighPart) << 32); + filetime_to_timespec((FILETIME *)&(fdata->LastAccessTime), + &(fse->u.s.st_atim)); + filetime_to_timespec((FILETIME *)&(fdata->LastWriteTime), + &(fse->u.s.st_mtim)); + filetime_to_timespec((FILETIME *)&(fdata->CreationTime), + &(fse->u.s.st_ctim)); + + return fse; +} + +/* + * Create an fsentry-based directory listing (similar to opendir / readdir). + * Dir should not contain trailing '/'. Use an empty string for the current + * directory (not "."!). + */ +static struct fsentry *fsentry_create_list(struct fscache *cache, const struct fsentry *dir, + int *dir_not_found) +{ + wchar_t pattern[MAX_LONG_PATH]; + NTSTATUS status; + IO_STATUS_BLOCK iosb; + PFILE_FULL_DIR_INFORMATION di; + HANDLE h; + int wlen; + struct fsentry *list, **phead; + DWORD err; + + *dir_not_found = 0; + + /* convert name to UTF-16 and check length */ + if ((wlen = xutftowcs_path_ex(pattern, dir->dirent.d_name, + MAX_LONG_PATH, dir->len, MAX_PATH - 2, + are_long_paths_enabled())) < 0) + return NULL; + + /* handle CWD */ + if (!wlen) { + wlen = GetCurrentDirectoryW(ARRAY_SIZE(pattern), pattern); + if (!wlen || wlen >= ARRAY_SIZE(pattern)) { + errno = wlen ? ENAMETOOLONG : err_win_to_posix(GetLastError()); + return NULL; + } + } + + h = CreateFileW(pattern, FILE_LIST_DIRECTORY, + FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE, + NULL, OPEN_EXISTING, FILE_FLAG_BACKUP_SEMANTICS, NULL); + if (h == INVALID_HANDLE_VALUE) { + err = GetLastError(); + *dir_not_found = 1; /* or empty directory */ + errno = (err == ERROR_DIRECTORY) ? ENOTDIR : err_win_to_posix(err); + trace_printf_key(&trace_fscache, "fscache: error(%d) '%s'\n", + errno, dir->dirent.d_name); + return NULL; + } + + /* allocate object to hold directory listing */ + list = fsentry_alloc(cache, NULL, dir->dirent.d_name, dir->len); + list->st_mode = S_IFDIR; + list->dirent.d_type = DT_DIR; + + /* walk directory and build linked list of fsentry structures */ + phead = &list->next; + status = NtQueryDirectoryFile(h, NULL, 0, 0, &iosb, cache->buffer, + sizeof(cache->buffer), FileFullDirectoryInformation, FALSE, NULL, FALSE); + if (!NT_SUCCESS(status)) { + /* + * NtQueryDirectoryFile returns STATUS_INVALID_PARAMETER when + * asked to enumerate an invalid directory (ie it is a file + * instead of a directory). Verify that is the actual cause + * of the error. + */ + if (status == STATUS_INVALID_PARAMETER) { + DWORD attributes = GetFileAttributesW(pattern); + if (!(attributes & FILE_ATTRIBUTE_DIRECTORY)) + status = ERROR_DIRECTORY; + } + goto Error; + } + di = (PFILE_FULL_DIR_INFORMATION)(cache->buffer); + for (;;) { + + *phead = fseentry_create_entry(cache, list, di); + phead = &(*phead)->next; + + /* If there is no offset in the entry, the buffer has been exhausted. */ + if (di->NextEntryOffset == 0) { + status = NtQueryDirectoryFile(h, NULL, 0, 0, &iosb, cache->buffer, + sizeof(cache->buffer), FileFullDirectoryInformation, FALSE, NULL, FALSE); + if (!NT_SUCCESS(status)) { + if (status == STATUS_NO_MORE_FILES) + break; + goto Error; + } + + di = (PFILE_FULL_DIR_INFORMATION)(cache->buffer); + continue; + } + + /* Advance to the next entry. */ + di = (PFILE_FULL_DIR_INFORMATION)(((PUCHAR)di) + di->NextEntryOffset); + } + + CloseHandle(h); + return list; + +Error: + trace_printf_key(&trace_fscache, + "fscache: status(%ld) unable to query directory " + "contents '%s'\n", status, dir->dirent.d_name); + CloseHandle(h); + fsentry_release(list); + return NULL; +} + +/* + * Adds a directory listing to the cache. + */ +static void fscache_add(struct fscache *cache, struct fsentry *fse) +{ + if (fse->list) + fse = fse->list; + + for (; fse; fse = fse->next) + hashmap_add(&cache->map, &fse->ent); +} + +/* + * Clears the cache. + */ +static void fscache_clear(struct fscache *cache) +{ + mem_pool_discard(&cache->mem_pool, 0); + mem_pool_init(&cache->mem_pool, 0); + hashmap_clear(&cache->map); + hashmap_init(&cache->map, (hashmap_cmp_fn)fsentry_cmp, NULL, 0); + cache->lstat_requests = cache->opendir_requests = 0; + cache->fscache_misses = cache->fscache_requests = 0; +} + +/* + * Checks if the cache is enabled for the given path. + */ +static int do_fscache_enabled(struct fscache *cache, const char *path) +{ + return cache->enabled > 0 && !is_absolute_path(path); +} + +int fscache_enabled(const char *path) +{ + struct fscache *cache = fscache_getcache(); + + return cache ? do_fscache_enabled(cache, path) : 0; +} + +/* + * Looks up or creates a cache entry for the specified key. + */ +static struct fsentry *fscache_get(struct fscache *cache, struct fsentry *key) +{ + struct fsentry *fse; + int dir_not_found; + + cache->fscache_requests++; + /* check if entry is in cache */ + fse = hashmap_get_entry(&cache->map, key, ent, NULL); + if (fse) { + if (fse->st_mode) + fsentry_addref(fse); + else + fse = NULL; /* non-existing directory */ + return fse; + } + /* if looking for a file, check if directory listing is in cache */ + if (!fse && key->list) { + fse = hashmap_get_entry(&cache->map, key->list, ent, NULL); + if (fse) { + /* + * dir entry without file entry, or dir does not + * exist -> file doesn't exist + */ + errno = ENOENT; + return NULL; + } + } + + /* create the directory listing */ + fse = fsentry_create_list(cache, key->list ? key->list : key, &dir_not_found); + + /* leave on error (errno set by fsentry_create_list) */ + if (!fse) { + if (dir_not_found && key->list) { + /* + * Record that the directory does not exist (or is + * empty, which for all practical matters is the same + * thing as far as fscache is concerned). + */ + fse = fsentry_alloc(cache, key->list->list, + key->list->dirent.d_name, + key->list->len); + fse->st_mode = 0; + hashmap_add(&cache->map, &fse->ent); + } + return NULL; + } + + /* add directory listing to the cache */ + cache->fscache_misses++; + fscache_add(cache, fse); + + /* lookup file entry if requested (fse already points to directory) */ + if (key->list) + fse = hashmap_get_entry(&cache->map, key, ent, NULL); + + if (fse && !fse->st_mode) + fse = NULL; /* non-existing directory */ + + /* return entry or ENOENT */ + if (fse) + fsentry_addref(fse); + else + errno = ENOENT; + + return fse; +} + +/* + * Enables the cache. Note that the cache is read-only, changes to + * the working directory are NOT reflected in the cache while enabled. + */ +int fscache_enable(size_t initial_size) +{ + int fscache; + struct fscache *cache; + int result = 0; + + /* allow the cache to be disabled entirely */ + fscache = git_env_bool("GIT_TEST_FSCACHE", -1); + if (fscache != -1) + core_fscache = fscache; + if (!core_fscache) + return 0; + + /* + * refcount the global fscache initialization so that the + * opendir and lstat function pointers are redirected if + * any threads are using the fscache. + */ + EnterCriticalSection(&fscache_cs); + if (!initialized) { + if (!dwTlsIndex) { + dwTlsIndex = TlsAlloc(); + if (dwTlsIndex == TLS_OUT_OF_INDEXES) { + LeaveCriticalSection(&fscache_cs); + return 0; + } + } + + /* redirect opendir and lstat to the fscache implementations */ + opendir = fscache_opendir; + lstat = fscache_lstat; + win32_is_mount_point = fscache_is_mount_point; + } + initialized++; + LeaveCriticalSection(&fscache_cs); + + /* refcount the thread specific initialization */ + cache = fscache_getcache(); + if (cache) { + cache->enabled++; + } else { + cache = (struct fscache *)xcalloc(1, sizeof(*cache)); + cache->enabled = 1; + /* + * avoid having to rehash by leaving room for the parent dirs. + * '4' was determined empirically by testing several repos + */ + hashmap_init(&cache->map, (hashmap_cmp_fn)fsentry_cmp, NULL, initial_size * 4); + mem_pool_init(&cache->mem_pool, 0); + if (!TlsSetValue(dwTlsIndex, cache)) + BUG("TlsSetValue error"); + } + + trace_printf_key(&trace_fscache, "fscache: enable\n"); + return result; +} + +/* + * Disables the cache. + */ +void fscache_disable(void) +{ + struct fscache *cache; + + if (!core_fscache) + return; + + /* update the thread specific fscache initialization */ + cache = fscache_getcache(); + if (!cache) + BUG("fscache_disable() called on a thread where fscache has not been initialized"); + if (!cache->enabled) + BUG("fscache_disable() called on an fscache that is already disabled"); + cache->enabled--; + if (!cache->enabled) { + TlsSetValue(dwTlsIndex, NULL); + trace_printf_key(&trace_fscache, "fscache_disable: lstat %u, opendir %u, " + "total requests/misses %u/%u\n", + cache->lstat_requests, cache->opendir_requests, + cache->fscache_requests, cache->fscache_misses); + mem_pool_discard(&cache->mem_pool, 0); + hashmap_clear(&cache->map); + free(cache); + } + + /* update the global fscache initialization */ + EnterCriticalSection(&fscache_cs); + initialized--; + if (!initialized) { + /* reset opendir and lstat to the original implementations */ + opendir = dirent_opendir; + lstat = mingw_lstat; + win32_is_mount_point = mingw_is_mount_point; + } + LeaveCriticalSection(&fscache_cs); + + trace_printf_key(&trace_fscache, "fscache: disable\n"); + return; +} + +/* + * Flush cached stats result when fscache is enabled. + */ +void fscache_flush(void) +{ + struct fscache *cache = fscache_getcache(); + + if (cache && cache->enabled) { + fscache_clear(cache); + } +} + +/* + * Lstat replacement, uses the cache if enabled, otherwise redirects to + * mingw_lstat. + */ +int fscache_lstat(const char *filename, struct stat *st) +{ + int dirlen, base, len; + struct heap_fsentry key[2]; + struct fsentry *fse; + struct fscache *cache = fscache_getcache(); + + if (!cache || !do_fscache_enabled(cache, filename)) + return mingw_lstat(filename, st); + + cache->lstat_requests++; + /* split filename into path + name */ + len = strlen(filename); + if (len && is_dir_sep(filename[len - 1])) + len--; + base = len; + while (base && !is_dir_sep(filename[base - 1])) + base--; + dirlen = base ? base - 1 : 0; + + /* lookup entry for path + name in cache */ + fsentry_init(&key[0].u.ent, NULL, filename, dirlen); + fsentry_init(&key[1].u.ent, &key[0].u.ent, filename + base, len - base); + fse = fscache_get(cache, &key[1].u.ent); + if (!fse) { + errno = ENOENT; + return -1; + } + + /* + * Special case symbolic links: FindFirstFile()/FindNextFile() did not + * provide us with the length of the target path. + */ + if (fse->u.s.st_size == MAX_LONG_PATH && S_ISLNK(fse->st_mode)) { + char buf[MAX_LONG_PATH]; + int len = readlink(filename, buf, sizeof(buf) - 1); + + if (len > 0) + fse->u.s.st_size = len; + } + + /* copy stat data */ + st->st_ino = 0; + st->st_gid = 0; + st->st_uid = 0; + st->st_dev = 0; + st->st_rdev = 0; + st->st_nlink = 1; + st->st_mode = fse->st_mode; + st->st_size = fse->u.s.st_size; + st->st_atim = fse->u.s.st_atim; + st->st_mtim = fse->u.s.st_mtim; + st->st_ctim = fse->u.s.st_ctim; + + /* don't forget to release fsentry */ + fsentry_release(fse); + return 0; +} + +/* + * is_mount_point() replacement, uses cache if enabled, otherwise falls + * back to mingw_is_mount_point(). + */ +int fscache_is_mount_point(struct strbuf *path) +{ + int dirlen, base, len; + struct heap_fsentry key[2]; + struct fsentry *fse; + struct fscache *cache = fscache_getcache(); + + if (!cache || !do_fscache_enabled(cache, path->buf)) + return mingw_is_mount_point(path); + + cache->lstat_requests++; + /* split path into path + name */ + len = path->len; + if (len && is_dir_sep(path->buf[len - 1])) + len--; + base = len; + while (base && !is_dir_sep(path->buf[base - 1])) + base--; + dirlen = base ? base - 1 : 0; + + /* lookup entry for path + name in cache */ + fsentry_init(&key[0].u.ent, NULL, path->buf, dirlen); + fsentry_init(&key[1].u.ent, &key[0].u.ent, path->buf + base, len - base); + fse = fscache_get(cache, &key[1].u.ent); + if (!fse) + return mingw_is_mount_point(path); + return fse->reparse_tag == IO_REPARSE_TAG_MOUNT_POINT; +} + +typedef struct fscache_DIR { + struct DIR base_dir; /* extend base struct DIR */ + struct fsentry *pfsentry; + struct dirent *dirent; +} fscache_DIR; + +/* + * Readdir replacement. + */ +static struct dirent *fscache_readdir(DIR *base_dir) +{ + fscache_DIR *dir = (fscache_DIR*) base_dir; + struct fsentry *next = dir->pfsentry->next; + if (!next) + return NULL; + dir->pfsentry = next; + dir->dirent = &next->dirent; + return dir->dirent; +} + +/* + * Closedir replacement. + */ +static int fscache_closedir(DIR *base_dir) +{ + fscache_DIR *dir = (fscache_DIR*) base_dir; + fsentry_release(dir->pfsentry); + free(dir); + return 0; +} + +/* + * Opendir replacement, uses a directory listing from the cache if enabled, + * otherwise calls original dirent implementation. + */ +DIR *fscache_opendir(const char *dirname) +{ + struct heap_fsentry key; + struct fsentry *list; + fscache_DIR *dir; + int len; + struct fscache *cache = fscache_getcache(); + + if (!cache || !do_fscache_enabled(cache, dirname)) + return dirent_opendir(dirname); + + cache->opendir_requests++; + /* prepare name (strip trailing '/', replace '.') */ + len = strlen(dirname); + if ((len == 1 && dirname[0] == '.') || + (len && is_dir_sep(dirname[len - 1]))) + len--; + + /* get directory listing from cache */ + fsentry_init(&key.u.ent, NULL, dirname, len); + list = fscache_get(cache, &key.u.ent); + if (!list) + return NULL; + + /* alloc and return DIR structure */ + dir = (fscache_DIR*) xmalloc(sizeof(fscache_DIR)); + dir->base_dir.preaddir = fscache_readdir; + dir->base_dir.pclosedir = fscache_closedir; + dir->pfsentry = list; + return (DIR*) dir; +} + +struct fscache *fscache_getcache(void) +{ + return (struct fscache *)TlsGetValue(dwTlsIndex); +} + +void fscache_merge(struct fscache *dest) +{ + struct hashmap_iter iter; + struct hashmap_entry *e; + struct fscache *cache = fscache_getcache(); + + /* + * Only do the merge if fscache was enabled and we have a dest + * cache to merge into. + */ + if (!dest) { + fscache_enable(0); + return; + } + if (!cache) + BUG("fscache_merge() called on a thread where fscache has not been initialized"); + + TlsSetValue(dwTlsIndex, NULL); + trace_printf_key(&trace_fscache, "fscache_merge: lstat %u, opendir %u, " + "total requests/misses %u/%u\n", + cache->lstat_requests, cache->opendir_requests, + cache->fscache_requests, cache->fscache_misses); + + /* + * This is only safe because the primary thread we're merging into + * isn't being used so the critical section only needs to prevent + * the the child threads from stomping on each other. + */ + EnterCriticalSection(&fscache_cs); + + hashmap_iter_init(&cache->map, &iter); + while ((e = hashmap_iter_next(&iter))) + hashmap_add(&dest->map, e); + + mem_pool_combine(&dest->mem_pool, &cache->mem_pool); + + dest->lstat_requests += cache->lstat_requests; + dest->opendir_requests += cache->opendir_requests; + dest->fscache_requests += cache->fscache_requests; + dest->fscache_misses += cache->fscache_misses; + initialized--; + LeaveCriticalSection(&fscache_cs); + + free(cache); + +} diff --git a/compat/win32/fscache.h b/compat/win32/fscache.h new file mode 100644 index 00000000000000..386c770a85d321 --- /dev/null +++ b/compat/win32/fscache.h @@ -0,0 +1,36 @@ +#ifndef FSCACHE_H +#define FSCACHE_H + +/* + * The fscache is thread specific. enable_fscache() must be called + * for each thread where caching is desired. + */ + +extern CRITICAL_SECTION fscache_cs; + +int fscache_enable(size_t initial_size); +#define enable_fscache(initial_size) fscache_enable(initial_size) + +void fscache_disable(void); +#define disable_fscache() fscache_disable() + +int fscache_enabled(const char *path); +#define is_fscache_enabled(path) fscache_enabled(path) + +void fscache_flush(void); +#define flush_fscache() fscache_flush() + +DIR *fscache_opendir(const char *dir); +int fscache_lstat(const char *file_name, struct stat *buf); +int fscache_is_mount_point(struct strbuf *path); + +/* opaque fscache structure */ +struct fscache; + +struct fscache *fscache_getcache(void); +#define getcache_fscache() fscache_getcache() + +void fscache_merge(struct fscache *dest); +#define merge_fscache(dest) fscache_merge(dest) + +#endif diff --git a/compat/win32/headless.c b/compat/win32/headless.c new file mode 100644 index 00000000000000..8b00dfe3bd5d00 --- /dev/null +++ b/compat/win32/headless.c @@ -0,0 +1,115 @@ +/* + * headless Git - run Git without opening a console window on Windows + */ + +#define STRICT +#define WIN32_LEAN_AND_MEAN +#define UNICODE +#define _UNICODE +#include +#include +#include +#include + +/* + * If `dir` contains the path to a Git exec directory, extend `PATH` to + * include the corresponding `bin/` directory (which is where all those + * `.dll` files needed by `git.exe` are, on Windows). + */ +static int extend_path(wchar_t *dir, size_t dir_len) +{ + const wchar_t *suffix = L"\\libexec\\git-core"; + size_t suffix_len = wcslen(suffix); + wchar_t *env; + DWORD len; + + if (dir_len < suffix_len) + return 0; + + dir_len -= suffix_len; + if (memcmp(dir + dir_len, suffix, suffix_len * sizeof(wchar_t))) + return 0; + + len = GetEnvironmentVariableW(L"PATH", NULL, 0); + if (!len) + return 0; + + env = _alloca((dir_len + 5 + len) * sizeof(wchar_t)); + wcsncpy(env, dir, dir_len); + wcscpy(env + dir_len, L"\\bin;"); + if (!GetEnvironmentVariableW(L"PATH", env + dir_len + 5, len)) + return 0; + + SetEnvironmentVariableW(L"PATH", env); + return 1; +} + +int WINAPI wWinMain(_In_ HINSTANCE instance, + _In_opt_ HINSTANCE previous_instance, + _In_ LPWSTR command_line, _In_ int show) +{ + wchar_t git_command_line[32768]; + size_t size = sizeof(git_command_line) / sizeof(wchar_t); + const wchar_t *needs_quotes = L""; + int slash = 0, i; + + STARTUPINFO startup_info = { + .cb = sizeof(STARTUPINFO), + .dwFlags = STARTF_USESHOWWINDOW, + .wShowWindow = SW_HIDE, + }; + PROCESS_INFORMATION process_info = { 0 }; + DWORD creation_flags = CREATE_UNICODE_ENVIRONMENT | + CREATE_NEW_CONSOLE | CREATE_NO_WINDOW; + DWORD exit_code; + + /* First, determine the full path of argv[0] */ + for (i = 0; _wpgmptr[i]; i++) + if (_wpgmptr[i] == L' ') + needs_quotes = L"\""; + else if (_wpgmptr[i] == L'\\') + slash = i; + + if (slash >= size - 11) + return 127; /* Too long path */ + + /* If it is in Git's exec path, add the bin/ directory to the PATH */ + extend_path(_wpgmptr, slash); + + /* Then, add the full path of `git.exe` as argv[0] */ + i = swprintf_s(git_command_line, size, L"%ls%.*ls\\git.exe%ls", + needs_quotes, slash, _wpgmptr, needs_quotes); + if (i < 0) + return 127; /* Too long path */ + + if (*command_line) { + /* Now, append the command-line arguments */ + i = swprintf_s(git_command_line + i, size - i, + L" %ls", command_line); + if (i < 0) + return 127; + } + + startup_info.hStdInput = GetStdHandle(STD_INPUT_HANDLE); + startup_info.hStdOutput = GetStdHandle(STD_OUTPUT_HANDLE); + startup_info.hStdError = GetStdHandle(STD_ERROR_HANDLE); + + if (!CreateProcess(NULL, /* infer argv[0] from the command line */ + git_command_line, /* modified command line */ + NULL, /* inherit process handles? */ + NULL, /* inherit thread handles? */ + FALSE, /* handles inheritable? */ + creation_flags, + NULL, /* use this process' environment */ + NULL, /* use this process' working directory */ + &startup_info, &process_info)) + return 129; /* could not start */ + WaitForSingleObject(process_info.hProcess, INFINITE); + if (!GetExitCodeProcess(process_info.hProcess, &exit_code)) + exit_code = 130; /* Could not determine exit code? */ + + CloseHandle(process_info.hProcess); + CloseHandle(process_info.hThread); + + return (int)exit_code; +} diff --git a/compat/win32/ntifs.h b/compat/win32/ntifs.h new file mode 100644 index 00000000000000..64ed792c52f352 --- /dev/null +++ b/compat/win32/ntifs.h @@ -0,0 +1,131 @@ +#ifndef _NTIFS_ +#define _NTIFS_ + +/* + * Copy necessary structures and definitions out of the Windows DDK + * to enable calling NtQueryDirectoryFile() + */ + +typedef _Return_type_success_(return >= 0) LONG NTSTATUS; +#define NT_SUCCESS(Status) (((NTSTATUS)(Status)) >= 0) + +#if !defined(_NTSECAPI_) && !defined(_WINTERNL_) && \ + !defined(__UNICODE_STRING_DEFINED) +#define __UNICODE_STRING_DEFINED +typedef struct _UNICODE_STRING { + USHORT Length; + USHORT MaximumLength; + PWSTR Buffer; +} UNICODE_STRING; +typedef UNICODE_STRING *PUNICODE_STRING; +typedef const UNICODE_STRING *PCUNICODE_STRING; +#endif /* !_NTSECAPI_ && !_WINTERNL_ && !__UNICODE_STRING_DEFINED */ + +typedef enum _FILE_INFORMATION_CLASS { + FileDirectoryInformation = 1, + FileFullDirectoryInformation, + FileBothDirectoryInformation, + FileBasicInformation, + FileStandardInformation, + FileInternalInformation, + FileEaInformation, + FileAccessInformation, + FileNameInformation, + FileRenameInformation, + FileLinkInformation, + FileNamesInformation, + FileDispositionInformation, + FilePositionInformation, + FileFullEaInformation, + FileModeInformation, + FileAlignmentInformation, + FileAllInformation, + FileAllocationInformation, + FileEndOfFileInformation, + FileAlternateNameInformation, + FileStreamInformation, + FilePipeInformation, + FilePipeLocalInformation, + FilePipeRemoteInformation, + FileMailslotQueryInformation, + FileMailslotSetInformation, + FileCompressionInformation, + FileObjectIdInformation, + FileCompletionInformation, + FileMoveClusterInformation, + FileQuotaInformation, + FileReparsePointInformation, + FileNetworkOpenInformation, + FileAttributeTagInformation, + FileTrackingInformation, + FileIdBothDirectoryInformation, + FileIdFullDirectoryInformation, + FileValidDataLengthInformation, + FileShortNameInformation, + FileIoCompletionNotificationInformation, + FileIoStatusBlockRangeInformation, + FileIoPriorityHintInformation, + FileSfioReserveInformation, + FileSfioVolumeInformation, + FileHardLinkInformation, + FileProcessIdsUsingFileInformation, + FileNormalizedNameInformation, + FileNetworkPhysicalNameInformation, + FileIdGlobalTxDirectoryInformation, + FileIsRemoteDeviceInformation, + FileAttributeCacheInformation, + FileNumaNodeInformation, + FileStandardLinkInformation, + FileRemoteProtocolInformation, + FileMaximumInformation +} FILE_INFORMATION_CLASS, *PFILE_INFORMATION_CLASS; + +typedef struct _FILE_FULL_DIR_INFORMATION { + ULONG NextEntryOffset; + ULONG FileIndex; + LARGE_INTEGER CreationTime; + LARGE_INTEGER LastAccessTime; + LARGE_INTEGER LastWriteTime; + LARGE_INTEGER ChangeTime; + LARGE_INTEGER EndOfFile; + LARGE_INTEGER AllocationSize; + ULONG FileAttributes; + ULONG FileNameLength; + ULONG EaSize; + WCHAR FileName[1]; +} FILE_FULL_DIR_INFORMATION, *PFILE_FULL_DIR_INFORMATION; + +typedef struct _IO_STATUS_BLOCK { + union { + NTSTATUS Status; + PVOID Pointer; + } u; + ULONG_PTR Information; +} IO_STATUS_BLOCK, *PIO_STATUS_BLOCK; + +typedef VOID +(NTAPI *PIO_APC_ROUTINE)( + IN PVOID ApcContext, + IN PIO_STATUS_BLOCK IoStatusBlock, + IN ULONG Reserved); + +NTSYSCALLAPI +NTSTATUS +NTAPI +NtQueryDirectoryFile( + _In_ HANDLE FileHandle, + _In_opt_ HANDLE Event, + _In_opt_ PIO_APC_ROUTINE ApcRoutine, + _In_opt_ PVOID ApcContext, + _Out_ PIO_STATUS_BLOCK IoStatusBlock, + _Out_writes_bytes_(Length) PVOID FileInformation, + _In_ ULONG Length, + _In_ FILE_INFORMATION_CLASS FileInformationClass, + _In_ BOOLEAN ReturnSingleEntry, + _In_opt_ PUNICODE_STRING FileName, + _In_ BOOLEAN RestartScan +); + +#define STATUS_NO_MORE_FILES ((NTSTATUS)0x80000006L) + +#endif diff --git a/compat/win32/pthread.c b/compat/win32/pthread.c index 2e7eead42cb008..9934c7ac6fa151 100644 --- a/compat/win32/pthread.c +++ b/compat/win32/pthread.c @@ -21,7 +21,7 @@ static unsigned __stdcall win32_start_routine(void *arg) return 0; } -int pthread_create(pthread_t *thread, const void *unused, +int win32_pthread_create(pthread_t *thread, const void *unused, void *(*start_routine)(void*), void *arg) { thread->arg = arg; @@ -50,7 +50,7 @@ int win32_pthread_join(pthread_t *thread, void **value_ptr) } } -pthread_t pthread_self(void) +pthread_t win32_pthread_self(void) { pthread_t t = { NULL }; t.tid = GetCurrentThreadId(); diff --git a/compat/win32/pthread.h b/compat/win32/pthread.h index 737983d00bae91..4e57147da2e57a 100644 --- a/compat/win32/pthread.h +++ b/compat/win32/pthread.h @@ -50,8 +50,9 @@ typedef struct { DWORD tid; } pthread_t; -int pthread_create(pthread_t *thread, const void *unused, - void *(*start_routine)(void*), void *arg); +int win32_pthread_create(pthread_t *thread, const void *unused, + void *(*start_routine)(void*), void *arg); +#define pthread_create win32_pthread_create /* * To avoid the need of copying a struct, we use small macro wrapper to pass @@ -62,7 +63,8 @@ int pthread_create(pthread_t *thread, const void *unused, int win32_pthread_join(pthread_t *thread, void **value_ptr); #define pthread_equal(t1, t2) ((t1).tid == (t2).tid) -pthread_t pthread_self(void); +pthread_t win32_pthread_self(void); +#define pthread_self win32_pthread_self static inline void NORETURN pthread_exit(void *ret) { diff --git a/compat/winansi.c b/compat/winansi.c index 3abe8dd5a2711b..945fe6451daff3 100644 --- a/compat/winansi.c +++ b/compat/winansi.c @@ -573,6 +573,9 @@ static void detect_msys_tty(int fd) if (!NT_SUCCESS(NtQueryObject(h, ObjectNameInformation, buffer, sizeof(buffer) - 2, &result))) return; + if (result < sizeof(*nameinfo) || !nameinfo->Name.Buffer || + !nameinfo->Name.Length) + return; name = nameinfo->Name.Buffer; name[nameinfo->Name.Length / sizeof(*name)] = 0; diff --git a/config.c b/config.c index cbb5a3bab74f6f..7176f800e1607b 100644 --- a/config.c +++ b/config.c @@ -1475,7 +1475,7 @@ int git_config_color(char *dest, const char *var, const char *value) return 0; } -static int git_default_core_config(const char *var, const char *value, void *cb) +int git_default_core_config(const char *var, const char *value, void *cb) { /* This needs a better name */ if (!strcmp(var, "core.filemode")) { diff --git a/config.h b/config.h index ca994d771475a9..fe5b7a1ae639b5 100644 --- a/config.h +++ b/config.h @@ -131,6 +131,7 @@ struct config_options { typedef int (*config_fn_t)(const char *, const char *, void *); int git_default_config(const char *, const char *, void *); +int git_default_core_config(const char *var, const char *value, void *cb); /** * Read a specific file in git-config format. diff --git a/config.mak.dev b/config.mak.dev index 4fa19d361b7837..8e99287376b4e0 100644 --- a/config.mak.dev +++ b/config.mak.dev @@ -22,8 +22,10 @@ endif ifneq ($(uname_S),FreeBSD) ifneq ($(or $(filter gcc6,$(COMPILER_FEATURES)),$(filter clang7,$(COMPILER_FEATURES))),) +ifndef USE_MIMALLOC DEVELOPER_CFLAGS += -std=gnu99 endif +endif else # FreeBSD cannot limit to C99 because its system headers unconditionally # rely on C11 features. diff --git a/config.mak.uname b/config.mak.uname index d63629fe807f59..307ec88e2dfdee 100644 --- a/config.mak.uname +++ b/config.mak.uname @@ -425,7 +425,7 @@ ifeq ($(uname_S),Windows) # link.exe next to, and required by, cl.exe, we have to prepend this # onto the existing $PATH. # - SANE_TOOL_PATH ?= $(msvc_bin_dir_msys) + SANE_TOOL_PATH ?= $(msvc_bin_dir_msys):$(sdk_ver_bin_dir_msys) HAVE_ALLOCA_H = YesPlease NO_PREAD = YesPlease NEEDS_CRYPTO_WITH_SSL = YesPlease @@ -480,20 +480,22 @@ endif CC = compat/vcbuild/scripts/clink.pl AR = compat/vcbuild/scripts/lib.pl CFLAGS = - BASIC_CFLAGS = -nologo -I. -Icompat/vcbuild/include -DWIN32 -D_CONSOLE -DHAVE_STRING_H -D_CRT_SECURE_NO_WARNINGS -D_CRT_NONSTDC_NO_DEPRECATE + BASIC_CFLAGS = -nologo -I. -Icompat/vcbuild/include -DWIN32 -D_CONSOLE -DHAVE_STRING_H -D_CRT_SECURE_NO_WARNINGS -D_CRT_NONSTDC_NO_DEPRECATE -MP -std:c11 COMPAT_OBJS = compat/msvc.o compat/winansi.o \ compat/win32/flush.o \ compat/win32/path-utils.o \ compat/win32/pthread.o compat/win32/syslog.o \ compat/win32/trace2_win32_process_info.o \ - compat/win32/dirent.o - COMPAT_CFLAGS = -D__USE_MINGW_ACCESS -DDETECT_MSYS_TTY -DNOGDI -DHAVE_STRING_H -Icompat -Icompat/regex -Icompat/win32 -DSTRIP_EXTENSION=\".exe\" - BASIC_LDFLAGS = -IGNORE:4217 -IGNORE:4049 -NOLOGO -ENTRY:wmainCRTStartup -SUBSYSTEM:CONSOLE + compat/win32/dirent.o compat/win32/fscache.o + COMPAT_CFLAGS = -D__USE_MINGW_ACCESS -DDETECT_MSYS_TTY -DENSURE_MSYSTEM_IS_SET -DNOGDI -DHAVE_STRING_H -Icompat -Icompat/regex -Icompat/win32 -DSTRIP_EXTENSION=\".exe\" + BASIC_LDFLAGS = -IGNORE:4217 -IGNORE:4049 -NOLOGO # invalidcontinue.obj allows Git's source code to close the same file # handle twice, or to access the osfhandle of an already-closed stdout # See https://msdn.microsoft.com/en-us/library/ms235330.aspx EXTLIBS = user32.lib advapi32.lib shell32.lib wininet.lib ws2_32.lib invalidcontinue.obj kernel32.lib ntdll.lib + GITLIBS += git.res PTHREAD_LIBS = + RC = compat/vcbuild/scripts/rc.pl lib = BASIC_CFLAGS += $(vcpkg_inc) $(sdk_includes) $(msvc_includes) ifndef DEBUG @@ -525,6 +527,8 @@ else endif X = .exe + EXTRA_PROGRAMS += headless-git$X + compat/msvc.o: compat/msvc.c compat/mingw.c GIT-CFLAGS endif ifeq ($(uname_S),Interix) @@ -652,7 +656,7 @@ ifeq ($(uname_S),MINGW) USE_WIN32_IPC = YesPlease USE_WIN32_MMAP = YesPlease MMAP_PREVENTS_DELETE = UnfortunatelyYes - USE_NED_ALLOCATOR = YesPlease + USE_MIMALLOC = YesPlease UNRELIABLE_FSTAT = UnfortunatelyYes OBJECT_CREATION_USES_RENAMES = UnfortunatelyNeedsTo NO_REGEX = YesPlease @@ -661,7 +665,7 @@ ifeq ($(uname_S),MINGW) DEFAULT_HELP_FORMAT = html HAVE_PLATFORM_PROCINFO = YesPlease CSPRNG_METHOD = rtlgenrandom - BASIC_LDFLAGS += -municode + BASIC_LDFLAGS += -municode -Wl,--tsaware COMPAT_CFLAGS += -DNOGDI -Icompat -Icompat/win32 COMPAT_CFLAGS += -DSTRIP_EXTENSION=\".exe\" COMPAT_OBJS += compat/mingw.o compat/winansi.o \ @@ -669,7 +673,7 @@ ifeq ($(uname_S),MINGW) compat/win32/flush.o \ compat/win32/path-utils.o \ compat/win32/pthread.o compat/win32/syslog.o \ - compat/win32/dirent.o + compat/win32/dirent.o compat/win32/fscache.o BASIC_CFLAGS += -DWIN32 EXTLIBS += -lws2_32 GITLIBS += git.res @@ -677,6 +681,7 @@ ifeq ($(uname_S),MINGW) RC = windres -O coff NATIVE_CRLF = YesPlease X = .exe + EXTRA_PROGRAMS += headless-git$X ifneq (,$(wildcard ../THIS_IS_MSYSGIT)) htmldir = doc/git/html/ prefix = @@ -706,20 +711,24 @@ else prefix = /mingw64 HOST_CPU = x86_64 BASIC_LDFLAGS += -Wl,--pic-executable,-e,mainCRTStartup + else ifeq (CLANGARM64,$(MSYSTEM)) + prefix = /clangarm64 + HOST_CPU = aarch64 + BASIC_LDFLAGS += -Wl,--pic-executable,-e,mainCRTStartup else COMPAT_CFLAGS += -D_USE_32BIT_TIME_T BASIC_LDFLAGS += -Wl,--large-address-aware endif CC = gcc COMPAT_CFLAGS += -D__USE_MINGW_ANSI_STDIO=0 -DDETECT_MSYS_TTY \ - -fstack-protector-strong + -DENSURE_MSYSTEM_IS_SET -fstack-protector-strong EXTLIBS += -lntdll INSTALL = /bin/install INTERNAL_QSORT = YesPlease HAVE_LIBCHARSET_H = YesPlease USE_GETTEXT_SCHEME = fallthrough USE_LIBPCRE = YesPlease - USE_NED_ALLOCATOR = YesPlease + NO_PYTHON = ifeq (/mingw64,$(subst 32,64,$(prefix))) # Move system config into top-level /etc/ ETC_GITCONFIG = ../etc/gitconfig @@ -731,6 +740,62 @@ else NO_PYTHON = YesPlease endif endif +ifeq (i686,$(uname_M)) + MINGW_PREFIX := mingw32 +endif +ifeq (x86_64,$(uname_M)) + MINGW_PREFIX := mingw64 +endif + + DESTDIR_WINDOWS = $(shell cygpath -aw '$(DESTDIR_SQ)') + DESTDIR_MIXED = $(shell cygpath -am '$(DESTDIR_SQ)') +install-mingit-test-artifacts: + install -m755 -d '$(DESTDIR_SQ)/usr/bin' + printf '%s\n%s\n' >'$(DESTDIR_SQ)/usr/bin/perl' \ + "#!/mingw64/bin/busybox sh" \ + "exec \"$(shell cygpath -am /usr/bin/perl.exe)\" \"\$$@\"" + + install -m755 -d '$(DESTDIR_SQ)' + printf '%s%s\n%s\n%s\n%s\n%s\n' >'$(DESTDIR_SQ)/init.bat' \ + "PATH=$(DESTDIR_WINDOWS)\\$(MINGW_PREFIX)\\bin;" \ + "C:\\WINDOWS;C:\\WINDOWS\\system32" \ + "@set GIT_TEST_INSTALLED=$(DESTDIR_MIXED)/$(MINGW_PREFIX)/bin" \ + "@`echo "$(DESTDIR_WINDOWS)" | sed 's/:.*/:/'`" \ + "@cd `echo "$(DESTDIR_WINDOWS)" | sed 's/^.://'`\\test-git\\t" \ + "@echo Now, run 'helper\\test-run-command testsuite'" + + install -m755 -d '$(DESTDIR_SQ)/test-git' + sed 's/^\(NO_PERL\|NO_PYTHON\)=.*/\1=YesPlease/' \ + '$(DESTDIR_SQ)/test-git/GIT-BUILD-OPTIONS' + + install -m755 -d '$(DESTDIR_SQ)/test-git/t/helper' + install -m755 $(TEST_PROGRAMS) '$(DESTDIR_SQ)/test-git/t/helper' + (cd t && $(TAR) cf - t[0-9][0-9][0-9][0-9] lib-diff) | \ + (cd '$(DESTDIR_SQ)/test-git/t' && $(TAR) xf -) + install -m755 t/t556x_common t/*.sh '$(DESTDIR_SQ)/test-git/t' + + install -m755 -d '$(DESTDIR_SQ)/test-git/templates' + (cd templates && $(TAR) cf - blt) | \ + (cd '$(DESTDIR_SQ)/test-git/templates' && $(TAR) xf -) + + # po/build/locale for t0200 + install -m755 -d '$(DESTDIR_SQ)/test-git/po/build/locale' + (cd po/build/locale && $(TAR) cf - .) | \ + (cd '$(DESTDIR_SQ)/test-git/po/build/locale' && $(TAR) xf -) + + # git-daemon.exe for t5802, git-http-backend.exe for t5560 + install -m755 -d '$(DESTDIR_SQ)/$(MINGW_PREFIX)/bin' + install -m755 git-daemon.exe git-http-backend.exe \ + '$(DESTDIR_SQ)/$(MINGW_PREFIX)/bin' + + # git-upload-archive (dashed) for t5000 + install -m755 -d '$(DESTDIR_SQ)/$(MINGW_PREFIX)/bin' + install -m755 git-upload-archive.exe '$(DESTDIR_SQ)/$(MINGW_PREFIX)/bin' + + # git-difftool--helper for t7800 + install -m755 -d '$(DESTDIR_SQ)/$(MINGW_PREFIX)/libexec/git-core' + install -m755 git-difftool--helper \ + '$(DESTDIR_SQ)/$(MINGW_PREFIX)/libexec/git-core' endif ifeq ($(uname_S),QNX) COMPAT_CFLAGS += -DSA_RESTART=0 @@ -755,7 +820,7 @@ vcxproj: # Make .vcxproj files and add them perl contrib/buildsystems/generate -g Vcxproj - git add -f git.sln {*,*/lib,t/helper/*}/*.vcxproj + git add -f git.sln {*,*/lib.proj,t/helper/*,reftable/libreftable{,_test}.proj}/*.vcxproj # Generate the LinkOrCopyBuiltins.targets and LinkOrCopyRemoteHttp.targets file (echo '' && \ @@ -765,7 +830,7 @@ vcxproj: echo ' '; \ done && \ echo ' ' && \ - echo '') >git/LinkOrCopyBuiltins.targets + echo '') >git.proj/LinkOrCopyBuiltins.targets (echo '' && \ echo ' ' && \ for name in $(REMOTE_CURL_ALIASES); \ @@ -773,8 +838,8 @@ vcxproj: echo ' '; \ done && \ echo ' ' && \ - echo '') >git-remote-http/LinkOrCopyRemoteHttp.targets - git add -f git/LinkOrCopyBuiltins.targets git-remote-http/LinkOrCopyRemoteHttp.targets + echo '') >git-remote-http.proj/LinkOrCopyRemoteHttp.targets + git add -f git.proj/LinkOrCopyBuiltins.targets git-remote-http.proj/LinkOrCopyRemoteHttp.targets # Add generated headers $(MAKE) MSVC=1 SKIP_VCPKG=1 prefix=/mingw64 $(GENERATED_H) @@ -787,9 +852,11 @@ vcxproj: sed -i '/^git_broken_path_fix ".*/d' git-sh-setup git add -f $(SCRIPT_LIB) $(SCRIPTS) +ifndef NO_PERL # Add Perl module $(MAKE) $(LIB_PERL_GEN) git add -f perl/build +endif # Add bin-wrappers, for testing rm -rf bin-wrappers/ diff --git a/contrib/buildsystems/CMakeLists.txt b/contrib/buildsystems/CMakeLists.txt index ea2a531be87494..38bc1f9a71e5e5 100644 --- a/contrib/buildsystems/CMakeLists.txt +++ b/contrib/buildsystems/CMakeLists.txt @@ -14,6 +14,11 @@ Note: Visual Studio also has the option of opening `CMakeLists.txt` directly; Using this option, Visual Studio will not find the source code, though, therefore the `File>Open>Folder...` option is preferred. +Visual Studio does not produce a .sln solution file nor the .vcxproj files +that may be required by VS extension tools. + +To generate the .sln/.vcxproj files run CMake manually, as described below. + Instructions to run CMake manually: mkdir -p contrib/buildsystems/out @@ -22,7 +27,7 @@ Instructions to run CMake manually: This will build the git binaries in contrib/buildsystems/out directory (our top-level .gitignore file knows to ignore contents of -this directory). +this directory). The project .sln and .vcxproj files are also generated. Possible build configurations(-DCMAKE_BUILD_TYPE) with corresponding compiler flags @@ -35,17 +40,16 @@ empty(default) : NOTE: -DCMAKE_BUILD_TYPE is optional. For multi-config generators like Visual Studio this option is ignored -This process generates a Makefile(Linux/*BSD/MacOS) , Visual Studio solution(Windows) by default. +This process generates a Makefile(Linux/*BSD/MacOS), Visual Studio solution(Windows) by default. Run `make` to build Git on Linux/*BSD/MacOS. Open git.sln on Windows and build Git. -NOTE: By default CMake uses Makefile as the build tool on Linux and Visual Studio in Windows, -to use another tool say `ninja` add this to the command line when configuring. -`-G Ninja` - NOTE: By default CMake will install vcpkg locally to your source tree on configuration, to avoid this, add `-DNO_VCPKG=TRUE` to the command line when configuring. +The Visual Studio default generator changed in v16.6 from its Visual Studio +implemenation to `Ninja` This required changes to many CMake scripts. + ]] cmake_minimum_required(VERSION 3.14) @@ -59,15 +63,29 @@ endif() if(NOT DEFINED CMAKE_EXPORT_COMPILE_COMMANDS) set(CMAKE_EXPORT_COMPILE_COMMANDS TRUE) + message("settting CMAKE_EXPORT_COMPILE_COMMANDS: ${CMAKE_EXPORT_COMPILE_COMMANDS}") endif() if(USE_VCPKG) set(VCPKG_DIR "${CMAKE_SOURCE_DIR}/compat/vcbuild/vcpkg") + message("WIN32: ${WIN32}") # show its underlying text values + message("VCPKG_DIR: ${VCPKG_DIR}") + message("VCPKG_ARCH: ${VCPKG_ARCH}") # maybe unset + message("MSVC: ${MSVC}") + message("CMAKE_GENERATOR: ${CMAKE_GENERATOR}") + message("CMAKE_CXX_COMPILER_ID: ${CMAKE_CXX_COMPILER_ID}") + message("CMAKE_GENERATOR_PLATFORM: ${CMAKE_GENERATOR_PLATFORM}") + message("CMAKE_EXPORT_COMPILE_COMMANDS: ${CMAKE_EXPORT_COMPILE_COMMANDS}") + message("ENV(CMAKE_EXPORT_COMPILE_COMMANDS): $ENV{CMAKE_EXPORT_COMPILE_COMMANDS}") if(NOT EXISTS ${VCPKG_DIR}) message("Initializing vcpkg and building the Git's dependencies (this will take a while...)") - execute_process(COMMAND ${CMAKE_SOURCE_DIR}/compat/vcbuild/vcpkg_install.bat) + execute_process(COMMAND ${CMAKE_SOURCE_DIR}/compat/vcbuild/vcpkg_install.bat ${VCPKG_ARCH}) endif() - list(APPEND CMAKE_PREFIX_PATH "${VCPKG_DIR}/installed/x64-windows") + if(NOT EXISTS ${VCPKG_ARCH}) + message("VCPKG_ARCH: unset, using 'x64-windows'") + set(VCPKG_ARCH "x64-windows") # default from vcpkg_install.bat + endif() + list(APPEND CMAKE_PREFIX_PATH "${VCPKG_DIR}/installed/${VCPKG_ARCH}") # In the vcpkg edition, we need this to be able to link to libcurl set(CURL_NO_CURL_CMAKE ON) @@ -223,7 +241,14 @@ endif() #default behaviour include_directories(${CMAKE_SOURCE_DIR}) -add_compile_definitions(GIT_HOST_CPU="${CMAKE_SYSTEM_PROCESSOR}") + +# When cross-compiling, define HOST_CPU as the canonical name of the CPU on +# which the built Git will run (for instance "x86_64"). +if(NOT HOST_CPU) + add_compile_definitions(GIT_HOST_CPU="${CMAKE_SYSTEM_PROCESSOR}") +else() + add_compile_definitions(GIT_HOST_CPU="${HOST_CPU}") +endif() add_compile_definitions(SHA256_BLK INTERNAL_QSORT RUNTIME_PREFIX) add_compile_definitions(NO_OPENSSL SHA1_DC SHA1DC_NO_STANDARD_INCLUDES SHA1DC_INIT_SAFE_HASH_DEFAULT=0 @@ -282,7 +307,8 @@ if(CMAKE_SYSTEM_NAME STREQUAL "Windows") compat/win32/trace2_win32_process_info.c compat/win32/dirent.c compat/nedmalloc/nedmalloc.c - compat/strdup.c) + compat/strdup.c + compat/win32/fscache.c) set(NO_UNIX_SOCKETS 1) elseif(CMAKE_SYSTEM_NAME STREQUAL "Linux") @@ -734,6 +760,15 @@ if(WIN32) else() message(FATAL_ERROR "Unhandled compiler: ${CMAKE_C_COMPILER_ID}") endif() + + add_executable(headless-git ${CMAKE_SOURCE_DIR}/compat/win32/headless.c) + if(CMAKE_C_COMPILER_ID STREQUAL "GNU" OR CMAKE_C_COMPILER_ID STREQUAL "Clang") + target_link_options(headless-git PUBLIC -municode -Wl,-subsystem,windows) + elseif(CMAKE_C_COMPILER_ID STREQUAL "MSVC") + target_link_options(headless-git PUBLIC /NOLOGO /ENTRY:wWinMainCRTStartup /SUBSYSTEM:WINDOWS) + else() + message(FATAL_ERROR "Unhandled compiler: ${CMAKE_C_COMPILER_ID}") + endif() elseif(UNIX) target_link_libraries(common-main pthread rt) endif() @@ -1064,24 +1099,20 @@ file(APPEND ${CMAKE_BINARY_DIR}/GIT-BUILD-OPTIONS "RUNTIME_PREFIX='${RUNTIME_PRE file(APPEND ${CMAKE_BINARY_DIR}/GIT-BUILD-OPTIONS "NO_PYTHON='${NO_PYTHON}'\n") file(APPEND ${CMAKE_BINARY_DIR}/GIT-BUILD-OPTIONS "SUPPORTS_SIMPLE_IPC='${SUPPORTS_SIMPLE_IPC}'\n") if(USE_VCPKG) - file(APPEND ${CMAKE_BINARY_DIR}/GIT-BUILD-OPTIONS "PATH=\"$PATH:$TEST_DIRECTORY/../compat/vcbuild/vcpkg/installed/x64-windows/bin\"\n") + file(APPEND ${CMAKE_BINARY_DIR}/GIT-BUILD-OPTIONS "PATH=\"$PATH:$TEST_DIRECTORY/../compat/vcbuild/vcpkg/installed/${VCPKG_ARCH}/bin\"\n") endif() #Make the tests work when building out of the source tree get_filename_component(CACHE_PATH ${CMAKE_CURRENT_LIST_DIR}/../../CMakeCache.txt ABSOLUTE) if(NOT ${CMAKE_BINARY_DIR}/CMakeCache.txt STREQUAL ${CACHE_PATH}) - file(RELATIVE_PATH BUILD_DIR_RELATIVE ${CMAKE_SOURCE_DIR} ${CMAKE_BINARY_DIR}/CMakeCache.txt) - string(REPLACE "/CMakeCache.txt" "" BUILD_DIR_RELATIVE ${BUILD_DIR_RELATIVE}) #Setting the build directory in test-lib.sh before running tests file(WRITE ${CMAKE_BINARY_DIR}/CTestCustom.cmake - "file(STRINGS ${CMAKE_SOURCE_DIR}/t/test-lib.sh GIT_BUILD_DIR_REPL REGEX \"GIT_BUILD_DIR=(.*)\")\n" - "file(STRINGS ${CMAKE_SOURCE_DIR}/t/test-lib.sh content NEWLINE_CONSUME)\n" - "string(REPLACE \"\${GIT_BUILD_DIR_REPL}\" \"GIT_BUILD_DIR=\\\"$TEST_DIRECTORY/../${BUILD_DIR_RELATIVE}\\\"\" content \"\${content}\")\n" - "file(WRITE ${CMAKE_SOURCE_DIR}/t/test-lib.sh \${content})") + "file(WRITE ${CMAKE_SOURCE_DIR}/GIT-BUILD-DIR \"${CMAKE_BINARY_DIR}\")") #misc copies file(COPY ${CMAKE_SOURCE_DIR}/t/chainlint.pl DESTINATION ${CMAKE_BINARY_DIR}/t/) file(COPY ${CMAKE_SOURCE_DIR}/po/is.po DESTINATION ${CMAKE_BINARY_DIR}/po/) - file(COPY ${CMAKE_SOURCE_DIR}/mergetools/tkdiff DESTINATION ${CMAKE_BINARY_DIR}/mergetools/) + file(GLOB mergetools "${CMAKE_SOURCE_DIR}/mergetools/*") + file(COPY ${mergetools} DESTINATION ${CMAKE_BINARY_DIR}/mergetools/) file(COPY ${CMAKE_SOURCE_DIR}/contrib/completion/git-prompt.sh DESTINATION ${CMAKE_BINARY_DIR}/contrib/completion/) file(COPY ${CMAKE_SOURCE_DIR}/contrib/completion/git-completion.bash DESTINATION ${CMAKE_BINARY_DIR}/contrib/completion/) endif() @@ -1091,8 +1122,12 @@ file(GLOB test_scipts "${CMAKE_SOURCE_DIR}/t/t[0-9]*.sh") #test foreach(tsh ${test_scipts}) add_test(NAME ${tsh} - COMMAND ${SH_EXE} ${tsh} + COMMAND ${SH_EXE} ${tsh} --no-bin-wrappers --no-chain-lint -vx WORKING_DIRECTORY ${CMAKE_SOURCE_DIR}/t) endforeach() +# This test script takes an extremely long time and is known to time out even +# on fast machines because it requires in excess of one hour to run +set_tests_properties("${CMAKE_SOURCE_DIR}/t/t7112-reset-submodule.sh" PROPERTIES TIMEOUT 4000) + endif()#BUILD_TESTING diff --git a/contrib/buildsystems/Generators/Vcxproj.pm b/contrib/buildsystems/Generators/Vcxproj.pm index 1a25789d28513b..52355932d28b9c 100644 --- a/contrib/buildsystems/Generators/Vcxproj.pm +++ b/contrib/buildsystems/Generators/Vcxproj.pm @@ -58,8 +58,8 @@ sub createProject { my $uuid = generate_guid($name); $$build_structure{"$prefix${target}_GUID"} = $uuid; my $vcxproj = $target; - $vcxproj =~ s/(.*\/)?(.*)/$&\/$2.vcxproj/; - $vcxproj =~ s/([^\/]*)(\/lib)\/(lib.vcxproj)/$1$2\/$1_$3/; + $vcxproj =~ s/(.*\/)?(.*)/$&.proj\/$2.vcxproj/; + $vcxproj =~ s/([^\/]*)(\/lib\.proj)\/(lib.vcxproj)/$1$2\/$1_$3/; $$build_structure{"$prefix${target}_VCXPROJ"} = $vcxproj; my @srcs = sort(map("$rel_dir\\$_", @{$$build_structure{"$prefix${name}_SOURCES"}})); @@ -76,8 +76,8 @@ sub createProject { my $libs_release = "\n "; my $libs_debug = "\n "; - if (!$static_library) { - $libs_release = join(";", sort(grep /^(?!libgit\.lib|xdiff\/lib\.lib|vcs-svn\/lib\.lib|reftable\/libreftable\.lib)/, @{$$build_structure{"$prefix${name}_LIBS"}})); + if (!$static_library && $name ne 'headless-git') { + $libs_release = join(";", sort(grep /^(?!libgit\.lib|xdiff\/lib\.lib|vcs-svn\/lib\.lib|reftable\/libreftable(_test)?\.lib)/, @{$$build_structure{"$prefix${name}_LIBS"}})); $libs_debug = $libs_release; $libs_debug =~ s/zlib\.lib/zlibd\.lib/g; $libs_debug =~ s/libexpat\.lib/libexpatd\.lib/g; @@ -88,8 +88,21 @@ sub createProject { $defines =~ s//>/g; $defines =~ s/\'//g; + $defines =~ s/\\"/"/g; + + my $rcdefines = $defines; + $rcdefines =~ s/(?$vcxproj" or die "Could not open $vcxproj for writing!\n"; binmode F, ":crlf :utf8"; @@ -114,12 +127,21 @@ sub createProject { Release x64 + + Debug + ARM64 + + + Release + ARM64 + $uuid Win32Proj x86-windows - x64-windows + x64-windows + arm64-windows $cdup\\compat\\vcbuild\\vcpkg\\installed\\\$(VCPKGArch) \$(VCPKGArchDirectory)\\debug\\bin \$(VCPKGArchDirectory)\\debug\\lib @@ -140,7 +162,7 @@ sub createProject { $config_type - v140 + v142 ..\\ @@ -166,6 +188,7 @@ sub createProject { OnlyExplicitInline ProgramDatabase + stdc11 true @@ -174,9 +197,9 @@ sub createProject { \$(VCPKGLibDirectory);%(AdditionalLibraryDirectories) \$(VCPKGLibs);\$(AdditionalDependencies) invalidcontinue.obj %(AdditionalOptions) - wmainCRTStartup + $entrypoint $cdup\\compat\\win32\\git.manifest - Console + $subsystem EOM if ($target eq 'libgit') { @@ -184,7 +207,7 @@ EOM Initialize VCPKG del "$cdup\\compat\\vcbuild\\vcpkg" - call "$cdup\\compat\\vcbuild\\vcpkg_install.bat" + call "$cdup\\compat\\vcbuild\\vcpkg_install.bat" \$(VCPKGArch) EOM } @@ -201,6 +224,9 @@ EOM WIN32;_DEBUG;$defines;%(PreprocessorDefinitions) MultiThreadedDebugDLL + + WIN32;_DEBUG;$rcdefines;%(PreprocessorDefinitions) + true @@ -214,6 +240,9 @@ EOM true Speed + + WIN32;NDEBUG;$rcdefines;%(PreprocessorDefinitions) + true true @@ -223,36 +252,47 @@ EOM EOM foreach(@sources) { - print F << "EOM"; + if (/\.rc$/) { + print F << "EOM"; + +EOM + } else { + print F << "EOM"; EOM + } } print F << "EOM"; EOM - if (!$static_library || $target =~ 'vcs-svn' || $target =~ 'xdiff') { + if ((!$static_library || $target =~ 'vcs-svn' || $target =~ 'xdiff') && !($name =~ /headless-git/)) { my $uuid_libgit = $$build_structure{"LIBS_libgit_GUID"}; my $uuid_libreftable = $$build_structure{"LIBS_reftable/libreftable_GUID"}; + my $uuid_libreftable_test = $$build_structure{"LIBS_reftable/libreftable_test_GUID"}; my $uuid_xdiff_lib = $$build_structure{"LIBS_xdiff/lib_GUID"}; print F << "EOM"; - + $uuid_libgit false EOM if (!($name =~ /xdiff|libreftable/)) { print F << "EOM"; - + $uuid_libreftable false + + $uuid_libreftable_test + false + EOM } if (!($name =~ 'xdiff')) { print F << "EOM"; - + $uuid_xdiff_lib false @@ -261,7 +301,7 @@ EOM if ($name =~ /(test-(line-buffer|svn-fe)|^git-remote-testsvn)\.exe$/) { my $uuid_vcs_svn_lib = $$build_structure{"LIBS_vcs-svn/lib_GUID"}; print F << "EOM"; - + $uuid_vcs_svn_lib false @@ -338,7 +378,7 @@ sub createGlueProject { my $vcxproj = $build_structure{"APPS_${appname}_VCXPROJ"}; $vcxproj =~ s/\//\\/g; $appname =~ s/.*\///; - print F "\"${appname}\", \"${vcxproj}\", \"${uuid}\""; + print F "\"${appname}.proj\", \"${vcxproj}\", \"${uuid}\""; print F "$SLN_POST"; } foreach (@libs) { @@ -348,15 +388,17 @@ sub createGlueProject { my $vcxproj = $build_structure{"LIBS_${libname}_VCXPROJ"}; $vcxproj =~ s/\//\\/g; $libname =~ s/\//_/g; - print F "\"${libname}\", \"${vcxproj}\", \"${uuid}\""; + print F "\"${libname}.proj\", \"${vcxproj}\", \"${uuid}\""; print F "$SLN_POST"; } print F << "EOM"; Global GlobalSection(SolutionConfigurationPlatforms) = preSolution + Debug|ARM64 = Debug|ARM64 Debug|x64 = Debug|x64 Debug|x86 = Debug|x86 + Release|ARM64 = Release|ARM64 Release|x64 = Release|x64 Release|x86 = Release|x86 EndGlobalSection @@ -367,10 +409,14 @@ EOM foreach (@apps) { my $appname = $_; my $uuid = $build_structure{"APPS_${appname}_GUID"}; + print F "\t\t${uuid}.Debug|ARM64.ActiveCfg = Debug|ARM64\n"; + print F "\t\t${uuid}.Debug|ARM64.Build.0 = Debug|ARM64\n"; print F "\t\t${uuid}.Debug|x64.ActiveCfg = Debug|x64\n"; print F "\t\t${uuid}.Debug|x64.Build.0 = Debug|x64\n"; print F "\t\t${uuid}.Debug|x86.ActiveCfg = Debug|Win32\n"; print F "\t\t${uuid}.Debug|x86.Build.0 = Debug|Win32\n"; + print F "\t\t${uuid}.Release|ARM64.ActiveCfg = Release|ARM64\n"; + print F "\t\t${uuid}.Release|ARM64.Build.0 = Release|ARM64\n"; print F "\t\t${uuid}.Release|x64.ActiveCfg = Release|x64\n"; print F "\t\t${uuid}.Release|x64.Build.0 = Release|x64\n"; print F "\t\t${uuid}.Release|x86.ActiveCfg = Release|Win32\n"; @@ -379,10 +425,14 @@ EOM foreach (@libs) { my $libname = $_; my $uuid = $build_structure{"LIBS_${libname}_GUID"}; + print F "\t\t${uuid}.Debug|ARM64.ActiveCfg = Debug|ARM64\n"; + print F "\t\t${uuid}.Debug|ARM64.Build.0 = Debug|ARM64\n"; print F "\t\t${uuid}.Debug|x64.ActiveCfg = Debug|x64\n"; print F "\t\t${uuid}.Debug|x64.Build.0 = Debug|x64\n"; print F "\t\t${uuid}.Debug|x86.ActiveCfg = Debug|Win32\n"; print F "\t\t${uuid}.Debug|x86.Build.0 = Debug|Win32\n"; + print F "\t\t${uuid}.Release|ARM64.ActiveCfg = Release|ARM64\n"; + print F "\t\t${uuid}.Release|ARM64.Build.0 = Release|ARM64\n"; print F "\t\t${uuid}.Release|x64.ActiveCfg = Release|x64\n"; print F "\t\t${uuid}.Release|x64.Build.0 = Release|x64\n"; print F "\t\t${uuid}.Release|x86.ActiveCfg = Release|Win32\n"; diff --git a/contrib/buildsystems/engine.pl b/contrib/buildsystems/engine.pl index ed6c45988a38b0..ee4fca200cc506 100755 --- a/contrib/buildsystems/engine.pl +++ b/contrib/buildsystems/engine.pl @@ -165,7 +165,7 @@ sub parseMakeOutput next; } - if($text =~ / -c /) { + if($text =~ / -c / || $text =~ / -i \S+\.rc /) { # compilation handleCompileLine($text, $line); @@ -263,16 +263,15 @@ sub handleCompileLine if ("$part" eq "-o") { # ignore object file shift @parts; - } elsif ("$part" eq "-c") { + } elsif ("$part" eq "-c" || "$part" eq "-i" || "$part" =~ /^-fno-/ || "$part" eq '-pedantic') { # ignore compile flag - } elsif ("$part" eq "-c") { } elsif ($part =~ /^.?-I/) { push(@incpaths, $part); } elsif ($part =~ /^.?-D/) { push(@defines, $part); } elsif ($part =~ /^-/) { push(@cflags, $part); - } elsif ($part =~ /\.(c|cc|cpp)$/) { + } elsif ($part =~ /\.(c|cc|cpp|rc)$/) { $sourcefile = $part; } else { die "Unhandled compiler option @ line $lineno: $part"; @@ -359,7 +358,7 @@ sub handleLinkLine push(@libs, $part); } elsif ($part eq 'invalidcontinue.obj') { # ignore - known to MSVC - } elsif ($part =~ /\.o$/) { + } elsif ($part =~ /\.(o|res)$/) { push(@objfiles, $part); } elsif ($part =~ /\.obj$/) { # do nothing, 'make' should not be producing .obj, only .o files @@ -371,7 +370,9 @@ sub handleLinkLine # exit(1); foreach (@objfiles) { my $sourcefile = $_; + $sourcefile =~ s/^headless-git\.o$/compat\/win32\/headless.c/; $sourcefile =~ s/\.o$/.c/; + $sourcefile =~ s/\.res$/.rc/; push(@sources, $sourcefile); push(@cflags, @{$compile_options{"${sourcefile}_CFLAGS"}}); push(@defines, @{$compile_options{"${sourcefile}_DEFINES"}}); diff --git a/contrib/subtree/Makefile b/contrib/subtree/Makefile index 6fa7496bfdb3fd..6f6e90c4cb49b6 100644 --- a/contrib/subtree/Makefile +++ b/contrib/subtree/Makefile @@ -94,7 +94,7 @@ $(GIT_SUBTREE_TEST): $(GIT_SUBTREE) cp $< $@ test: $(GIT_SUBTREE_TEST) - $(MAKE) -C t/ test + $(MAKE) -C t/ all clean: $(RM) $(GIT_SUBTREE) diff --git a/dir.c b/dir.c index 75429508200a9d..08d560657486b0 100644 --- a/dir.c +++ b/dir.c @@ -1063,16 +1063,59 @@ static int add_patterns(const char *fname, const char *base, int baselen, size_t size = 0; char *buf; - if (flags & PATTERN_NOFOLLOW) - fd = open_nofollow(fname, O_RDONLY); - else - fd = open(fname, O_RDONLY); - - if (fd < 0 || fstat(fd, &st) < 0) { - if (fd < 0) - warn_on_fopen_errors(fname); + /* + * A performance optimization for status. + * + * During a status scan, git looks in each directory for a .gitignore + * file before scanning the directory. Since .gitignore files are not + * that common, we can waste a lot of time looking for files that are + * not there. Fortunately, the fscache already knows if the directory + * contains a .gitignore file, since it has already read the directory + * and it already has the stat-data. + * + * If the fscache is enabled, use the fscache-lstat() interlude to see + * if the file exists (in the fscache hash maps) before trying to open() + * it. + * + * This causes problem when the .gitignore file is a symlink, because + * we call lstat() rather than stat() on the symlnk and the resulting + * stat-data is for the symlink itself rather than the target file. + * We CANNOT use stat() here because the fscache DOES NOT install an + * interlude for stat() and mingw_stat() always calls "open-fstat-close" + * on the file and defeats the purpose of the optimization here. Since + * symlinks are even more rare than .gitignore files, we force a fstat() + * after our open() to get stat-data for the target file. + */ + if (is_fscache_enabled(fname)) { + if (lstat(fname, &st) < 0) { + fd = -1; + } else { + fd = open(fname, O_RDONLY); + if (fd < 0) + warn_on_fopen_errors(fname); + else if (S_ISLNK(st.st_mode) && fstat(fd, &st) < 0) { + warn_on_fopen_errors(fname); + close(fd); + fd = -1; + } + } + } else { + if (flags & PATTERN_NOFOLLOW) + fd = open_nofollow(fname, O_RDONLY); else - close(fd); + fd = open(fname, O_RDONLY); + + if (fd < 0 || fstat(fd, &st) < 0) { + if (fd < 0) + warn_on_fopen_errors(fname); + else { + close(fd); + fd = -1; + } + } + } + + if (fd < 0) { if (!istate) return -1; r = read_skip_worktree_file_from_index(istate, fname, diff --git a/editor.c b/editor.c index 008c04fe2f6e0a..ba31e84ece5461 100644 --- a/editor.c +++ b/editor.c @@ -4,6 +4,7 @@ #include "strvec.h" #include "run-command.h" #include "sigchain.h" +#include "compat/terminal.h" #ifndef DEFAULT_EDITOR #define DEFAULT_EDITOR "vi" @@ -55,6 +56,7 @@ static int launch_specified_editor(const char *editor, const char *path, return error("Terminal is dumb, but EDITOR unset"); if (strcmp(editor, ":")) { + int save_and_restore_term = !strcmp(editor, "vi") || !strcmp(editor, "vim"); struct strbuf realpath = STRBUF_INIT; struct child_process p = CHILD_PROCESS_INIT; int ret, sig; @@ -83,7 +85,11 @@ static int launch_specified_editor(const char *editor, const char *path, strvec_pushv(&p.env, (const char **)env); p.use_shell = 1; p.trace2_child_class = "editor"; + if (save_and_restore_term) + save_and_restore_term = !save_term(1); if (start_command(&p) < 0) { + if (save_and_restore_term) + restore_term(); strbuf_release(&realpath); return error("unable to start editor '%s'", editor); } @@ -91,6 +97,8 @@ static int launch_specified_editor(const char *editor, const char *path, sigchain_push(SIGINT, SIG_IGN); sigchain_push(SIGQUIT, SIG_IGN); ret = finish_command(&p); + if (save_and_restore_term) + restore_term(); strbuf_release(&realpath); sig = ret - 128; sigchain_pop(SIGINT); diff --git a/entry.c b/entry.c index 616e4f073c1d6b..c9a701db189cfe 100644 --- a/entry.c +++ b/entry.c @@ -306,7 +306,7 @@ static int write_entry(struct cache_entry *ce, char *path, struct conv_attrs *ca if (!has_symlinks || to_tempfile) goto write_file_entry; - ret = symlink(new_blob, path); + ret = create_symlink(state->istate, new_blob, path); free(new_blob); if (ret) return error_errno("unable to create symlink %s", path); @@ -393,6 +393,9 @@ static int write_entry(struct cache_entry *ce, char *path, struct conv_attrs *ca } finish: + /* Flush cached lstat in fscache after writing to disk. */ + flush_fscache(); + if (state->refresh_cache) { if (!fstat_done && lstat(ce->name, &st) < 0) return error_errno("unable to stat just-written file %s", diff --git a/fetch-pack.c b/fetch-pack.c index 998fc2fa1ed4ab..007e2894dd991b 100644 --- a/fetch-pack.c +++ b/fetch-pack.c @@ -755,6 +755,7 @@ static void mark_complete_and_common_ref(struct fetch_negotiator *negotiator, save_commit_buffer = 0; trace2_region_enter("fetch-pack", "parse_remote_refs_and_find_cutoff", NULL); + enable_fscache(0); for (ref = *refs; ref; ref = ref->next) { struct commit *commit; @@ -781,6 +782,7 @@ static void mark_complete_and_common_ref(struct fetch_negotiator *negotiator, if (!cutoff || cutoff < commit->date) cutoff = commit->date; } + disable_fscache(); trace2_region_leave("fetch-pack", "parse_remote_refs_and_find_cutoff", NULL); /* diff --git a/fsmonitor-settings.c b/fsmonitor-settings.c index 464424a1e924c6..5377222651d09e 100644 --- a/fsmonitor-settings.c +++ b/fsmonitor-settings.c @@ -47,6 +47,31 @@ static struct fsmonitor_settings *alloc_settings(void) return s; } +static int check_deprecated_builtin_config(struct repository *r) +{ + int core_use_builtin_fsmonitor = 0; + + /* + * If 'core.useBuiltinFSMonitor' is set, print a deprecation warning + * suggesting the use of 'core.fsmonitor' instead. If the config is + * set to true, set the appropriate mode and return 1 indicating that + * the check resulted the config being set by this (deprecated) setting. + */ + if(!repo_config_get_bool(r, "core.useBuiltinFSMonitor", &core_use_builtin_fsmonitor) && + core_use_builtin_fsmonitor) { + if (!git_env_bool("GIT_SUPPRESS_USEBUILTINFSMONITOR_ADVICE", 0)) { + advise_if_enabled(ADVICE_USE_CORE_FSMONITOR_CONFIG, + _("core.useBuiltinFSMonitor=true is deprecated;" + "please set core.fsmonitor=true instead")); + setenv("GIT_SUPPRESS_USEBUILTINFSMONITOR_ADVICE", "1", 1); + } + fsm_settings__set_ipc(r); + return 1; + } + + return 0; +} + static void lookup_fsmonitor_settings(struct repository *r) { const char *const_str; @@ -72,12 +97,16 @@ static void lookup_fsmonitor_settings(struct repository *r) return; case 1: /* config value was unset */ + if (check_deprecated_builtin_config(r)) + return; + const_str = getenv("GIT_TEST_FSMONITOR"); break; case -1: /* config value set to an arbitrary string */ - if (repo_config_get_pathname(r, "core.fsmonitor", &const_str)) - return; /* should not happen */ + if (check_deprecated_builtin_config(r) || + repo_config_get_pathname(r, "core.fsmonitor", &const_str)) + return; break; default: /* should not happen */ diff --git a/git-add--interactive.perl b/git-add--interactive.perl index 95887fd8e52ae9..63dd89a6513bcd 100755 --- a/git-add--interactive.perl +++ b/git-add--interactive.perl @@ -174,6 +174,24 @@ sub run_cmd_pipe { die "$^O does not support: @invalid\n" if @invalid; my @args = map { m/ /o ? "\"$_\"": $_ } @_; return qx{@args}; + } elsif (($^O eq 'MSWin32' || $^O eq 'msys') && (scalar @_ > 200) && + grep $_ eq '--', @_) { + use File::Temp qw(tempfile); + my ($fhargs, $filename) = + tempfile('git-args-XXXXXX', UNLINK => 1); + + my $cmd = 'cat '.$filename.' | xargs -0 -s 20000 '; + while ($_[0] ne '--') { + $cmd = $cmd . shift(@_) . ' '; + } + + shift(@_); + print $fhargs join("\0", @_); + close($fhargs); + + my $fh = undef; + open($fh, '-|', $cmd) or die; + return <$fh>; } else { my $fh = undef; open($fh, '-|', @_) or die; diff --git a/git-compat-util.h b/git-compat-util.h index b90b64718eb610..cbaf0a4276dc69 100644 --- a/git-compat-util.h +++ b/git-compat-util.h @@ -185,7 +185,9 @@ struct strbuf; #define _ALL_SOURCE 1 #define _GNU_SOURCE 1 #define _BSD_SOURCE 1 +#ifndef _DEFAULT_SOURCE #define _DEFAULT_SOURCE 1 +#endif #define _NETBSD_SOURCE 1 #define _SGI_SOURCE 1 @@ -259,9 +261,11 @@ static inline int is_xplatform_dir_sep(int c) /* pull in Windows compatibility stuff */ #include "compat/win32/path-utils.h" #include "compat/mingw.h" +#include "compat/win32/fscache.h" #elif defined(_MSC_VER) #include "compat/win32/path-utils.h" #include "compat/msvc.h" +#include "compat/win32/fscache.h" #else #include #include @@ -369,6 +373,16 @@ char *gitdirname(char *); # include #endif +#ifdef USE_MIMALLOC +#include "mimalloc.h" +#define malloc mi_malloc +#define calloc mi_calloc +#define realloc mi_realloc +#define free mi_free +#define strdup mi_strdup +#define strndup mi_strndup +#endif + /* On most systems would have given us this, but * not on some systems (e.g. z/OS). */ @@ -540,10 +554,27 @@ static inline int git_has_dir_sep(const char *path) #define has_dir_sep(path) git_has_dir_sep(path) #endif +#ifndef is_mount_point +#define is_mount_point is_mount_point_via_stat +#endif + +#ifndef create_symlink +struct index_state; +static inline int git_create_symlink(struct index_state *index, const char *target, const char *link) +{ + return symlink(target, link); +} +#define create_symlink git_create_symlink +#endif + #ifndef query_user_email #define query_user_email() NULL #endif +#ifndef platform_strbuf_realpath +#define platform_strbuf_realpath(resolved, path) NULL +#endif + #ifdef __TANDEM #include #include @@ -1507,6 +1538,45 @@ static inline int is_missing_file_error(int errno_) return (errno_ == ENOENT || errno_ == ENOTDIR); } +/* + * Enable/disable a read-only cache for file system data on platforms that + * support it. + * + * Implementing a live-cache is complicated and requires special platform + * support (inotify, ReadDirectoryChangesW...). enable_fscache shall be used + * to mark sections of git code that extensively read from the file system + * without modifying anything. Implementations can use this to cache e.g. stat + * data or even file content without the need to synchronize with the file + * system. + */ + + /* opaque fscache structure */ +struct fscache; + +#ifndef enable_fscache +#define enable_fscache(x) /* noop */ +#endif + +#ifndef disable_fscache +#define disable_fscache() /* noop */ +#endif + +#ifndef is_fscache_enabled +#define is_fscache_enabled(path) (0) +#endif + +#ifndef flush_fscache +#define flush_fscache() /* noop */ +#endif + +#ifndef getcache_fscache +#define getcache_fscache() (NULL) /* noop */ +#endif + +#ifndef merge_fscache +#define merge_fscache(dest) /* noop */ +#endif + int cmd_main(int, const char **); /* diff --git a/git-curl-compat.h b/git-curl-compat.h index 56a83b6bbd8c43..f9f784220f83e6 100644 --- a/git-curl-compat.h +++ b/git-curl-compat.h @@ -126,4 +126,12 @@ #define GIT_CURL_HAVE_CURLSSLSET_NO_BACKENDS #endif +/** + * CURLSSLOPT_AUTO_CLIENT_CERT was added in 7.77.0, released in May + * 2021. + */ +#if LIBCURL_VERSION_NUM >= 0x074d00 +#define GIT_CURL_HAVE_CURLSSLOPT_AUTO_CLIENT_CERT +#endif + #endif diff --git a/git-gui/Makefile b/git-gui/Makefile index 56c85a85c1e493..8bc0789d7dfded 100644 --- a/git-gui/Makefile +++ b/git-gui/Makefile @@ -295,6 +295,7 @@ install: all $(QUIET)$(INSTALL_D0)'$(DESTDIR_SQ)$(gitexecdir_SQ)' $(INSTALL_D1) $(QUIET)$(INSTALL_X0)git-gui $(INSTALL_X1) '$(DESTDIR_SQ)$(gitexecdir_SQ)' $(QUIET)$(INSTALL_X0)git-gui--askpass $(INSTALL_X1) '$(DESTDIR_SQ)$(gitexecdir_SQ)' + $(QUIET)$(INSTALL_X0)git-gui--askyesno $(INSTALL_X1) '$(DESTDIR_SQ)$(gitexecdir_SQ)' $(QUIET)$(foreach p,$(GITGUI_BUILT_INS), $(INSTALL_L0)'$(DESTDIR_SQ)$(gitexecdir_SQ)/$p' $(INSTALL_L1)'$(DESTDIR_SQ)$(gitexecdir_SQ)/git-gui' $(INSTALL_L2)'$(DESTDIR_SQ)$(gitexecdir_SQ)/$p' $(INSTALL_L3) &&) true ifdef GITGUI_WINDOWS_WRAPPER $(QUIET)$(INSTALL_R0)git-gui.tcl $(INSTALL_R1) '$(DESTDIR_SQ)$(gitexecdir_SQ)' @@ -313,6 +314,7 @@ uninstall: $(QUIET)$(CLEAN_DST) '$(DESTDIR_SQ)$(gitexecdir_SQ)' $(QUIET)$(REMOVE_F0)'$(DESTDIR_SQ)$(gitexecdir_SQ)'/git-gui $(REMOVE_F1) $(QUIET)$(REMOVE_F0)'$(DESTDIR_SQ)$(gitexecdir_SQ)'/git-gui--askpass $(REMOVE_F1) + $(QUIET)$(REMOVE_F0)'$(DESTDIR_SQ)$(gitexecdir_SQ)'/git-gui--askyesno $(REMOVE_F1) $(QUIET)$(foreach p,$(GITGUI_BUILT_INS), $(REMOVE_F0)'$(DESTDIR_SQ)$(gitexecdir_SQ)'/$p $(REMOVE_F1) &&) true ifdef GITGUI_WINDOWS_WRAPPER $(QUIET)$(REMOVE_F0)'$(DESTDIR_SQ)$(gitexecdir_SQ)'/git-gui.tcl $(REMOVE_F1) diff --git a/git-gui/git-gui--askyesno b/git-gui/git-gui--askyesno new file mode 100755 index 00000000000000..c0c82e7cbd01d6 --- /dev/null +++ b/git-gui/git-gui--askyesno @@ -0,0 +1,68 @@ +#!/bin/sh +# Tcl ignores the next line -*- tcl -*- \ +exec wish "$0" -- "$@" + +# This is an implementation of a simple yes no dialog +# which is injected into the git commandline by git gui +# in case a yesno question needs to be answered. + +set NS {} +set use_ttk [package vsatisfies [package provide Tk] 8.5] +if {$use_ttk} { + set NS ttk +} + +set title "Question?" +if {$argc < 1} { + puts stderr "Usage: $argv0 " + exit 1 +} else { + if {$argc > 2 && [lindex $argv 0] == "--title"} { + set title [lindex $argv 1] + set argv [lreplace $argv 0 1] + } + set prompt [join $argv " "] +} + +${NS}::frame .t +${NS}::label .t.m -text $prompt -justify center -width 400px +.t.m configure -wraplength 400px +pack .t.m -side top -fill x -padx 20 -pady 20 -expand 1 +pack .t -side top -fill x -ipadx 20 -ipady 20 -expand 1 + +${NS}::frame .b +${NS}::frame .b.left -width 200 +${NS}::button .b.yes -text Yes -command yes +${NS}::button .b.no -text No -command no + + +pack .b.left -side left -expand 1 -fill x +pack .b.yes -side left -expand 1 +pack .b.no -side right -expand 1 -ipadx 5 +pack .b -side bottom -fill x -ipadx 20 -ipady 15 + +bind . {exit 0} +bind . {exit 1} + +proc no {} { + exit 1 +} + +proc yes {} { + exit 0 +} + +if {$::tcl_platform(platform) eq {windows}} { + set icopath [file dirname [file normalize $argv0]] + if {[file tail $icopath] eq {git-core}} { + set icopath [file dirname $icopath] + } + set icopath [file dirname $icopath] + set icopath [file join $icopath share git git-for-windows.ico] + if {[file exists $icopath]} { + wm iconbitmap . -default $icopath + } +} + +wm title . $title +tk::PlaceWindow . diff --git a/git-gui/git-gui.sh b/git-gui/git-gui.sh index 201524c34edac0..4c1195c71a0e54 100755 --- a/git-gui/git-gui.sh +++ b/git-gui/git-gui.sh @@ -623,7 +623,11 @@ proc git_write {args} { } proc githook_read {hook_name args} { - set pchook [gitdir hooks $hook_name] + if {[package vcompare $::_git_version 2.5.0] >= 0} { + set pchook [git rev-parse --git-path "hooks/$hook_name"] + } else { + set pchook [gitdir hooks $hook_name] + } lappend args 2>@1 # On Windows [file executable] might lie so we need to ask @@ -1230,6 +1234,12 @@ set have_tk85 [expr {[package vcompare $tk_version "8.5"] >= 0}] if {![info exists env(SSH_ASKPASS)]} { set env(SSH_ASKPASS) [gitexec git-gui--askpass] } +if {![info exists env(GIT_ASKPASS)]} { + set env(GIT_ASKPASS) [gitexec git-gui--askpass] +} +if {![info exists env(GIT_ASK_YESNO)]} { + set env(GIT_ASK_YESNO) [gitexec git-gui--askyesno] +} ###################################################################### ## @@ -2087,6 +2097,7 @@ set all_icons(U$ui_index) file_merge set all_icons(T$ui_index) file_statechange set all_icons(_$ui_workdir) file_plain +set all_icons(A$ui_workdir) file_plain set all_icons(M$ui_workdir) file_mod set all_icons(D$ui_workdir) file_question set all_icons(U$ui_workdir) file_merge @@ -2113,6 +2124,7 @@ foreach i { {A_ {mc "Staged for commit"}} {AM {mc "Portions staged for commit"}} {AD {mc "Staged for commit, missing"}} + {AA {mc "Intended to be added"}} {_D {mc "Missing"}} {D_ {mc "Staged for removal"}} diff --git a/git-gui/lib/diff.tcl b/git-gui/lib/diff.tcl index 871ad488c2a1c0..36d3715f7b25a2 100644 --- a/git-gui/lib/diff.tcl +++ b/git-gui/lib/diff.tcl @@ -582,7 +582,8 @@ proc apply_or_revert_hunk {x y revert} { if {$current_diff_side eq $ui_index} { set failed_msg [mc "Failed to unstage selected hunk."] lappend apply_cmd --reverse --cached - if {[string index $mi 0] ne {M}} { + set file_state [string index $mi 0] + if {$file_state ne {M} && $file_state ne {A}} { unlock_index return } @@ -595,7 +596,8 @@ proc apply_or_revert_hunk {x y revert} { lappend apply_cmd --cached } - if {[string index $mi 1] ne {M}} { + set file_state [string index $mi 1] + if {$file_state ne {M} && $file_state ne {A}} { unlock_index return } @@ -687,7 +689,8 @@ proc apply_or_revert_range_or_line {x y revert} { set failed_msg [mc "Failed to unstage selected line."] set to_context {+} lappend apply_cmd --reverse --cached - if {[string index $mi 0] ne {M}} { + set file_state [string index $mi 0] + if {$file_state ne {M} && $file_state ne {A}} { unlock_index return } @@ -702,7 +705,8 @@ proc apply_or_revert_range_or_line {x y revert} { lappend apply_cmd --cached } - if {[string index $mi 1] ne {M}} { + set file_state [string index $mi 1] + if {$file_state ne {M} && $file_state ne {A}} { unlock_index return } diff --git a/git-sh-setup.sh b/git-sh-setup.sh index ce273fe0e48d99..b53539cd0e556a 100644 --- a/git-sh-setup.sh +++ b/git-sh-setup.sh @@ -292,17 +292,30 @@ create_virtual_base() { # Platform specific tweaks to work around some commands case $(uname -s) in *MINGW*) - # Windows has its own (incompatible) sort and find - sort () { - /usr/bin/sort "$@" - } - find () { - /usr/bin/find "$@" - } - # git sees Windows-style pwd - pwd () { - builtin pwd -W - } + if test -x /usr/bin/sort + then + # Windows has its own (incompatible) sort; override + sort () { + /usr/bin/sort "$@" + } + fi + if test -x /usr/bin/find + then + # Windows has its own (incompatible) find; override + find () { + /usr/bin/find "$@" + } + fi + # On Windows, Git wants Windows paths. But /usr/bin/pwd spits out + # Unix-style paths. At least in Bash, we have a builtin pwd that + # understands the -W option to force "mixed" paths, i.e. with drive + # prefix but still with forward slashes. Let's use that, if available. + if type builtin >/dev/null 2>&1 + then + pwd () { + builtin pwd -W + } + fi is_absolute_path () { case "$1" in [/\\]* | [A-Za-z]:*) diff --git a/git.rc b/git.rc index cc3fdc6cc6cb83..9c08cb214ac4a9 100644 --- a/git.rc +++ b/git.rc @@ -12,6 +12,7 @@ BEGIN VALUE "OriginalFilename", "git.exe\0" VALUE "ProductName", "Git\0" VALUE "ProductVersion", GIT_VERSION "\0" + VALUE "FileVersion", GIT_VERSION "\0" END END diff --git a/gitk-git/gitk b/gitk-git/gitk index 0ae7d685904b85..817a7878b1eaee 100755 --- a/gitk-git/gitk +++ b/gitk-git/gitk @@ -353,6 +353,16 @@ proc parseviewrevs {view revs} { return $ret } +# Escapes a list of filter paths to be passed to git log via stdin. Note that +# paths must not be quoted. +proc escape_filter_paths {paths} { + set escaped [list] + foreach path $paths { + lappend escaped [string map {\\ \\\\ "\ " "\\\ "} $path] + } + return $escaped +} + # Start off a git log process and arrange to read its output proc start_rev_list {view} { global startmsecs commitidx viewcomplete curview @@ -405,14 +415,17 @@ proc start_rev_list {view} { if {$revs eq {}} { return 0 } - set args [concat $vflags($view) $revs] + set args $vflags($view) } else { + set revs {} set args $vorigargs($view) } if {[catch { set fd [open [concat | git log --no-color -z --pretty=raw $show_notes \ - --parents --boundary $args "--" $files] r] + --parents --boundary $args --stdin \ + "<<[join [concat $revs "--" \ + [escape_filter_paths $files]] "\\n"]"] r] } err]} { error_popup "[mc "Error executing git log:"] $err" return 0 @@ -554,13 +567,20 @@ proc updatecommits {} { set revs $newrevs set vposids($view) [lsort -unique [concat $oldpos $vposids($view)]] } - set args [concat $vflags($view) $revs --not $oldpos] + set args $vflags($view) + foreach r $oldpos { + lappend revs "^$r" + } } else { + set revs {} set args $vorigargs($view) } if {[catch { set fd [open [concat | git log --no-color -z --pretty=raw $show_notes \ - --parents --boundary $args "--" $vfilelimit($view)] r] + --parents --boundary $args --stdin \ + "<<[join [concat $revs "--" \ + [escape_filter_paths \ + $vfilelimit($view)]] "\\n"]"] r] } err]} { error_popup "[mc "Error executing git log:"] $err" return @@ -2079,7 +2099,7 @@ proc makewindow {} { global headctxmenu progresscanv progressitem progresscoords statusw global fprogitem fprogcoord lastprogupdate progupdatepending global rprogitem rprogcoord rownumsel numcommits - global have_tk85 use_ttk NS + global have_tk85 have_tk86 use_ttk NS global git_version global worddiff @@ -2577,8 +2597,13 @@ proc makewindow {} { bind . "selnextline 1" bind . "dofind -1 0" bind . "dofind 1 0" - bindkey "goforw" - bindkey "goback" + if {$have_tk86} { + bindkey <> "goforw" + bindkey <> "goback" + } else { + bindkey "goforw" + bindkey "goback" + } bind . "selnextpage -1" bind . "selnextpage 1" bind . <$M1B-Home> "allcanvs yview moveto 0.0" @@ -7692,7 +7717,7 @@ proc gettreeline {gtf id} { if {[string index $fname 0] eq "\""} { set fname [lindex $fname 0] } - set fname [encoding convertfrom $fname] + set fname [encoding convertfrom utf-8 $fname] lappend treefilelist($id) $fname } if {![eof $gtf]} { @@ -7954,7 +7979,7 @@ proc gettreediffline {gdtf ids} { if {[string index $file 0] eq "\""} { set file [lindex $file 0] } - set file [encoding convertfrom $file] + set file [encoding convertfrom utf-8 $file] if {$file ne [lindex $treediff end]} { lappend treediff $file lappend sublist $file @@ -8099,7 +8124,7 @@ proc makediffhdr {fname ids} { global ctext curdiffstart treediffs diffencoding global ctext_file_names jump_to_here targetline diffline - set fname [encoding convertfrom $fname] + set fname [encoding convertfrom utf-8 $fname] set diffencoding [get_path_encoding $fname] set i [lsearch -exact $treediffs($ids) $fname] if {$i >= 0} { @@ -8161,7 +8186,7 @@ proc parseblobdiffline {ids line} { if {![string compare -length 5 "diff " $line]} { if {![regexp {^diff (--cc|--git) } $line m type]} { - set line [encoding convertfrom $line] + set line [encoding convertfrom utf-8 $line] $ctext insert end "$line\n" hunksep continue } @@ -8210,7 +8235,7 @@ proc parseblobdiffline {ids line} { makediffhdr $fname $ids } elseif {![string compare -length 16 "* Unmerged path " $line]} { - set fname [encoding convertfrom [string range $line 16 end]] + set fname [encoding convertfrom utf-8 [string range $line 16 end]] $ctext insert end "\n" set curdiffstart [$ctext index "end - 1c"] lappend ctext_file_names $fname @@ -8263,7 +8288,7 @@ proc parseblobdiffline {ids line} { if {[string index $fname 0] eq "\""} { set fname [lindex $fname 0] } - set fname [encoding convertfrom $fname] + set fname [encoding convertfrom utf-8 $fname] set i [lsearch -exact $treediffs($ids) $fname] if {$i >= 0} { setinlist difffilestart $i $curdiffstart @@ -8282,6 +8307,7 @@ proc parseblobdiffline {ids line} { set diffinhdr 0 return } + set line [encoding convertfrom utf-8 $line] $ctext insert end "$line\n" filesep } else { @@ -10040,7 +10066,7 @@ proc showrefs {} { text $top.list -background $bgcolor -foreground $fgcolor \ -selectbackground $selectbgcolor -font mainfont \ -xscrollcommand "$top.xsb set" -yscrollcommand "$top.ysb set" \ - -width 30 -height 20 -cursor $maincursor \ + -width 60 -height 20 -cursor $maincursor \ -spacing1 1 -spacing3 1 -state disabled $top.list tag configure highlight -background $selectbgcolor if {![lsearch -exact $bglist $top.list]} { @@ -10231,10 +10257,16 @@ proc getallcommits {} { foreach id $seeds { lappend ids "^$id" } + lappend ids "--" } } if {$ids ne {}} { - set fd [open [concat $cmd $ids] r] + if {$ids eq "--all"} { + set cmd [concat $cmd "--all"] + } else { + set cmd [concat $cmd --stdin "<<[join $ids "\\n"]"] + } + set fd [open $cmd r] fconfigure $fd -blocking 0 incr allcommits nowbusy allcommits @@ -12244,7 +12276,7 @@ proc cache_gitattr {attr pathlist} { foreach row [split $rlist "\n"] { if {[regexp "(.*): $attr: (.*)" $row m path value]} { if {[string index $path 0] eq "\""} { - set path [encoding convertfrom [lindex $path 0]] + set path [encoding convertfrom utf-8 [lindex $path 0]] } set path_attr_cache($attr,$path) $value } @@ -12274,7 +12306,6 @@ if { [info exists ::env(GITK_MSGSDIR)] } { set gitk_prefix [file dirname [file dirname [file normalize $argv0]]] set gitk_libdir [file join $gitk_prefix share gitk lib] set gitk_msgsdir [file join $gitk_libdir msgs] - unset gitk_prefix } ## Internationalization (i18n) through msgcat and gettext. See @@ -12573,6 +12604,7 @@ set nullid2 "0000000000000000000000000000000000000001" set nullfile "/dev/null" set have_tk85 [expr {[package vcompare $tk_version "8.5"] >= 0}] +set have_tk86 [expr {[package vcompare $tk_version "8.6"] >= 0}] if {![info exists have_ttk]} { set have_ttk [llength [info commands ::ttk::style]] } @@ -12637,28 +12669,32 @@ if {[expr {[exec git rev-parse --is-inside-work-tree] == "true"}]} { set worktree [gitworktree] setcoords makewindow -catch { - image create photo gitlogo -width 16 -height 16 - - image create photo gitlogominus -width 4 -height 2 - gitlogominus put #C00000 -to 0 0 4 2 - gitlogo copy gitlogominus -to 1 5 - gitlogo copy gitlogominus -to 6 5 - gitlogo copy gitlogominus -to 11 5 - image delete gitlogominus - - image create photo gitlogoplus -width 4 -height 4 - gitlogoplus put #008000 -to 1 0 3 4 - gitlogoplus put #008000 -to 0 1 4 3 - gitlogo copy gitlogoplus -to 1 9 - gitlogo copy gitlogoplus -to 6 9 - gitlogo copy gitlogoplus -to 11 9 - image delete gitlogoplus - - image create photo gitlogo32 -width 32 -height 32 - gitlogo32 copy gitlogo -zoom 2 2 - - wm iconphoto . -default gitlogo gitlogo32 +if {$::tcl_platform(platform) eq {windows} && [file exists $gitk_prefix/etc/git.ico]} { + wm iconbitmap . -default $gitk_prefix/etc/git.ico +} else { + catch { + image create photo gitlogo -width 16 -height 16 + + image create photo gitlogominus -width 4 -height 2 + gitlogominus put #C00000 -to 0 0 4 2 + gitlogo copy gitlogominus -to 1 5 + gitlogo copy gitlogominus -to 6 5 + gitlogo copy gitlogominus -to 11 5 + image delete gitlogominus + + image create photo gitlogoplus -width 4 -height 4 + gitlogoplus put #008000 -to 1 0 3 4 + gitlogoplus put #008000 -to 0 1 4 3 + gitlogo copy gitlogoplus -to 1 9 + gitlogo copy gitlogoplus -to 6 9 + gitlogo copy gitlogoplus -to 11 9 + image delete gitlogoplus + + image create photo gitlogo32 -width 32 -height 32 + gitlogo32 copy gitlogo -zoom 2 2 + + wm iconphoto . -default gitlogo gitlogo32 + } } # wait for the window to become visible tkwait visibility . diff --git a/gpg-interface.c b/gpg-interface.c index 9aa714bdeea81e..cfa6bd4de70796 100644 --- a/gpg-interface.c +++ b/gpg-interface.c @@ -950,12 +950,9 @@ static int sign_buffer_gpg(struct strbuf *buffer, struct strbuf *signature, struct child_process gpg = CHILD_PROCESS_INIT; int ret; size_t bottom; - const char *cp; - struct strbuf gpg_status = STRBUF_INIT; strvec_pushl(&gpg.args, use_format->program, - "--status-fd=2", "-bsau", signing_key, NULL); @@ -967,18 +964,10 @@ static int sign_buffer_gpg(struct strbuf *buffer, struct strbuf *signature, */ sigchain_push(SIGPIPE, SIG_IGN); ret = pipe_command(&gpg, buffer->buf, buffer->len, - signature, 1024, &gpg_status, 0); + signature, 1024, NULL, 0); sigchain_pop(SIGPIPE); - for (cp = gpg_status.buf; - cp && (cp = strstr(cp, "[GNUPG:] SIG_CREATED ")); - cp++) { - if (cp == gpg_status.buf || cp[-1] == '\n') - break; /* found */ - } - ret |= !cp; - strbuf_release(&gpg_status); - if (ret) + if (ret || signature->len == bottom) return error(_("gpg failed to sign the data")); /* Strip CR from the line endings, in case we are on Windows. */ diff --git a/http.c b/http.c index 5d0502f51fd85d..2a722f63579840 100644 --- a/http.c +++ b/http.c @@ -136,7 +136,13 @@ static char *cached_accept_language; static char *http_ssl_backend; -static int http_schannel_check_revoke = 1; +static int http_schannel_check_revoke_mode = +#ifdef CURLSSLOPT_REVOKE_BEST_EFFORT + CURLSSLOPT_REVOKE_BEST_EFFORT; +#else + CURLSSLOPT_NO_REVOKE; +#endif + /* * With the backend being set to `schannel`, setting sslCAinfo would override * the Certificate Store in cURL v7.60.0 and later, which is not what we want @@ -144,6 +150,8 @@ static int http_schannel_check_revoke = 1; */ static int http_schannel_use_ssl_cainfo; +static int http_auto_client_cert; + size_t fread_buffer(char *ptr, size_t eltsize, size_t nmemb, void *buffer_) { size_t size = eltsize * nmemb; @@ -287,7 +295,19 @@ static int http_options(const char *var, const char *value, void *cb) } if (!strcmp("http.schannelcheckrevoke", var)) { - http_schannel_check_revoke = git_config_bool(var, value); + if (value && !strcmp(value, "best-effort")) { + http_schannel_check_revoke_mode = +#ifdef CURLSSLOPT_REVOKE_BEST_EFFORT + CURLSSLOPT_REVOKE_BEST_EFFORT; +#else + CURLSSLOPT_NO_REVOKE; + warning(_("%s=%s unsupported by current cURL"), + var, value); +#endif + } else + http_schannel_check_revoke_mode = + (git_config_bool(var, value) ? + 0 : CURLSSLOPT_NO_REVOKE); return 0; } @@ -296,6 +316,11 @@ static int http_options(const char *var, const char *value, void *cb) return 0; } + if (!strcmp("http.sslautoclientcert", var)) { + http_auto_client_cert = git_config_bool(var, value); + return 0; + } + if (!strcmp("http.minsessions", var)) { min_curl_sessions = git_config_int(var, value); if (min_curl_sessions > 1) @@ -818,13 +843,24 @@ static CURL *get_curl_handle(void) } #endif - if (http_ssl_backend && !strcmp("schannel", http_ssl_backend) && - !http_schannel_check_revoke) { + if (http_ssl_backend && !strcmp("schannel", http_ssl_backend)) { + long ssl_options = 0; + if (http_schannel_check_revoke_mode) { #ifdef GIT_CURL_HAVE_CURLSSLOPT_NO_REVOKE - curl_easy_setopt(result, CURLOPT_SSL_OPTIONS, CURLSSLOPT_NO_REVOKE); + ssl_options |= http_schannel_check_revoke_mode; #else - warning(_("CURLSSLOPT_NO_REVOKE not supported with cURL < 7.44.0")); + warning(_("CURLSSLOPT_NO_REVOKE not supported with cURL < 7.44.0")); +#endif + } + + if (http_auto_client_cert) { +#ifdef GIT_CURL_HAVE_CURLSSLOPT_AUTO_CLIENT_CERT + ssl_options |= CURLSSLOPT_AUTO_CLIENT_CERT; #endif + } + + if (ssl_options) + curl_easy_setopt(result, CURLOPT_SSL_OPTIONS, ssl_options); } if (http_proactive_auth) diff --git a/lockfile.c b/lockfile.c index cc9a4b84283be3..62e4c9a14f0060 100644 --- a/lockfile.c +++ b/lockfile.c @@ -17,14 +17,14 @@ static void trim_last_path_component(struct strbuf *path) int i = path->len; /* back up past trailing slashes, if any */ - while (i && path->buf[i - 1] == '/') + while (i && is_dir_sep(path->buf[i - 1])) i--; /* * then go backwards until a slash, or the beginning of the * string */ - while (i && path->buf[i - 1] != '/') + while (i && !is_dir_sep(path->buf[i - 1])) i--; strbuf_setlen(path, i); diff --git a/mem-pool.c b/mem-pool.c index 599d8e895f8121..a961d71f2ab10c 100644 --- a/mem-pool.c +++ b/mem-pool.c @@ -5,6 +5,7 @@ #include "cache.h" #include "mem-pool.h" +static struct trace_key trace_mem_pool = TRACE_KEY_INIT(MEMPOOL); #define BLOCK_GROWTH_SIZE (1024 * 1024 - sizeof(struct mp_block)) /* @@ -62,12 +63,20 @@ void mem_pool_init(struct mem_pool *pool, size_t initial_size) if (initial_size > 0) mem_pool_alloc_block(pool, initial_size, NULL); + + trace_printf_key(&trace_mem_pool, + "mem_pool (%p): init (%"PRIuMAX") initial size\n", + (void *)pool, (uintmax_t)initial_size); } void mem_pool_discard(struct mem_pool *pool, int invalidate_memory) { struct mp_block *block, *block_to_free; + trace_printf_key(&trace_mem_pool, + "mem_pool (%p): discard (%"PRIuMAX") unused\n", + (void *)pool, + (uintmax_t)(pool->mp_block->end - pool->mp_block->next_free)); block = pool->mp_block; while (block) { diff --git a/merge-recursive.c b/merge-recursive.c index 4ddd3adea003e3..005b0cfd54b3b1 100644 --- a/merge-recursive.c +++ b/merge-recursive.c @@ -993,7 +993,7 @@ static int update_file_flags(struct merge_options *opt, char *lnk = xmemdupz(buf, size); safe_create_leading_directories_const(path); unlink(path); - if (symlink(lnk, path)) + if (create_symlink(&opt->priv->orig_index, lnk, path)) ret = err(opt, _("failed to symlink '%s': %s"), path, strerror(errno)); free(lnk); diff --git a/object-file.c b/object-file.c index 5b270f046dda1d..0071655b698063 100644 --- a/object-file.c +++ b/object-file.c @@ -1785,9 +1785,9 @@ void *read_object_with_reference(struct repository *r, } static void hash_object_body(const struct git_hash_algo *algo, git_hash_ctx *c, - const void *buf, unsigned long len, + const void *buf, size_t len, struct object_id *oid, - char *hdr, int *hdrlen) + char *hdr, size_t *hdrlen) { algo->init_fn(c); algo->update_fn(c, hdr, *hdrlen); @@ -1796,23 +1796,23 @@ static void hash_object_body(const struct git_hash_algo *algo, git_hash_ctx *c, } static void write_object_file_prepare(const struct git_hash_algo *algo, - const void *buf, unsigned long len, + const void *buf, size_t len, enum object_type type, struct object_id *oid, - char *hdr, int *hdrlen) + char *hdr, size_t *hdrlen) { git_hash_ctx c; /* Generate the header */ *hdrlen = format_object_header(hdr, *hdrlen, type, len); - /* Sha1.. */ + /* Hash (function pointers) computation */ hash_object_body(algo, &c, buf, len, oid, hdr, hdrlen); } static void write_object_file_prepare_literally(const struct git_hash_algo *algo, - const void *buf, unsigned long len, + const void *buf, size_t len, const char *type, struct object_id *oid, - char *hdr, int *hdrlen) + char *hdr, size_t *hdrlen) { git_hash_ctx c; @@ -1871,17 +1871,17 @@ static int write_buffer(int fd, const void *buf, size_t len) } static void hash_object_file_literally(const struct git_hash_algo *algo, - const void *buf, unsigned long len, + const void *buf, size_t len, const char *type, struct object_id *oid) { char hdr[MAX_HEADER_LEN]; - int hdrlen = sizeof(hdr); + size_t hdrlen = sizeof(hdr); write_object_file_prepare_literally(algo, buf, len, type, oid, hdr, &hdrlen); } void hash_object_file(const struct git_hash_algo *algo, const void *buf, - unsigned long len, enum object_type type, + size_t len, enum object_type type, struct object_id *oid) { hash_object_file_literally(algo, buf, len, type_name(type), oid); @@ -2223,12 +2223,12 @@ int stream_loose_object(struct input_stream *in_stream, size_t len, return err; } -int write_object_file_flags(const void *buf, unsigned long len, +int write_object_file_flags(const void *buf, size_t len, enum object_type type, struct object_id *oid, unsigned flags) { char hdr[MAX_HEADER_LEN]; - int hdrlen = sizeof(hdr); + size_t hdrlen = sizeof(hdr); /* Normally if we have it in the pack then we do not bother writing * it out into .git/objects/??/?{38} file. @@ -2240,12 +2240,13 @@ int write_object_file_flags(const void *buf, unsigned long len, return write_loose_object(oid, hdr, hdrlen, buf, len, 0, flags); } -int write_object_file_literally(const void *buf, unsigned long len, +int write_object_file_literally(const void *buf, size_t len, const char *type, struct object_id *oid, unsigned flags) { char *header; - int hdrlen, status = 0; + size_t hdrlen; + int status = 0; /* type string, SP, %lu of the length plus NUL must fit this */ hdrlen = strlen(type) + MAX_HEADER_LEN; diff --git a/object-store.h b/object-store.h index 1be57abaf10d7a..de48c9a78d76b1 100644 --- a/object-store.h +++ b/object-store.h @@ -260,10 +260,10 @@ static inline void *repo_read_object_file(struct repository *r, int oid_object_info(struct repository *r, const struct object_id *, unsigned long *); void hash_object_file(const struct git_hash_algo *algo, const void *buf, - unsigned long len, enum object_type type, + size_t len, enum object_type type, struct object_id *oid); -int write_object_file_flags(const void *buf, unsigned long len, +int write_object_file_flags(const void *buf, size_t len, enum object_type type, struct object_id *oid, unsigned flags); static inline int write_object_file(const void *buf, unsigned long len, @@ -272,7 +272,7 @@ static inline int write_object_file(const void *buf, unsigned long len, return write_object_file_flags(buf, len, type, oid, 0); } -int write_object_file_literally(const void *buf, unsigned long len, +int write_object_file_literally(const void *buf, size_t len, const char *type, struct object_id *oid, unsigned flags); int stream_loose_object(struct input_stream *in_stream, size_t len, diff --git a/parallel-checkout.c b/parallel-checkout.c index 4f6819f2406ea8..0679a80ea44621 100644 --- a/parallel-checkout.c +++ b/parallel-checkout.c @@ -637,6 +637,7 @@ static void write_items_sequentially(struct checkout *state) { size_t i; + flush_fscache(); for (i = 0; i < parallel_checkout.nr; i++) { struct parallel_checkout_item *pc_item = ¶llel_checkout.items[i]; write_pc_item(pc_item, state); diff --git a/path.c b/path.c index a3cfcd8a6e95b3..f1db2f92a48e30 100644 --- a/path.c +++ b/path.c @@ -739,6 +739,12 @@ char *interpolate_path(const char *path, int real_home) if (skip_prefix(path, "%(prefix)/", &path)) return system_path(path); +#ifdef __MINGW32__ + if (path[0] == '/') { + warning(_("encountered old-style '%s' that should be '%%(prefix)/%s'"), path, path); + return system_path(path + 1); + } +#endif if (path[0] == '~') { const char *first_slash = strchrnul(path, '/'); const char *username = path + 1; @@ -1300,6 +1306,45 @@ char *strip_path_suffix(const char *path, const char *suffix) return offset == -1 ? NULL : xstrndup(path, offset); } +int is_mount_point_via_stat(struct strbuf *path) +{ + size_t len = path->len; + unsigned int current_dev; + struct stat st; + + if (!strcmp("/", path->buf)) + return 1; + + strbuf_addstr(path, "/."); + if (lstat(path->buf, &st)) { + /* + * If we cannot access the current directory, we cannot say + * that it is a bind mount. + */ + strbuf_setlen(path, len); + return 0; + } + current_dev = st.st_dev; + + /* Now look at the parent directory */ + strbuf_addch(path, '.'); + if (lstat(path->buf, &st)) { + /* + * If we cannot access the parent directory, we cannot say + * that it is a bind mount. + */ + strbuf_setlen(path, len); + return 0; + } + strbuf_setlen(path, len); + + /* + * If the device ID differs between current and parent directory, + * then it is a bind mount. + */ + return current_dev != st.st_dev; +} + int daemon_avoid_alias(const char *p) { int sl, ndot; diff --git a/preload-index.c b/preload-index.c index 100f7a374dca1b..0fc9423369d503 100644 --- a/preload-index.c +++ b/preload-index.c @@ -10,6 +10,8 @@ #include "thread-utils.h" #include "repository.h" +static struct fscache *fscache; + /* * Mostly randomly chosen maximum thread counts: we * cap the parallelism to 20 threads, and we want @@ -47,6 +49,7 @@ static void *preload_thread(void *_data) nr = index->cache_nr - p->offset; last_nr = nr; + enable_fscache(nr); do { struct cache_entry *ce = *cep++; struct stat st; @@ -90,6 +93,7 @@ static void *preload_thread(void *_data) pthread_mutex_unlock(&pd->mutex); } cache_def_clear(&cache); + merge_fscache(fscache); return NULL; } @@ -105,6 +109,7 @@ void preload_index(struct index_state *index, if (!HAVE_THREADS || !core_preload_index) return; + fscache = getcache_fscache(); threads = index->cache_nr / THREAD_COST; if ((index->cache_nr > 1) && (threads < 2) && git_env_bool("GIT_TEST_PRELOAD_INDEX", 0)) threads = 2; diff --git a/prompt.c b/prompt.c index 50df17279d1d5b..c546e4373774d6 100644 --- a/prompt.c +++ b/prompt.c @@ -77,7 +77,7 @@ int git_read_line_interactively(struct strbuf *line) int ret; fflush(stdout); - ret = strbuf_getline_lf(line, stdin); + ret = strbuf_getline(line, stdin); if (ret != EOF) strbuf_trim_trailing_newline(line); diff --git a/read-cache.c b/read-cache.c index b09128b1884318..7f94bc68d060aa 100644 --- a/read-cache.c +++ b/read-cache.c @@ -479,6 +479,17 @@ int ie_modified(struct index_state *istate, * then we know it is. */ if ((changed & DATA_CHANGED) && +#ifdef GIT_WINDOWS_NATIVE + /* + * Work around Git for Windows v2.27.0 fixing a bug where symlinks' + * target path lengths were not read at all, and instead recorded + * as 4096: now, all symlinks would appear as modified. + * + * So let's just special-case symlinks with a target path length + * (i.e. `sd_size`) of 4096 and force them to be re-checked. + */ + (!S_ISLNK(st->st_mode) || ce->ce_stat_data.sd_size != MAX_LONG_PATH) && +#endif (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0)) return changed; @@ -1630,6 +1641,7 @@ int refresh_index(struct index_state *istate, unsigned int flags, typechange_fmt = in_porcelain ? "T\t%s\n" : "%s: needs update\n"; added_fmt = in_porcelain ? "A\t%s\n" : "%s: needs update\n"; unmerged_fmt = in_porcelain ? "U\t%s\n" : "%s: needs merge\n"; + enable_fscache(0); /* * Use the multi-threaded preload_index() to refresh most of the * cache entries quickly then in the single threaded loop below, @@ -1724,6 +1736,7 @@ int refresh_index(struct index_state *istate, unsigned int flags, display_progress(progress, istate->cache_nr); stop_progress(&progress); trace_performance_leave("refresh index"); + disable_fscache(); return has_errors; } diff --git a/refs/files-backend.c b/refs/files-backend.c index e4009b3c421f5b..2ed8f890dafa2d 100644 --- a/refs/files-backend.c +++ b/refs/files-backend.c @@ -1880,7 +1880,7 @@ static int create_ref_symlink(struct ref_lock *lock, const char *target) #ifndef NO_SYMLINK_HEAD char *ref_path = get_locked_file_path(&lock->lk); unlink(ref_path); - ret = symlink(target, ref_path); + ret = create_symlink(NULL, target, ref_path); free(ref_path); if (ret) diff --git a/repo-settings.c b/repo-settings.c index e8b58151bc4a01..80157597c93f17 100644 --- a/repo-settings.c +++ b/repo-settings.c @@ -2,7 +2,8 @@ #include "config.h" #include "repository.h" #include "midx.h" -#include "compat/fsmonitor/fsm-listen.h" +#include "fsmonitor-ipc.h" +#include "fsmonitor-settings.h" static void repo_cfg_bool(struct repository *r, const char *key, int *dest, int def) @@ -43,6 +44,30 @@ void prepare_repo_settings(struct repository *r) /* Defaults modified by feature.* */ if (experimental) { r->settings.fetch_negotiation_algorithm = FETCH_NEGOTIATION_SKIPPING; + + /* + * Force enable the builtin FSMonitor (unless the repo + * is incompatible or they've already selected it or + * the hook version). But only if they haven't + * explicitly turned it off -- so only if our config + * value is UNSET. + * + * lookup_fsmonitor_settings() and check_for_ipc() do + * not distinguish between explicitly set FALSE and + * UNSET, so we re-test for an UNSET config key here. + * + * I'm not sure I want to fix fsmonitor-settings.c to + * have more than one _DISABLED state since our usage + * here is only to support this experimental period + * (and I don't want to overload the _reason field + * because it describes incompabilities). + */ + if (manyfiles && + fsmonitor_ipc__is_supported() && + fsm_settings__get_mode(r) == FSMONITOR_MODE_DISABLED && + repo_config_get_maybe_bool(r, "core.fsmonitor", &value) > 0 && + repo_config_get_bool(r, "core.useBuiltinFSMonitor", &value)) + fsm_settings__set_ipc(r); } if (manyfiles) { r->settings.index_version = 4; diff --git a/send-pack.c b/send-pack.c index f2e19838c9c342..03356faae66710 100644 --- a/send-pack.c +++ b/send-pack.c @@ -39,6 +39,16 @@ int option_parse_push_signed(const struct option *opt, die("bad %s argument: %s", opt->long_name, arg); } +static int config_use_sideband = 1; + +static int send_pack_config(const char *var, const char *value, void *unused) +{ + if (!strcmp("sendpack.sideband", var)) + config_use_sideband = git_config_bool(var, value); + + return 0; +} + static void feed_object(const struct object_id *oid, FILE *fh, int negative) { if (negative && @@ -497,6 +507,8 @@ int send_pack(struct send_pack_args *args, return 0; } + git_config(send_pack_config, NULL); + git_config_get_bool("push.negotiate", &push_negotiate); if (push_negotiate) get_commons_through_negotiation(args->url, remote_refs, &commons); @@ -515,7 +527,7 @@ int send_pack(struct send_pack_args *args, allow_deleting_refs = 1; if (server_supports("ofs-delta")) args->use_ofs_delta = 1; - if (server_supports("side-band-64k")) + if (config_use_sideband && server_supports("side-band-64k")) use_sideband = 1; if (server_supports("quiet")) quiet_supported = 1; diff --git a/setup.c b/setup.c index cefd5f63c4680f..852b3d9141c6ed 100644 --- a/setup.c +++ b/setup.c @@ -1482,10 +1482,19 @@ const char *setup_git_directory_gently(int *nongit_ok) break; case GIT_DIR_INVALID_OWNERSHIP: if (!nongit_ok) { + struct strbuf prequoted = STRBUF_INIT; struct strbuf quoted = STRBUF_INIT; strbuf_complete(&report, '\n'); - sq_quote_buf_pretty("ed, dir.buf); + +#ifdef __MINGW32__ + if (dir.buf[0] == '/') + strbuf_addstr(&prequoted, "%(prefix)/"); +#endif + + strbuf_add(&prequoted, dir.buf, dir.len); + sq_quote_buf_pretty("ed, prequoted.buf); + die(_("detected dubious ownership in repository at '%s'\n" "%s" "To add an exception for this directory, call:\n" diff --git a/sha1dc_git.c b/sha1dc_git.c index 5c300e812e0a11..65a6f1f2cd1bc2 100644 --- a/sha1dc_git.c +++ b/sha1dc_git.c @@ -25,10 +25,9 @@ void git_SHA1DCFinal(unsigned char hash[20], SHA1_CTX *ctx) /* * Same as SHA1DCUpdate, but adjust types to match git's usual interface. */ -void git_SHA1DCUpdate(SHA1_CTX *ctx, const void *vdata, unsigned long len) +void git_SHA1DCUpdate(SHA1_CTX *ctx, const void *vdata, size_t len) { const char *data = vdata; - /* We expect an unsigned long, but sha1dc only takes an int */ while (len > INT_MAX) { SHA1DCUpdate(ctx, data, INT_MAX); data += INT_MAX; diff --git a/sha1dc_git.h b/sha1dc_git.h index 41e1c3fd3f787e..847b38fb4b9f23 100644 --- a/sha1dc_git.h +++ b/sha1dc_git.h @@ -15,7 +15,7 @@ void git_SHA1DCInit(SHA1_CTX *); #endif void git_SHA1DCFinal(unsigned char [20], SHA1_CTX *); -void git_SHA1DCUpdate(SHA1_CTX *ctx, const void *data, unsigned long len); +void git_SHA1DCUpdate(SHA1_CTX *ctx, const void *data, size_t len); #define platform_SHA_CTX SHA1_CTX #define platform_SHA1_Init git_SHA1DCInit diff --git a/strbuf.c b/strbuf.c index 0890b1405c5cc6..8da139b1126ab5 100644 --- a/strbuf.c +++ b/strbuf.c @@ -559,8 +559,6 @@ ssize_t strbuf_write(struct strbuf *sb, FILE *f) return sb->len ? fwrite(sb->buf, 1, sb->len, f) : 0; } -#define STRBUF_MAXLINK (2*PATH_MAX) - int strbuf_readlink(struct strbuf *sb, const char *path, size_t hint) { size_t oldalloc = sb->alloc; @@ -568,15 +566,15 @@ int strbuf_readlink(struct strbuf *sb, const char *path, size_t hint) if (hint < 32) hint = 32; - while (hint < STRBUF_MAXLINK) { + for (;;) { ssize_t len; - strbuf_grow(sb, hint); - len = readlink(path, sb->buf, hint); + strbuf_grow(sb, hint + 1); + len = readlink(path, sb->buf, hint + 1); if (len < 0) { if (errno != ERANGE) break; - } else if (len < hint) { + } else if (len <= hint) { strbuf_setlen(sb, len); return 0; } diff --git a/t/README b/t/README index 979b2d4833d983..8112e711b57478 100644 --- a/t/README +++ b/t/README @@ -502,6 +502,9 @@ a test and then fails then the whole test run will abort. This can help to make sure the expected tests are executed and not silently skipped when their dependency breaks or is simply not present in a new environment. +GIT_TEST_FSCACHE= exercises the uncommon fscache code path +which adds a cache below mingw's lstat and dirent implementations. + Naming Tests ------------ diff --git a/t/helper/test-cmp.c b/t/helper/test-cmp.c new file mode 100644 index 00000000000000..1c646a54bf609b --- /dev/null +++ b/t/helper/test-cmp.c @@ -0,0 +1,73 @@ +#include "test-tool.h" +#include "git-compat-util.h" +#include "strbuf.h" +#include "gettext.h" +#include "parse-options.h" +#include "run-command.h" + +#ifdef WIN32 +#define NO_SUCH_DIR "\\\\.\\GLOBALROOT\\invalid" +#else +#define NO_SUCH_DIR "/dev/null" +#endif + +static int run_diff(const char *path1, const char *path2) +{ + const char *argv[] = { + "diff", "--no-index", NULL, NULL, NULL + }; + const char *env[] = { + "GIT_PAGER=cat", + "GIT_DIR=" NO_SUCH_DIR, + "HOME=" NO_SUCH_DIR, + NULL + }; + + argv[2] = path1; + argv[3] = path2; + return run_command_v_opt_cd_env(argv, + RUN_COMMAND_NO_STDIN | RUN_GIT_CMD, + NULL, env); +} + +int cmd__cmp(int argc, const char **argv) +{ + FILE *f0, *f1; + struct strbuf b0 = STRBUF_INIT, b1 = STRBUF_INIT; + + if (argc != 3) + die("Require exactly 2 arguments, got %d", argc); + + if (!(f0 = !strcmp(argv[1], "-") ? stdin : fopen(argv[1], "r"))) + return error_errno("could not open '%s'", argv[1]); + if (!(f1 = !strcmp(argv[2], "-") ? stdin : fopen(argv[2], "r"))) { + fclose(f0); + return error_errno("could not open '%s'", argv[2]); + } + + for (;;) { + int r0 = strbuf_getline(&b0, f0); + int r1 = strbuf_getline(&b1, f1); + + if (r0 == EOF) { + fclose(f0); + fclose(f1); + strbuf_release(&b0); + strbuf_release(&b1); + if (r1 == EOF) + return 0; +cmp_failed: + if (!run_diff(argv[1], argv[2])) + die("Huh? 'diff --no-index %s %s' succeeded", + argv[1], argv[2]); + return 1; + } + if (r1 == EOF || strbuf_cmp(&b0, &b1)) { + fclose(f0); + fclose(f1); + strbuf_release(&b0); + strbuf_release(&b1); + goto cmp_failed; + } + } +} diff --git a/t/helper/test-iconv.c b/t/helper/test-iconv.c new file mode 100644 index 00000000000000..d3c772fddf990b --- /dev/null +++ b/t/helper/test-iconv.c @@ -0,0 +1,47 @@ +#include "test-tool.h" +#include "git-compat-util.h" +#include "strbuf.h" +#include "gettext.h" +#include "parse-options.h" +#include "utf8.h" + +int cmd__iconv(int argc, const char **argv) +{ + struct strbuf buf = STRBUF_INIT; + char *from = NULL, *to = NULL, *p; + size_t len; + int ret = 0; + const char * const iconv_usage[] = { + N_("test-helper --iconv []"), + NULL + }; + struct option options[] = { + OPT_STRING('f', "from-code", &from, "encoding", "from"), + OPT_STRING('t', "to-code", &to, "encoding", "to"), + OPT_END() + }; + + argc = parse_options(argc, argv, NULL, options, + iconv_usage, 0); + + if (argc > 1 || !from || !to) + usage_with_options(iconv_usage, options); + + if (!argc) { + if (strbuf_read(&buf, 0, 2048) < 0) + die_errno("Could not read from stdin"); + } else if (strbuf_read_file(&buf, argv[0], 2048) < 0) + die_errno("Could not read from '%s'", argv[0]); + + p = reencode_string_len(buf.buf, buf.len, to, from, &len); + if (!p) + die_errno("Could not reencode"); + if (write(1, p, len) < 0) + ret = !!error_errno("Could not write %"PRIuMAX" bytes", + (uintmax_t)len); + + strbuf_release(&buf); + free(p); + + return ret; +} diff --git a/t/helper/test-tool.c b/t/helper/test-tool.c index d1d013bcd920b1..a64bfe0ebc1da0 100644 --- a/t/helper/test-tool.c +++ b/t/helper/test-tool.c @@ -14,6 +14,7 @@ static struct test_cmd cmds[] = { { "bitmap", cmd__bitmap }, { "bloom", cmd__bloom }, { "chmtime", cmd__chmtime }, + { "cmp", cmd__cmp }, { "config", cmd__config }, { "crontab", cmd__crontab }, { "csprng", cmd__csprng }, @@ -35,6 +36,7 @@ static struct test_cmd cmds[] = { { "hashmap", cmd__hashmap }, { "hash-speed", cmd__hash_speed }, { "hexdump", cmd__hexdump }, + { "iconv", cmd__iconv }, { "index-version", cmd__index_version }, { "json-writer", cmd__json_writer }, { "lazy-init-name-hash", cmd__lazy_init_name_hash }, diff --git a/t/helper/test-tool.h b/t/helper/test-tool.h index 6b46b6444b657c..d98e57710b5370 100644 --- a/t/helper/test-tool.h +++ b/t/helper/test-tool.h @@ -8,6 +8,7 @@ int cmd__advise_if_enabled(int argc, const char **argv); int cmd__bitmap(int argc, const char **argv); int cmd__bloom(int argc, const char **argv); int cmd__chmtime(int argc, const char **argv); +int cmd__cmp(int argc, const char **argv); int cmd__config(int argc, const char **argv); int cmd__crontab(int argc, const char **argv); int cmd__csprng(int argc, const char **argv); @@ -30,6 +31,7 @@ int cmd__getcwd(int argc, const char **argv); int cmd__hashmap(int argc, const char **argv); int cmd__hash_speed(int argc, const char **argv); int cmd__hexdump(int argc, const char **argv); +int cmd__iconv(int argc, const char **argv); int cmd__index_version(int argc, const char **argv); int cmd__json_writer(int argc, const char **argv); int cmd__lazy_init_name_hash(int argc, const char **argv); diff --git a/t/interop/interop-lib.sh b/t/interop/interop-lib.sh index 3e0a2911d4f9ba..dea8883821fe3e 100644 --- a/t/interop/interop-lib.sh +++ b/t/interop/interop-lib.sh @@ -4,6 +4,10 @@ . ../../GIT-BUILD-OPTIONS INTEROP_ROOT=$(pwd) BUILD_ROOT=$INTEROP_ROOT/build +case "$PATH" in +*\;*) PATH_SEP=\; ;; +*) PATH_SEP=: ;; +esac build_version () { if test -z "$1" @@ -57,7 +61,7 @@ wrap_git () { write_script "$1" <<-EOF GIT_EXEC_PATH="$2" export GIT_EXEC_PATH - PATH="$2:\$PATH" + PATH="$2$PATH_SEP\$PATH" export GIT_EXEC_PATH exec git "\$@" EOF @@ -71,7 +75,7 @@ generate_wrappers () { echo >&2 fatal: test tried to run generic git exit 1 EOF - PATH=$(pwd)/.bin:$PATH + PATH=$(pwd)/.bin$PATH_SEP$PATH } VERSION_A=${GIT_TEST_VERSION_A:-$VERSION_A} diff --git a/t/test-binary-1.png b/t/lib-diff/test-binary-1.png similarity index 100% rename from t/test-binary-1.png rename to t/lib-diff/test-binary-1.png diff --git a/t/test-binary-2.png b/t/lib-diff/test-binary-2.png similarity index 100% rename from t/test-binary-2.png rename to t/lib-diff/test-binary-2.png diff --git a/t/lib-proto-disable.sh b/t/lib-proto-disable.sh index 890622be81642b..9db481e1be15b2 100644 --- a/t/lib-proto-disable.sh +++ b/t/lib-proto-disable.sh @@ -214,7 +214,7 @@ setup_ext_wrapper () { cd "$TRASH_DIRECTORY/remote" && eval "$*" EOF - PATH=$TRASH_DIRECTORY:$PATH && + PATH=$TRASH_DIRECTORY$PATH_SEP$PATH && export TRASH_DIRECTORY ' } diff --git a/t/t0014-alias.sh b/t/t0014-alias.sh index 8d3d9144c06d61..06669cab78eec6 100755 --- a/t/t0014-alias.sh +++ b/t/t0014-alias.sh @@ -38,10 +38,10 @@ test_expect_success 'looping aliases - internal execution' ' #' test_expect_success 'run-command formats empty args properly' ' - test_must_fail env GIT_TRACE=1 git frotz a "" b " " c 2>actual.raw && - sed -ne "/run_command:/s/.*trace: run_command: //p" actual.raw >actual && - echo "git-frotz a '\'''\'' b '\'' '\'' c" >expect && - test_cmp expect actual + test_must_fail env GIT_TRACE=1 git frotz a "" b " " c 2>actual.raw && + sed -ne "/run_command: git-frotz/s/.*trace: run_command: //p" actual.raw >actual && + echo "git-frotz a '\'''\'' b '\'' '\'' c" >expect && + test_cmp expect actual ' test_done diff --git a/t/t0021-conversion.sh b/t/t0021-conversion.sh index abecd75e4e430b..408aeededc4295 100755 --- a/t/t0021-conversion.sh +++ b/t/t0021-conversion.sh @@ -8,8 +8,8 @@ export GIT_TEST_DEFAULT_INITIAL_BRANCH_NAME . ./test-lib.sh . "$TEST_DIRECTORY"/lib-terminal.sh -TEST_ROOT="$PWD" -PATH=$TEST_ROOT:$PATH +TEST_ROOT="$(pwd)" +PATH=$PWD$PATH_SEP$PATH write_script <<\EOF "$TEST_ROOT/rot13.sh" tr \ diff --git a/t/t0060-path-utils.sh b/t/t0060-path-utils.sh index 68e29c904a62c9..48433abca1cd98 100755 --- a/t/t0060-path-utils.sh +++ b/t/t0060-path-utils.sh @@ -139,25 +139,25 @@ ancestor /foo /fo -1 ancestor /foo /foo -1 ancestor /foo /bar -1 ancestor /foo /foo/bar -1 -ancestor /foo /foo:/bar -1 -ancestor /foo /:/foo:/bar 0 -ancestor /foo /foo:/:/bar 0 -ancestor /foo /:/bar:/foo 0 +ancestor /foo "/foo$PATH_SEP/bar" -1 +ancestor /foo "/$PATH_SEP/foo$PATH_SEP/bar" 0 +ancestor /foo "/foo$PATH_SEP/$PATH_SEP/bar" 0 +ancestor /foo "/$PATH_SEP/bar$PATH_SEP/foo" 0 ancestor /foo/bar / 0 ancestor /foo/bar /fo -1 ancestor /foo/bar /foo 4 ancestor /foo/bar /foo/ba -1 -ancestor /foo/bar /:/fo 0 -ancestor /foo/bar /foo:/foo/ba 4 +ancestor /foo/bar "/$PATH_SEP/fo" 0 +ancestor /foo/bar "/foo$PATH_SEP/foo/ba" 4 ancestor /foo/bar /bar -1 ancestor /foo/bar /fo -1 -ancestor /foo/bar /foo:/bar 4 -ancestor /foo/bar /:/foo:/bar 4 -ancestor /foo/bar /foo:/:/bar 4 -ancestor /foo/bar /:/bar:/fo 0 -ancestor /foo/bar /:/bar 0 +ancestor /foo/bar "/foo$PATH_SEP/bar" 4 +ancestor /foo/bar "/$PATH_SEP/foo$PATH_SEP/bar" 4 +ancestor /foo/bar "/foo$PATH_SEP/$PATH_SEP/bar" 4 +ancestor /foo/bar "/$PATH_SEP/bar$PATH_SEP/fo" 0 +ancestor /foo/bar "/$PATH_SEP/bar" 0 ancestor /foo/bar /foo 4 -ancestor /foo/bar /foo:/bar 4 +ancestor /foo/bar "/foo$PATH_SEP/bar" 4 ancestor /foo/bar /bar -1 # Windows-specific: DOS drives, network shares @@ -549,7 +549,8 @@ test_expect_success !VALGRIND,RUNTIME_PREFIX,CAN_EXEC_IN_PWD 'RUNTIME_PREFIX wor cp "$GIT_EXEC_PATH"/git$X pretend/bin/ && GIT_EXEC_PATH= ./pretend/bin/git here >actual && echo HERE >expect && - test_cmp expect actual' + test_cmp expect actual +' test_expect_success !VALGRIND,RUNTIME_PREFIX,CAN_EXEC_IN_PWD '%(prefix)/ works' ' mkdir -p pretend/bin && @@ -560,4 +561,32 @@ test_expect_success !VALGRIND,RUNTIME_PREFIX,CAN_EXEC_IN_PWD '%(prefix)/ works' test_cmp expect actual ' +test_expect_success MINGW 'MSYSTEM/PATH is adjusted if necessary' ' + mkdir -p "$HOME"/bin pretend/mingw64/bin \ + pretend/mingw64/libexec/git-core pretend/usr/bin && + cp "$GIT_EXEC_PATH"/git.exe pretend/mingw64/bin/ && + cp "$GIT_EXEC_PATH"/git.exe pretend/mingw64/libexec/git-core/ && + # copy the .dll files, if any (happens when building via CMake) + case "$GIT_EXEC_PATH"/*.dll in + */"*.dll") ;; # no `.dll` files to be copied + *) + cp "$GIT_EXEC_PATH"/*.dll pretend/mingw64/bin/ && + cp "$GIT_EXEC_PATH"/*.dll pretend/mingw64/libexec/git-core/ + ;; + esac && + echo "env | grep MSYSTEM=" | write_script "$HOME"/bin/git-test-home && + echo "echo mingw64" | write_script pretend/mingw64/bin/git-test-bin && + echo "echo usr" | write_script pretend/usr/bin/git-test-bin2 && + + ( + MSYSTEM= && + GIT_EXEC_PATH= && + pretend/mingw64/libexec/git-core/git.exe test-home >actual && + pretend/mingw64/libexec/git-core/git.exe test-bin >>actual && + pretend/mingw64/bin/git.exe test-bin2 >>actual + ) && + test_write_lines MSYSTEM=$MSYSTEM mingw64 usr >expect && + test_cmp expect actual +' + test_done diff --git a/t/t0061-run-command.sh b/t/t0061-run-command.sh index 7b5423eebdafa4..9e55358742e464 100755 --- a/t/t0061-run-command.sh +++ b/t/t0061-run-command.sh @@ -70,7 +70,7 @@ test_expect_success 'run_command does not try to execute a directory' ' cat bin2/greet EOF - PATH=$PWD/bin1:$PWD/bin2:$PATH \ + PATH=$PWD/bin1$PATH_SEP$PWD/bin2$PATH_SEP$PATH \ test-tool run-command run-command greet >actual 2>err && test_cmp bin2/greet actual && test_must_be_empty err @@ -87,7 +87,7 @@ test_expect_success POSIXPERM 'run_command passes over non-executable file' ' cat bin2/greet EOF - PATH=$PWD/bin1:$PWD/bin2:$PATH \ + PATH=$PWD/bin1$PATH_SEP$PWD/bin2$PATH_SEP$PATH \ test-tool run-command run-command greet >actual 2>err && test_cmp bin2/greet actual && test_must_be_empty err @@ -107,7 +107,7 @@ test_expect_success POSIXPERM,SANITY 'unreadable directory in PATH' ' git config alias.nitfol "!echo frotz" && chmod a-rx local-command && ( - PATH=./local-command:$PATH && + PATH=./local-command$PATH_SEP$PATH && git nitfol >actual ) && echo frotz >expect && diff --git a/t/t0300-credentials.sh b/t/t0300-credentials.sh index 3485c0534e6d39..ed045188e0f124 100755 --- a/t/t0300-credentials.sh +++ b/t/t0300-credentials.sh @@ -35,7 +35,7 @@ test_expect_success 'setup helper scripts' ' test -z "$pass" || echo password=$pass EOF - PATH="$PWD:$PATH" + PATH="$PWD$PATH_SEP$PATH" ' test_expect_success 'credential_fill invokes helper' ' diff --git a/t/t1007-hash-object.sh b/t/t1007-hash-object.sh index ac5ad8c7402d2b..df78646b073760 100755 --- a/t/t1007-hash-object.sh +++ b/t/t1007-hash-object.sh @@ -50,6 +50,9 @@ test_expect_success 'setup' ' example sha1:ddd3f836d3e3fbb7ae289aa9ae83536f76956399 example sha256:b44fe1fe65589848253737db859bd490453510719d7424daab03daf0767b85ae + + large5GB sha1:0be2be10a4c8764f32c4bf372a98edc731a4b204 + large5GB sha256:dc18ca621300c8d3cfa505a275641ebab00de189859e022a975056882d313e64 EOF ' @@ -249,4 +252,40 @@ test_expect_success '--literally with extra-long type' ' echo example | git hash-object -t $t --literally --stdin ' +test_expect_success EXPENSIVE,SIZE_T_IS_64BIT,!LONG_IS_64BIT \ + 'files over 4GB hash literally' ' + test-tool genzeros $((5*1024*1024*1024)) >big && + test_oid large5GB >expect && + git hash-object --stdin --literally actual && + test_cmp expect actual +' + +test_expect_success EXPENSIVE,SIZE_T_IS_64BIT,!LONG_IS_64BIT \ + 'files over 4GB hash correctly via --stdin' ' + { test -f big || test-tool genzeros $((5*1024*1024*1024)) >big; } && + test_oid large5GB >expect && + git hash-object --stdin actual && + test_cmp expect actual +' + +test_expect_success EXPENSIVE,SIZE_T_IS_64BIT,!LONG_IS_64BIT \ + 'files over 4GB hash correctly' ' + { test -f big || test-tool genzeros $((5*1024*1024*1024)) >big; } && + test_oid large5GB >expect && + git hash-object -- big >actual && + test_cmp expect actual +' + +# This clean filter does nothing, other than excercising the interface. +# We ensure that cleaning doesn't mangle large files on 64-bit Windows. +test_expect_success EXPENSIVE,SIZE_T_IS_64BIT,!LONG_IS_64BIT \ + 'hash filtered files over 4GB correctly' ' + { test -f big || test-tool genzeros $((5*1024*1024*1024)) >big; } && + test_oid large5GB >expect && + test_config filter.null-filter.clean "cat" && + echo "big filter=null-filter" >.gitattributes && + git hash-object -- big >actual && + test_cmp expect actual +' + test_done diff --git a/t/t1090-sparse-checkout-scope.sh b/t/t1090-sparse-checkout-scope.sh index 3a14218b245d4c..529844e2862c74 100755 --- a/t/t1090-sparse-checkout-scope.sh +++ b/t/t1090-sparse-checkout-scope.sh @@ -106,4 +106,24 @@ test_expect_success 'in partial clone, sparse checkout only fetches needed blobs test_cmp expect actual ' +test_expect_success MINGW 'no unnecessary opendir() with fscache' ' + git clone . fscache-test && + ( + cd fscache-test && + git config core.fscache 1 && + echo "/excluded/*" >.git/info/sparse-checkout && + for f in $(test_seq 10) + do + sha1=$(echo $f | git hash-object -w --stdin) && + git update-index --add \ + --cacheinfo 100644,$sha1,excluded/$f || exit 1 + done && + test_tick && + git commit -m excluded && + GIT_TRACE_FSCACHE=1 git status >out 2>err && + grep excluded err >grep.out && + test_line_count = 1 grep.out + ) +' + test_done diff --git a/t/t1504-ceiling-dirs.sh b/t/t1504-ceiling-dirs.sh index 0fafcf9dde385f..060296febcce1c 100755 --- a/t/t1504-ceiling-dirs.sh +++ b/t/t1504-ceiling-dirs.sh @@ -81,9 +81,9 @@ then GIT_CEILING_DIRECTORIES="$TRASH_ROOT/top/" test_fail subdir_ceil_at_top_slash - GIT_CEILING_DIRECTORIES=":$TRASH_ROOT/top" + GIT_CEILING_DIRECTORIES="$PATH_SEP$TRASH_ROOT/top" test_prefix subdir_ceil_at_top_no_resolve "sub/dir/" - GIT_CEILING_DIRECTORIES=":$TRASH_ROOT/top/" + GIT_CEILING_DIRECTORIES="$PATH_SEP$TRASH_ROOT/top/" test_prefix subdir_ceil_at_top_slash_no_resolve "sub/dir/" fi @@ -113,13 +113,13 @@ GIT_CEILING_DIRECTORIES="$TRASH_ROOT/subdi" test_prefix subdir_ceil_at_subdi_slash "sub/dir/" -GIT_CEILING_DIRECTORIES="/foo:$TRASH_ROOT/sub" +GIT_CEILING_DIRECTORIES="/foo$PATH_SEP$TRASH_ROOT/sub" test_fail second_of_two -GIT_CEILING_DIRECTORIES="$TRASH_ROOT/sub:/bar" +GIT_CEILING_DIRECTORIES="$TRASH_ROOT/sub$PATH_SEP/bar" test_fail first_of_two -GIT_CEILING_DIRECTORIES="/foo:$TRASH_ROOT/sub:/bar" +GIT_CEILING_DIRECTORIES="/foo$PATH_SEP$TRASH_ROOT/sub$PATH_SEP/bar" test_fail second_of_three diff --git a/t/t2031-checkout-long-paths.sh b/t/t2031-checkout-long-paths.sh new file mode 100755 index 00000000000000..f30f8920ca689c --- /dev/null +++ b/t/t2031-checkout-long-paths.sh @@ -0,0 +1,102 @@ +#!/bin/sh + +test_description='checkout long paths on Windows + +Ensures that Git for Windows can deal with long paths (>260) enabled via core.longpaths' + +. ./test-lib.sh + +if test_have_prereq !MINGW +then + skip_all='skipping MINGW specific long paths test' + test_done +fi + +test_expect_success setup ' + p=longpathxx && # -> 10 + p=$p$p$p$p$p && # -> 50 + p=$p$p$p$p$p && # -> 250 + + path=${p}/longtestfile && # -> 263 (MAX_PATH = 260) + + blob=$(echo foobar | git hash-object -w --stdin) && + + printf "100644 %s 0\t%s\n" "$blob" "$path" | + git update-index --add --index-info && + git commit -m initial -q +' + +test_expect_success 'checkout of long paths without core.longpaths fails' ' + git config core.longpaths false && + test_must_fail git checkout -f 2>error && + grep -q "Filename too long" error && + test ! -d longpa* +' + +test_expect_success 'checkout of long paths with core.longpaths works' ' + git config core.longpaths true && + git checkout -f && + test_path_is_file longpa*/longtestfile +' + +test_expect_success 'update of long paths' ' + echo frotz >>$(ls longpa*/longtestfile) && + echo $path > expect && + git ls-files -m > actual && + test_cmp expect actual && + git add $path && + git commit -m second && + git grep "frotz" HEAD -- $path +' + +test_expect_success cleanup ' + # bash cannot delete the trash dir if it contains a long path + # lets help cleaning up (unless in debug mode) + if test -z "$debug" + then + rm -rf longpa~1 + fi +' + +# check that the template used in the test won't be too long: +abspath="$(pwd)"/testdir +test ${#abspath} -gt 230 || +test_set_prereq SHORTABSPATH + +test_expect_success SHORTABSPATH 'clean up path close to MAX_PATH' ' + p=/123456789abcdef/123456789abcdef/123456789abcdef/123456789abc/ef && + p=y$p$p$p$p && + subdir="x$(echo "$p" | tail -c $((253 - ${#abspath})) - )" && + # Now, $abspath/$subdir has exactly 254 characters, and is inside CWD + p2="$abspath/$subdir" && + test 254 = ${#p2} && + + # Be careful to overcome path limitations of the MSys tools and split + # the $subdir into two parts. ($subdir2 has to contain 16 chars and a + # slash somewhere following; that is why we asked for abspath <= 230 and + # why we placed a slash near the end of the $subdir template.) + subdir2=${subdir#????????????????*/} && + subdir1=testdir/${subdir%/$subdir2} && + mkdir -p "$subdir1" && + i=0 && + # The most important case is when absolute path is 258 characters long, + # and that will be when i == 4. + while test $i -le 7 + do + mkdir -p $subdir2 && + touch $subdir2/one-file && + mv ${subdir2%%/*} "$subdir1/" && + subdir2=z${subdir2} && + i=$(($i+1)) || + exit 1 + done && + + # now check that git is able to clear the tree: + (cd testdir && + git init && + git config core.longpaths yes && + git clean -fdx) && + test ! -d "$subdir1" +' + +test_done diff --git a/t/t2040-checkout-symlink-attr.sh b/t/t2040-checkout-symlink-attr.sh new file mode 100755 index 00000000000000..e00c31d096ce88 --- /dev/null +++ b/t/t2040-checkout-symlink-attr.sh @@ -0,0 +1,46 @@ +#!/bin/sh + +test_description='checkout symlinks with `symlink` attribute on Windows + +Ensures that Git for Windows creates symlinks of the right type, +as specified by the `symlink` attribute in `.gitattributes`.' + +# Tell MSYS to create native symlinks. Without this flag test-lib's +# prerequisite detection for SYMLINKS doesn't detect the right thing. +MSYS=winsymlinks:nativestrict && export MSYS + +. ./test-lib.sh + +if ! test_have_prereq MINGW,SYMLINKS +then + skip_all='skipping $0: MinGW-only test, which requires symlink support.' + test_done +fi + +# Adds a symlink to the index without clobbering the work tree. +cache_symlink () { + sha=$(printf '%s' "$1" | git hash-object --stdin -w) && + git update-index --add --cacheinfo 120000,$sha,"$2" +} + +test_expect_success 'checkout symlinks with attr' ' + cache_symlink file1 file-link && + cache_symlink dir dir-link && + + printf "file-link symlink=file\ndir-link symlink=dir\n" >.gitattributes && + git add .gitattributes && + + git checkout . && + + mkdir dir && + echo "[a]b=c" >file1 && + echo "[x]y=z" >dir/file2 && + + # MSYS2 is very forgiving, it will resolve symlinks even if the + # symlink type is incorrect. To make this test meaningful, try + # them with a native, non-MSYS executable, such as `git config`. + test "$(git config -f file-link a.b)" = "c" && + test "$(git config -f dir-link/file2 x.y)" = "z" +' + +test_done diff --git a/t/t2300-cd-to-toplevel.sh b/t/t2300-cd-to-toplevel.sh index b40eeb263fe896..4328b9172a5e6e 100755 --- a/t/t2300-cd-to-toplevel.sh +++ b/t/t2300-cd-to-toplevel.sh @@ -17,7 +17,7 @@ test_cd_to_toplevel () { test_expect_success $3 "$2" ' ( cd '"'$1'"' && - PATH="$EXEC_PATH:$PATH" && + PATH="$EXEC_PATH$PATH_SEP$PATH" && . git-sh-setup && cd_to_toplevel && [ "$(pwd -P)" = "$TOPLEVEL" ] diff --git a/t/t3307-notes-man.sh b/t/t3307-notes-man.sh index ae316502c4531b..2b6894df0a247a 100755 --- a/t/t3307-notes-man.sh +++ b/t/t3307-notes-man.sh @@ -27,7 +27,7 @@ test_expect_success 'example 1: notes to add an Acked-by line' ' ' test_expect_success 'example 2: binary notes' ' - cp "$TEST_DIRECTORY"/test-binary-1.png . && + cp "$TEST_DIRECTORY"/lib-diff/test-binary-1.png . && git checkout B && blob=$(git hash-object -w test-binary-1.png) && git notes --ref=logo add -C "$blob" && diff --git a/t/t3402-rebase-merge.sh b/t/t3402-rebase-merge.sh index 7e46f4ca850616..ce553b6edd50e4 100755 --- a/t/t3402-rebase-merge.sh +++ b/t/t3402-rebase-merge.sh @@ -146,7 +146,7 @@ test_expect_success 'rebase -s funny -Xopt' ' git checkout -b test-funny main^ && test_commit funny && ( - PATH=./test-bin:$PATH && + PATH=./test-bin$PATH_SEP$PATH && git rebase -s funny -Xopt main ) && test -f funny.was.run diff --git a/t/t3418-rebase-continue.sh b/t/t3418-rebase-continue.sh index 130e2f9b553003..f1c1eff4eed4f5 100755 --- a/t/t3418-rebase-continue.sh +++ b/t/t3418-rebase-continue.sh @@ -73,7 +73,7 @@ test_expect_success 'rebase --continue remembers merge strategy and options' ' EOF chmod +x test-bin/git-merge-funny && ( - PATH=./test-bin:$PATH && + PATH=./test-bin$PATH_SEP$PATH && test_must_fail git rebase -s funny -Xopt main topic ) && test -f funny.was.run && @@ -81,7 +81,7 @@ test_expect_success 'rebase --continue remembers merge strategy and options' ' echo "Resolved" >F2 && git add F2 && ( - PATH=./test-bin:$PATH && + PATH=./test-bin$PATH_SEP$PATH && git rebase --continue ) && test -f funny.was.run @@ -105,7 +105,7 @@ test_expect_success 'rebase -i --continue handles merge strategy and options' ' EOF chmod +x test-bin/git-merge-funny && ( - PATH=./test-bin:$PATH && + PATH=./test-bin$PATH_SEP$PATH && test_must_fail git rebase -i -s funny -Xopt -Xfoo main topic ) && test -f funny.was.run && @@ -113,7 +113,7 @@ test_expect_success 'rebase -i --continue handles merge strategy and options' ' echo "Resolved" >F2 && git add F2 && ( - PATH=./test-bin:$PATH && + PATH=./test-bin$PATH_SEP$PATH && git rebase --continue ) && test -f funny.was.run diff --git a/t/t3700-add.sh b/t/t3700-add.sh index 8689b48589c0d5..648019d38e7a23 100755 --- a/t/t3700-add.sh +++ b/t/t3700-add.sh @@ -498,4 +498,15 @@ test_expect_success CASE_INSENSITIVE_FS 'path is case-insensitive' ' git add "$downcased" ' +test_expect_success MINGW 'can add files via NTFS junctions' ' + test_when_finished "cmd //c rmdir junction && rm -rf target" && + test_create_repo target && + cmd //c "mklink /j junction target" && + >target/via-junction && + git -C junction add "$(pwd)/junction/via-junction" && + echo via-junction >expect && + git -C target diff --cached --name-only >actual && + test_cmp expect actual +' + test_done diff --git a/t/t3701-add-interactive.sh b/t/t3701-add-interactive.sh index 5841f280fb2d4c..0c6d914b9b3b8c 100755 --- a/t/t3701-add-interactive.sh +++ b/t/t3701-add-interactive.sh @@ -1044,6 +1044,27 @@ test_expect_success 'checkout -p patch editing of added file' ' ) ' +test_expect_success EXPENSIVE 'add -i with a lot of files' ' + git reset --hard && + x160=0123456789012345678901234567890123456789 && + x160=$x160$x160$x160$x160 && + y= && + i=0 && + while test $i -le 200 + do + name=$(printf "%s%03d" $x160 $i) && + echo $name >$name && + git add -N $name && + y="${y}y$LF" && + i=$(($i+1)) || + exit 1 + done && + echo "$y" | git add -p -- . && + git diff --cached >staged && + test_line_count = 1407 staged && + git reset --hard +' + test_expect_success 'show help from add--helper' ' git reset --hard && cat >expect <<-EOF && diff --git a/t/t3903-stash.sh b/t/t3903-stash.sh index 376cc8f4ab8429..4e49ac1b114a3b 100755 --- a/t/t3903-stash.sh +++ b/t/t3903-stash.sh @@ -1332,7 +1332,7 @@ test_expect_success 'stash -- works with binary files' ' git reset && >subdir/untracked && >subdir/tracked && - cp "$TEST_DIRECTORY"/test-binary-1.png subdir/tracked-binary && + cp "$TEST_DIRECTORY"/lib-diff/test-binary-1.png subdir/tracked-binary && git add subdir/tracked* && git stash -- subdir/ && test_path_is_missing subdir/tracked && diff --git a/t/t4012-diff-binary.sh b/t/t4012-diff-binary.sh index c509143c8141e0..c2c7c021346421 100755 --- a/t/t4012-diff-binary.sh +++ b/t/t4012-diff-binary.sh @@ -20,7 +20,7 @@ test_expect_success 'prepare repository' ' echo AIT >a && echo BIT >b && echo CIT >c && echo DIT >d && git update-index --add a b c d && echo git >a && - cat "$TEST_DIRECTORY"/test-binary-1.png >b && + cat "$TEST_DIRECTORY"/lib-diff/test-binary-1.png >b && echo git >c && cat b b >d ' diff --git a/t/t4049-diff-stat-count.sh b/t/t4049-diff-stat-count.sh index 0a4fc735d44ad5..6c028646485971 100755 --- a/t/t4049-diff-stat-count.sh +++ b/t/t4049-diff-stat-count.sh @@ -34,7 +34,7 @@ test_expect_success 'binary changes do not count in lines' ' git reset --hard && echo a >a && echo c >c && - cat "$TEST_DIRECTORY"/test-binary-1.png >d && + cat "$TEST_DIRECTORY"/lib-diff/test-binary-1.png >d && cat >expect <<-\EOF && a | 1 + c | 1 + diff --git a/t/t4108-apply-threeway.sh b/t/t4108-apply-threeway.sh index c558282bc09475..1c85e7051d5f72 100755 --- a/t/t4108-apply-threeway.sh +++ b/t/t4108-apply-threeway.sh @@ -232,11 +232,11 @@ test_expect_success 'apply with --3way --cached and conflicts' ' test_expect_success 'apply binary file patch' ' git reset --hard main && - cp "$TEST_DIRECTORY/test-binary-1.png" bin.png && + cp "$TEST_DIRECTORY/lib-diff/test-binary-1.png" bin.png && git add bin.png && git commit -m "add binary file" && - cp "$TEST_DIRECTORY/test-binary-2.png" bin.png && + cp "$TEST_DIRECTORY/lib-diff/test-binary-2.png" bin.png && git diff --binary >bin.diff && git reset --hard && @@ -247,11 +247,11 @@ test_expect_success 'apply binary file patch' ' test_expect_success 'apply binary file patch with 3way' ' git reset --hard main && - cp "$TEST_DIRECTORY/test-binary-1.png" bin.png && + cp "$TEST_DIRECTORY/lib-diff/test-binary-1.png" bin.png && git add bin.png && git commit -m "add binary file" && - cp "$TEST_DIRECTORY/test-binary-2.png" bin.png && + cp "$TEST_DIRECTORY/lib-diff/test-binary-2.png" bin.png && git diff --binary >bin.diff && git reset --hard && @@ -262,11 +262,11 @@ test_expect_success 'apply binary file patch with 3way' ' test_expect_success 'apply full-index patch with 3way' ' git reset --hard main && - cp "$TEST_DIRECTORY/test-binary-1.png" bin.png && + cp "$TEST_DIRECTORY/lib-diff/test-binary-1.png" bin.png && git add bin.png && git commit -m "add binary file" && - cp "$TEST_DIRECTORY/test-binary-2.png" bin.png && + cp "$TEST_DIRECTORY/lib-diff/test-binary-2.png" bin.png && git diff --full-index >bin.diff && git reset --hard && diff --git a/t/t5003-archive-zip.sh b/t/t5003-archive-zip.sh index fc499cdff01d01..2bfa0c0a285689 100755 --- a/t/t5003-archive-zip.sh +++ b/t/t5003-archive-zip.sh @@ -88,7 +88,7 @@ test_expect_success \ 'mkdir a && echo simple textfile >a/a && mkdir a/bin && - cp /bin/sh a/bin && + cp "$TEST_DIRECTORY/lib-diff/test-binary-1.png" a/bin && printf "text\r" >a/text.cr && printf "text\r\n" >a/text.crlf && printf "text\n" >a/text.lf && diff --git a/t/t5505-remote.sh b/t/t5505-remote.sh index 9006196ac601e8..c40fe351977568 100755 --- a/t/t5505-remote.sh +++ b/t/t5505-remote.sh @@ -835,8 +835,8 @@ test_expect_success '"remote show" does not show symbolic refs' ' ( cd three && git remote show origin >output && - ! grep "^ *HEAD$" < output && - ! grep -i stale < output + ! grep "^ *HEAD$" .git/branches/origin && git remote rename origin origin && test_path_is_missing .git/branches/origin && @@ -1043,8 +1043,8 @@ test_expect_success 'migrate a remote from named file in $GIT_DIR/branches (2)' ( cd seven && git remote rm origin && - mkdir .git/branches && - echo "quux#foom" > .git/branches/origin && + mkdir -p .git/branches && + echo "quux#foom" >.git/branches/origin && git remote rename origin origin && test_path_is_missing .git/branches/origin && test "$(git config remote.origin.url)" = "quux" && diff --git a/t/t5516-fetch-push.sh b/t/t5516-fetch-push.sh index 79dc470c014a15..626972f07567b3 100755 --- a/t/t5516-fetch-push.sh +++ b/t/t5516-fetch-push.sh @@ -946,8 +946,8 @@ test_expect_success 'fetch with branches' ' mk_empty testrepo && git branch second $the_first_commit && git checkout second && - mkdir testrepo/.git/branches && - echo ".." > testrepo/.git/branches/branch1 && + mkdir -p testrepo/.git/branches && + echo ".." >testrepo/.git/branches/branch1 && ( cd testrepo && git fetch branch1 && @@ -960,8 +960,8 @@ test_expect_success 'fetch with branches' ' test_expect_success 'fetch with branches containing #' ' mk_empty testrepo && - mkdir testrepo/.git/branches && - echo "..#second" > testrepo/.git/branches/branch2 && + mkdir -p testrepo/.git/branches && + echo "..#second" >testrepo/.git/branches/branch2 && ( cd testrepo && git fetch branch2 && @@ -977,8 +977,8 @@ test_expect_success 'push with branches' ' git checkout second && test_when_finished "rm -rf .git/branches" && - mkdir .git/branches && - echo "testrepo" > .git/branches/branch1 && + mkdir -p .git/branches && + echo "testrepo" >.git/branches/branch1 && git push branch1 && ( @@ -993,8 +993,8 @@ test_expect_success 'push with branches containing #' ' mk_empty testrepo && test_when_finished "rm -rf .git/branches" && - mkdir .git/branches && - echo "testrepo#branch3" > .git/branches/branch2 && + mkdir -p .git/branches && + echo "testrepo#branch3" >.git/branches/branch2 && git push branch2 && ( @@ -1523,7 +1523,7 @@ EOF git init no-thin && git --git-dir=no-thin/.git config receive.unpacklimit 0 && git push no-thin/.git refs/heads/main:refs/heads/foo && - echo modified >> path1 && + echo modified >>path1 && git commit -am modified && git repack -adf && rcvpck="git receive-pack --reject-thin-pack-for-testing" && diff --git a/t/t5532-fetch-proxy.sh b/t/t5532-fetch-proxy.sh index d664912799b43a..55bc57c138ddc2 100755 --- a/t/t5532-fetch-proxy.sh +++ b/t/t5532-fetch-proxy.sh @@ -27,7 +27,7 @@ test_expect_success 'setup proxy script' ' write_script proxy <<-\EOF echo >&2 "proxying for $*" - cmd=$(./proxy-get-cmd) + cmd=$("$PERL_PATH" ./proxy-get-cmd) echo >&2 "Running $cmd" exec $cmd EOF diff --git a/t/t5580-unc-paths.sh b/t/t5580-unc-paths.sh index cd7604fff93a55..2c5b410048952d 100755 --- a/t/t5580-unc-paths.sh +++ b/t/t5580-unc-paths.sh @@ -21,14 +21,11 @@ fi UNCPATH="$(winpwd)" case "$UNCPATH" in [A-Z]:*) + WITHOUTDRIVE="${UNCPATH#?:}" # Use administrative share e.g. \\localhost\C$\git-sdk-64\usr\src\git # (we use forward slashes here because MSYS2 and Git accept them, and # they are easier on the eyes) - UNCPATH="//localhost/${UNCPATH%%:*}\$/${UNCPATH#?:}" - test -d "$UNCPATH" || { - skip_all='could not access administrative share; skipping' - test_done - } + UNCPATH="//localhost/${UNCPATH%%:*}\$$WITHOUTDRIVE" ;; *) skip_all='skipping UNC path tests, cannot determine current path as UNC' @@ -36,6 +33,18 @@ case "$UNCPATH" in ;; esac +test_expect_success 'clone into absolute path lacking a drive prefix' ' + USINGBACKSLASHES="$(echo "$WITHOUTDRIVE"/without-drive-prefix | + tr / \\\\)" && + git clone . "$USINGBACKSLASHES" && + test -f without-drive-prefix/.git/HEAD +' + +test -d "$UNCPATH" || { + skip_all='could not access administrative share; skipping' + test_done +} + test_expect_success setup ' test_commit initial ' diff --git a/t/t5601-clone.sh b/t/t5601-clone.sh index 45f0803ed4dccd..cdf8eb0ab5d27e 100755 --- a/t/t5601-clone.sh +++ b/t/t5601-clone.sh @@ -94,6 +94,13 @@ test_expect_success LIBCURL 'clone does not detect username:password when it is ! grep "uses plaintext credentials" err ' +test_expect_success CASE_INSENSITIVE_FS 'core.worktree is not added due to path case' ' + + mkdir UPPERCASE && + git clone src "$(pwd)/uppercase" && + test "unset" = "$(git -C UPPERCASE config --default unset core.worktree)" +' + test_expect_success 'clone from hooks' ' test_create_repo r0 && diff --git a/t/t5605-clone-local.sh b/t/t5605-clone-local.sh index 38b850c10ef8cd..829d54edc0e2dd 100755 --- a/t/t5605-clone-local.sh +++ b/t/t5605-clone-local.sh @@ -11,6 +11,21 @@ repo_is_hardlinked() { test_line_count = 0 output } +if test_have_prereq MINGW,BUSYBOX +then + # BusyBox' `find` does not support `-links`. Besides, BusyBox-w32's + # lstat() does not report hard links, just like Git's mingw_lstat() + # (from where BusyBox-w32 got its initial implementation). + repo_is_hardlinked() { + for f in $(find "$1/objects" -type f) + do + "$SYSTEMROOT"/system32/fsutil.exe \ + hardlink list $f >links && + test_line_count -gt 1 links || return 1 + done + } +fi + test_expect_success 'preparing origin repository' ' : >file && git add . && git commit -m1 && git clone --bare . a.git && diff --git a/t/t5615-alternate-env.sh b/t/t5615-alternate-env.sh index 83513e46a3556b..fa2dd08084d21c 100755 --- a/t/t5615-alternate-env.sh +++ b/t/t5615-alternate-env.sh @@ -40,7 +40,7 @@ test_expect_success 'access alternate via absolute path' ' ' test_expect_success 'access multiple alternates' ' - check_obj "$PWD/one.git/objects:$PWD/two.git/objects" <<-EOF + check_obj "$PWD/one.git/objects$PATH_SEP$PWD/two.git/objects" <<-EOF $one blob $two blob EOF @@ -76,7 +76,7 @@ test_expect_success 'access alternate via relative path (subdir)' ' quoted='"one.git\057objects"' unquoted='two.git/objects' test_expect_success 'mix of quoted and unquoted alternates' ' - check_obj "$quoted:$unquoted" <<-EOF + check_obj "$quoted$PATH_SEP$unquoted" <<-EOF $one blob $two blob EOF diff --git a/t/t5801-remote-helpers.sh b/t/t5801-remote-helpers.sh index d386076dbd3f8d..b5d69b0c50e060 100755 --- a/t/t5801-remote-helpers.sh +++ b/t/t5801-remote-helpers.sh @@ -239,7 +239,7 @@ test_expect_success 'push update refs failure' ' echo "update fail" >>file && git commit -a -m "update fail" && git rev-parse --verify testgit/origin/heads/update >expect && - test_expect_code 1 env GIT_REMOTE_TESTGIT_FAILURE="non-fast forward" \ + test_must_fail env GIT_REMOTE_TESTGIT_FAILURE="non-fast forward" \ git push origin update && git rev-parse --verify testgit/origin/heads/update >actual && test_cmp expect actual diff --git a/t/t5802-connect-helper.sh b/t/t5802-connect-helper.sh index c6c2661878c0ca..a096eeeeb427cf 100755 --- a/t/t5802-connect-helper.sh +++ b/t/t5802-connect-helper.sh @@ -85,7 +85,7 @@ test_expect_success 'set up fake git-daemon' ' "$TRASH_DIRECTORY/remote" EOF export TRASH_DIRECTORY && - PATH=$TRASH_DIRECTORY:$PATH + PATH=$TRASH_DIRECTORY$PATH_SEP$PATH ' test_expect_success 'ext command can connect to git daemon (no vhost)' ' diff --git a/t/t5813-proto-disable-ssh.sh b/t/t5813-proto-disable-ssh.sh index 3f084ee306517b..0a2c77093babad 100755 --- a/t/t5813-proto-disable-ssh.sh +++ b/t/t5813-proto-disable-ssh.sh @@ -14,8 +14,23 @@ test_expect_success 'setup repository to clone' ' ' test_proto "host:path" ssh "remote:repo.git" -test_proto "ssh://" ssh "ssh://remote$PWD/remote/repo.git" -test_proto "git+ssh://" ssh "git+ssh://remote$PWD/remote/repo.git" + +hostdir="$PWD" +if test_have_prereq MINGW && test "/${PWD#/}" != "$PWD" +then + case "$PWD" in + [A-Za-z]:/*) + hostdir="${PWD#?:}" + ;; + *) + skip_all="Unhandled PWD '$PWD'; skipping rest" + test_done + ;; + esac +fi + +test_proto "ssh://" ssh "ssh://remote$hostdir/remote/repo.git" +test_proto "git+ssh://" ssh "git+ssh://remote$hostdir/remote/repo.git" # Don't even bother setting up a "-remote" directory, as ssh would generally # complain about the bogus option rather than completing our request. Our diff --git a/t/t6403-merge-file.sh b/t/t6403-merge-file.sh index 1a7082323dddfc..a6854952c9ab9e 100755 --- a/t/t6403-merge-file.sh +++ b/t/t6403-merge-file.sh @@ -252,7 +252,7 @@ test_expect_success "expected conflict markers" ' test_expect_success 'binary files cannot be merged' ' test_must_fail git merge-file -p \ - orig.txt "$TEST_DIRECTORY"/test-binary-1.png new1.txt 2> merge.err && + orig.txt "$TEST_DIRECTORY"/lib-diff/test-binary-1.png new1.txt 2> merge.err && grep "Cannot merge binary files" merge.err ' diff --git a/t/t6407-merge-binary.sh b/t/t6407-merge-binary.sh index e8a28717cece32..2547f1d504a2c5 100755 --- a/t/t6407-merge-binary.sh +++ b/t/t6407-merge-binary.sh @@ -9,7 +9,7 @@ export GIT_TEST_DEFAULT_INITIAL_BRANCH_NAME test_expect_success setup ' - cat "$TEST_DIRECTORY"/test-binary-1.png >m && + cat "$TEST_DIRECTORY"/lib-diff/test-binary-1.png >m && git add m && git ls-files -s | sed -e "s/ 0 / 1 /" >E1 && test_tick && diff --git a/t/t7004-tag.sh b/t/t7004-tag.sh index 9aa1660651b8a9..4e56d5c70124d0 100755 --- a/t/t7004-tag.sh +++ b/t/t7004-tag.sh @@ -1375,30 +1375,6 @@ test_expect_success GPG \ 'test_config user.signingkey BobTheMouse && test_must_fail git tag -s -m tail tag-gpg-failure' -# try to produce invalid signature -test_expect_success GPG \ - 'git tag -s fails if gpg is misconfigured (bad signature format)' \ - 'test_config gpg.program echo && - test_must_fail git tag -s -m tail tag-gpg-failure' - -# try to produce invalid signature -test_expect_success GPG 'git verifies tag is valid with double signature' ' - git tag -s -m tail tag-gpg-double-sig && - git cat-file tag tag-gpg-double-sig >tag && - othersigheader=$(test_oid othersigheader) && - sed -ne "/^\$/q;p" tag >new-tag && - cat <<-EOM >>new-tag && - $othersigheader -----BEGIN PGP SIGNATURE----- - someinvaliddata - -----END PGP SIGNATURE----- - EOM - sed -e "1,/^tagger/d" tag >>new-tag && - new_tag=$(git hash-object -t tag -w new-tag) && - git update-ref refs/tags/tag-gpg-double-sig $new_tag && - git verify-tag tag-gpg-double-sig && - git fsck -' - # try to sign with bad user.signingkey test_expect_success GPGSM \ 'git tag -s fails if gpgsm is misconfigured (bad key)' \ @@ -1406,13 +1382,6 @@ test_expect_success GPGSM \ test_config gpg.format x509 && test_must_fail git tag -s -m tail tag-gpg-failure' -# try to produce invalid signature -test_expect_success GPGSM \ - 'git tag -s fails if gpgsm is misconfigured (bad signature format)' \ - 'test_config gpg.x509.program echo && - test_config gpg.format x509 && - test_must_fail git tag -s -m tail tag-gpg-failure' - # try to verify without gpg: rm -rf gpghome diff --git a/t/t7006-pager.sh b/t/t7006-pager.sh index e56ca5b0fa8d47..2a90dc74f2b73d 100755 --- a/t/t7006-pager.sh +++ b/t/t7006-pager.sh @@ -54,7 +54,7 @@ test_expect_success !MINGW,TTY 'LESS and LV envvars set by git-sh-setup' ' sane_unset LESS LV && PAGER="env >pager-env.out; wc" && export PAGER && - PATH="$(git --exec-path):$PATH" && + PATH="$(git --exec-path)$PATH_SEP$PATH" && export PATH && test_terminal sh -c ". git-sh-setup && git_pager" ) && @@ -388,7 +388,7 @@ test_default_pager() { EOF chmod +x \$less && ( - PATH=.:\$PATH && + PATH=.$PATH_SEP\$PATH && export PATH && $full_command ) && diff --git a/t/t7108-reset-stdin.sh b/t/t7108-reset-stdin.sh new file mode 100755 index 00000000000000..b7cbcbf869296c --- /dev/null +++ b/t/t7108-reset-stdin.sh @@ -0,0 +1,32 @@ +#!/bin/sh + +test_description='reset --stdin' + +. ./test-lib.sh + +test_expect_success 'reset --stdin' ' + test_commit hello && + git rm hello.t && + test -z "$(git ls-files hello.t)" && + echo hello.t | git reset --stdin && + test hello.t = "$(git ls-files hello.t)" +' + +test_expect_success 'reset --stdin -z' ' + test_commit world && + git rm hello.t world.t && + test -z "$(git ls-files hello.t world.t)" && + printf world.tQworld.tQhello.tQ | q_to_nul | git reset --stdin -z && + printf "hello.t\nworld.t\n" >expect && + git ls-files >actual && + test_cmp expect actual +' + +test_expect_success '--stdin requires --mixed' ' + echo hello.t >list && + test_must_fail git reset --soft --stdin test.txt && + git add test.txt && + git commit -m A && + echo B >> test.txt && + git checkout . && + test -z "$(git status -s)" && + echo A > expect.txt && + test_cmp expect.txt test.txt && + cd .. && + rm -rf fscache-test +' + +test_expect_success MINGW 'fscache flush cache dir' ' + + git init fscache-test && + cd fscache-test && + git config core.fscache 1 && + echo A > test.txt && + git add test.txt && + git commit -m A && + rm test.txt && + mkdir test.txt && + touch test.txt/test.txt && + git checkout . && + test -z "$(git status -s)" && + echo A > expect.txt && + test_cmp expect.txt test.txt && + cd .. && + rm -rf fscache-test +' + test_expect_success setup ' fill x y z >same && fill 1 2 3 4 5 6 7 8 >one && diff --git a/t/t7300-clean.sh b/t/t7300-clean.sh index c975eb54d234d7..121048e201064d 100755 --- a/t/t7300-clean.sh +++ b/t/t7300-clean.sh @@ -789,4 +789,14 @@ test_expect_success 'traverse into directories that may have ignored entries' ' ) ' +test_expect_success MINGW 'clean does not traverse mount points' ' + mkdir target && + >target/dont-clean-me && + git init with-mountpoint && + cmd //c "mklink /j with-mountpoint\\mountpoint target" && + git -C with-mountpoint clean -dfx && + test_path_is_missing with-mountpoint/mountpoint && + test_path_is_file target/dont-clean-me +' + test_done diff --git a/t/t7422-submodule-long-path.sh b/t/t7422-submodule-long-path.sh new file mode 100755 index 00000000000000..458519eafd6f03 --- /dev/null +++ b/t/t7422-submodule-long-path.sh @@ -0,0 +1,110 @@ +#!/bin/sh +# +# Copyright (c) 2013 Doug Kelly +# + +test_description='Test submodules with a path near PATH_MAX + +This test verifies that "git submodule" initialization, update and clones work, including with recursive submodules and paths approaching PATH_MAX (260 characters on Windows) +' + +TEST_NO_CREATE_REPO=1 +. ./test-lib.sh + +# cloning a submodule calls is_git_directory("$path/../.git/modules/$path"), +# which effectively limits the maximum length to PATH_MAX / 2 minus some +# overhead; start with 3 * 36 = 108 chars (test 2 fails if >= 110) +longpath36=0123456789abcdefghijklmnopqrstuvwxyz +longpath180=$longpath36$longpath36$longpath36$longpath36$longpath36 + +# the git database must fit within PATH_MAX, which limits the submodule name +# to PATH_MAX - len(pwd) - ~90 (= len("/objects//") + 40-byte sha1 + some +# overhead from the test case) +pwd=$(pwd) +pwdlen=$(echo "$pwd" | wc -c) +longpath=$(echo $longpath180 | cut -c 1-$((170-$pwdlen))) + +test_expect_success 'submodule with a long path' ' + git config --global protocol.file.allow always && + GIT_TEST_DEFAULT_INITIAL_BRANCH_NAME= \ + git -c init.defaultBranch=long init --bare remote && + test_create_repo bundle1 && + ( + cd bundle1 && + test_commit "shoot" && + git rev-parse --verify HEAD >../expect + ) && + mkdir home && + ( + cd home && + git clone ../remote test && + cd test && + git checkout -B long && + git submodule add ../bundle1 $longpath && + test_commit "sogood" && + ( + cd $longpath && + git rev-parse --verify HEAD >actual && + test_cmp ../../../expect actual + ) && + git push origin long + ) && + mkdir home2 && + ( + cd home2 && + git clone ../remote test && + cd test && + git checkout long && + git submodule update --init && + ( + cd $longpath && + git rev-parse --verify HEAD >actual && + test_cmp ../../../expect actual + ) + ) +' + +test_expect_success 'recursive submodule with a long path' ' + GIT_TEST_DEFAULT_INITIAL_BRANCH_NAME= \ + git -c init.defaultBranch=long init --bare super && + test_create_repo child && + ( + cd child && + test_commit "shoot" && + git rev-parse --verify HEAD >../expect + ) && + test_create_repo parent && + ( + cd parent && + git submodule add ../child $longpath && + test_commit "aim" + ) && + mkdir home3 && + ( + cd home3 && + git clone ../super test && + cd test && + git checkout -B long && + git submodule add ../parent foo && + git submodule update --init --recursive && + test_commit "sogood" && + ( + cd foo/$longpath && + git rev-parse --verify HEAD >actual && + test_cmp ../../../../expect actual + ) && + git push origin long + ) && + mkdir home4 && + ( + cd home4 && + git clone ../super test --recursive && + ( + cd test/foo/$longpath && + git rev-parse --verify HEAD >actual && + test_cmp ../../../../expect actual + ) + ) +' + +test_done diff --git a/t/t7502-commit-porcelain.sh b/t/t7502-commit-porcelain.sh index 38a532d81ccd06..b643375fa59da0 100755 --- a/t/t7502-commit-porcelain.sh +++ b/t/t7502-commit-porcelain.sh @@ -586,6 +586,48 @@ test_expect_success 'cleanup commit messages (scissors option,-F,-e, scissors on test_must_be_empty actual ' +test_expect_success 'helper-editor' ' + + write_script lf-to-crlf.sh <<-\EOF + sed "s/\$/Q/" <"$1" | tr Q "\\015" >"$1".new && + mv -f "$1".new "$1" + EOF +' + +test_expect_success 'cleanup commit messages (scissors option,-F,-e, CR/LF line endings)' ' + + test_config core.editor "\"$PWD/lf-to-crlf.sh\"" && + scissors="# ------------------------ >8 ------------------------" && + + test_write_lines >text \ + "# Keep this comment" "" " $scissors" \ + "# Keep this comment, too" "$scissors" \ + "# Remove this comment" "$scissors" \ + "Remove this comment, too" && + + test_write_lines >expect \ + "# Keep this comment" "" " $scissors" \ + "# Keep this comment, too" && + + git commit --cleanup=scissors -e -F text --allow-empty && + git cat-file -p HEAD >raw && + sed -e "1,/^\$/d" raw >actual && + test_cmp expect actual +' + +test_expect_success 'cleanup commit messages (scissors option,-F,-e, scissors on first line, CR/LF line endings)' ' + + scissors="# ------------------------ >8 ------------------------" && + test_write_lines >text \ + "$scissors" \ + "# Remove this comment and any following lines" && + cp text /tmp/test2-text && + git commit --cleanup=scissors -e -F text --allow-empty --allow-empty-message && + git cat-file -p HEAD >raw && + sed -e "1,/^\$/d" raw >actual && + test_must_be_empty actual +' + test_expect_success 'cleanup commit messages (strip option,-F)' ' echo >>negative && diff --git a/t/t7606-merge-custom.sh b/t/t7606-merge-custom.sh index 81fb7c474c14c1..8197a1c46bb5b6 100755 --- a/t/t7606-merge-custom.sh +++ b/t/t7606-merge-custom.sh @@ -23,7 +23,7 @@ test_expect_success 'set up custom strategy' ' EOF chmod +x git-merge-theirs && - PATH=.:$PATH && + PATH=.$PATH_SEP$PATH && export PATH ' diff --git a/t/t7811-grep-open.sh b/t/t7811-grep-open.sh index 1dd07141a7df9f..668de47c42e226 100755 --- a/t/t7811-grep-open.sh +++ b/t/t7811-grep-open.sh @@ -53,7 +53,7 @@ test_expect_success SIMPLEPAGER 'git grep -O' ' EOF echo grep.h >expect.notless && - PATH=.:$PATH git grep -O GREP_PATTERN >out && + PATH=.$PATH_SEP$PATH git grep -O GREP_PATTERN >out && { test_cmp expect.less pager-args || test_cmp expect.notless pager-args diff --git a/t/t9003-help-autocorrect.sh b/t/t9003-help-autocorrect.sh index f00deaf3815f3b..982aac7d169f55 100755 --- a/t/t9003-help-autocorrect.sh +++ b/t/t9003-help-autocorrect.sh @@ -12,7 +12,7 @@ test_expect_success 'setup' ' echo distimdistim was called EOF - PATH="$PATH:." && + PATH="$PATH$PATH_SEP." && export PATH && git commit --allow-empty -m "a single log entry" && diff --git a/t/t9200-git-cvsexportcommit.sh b/t/t9200-git-cvsexportcommit.sh index c5946cb0b8a94c..450ff1c2bc773d 100755 --- a/t/t9200-git-cvsexportcommit.sh +++ b/t/t9200-git-cvsexportcommit.sh @@ -11,6 +11,13 @@ if ! test_have_prereq PERL; then test_done fi +case "$PWD" in +*:*) + skip_all='cvs would get confused by the colon in `pwd`; skipping tests' + test_done + ;; +esac + cvs >/dev/null 2>&1 if test $? -ne 1 then @@ -55,8 +62,8 @@ test_expect_success \ 'mkdir A B C D E F && echo hello1 >A/newfile1.txt && echo hello2 >B/newfile2.txt && - cp "$TEST_DIRECTORY"/test-binary-1.png C/newfile3.png && - cp "$TEST_DIRECTORY"/test-binary-1.png D/newfile4.png && + cp "$TEST_DIRECTORY"/lib-diff/test-binary-1.png C/newfile3.png && + cp "$TEST_DIRECTORY"/lib-diff/test-binary-1.png D/newfile4.png && git add A/newfile1.txt && git add B/newfile2.txt && git add C/newfile3.png && @@ -81,8 +88,8 @@ test_expect_success \ rm -f B/newfile2.txt && rm -f C/newfile3.png && echo Hello5 >E/newfile5.txt && - cp "$TEST_DIRECTORY"/test-binary-2.png D/newfile4.png && - cp "$TEST_DIRECTORY"/test-binary-1.png F/newfile6.png && + cp "$TEST_DIRECTORY"/lib-diff/test-binary-2.png D/newfile4.png && + cp "$TEST_DIRECTORY"/lib-diff/test-binary-1.png F/newfile6.png && git add E/newfile5.txt && git add F/newfile6.png && git commit -a -m "Test: Remove, add and update" && @@ -170,7 +177,7 @@ test_expect_success \ 'mkdir "G g" && echo ok then >"G g/with spaces.txt" && git add "G g/with spaces.txt" && \ - cp "$TEST_DIRECTORY"/test-binary-1.png "G g/with spaces.png" && \ + cp "$TEST_DIRECTORY"/lib-diff/test-binary-1.png "G g/with spaces.png" && \ git add "G g/with spaces.png" && git commit -a -m "With spaces" && id=$(git rev-list --max-count=1 HEAD) && @@ -182,7 +189,8 @@ test_expect_success \ test_expect_success \ 'Update file with spaces in file name' \ 'echo Ok then >>"G g/with spaces.txt" && - cat "$TEST_DIRECTORY"/test-binary-1.png >>"G g/with spaces.png" && \ + cat "$TEST_DIRECTORY"/lib-diff/test-binary-1.png \ + >>"G g/with spaces.png" && \ git add "G g/with spaces.png" && git commit -a -m "Update with spaces" && id=$(git rev-list --max-count=1 HEAD) && @@ -207,7 +215,7 @@ test_expect_success !MINGW \ 'mkdir -p Å/goo/a/b/c/d/e/f/g/h/i/j/k/l/m/n/o/p/q/r/s/t/u/v/w/x/y/z/å/ä/ö && echo Foo >Å/goo/a/b/c/d/e/f/g/h/i/j/k/l/m/n/o/p/q/r/s/t/u/v/w/x/y/z/å/ä/ö/gårdetsågårdet.txt && git add Å/goo/a/b/c/d/e/f/g/h/i/j/k/l/m/n/o/p/q/r/s/t/u/v/w/x/y/z/å/ä/ö/gårdetsågårdet.txt && - cp "$TEST_DIRECTORY"/test-binary-1.png Å/goo/a/b/c/d/e/f/g/h/i/j/k/l/m/n/o/p/q/r/s/t/u/v/w/x/y/z/å/ä/ö/gårdetsågårdet.png && + cp "$TEST_DIRECTORY"/lib-diff/test-binary-1.png Å/goo/a/b/c/d/e/f/g/h/i/j/k/l/m/n/o/p/q/r/s/t/u/v/w/x/y/z/å/ä/ö/gårdetsågårdet.png && git add Å/goo/a/b/c/d/e/f/g/h/i/j/k/l/m/n/o/p/q/r/s/t/u/v/w/x/y/z/å/ä/ö/gårdetsågårdet.png && git commit -a -m "Går det så går det" && \ id=$(git rev-list --max-count=1 HEAD) && diff --git a/t/t9350-fast-export.sh b/t/t9350-fast-export.sh index ff21a12ee6e4eb..139151a79d11bc 100755 --- a/t/t9350-fast-export.sh +++ b/t/t9350-fast-export.sh @@ -791,4 +791,15 @@ test_expect_success 'fast-export --first-parent outputs all revisions output by ) ' +cat > expected << EOF +reset refs/heads/master +from $(git rev-parse master) + +EOF + +test_expect_failure 'refs are updated even if no commits need to be exported' ' + git fast-export master..master > actual && + test_cmp expected actual +' + test_done diff --git a/t/t9800-git-p4-basic.sh b/t/t9800-git-p4-basic.sh index dc88d0e064931a..8a5233e3af800e 100755 --- a/t/t9800-git-p4-basic.sh +++ b/t/t9800-git-p4-basic.sh @@ -286,7 +286,7 @@ test_expect_success 'exit when p4 fails to produce marshaled output' ' EOF chmod 755 badp4dir/p4 && ( - PATH="$TRASH_DIRECTORY/badp4dir:$PATH" && + PATH="$TRASH_DIRECTORY/badp4dir$PATH_SEP$PATH" && export PATH && test_expect_code 1 git p4 clone --dest="$git" //depot >errs 2>&1 ) && diff --git a/t/t9902-completion.sh b/t/t9902-completion.sh index 43de868b8005d3..d1e93f9529adbc 100755 --- a/t/t9902-completion.sh +++ b/t/t9902-completion.sh @@ -126,12 +126,7 @@ invalid_variable_name='${foo.bar}' actual="$TRASH_DIRECTORY/actual" -if test_have_prereq MINGW -then - ROOT="$(pwd -W)" -else - ROOT="$(pwd)" -fi +ROOT="$(pwd)" test_expect_success 'setup for __git_find_repo_path/__gitdir tests' ' mkdir -p subdir/subsubdir && diff --git a/t/test-lib-functions.sh b/t/test-lib-functions.sh index c6479f24eb5ac2..1c77050849e8ca 100644 --- a/t/test-lib-functions.sh +++ b/t/test-lib-functions.sh @@ -1153,7 +1153,7 @@ test_expect_code () { test_cmp () { test "$#" -ne 2 && BUG "2 param" - eval "$GIT_TEST_CMP" '"$@"' + GIT_ALLOC_LIMIT=0 eval "$GIT_TEST_CMP" '"$@"' } # Check that the given config key has the expected value. @@ -1454,72 +1454,6 @@ test_skip_or_die () { error "$2" } -# The following mingw_* functions obey POSIX shell syntax, but are actually -# bash scripts, and are meant to be used only with bash on Windows. - -# A test_cmp function that treats LF and CRLF equal and avoids to fork -# diff when possible. -mingw_test_cmp () { - # Read text into shell variables and compare them. If the results - # are different, use regular diff to report the difference. - local test_cmp_a= test_cmp_b= - - # When text came from stdin (one argument is '-') we must feed it - # to diff. - local stdin_for_diff= - - # Since it is difficult to detect the difference between an - # empty input file and a failure to read the files, we go straight - # to diff if one of the inputs is empty. - if test -s "$1" && test -s "$2" - then - # regular case: both files non-empty - mingw_read_file_strip_cr_ test_cmp_a <"$1" - mingw_read_file_strip_cr_ test_cmp_b <"$2" - elif test -s "$1" && test "$2" = - - then - # read 2nd file from stdin - mingw_read_file_strip_cr_ test_cmp_a <"$1" - mingw_read_file_strip_cr_ test_cmp_b - stdin_for_diff='<<<"$test_cmp_b"' - elif test "$1" = - && test -s "$2" - then - # read 1st file from stdin - mingw_read_file_strip_cr_ test_cmp_a - mingw_read_file_strip_cr_ test_cmp_b <"$2" - stdin_for_diff='<<<"$test_cmp_a"' - fi - test -n "$test_cmp_a" && - test -n "$test_cmp_b" && - test "$test_cmp_a" = "$test_cmp_b" || - eval "diff -u \"\$@\" $stdin_for_diff" -} - -# $1 is the name of the shell variable to fill in -mingw_read_file_strip_cr_ () { - # Read line-wise using LF as the line separator - # and use IFS to strip CR. - local line - while : - do - if IFS=$'\r' read -r -d $'\n' line - then - # good - line=$line$'\n' - else - # we get here at EOF, but also if the last line - # was not terminated by LF; in the latter case, - # some text was read - if test -z "$line" - then - # EOF, really - break - fi - fi - eval "$1=\$$1\$line" - done -} - # Like "env FOO=BAR some-program", but run inside a subshell, which means # it also works for shell functions (though those functions cannot impact # the environment outside of the test_env invocation). diff --git a/t/test-lib.sh b/t/test-lib.sh index a65df2fd220465..3d998d998e6cd3 100644 --- a/t/test-lib.sh +++ b/t/test-lib.sh @@ -15,6 +15,15 @@ # You should have received a copy of the GNU General Public License # along with this program. If not, see http://www.gnu.org/licenses/ . +# On Unix/Linux, the path separator is the colon, on other systems it +# may be different, though. On Windows, for example, it is a semicolon. +# If the PATH variable contains semicolons, it is pretty safe to assume +# that the path separator is a semicolon. +case "$PATH" in +*\;*) PATH_SEP=\; ;; +*) PATH_SEP=: ;; +esac + # Test the binaries we have just built. The tests are kept in # t/ subdirectory and are run in 'trash directory' subdirectory. if test -z "$TEST_DIRECTORY" @@ -42,11 +51,30 @@ then TEST_OUTPUT_DIRECTORY=$TEST_DIRECTORY fi GIT_BUILD_DIR="${TEST_DIRECTORY%/t}" -if test "$TEST_DIRECTORY" = "$GIT_BUILD_DIR" +if test -f "$GIT_BUILD_DIR/GIT-BUILD-DIR" +then + GIT_BUILD_DIR="$(cat "$GIT_BUILD_DIR/GIT-BUILD-DIR")" || exit 1 + # On Windows, we must convert Windows paths lest they contain a colon + case "$(uname -s)" in + *MINGW*) + GIT_BUILD_DIR="$(cygpath -au "$GIT_BUILD_DIR")" + ;; + esac +elif test "$TEST_DIRECTORY" = "$GIT_BUILD_DIR" then echo "PANIC: Running in a $TEST_DIRECTORY that doesn't end in '/t'?" >&2 exit 1 fi +if test -f "$GIT_BUILD_DIR/GIT-BUILD-DIR" +then + GIT_BUILD_DIR="$(cat "$GIT_BUILD_DIR/GIT-BUILD-DIR")" || exit 1 + # On Windows, we must convert Windows paths lest they contain a colon + case "$(uname -s)" in + *MINGW*) + GIT_BUILD_DIR="$(cygpath -au "$GIT_BUILD_DIR")" + ;; + esac +fi # Prepend a string to a VAR using an arbitrary ":" delimiter, not # adding the delimiter if VAR or VALUE is empty. I.e. a generalized: @@ -486,23 +514,18 @@ EDITOR=: # /usr/xpg4/bin/sh and /bin/ksh to bail out. So keep the unsets # deriving from the command substitution clustered with the other # ones. -unset VISUAL EMAIL LANGUAGE $("$PERL_PATH" -e ' - my @env = keys %ENV; - my $ok = join("|", qw( - TRACE - DEBUG - TEST - .*_TEST - PROVE - VALGRIND - UNZIP - PERF_ - CURL_VERBOSE - TRACE_CURL - )); - my @vars = grep(/^GIT_/ && !/^GIT_($ok)/o, @env); - print join("\n", @vars); -') +unset VISUAL EMAIL LANGUAGE $(env | sed -n \ + -e '/^GIT_TRACE/d' \ + -e '/^GIT_DEBUG/d' \ + -e '/^GIT_TEST/d' \ + -e '/^GIT_.*_TEST/d' \ + -e '/^GIT_PROVE/d' \ + -e '/^GIT_VALGRIND/d' \ + -e '/^GIT_UNZIP/d' \ + -e '/^GIT_PERF_/d' \ + -e '/^GIT_CURL_VERBOSE/d' \ + -e '/^GIT_TRACE_CURL/d' \ + -e 's/^\(GIT_[^=]*\)=.*/\1/p') unset XDG_CACHE_HOME unset XDG_CONFIG_HOME unset GITPERLLIB @@ -1447,7 +1470,7 @@ then done done IFS=$OLDIFS - PATH=$GIT_VALGRIND/bin:$PATH + PATH=$GIT_VALGRIND/bin$PATH_SEP$PATH GIT_EXEC_PATH=$GIT_VALGRIND/bin export GIT_VALGRIND GIT_VALGRIND_MODE="$valgrind" @@ -1459,7 +1482,7 @@ elif test -n "$GIT_TEST_INSTALLED" then GIT_EXEC_PATH=$($GIT_TEST_INSTALLED/git --exec-path) || error "Cannot run git from $GIT_TEST_INSTALLED." - PATH=$GIT_TEST_INSTALLED:$GIT_BUILD_DIR/t/helper:$PATH + PATH=$GIT_TEST_INSTALLED$PATH_SEP$GIT_BUILD_DIR/t/helper$PATH_SEP$PATH GIT_EXEC_PATH=${GIT_TEST_EXEC_PATH:-$GIT_EXEC_PATH} else # normal case, use ../bin-wrappers only unless $with_dashes: if test -n "$no_bin_wrappers" @@ -1475,12 +1498,12 @@ else # normal case, use ../bin-wrappers only unless $with_dashes: fi with_dashes=t fi - PATH="$git_bin_dir:$PATH" + PATH="$git_bin_dir$PATH_SEP$PATH" fi GIT_EXEC_PATH=$GIT_BUILD_DIR if test -n "$with_dashes" then - PATH="$GIT_BUILD_DIR:$GIT_BUILD_DIR/t/helper:$PATH" + PATH="$GIT_BUILD_DIR$PATH_SEP$GIT_BUILD_DIR/t/helper$PATH_SEP$PATH" fi fi GIT_TEMPLATE_DIR="$GIT_BUILD_DIR"/templates/blt @@ -1690,17 +1713,30 @@ fi uname_s=$(uname -s) case $uname_s in *MINGW*) - # Windows has its own (incompatible) sort and find - sort () { - /usr/bin/sort "$@" - } - find () { - /usr/bin/find "$@" - } - # git sees Windows-style pwd - pwd () { - builtin pwd -W - } + if test -x /usr/bin/sort + then + # Windows has its own (incompatible) sort; override + sort () { + /usr/bin/sort "$@" + } + fi + if test -x /usr/bin/find + then + # Windows has its own (incompatible) find; override + find () { + /usr/bin/find "$@" + } + fi + # On Windows, Git wants Windows paths. But /usr/bin/pwd spits out + # Unix-style paths. At least in Bash, we have a builtin pwd that + # understands the -W option to force "mixed" paths, i.e. with drive + # prefix but still with forward slashes. Let's use that, if available. + if type builtin >/dev/null 2>&1 + then + pwd () { + builtin pwd -W + } + fi # no POSIX permissions # backslashes in pathspec are converted to '/' # exec does not inherit the PID @@ -1709,7 +1745,13 @@ case $uname_s in test_set_prereq SED_STRIPS_CR test_set_prereq GREP_STRIPS_CR test_set_prereq WINDOWS - GIT_TEST_CMP=mingw_test_cmp + GIT_TEST_CMP="test-tool cmp" + if ! type iconv >/dev/null 2>&1 + then + iconv () { + test-tool iconv "$@" + } + fi ;; *CYGWIN*) test_set_prereq POSIXPERM @@ -1867,6 +1909,10 @@ test_lazy_prereq UNZIP ' test $? -ne 127 ' +test_lazy_prereq BUSYBOX ' + case "$($SHELL --help 2>&1)" in *BusyBox*) true;; *) false;; esac +' + run_with_limited_cmdline () { (ulimit -s 128 && "$@") } diff --git a/transport-helper.c b/transport-helper.c index e95267a4ab54dc..eada58c8ef0391 100644 --- a/transport-helper.c +++ b/transport-helper.c @@ -16,6 +16,8 @@ #include "protocol.h" static int debug; +/* TODO: put somewhere sensible, e.g. git_transport_options? */ +static int auto_gc = 1; struct helper_data { const char *name; @@ -478,10 +480,25 @@ static int get_exporter(struct transport *transport, for (i = 0; i < revlist_args->nr; i++) strvec_push(&fastexport->args, revlist_args->items[i].string); + strvec_push(&fastexport->args, "--"); + fastexport->git_cmd = 1; return start_command(fastexport); } +static void check_helper_status(struct helper_data *data) +{ + int pid, status; + + pid = waitpid(data->helper->pid, &status, WNOHANG); + if (pid < 0) + die("Could not retrieve status of remote helper '%s'", + data->name); + if (pid > 0 && WIFEXITED(status)) + die("Remote helper '%s' died with %d", + data->name, WEXITSTATUS(status)); +} + static int fetch_with_import(struct transport *transport, int nr_heads, struct ref **to_fetch) { @@ -518,6 +535,7 @@ static int fetch_with_import(struct transport *transport, if (finish_command(&fastimport)) die(_("error while running fast-import")); + check_helper_status(data); /* * The fast-import stream of a remote helper that advertises @@ -551,6 +569,12 @@ static int fetch_with_import(struct transport *transport, } } strbuf_release(&buf); + if (auto_gc) { + const char *argv_gc_auto[] = { + "gc", "--auto", "--quiet", NULL, + }; + run_command_v_opt(argv_gc_auto, RUN_GIT_CMD); + } return 0; } @@ -1122,6 +1146,7 @@ static int push_refs_with_export(struct transport *transport, if (finish_command(&exporter)) die(_("error while running fast-export")); + check_helper_status(data); if (push_update_refs_status(data, remote_refs, flags)) return 1; diff --git a/unpack-trees.c b/unpack-trees.c index bae812156c4fed..98945c53e9792f 100644 --- a/unpack-trees.c +++ b/unpack-trees.c @@ -1797,7 +1797,9 @@ static void mark_new_skip_worktree(struct pattern_list *pl, * 2. Widen worktree according to sparse-checkout file. * Matched entries will have skip_wt_flag cleared (i.e. "in") */ + enable_fscache(istate->cache_nr); clear_ce_flags(istate, select_flag, skip_wt_flag, pl, show_progress); + disable_fscache(); } static void populate_from_existing_patterns(struct unpack_trees_options *o, diff --git a/wt-status.c b/wt-status.c index 5813174896cc9a..eebf26370753c2 100644 --- a/wt-status.c +++ b/wt-status.c @@ -22,7 +22,7 @@ #define AB_DELAY_WARNING_IN_MS (2 * 1000) static const char cut_line[] = -"------------------------ >8 ------------------------\n"; +"------------------------ >8 ------------------------"; static char default_wt_status_colors[][COLOR_MAXLEN] = { GIT_COLOR_NORMAL, /* WT_STATUS_HEADER */ @@ -1070,15 +1070,22 @@ static void wt_longstatus_print_other(struct wt_status *s, status_printf_ln(s, GIT_COLOR_NORMAL, "%s", ""); } +static inline int starts_with_newline(const char *p) +{ + return *p == '\n' || (*p == '\r' && p[1] == '\n'); +} + size_t wt_status_locate_end(const char *s, size_t len) { const char *p; struct strbuf pattern = STRBUF_INIT; strbuf_addf(&pattern, "\n%c %s", comment_line_char, cut_line); - if (starts_with(s, pattern.buf + 1)) + if (starts_with(s, pattern.buf + 1) && + starts_with_newline(s + pattern.len - 1)) len = 0; - else if ((p = strstr(s, pattern.buf))) + else if ((p = strstr(s, pattern.buf)) && + starts_with_newline(p + pattern.len)) len = p - s + 1; strbuf_release(&pattern); return len;