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/Basic/CharInfo.h"
18 #include "clang/StaticAnalyzer/Core/BugReporter/BugType.h"
19 #include "clang/StaticAnalyzer/Core/Checker.h"
20 #include "clang/StaticAnalyzer/Core/CheckerManager.h"
21 #include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h"
22 #include "clang/StaticAnalyzer/Core/PathSensitive/ProgramStateTrait.h"
23 #include "llvm/ADT/STLExtras.h"
24 #include "llvm/ADT/SmallString.h"
25 #include "llvm/Support/raw_ostream.h"
27 using namespace clang;
31 class CStringChecker : public Checker< eval::Call,
32 check::PreStmt<DeclStmt>,
37 mutable std::unique_ptr<BugType> BT_Null, BT_Bounds, BT_Overlap,
38 BT_NotCString, BT_AdditionOverflow;
40 mutable const char *CurrentFunctionDescription;
43 /// The filter is used to filter out the diagnostics which are not enabled by
45 struct CStringChecksFilter {
46 DefaultBool CheckCStringNullArg;
47 DefaultBool CheckCStringOutOfBounds;
48 DefaultBool CheckCStringBufferOverlap;
49 DefaultBool CheckCStringNotNullTerm;
51 CheckName CheckNameCStringNullArg;
52 CheckName CheckNameCStringOutOfBounds;
53 CheckName CheckNameCStringBufferOverlap;
54 CheckName CheckNameCStringNotNullTerm;
57 CStringChecksFilter Filter;
59 static void *getTag() { static int tag; return &tag; }
61 bool evalCall(const CallExpr *CE, CheckerContext &C) const;
62 void checkPreStmt(const DeclStmt *DS, CheckerContext &C) const;
63 void checkLiveSymbols(ProgramStateRef state, SymbolReaper &SR) const;
64 void checkDeadSymbols(SymbolReaper &SR, CheckerContext &C) const;
67 checkRegionChanges(ProgramStateRef state,
68 const InvalidatedSymbols *,
69 ArrayRef<const MemRegion *> ExplicitRegions,
70 ArrayRef<const MemRegion *> Regions,
71 const LocationContext *LCtx,
72 const CallEvent *Call) const;
74 typedef void (CStringChecker::*FnCheck)(CheckerContext &,
75 const CallExpr *) const;
77 void evalMemcpy(CheckerContext &C, const CallExpr *CE) const;
78 void evalMempcpy(CheckerContext &C, const CallExpr *CE) const;
79 void evalMemmove(CheckerContext &C, const CallExpr *CE) const;
80 void evalBcopy(CheckerContext &C, const CallExpr *CE) const;
81 void evalCopyCommon(CheckerContext &C, const CallExpr *CE,
82 ProgramStateRef state,
86 bool Restricted = false,
87 bool IsMempcpy = false) const;
89 void evalMemcmp(CheckerContext &C, const CallExpr *CE) const;
91 void evalstrLength(CheckerContext &C, const CallExpr *CE) const;
92 void evalstrnLength(CheckerContext &C, const CallExpr *CE) const;
93 void evalstrLengthCommon(CheckerContext &C,
95 bool IsStrnlen = false) const;
97 void evalStrcpy(CheckerContext &C, const CallExpr *CE) const;
98 void evalStrncpy(CheckerContext &C, const CallExpr *CE) const;
99 void evalStpcpy(CheckerContext &C, const CallExpr *CE) const;
100 void evalStrcpyCommon(CheckerContext &C,
104 bool isAppending) const;
106 void evalStrcat(CheckerContext &C, const CallExpr *CE) const;
107 void evalStrncat(CheckerContext &C, const CallExpr *CE) const;
109 void evalStrcmp(CheckerContext &C, const CallExpr *CE) const;
110 void evalStrncmp(CheckerContext &C, const CallExpr *CE) const;
111 void evalStrcasecmp(CheckerContext &C, const CallExpr *CE) const;
112 void evalStrncasecmp(CheckerContext &C, const CallExpr *CE) const;
113 void evalStrcmpCommon(CheckerContext &C,
115 bool isBounded = false,
116 bool ignoreCase = false) const;
118 void evalStrsep(CheckerContext &C, const CallExpr *CE) const;
120 void evalStdCopy(CheckerContext &C, const CallExpr *CE) const;
121 void evalStdCopyBackward(CheckerContext &C, const CallExpr *CE) const;
122 void evalStdCopyCommon(CheckerContext &C, const CallExpr *CE) const;
123 void evalMemset(CheckerContext &C, const CallExpr *CE) const;
126 std::pair<ProgramStateRef , ProgramStateRef >
127 static assumeZero(CheckerContext &C,
128 ProgramStateRef state, SVal V, QualType Ty);
130 static ProgramStateRef setCStringLength(ProgramStateRef state,
133 static SVal getCStringLengthForRegion(CheckerContext &C,
134 ProgramStateRef &state,
138 SVal getCStringLength(CheckerContext &C,
139 ProgramStateRef &state,
142 bool hypothetical = false) const;
144 const StringLiteral *getCStringLiteral(CheckerContext &C,
145 ProgramStateRef &state,
149 static ProgramStateRef InvalidateBuffer(CheckerContext &C,
150 ProgramStateRef state,
151 const Expr *Ex, SVal V,
155 static bool SummarizeRegion(raw_ostream &os, ASTContext &Ctx,
156 const MemRegion *MR);
159 ProgramStateRef checkNonNull(CheckerContext &C,
160 ProgramStateRef state,
163 ProgramStateRef CheckLocation(CheckerContext &C,
164 ProgramStateRef state,
167 const char *message = nullptr) const;
168 ProgramStateRef CheckBufferAccess(CheckerContext &C,
169 ProgramStateRef state,
171 const Expr *FirstBuf,
172 const Expr *SecondBuf,
173 const char *firstMessage = nullptr,
174 const char *secondMessage = nullptr,
175 bool WarnAboutSize = false) const;
177 ProgramStateRef CheckBufferAccess(CheckerContext &C,
178 ProgramStateRef state,
181 const char *message = nullptr,
182 bool WarnAboutSize = false) const {
183 // This is a convenience override.
184 return CheckBufferAccess(C, state, Size, Buf, nullptr, message, nullptr,
187 ProgramStateRef CheckOverlap(CheckerContext &C,
188 ProgramStateRef state,
191 const Expr *Second) const;
192 void emitOverlapBug(CheckerContext &C,
193 ProgramStateRef state,
195 const Stmt *Second) const;
197 ProgramStateRef checkAdditionOverflow(CheckerContext &C,
198 ProgramStateRef state,
202 // Return true if the destination buffer of the copy function may be in bound.
203 // Expects SVal of Size to be positive and unsigned.
204 // Expects SVal of FirstBuf to be a FieldRegion.
205 static bool IsFirstBufInBound(CheckerContext &C,
206 ProgramStateRef state,
207 const Expr *FirstBuf,
211 } //end anonymous namespace
213 REGISTER_MAP_WITH_PROGRAMSTATE(CStringLength, const MemRegion *, SVal)
215 //===----------------------------------------------------------------------===//
216 // Individual checks and utility methods.
217 //===----------------------------------------------------------------------===//
219 std::pair<ProgramStateRef , ProgramStateRef >
220 CStringChecker::assumeZero(CheckerContext &C, ProgramStateRef state, SVal V,
222 Optional<DefinedSVal> val = V.getAs<DefinedSVal>();
224 return std::pair<ProgramStateRef , ProgramStateRef >(state, state);
226 SValBuilder &svalBuilder = C.getSValBuilder();
227 DefinedOrUnknownSVal zero = svalBuilder.makeZeroVal(Ty);
228 return state->assume(svalBuilder.evalEQ(state, *val, zero));
231 ProgramStateRef CStringChecker::checkNonNull(CheckerContext &C,
232 ProgramStateRef state,
233 const Expr *S, SVal l) const {
234 // If a previous check has failed, propagate the failure.
238 ProgramStateRef stateNull, stateNonNull;
239 std::tie(stateNull, stateNonNull) = assumeZero(C, state, l, S->getType());
241 if (stateNull && !stateNonNull) {
242 if (!Filter.CheckCStringNullArg)
245 ExplodedNode *N = C.generateErrorNode(stateNull);
250 BT_Null.reset(new BuiltinBug(
251 Filter.CheckNameCStringNullArg, categories::UnixAPI,
252 "Null pointer argument in call to byte string function"));
255 llvm::raw_svector_ostream os(buf);
256 assert(CurrentFunctionDescription);
257 os << "Null pointer argument in call to " << CurrentFunctionDescription;
259 // Generate a report for this bug.
260 BuiltinBug *BT = static_cast<BuiltinBug*>(BT_Null.get());
261 auto report = llvm::make_unique<BugReport>(*BT, os.str(), N);
263 report->addRange(S->getSourceRange());
264 bugreporter::trackNullOrUndefValue(N, S, *report);
265 C.emitReport(std::move(report));
269 // From here on, assume that the value is non-null.
270 assert(stateNonNull);
274 // FIXME: This was originally copied from ArrayBoundChecker.cpp. Refactor?
275 ProgramStateRef CStringChecker::CheckLocation(CheckerContext &C,
276 ProgramStateRef state,
277 const Expr *S, SVal l,
278 const char *warningMsg) const {
279 // If a previous check has failed, propagate the failure.
283 // Check for out of bound array element access.
284 const MemRegion *R = l.getAsRegion();
288 const ElementRegion *ER = dyn_cast<ElementRegion>(R);
292 assert(ER->getValueType() == C.getASTContext().CharTy &&
293 "CheckLocation should only be called with char* ElementRegions");
295 // Get the size of the array.
296 const SubRegion *superReg = cast<SubRegion>(ER->getSuperRegion());
297 SValBuilder &svalBuilder = C.getSValBuilder();
299 svalBuilder.convertToArrayIndex(superReg->getExtent(svalBuilder));
300 DefinedOrUnknownSVal Size = Extent.castAs<DefinedOrUnknownSVal>();
302 // Get the index of the accessed element.
303 DefinedOrUnknownSVal Idx = ER->getIndex().castAs<DefinedOrUnknownSVal>();
305 ProgramStateRef StInBound = state->assumeInBound(Idx, Size, true);
306 ProgramStateRef StOutBound = state->assumeInBound(Idx, Size, false);
307 if (StOutBound && !StInBound) {
308 ExplodedNode *N = C.generateErrorNode(StOutBound);
313 BT_Bounds.reset(new BuiltinBug(
314 Filter.CheckNameCStringOutOfBounds, "Out-of-bound array access",
315 "Byte string function accesses out-of-bound array element"));
317 BuiltinBug *BT = static_cast<BuiltinBug*>(BT_Bounds.get());
319 // Generate a report for this bug.
320 std::unique_ptr<BugReport> report;
322 report = llvm::make_unique<BugReport>(*BT, warningMsg, N);
324 assert(CurrentFunctionDescription);
325 assert(CurrentFunctionDescription[0] != '\0');
328 llvm::raw_svector_ostream os(buf);
329 os << toUppercase(CurrentFunctionDescription[0])
330 << &CurrentFunctionDescription[1]
331 << " accesses out-of-bound array element";
332 report = llvm::make_unique<BugReport>(*BT, os.str(), N);
335 // FIXME: It would be nice to eventually make this diagnostic more clear,
336 // e.g., by referencing the original declaration or by saying *why* this
337 // reference is outside the range.
339 report->addRange(S->getSourceRange());
340 C.emitReport(std::move(report));
344 // Array bound check succeeded. From this point forward the array bound
345 // should always succeed.
349 ProgramStateRef CStringChecker::CheckBufferAccess(CheckerContext &C,
350 ProgramStateRef state,
352 const Expr *FirstBuf,
353 const Expr *SecondBuf,
354 const char *firstMessage,
355 const char *secondMessage,
356 bool WarnAboutSize) const {
357 // If a previous check has failed, propagate the failure.
361 SValBuilder &svalBuilder = C.getSValBuilder();
362 ASTContext &Ctx = svalBuilder.getContext();
363 const LocationContext *LCtx = C.getLocationContext();
365 QualType sizeTy = Size->getType();
366 QualType PtrTy = Ctx.getPointerType(Ctx.CharTy);
368 // Check that the first buffer is non-null.
369 SVal BufVal = state->getSVal(FirstBuf, LCtx);
370 state = checkNonNull(C, state, FirstBuf, BufVal);
374 // If out-of-bounds checking is turned off, skip the rest.
375 if (!Filter.CheckCStringOutOfBounds)
378 // Get the access length and make sure it is known.
379 // FIXME: This assumes the caller has already checked that the access length
380 // is positive. And that it's unsigned.
381 SVal LengthVal = state->getSVal(Size, LCtx);
382 Optional<NonLoc> Length = LengthVal.getAs<NonLoc>();
386 // Compute the offset of the last element to be accessed: size-1.
387 NonLoc One = svalBuilder.makeIntVal(1, sizeTy).castAs<NonLoc>();
388 NonLoc LastOffset = svalBuilder
389 .evalBinOpNN(state, BO_Sub, *Length, One, sizeTy).castAs<NonLoc>();
391 // Check that the first buffer is sufficiently long.
392 SVal BufStart = svalBuilder.evalCast(BufVal, PtrTy, FirstBuf->getType());
393 if (Optional<Loc> BufLoc = BufStart.getAs<Loc>()) {
394 const Expr *warningExpr = (WarnAboutSize ? Size : FirstBuf);
396 SVal BufEnd = svalBuilder.evalBinOpLN(state, BO_Add, *BufLoc,
398 state = CheckLocation(C, state, warningExpr, BufEnd, firstMessage);
400 // If the buffer isn't large enough, abort.
405 // If there's a second buffer, check it as well.
407 BufVal = state->getSVal(SecondBuf, LCtx);
408 state = checkNonNull(C, state, SecondBuf, BufVal);
412 BufStart = svalBuilder.evalCast(BufVal, PtrTy, SecondBuf->getType());
413 if (Optional<Loc> BufLoc = BufStart.getAs<Loc>()) {
414 const Expr *warningExpr = (WarnAboutSize ? Size : SecondBuf);
416 SVal BufEnd = svalBuilder.evalBinOpLN(state, BO_Add, *BufLoc,
418 state = CheckLocation(C, state, warningExpr, BufEnd, secondMessage);
422 // Large enough or not, return this state!
426 ProgramStateRef CStringChecker::CheckOverlap(CheckerContext &C,
427 ProgramStateRef state,
430 const Expr *Second) const {
431 if (!Filter.CheckCStringBufferOverlap)
434 // Do a simple check for overlap: if the two arguments are from the same
435 // buffer, see if the end of the first is greater than the start of the second
438 // If a previous check has failed, propagate the failure.
442 ProgramStateRef stateTrue, stateFalse;
444 // Get the buffer values and make sure they're known locations.
445 const LocationContext *LCtx = C.getLocationContext();
446 SVal firstVal = state->getSVal(First, LCtx);
447 SVal secondVal = state->getSVal(Second, LCtx);
449 Optional<Loc> firstLoc = firstVal.getAs<Loc>();
453 Optional<Loc> secondLoc = secondVal.getAs<Loc>();
457 // Are the two values the same?
458 SValBuilder &svalBuilder = C.getSValBuilder();
459 std::tie(stateTrue, stateFalse) =
460 state->assume(svalBuilder.evalEQ(state, *firstLoc, *secondLoc));
462 if (stateTrue && !stateFalse) {
463 // If the values are known to be equal, that's automatically an overlap.
464 emitOverlapBug(C, stateTrue, First, Second);
468 // assume the two expressions are not equal.
472 // Which value comes first?
473 QualType cmpTy = svalBuilder.getConditionType();
474 SVal reverse = svalBuilder.evalBinOpLL(state, BO_GT,
475 *firstLoc, *secondLoc, cmpTy);
476 Optional<DefinedOrUnknownSVal> reverseTest =
477 reverse.getAs<DefinedOrUnknownSVal>();
481 std::tie(stateTrue, stateFalse) = state->assume(*reverseTest);
484 // If we don't know which one comes first, we can't perform this test.
487 // Switch the values so that firstVal is before secondVal.
488 std::swap(firstLoc, secondLoc);
490 // Switch the Exprs as well, so that they still correspond.
491 std::swap(First, Second);
495 // Get the length, and make sure it too is known.
496 SVal LengthVal = state->getSVal(Size, LCtx);
497 Optional<NonLoc> Length = LengthVal.getAs<NonLoc>();
501 // Convert the first buffer's start address to char*.
502 // Bail out if the cast fails.
503 ASTContext &Ctx = svalBuilder.getContext();
504 QualType CharPtrTy = Ctx.getPointerType(Ctx.CharTy);
505 SVal FirstStart = svalBuilder.evalCast(*firstLoc, CharPtrTy,
507 Optional<Loc> FirstStartLoc = FirstStart.getAs<Loc>();
511 // Compute the end of the first buffer. Bail out if THAT fails.
512 SVal FirstEnd = svalBuilder.evalBinOpLN(state, BO_Add,
513 *FirstStartLoc, *Length, CharPtrTy);
514 Optional<Loc> FirstEndLoc = FirstEnd.getAs<Loc>();
518 // Is the end of the first buffer past the start of the second buffer?
519 SVal Overlap = svalBuilder.evalBinOpLL(state, BO_GT,
520 *FirstEndLoc, *secondLoc, cmpTy);
521 Optional<DefinedOrUnknownSVal> OverlapTest =
522 Overlap.getAs<DefinedOrUnknownSVal>();
526 std::tie(stateTrue, stateFalse) = state->assume(*OverlapTest);
528 if (stateTrue && !stateFalse) {
530 emitOverlapBug(C, stateTrue, First, Second);
534 // assume the two expressions don't overlap.
539 void CStringChecker::emitOverlapBug(CheckerContext &C, ProgramStateRef state,
540 const Stmt *First, const Stmt *Second) const {
541 ExplodedNode *N = C.generateErrorNode(state);
546 BT_Overlap.reset(new BugType(Filter.CheckNameCStringBufferOverlap,
547 categories::UnixAPI, "Improper arguments"));
549 // Generate a report for this bug.
550 auto report = llvm::make_unique<BugReport>(
551 *BT_Overlap, "Arguments must not be overlapping buffers", N);
552 report->addRange(First->getSourceRange());
553 report->addRange(Second->getSourceRange());
555 C.emitReport(std::move(report));
558 ProgramStateRef CStringChecker::checkAdditionOverflow(CheckerContext &C,
559 ProgramStateRef state,
561 NonLoc right) const {
562 // If out-of-bounds checking is turned off, skip the rest.
563 if (!Filter.CheckCStringOutOfBounds)
566 // If a previous check has failed, propagate the failure.
570 SValBuilder &svalBuilder = C.getSValBuilder();
571 BasicValueFactory &BVF = svalBuilder.getBasicValueFactory();
573 QualType sizeTy = svalBuilder.getContext().getSizeType();
574 const llvm::APSInt &maxValInt = BVF.getMaxValue(sizeTy);
575 NonLoc maxVal = svalBuilder.makeIntVal(maxValInt);
578 if (right.getAs<nonloc::ConcreteInt>()) {
579 maxMinusRight = svalBuilder.evalBinOpNN(state, BO_Sub, maxVal, right,
582 // Try switching the operands. (The order of these two assignments is
584 maxMinusRight = svalBuilder.evalBinOpNN(state, BO_Sub, maxVal, left,
589 if (Optional<NonLoc> maxMinusRightNL = maxMinusRight.getAs<NonLoc>()) {
590 QualType cmpTy = svalBuilder.getConditionType();
591 // If left > max - right, we have an overflow.
592 SVal willOverflow = svalBuilder.evalBinOpNN(state, BO_GT, left,
593 *maxMinusRightNL, cmpTy);
595 ProgramStateRef stateOverflow, stateOkay;
596 std::tie(stateOverflow, stateOkay) =
597 state->assume(willOverflow.castAs<DefinedOrUnknownSVal>());
599 if (stateOverflow && !stateOkay) {
600 // We have an overflow. Emit a bug report.
601 ExplodedNode *N = C.generateErrorNode(stateOverflow);
605 if (!BT_AdditionOverflow)
606 BT_AdditionOverflow.reset(
607 new BuiltinBug(Filter.CheckNameCStringOutOfBounds, "API",
608 "Sum of expressions causes overflow"));
610 // This isn't a great error message, but this should never occur in real
611 // code anyway -- you'd have to create a buffer longer than a size_t can
612 // represent, which is sort of a contradiction.
613 const char *warning =
614 "This expression will create a string whose length is too big to "
615 "be represented as a size_t";
617 // Generate a report for this bug.
619 llvm::make_unique<BugReport>(*BT_AdditionOverflow, warning, N));
624 // From now on, assume an overflow didn't occur.
632 ProgramStateRef CStringChecker::setCStringLength(ProgramStateRef state,
635 assert(!strLength.isUndef() && "Attempt to set an undefined string length");
637 MR = MR->StripCasts();
639 switch (MR->getKind()) {
640 case MemRegion::StringRegionKind:
641 // FIXME: This can happen if we strcpy() into a string region. This is
642 // undefined [C99 6.4.5p6], but we should still warn about it.
645 case MemRegion::SymbolicRegionKind:
646 case MemRegion::AllocaRegionKind:
647 case MemRegion::VarRegionKind:
648 case MemRegion::FieldRegionKind:
649 case MemRegion::ObjCIvarRegionKind:
650 // These are the types we can currently track string lengths for.
653 case MemRegion::ElementRegionKind:
654 // FIXME: Handle element regions by upper-bounding the parent region's
659 // Other regions (mostly non-data) can't have a reliable C string length.
660 // For now, just ignore the change.
661 // FIXME: These are rare but not impossible. We should output some kind of
662 // warning for things like strcpy((char[]){'a', 0}, "b");
666 if (strLength.isUnknown())
667 return state->remove<CStringLength>(MR);
669 return state->set<CStringLength>(MR, strLength);
672 SVal CStringChecker::getCStringLengthForRegion(CheckerContext &C,
673 ProgramStateRef &state,
678 // If there's a recorded length, go ahead and return it.
679 const SVal *Recorded = state->get<CStringLength>(MR);
684 // Otherwise, get a new symbol and update the state.
685 SValBuilder &svalBuilder = C.getSValBuilder();
686 QualType sizeTy = svalBuilder.getContext().getSizeType();
687 SVal strLength = svalBuilder.getMetadataSymbolVal(CStringChecker::getTag(),
689 C.getLocationContext(),
693 if (Optional<NonLoc> strLn = strLength.getAs<NonLoc>()) {
694 // In case of unbounded calls strlen etc bound the range to SIZE_MAX/4
695 BasicValueFactory &BVF = svalBuilder.getBasicValueFactory();
696 const llvm::APSInt &maxValInt = BVF.getMaxValue(sizeTy);
697 llvm::APSInt fourInt = APSIntType(maxValInt).getValue(4);
698 const llvm::APSInt *maxLengthInt = BVF.evalAPSInt(BO_Div, maxValInt,
700 NonLoc maxLength = svalBuilder.makeIntVal(*maxLengthInt);
701 SVal evalLength = svalBuilder.evalBinOpNN(state, BO_LE, *strLn,
703 state = state->assume(evalLength.castAs<DefinedOrUnknownSVal>(), true);
705 state = state->set<CStringLength>(MR, strLength);
711 SVal CStringChecker::getCStringLength(CheckerContext &C, ProgramStateRef &state,
712 const Expr *Ex, SVal Buf,
713 bool hypothetical) const {
714 const MemRegion *MR = Buf.getAsRegion();
716 // If we can't get a region, see if it's something we /know/ isn't a
717 // C string. In the context of locations, the only time we can issue such
718 // a warning is for labels.
719 if (Optional<loc::GotoLabel> Label = Buf.getAs<loc::GotoLabel>()) {
720 if (!Filter.CheckCStringNotNullTerm)
721 return UndefinedVal();
723 if (ExplodedNode *N = C.generateNonFatalErrorNode(state)) {
725 BT_NotCString.reset(new BuiltinBug(
726 Filter.CheckNameCStringNotNullTerm, categories::UnixAPI,
727 "Argument is not a null-terminated string."));
729 SmallString<120> buf;
730 llvm::raw_svector_ostream os(buf);
731 assert(CurrentFunctionDescription);
732 os << "Argument to " << CurrentFunctionDescription
733 << " is the address of the label '" << Label->getLabel()->getName()
734 << "', which is not a null-terminated string";
736 // Generate a report for this bug.
737 auto report = llvm::make_unique<BugReport>(*BT_NotCString, os.str(), N);
739 report->addRange(Ex->getSourceRange());
740 C.emitReport(std::move(report));
742 return UndefinedVal();
746 // If it's not a region and not a label, give up.
750 // If we have a region, strip casts from it and see if we can figure out
751 // its length. For anything we can't figure out, just return UnknownVal.
752 MR = MR->StripCasts();
754 switch (MR->getKind()) {
755 case MemRegion::StringRegionKind: {
756 // Modifying the contents of string regions is undefined [C99 6.4.5p6],
757 // so we can assume that the byte length is the correct C string length.
758 SValBuilder &svalBuilder = C.getSValBuilder();
759 QualType sizeTy = svalBuilder.getContext().getSizeType();
760 const StringLiteral *strLit = cast<StringRegion>(MR)->getStringLiteral();
761 return svalBuilder.makeIntVal(strLit->getByteLength(), sizeTy);
763 case MemRegion::SymbolicRegionKind:
764 case MemRegion::AllocaRegionKind:
765 case MemRegion::VarRegionKind:
766 case MemRegion::FieldRegionKind:
767 case MemRegion::ObjCIvarRegionKind:
768 return getCStringLengthForRegion(C, state, Ex, MR, hypothetical);
769 case MemRegion::CompoundLiteralRegionKind:
770 // FIXME: Can we track this? Is it necessary?
772 case MemRegion::ElementRegionKind:
773 // FIXME: How can we handle this? It's not good enough to subtract the
774 // offset from the base string length; consider "123\x00567" and &a[5].
777 // Other regions (mostly non-data) can't have a reliable C string length.
778 // In this case, an error is emitted and UndefinedVal is returned.
779 // The caller should always be prepared to handle this case.
780 if (!Filter.CheckCStringNotNullTerm)
781 return UndefinedVal();
783 if (ExplodedNode *N = C.generateNonFatalErrorNode(state)) {
785 BT_NotCString.reset(new BuiltinBug(
786 Filter.CheckNameCStringNotNullTerm, categories::UnixAPI,
787 "Argument is not a null-terminated string."));
789 SmallString<120> buf;
790 llvm::raw_svector_ostream os(buf);
792 assert(CurrentFunctionDescription);
793 os << "Argument to " << CurrentFunctionDescription << " is ";
795 if (SummarizeRegion(os, C.getASTContext(), MR))
796 os << ", which is not a null-terminated string";
798 os << "not a null-terminated string";
800 // Generate a report for this bug.
801 auto report = llvm::make_unique<BugReport>(*BT_NotCString, os.str(), N);
803 report->addRange(Ex->getSourceRange());
804 C.emitReport(std::move(report));
807 return UndefinedVal();
811 const StringLiteral *CStringChecker::getCStringLiteral(CheckerContext &C,
812 ProgramStateRef &state, const Expr *expr, SVal val) const {
814 // Get the memory region pointed to by the val.
815 const MemRegion *bufRegion = val.getAsRegion();
819 // Strip casts off the memory region.
820 bufRegion = bufRegion->StripCasts();
822 // Cast the memory region to a string region.
823 const StringRegion *strRegion= dyn_cast<StringRegion>(bufRegion);
827 // Return the actual string in the string region.
828 return strRegion->getStringLiteral();
831 bool CStringChecker::IsFirstBufInBound(CheckerContext &C,
832 ProgramStateRef state,
833 const Expr *FirstBuf,
835 // If we do not know that the buffer is long enough we return 'true'.
836 // Otherwise the parent region of this field region would also get
837 // invalidated, which would lead to warnings based on an unknown state.
839 // Originally copied from CheckBufferAccess and CheckLocation.
840 SValBuilder &svalBuilder = C.getSValBuilder();
841 ASTContext &Ctx = svalBuilder.getContext();
842 const LocationContext *LCtx = C.getLocationContext();
844 QualType sizeTy = Size->getType();
845 QualType PtrTy = Ctx.getPointerType(Ctx.CharTy);
846 SVal BufVal = state->getSVal(FirstBuf, LCtx);
848 SVal LengthVal = state->getSVal(Size, LCtx);
849 Optional<NonLoc> Length = LengthVal.getAs<NonLoc>();
851 return true; // cf top comment.
853 // Compute the offset of the last element to be accessed: size-1.
854 NonLoc One = svalBuilder.makeIntVal(1, sizeTy).castAs<NonLoc>();
856 svalBuilder.evalBinOpNN(state, BO_Sub, *Length, One, sizeTy)
859 // Check that the first buffer is sufficiently long.
860 SVal BufStart = svalBuilder.evalCast(BufVal, PtrTy, FirstBuf->getType());
861 Optional<Loc> BufLoc = BufStart.getAs<Loc>();
863 return true; // cf top comment.
866 svalBuilder.evalBinOpLN(state, BO_Add, *BufLoc, LastOffset, PtrTy);
868 // Check for out of bound array element access.
869 const MemRegion *R = BufEnd.getAsRegion();
871 return true; // cf top comment.
873 const ElementRegion *ER = dyn_cast<ElementRegion>(R);
875 return true; // cf top comment.
877 assert(ER->getValueType() == C.getASTContext().CharTy &&
878 "IsFirstBufInBound should only be called with char* ElementRegions");
880 // Get the size of the array.
881 const SubRegion *superReg = cast<SubRegion>(ER->getSuperRegion());
883 svalBuilder.convertToArrayIndex(superReg->getExtent(svalBuilder));
884 DefinedOrUnknownSVal ExtentSize = Extent.castAs<DefinedOrUnknownSVal>();
886 // Get the index of the accessed element.
887 DefinedOrUnknownSVal Idx = ER->getIndex().castAs<DefinedOrUnknownSVal>();
889 ProgramStateRef StInBound = state->assumeInBound(Idx, ExtentSize, true);
891 return static_cast<bool>(StInBound);
894 ProgramStateRef CStringChecker::InvalidateBuffer(CheckerContext &C,
895 ProgramStateRef state,
896 const Expr *E, SVal V,
899 Optional<Loc> L = V.getAs<Loc>();
903 // FIXME: This is a simplified version of what's in CFRefCount.cpp -- it makes
904 // some assumptions about the value that CFRefCount can't. Even so, it should
905 // probably be refactored.
906 if (Optional<loc::MemRegionVal> MR = L->getAs<loc::MemRegionVal>()) {
907 const MemRegion *R = MR->getRegion()->StripCasts();
909 // Are we dealing with an ElementRegion? If so, we should be invalidating
911 if (const ElementRegion *ER = dyn_cast<ElementRegion>(R)) {
912 R = ER->getSuperRegion();
913 // FIXME: What about layers of ElementRegions?
916 // Invalidate this region.
917 const LocationContext *LCtx = C.getPredecessor()->getLocationContext();
919 bool CausesPointerEscape = false;
920 RegionAndSymbolInvalidationTraits ITraits;
921 // Invalidate and escape only indirect regions accessible through the source
923 if (IsSourceBuffer) {
924 ITraits.setTrait(R->getBaseRegion(),
925 RegionAndSymbolInvalidationTraits::TK_PreserveContents);
926 ITraits.setTrait(R, RegionAndSymbolInvalidationTraits::TK_SuppressEscape);
927 CausesPointerEscape = true;
929 const MemRegion::Kind& K = R->getKind();
930 if (K == MemRegion::FieldRegionKind)
931 if (Size && IsFirstBufInBound(C, state, E, Size)) {
932 // If destination buffer is a field region and access is in bound,
933 // do not invalidate its super region.
936 RegionAndSymbolInvalidationTraits::TK_DoNotInvalidateSuperRegion);
940 return state->invalidateRegions(R, E, C.blockCount(), LCtx,
941 CausesPointerEscape, nullptr, nullptr,
945 // If we have a non-region value by chance, just remove the binding.
946 // FIXME: is this necessary or correct? This handles the non-Region
947 // cases. Is it ever valid to store to these?
948 return state->killBinding(*L);
951 bool CStringChecker::SummarizeRegion(raw_ostream &os, ASTContext &Ctx,
952 const MemRegion *MR) {
953 const TypedValueRegion *TVR = dyn_cast<TypedValueRegion>(MR);
955 switch (MR->getKind()) {
956 case MemRegion::FunctionCodeRegionKind: {
957 const NamedDecl *FD = cast<FunctionCodeRegion>(MR)->getDecl();
959 os << "the address of the function '" << *FD << '\'';
961 os << "the address of a function";
964 case MemRegion::BlockCodeRegionKind:
967 case MemRegion::BlockDataRegionKind:
970 case MemRegion::CXXThisRegionKind:
971 case MemRegion::CXXTempObjectRegionKind:
972 os << "a C++ temp object of type " << TVR->getValueType().getAsString();
974 case MemRegion::VarRegionKind:
975 os << "a variable of type" << TVR->getValueType().getAsString();
977 case MemRegion::FieldRegionKind:
978 os << "a field of type " << TVR->getValueType().getAsString();
980 case MemRegion::ObjCIvarRegionKind:
981 os << "an instance variable of type " << TVR->getValueType().getAsString();
988 //===----------------------------------------------------------------------===//
989 // evaluation of individual function calls.
990 //===----------------------------------------------------------------------===//
992 void CStringChecker::evalCopyCommon(CheckerContext &C,
994 ProgramStateRef state,
995 const Expr *Size, const Expr *Dest,
996 const Expr *Source, bool Restricted,
997 bool IsMempcpy) const {
998 CurrentFunctionDescription = "memory copy function";
1000 // See if the size argument is zero.
1001 const LocationContext *LCtx = C.getLocationContext();
1002 SVal sizeVal = state->getSVal(Size, LCtx);
1003 QualType sizeTy = Size->getType();
1005 ProgramStateRef stateZeroSize, stateNonZeroSize;
1006 std::tie(stateZeroSize, stateNonZeroSize) =
1007 assumeZero(C, state, sizeVal, sizeTy);
1009 // Get the value of the Dest.
1010 SVal destVal = state->getSVal(Dest, LCtx);
1012 // If the size is zero, there won't be any actual memory access, so
1013 // just bind the return value to the destination buffer and return.
1014 if (stateZeroSize && !stateNonZeroSize) {
1015 stateZeroSize = stateZeroSize->BindExpr(CE, LCtx, destVal);
1016 C.addTransition(stateZeroSize);
1020 // If the size can be nonzero, we have to check the other arguments.
1021 if (stateNonZeroSize) {
1022 state = stateNonZeroSize;
1024 // Ensure the destination is not null. If it is NULL there will be a
1025 // NULL pointer dereference.
1026 state = checkNonNull(C, state, Dest, destVal);
1030 // Get the value of the Src.
1031 SVal srcVal = state->getSVal(Source, LCtx);
1033 // Ensure the source is not null. If it is NULL there will be a
1034 // NULL pointer dereference.
1035 state = checkNonNull(C, state, Source, srcVal);
1039 // Ensure the accesses are valid and that the buffers do not overlap.
1040 const char * const writeWarning =
1041 "Memory copy function overflows destination buffer";
1042 state = CheckBufferAccess(C, state, Size, Dest, Source,
1043 writeWarning, /* sourceWarning = */ nullptr);
1045 state = CheckOverlap(C, state, Size, Dest, Source);
1050 // If this is mempcpy, get the byte after the last byte copied and
1053 loc::MemRegionVal destRegVal = destVal.castAs<loc::MemRegionVal>();
1055 // Get the length to copy.
1056 if (Optional<NonLoc> lenValNonLoc = sizeVal.getAs<NonLoc>()) {
1057 // Get the byte after the last byte copied.
1058 SValBuilder &SvalBuilder = C.getSValBuilder();
1059 ASTContext &Ctx = SvalBuilder.getContext();
1060 QualType CharPtrTy = Ctx.getPointerType(Ctx.CharTy);
1061 loc::MemRegionVal DestRegCharVal = SvalBuilder.evalCast(destRegVal,
1062 CharPtrTy, Dest->getType()).castAs<loc::MemRegionVal>();
1063 SVal lastElement = C.getSValBuilder().evalBinOpLN(state, BO_Add,
1068 // The byte after the last byte copied is the return value.
1069 state = state->BindExpr(CE, LCtx, lastElement);
1071 // If we don't know how much we copied, we can at least
1072 // conjure a return value for later.
1073 SVal result = C.getSValBuilder().conjureSymbolVal(nullptr, CE, LCtx,
1075 state = state->BindExpr(CE, LCtx, result);
1079 // All other copies return the destination buffer.
1080 // (Well, bcopy() has a void return type, but this won't hurt.)
1081 state = state->BindExpr(CE, LCtx, destVal);
1084 // Invalidate the destination (regular invalidation without pointer-escaping
1085 // the address of the top-level region).
1086 // FIXME: Even if we can't perfectly model the copy, we should see if we
1087 // can use LazyCompoundVals to copy the source values into the destination.
1088 // This would probably remove any existing bindings past the end of the
1089 // copied region, but that's still an improvement over blank invalidation.
1090 state = InvalidateBuffer(C, state, Dest, C.getSVal(Dest),
1091 /*IsSourceBuffer*/false, Size);
1093 // Invalidate the source (const-invalidation without const-pointer-escaping
1094 // the address of the top-level region).
1095 state = InvalidateBuffer(C, state, Source, C.getSVal(Source),
1096 /*IsSourceBuffer*/true, nullptr);
1098 C.addTransition(state);
1103 void CStringChecker::evalMemcpy(CheckerContext &C, const CallExpr *CE) const {
1104 if (CE->getNumArgs() < 3)
1107 // void *memcpy(void *restrict dst, const void *restrict src, size_t n);
1108 // The return value is the address of the destination buffer.
1109 const Expr *Dest = CE->getArg(0);
1110 ProgramStateRef state = C.getState();
1112 evalCopyCommon(C, CE, state, CE->getArg(2), Dest, CE->getArg(1), true);
1115 void CStringChecker::evalMempcpy(CheckerContext &C, const CallExpr *CE) const {
1116 if (CE->getNumArgs() < 3)
1119 // void *mempcpy(void *restrict dst, const void *restrict src, size_t n);
1120 // The return value is a pointer to the byte following the last written byte.
1121 const Expr *Dest = CE->getArg(0);
1122 ProgramStateRef state = C.getState();
1124 evalCopyCommon(C, CE, state, CE->getArg(2), Dest, CE->getArg(1), true, true);
1127 void CStringChecker::evalMemmove(CheckerContext &C, const CallExpr *CE) const {
1128 if (CE->getNumArgs() < 3)
1131 // void *memmove(void *dst, const void *src, size_t n);
1132 // The return value is the address of the destination buffer.
1133 const Expr *Dest = CE->getArg(0);
1134 ProgramStateRef state = C.getState();
1136 evalCopyCommon(C, CE, state, CE->getArg(2), Dest, CE->getArg(1));
1139 void CStringChecker::evalBcopy(CheckerContext &C, const CallExpr *CE) const {
1140 if (CE->getNumArgs() < 3)
1143 // void bcopy(const void *src, void *dst, size_t n);
1144 evalCopyCommon(C, CE, C.getState(),
1145 CE->getArg(2), CE->getArg(1), CE->getArg(0));
1148 void CStringChecker::evalMemcmp(CheckerContext &C, const CallExpr *CE) const {
1149 if (CE->getNumArgs() < 3)
1152 // int memcmp(const void *s1, const void *s2, size_t n);
1153 CurrentFunctionDescription = "memory comparison function";
1155 const Expr *Left = CE->getArg(0);
1156 const Expr *Right = CE->getArg(1);
1157 const Expr *Size = CE->getArg(2);
1159 ProgramStateRef state = C.getState();
1160 SValBuilder &svalBuilder = C.getSValBuilder();
1162 // See if the size argument is zero.
1163 const LocationContext *LCtx = C.getLocationContext();
1164 SVal sizeVal = state->getSVal(Size, LCtx);
1165 QualType sizeTy = Size->getType();
1167 ProgramStateRef stateZeroSize, stateNonZeroSize;
1168 std::tie(stateZeroSize, stateNonZeroSize) =
1169 assumeZero(C, state, sizeVal, sizeTy);
1171 // If the size can be zero, the result will be 0 in that case, and we don't
1172 // have to check either of the buffers.
1173 if (stateZeroSize) {
1174 state = stateZeroSize;
1175 state = state->BindExpr(CE, LCtx,
1176 svalBuilder.makeZeroVal(CE->getType()));
1177 C.addTransition(state);
1180 // If the size can be nonzero, we have to check the other arguments.
1181 if (stateNonZeroSize) {
1182 state = stateNonZeroSize;
1183 // If we know the two buffers are the same, we know the result is 0.
1184 // First, get the two buffers' addresses. Another checker will have already
1185 // made sure they're not undefined.
1186 DefinedOrUnknownSVal LV =
1187 state->getSVal(Left, LCtx).castAs<DefinedOrUnknownSVal>();
1188 DefinedOrUnknownSVal RV =
1189 state->getSVal(Right, LCtx).castAs<DefinedOrUnknownSVal>();
1191 // See if they are the same.
1192 DefinedOrUnknownSVal SameBuf = svalBuilder.evalEQ(state, LV, RV);
1193 ProgramStateRef StSameBuf, StNotSameBuf;
1194 std::tie(StSameBuf, StNotSameBuf) = state->assume(SameBuf);
1196 // If the two arguments might be the same buffer, we know the result is 0,
1197 // and we only need to check one size.
1200 state = CheckBufferAccess(C, state, Size, Left);
1202 state = StSameBuf->BindExpr(CE, LCtx,
1203 svalBuilder.makeZeroVal(CE->getType()));
1204 C.addTransition(state);
1208 // If the two arguments might be different buffers, we have to check the
1209 // size of both of them.
1211 state = StNotSameBuf;
1212 state = CheckBufferAccess(C, state, Size, Left, Right);
1214 // The return value is the comparison result, which we don't know.
1215 SVal CmpV = svalBuilder.conjureSymbolVal(nullptr, CE, LCtx,
1217 state = state->BindExpr(CE, LCtx, CmpV);
1218 C.addTransition(state);
1224 void CStringChecker::evalstrLength(CheckerContext &C,
1225 const CallExpr *CE) const {
1226 if (CE->getNumArgs() < 1)
1229 // size_t strlen(const char *s);
1230 evalstrLengthCommon(C, CE, /* IsStrnlen = */ false);
1233 void CStringChecker::evalstrnLength(CheckerContext &C,
1234 const CallExpr *CE) const {
1235 if (CE->getNumArgs() < 2)
1238 // size_t strnlen(const char *s, size_t maxlen);
1239 evalstrLengthCommon(C, CE, /* IsStrnlen = */ true);
1242 void CStringChecker::evalstrLengthCommon(CheckerContext &C, const CallExpr *CE,
1243 bool IsStrnlen) const {
1244 CurrentFunctionDescription = "string length function";
1245 ProgramStateRef state = C.getState();
1246 const LocationContext *LCtx = C.getLocationContext();
1249 const Expr *maxlenExpr = CE->getArg(1);
1250 SVal maxlenVal = state->getSVal(maxlenExpr, LCtx);
1252 ProgramStateRef stateZeroSize, stateNonZeroSize;
1253 std::tie(stateZeroSize, stateNonZeroSize) =
1254 assumeZero(C, state, maxlenVal, maxlenExpr->getType());
1256 // If the size can be zero, the result will be 0 in that case, and we don't
1257 // have to check the string itself.
1258 if (stateZeroSize) {
1259 SVal zero = C.getSValBuilder().makeZeroVal(CE->getType());
1260 stateZeroSize = stateZeroSize->BindExpr(CE, LCtx, zero);
1261 C.addTransition(stateZeroSize);
1264 // If the size is GUARANTEED to be zero, we're done!
1265 if (!stateNonZeroSize)
1268 // Otherwise, record the assumption that the size is nonzero.
1269 state = stateNonZeroSize;
1272 // Check that the string argument is non-null.
1273 const Expr *Arg = CE->getArg(0);
1274 SVal ArgVal = state->getSVal(Arg, LCtx);
1276 state = checkNonNull(C, state, Arg, ArgVal);
1281 SVal strLength = getCStringLength(C, state, Arg, ArgVal);
1283 // If the argument isn't a valid C string, there's no valid state to
1285 if (strLength.isUndef())
1288 DefinedOrUnknownSVal result = UnknownVal();
1290 // If the check is for strnlen() then bind the return value to no more than
1291 // the maxlen value.
1293 QualType cmpTy = C.getSValBuilder().getConditionType();
1295 // It's a little unfortunate to be getting this again,
1296 // but it's not that expensive...
1297 const Expr *maxlenExpr = CE->getArg(1);
1298 SVal maxlenVal = state->getSVal(maxlenExpr, LCtx);
1300 Optional<NonLoc> strLengthNL = strLength.getAs<NonLoc>();
1301 Optional<NonLoc> maxlenValNL = maxlenVal.getAs<NonLoc>();
1303 if (strLengthNL && maxlenValNL) {
1304 ProgramStateRef stateStringTooLong, stateStringNotTooLong;
1306 // Check if the strLength is greater than the maxlen.
1307 std::tie(stateStringTooLong, stateStringNotTooLong) = state->assume(
1309 .evalBinOpNN(state, BO_GT, *strLengthNL, *maxlenValNL, cmpTy)
1310 .castAs<DefinedOrUnknownSVal>());
1312 if (stateStringTooLong && !stateStringNotTooLong) {
1313 // If the string is longer than maxlen, return maxlen.
1314 result = *maxlenValNL;
1315 } else if (stateStringNotTooLong && !stateStringTooLong) {
1316 // If the string is shorter than maxlen, return its length.
1317 result = *strLengthNL;
1321 if (result.isUnknown()) {
1322 // If we don't have enough information for a comparison, there's
1323 // no guarantee the full string length will actually be returned.
1324 // All we know is the return value is the min of the string length
1325 // and the limit. This is better than nothing.
1326 result = C.getSValBuilder().conjureSymbolVal(nullptr, CE, LCtx,
1328 NonLoc resultNL = result.castAs<NonLoc>();
1331 state = state->assume(C.getSValBuilder().evalBinOpNN(
1332 state, BO_LE, resultNL, *strLengthNL, cmpTy)
1333 .castAs<DefinedOrUnknownSVal>(), true);
1337 state = state->assume(C.getSValBuilder().evalBinOpNN(
1338 state, BO_LE, resultNL, *maxlenValNL, cmpTy)
1339 .castAs<DefinedOrUnknownSVal>(), true);
1344 // This is a plain strlen(), not strnlen().
1345 result = strLength.castAs<DefinedOrUnknownSVal>();
1347 // If we don't know the length of the string, conjure a return
1348 // value, so it can be used in constraints, at least.
1349 if (result.isUnknown()) {
1350 result = C.getSValBuilder().conjureSymbolVal(nullptr, CE, LCtx,
1355 // Bind the return value.
1356 assert(!result.isUnknown() && "Should have conjured a value by now");
1357 state = state->BindExpr(CE, LCtx, result);
1358 C.addTransition(state);
1361 void CStringChecker::evalStrcpy(CheckerContext &C, const CallExpr *CE) const {
1362 if (CE->getNumArgs() < 2)
1365 // char *strcpy(char *restrict dst, const char *restrict src);
1366 evalStrcpyCommon(C, CE,
1367 /* returnEnd = */ false,
1368 /* isBounded = */ false,
1369 /* isAppending = */ false);
1372 void CStringChecker::evalStrncpy(CheckerContext &C, const CallExpr *CE) const {
1373 if (CE->getNumArgs() < 3)
1376 // char *strncpy(char *restrict dst, const char *restrict src, size_t n);
1377 evalStrcpyCommon(C, CE,
1378 /* returnEnd = */ false,
1379 /* isBounded = */ true,
1380 /* isAppending = */ false);
1383 void CStringChecker::evalStpcpy(CheckerContext &C, const CallExpr *CE) const {
1384 if (CE->getNumArgs() < 2)
1387 // char *stpcpy(char *restrict dst, const char *restrict src);
1388 evalStrcpyCommon(C, CE,
1389 /* returnEnd = */ true,
1390 /* isBounded = */ false,
1391 /* isAppending = */ false);
1394 void CStringChecker::evalStrcat(CheckerContext &C, const CallExpr *CE) const {
1395 if (CE->getNumArgs() < 2)
1398 //char *strcat(char *restrict s1, const char *restrict s2);
1399 evalStrcpyCommon(C, CE,
1400 /* returnEnd = */ false,
1401 /* isBounded = */ false,
1402 /* isAppending = */ true);
1405 void CStringChecker::evalStrncat(CheckerContext &C, const CallExpr *CE) const {
1406 if (CE->getNumArgs() < 3)
1409 //char *strncat(char *restrict s1, const char *restrict s2, size_t n);
1410 evalStrcpyCommon(C, CE,
1411 /* returnEnd = */ false,
1412 /* isBounded = */ true,
1413 /* isAppending = */ true);
1416 void CStringChecker::evalStrcpyCommon(CheckerContext &C, const CallExpr *CE,
1417 bool returnEnd, bool isBounded,
1418 bool isAppending) const {
1419 CurrentFunctionDescription = "string copy function";
1420 ProgramStateRef state = C.getState();
1421 const LocationContext *LCtx = C.getLocationContext();
1423 // Check that the destination is non-null.
1424 const Expr *Dst = CE->getArg(0);
1425 SVal DstVal = state->getSVal(Dst, LCtx);
1427 state = checkNonNull(C, state, Dst, DstVal);
1431 // Check that the source is non-null.
1432 const Expr *srcExpr = CE->getArg(1);
1433 SVal srcVal = state->getSVal(srcExpr, LCtx);
1434 state = checkNonNull(C, state, srcExpr, srcVal);
1438 // Get the string length of the source.
1439 SVal strLength = getCStringLength(C, state, srcExpr, srcVal);
1441 // If the source isn't a valid C string, give up.
1442 if (strLength.isUndef())
1445 SValBuilder &svalBuilder = C.getSValBuilder();
1446 QualType cmpTy = svalBuilder.getConditionType();
1447 QualType sizeTy = svalBuilder.getContext().getSizeType();
1449 // These two values allow checking two kinds of errors:
1450 // - actual overflows caused by a source that doesn't fit in the destination
1451 // - potential overflows caused by a bound that could exceed the destination
1452 SVal amountCopied = UnknownVal();
1453 SVal maxLastElementIndex = UnknownVal();
1454 const char *boundWarning = nullptr;
1456 // If the function is strncpy, strncat, etc... it is bounded.
1458 // Get the max number of characters to copy.
1459 const Expr *lenExpr = CE->getArg(2);
1460 SVal lenVal = state->getSVal(lenExpr, LCtx);
1462 // Protect against misdeclared strncpy().
1463 lenVal = svalBuilder.evalCast(lenVal, sizeTy, lenExpr->getType());
1465 Optional<NonLoc> strLengthNL = strLength.getAs<NonLoc>();
1466 Optional<NonLoc> lenValNL = lenVal.getAs<NonLoc>();
1468 // If we know both values, we might be able to figure out how much
1470 if (strLengthNL && lenValNL) {
1471 ProgramStateRef stateSourceTooLong, stateSourceNotTooLong;
1473 // Check if the max number to copy is less than the length of the src.
1474 // If the bound is equal to the source length, strncpy won't null-
1475 // terminate the result!
1476 std::tie(stateSourceTooLong, stateSourceNotTooLong) = state->assume(
1477 svalBuilder.evalBinOpNN(state, BO_GE, *strLengthNL, *lenValNL, cmpTy)
1478 .castAs<DefinedOrUnknownSVal>());
1480 if (stateSourceTooLong && !stateSourceNotTooLong) {
1481 // Max number to copy is less than the length of the src, so the actual
1482 // strLength copied is the max number arg.
1483 state = stateSourceTooLong;
1484 amountCopied = lenVal;
1486 } else if (!stateSourceTooLong && stateSourceNotTooLong) {
1487 // The source buffer entirely fits in the bound.
1488 state = stateSourceNotTooLong;
1489 amountCopied = strLength;
1493 // We still want to know if the bound is known to be too large.
1496 // For strncat, the check is strlen(dst) + lenVal < sizeof(dst)
1498 // Get the string length of the destination. If the destination is
1499 // memory that can't have a string length, we shouldn't be copying
1501 SVal dstStrLength = getCStringLength(C, state, Dst, DstVal);
1502 if (dstStrLength.isUndef())
1505 if (Optional<NonLoc> dstStrLengthNL = dstStrLength.getAs<NonLoc>()) {
1506 maxLastElementIndex = svalBuilder.evalBinOpNN(state, BO_Add,
1510 boundWarning = "Size argument is greater than the free space in the "
1511 "destination buffer";
1515 // For strncpy, this is just checking that lenVal <= sizeof(dst)
1516 // (Yes, strncpy and strncat differ in how they treat termination.
1517 // strncat ALWAYS terminates, but strncpy doesn't.)
1519 // We need a special case for when the copy size is zero, in which
1520 // case strncpy will do no work at all. Our bounds check uses n-1
1521 // as the last element accessed, so n == 0 is problematic.
1522 ProgramStateRef StateZeroSize, StateNonZeroSize;
1523 std::tie(StateZeroSize, StateNonZeroSize) =
1524 assumeZero(C, state, *lenValNL, sizeTy);
1526 // If the size is known to be zero, we're done.
1527 if (StateZeroSize && !StateNonZeroSize) {
1528 StateZeroSize = StateZeroSize->BindExpr(CE, LCtx, DstVal);
1529 C.addTransition(StateZeroSize);
1533 // Otherwise, go ahead and figure out the last element we'll touch.
1534 // We don't record the non-zero assumption here because we can't
1535 // be sure. We won't warn on a possible zero.
1536 NonLoc one = svalBuilder.makeIntVal(1, sizeTy).castAs<NonLoc>();
1537 maxLastElementIndex = svalBuilder.evalBinOpNN(state, BO_Sub, *lenValNL,
1539 boundWarning = "Size argument is greater than the length of the "
1540 "destination buffer";
1544 // If we couldn't pin down the copy length, at least bound it.
1545 // FIXME: We should actually run this code path for append as well, but
1546 // right now it creates problems with constraints (since we can end up
1547 // trying to pass constraints from symbol to symbol).
1548 if (amountCopied.isUnknown() && !isAppending) {
1549 // Try to get a "hypothetical" string length symbol, which we can later
1550 // set as a real value if that turns out to be the case.
1551 amountCopied = getCStringLength(C, state, lenExpr, srcVal, true);
1552 assert(!amountCopied.isUndef());
1554 if (Optional<NonLoc> amountCopiedNL = amountCopied.getAs<NonLoc>()) {
1556 // amountCopied <= lenVal
1557 SVal copiedLessThanBound = svalBuilder.evalBinOpNN(state, BO_LE,
1561 state = state->assume(
1562 copiedLessThanBound.castAs<DefinedOrUnknownSVal>(), true);
1568 // amountCopied <= strlen(source)
1569 SVal copiedLessThanSrc = svalBuilder.evalBinOpNN(state, BO_LE,
1573 state = state->assume(
1574 copiedLessThanSrc.castAs<DefinedOrUnknownSVal>(), true);
1582 // The function isn't bounded. The amount copied should match the length
1583 // of the source buffer.
1584 amountCopied = strLength;
1589 // This represents the number of characters copied into the destination
1590 // buffer. (It may not actually be the strlen if the destination buffer
1591 // is not terminated.)
1592 SVal finalStrLength = UnknownVal();
1594 // If this is an appending function (strcat, strncat...) then set the
1595 // string length to strlen(src) + strlen(dst) since the buffer will
1596 // ultimately contain both.
1598 // Get the string length of the destination. If the destination is memory
1599 // that can't have a string length, we shouldn't be copying into it anyway.
1600 SVal dstStrLength = getCStringLength(C, state, Dst, DstVal);
1601 if (dstStrLength.isUndef())
1604 Optional<NonLoc> srcStrLengthNL = amountCopied.getAs<NonLoc>();
1605 Optional<NonLoc> dstStrLengthNL = dstStrLength.getAs<NonLoc>();
1607 // If we know both string lengths, we might know the final string length.
1608 if (srcStrLengthNL && dstStrLengthNL) {
1609 // Make sure the two lengths together don't overflow a size_t.
1610 state = checkAdditionOverflow(C, state, *srcStrLengthNL, *dstStrLengthNL);
1614 finalStrLength = svalBuilder.evalBinOpNN(state, BO_Add, *srcStrLengthNL,
1615 *dstStrLengthNL, sizeTy);
1618 // If we couldn't get a single value for the final string length,
1619 // we can at least bound it by the individual lengths.
1620 if (finalStrLength.isUnknown()) {
1621 // Try to get a "hypothetical" string length symbol, which we can later
1622 // set as a real value if that turns out to be the case.
1623 finalStrLength = getCStringLength(C, state, CE, DstVal, true);
1624 assert(!finalStrLength.isUndef());
1626 if (Optional<NonLoc> finalStrLengthNL = finalStrLength.getAs<NonLoc>()) {
1627 if (srcStrLengthNL) {
1628 // finalStrLength >= srcStrLength
1629 SVal sourceInResult = svalBuilder.evalBinOpNN(state, BO_GE,
1633 state = state->assume(sourceInResult.castAs<DefinedOrUnknownSVal>(),
1639 if (dstStrLengthNL) {
1640 // finalStrLength >= dstStrLength
1641 SVal destInResult = svalBuilder.evalBinOpNN(state, BO_GE,
1646 state->assume(destInResult.castAs<DefinedOrUnknownSVal>(), true);
1654 // Otherwise, this is a copy-over function (strcpy, strncpy, ...), and
1655 // the final string length will match the input string length.
1656 finalStrLength = amountCopied;
1659 // The final result of the function will either be a pointer past the last
1660 // copied element, or a pointer to the start of the destination buffer.
1661 SVal Result = (returnEnd ? UnknownVal() : DstVal);
1665 // If the destination is a MemRegion, try to check for a buffer overflow and
1666 // record the new string length.
1667 if (Optional<loc::MemRegionVal> dstRegVal =
1668 DstVal.getAs<loc::MemRegionVal>()) {
1669 QualType ptrTy = Dst->getType();
1671 // If we have an exact value on a bounded copy, use that to check for
1672 // overflows, rather than our estimate about how much is actually copied.
1674 if (Optional<NonLoc> maxLastNL = maxLastElementIndex.getAs<NonLoc>()) {
1675 SVal maxLastElement = svalBuilder.evalBinOpLN(state, BO_Add, *dstRegVal,
1677 state = CheckLocation(C, state, CE->getArg(2), maxLastElement,
1684 // Then, if the final length is known...
1685 if (Optional<NonLoc> knownStrLength = finalStrLength.getAs<NonLoc>()) {
1686 SVal lastElement = svalBuilder.evalBinOpLN(state, BO_Add, *dstRegVal,
1687 *knownStrLength, ptrTy);
1689 // ...and we haven't checked the bound, we'll check the actual copy.
1690 if (!boundWarning) {
1691 const char * const warningMsg =
1692 "String copy function overflows destination buffer";
1693 state = CheckLocation(C, state, Dst, lastElement, warningMsg);
1698 // If this is a stpcpy-style copy, the last element is the return value.
1700 Result = lastElement;
1703 // Invalidate the destination (regular invalidation without pointer-escaping
1704 // the address of the top-level region). This must happen before we set the
1705 // C string length because invalidation will clear the length.
1706 // FIXME: Even if we can't perfectly model the copy, we should see if we
1707 // can use LazyCompoundVals to copy the source values into the destination.
1708 // This would probably remove any existing bindings past the end of the
1709 // string, but that's still an improvement over blank invalidation.
1710 state = InvalidateBuffer(C, state, Dst, *dstRegVal,
1711 /*IsSourceBuffer*/false, nullptr);
1713 // Invalidate the source (const-invalidation without const-pointer-escaping
1714 // the address of the top-level region).
1715 state = InvalidateBuffer(C, state, srcExpr, srcVal, /*IsSourceBuffer*/true,
1718 // Set the C string length of the destination, if we know it.
1719 if (isBounded && !isAppending) {
1720 // strncpy is annoying in that it doesn't guarantee to null-terminate
1721 // the result string. If the original string didn't fit entirely inside
1722 // the bound (including the null-terminator), we don't know how long the
1724 if (amountCopied != strLength)
1725 finalStrLength = UnknownVal();
1727 state = setCStringLength(state, dstRegVal->getRegion(), finalStrLength);
1732 // If this is a stpcpy-style copy, but we were unable to check for a buffer
1733 // overflow, we still need a result. Conjure a return value.
1734 if (returnEnd && Result.isUnknown()) {
1735 Result = svalBuilder.conjureSymbolVal(nullptr, CE, LCtx, C.blockCount());
1738 // Set the return value.
1739 state = state->BindExpr(CE, LCtx, Result);
1740 C.addTransition(state);
1743 void CStringChecker::evalStrcmp(CheckerContext &C, const CallExpr *CE) const {
1744 if (CE->getNumArgs() < 2)
1747 //int strcmp(const char *s1, const char *s2);
1748 evalStrcmpCommon(C, CE, /* isBounded = */ false, /* ignoreCase = */ false);
1751 void CStringChecker::evalStrncmp(CheckerContext &C, const CallExpr *CE) const {
1752 if (CE->getNumArgs() < 3)
1755 //int strncmp(const char *s1, const char *s2, size_t n);
1756 evalStrcmpCommon(C, CE, /* isBounded = */ true, /* ignoreCase = */ false);
1759 void CStringChecker::evalStrcasecmp(CheckerContext &C,
1760 const CallExpr *CE) const {
1761 if (CE->getNumArgs() < 2)
1764 //int strcasecmp(const char *s1, const char *s2);
1765 evalStrcmpCommon(C, CE, /* isBounded = */ false, /* ignoreCase = */ true);
1768 void CStringChecker::evalStrncasecmp(CheckerContext &C,
1769 const CallExpr *CE) const {
1770 if (CE->getNumArgs() < 3)
1773 //int strncasecmp(const char *s1, const char *s2, size_t n);
1774 evalStrcmpCommon(C, CE, /* isBounded = */ true, /* ignoreCase = */ true);
1777 void CStringChecker::evalStrcmpCommon(CheckerContext &C, const CallExpr *CE,
1778 bool isBounded, bool ignoreCase) const {
1779 CurrentFunctionDescription = "string comparison function";
1780 ProgramStateRef state = C.getState();
1781 const LocationContext *LCtx = C.getLocationContext();
1783 // Check that the first string is non-null
1784 const Expr *s1 = CE->getArg(0);
1785 SVal s1Val = state->getSVal(s1, LCtx);
1786 state = checkNonNull(C, state, s1, s1Val);
1790 // Check that the second string is non-null.
1791 const Expr *s2 = CE->getArg(1);
1792 SVal s2Val = state->getSVal(s2, LCtx);
1793 state = checkNonNull(C, state, s2, s2Val);
1797 // Get the string length of the first string or give up.
1798 SVal s1Length = getCStringLength(C, state, s1, s1Val);
1799 if (s1Length.isUndef())
1802 // Get the string length of the second string or give up.
1803 SVal s2Length = getCStringLength(C, state, s2, s2Val);
1804 if (s2Length.isUndef())
1807 // If we know the two buffers are the same, we know the result is 0.
1808 // First, get the two buffers' addresses. Another checker will have already
1809 // made sure they're not undefined.
1810 DefinedOrUnknownSVal LV = s1Val.castAs<DefinedOrUnknownSVal>();
1811 DefinedOrUnknownSVal RV = s2Val.castAs<DefinedOrUnknownSVal>();
1813 // See if they are the same.
1814 SValBuilder &svalBuilder = C.getSValBuilder();
1815 DefinedOrUnknownSVal SameBuf = svalBuilder.evalEQ(state, LV, RV);
1816 ProgramStateRef StSameBuf, StNotSameBuf;
1817 std::tie(StSameBuf, StNotSameBuf) = state->assume(SameBuf);
1819 // If the two arguments might be the same buffer, we know the result is 0,
1820 // and we only need to check one size.
1822 StSameBuf = StSameBuf->BindExpr(CE, LCtx,
1823 svalBuilder.makeZeroVal(CE->getType()));
1824 C.addTransition(StSameBuf);
1826 // If the two arguments are GUARANTEED to be the same, we're done!
1831 assert(StNotSameBuf);
1832 state = StNotSameBuf;
1834 // At this point we can go about comparing the two buffers.
1835 // For now, we only do this if they're both known string literals.
1837 // Attempt to extract string literals from both expressions.
1838 const StringLiteral *s1StrLiteral = getCStringLiteral(C, state, s1, s1Val);
1839 const StringLiteral *s2StrLiteral = getCStringLiteral(C, state, s2, s2Val);
1840 bool canComputeResult = false;
1841 SVal resultVal = svalBuilder.conjureSymbolVal(nullptr, CE, LCtx,
1844 if (s1StrLiteral && s2StrLiteral) {
1845 StringRef s1StrRef = s1StrLiteral->getString();
1846 StringRef s2StrRef = s2StrLiteral->getString();
1849 // Get the max number of characters to compare.
1850 const Expr *lenExpr = CE->getArg(2);
1851 SVal lenVal = state->getSVal(lenExpr, LCtx);
1853 // If the length is known, we can get the right substrings.
1854 if (const llvm::APSInt *len = svalBuilder.getKnownValue(state, lenVal)) {
1855 // Create substrings of each to compare the prefix.
1856 s1StrRef = s1StrRef.substr(0, (size_t)len->getZExtValue());
1857 s2StrRef = s2StrRef.substr(0, (size_t)len->getZExtValue());
1858 canComputeResult = true;
1861 // This is a normal, unbounded strcmp.
1862 canComputeResult = true;
1865 if (canComputeResult) {
1866 // Real strcmp stops at null characters.
1867 size_t s1Term = s1StrRef.find('\0');
1868 if (s1Term != StringRef::npos)
1869 s1StrRef = s1StrRef.substr(0, s1Term);
1871 size_t s2Term = s2StrRef.find('\0');
1872 if (s2Term != StringRef::npos)
1873 s2StrRef = s2StrRef.substr(0, s2Term);
1875 // Use StringRef's comparison methods to compute the actual result.
1876 int compareRes = ignoreCase ? s1StrRef.compare_lower(s2StrRef)
1877 : s1StrRef.compare(s2StrRef);
1879 // The strcmp function returns an integer greater than, equal to, or less
1880 // than zero, [c11, p7.24.4.2].
1881 if (compareRes == 0) {
1882 resultVal = svalBuilder.makeIntVal(compareRes, CE->getType());
1885 DefinedSVal zeroVal = svalBuilder.makeIntVal(0, CE->getType());
1886 // Constrain strcmp's result range based on the result of StringRef's
1887 // comparison methods.
1888 BinaryOperatorKind op = (compareRes == 1) ? BO_GT : BO_LT;
1889 SVal compareWithZero =
1890 svalBuilder.evalBinOp(state, op, resultVal, zeroVal,
1891 svalBuilder.getConditionType());
1892 DefinedSVal compareWithZeroVal = compareWithZero.castAs<DefinedSVal>();
1893 state = state->assume(compareWithZeroVal, true);
1898 state = state->BindExpr(CE, LCtx, resultVal);
1900 // Record this as a possible path.
1901 C.addTransition(state);
1904 void CStringChecker::evalStrsep(CheckerContext &C, const CallExpr *CE) const {
1905 //char *strsep(char **stringp, const char *delim);
1906 if (CE->getNumArgs() < 2)
1909 // Sanity: does the search string parameter match the return type?
1910 const Expr *SearchStrPtr = CE->getArg(0);
1911 QualType CharPtrTy = SearchStrPtr->getType()->getPointeeType();
1912 if (CharPtrTy.isNull() ||
1913 CE->getType().getUnqualifiedType() != CharPtrTy.getUnqualifiedType())
1916 CurrentFunctionDescription = "strsep()";
1917 ProgramStateRef State = C.getState();
1918 const LocationContext *LCtx = C.getLocationContext();
1920 // Check that the search string pointer is non-null (though it may point to
1922 SVal SearchStrVal = State->getSVal(SearchStrPtr, LCtx);
1923 State = checkNonNull(C, State, SearchStrPtr, SearchStrVal);
1927 // Check that the delimiter string is non-null.
1928 const Expr *DelimStr = CE->getArg(1);
1929 SVal DelimStrVal = State->getSVal(DelimStr, LCtx);
1930 State = checkNonNull(C, State, DelimStr, DelimStrVal);
1934 SValBuilder &SVB = C.getSValBuilder();
1936 if (Optional<Loc> SearchStrLoc = SearchStrVal.getAs<Loc>()) {
1937 // Get the current value of the search string pointer, as a char*.
1938 Result = State->getSVal(*SearchStrLoc, CharPtrTy);
1940 // Invalidate the search string, representing the change of one delimiter
1941 // character to NUL.
1942 State = InvalidateBuffer(C, State, SearchStrPtr, Result,
1943 /*IsSourceBuffer*/false, nullptr);
1945 // Overwrite the search string pointer. The new value is either an address
1946 // further along in the same string, or NULL if there are no more tokens.
1947 State = State->bindLoc(*SearchStrLoc,
1948 SVB.conjureSymbolVal(getTag(),
1955 assert(SearchStrVal.isUnknown());
1956 // Conjure a symbolic value. It's the best we can do.
1957 Result = SVB.conjureSymbolVal(nullptr, CE, LCtx, C.blockCount());
1960 // Set the return value, and finish.
1961 State = State->BindExpr(CE, LCtx, Result);
1962 C.addTransition(State);
1965 // These should probably be moved into a C++ standard library checker.
1966 void CStringChecker::evalStdCopy(CheckerContext &C, const CallExpr *CE) const {
1967 evalStdCopyCommon(C, CE);
1970 void CStringChecker::evalStdCopyBackward(CheckerContext &C,
1971 const CallExpr *CE) const {
1972 evalStdCopyCommon(C, CE);
1975 void CStringChecker::evalStdCopyCommon(CheckerContext &C,
1976 const CallExpr *CE) const {
1977 if (CE->getNumArgs() < 3)
1980 ProgramStateRef State = C.getState();
1982 const LocationContext *LCtx = C.getLocationContext();
1984 // template <class _InputIterator, class _OutputIterator>
1986 // copy(_InputIterator __first, _InputIterator __last,
1987 // _OutputIterator __result)
1989 // Invalidate the destination buffer
1990 const Expr *Dst = CE->getArg(2);
1991 SVal DstVal = State->getSVal(Dst, LCtx);
1992 State = InvalidateBuffer(C, State, Dst, DstVal, /*IsSource=*/false,
1995 SValBuilder &SVB = C.getSValBuilder();
1997 SVal ResultVal = SVB.conjureSymbolVal(nullptr, CE, LCtx, C.blockCount());
1998 State = State->BindExpr(CE, LCtx, ResultVal);
2000 C.addTransition(State);
2003 void CStringChecker::evalMemset(CheckerContext &C, const CallExpr *CE) const {
2004 if (CE->getNumArgs() != 3)
2007 CurrentFunctionDescription = "memory set function";
2009 const Expr *Mem = CE->getArg(0);
2010 const Expr *Size = CE->getArg(2);
2011 ProgramStateRef State = C.getState();
2013 // See if the size argument is zero.
2014 const LocationContext *LCtx = C.getLocationContext();
2015 SVal SizeVal = State->getSVal(Size, LCtx);
2016 QualType SizeTy = Size->getType();
2018 ProgramStateRef StateZeroSize, StateNonZeroSize;
2019 std::tie(StateZeroSize, StateNonZeroSize) =
2020 assumeZero(C, State, SizeVal, SizeTy);
2022 // Get the value of the memory area.
2023 SVal MemVal = State->getSVal(Mem, LCtx);
2025 // If the size is zero, there won't be any actual memory access, so
2026 // just bind the return value to the Mem buffer and return.
2027 if (StateZeroSize && !StateNonZeroSize) {
2028 StateZeroSize = StateZeroSize->BindExpr(CE, LCtx, MemVal);
2029 C.addTransition(StateZeroSize);
2033 // Ensure the memory area is not null.
2034 // If it is NULL there will be a NULL pointer dereference.
2035 State = checkNonNull(C, StateNonZeroSize, Mem, MemVal);
2039 State = CheckBufferAccess(C, State, Size, Mem);
2042 State = InvalidateBuffer(C, State, Mem, C.getSVal(Mem),
2043 /*IsSourceBuffer*/false, Size);
2047 State = State->BindExpr(CE, LCtx, MemVal);
2048 C.addTransition(State);
2051 static bool isCPPStdLibraryFunction(const FunctionDecl *FD, StringRef Name) {
2052 IdentifierInfo *II = FD->getIdentifier();
2056 if (!AnalysisDeclContext::isInStdNamespace(FD))
2059 if (II->getName().equals(Name))
2064 //===----------------------------------------------------------------------===//
2065 // The driver method, and other Checker callbacks.
2066 //===----------------------------------------------------------------------===//
2068 bool CStringChecker::evalCall(const CallExpr *CE, CheckerContext &C) const {
2069 const FunctionDecl *FDecl = C.getCalleeDecl(CE);
2074 // FIXME: Poorly-factored string switches are slow.
2075 FnCheck evalFunction = nullptr;
2076 if (C.isCLibraryFunction(FDecl, "memcpy"))
2077 evalFunction = &CStringChecker::evalMemcpy;
2078 else if (C.isCLibraryFunction(FDecl, "mempcpy"))
2079 evalFunction = &CStringChecker::evalMempcpy;
2080 else if (C.isCLibraryFunction(FDecl, "memcmp"))
2081 evalFunction = &CStringChecker::evalMemcmp;
2082 else if (C.isCLibraryFunction(FDecl, "memmove"))
2083 evalFunction = &CStringChecker::evalMemmove;
2084 else if (C.isCLibraryFunction(FDecl, "memset"))
2085 evalFunction = &CStringChecker::evalMemset;
2086 else if (C.isCLibraryFunction(FDecl, "strcpy"))
2087 evalFunction = &CStringChecker::evalStrcpy;
2088 else if (C.isCLibraryFunction(FDecl, "strncpy"))
2089 evalFunction = &CStringChecker::evalStrncpy;
2090 else if (C.isCLibraryFunction(FDecl, "stpcpy"))
2091 evalFunction = &CStringChecker::evalStpcpy;
2092 else if (C.isCLibraryFunction(FDecl, "strcat"))
2093 evalFunction = &CStringChecker::evalStrcat;
2094 else if (C.isCLibraryFunction(FDecl, "strncat"))
2095 evalFunction = &CStringChecker::evalStrncat;
2096 else if (C.isCLibraryFunction(FDecl, "strlen"))
2097 evalFunction = &CStringChecker::evalstrLength;
2098 else if (C.isCLibraryFunction(FDecl, "strnlen"))
2099 evalFunction = &CStringChecker::evalstrnLength;
2100 else if (C.isCLibraryFunction(FDecl, "strcmp"))
2101 evalFunction = &CStringChecker::evalStrcmp;
2102 else if (C.isCLibraryFunction(FDecl, "strncmp"))
2103 evalFunction = &CStringChecker::evalStrncmp;
2104 else if (C.isCLibraryFunction(FDecl, "strcasecmp"))
2105 evalFunction = &CStringChecker::evalStrcasecmp;
2106 else if (C.isCLibraryFunction(FDecl, "strncasecmp"))
2107 evalFunction = &CStringChecker::evalStrncasecmp;
2108 else if (C.isCLibraryFunction(FDecl, "strsep"))
2109 evalFunction = &CStringChecker::evalStrsep;
2110 else if (C.isCLibraryFunction(FDecl, "bcopy"))
2111 evalFunction = &CStringChecker::evalBcopy;
2112 else if (C.isCLibraryFunction(FDecl, "bcmp"))
2113 evalFunction = &CStringChecker::evalMemcmp;
2114 else if (isCPPStdLibraryFunction(FDecl, "copy"))
2115 evalFunction = &CStringChecker::evalStdCopy;
2116 else if (isCPPStdLibraryFunction(FDecl, "copy_backward"))
2117 evalFunction = &CStringChecker::evalStdCopyBackward;
2119 // If the callee isn't a string function, let another checker handle it.
2123 // Check and evaluate the call.
2124 (this->*evalFunction)(C, CE);
2126 // If the evaluate call resulted in no change, chain to the next eval call
2128 // Note, the custom CString evaluation calls assume that basic safety
2129 // properties are held. However, if the user chooses to turn off some of these
2130 // checks, we ignore the issues and leave the call evaluation to a generic
2132 return C.isDifferent();
2135 void CStringChecker::checkPreStmt(const DeclStmt *DS, CheckerContext &C) const {
2136 // Record string length for char a[] = "abc";
2137 ProgramStateRef state = C.getState();
2139 for (const auto *I : DS->decls()) {
2140 const VarDecl *D = dyn_cast<VarDecl>(I);
2144 // FIXME: Handle array fields of structs.
2145 if (!D->getType()->isArrayType())
2148 const Expr *Init = D->getInit();
2151 if (!isa<StringLiteral>(Init))
2154 Loc VarLoc = state->getLValue(D, C.getLocationContext());
2155 const MemRegion *MR = VarLoc.getAsRegion();
2159 SVal StrVal = state->getSVal(Init, C.getLocationContext());
2160 assert(StrVal.isValid() && "Initializer string is unknown or undefined");
2161 DefinedOrUnknownSVal strLength =
2162 getCStringLength(C, state, Init, StrVal).castAs<DefinedOrUnknownSVal>();
2164 state = state->set<CStringLength>(MR, strLength);
2167 C.addTransition(state);
2171 CStringChecker::checkRegionChanges(ProgramStateRef state,
2172 const InvalidatedSymbols *,
2173 ArrayRef<const MemRegion *> ExplicitRegions,
2174 ArrayRef<const MemRegion *> Regions,
2175 const LocationContext *LCtx,
2176 const CallEvent *Call) const {
2177 CStringLengthTy Entries = state->get<CStringLength>();
2178 if (Entries.isEmpty())
2181 llvm::SmallPtrSet<const MemRegion *, 8> Invalidated;
2182 llvm::SmallPtrSet<const MemRegion *, 32> SuperRegions;
2184 // First build sets for the changed regions and their super-regions.
2185 for (ArrayRef<const MemRegion *>::iterator
2186 I = Regions.begin(), E = Regions.end(); I != E; ++I) {
2187 const MemRegion *MR = *I;
2188 Invalidated.insert(MR);
2190 SuperRegions.insert(MR);
2191 while (const SubRegion *SR = dyn_cast<SubRegion>(MR)) {
2192 MR = SR->getSuperRegion();
2193 SuperRegions.insert(MR);
2197 CStringLengthTy::Factory &F = state->get_context<CStringLength>();
2199 // Then loop over the entries in the current state.
2200 for (CStringLengthTy::iterator I = Entries.begin(),
2201 E = Entries.end(); I != E; ++I) {
2202 const MemRegion *MR = I.getKey();
2204 // Is this entry for a super-region of a changed region?
2205 if (SuperRegions.count(MR)) {
2206 Entries = F.remove(Entries, MR);
2210 // Is this entry for a sub-region of a changed region?
2211 const MemRegion *Super = MR;
2212 while (const SubRegion *SR = dyn_cast<SubRegion>(Super)) {
2213 Super = SR->getSuperRegion();
2214 if (Invalidated.count(Super)) {
2215 Entries = F.remove(Entries, MR);
2221 return state->set<CStringLength>(Entries);
2224 void CStringChecker::checkLiveSymbols(ProgramStateRef state,
2225 SymbolReaper &SR) const {
2226 // Mark all symbols in our string length map as valid.
2227 CStringLengthTy Entries = state->get<CStringLength>();
2229 for (CStringLengthTy::iterator I = Entries.begin(), E = Entries.end();
2231 SVal Len = I.getData();
2233 for (SymExpr::symbol_iterator si = Len.symbol_begin(),
2234 se = Len.symbol_end(); si != se; ++si)
2239 void CStringChecker::checkDeadSymbols(SymbolReaper &SR,
2240 CheckerContext &C) const {
2241 if (!SR.hasDeadSymbols())
2244 ProgramStateRef state = C.getState();
2245 CStringLengthTy Entries = state->get<CStringLength>();
2246 if (Entries.isEmpty())
2249 CStringLengthTy::Factory &F = state->get_context<CStringLength>();
2250 for (CStringLengthTy::iterator I = Entries.begin(), E = Entries.end();
2252 SVal Len = I.getData();
2253 if (SymbolRef Sym = Len.getAsSymbol()) {
2255 Entries = F.remove(Entries, I.getKey());
2259 state = state->set<CStringLength>(Entries);
2260 C.addTransition(state);
2263 #define REGISTER_CHECKER(name) \
2264 void ento::register##name(CheckerManager &mgr) { \
2265 CStringChecker *checker = mgr.registerChecker<CStringChecker>(); \
2266 checker->Filter.Check##name = true; \
2267 checker->Filter.CheckName##name = mgr.getCurrentCheckName(); \
2270 REGISTER_CHECKER(CStringNullArg)
2271 REGISTER_CHECKER(CStringOutOfBounds)
2272 REGISTER_CHECKER(CStringBufferOverlap)
2273 REGISTER_CHECKER(CStringNotNullTerm)
2275 void ento::registerCStringCheckerBasic(CheckerManager &Mgr) {
2276 registerCStringNullArg(Mgr);