1 //===------ MemoryBuiltins.cpp - Identify calls to memory builtins --------===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This family of functions identifies calls to builtin functions that allocate
11 // or free memory.
12 //
13 //===----------------------------------------------------------------------===//
14 
15 #include "llvm/Analysis/MemoryBuiltins.h"
16 #include "llvm/ADT/STLExtras.h"
17 #include "llvm/ADT/Statistic.h"
18 #include "llvm/Analysis/TargetLibraryInfo.h"
19 #include "llvm/Analysis/ValueTracking.h"
20 #include "llvm/IR/DataLayout.h"
21 #include "llvm/IR/GlobalVariable.h"
22 #include "llvm/IR/Instructions.h"
23 #include "llvm/IR/Intrinsics.h"
24 #include "llvm/IR/Metadata.h"
25 #include "llvm/IR/Module.h"
26 #include "llvm/Support/Debug.h"
27 #include "llvm/Support/MathExtras.h"
28 #include "llvm/Support/raw_ostream.h"
29 #include "llvm/Transforms/Utils/Local.h"
30 using namespace llvm;
31 
32 #define DEBUG_TYPE "memory-builtins"
33 
34 enum AllocType {
35   OpNewLike          = 1<<0, // allocates; never returns null
36   MallocLike         = 1<<1 | OpNewLike, // allocates; may return null
37   CallocLike         = 1<<2, // allocates + bzero
38   ReallocLike        = 1<<3, // reallocates
39   StrDupLike         = 1<<4,
40   AllocLike          = MallocLike | CallocLike | StrDupLike,
41   AnyAlloc           = AllocLike | ReallocLike
42 };
43 
44 struct AllocFnsTy {
45   LibFunc::Func Func;
46   AllocType AllocTy;
47   unsigned char NumParams;
48   // First and Second size parameters (or -1 if unused)
49   signed char FstParam, SndParam;
50 };
51 
52 // FIXME: certain users need more information. E.g., SimplifyLibCalls needs to
53 // know which functions are nounwind, noalias, nocapture parameters, etc.
54 static const AllocFnsTy AllocationFnData[] = {
55   {LibFunc::malloc,              MallocLike,  1, 0,  -1},
56   {LibFunc::valloc,              MallocLike,  1, 0,  -1},
57   {LibFunc::Znwj,                OpNewLike,   1, 0,  -1}, // new(unsigned int)
58   {LibFunc::ZnwjRKSt9nothrow_t,  MallocLike,  2, 0,  -1}, // new(unsigned int, nothrow)
59   {LibFunc::Znwm,                OpNewLike,   1, 0,  -1}, // new(unsigned long)
60   {LibFunc::ZnwmRKSt9nothrow_t,  MallocLike,  2, 0,  -1}, // new(unsigned long, nothrow)
61   {LibFunc::Znaj,                OpNewLike,   1, 0,  -1}, // new[](unsigned int)
62   {LibFunc::ZnajRKSt9nothrow_t,  MallocLike,  2, 0,  -1}, // new[](unsigned int, nothrow)
63   {LibFunc::Znam,                OpNewLike,   1, 0,  -1}, // new[](unsigned long)
64   {LibFunc::ZnamRKSt9nothrow_t,  MallocLike,  2, 0,  -1}, // new[](unsigned long, nothrow)
65   {LibFunc::calloc,              CallocLike,  2, 0,   1},
66   {LibFunc::realloc,             ReallocLike, 2, 1,  -1},
67   {LibFunc::reallocf,            ReallocLike, 2, 1,  -1},
68   {LibFunc::strdup,              StrDupLike,  1, -1, -1},
69   {LibFunc::strndup,             StrDupLike,  2, 1,  -1}
70   // TODO: Handle "int posix_memalign(void **, size_t, size_t)"
71 };
72 
73 
getCalledFunction(const Value * V,bool LookThroughBitCast)74 static Function *getCalledFunction(const Value *V, bool LookThroughBitCast) {
75   if (LookThroughBitCast)
76     V = V->stripPointerCasts();
77 
78   CallSite CS(const_cast<Value*>(V));
79   if (!CS.getInstruction())
80     return nullptr;
81 
82   if (CS.isNoBuiltin())
83     return nullptr;
84 
85   Function *Callee = CS.getCalledFunction();
86   if (!Callee || !Callee->isDeclaration())
87     return nullptr;
88   return Callee;
89 }
90 
91 /// \brief Returns the allocation data for the given value if it is a call to a
92 /// known allocation function, and NULL otherwise.
getAllocationData(const Value * V,AllocType AllocTy,const TargetLibraryInfo * TLI,bool LookThroughBitCast=false)93 static const AllocFnsTy *getAllocationData(const Value *V, AllocType AllocTy,
94                                            const TargetLibraryInfo *TLI,
95                                            bool LookThroughBitCast = false) {
96   // Skip intrinsics
97   if (isa<IntrinsicInst>(V))
98     return nullptr;
99 
100   Function *Callee = getCalledFunction(V, LookThroughBitCast);
101   if (!Callee)
102     return nullptr;
103 
104   // Make sure that the function is available.
105   StringRef FnName = Callee->getName();
106   LibFunc::Func TLIFn;
107   if (!TLI || !TLI->getLibFunc(FnName, TLIFn) || !TLI->has(TLIFn))
108     return nullptr;
109 
110   unsigned i = 0;
111   bool found = false;
112   for ( ; i < array_lengthof(AllocationFnData); ++i) {
113     if (AllocationFnData[i].Func == TLIFn) {
114       found = true;
115       break;
116     }
117   }
118   if (!found)
119     return nullptr;
120 
121   const AllocFnsTy *FnData = &AllocationFnData[i];
122   if ((FnData->AllocTy & AllocTy) != FnData->AllocTy)
123     return nullptr;
124 
125   // Check function prototype.
126   int FstParam = FnData->FstParam;
127   int SndParam = FnData->SndParam;
128   FunctionType *FTy = Callee->getFunctionType();
129 
130   if (FTy->getReturnType() == Type::getInt8PtrTy(FTy->getContext()) &&
131       FTy->getNumParams() == FnData->NumParams &&
132       (FstParam < 0 ||
133        (FTy->getParamType(FstParam)->isIntegerTy(32) ||
134         FTy->getParamType(FstParam)->isIntegerTy(64))) &&
135       (SndParam < 0 ||
136        FTy->getParamType(SndParam)->isIntegerTy(32) ||
137        FTy->getParamType(SndParam)->isIntegerTy(64)))
138     return FnData;
139   return nullptr;
140 }
141 
hasNoAliasAttr(const Value * V,bool LookThroughBitCast)142 static bool hasNoAliasAttr(const Value *V, bool LookThroughBitCast) {
143   ImmutableCallSite CS(LookThroughBitCast ? V->stripPointerCasts() : V);
144   return CS && CS.hasFnAttr(Attribute::NoAlias);
145 }
146 
147 
148 /// \brief Tests if a value is a call or invoke to a library function that
149 /// allocates or reallocates memory (either malloc, calloc, realloc, or strdup
150 /// like).
isAllocationFn(const Value * V,const TargetLibraryInfo * TLI,bool LookThroughBitCast)151 bool llvm::isAllocationFn(const Value *V, const TargetLibraryInfo *TLI,
152                           bool LookThroughBitCast) {
153   return getAllocationData(V, AnyAlloc, TLI, LookThroughBitCast);
154 }
155 
156 /// \brief Tests if a value is a call or invoke to a function that returns a
157 /// NoAlias pointer (including malloc/calloc/realloc/strdup-like functions).
isNoAliasFn(const Value * V,const TargetLibraryInfo * TLI,bool LookThroughBitCast)158 bool llvm::isNoAliasFn(const Value *V, const TargetLibraryInfo *TLI,
159                        bool LookThroughBitCast) {
160   // it's safe to consider realloc as noalias since accessing the original
161   // pointer is undefined behavior
162   return isAllocationFn(V, TLI, LookThroughBitCast) ||
163          hasNoAliasAttr(V, LookThroughBitCast);
164 }
165 
166 /// \brief Tests if a value is a call or invoke to a library function that
167 /// allocates uninitialized memory (such as malloc).
isMallocLikeFn(const Value * V,const TargetLibraryInfo * TLI,bool LookThroughBitCast)168 bool llvm::isMallocLikeFn(const Value *V, const TargetLibraryInfo *TLI,
169                           bool LookThroughBitCast) {
170   return getAllocationData(V, MallocLike, TLI, LookThroughBitCast);
171 }
172 
173 /// \brief Tests if a value is a call or invoke to a library function that
174 /// allocates zero-filled memory (such as calloc).
isCallocLikeFn(const Value * V,const TargetLibraryInfo * TLI,bool LookThroughBitCast)175 bool llvm::isCallocLikeFn(const Value *V, const TargetLibraryInfo *TLI,
176                           bool LookThroughBitCast) {
177   return getAllocationData(V, CallocLike, TLI, LookThroughBitCast);
178 }
179 
180 /// \brief Tests if a value is a call or invoke to a library function that
181 /// allocates memory (either malloc, calloc, or strdup like).
isAllocLikeFn(const Value * V,const TargetLibraryInfo * TLI,bool LookThroughBitCast)182 bool llvm::isAllocLikeFn(const Value *V, const TargetLibraryInfo *TLI,
183                          bool LookThroughBitCast) {
184   return getAllocationData(V, AllocLike, TLI, LookThroughBitCast);
185 }
186 
187 /// \brief Tests if a value is a call or invoke to a library function that
188 /// reallocates memory (such as realloc).
isReallocLikeFn(const Value * V,const TargetLibraryInfo * TLI,bool LookThroughBitCast)189 bool llvm::isReallocLikeFn(const Value *V, const TargetLibraryInfo *TLI,
190                            bool LookThroughBitCast) {
191   return getAllocationData(V, ReallocLike, TLI, LookThroughBitCast);
192 }
193 
194 /// \brief Tests if a value is a call or invoke to a library function that
195 /// allocates memory and never returns null (such as operator new).
isOperatorNewLikeFn(const Value * V,const TargetLibraryInfo * TLI,bool LookThroughBitCast)196 bool llvm::isOperatorNewLikeFn(const Value *V, const TargetLibraryInfo *TLI,
197                                bool LookThroughBitCast) {
198   return getAllocationData(V, OpNewLike, TLI, LookThroughBitCast);
199 }
200 
201 /// extractMallocCall - Returns the corresponding CallInst if the instruction
202 /// is a malloc call.  Since CallInst::CreateMalloc() only creates calls, we
203 /// ignore InvokeInst here.
extractMallocCall(const Value * I,const TargetLibraryInfo * TLI)204 const CallInst *llvm::extractMallocCall(const Value *I,
205                                         const TargetLibraryInfo *TLI) {
206   return isMallocLikeFn(I, TLI) ? dyn_cast<CallInst>(I) : nullptr;
207 }
208 
computeArraySize(const CallInst * CI,const DataLayout & DL,const TargetLibraryInfo * TLI,bool LookThroughSExt=false)209 static Value *computeArraySize(const CallInst *CI, const DataLayout &DL,
210                                const TargetLibraryInfo *TLI,
211                                bool LookThroughSExt = false) {
212   if (!CI)
213     return nullptr;
214 
215   // The size of the malloc's result type must be known to determine array size.
216   Type *T = getMallocAllocatedType(CI, TLI);
217   if (!T || !T->isSized())
218     return nullptr;
219 
220   unsigned ElementSize = DL.getTypeAllocSize(T);
221   if (StructType *ST = dyn_cast<StructType>(T))
222     ElementSize = DL.getStructLayout(ST)->getSizeInBytes();
223 
224   // If malloc call's arg can be determined to be a multiple of ElementSize,
225   // return the multiple.  Otherwise, return NULL.
226   Value *MallocArg = CI->getArgOperand(0);
227   Value *Multiple = nullptr;
228   if (ComputeMultiple(MallocArg, ElementSize, Multiple,
229                       LookThroughSExt))
230     return Multiple;
231 
232   return nullptr;
233 }
234 
235 /// getMallocType - Returns the PointerType resulting from the malloc call.
236 /// The PointerType depends on the number of bitcast uses of the malloc call:
237 ///   0: PointerType is the calls' return type.
238 ///   1: PointerType is the bitcast's result type.
239 ///  >1: Unique PointerType cannot be determined, return NULL.
getMallocType(const CallInst * CI,const TargetLibraryInfo * TLI)240 PointerType *llvm::getMallocType(const CallInst *CI,
241                                  const TargetLibraryInfo *TLI) {
242   assert(isMallocLikeFn(CI, TLI) && "getMallocType and not malloc call");
243 
244   PointerType *MallocType = nullptr;
245   unsigned NumOfBitCastUses = 0;
246 
247   // Determine if CallInst has a bitcast use.
248   for (Value::const_user_iterator UI = CI->user_begin(), E = CI->user_end();
249        UI != E;)
250     if (const BitCastInst *BCI = dyn_cast<BitCastInst>(*UI++)) {
251       MallocType = cast<PointerType>(BCI->getDestTy());
252       NumOfBitCastUses++;
253     }
254 
255   // Malloc call has 1 bitcast use, so type is the bitcast's destination type.
256   if (NumOfBitCastUses == 1)
257     return MallocType;
258 
259   // Malloc call was not bitcast, so type is the malloc function's return type.
260   if (NumOfBitCastUses == 0)
261     return cast<PointerType>(CI->getType());
262 
263   // Type could not be determined.
264   return nullptr;
265 }
266 
267 /// getMallocAllocatedType - Returns the Type allocated by malloc call.
268 /// The Type depends on the number of bitcast uses of the malloc call:
269 ///   0: PointerType is the malloc calls' return type.
270 ///   1: PointerType is the bitcast's result type.
271 ///  >1: Unique PointerType cannot be determined, return NULL.
getMallocAllocatedType(const CallInst * CI,const TargetLibraryInfo * TLI)272 Type *llvm::getMallocAllocatedType(const CallInst *CI,
273                                    const TargetLibraryInfo *TLI) {
274   PointerType *PT = getMallocType(CI, TLI);
275   return PT ? PT->getElementType() : nullptr;
276 }
277 
278 /// getMallocArraySize - Returns the array size of a malloc call.  If the
279 /// argument passed to malloc is a multiple of the size of the malloced type,
280 /// then return that multiple.  For non-array mallocs, the multiple is
281 /// constant 1.  Otherwise, return NULL for mallocs whose array size cannot be
282 /// determined.
getMallocArraySize(CallInst * CI,const DataLayout & DL,const TargetLibraryInfo * TLI,bool LookThroughSExt)283 Value *llvm::getMallocArraySize(CallInst *CI, const DataLayout &DL,
284                                 const TargetLibraryInfo *TLI,
285                                 bool LookThroughSExt) {
286   assert(isMallocLikeFn(CI, TLI) && "getMallocArraySize and not malloc call");
287   return computeArraySize(CI, DL, TLI, LookThroughSExt);
288 }
289 
290 
291 /// extractCallocCall - Returns the corresponding CallInst if the instruction
292 /// is a calloc call.
extractCallocCall(const Value * I,const TargetLibraryInfo * TLI)293 const CallInst *llvm::extractCallocCall(const Value *I,
294                                         const TargetLibraryInfo *TLI) {
295   return isCallocLikeFn(I, TLI) ? cast<CallInst>(I) : nullptr;
296 }
297 
298 
299 /// isFreeCall - Returns non-null if the value is a call to the builtin free()
isFreeCall(const Value * I,const TargetLibraryInfo * TLI)300 const CallInst *llvm::isFreeCall(const Value *I, const TargetLibraryInfo *TLI) {
301   const CallInst *CI = dyn_cast<CallInst>(I);
302   if (!CI || isa<IntrinsicInst>(CI))
303     return nullptr;
304   Function *Callee = CI->getCalledFunction();
305   if (Callee == nullptr)
306     return nullptr;
307 
308   StringRef FnName = Callee->getName();
309   LibFunc::Func TLIFn;
310   if (!TLI || !TLI->getLibFunc(FnName, TLIFn) || !TLI->has(TLIFn))
311     return nullptr;
312 
313   unsigned ExpectedNumParams;
314   if (TLIFn == LibFunc::free ||
315       TLIFn == LibFunc::ZdlPv || // operator delete(void*)
316       TLIFn == LibFunc::ZdaPv)   // operator delete[](void*)
317     ExpectedNumParams = 1;
318   else if (TLIFn == LibFunc::ZdlPvj ||              // delete(void*, uint)
319            TLIFn == LibFunc::ZdlPvm ||              // delete(void*, ulong)
320            TLIFn == LibFunc::ZdlPvRKSt9nothrow_t || // delete(void*, nothrow)
321            TLIFn == LibFunc::ZdaPvj ||              // delete[](void*, uint)
322            TLIFn == LibFunc::ZdaPvm ||              // delete[](void*, ulong)
323            TLIFn == LibFunc::ZdaPvRKSt9nothrow_t)   // delete[](void*, nothrow)
324     ExpectedNumParams = 2;
325   else
326     return nullptr;
327 
328   // Check free prototype.
329   // FIXME: workaround for PR5130, this will be obsolete when a nobuiltin
330   // attribute will exist.
331   FunctionType *FTy = Callee->getFunctionType();
332   if (!FTy->getReturnType()->isVoidTy())
333     return nullptr;
334   if (FTy->getNumParams() != ExpectedNumParams)
335     return nullptr;
336   if (FTy->getParamType(0) != Type::getInt8PtrTy(Callee->getContext()))
337     return nullptr;
338 
339   return CI;
340 }
341 
342 
343 
344 //===----------------------------------------------------------------------===//
345 //  Utility functions to compute size of objects.
346 //
347 
348 
349 /// \brief Compute the size of the object pointed by Ptr. Returns true and the
350 /// object size in Size if successful, and false otherwise.
351 /// If RoundToAlign is true, then Size is rounded up to the aligment of allocas,
352 /// byval arguments, and global variables.
getObjectSize(const Value * Ptr,uint64_t & Size,const DataLayout & DL,const TargetLibraryInfo * TLI,bool RoundToAlign)353 bool llvm::getObjectSize(const Value *Ptr, uint64_t &Size, const DataLayout &DL,
354                          const TargetLibraryInfo *TLI, bool RoundToAlign) {
355   ObjectSizeOffsetVisitor Visitor(DL, TLI, Ptr->getContext(), RoundToAlign);
356   SizeOffsetType Data = Visitor.compute(const_cast<Value*>(Ptr));
357   if (!Visitor.bothKnown(Data))
358     return false;
359 
360   APInt ObjSize = Data.first, Offset = Data.second;
361   // check for overflow
362   if (Offset.slt(0) || ObjSize.ult(Offset))
363     Size = 0;
364   else
365     Size = (ObjSize - Offset).getZExtValue();
366   return true;
367 }
368 
369 
370 STATISTIC(ObjectVisitorArgument,
371           "Number of arguments with unsolved size and offset");
372 STATISTIC(ObjectVisitorLoad,
373           "Number of load instructions with unsolved size and offset");
374 
375 
align(APInt Size,uint64_t Align)376 APInt ObjectSizeOffsetVisitor::align(APInt Size, uint64_t Align) {
377   if (RoundToAlign && Align)
378     return APInt(IntTyBits, RoundUpToAlignment(Size.getZExtValue(), Align));
379   return Size;
380 }
381 
ObjectSizeOffsetVisitor(const DataLayout & DL,const TargetLibraryInfo * TLI,LLVMContext & Context,bool RoundToAlign)382 ObjectSizeOffsetVisitor::ObjectSizeOffsetVisitor(const DataLayout &DL,
383                                                  const TargetLibraryInfo *TLI,
384                                                  LLVMContext &Context,
385                                                  bool RoundToAlign)
386     : DL(DL), TLI(TLI), RoundToAlign(RoundToAlign) {
387   // Pointer size must be rechecked for each object visited since it could have
388   // a different address space.
389 }
390 
compute(Value * V)391 SizeOffsetType ObjectSizeOffsetVisitor::compute(Value *V) {
392   IntTyBits = DL.getPointerTypeSizeInBits(V->getType());
393   Zero = APInt::getNullValue(IntTyBits);
394 
395   V = V->stripPointerCasts();
396   if (Instruction *I = dyn_cast<Instruction>(V)) {
397     // If we have already seen this instruction, bail out. Cycles can happen in
398     // unreachable code after constant propagation.
399     if (!SeenInsts.insert(I).second)
400       return unknown();
401 
402     if (GEPOperator *GEP = dyn_cast<GEPOperator>(V))
403       return visitGEPOperator(*GEP);
404     return visit(*I);
405   }
406   if (Argument *A = dyn_cast<Argument>(V))
407     return visitArgument(*A);
408   if (ConstantPointerNull *P = dyn_cast<ConstantPointerNull>(V))
409     return visitConstantPointerNull(*P);
410   if (GlobalAlias *GA = dyn_cast<GlobalAlias>(V))
411     return visitGlobalAlias(*GA);
412   if (GlobalVariable *GV = dyn_cast<GlobalVariable>(V))
413     return visitGlobalVariable(*GV);
414   if (UndefValue *UV = dyn_cast<UndefValue>(V))
415     return visitUndefValue(*UV);
416   if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V)) {
417     if (CE->getOpcode() == Instruction::IntToPtr)
418       return unknown(); // clueless
419     if (CE->getOpcode() == Instruction::GetElementPtr)
420       return visitGEPOperator(cast<GEPOperator>(*CE));
421   }
422 
423   DEBUG(dbgs() << "ObjectSizeOffsetVisitor::compute() unhandled value: " << *V
424         << '\n');
425   return unknown();
426 }
427 
visitAllocaInst(AllocaInst & I)428 SizeOffsetType ObjectSizeOffsetVisitor::visitAllocaInst(AllocaInst &I) {
429   if (!I.getAllocatedType()->isSized())
430     return unknown();
431 
432   APInt Size(IntTyBits, DL.getTypeAllocSize(I.getAllocatedType()));
433   if (!I.isArrayAllocation())
434     return std::make_pair(align(Size, I.getAlignment()), Zero);
435 
436   Value *ArraySize = I.getArraySize();
437   if (const ConstantInt *C = dyn_cast<ConstantInt>(ArraySize)) {
438     Size *= C->getValue().zextOrSelf(IntTyBits);
439     return std::make_pair(align(Size, I.getAlignment()), Zero);
440   }
441   return unknown();
442 }
443 
visitArgument(Argument & A)444 SizeOffsetType ObjectSizeOffsetVisitor::visitArgument(Argument &A) {
445   // no interprocedural analysis is done at the moment
446   if (!A.hasByValOrInAllocaAttr()) {
447     ++ObjectVisitorArgument;
448     return unknown();
449   }
450   PointerType *PT = cast<PointerType>(A.getType());
451   APInt Size(IntTyBits, DL.getTypeAllocSize(PT->getElementType()));
452   return std::make_pair(align(Size, A.getParamAlignment()), Zero);
453 }
454 
visitCallSite(CallSite CS)455 SizeOffsetType ObjectSizeOffsetVisitor::visitCallSite(CallSite CS) {
456   const AllocFnsTy *FnData = getAllocationData(CS.getInstruction(), AnyAlloc,
457                                                TLI);
458   if (!FnData)
459     return unknown();
460 
461   // handle strdup-like functions separately
462   if (FnData->AllocTy == StrDupLike) {
463     APInt Size(IntTyBits, GetStringLength(CS.getArgument(0)));
464     if (!Size)
465       return unknown();
466 
467     // strndup limits strlen
468     if (FnData->FstParam > 0) {
469       ConstantInt *Arg= dyn_cast<ConstantInt>(CS.getArgument(FnData->FstParam));
470       if (!Arg)
471         return unknown();
472 
473       APInt MaxSize = Arg->getValue().zextOrSelf(IntTyBits);
474       if (Size.ugt(MaxSize))
475         Size = MaxSize + 1;
476     }
477     return std::make_pair(Size, Zero);
478   }
479 
480   ConstantInt *Arg = dyn_cast<ConstantInt>(CS.getArgument(FnData->FstParam));
481   if (!Arg)
482     return unknown();
483 
484   APInt Size = Arg->getValue().zextOrSelf(IntTyBits);
485   // size determined by just 1 parameter
486   if (FnData->SndParam < 0)
487     return std::make_pair(Size, Zero);
488 
489   Arg = dyn_cast<ConstantInt>(CS.getArgument(FnData->SndParam));
490   if (!Arg)
491     return unknown();
492 
493   Size *= Arg->getValue().zextOrSelf(IntTyBits);
494   return std::make_pair(Size, Zero);
495 
496   // TODO: handle more standard functions (+ wchar cousins):
497   // - strdup / strndup
498   // - strcpy / strncpy
499   // - strcat / strncat
500   // - memcpy / memmove
501   // - strcat / strncat
502   // - memset
503 }
504 
505 SizeOffsetType
visitConstantPointerNull(ConstantPointerNull &)506 ObjectSizeOffsetVisitor::visitConstantPointerNull(ConstantPointerNull&) {
507   return std::make_pair(Zero, Zero);
508 }
509 
510 SizeOffsetType
visitExtractElementInst(ExtractElementInst &)511 ObjectSizeOffsetVisitor::visitExtractElementInst(ExtractElementInst&) {
512   return unknown();
513 }
514 
515 SizeOffsetType
visitExtractValueInst(ExtractValueInst &)516 ObjectSizeOffsetVisitor::visitExtractValueInst(ExtractValueInst&) {
517   // Easy cases were already folded by previous passes.
518   return unknown();
519 }
520 
visitGEPOperator(GEPOperator & GEP)521 SizeOffsetType ObjectSizeOffsetVisitor::visitGEPOperator(GEPOperator &GEP) {
522   SizeOffsetType PtrData = compute(GEP.getPointerOperand());
523   APInt Offset(IntTyBits, 0);
524   if (!bothKnown(PtrData) || !GEP.accumulateConstantOffset(DL, Offset))
525     return unknown();
526 
527   return std::make_pair(PtrData.first, PtrData.second + Offset);
528 }
529 
visitGlobalAlias(GlobalAlias & GA)530 SizeOffsetType ObjectSizeOffsetVisitor::visitGlobalAlias(GlobalAlias &GA) {
531   if (GA.mayBeOverridden())
532     return unknown();
533   return compute(GA.getAliasee());
534 }
535 
visitGlobalVariable(GlobalVariable & GV)536 SizeOffsetType ObjectSizeOffsetVisitor::visitGlobalVariable(GlobalVariable &GV){
537   if (!GV.hasDefinitiveInitializer())
538     return unknown();
539 
540   APInt Size(IntTyBits, DL.getTypeAllocSize(GV.getType()->getElementType()));
541   return std::make_pair(align(Size, GV.getAlignment()), Zero);
542 }
543 
visitIntToPtrInst(IntToPtrInst &)544 SizeOffsetType ObjectSizeOffsetVisitor::visitIntToPtrInst(IntToPtrInst&) {
545   // clueless
546   return unknown();
547 }
548 
visitLoadInst(LoadInst &)549 SizeOffsetType ObjectSizeOffsetVisitor::visitLoadInst(LoadInst&) {
550   ++ObjectVisitorLoad;
551   return unknown();
552 }
553 
visitPHINode(PHINode &)554 SizeOffsetType ObjectSizeOffsetVisitor::visitPHINode(PHINode&) {
555   // too complex to analyze statically.
556   return unknown();
557 }
558 
visitSelectInst(SelectInst & I)559 SizeOffsetType ObjectSizeOffsetVisitor::visitSelectInst(SelectInst &I) {
560   SizeOffsetType TrueSide  = compute(I.getTrueValue());
561   SizeOffsetType FalseSide = compute(I.getFalseValue());
562   if (bothKnown(TrueSide) && bothKnown(FalseSide) && TrueSide == FalseSide)
563     return TrueSide;
564   return unknown();
565 }
566 
visitUndefValue(UndefValue &)567 SizeOffsetType ObjectSizeOffsetVisitor::visitUndefValue(UndefValue&) {
568   return std::make_pair(Zero, Zero);
569 }
570 
visitInstruction(Instruction & I)571 SizeOffsetType ObjectSizeOffsetVisitor::visitInstruction(Instruction &I) {
572   DEBUG(dbgs() << "ObjectSizeOffsetVisitor unknown instruction:" << I << '\n');
573   return unknown();
574 }
575 
ObjectSizeOffsetEvaluator(const DataLayout & DL,const TargetLibraryInfo * TLI,LLVMContext & Context,bool RoundToAlign)576 ObjectSizeOffsetEvaluator::ObjectSizeOffsetEvaluator(
577     const DataLayout &DL, const TargetLibraryInfo *TLI, LLVMContext &Context,
578     bool RoundToAlign)
579     : DL(DL), TLI(TLI), Context(Context), Builder(Context, TargetFolder(DL)),
580       RoundToAlign(RoundToAlign) {
581   // IntTy and Zero must be set for each compute() since the address space may
582   // be different for later objects.
583 }
584 
compute(Value * V)585 SizeOffsetEvalType ObjectSizeOffsetEvaluator::compute(Value *V) {
586   // XXX - Are vectors of pointers possible here?
587   IntTy = cast<IntegerType>(DL.getIntPtrType(V->getType()));
588   Zero = ConstantInt::get(IntTy, 0);
589 
590   SizeOffsetEvalType Result = compute_(V);
591 
592   if (!bothKnown(Result)) {
593     // erase everything that was computed in this iteration from the cache, so
594     // that no dangling references are left behind. We could be a bit smarter if
595     // we kept a dependency graph. It's probably not worth the complexity.
596     for (PtrSetTy::iterator I=SeenVals.begin(), E=SeenVals.end(); I != E; ++I) {
597       CacheMapTy::iterator CacheIt = CacheMap.find(*I);
598       // non-computable results can be safely cached
599       if (CacheIt != CacheMap.end() && anyKnown(CacheIt->second))
600         CacheMap.erase(CacheIt);
601     }
602   }
603 
604   SeenVals.clear();
605   return Result;
606 }
607 
compute_(Value * V)608 SizeOffsetEvalType ObjectSizeOffsetEvaluator::compute_(Value *V) {
609   ObjectSizeOffsetVisitor Visitor(DL, TLI, Context, RoundToAlign);
610   SizeOffsetType Const = Visitor.compute(V);
611   if (Visitor.bothKnown(Const))
612     return std::make_pair(ConstantInt::get(Context, Const.first),
613                           ConstantInt::get(Context, Const.second));
614 
615   V = V->stripPointerCasts();
616 
617   // check cache
618   CacheMapTy::iterator CacheIt = CacheMap.find(V);
619   if (CacheIt != CacheMap.end())
620     return CacheIt->second;
621 
622   // always generate code immediately before the instruction being
623   // processed, so that the generated code dominates the same BBs
624   Instruction *PrevInsertPoint = Builder.GetInsertPoint();
625   if (Instruction *I = dyn_cast<Instruction>(V))
626     Builder.SetInsertPoint(I);
627 
628   // now compute the size and offset
629   SizeOffsetEvalType Result;
630 
631   // Record the pointers that were handled in this run, so that they can be
632   // cleaned later if something fails. We also use this set to break cycles that
633   // can occur in dead code.
634   if (!SeenVals.insert(V).second) {
635     Result = unknown();
636   } else if (GEPOperator *GEP = dyn_cast<GEPOperator>(V)) {
637     Result = visitGEPOperator(*GEP);
638   } else if (Instruction *I = dyn_cast<Instruction>(V)) {
639     Result = visit(*I);
640   } else if (isa<Argument>(V) ||
641              (isa<ConstantExpr>(V) &&
642               cast<ConstantExpr>(V)->getOpcode() == Instruction::IntToPtr) ||
643              isa<GlobalAlias>(V) ||
644              isa<GlobalVariable>(V)) {
645     // ignore values where we cannot do more than what ObjectSizeVisitor can
646     Result = unknown();
647   } else {
648     DEBUG(dbgs() << "ObjectSizeOffsetEvaluator::compute() unhandled value: "
649           << *V << '\n');
650     Result = unknown();
651   }
652 
653   if (PrevInsertPoint)
654     Builder.SetInsertPoint(PrevInsertPoint);
655 
656   // Don't reuse CacheIt since it may be invalid at this point.
657   CacheMap[V] = Result;
658   return Result;
659 }
660 
visitAllocaInst(AllocaInst & I)661 SizeOffsetEvalType ObjectSizeOffsetEvaluator::visitAllocaInst(AllocaInst &I) {
662   if (!I.getAllocatedType()->isSized())
663     return unknown();
664 
665   // must be a VLA
666   assert(I.isArrayAllocation());
667   Value *ArraySize = I.getArraySize();
668   Value *Size = ConstantInt::get(ArraySize->getType(),
669                                  DL.getTypeAllocSize(I.getAllocatedType()));
670   Size = Builder.CreateMul(Size, ArraySize);
671   return std::make_pair(Size, Zero);
672 }
673 
visitCallSite(CallSite CS)674 SizeOffsetEvalType ObjectSizeOffsetEvaluator::visitCallSite(CallSite CS) {
675   const AllocFnsTy *FnData = getAllocationData(CS.getInstruction(), AnyAlloc,
676                                                TLI);
677   if (!FnData)
678     return unknown();
679 
680   // handle strdup-like functions separately
681   if (FnData->AllocTy == StrDupLike) {
682     // TODO
683     return unknown();
684   }
685 
686   Value *FirstArg = CS.getArgument(FnData->FstParam);
687   FirstArg = Builder.CreateZExt(FirstArg, IntTy);
688   if (FnData->SndParam < 0)
689     return std::make_pair(FirstArg, Zero);
690 
691   Value *SecondArg = CS.getArgument(FnData->SndParam);
692   SecondArg = Builder.CreateZExt(SecondArg, IntTy);
693   Value *Size = Builder.CreateMul(FirstArg, SecondArg);
694   return std::make_pair(Size, Zero);
695 
696   // TODO: handle more standard functions (+ wchar cousins):
697   // - strdup / strndup
698   // - strcpy / strncpy
699   // - strcat / strncat
700   // - memcpy / memmove
701   // - strcat / strncat
702   // - memset
703 }
704 
705 SizeOffsetEvalType
visitExtractElementInst(ExtractElementInst &)706 ObjectSizeOffsetEvaluator::visitExtractElementInst(ExtractElementInst&) {
707   return unknown();
708 }
709 
710 SizeOffsetEvalType
visitExtractValueInst(ExtractValueInst &)711 ObjectSizeOffsetEvaluator::visitExtractValueInst(ExtractValueInst&) {
712   return unknown();
713 }
714 
715 SizeOffsetEvalType
visitGEPOperator(GEPOperator & GEP)716 ObjectSizeOffsetEvaluator::visitGEPOperator(GEPOperator &GEP) {
717   SizeOffsetEvalType PtrData = compute_(GEP.getPointerOperand());
718   if (!bothKnown(PtrData))
719     return unknown();
720 
721   Value *Offset = EmitGEPOffset(&Builder, DL, &GEP, /*NoAssumptions=*/true);
722   Offset = Builder.CreateAdd(PtrData.second, Offset);
723   return std::make_pair(PtrData.first, Offset);
724 }
725 
visitIntToPtrInst(IntToPtrInst &)726 SizeOffsetEvalType ObjectSizeOffsetEvaluator::visitIntToPtrInst(IntToPtrInst&) {
727   // clueless
728   return unknown();
729 }
730 
visitLoadInst(LoadInst &)731 SizeOffsetEvalType ObjectSizeOffsetEvaluator::visitLoadInst(LoadInst&) {
732   return unknown();
733 }
734 
visitPHINode(PHINode & PHI)735 SizeOffsetEvalType ObjectSizeOffsetEvaluator::visitPHINode(PHINode &PHI) {
736   // create 2 PHIs: one for size and another for offset
737   PHINode *SizePHI   = Builder.CreatePHI(IntTy, PHI.getNumIncomingValues());
738   PHINode *OffsetPHI = Builder.CreatePHI(IntTy, PHI.getNumIncomingValues());
739 
740   // insert right away in the cache to handle recursive PHIs
741   CacheMap[&PHI] = std::make_pair(SizePHI, OffsetPHI);
742 
743   // compute offset/size for each PHI incoming pointer
744   for (unsigned i = 0, e = PHI.getNumIncomingValues(); i != e; ++i) {
745     Builder.SetInsertPoint(PHI.getIncomingBlock(i)->getFirstInsertionPt());
746     SizeOffsetEvalType EdgeData = compute_(PHI.getIncomingValue(i));
747 
748     if (!bothKnown(EdgeData)) {
749       OffsetPHI->replaceAllUsesWith(UndefValue::get(IntTy));
750       OffsetPHI->eraseFromParent();
751       SizePHI->replaceAllUsesWith(UndefValue::get(IntTy));
752       SizePHI->eraseFromParent();
753       return unknown();
754     }
755     SizePHI->addIncoming(EdgeData.first, PHI.getIncomingBlock(i));
756     OffsetPHI->addIncoming(EdgeData.second, PHI.getIncomingBlock(i));
757   }
758 
759   Value *Size = SizePHI, *Offset = OffsetPHI, *Tmp;
760   if ((Tmp = SizePHI->hasConstantValue())) {
761     Size = Tmp;
762     SizePHI->replaceAllUsesWith(Size);
763     SizePHI->eraseFromParent();
764   }
765   if ((Tmp = OffsetPHI->hasConstantValue())) {
766     Offset = Tmp;
767     OffsetPHI->replaceAllUsesWith(Offset);
768     OffsetPHI->eraseFromParent();
769   }
770   return std::make_pair(Size, Offset);
771 }
772 
visitSelectInst(SelectInst & I)773 SizeOffsetEvalType ObjectSizeOffsetEvaluator::visitSelectInst(SelectInst &I) {
774   SizeOffsetEvalType TrueSide  = compute_(I.getTrueValue());
775   SizeOffsetEvalType FalseSide = compute_(I.getFalseValue());
776 
777   if (!bothKnown(TrueSide) || !bothKnown(FalseSide))
778     return unknown();
779   if (TrueSide == FalseSide)
780     return TrueSide;
781 
782   Value *Size = Builder.CreateSelect(I.getCondition(), TrueSide.first,
783                                      FalseSide.first);
784   Value *Offset = Builder.CreateSelect(I.getCondition(), TrueSide.second,
785                                        FalseSide.second);
786   return std::make_pair(Size, Offset);
787 }
788 
visitInstruction(Instruction & I)789 SizeOffsetEvalType ObjectSizeOffsetEvaluator::visitInstruction(Instruction &I) {
790   DEBUG(dbgs() << "ObjectSizeOffsetEvaluator unknown instruction:" << I <<'\n');
791   return unknown();
792 }
793