xref: /llvm-project/clang/lib/StaticAnalyzer/Checkers/CStringChecker.cpp (revision 45d8c12b6862a569e0066f9e8e682ee1d6f09ff7)
1 //= CStringChecker.h - Checks calls to C string functions ----------*- C++ -*-//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This defines CStringChecker, which is an assortment of checks on calls
11 // to functions in <string.h>.
12 //
13 //===----------------------------------------------------------------------===//
14 
15 #include "ClangSACheckers.h"
16 #include "clang/StaticAnalyzer/Core/Checker.h"
17 #include "clang/StaticAnalyzer/Core/CheckerManager.h"
18 #include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h"
19 #include "clang/StaticAnalyzer/Core/BugReporter/BugType.h"
20 #include "clang/StaticAnalyzer/Core/PathSensitive/GRStateTrait.h"
21 #include "llvm/ADT/StringSwitch.h"
22 
23 using namespace clang;
24 using namespace ento;
25 
26 namespace {
27 class CStringChecker : public Checker< eval::Call,
28                                          check::PreStmt<DeclStmt>,
29                                          check::LiveSymbols,
30                                          check::DeadSymbols,
31                                          check::RegionChanges
32                                          > {
33   mutable llvm::OwningPtr<BugType> BT_Null, BT_Bounds, BT_BoundsWrite,
34                                    BT_Overlap, BT_NotCString;
35 public:
36   static void *getTag() { static int tag; return &tag; }
37 
38   bool evalCall(const CallExpr *CE, CheckerContext &C) const;
39   void checkPreStmt(const DeclStmt *DS, CheckerContext &C) const;
40   void checkLiveSymbols(const GRState *state, SymbolReaper &SR) const;
41   void checkDeadSymbols(SymbolReaper &SR, CheckerContext &C) const;
42   bool wantsRegionChangeUpdate(const GRState *state) const;
43 
44   const GRState *checkRegionChanges(const GRState *state,
45                                     const StoreManager::InvalidatedSymbols *,
46                                     const MemRegion * const *Begin,
47                                     const MemRegion * const *End) const;
48 
49   typedef void (CStringChecker::*FnCheck)(CheckerContext &,
50                                           const CallExpr *) const;
51 
52   void evalMemcpy(CheckerContext &C, const CallExpr *CE) const;
53   void evalMempcpy(CheckerContext &C, const CallExpr *CE) const;
54   void evalMemmove(CheckerContext &C, const CallExpr *CE) const;
55   void evalBcopy(CheckerContext &C, const CallExpr *CE) const;
56   void evalCopyCommon(CheckerContext &C, const CallExpr *CE,
57                       const GRState *state,
58                       const Expr *Size, const Expr *Source, const Expr *Dest,
59                       bool Restricted = false,
60                       bool IsMempcpy = false) const;
61 
62   void evalMemcmp(CheckerContext &C, const CallExpr *CE) const;
63 
64   void evalstrLength(CheckerContext &C, const CallExpr *CE) const;
65   void evalstrnLength(CheckerContext &C, const CallExpr *CE) const;
66   void evalstrLengthCommon(CheckerContext &C, const CallExpr *CE,
67                            bool IsStrnlen = false) const;
68 
69   void evalStrcpy(CheckerContext &C, const CallExpr *CE) const;
70   void evalStrncpy(CheckerContext &C, const CallExpr *CE) const;
71   void evalStpcpy(CheckerContext &C, const CallExpr *CE) const;
72   void evalStrcpyCommon(CheckerContext &C, const CallExpr *CE, bool returnEnd,
73                         bool isBounded, bool isAppending) const;
74 
75   void evalStrcat(CheckerContext &C, const CallExpr *CE) const;
76   void evalStrncat(CheckerContext &C, const CallExpr *CE) const;
77 
78   void evalStrcmp(CheckerContext &C, const CallExpr *CE) const;
79   void evalStrncmp(CheckerContext &C, const CallExpr *CE) const;
80   void evalStrcasecmp(CheckerContext &C, const CallExpr *CE) const;
81   void evalStrncasecmp(CheckerContext &C, const CallExpr *CE) const;
82   void evalStrcmpCommon(CheckerContext &C, const CallExpr *CE,
83                         bool isBounded = false, bool ignoreCase = false) const;
84 
85   // Utility methods
86   std::pair<const GRState*, const GRState*>
87   static assumeZero(CheckerContext &C,
88                     const GRState *state, SVal V, QualType Ty);
89 
90   static const GRState *setCStringLength(const GRState *state,
91                                          const MemRegion *MR, SVal strLength);
92   static SVal getCStringLengthForRegion(CheckerContext &C,
93                                         const GRState *&state,
94                                         const Expr *Ex, const MemRegion *MR);
95   SVal getCStringLength(CheckerContext &C, const GRState *&state,
96                         const Expr *Ex, SVal Buf) const;
97 
98   const StringLiteral *getCStringLiteral(CheckerContext &C,
99                                          const GRState *&state,
100                                          const Expr *expr,
101                                          SVal val) const;
102 
103   static const GRState *InvalidateBuffer(CheckerContext &C,
104                                          const GRState *state,
105                                          const Expr *Ex, SVal V);
106 
107   static bool SummarizeRegion(llvm::raw_ostream& os, ASTContext& Ctx,
108                               const MemRegion *MR);
109 
110   // Re-usable checks
111   const GRState *checkNonNull(CheckerContext &C, const GRState *state,
112                                const Expr *S, SVal l) const;
113   const GRState *CheckLocation(CheckerContext &C, const GRState *state,
114                                const Expr *S, SVal l,
115                                bool IsDestination = false) const;
116   const GRState *CheckBufferAccess(CheckerContext &C, const GRState *state,
117                                    const Expr *Size,
118                                    const Expr *FirstBuf,
119                                    const Expr *SecondBuf = NULL,
120                                    bool FirstIsDestination = false) const;
121   const GRState *CheckOverlap(CheckerContext &C, const GRState *state,
122                               const Expr *Size, const Expr *First,
123                               const Expr *Second) const;
124   void emitOverlapBug(CheckerContext &C, const GRState *state,
125                       const Stmt *First, const Stmt *Second) const;
126 };
127 
128 class CStringLength {
129 public:
130   typedef llvm::ImmutableMap<const MemRegion *, SVal> EntryMap;
131 };
132 } //end anonymous namespace
133 
134 namespace clang {
135 namespace ento {
136   template <>
137   struct GRStateTrait<CStringLength>
138     : public GRStatePartialTrait<CStringLength::EntryMap> {
139     static void *GDMIndex() { return CStringChecker::getTag(); }
140   };
141 }
142 }
143 
144 //===----------------------------------------------------------------------===//
145 // Individual checks and utility methods.
146 //===----------------------------------------------------------------------===//
147 
148 std::pair<const GRState*, const GRState*>
149 CStringChecker::assumeZero(CheckerContext &C, const GRState *state, SVal V,
150                            QualType Ty) {
151   DefinedSVal *val = dyn_cast<DefinedSVal>(&V);
152   if (!val)
153     return std::pair<const GRState*, const GRState *>(state, state);
154 
155   SValBuilder &svalBuilder = C.getSValBuilder();
156   DefinedOrUnknownSVal zero = svalBuilder.makeZeroVal(Ty);
157   return state->assume(svalBuilder.evalEQ(state, *val, zero));
158 }
159 
160 const GRState *CStringChecker::checkNonNull(CheckerContext &C,
161                                             const GRState *state,
162                                             const Expr *S, SVal l) const {
163   // If a previous check has failed, propagate the failure.
164   if (!state)
165     return NULL;
166 
167   const GRState *stateNull, *stateNonNull;
168   llvm::tie(stateNull, stateNonNull) = assumeZero(C, state, l, S->getType());
169 
170   if (stateNull && !stateNonNull) {
171     ExplodedNode *N = C.generateSink(stateNull);
172     if (!N)
173       return NULL;
174 
175     if (!BT_Null)
176       BT_Null.reset(new BuiltinBug("API",
177         "Null pointer argument in call to byte string function"));
178 
179     // Generate a report for this bug.
180     BuiltinBug *BT = static_cast<BuiltinBug*>(BT_Null.get());
181     EnhancedBugReport *report = new EnhancedBugReport(*BT,
182                                                       BT->getDescription(), N);
183 
184     report->addRange(S->getSourceRange());
185     report->addVisitorCreator(bugreporter::registerTrackNullOrUndefValue, S);
186     C.EmitReport(report);
187     return NULL;
188   }
189 
190   // From here on, assume that the value is non-null.
191   assert(stateNonNull);
192   return stateNonNull;
193 }
194 
195 // FIXME: This was originally copied from ArrayBoundChecker.cpp. Refactor?
196 const GRState *CStringChecker::CheckLocation(CheckerContext &C,
197                                              const GRState *state,
198                                              const Expr *S, SVal l,
199                                              bool IsDestination) const {
200   // If a previous check has failed, propagate the failure.
201   if (!state)
202     return NULL;
203 
204   // Check for out of bound array element access.
205   const MemRegion *R = l.getAsRegion();
206   if (!R)
207     return state;
208 
209   const ElementRegion *ER = dyn_cast<ElementRegion>(R);
210   if (!ER)
211     return state;
212 
213   assert(ER->getValueType() == C.getASTContext().CharTy &&
214     "CheckLocation should only be called with char* ElementRegions");
215 
216   // Get the size of the array.
217   const SubRegion *superReg = cast<SubRegion>(ER->getSuperRegion());
218   SValBuilder &svalBuilder = C.getSValBuilder();
219   SVal Extent = svalBuilder.convertToArrayIndex(superReg->getExtent(svalBuilder));
220   DefinedOrUnknownSVal Size = cast<DefinedOrUnknownSVal>(Extent);
221 
222   // Get the index of the accessed element.
223   DefinedOrUnknownSVal Idx = cast<DefinedOrUnknownSVal>(ER->getIndex());
224 
225   const GRState *StInBound = state->assumeInBound(Idx, Size, true);
226   const GRState *StOutBound = state->assumeInBound(Idx, Size, false);
227   if (StOutBound && !StInBound) {
228     ExplodedNode *N = C.generateSink(StOutBound);
229     if (!N)
230       return NULL;
231 
232     BuiltinBug *BT;
233     if (IsDestination) {
234       if (!BT_BoundsWrite) {
235         BT_BoundsWrite.reset(new BuiltinBug("Out-of-bound array access",
236           "Byte string function overflows destination buffer"));
237       }
238       BT = static_cast<BuiltinBug*>(BT_BoundsWrite.get());
239     } else {
240       if (!BT_Bounds) {
241         BT_Bounds.reset(new BuiltinBug("Out-of-bound array access",
242           "Byte string function accesses out-of-bound array element"));
243       }
244       BT = static_cast<BuiltinBug*>(BT_Bounds.get());
245     }
246 
247     // FIXME: It would be nice to eventually make this diagnostic more clear,
248     // e.g., by referencing the original declaration or by saying *why* this
249     // reference is outside the range.
250 
251     // Generate a report for this bug.
252     RangedBugReport *report = new RangedBugReport(*BT, BT->getDescription(), N);
253 
254     report->addRange(S->getSourceRange());
255     C.EmitReport(report);
256     return NULL;
257   }
258 
259   // Array bound check succeeded.  From this point forward the array bound
260   // should always succeed.
261   return StInBound;
262 }
263 
264 const GRState *CStringChecker::CheckBufferAccess(CheckerContext &C,
265                                                  const GRState *state,
266                                                  const Expr *Size,
267                                                  const Expr *FirstBuf,
268                                                  const Expr *SecondBuf,
269                                                 bool FirstIsDestination) const {
270   // If a previous check has failed, propagate the failure.
271   if (!state)
272     return NULL;
273 
274   SValBuilder &svalBuilder = C.getSValBuilder();
275   ASTContext &Ctx = C.getASTContext();
276 
277   QualType sizeTy = Size->getType();
278   QualType PtrTy = Ctx.getPointerType(Ctx.CharTy);
279 
280   // Check that the first buffer is non-null.
281   SVal BufVal = state->getSVal(FirstBuf);
282   state = checkNonNull(C, state, FirstBuf, BufVal);
283   if (!state)
284     return NULL;
285 
286   // Get the access length and make sure it is known.
287   SVal LengthVal = state->getSVal(Size);
288   NonLoc *Length = dyn_cast<NonLoc>(&LengthVal);
289   if (!Length)
290     return state;
291 
292   // Compute the offset of the last element to be accessed: size-1.
293   NonLoc One = cast<NonLoc>(svalBuilder.makeIntVal(1, sizeTy));
294   NonLoc LastOffset = cast<NonLoc>(svalBuilder.evalBinOpNN(state, BO_Sub,
295                                                     *Length, One, sizeTy));
296 
297   // Check that the first buffer is sufficiently long.
298   SVal BufStart = svalBuilder.evalCast(BufVal, PtrTy, FirstBuf->getType());
299   if (Loc *BufLoc = dyn_cast<Loc>(&BufStart)) {
300     SVal BufEnd = svalBuilder.evalBinOpLN(state, BO_Add, *BufLoc,
301                                           LastOffset, PtrTy);
302     state = CheckLocation(C, state, FirstBuf, BufEnd, FirstIsDestination);
303 
304     // If the buffer isn't large enough, abort.
305     if (!state)
306       return NULL;
307   }
308 
309   // If there's a second buffer, check it as well.
310   if (SecondBuf) {
311     BufVal = state->getSVal(SecondBuf);
312     state = checkNonNull(C, state, SecondBuf, BufVal);
313     if (!state)
314       return NULL;
315 
316     BufStart = svalBuilder.evalCast(BufVal, PtrTy, SecondBuf->getType());
317     if (Loc *BufLoc = dyn_cast<Loc>(&BufStart)) {
318       SVal BufEnd = svalBuilder.evalBinOpLN(state, BO_Add, *BufLoc,
319                                             LastOffset, PtrTy);
320       state = CheckLocation(C, state, SecondBuf, BufEnd);
321     }
322   }
323 
324   // Large enough or not, return this state!
325   return state;
326 }
327 
328 const GRState *CStringChecker::CheckOverlap(CheckerContext &C,
329                                             const GRState *state,
330                                             const Expr *Size,
331                                             const Expr *First,
332                                             const Expr *Second) const {
333   // Do a simple check for overlap: if the two arguments are from the same
334   // buffer, see if the end of the first is greater than the start of the second
335   // or vice versa.
336 
337   // If a previous check has failed, propagate the failure.
338   if (!state)
339     return NULL;
340 
341   const GRState *stateTrue, *stateFalse;
342 
343   // Get the buffer values and make sure they're known locations.
344   SVal firstVal = state->getSVal(First);
345   SVal secondVal = state->getSVal(Second);
346 
347   Loc *firstLoc = dyn_cast<Loc>(&firstVal);
348   if (!firstLoc)
349     return state;
350 
351   Loc *secondLoc = dyn_cast<Loc>(&secondVal);
352   if (!secondLoc)
353     return state;
354 
355   // Are the two values the same?
356   SValBuilder &svalBuilder = C.getSValBuilder();
357   llvm::tie(stateTrue, stateFalse) =
358     state->assume(svalBuilder.evalEQ(state, *firstLoc, *secondLoc));
359 
360   if (stateTrue && !stateFalse) {
361     // If the values are known to be equal, that's automatically an overlap.
362     emitOverlapBug(C, stateTrue, First, Second);
363     return NULL;
364   }
365 
366   // assume the two expressions are not equal.
367   assert(stateFalse);
368   state = stateFalse;
369 
370   // Which value comes first?
371   ASTContext &Ctx = svalBuilder.getContext();
372   QualType cmpTy = Ctx.IntTy;
373   SVal reverse = svalBuilder.evalBinOpLL(state, BO_GT,
374                                          *firstLoc, *secondLoc, cmpTy);
375   DefinedOrUnknownSVal *reverseTest = dyn_cast<DefinedOrUnknownSVal>(&reverse);
376   if (!reverseTest)
377     return state;
378 
379   llvm::tie(stateTrue, stateFalse) = state->assume(*reverseTest);
380   if (stateTrue) {
381     if (stateFalse) {
382       // If we don't know which one comes first, we can't perform this test.
383       return state;
384     } else {
385       // Switch the values so that firstVal is before secondVal.
386       Loc *tmpLoc = firstLoc;
387       firstLoc = secondLoc;
388       secondLoc = tmpLoc;
389 
390       // Switch the Exprs as well, so that they still correspond.
391       const Expr *tmpExpr = First;
392       First = Second;
393       Second = tmpExpr;
394     }
395   }
396 
397   // Get the length, and make sure it too is known.
398   SVal LengthVal = state->getSVal(Size);
399   NonLoc *Length = dyn_cast<NonLoc>(&LengthVal);
400   if (!Length)
401     return state;
402 
403   // Convert the first buffer's start address to char*.
404   // Bail out if the cast fails.
405   QualType CharPtrTy = Ctx.getPointerType(Ctx.CharTy);
406   SVal FirstStart = svalBuilder.evalCast(*firstLoc, CharPtrTy, First->getType());
407   Loc *FirstStartLoc = dyn_cast<Loc>(&FirstStart);
408   if (!FirstStartLoc)
409     return state;
410 
411   // Compute the end of the first buffer. Bail out if THAT fails.
412   SVal FirstEnd = svalBuilder.evalBinOpLN(state, BO_Add,
413                                  *FirstStartLoc, *Length, CharPtrTy);
414   Loc *FirstEndLoc = dyn_cast<Loc>(&FirstEnd);
415   if (!FirstEndLoc)
416     return state;
417 
418   // Is the end of the first buffer past the start of the second buffer?
419   SVal Overlap = svalBuilder.evalBinOpLL(state, BO_GT,
420                                 *FirstEndLoc, *secondLoc, cmpTy);
421   DefinedOrUnknownSVal *OverlapTest = dyn_cast<DefinedOrUnknownSVal>(&Overlap);
422   if (!OverlapTest)
423     return state;
424 
425   llvm::tie(stateTrue, stateFalse) = state->assume(*OverlapTest);
426 
427   if (stateTrue && !stateFalse) {
428     // Overlap!
429     emitOverlapBug(C, stateTrue, First, Second);
430     return NULL;
431   }
432 
433   // assume the two expressions don't overlap.
434   assert(stateFalse);
435   return stateFalse;
436 }
437 
438 void CStringChecker::emitOverlapBug(CheckerContext &C, const GRState *state,
439                                   const Stmt *First, const Stmt *Second) const {
440   ExplodedNode *N = C.generateSink(state);
441   if (!N)
442     return;
443 
444   if (!BT_Overlap)
445     BT_Overlap.reset(new BugType("Unix API", "Improper arguments"));
446 
447   // Generate a report for this bug.
448   RangedBugReport *report =
449     new RangedBugReport(*BT_Overlap,
450       "Arguments must not be overlapping buffers", N);
451   report->addRange(First->getSourceRange());
452   report->addRange(Second->getSourceRange());
453 
454   C.EmitReport(report);
455 }
456 
457 const GRState *CStringChecker::setCStringLength(const GRState *state,
458                                                 const MemRegion *MR,
459                                                 SVal strLength) {
460   assert(!strLength.isUndef() && "Attempt to set an undefined string length");
461   if (strLength.isUnknown())
462     return state;
463 
464   MR = MR->StripCasts();
465 
466   switch (MR->getKind()) {
467   case MemRegion::StringRegionKind:
468     // FIXME: This can happen if we strcpy() into a string region. This is
469     // undefined [C99 6.4.5p6], but we should still warn about it.
470     return state;
471 
472   case MemRegion::SymbolicRegionKind:
473   case MemRegion::AllocaRegionKind:
474   case MemRegion::VarRegionKind:
475   case MemRegion::FieldRegionKind:
476   case MemRegion::ObjCIvarRegionKind:
477     return state->set<CStringLength>(MR, strLength);
478 
479   case MemRegion::ElementRegionKind:
480     // FIXME: Handle element regions by upper-bounding the parent region's
481     // string length.
482     return state;
483 
484   default:
485     // Other regions (mostly non-data) can't have a reliable C string length.
486     // For now, just ignore the change.
487     // FIXME: These are rare but not impossible. We should output some kind of
488     // warning for things like strcpy((char[]){'a', 0}, "b");
489     return state;
490   }
491 }
492 
493 SVal CStringChecker::getCStringLengthForRegion(CheckerContext &C,
494                                                const GRState *&state,
495                                                const Expr *Ex,
496                                                const MemRegion *MR) {
497   // If there's a recorded length, go ahead and return it.
498   const SVal *Recorded = state->get<CStringLength>(MR);
499   if (Recorded)
500     return *Recorded;
501 
502   // Otherwise, get a new symbol and update the state.
503   unsigned Count = C.getNodeBuilder().getCurrentBlockCount();
504   SValBuilder &svalBuilder = C.getSValBuilder();
505   QualType sizeTy = svalBuilder.getContext().getSizeType();
506   SVal strLength = svalBuilder.getMetadataSymbolVal(CStringChecker::getTag(),
507                                                     MR, Ex, sizeTy, Count);
508   state = state->set<CStringLength>(MR, strLength);
509   return strLength;
510 }
511 
512 SVal CStringChecker::getCStringLength(CheckerContext &C, const GRState *&state,
513                                       const Expr *Ex, SVal Buf) const {
514   const MemRegion *MR = Buf.getAsRegion();
515   if (!MR) {
516     // If we can't get a region, see if it's something we /know/ isn't a
517     // C string. In the context of locations, the only time we can issue such
518     // a warning is for labels.
519     if (loc::GotoLabel *Label = dyn_cast<loc::GotoLabel>(&Buf)) {
520       if (ExplodedNode *N = C.generateNode(state)) {
521         if (!BT_NotCString)
522           BT_NotCString.reset(new BuiltinBug("API",
523             "Argument is not a null-terminated string."));
524 
525         llvm::SmallString<120> buf;
526         llvm::raw_svector_ostream os(buf);
527         os << "Argument to byte string function is the address of the label '"
528            << Label->getLabel()->getName()
529            << "', which is not a null-terminated string";
530 
531         // Generate a report for this bug.
532         EnhancedBugReport *report = new EnhancedBugReport(*BT_NotCString,
533                                                           os.str(), N);
534 
535         report->addRange(Ex->getSourceRange());
536         C.EmitReport(report);
537       }
538 
539       return UndefinedVal();
540     }
541 
542     // If it's not a region and not a label, give up.
543     return UnknownVal();
544   }
545 
546   // If we have a region, strip casts from it and see if we can figure out
547   // its length. For anything we can't figure out, just return UnknownVal.
548   MR = MR->StripCasts();
549 
550   switch (MR->getKind()) {
551   case MemRegion::StringRegionKind: {
552     // Modifying the contents of string regions is undefined [C99 6.4.5p6],
553     // so we can assume that the byte length is the correct C string length.
554     SValBuilder &svalBuilder = C.getSValBuilder();
555     QualType sizeTy = svalBuilder.getContext().getSizeType();
556     const StringLiteral *strLit = cast<StringRegion>(MR)->getStringLiteral();
557     return svalBuilder.makeIntVal(strLit->getByteLength(), sizeTy);
558   }
559   case MemRegion::SymbolicRegionKind:
560   case MemRegion::AllocaRegionKind:
561   case MemRegion::VarRegionKind:
562   case MemRegion::FieldRegionKind:
563   case MemRegion::ObjCIvarRegionKind:
564     return getCStringLengthForRegion(C, state, Ex, MR);
565   case MemRegion::CompoundLiteralRegionKind:
566     // FIXME: Can we track this? Is it necessary?
567     return UnknownVal();
568   case MemRegion::ElementRegionKind:
569     // FIXME: How can we handle this? It's not good enough to subtract the
570     // offset from the base string length; consider "123\x00567" and &a[5].
571     return UnknownVal();
572   default:
573     // Other regions (mostly non-data) can't have a reliable C string length.
574     // In this case, an error is emitted and UndefinedVal is returned.
575     // The caller should always be prepared to handle this case.
576     if (ExplodedNode *N = C.generateNode(state)) {
577       if (!BT_NotCString)
578         BT_NotCString.reset(new BuiltinBug("API",
579           "Argument is not a null-terminated string."));
580 
581       llvm::SmallString<120> buf;
582       llvm::raw_svector_ostream os(buf);
583 
584       os << "Argument to byte string function is ";
585 
586       if (SummarizeRegion(os, C.getASTContext(), MR))
587         os << ", which is not a null-terminated string";
588       else
589         os << "not a null-terminated string";
590 
591       // Generate a report for this bug.
592       EnhancedBugReport *report = new EnhancedBugReport(*BT_NotCString,
593                                                         os.str(), N);
594 
595       report->addRange(Ex->getSourceRange());
596       C.EmitReport(report);
597     }
598 
599     return UndefinedVal();
600   }
601 }
602 
603 const StringLiteral *CStringChecker::getCStringLiteral(CheckerContext &C,
604   const GRState *&state, const Expr *expr, SVal val) const {
605 
606   // Get the memory region pointed to by the val.
607   const MemRegion *bufRegion = val.getAsRegion();
608   if (!bufRegion)
609     return NULL;
610 
611   // Strip casts off the memory region.
612   bufRegion = bufRegion->StripCasts();
613 
614   // Cast the memory region to a string region.
615   const StringRegion *strRegion= dyn_cast<StringRegion>(bufRegion);
616   if (!strRegion)
617     return NULL;
618 
619   // Return the actual string in the string region.
620   return strRegion->getStringLiteral();
621 }
622 
623 const GRState *CStringChecker::InvalidateBuffer(CheckerContext &C,
624                                                 const GRState *state,
625                                                 const Expr *E, SVal V) {
626   Loc *L = dyn_cast<Loc>(&V);
627   if (!L)
628     return state;
629 
630   // FIXME: This is a simplified version of what's in CFRefCount.cpp -- it makes
631   // some assumptions about the value that CFRefCount can't. Even so, it should
632   // probably be refactored.
633   if (loc::MemRegionVal* MR = dyn_cast<loc::MemRegionVal>(L)) {
634     const MemRegion *R = MR->getRegion()->StripCasts();
635 
636     // Are we dealing with an ElementRegion?  If so, we should be invalidating
637     // the super-region.
638     if (const ElementRegion *ER = dyn_cast<ElementRegion>(R)) {
639       R = ER->getSuperRegion();
640       // FIXME: What about layers of ElementRegions?
641     }
642 
643     // Invalidate this region.
644     unsigned Count = C.getNodeBuilder().getCurrentBlockCount();
645     return state->invalidateRegion(R, E, Count, NULL);
646   }
647 
648   // If we have a non-region value by chance, just remove the binding.
649   // FIXME: is this necessary or correct? This handles the non-Region
650   //  cases.  Is it ever valid to store to these?
651   return state->unbindLoc(*L);
652 }
653 
654 bool CStringChecker::SummarizeRegion(llvm::raw_ostream& os, ASTContext& Ctx,
655                                      const MemRegion *MR) {
656   const TypedRegion *TR = dyn_cast<TypedRegion>(MR);
657   if (!TR)
658     return false;
659 
660   switch (TR->getKind()) {
661   case MemRegion::FunctionTextRegionKind: {
662     const FunctionDecl *FD = cast<FunctionTextRegion>(TR)->getDecl();
663     if (FD)
664       os << "the address of the function '" << FD << "'";
665     else
666       os << "the address of a function";
667     return true;
668   }
669   case MemRegion::BlockTextRegionKind:
670     os << "block text";
671     return true;
672   case MemRegion::BlockDataRegionKind:
673     os << "a block";
674     return true;
675   case MemRegion::CXXThisRegionKind:
676   case MemRegion::CXXTempObjectRegionKind:
677     os << "a C++ temp object of type " << TR->getValueType().getAsString();
678     return true;
679   case MemRegion::VarRegionKind:
680     os << "a variable of type" << TR->getValueType().getAsString();
681     return true;
682   case MemRegion::FieldRegionKind:
683     os << "a field of type " << TR->getValueType().getAsString();
684     return true;
685   case MemRegion::ObjCIvarRegionKind:
686     os << "an instance variable of type " << TR->getValueType().getAsString();
687     return true;
688   default:
689     return false;
690   }
691 }
692 
693 //===----------------------------------------------------------------------===//
694 // evaluation of individual function calls.
695 //===----------------------------------------------------------------------===//
696 
697 void CStringChecker::evalCopyCommon(CheckerContext &C,
698                                     const CallExpr *CE,
699                                     const GRState *state,
700                                     const Expr *Size, const Expr *Dest,
701                                     const Expr *Source, bool Restricted,
702                                     bool IsMempcpy) const {
703   // See if the size argument is zero.
704   SVal sizeVal = state->getSVal(Size);
705   QualType sizeTy = Size->getType();
706 
707   const GRState *stateZeroSize, *stateNonZeroSize;
708   llvm::tie(stateZeroSize, stateNonZeroSize) = assumeZero(C, state, sizeVal, sizeTy);
709 
710   // Get the value of the Dest.
711   SVal destVal = state->getSVal(Dest);
712 
713   // If the size is zero, there won't be any actual memory access, so
714   // just bind the return value to the destination buffer and return.
715   if (stateZeroSize) {
716     stateZeroSize = stateZeroSize->BindExpr(CE, destVal);
717     C.addTransition(stateZeroSize);
718   }
719 
720   // If the size can be nonzero, we have to check the other arguments.
721   if (stateNonZeroSize) {
722     state = stateNonZeroSize;
723 
724     // Ensure the destination is not null. If it is NULL there will be a
725     // NULL pointer dereference.
726     state = checkNonNull(C, state, Dest, destVal);
727     if (!state)
728       return;
729 
730     // Get the value of the Src.
731     SVal srcVal = state->getSVal(Source);
732 
733     // Ensure the source is not null. If it is NULL there will be a
734     // NULL pointer dereference.
735     state = checkNonNull(C, state, Source, srcVal);
736     if (!state)
737       return;
738 
739     // Ensure the accesses are valid and that the buffers do not overlap.
740     state = CheckBufferAccess(C, state, Size, Dest, Source,
741                               /* FirstIsDst = */ true);
742     if (Restricted)
743       state = CheckOverlap(C, state, Size, Dest, Source);
744 
745     if (!state)
746       return;
747 
748     // If this is mempcpy, get the byte after the last byte copied and
749     // bind the expr.
750     if (IsMempcpy) {
751       loc::MemRegionVal *destRegVal = dyn_cast<loc::MemRegionVal>(&destVal);
752       assert(destRegVal && "Destination should be a known MemRegionVal here");
753 
754       // Get the length to copy.
755       NonLoc *lenValNonLoc = dyn_cast<NonLoc>(&sizeVal);
756 
757       if (lenValNonLoc) {
758         // Get the byte after the last byte copied.
759         SVal lastElement = C.getSValBuilder().evalBinOpLN(state, BO_Add,
760                                                           *destRegVal,
761                                                           *lenValNonLoc,
762                                                           Dest->getType());
763 
764         // The byte after the last byte copied is the return value.
765         state = state->BindExpr(CE, lastElement);
766       } else {
767         // If we don't know how much we copied, we can at least
768         // conjure a return value for later.
769         unsigned Count = C.getNodeBuilder().getCurrentBlockCount();
770         SVal result =
771           C.getSValBuilder().getConjuredSymbolVal(NULL, CE, Count);
772         state = state->BindExpr(CE, result);
773       }
774 
775     } else {
776       // All other copies return the destination buffer.
777       // (Well, bcopy() has a void return type, but this won't hurt.)
778       state = state->BindExpr(CE, destVal);
779     }
780 
781     // Invalidate the destination.
782     // FIXME: Even if we can't perfectly model the copy, we should see if we
783     // can use LazyCompoundVals to copy the source values into the destination.
784     // This would probably remove any existing bindings past the end of the
785     // copied region, but that's still an improvement over blank invalidation.
786     state = InvalidateBuffer(C, state, Dest, state->getSVal(Dest));
787     C.addTransition(state);
788   }
789 }
790 
791 
792 void CStringChecker::evalMemcpy(CheckerContext &C, const CallExpr *CE) const {
793   // void *memcpy(void *restrict dst, const void *restrict src, size_t n);
794   // The return value is the address of the destination buffer.
795   const Expr *Dest = CE->getArg(0);
796   const GRState *state = C.getState();
797 
798   evalCopyCommon(C, CE, state, CE->getArg(2), Dest, CE->getArg(1), true);
799 }
800 
801 void CStringChecker::evalMempcpy(CheckerContext &C, const CallExpr *CE) const {
802   // void *mempcpy(void *restrict dst, const void *restrict src, size_t n);
803   // The return value is a pointer to the byte following the last written byte.
804   const Expr *Dest = CE->getArg(0);
805   const GRState *state = C.getState();
806 
807   evalCopyCommon(C, CE, state, CE->getArg(2), Dest, CE->getArg(1), true, true);
808 }
809 
810 void CStringChecker::evalMemmove(CheckerContext &C, const CallExpr *CE) const {
811   // void *memmove(void *dst, const void *src, size_t n);
812   // The return value is the address of the destination buffer.
813   const Expr *Dest = CE->getArg(0);
814   const GRState *state = C.getState();
815 
816   evalCopyCommon(C, CE, state, CE->getArg(2), Dest, CE->getArg(1));
817 }
818 
819 void CStringChecker::evalBcopy(CheckerContext &C, const CallExpr *CE) const {
820   // void bcopy(const void *src, void *dst, size_t n);
821   evalCopyCommon(C, CE, C.getState(),
822                  CE->getArg(2), CE->getArg(1), CE->getArg(0));
823 }
824 
825 void CStringChecker::evalMemcmp(CheckerContext &C, const CallExpr *CE) const {
826   // int memcmp(const void *s1, const void *s2, size_t n);
827   const Expr *Left = CE->getArg(0);
828   const Expr *Right = CE->getArg(1);
829   const Expr *Size = CE->getArg(2);
830 
831   const GRState *state = C.getState();
832   SValBuilder &svalBuilder = C.getSValBuilder();
833 
834   // See if the size argument is zero.
835   SVal sizeVal = state->getSVal(Size);
836   QualType sizeTy = Size->getType();
837 
838   const GRState *stateZeroSize, *stateNonZeroSize;
839   llvm::tie(stateZeroSize, stateNonZeroSize) =
840     assumeZero(C, state, sizeVal, sizeTy);
841 
842   // If the size can be zero, the result will be 0 in that case, and we don't
843   // have to check either of the buffers.
844   if (stateZeroSize) {
845     state = stateZeroSize;
846     state = state->BindExpr(CE, svalBuilder.makeZeroVal(CE->getType()));
847     C.addTransition(state);
848   }
849 
850   // If the size can be nonzero, we have to check the other arguments.
851   if (stateNonZeroSize) {
852     state = stateNonZeroSize;
853     // If we know the two buffers are the same, we know the result is 0.
854     // First, get the two buffers' addresses. Another checker will have already
855     // made sure they're not undefined.
856     DefinedOrUnknownSVal LV = cast<DefinedOrUnknownSVal>(state->getSVal(Left));
857     DefinedOrUnknownSVal RV = cast<DefinedOrUnknownSVal>(state->getSVal(Right));
858 
859     // See if they are the same.
860     DefinedOrUnknownSVal SameBuf = svalBuilder.evalEQ(state, LV, RV);
861     const GRState *StSameBuf, *StNotSameBuf;
862     llvm::tie(StSameBuf, StNotSameBuf) = state->assume(SameBuf);
863 
864     // If the two arguments might be the same buffer, we know the result is zero,
865     // and we only need to check one size.
866     if (StSameBuf) {
867       state = StSameBuf;
868       state = CheckBufferAccess(C, state, Size, Left);
869       if (state) {
870         state = StSameBuf->BindExpr(CE, svalBuilder.makeZeroVal(CE->getType()));
871         C.addTransition(state);
872       }
873     }
874 
875     // If the two arguments might be different buffers, we have to check the
876     // size of both of them.
877     if (StNotSameBuf) {
878       state = StNotSameBuf;
879       state = CheckBufferAccess(C, state, Size, Left, Right);
880       if (state) {
881         // The return value is the comparison result, which we don't know.
882         unsigned Count = C.getNodeBuilder().getCurrentBlockCount();
883         SVal CmpV = svalBuilder.getConjuredSymbolVal(NULL, CE, Count);
884         state = state->BindExpr(CE, CmpV);
885         C.addTransition(state);
886       }
887     }
888   }
889 }
890 
891 void CStringChecker::evalstrLength(CheckerContext &C,
892                                    const CallExpr *CE) const {
893   // size_t strlen(const char *s);
894   evalstrLengthCommon(C, CE, /* IsStrnlen = */ false);
895 }
896 
897 void CStringChecker::evalstrnLength(CheckerContext &C,
898                                     const CallExpr *CE) const {
899   // size_t strnlen(const char *s, size_t maxlen);
900   evalstrLengthCommon(C, CE, /* IsStrnlen = */ true);
901 }
902 
903 void CStringChecker::evalstrLengthCommon(CheckerContext &C, const CallExpr *CE,
904                                          bool IsStrnlen) const {
905   const GRState *state = C.getState();
906 
907   if (IsStrnlen) {
908     const Expr *maxlenExpr = CE->getArg(1);
909     SVal maxlenVal = state->getSVal(maxlenExpr);
910 
911     const GRState *stateZeroSize, *stateNonZeroSize;
912     llvm::tie(stateZeroSize, stateNonZeroSize) =
913       assumeZero(C, state, maxlenVal, maxlenExpr->getType());
914 
915     // If the size can be zero, the result will be 0 in that case, and we don't
916     // have to check the string itself.
917     if (stateZeroSize) {
918       SVal zero = C.getSValBuilder().makeZeroVal(CE->getType());
919       stateZeroSize = stateZeroSize->BindExpr(CE, zero);
920       C.addTransition(stateZeroSize);
921     }
922 
923     // If the size is GUARANTEED to be zero, we're done!
924     if (!stateNonZeroSize)
925       return;
926 
927     // Otherwise, record the assumption that the size is nonzero.
928     state = stateNonZeroSize;
929   }
930 
931   // Check that the string argument is non-null.
932   const Expr *Arg = CE->getArg(0);
933   SVal ArgVal = state->getSVal(Arg);
934 
935   state = checkNonNull(C, state, Arg, ArgVal);
936 
937   if (!state)
938     return;
939 
940   SVal strLength = getCStringLength(C, state, Arg, ArgVal);
941 
942   // If the argument isn't a valid C string, there's no valid state to
943   // transition to.
944   if (strLength.isUndef())
945     return;
946 
947   DefinedOrUnknownSVal result = UnknownVal();
948 
949   // If the check is for strnlen() then bind the return value to no more than
950   // the maxlen value.
951   if (IsStrnlen) {
952     QualType cmpTy = C.getSValBuilder().getContext().IntTy;
953 
954     // It's a little unfortunate to be getting this again,
955     // but it's not that expensive...
956     const Expr *maxlenExpr = CE->getArg(1);
957     SVal maxlenVal = state->getSVal(maxlenExpr);
958 
959     NonLoc *strLengthNL = dyn_cast<NonLoc>(&strLength);
960     NonLoc *maxlenValNL = dyn_cast<NonLoc>(&maxlenVal);
961 
962     if (strLengthNL && maxlenValNL) {
963       const GRState *stateStringTooLong, *stateStringNotTooLong;
964 
965       // Check if the strLength is greater than the maxlen.
966       llvm::tie(stateStringTooLong, stateStringNotTooLong) =
967         state->assume(cast<DefinedOrUnknownSVal>
968                       (C.getSValBuilder().evalBinOpNN(state, BO_GT,
969                                                       *strLengthNL,
970                                                       *maxlenValNL,
971                                                       cmpTy)));
972 
973       if (stateStringTooLong && !stateStringNotTooLong) {
974         // If the string is longer than maxlen, return maxlen.
975         result = *maxlenValNL;
976       } else if (stateStringNotTooLong && !stateStringTooLong) {
977         // If the string is shorter than maxlen, return its length.
978         result = *strLengthNL;
979       }
980     }
981 
982     if (result.isUnknown()) {
983       // If we don't have enough information for a comparison, there's
984       // no guarantee the full string length will actually be returned.
985       // All we know is the return value is the min of the string length
986       // and the limit. This is better than nothing.
987       unsigned Count = C.getNodeBuilder().getCurrentBlockCount();
988       result = C.getSValBuilder().getConjuredSymbolVal(NULL, CE, Count);
989       NonLoc *resultNL = cast<NonLoc>(&result);
990 
991       if (strLengthNL) {
992         state = state->assume(cast<DefinedOrUnknownSVal>
993                               (C.getSValBuilder().evalBinOpNN(state, BO_LE,
994                                                               *resultNL,
995                                                               *strLengthNL,
996                                                               cmpTy)), true);
997       }
998 
999       if (maxlenValNL) {
1000         state = state->assume(cast<DefinedOrUnknownSVal>
1001                               (C.getSValBuilder().evalBinOpNN(state, BO_LE,
1002                                                               *resultNL,
1003                                                               *maxlenValNL,
1004                                                               cmpTy)), true);
1005       }
1006     }
1007 
1008   } else {
1009     // This is a plain strlen(), not strnlen().
1010     result = cast<DefinedOrUnknownSVal>(strLength);
1011 
1012     // If we don't know the length of the string, conjure a return
1013     // value, so it can be used in constraints, at least.
1014     if (result.isUnknown()) {
1015       unsigned Count = C.getNodeBuilder().getCurrentBlockCount();
1016       result = C.getSValBuilder().getConjuredSymbolVal(NULL, CE, Count);
1017     }
1018   }
1019 
1020   // Bind the return value.
1021   assert(!result.isUnknown() && "Should have conjured a value by now");
1022   state = state->BindExpr(CE, result);
1023   C.addTransition(state);
1024 }
1025 
1026 void CStringChecker::evalStrcpy(CheckerContext &C, const CallExpr *CE) const {
1027   // char *strcpy(char *restrict dst, const char *restrict src);
1028   evalStrcpyCommon(C, CE,
1029                    /* returnEnd = */ false,
1030                    /* isBounded = */ false,
1031                    /* isAppending = */ false);
1032 }
1033 
1034 void CStringChecker::evalStrncpy(CheckerContext &C, const CallExpr *CE) const {
1035   // char *strncpy(char *restrict dst, const char *restrict src, size_t n);
1036   evalStrcpyCommon(C, CE,
1037                    /* returnEnd = */ false,
1038                    /* isBounded = */ true,
1039                    /* isAppending = */ false);
1040 }
1041 
1042 void CStringChecker::evalStpcpy(CheckerContext &C, const CallExpr *CE) const {
1043   // char *stpcpy(char *restrict dst, const char *restrict src);
1044   evalStrcpyCommon(C, CE,
1045                    /* returnEnd = */ true,
1046                    /* isBounded = */ false,
1047                    /* isAppending = */ false);
1048 }
1049 
1050 void CStringChecker::evalStrcat(CheckerContext &C, const CallExpr *CE) const {
1051   //char *strcat(char *restrict s1, const char *restrict s2);
1052   evalStrcpyCommon(C, CE,
1053                    /* returnEnd = */ false,
1054                    /* isBounded = */ false,
1055                    /* isAppending = */ true);
1056 }
1057 
1058 void CStringChecker::evalStrncat(CheckerContext &C, const CallExpr *CE) const {
1059   //char *strncat(char *restrict s1, const char *restrict s2, size_t n);
1060   evalStrcpyCommon(C, CE,
1061                    /* returnEnd = */ false,
1062                    /* isBounded = */ true,
1063                    /* isAppending = */ true);
1064 }
1065 
1066 void CStringChecker::evalStrcpyCommon(CheckerContext &C, const CallExpr *CE,
1067                                       bool returnEnd, bool isBounded,
1068                                       bool isAppending) const {
1069   const GRState *state = C.getState();
1070 
1071   // Check that the destination is non-null.
1072   const Expr *Dst = CE->getArg(0);
1073   SVal DstVal = state->getSVal(Dst);
1074 
1075   state = checkNonNull(C, state, Dst, DstVal);
1076   if (!state)
1077     return;
1078 
1079   // Check that the source is non-null.
1080   const Expr *srcExpr = CE->getArg(1);
1081   SVal srcVal = state->getSVal(srcExpr);
1082   state = checkNonNull(C, state, srcExpr, srcVal);
1083   if (!state)
1084     return;
1085 
1086   // Get the string length of the source.
1087   SVal strLength = getCStringLength(C, state, srcExpr, srcVal);
1088 
1089   // If the source isn't a valid C string, give up.
1090   if (strLength.isUndef())
1091     return;
1092 
1093   // If the function is strncpy, strncat, etc... it is bounded.
1094   if (isBounded) {
1095     // Get the max number of characters to copy.
1096     const Expr *lenExpr = CE->getArg(2);
1097     SVal lenVal = state->getSVal(lenExpr);
1098 
1099     // Cast the length to a NonLoc SVal. If it is not a NonLoc then give up.
1100     NonLoc *strLengthNL = dyn_cast<NonLoc>(&strLength);
1101     if (!strLengthNL)
1102       return;
1103 
1104     // Cast the max length to a NonLoc SVal. If it is not a NonLoc then give up.
1105     NonLoc *lenValNL = dyn_cast<NonLoc>(&lenVal);
1106     if (!lenValNL)
1107       return;
1108 
1109     QualType cmpTy = C.getSValBuilder().getContext().IntTy;
1110     const GRState *stateTrue, *stateFalse;
1111 
1112     // Check if the max number to copy is less than the length of the src.
1113     llvm::tie(stateTrue, stateFalse) =
1114       state->assume(cast<DefinedOrUnknownSVal>
1115                     (C.getSValBuilder().evalBinOpNN(state, BO_GT,
1116                                                     *strLengthNL, *lenValNL,
1117                                                     cmpTy)));
1118 
1119     if (stateTrue) {
1120       // Max number to copy is less than the length of the src, so the actual
1121       // strLength copied is the max number arg.
1122       strLength = lenVal;
1123     }
1124   }
1125 
1126   // If this is an appending function (strcat, strncat...) then set the
1127   // string length to strlen(src) + strlen(dst) since the buffer will
1128   // ultimately contain both.
1129   if (isAppending) {
1130     // Get the string length of the destination, or give up.
1131     SVal dstStrLength = getCStringLength(C, state, Dst, DstVal);
1132     if (dstStrLength.isUndef())
1133       return;
1134 
1135     NonLoc *srcStrLengthNL = dyn_cast<NonLoc>(&strLength);
1136     NonLoc *dstStrLengthNL = dyn_cast<NonLoc>(&dstStrLength);
1137 
1138     // If src or dst cast to NonLoc is NULL, give up.
1139     if ((!srcStrLengthNL) || (!dstStrLengthNL))
1140       return;
1141 
1142     QualType addTy = C.getSValBuilder().getContext().getSizeType();
1143 
1144     strLength = C.getSValBuilder().evalBinOpNN(state, BO_Add,
1145                                                *srcStrLengthNL, *dstStrLengthNL,
1146                                                addTy);
1147   }
1148 
1149   SVal Result = (returnEnd ? UnknownVal() : DstVal);
1150 
1151   // If the destination is a MemRegion, try to check for a buffer overflow and
1152   // record the new string length.
1153   if (loc::MemRegionVal *dstRegVal = dyn_cast<loc::MemRegionVal>(&DstVal)) {
1154     // If the length is known, we can check for an overflow.
1155     if (NonLoc *knownStrLength = dyn_cast<NonLoc>(&strLength)) {
1156       SVal lastElement =
1157         C.getSValBuilder().evalBinOpLN(state, BO_Add, *dstRegVal,
1158                                        *knownStrLength, Dst->getType());
1159 
1160       state = CheckLocation(C, state, Dst, lastElement, /* IsDst = */ true);
1161       if (!state)
1162         return;
1163 
1164       // If this is a stpcpy-style copy, the last element is the return value.
1165       if (returnEnd)
1166         Result = lastElement;
1167     }
1168 
1169     // Invalidate the destination. This must happen before we set the C string
1170     // length because invalidation will clear the length.
1171     // FIXME: Even if we can't perfectly model the copy, we should see if we
1172     // can use LazyCompoundVals to copy the source values into the destination.
1173     // This would probably remove any existing bindings past the end of the
1174     // string, but that's still an improvement over blank invalidation.
1175     state = InvalidateBuffer(C, state, Dst, *dstRegVal);
1176 
1177     // Set the C string length of the destination.
1178     state = setCStringLength(state, dstRegVal->getRegion(), strLength);
1179   }
1180 
1181   // If this is a stpcpy-style copy, but we were unable to check for a buffer
1182   // overflow, we still need a result. Conjure a return value.
1183   if (returnEnd && Result.isUnknown()) {
1184     SValBuilder &svalBuilder = C.getSValBuilder();
1185     unsigned Count = C.getNodeBuilder().getCurrentBlockCount();
1186     strLength = svalBuilder.getConjuredSymbolVal(NULL, CE, Count);
1187   }
1188 
1189   // Set the return value.
1190   state = state->BindExpr(CE, Result);
1191   C.addTransition(state);
1192 }
1193 
1194 void CStringChecker::evalStrcmp(CheckerContext &C, const CallExpr *CE) const {
1195   //int strcmp(const char *restrict s1, const char *restrict s2);
1196   evalStrcmpCommon(C, CE, /* isBounded = */ false, /* ignoreCase = */ false);
1197 }
1198 
1199 void CStringChecker::evalStrncmp(CheckerContext &C, const CallExpr *CE) const {
1200   //int strncmp(const char *restrict s1, const char *restrict s2, size_t n);
1201   evalStrcmpCommon(C, CE, /* isBounded = */ true, /* ignoreCase = */ false);
1202 }
1203 
1204 void CStringChecker::evalStrcasecmp(CheckerContext &C,
1205                                     const CallExpr *CE) const {
1206   //int strcasecmp(const char *restrict s1, const char *restrict s2);
1207   evalStrcmpCommon(C, CE, /* isBounded = */ false, /* ignoreCase = */ true);
1208 }
1209 
1210 void CStringChecker::evalStrncasecmp(CheckerContext &C,
1211                                      const CallExpr *CE) const {
1212   //int strncasecmp(const char *restrict s1, const char *restrict s2, size_t n);
1213   evalStrcmpCommon(C, CE, /* isBounded = */ true, /* ignoreCase = */ true);
1214 }
1215 
1216 void CStringChecker::evalStrcmpCommon(CheckerContext &C, const CallExpr *CE,
1217                                       bool isBounded, bool ignoreCase) const {
1218   const GRState *state = C.getState();
1219 
1220   // Check that the first string is non-null
1221   const Expr *s1 = CE->getArg(0);
1222   SVal s1Val = state->getSVal(s1);
1223   state = checkNonNull(C, state, s1, s1Val);
1224   if (!state)
1225     return;
1226 
1227   // Check that the second string is non-null.
1228   const Expr *s2 = CE->getArg(1);
1229   SVal s2Val = state->getSVal(s2);
1230   state = checkNonNull(C, state, s2, s2Val);
1231   if (!state)
1232     return;
1233 
1234   // Get the string length of the first string or give up.
1235   SVal s1Length = getCStringLength(C, state, s1, s1Val);
1236   if (s1Length.isUndef())
1237     return;
1238 
1239   // Get the string length of the second string or give up.
1240   SVal s2Length = getCStringLength(C, state, s2, s2Val);
1241   if (s2Length.isUndef())
1242     return;
1243 
1244   // Get the string literal of the first string.
1245   const StringLiteral *s1StrLiteral = getCStringLiteral(C, state, s1, s1Val);
1246   if (!s1StrLiteral)
1247     return;
1248   llvm::StringRef s1StrRef = s1StrLiteral->getString();
1249 
1250   // Get the string literal of the second string.
1251   const StringLiteral *s2StrLiteral = getCStringLiteral(C, state, s2, s2Val);
1252   if (!s2StrLiteral)
1253     return;
1254   llvm::StringRef s2StrRef = s2StrLiteral->getString();
1255 
1256   int result;
1257   if (isBounded) {
1258     // Get the max number of characters to compare.
1259     const Expr *lenExpr = CE->getArg(2);
1260     SVal lenVal = state->getSVal(lenExpr);
1261 
1262     // Dynamically cast the length to a ConcreteInt. If it is not a ConcreteInt
1263     // then give up, otherwise get the value and use it as the bounds.
1264     nonloc::ConcreteInt *CI = dyn_cast<nonloc::ConcreteInt>(&lenVal);
1265     if (!CI)
1266       return;
1267     llvm::APSInt lenInt(CI->getValue());
1268 
1269     // Create substrings of each to compare the prefix.
1270     s1StrRef = s1StrRef.substr(0, (size_t)lenInt.getLimitedValue());
1271     s2StrRef = s2StrRef.substr(0, (size_t)lenInt.getLimitedValue());
1272   }
1273 
1274   if (ignoreCase) {
1275     // Compare string 1 to string 2 the same way strcasecmp() does.
1276     result = s1StrRef.compare_lower(s2StrRef);
1277   } else {
1278     // Compare string 1 to string 2 the same way strcmp() does.
1279     result = s1StrRef.compare(s2StrRef);
1280   }
1281 
1282   // Build the SVal of the comparison to bind the return value.
1283   SValBuilder &svalBuilder = C.getSValBuilder();
1284   QualType intTy = svalBuilder.getContext().IntTy;
1285   SVal resultVal = svalBuilder.makeIntVal(result, intTy);
1286 
1287   // Bind the return value of the expression.
1288   // Set the return value.
1289   state = state->BindExpr(CE, resultVal);
1290   C.addTransition(state);
1291 }
1292 
1293 //===----------------------------------------------------------------------===//
1294 // The driver method, and other Checker callbacks.
1295 //===----------------------------------------------------------------------===//
1296 
1297 bool CStringChecker::evalCall(const CallExpr *CE, CheckerContext &C) const {
1298   // Get the callee.  All the functions we care about are C functions
1299   // with simple identifiers.
1300   const GRState *state = C.getState();
1301   const Expr *Callee = CE->getCallee();
1302   const FunctionDecl *FD = state->getSVal(Callee).getAsFunctionDecl();
1303 
1304   if (!FD)
1305     return false;
1306 
1307   // Get the name of the callee. If it's a builtin, strip off the prefix.
1308   IdentifierInfo *II = FD->getIdentifier();
1309   if (!II)   // if no identifier, not a simple C function
1310     return false;
1311   llvm::StringRef Name = II->getName();
1312   if (Name.startswith("__builtin_"))
1313     Name = Name.substr(10);
1314 
1315   FnCheck evalFunction = llvm::StringSwitch<FnCheck>(Name)
1316     .Cases("memcpy", "__memcpy_chk", &CStringChecker::evalMemcpy)
1317     .Cases("mempcpy", "__mempcpy_chk", &CStringChecker::evalMempcpy)
1318     .Cases("memcmp", "bcmp", &CStringChecker::evalMemcmp)
1319     .Cases("memmove", "__memmove_chk", &CStringChecker::evalMemmove)
1320     .Cases("strcpy", "__strcpy_chk", &CStringChecker::evalStrcpy)
1321     //.Cases("strncpy", "__strncpy_chk", &CStringChecker::evalStrncpy)
1322     .Cases("stpcpy", "__stpcpy_chk", &CStringChecker::evalStpcpy)
1323     .Cases("strcat", "__strcat_chk", &CStringChecker::evalStrcat)
1324     .Cases("strncat", "__strncat_chk", &CStringChecker::evalStrncat)
1325     .Case("strlen", &CStringChecker::evalstrLength)
1326     .Case("strnlen", &CStringChecker::evalstrnLength)
1327     .Case("strcmp", &CStringChecker::evalStrcmp)
1328     .Case("strncmp", &CStringChecker::evalStrncmp)
1329     .Case("strcasecmp", &CStringChecker::evalStrcasecmp)
1330     .Case("strncasecmp", &CStringChecker::evalStrncasecmp)
1331     .Case("bcopy", &CStringChecker::evalBcopy)
1332     .Default(NULL);
1333 
1334   // If the callee isn't a string function, let another checker handle it.
1335   if (!evalFunction)
1336     return false;
1337 
1338   // Check and evaluate the call.
1339   (this->*evalFunction)(C, CE);
1340   return true;
1341 }
1342 
1343 void CStringChecker::checkPreStmt(const DeclStmt *DS, CheckerContext &C) const {
1344   // Record string length for char a[] = "abc";
1345   const GRState *state = C.getState();
1346 
1347   for (DeclStmt::const_decl_iterator I = DS->decl_begin(), E = DS->decl_end();
1348        I != E; ++I) {
1349     const VarDecl *D = dyn_cast<VarDecl>(*I);
1350     if (!D)
1351       continue;
1352 
1353     // FIXME: Handle array fields of structs.
1354     if (!D->getType()->isArrayType())
1355       continue;
1356 
1357     const Expr *Init = D->getInit();
1358     if (!Init)
1359       continue;
1360     if (!isa<StringLiteral>(Init))
1361       continue;
1362 
1363     Loc VarLoc = state->getLValue(D, C.getPredecessor()->getLocationContext());
1364     const MemRegion *MR = VarLoc.getAsRegion();
1365     if (!MR)
1366       continue;
1367 
1368     SVal StrVal = state->getSVal(Init);
1369     assert(StrVal.isValid() && "Initializer string is unknown or undefined");
1370     DefinedOrUnknownSVal strLength
1371       = cast<DefinedOrUnknownSVal>(getCStringLength(C, state, Init, StrVal));
1372 
1373     state = state->set<CStringLength>(MR, strLength);
1374   }
1375 
1376   C.addTransition(state);
1377 }
1378 
1379 bool CStringChecker::wantsRegionChangeUpdate(const GRState *state) const {
1380   CStringLength::EntryMap Entries = state->get<CStringLength>();
1381   return !Entries.isEmpty();
1382 }
1383 
1384 const GRState *
1385 CStringChecker::checkRegionChanges(const GRState *state,
1386                                    const StoreManager::InvalidatedSymbols *,
1387                                    const MemRegion * const *Begin,
1388                                    const MemRegion * const *End) const {
1389   CStringLength::EntryMap Entries = state->get<CStringLength>();
1390   if (Entries.isEmpty())
1391     return state;
1392 
1393   llvm::SmallPtrSet<const MemRegion *, 8> Invalidated;
1394   llvm::SmallPtrSet<const MemRegion *, 32> SuperRegions;
1395 
1396   // First build sets for the changed regions and their super-regions.
1397   for ( ; Begin != End; ++Begin) {
1398     const MemRegion *MR = *Begin;
1399     Invalidated.insert(MR);
1400 
1401     SuperRegions.insert(MR);
1402     while (const SubRegion *SR = dyn_cast<SubRegion>(MR)) {
1403       MR = SR->getSuperRegion();
1404       SuperRegions.insert(MR);
1405     }
1406   }
1407 
1408   CStringLength::EntryMap::Factory &F = state->get_context<CStringLength>();
1409 
1410   // Then loop over the entries in the current state.
1411   for (CStringLength::EntryMap::iterator I = Entries.begin(),
1412        E = Entries.end(); I != E; ++I) {
1413     const MemRegion *MR = I.getKey();
1414 
1415     // Is this entry for a super-region of a changed region?
1416     if (SuperRegions.count(MR)) {
1417       Entries = F.remove(Entries, MR);
1418       continue;
1419     }
1420 
1421     // Is this entry for a sub-region of a changed region?
1422     const MemRegion *Super = MR;
1423     while (const SubRegion *SR = dyn_cast<SubRegion>(Super)) {
1424       Super = SR->getSuperRegion();
1425       if (Invalidated.count(Super)) {
1426         Entries = F.remove(Entries, MR);
1427         break;
1428       }
1429     }
1430   }
1431 
1432   return state->set<CStringLength>(Entries);
1433 }
1434 
1435 void CStringChecker::checkLiveSymbols(const GRState *state,
1436                                       SymbolReaper &SR) const {
1437   // Mark all symbols in our string length map as valid.
1438   CStringLength::EntryMap Entries = state->get<CStringLength>();
1439 
1440   for (CStringLength::EntryMap::iterator I = Entries.begin(), E = Entries.end();
1441        I != E; ++I) {
1442     SVal Len = I.getData();
1443     if (SymbolRef Sym = Len.getAsSymbol())
1444       SR.markInUse(Sym);
1445   }
1446 }
1447 
1448 void CStringChecker::checkDeadSymbols(SymbolReaper &SR,
1449                                       CheckerContext &C) const {
1450   if (!SR.hasDeadSymbols())
1451     return;
1452 
1453   const GRState *state = C.getState();
1454   CStringLength::EntryMap Entries = state->get<CStringLength>();
1455   if (Entries.isEmpty())
1456     return;
1457 
1458   CStringLength::EntryMap::Factory &F = state->get_context<CStringLength>();
1459   for (CStringLength::EntryMap::iterator I = Entries.begin(), E = Entries.end();
1460        I != E; ++I) {
1461     SVal Len = I.getData();
1462     if (SymbolRef Sym = Len.getAsSymbol()) {
1463       if (SR.isDead(Sym))
1464         Entries = F.remove(Entries, I.getKey());
1465     }
1466   }
1467 
1468   state = state->set<CStringLength>(Entries);
1469   C.generateNode(state);
1470 }
1471 
1472 void ento::registerCStringChecker(CheckerManager &mgr) {
1473   mgr.registerChecker<CStringChecker>();
1474 }
1475