[llvm][cas] Fix llvm-cas --ingest symlink issues

* Allow ingesting a symlink to a non-existent file. While this may be
  unexpected, it is not a filesystem error and we should not assume it
  will not occur in practice in llvm-cas.
* Avoid walking into directories repeatedly when there are symlinks,
  which could cause stack overflow from recursion or a filesystem error
  about a path with too many symlinks.

(cherry picked from commit 23a1a10)

Author: benlangmuir

llvm/test/tools/llvm-cas/Inputs/broken_symlink

@@ -0,0 +1 @@
+missing

llvm/test/tools/llvm-cas/Inputs/self

@@ -0,0 +1 @@
+.

llvm/test/tools/llvm-cas/ingest-remap.test

@@ -3,6 +3,8 @@ RUN: rm -rf %t && mkdir -p %t
 RUN: llvm-cas --cas %t/cas --ingest --prefix-map %S/Inputs=/^input --data %S/Inputs > %t/cas.id
 RUN: llvm-cas --cas %t/cas --ls-tree-recursive @%t/cas.id | FileCheck %s
 
+CHECK: syml
+CHECK-SAME: /^input/broken_symlink -> missing
 CHECK: file
 CHECK-SAME: /^input/directory/file
 CHECK: syml
@@ -12,4 +14,6 @@ CHECK-SAME: /^input/oneline
 CHECK: file
 CHECK-SAME: /^input/oneline-nonewline
 CHECK: syml
-CHECK-SAME: /^input/sym_dir
+CHECK-SAME: /^input/self -> .
+CHECK: syml
+CHECK-SAME: /^input/sym_dir -> directory

llvm/test/tools/llvm-cas/ingest.test

@@ -4,6 +4,8 @@ RUN: mkdir %t
 RUN: llvm-cas --cas %t/cas --ingest --data %S/Inputs > %t/cas.id
 RUN: llvm-cas --cas %t/cas --ls-tree-recursive @%t/cas.id | FileCheck %s
 
+CHECK: syml
+CHECK-SAME: broken_symlink -> missing
 CHECK: file
 CHECK-SAME: directory/file
 CHECK: syml
@@ -13,7 +15,9 @@ CHECK-SAME: oneline
 CHECK: file
 CHECK-SAME: oneline-nonewline
 CHECK: syml
-CHECK-SAME: sym_dir
+CHECK-SAME: self -> .
+CHECK: syml
+CHECK-SAME: sym_dir -> directory
 
 RUN: llvm-cas --cas %t/cas --get-cas-id --data %S/Inputs/directory/file @%t/cas.id > %t/file.casid
 RUN: llvm-cas --cas %t/cas --cat-blob @%t/file.casid | FileCheck %s --check-prefix=CHECK-TEST-FILE

llvm/tools/llvm-cas/llvm-cas.cpp

@@ -426,16 +426,31 @@ int traverseGraph(ObjectStore &CAS, const CASID &ID) {
   return 0;
 }
 
-static Error recursiveAccess(CachingOnDiskFileSystem &FS, StringRef Path) {
+static Error
+recursiveAccess(CachingOnDiskFileSystem &FS, StringRef Path,
+                llvm::DenseSet<llvm::sys::fs::UniqueID> &SeenDirectories) {
   auto ST = FS.status(Path);
+
+  // Ignore missing entries, which can be a symlink to a missing file, which is
+  // not an error in the filesystem itself.
+  // FIXME: add status(follow=false) to VFS instead, which would let us detect
+  // this case directly.
+  if (ST.getError() == llvm::errc::no_such_file_or_directory)
+    return Error::success();
+
   if (!ST)
     return createFileError(Path, ST.getError());
 
-  if (ST->isDirectory()) {
+  // Check that this is the first time we see the directory to prevent infinite
+  // recursion into symlinks. The status() above will ensure all symlinks are
+  // ingested.
+  // FIXME: add status(follow=false) to VFS instead, and then only traverse
+  // a directory and not a symlink to a directory.
+  if (ST->isDirectory() && SeenDirectories.insert(ST->getUniqueID()).second) {
     std::error_code EC;
     for (llvm::vfs::directory_iterator I = FS.dir_begin(Path, EC), IE;
          !EC && I != IE; I.increment(EC)) {
-      auto Err = recursiveAccess(FS, I->path());
+      auto Err = recursiveAccess(FS, I->path(), SeenDirectories);
       if (Err)
         return Err;
     }
@@ -460,7 +475,8 @@ static Expected<ObjectProxy> ingestFileSystemImpl(ObjectStore &CAS,
 
   (*FS)->trackNewAccesses();
 
-  if (Error E = recursiveAccess(**FS, Path))
+  llvm::DenseSet<llvm::sys::fs::UniqueID> SeenDirectories;
+  if (Error E = recursiveAccess(**FS, Path, SeenDirectories))
     return std::move(E);
 
   return (*FS)->createTreeFromNewAccesses(