1 //= CStringChecker.cpp - 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 "InterCheckerAPI.h"
17 #include "clang/StaticAnalyzer/Core/Checker.h"
18 #include "clang/StaticAnalyzer/Core/CheckerManager.h"
19 #include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h"
20 #include "clang/StaticAnalyzer/Core/BugReporter/BugType.h"
21 #include "clang/StaticAnalyzer/Core/PathSensitive/ProgramStateTrait.h"
22 #include "llvm/ADT/SmallString.h"
23 #include "llvm/ADT/STLExtras.h"
24 #include "llvm/ADT/StringSwitch.h"
26 using namespace clang;
30 class CStringChecker : public Checker< eval::Call,
31 check::PreStmt<DeclStmt>,
36 mutable OwningPtr<BugType> BT_Null,
42 mutable const char *CurrentFunctionDescription;
45 /// The filter is used to filter out the diagnostics which are not enabled by
47 struct CStringChecksFilter {
48 DefaultBool CheckCStringNullArg;
49 DefaultBool CheckCStringOutOfBounds;
50 DefaultBool CheckCStringBufferOverlap;
51 DefaultBool CheckCStringNotNullTerm;
54 CStringChecksFilter Filter;
56 static void *getTag() { static int tag; return &tag; }
58 bool evalCall(const CallExpr *CE, CheckerContext &C) const;
59 void checkPreStmt(const DeclStmt *DS, CheckerContext &C) const;
60 void checkLiveSymbols(ProgramStateRef state, SymbolReaper &SR) const;
61 void checkDeadSymbols(SymbolReaper &SR, CheckerContext &C) const;
62 bool wantsRegionChangeUpdate(ProgramStateRef state) const;
65 checkRegionChanges(ProgramStateRef state,
66 const StoreManager::InvalidatedSymbols *,
67 ArrayRef<const MemRegion *> ExplicitRegions,
68 ArrayRef<const MemRegion *> Regions,
69 const CallEvent *Call) const;
71 typedef void (CStringChecker::*FnCheck)(CheckerContext &,
72 const CallExpr *) const;
74 void evalMemcpy(CheckerContext &C, const CallExpr *CE) const;
75 void evalMempcpy(CheckerContext &C, const CallExpr *CE) const;
76 void evalMemmove(CheckerContext &C, const CallExpr *CE) const;
77 void evalBcopy(CheckerContext &C, const CallExpr *CE) const;
78 void evalCopyCommon(CheckerContext &C, const CallExpr *CE,
79 ProgramStateRef state,
83 bool Restricted = false,
84 bool IsMempcpy = false) const;
86 void evalMemcmp(CheckerContext &C, const CallExpr *CE) const;
88 void evalstrLength(CheckerContext &C, const CallExpr *CE) const;
89 void evalstrnLength(CheckerContext &C, const CallExpr *CE) const;
90 void evalstrLengthCommon(CheckerContext &C,
92 bool IsStrnlen = false) const;
94 void evalStrcpy(CheckerContext &C, const CallExpr *CE) const;
95 void evalStrncpy(CheckerContext &C, const CallExpr *CE) const;
96 void evalStpcpy(CheckerContext &C, const CallExpr *CE) const;
97 void evalStrcpyCommon(CheckerContext &C,
101 bool isAppending) const;
103 void evalStrcat(CheckerContext &C, const CallExpr *CE) const;
104 void evalStrncat(CheckerContext &C, const CallExpr *CE) const;
106 void evalStrcmp(CheckerContext &C, const CallExpr *CE) const;
107 void evalStrncmp(CheckerContext &C, const CallExpr *CE) const;
108 void evalStrcasecmp(CheckerContext &C, const CallExpr *CE) const;
109 void evalStrncasecmp(CheckerContext &C, const CallExpr *CE) const;
110 void evalStrcmpCommon(CheckerContext &C,
112 bool isBounded = false,
113 bool ignoreCase = false) const;
116 std::pair<ProgramStateRef , ProgramStateRef >
117 static assumeZero(CheckerContext &C,
118 ProgramStateRef state, SVal V, QualType Ty);
120 static ProgramStateRef setCStringLength(ProgramStateRef state,
123 static SVal getCStringLengthForRegion(CheckerContext &C,
124 ProgramStateRef &state,
128 SVal getCStringLength(CheckerContext &C,
129 ProgramStateRef &state,
132 bool hypothetical = false) const;
134 const StringLiteral *getCStringLiteral(CheckerContext &C,
135 ProgramStateRef &state,
139 static ProgramStateRef InvalidateBuffer(CheckerContext &C,
140 ProgramStateRef state,
141 const Expr *Ex, SVal V);
143 static bool SummarizeRegion(raw_ostream &os, ASTContext &Ctx,
144 const MemRegion *MR);
147 ProgramStateRef checkNonNull(CheckerContext &C,
148 ProgramStateRef state,
151 ProgramStateRef CheckLocation(CheckerContext &C,
152 ProgramStateRef state,
155 const char *message = NULL) const;
156 ProgramStateRef CheckBufferAccess(CheckerContext &C,
157 ProgramStateRef state,
159 const Expr *FirstBuf,
160 const Expr *SecondBuf,
161 const char *firstMessage = NULL,
162 const char *secondMessage = NULL,
163 bool WarnAboutSize = false) const;
165 ProgramStateRef CheckBufferAccess(CheckerContext &C,
166 ProgramStateRef state,
169 const char *message = NULL,
170 bool WarnAboutSize = false) const {
171 // This is a convenience override.
172 return CheckBufferAccess(C, state, Size, Buf, NULL, message, NULL,
175 ProgramStateRef CheckOverlap(CheckerContext &C,
176 ProgramStateRef state,
179 const Expr *Second) const;
180 void emitOverlapBug(CheckerContext &C,
181 ProgramStateRef state,
183 const Stmt *Second) const;
185 ProgramStateRef checkAdditionOverflow(CheckerContext &C,
186 ProgramStateRef state,
191 class CStringLength {
193 typedef llvm::ImmutableMap<const MemRegion *, SVal> EntryMap;
195 } //end anonymous namespace
200 struct ProgramStateTrait<CStringLength>
201 : public ProgramStatePartialTrait<CStringLength::EntryMap> {
202 static void *GDMIndex() { return CStringChecker::getTag(); }
207 //===----------------------------------------------------------------------===//
208 // Individual checks and utility methods.
209 //===----------------------------------------------------------------------===//
211 std::pair<ProgramStateRef , ProgramStateRef >
212 CStringChecker::assumeZero(CheckerContext &C, ProgramStateRef state, SVal V,
214 DefinedSVal *val = dyn_cast<DefinedSVal>(&V);
216 return std::pair<ProgramStateRef , ProgramStateRef >(state, state);
218 SValBuilder &svalBuilder = C.getSValBuilder();
219 DefinedOrUnknownSVal zero = svalBuilder.makeZeroVal(Ty);
220 return state->assume(svalBuilder.evalEQ(state, *val, zero));
223 ProgramStateRef CStringChecker::checkNonNull(CheckerContext &C,
224 ProgramStateRef state,
225 const Expr *S, SVal l) const {
226 // If a previous check has failed, propagate the failure.
230 ProgramStateRef stateNull, stateNonNull;
231 llvm::tie(stateNull, stateNonNull) = assumeZero(C, state, l, S->getType());
233 if (stateNull && !stateNonNull) {
234 if (!Filter.CheckCStringNullArg)
237 ExplodedNode *N = C.generateSink(stateNull);
242 BT_Null.reset(new BuiltinBug("Unix API",
243 "Null pointer argument in call to byte string function"));
246 llvm::raw_svector_ostream os(buf);
247 assert(CurrentFunctionDescription);
248 os << "Null pointer argument in call to " << CurrentFunctionDescription;
250 // Generate a report for this bug.
251 BuiltinBug *BT = static_cast<BuiltinBug*>(BT_Null.get());
252 BugReport *report = new BugReport(*BT, os.str(), N);
254 report->addRange(S->getSourceRange());
255 bugreporter::addTrackNullOrUndefValueVisitor(N, S, report);
256 C.EmitReport(report);
260 // From here on, assume that the value is non-null.
261 assert(stateNonNull);
265 // FIXME: This was originally copied from ArrayBoundChecker.cpp. Refactor?
266 ProgramStateRef CStringChecker::CheckLocation(CheckerContext &C,
267 ProgramStateRef state,
268 const Expr *S, SVal l,
269 const char *warningMsg) const {
270 // If a previous check has failed, propagate the failure.
274 // Check for out of bound array element access.
275 const MemRegion *R = l.getAsRegion();
279 const ElementRegion *ER = dyn_cast<ElementRegion>(R);
283 assert(ER->getValueType() == C.getASTContext().CharTy &&
284 "CheckLocation should only be called with char* ElementRegions");
286 // Get the size of the array.
287 const SubRegion *superReg = cast<SubRegion>(ER->getSuperRegion());
288 SValBuilder &svalBuilder = C.getSValBuilder();
290 svalBuilder.convertToArrayIndex(superReg->getExtent(svalBuilder));
291 DefinedOrUnknownSVal Size = cast<DefinedOrUnknownSVal>(Extent);
293 // Get the index of the accessed element.
294 DefinedOrUnknownSVal Idx = cast<DefinedOrUnknownSVal>(ER->getIndex());
296 ProgramStateRef StInBound = state->assumeInBound(Idx, Size, true);
297 ProgramStateRef StOutBound = state->assumeInBound(Idx, Size, false);
298 if (StOutBound && !StInBound) {
299 ExplodedNode *N = C.generateSink(StOutBound);
304 BT_Bounds.reset(new BuiltinBug("Out-of-bound array access",
305 "Byte string function accesses out-of-bound array element"));
307 BuiltinBug *BT = static_cast<BuiltinBug*>(BT_Bounds.get());
309 // Generate a report for this bug.
312 report = new BugReport(*BT, warningMsg, N);
314 assert(CurrentFunctionDescription);
315 assert(CurrentFunctionDescription[0] != '\0');
318 llvm::raw_svector_ostream os(buf);
319 os << (char)toupper(CurrentFunctionDescription[0])
320 << &CurrentFunctionDescription[1]
321 << " accesses out-of-bound array element";
322 report = new BugReport(*BT, os.str(), N);
325 // FIXME: It would be nice to eventually make this diagnostic more clear,
326 // e.g., by referencing the original declaration or by saying *why* this
327 // reference is outside the range.
329 report->addRange(S->getSourceRange());
330 C.EmitReport(report);
334 // Array bound check succeeded. From this point forward the array bound
335 // should always succeed.
339 ProgramStateRef CStringChecker::CheckBufferAccess(CheckerContext &C,
340 ProgramStateRef state,
342 const Expr *FirstBuf,
343 const Expr *SecondBuf,
344 const char *firstMessage,
345 const char *secondMessage,
346 bool WarnAboutSize) const {
347 // If a previous check has failed, propagate the failure.
351 SValBuilder &svalBuilder = C.getSValBuilder();
352 ASTContext &Ctx = svalBuilder.getContext();
353 const LocationContext *LCtx = C.getLocationContext();
355 QualType sizeTy = Size->getType();
356 QualType PtrTy = Ctx.getPointerType(Ctx.CharTy);
358 // Check that the first buffer is non-null.
359 SVal BufVal = state->getSVal(FirstBuf, LCtx);
360 state = checkNonNull(C, state, FirstBuf, BufVal);
364 // If out-of-bounds checking is turned off, skip the rest.
365 if (!Filter.CheckCStringOutOfBounds)
368 // Get the access length and make sure it is known.
369 // FIXME: This assumes the caller has already checked that the access length
370 // is positive. And that it's unsigned.
371 SVal LengthVal = state->getSVal(Size, LCtx);
372 NonLoc *Length = dyn_cast<NonLoc>(&LengthVal);
376 // Compute the offset of the last element to be accessed: size-1.
377 NonLoc One = cast<NonLoc>(svalBuilder.makeIntVal(1, sizeTy));
378 NonLoc LastOffset = cast<NonLoc>(svalBuilder.evalBinOpNN(state, BO_Sub,
379 *Length, One, sizeTy));
381 // Check that the first buffer is sufficiently long.
382 SVal BufStart = svalBuilder.evalCast(BufVal, PtrTy, FirstBuf->getType());
383 if (Loc *BufLoc = dyn_cast<Loc>(&BufStart)) {
384 const Expr *warningExpr = (WarnAboutSize ? Size : FirstBuf);
386 SVal BufEnd = svalBuilder.evalBinOpLN(state, BO_Add, *BufLoc,
388 state = CheckLocation(C, state, warningExpr, BufEnd, firstMessage);
390 // If the buffer isn't large enough, abort.
395 // If there's a second buffer, check it as well.
397 BufVal = state->getSVal(SecondBuf, LCtx);
398 state = checkNonNull(C, state, SecondBuf, BufVal);
402 BufStart = svalBuilder.evalCast(BufVal, PtrTy, SecondBuf->getType());
403 if (Loc *BufLoc = dyn_cast<Loc>(&BufStart)) {
404 const Expr *warningExpr = (WarnAboutSize ? Size : SecondBuf);
406 SVal BufEnd = svalBuilder.evalBinOpLN(state, BO_Add, *BufLoc,
408 state = CheckLocation(C, state, warningExpr, BufEnd, secondMessage);
412 // Large enough or not, return this state!
416 ProgramStateRef CStringChecker::CheckOverlap(CheckerContext &C,
417 ProgramStateRef state,
420 const Expr *Second) const {
421 if (!Filter.CheckCStringBufferOverlap)
424 // Do a simple check for overlap: if the two arguments are from the same
425 // buffer, see if the end of the first is greater than the start of the second
428 // If a previous check has failed, propagate the failure.
432 ProgramStateRef stateTrue, stateFalse;
434 // Get the buffer values and make sure they're known locations.
435 const LocationContext *LCtx = C.getLocationContext();
436 SVal firstVal = state->getSVal(First, LCtx);
437 SVal secondVal = state->getSVal(Second, LCtx);
439 Loc *firstLoc = dyn_cast<Loc>(&firstVal);
443 Loc *secondLoc = dyn_cast<Loc>(&secondVal);
447 // Are the two values the same?
448 SValBuilder &svalBuilder = C.getSValBuilder();
449 llvm::tie(stateTrue, stateFalse) =
450 state->assume(svalBuilder.evalEQ(state, *firstLoc, *secondLoc));
452 if (stateTrue && !stateFalse) {
453 // If the values are known to be equal, that's automatically an overlap.
454 emitOverlapBug(C, stateTrue, First, Second);
458 // assume the two expressions are not equal.
462 // Which value comes first?
463 QualType cmpTy = svalBuilder.getConditionType();
464 SVal reverse = svalBuilder.evalBinOpLL(state, BO_GT,
465 *firstLoc, *secondLoc, cmpTy);
466 DefinedOrUnknownSVal *reverseTest = dyn_cast<DefinedOrUnknownSVal>(&reverse);
470 llvm::tie(stateTrue, stateFalse) = state->assume(*reverseTest);
473 // If we don't know which one comes first, we can't perform this test.
476 // Switch the values so that firstVal is before secondVal.
477 Loc *tmpLoc = firstLoc;
478 firstLoc = secondLoc;
481 // Switch the Exprs as well, so that they still correspond.
482 const Expr *tmpExpr = First;
488 // Get the length, and make sure it too is known.
489 SVal LengthVal = state->getSVal(Size, LCtx);
490 NonLoc *Length = dyn_cast<NonLoc>(&LengthVal);
494 // Convert the first buffer's start address to char*.
495 // Bail out if the cast fails.
496 ASTContext &Ctx = svalBuilder.getContext();
497 QualType CharPtrTy = Ctx.getPointerType(Ctx.CharTy);
498 SVal FirstStart = svalBuilder.evalCast(*firstLoc, CharPtrTy,
500 Loc *FirstStartLoc = dyn_cast<Loc>(&FirstStart);
504 // Compute the end of the first buffer. Bail out if THAT fails.
505 SVal FirstEnd = svalBuilder.evalBinOpLN(state, BO_Add,
506 *FirstStartLoc, *Length, CharPtrTy);
507 Loc *FirstEndLoc = dyn_cast<Loc>(&FirstEnd);
511 // Is the end of the first buffer past the start of the second buffer?
512 SVal Overlap = svalBuilder.evalBinOpLL(state, BO_GT,
513 *FirstEndLoc, *secondLoc, cmpTy);
514 DefinedOrUnknownSVal *OverlapTest = dyn_cast<DefinedOrUnknownSVal>(&Overlap);
518 llvm::tie(stateTrue, stateFalse) = state->assume(*OverlapTest);
520 if (stateTrue && !stateFalse) {
522 emitOverlapBug(C, stateTrue, First, Second);
526 // assume the two expressions don't overlap.
531 void CStringChecker::emitOverlapBug(CheckerContext &C, ProgramStateRef state,
532 const Stmt *First, const Stmt *Second) const {
533 ExplodedNode *N = C.generateSink(state);
538 BT_Overlap.reset(new BugType("Unix API", "Improper arguments"));
540 // Generate a report for this bug.
542 new BugReport(*BT_Overlap,
543 "Arguments must not be overlapping buffers", N);
544 report->addRange(First->getSourceRange());
545 report->addRange(Second->getSourceRange());
547 C.EmitReport(report);
550 ProgramStateRef CStringChecker::checkAdditionOverflow(CheckerContext &C,
551 ProgramStateRef state,
553 NonLoc right) const {
554 // If out-of-bounds checking is turned off, skip the rest.
555 if (!Filter.CheckCStringOutOfBounds)
558 // If a previous check has failed, propagate the failure.
562 SValBuilder &svalBuilder = C.getSValBuilder();
563 BasicValueFactory &BVF = svalBuilder.getBasicValueFactory();
565 QualType sizeTy = svalBuilder.getContext().getSizeType();
566 const llvm::APSInt &maxValInt = BVF.getMaxValue(sizeTy);
567 NonLoc maxVal = svalBuilder.makeIntVal(maxValInt);
570 if (isa<nonloc::ConcreteInt>(right)) {
571 maxMinusRight = svalBuilder.evalBinOpNN(state, BO_Sub, maxVal, right,
574 // Try switching the operands. (The order of these two assignments is
576 maxMinusRight = svalBuilder.evalBinOpNN(state, BO_Sub, maxVal, left,
581 if (NonLoc *maxMinusRightNL = dyn_cast<NonLoc>(&maxMinusRight)) {
582 QualType cmpTy = svalBuilder.getConditionType();
583 // If left > max - right, we have an overflow.
584 SVal willOverflow = svalBuilder.evalBinOpNN(state, BO_GT, left,
585 *maxMinusRightNL, cmpTy);
587 ProgramStateRef stateOverflow, stateOkay;
588 llvm::tie(stateOverflow, stateOkay) =
589 state->assume(cast<DefinedOrUnknownSVal>(willOverflow));
591 if (stateOverflow && !stateOkay) {
592 // We have an overflow. Emit a bug report.
593 ExplodedNode *N = C.generateSink(stateOverflow);
597 if (!BT_AdditionOverflow)
598 BT_AdditionOverflow.reset(new BuiltinBug("API",
599 "Sum of expressions causes overflow"));
601 // This isn't a great error message, but this should never occur in real
602 // code anyway -- you'd have to create a buffer longer than a size_t can
603 // represent, which is sort of a contradiction.
604 const char *warning =
605 "This expression will create a string whose length is too big to "
606 "be represented as a size_t";
608 // Generate a report for this bug.
609 BugReport *report = new BugReport(*BT_AdditionOverflow, warning, N);
610 C.EmitReport(report);
615 // From now on, assume an overflow didn't occur.
623 ProgramStateRef CStringChecker::setCStringLength(ProgramStateRef state,
626 assert(!strLength.isUndef() && "Attempt to set an undefined string length");
628 MR = MR->StripCasts();
630 switch (MR->getKind()) {
631 case MemRegion::StringRegionKind:
632 // FIXME: This can happen if we strcpy() into a string region. This is
633 // undefined [C99 6.4.5p6], but we should still warn about it.
636 case MemRegion::SymbolicRegionKind:
637 case MemRegion::AllocaRegionKind:
638 case MemRegion::VarRegionKind:
639 case MemRegion::FieldRegionKind:
640 case MemRegion::ObjCIvarRegionKind:
641 // These are the types we can currently track string lengths for.
644 case MemRegion::ElementRegionKind:
645 // FIXME: Handle element regions by upper-bounding the parent region's
650 // Other regions (mostly non-data) can't have a reliable C string length.
651 // For now, just ignore the change.
652 // FIXME: These are rare but not impossible. We should output some kind of
653 // warning for things like strcpy((char[]){'a', 0}, "b");
657 if (strLength.isUnknown())
658 return state->remove<CStringLength>(MR);
660 return state->set<CStringLength>(MR, strLength);
663 SVal CStringChecker::getCStringLengthForRegion(CheckerContext &C,
664 ProgramStateRef &state,
669 // If there's a recorded length, go ahead and return it.
670 const SVal *Recorded = state->get<CStringLength>(MR);
675 // Otherwise, get a new symbol and update the state.
676 unsigned Count = C.getCurrentBlockCount();
677 SValBuilder &svalBuilder = C.getSValBuilder();
678 QualType sizeTy = svalBuilder.getContext().getSizeType();
679 SVal strLength = svalBuilder.getMetadataSymbolVal(CStringChecker::getTag(),
680 MR, Ex, sizeTy, Count);
683 state = state->set<CStringLength>(MR, strLength);
688 SVal CStringChecker::getCStringLength(CheckerContext &C, ProgramStateRef &state,
689 const Expr *Ex, SVal Buf,
690 bool hypothetical) const {
691 const MemRegion *MR = Buf.getAsRegion();
693 // If we can't get a region, see if it's something we /know/ isn't a
694 // C string. In the context of locations, the only time we can issue such
695 // a warning is for labels.
696 if (loc::GotoLabel *Label = dyn_cast<loc::GotoLabel>(&Buf)) {
697 if (!Filter.CheckCStringNotNullTerm)
698 return UndefinedVal();
700 if (ExplodedNode *N = C.addTransition(state)) {
702 BT_NotCString.reset(new BuiltinBug("Unix API",
703 "Argument is not a null-terminated string."));
705 SmallString<120> buf;
706 llvm::raw_svector_ostream os(buf);
707 assert(CurrentFunctionDescription);
708 os << "Argument to " << CurrentFunctionDescription
709 << " is the address of the label '" << Label->getLabel()->getName()
710 << "', which is not a null-terminated string";
712 // Generate a report for this bug.
713 BugReport *report = new BugReport(*BT_NotCString,
716 report->addRange(Ex->getSourceRange());
717 C.EmitReport(report);
719 return UndefinedVal();
723 // If it's not a region and not a label, give up.
727 // If we have a region, strip casts from it and see if we can figure out
728 // its length. For anything we can't figure out, just return UnknownVal.
729 MR = MR->StripCasts();
731 switch (MR->getKind()) {
732 case MemRegion::StringRegionKind: {
733 // Modifying the contents of string regions is undefined [C99 6.4.5p6],
734 // so we can assume that the byte length is the correct C string length.
735 SValBuilder &svalBuilder = C.getSValBuilder();
736 QualType sizeTy = svalBuilder.getContext().getSizeType();
737 const StringLiteral *strLit = cast<StringRegion>(MR)->getStringLiteral();
738 return svalBuilder.makeIntVal(strLit->getByteLength(), sizeTy);
740 case MemRegion::SymbolicRegionKind:
741 case MemRegion::AllocaRegionKind:
742 case MemRegion::VarRegionKind:
743 case MemRegion::FieldRegionKind:
744 case MemRegion::ObjCIvarRegionKind:
745 return getCStringLengthForRegion(C, state, Ex, MR, hypothetical);
746 case MemRegion::CompoundLiteralRegionKind:
747 // FIXME: Can we track this? Is it necessary?
749 case MemRegion::ElementRegionKind:
750 // FIXME: How can we handle this? It's not good enough to subtract the
751 // offset from the base string length; consider "123\x00567" and &a[5].
754 // Other regions (mostly non-data) can't have a reliable C string length.
755 // In this case, an error is emitted and UndefinedVal is returned.
756 // The caller should always be prepared to handle this case.
757 if (!Filter.CheckCStringNotNullTerm)
758 return UndefinedVal();
760 if (ExplodedNode *N = C.addTransition(state)) {
762 BT_NotCString.reset(new BuiltinBug("Unix API",
763 "Argument is not a null-terminated string."));
765 SmallString<120> buf;
766 llvm::raw_svector_ostream os(buf);
768 assert(CurrentFunctionDescription);
769 os << "Argument to " << CurrentFunctionDescription << " is ";
771 if (SummarizeRegion(os, C.getASTContext(), MR))
772 os << ", which is not a null-terminated string";
774 os << "not a null-terminated string";
776 // Generate a report for this bug.
777 BugReport *report = new BugReport(*BT_NotCString,
780 report->addRange(Ex->getSourceRange());
781 C.EmitReport(report);
784 return UndefinedVal();
788 const StringLiteral *CStringChecker::getCStringLiteral(CheckerContext &C,
789 ProgramStateRef &state, const Expr *expr, SVal val) const {
791 // Get the memory region pointed to by the val.
792 const MemRegion *bufRegion = val.getAsRegion();
796 // Strip casts off the memory region.
797 bufRegion = bufRegion->StripCasts();
799 // Cast the memory region to a string region.
800 const StringRegion *strRegion= dyn_cast<StringRegion>(bufRegion);
804 // Return the actual string in the string region.
805 return strRegion->getStringLiteral();
808 ProgramStateRef CStringChecker::InvalidateBuffer(CheckerContext &C,
809 ProgramStateRef state,
810 const Expr *E, SVal V) {
811 Loc *L = dyn_cast<Loc>(&V);
815 // FIXME: This is a simplified version of what's in CFRefCount.cpp -- it makes
816 // some assumptions about the value that CFRefCount can't. Even so, it should
817 // probably be refactored.
818 if (loc::MemRegionVal* MR = dyn_cast<loc::MemRegionVal>(L)) {
819 const MemRegion *R = MR->getRegion()->StripCasts();
821 // Are we dealing with an ElementRegion? If so, we should be invalidating
823 if (const ElementRegion *ER = dyn_cast<ElementRegion>(R)) {
824 R = ER->getSuperRegion();
825 // FIXME: What about layers of ElementRegions?
828 // Invalidate this region.
829 unsigned Count = C.getCurrentBlockCount();
830 const LocationContext *LCtx = C.getPredecessor()->getLocationContext();
831 return state->invalidateRegions(R, E, Count, LCtx);
834 // If we have a non-region value by chance, just remove the binding.
835 // FIXME: is this necessary or correct? This handles the non-Region
836 // cases. Is it ever valid to store to these?
837 return state->unbindLoc(*L);
840 bool CStringChecker::SummarizeRegion(raw_ostream &os, ASTContext &Ctx,
841 const MemRegion *MR) {
842 const TypedValueRegion *TVR = dyn_cast<TypedValueRegion>(MR);
844 switch (MR->getKind()) {
845 case MemRegion::FunctionTextRegionKind: {
846 const FunctionDecl *FD = cast<FunctionTextRegion>(MR)->getDecl();
848 os << "the address of the function '" << *FD << '\'';
850 os << "the address of a function";
853 case MemRegion::BlockTextRegionKind:
856 case MemRegion::BlockDataRegionKind:
859 case MemRegion::CXXThisRegionKind:
860 case MemRegion::CXXTempObjectRegionKind:
861 os << "a C++ temp object of type " << TVR->getValueType().getAsString();
863 case MemRegion::VarRegionKind:
864 os << "a variable of type" << TVR->getValueType().getAsString();
866 case MemRegion::FieldRegionKind:
867 os << "a field of type " << TVR->getValueType().getAsString();
869 case MemRegion::ObjCIvarRegionKind:
870 os << "an instance variable of type " << TVR->getValueType().getAsString();
877 //===----------------------------------------------------------------------===//
878 // evaluation of individual function calls.
879 //===----------------------------------------------------------------------===//
881 void CStringChecker::evalCopyCommon(CheckerContext &C,
883 ProgramStateRef state,
884 const Expr *Size, const Expr *Dest,
885 const Expr *Source, bool Restricted,
886 bool IsMempcpy) const {
887 CurrentFunctionDescription = "memory copy function";
889 // See if the size argument is zero.
890 const LocationContext *LCtx = C.getLocationContext();
891 SVal sizeVal = state->getSVal(Size, LCtx);
892 QualType sizeTy = Size->getType();
894 ProgramStateRef stateZeroSize, stateNonZeroSize;
895 llvm::tie(stateZeroSize, stateNonZeroSize) =
896 assumeZero(C, state, sizeVal, sizeTy);
898 // Get the value of the Dest.
899 SVal destVal = state->getSVal(Dest, LCtx);
901 // If the size is zero, there won't be any actual memory access, so
902 // just bind the return value to the destination buffer and return.
903 if (stateZeroSize && !stateNonZeroSize) {
904 stateZeroSize = stateZeroSize->BindExpr(CE, LCtx, destVal);
905 C.addTransition(stateZeroSize);
909 // If the size can be nonzero, we have to check the other arguments.
910 if (stateNonZeroSize) {
911 state = stateNonZeroSize;
913 // Ensure the destination is not null. If it is NULL there will be a
914 // NULL pointer dereference.
915 state = checkNonNull(C, state, Dest, destVal);
919 // Get the value of the Src.
920 SVal srcVal = state->getSVal(Source, LCtx);
922 // Ensure the source is not null. If it is NULL there will be a
923 // NULL pointer dereference.
924 state = checkNonNull(C, state, Source, srcVal);
928 // Ensure the accesses are valid and that the buffers do not overlap.
929 const char * const writeWarning =
930 "Memory copy function overflows destination buffer";
931 state = CheckBufferAccess(C, state, Size, Dest, Source,
932 writeWarning, /* sourceWarning = */ NULL);
934 state = CheckOverlap(C, state, Size, Dest, Source);
939 // If this is mempcpy, get the byte after the last byte copied and
942 loc::MemRegionVal *destRegVal = dyn_cast<loc::MemRegionVal>(&destVal);
943 assert(destRegVal && "Destination should be a known MemRegionVal here");
945 // Get the length to copy.
946 NonLoc *lenValNonLoc = dyn_cast<NonLoc>(&sizeVal);
949 // Get the byte after the last byte copied.
950 SVal lastElement = C.getSValBuilder().evalBinOpLN(state, BO_Add,
955 // The byte after the last byte copied is the return value.
956 state = state->BindExpr(CE, LCtx, lastElement);
958 // If we don't know how much we copied, we can at least
959 // conjure a return value for later.
960 unsigned Count = C.getCurrentBlockCount();
962 C.getSValBuilder().getConjuredSymbolVal(NULL, CE, LCtx, Count);
963 state = state->BindExpr(CE, LCtx, result);
967 // All other copies return the destination buffer.
968 // (Well, bcopy() has a void return type, but this won't hurt.)
969 state = state->BindExpr(CE, LCtx, destVal);
972 // Invalidate the destination.
973 // FIXME: Even if we can't perfectly model the copy, we should see if we
974 // can use LazyCompoundVals to copy the source values into the destination.
975 // This would probably remove any existing bindings past the end of the
976 // copied region, but that's still an improvement over blank invalidation.
977 state = InvalidateBuffer(C, state, Dest,
978 state->getSVal(Dest, C.getLocationContext()));
979 C.addTransition(state);
984 void CStringChecker::evalMemcpy(CheckerContext &C, const CallExpr *CE) const {
985 if (CE->getNumArgs() < 3)
988 // void *memcpy(void *restrict dst, const void *restrict src, size_t n);
989 // The return value is the address of the destination buffer.
990 const Expr *Dest = CE->getArg(0);
991 ProgramStateRef state = C.getState();
993 evalCopyCommon(C, CE, state, CE->getArg(2), Dest, CE->getArg(1), true);
996 void CStringChecker::evalMempcpy(CheckerContext &C, const CallExpr *CE) const {
997 if (CE->getNumArgs() < 3)
1000 // void *mempcpy(void *restrict dst, const void *restrict src, size_t n);
1001 // The return value is a pointer to the byte following the last written byte.
1002 const Expr *Dest = CE->getArg(0);
1003 ProgramStateRef state = C.getState();
1005 evalCopyCommon(C, CE, state, CE->getArg(2), Dest, CE->getArg(1), true, true);
1008 void CStringChecker::evalMemmove(CheckerContext &C, const CallExpr *CE) const {
1009 if (CE->getNumArgs() < 3)
1012 // void *memmove(void *dst, const void *src, size_t n);
1013 // The return value is the address of the destination buffer.
1014 const Expr *Dest = CE->getArg(0);
1015 ProgramStateRef state = C.getState();
1017 evalCopyCommon(C, CE, state, CE->getArg(2), Dest, CE->getArg(1));
1020 void CStringChecker::evalBcopy(CheckerContext &C, const CallExpr *CE) const {
1021 if (CE->getNumArgs() < 3)
1024 // void bcopy(const void *src, void *dst, size_t n);
1025 evalCopyCommon(C, CE, C.getState(),
1026 CE->getArg(2), CE->getArg(1), CE->getArg(0));
1029 void CStringChecker::evalMemcmp(CheckerContext &C, const CallExpr *CE) const {
1030 if (CE->getNumArgs() < 3)
1033 // int memcmp(const void *s1, const void *s2, size_t n);
1034 CurrentFunctionDescription = "memory comparison function";
1036 const Expr *Left = CE->getArg(0);
1037 const Expr *Right = CE->getArg(1);
1038 const Expr *Size = CE->getArg(2);
1040 ProgramStateRef state = C.getState();
1041 SValBuilder &svalBuilder = C.getSValBuilder();
1043 // See if the size argument is zero.
1044 const LocationContext *LCtx = C.getLocationContext();
1045 SVal sizeVal = state->getSVal(Size, LCtx);
1046 QualType sizeTy = Size->getType();
1048 ProgramStateRef stateZeroSize, stateNonZeroSize;
1049 llvm::tie(stateZeroSize, stateNonZeroSize) =
1050 assumeZero(C, state, sizeVal, sizeTy);
1052 // If the size can be zero, the result will be 0 in that case, and we don't
1053 // have to check either of the buffers.
1054 if (stateZeroSize) {
1055 state = stateZeroSize;
1056 state = state->BindExpr(CE, LCtx,
1057 svalBuilder.makeZeroVal(CE->getType()));
1058 C.addTransition(state);
1061 // If the size can be nonzero, we have to check the other arguments.
1062 if (stateNonZeroSize) {
1063 state = stateNonZeroSize;
1064 // If we know the two buffers are the same, we know the result is 0.
1065 // First, get the two buffers' addresses. Another checker will have already
1066 // made sure they're not undefined.
1067 DefinedOrUnknownSVal LV =
1068 cast<DefinedOrUnknownSVal>(state->getSVal(Left, LCtx));
1069 DefinedOrUnknownSVal RV =
1070 cast<DefinedOrUnknownSVal>(state->getSVal(Right, LCtx));
1072 // See if they are the same.
1073 DefinedOrUnknownSVal SameBuf = svalBuilder.evalEQ(state, LV, RV);
1074 ProgramStateRef StSameBuf, StNotSameBuf;
1075 llvm::tie(StSameBuf, StNotSameBuf) = state->assume(SameBuf);
1077 // If the two arguments might be the same buffer, we know the result is 0,
1078 // and we only need to check one size.
1081 state = CheckBufferAccess(C, state, Size, Left);
1083 state = StSameBuf->BindExpr(CE, LCtx,
1084 svalBuilder.makeZeroVal(CE->getType()));
1085 C.addTransition(state);
1089 // If the two arguments might be different buffers, we have to check the
1090 // size of both of them.
1092 state = StNotSameBuf;
1093 state = CheckBufferAccess(C, state, Size, Left, Right);
1095 // The return value is the comparison result, which we don't know.
1096 unsigned Count = C.getCurrentBlockCount();
1097 SVal CmpV = svalBuilder.getConjuredSymbolVal(NULL, CE, LCtx, Count);
1098 state = state->BindExpr(CE, LCtx, CmpV);
1099 C.addTransition(state);
1105 void CStringChecker::evalstrLength(CheckerContext &C,
1106 const CallExpr *CE) const {
1107 if (CE->getNumArgs() < 1)
1110 // size_t strlen(const char *s);
1111 evalstrLengthCommon(C, CE, /* IsStrnlen = */ false);
1114 void CStringChecker::evalstrnLength(CheckerContext &C,
1115 const CallExpr *CE) const {
1116 if (CE->getNumArgs() < 2)
1119 // size_t strnlen(const char *s, size_t maxlen);
1120 evalstrLengthCommon(C, CE, /* IsStrnlen = */ true);
1123 void CStringChecker::evalstrLengthCommon(CheckerContext &C, const CallExpr *CE,
1124 bool IsStrnlen) const {
1125 CurrentFunctionDescription = "string length function";
1126 ProgramStateRef state = C.getState();
1127 const LocationContext *LCtx = C.getLocationContext();
1130 const Expr *maxlenExpr = CE->getArg(1);
1131 SVal maxlenVal = state->getSVal(maxlenExpr, LCtx);
1133 ProgramStateRef stateZeroSize, stateNonZeroSize;
1134 llvm::tie(stateZeroSize, stateNonZeroSize) =
1135 assumeZero(C, state, maxlenVal, maxlenExpr->getType());
1137 // If the size can be zero, the result will be 0 in that case, and we don't
1138 // have to check the string itself.
1139 if (stateZeroSize) {
1140 SVal zero = C.getSValBuilder().makeZeroVal(CE->getType());
1141 stateZeroSize = stateZeroSize->BindExpr(CE, LCtx, zero);
1142 C.addTransition(stateZeroSize);
1145 // If the size is GUARANTEED to be zero, we're done!
1146 if (!stateNonZeroSize)
1149 // Otherwise, record the assumption that the size is nonzero.
1150 state = stateNonZeroSize;
1153 // Check that the string argument is non-null.
1154 const Expr *Arg = CE->getArg(0);
1155 SVal ArgVal = state->getSVal(Arg, LCtx);
1157 state = checkNonNull(C, state, Arg, ArgVal);
1162 SVal strLength = getCStringLength(C, state, Arg, ArgVal);
1164 // If the argument isn't a valid C string, there's no valid state to
1166 if (strLength.isUndef())
1169 DefinedOrUnknownSVal result = UnknownVal();
1171 // If the check is for strnlen() then bind the return value to no more than
1172 // the maxlen value.
1174 QualType cmpTy = C.getSValBuilder().getConditionType();
1176 // It's a little unfortunate to be getting this again,
1177 // but it's not that expensive...
1178 const Expr *maxlenExpr = CE->getArg(1);
1179 SVal maxlenVal = state->getSVal(maxlenExpr, LCtx);
1181 NonLoc *strLengthNL = dyn_cast<NonLoc>(&strLength);
1182 NonLoc *maxlenValNL = dyn_cast<NonLoc>(&maxlenVal);
1184 if (strLengthNL && maxlenValNL) {
1185 ProgramStateRef stateStringTooLong, stateStringNotTooLong;
1187 // Check if the strLength is greater than the maxlen.
1188 llvm::tie(stateStringTooLong, stateStringNotTooLong) =
1189 state->assume(cast<DefinedOrUnknownSVal>
1190 (C.getSValBuilder().evalBinOpNN(state, BO_GT,
1195 if (stateStringTooLong && !stateStringNotTooLong) {
1196 // If the string is longer than maxlen, return maxlen.
1197 result = *maxlenValNL;
1198 } else if (stateStringNotTooLong && !stateStringTooLong) {
1199 // If the string is shorter than maxlen, return its length.
1200 result = *strLengthNL;
1204 if (result.isUnknown()) {
1205 // If we don't have enough information for a comparison, there's
1206 // no guarantee the full string length will actually be returned.
1207 // All we know is the return value is the min of the string length
1208 // and the limit. This is better than nothing.
1209 unsigned Count = C.getCurrentBlockCount();
1210 result = C.getSValBuilder().getConjuredSymbolVal(NULL, CE, LCtx, Count);
1211 NonLoc *resultNL = cast<NonLoc>(&result);
1214 state = state->assume(cast<DefinedOrUnknownSVal>
1215 (C.getSValBuilder().evalBinOpNN(state, BO_LE,
1222 state = state->assume(cast<DefinedOrUnknownSVal>
1223 (C.getSValBuilder().evalBinOpNN(state, BO_LE,
1231 // This is a plain strlen(), not strnlen().
1232 result = cast<DefinedOrUnknownSVal>(strLength);
1234 // If we don't know the length of the string, conjure a return
1235 // value, so it can be used in constraints, at least.
1236 if (result.isUnknown()) {
1237 unsigned Count = C.getCurrentBlockCount();
1238 result = C.getSValBuilder().getConjuredSymbolVal(NULL, CE, LCtx, Count);
1242 // Bind the return value.
1243 assert(!result.isUnknown() && "Should have conjured a value by now");
1244 state = state->BindExpr(CE, LCtx, result);
1245 C.addTransition(state);
1248 void CStringChecker::evalStrcpy(CheckerContext &C, const CallExpr *CE) const {
1249 if (CE->getNumArgs() < 2)
1252 // char *strcpy(char *restrict dst, const char *restrict src);
1253 evalStrcpyCommon(C, CE,
1254 /* returnEnd = */ false,
1255 /* isBounded = */ false,
1256 /* isAppending = */ false);
1259 void CStringChecker::evalStrncpy(CheckerContext &C, const CallExpr *CE) const {
1260 if (CE->getNumArgs() < 3)
1263 // char *strncpy(char *restrict dst, const char *restrict src, size_t n);
1264 evalStrcpyCommon(C, CE,
1265 /* returnEnd = */ false,
1266 /* isBounded = */ true,
1267 /* isAppending = */ false);
1270 void CStringChecker::evalStpcpy(CheckerContext &C, const CallExpr *CE) const {
1271 if (CE->getNumArgs() < 2)
1274 // char *stpcpy(char *restrict dst, const char *restrict src);
1275 evalStrcpyCommon(C, CE,
1276 /* returnEnd = */ true,
1277 /* isBounded = */ false,
1278 /* isAppending = */ false);
1281 void CStringChecker::evalStrcat(CheckerContext &C, const CallExpr *CE) const {
1282 if (CE->getNumArgs() < 2)
1285 //char *strcat(char *restrict s1, const char *restrict s2);
1286 evalStrcpyCommon(C, CE,
1287 /* returnEnd = */ false,
1288 /* isBounded = */ false,
1289 /* isAppending = */ true);
1292 void CStringChecker::evalStrncat(CheckerContext &C, const CallExpr *CE) const {
1293 if (CE->getNumArgs() < 3)
1296 //char *strncat(char *restrict s1, const char *restrict s2, size_t n);
1297 evalStrcpyCommon(C, CE,
1298 /* returnEnd = */ false,
1299 /* isBounded = */ true,
1300 /* isAppending = */ true);
1303 void CStringChecker::evalStrcpyCommon(CheckerContext &C, const CallExpr *CE,
1304 bool returnEnd, bool isBounded,
1305 bool isAppending) const {
1306 CurrentFunctionDescription = "string copy function";
1307 ProgramStateRef state = C.getState();
1308 const LocationContext *LCtx = C.getLocationContext();
1310 // Check that the destination is non-null.
1311 const Expr *Dst = CE->getArg(0);
1312 SVal DstVal = state->getSVal(Dst, LCtx);
1314 state = checkNonNull(C, state, Dst, DstVal);
1318 // Check that the source is non-null.
1319 const Expr *srcExpr = CE->getArg(1);
1320 SVal srcVal = state->getSVal(srcExpr, LCtx);
1321 state = checkNonNull(C, state, srcExpr, srcVal);
1325 // Get the string length of the source.
1326 SVal strLength = getCStringLength(C, state, srcExpr, srcVal);
1328 // If the source isn't a valid C string, give up.
1329 if (strLength.isUndef())
1332 SValBuilder &svalBuilder = C.getSValBuilder();
1333 QualType cmpTy = svalBuilder.getConditionType();
1334 QualType sizeTy = svalBuilder.getContext().getSizeType();
1336 // These two values allow checking two kinds of errors:
1337 // - actual overflows caused by a source that doesn't fit in the destination
1338 // - potential overflows caused by a bound that could exceed the destination
1339 SVal amountCopied = UnknownVal();
1340 SVal maxLastElementIndex = UnknownVal();
1341 const char *boundWarning = NULL;
1343 // If the function is strncpy, strncat, etc... it is bounded.
1345 // Get the max number of characters to copy.
1346 const Expr *lenExpr = CE->getArg(2);
1347 SVal lenVal = state->getSVal(lenExpr, LCtx);
1349 // Protect against misdeclared strncpy().
1350 lenVal = svalBuilder.evalCast(lenVal, sizeTy, lenExpr->getType());
1352 NonLoc *strLengthNL = dyn_cast<NonLoc>(&strLength);
1353 NonLoc *lenValNL = dyn_cast<NonLoc>(&lenVal);
1355 // If we know both values, we might be able to figure out how much
1357 if (strLengthNL && lenValNL) {
1358 ProgramStateRef stateSourceTooLong, stateSourceNotTooLong;
1360 // Check if the max number to copy is less than the length of the src.
1361 // If the bound is equal to the source length, strncpy won't null-
1362 // terminate the result!
1363 llvm::tie(stateSourceTooLong, stateSourceNotTooLong) =
1364 state->assume(cast<DefinedOrUnknownSVal>
1365 (svalBuilder.evalBinOpNN(state, BO_GE, *strLengthNL,
1366 *lenValNL, cmpTy)));
1368 if (stateSourceTooLong && !stateSourceNotTooLong) {
1369 // Max number to copy is less than the length of the src, so the actual
1370 // strLength copied is the max number arg.
1371 state = stateSourceTooLong;
1372 amountCopied = lenVal;
1374 } else if (!stateSourceTooLong && stateSourceNotTooLong) {
1375 // The source buffer entirely fits in the bound.
1376 state = stateSourceNotTooLong;
1377 amountCopied = strLength;
1381 // We still want to know if the bound is known to be too large.
1384 // For strncat, the check is strlen(dst) + lenVal < sizeof(dst)
1386 // Get the string length of the destination. If the destination is
1387 // memory that can't have a string length, we shouldn't be copying
1389 SVal dstStrLength = getCStringLength(C, state, Dst, DstVal);
1390 if (dstStrLength.isUndef())
1393 if (NonLoc *dstStrLengthNL = dyn_cast<NonLoc>(&dstStrLength)) {
1394 maxLastElementIndex = svalBuilder.evalBinOpNN(state, BO_Add,
1398 boundWarning = "Size argument is greater than the free space in the "
1399 "destination buffer";
1403 // For strncpy, this is just checking that lenVal <= sizeof(dst)
1404 // (Yes, strncpy and strncat differ in how they treat termination.
1405 // strncat ALWAYS terminates, but strncpy doesn't.)
1407 // We need a special case for when the copy size is zero, in which
1408 // case strncpy will do no work at all. Our bounds check uses n-1
1409 // as the last element accessed, so n == 0 is problematic.
1410 ProgramStateRef StateZeroSize, StateNonZeroSize;
1411 llvm::tie(StateZeroSize, StateNonZeroSize) =
1412 assumeZero(C, state, *lenValNL, sizeTy);
1414 // If the size is known to be zero, we're done.
1415 if (StateZeroSize && !StateNonZeroSize) {
1416 StateZeroSize = StateZeroSize->BindExpr(CE, LCtx, DstVal);
1417 C.addTransition(StateZeroSize);
1421 // Otherwise, go ahead and figure out the last element we'll touch.
1422 // We don't record the non-zero assumption here because we can't
1423 // be sure. We won't warn on a possible zero.
1424 NonLoc one = cast<NonLoc>(svalBuilder.makeIntVal(1, sizeTy));
1425 maxLastElementIndex = svalBuilder.evalBinOpNN(state, BO_Sub, *lenValNL,
1427 boundWarning = "Size argument is greater than the length of the "
1428 "destination buffer";
1432 // If we couldn't pin down the copy length, at least bound it.
1433 // FIXME: We should actually run this code path for append as well, but
1434 // right now it creates problems with constraints (since we can end up
1435 // trying to pass constraints from symbol to symbol).
1436 if (amountCopied.isUnknown() && !isAppending) {
1437 // Try to get a "hypothetical" string length symbol, which we can later
1438 // set as a real value if that turns out to be the case.
1439 amountCopied = getCStringLength(C, state, lenExpr, srcVal, true);
1440 assert(!amountCopied.isUndef());
1442 if (NonLoc *amountCopiedNL = dyn_cast<NonLoc>(&amountCopied)) {
1444 // amountCopied <= lenVal
1445 SVal copiedLessThanBound = svalBuilder.evalBinOpNN(state, BO_LE,
1449 state = state->assume(cast<DefinedOrUnknownSVal>(copiedLessThanBound),
1456 // amountCopied <= strlen(source)
1457 SVal copiedLessThanSrc = svalBuilder.evalBinOpNN(state, BO_LE,
1461 state = state->assume(cast<DefinedOrUnknownSVal>(copiedLessThanSrc),
1470 // The function isn't bounded. The amount copied should match the length
1471 // of the source buffer.
1472 amountCopied = strLength;
1477 // This represents the number of characters copied into the destination
1478 // buffer. (It may not actually be the strlen if the destination buffer
1479 // is not terminated.)
1480 SVal finalStrLength = UnknownVal();
1482 // If this is an appending function (strcat, strncat...) then set the
1483 // string length to strlen(src) + strlen(dst) since the buffer will
1484 // ultimately contain both.
1486 // Get the string length of the destination. If the destination is memory
1487 // that can't have a string length, we shouldn't be copying into it anyway.
1488 SVal dstStrLength = getCStringLength(C, state, Dst, DstVal);
1489 if (dstStrLength.isUndef())
1492 NonLoc *srcStrLengthNL = dyn_cast<NonLoc>(&amountCopied);
1493 NonLoc *dstStrLengthNL = dyn_cast<NonLoc>(&dstStrLength);
1495 // If we know both string lengths, we might know the final string length.
1496 if (srcStrLengthNL && dstStrLengthNL) {
1497 // Make sure the two lengths together don't overflow a size_t.
1498 state = checkAdditionOverflow(C, state, *srcStrLengthNL, *dstStrLengthNL);
1502 finalStrLength = svalBuilder.evalBinOpNN(state, BO_Add, *srcStrLengthNL,
1503 *dstStrLengthNL, sizeTy);
1506 // If we couldn't get a single value for the final string length,
1507 // we can at least bound it by the individual lengths.
1508 if (finalStrLength.isUnknown()) {
1509 // Try to get a "hypothetical" string length symbol, which we can later
1510 // set as a real value if that turns out to be the case.
1511 finalStrLength = getCStringLength(C, state, CE, DstVal, true);
1512 assert(!finalStrLength.isUndef());
1514 if (NonLoc *finalStrLengthNL = dyn_cast<NonLoc>(&finalStrLength)) {
1515 if (srcStrLengthNL) {
1516 // finalStrLength >= srcStrLength
1517 SVal sourceInResult = svalBuilder.evalBinOpNN(state, BO_GE,
1521 state = state->assume(cast<DefinedOrUnknownSVal>(sourceInResult),
1527 if (dstStrLengthNL) {
1528 // finalStrLength >= dstStrLength
1529 SVal destInResult = svalBuilder.evalBinOpNN(state, BO_GE,
1533 state = state->assume(cast<DefinedOrUnknownSVal>(destInResult),
1542 // Otherwise, this is a copy-over function (strcpy, strncpy, ...), and
1543 // the final string length will match the input string length.
1544 finalStrLength = amountCopied;
1547 // The final result of the function will either be a pointer past the last
1548 // copied element, or a pointer to the start of the destination buffer.
1549 SVal Result = (returnEnd ? UnknownVal() : DstVal);
1553 // If the destination is a MemRegion, try to check for a buffer overflow and
1554 // record the new string length.
1555 if (loc::MemRegionVal *dstRegVal = dyn_cast<loc::MemRegionVal>(&DstVal)) {
1556 QualType ptrTy = Dst->getType();
1558 // If we have an exact value on a bounded copy, use that to check for
1559 // overflows, rather than our estimate about how much is actually copied.
1561 if (NonLoc *maxLastNL = dyn_cast<NonLoc>(&maxLastElementIndex)) {
1562 SVal maxLastElement = svalBuilder.evalBinOpLN(state, BO_Add, *dstRegVal,
1564 state = CheckLocation(C, state, CE->getArg(2), maxLastElement,
1571 // Then, if the final length is known...
1572 if (NonLoc *knownStrLength = dyn_cast<NonLoc>(&finalStrLength)) {
1573 SVal lastElement = svalBuilder.evalBinOpLN(state, BO_Add, *dstRegVal,
1574 *knownStrLength, ptrTy);
1576 // ...and we haven't checked the bound, we'll check the actual copy.
1577 if (!boundWarning) {
1578 const char * const warningMsg =
1579 "String copy function overflows destination buffer";
1580 state = CheckLocation(C, state, Dst, lastElement, warningMsg);
1585 // If this is a stpcpy-style copy, the last element is the return value.
1587 Result = lastElement;
1590 // Invalidate the destination. This must happen before we set the C string
1591 // length because invalidation will clear the length.
1592 // FIXME: Even if we can't perfectly model the copy, we should see if we
1593 // can use LazyCompoundVals to copy the source values into the destination.
1594 // This would probably remove any existing bindings past the end of the
1595 // string, but that's still an improvement over blank invalidation.
1596 state = InvalidateBuffer(C, state, Dst, *dstRegVal);
1598 // Set the C string length of the destination, if we know it.
1599 if (isBounded && !isAppending) {
1600 // strncpy is annoying in that it doesn't guarantee to null-terminate
1601 // the result string. If the original string didn't fit entirely inside
1602 // the bound (including the null-terminator), we don't know how long the
1604 if (amountCopied != strLength)
1605 finalStrLength = UnknownVal();
1607 state = setCStringLength(state, dstRegVal->getRegion(), finalStrLength);
1612 // If this is a stpcpy-style copy, but we were unable to check for a buffer
1613 // overflow, we still need a result. Conjure a return value.
1614 if (returnEnd && Result.isUnknown()) {
1615 unsigned Count = C.getCurrentBlockCount();
1616 Result = svalBuilder.getConjuredSymbolVal(NULL, CE, LCtx, Count);
1619 // Set the return value.
1620 state = state->BindExpr(CE, LCtx, Result);
1621 C.addTransition(state);
1624 void CStringChecker::evalStrcmp(CheckerContext &C, const CallExpr *CE) const {
1625 if (CE->getNumArgs() < 2)
1628 //int strcmp(const char *s1, const char *s2);
1629 evalStrcmpCommon(C, CE, /* isBounded = */ false, /* ignoreCase = */ false);
1632 void CStringChecker::evalStrncmp(CheckerContext &C, const CallExpr *CE) const {
1633 if (CE->getNumArgs() < 3)
1636 //int strncmp(const char *s1, const char *s2, size_t n);
1637 evalStrcmpCommon(C, CE, /* isBounded = */ true, /* ignoreCase = */ false);
1640 void CStringChecker::evalStrcasecmp(CheckerContext &C,
1641 const CallExpr *CE) const {
1642 if (CE->getNumArgs() < 2)
1645 //int strcasecmp(const char *s1, const char *s2);
1646 evalStrcmpCommon(C, CE, /* isBounded = */ false, /* ignoreCase = */ true);
1649 void CStringChecker::evalStrncasecmp(CheckerContext &C,
1650 const CallExpr *CE) const {
1651 if (CE->getNumArgs() < 3)
1654 //int strncasecmp(const char *s1, const char *s2, size_t n);
1655 evalStrcmpCommon(C, CE, /* isBounded = */ true, /* ignoreCase = */ true);
1658 void CStringChecker::evalStrcmpCommon(CheckerContext &C, const CallExpr *CE,
1659 bool isBounded, bool ignoreCase) const {
1660 CurrentFunctionDescription = "string comparison function";
1661 ProgramStateRef state = C.getState();
1662 const LocationContext *LCtx = C.getLocationContext();
1664 // Check that the first string is non-null
1665 const Expr *s1 = CE->getArg(0);
1666 SVal s1Val = state->getSVal(s1, LCtx);
1667 state = checkNonNull(C, state, s1, s1Val);
1671 // Check that the second string is non-null.
1672 const Expr *s2 = CE->getArg(1);
1673 SVal s2Val = state->getSVal(s2, LCtx);
1674 state = checkNonNull(C, state, s2, s2Val);
1678 // Get the string length of the first string or give up.
1679 SVal s1Length = getCStringLength(C, state, s1, s1Val);
1680 if (s1Length.isUndef())
1683 // Get the string length of the second string or give up.
1684 SVal s2Length = getCStringLength(C, state, s2, s2Val);
1685 if (s2Length.isUndef())
1688 // If we know the two buffers are the same, we know the result is 0.
1689 // First, get the two buffers' addresses. Another checker will have already
1690 // made sure they're not undefined.
1691 DefinedOrUnknownSVal LV = cast<DefinedOrUnknownSVal>(s1Val);
1692 DefinedOrUnknownSVal RV = cast<DefinedOrUnknownSVal>(s2Val);
1694 // See if they are the same.
1695 SValBuilder &svalBuilder = C.getSValBuilder();
1696 DefinedOrUnknownSVal SameBuf = svalBuilder.evalEQ(state, LV, RV);
1697 ProgramStateRef StSameBuf, StNotSameBuf;
1698 llvm::tie(StSameBuf, StNotSameBuf) = state->assume(SameBuf);
1700 // If the two arguments might be the same buffer, we know the result is 0,
1701 // and we only need to check one size.
1703 StSameBuf = StSameBuf->BindExpr(CE, LCtx,
1704 svalBuilder.makeZeroVal(CE->getType()));
1705 C.addTransition(StSameBuf);
1707 // If the two arguments are GUARANTEED to be the same, we're done!
1712 assert(StNotSameBuf);
1713 state = StNotSameBuf;
1715 // At this point we can go about comparing the two buffers.
1716 // For now, we only do this if they're both known string literals.
1718 // Attempt to extract string literals from both expressions.
1719 const StringLiteral *s1StrLiteral = getCStringLiteral(C, state, s1, s1Val);
1720 const StringLiteral *s2StrLiteral = getCStringLiteral(C, state, s2, s2Val);
1721 bool canComputeResult = false;
1723 if (s1StrLiteral && s2StrLiteral) {
1724 StringRef s1StrRef = s1StrLiteral->getString();
1725 StringRef s2StrRef = s2StrLiteral->getString();
1728 // Get the max number of characters to compare.
1729 const Expr *lenExpr = CE->getArg(2);
1730 SVal lenVal = state->getSVal(lenExpr, LCtx);
1732 // If the length is known, we can get the right substrings.
1733 if (const llvm::APSInt *len = svalBuilder.getKnownValue(state, lenVal)) {
1734 // Create substrings of each to compare the prefix.
1735 s1StrRef = s1StrRef.substr(0, (size_t)len->getZExtValue());
1736 s2StrRef = s2StrRef.substr(0, (size_t)len->getZExtValue());
1737 canComputeResult = true;
1740 // This is a normal, unbounded strcmp.
1741 canComputeResult = true;
1744 if (canComputeResult) {
1745 // Real strcmp stops at null characters.
1746 size_t s1Term = s1StrRef.find('\0');
1747 if (s1Term != StringRef::npos)
1748 s1StrRef = s1StrRef.substr(0, s1Term);
1750 size_t s2Term = s2StrRef.find('\0');
1751 if (s2Term != StringRef::npos)
1752 s2StrRef = s2StrRef.substr(0, s2Term);
1754 // Use StringRef's comparison methods to compute the actual result.
1758 // Compare string 1 to string 2 the same way strcasecmp() does.
1759 result = s1StrRef.compare_lower(s2StrRef);
1761 // Compare string 1 to string 2 the same way strcmp() does.
1762 result = s1StrRef.compare(s2StrRef);
1765 // Build the SVal of the comparison and bind the return value.
1766 SVal resultVal = svalBuilder.makeIntVal(result, CE->getType());
1767 state = state->BindExpr(CE, LCtx, resultVal);
1771 if (!canComputeResult) {
1772 // Conjure a symbolic value. It's the best we can do.
1773 unsigned Count = C.getCurrentBlockCount();
1774 SVal resultVal = svalBuilder.getConjuredSymbolVal(NULL, CE, LCtx, Count);
1775 state = state->BindExpr(CE, LCtx, resultVal);
1778 // Record this as a possible path.
1779 C.addTransition(state);
1782 //===----------------------------------------------------------------------===//
1783 // The driver method, and other Checker callbacks.
1784 //===----------------------------------------------------------------------===//
1786 bool CStringChecker::evalCall(const CallExpr *CE, CheckerContext &C) const {
1787 const FunctionDecl *FDecl = C.getCalleeDecl(CE);
1792 FnCheck evalFunction = 0;
1793 if (C.isCLibraryFunction(FDecl, "memcpy"))
1794 evalFunction = &CStringChecker::evalMemcpy;
1795 else if (C.isCLibraryFunction(FDecl, "mempcpy"))
1796 evalFunction = &CStringChecker::evalMempcpy;
1797 else if (C.isCLibraryFunction(FDecl, "memcmp"))
1798 evalFunction = &CStringChecker::evalMemcmp;
1799 else if (C.isCLibraryFunction(FDecl, "memmove"))
1800 evalFunction = &CStringChecker::evalMemmove;
1801 else if (C.isCLibraryFunction(FDecl, "strcpy"))
1802 evalFunction = &CStringChecker::evalStrcpy;
1803 else if (C.isCLibraryFunction(FDecl, "strncpy"))
1804 evalFunction = &CStringChecker::evalStrncpy;
1805 else if (C.isCLibraryFunction(FDecl, "stpcpy"))
1806 evalFunction = &CStringChecker::evalStpcpy;
1807 else if (C.isCLibraryFunction(FDecl, "strcat"))
1808 evalFunction = &CStringChecker::evalStrcat;
1809 else if (C.isCLibraryFunction(FDecl, "strncat"))
1810 evalFunction = &CStringChecker::evalStrncat;
1811 else if (C.isCLibraryFunction(FDecl, "strlen"))
1812 evalFunction = &CStringChecker::evalstrLength;
1813 else if (C.isCLibraryFunction(FDecl, "strnlen"))
1814 evalFunction = &CStringChecker::evalstrnLength;
1815 else if (C.isCLibraryFunction(FDecl, "strcmp"))
1816 evalFunction = &CStringChecker::evalStrcmp;
1817 else if (C.isCLibraryFunction(FDecl, "strncmp"))
1818 evalFunction = &CStringChecker::evalStrncmp;
1819 else if (C.isCLibraryFunction(FDecl, "strcasecmp"))
1820 evalFunction = &CStringChecker::evalStrcasecmp;
1821 else if (C.isCLibraryFunction(FDecl, "strncasecmp"))
1822 evalFunction = &CStringChecker::evalStrncasecmp;
1823 else if (C.isCLibraryFunction(FDecl, "bcopy"))
1824 evalFunction = &CStringChecker::evalBcopy;
1825 else if (C.isCLibraryFunction(FDecl, "bcmp"))
1826 evalFunction = &CStringChecker::evalMemcmp;
1828 // If the callee isn't a string function, let another checker handle it.
1832 // Make sure each function sets its own description.
1833 // (But don't bother in a release build.)
1834 assert(!(CurrentFunctionDescription = NULL));
1836 // Check and evaluate the call.
1837 (this->*evalFunction)(C, CE);
1839 // If the evaluate call resulted in no change, chain to the next eval call
1841 // Note, the custom CString evaluation calls assume that basic safety
1842 // properties are held. However, if the user chooses to turn off some of these
1843 // checks, we ignore the issues and leave the call evaluation to a generic
1845 if (!C.isDifferent())
1851 void CStringChecker::checkPreStmt(const DeclStmt *DS, CheckerContext &C) const {
1852 // Record string length for char a[] = "abc";
1853 ProgramStateRef state = C.getState();
1855 for (DeclStmt::const_decl_iterator I = DS->decl_begin(), E = DS->decl_end();
1857 const VarDecl *D = dyn_cast<VarDecl>(*I);
1861 // FIXME: Handle array fields of structs.
1862 if (!D->getType()->isArrayType())
1865 const Expr *Init = D->getInit();
1868 if (!isa<StringLiteral>(Init))
1871 Loc VarLoc = state->getLValue(D, C.getLocationContext());
1872 const MemRegion *MR = VarLoc.getAsRegion();
1876 SVal StrVal = state->getSVal(Init, C.getLocationContext());
1877 assert(StrVal.isValid() && "Initializer string is unknown or undefined");
1878 DefinedOrUnknownSVal strLength
1879 = cast<DefinedOrUnknownSVal>(getCStringLength(C, state, Init, StrVal));
1881 state = state->set<CStringLength>(MR, strLength);
1884 C.addTransition(state);
1887 bool CStringChecker::wantsRegionChangeUpdate(ProgramStateRef state) const {
1888 CStringLength::EntryMap Entries = state->get<CStringLength>();
1889 return !Entries.isEmpty();
1893 CStringChecker::checkRegionChanges(ProgramStateRef state,
1894 const StoreManager::InvalidatedSymbols *,
1895 ArrayRef<const MemRegion *> ExplicitRegions,
1896 ArrayRef<const MemRegion *> Regions,
1897 const CallEvent *Call) const {
1898 CStringLength::EntryMap Entries = state->get<CStringLength>();
1899 if (Entries.isEmpty())
1902 llvm::SmallPtrSet<const MemRegion *, 8> Invalidated;
1903 llvm::SmallPtrSet<const MemRegion *, 32> SuperRegions;
1905 // First build sets for the changed regions and their super-regions.
1906 for (ArrayRef<const MemRegion *>::iterator
1907 I = Regions.begin(), E = Regions.end(); I != E; ++I) {
1908 const MemRegion *MR = *I;
1909 Invalidated.insert(MR);
1911 SuperRegions.insert(MR);
1912 while (const SubRegion *SR = dyn_cast<SubRegion>(MR)) {
1913 MR = SR->getSuperRegion();
1914 SuperRegions.insert(MR);
1918 CStringLength::EntryMap::Factory &F = state->get_context<CStringLength>();
1920 // Then loop over the entries in the current state.
1921 for (CStringLength::EntryMap::iterator I = Entries.begin(),
1922 E = Entries.end(); I != E; ++I) {
1923 const MemRegion *MR = I.getKey();
1925 // Is this entry for a super-region of a changed region?
1926 if (SuperRegions.count(MR)) {
1927 Entries = F.remove(Entries, MR);
1931 // Is this entry for a sub-region of a changed region?
1932 const MemRegion *Super = MR;
1933 while (const SubRegion *SR = dyn_cast<SubRegion>(Super)) {
1934 Super = SR->getSuperRegion();
1935 if (Invalidated.count(Super)) {
1936 Entries = F.remove(Entries, MR);
1942 return state->set<CStringLength>(Entries);
1945 void CStringChecker::checkLiveSymbols(ProgramStateRef state,
1946 SymbolReaper &SR) const {
1947 // Mark all symbols in our string length map as valid.
1948 CStringLength::EntryMap Entries = state->get<CStringLength>();
1950 for (CStringLength::EntryMap::iterator I = Entries.begin(), E = Entries.end();
1952 SVal Len = I.getData();
1954 for (SymExpr::symbol_iterator si = Len.symbol_begin(),
1955 se = Len.symbol_end(); si != se; ++si)
1960 void CStringChecker::checkDeadSymbols(SymbolReaper &SR,
1961 CheckerContext &C) const {
1962 if (!SR.hasDeadSymbols())
1965 ProgramStateRef state = C.getState();
1966 CStringLength::EntryMap Entries = state->get<CStringLength>();
1967 if (Entries.isEmpty())
1970 CStringLength::EntryMap::Factory &F = state->get_context<CStringLength>();
1971 for (CStringLength::EntryMap::iterator I = Entries.begin(), E = Entries.end();
1973 SVal Len = I.getData();
1974 if (SymbolRef Sym = Len.getAsSymbol()) {
1976 Entries = F.remove(Entries, I.getKey());
1980 state = state->set<CStringLength>(Entries);
1981 C.addTransition(state);
1984 #define REGISTER_CHECKER(name) \
1985 void ento::register##name(CheckerManager &mgr) {\
1986 static CStringChecker *TheChecker = 0; \
1987 if (TheChecker == 0) \
1988 TheChecker = mgr.registerChecker<CStringChecker>(); \
1989 TheChecker->Filter.Check##name = true; \
1992 REGISTER_CHECKER(CStringNullArg)
1993 REGISTER_CHECKER(CStringOutOfBounds)
1994 REGISTER_CHECKER(CStringBufferOverlap)
1995 REGISTER_CHECKER(CStringNotNullTerm)
1997 void ento::registerCStringCheckerBasic(CheckerManager &Mgr) {
1998 registerCStringNullArg(Mgr);