1 //= CStringChecker.h - Checks calls to C string functions ----------*- C++ -*-//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This defines CStringChecker, which is an assortment of checks on calls
11 // to functions in <string.h>.
13 //===----------------------------------------------------------------------===//
15 #include "ClangSACheckers.h"
16 #include "clang/StaticAnalyzer/Core/CheckerV2.h"
17 #include "clang/StaticAnalyzer/Core/CheckerManager.h"
18 #include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h"
19 #include "clang/StaticAnalyzer/Core/BugReporter/BugType.h"
20 #include "clang/StaticAnalyzer/Core/PathSensitive/GRStateTrait.h"
21 #include "llvm/ADT/StringSwitch.h"
23 using namespace clang;
27 class CStringChecker : public CheckerV2< eval::Call,
28 check::PreStmt<DeclStmt>,
33 mutable llvm::OwningPtr<BugType> BT_Null, BT_Bounds, BT_BoundsWrite,
34 BT_Overlap, BT_NotCString;
36 static void *getTag() { static int tag; return &tag; }
38 bool evalCall(const CallExpr *CE, CheckerContext &C) const;
39 void checkPreStmt(const DeclStmt *DS, CheckerContext &C) const;
40 void checkLiveSymbols(const GRState *state, SymbolReaper &SR) const;
41 void checkDeadSymbols(SymbolReaper &SR, CheckerContext &C) const;
42 bool wantsRegionChangeUpdate(const GRState *state) const;
44 const GRState *checkRegionChanges(const GRState *state,
45 const MemRegion * const *Begin,
46 const MemRegion * const *End) const;
48 typedef void (CStringChecker::*FnCheck)(CheckerContext &,
49 const CallExpr *) const;
51 void evalMemcpy(CheckerContext &C, const CallExpr *CE) const;
52 void evalMemmove(CheckerContext &C, const CallExpr *CE) const;
53 void evalBcopy(CheckerContext &C, const CallExpr *CE) const;
54 void evalCopyCommon(CheckerContext &C, const GRState *state,
55 const Expr *Size, const Expr *Source, const Expr *Dest,
56 bool Restricted = false) const;
58 void evalMemcmp(CheckerContext &C, const CallExpr *CE) const;
60 void evalstrLength(CheckerContext &C, const CallExpr *CE) const;
61 void evalstrnLength(CheckerContext &C, const CallExpr *CE) const;
62 void evalstrLengthCommon(CheckerContext &C, const CallExpr *CE,
63 bool IsStrnlen = false) const;
65 void evalStrcpy(CheckerContext &C, const CallExpr *CE) const;
66 void evalStrncpy(CheckerContext &C, const CallExpr *CE) const;
67 void evalStpcpy(CheckerContext &C, const CallExpr *CE) const;
68 void evalStrcpyCommon(CheckerContext &C, const CallExpr *CE, bool returnEnd,
69 bool isStrncpy) const;
72 std::pair<const GRState*, const GRState*>
73 static assumeZero(CheckerContext &C,
74 const GRState *state, SVal V, QualType Ty);
76 static const GRState *setCStringLength(const GRState *state,
77 const MemRegion *MR, SVal strLength);
78 static SVal getCStringLengthForRegion(CheckerContext &C,
79 const GRState *&state,
80 const Expr *Ex, const MemRegion *MR);
81 SVal getCStringLength(CheckerContext &C, const GRState *&state,
82 const Expr *Ex, SVal Buf) const;
84 static const GRState *InvalidateBuffer(CheckerContext &C,
86 const Expr *Ex, SVal V);
88 static bool SummarizeRegion(llvm::raw_ostream& os, ASTContext& Ctx,
92 const GRState *checkNonNull(CheckerContext &C, const GRState *state,
93 const Expr *S, SVal l) const;
94 const GRState *CheckLocation(CheckerContext &C, const GRState *state,
95 const Expr *S, SVal l,
96 bool IsDestination = false) const;
97 const GRState *CheckBufferAccess(CheckerContext &C, const GRState *state,
100 const Expr *SecondBuf = NULL,
101 bool FirstIsDestination = false) const;
102 const GRState *CheckOverlap(CheckerContext &C, const GRState *state,
103 const Expr *Size, const Expr *First,
104 const Expr *Second) const;
105 void emitOverlapBug(CheckerContext &C, const GRState *state,
106 const Stmt *First, const Stmt *Second) const;
109 class CStringLength {
111 typedef llvm::ImmutableMap<const MemRegion *, SVal> EntryMap;
113 } //end anonymous namespace
118 struct GRStateTrait<CStringLength>
119 : public GRStatePartialTrait<CStringLength::EntryMap> {
120 static void *GDMIndex() { return CStringChecker::getTag(); }
125 //===----------------------------------------------------------------------===//
126 // Individual checks and utility methods.
127 //===----------------------------------------------------------------------===//
129 std::pair<const GRState*, const GRState*>
130 CStringChecker::assumeZero(CheckerContext &C, const GRState *state, SVal V,
132 DefinedSVal *val = dyn_cast<DefinedSVal>(&V);
134 return std::pair<const GRState*, const GRState *>(state, state);
136 SValBuilder &svalBuilder = C.getSValBuilder();
137 DefinedOrUnknownSVal zero = svalBuilder.makeZeroVal(Ty);
138 return state->assume(svalBuilder.evalEQ(state, *val, zero));
141 const GRState *CStringChecker::checkNonNull(CheckerContext &C,
142 const GRState *state,
143 const Expr *S, SVal l) const {
144 // If a previous check has failed, propagate the failure.
148 const GRState *stateNull, *stateNonNull;
149 llvm::tie(stateNull, stateNonNull) = assumeZero(C, state, l, S->getType());
151 if (stateNull && !stateNonNull) {
152 ExplodedNode *N = C.generateSink(stateNull);
157 BT_Null.reset(new BuiltinBug("API",
158 "Null pointer argument in call to byte string function"));
160 // Generate a report for this bug.
161 BuiltinBug *BT = static_cast<BuiltinBug*>(BT_Null.get());
162 EnhancedBugReport *report = new EnhancedBugReport(*BT,
163 BT->getDescription(), N);
165 report->addRange(S->getSourceRange());
166 report->addVisitorCreator(bugreporter::registerTrackNullOrUndefValue, S);
167 C.EmitReport(report);
171 // From here on, assume that the value is non-null.
172 assert(stateNonNull);
176 // FIXME: This was originally copied from ArrayBoundChecker.cpp. Refactor?
177 const GRState *CStringChecker::CheckLocation(CheckerContext &C,
178 const GRState *state,
179 const Expr *S, SVal l,
180 bool IsDestination) const {
181 // If a previous check has failed, propagate the failure.
185 // Check for out of bound array element access.
186 const MemRegion *R = l.getAsRegion();
190 const ElementRegion *ER = dyn_cast<ElementRegion>(R);
194 assert(ER->getValueType() == C.getASTContext().CharTy &&
195 "CheckLocation should only be called with char* ElementRegions");
197 // Get the size of the array.
198 const SubRegion *superReg = cast<SubRegion>(ER->getSuperRegion());
199 SValBuilder &svalBuilder = C.getSValBuilder();
200 SVal Extent = svalBuilder.convertToArrayIndex(superReg->getExtent(svalBuilder));
201 DefinedOrUnknownSVal Size = cast<DefinedOrUnknownSVal>(Extent);
203 // Get the index of the accessed element.
204 DefinedOrUnknownSVal Idx = cast<DefinedOrUnknownSVal>(ER->getIndex());
206 const GRState *StInBound = state->assumeInBound(Idx, Size, true);
207 const GRState *StOutBound = state->assumeInBound(Idx, Size, false);
208 if (StOutBound && !StInBound) {
209 ExplodedNode *N = C.generateSink(StOutBound);
215 if (!BT_BoundsWrite) {
216 BT_BoundsWrite.reset(new BuiltinBug("Out-of-bound array access",
217 "Byte string function overflows destination buffer"));
219 BT = static_cast<BuiltinBug*>(BT_BoundsWrite.get());
222 BT_Bounds.reset(new BuiltinBug("Out-of-bound array access",
223 "Byte string function accesses out-of-bound array element"));
225 BT = static_cast<BuiltinBug*>(BT_Bounds.get());
228 // FIXME: It would be nice to eventually make this diagnostic more clear,
229 // e.g., by referencing the original declaration or by saying *why* this
230 // reference is outside the range.
232 // Generate a report for this bug.
233 RangedBugReport *report = new RangedBugReport(*BT, BT->getDescription(), N);
235 report->addRange(S->getSourceRange());
236 C.EmitReport(report);
240 // Array bound check succeeded. From this point forward the array bound
241 // should always succeed.
245 const GRState *CStringChecker::CheckBufferAccess(CheckerContext &C,
246 const GRState *state,
248 const Expr *FirstBuf,
249 const Expr *SecondBuf,
250 bool FirstIsDestination) const {
251 // If a previous check has failed, propagate the failure.
255 SValBuilder &svalBuilder = C.getSValBuilder();
256 ASTContext &Ctx = C.getASTContext();
258 QualType sizeTy = Size->getType();
259 QualType PtrTy = Ctx.getPointerType(Ctx.CharTy);
261 // Check that the first buffer is non-null.
262 SVal BufVal = state->getSVal(FirstBuf);
263 state = checkNonNull(C, state, FirstBuf, BufVal);
267 // Get the access length and make sure it is known.
268 SVal LengthVal = state->getSVal(Size);
269 NonLoc *Length = dyn_cast<NonLoc>(&LengthVal);
273 // Compute the offset of the last element to be accessed: size-1.
274 NonLoc One = cast<NonLoc>(svalBuilder.makeIntVal(1, sizeTy));
275 NonLoc LastOffset = cast<NonLoc>(svalBuilder.evalBinOpNN(state, BO_Sub,
276 *Length, One, sizeTy));
278 // Check that the first buffer is sufficently long.
279 SVal BufStart = svalBuilder.evalCast(BufVal, PtrTy, FirstBuf->getType());
280 if (Loc *BufLoc = dyn_cast<Loc>(&BufStart)) {
281 SVal BufEnd = svalBuilder.evalBinOpLN(state, BO_Add, *BufLoc,
283 state = CheckLocation(C, state, FirstBuf, BufEnd, FirstIsDestination);
285 // If the buffer isn't large enough, abort.
290 // If there's a second buffer, check it as well.
292 BufVal = state->getSVal(SecondBuf);
293 state = checkNonNull(C, state, SecondBuf, BufVal);
297 BufStart = svalBuilder.evalCast(BufVal, PtrTy, SecondBuf->getType());
298 if (Loc *BufLoc = dyn_cast<Loc>(&BufStart)) {
299 SVal BufEnd = svalBuilder.evalBinOpLN(state, BO_Add, *BufLoc,
301 state = CheckLocation(C, state, SecondBuf, BufEnd);
305 // Large enough or not, return this state!
309 const GRState *CStringChecker::CheckOverlap(CheckerContext &C,
310 const GRState *state,
313 const Expr *Second) const {
314 // Do a simple check for overlap: if the two arguments are from the same
315 // buffer, see if the end of the first is greater than the start of the second
318 // If a previous check has failed, propagate the failure.
322 const GRState *stateTrue, *stateFalse;
324 // Get the buffer values and make sure they're known locations.
325 SVal firstVal = state->getSVal(First);
326 SVal secondVal = state->getSVal(Second);
328 Loc *firstLoc = dyn_cast<Loc>(&firstVal);
332 Loc *secondLoc = dyn_cast<Loc>(&secondVal);
336 // Are the two values the same?
337 SValBuilder &svalBuilder = C.getSValBuilder();
338 llvm::tie(stateTrue, stateFalse) =
339 state->assume(svalBuilder.evalEQ(state, *firstLoc, *secondLoc));
341 if (stateTrue && !stateFalse) {
342 // If the values are known to be equal, that's automatically an overlap.
343 emitOverlapBug(C, stateTrue, First, Second);
347 // assume the two expressions are not equal.
351 // Which value comes first?
352 ASTContext &Ctx = svalBuilder.getContext();
353 QualType cmpTy = Ctx.IntTy;
354 SVal reverse = svalBuilder.evalBinOpLL(state, BO_GT,
355 *firstLoc, *secondLoc, cmpTy);
356 DefinedOrUnknownSVal *reverseTest = dyn_cast<DefinedOrUnknownSVal>(&reverse);
360 llvm::tie(stateTrue, stateFalse) = state->assume(*reverseTest);
363 // If we don't know which one comes first, we can't perform this test.
366 // Switch the values so that firstVal is before secondVal.
367 Loc *tmpLoc = firstLoc;
368 firstLoc = secondLoc;
371 // Switch the Exprs as well, so that they still correspond.
372 const Expr *tmpExpr = First;
378 // Get the length, and make sure it too is known.
379 SVal LengthVal = state->getSVal(Size);
380 NonLoc *Length = dyn_cast<NonLoc>(&LengthVal);
384 // Convert the first buffer's start address to char*.
385 // Bail out if the cast fails.
386 QualType CharPtrTy = Ctx.getPointerType(Ctx.CharTy);
387 SVal FirstStart = svalBuilder.evalCast(*firstLoc, CharPtrTy, First->getType());
388 Loc *FirstStartLoc = dyn_cast<Loc>(&FirstStart);
392 // Compute the end of the first buffer. Bail out if THAT fails.
393 SVal FirstEnd = svalBuilder.evalBinOpLN(state, BO_Add,
394 *FirstStartLoc, *Length, CharPtrTy);
395 Loc *FirstEndLoc = dyn_cast<Loc>(&FirstEnd);
399 // Is the end of the first buffer past the start of the second buffer?
400 SVal Overlap = svalBuilder.evalBinOpLL(state, BO_GT,
401 *FirstEndLoc, *secondLoc, cmpTy);
402 DefinedOrUnknownSVal *OverlapTest = dyn_cast<DefinedOrUnknownSVal>(&Overlap);
406 llvm::tie(stateTrue, stateFalse) = state->assume(*OverlapTest);
408 if (stateTrue && !stateFalse) {
410 emitOverlapBug(C, stateTrue, First, Second);
414 // assume the two expressions don't overlap.
419 void CStringChecker::emitOverlapBug(CheckerContext &C, const GRState *state,
420 const Stmt *First, const Stmt *Second) const {
421 ExplodedNode *N = C.generateSink(state);
426 BT_Overlap.reset(new BugType("Unix API", "Improper arguments"));
428 // Generate a report for this bug.
429 RangedBugReport *report =
430 new RangedBugReport(*BT_Overlap,
431 "Arguments must not be overlapping buffers", N);
432 report->addRange(First->getSourceRange());
433 report->addRange(Second->getSourceRange());
435 C.EmitReport(report);
438 const GRState *CStringChecker::setCStringLength(const GRState *state,
441 assert(!strLength.isUndef() && "Attempt to set an undefined string length");
442 if (strLength.isUnknown())
445 MR = MR->StripCasts();
447 switch (MR->getKind()) {
448 case MemRegion::StringRegionKind:
449 // FIXME: This can happen if we strcpy() into a string region. This is
450 // undefined [C99 6.4.5p6], but we should still warn about it.
453 case MemRegion::SymbolicRegionKind:
454 case MemRegion::AllocaRegionKind:
455 case MemRegion::VarRegionKind:
456 case MemRegion::FieldRegionKind:
457 case MemRegion::ObjCIvarRegionKind:
458 return state->set<CStringLength>(MR, strLength);
460 case MemRegion::ElementRegionKind:
461 // FIXME: Handle element regions by upper-bounding the parent region's
466 // Other regions (mostly non-data) can't have a reliable C string length.
467 // For now, just ignore the change.
468 // FIXME: These are rare but not impossible. We should output some kind of
469 // warning for things like strcpy((char[]){'a', 0}, "b");
474 SVal CStringChecker::getCStringLengthForRegion(CheckerContext &C,
475 const GRState *&state,
477 const MemRegion *MR) {
478 // If there's a recorded length, go ahead and return it.
479 const SVal *Recorded = state->get<CStringLength>(MR);
483 // Otherwise, get a new symbol and update the state.
484 unsigned Count = C.getNodeBuilder().getCurrentBlockCount();
485 SValBuilder &svalBuilder = C.getSValBuilder();
486 QualType sizeTy = svalBuilder.getContext().getSizeType();
487 SVal strLength = svalBuilder.getMetadataSymbolVal(CStringChecker::getTag(),
488 MR, Ex, sizeTy, Count);
489 state = state->set<CStringLength>(MR, strLength);
493 SVal CStringChecker::getCStringLength(CheckerContext &C, const GRState *&state,
494 const Expr *Ex, SVal Buf) const {
495 const MemRegion *MR = Buf.getAsRegion();
497 // If we can't get a region, see if it's something we /know/ isn't a
498 // C string. In the context of locations, the only time we can issue such
499 // a warning is for labels.
500 if (loc::GotoLabel *Label = dyn_cast<loc::GotoLabel>(&Buf)) {
501 if (ExplodedNode *N = C.generateNode(state)) {
503 BT_NotCString.reset(new BuiltinBug("API",
504 "Argument is not a null-terminated string."));
506 llvm::SmallString<120> buf;
507 llvm::raw_svector_ostream os(buf);
508 os << "Argument to byte string function is the address of the label '"
509 << Label->getLabel()->getName()
510 << "', which is not a null-terminated string";
512 // Generate a report for this bug.
513 EnhancedBugReport *report = new EnhancedBugReport(*BT_NotCString,
516 report->addRange(Ex->getSourceRange());
517 C.EmitReport(report);
520 return UndefinedVal();
523 // If it's not a region and not a label, give up.
527 // If we have a region, strip casts from it and see if we can figure out
528 // its length. For anything we can't figure out, just return UnknownVal.
529 MR = MR->StripCasts();
531 switch (MR->getKind()) {
532 case MemRegion::StringRegionKind: {
533 // Modifying the contents of string regions is undefined [C99 6.4.5p6],
534 // so we can assume that the byte length is the correct C string length.
535 SValBuilder &svalBuilder = C.getSValBuilder();
536 QualType sizeTy = svalBuilder.getContext().getSizeType();
537 const StringLiteral *strLit = cast<StringRegion>(MR)->getStringLiteral();
538 return svalBuilder.makeIntVal(strLit->getByteLength(), sizeTy);
540 case MemRegion::SymbolicRegionKind:
541 case MemRegion::AllocaRegionKind:
542 case MemRegion::VarRegionKind:
543 case MemRegion::FieldRegionKind:
544 case MemRegion::ObjCIvarRegionKind:
545 return getCStringLengthForRegion(C, state, Ex, MR);
546 case MemRegion::CompoundLiteralRegionKind:
547 // FIXME: Can we track this? Is it necessary?
549 case MemRegion::ElementRegionKind:
550 // FIXME: How can we handle this? It's not good enough to subtract the
551 // offset from the base string length; consider "123\x00567" and &a[5].
554 // Other regions (mostly non-data) can't have a reliable C string length.
555 // In this case, an error is emitted and UndefinedVal is returned.
556 // The caller should always be prepared to handle this case.
557 if (ExplodedNode *N = C.generateNode(state)) {
559 BT_NotCString.reset(new BuiltinBug("API",
560 "Argument is not a null-terminated string."));
562 llvm::SmallString<120> buf;
563 llvm::raw_svector_ostream os(buf);
565 os << "Argument to byte string function is ";
567 if (SummarizeRegion(os, C.getASTContext(), MR))
568 os << ", which is not a null-terminated string";
570 os << "not a null-terminated string";
572 // Generate a report for this bug.
573 EnhancedBugReport *report = new EnhancedBugReport(*BT_NotCString,
576 report->addRange(Ex->getSourceRange());
577 C.EmitReport(report);
580 return UndefinedVal();
584 const GRState *CStringChecker::InvalidateBuffer(CheckerContext &C,
585 const GRState *state,
586 const Expr *E, SVal V) {
587 Loc *L = dyn_cast<Loc>(&V);
591 // FIXME: This is a simplified version of what's in CFRefCount.cpp -- it makes
592 // some assumptions about the value that CFRefCount can't. Even so, it should
593 // probably be refactored.
594 if (loc::MemRegionVal* MR = dyn_cast<loc::MemRegionVal>(L)) {
595 const MemRegion *R = MR->getRegion()->StripCasts();
597 // Are we dealing with an ElementRegion? If so, we should be invalidating
599 if (const ElementRegion *ER = dyn_cast<ElementRegion>(R)) {
600 R = ER->getSuperRegion();
601 // FIXME: What about layers of ElementRegions?
604 // Invalidate this region.
605 unsigned Count = C.getNodeBuilder().getCurrentBlockCount();
606 return state->invalidateRegion(R, E, Count, NULL);
609 // If we have a non-region value by chance, just remove the binding.
610 // FIXME: is this necessary or correct? This handles the non-Region
611 // cases. Is it ever valid to store to these?
612 return state->unbindLoc(*L);
615 bool CStringChecker::SummarizeRegion(llvm::raw_ostream& os, ASTContext& Ctx,
616 const MemRegion *MR) {
617 const TypedRegion *TR = dyn_cast<TypedRegion>(MR);
621 switch (TR->getKind()) {
622 case MemRegion::FunctionTextRegionKind: {
623 const FunctionDecl *FD = cast<FunctionTextRegion>(TR)->getDecl();
625 os << "the address of the function '" << FD << "'";
627 os << "the address of a function";
630 case MemRegion::BlockTextRegionKind:
633 case MemRegion::BlockDataRegionKind:
636 case MemRegion::CXXThisRegionKind:
637 case MemRegion::CXXTempObjectRegionKind:
638 os << "a C++ temp object of type " << TR->getValueType().getAsString();
640 case MemRegion::VarRegionKind:
641 os << "a variable of type" << TR->getValueType().getAsString();
643 case MemRegion::FieldRegionKind:
644 os << "a field of type " << TR->getValueType().getAsString();
646 case MemRegion::ObjCIvarRegionKind:
647 os << "an instance variable of type " << TR->getValueType().getAsString();
654 //===----------------------------------------------------------------------===//
655 // evaluation of individual function calls.
656 //===----------------------------------------------------------------------===//
658 void CStringChecker::evalCopyCommon(CheckerContext &C, const GRState *state,
659 const Expr *Size, const Expr *Dest,
660 const Expr *Source, bool Restricted) const {
661 // See if the size argument is zero.
662 SVal sizeVal = state->getSVal(Size);
663 QualType sizeTy = Size->getType();
665 const GRState *stateZeroSize, *stateNonZeroSize;
666 llvm::tie(stateZeroSize, stateNonZeroSize) = assumeZero(C, state, sizeVal, sizeTy);
668 // If the size is zero, there won't be any actual memory access.
670 C.addTransition(stateZeroSize);
672 // If the size can be nonzero, we have to check the other arguments.
673 if (stateNonZeroSize) {
674 state = stateNonZeroSize;
675 state = CheckBufferAccess(C, state, Size, Dest, Source,
676 /* FirstIsDst = */ true);
678 state = CheckOverlap(C, state, Size, Dest, Source);
681 // Invalidate the destination.
682 // FIXME: Even if we can't perfectly model the copy, we should see if we
683 // can use LazyCompoundVals to copy the source values into the destination.
684 // This would probably remove any existing bindings past the end of the
685 // copied region, but that's still an improvement over blank invalidation.
686 state = InvalidateBuffer(C, state, Dest, state->getSVal(Dest));
687 C.addTransition(state);
693 void CStringChecker::evalMemcpy(CheckerContext &C, const CallExpr *CE) const {
694 // void *memcpy(void *restrict dst, const void *restrict src, size_t n);
695 // The return value is the address of the destination buffer.
696 const Expr *Dest = CE->getArg(0);
697 const GRState *state = C.getState();
698 state = state->BindExpr(CE, state->getSVal(Dest));
699 evalCopyCommon(C, state, CE->getArg(2), Dest, CE->getArg(1), true);
702 void CStringChecker::evalMemmove(CheckerContext &C, const CallExpr *CE) const {
703 // void *memmove(void *dst, const void *src, size_t n);
704 // The return value is the address of the destination buffer.
705 const Expr *Dest = CE->getArg(0);
706 const GRState *state = C.getState();
707 state = state->BindExpr(CE, state->getSVal(Dest));
708 evalCopyCommon(C, state, CE->getArg(2), Dest, CE->getArg(1));
711 void CStringChecker::evalBcopy(CheckerContext &C, const CallExpr *CE) const {
712 // void bcopy(const void *src, void *dst, size_t n);
713 evalCopyCommon(C, C.getState(), CE->getArg(2), CE->getArg(1), CE->getArg(0));
716 void CStringChecker::evalMemcmp(CheckerContext &C, const CallExpr *CE) const {
717 // int memcmp(const void *s1, const void *s2, size_t n);
718 const Expr *Left = CE->getArg(0);
719 const Expr *Right = CE->getArg(1);
720 const Expr *Size = CE->getArg(2);
722 const GRState *state = C.getState();
723 SValBuilder &svalBuilder = C.getSValBuilder();
725 // See if the size argument is zero.
726 SVal sizeVal = state->getSVal(Size);
727 QualType sizeTy = Size->getType();
729 const GRState *stateZeroSize, *stateNonZeroSize;
730 llvm::tie(stateZeroSize, stateNonZeroSize) =
731 assumeZero(C, state, sizeVal, sizeTy);
733 // If the size can be zero, the result will be 0 in that case, and we don't
734 // have to check either of the buffers.
736 state = stateZeroSize;
737 state = state->BindExpr(CE, svalBuilder.makeZeroVal(CE->getType()));
738 C.addTransition(state);
741 // If the size can be nonzero, we have to check the other arguments.
742 if (stateNonZeroSize) {
743 state = stateNonZeroSize;
744 // If we know the two buffers are the same, we know the result is 0.
745 // First, get the two buffers' addresses. Another checker will have already
746 // made sure they're not undefined.
747 DefinedOrUnknownSVal LV = cast<DefinedOrUnknownSVal>(state->getSVal(Left));
748 DefinedOrUnknownSVal RV = cast<DefinedOrUnknownSVal>(state->getSVal(Right));
750 // See if they are the same.
751 DefinedOrUnknownSVal SameBuf = svalBuilder.evalEQ(state, LV, RV);
752 const GRState *StSameBuf, *StNotSameBuf;
753 llvm::tie(StSameBuf, StNotSameBuf) = state->assume(SameBuf);
755 // If the two arguments might be the same buffer, we know the result is zero,
756 // and we only need to check one size.
759 state = CheckBufferAccess(C, state, Size, Left);
761 state = StSameBuf->BindExpr(CE, svalBuilder.makeZeroVal(CE->getType()));
762 C.addTransition(state);
766 // If the two arguments might be different buffers, we have to check the
767 // size of both of them.
769 state = StNotSameBuf;
770 state = CheckBufferAccess(C, state, Size, Left, Right);
772 // The return value is the comparison result, which we don't know.
773 unsigned Count = C.getNodeBuilder().getCurrentBlockCount();
774 SVal CmpV = svalBuilder.getConjuredSymbolVal(NULL, CE, Count);
775 state = state->BindExpr(CE, CmpV);
776 C.addTransition(state);
782 void CStringChecker::evalstrLength(CheckerContext &C,
783 const CallExpr *CE) const {
784 // size_t strlen(const char *s);
785 evalstrLengthCommon(C, CE, /* IsStrnlen = */ false);
788 void CStringChecker::evalstrnLength(CheckerContext &C,
789 const CallExpr *CE) const {
790 // size_t strnlen(const char *s, size_t maxlen);
791 evalstrLengthCommon(C, CE, /* IsStrnlen = */ true);
794 void CStringChecker::evalstrLengthCommon(CheckerContext &C, const CallExpr *CE,
795 bool IsStrnlen) const {
796 const GRState *state = C.getState();
797 const Expr *Arg = CE->getArg(0);
798 SVal ArgVal = state->getSVal(Arg);
800 // Check that the argument is non-null.
801 state = checkNonNull(C, state, Arg, ArgVal);
804 SVal strLength = getCStringLength(C, state, Arg, ArgVal);
806 // If the argument isn't a valid C string, there's no valid state to
808 if (strLength.isUndef())
811 // If the check is for strnlen() then bind the return value to no more than
814 const Expr *maxlenExpr = CE->getArg(1);
815 SVal maxlenVal = state->getSVal(maxlenExpr);
817 NonLoc *strLengthNL = dyn_cast<NonLoc>(&strLength);
818 NonLoc *maxlenValNL = dyn_cast<NonLoc>(&maxlenVal);
820 QualType cmpTy = C.getSValBuilder().getContext().IntTy;
821 const GRState *stateTrue, *stateFalse;
823 // Check if the strLength is greater than or equal to the maxlen
824 llvm::tie(stateTrue, stateFalse) =
825 state->assume(cast<DefinedOrUnknownSVal>
826 (C.getSValBuilder().evalBinOpNN(state, BO_GE,
827 *strLengthNL, *maxlenValNL,
830 // If the strLength is greater than or equal to the maxlen, set strLength
832 if (stateTrue && !stateFalse) {
833 strLength = maxlenVal;
837 // If getCStringLength couldn't figure out the length, conjure a return
838 // value, so it can be used in constraints, at least.
839 if (strLength.isUnknown()) {
840 unsigned Count = C.getNodeBuilder().getCurrentBlockCount();
841 strLength = C.getSValBuilder().getConjuredSymbolVal(NULL, CE, Count);
844 // Bind the return value.
845 state = state->BindExpr(CE, strLength);
846 C.addTransition(state);
850 void CStringChecker::evalStrcpy(CheckerContext &C, const CallExpr *CE) const {
851 // char *strcpy(char *restrict dst, const char *restrict src);
852 evalStrcpyCommon(C, CE, /* returnEnd = */ false, /* isStrncpy = */ false);
855 void CStringChecker::evalStrncpy(CheckerContext &C, const CallExpr *CE) const {
856 // char *strcpy(char *restrict dst, const char *restrict src);
857 evalStrcpyCommon(C, CE, /* returnEnd = */ false, /* isStrncpy = */ true);
860 void CStringChecker::evalStpcpy(CheckerContext &C, const CallExpr *CE) const {
861 // char *stpcpy(char *restrict dst, const char *restrict src);
862 evalStrcpyCommon(C, CE, /* returnEnd = */ true, /* isStrncpy = */ false);
865 void CStringChecker::evalStrcpyCommon(CheckerContext &C, const CallExpr *CE,
866 bool returnEnd, bool isStrncpy) const {
867 const GRState *state = C.getState();
869 // Check that the destination is non-null
870 const Expr *Dst = CE->getArg(0);
871 SVal DstVal = state->getSVal(Dst);
873 state = checkNonNull(C, state, Dst, DstVal);
877 // Check that the source is non-null.
878 const Expr *srcExpr = CE->getArg(1);
879 SVal srcVal = state->getSVal(srcExpr);
880 state = checkNonNull(C, state, srcExpr, srcVal);
884 // Get the string length of the source.
885 SVal strLength = getCStringLength(C, state, srcExpr, srcVal);
887 // If the source isn't a valid C string, give up.
888 if (strLength.isUndef())
892 // Get the max number of characters to copy
893 const Expr *lenExpr = CE->getArg(2);
894 SVal lenVal = state->getSVal(lenExpr);
896 NonLoc *strLengthNL = dyn_cast<NonLoc>(&strLength);
897 NonLoc *lenValNL = dyn_cast<NonLoc>(&lenVal);
899 QualType cmpTy = C.getSValBuilder().getContext().IntTy;
900 const GRState *stateTrue, *stateFalse;
902 // Check if the max number to copy is less than the length of the src
903 llvm::tie(stateTrue, stateFalse) =
904 state->assume(cast<DefinedOrUnknownSVal>
905 (C.getSValBuilder().evalBinOpNN(state, BO_GT,
906 *strLengthNL, *lenValNL,
910 // Max number to copy is less than the length of the src, so the actual
911 // strLength copied is the max number arg.
916 SVal Result = (returnEnd ? UnknownVal() : DstVal);
918 // If the destination is a MemRegion, try to check for a buffer overflow and
919 // record the new string length.
920 if (loc::MemRegionVal *dstRegVal = dyn_cast<loc::MemRegionVal>(&DstVal)) {
921 // If the length is known, we can check for an overflow.
922 if (NonLoc *knownStrLength = dyn_cast<NonLoc>(&strLength)) {
924 C.getSValBuilder().evalBinOpLN(state, BO_Add, *dstRegVal,
925 *knownStrLength, Dst->getType());
927 state = CheckLocation(C, state, Dst, lastElement, /* IsDst = */ true);
931 // If this is a stpcpy-style copy, the last element is the return value.
933 Result = lastElement;
936 // Invalidate the destination. This must happen before we set the C string
937 // length because invalidation will clear the length.
938 // FIXME: Even if we can't perfectly model the copy, we should see if we
939 // can use LazyCompoundVals to copy the source values into the destination.
940 // This would probably remove any existing bindings past the end of the
941 // string, but that's still an improvement over blank invalidation.
942 state = InvalidateBuffer(C, state, Dst, *dstRegVal);
944 // Set the C string length of the destination.
945 state = setCStringLength(state, dstRegVal->getRegion(), strLength);
948 // If this is a stpcpy-style copy, but we were unable to check for a buffer
949 // overflow, we still need a result. Conjure a return value.
950 if (returnEnd && Result.isUnknown()) {
951 SValBuilder &svalBuilder = C.getSValBuilder();
952 unsigned Count = C.getNodeBuilder().getCurrentBlockCount();
953 strLength = svalBuilder.getConjuredSymbolVal(NULL, CE, Count);
956 // Set the return value.
957 state = state->BindExpr(CE, Result);
958 C.addTransition(state);
961 //===----------------------------------------------------------------------===//
962 // The driver method, and other Checker callbacks.
963 //===----------------------------------------------------------------------===//
965 bool CStringChecker::evalCall(const CallExpr *CE, CheckerContext &C) const {
966 // Get the callee. All the functions we care about are C functions
967 // with simple identifiers.
968 const GRState *state = C.getState();
969 const Expr *Callee = CE->getCallee();
970 const FunctionDecl *FD = state->getSVal(Callee).getAsFunctionDecl();
975 // Get the name of the callee. If it's a builtin, strip off the prefix.
976 IdentifierInfo *II = FD->getIdentifier();
977 if (!II) // if no identifier, not a simple C function
979 llvm::StringRef Name = II->getName();
980 if (Name.startswith("__builtin_"))
981 Name = Name.substr(10);
983 FnCheck evalFunction = llvm::StringSwitch<FnCheck>(Name)
984 .Cases("memcpy", "__memcpy_chk", &CStringChecker::evalMemcpy)
985 .Cases("memcmp", "bcmp", &CStringChecker::evalMemcmp)
986 .Cases("memmove", "__memmove_chk", &CStringChecker::evalMemmove)
987 .Cases("strcpy", "__strcpy_chk", &CStringChecker::evalStrcpy)
988 .Cases("strncpy", "__strncpy_chk", &CStringChecker::evalStrncpy)
989 .Cases("stpcpy", "__stpcpy_chk", &CStringChecker::evalStpcpy)
990 .Case("strlen", &CStringChecker::evalstrLength)
991 .Case("strnlen", &CStringChecker::evalstrnLength)
992 .Case("bcopy", &CStringChecker::evalBcopy)
995 // If the callee isn't a string function, let another checker handle it.
999 // Check and evaluate the call.
1000 (this->*evalFunction)(C, CE);
1004 void CStringChecker::checkPreStmt(const DeclStmt *DS, CheckerContext &C) const {
1005 // Record string length for char a[] = "abc";
1006 const GRState *state = C.getState();
1008 for (DeclStmt::const_decl_iterator I = DS->decl_begin(), E = DS->decl_end();
1010 const VarDecl *D = dyn_cast<VarDecl>(*I);
1014 // FIXME: Handle array fields of structs.
1015 if (!D->getType()->isArrayType())
1018 const Expr *Init = D->getInit();
1021 if (!isa<StringLiteral>(Init))
1024 Loc VarLoc = state->getLValue(D, C.getPredecessor()->getLocationContext());
1025 const MemRegion *MR = VarLoc.getAsRegion();
1029 SVal StrVal = state->getSVal(Init);
1030 assert(StrVal.isValid() && "Initializer string is unknown or undefined");
1031 DefinedOrUnknownSVal strLength
1032 = cast<DefinedOrUnknownSVal>(getCStringLength(C, state, Init, StrVal));
1034 state = state->set<CStringLength>(MR, strLength);
1037 C.addTransition(state);
1040 bool CStringChecker::wantsRegionChangeUpdate(const GRState *state) const {
1041 CStringLength::EntryMap Entries = state->get<CStringLength>();
1042 return !Entries.isEmpty();
1046 CStringChecker::checkRegionChanges(const GRState *state,
1047 const MemRegion * const *Begin,
1048 const MemRegion * const *End) const {
1049 CStringLength::EntryMap Entries = state->get<CStringLength>();
1050 if (Entries.isEmpty())
1053 llvm::SmallPtrSet<const MemRegion *, 8> Invalidated;
1054 llvm::SmallPtrSet<const MemRegion *, 32> SuperRegions;
1056 // First build sets for the changed regions and their super-regions.
1057 for ( ; Begin != End; ++Begin) {
1058 const MemRegion *MR = *Begin;
1059 Invalidated.insert(MR);
1061 SuperRegions.insert(MR);
1062 while (const SubRegion *SR = dyn_cast<SubRegion>(MR)) {
1063 MR = SR->getSuperRegion();
1064 SuperRegions.insert(MR);
1068 CStringLength::EntryMap::Factory &F = state->get_context<CStringLength>();
1070 // Then loop over the entries in the current state.
1071 for (CStringLength::EntryMap::iterator I = Entries.begin(),
1072 E = Entries.end(); I != E; ++I) {
1073 const MemRegion *MR = I.getKey();
1075 // Is this entry for a super-region of a changed region?
1076 if (SuperRegions.count(MR)) {
1077 Entries = F.remove(Entries, MR);
1081 // Is this entry for a sub-region of a changed region?
1082 const MemRegion *Super = MR;
1083 while (const SubRegion *SR = dyn_cast<SubRegion>(Super)) {
1084 Super = SR->getSuperRegion();
1085 if (Invalidated.count(Super)) {
1086 Entries = F.remove(Entries, MR);
1092 return state->set<CStringLength>(Entries);
1095 void CStringChecker::checkLiveSymbols(const GRState *state,
1096 SymbolReaper &SR) const {
1097 // Mark all symbols in our string length map as valid.
1098 CStringLength::EntryMap Entries = state->get<CStringLength>();
1100 for (CStringLength::EntryMap::iterator I = Entries.begin(), E = Entries.end();
1102 SVal Len = I.getData();
1103 if (SymbolRef Sym = Len.getAsSymbol())
1108 void CStringChecker::checkDeadSymbols(SymbolReaper &SR,
1109 CheckerContext &C) const {
1110 if (!SR.hasDeadSymbols())
1113 const GRState *state = C.getState();
1114 CStringLength::EntryMap Entries = state->get<CStringLength>();
1115 if (Entries.isEmpty())
1118 CStringLength::EntryMap::Factory &F = state->get_context<CStringLength>();
1119 for (CStringLength::EntryMap::iterator I = Entries.begin(), E = Entries.end();
1121 SVal Len = I.getData();
1122 if (SymbolRef Sym = Len.getAsSymbol()) {
1124 Entries = F.remove(Entries, I.getKey());
1128 state = state->set<CStringLength>(Entries);
1129 C.generateNode(state);
1132 void ento::registerCStringChecker(CheckerManager &mgr) {
1133 mgr.registerChecker<CStringChecker>();