#define DEBUG_TYPE "asan"
-#include "FunctionBlackList.h"
+#include "BlackList.h"
#include "llvm/Function.h"
#include "llvm/IRBuilder.h"
+#include "llvm/InlineAsm.h"
#include "llvm/IntrinsicInst.h"
#include "llvm/LLVMContext.h"
#include "llvm/Module.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Support/system_error.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/DataLayout.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Transforms/Instrumentation.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
static const char *kAsanReportErrorTemplate = "__asan_report_";
static const char *kAsanRegisterGlobalsName = "__asan_register_globals";
static const char *kAsanUnregisterGlobalsName = "__asan_unregister_globals";
+static const char *kAsanPoisonGlobalsName = "__asan_before_dynamic_init";
+static const char *kAsanUnpoisonGlobalsName = "__asan_after_dynamic_init";
static const char *kAsanInitName = "__asan_init";
static const char *kAsanHandleNoReturnName = "__asan_handle_no_return";
static const char *kAsanMappingOffsetName = "__asan_mapping_offset";
static const char *kAsanMappingScaleName = "__asan_mapping_scale";
static const char *kAsanStackMallocName = "__asan_stack_malloc";
static const char *kAsanStackFreeName = "__asan_stack_free";
+static const char *kAsanGenPrefix = "__asan_gen_";
static const int kAsanStackLeftRedzoneMagic = 0xf1;
static const int kAsanStackMidRedzoneMagic = 0xf2;
static const int kAsanStackRightRedzoneMagic = 0xf3;
static const int kAsanStackPartialRedzoneMagic = 0xf4;
+// Accesses sizes are powers of two: 1, 2, 4, 8, 16.
+static const size_t kNumberOfAccessSizes = 5;
+
// Command-line flags.
// This flag may need to be replaced with -f[no-]asan-reads.
static cl::opt<bool> ClInstrumentAtomics("asan-instrument-atomics",
cl::desc("instrument atomic instructions (rmw, cmpxchg)"),
cl::Hidden, cl::init(true));
-// This flags limits the number of instructions to be instrumented
+static cl::opt<bool> ClAlwaysSlowPath("asan-always-slow-path",
+ cl::desc("use instrumentation with slow path for all accesses"),
+ cl::Hidden, cl::init(false));
+// This flag limits the number of instructions to be instrumented
// in any given BB. Normally, this should be set to unlimited (INT_MAX),
// but due to http://llvm.org/bugs/show_bug.cgi?id=12652 we temporary
// set it to 10000.
// This flag may need to be replaced with -f[no]asan-globals.
static cl::opt<bool> ClGlobals("asan-globals",
cl::desc("Handle global objects"), cl::Hidden, cl::init(true));
+static cl::opt<bool> ClInitializers("asan-initialization-order",
+ cl::desc("Handle C++ initializer order"), cl::Hidden, cl::init(false));
static cl::opt<bool> ClMemIntrin("asan-memintrin",
cl::desc("Handle memset/memcpy/memmove"), cl::Hidden, cl::init(true));
// This flag may need to be replaced with -fasan-blacklist.
cl::Hidden, cl::init(-1));
namespace {
+/// A set of dynamically initialized globals extracted from metadata.
+class SetOfDynamicallyInitializedGlobals {
+ public:
+ void Init(Module& M) {
+ // Clang generates metadata identifying all dynamically initialized globals.
+ NamedMDNode *DynamicGlobals =
+ M.getNamedMetadata("llvm.asan.dynamically_initialized_globals");
+ if (!DynamicGlobals)
+ return;
+ for (int i = 0, n = DynamicGlobals->getNumOperands(); i < n; ++i) {
+ MDNode *MDN = DynamicGlobals->getOperand(i);
+ assert(MDN->getNumOperands() == 1);
+ Value *VG = MDN->getOperand(0);
+ // The optimizer may optimize away a global entirely, in which case we
+ // cannot instrument access to it.
+ if (!VG)
+ continue;
+ DynInitGlobals.insert(cast<GlobalVariable>(VG));
+ }
+ }
+ bool Contains(GlobalVariable *G) { return DynInitGlobals.count(G) != 0; }
+ private:
+ SmallSet<GlobalValue*, 32> DynInitGlobals;
+};
+
+static int MappingScale() {
+ return ClMappingScale ? ClMappingScale : kDefaultShadowScale;
+}
+
+static size_t RedzoneSize() {
+ // Redzone used for stack and globals is at least 32 bytes.
+ // For scales 6 and 7, the redzone has to be 64 and 128 bytes respectively.
+ return std::max(32U, 1U << MappingScale());
+}
/// AddressSanitizer: instrument the code in module to find memory bugs.
-struct AddressSanitizer : public ModulePass {
+struct AddressSanitizer : public FunctionPass {
AddressSanitizer();
virtual const char *getPassName() const;
void instrumentMop(Instruction *I);
void instrumentAddress(Instruction *OrigIns, IRBuilder<> &IRB,
Value *Addr, uint32_t TypeSize, bool IsWrite);
- Instruction *generateCrashCode(IRBuilder<> &IRB, Value *Addr,
- bool IsWrite, uint32_t TypeSize);
+ Value *createSlowPathCmp(IRBuilder<> &IRB, Value *AddrLong,
+ Value *ShadowValue, uint32_t TypeSize);
+ Instruction *generateCrashCode(Instruction *InsertBefore, Value *Addr,
+ bool IsWrite, size_t AccessSizeIndex);
bool instrumentMemIntrinsic(MemIntrinsic *MI);
void instrumentMemIntrinsicParam(Instruction *OrigIns, Value *Addr,
- Value *Size,
+ Value *Size,
Instruction *InsertBefore, bool IsWrite);
Value *memToShadow(Value *Shadow, IRBuilder<> &IRB);
- bool handleFunction(Module &M, Function &F);
+ bool runOnFunction(Function &F);
+ void createInitializerPoisonCalls(Module &M,
+ Value *FirstAddr, Value *LastAddr);
bool maybeInsertAsanInitAtFunctionEntry(Function &F);
- bool poisonStackInFunction(Module &M, Function &F);
- virtual bool runOnModule(Module &M);
- bool insertGlobalRedzones(Module &M);
+ bool poisonStackInFunction(Function &F);
+ virtual bool doInitialization(Module &M);
+ virtual bool doFinalization(Module &M);
static char ID; // Pass identification, replacement for typeid
private:
-
uint64_t getAllocaSizeInBytes(AllocaInst *AI) {
Type *Ty = AI->getAllocatedType();
uint64_t SizeInBytes = TD->getTypeAllocSize(Ty);
return SizeInBytes;
}
uint64_t getAlignedSize(uint64_t SizeInBytes) {
- return ((SizeInBytes + RedzoneSize - 1)
- / RedzoneSize) * RedzoneSize;
+ size_t RZ = RedzoneSize();
+ return ((SizeInBytes + RZ - 1) / RZ) * RZ;
}
uint64_t getAlignedAllocaSize(AllocaInst *AI) {
uint64_t SizeInBytes = getAllocaSizeInBytes(AI);
return getAlignedSize(SizeInBytes);
}
- Function *checkInterfaceFunction(Constant *FuncOrBitcast);
+ bool ShouldInstrumentGlobal(GlobalVariable *G);
void PoisonStack(const ArrayRef<AllocaInst*> &AllocaVec, IRBuilder<> IRB,
Value *ShadowBase, bool DoPoison);
bool LooksLikeCodeInBug11395(Instruction *I);
+ void FindDynamicInitializers(Module &M);
LLVMContext *C;
- TargetData *TD;
+ DataLayout *TD;
uint64_t MappingOffset;
- int MappingScale;
- size_t RedzoneSize;
int LongSize;
Type *IntptrTy;
Type *IntptrPtrTy;
Function *AsanCtorFunction;
Function *AsanInitFunction;
- Instruction *CtorInsertBefore;
- OwningPtr<FunctionBlackList> BL;
- // Accesses sizes are powers of two: 1, 2, 4, 8, 16.
- static const size_t kNumberOfAccessSizes = 5;
+ Function *AsanStackMallocFunc, *AsanStackFreeFunc;
+ Function *AsanHandleNoReturnFunc;
+ OwningPtr<BlackList> BL;
// This array is indexed by AccessIsWrite and log2(AccessSize).
Function *AsanErrorCallback[2][kNumberOfAccessSizes];
+ InlineAsm *EmptyAsm;
+ SetOfDynamicallyInitializedGlobals DynamicallyInitializedGlobals;
};
+
+// FIXME: inherit this from ModulePass and actually use it as a ModulePass.
+class AddressSanitizerCreateGlobalRedzonesPass {
+ public:
+ bool runOnModule(Module &M, DataLayout *TD);
+ private:
+ bool ShouldInstrumentGlobal(GlobalVariable *G);
+ void createInitializerPoisonCalls(Module &M, Value *FirstAddr,
+ Value *LastAddr);
+
+ OwningPtr<BlackList> BL;
+ SetOfDynamicallyInitializedGlobals DynamicallyInitializedGlobals;
+ Type *IntptrTy;
+ LLVMContext *C;
+};
+
} // namespace
char AddressSanitizer::ID = 0;
INITIALIZE_PASS(AddressSanitizer, "asan",
"AddressSanitizer: detects use-after-free and out-of-bounds bugs.",
false, false)
-AddressSanitizer::AddressSanitizer() : ModulePass(ID) { }
-ModulePass *llvm::createAddressSanitizerPass() {
+AddressSanitizer::AddressSanitizer() : FunctionPass(ID) { }
+FunctionPass *llvm::createAddressSanitizerPass() {
return new AddressSanitizer();
}
return "AddressSanitizer";
}
+static size_t TypeSizeToSizeIndex(uint32_t TypeSize) {
+ size_t Res = CountTrailingZeros_32(TypeSize / 8);
+ assert(Res < kNumberOfAccessSizes);
+ return Res;
+}
+
// Create a constant for Str so that we can pass it to the run-time lib.
static GlobalVariable *createPrivateGlobalForString(Module &M, StringRef Str) {
Constant *StrConst = ConstantDataArray::getString(M.getContext(), Str);
return new GlobalVariable(M, StrConst->getType(), true,
- GlobalValue::PrivateLinkage, StrConst, "");
+ GlobalValue::PrivateLinkage, StrConst,
+ kAsanGenPrefix);
}
-// Split the basic block and insert an if-then code.
-// Before:
-// Head
-// Cmp
-// Tail
-// After:
-// Head
-// if (Cmp)
-// ThenBlock
-// Tail
-//
-// Returns the ThenBlock's terminator.
-static BranchInst *splitBlockAndInsertIfThen(Value *Cmp) {
- Instruction *SplitBefore = cast<Instruction>(Cmp)->getNextNode();
- BasicBlock *Head = SplitBefore->getParent();
- BasicBlock *Tail = Head->splitBasicBlock(SplitBefore);
- TerminatorInst *HeadOldTerm = Head->getTerminator();
- LLVMContext &C = Head->getParent()->getParent()->getContext();
- BasicBlock *ThenBlock = BasicBlock::Create(C, "", Head->getParent());
- BranchInst *HeadNewTerm =
- BranchInst::Create(/*ifTrue*/ThenBlock, /*ifFalse*/Tail, Cmp);
- ReplaceInstWithInst(HeadOldTerm, HeadNewTerm);
-
- BranchInst *CheckTerm = BranchInst::Create(Tail, ThenBlock);
- return CheckTerm;
+static bool GlobalWasGeneratedByAsan(GlobalVariable *G) {
+ return G->getName().find(kAsanGenPrefix) == 0;
}
Value *AddressSanitizer::memToShadow(Value *Shadow, IRBuilder<> &IRB) {
// Shadow >> scale
- Shadow = IRB.CreateLShr(Shadow, MappingScale);
+ Shadow = IRB.CreateLShr(Shadow, MappingScale());
if (MappingOffset == 0)
return Shadow;
// (Shadow >> scale) | offset
MappingOffset));
}
-void AddressSanitizer::instrumentMemIntrinsicParam(Instruction *OrigIns,
+void AddressSanitizer::instrumentMemIntrinsicParam(
+ Instruction *OrigIns,
Value *Addr, Value *Size, Instruction *InsertBefore, bool IsWrite) {
// Check the first byte.
{
bool AddressSanitizer::instrumentMemIntrinsic(MemIntrinsic *MI) {
Value *Dst = MI->getDest();
MemTransferInst *MemTran = dyn_cast<MemTransferInst>(MI);
- Value *Src = MemTran ? MemTran->getSource() : NULL;
+ Value *Src = MemTran ? MemTran->getSource() : 0;
Value *Length = MI->getLength();
Constant *ConstLength = dyn_cast<Constant>(Length);
Value *Cmp = IRB.CreateICmpNE(Length,
Constant::getNullValue(Length->getType()));
- InsertBefore = splitBlockAndInsertIfThen(Cmp);
+ InsertBefore = SplitBlockAndInsertIfThen(cast<Instruction>(Cmp), false);
}
instrumentMemIntrinsicParam(MI, Dst, Length, InsertBefore, true);
}
void AddressSanitizer::instrumentMop(Instruction *I) {
- bool IsWrite;
+ bool IsWrite = false;
Value *Addr = isInterestingMemoryAccess(I, &IsWrite);
assert(Addr);
- if (ClOpt && ClOptGlobals && isa<GlobalVariable>(Addr)) {
- // We are accessing a global scalar variable. Nothing to catch here.
- return;
+ if (ClOpt && ClOptGlobals) {
+ if (GlobalVariable *G = dyn_cast<GlobalVariable>(Addr)) {
+ // If initialization order checking is disabled, a simple access to a
+ // dynamically initialized global is always valid.
+ if (!ClInitializers)
+ return;
+ // If a global variable does not have dynamic initialization we don't
+ // have to instrument it. However, if a global does not have initailizer
+ // at all, we assume it has dynamic initializer (in other TU).
+ if (G->hasInitializer() && !DynamicallyInitializedGlobals.Contains(G))
+ return;
+ }
}
+
Type *OrigPtrTy = Addr->getType();
Type *OrigTy = cast<PointerType>(OrigPtrTy)->getElementType();
// function of AddressSanitizer. If the instrumented module defines a function
// with the same name, their prototypes must match, otherwise
// getOrInsertFunction returns a bitcast.
-Function *AddressSanitizer::checkInterfaceFunction(Constant *FuncOrBitcast) {
+static Function *checkInterfaceFunction(Constant *FuncOrBitcast) {
if (isa<Function>(FuncOrBitcast)) return cast<Function>(FuncOrBitcast);
FuncOrBitcast->dump();
report_fatal_error("trying to redefine an AddressSanitizer "
}
Instruction *AddressSanitizer::generateCrashCode(
- IRBuilder<> &IRB, Value *Addr, bool IsWrite, uint32_t TypeSize) {
- size_t AccessSizeIndex = CountTrailingZeros_32(TypeSize / 8);
- assert(AccessSizeIndex < kNumberOfAccessSizes);
+ Instruction *InsertBefore, Value *Addr,
+ bool IsWrite, size_t AccessSizeIndex) {
+ IRBuilder<> IRB(InsertBefore);
CallInst *Call = IRB.CreateCall(AsanErrorCallback[IsWrite][AccessSizeIndex],
Addr);
- Call->setDoesNotReturn();
+ // We don't do Call->setDoesNotReturn() because the BB already has
+ // UnreachableInst at the end.
+ // This EmptyAsm is required to avoid callback merge.
+ IRB.CreateCall(EmptyAsm);
return Call;
}
+Value *AddressSanitizer::createSlowPathCmp(IRBuilder<> &IRB, Value *AddrLong,
+ Value *ShadowValue,
+ uint32_t TypeSize) {
+ size_t Granularity = 1 << MappingScale();
+ // Addr & (Granularity - 1)
+ Value *LastAccessedByte = IRB.CreateAnd(
+ AddrLong, ConstantInt::get(IntptrTy, Granularity - 1));
+ // (Addr & (Granularity - 1)) + size - 1
+ if (TypeSize / 8 > 1)
+ LastAccessedByte = IRB.CreateAdd(
+ LastAccessedByte, ConstantInt::get(IntptrTy, TypeSize / 8 - 1));
+ // (uint8_t) ((Addr & (Granularity-1)) + size - 1)
+ LastAccessedByte = IRB.CreateIntCast(
+ LastAccessedByte, ShadowValue->getType(), false);
+ // ((uint8_t) ((Addr & (Granularity-1)) + size - 1)) >= ShadowValue
+ return IRB.CreateICmpSGE(LastAccessedByte, ShadowValue);
+}
+
void AddressSanitizer::instrumentAddress(Instruction *OrigIns,
IRBuilder<> &IRB, Value *Addr,
uint32_t TypeSize, bool IsWrite) {
Value *AddrLong = IRB.CreatePointerCast(Addr, IntptrTy);
Type *ShadowTy = IntegerType::get(
- *C, std::max(8U, TypeSize >> MappingScale));
+ *C, std::max(8U, TypeSize >> MappingScale()));
Type *ShadowPtrTy = PointerType::get(ShadowTy, 0);
Value *ShadowPtr = memToShadow(AddrLong, IRB);
Value *CmpVal = Constant::getNullValue(ShadowTy);
IRB.CreateIntToPtr(ShadowPtr, ShadowPtrTy));
Value *Cmp = IRB.CreateICmpNE(ShadowValue, CmpVal);
+ size_t AccessSizeIndex = TypeSizeToSizeIndex(TypeSize);
+ size_t Granularity = 1 << MappingScale();
+ TerminatorInst *CrashTerm = 0;
+
+ if (ClAlwaysSlowPath || (TypeSize < 8 * Granularity)) {
+ TerminatorInst *CheckTerm =
+ SplitBlockAndInsertIfThen(cast<Instruction>(Cmp), false);
+ assert(dyn_cast<BranchInst>(CheckTerm)->isUnconditional());
+ BasicBlock *NextBB = CheckTerm->getSuccessor(0);
+ IRB.SetInsertPoint(CheckTerm);
+ Value *Cmp2 = createSlowPathCmp(IRB, AddrLong, ShadowValue, TypeSize);
+ BasicBlock *CrashBlock =
+ BasicBlock::Create(*C, "", NextBB->getParent(), NextBB);
+ CrashTerm = new UnreachableInst(*C, CrashBlock);
+ BranchInst *NewTerm = BranchInst::Create(CrashBlock, NextBB, Cmp2);
+ ReplaceInstWithInst(CheckTerm, NewTerm);
+ } else {
+ CrashTerm = SplitBlockAndInsertIfThen(cast<Instruction>(Cmp), true);
+ }
- Instruction *CheckTerm = splitBlockAndInsertIfThen(Cmp);
- IRBuilder<> IRB2(CheckTerm);
-
- size_t Granularity = 1 << MappingScale;
- if (TypeSize < 8 * Granularity) {
- // Addr & (Granularity - 1)
- Value *LastAccessedByte = IRB2.CreateAnd(
- AddrLong, ConstantInt::get(IntptrTy, Granularity - 1));
- // (Addr & (Granularity - 1)) + size - 1
- if (TypeSize / 8 > 1)
- LastAccessedByte = IRB2.CreateAdd(
- LastAccessedByte, ConstantInt::get(IntptrTy, TypeSize / 8 - 1));
- // (uint8_t) ((Addr & (Granularity-1)) + size - 1)
- LastAccessedByte = IRB2.CreateIntCast(
- LastAccessedByte, IRB.getInt8Ty(), false);
- // ((uint8_t) ((Addr & (Granularity-1)) + size - 1)) >= ShadowValue
- Value *Cmp2 = IRB2.CreateICmpSGE(LastAccessedByte, ShadowValue);
-
- CheckTerm = splitBlockAndInsertIfThen(Cmp2);
- }
-
- IRBuilder<> IRB1(CheckTerm);
- Instruction *Crash = generateCrashCode(IRB1, AddrLong, IsWrite, TypeSize);
+ Instruction *Crash =
+ generateCrashCode(CrashTerm, AddrLong, IsWrite, AccessSizeIndex);
Crash->setDebugLoc(OrigIns->getDebugLoc());
- ReplaceInstWithInst(CheckTerm, new UnreachableInst(*C));
}
-// This function replaces all global variables with new variables that have
-// trailing redzones. It also creates a function that poisons
-// redzones and inserts this function into llvm.global_ctors.
-bool AddressSanitizer::insertGlobalRedzones(Module &M) {
- SmallVector<GlobalVariable *, 16> GlobalsToChange;
+void AddressSanitizerCreateGlobalRedzonesPass::createInitializerPoisonCalls(
+ Module &M, Value *FirstAddr, Value *LastAddr) {
+ // We do all of our poisoning and unpoisoning within _GLOBAL__I_a.
+ Function *GlobalInit = M.getFunction("_GLOBAL__I_a");
+ // If that function is not present, this TU contains no globals, or they have
+ // all been optimized away
+ if (!GlobalInit)
+ return;
- for (Module::GlobalListType::iterator G = M.getGlobalList().begin(),
- E = M.getGlobalList().end(); G != E; ++G) {
- Type *Ty = cast<PointerType>(G->getType())->getElementType();
- DEBUG(dbgs() << "GLOBAL: " << *G);
-
- if (!Ty->isSized()) continue;
- if (!G->hasInitializer()) continue;
- // Touch only those globals that will not be defined in other modules.
- // Don't handle ODR type linkages since other modules may be built w/o asan.
- if (G->getLinkage() != GlobalVariable::ExternalLinkage &&
- G->getLinkage() != GlobalVariable::PrivateLinkage &&
- G->getLinkage() != GlobalVariable::InternalLinkage)
- continue;
- // Two problems with thread-locals:
- // - The address of the main thread's copy can't be computed at link-time.
- // - Need to poison all copies, not just the main thread's one.
- if (G->isThreadLocal())
- continue;
- // For now, just ignore this Alloca if the alignment is large.
- if (G->getAlignment() > RedzoneSize) continue;
-
- // Ignore all the globals with the names starting with "\01L_OBJC_".
- // Many of those are put into the .cstring section. The linker compresses
- // that section by removing the spare \0s after the string terminator, so
- // our redzones get broken.
- if ((G->getName().find("\01L_OBJC_") == 0) ||
- (G->getName().find("\01l_OBJC_") == 0)) {
- DEBUG(dbgs() << "Ignoring \\01L_OBJC_* global: " << *G);
- continue;
+ // Set up the arguments to our poison/unpoison functions.
+ IRBuilder<> IRB(GlobalInit->begin()->getFirstInsertionPt());
+
+ // Declare our poisoning and unpoisoning functions.
+ Function *AsanPoisonGlobals = checkInterfaceFunction(M.getOrInsertFunction(
+ kAsanPoisonGlobalsName, IRB.getVoidTy(), IntptrTy, IntptrTy, NULL));
+ AsanPoisonGlobals->setLinkage(Function::ExternalLinkage);
+ Function *AsanUnpoisonGlobals = checkInterfaceFunction(M.getOrInsertFunction(
+ kAsanUnpoisonGlobalsName, IRB.getVoidTy(), NULL));
+ AsanUnpoisonGlobals->setLinkage(Function::ExternalLinkage);
+
+ // Add a call to poison all external globals before the given function starts.
+ IRB.CreateCall2(AsanPoisonGlobals, FirstAddr, LastAddr);
+
+ // Add calls to unpoison all globals before each return instruction.
+ for (Function::iterator I = GlobalInit->begin(), E = GlobalInit->end();
+ I != E; ++I) {
+ if (ReturnInst *RI = dyn_cast<ReturnInst>(I->getTerminator())) {
+ CallInst::Create(AsanUnpoisonGlobals, "", RI);
}
+ }
+}
- if (G->hasSection()) {
- StringRef Section(G->getSection());
- // Ignore the globals from the __OBJC section. The ObjC runtime assumes
- // those conform to /usr/lib/objc/runtime.h, so we can't add redzones to
- // them.
- if ((Section.find("__OBJC,") == 0) ||
- (Section.find("__DATA, __objc_") == 0)) {
- DEBUG(dbgs() << "Ignoring ObjC runtime global: " << *G);
- continue;
- }
- // See http://code.google.com/p/address-sanitizer/issues/detail?id=32
- // Constant CFString instances are compiled in the following way:
- // -- the string buffer is emitted into
- // __TEXT,__cstring,cstring_literals
- // -- the constant NSConstantString structure referencing that buffer
- // is placed into __DATA,__cfstring
- // Therefore there's no point in placing redzones into __DATA,__cfstring.
- // Moreover, it causes the linker to crash on OS X 10.7
- if (Section.find("__DATA,__cfstring") == 0) {
- DEBUG(dbgs() << "Ignoring CFString: " << *G);
- continue;
- }
+bool AddressSanitizerCreateGlobalRedzonesPass::ShouldInstrumentGlobal(
+ GlobalVariable *G) {
+ Type *Ty = cast<PointerType>(G->getType())->getElementType();
+ DEBUG(dbgs() << "GLOBAL: " << *G << "\n");
+
+ if (BL->isIn(*G)) return false;
+ if (!Ty->isSized()) return false;
+ if (!G->hasInitializer()) return false;
+ if (GlobalWasGeneratedByAsan(G)) return false; // Our own global.
+ // Touch only those globals that will not be defined in other modules.
+ // Don't handle ODR type linkages since other modules may be built w/o asan.
+ if (G->getLinkage() != GlobalVariable::ExternalLinkage &&
+ G->getLinkage() != GlobalVariable::PrivateLinkage &&
+ G->getLinkage() != GlobalVariable::InternalLinkage)
+ return false;
+ // Two problems with thread-locals:
+ // - The address of the main thread's copy can't be computed at link-time.
+ // - Need to poison all copies, not just the main thread's one.
+ if (G->isThreadLocal())
+ return false;
+ // For now, just ignore this Alloca if the alignment is large.
+ if (G->getAlignment() > RedzoneSize()) return false;
+
+ // Ignore all the globals with the names starting with "\01L_OBJC_".
+ // Many of those are put into the .cstring section. The linker compresses
+ // that section by removing the spare \0s after the string terminator, so
+ // our redzones get broken.
+ if ((G->getName().find("\01L_OBJC_") == 0) ||
+ (G->getName().find("\01l_OBJC_") == 0)) {
+ DEBUG(dbgs() << "Ignoring \\01L_OBJC_* global: " << *G);
+ return false;
+ }
+
+ if (G->hasSection()) {
+ StringRef Section(G->getSection());
+ // Ignore the globals from the __OBJC section. The ObjC runtime assumes
+ // those conform to /usr/lib/objc/runtime.h, so we can't add redzones to
+ // them.
+ if ((Section.find("__OBJC,") == 0) ||
+ (Section.find("__DATA, __objc_") == 0)) {
+ DEBUG(dbgs() << "Ignoring ObjC runtime global: " << *G);
+ return false;
+ }
+ // See http://code.google.com/p/address-sanitizer/issues/detail?id=32
+ // Constant CFString instances are compiled in the following way:
+ // -- the string buffer is emitted into
+ // __TEXT,__cstring,cstring_literals
+ // -- the constant NSConstantString structure referencing that buffer
+ // is placed into __DATA,__cfstring
+ // Therefore there's no point in placing redzones into __DATA,__cfstring.
+ // Moreover, it causes the linker to crash on OS X 10.7
+ if (Section.find("__DATA,__cfstring") == 0) {
+ DEBUG(dbgs() << "Ignoring CFString: " << *G);
+ return false;
}
+ }
+
+ return true;
+}
+
+// This function replaces all global variables with new variables that have
+// trailing redzones. It also creates a function that poisons
+// redzones and inserts this function into llvm.global_ctors.
+bool AddressSanitizerCreateGlobalRedzonesPass::runOnModule(Module &M,
+ DataLayout *TD) {
+ BL.reset(new BlackList(ClBlackListFile));
+ DynamicallyInitializedGlobals.Init(M);
+ C = &(M.getContext());
+ IntptrTy = Type::getIntNTy(*C, TD->getPointerSizeInBits());
- GlobalsToChange.push_back(G);
+ SmallVector<GlobalVariable *, 16> GlobalsToChange;
+
+ for (Module::GlobalListType::iterator G = M.global_begin(),
+ E = M.global_end(); G != E; ++G) {
+ if (ShouldInstrumentGlobal(G))
+ GlobalsToChange.push_back(G);
}
size_t n = GlobalsToChange.size();
// size_t size;
// size_t size_with_redzone;
// const char *name;
+ // size_t has_dynamic_init;
// We initialize an array of such structures and pass it to a run-time call.
StructType *GlobalStructTy = StructType::get(IntptrTy, IntptrTy,
- IntptrTy, IntptrTy, NULL);
- SmallVector<Constant *, 16> Initializers(n);
+ IntptrTy, IntptrTy,
+ IntptrTy, NULL);
+ SmallVector<Constant *, 16> Initializers(n), DynamicInit;
+
- IRBuilder<> IRB(CtorInsertBefore);
+ Function *CtorFunc = M.getFunction(kAsanModuleCtorName);
+ assert(CtorFunc);
+ IRBuilder<> IRB(CtorFunc->getEntryBlock().getTerminator());
+
+ // The addresses of the first and last dynamically initialized globals in
+ // this TU. Used in initialization order checking.
+ Value *FirstDynamic = 0, *LastDynamic = 0;
for (size_t i = 0; i < n; i++) {
GlobalVariable *G = GlobalsToChange[i];
PointerType *PtrTy = cast<PointerType>(G->getType());
Type *Ty = PtrTy->getElementType();
uint64_t SizeInBytes = TD->getTypeAllocSize(Ty);
- uint64_t RightRedzoneSize = RedzoneSize +
- (RedzoneSize - (SizeInBytes % RedzoneSize));
+ size_t RZ = RedzoneSize();
+ uint64_t RightRedzoneSize = RZ + (RZ - (SizeInBytes % RZ));
Type *RightRedZoneTy = ArrayType::get(IRB.getInt8Ty(), RightRedzoneSize);
+ // Determine whether this global should be poisoned in initialization.
+ bool GlobalHasDynamicInitializer =
+ DynamicallyInitializedGlobals.Contains(G);
+ // Don't check initialization order if this global is blacklisted.
+ GlobalHasDynamicInitializer &= !BL->isInInit(*G);
StructType *NewTy = StructType::get(Ty, RightRedZoneTy, NULL);
Constant *NewInitializer = ConstantStruct::get(
M, NewTy, G->isConstant(), G->getLinkage(),
NewInitializer, "", G, G->getThreadLocalMode());
NewGlobal->copyAttributesFrom(G);
- NewGlobal->setAlignment(RedzoneSize);
+ NewGlobal->setAlignment(RZ);
Value *Indices2[2];
Indices2[0] = IRB.getInt32(0);
ConstantInt::get(IntptrTy, SizeInBytes),
ConstantInt::get(IntptrTy, SizeInBytes + RightRedzoneSize),
ConstantExpr::getPointerCast(Name, IntptrTy),
+ ConstantInt::get(IntptrTy, GlobalHasDynamicInitializer),
NULL);
- DEBUG(dbgs() << "NEW GLOBAL:\n" << *NewGlobal);
+
+ // Populate the first and last globals declared in this TU.
+ if (ClInitializers && GlobalHasDynamicInitializer) {
+ LastDynamic = ConstantExpr::getPointerCast(NewGlobal, IntptrTy);
+ if (FirstDynamic == 0)
+ FirstDynamic = LastDynamic;
+ }
+
+ DEBUG(dbgs() << "NEW GLOBAL: " << *NewGlobal << "\n");
}
ArrayType *ArrayOfGlobalStructTy = ArrayType::get(GlobalStructTy, n);
M, ArrayOfGlobalStructTy, false, GlobalVariable::PrivateLinkage,
ConstantArray::get(ArrayOfGlobalStructTy, Initializers), "");
+ // Create calls for poisoning before initializers run and unpoisoning after.
+ if (ClInitializers && FirstDynamic && LastDynamic)
+ createInitializerPoisonCalls(M, FirstDynamic, LastDynamic);
+
Function *AsanRegisterGlobals = checkInterfaceFunction(M.getOrInsertFunction(
- kAsanRegisterGlobalsName, IRB.getVoidTy(), IntptrTy, IntptrTy, NULL));
+ kAsanRegisterGlobalsName, IRB.getVoidTy(),
+ IntptrTy, IntptrTy, NULL));
AsanRegisterGlobals->setLinkage(Function::ExternalLinkage);
IRB.CreateCall2(AsanRegisterGlobals,
}
// virtual
-bool AddressSanitizer::runOnModule(Module &M) {
+bool AddressSanitizer::doInitialization(Module &M) {
// Initialize the private fields. No one has accessed them before.
- TD = getAnalysisIfAvailable<TargetData>();
+ TD = getAnalysisIfAvailable<DataLayout>();
+
if (!TD)
return false;
- BL.reset(new FunctionBlackList(ClBlackListFile));
+ BL.reset(new BlackList(ClBlackListFile));
+ DynamicallyInitializedGlobals.Init(M);
C = &(M.getContext());
LongSize = TD->getPointerSizeInBits();
FunctionType::get(Type::getVoidTy(*C), false),
GlobalValue::InternalLinkage, kAsanModuleCtorName, &M);
BasicBlock *AsanCtorBB = BasicBlock::Create(*C, "", AsanCtorFunction);
- CtorInsertBefore = ReturnInst::Create(*C, AsanCtorBB);
-
// call __asan_init in the module ctor.
- IRBuilder<> IRB(CtorInsertBefore);
+ IRBuilder<> IRB(ReturnInst::Create(*C, AsanCtorBB));
AsanInitFunction = checkInterfaceFunction(
M.getOrInsertFunction(kAsanInitName, IRB.getVoidTy(), NULL));
AsanInitFunction->setLinkage(Function::ExternalLinkage);
// IsWrite and TypeSize are encoded in the function name.
std::string FunctionName = std::string(kAsanReportErrorTemplate) +
(AccessIsWrite ? "store" : "load") + itostr(1 << AccessSizeIndex);
- AsanErrorCallback[AccessIsWrite][AccessSizeIndex] = cast<Function>(
- M.getOrInsertFunction(FunctionName, IRB.getVoidTy(), IntptrTy, NULL));
+ // If we are merging crash callbacks, they have two parameters.
+ AsanErrorCallback[AccessIsWrite][AccessSizeIndex] =
+ checkInterfaceFunction(M.getOrInsertFunction(
+ FunctionName, IRB.getVoidTy(), IntptrTy, NULL));
}
}
+ AsanStackMallocFunc = checkInterfaceFunction(M.getOrInsertFunction(
+ kAsanStackMallocName, IntptrTy, IntptrTy, IntptrTy, NULL));
+ AsanStackFreeFunc = checkInterfaceFunction(M.getOrInsertFunction(
+ kAsanStackFreeName, IRB.getVoidTy(),
+ IntptrTy, IntptrTy, IntptrTy, NULL));
+ AsanHandleNoReturnFunc = checkInterfaceFunction(M.getOrInsertFunction(
+ kAsanHandleNoReturnName, IRB.getVoidTy(), NULL));
+
+ // We insert an empty inline asm after __asan_report* to avoid callback merge.
+ EmptyAsm = InlineAsm::get(FunctionType::get(IRB.getVoidTy(), false),
+ StringRef(""), StringRef(""),
+ /*hasSideEffects=*/true);
+
llvm::Triple targetTriple(M.getTargetTriple());
- bool isAndroid = targetTriple.getEnvironment() == llvm::Triple::ANDROIDEABI;
+ bool isAndroid = targetTriple.getEnvironment() == llvm::Triple::Android;
MappingOffset = isAndroid ? kDefaultShadowOffsetAndroid :
(LongSize == 32 ? kDefaultShadowOffset32 : kDefaultShadowOffset64);
MappingOffset = 1ULL << ClMappingOffsetLog;
}
}
- MappingScale = kDefaultShadowScale;
- if (ClMappingScale) {
- MappingScale = ClMappingScale;
- }
- // Redzone used for stack and globals is at least 32 bytes.
- // For scales 6 and 7, the redzone has to be 64 and 128 bytes respectively.
- RedzoneSize = std::max(32, (int)(1 << MappingScale));
-
- bool Res = false;
- if (ClGlobals)
- Res |= insertGlobalRedzones(M);
if (ClMappingOffsetLog >= 0) {
// Tell the run-time the current values of mapping offset and scale.
if (ClMappingScale) {
GlobalValue *asan_mapping_scale =
new GlobalVariable(M, IntptrTy, true, GlobalValue::LinkOnceODRLinkage,
- ConstantInt::get(IntptrTy, MappingScale),
+ ConstantInt::get(IntptrTy, MappingScale()),
kAsanMappingScaleName);
// Read the global, otherwise it may be optimized away.
IRB.CreateLoad(asan_mapping_scale, true);
}
-
- for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
- if (F->isDeclaration()) continue;
- Res |= handleFunction(M, *F);
- }
-
appendToGlobalCtors(M, AsanCtorFunction, kAsanCtorAndCtorPriority);
- return Res;
+ return true;
}
+bool AddressSanitizer::doFinalization(Module &M) {
+ // We transform the globals at the very end so that the optimization analysis
+ // works on the original globals.
+ if (ClGlobals) {
+ // FIXME: instead of doFinalization, run this as a true ModulePass.
+ AddressSanitizerCreateGlobalRedzonesPass Pass;
+ return Pass.runOnModule(M, TD);
+ }
+ return false;
+}
+
+
bool AddressSanitizer::maybeInsertAsanInitAtFunctionEntry(Function &F) {
// For each NSObject descendant having a +load method, this method is invoked
// by the ObjC runtime before any of the static constructors is called.
return false;
}
-bool AddressSanitizer::handleFunction(Module &M, Function &F) {
+bool AddressSanitizer::runOnFunction(Function &F) {
if (BL->isIn(F)) return false;
if (&F == AsanCtorFunction) return false;
+ DEBUG(dbgs() << "ASAN instrumenting:\n" << F << "\n");
// If needed, insert __asan_init before checking for AddressSafety attr.
maybeInsertAsanInitAtFunctionEntry(F);
- if (!F.hasFnAttr(Attribute::AddressSafety)) return false;
+ if (!F.getFnAttributes().hasAttribute(Attributes::AddressSafety))
+ return false;
if (!ClDebugFunc.empty() && ClDebugFunc != F.getName())
return false;
- // We want to instrument every address only once per basic block
- // (unless there are calls between uses).
+
+ // We want to instrument every address only once per basic block (unless there
+ // are calls between uses).
SmallSet<Value*, 16> TempsToInstrument;
SmallVector<Instruction*, 16> ToInstrument;
SmallVector<Instruction*, 8> NoReturnCalls;
NumInstrumented++;
}
- DEBUG(dbgs() << F);
-
- bool ChangedStack = poisonStackInFunction(M, F);
+ bool ChangedStack = poisonStackInFunction(F);
// We must unpoison the stack before every NoReturn call (throw, _exit, etc).
// See e.g. http://code.google.com/p/address-sanitizer/issues/detail?id=37
for (size_t i = 0, n = NoReturnCalls.size(); i != n; i++) {
Instruction *CI = NoReturnCalls[i];
IRBuilder<> IRB(CI);
- IRB.CreateCall(M.getOrInsertFunction(kAsanHandleNoReturnName,
- IRB.getVoidTy(), NULL));
+ IRB.CreateCall(AsanHandleNoReturnFunc);
}
+ DEBUG(dbgs() << "ASAN done instrumenting:\n" << F << "\n");
return NumInstrumented > 0 || ChangedStack || !NoReturnCalls.empty();
}
static void PoisonShadowPartialRightRedzone(uint8_t *Shadow,
size_t Size,
- size_t RedzoneSize,
+ size_t RZSize,
size_t ShadowGranularity,
uint8_t Magic) {
- for (size_t i = 0; i < RedzoneSize;
+ for (size_t i = 0; i < RZSize;
i+= ShadowGranularity, Shadow++) {
if (i + ShadowGranularity <= Size) {
*Shadow = 0; // fully addressable
void AddressSanitizer::PoisonStack(const ArrayRef<AllocaInst*> &AllocaVec,
IRBuilder<> IRB,
Value *ShadowBase, bool DoPoison) {
- size_t ShadowRZSize = RedzoneSize >> MappingScale;
+ size_t ShadowRZSize = RedzoneSize() >> MappingScale();
assert(ShadowRZSize >= 1 && ShadowRZSize <= 4);
Type *RZTy = Type::getIntNTy(*C, ShadowRZSize * 8);
Type *RZPtrTy = PointerType::get(RZTy, 0);
IRB.CreateStore(PoisonLeft, IRB.CreateIntToPtr(ShadowBase, RZPtrTy));
// poison all other red zones.
- uint64_t Pos = RedzoneSize;
+ uint64_t Pos = RedzoneSize();
for (size_t i = 0, n = AllocaVec.size(); i < n; i++) {
AllocaInst *AI = AllocaVec[i];
uint64_t SizeInBytes = getAllocaSizeInBytes(AI);
uint64_t AlignedSize = getAlignedAllocaSize(AI);
- assert(AlignedSize - SizeInBytes < RedzoneSize);
+ assert(AlignedSize - SizeInBytes < RedzoneSize());
Value *Ptr = NULL;
Pos += AlignedSize;
// Poison the partial redzone at right
Ptr = IRB.CreateAdd(
ShadowBase, ConstantInt::get(IntptrTy,
- (Pos >> MappingScale) - ShadowRZSize));
- size_t AddressableBytes = RedzoneSize - (AlignedSize - SizeInBytes);
+ (Pos >> MappingScale()) - ShadowRZSize));
+ size_t AddressableBytes = RedzoneSize() - (AlignedSize - SizeInBytes);
uint32_t Poison = 0;
if (DoPoison) {
PoisonShadowPartialRightRedzone((uint8_t*)&Poison, AddressableBytes,
- RedzoneSize,
- 1ULL << MappingScale,
+ RedzoneSize(),
+ 1ULL << MappingScale(),
kAsanStackPartialRedzoneMagic);
}
Value *PartialPoison = ConstantInt::get(RZTy, Poison);
// Poison the full redzone at right.
Ptr = IRB.CreateAdd(ShadowBase,
- ConstantInt::get(IntptrTy, Pos >> MappingScale));
+ ConstantInt::get(IntptrTy, Pos >> MappingScale()));
Value *Poison = i == AllocaVec.size() - 1 ? PoisonRight : PoisonMid;
IRB.CreateStore(Poison, IRB.CreateIntToPtr(Ptr, RZPtrTy));
- Pos += RedzoneSize;
+ Pos += RedzoneSize();
}
}
// compiler hoists the load of the shadow value somewhere too high.
// This causes asan to report a non-existing bug on 453.povray.
// It sounds like an LLVM bug.
-bool AddressSanitizer::poisonStackInFunction(Module &M, Function &F) {
+bool AddressSanitizer::poisonStackInFunction(Function &F) {
if (!ClStack) return false;
SmallVector<AllocaInst*, 16> AllocaVec;
SmallVector<Instruction*, 8> RetVec;
if (AI->isArrayAllocation()) continue;
if (!AI->isStaticAlloca()) continue;
if (!AI->getAllocatedType()->isSized()) continue;
- if (AI->getAlignment() > RedzoneSize) continue;
+ if (AI->getAlignment() > RedzoneSize()) continue;
AllocaVec.push_back(AI);
uint64_t AlignedSize = getAlignedAllocaSize(AI);
TotalSize += AlignedSize;
if (AllocaVec.empty()) return false;
- uint64_t LocalStackSize = TotalSize + (AllocaVec.size() + 1) * RedzoneSize;
+ uint64_t LocalStackSize = TotalSize + (AllocaVec.size() + 1) * RedzoneSize();
bool DoStackMalloc = ClUseAfterReturn
&& LocalStackSize <= kMaxStackMallocSize;
Type *ByteArrayTy = ArrayType::get(IRB.getInt8Ty(), LocalStackSize);
AllocaInst *MyAlloca =
new AllocaInst(ByteArrayTy, "MyAlloca", InsBefore);
- MyAlloca->setAlignment(RedzoneSize);
+ MyAlloca->setAlignment(RedzoneSize());
assert(MyAlloca->isStaticAlloca());
Value *OrigStackBase = IRB.CreatePointerCast(MyAlloca, IntptrTy);
Value *LocalStackBase = OrigStackBase;
if (DoStackMalloc) {
- Value *AsanStackMallocFunc = M.getOrInsertFunction(
- kAsanStackMallocName, IntptrTy, IntptrTy, IntptrTy, NULL);
LocalStackBase = IRB.CreateCall2(AsanStackMallocFunc,
ConstantInt::get(IntptrTy, LocalStackSize), OrigStackBase);
}
raw_svector_ostream StackDescription(StackDescriptionStorage);
StackDescription << F.getName() << " " << AllocaVec.size() << " ";
- uint64_t Pos = RedzoneSize;
+ uint64_t Pos = RedzoneSize();
// Replace Alloca instructions with base+offset.
for (size_t i = 0, n = AllocaVec.size(); i < n; i++) {
AllocaInst *AI = AllocaVec[i];
StackDescription << Pos << " " << SizeInBytes << " "
<< Name.size() << " " << Name << " ";
uint64_t AlignedSize = getAlignedAllocaSize(AI);
- assert((AlignedSize % RedzoneSize) == 0);
+ assert((AlignedSize % RedzoneSize()) == 0);
AI->replaceAllUsesWith(
IRB.CreateIntToPtr(
IRB.CreateAdd(LocalStackBase, ConstantInt::get(IntptrTy, Pos)),
AI->getType()));
- Pos += AlignedSize + RedzoneSize;
+ Pos += AlignedSize + RedzoneSize();
}
assert(Pos == LocalStackSize);
Value *BasePlus1 = IRB.CreateAdd(LocalStackBase,
ConstantInt::get(IntptrTy, LongSize/8));
BasePlus1 = IRB.CreateIntToPtr(BasePlus1, IntptrPtrTy);
- Value *Description = IRB.CreatePointerCast(
- createPrivateGlobalForString(M, StackDescription.str()),
- IntptrTy);
+ GlobalVariable *StackDescriptionGlobal =
+ createPrivateGlobalForString(*F.getParent(), StackDescription.str());
+ Value *Description = IRB.CreatePointerCast(StackDescriptionGlobal, IntptrTy);
IRB.CreateStore(Description, BasePlus1);
// Poison the stack redzones at the entry.
Value *ShadowBase = memToShadow(LocalStackBase, IRB);
PoisonStack(ArrayRef<AllocaInst*>(AllocaVec), IRB, ShadowBase, true);
- Value *AsanStackFreeFunc = NULL;
- if (DoStackMalloc) {
- AsanStackFreeFunc = M.getOrInsertFunction(
- kAsanStackFreeName, IRB.getVoidTy(),
- IntptrTy, IntptrTy, IntptrTy, NULL);
- }
-
// Unpoison the stack before all ret instructions.
for (size_t i = 0, n = RetVec.size(); i < n; i++) {
Instruction *Ret = RetVec[i];
}
}
+ // We are done. Remove the old unused alloca instructions.
+ for (size_t i = 0, n = AllocaVec.size(); i < n; i++)
+ AllocaVec[i]->eraseFromParent();
+
if (ClDebugStack) {
DEBUG(dbgs() << F);
}