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;
125 std::pair<ProgramStateRef , ProgramStateRef >
126 static assumeZero(CheckerContext &C,
127 ProgramStateRef state, SVal V, QualType Ty);
129 static ProgramStateRef setCStringLength(ProgramStateRef state,
132 static SVal getCStringLengthForRegion(CheckerContext &C,
133 ProgramStateRef &state,
137 SVal getCStringLength(CheckerContext &C,
138 ProgramStateRef &state,
141 bool hypothetical = false) const;
143 const StringLiteral *getCStringLiteral(CheckerContext &C,
144 ProgramStateRef &state,
148 static ProgramStateRef InvalidateBuffer(CheckerContext &C,
149 ProgramStateRef state,
150 const Expr *Ex, SVal V,
154 static bool SummarizeRegion(raw_ostream &os, ASTContext &Ctx,
155 const MemRegion *MR);
158 ProgramStateRef checkNonNull(CheckerContext &C,
159 ProgramStateRef state,
162 ProgramStateRef CheckLocation(CheckerContext &C,
163 ProgramStateRef state,
166 const char *message = nullptr) const;
167 ProgramStateRef CheckBufferAccess(CheckerContext &C,
168 ProgramStateRef state,
170 const Expr *FirstBuf,
171 const Expr *SecondBuf,
172 const char *firstMessage = nullptr,
173 const char *secondMessage = nullptr,
174 bool WarnAboutSize = false) const;
176 ProgramStateRef CheckBufferAccess(CheckerContext &C,
177 ProgramStateRef state,
180 const char *message = nullptr,
181 bool WarnAboutSize = false) const {
182 // This is a convenience override.
183 return CheckBufferAccess(C, state, Size, Buf, nullptr, message, nullptr,
186 ProgramStateRef CheckOverlap(CheckerContext &C,
187 ProgramStateRef state,
190 const Expr *Second) const;
191 void emitOverlapBug(CheckerContext &C,
192 ProgramStateRef state,
194 const Stmt *Second) const;
196 ProgramStateRef checkAdditionOverflow(CheckerContext &C,
197 ProgramStateRef state,
201 // Return true if the destination buffer of the copy function may be in bound.
202 // Expects SVal of Size to be positive and unsigned.
203 // Expects SVal of FirstBuf to be a FieldRegion.
204 static bool IsFirstBufInBound(CheckerContext &C,
205 ProgramStateRef state,
206 const Expr *FirstBuf,
210 } //end anonymous namespace
212 REGISTER_MAP_WITH_PROGRAMSTATE(CStringLength, const MemRegion *, SVal)
214 //===----------------------------------------------------------------------===//
215 // Individual checks and utility methods.
216 //===----------------------------------------------------------------------===//
218 std::pair<ProgramStateRef , ProgramStateRef >
219 CStringChecker::assumeZero(CheckerContext &C, ProgramStateRef state, SVal V,
221 Optional<DefinedSVal> val = V.getAs<DefinedSVal>();
223 return std::pair<ProgramStateRef , ProgramStateRef >(state, state);
225 SValBuilder &svalBuilder = C.getSValBuilder();
226 DefinedOrUnknownSVal zero = svalBuilder.makeZeroVal(Ty);
227 return state->assume(svalBuilder.evalEQ(state, *val, zero));
230 ProgramStateRef CStringChecker::checkNonNull(CheckerContext &C,
231 ProgramStateRef state,
232 const Expr *S, SVal l) const {
233 // If a previous check has failed, propagate the failure.
237 ProgramStateRef stateNull, stateNonNull;
238 std::tie(stateNull, stateNonNull) = assumeZero(C, state, l, S->getType());
240 if (stateNull && !stateNonNull) {
241 if (!Filter.CheckCStringNullArg)
244 ExplodedNode *N = C.generateErrorNode(stateNull);
249 BT_Null.reset(new BuiltinBug(
250 Filter.CheckNameCStringNullArg, categories::UnixAPI,
251 "Null pointer argument in call to byte string function"));
254 llvm::raw_svector_ostream os(buf);
255 assert(CurrentFunctionDescription);
256 os << "Null pointer argument in call to " << CurrentFunctionDescription;
258 // Generate a report for this bug.
259 BuiltinBug *BT = static_cast<BuiltinBug*>(BT_Null.get());
260 auto report = llvm::make_unique<BugReport>(*BT, os.str(), N);
262 report->addRange(S->getSourceRange());
263 bugreporter::trackNullOrUndefValue(N, S, *report);
264 C.emitReport(std::move(report));
268 // From here on, assume that the value is non-null.
269 assert(stateNonNull);
273 // FIXME: This was originally copied from ArrayBoundChecker.cpp. Refactor?
274 ProgramStateRef CStringChecker::CheckLocation(CheckerContext &C,
275 ProgramStateRef state,
276 const Expr *S, SVal l,
277 const char *warningMsg) const {
278 // If a previous check has failed, propagate the failure.
282 // Check for out of bound array element access.
283 const MemRegion *R = l.getAsRegion();
287 const ElementRegion *ER = dyn_cast<ElementRegion>(R);
291 assert(ER->getValueType() == C.getASTContext().CharTy &&
292 "CheckLocation should only be called with char* ElementRegions");
294 // Get the size of the array.
295 const SubRegion *superReg = cast<SubRegion>(ER->getSuperRegion());
296 SValBuilder &svalBuilder = C.getSValBuilder();
298 svalBuilder.convertToArrayIndex(superReg->getExtent(svalBuilder));
299 DefinedOrUnknownSVal Size = Extent.castAs<DefinedOrUnknownSVal>();
301 // Get the index of the accessed element.
302 DefinedOrUnknownSVal Idx = ER->getIndex().castAs<DefinedOrUnknownSVal>();
304 ProgramStateRef StInBound = state->assumeInBound(Idx, Size, true);
305 ProgramStateRef StOutBound = state->assumeInBound(Idx, Size, false);
306 if (StOutBound && !StInBound) {
307 ExplodedNode *N = C.generateErrorNode(StOutBound);
312 BT_Bounds.reset(new BuiltinBug(
313 Filter.CheckNameCStringOutOfBounds, "Out-of-bound array access",
314 "Byte string function accesses out-of-bound array element"));
316 BuiltinBug *BT = static_cast<BuiltinBug*>(BT_Bounds.get());
318 // Generate a report for this bug.
319 std::unique_ptr<BugReport> report;
321 report = llvm::make_unique<BugReport>(*BT, warningMsg, N);
323 assert(CurrentFunctionDescription);
324 assert(CurrentFunctionDescription[0] != '\0');
327 llvm::raw_svector_ostream os(buf);
328 os << toUppercase(CurrentFunctionDescription[0])
329 << &CurrentFunctionDescription[1]
330 << " accesses out-of-bound array element";
331 report = llvm::make_unique<BugReport>(*BT, os.str(), N);
334 // FIXME: It would be nice to eventually make this diagnostic more clear,
335 // e.g., by referencing the original declaration or by saying *why* this
336 // reference is outside the range.
338 report->addRange(S->getSourceRange());
339 C.emitReport(std::move(report));
343 // Array bound check succeeded. From this point forward the array bound
344 // should always succeed.
348 ProgramStateRef CStringChecker::CheckBufferAccess(CheckerContext &C,
349 ProgramStateRef state,
351 const Expr *FirstBuf,
352 const Expr *SecondBuf,
353 const char *firstMessage,
354 const char *secondMessage,
355 bool WarnAboutSize) const {
356 // If a previous check has failed, propagate the failure.
360 SValBuilder &svalBuilder = C.getSValBuilder();
361 ASTContext &Ctx = svalBuilder.getContext();
362 const LocationContext *LCtx = C.getLocationContext();
364 QualType sizeTy = Size->getType();
365 QualType PtrTy = Ctx.getPointerType(Ctx.CharTy);
367 // Check that the first buffer is non-null.
368 SVal BufVal = state->getSVal(FirstBuf, LCtx);
369 state = checkNonNull(C, state, FirstBuf, BufVal);
373 // If out-of-bounds checking is turned off, skip the rest.
374 if (!Filter.CheckCStringOutOfBounds)
377 // Get the access length and make sure it is known.
378 // FIXME: This assumes the caller has already checked that the access length
379 // is positive. And that it's unsigned.
380 SVal LengthVal = state->getSVal(Size, LCtx);
381 Optional<NonLoc> Length = LengthVal.getAs<NonLoc>();
385 // Compute the offset of the last element to be accessed: size-1.
386 NonLoc One = svalBuilder.makeIntVal(1, sizeTy).castAs<NonLoc>();
387 NonLoc LastOffset = svalBuilder
388 .evalBinOpNN(state, BO_Sub, *Length, One, sizeTy).castAs<NonLoc>();
390 // Check that the first buffer is sufficiently long.
391 SVal BufStart = svalBuilder.evalCast(BufVal, PtrTy, FirstBuf->getType());
392 if (Optional<Loc> BufLoc = BufStart.getAs<Loc>()) {
393 const Expr *warningExpr = (WarnAboutSize ? Size : FirstBuf);
395 SVal BufEnd = svalBuilder.evalBinOpLN(state, BO_Add, *BufLoc,
397 state = CheckLocation(C, state, warningExpr, BufEnd, firstMessage);
399 // If the buffer isn't large enough, abort.
404 // If there's a second buffer, check it as well.
406 BufVal = state->getSVal(SecondBuf, LCtx);
407 state = checkNonNull(C, state, SecondBuf, BufVal);
411 BufStart = svalBuilder.evalCast(BufVal, PtrTy, SecondBuf->getType());
412 if (Optional<Loc> BufLoc = BufStart.getAs<Loc>()) {
413 const Expr *warningExpr = (WarnAboutSize ? Size : SecondBuf);
415 SVal BufEnd = svalBuilder.evalBinOpLN(state, BO_Add, *BufLoc,
417 state = CheckLocation(C, state, warningExpr, BufEnd, secondMessage);
421 // Large enough or not, return this state!
425 ProgramStateRef CStringChecker::CheckOverlap(CheckerContext &C,
426 ProgramStateRef state,
429 const Expr *Second) const {
430 if (!Filter.CheckCStringBufferOverlap)
433 // Do a simple check for overlap: if the two arguments are from the same
434 // buffer, see if the end of the first is greater than the start of the second
437 // If a previous check has failed, propagate the failure.
441 ProgramStateRef stateTrue, stateFalse;
443 // Get the buffer values and make sure they're known locations.
444 const LocationContext *LCtx = C.getLocationContext();
445 SVal firstVal = state->getSVal(First, LCtx);
446 SVal secondVal = state->getSVal(Second, LCtx);
448 Optional<Loc> firstLoc = firstVal.getAs<Loc>();
452 Optional<Loc> secondLoc = secondVal.getAs<Loc>();
456 // Are the two values the same?
457 SValBuilder &svalBuilder = C.getSValBuilder();
458 std::tie(stateTrue, stateFalse) =
459 state->assume(svalBuilder.evalEQ(state, *firstLoc, *secondLoc));
461 if (stateTrue && !stateFalse) {
462 // If the values are known to be equal, that's automatically an overlap.
463 emitOverlapBug(C, stateTrue, First, Second);
467 // assume the two expressions are not equal.
471 // Which value comes first?
472 QualType cmpTy = svalBuilder.getConditionType();
473 SVal reverse = svalBuilder.evalBinOpLL(state, BO_GT,
474 *firstLoc, *secondLoc, cmpTy);
475 Optional<DefinedOrUnknownSVal> reverseTest =
476 reverse.getAs<DefinedOrUnknownSVal>();
480 std::tie(stateTrue, stateFalse) = state->assume(*reverseTest);
483 // If we don't know which one comes first, we can't perform this test.
486 // Switch the values so that firstVal is before secondVal.
487 std::swap(firstLoc, secondLoc);
489 // Switch the Exprs as well, so that they still correspond.
490 std::swap(First, Second);
494 // Get the length, and make sure it too is known.
495 SVal LengthVal = state->getSVal(Size, LCtx);
496 Optional<NonLoc> Length = LengthVal.getAs<NonLoc>();
500 // Convert the first buffer's start address to char*.
501 // Bail out if the cast fails.
502 ASTContext &Ctx = svalBuilder.getContext();
503 QualType CharPtrTy = Ctx.getPointerType(Ctx.CharTy);
504 SVal FirstStart = svalBuilder.evalCast(*firstLoc, CharPtrTy,
506 Optional<Loc> FirstStartLoc = FirstStart.getAs<Loc>();
510 // Compute the end of the first buffer. Bail out if THAT fails.
511 SVal FirstEnd = svalBuilder.evalBinOpLN(state, BO_Add,
512 *FirstStartLoc, *Length, CharPtrTy);
513 Optional<Loc> FirstEndLoc = FirstEnd.getAs<Loc>();
517 // Is the end of the first buffer past the start of the second buffer?
518 SVal Overlap = svalBuilder.evalBinOpLL(state, BO_GT,
519 *FirstEndLoc, *secondLoc, cmpTy);
520 Optional<DefinedOrUnknownSVal> OverlapTest =
521 Overlap.getAs<DefinedOrUnknownSVal>();
525 std::tie(stateTrue, stateFalse) = state->assume(*OverlapTest);
527 if (stateTrue && !stateFalse) {
529 emitOverlapBug(C, stateTrue, First, Second);
533 // assume the two expressions don't overlap.
538 void CStringChecker::emitOverlapBug(CheckerContext &C, ProgramStateRef state,
539 const Stmt *First, const Stmt *Second) const {
540 ExplodedNode *N = C.generateErrorNode(state);
545 BT_Overlap.reset(new BugType(Filter.CheckNameCStringBufferOverlap,
546 categories::UnixAPI, "Improper arguments"));
548 // Generate a report for this bug.
549 auto report = llvm::make_unique<BugReport>(
550 *BT_Overlap, "Arguments must not be overlapping buffers", N);
551 report->addRange(First->getSourceRange());
552 report->addRange(Second->getSourceRange());
554 C.emitReport(std::move(report));
557 ProgramStateRef CStringChecker::checkAdditionOverflow(CheckerContext &C,
558 ProgramStateRef state,
560 NonLoc right) const {
561 // If out-of-bounds checking is turned off, skip the rest.
562 if (!Filter.CheckCStringOutOfBounds)
565 // If a previous check has failed, propagate the failure.
569 SValBuilder &svalBuilder = C.getSValBuilder();
570 BasicValueFactory &BVF = svalBuilder.getBasicValueFactory();
572 QualType sizeTy = svalBuilder.getContext().getSizeType();
573 const llvm::APSInt &maxValInt = BVF.getMaxValue(sizeTy);
574 NonLoc maxVal = svalBuilder.makeIntVal(maxValInt);
577 if (right.getAs<nonloc::ConcreteInt>()) {
578 maxMinusRight = svalBuilder.evalBinOpNN(state, BO_Sub, maxVal, right,
581 // Try switching the operands. (The order of these two assignments is
583 maxMinusRight = svalBuilder.evalBinOpNN(state, BO_Sub, maxVal, left,
588 if (Optional<NonLoc> maxMinusRightNL = maxMinusRight.getAs<NonLoc>()) {
589 QualType cmpTy = svalBuilder.getConditionType();
590 // If left > max - right, we have an overflow.
591 SVal willOverflow = svalBuilder.evalBinOpNN(state, BO_GT, left,
592 *maxMinusRightNL, cmpTy);
594 ProgramStateRef stateOverflow, stateOkay;
595 std::tie(stateOverflow, stateOkay) =
596 state->assume(willOverflow.castAs<DefinedOrUnknownSVal>());
598 if (stateOverflow && !stateOkay) {
599 // We have an overflow. Emit a bug report.
600 ExplodedNode *N = C.generateErrorNode(stateOverflow);
604 if (!BT_AdditionOverflow)
605 BT_AdditionOverflow.reset(
606 new BuiltinBug(Filter.CheckNameCStringOutOfBounds, "API",
607 "Sum of expressions causes overflow"));
609 // This isn't a great error message, but this should never occur in real
610 // code anyway -- you'd have to create a buffer longer than a size_t can
611 // represent, which is sort of a contradiction.
612 const char *warning =
613 "This expression will create a string whose length is too big to "
614 "be represented as a size_t";
616 // Generate a report for this bug.
618 llvm::make_unique<BugReport>(*BT_AdditionOverflow, warning, N));
623 // From now on, assume an overflow didn't occur.
631 ProgramStateRef CStringChecker::setCStringLength(ProgramStateRef state,
634 assert(!strLength.isUndef() && "Attempt to set an undefined string length");
636 MR = MR->StripCasts();
638 switch (MR->getKind()) {
639 case MemRegion::StringRegionKind:
640 // FIXME: This can happen if we strcpy() into a string region. This is
641 // undefined [C99 6.4.5p6], but we should still warn about it.
644 case MemRegion::SymbolicRegionKind:
645 case MemRegion::AllocaRegionKind:
646 case MemRegion::VarRegionKind:
647 case MemRegion::FieldRegionKind:
648 case MemRegion::ObjCIvarRegionKind:
649 // These are the types we can currently track string lengths for.
652 case MemRegion::ElementRegionKind:
653 // FIXME: Handle element regions by upper-bounding the parent region's
658 // Other regions (mostly non-data) can't have a reliable C string length.
659 // For now, just ignore the change.
660 // FIXME: These are rare but not impossible. We should output some kind of
661 // warning for things like strcpy((char[]){'a', 0}, "b");
665 if (strLength.isUnknown())
666 return state->remove<CStringLength>(MR);
668 return state->set<CStringLength>(MR, strLength);
671 SVal CStringChecker::getCStringLengthForRegion(CheckerContext &C,
672 ProgramStateRef &state,
677 // If there's a recorded length, go ahead and return it.
678 const SVal *Recorded = state->get<CStringLength>(MR);
683 // Otherwise, get a new symbol and update the state.
684 SValBuilder &svalBuilder = C.getSValBuilder();
685 QualType sizeTy = svalBuilder.getContext().getSizeType();
686 SVal strLength = svalBuilder.getMetadataSymbolVal(CStringChecker::getTag(),
688 C.getLocationContext(),
692 if (Optional<NonLoc> strLn = strLength.getAs<NonLoc>()) {
693 // In case of unbounded calls strlen etc bound the range to SIZE_MAX/4
694 BasicValueFactory &BVF = svalBuilder.getBasicValueFactory();
695 const llvm::APSInt &maxValInt = BVF.getMaxValue(sizeTy);
696 llvm::APSInt fourInt = APSIntType(maxValInt).getValue(4);
697 const llvm::APSInt *maxLengthInt = BVF.evalAPSInt(BO_Div, maxValInt,
699 NonLoc maxLength = svalBuilder.makeIntVal(*maxLengthInt);
700 SVal evalLength = svalBuilder.evalBinOpNN(state, BO_LE, *strLn,
702 state = state->assume(evalLength.castAs<DefinedOrUnknownSVal>(), true);
704 state = state->set<CStringLength>(MR, strLength);
710 SVal CStringChecker::getCStringLength(CheckerContext &C, ProgramStateRef &state,
711 const Expr *Ex, SVal Buf,
712 bool hypothetical) const {
713 const MemRegion *MR = Buf.getAsRegion();
715 // If we can't get a region, see if it's something we /know/ isn't a
716 // C string. In the context of locations, the only time we can issue such
717 // a warning is for labels.
718 if (Optional<loc::GotoLabel> Label = Buf.getAs<loc::GotoLabel>()) {
719 if (!Filter.CheckCStringNotNullTerm)
720 return UndefinedVal();
722 if (ExplodedNode *N = C.generateNonFatalErrorNode(state)) {
724 BT_NotCString.reset(new BuiltinBug(
725 Filter.CheckNameCStringNotNullTerm, categories::UnixAPI,
726 "Argument is not a null-terminated string."));
728 SmallString<120> buf;
729 llvm::raw_svector_ostream os(buf);
730 assert(CurrentFunctionDescription);
731 os << "Argument to " << CurrentFunctionDescription
732 << " is the address of the label '" << Label->getLabel()->getName()
733 << "', which is not a null-terminated string";
735 // Generate a report for this bug.
736 auto report = llvm::make_unique<BugReport>(*BT_NotCString, os.str(), N);
738 report->addRange(Ex->getSourceRange());
739 C.emitReport(std::move(report));
741 return UndefinedVal();
745 // If it's not a region and not a label, give up.
749 // If we have a region, strip casts from it and see if we can figure out
750 // its length. For anything we can't figure out, just return UnknownVal.
751 MR = MR->StripCasts();
753 switch (MR->getKind()) {
754 case MemRegion::StringRegionKind: {
755 // Modifying the contents of string regions is undefined [C99 6.4.5p6],
756 // so we can assume that the byte length is the correct C string length.
757 SValBuilder &svalBuilder = C.getSValBuilder();
758 QualType sizeTy = svalBuilder.getContext().getSizeType();
759 const StringLiteral *strLit = cast<StringRegion>(MR)->getStringLiteral();
760 return svalBuilder.makeIntVal(strLit->getByteLength(), sizeTy);
762 case MemRegion::SymbolicRegionKind:
763 case MemRegion::AllocaRegionKind:
764 case MemRegion::VarRegionKind:
765 case MemRegion::FieldRegionKind:
766 case MemRegion::ObjCIvarRegionKind:
767 return getCStringLengthForRegion(C, state, Ex, MR, hypothetical);
768 case MemRegion::CompoundLiteralRegionKind:
769 // FIXME: Can we track this? Is it necessary?
771 case MemRegion::ElementRegionKind:
772 // FIXME: How can we handle this? It's not good enough to subtract the
773 // offset from the base string length; consider "123\x00567" and &a[5].
776 // Other regions (mostly non-data) can't have a reliable C string length.
777 // In this case, an error is emitted and UndefinedVal is returned.
778 // The caller should always be prepared to handle this case.
779 if (!Filter.CheckCStringNotNullTerm)
780 return UndefinedVal();
782 if (ExplodedNode *N = C.generateNonFatalErrorNode(state)) {
784 BT_NotCString.reset(new BuiltinBug(
785 Filter.CheckNameCStringNotNullTerm, categories::UnixAPI,
786 "Argument is not a null-terminated string."));
788 SmallString<120> buf;
789 llvm::raw_svector_ostream os(buf);
791 assert(CurrentFunctionDescription);
792 os << "Argument to " << CurrentFunctionDescription << " is ";
794 if (SummarizeRegion(os, C.getASTContext(), MR))
795 os << ", which is not a null-terminated string";
797 os << "not a null-terminated string";
799 // Generate a report for this bug.
800 auto report = llvm::make_unique<BugReport>(*BT_NotCString, os.str(), N);
802 report->addRange(Ex->getSourceRange());
803 C.emitReport(std::move(report));
806 return UndefinedVal();
810 const StringLiteral *CStringChecker::getCStringLiteral(CheckerContext &C,
811 ProgramStateRef &state, const Expr *expr, SVal val) const {
813 // Get the memory region pointed to by the val.
814 const MemRegion *bufRegion = val.getAsRegion();
818 // Strip casts off the memory region.
819 bufRegion = bufRegion->StripCasts();
821 // Cast the memory region to a string region.
822 const StringRegion *strRegion= dyn_cast<StringRegion>(bufRegion);
826 // Return the actual string in the string region.
827 return strRegion->getStringLiteral();
830 bool CStringChecker::IsFirstBufInBound(CheckerContext &C,
831 ProgramStateRef state,
832 const Expr *FirstBuf,
834 // If we do not know that the buffer is long enough we return 'true'.
835 // Otherwise the parent region of this field region would also get
836 // invalidated, which would lead to warnings based on an unknown state.
838 // Originally copied from CheckBufferAccess and CheckLocation.
839 SValBuilder &svalBuilder = C.getSValBuilder();
840 ASTContext &Ctx = svalBuilder.getContext();
841 const LocationContext *LCtx = C.getLocationContext();
843 QualType sizeTy = Size->getType();
844 QualType PtrTy = Ctx.getPointerType(Ctx.CharTy);
845 SVal BufVal = state->getSVal(FirstBuf, LCtx);
847 SVal LengthVal = state->getSVal(Size, LCtx);
848 Optional<NonLoc> Length = LengthVal.getAs<NonLoc>();
850 return true; // cf top comment.
852 // Compute the offset of the last element to be accessed: size-1.
853 NonLoc One = svalBuilder.makeIntVal(1, sizeTy).castAs<NonLoc>();
855 svalBuilder.evalBinOpNN(state, BO_Sub, *Length, One, sizeTy)
858 // Check that the first buffer is sufficiently long.
859 SVal BufStart = svalBuilder.evalCast(BufVal, PtrTy, FirstBuf->getType());
860 Optional<Loc> BufLoc = BufStart.getAs<Loc>();
862 return true; // cf top comment.
865 svalBuilder.evalBinOpLN(state, BO_Add, *BufLoc, LastOffset, PtrTy);
867 // Check for out of bound array element access.
868 const MemRegion *R = BufEnd.getAsRegion();
870 return true; // cf top comment.
872 const ElementRegion *ER = dyn_cast<ElementRegion>(R);
874 return true; // cf top comment.
876 assert(ER->getValueType() == C.getASTContext().CharTy &&
877 "IsFirstBufInBound should only be called with char* ElementRegions");
879 // Get the size of the array.
880 const SubRegion *superReg = cast<SubRegion>(ER->getSuperRegion());
882 svalBuilder.convertToArrayIndex(superReg->getExtent(svalBuilder));
883 DefinedOrUnknownSVal ExtentSize = Extent.castAs<DefinedOrUnknownSVal>();
885 // Get the index of the accessed element.
886 DefinedOrUnknownSVal Idx = ER->getIndex().castAs<DefinedOrUnknownSVal>();
888 ProgramStateRef StInBound = state->assumeInBound(Idx, ExtentSize, true);
890 return static_cast<bool>(StInBound);
893 ProgramStateRef CStringChecker::InvalidateBuffer(CheckerContext &C,
894 ProgramStateRef state,
895 const Expr *E, SVal V,
898 Optional<Loc> L = V.getAs<Loc>();
902 // FIXME: This is a simplified version of what's in CFRefCount.cpp -- it makes
903 // some assumptions about the value that CFRefCount can't. Even so, it should
904 // probably be refactored.
905 if (Optional<loc::MemRegionVal> MR = L->getAs<loc::MemRegionVal>()) {
906 const MemRegion *R = MR->getRegion()->StripCasts();
908 // Are we dealing with an ElementRegion? If so, we should be invalidating
910 if (const ElementRegion *ER = dyn_cast<ElementRegion>(R)) {
911 R = ER->getSuperRegion();
912 // FIXME: What about layers of ElementRegions?
915 // Invalidate this region.
916 const LocationContext *LCtx = C.getPredecessor()->getLocationContext();
918 bool CausesPointerEscape = false;
919 RegionAndSymbolInvalidationTraits ITraits;
920 // Invalidate and escape only indirect regions accessible through the source
922 if (IsSourceBuffer) {
923 ITraits.setTrait(R->getBaseRegion(),
924 RegionAndSymbolInvalidationTraits::TK_PreserveContents);
925 ITraits.setTrait(R, RegionAndSymbolInvalidationTraits::TK_SuppressEscape);
926 CausesPointerEscape = true;
928 const MemRegion::Kind& K = R->getKind();
929 if (K == MemRegion::FieldRegionKind)
930 if (Size && IsFirstBufInBound(C, state, E, Size)) {
931 // If destination buffer is a field region and access is in bound,
932 // do not invalidate its super region.
935 RegionAndSymbolInvalidationTraits::TK_DoNotInvalidateSuperRegion);
939 return state->invalidateRegions(R, E, C.blockCount(), LCtx,
940 CausesPointerEscape, nullptr, nullptr,
944 // If we have a non-region value by chance, just remove the binding.
945 // FIXME: is this necessary or correct? This handles the non-Region
946 // cases. Is it ever valid to store to these?
947 return state->killBinding(*L);
950 bool CStringChecker::SummarizeRegion(raw_ostream &os, ASTContext &Ctx,
951 const MemRegion *MR) {
952 const TypedValueRegion *TVR = dyn_cast<TypedValueRegion>(MR);
954 switch (MR->getKind()) {
955 case MemRegion::FunctionCodeRegionKind: {
956 const NamedDecl *FD = cast<FunctionCodeRegion>(MR)->getDecl();
958 os << "the address of the function '" << *FD << '\'';
960 os << "the address of a function";
963 case MemRegion::BlockCodeRegionKind:
966 case MemRegion::BlockDataRegionKind:
969 case MemRegion::CXXThisRegionKind:
970 case MemRegion::CXXTempObjectRegionKind:
971 os << "a C++ temp object of type " << TVR->getValueType().getAsString();
973 case MemRegion::VarRegionKind:
974 os << "a variable of type" << TVR->getValueType().getAsString();
976 case MemRegion::FieldRegionKind:
977 os << "a field of type " << TVR->getValueType().getAsString();
979 case MemRegion::ObjCIvarRegionKind:
980 os << "an instance variable of type " << TVR->getValueType().getAsString();
987 //===----------------------------------------------------------------------===//
988 // evaluation of individual function calls.
989 //===----------------------------------------------------------------------===//
991 void CStringChecker::evalCopyCommon(CheckerContext &C,
993 ProgramStateRef state,
994 const Expr *Size, const Expr *Dest,
995 const Expr *Source, bool Restricted,
996 bool IsMempcpy) const {
997 CurrentFunctionDescription = "memory copy function";
999 // See if the size argument is zero.
1000 const LocationContext *LCtx = C.getLocationContext();
1001 SVal sizeVal = state->getSVal(Size, LCtx);
1002 QualType sizeTy = Size->getType();
1004 ProgramStateRef stateZeroSize, stateNonZeroSize;
1005 std::tie(stateZeroSize, stateNonZeroSize) =
1006 assumeZero(C, state, sizeVal, sizeTy);
1008 // Get the value of the Dest.
1009 SVal destVal = state->getSVal(Dest, LCtx);
1011 // If the size is zero, there won't be any actual memory access, so
1012 // just bind the return value to the destination buffer and return.
1013 if (stateZeroSize && !stateNonZeroSize) {
1014 stateZeroSize = stateZeroSize->BindExpr(CE, LCtx, destVal);
1015 C.addTransition(stateZeroSize);
1019 // If the size can be nonzero, we have to check the other arguments.
1020 if (stateNonZeroSize) {
1021 state = stateNonZeroSize;
1023 // Ensure the destination is not null. If it is NULL there will be a
1024 // NULL pointer dereference.
1025 state = checkNonNull(C, state, Dest, destVal);
1029 // Get the value of the Src.
1030 SVal srcVal = state->getSVal(Source, LCtx);
1032 // Ensure the source is not null. If it is NULL there will be a
1033 // NULL pointer dereference.
1034 state = checkNonNull(C, state, Source, srcVal);
1038 // Ensure the accesses are valid and that the buffers do not overlap.
1039 const char * const writeWarning =
1040 "Memory copy function overflows destination buffer";
1041 state = CheckBufferAccess(C, state, Size, Dest, Source,
1042 writeWarning, /* sourceWarning = */ nullptr);
1044 state = CheckOverlap(C, state, Size, Dest, Source);
1049 // If this is mempcpy, get the byte after the last byte copied and
1052 loc::MemRegionVal destRegVal = destVal.castAs<loc::MemRegionVal>();
1054 // Get the length to copy.
1055 if (Optional<NonLoc> lenValNonLoc = sizeVal.getAs<NonLoc>()) {
1056 // Get the byte after the last byte copied.
1057 SValBuilder &SvalBuilder = C.getSValBuilder();
1058 ASTContext &Ctx = SvalBuilder.getContext();
1059 QualType CharPtrTy = Ctx.getPointerType(Ctx.CharTy);
1060 loc::MemRegionVal DestRegCharVal = SvalBuilder.evalCast(destRegVal,
1061 CharPtrTy, Dest->getType()).castAs<loc::MemRegionVal>();
1062 SVal lastElement = C.getSValBuilder().evalBinOpLN(state, BO_Add,
1067 // The byte after the last byte copied is the return value.
1068 state = state->BindExpr(CE, LCtx, lastElement);
1070 // If we don't know how much we copied, we can at least
1071 // conjure a return value for later.
1072 SVal result = C.getSValBuilder().conjureSymbolVal(nullptr, CE, LCtx,
1074 state = state->BindExpr(CE, LCtx, result);
1078 // All other copies return the destination buffer.
1079 // (Well, bcopy() has a void return type, but this won't hurt.)
1080 state = state->BindExpr(CE, LCtx, destVal);
1083 // Invalidate the destination (regular invalidation without pointer-escaping
1084 // the address of the top-level region).
1085 // FIXME: Even if we can't perfectly model the copy, we should see if we
1086 // can use LazyCompoundVals to copy the source values into the destination.
1087 // This would probably remove any existing bindings past the end of the
1088 // copied region, but that's still an improvement over blank invalidation.
1089 state = InvalidateBuffer(C, state, Dest, C.getSVal(Dest),
1090 /*IsSourceBuffer*/false, Size);
1092 // Invalidate the source (const-invalidation without const-pointer-escaping
1093 // the address of the top-level region).
1094 state = InvalidateBuffer(C, state, Source, C.getSVal(Source),
1095 /*IsSourceBuffer*/true, nullptr);
1097 C.addTransition(state);
1102 void CStringChecker::evalMemcpy(CheckerContext &C, const CallExpr *CE) const {
1103 if (CE->getNumArgs() < 3)
1106 // void *memcpy(void *restrict dst, const void *restrict src, size_t n);
1107 // The return value is the address of the destination buffer.
1108 const Expr *Dest = CE->getArg(0);
1109 ProgramStateRef state = C.getState();
1111 evalCopyCommon(C, CE, state, CE->getArg(2), Dest, CE->getArg(1), true);
1114 void CStringChecker::evalMempcpy(CheckerContext &C, const CallExpr *CE) const {
1115 if (CE->getNumArgs() < 3)
1118 // void *mempcpy(void *restrict dst, const void *restrict src, size_t n);
1119 // The return value is a pointer to the byte following the last written byte.
1120 const Expr *Dest = CE->getArg(0);
1121 ProgramStateRef state = C.getState();
1123 evalCopyCommon(C, CE, state, CE->getArg(2), Dest, CE->getArg(1), true, true);
1126 void CStringChecker::evalMemmove(CheckerContext &C, const CallExpr *CE) const {
1127 if (CE->getNumArgs() < 3)
1130 // void *memmove(void *dst, const void *src, size_t n);
1131 // The return value is the address of the destination buffer.
1132 const Expr *Dest = CE->getArg(0);
1133 ProgramStateRef state = C.getState();
1135 evalCopyCommon(C, CE, state, CE->getArg(2), Dest, CE->getArg(1));
1138 void CStringChecker::evalBcopy(CheckerContext &C, const CallExpr *CE) const {
1139 if (CE->getNumArgs() < 3)
1142 // void bcopy(const void *src, void *dst, size_t n);
1143 evalCopyCommon(C, CE, C.getState(),
1144 CE->getArg(2), CE->getArg(1), CE->getArg(0));
1147 void CStringChecker::evalMemcmp(CheckerContext &C, const CallExpr *CE) const {
1148 if (CE->getNumArgs() < 3)
1151 // int memcmp(const void *s1, const void *s2, size_t n);
1152 CurrentFunctionDescription = "memory comparison function";
1154 const Expr *Left = CE->getArg(0);
1155 const Expr *Right = CE->getArg(1);
1156 const Expr *Size = CE->getArg(2);
1158 ProgramStateRef state = C.getState();
1159 SValBuilder &svalBuilder = C.getSValBuilder();
1161 // See if the size argument is zero.
1162 const LocationContext *LCtx = C.getLocationContext();
1163 SVal sizeVal = state->getSVal(Size, LCtx);
1164 QualType sizeTy = Size->getType();
1166 ProgramStateRef stateZeroSize, stateNonZeroSize;
1167 std::tie(stateZeroSize, stateNonZeroSize) =
1168 assumeZero(C, state, sizeVal, sizeTy);
1170 // If the size can be zero, the result will be 0 in that case, and we don't
1171 // have to check either of the buffers.
1172 if (stateZeroSize) {
1173 state = stateZeroSize;
1174 state = state->BindExpr(CE, LCtx,
1175 svalBuilder.makeZeroVal(CE->getType()));
1176 C.addTransition(state);
1179 // If the size can be nonzero, we have to check the other arguments.
1180 if (stateNonZeroSize) {
1181 state = stateNonZeroSize;
1182 // If we know the two buffers are the same, we know the result is 0.
1183 // First, get the two buffers' addresses. Another checker will have already
1184 // made sure they're not undefined.
1185 DefinedOrUnknownSVal LV =
1186 state->getSVal(Left, LCtx).castAs<DefinedOrUnknownSVal>();
1187 DefinedOrUnknownSVal RV =
1188 state->getSVal(Right, LCtx).castAs<DefinedOrUnknownSVal>();
1190 // See if they are the same.
1191 DefinedOrUnknownSVal SameBuf = svalBuilder.evalEQ(state, LV, RV);
1192 ProgramStateRef StSameBuf, StNotSameBuf;
1193 std::tie(StSameBuf, StNotSameBuf) = state->assume(SameBuf);
1195 // If the two arguments might be the same buffer, we know the result is 0,
1196 // and we only need to check one size.
1199 state = CheckBufferAccess(C, state, Size, Left);
1201 state = StSameBuf->BindExpr(CE, LCtx,
1202 svalBuilder.makeZeroVal(CE->getType()));
1203 C.addTransition(state);
1207 // If the two arguments might be different buffers, we have to check the
1208 // size of both of them.
1210 state = StNotSameBuf;
1211 state = CheckBufferAccess(C, state, Size, Left, Right);
1213 // The return value is the comparison result, which we don't know.
1214 SVal CmpV = svalBuilder.conjureSymbolVal(nullptr, CE, LCtx,
1216 state = state->BindExpr(CE, LCtx, CmpV);
1217 C.addTransition(state);
1223 void CStringChecker::evalstrLength(CheckerContext &C,
1224 const CallExpr *CE) const {
1225 if (CE->getNumArgs() < 1)
1228 // size_t strlen(const char *s);
1229 evalstrLengthCommon(C, CE, /* IsStrnlen = */ false);
1232 void CStringChecker::evalstrnLength(CheckerContext &C,
1233 const CallExpr *CE) const {
1234 if (CE->getNumArgs() < 2)
1237 // size_t strnlen(const char *s, size_t maxlen);
1238 evalstrLengthCommon(C, CE, /* IsStrnlen = */ true);
1241 void CStringChecker::evalstrLengthCommon(CheckerContext &C, const CallExpr *CE,
1242 bool IsStrnlen) const {
1243 CurrentFunctionDescription = "string length function";
1244 ProgramStateRef state = C.getState();
1245 const LocationContext *LCtx = C.getLocationContext();
1248 const Expr *maxlenExpr = CE->getArg(1);
1249 SVal maxlenVal = state->getSVal(maxlenExpr, LCtx);
1251 ProgramStateRef stateZeroSize, stateNonZeroSize;
1252 std::tie(stateZeroSize, stateNonZeroSize) =
1253 assumeZero(C, state, maxlenVal, maxlenExpr->getType());
1255 // If the size can be zero, the result will be 0 in that case, and we don't
1256 // have to check the string itself.
1257 if (stateZeroSize) {
1258 SVal zero = C.getSValBuilder().makeZeroVal(CE->getType());
1259 stateZeroSize = stateZeroSize->BindExpr(CE, LCtx, zero);
1260 C.addTransition(stateZeroSize);
1263 // If the size is GUARANTEED to be zero, we're done!
1264 if (!stateNonZeroSize)
1267 // Otherwise, record the assumption that the size is nonzero.
1268 state = stateNonZeroSize;
1271 // Check that the string argument is non-null.
1272 const Expr *Arg = CE->getArg(0);
1273 SVal ArgVal = state->getSVal(Arg, LCtx);
1275 state = checkNonNull(C, state, Arg, ArgVal);
1280 SVal strLength = getCStringLength(C, state, Arg, ArgVal);
1282 // If the argument isn't a valid C string, there's no valid state to
1284 if (strLength.isUndef())
1287 DefinedOrUnknownSVal result = UnknownVal();
1289 // If the check is for strnlen() then bind the return value to no more than
1290 // the maxlen value.
1292 QualType cmpTy = C.getSValBuilder().getConditionType();
1294 // It's a little unfortunate to be getting this again,
1295 // but it's not that expensive...
1296 const Expr *maxlenExpr = CE->getArg(1);
1297 SVal maxlenVal = state->getSVal(maxlenExpr, LCtx);
1299 Optional<NonLoc> strLengthNL = strLength.getAs<NonLoc>();
1300 Optional<NonLoc> maxlenValNL = maxlenVal.getAs<NonLoc>();
1302 if (strLengthNL && maxlenValNL) {
1303 ProgramStateRef stateStringTooLong, stateStringNotTooLong;
1305 // Check if the strLength is greater than the maxlen.
1306 std::tie(stateStringTooLong, stateStringNotTooLong) = state->assume(
1308 .evalBinOpNN(state, BO_GT, *strLengthNL, *maxlenValNL, cmpTy)
1309 .castAs<DefinedOrUnknownSVal>());
1311 if (stateStringTooLong && !stateStringNotTooLong) {
1312 // If the string is longer than maxlen, return maxlen.
1313 result = *maxlenValNL;
1314 } else if (stateStringNotTooLong && !stateStringTooLong) {
1315 // If the string is shorter than maxlen, return its length.
1316 result = *strLengthNL;
1320 if (result.isUnknown()) {
1321 // If we don't have enough information for a comparison, there's
1322 // no guarantee the full string length will actually be returned.
1323 // All we know is the return value is the min of the string length
1324 // and the limit. This is better than nothing.
1325 result = C.getSValBuilder().conjureSymbolVal(nullptr, CE, LCtx,
1327 NonLoc resultNL = result.castAs<NonLoc>();
1330 state = state->assume(C.getSValBuilder().evalBinOpNN(
1331 state, BO_LE, resultNL, *strLengthNL, cmpTy)
1332 .castAs<DefinedOrUnknownSVal>(), true);
1336 state = state->assume(C.getSValBuilder().evalBinOpNN(
1337 state, BO_LE, resultNL, *maxlenValNL, cmpTy)
1338 .castAs<DefinedOrUnknownSVal>(), true);
1343 // This is a plain strlen(), not strnlen().
1344 result = strLength.castAs<DefinedOrUnknownSVal>();
1346 // If we don't know the length of the string, conjure a return
1347 // value, so it can be used in constraints, at least.
1348 if (result.isUnknown()) {
1349 result = C.getSValBuilder().conjureSymbolVal(nullptr, CE, LCtx,
1354 // Bind the return value.
1355 assert(!result.isUnknown() && "Should have conjured a value by now");
1356 state = state->BindExpr(CE, LCtx, result);
1357 C.addTransition(state);
1360 void CStringChecker::evalStrcpy(CheckerContext &C, const CallExpr *CE) const {
1361 if (CE->getNumArgs() < 2)
1364 // char *strcpy(char *restrict dst, const char *restrict src);
1365 evalStrcpyCommon(C, CE,
1366 /* returnEnd = */ false,
1367 /* isBounded = */ false,
1368 /* isAppending = */ false);
1371 void CStringChecker::evalStrncpy(CheckerContext &C, const CallExpr *CE) const {
1372 if (CE->getNumArgs() < 3)
1375 // char *strncpy(char *restrict dst, const char *restrict src, size_t n);
1376 evalStrcpyCommon(C, CE,
1377 /* returnEnd = */ false,
1378 /* isBounded = */ true,
1379 /* isAppending = */ false);
1382 void CStringChecker::evalStpcpy(CheckerContext &C, const CallExpr *CE) const {
1383 if (CE->getNumArgs() < 2)
1386 // char *stpcpy(char *restrict dst, const char *restrict src);
1387 evalStrcpyCommon(C, CE,
1388 /* returnEnd = */ true,
1389 /* isBounded = */ false,
1390 /* isAppending = */ false);
1393 void CStringChecker::evalStrcat(CheckerContext &C, const CallExpr *CE) const {
1394 if (CE->getNumArgs() < 2)
1397 //char *strcat(char *restrict s1, const char *restrict s2);
1398 evalStrcpyCommon(C, CE,
1399 /* returnEnd = */ false,
1400 /* isBounded = */ false,
1401 /* isAppending = */ true);
1404 void CStringChecker::evalStrncat(CheckerContext &C, const CallExpr *CE) const {
1405 if (CE->getNumArgs() < 3)
1408 //char *strncat(char *restrict s1, const char *restrict s2, size_t n);
1409 evalStrcpyCommon(C, CE,
1410 /* returnEnd = */ false,
1411 /* isBounded = */ true,
1412 /* isAppending = */ true);
1415 void CStringChecker::evalStrcpyCommon(CheckerContext &C, const CallExpr *CE,
1416 bool returnEnd, bool isBounded,
1417 bool isAppending) const {
1418 CurrentFunctionDescription = "string copy function";
1419 ProgramStateRef state = C.getState();
1420 const LocationContext *LCtx = C.getLocationContext();
1422 // Check that the destination is non-null.
1423 const Expr *Dst = CE->getArg(0);
1424 SVal DstVal = state->getSVal(Dst, LCtx);
1426 state = checkNonNull(C, state, Dst, DstVal);
1430 // Check that the source is non-null.
1431 const Expr *srcExpr = CE->getArg(1);
1432 SVal srcVal = state->getSVal(srcExpr, LCtx);
1433 state = checkNonNull(C, state, srcExpr, srcVal);
1437 // Get the string length of the source.
1438 SVal strLength = getCStringLength(C, state, srcExpr, srcVal);
1440 // If the source isn't a valid C string, give up.
1441 if (strLength.isUndef())
1444 SValBuilder &svalBuilder = C.getSValBuilder();
1445 QualType cmpTy = svalBuilder.getConditionType();
1446 QualType sizeTy = svalBuilder.getContext().getSizeType();
1448 // These two values allow checking two kinds of errors:
1449 // - actual overflows caused by a source that doesn't fit in the destination
1450 // - potential overflows caused by a bound that could exceed the destination
1451 SVal amountCopied = UnknownVal();
1452 SVal maxLastElementIndex = UnknownVal();
1453 const char *boundWarning = nullptr;
1455 // If the function is strncpy, strncat, etc... it is bounded.
1457 // Get the max number of characters to copy.
1458 const Expr *lenExpr = CE->getArg(2);
1459 SVal lenVal = state->getSVal(lenExpr, LCtx);
1461 // Protect against misdeclared strncpy().
1462 lenVal = svalBuilder.evalCast(lenVal, sizeTy, lenExpr->getType());
1464 Optional<NonLoc> strLengthNL = strLength.getAs<NonLoc>();
1465 Optional<NonLoc> lenValNL = lenVal.getAs<NonLoc>();
1467 // If we know both values, we might be able to figure out how much
1469 if (strLengthNL && lenValNL) {
1470 ProgramStateRef stateSourceTooLong, stateSourceNotTooLong;
1472 // Check if the max number to copy is less than the length of the src.
1473 // If the bound is equal to the source length, strncpy won't null-
1474 // terminate the result!
1475 std::tie(stateSourceTooLong, stateSourceNotTooLong) = state->assume(
1476 svalBuilder.evalBinOpNN(state, BO_GE, *strLengthNL, *lenValNL, cmpTy)
1477 .castAs<DefinedOrUnknownSVal>());
1479 if (stateSourceTooLong && !stateSourceNotTooLong) {
1480 // Max number to copy is less than the length of the src, so the actual
1481 // strLength copied is the max number arg.
1482 state = stateSourceTooLong;
1483 amountCopied = lenVal;
1485 } else if (!stateSourceTooLong && stateSourceNotTooLong) {
1486 // The source buffer entirely fits in the bound.
1487 state = stateSourceNotTooLong;
1488 amountCopied = strLength;
1492 // We still want to know if the bound is known to be too large.
1495 // For strncat, the check is strlen(dst) + lenVal < sizeof(dst)
1497 // Get the string length of the destination. If the destination is
1498 // memory that can't have a string length, we shouldn't be copying
1500 SVal dstStrLength = getCStringLength(C, state, Dst, DstVal);
1501 if (dstStrLength.isUndef())
1504 if (Optional<NonLoc> dstStrLengthNL = dstStrLength.getAs<NonLoc>()) {
1505 maxLastElementIndex = svalBuilder.evalBinOpNN(state, BO_Add,
1509 boundWarning = "Size argument is greater than the free space in the "
1510 "destination buffer";
1514 // For strncpy, this is just checking that lenVal <= sizeof(dst)
1515 // (Yes, strncpy and strncat differ in how they treat termination.
1516 // strncat ALWAYS terminates, but strncpy doesn't.)
1518 // We need a special case for when the copy size is zero, in which
1519 // case strncpy will do no work at all. Our bounds check uses n-1
1520 // as the last element accessed, so n == 0 is problematic.
1521 ProgramStateRef StateZeroSize, StateNonZeroSize;
1522 std::tie(StateZeroSize, StateNonZeroSize) =
1523 assumeZero(C, state, *lenValNL, sizeTy);
1525 // If the size is known to be zero, we're done.
1526 if (StateZeroSize && !StateNonZeroSize) {
1527 StateZeroSize = StateZeroSize->BindExpr(CE, LCtx, DstVal);
1528 C.addTransition(StateZeroSize);
1532 // Otherwise, go ahead and figure out the last element we'll touch.
1533 // We don't record the non-zero assumption here because we can't
1534 // be sure. We won't warn on a possible zero.
1535 NonLoc one = svalBuilder.makeIntVal(1, sizeTy).castAs<NonLoc>();
1536 maxLastElementIndex = svalBuilder.evalBinOpNN(state, BO_Sub, *lenValNL,
1538 boundWarning = "Size argument is greater than the length of the "
1539 "destination buffer";
1543 // If we couldn't pin down the copy length, at least bound it.
1544 // FIXME: We should actually run this code path for append as well, but
1545 // right now it creates problems with constraints (since we can end up
1546 // trying to pass constraints from symbol to symbol).
1547 if (amountCopied.isUnknown() && !isAppending) {
1548 // Try to get a "hypothetical" string length symbol, which we can later
1549 // set as a real value if that turns out to be the case.
1550 amountCopied = getCStringLength(C, state, lenExpr, srcVal, true);
1551 assert(!amountCopied.isUndef());
1553 if (Optional<NonLoc> amountCopiedNL = amountCopied.getAs<NonLoc>()) {
1555 // amountCopied <= lenVal
1556 SVal copiedLessThanBound = svalBuilder.evalBinOpNN(state, BO_LE,
1560 state = state->assume(
1561 copiedLessThanBound.castAs<DefinedOrUnknownSVal>(), true);
1567 // amountCopied <= strlen(source)
1568 SVal copiedLessThanSrc = svalBuilder.evalBinOpNN(state, BO_LE,
1572 state = state->assume(
1573 copiedLessThanSrc.castAs<DefinedOrUnknownSVal>(), true);
1581 // The function isn't bounded. The amount copied should match the length
1582 // of the source buffer.
1583 amountCopied = strLength;
1588 // This represents the number of characters copied into the destination
1589 // buffer. (It may not actually be the strlen if the destination buffer
1590 // is not terminated.)
1591 SVal finalStrLength = UnknownVal();
1593 // If this is an appending function (strcat, strncat...) then set the
1594 // string length to strlen(src) + strlen(dst) since the buffer will
1595 // ultimately contain both.
1597 // Get the string length of the destination. If the destination is memory
1598 // that can't have a string length, we shouldn't be copying into it anyway.
1599 SVal dstStrLength = getCStringLength(C, state, Dst, DstVal);
1600 if (dstStrLength.isUndef())
1603 Optional<NonLoc> srcStrLengthNL = amountCopied.getAs<NonLoc>();
1604 Optional<NonLoc> dstStrLengthNL = dstStrLength.getAs<NonLoc>();
1606 // If we know both string lengths, we might know the final string length.
1607 if (srcStrLengthNL && dstStrLengthNL) {
1608 // Make sure the two lengths together don't overflow a size_t.
1609 state = checkAdditionOverflow(C, state, *srcStrLengthNL, *dstStrLengthNL);
1613 finalStrLength = svalBuilder.evalBinOpNN(state, BO_Add, *srcStrLengthNL,
1614 *dstStrLengthNL, sizeTy);
1617 // If we couldn't get a single value for the final string length,
1618 // we can at least bound it by the individual lengths.
1619 if (finalStrLength.isUnknown()) {
1620 // Try to get a "hypothetical" string length symbol, which we can later
1621 // set as a real value if that turns out to be the case.
1622 finalStrLength = getCStringLength(C, state, CE, DstVal, true);
1623 assert(!finalStrLength.isUndef());
1625 if (Optional<NonLoc> finalStrLengthNL = finalStrLength.getAs<NonLoc>()) {
1626 if (srcStrLengthNL) {
1627 // finalStrLength >= srcStrLength
1628 SVal sourceInResult = svalBuilder.evalBinOpNN(state, BO_GE,
1632 state = state->assume(sourceInResult.castAs<DefinedOrUnknownSVal>(),
1638 if (dstStrLengthNL) {
1639 // finalStrLength >= dstStrLength
1640 SVal destInResult = svalBuilder.evalBinOpNN(state, BO_GE,
1645 state->assume(destInResult.castAs<DefinedOrUnknownSVal>(), true);
1653 // Otherwise, this is a copy-over function (strcpy, strncpy, ...), and
1654 // the final string length will match the input string length.
1655 finalStrLength = amountCopied;
1658 // The final result of the function will either be a pointer past the last
1659 // copied element, or a pointer to the start of the destination buffer.
1660 SVal Result = (returnEnd ? UnknownVal() : DstVal);
1664 // If the destination is a MemRegion, try to check for a buffer overflow and
1665 // record the new string length.
1666 if (Optional<loc::MemRegionVal> dstRegVal =
1667 DstVal.getAs<loc::MemRegionVal>()) {
1668 QualType ptrTy = Dst->getType();
1670 // If we have an exact value on a bounded copy, use that to check for
1671 // overflows, rather than our estimate about how much is actually copied.
1673 if (Optional<NonLoc> maxLastNL = maxLastElementIndex.getAs<NonLoc>()) {
1674 SVal maxLastElement = svalBuilder.evalBinOpLN(state, BO_Add, *dstRegVal,
1676 state = CheckLocation(C, state, CE->getArg(2), maxLastElement,
1683 // Then, if the final length is known...
1684 if (Optional<NonLoc> knownStrLength = finalStrLength.getAs<NonLoc>()) {
1685 SVal lastElement = svalBuilder.evalBinOpLN(state, BO_Add, *dstRegVal,
1686 *knownStrLength, ptrTy);
1688 // ...and we haven't checked the bound, we'll check the actual copy.
1689 if (!boundWarning) {
1690 const char * const warningMsg =
1691 "String copy function overflows destination buffer";
1692 state = CheckLocation(C, state, Dst, lastElement, warningMsg);
1697 // If this is a stpcpy-style copy, the last element is the return value.
1699 Result = lastElement;
1702 // Invalidate the destination (regular invalidation without pointer-escaping
1703 // the address of the top-level region). This must happen before we set the
1704 // C string length because invalidation will clear the length.
1705 // FIXME: Even if we can't perfectly model the copy, we should see if we
1706 // can use LazyCompoundVals to copy the source values into the destination.
1707 // This would probably remove any existing bindings past the end of the
1708 // string, but that's still an improvement over blank invalidation.
1709 state = InvalidateBuffer(C, state, Dst, *dstRegVal,
1710 /*IsSourceBuffer*/false, nullptr);
1712 // Invalidate the source (const-invalidation without const-pointer-escaping
1713 // the address of the top-level region).
1714 state = InvalidateBuffer(C, state, srcExpr, srcVal, /*IsSourceBuffer*/true,
1717 // Set the C string length of the destination, if we know it.
1718 if (isBounded && !isAppending) {
1719 // strncpy is annoying in that it doesn't guarantee to null-terminate
1720 // the result string. If the original string didn't fit entirely inside
1721 // the bound (including the null-terminator), we don't know how long the
1723 if (amountCopied != strLength)
1724 finalStrLength = UnknownVal();
1726 state = setCStringLength(state, dstRegVal->getRegion(), finalStrLength);
1731 // If this is a stpcpy-style copy, but we were unable to check for a buffer
1732 // overflow, we still need a result. Conjure a return value.
1733 if (returnEnd && Result.isUnknown()) {
1734 Result = svalBuilder.conjureSymbolVal(nullptr, CE, LCtx, C.blockCount());
1737 // Set the return value.
1738 state = state->BindExpr(CE, LCtx, Result);
1739 C.addTransition(state);
1742 void CStringChecker::evalStrcmp(CheckerContext &C, const CallExpr *CE) const {
1743 if (CE->getNumArgs() < 2)
1746 //int strcmp(const char *s1, const char *s2);
1747 evalStrcmpCommon(C, CE, /* isBounded = */ false, /* ignoreCase = */ false);
1750 void CStringChecker::evalStrncmp(CheckerContext &C, const CallExpr *CE) const {
1751 if (CE->getNumArgs() < 3)
1754 //int strncmp(const char *s1, const char *s2, size_t n);
1755 evalStrcmpCommon(C, CE, /* isBounded = */ true, /* ignoreCase = */ false);
1758 void CStringChecker::evalStrcasecmp(CheckerContext &C,
1759 const CallExpr *CE) const {
1760 if (CE->getNumArgs() < 2)
1763 //int strcasecmp(const char *s1, const char *s2);
1764 evalStrcmpCommon(C, CE, /* isBounded = */ false, /* ignoreCase = */ true);
1767 void CStringChecker::evalStrncasecmp(CheckerContext &C,
1768 const CallExpr *CE) const {
1769 if (CE->getNumArgs() < 3)
1772 //int strncasecmp(const char *s1, const char *s2, size_t n);
1773 evalStrcmpCommon(C, CE, /* isBounded = */ true, /* ignoreCase = */ true);
1776 void CStringChecker::evalStrcmpCommon(CheckerContext &C, const CallExpr *CE,
1777 bool isBounded, bool ignoreCase) const {
1778 CurrentFunctionDescription = "string comparison function";
1779 ProgramStateRef state = C.getState();
1780 const LocationContext *LCtx = C.getLocationContext();
1782 // Check that the first string is non-null
1783 const Expr *s1 = CE->getArg(0);
1784 SVal s1Val = state->getSVal(s1, LCtx);
1785 state = checkNonNull(C, state, s1, s1Val);
1789 // Check that the second string is non-null.
1790 const Expr *s2 = CE->getArg(1);
1791 SVal s2Val = state->getSVal(s2, LCtx);
1792 state = checkNonNull(C, state, s2, s2Val);
1796 // Get the string length of the first string or give up.
1797 SVal s1Length = getCStringLength(C, state, s1, s1Val);
1798 if (s1Length.isUndef())
1801 // Get the string length of the second string or give up.
1802 SVal s2Length = getCStringLength(C, state, s2, s2Val);
1803 if (s2Length.isUndef())
1806 // If we know the two buffers are the same, we know the result is 0.
1807 // First, get the two buffers' addresses. Another checker will have already
1808 // made sure they're not undefined.
1809 DefinedOrUnknownSVal LV = s1Val.castAs<DefinedOrUnknownSVal>();
1810 DefinedOrUnknownSVal RV = s2Val.castAs<DefinedOrUnknownSVal>();
1812 // See if they are the same.
1813 SValBuilder &svalBuilder = C.getSValBuilder();
1814 DefinedOrUnknownSVal SameBuf = svalBuilder.evalEQ(state, LV, RV);
1815 ProgramStateRef StSameBuf, StNotSameBuf;
1816 std::tie(StSameBuf, StNotSameBuf) = state->assume(SameBuf);
1818 // If the two arguments might be the same buffer, we know the result is 0,
1819 // and we only need to check one size.
1821 StSameBuf = StSameBuf->BindExpr(CE, LCtx,
1822 svalBuilder.makeZeroVal(CE->getType()));
1823 C.addTransition(StSameBuf);
1825 // If the two arguments are GUARANTEED to be the same, we're done!
1830 assert(StNotSameBuf);
1831 state = StNotSameBuf;
1833 // At this point we can go about comparing the two buffers.
1834 // For now, we only do this if they're both known string literals.
1836 // Attempt to extract string literals from both expressions.
1837 const StringLiteral *s1StrLiteral = getCStringLiteral(C, state, s1, s1Val);
1838 const StringLiteral *s2StrLiteral = getCStringLiteral(C, state, s2, s2Val);
1839 bool canComputeResult = false;
1840 SVal resultVal = svalBuilder.conjureSymbolVal(nullptr, CE, LCtx,
1843 if (s1StrLiteral && s2StrLiteral) {
1844 StringRef s1StrRef = s1StrLiteral->getString();
1845 StringRef s2StrRef = s2StrLiteral->getString();
1848 // Get the max number of characters to compare.
1849 const Expr *lenExpr = CE->getArg(2);
1850 SVal lenVal = state->getSVal(lenExpr, LCtx);
1852 // If the length is known, we can get the right substrings.
1853 if (const llvm::APSInt *len = svalBuilder.getKnownValue(state, lenVal)) {
1854 // Create substrings of each to compare the prefix.
1855 s1StrRef = s1StrRef.substr(0, (size_t)len->getZExtValue());
1856 s2StrRef = s2StrRef.substr(0, (size_t)len->getZExtValue());
1857 canComputeResult = true;
1860 // This is a normal, unbounded strcmp.
1861 canComputeResult = true;
1864 if (canComputeResult) {
1865 // Real strcmp stops at null characters.
1866 size_t s1Term = s1StrRef.find('\0');
1867 if (s1Term != StringRef::npos)
1868 s1StrRef = s1StrRef.substr(0, s1Term);
1870 size_t s2Term = s2StrRef.find('\0');
1871 if (s2Term != StringRef::npos)
1872 s2StrRef = s2StrRef.substr(0, s2Term);
1874 // Use StringRef's comparison methods to compute the actual result.
1875 int compareRes = ignoreCase ? s1StrRef.compare_lower(s2StrRef)
1876 : s1StrRef.compare(s2StrRef);
1878 // The strcmp function returns an integer greater than, equal to, or less
1879 // than zero, [c11, p7.24.4.2].
1880 if (compareRes == 0) {
1881 resultVal = svalBuilder.makeIntVal(compareRes, CE->getType());
1884 DefinedSVal zeroVal = svalBuilder.makeIntVal(0, CE->getType());
1885 // Constrain strcmp's result range based on the result of StringRef's
1886 // comparison methods.
1887 BinaryOperatorKind op = (compareRes == 1) ? BO_GT : BO_LT;
1888 SVal compareWithZero =
1889 svalBuilder.evalBinOp(state, op, resultVal, zeroVal,
1890 svalBuilder.getConditionType());
1891 DefinedSVal compareWithZeroVal = compareWithZero.castAs<DefinedSVal>();
1892 state = state->assume(compareWithZeroVal, true);
1897 state = state->BindExpr(CE, LCtx, resultVal);
1899 // Record this as a possible path.
1900 C.addTransition(state);
1903 void CStringChecker::evalStrsep(CheckerContext &C, const CallExpr *CE) const {
1904 //char *strsep(char **stringp, const char *delim);
1905 if (CE->getNumArgs() < 2)
1908 // Sanity: does the search string parameter match the return type?
1909 const Expr *SearchStrPtr = CE->getArg(0);
1910 QualType CharPtrTy = SearchStrPtr->getType()->getPointeeType();
1911 if (CharPtrTy.isNull() ||
1912 CE->getType().getUnqualifiedType() != CharPtrTy.getUnqualifiedType())
1915 CurrentFunctionDescription = "strsep()";
1916 ProgramStateRef State = C.getState();
1917 const LocationContext *LCtx = C.getLocationContext();
1919 // Check that the search string pointer is non-null (though it may point to
1921 SVal SearchStrVal = State->getSVal(SearchStrPtr, LCtx);
1922 State = checkNonNull(C, State, SearchStrPtr, SearchStrVal);
1926 // Check that the delimiter string is non-null.
1927 const Expr *DelimStr = CE->getArg(1);
1928 SVal DelimStrVal = State->getSVal(DelimStr, LCtx);
1929 State = checkNonNull(C, State, DelimStr, DelimStrVal);
1933 SValBuilder &SVB = C.getSValBuilder();
1935 if (Optional<Loc> SearchStrLoc = SearchStrVal.getAs<Loc>()) {
1936 // Get the current value of the search string pointer, as a char*.
1937 Result = State->getSVal(*SearchStrLoc, CharPtrTy);
1939 // Invalidate the search string, representing the change of one delimiter
1940 // character to NUL.
1941 State = InvalidateBuffer(C, State, SearchStrPtr, Result,
1942 /*IsSourceBuffer*/false, nullptr);
1944 // Overwrite the search string pointer. The new value is either an address
1945 // further along in the same string, or NULL if there are no more tokens.
1946 State = State->bindLoc(*SearchStrLoc,
1947 SVB.conjureSymbolVal(getTag(),
1954 assert(SearchStrVal.isUnknown());
1955 // Conjure a symbolic value. It's the best we can do.
1956 Result = SVB.conjureSymbolVal(nullptr, CE, LCtx, C.blockCount());
1959 // Set the return value, and finish.
1960 State = State->BindExpr(CE, LCtx, Result);
1961 C.addTransition(State);
1964 // These should probably be moved into a C++ standard library checker.
1965 void CStringChecker::evalStdCopy(CheckerContext &C, const CallExpr *CE) const {
1966 evalStdCopyCommon(C, CE);
1969 void CStringChecker::evalStdCopyBackward(CheckerContext &C,
1970 const CallExpr *CE) const {
1971 evalStdCopyCommon(C, CE);
1974 void CStringChecker::evalStdCopyCommon(CheckerContext &C,
1975 const CallExpr *CE) const {
1976 if (CE->getNumArgs() < 3)
1979 ProgramStateRef State = C.getState();
1981 const LocationContext *LCtx = C.getLocationContext();
1983 // template <class _InputIterator, class _OutputIterator>
1985 // copy(_InputIterator __first, _InputIterator __last,
1986 // _OutputIterator __result)
1988 // Invalidate the destination buffer
1989 const Expr *Dst = CE->getArg(2);
1990 SVal DstVal = State->getSVal(Dst, LCtx);
1991 State = InvalidateBuffer(C, State, Dst, DstVal, /*IsSource=*/false,
1994 SValBuilder &SVB = C.getSValBuilder();
1996 SVal ResultVal = SVB.conjureSymbolVal(nullptr, CE, LCtx, C.blockCount());
1997 State = State->BindExpr(CE, LCtx, ResultVal);
1999 C.addTransition(State);
2002 static bool isCPPStdLibraryFunction(const FunctionDecl *FD, StringRef Name) {
2003 IdentifierInfo *II = FD->getIdentifier();
2007 if (!AnalysisDeclContext::isInStdNamespace(FD))
2010 if (II->getName().equals(Name))
2015 //===----------------------------------------------------------------------===//
2016 // The driver method, and other Checker callbacks.
2017 //===----------------------------------------------------------------------===//
2019 bool CStringChecker::evalCall(const CallExpr *CE, CheckerContext &C) const {
2020 const FunctionDecl *FDecl = C.getCalleeDecl(CE);
2025 // FIXME: Poorly-factored string switches are slow.
2026 FnCheck evalFunction = nullptr;
2027 if (C.isCLibraryFunction(FDecl, "memcpy"))
2028 evalFunction = &CStringChecker::evalMemcpy;
2029 else if (C.isCLibraryFunction(FDecl, "mempcpy"))
2030 evalFunction = &CStringChecker::evalMempcpy;
2031 else if (C.isCLibraryFunction(FDecl, "memcmp"))
2032 evalFunction = &CStringChecker::evalMemcmp;
2033 else if (C.isCLibraryFunction(FDecl, "memmove"))
2034 evalFunction = &CStringChecker::evalMemmove;
2035 else if (C.isCLibraryFunction(FDecl, "strcpy"))
2036 evalFunction = &CStringChecker::evalStrcpy;
2037 else if (C.isCLibraryFunction(FDecl, "strncpy"))
2038 evalFunction = &CStringChecker::evalStrncpy;
2039 else if (C.isCLibraryFunction(FDecl, "stpcpy"))
2040 evalFunction = &CStringChecker::evalStpcpy;
2041 else if (C.isCLibraryFunction(FDecl, "strcat"))
2042 evalFunction = &CStringChecker::evalStrcat;
2043 else if (C.isCLibraryFunction(FDecl, "strncat"))
2044 evalFunction = &CStringChecker::evalStrncat;
2045 else if (C.isCLibraryFunction(FDecl, "strlen"))
2046 evalFunction = &CStringChecker::evalstrLength;
2047 else if (C.isCLibraryFunction(FDecl, "strnlen"))
2048 evalFunction = &CStringChecker::evalstrnLength;
2049 else if (C.isCLibraryFunction(FDecl, "strcmp"))
2050 evalFunction = &CStringChecker::evalStrcmp;
2051 else if (C.isCLibraryFunction(FDecl, "strncmp"))
2052 evalFunction = &CStringChecker::evalStrncmp;
2053 else if (C.isCLibraryFunction(FDecl, "strcasecmp"))
2054 evalFunction = &CStringChecker::evalStrcasecmp;
2055 else if (C.isCLibraryFunction(FDecl, "strncasecmp"))
2056 evalFunction = &CStringChecker::evalStrncasecmp;
2057 else if (C.isCLibraryFunction(FDecl, "strsep"))
2058 evalFunction = &CStringChecker::evalStrsep;
2059 else if (C.isCLibraryFunction(FDecl, "bcopy"))
2060 evalFunction = &CStringChecker::evalBcopy;
2061 else if (C.isCLibraryFunction(FDecl, "bcmp"))
2062 evalFunction = &CStringChecker::evalMemcmp;
2063 else if (isCPPStdLibraryFunction(FDecl, "copy"))
2064 evalFunction = &CStringChecker::evalStdCopy;
2065 else if (isCPPStdLibraryFunction(FDecl, "copy_backward"))
2066 evalFunction = &CStringChecker::evalStdCopyBackward;
2068 // If the callee isn't a string function, let another checker handle it.
2072 // Check and evaluate the call.
2073 (this->*evalFunction)(C, CE);
2075 // If the evaluate call resulted in no change, chain to the next eval call
2077 // Note, the custom CString evaluation calls assume that basic safety
2078 // properties are held. However, if the user chooses to turn off some of these
2079 // checks, we ignore the issues and leave the call evaluation to a generic
2081 return C.isDifferent();
2084 void CStringChecker::checkPreStmt(const DeclStmt *DS, CheckerContext &C) const {
2085 // Record string length for char a[] = "abc";
2086 ProgramStateRef state = C.getState();
2088 for (const auto *I : DS->decls()) {
2089 const VarDecl *D = dyn_cast<VarDecl>(I);
2093 // FIXME: Handle array fields of structs.
2094 if (!D->getType()->isArrayType())
2097 const Expr *Init = D->getInit();
2100 if (!isa<StringLiteral>(Init))
2103 Loc VarLoc = state->getLValue(D, C.getLocationContext());
2104 const MemRegion *MR = VarLoc.getAsRegion();
2108 SVal StrVal = state->getSVal(Init, C.getLocationContext());
2109 assert(StrVal.isValid() && "Initializer string is unknown or undefined");
2110 DefinedOrUnknownSVal strLength =
2111 getCStringLength(C, state, Init, StrVal).castAs<DefinedOrUnknownSVal>();
2113 state = state->set<CStringLength>(MR, strLength);
2116 C.addTransition(state);
2120 CStringChecker::checkRegionChanges(ProgramStateRef state,
2121 const InvalidatedSymbols *,
2122 ArrayRef<const MemRegion *> ExplicitRegions,
2123 ArrayRef<const MemRegion *> Regions,
2124 const LocationContext *LCtx,
2125 const CallEvent *Call) const {
2126 CStringLengthTy Entries = state->get<CStringLength>();
2127 if (Entries.isEmpty())
2130 llvm::SmallPtrSet<const MemRegion *, 8> Invalidated;
2131 llvm::SmallPtrSet<const MemRegion *, 32> SuperRegions;
2133 // First build sets for the changed regions and their super-regions.
2134 for (ArrayRef<const MemRegion *>::iterator
2135 I = Regions.begin(), E = Regions.end(); I != E; ++I) {
2136 const MemRegion *MR = *I;
2137 Invalidated.insert(MR);
2139 SuperRegions.insert(MR);
2140 while (const SubRegion *SR = dyn_cast<SubRegion>(MR)) {
2141 MR = SR->getSuperRegion();
2142 SuperRegions.insert(MR);
2146 CStringLengthTy::Factory &F = state->get_context<CStringLength>();
2148 // Then loop over the entries in the current state.
2149 for (CStringLengthTy::iterator I = Entries.begin(),
2150 E = Entries.end(); I != E; ++I) {
2151 const MemRegion *MR = I.getKey();
2153 // Is this entry for a super-region of a changed region?
2154 if (SuperRegions.count(MR)) {
2155 Entries = F.remove(Entries, MR);
2159 // Is this entry for a sub-region of a changed region?
2160 const MemRegion *Super = MR;
2161 while (const SubRegion *SR = dyn_cast<SubRegion>(Super)) {
2162 Super = SR->getSuperRegion();
2163 if (Invalidated.count(Super)) {
2164 Entries = F.remove(Entries, MR);
2170 return state->set<CStringLength>(Entries);
2173 void CStringChecker::checkLiveSymbols(ProgramStateRef state,
2174 SymbolReaper &SR) const {
2175 // Mark all symbols in our string length map as valid.
2176 CStringLengthTy Entries = state->get<CStringLength>();
2178 for (CStringLengthTy::iterator I = Entries.begin(), E = Entries.end();
2180 SVal Len = I.getData();
2182 for (SymExpr::symbol_iterator si = Len.symbol_begin(),
2183 se = Len.symbol_end(); si != se; ++si)
2188 void CStringChecker::checkDeadSymbols(SymbolReaper &SR,
2189 CheckerContext &C) const {
2190 if (!SR.hasDeadSymbols())
2193 ProgramStateRef state = C.getState();
2194 CStringLengthTy Entries = state->get<CStringLength>();
2195 if (Entries.isEmpty())
2198 CStringLengthTy::Factory &F = state->get_context<CStringLength>();
2199 for (CStringLengthTy::iterator I = Entries.begin(), E = Entries.end();
2201 SVal Len = I.getData();
2202 if (SymbolRef Sym = Len.getAsSymbol()) {
2204 Entries = F.remove(Entries, I.getKey());
2208 state = state->set<CStringLength>(Entries);
2209 C.addTransition(state);
2212 #define REGISTER_CHECKER(name) \
2213 void ento::register##name(CheckerManager &mgr) { \
2214 CStringChecker *checker = mgr.registerChecker<CStringChecker>(); \
2215 checker->Filter.Check##name = true; \
2216 checker->Filter.CheckName##name = mgr.getCurrentCheckName(); \
2219 REGISTER_CHECKER(CStringNullArg)
2220 REGISTER_CHECKER(CStringOutOfBounds)
2221 REGISTER_CHECKER(CStringBufferOverlap)
2222 REGISTER_CHECKER(CStringNotNullTerm)
2224 void ento::registerCStringCheckerBasic(CheckerManager &Mgr) {
2225 registerCStringNullArg(Mgr);