[llvm-objcopy][ELF] Add an option to remove notes (#118739)

This adds an option `--remove-note=[name/]type` to selectively delete
notes in ELF files, where `type` is the numeric value of the note type
and `name` is the name of the originator. The name can be omitted, in
which case all notes of the specified type will be removed. For now,
only `SHT_NOTE` sections that are not associated with segments are
handled. The implementation can be extended later as needed.


RFC: https://discourse.llvm.org/t/rfc-llvm-objcopy-feature-for-editing-notes/83491

Author: igorkudrin

llvm/docs/CommandGuide/llvm-objcopy.rst

@@ -477,6 +477,11 @@ them.
 
  Preserve access and modification timestamps in the output.
 
+.. option:: --remove-note [<name>/]<type>
+
+ Remove notes of integer type ``<type>`` and name ``<name>`` from SHT_NOTE
+ sections that are not in a segment. Can be specified multiple times.
+
 .. option:: --rename-section <old>=<new>[,<flag>,...]
 
  Rename sections called ``<old>`` to ``<new>`` in the output, and apply any

llvm/include/llvm/ObjCopy/CommonConfig.h

@@ -281,6 +281,11 @@ struct CommonConfig {
 
   SmallVector<std::pair<NameMatcher, llvm::DebugCompressionType>, 0>
       compressSections;
+
+  // ErrorCallback is used to handle recoverable errors. An Error returned
+  // by the callback aborts the execution and is then returned to the caller.
+  // If the callback is not set, the errors are not issued.
+  std::function<Error(Error)> ErrorCallback;
 };
 
 } // namespace objcopy

llvm/include/llvm/ObjCopy/ELF/ELFConfig.h

@@ -15,6 +15,12 @@
 namespace llvm {
 namespace objcopy {
 
+// Note to remove info specified by --remove-note option.
+struct RemoveNoteInfo {
+  StringRef Name;
+  uint32_t TypeId;
+};
+
 // ELF specific configuration for copying/stripping a single file.
 struct ELFConfig {
   uint8_t NewSymbolVisibility = (uint8_t)ELF::STV_DEFAULT;
@@ -31,6 +37,9 @@ struct ELFConfig {
   bool KeepFileSymbols = false;
   bool LocalizeHidden = false;
   bool VerifyNoteSections = true;
+
+  // Notes specified by --remove-note option.
+  SmallVector<RemoveNoteInfo, 0> NotesToRemove;
 };
 
 } // namespace objcopy

llvm/lib/ObjCopy/ELF/ELFObjcopy.cpp

@@ -609,6 +609,112 @@ static void addSymbol(Object &Obj, const NewSymbolInfo &SymInfo,
       Sec ? (uint16_t)SYMBOL_SIMPLE_INDEX : (uint16_t)SHN_ABS, 0);
 }
 
+namespace {
+struct RemoveNoteDetail {
+  struct DeletedRange {
+    uint64_t OldFrom;
+    uint64_t OldTo;
+  };
+
+  template <class ELFT>
+  static std::vector<DeletedRange>
+  findNotesToRemove(ArrayRef<uint8_t> Data, size_t Align,
+                    ArrayRef<RemoveNoteInfo> NotesToRemove);
+  static std::vector<uint8_t> updateData(ArrayRef<uint8_t> OldData,
+                                         ArrayRef<DeletedRange> ToRemove);
+};
+} // namespace
+
+template <class ELFT>
+std::vector<RemoveNoteDetail::DeletedRange>
+RemoveNoteDetail::findNotesToRemove(ArrayRef<uint8_t> Data, size_t Align,
+                                    ArrayRef<RemoveNoteInfo> NotesToRemove) {
+  LLVM_ELF_IMPORT_TYPES_ELFT(ELFT);
+  std::vector<DeletedRange> ToRemove;
+  uint64_t CurPos = 0;
+  while (CurPos + sizeof(Elf_Nhdr) <= Data.size()) {
+    auto Nhdr = reinterpret_cast<const Elf_Nhdr *>(Data.data() + CurPos);
+    size_t FullSize = Nhdr->getSize(Align);
+    if (CurPos + FullSize > Data.size())
+      break;
+    Elf_Note Note(*Nhdr);
+    bool ShouldRemove =
+        llvm::any_of(NotesToRemove, [&Note](const RemoveNoteInfo &NoteInfo) {
+          return NoteInfo.TypeId == Note.getType() &&
+                 (NoteInfo.Name.empty() || NoteInfo.Name == Note.getName());
+        });
+    if (ShouldRemove)
+      ToRemove.push_back({CurPos, CurPos + FullSize});
+    CurPos += FullSize;
+  }
+  return ToRemove;
+}
+
+std::vector<uint8_t>
+RemoveNoteDetail::updateData(ArrayRef<uint8_t> OldData,
+                             ArrayRef<DeletedRange> ToRemove) {
+  std::vector<uint8_t> NewData;
+  NewData.reserve(OldData.size());
+  uint64_t CurPos = 0;
+  for (const DeletedRange &RemRange : ToRemove) {
+    if (CurPos < RemRange.OldFrom) {
+      auto Slice = OldData.slice(CurPos, RemRange.OldFrom - CurPos);
+      NewData.insert(NewData.end(), Slice.begin(), Slice.end());
+    }
+    CurPos = RemRange.OldTo;
+  }
+  if (CurPos < OldData.size()) {
+    auto Slice = OldData.slice(CurPos);
+    NewData.insert(NewData.end(), Slice.begin(), Slice.end());
+  }
+  return NewData;
+}
+
+static Error removeNotes(Object &Obj, endianness Endianness,
+                         ArrayRef<RemoveNoteInfo> NotesToRemove,
+                         function_ref<Error(Error)> ErrorCallback) {
+  // TODO: Support note segments.
+  if (ErrorCallback) {
+    for (Segment &Seg : Obj.segments()) {
+      if (Seg.Type == PT_NOTE) {
+        if (Error E = ErrorCallback(createStringError(
+                errc::not_supported, "note segments are not supported")))
+          return E;
+        break;
+      }
+    }
+  }
+  for (auto &Sec : Obj.sections()) {
+    if (Sec.Type != SHT_NOTE || !Sec.hasContents())
+      continue;
+    // TODO: Support note sections in segments.
+    if (Sec.ParentSegment) {
+      if (ErrorCallback)
+        if (Error E = ErrorCallback(createStringError(
+                errc::not_supported,
+                "cannot remove note(s) from " + Sec.Name +
+                    ": sections in segments are not supported")))
+          return E;
+      continue;
+    }
+    ArrayRef<uint8_t> OldData = Sec.getContents();
+    size_t Align = std::max<size_t>(4, Sec.Align);
+    // Note: notes for both 32-bit and 64-bit ELF files use 4-byte words in the
+    // header, so the parsers are the same.
+    auto ToRemove = (Endianness == endianness::little)
+                        ? RemoveNoteDetail::findNotesToRemove<ELF64LE>(
+                              OldData, Align, NotesToRemove)
+                        : RemoveNoteDetail::findNotesToRemove<ELF64BE>(
+                              OldData, Align, NotesToRemove);
+    if (!ToRemove.empty()) {
+      if (Error E = Obj.updateSectionData(
+              Sec, RemoveNoteDetail::updateData(OldData, ToRemove)))
+        return E;
+    }
+  }
+  return Error::success();
+}
+
 static Error
 handleUserSection(const NewSectionInfo &NewSection,
                   function_ref<Error(StringRef, ArrayRef<uint8_t>)> F) {
@@ -799,6 +905,12 @@ static Error handleArgs(const CommonConfig &Config, const ELFConfig &ELFConfig,
                      ? endianness::little
                      : endianness::big;
 
+  if (!ELFConfig.NotesToRemove.empty()) {
+    if (Error Err =
+            removeNotes(Obj, E, ELFConfig.NotesToRemove, Config.ErrorCallback))
+      return Err;
+  }
+
   for (const NewSectionInfo &AddedSection : Config.AddSection) {
     auto AddSection = [&](StringRef Name, ArrayRef<uint8_t> Data) -> Error {
       OwnedDataSection &NewSection =

llvm/lib/ObjCopy/ELF/ELFObject.cpp

@@ -2154,37 +2154,46 @@ ELFWriter<ELFT>::ELFWriter(Object &Obj, raw_ostream &Buf, bool WSH,
     : Writer(Obj, Buf), WriteSectionHeaders(WSH && Obj.HadShdrs),
       OnlyKeepDebug(OnlyKeepDebug) {}
 
-Error Object::updateSection(StringRef Name, ArrayRef<uint8_t> Data) {
-  auto It = llvm::find_if(Sections,
-                          [&](const SecPtr &Sec) { return Sec->Name == Name; });
-  if (It == Sections.end())
-    return createStringError(errc::invalid_argument, "section '%s' not found",
-                             Name.str().c_str());
-
-  auto *OldSec = It->get();
-  if (!OldSec->hasContents())
+Error Object::updateSectionData(SecPtr &Sec, ArrayRef<uint8_t> Data) {
+  if (!Sec->hasContents())
     return createStringError(
         errc::invalid_argument,
         "section '%s' cannot be updated because it does not have contents",
-        Name.str().c_str());
+        Sec->Name.c_str());
 
-  if (Data.size() > OldSec->Size && OldSec->ParentSegment)
+  if (Data.size() > Sec->Size && Sec->ParentSegment)
     return createStringError(errc::invalid_argument,
                              "cannot fit data of size %zu into section '%s' "
                              "with size %" PRIu64 " that is part of a segment",
-                             Data.size(), Name.str().c_str(), OldSec->Size);
+                             Data.size(), Sec->Name.c_str(), Sec->Size);
 
-  if (!OldSec->ParentSegment) {
-    *It = std::make_unique<OwnedDataSection>(*OldSec, Data);
+  if (!Sec->ParentSegment) {
+    Sec = std::make_unique<OwnedDataSection>(*Sec, Data);
   } else {
     // The segment writer will be in charge of updating these contents.
-    OldSec->Size = Data.size();
-    UpdatedSections[OldSec] = Data;
+    Sec->Size = Data.size();
+    UpdatedSections[Sec.get()] = Data;
   }
 
   return Error::success();
 }
 
+Error Object::updateSection(StringRef Name, ArrayRef<uint8_t> Data) {
+  auto It = llvm::find_if(Sections,
+                          [&](const SecPtr &Sec) { return Sec->Name == Name; });
+  if (It == Sections.end())
+    return createStringError(errc::invalid_argument, "section '%s' not found",
+                             Name.str().c_str());
+  return updateSectionData(*It, Data);
+}
+
+Error Object::updateSectionData(SectionBase &S, ArrayRef<uint8_t> Data) {
+  auto It = llvm::find_if(Sections,
+                          [&](const SecPtr &Sec) { return Sec.get() == &S; });
+  assert(It != Sections.end() && "The section should belong to the object");
+  return updateSectionData(*It, Data);
+}
+
 Error Object::removeSections(
     bool AllowBrokenLinks, std::function<bool(const SectionBase &)> ToRemove) {
 

llvm/lib/ObjCopy/ELF/ELFObject.h

@@ -549,6 +549,7 @@ class SectionBase {
   virtual void
   replaceSectionReferences(const DenseMap<SectionBase *, SectionBase *> &);
   virtual bool hasContents() const { return false; }
+  virtual ArrayRef<uint8_t> getContents() const { return {}; }
   // Notify the section that it is subject to removal.
   virtual void onRemove();
 
@@ -619,6 +620,8 @@ class Section : public SectionBase {
   bool hasContents() const override {
     return Type != ELF::SHT_NOBITS && Type != ELF::SHT_NULL;
   }
+  ArrayRef<uint8_t> getContents() const override { return Contents; }
+
   void restoreSymTabLink(SymbolTableSection &SymTab) override;
 };
 
@@ -654,6 +657,7 @@ class OwnedDataSection : public SectionBase {
   Error accept(SectionVisitor &Sec) const override;
   Error accept(MutableSectionVisitor &Visitor) override;
   bool hasContents() const override { return true; }
+  ArrayRef<uint8_t> getContents() const override { return Data; }
 };
 
 class CompressedSection : public SectionBase {
@@ -1164,6 +1168,8 @@ class Object {
     return Sec.Flags & ELF::SHF_ALLOC;
   };
 
+  Error updateSectionData(SecPtr &Sec, ArrayRef<uint8_t> Data);
+
 public:
   template <class T>
   using ConstRange = iterator_range<pointee_iterator<
@@ -1206,6 +1212,7 @@ class Object {
 
   const auto &getUpdatedSections() const { return UpdatedSections; }
   Error updateSection(StringRef Name, ArrayRef<uint8_t> Data);
+  Error updateSectionData(SectionBase &S, ArrayRef<uint8_t> Data);
 
   SectionBase *findSection(StringRef Name) {
     auto SecIt =