Add sanitizer-specific GlobalValue attributes.

Plan is the migrate the global variable metadata for sanitizers, that's
currently carried around generally in the 'llvm.asan.globals' section,
onto the global variable itself.

This patch adds the attribute and plumbs it through the LLVM IR and
bitcode formats, but is a no-op other than that so far.

Reviewed By: vitalybuka, kstoimenov

Differential Revision: https://reviews.llvm.org/D126100

llvm/include/llvm/AsmParser/LLParser.h

@@ -515,6 +515,7 @@ namespace llvm {
     bool parseGlobalValueVector(SmallVectorImpl<Constant *> &Elts,
                                 Optional<unsigned> *InRangeOp = nullptr);
     bool parseOptionalComdat(StringRef GlobalName, Comdat *&C);
+    bool parseSanitizer(GlobalVariable *GV);
     bool parseMetadataAsValue(Value *&V, PerFunctionState &PFS);
     bool parseValueAsMetadata(Metadata *&MD, const Twine &TypeMsg,
                               PerFunctionState *PFS);

llvm/include/llvm/AsmParser/LLToken.h

@@ -391,6 +391,19 @@ enum Kind {
   kw_bit,
   kw_varFlags,
 
+  // GV's with __attribute__((no_sanitize("address"))), or things in
+  // -fsanitize-ignorelist when built with ASan.
+  kw_no_sanitize_address,
+  // GV's with __attribute__((no_sanitize("hwaddress"))), or things in
+  // -fsanitize-ignorelist when built with HWASan.
+  kw_no_sanitize_hwaddress,
+  // GV's with __attribute__((no_sanitize("memtag"))), or things in
+  // -fsanitize-ignorelist when built with memory tagging.
+  kw_no_sanitize_memtag,
+  // GV's where the clang++ frontend (when ASan is used) notes that this is
+  // dynamically initialized, and thus needs ODR detection.
+  kw_sanitize_address_dyninit,
+
   // Unsigned Valued tokens (UIntVal).
   LabelID,    // 42:
   GlobalID,   // @42

llvm/include/llvm/IR/GlobalValue.h

@@ -79,14 +79,15 @@ protected:
         ValueType(Ty), Visibility(DefaultVisibility),
         UnnamedAddrVal(unsigned(UnnamedAddr::None)),
         DllStorageClass(DefaultStorageClass), ThreadLocal(NotThreadLocal),
-        HasLLVMReservedName(false), IsDSOLocal(false), HasPartition(false) {
+        HasLLVMReservedName(false), IsDSOLocal(false), HasPartition(false),
+        HasSanitizerMetadata(false) {
     setLinkage(Linkage);
     setName(Name);
   }
 
   Type *ValueType;
 
-  static const unsigned GlobalValueSubClassDataBits = 16;
+  static const unsigned GlobalValueSubClassDataBits = 15;
 
   // All bitfields use unsigned as the underlying type so that MSVC will pack
   // them.
@@ -111,9 +112,14 @@ protected:
   /// https://lld.llvm.org/Partitions.html).
   unsigned HasPartition : 1;
 
+  /// True if this symbol has sanitizer metadata available. Should only happen
+  /// if sanitizers were enabled when building the translation unit which
+  /// contains this GV.
+  unsigned HasSanitizerMetadata : 1;
+
 private:
   // Give subclasses access to what otherwise would be wasted padding.
-  // (16 + 4 + 2 + 2 + 2 + 3 + 1 + 1 + 1) == 32.
+  // (15 + 4 + 2 + 2 + 2 + 3 + 1 + 1 + 1 + 1) == 32.
   unsigned SubClassData : GlobalValueSubClassDataBits;
 
   friend class Constant;
@@ -288,6 +294,39 @@ public:
   StringRef getPartition() const;
   void setPartition(StringRef Part);
 
+  // ASan, HWASan and Memtag sanitizers have some instrumentation that applies
+  // specifically to global variables. This instrumentation is implicitly
+  // applied to all global variables when built with -fsanitize=*. What we need
+  // is a way to persist the information that a certain global variable should
+  // *not* have sanitizers applied, which occurs if:
+  //   1. The global variable is in the sanitizer ignore list, or
+  //   2. The global variable is created by the sanitizers itself for internal
+  //      usage, or
+  //   3. The global variable has __attribute__((no_sanitize("..."))) or
+  //      __attribute__((disable_sanitizer_instrumentation)).
+  //
+  // This is important, a some IR passes like GlobalMerge can delete global
+  // variables and replace them with new ones. If the old variables were marked
+  // to be unsanitized, then the new ones should also be.
+  struct SanitizerMetadata {
+    SanitizerMetadata()
+        : NoAddress(false), NoHWAddress(false), NoMemtag(false),
+          IsDynInit(false) {}
+    unsigned NoAddress : 1;
+    unsigned NoHWAddress : 1;
+    unsigned NoMemtag : 1;
+
+    // ASan-specific metadata. Is this global variable dynamically initialized
+    // (from a C++ language perspective), and should therefore be checked for
+    // ODR violations.
+    unsigned IsDynInit : 1;
+  };
+
+  bool hasSanitizerMetadata() const { return HasSanitizerMetadata; }
+  const SanitizerMetadata &getSanitizerMetadata() const;
+  void setSanitizerMetadata(const SanitizerMetadata &Meta);
+  void removeSanitizerMetadata();
+
   static LinkageTypes getLinkOnceLinkage(bool ODR) {
     return ODR ? LinkOnceODRLinkage : LinkOnceAnyLinkage;
   }

llvm/lib/AsmParser/LLLexer.cpp

@@ -580,6 +580,11 @@ lltok::Kind LLLexer::LexIdentifier() {
   KEYWORD(prefix);
   KEYWORD(prologue);
 
+  KEYWORD(no_sanitize_address);
+  KEYWORD(no_sanitize_hwaddress);
+  KEYWORD(no_sanitize_memtag);
+  KEYWORD(sanitize_address_dyninit);
+
   KEYWORD(ccc);
   KEYWORD(fastcc);
   KEYWORD(coldcc);

llvm/lib/AsmParser/LLParser.cpp

@@ -1103,6 +1103,45 @@ bool LLParser::parseAliasOrIFunc(const std::string &Name, LocTy NameLoc,
   return false;
 }
 
+static bool isSanitizer(lltok::Kind Kind) {
+  switch (Kind) {
+  case lltok::kw_no_sanitize_address:
+  case lltok::kw_no_sanitize_hwaddress:
+  case lltok::kw_no_sanitize_memtag:
+  case lltok::kw_sanitize_address_dyninit:
+    return true;
+  default:
+    return false;
+  }
+}
+
+bool LLParser::parseSanitizer(GlobalVariable *GV) {
+  using SanitizerMetadata = GlobalValue::SanitizerMetadata;
+  SanitizerMetadata Meta;
+  if (GV->hasSanitizerMetadata())
+    Meta = GV->getSanitizerMetadata();
+
+  switch (Lex.getKind()) {
+  case lltok::kw_no_sanitize_address:
+    Meta.NoAddress = true;
+    break;
+  case lltok::kw_no_sanitize_hwaddress:
+    Meta.NoHWAddress = true;
+    break;
+  case lltok::kw_no_sanitize_memtag:
+    Meta.NoMemtag = true;
+    break;
+  case lltok::kw_sanitize_address_dyninit:
+    Meta.IsDynInit = true;
+    break;
+  default:
+    return tokError("non-sanitizer token passed to LLParser::parseSanitizer()");
+  }
+  GV->setSanitizerMetadata(Meta);
+  Lex.Lex();
+  return false;
+}
+
 /// parseGlobal
 ///   ::= GlobalVar '=' OptionalLinkage OptionalPreemptionSpecifier
 ///       OptionalVisibility OptionalDLLStorageClass
@@ -1221,6 +1260,9 @@ bool LLParser::parseGlobal(const std::string &Name, LocTy NameLoc,
     } else if (Lex.getKind() == lltok::MetadataVar) {
       if (parseGlobalObjectMetadataAttachment(*GV))
         return true;
+    } else if (isSanitizer(Lex.getKind())) {
+      if (parseSanitizer(GV))
+        return true;
     } else {
       Comdat *C;
       if (parseOptionalComdat(Name, C))

llvm/lib/Bitcode/Reader/BitcodeReader.cpp

@@ -3441,6 +3441,19 @@ static void inferDSOLocal(GlobalValue *GV) {
     GV->setDSOLocal(true);
 }
 
+GlobalValue::SanitizerMetadata deserializeSanitizerMetadata(unsigned V) {
+  GlobalValue::SanitizerMetadata Meta;
+  if (V & (1 << 0))
+    Meta.NoAddress = true;
+  if (V & (1 << 1))
+    Meta.NoHWAddress = true;
+  if (V & (1 << 2))
+    Meta.NoMemtag = true;
+  if (V & (1 << 3))
+    Meta.IsDynInit = true;
+  return Meta;
+}
+
 Error BitcodeReader::parseGlobalVarRecord(ArrayRef<uint64_t> Record) {
   // v1: [pointer type, isconst, initid, linkage, alignment, section,
   // visibility, threadlocal, unnamed_addr, externally_initialized,
@@ -3544,6 +3557,12 @@ Error BitcodeReader::parseGlobalVarRecord(ArrayRef<uint64_t> Record) {
   if (Record.size() > 15)
     NewGV->setPartition(StringRef(Strtab.data() + Record[14], Record[15]));
 
+  if (Record.size() > 16 && Record[16] != UINT_MAX) {
+    llvm::GlobalValue::SanitizerMetadata Meta =
+        deserializeSanitizerMetadata(Record[16]);
+    NewGV->setSanitizerMetadata(Meta);
+  }
+
   return Error::success();
 }
 

llvm/lib/Bitcode/Writer/BitcodeWriter.cpp

@@ -1223,6 +1223,14 @@ static StringEncoding getStringEncoding(StringRef Str) {
   return SE_Fixed7;
 }
 
+static_assert(sizeof(GlobalValue::SanitizerMetadata) <= sizeof(unsigned),
+              "Sanitizer Metadata is too large for naive serialization.");
+static unsigned
+serializeSanitizerMetadata(const GlobalValue::SanitizerMetadata &Meta) {
+  return Meta.NoAddress | (Meta.NoHWAddress << 1) |
+         (Meta.NoMemtag << 2) | (Meta.IsDynInit << 3);
+}
+
 /// Emit top-level description of module, including target triple, inline asm,
 /// descriptors for global variables, and function prototype info.
 /// Returns the bit offset to backpatch with the location of the real VST.
@@ -1346,7 +1354,7 @@ void ModuleBitcodeWriter::writeModuleInfo() {
     // GLOBALVAR: [strtab offset, strtab size, type, isconst, initid,
     //             linkage, alignment, section, visibility, threadlocal,
     //             unnamed_addr, externally_initialized, dllstorageclass,
-    //             comdat, attributes, DSO_Local]
+    //             comdat, attributes, DSO_Local, GlobalSanitizer]
     Vals.push_back(addToStrtab(GV.getName()));
     Vals.push_back(GV.getName().size());
     Vals.push_back(VE.getTypeID(GV.getValueType()));
@@ -1363,7 +1371,7 @@ void ModuleBitcodeWriter::writeModuleInfo() {
         GV.isExternallyInitialized() ||
         GV.getDLLStorageClass() != GlobalValue::DefaultStorageClass ||
         GV.hasComdat() || GV.hasAttributes() || GV.isDSOLocal() ||
-        GV.hasPartition()) {
+        GV.hasPartition() || GV.hasSanitizerMetadata()) {
       Vals.push_back(getEncodedVisibility(GV));
       Vals.push_back(getEncodedThreadLocalMode(GV));
       Vals.push_back(getEncodedUnnamedAddr(GV));
@@ -1377,6 +1385,11 @@ void ModuleBitcodeWriter::writeModuleInfo() {
       Vals.push_back(GV.isDSOLocal());
       Vals.push_back(addToStrtab(GV.getPartition()));
       Vals.push_back(GV.getPartition().size());
+
+      if (GV.hasSanitizerMetadata())
+        Vals.push_back(serializeSanitizerMetadata(GV.getSanitizerMetadata()));
+      else
+        Vals.push_back(UINT_MAX);
     } else {
       AbbrevToUse = SimpleGVarAbbrev;
     }

llvm/lib/IR/AsmWriter.cpp

@@ -3535,6 +3535,19 @@ void AssemblyWriter::printGlobal(const GlobalVariable *GV) {
     Out << '"';
   }
 
+  using SanitizerMetadata = llvm::GlobalValue::SanitizerMetadata;
+  if (GV->hasSanitizerMetadata()) {
+    SanitizerMetadata MD = GV->getSanitizerMetadata();
+    if (MD.NoAddress)
+      Out << ", no_sanitize_address";
+    if (MD.NoHWAddress)
+      Out << ", no_sanitize_hwaddress";
+    if (MD.NoMemtag)
+      Out << ", no_sanitize_memtag";
+    if (MD.IsDynInit)
+      Out << ", sanitize_address_dyninit";
+  }
+
   maybePrintComdat(Out, *GV);
   if (MaybeAlign A = GV->getAlign())
     Out << ", align " << A->value();

llvm/lib/IR/Globals.cpp

@@ -67,6 +67,10 @@ void GlobalValue::copyAttributesFrom(const GlobalValue *Src) {
   setDLLStorageClass(Src->getDLLStorageClass());
   setDSOLocal(Src->isDSOLocal());
   setPartition(Src->getPartition());
+  if (Src->hasSanitizerMetadata())
+    setSanitizerMetadata(Src->getSanitizerMetadata());
+  else
+    removeSanitizerMetadata();
 }
 
 void GlobalValue::removeFromParent() {
@@ -217,6 +221,27 @@ void GlobalValue::setPartition(StringRef S) {
   HasPartition = !S.empty();
 }
 
+using SanitizerMetadata = GlobalValue::SanitizerMetadata;
+const SanitizerMetadata &GlobalValue::getSanitizerMetadata() const {
+  assert(hasSanitizerMetadata());
+  assert(getContext().pImpl->GlobalValueSanitizerMetadata.count(this));
+  return getContext().pImpl->GlobalValueSanitizerMetadata[this];
+}
+
+void GlobalValue::setSanitizerMetadata(const SanitizerMetadata &Meta) {
+  getContext().pImpl->GlobalValueSanitizerMetadata[this] = Meta;
+  HasSanitizerMetadata = true;
+}
+
+void GlobalValue::removeSanitizerMetadata() {
+  DenseMap<const GlobalValue *, SanitizerMetadata> &MetadataMap =
+      getContext().pImpl->GlobalValueSanitizerMetadata;
+  auto It = MetadataMap.find(this);
+  if (It != MetadataMap.end())
+    MetadataMap.erase(It);
+  HasSanitizerMetadata = false;
+}
+
 StringRef GlobalObject::getSectionImpl() const {
   assert(hasSection());
   return getContext().pImpl->GlobalObjectSections[this];

llvm/lib/IR/LLVMContextImpl.h

@@ -1503,6 +1503,9 @@ public:
   /// Collection of per-GlobalValue partitions used in this context.
   DenseMap<const GlobalValue *, StringRef> GlobalValuePartitions;
 
+  DenseMap<const GlobalValue *, GlobalValue::SanitizerMetadata>
+      GlobalValueSanitizerMetadata;
+
   /// DiscriminatorTable - This table maps file:line locations to an
   /// integer representing the next DWARF path discriminator to assign to
   /// instructions in different blocks at the same location.

llvm/test/Assembler/globalvariable-attributes.ll

@@ -4,6 +4,11 @@
 @g2 = global i32 2, align 4 "key3" = "value3"
 @g3 = global i32 2 #0
 @g4 = global i32 2, align 4 "key5" = "value5" #0
+@g5 = global i32 2, no_sanitize_address, align 4
+@g6 = global i32 2, no_sanitize_hwaddress, align 4
+@g7 = global i32 2, no_sanitize_memtag, align 4
+@g8 = global i32 2, sanitize_address_dyninit, align 4
+@g9 = global i32 2, no_sanitize_address, no_sanitize_hwaddress, no_sanitize_memtag, align 4
 
 attributes #0 = { "string" = "value" nobuiltin norecurse }
 
@@ -11,6 +16,11 @@ attributes #0 = { "string" = "value" nobuiltin norecurse }
 ; CHECK: @g2 = global i32 2, align 4 #1
 ; CHECK: @g3 = global i32 2 #2
 ; CHECK: @g4 = global i32 2, align 4 #3
+; CHECK: @g5 = global i32 2, no_sanitize_address, align 4
+; CHECK: @g6 = global i32 2, no_sanitize_hwaddress, align 4
+; CHECK: @g7 = global i32 2, no_sanitize_memtag, align 4
+; CHECK: @g8 = global i32 2, sanitize_address_dyninit, align 4
+; CHECK: @g9 = global i32 2, no_sanitize_address, no_sanitize_hwaddress, no_sanitize_memtag, align 4
 
 ; CHECK: attributes #0 = { "key"="value" "key2"="value2" }
 ; CHECK: attributes #1 = { "key3"="value3" }

llvm/test/Bitcode/compatibility.ll

@@ -203,6 +203,18 @@ declare void @g.f1()
 @llvm.global_dtors = appending global [1 x %pri.func.data] [%pri.func.data { i32 0, void ()* @g.f1, i8* @g.used3 }], section "llvm.metadata"
 ; CHECK: @llvm.global_dtors = appending global [1 x %pri.func.data] [%pri.func.data { i32 0, void ()* @g.f1, i8* @g.used3 }], section "llvm.metadata"
 
+; Global Variables -- sanitizers
+@g.no_sanitize_address = global i32 0, no_sanitize_address
+@g.no_sanitize_hwaddress = global i32 0, no_sanitize_hwaddress
+@g.no_sanitize_memtag = global i32 0, no_sanitize_memtag
+@g.no_sanitize_multiple = global i32 0, no_sanitize_address, no_sanitize_hwaddress, no_sanitize_memtag
+@g.sanitize_address_dyninit = global i32 0, sanitize_address_dyninit
+; CHECK: @g.no_sanitize_address = global i32 0, no_sanitize_address
+; CHECK: @g.no_sanitize_hwaddress = global i32 0, no_sanitize_hwaddress
+; CHECK: @g.no_sanitize_memtag = global i32 0, no_sanitize_memtag
+; CHECK: @g.no_sanitize_multiple = global i32 0, no_sanitize_address, no_sanitize_hwaddress, no_sanitize_memtag
+; CHECK: @g.sanitize_address_dyninit = global i32 0, sanitize_address_dyninit
+
 ;; Aliases
 ; Format: @<Name> = [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal]
 ;                   [unnamed_addr] alias <AliaseeTy> @<Aliasee>