1 //= CStringChecker.cpp - 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 "clang/StaticAnalyzer/Checkers/BuiltinCheckerRegistration.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"
26
27 using namespace clang;
28 using namespace ento;
29
30 namespace {
31 class CStringChecker : public Checker< eval::Call,
32 check::PreStmt<DeclStmt>,
33 check::LiveSymbols,
34 check::DeadSymbols,
35 check::RegionChanges
36 > {
37 mutable std::unique_ptr<BugType> BT_Null, BT_Bounds, BT_Overlap,
38 BT_NotCString, BT_AdditionOverflow;
39
40 mutable const char *CurrentFunctionDescription;
41
42 public:
43 /// The filter is used to filter out the diagnostics which are not enabled by
44 /// the user.
45 struct CStringChecksFilter {
46 DefaultBool CheckCStringNullArg;
47 DefaultBool CheckCStringOutOfBounds;
48 DefaultBool CheckCStringBufferOverlap;
49 DefaultBool CheckCStringNotNullTerm;
50
51 CheckName CheckNameCStringNullArg;
52 CheckName CheckNameCStringOutOfBounds;
53 CheckName CheckNameCStringBufferOverlap;
54 CheckName CheckNameCStringNotNullTerm;
55 };
56
57 CStringChecksFilter Filter;
58
getTag()59 static void *getTag() { static int tag; return &tag; }
60
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;
65
66 ProgramStateRef
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;
73
74 typedef void (CStringChecker::*FnCheck)(CheckerContext &,
75 const CallExpr *) const;
76
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,
83 const Expr *Size,
84 const Expr *Source,
85 const Expr *Dest,
86 bool Restricted = false,
87 bool IsMempcpy = false) const;
88
89 void evalMemcmp(CheckerContext &C, const CallExpr *CE) const;
90
91 void evalstrLength(CheckerContext &C, const CallExpr *CE) const;
92 void evalstrnLength(CheckerContext &C, const CallExpr *CE) const;
93 void evalstrLengthCommon(CheckerContext &C,
94 const CallExpr *CE,
95 bool IsStrnlen = false) const;
96
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 evalStrlcpy(CheckerContext &C, const CallExpr *CE) const;
101 void evalStrcpyCommon(CheckerContext &C,
102 const CallExpr *CE,
103 bool returnEnd,
104 bool isBounded,
105 bool isAppending,
106 bool returnPtr = true) const;
107
108 void evalStrcat(CheckerContext &C, const CallExpr *CE) const;
109 void evalStrncat(CheckerContext &C, const CallExpr *CE) const;
110 void evalStrlcat(CheckerContext &C, const CallExpr *CE) const;
111
112 void evalStrcmp(CheckerContext &C, const CallExpr *CE) const;
113 void evalStrncmp(CheckerContext &C, const CallExpr *CE) const;
114 void evalStrcasecmp(CheckerContext &C, const CallExpr *CE) const;
115 void evalStrncasecmp(CheckerContext &C, const CallExpr *CE) const;
116 void evalStrcmpCommon(CheckerContext &C,
117 const CallExpr *CE,
118 bool isBounded = false,
119 bool ignoreCase = false) const;
120
121 void evalStrsep(CheckerContext &C, const CallExpr *CE) const;
122
123 void evalStdCopy(CheckerContext &C, const CallExpr *CE) const;
124 void evalStdCopyBackward(CheckerContext &C, const CallExpr *CE) const;
125 void evalStdCopyCommon(CheckerContext &C, const CallExpr *CE) const;
126 void evalMemset(CheckerContext &C, const CallExpr *CE) const;
127 void evalBzero(CheckerContext &C, const CallExpr *CE) const;
128
129 // Utility methods
130 std::pair<ProgramStateRef , ProgramStateRef >
131 static assumeZero(CheckerContext &C,
132 ProgramStateRef state, SVal V, QualType Ty);
133
134 static ProgramStateRef setCStringLength(ProgramStateRef state,
135 const MemRegion *MR,
136 SVal strLength);
137 static SVal getCStringLengthForRegion(CheckerContext &C,
138 ProgramStateRef &state,
139 const Expr *Ex,
140 const MemRegion *MR,
141 bool hypothetical);
142 SVal getCStringLength(CheckerContext &C,
143 ProgramStateRef &state,
144 const Expr *Ex,
145 SVal Buf,
146 bool hypothetical = false) const;
147
148 const StringLiteral *getCStringLiteral(CheckerContext &C,
149 ProgramStateRef &state,
150 const Expr *expr,
151 SVal val) const;
152
153 static ProgramStateRef InvalidateBuffer(CheckerContext &C,
154 ProgramStateRef state,
155 const Expr *Ex, SVal V,
156 bool IsSourceBuffer,
157 const Expr *Size);
158
159 static bool SummarizeRegion(raw_ostream &os, ASTContext &Ctx,
160 const MemRegion *MR);
161
162 static bool memsetAux(const Expr *DstBuffer, SVal CharE,
163 const Expr *Size, CheckerContext &C,
164 ProgramStateRef &State);
165
166 // Re-usable checks
167 ProgramStateRef checkNonNull(CheckerContext &C,
168 ProgramStateRef state,
169 const Expr *S,
170 SVal l) const;
171 ProgramStateRef CheckLocation(CheckerContext &C,
172 ProgramStateRef state,
173 const Expr *S,
174 SVal l,
175 const char *message = nullptr) const;
176 ProgramStateRef CheckBufferAccess(CheckerContext &C,
177 ProgramStateRef state,
178 const Expr *Size,
179 const Expr *FirstBuf,
180 const Expr *SecondBuf,
181 const char *firstMessage = nullptr,
182 const char *secondMessage = nullptr,
183 bool WarnAboutSize = false) const;
184
CheckBufferAccess(CheckerContext & C,ProgramStateRef state,const Expr * Size,const Expr * Buf,const char * message=nullptr,bool WarnAboutSize=false) const185 ProgramStateRef CheckBufferAccess(CheckerContext &C,
186 ProgramStateRef state,
187 const Expr *Size,
188 const Expr *Buf,
189 const char *message = nullptr,
190 bool WarnAboutSize = false) const {
191 // This is a convenience overload.
192 return CheckBufferAccess(C, state, Size, Buf, nullptr, message, nullptr,
193 WarnAboutSize);
194 }
195 ProgramStateRef CheckOverlap(CheckerContext &C,
196 ProgramStateRef state,
197 const Expr *Size,
198 const Expr *First,
199 const Expr *Second) const;
200 void emitOverlapBug(CheckerContext &C,
201 ProgramStateRef state,
202 const Stmt *First,
203 const Stmt *Second) const;
204
205 void emitNullArgBug(CheckerContext &C, ProgramStateRef State, const Stmt *S,
206 StringRef WarningMsg) const;
207 void emitOutOfBoundsBug(CheckerContext &C, ProgramStateRef State,
208 const Stmt *S, StringRef WarningMsg) const;
209 void emitNotCStringBug(CheckerContext &C, ProgramStateRef State,
210 const Stmt *S, StringRef WarningMsg) const;
211 void emitAdditionOverflowBug(CheckerContext &C, ProgramStateRef State) const;
212
213 ProgramStateRef checkAdditionOverflow(CheckerContext &C,
214 ProgramStateRef state,
215 NonLoc left,
216 NonLoc right) const;
217
218 // Return true if the destination buffer of the copy function may be in bound.
219 // Expects SVal of Size to be positive and unsigned.
220 // Expects SVal of FirstBuf to be a FieldRegion.
221 static bool IsFirstBufInBound(CheckerContext &C,
222 ProgramStateRef state,
223 const Expr *FirstBuf,
224 const Expr *Size);
225 };
226
227 } //end anonymous namespace
228
REGISTER_MAP_WITH_PROGRAMSTATE(CStringLength,const MemRegion *,SVal)229 REGISTER_MAP_WITH_PROGRAMSTATE(CStringLength, const MemRegion *, SVal)
230
231 //===----------------------------------------------------------------------===//
232 // Individual checks and utility methods.
233 //===----------------------------------------------------------------------===//
234
235 std::pair<ProgramStateRef , ProgramStateRef >
236 CStringChecker::assumeZero(CheckerContext &C, ProgramStateRef state, SVal V,
237 QualType Ty) {
238 Optional<DefinedSVal> val = V.getAs<DefinedSVal>();
239 if (!val)
240 return std::pair<ProgramStateRef , ProgramStateRef >(state, state);
241
242 SValBuilder &svalBuilder = C.getSValBuilder();
243 DefinedOrUnknownSVal zero = svalBuilder.makeZeroVal(Ty);
244 return state->assume(svalBuilder.evalEQ(state, *val, zero));
245 }
246
checkNonNull(CheckerContext & C,ProgramStateRef state,const Expr * S,SVal l) const247 ProgramStateRef CStringChecker::checkNonNull(CheckerContext &C,
248 ProgramStateRef state,
249 const Expr *S, SVal l) const {
250 // If a previous check has failed, propagate the failure.
251 if (!state)
252 return nullptr;
253
254 ProgramStateRef stateNull, stateNonNull;
255 std::tie(stateNull, stateNonNull) = assumeZero(C, state, l, S->getType());
256
257 if (stateNull && !stateNonNull) {
258 if (Filter.CheckCStringNullArg) {
259 SmallString<80> buf;
260 llvm::raw_svector_ostream os(buf);
261 assert(CurrentFunctionDescription);
262 os << "Null pointer argument in call to " << CurrentFunctionDescription;
263
264 emitNullArgBug(C, stateNull, S, os.str());
265 }
266 return nullptr;
267 }
268
269 // From here on, assume that the value is non-null.
270 assert(stateNonNull);
271 return stateNonNull;
272 }
273
274 // FIXME: This was originally copied from ArrayBoundChecker.cpp. Refactor?
CheckLocation(CheckerContext & C,ProgramStateRef state,const Expr * S,SVal l,const char * warningMsg) const275 ProgramStateRef CStringChecker::CheckLocation(CheckerContext &C,
276 ProgramStateRef state,
277 const Expr *S, SVal l,
278 const char *warningMsg) const {
279 // If a previous check has failed, propagate the failure.
280 if (!state)
281 return nullptr;
282
283 // Check for out of bound array element access.
284 const MemRegion *R = l.getAsRegion();
285 if (!R)
286 return state;
287
288 const ElementRegion *ER = dyn_cast<ElementRegion>(R);
289 if (!ER)
290 return state;
291
292 if (ER->getValueType() != C.getASTContext().CharTy)
293 return state;
294
295 // Get the size of the array.
296 const SubRegion *superReg = cast<SubRegion>(ER->getSuperRegion());
297 SValBuilder &svalBuilder = C.getSValBuilder();
298 SVal Extent =
299 svalBuilder.convertToArrayIndex(superReg->getExtent(svalBuilder));
300 DefinedOrUnknownSVal Size = Extent.castAs<DefinedOrUnknownSVal>();
301
302 // Get the index of the accessed element.
303 DefinedOrUnknownSVal Idx = ER->getIndex().castAs<DefinedOrUnknownSVal>();
304
305 ProgramStateRef StInBound = state->assumeInBound(Idx, Size, true);
306 ProgramStateRef StOutBound = state->assumeInBound(Idx, Size, false);
307 if (StOutBound && !StInBound) {
308 // These checks are either enabled by the CString out-of-bounds checker
309 // explicitly or implicitly by the Malloc checker.
310 // In the latter case we only do modeling but do not emit warning.
311 if (!Filter.CheckCStringOutOfBounds)
312 return nullptr;
313 // Emit a bug report.
314 if (warningMsg) {
315 emitOutOfBoundsBug(C, StOutBound, S, warningMsg);
316 } else {
317 assert(CurrentFunctionDescription);
318 assert(CurrentFunctionDescription[0] != '\0');
319
320 SmallString<80> buf;
321 llvm::raw_svector_ostream os(buf);
322 os << toUppercase(CurrentFunctionDescription[0])
323 << &CurrentFunctionDescription[1]
324 << " accesses out-of-bound array element";
325 emitOutOfBoundsBug(C, StOutBound, S, os.str());
326 }
327 return nullptr;
328 }
329
330 // Array bound check succeeded. From this point forward the array bound
331 // should always succeed.
332 return StInBound;
333 }
334
CheckBufferAccess(CheckerContext & C,ProgramStateRef state,const Expr * Size,const Expr * FirstBuf,const Expr * SecondBuf,const char * firstMessage,const char * secondMessage,bool WarnAboutSize) const335 ProgramStateRef CStringChecker::CheckBufferAccess(CheckerContext &C,
336 ProgramStateRef state,
337 const Expr *Size,
338 const Expr *FirstBuf,
339 const Expr *SecondBuf,
340 const char *firstMessage,
341 const char *secondMessage,
342 bool WarnAboutSize) const {
343 // If a previous check has failed, propagate the failure.
344 if (!state)
345 return nullptr;
346
347 SValBuilder &svalBuilder = C.getSValBuilder();
348 ASTContext &Ctx = svalBuilder.getContext();
349 const LocationContext *LCtx = C.getLocationContext();
350
351 QualType sizeTy = Size->getType();
352 QualType PtrTy = Ctx.getPointerType(Ctx.CharTy);
353
354 // Check that the first buffer is non-null.
355 SVal BufVal = C.getSVal(FirstBuf);
356 state = checkNonNull(C, state, FirstBuf, BufVal);
357 if (!state)
358 return nullptr;
359
360 // If out-of-bounds checking is turned off, skip the rest.
361 if (!Filter.CheckCStringOutOfBounds)
362 return state;
363
364 // Get the access length and make sure it is known.
365 // FIXME: This assumes the caller has already checked that the access length
366 // is positive. And that it's unsigned.
367 SVal LengthVal = C.getSVal(Size);
368 Optional<NonLoc> Length = LengthVal.getAs<NonLoc>();
369 if (!Length)
370 return state;
371
372 // Compute the offset of the last element to be accessed: size-1.
373 NonLoc One = svalBuilder.makeIntVal(1, sizeTy).castAs<NonLoc>();
374 SVal Offset = svalBuilder.evalBinOpNN(state, BO_Sub, *Length, One, sizeTy);
375 if (Offset.isUnknown())
376 return nullptr;
377 NonLoc LastOffset = Offset.castAs<NonLoc>();
378
379 // Check that the first buffer is sufficiently long.
380 SVal BufStart = svalBuilder.evalCast(BufVal, PtrTy, FirstBuf->getType());
381 if (Optional<Loc> BufLoc = BufStart.getAs<Loc>()) {
382 const Expr *warningExpr = (WarnAboutSize ? Size : FirstBuf);
383
384 SVal BufEnd = svalBuilder.evalBinOpLN(state, BO_Add, *BufLoc,
385 LastOffset, PtrTy);
386 state = CheckLocation(C, state, warningExpr, BufEnd, firstMessage);
387
388 // If the buffer isn't large enough, abort.
389 if (!state)
390 return nullptr;
391 }
392
393 // If there's a second buffer, check it as well.
394 if (SecondBuf) {
395 BufVal = state->getSVal(SecondBuf, LCtx);
396 state = checkNonNull(C, state, SecondBuf, BufVal);
397 if (!state)
398 return nullptr;
399
400 BufStart = svalBuilder.evalCast(BufVal, PtrTy, SecondBuf->getType());
401 if (Optional<Loc> BufLoc = BufStart.getAs<Loc>()) {
402 const Expr *warningExpr = (WarnAboutSize ? Size : SecondBuf);
403
404 SVal BufEnd = svalBuilder.evalBinOpLN(state, BO_Add, *BufLoc,
405 LastOffset, PtrTy);
406 state = CheckLocation(C, state, warningExpr, BufEnd, secondMessage);
407 }
408 }
409
410 // Large enough or not, return this state!
411 return state;
412 }
413
CheckOverlap(CheckerContext & C,ProgramStateRef state,const Expr * Size,const Expr * First,const Expr * Second) const414 ProgramStateRef CStringChecker::CheckOverlap(CheckerContext &C,
415 ProgramStateRef state,
416 const Expr *Size,
417 const Expr *First,
418 const Expr *Second) const {
419 if (!Filter.CheckCStringBufferOverlap)
420 return state;
421
422 // Do a simple check for overlap: if the two arguments are from the same
423 // buffer, see if the end of the first is greater than the start of the second
424 // or vice versa.
425
426 // If a previous check has failed, propagate the failure.
427 if (!state)
428 return nullptr;
429
430 ProgramStateRef stateTrue, stateFalse;
431
432 // Get the buffer values and make sure they're known locations.
433 const LocationContext *LCtx = C.getLocationContext();
434 SVal firstVal = state->getSVal(First, LCtx);
435 SVal secondVal = state->getSVal(Second, LCtx);
436
437 Optional<Loc> firstLoc = firstVal.getAs<Loc>();
438 if (!firstLoc)
439 return state;
440
441 Optional<Loc> secondLoc = secondVal.getAs<Loc>();
442 if (!secondLoc)
443 return state;
444
445 // Are the two values the same?
446 SValBuilder &svalBuilder = C.getSValBuilder();
447 std::tie(stateTrue, stateFalse) =
448 state->assume(svalBuilder.evalEQ(state, *firstLoc, *secondLoc));
449
450 if (stateTrue && !stateFalse) {
451 // If the values are known to be equal, that's automatically an overlap.
452 emitOverlapBug(C, stateTrue, First, Second);
453 return nullptr;
454 }
455
456 // assume the two expressions are not equal.
457 assert(stateFalse);
458 state = stateFalse;
459
460 // Which value comes first?
461 QualType cmpTy = svalBuilder.getConditionType();
462 SVal reverse = svalBuilder.evalBinOpLL(state, BO_GT,
463 *firstLoc, *secondLoc, cmpTy);
464 Optional<DefinedOrUnknownSVal> reverseTest =
465 reverse.getAs<DefinedOrUnknownSVal>();
466 if (!reverseTest)
467 return state;
468
469 std::tie(stateTrue, stateFalse) = state->assume(*reverseTest);
470 if (stateTrue) {
471 if (stateFalse) {
472 // If we don't know which one comes first, we can't perform this test.
473 return state;
474 } else {
475 // Switch the values so that firstVal is before secondVal.
476 std::swap(firstLoc, secondLoc);
477
478 // Switch the Exprs as well, so that they still correspond.
479 std::swap(First, Second);
480 }
481 }
482
483 // Get the length, and make sure it too is known.
484 SVal LengthVal = state->getSVal(Size, LCtx);
485 Optional<NonLoc> Length = LengthVal.getAs<NonLoc>();
486 if (!Length)
487 return state;
488
489 // Convert the first buffer's start address to char*.
490 // Bail out if the cast fails.
491 ASTContext &Ctx = svalBuilder.getContext();
492 QualType CharPtrTy = Ctx.getPointerType(Ctx.CharTy);
493 SVal FirstStart = svalBuilder.evalCast(*firstLoc, CharPtrTy,
494 First->getType());
495 Optional<Loc> FirstStartLoc = FirstStart.getAs<Loc>();
496 if (!FirstStartLoc)
497 return state;
498
499 // Compute the end of the first buffer. Bail out if THAT fails.
500 SVal FirstEnd = svalBuilder.evalBinOpLN(state, BO_Add,
501 *FirstStartLoc, *Length, CharPtrTy);
502 Optional<Loc> FirstEndLoc = FirstEnd.getAs<Loc>();
503 if (!FirstEndLoc)
504 return state;
505
506 // Is the end of the first buffer past the start of the second buffer?
507 SVal Overlap = svalBuilder.evalBinOpLL(state, BO_GT,
508 *FirstEndLoc, *secondLoc, cmpTy);
509 Optional<DefinedOrUnknownSVal> OverlapTest =
510 Overlap.getAs<DefinedOrUnknownSVal>();
511 if (!OverlapTest)
512 return state;
513
514 std::tie(stateTrue, stateFalse) = state->assume(*OverlapTest);
515
516 if (stateTrue && !stateFalse) {
517 // Overlap!
518 emitOverlapBug(C, stateTrue, First, Second);
519 return nullptr;
520 }
521
522 // assume the two expressions don't overlap.
523 assert(stateFalse);
524 return stateFalse;
525 }
526
emitOverlapBug(CheckerContext & C,ProgramStateRef state,const Stmt * First,const Stmt * Second) const527 void CStringChecker::emitOverlapBug(CheckerContext &C, ProgramStateRef state,
528 const Stmt *First, const Stmt *Second) const {
529 ExplodedNode *N = C.generateErrorNode(state);
530 if (!N)
531 return;
532
533 if (!BT_Overlap)
534 BT_Overlap.reset(new BugType(Filter.CheckNameCStringBufferOverlap,
535 categories::UnixAPI, "Improper arguments"));
536
537 // Generate a report for this bug.
538 auto report = llvm::make_unique<BugReport>(
539 *BT_Overlap, "Arguments must not be overlapping buffers", N);
540 report->addRange(First->getSourceRange());
541 report->addRange(Second->getSourceRange());
542
543 C.emitReport(std::move(report));
544 }
545
emitNullArgBug(CheckerContext & C,ProgramStateRef State,const Stmt * S,StringRef WarningMsg) const546 void CStringChecker::emitNullArgBug(CheckerContext &C, ProgramStateRef State,
547 const Stmt *S, StringRef WarningMsg) const {
548 if (ExplodedNode *N = C.generateErrorNode(State)) {
549 if (!BT_Null)
550 BT_Null.reset(new BuiltinBug(
551 Filter.CheckNameCStringNullArg, categories::UnixAPI,
552 "Null pointer argument in call to byte string function"));
553
554 BuiltinBug *BT = static_cast<BuiltinBug *>(BT_Null.get());
555 auto Report = llvm::make_unique<BugReport>(*BT, WarningMsg, N);
556 Report->addRange(S->getSourceRange());
557 if (const auto *Ex = dyn_cast<Expr>(S))
558 bugreporter::trackExpressionValue(N, Ex, *Report);
559 C.emitReport(std::move(Report));
560 }
561 }
562
emitOutOfBoundsBug(CheckerContext & C,ProgramStateRef State,const Stmt * S,StringRef WarningMsg) const563 void CStringChecker::emitOutOfBoundsBug(CheckerContext &C,
564 ProgramStateRef State, const Stmt *S,
565 StringRef WarningMsg) const {
566 if (ExplodedNode *N = C.generateErrorNode(State)) {
567 if (!BT_Bounds)
568 BT_Bounds.reset(new BuiltinBug(
569 Filter.CheckCStringOutOfBounds ? Filter.CheckNameCStringOutOfBounds
570 : Filter.CheckNameCStringNullArg,
571 "Out-of-bound array access",
572 "Byte string function accesses out-of-bound array element"));
573
574 BuiltinBug *BT = static_cast<BuiltinBug *>(BT_Bounds.get());
575
576 // FIXME: It would be nice to eventually make this diagnostic more clear,
577 // e.g., by referencing the original declaration or by saying *why* this
578 // reference is outside the range.
579 auto Report = llvm::make_unique<BugReport>(*BT, WarningMsg, N);
580 Report->addRange(S->getSourceRange());
581 C.emitReport(std::move(Report));
582 }
583 }
584
emitNotCStringBug(CheckerContext & C,ProgramStateRef State,const Stmt * S,StringRef WarningMsg) const585 void CStringChecker::emitNotCStringBug(CheckerContext &C, ProgramStateRef State,
586 const Stmt *S,
587 StringRef WarningMsg) const {
588 if (ExplodedNode *N = C.generateNonFatalErrorNode(State)) {
589 if (!BT_NotCString)
590 BT_NotCString.reset(new BuiltinBug(
591 Filter.CheckNameCStringNotNullTerm, categories::UnixAPI,
592 "Argument is not a null-terminated string."));
593
594 auto Report = llvm::make_unique<BugReport>(*BT_NotCString, WarningMsg, N);
595
596 Report->addRange(S->getSourceRange());
597 C.emitReport(std::move(Report));
598 }
599 }
600
emitAdditionOverflowBug(CheckerContext & C,ProgramStateRef State) const601 void CStringChecker::emitAdditionOverflowBug(CheckerContext &C,
602 ProgramStateRef State) const {
603 if (ExplodedNode *N = C.generateErrorNode(State)) {
604 if (!BT_NotCString)
605 BT_NotCString.reset(
606 new BuiltinBug(Filter.CheckNameCStringOutOfBounds, "API",
607 "Sum of expressions causes overflow."));
608
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 *WarningMsg =
613 "This expression will create a string whose length is too big to "
614 "be represented as a size_t";
615
616 auto Report = llvm::make_unique<BugReport>(*BT_NotCString, WarningMsg, N);
617 C.emitReport(std::move(Report));
618 }
619 }
620
checkAdditionOverflow(CheckerContext & C,ProgramStateRef state,NonLoc left,NonLoc right) const621 ProgramStateRef CStringChecker::checkAdditionOverflow(CheckerContext &C,
622 ProgramStateRef state,
623 NonLoc left,
624 NonLoc right) const {
625 // If out-of-bounds checking is turned off, skip the rest.
626 if (!Filter.CheckCStringOutOfBounds)
627 return state;
628
629 // If a previous check has failed, propagate the failure.
630 if (!state)
631 return nullptr;
632
633 SValBuilder &svalBuilder = C.getSValBuilder();
634 BasicValueFactory &BVF = svalBuilder.getBasicValueFactory();
635
636 QualType sizeTy = svalBuilder.getContext().getSizeType();
637 const llvm::APSInt &maxValInt = BVF.getMaxValue(sizeTy);
638 NonLoc maxVal = svalBuilder.makeIntVal(maxValInt);
639
640 SVal maxMinusRight;
641 if (right.getAs<nonloc::ConcreteInt>()) {
642 maxMinusRight = svalBuilder.evalBinOpNN(state, BO_Sub, maxVal, right,
643 sizeTy);
644 } else {
645 // Try switching the operands. (The order of these two assignments is
646 // important!)
647 maxMinusRight = svalBuilder.evalBinOpNN(state, BO_Sub, maxVal, left,
648 sizeTy);
649 left = right;
650 }
651
652 if (Optional<NonLoc> maxMinusRightNL = maxMinusRight.getAs<NonLoc>()) {
653 QualType cmpTy = svalBuilder.getConditionType();
654 // If left > max - right, we have an overflow.
655 SVal willOverflow = svalBuilder.evalBinOpNN(state, BO_GT, left,
656 *maxMinusRightNL, cmpTy);
657
658 ProgramStateRef stateOverflow, stateOkay;
659 std::tie(stateOverflow, stateOkay) =
660 state->assume(willOverflow.castAs<DefinedOrUnknownSVal>());
661
662 if (stateOverflow && !stateOkay) {
663 // We have an overflow. Emit a bug report.
664 emitAdditionOverflowBug(C, stateOverflow);
665 return nullptr;
666 }
667
668 // From now on, assume an overflow didn't occur.
669 assert(stateOkay);
670 state = stateOkay;
671 }
672
673 return state;
674 }
675
setCStringLength(ProgramStateRef state,const MemRegion * MR,SVal strLength)676 ProgramStateRef CStringChecker::setCStringLength(ProgramStateRef state,
677 const MemRegion *MR,
678 SVal strLength) {
679 assert(!strLength.isUndef() && "Attempt to set an undefined string length");
680
681 MR = MR->StripCasts();
682
683 switch (MR->getKind()) {
684 case MemRegion::StringRegionKind:
685 // FIXME: This can happen if we strcpy() into a string region. This is
686 // undefined [C99 6.4.5p6], but we should still warn about it.
687 return state;
688
689 case MemRegion::SymbolicRegionKind:
690 case MemRegion::AllocaRegionKind:
691 case MemRegion::VarRegionKind:
692 case MemRegion::FieldRegionKind:
693 case MemRegion::ObjCIvarRegionKind:
694 // These are the types we can currently track string lengths for.
695 break;
696
697 case MemRegion::ElementRegionKind:
698 // FIXME: Handle element regions by upper-bounding the parent region's
699 // string length.
700 return state;
701
702 default:
703 // Other regions (mostly non-data) can't have a reliable C string length.
704 // For now, just ignore the change.
705 // FIXME: These are rare but not impossible. We should output some kind of
706 // warning for things like strcpy((char[]){'a', 0}, "b");
707 return state;
708 }
709
710 if (strLength.isUnknown())
711 return state->remove<CStringLength>(MR);
712
713 return state->set<CStringLength>(MR, strLength);
714 }
715
getCStringLengthForRegion(CheckerContext & C,ProgramStateRef & state,const Expr * Ex,const MemRegion * MR,bool hypothetical)716 SVal CStringChecker::getCStringLengthForRegion(CheckerContext &C,
717 ProgramStateRef &state,
718 const Expr *Ex,
719 const MemRegion *MR,
720 bool hypothetical) {
721 if (!hypothetical) {
722 // If there's a recorded length, go ahead and return it.
723 const SVal *Recorded = state->get<CStringLength>(MR);
724 if (Recorded)
725 return *Recorded;
726 }
727
728 // Otherwise, get a new symbol and update the state.
729 SValBuilder &svalBuilder = C.getSValBuilder();
730 QualType sizeTy = svalBuilder.getContext().getSizeType();
731 SVal strLength = svalBuilder.getMetadataSymbolVal(CStringChecker::getTag(),
732 MR, Ex, sizeTy,
733 C.getLocationContext(),
734 C.blockCount());
735
736 if (!hypothetical) {
737 if (Optional<NonLoc> strLn = strLength.getAs<NonLoc>()) {
738 // In case of unbounded calls strlen etc bound the range to SIZE_MAX/4
739 BasicValueFactory &BVF = svalBuilder.getBasicValueFactory();
740 const llvm::APSInt &maxValInt = BVF.getMaxValue(sizeTy);
741 llvm::APSInt fourInt = APSIntType(maxValInt).getValue(4);
742 const llvm::APSInt *maxLengthInt = BVF.evalAPSInt(BO_Div, maxValInt,
743 fourInt);
744 NonLoc maxLength = svalBuilder.makeIntVal(*maxLengthInt);
745 SVal evalLength = svalBuilder.evalBinOpNN(state, BO_LE, *strLn,
746 maxLength, sizeTy);
747 state = state->assume(evalLength.castAs<DefinedOrUnknownSVal>(), true);
748 }
749 state = state->set<CStringLength>(MR, strLength);
750 }
751
752 return strLength;
753 }
754
getCStringLength(CheckerContext & C,ProgramStateRef & state,const Expr * Ex,SVal Buf,bool hypothetical) const755 SVal CStringChecker::getCStringLength(CheckerContext &C, ProgramStateRef &state,
756 const Expr *Ex, SVal Buf,
757 bool hypothetical) const {
758 const MemRegion *MR = Buf.getAsRegion();
759 if (!MR) {
760 // If we can't get a region, see if it's something we /know/ isn't a
761 // C string. In the context of locations, the only time we can issue such
762 // a warning is for labels.
763 if (Optional<loc::GotoLabel> Label = Buf.getAs<loc::GotoLabel>()) {
764 if (Filter.CheckCStringNotNullTerm) {
765 SmallString<120> buf;
766 llvm::raw_svector_ostream os(buf);
767 assert(CurrentFunctionDescription);
768 os << "Argument to " << CurrentFunctionDescription
769 << " is the address of the label '" << Label->getLabel()->getName()
770 << "', which is not a null-terminated string";
771
772 emitNotCStringBug(C, state, Ex, os.str());
773 }
774 return UndefinedVal();
775 }
776
777 // If it's not a region and not a label, give up.
778 return UnknownVal();
779 }
780
781 // If we have a region, strip casts from it and see if we can figure out
782 // its length. For anything we can't figure out, just return UnknownVal.
783 MR = MR->StripCasts();
784
785 switch (MR->getKind()) {
786 case MemRegion::StringRegionKind: {
787 // Modifying the contents of string regions is undefined [C99 6.4.5p6],
788 // so we can assume that the byte length is the correct C string length.
789 SValBuilder &svalBuilder = C.getSValBuilder();
790 QualType sizeTy = svalBuilder.getContext().getSizeType();
791 const StringLiteral *strLit = cast<StringRegion>(MR)->getStringLiteral();
792 return svalBuilder.makeIntVal(strLit->getByteLength(), sizeTy);
793 }
794 case MemRegion::SymbolicRegionKind:
795 case MemRegion::AllocaRegionKind:
796 case MemRegion::VarRegionKind:
797 case MemRegion::FieldRegionKind:
798 case MemRegion::ObjCIvarRegionKind:
799 return getCStringLengthForRegion(C, state, Ex, MR, hypothetical);
800 case MemRegion::CompoundLiteralRegionKind:
801 // FIXME: Can we track this? Is it necessary?
802 return UnknownVal();
803 case MemRegion::ElementRegionKind:
804 // FIXME: How can we handle this? It's not good enough to subtract the
805 // offset from the base string length; consider "123\x00567" and &a[5].
806 return UnknownVal();
807 default:
808 // Other regions (mostly non-data) can't have a reliable C string length.
809 // In this case, an error is emitted and UndefinedVal is returned.
810 // The caller should always be prepared to handle this case.
811 if (Filter.CheckCStringNotNullTerm) {
812 SmallString<120> buf;
813 llvm::raw_svector_ostream os(buf);
814
815 assert(CurrentFunctionDescription);
816 os << "Argument to " << CurrentFunctionDescription << " is ";
817
818 if (SummarizeRegion(os, C.getASTContext(), MR))
819 os << ", which is not a null-terminated string";
820 else
821 os << "not a null-terminated string";
822
823 emitNotCStringBug(C, state, Ex, os.str());
824 }
825 return UndefinedVal();
826 }
827 }
828
getCStringLiteral(CheckerContext & C,ProgramStateRef & state,const Expr * expr,SVal val) const829 const StringLiteral *CStringChecker::getCStringLiteral(CheckerContext &C,
830 ProgramStateRef &state, const Expr *expr, SVal val) const {
831
832 // Get the memory region pointed to by the val.
833 const MemRegion *bufRegion = val.getAsRegion();
834 if (!bufRegion)
835 return nullptr;
836
837 // Strip casts off the memory region.
838 bufRegion = bufRegion->StripCasts();
839
840 // Cast the memory region to a string region.
841 const StringRegion *strRegion= dyn_cast<StringRegion>(bufRegion);
842 if (!strRegion)
843 return nullptr;
844
845 // Return the actual string in the string region.
846 return strRegion->getStringLiteral();
847 }
848
IsFirstBufInBound(CheckerContext & C,ProgramStateRef state,const Expr * FirstBuf,const Expr * Size)849 bool CStringChecker::IsFirstBufInBound(CheckerContext &C,
850 ProgramStateRef state,
851 const Expr *FirstBuf,
852 const Expr *Size) {
853 // If we do not know that the buffer is long enough we return 'true'.
854 // Otherwise the parent region of this field region would also get
855 // invalidated, which would lead to warnings based on an unknown state.
856
857 // Originally copied from CheckBufferAccess and CheckLocation.
858 SValBuilder &svalBuilder = C.getSValBuilder();
859 ASTContext &Ctx = svalBuilder.getContext();
860 const LocationContext *LCtx = C.getLocationContext();
861
862 QualType sizeTy = Size->getType();
863 QualType PtrTy = Ctx.getPointerType(Ctx.CharTy);
864 SVal BufVal = state->getSVal(FirstBuf, LCtx);
865
866 SVal LengthVal = state->getSVal(Size, LCtx);
867 Optional<NonLoc> Length = LengthVal.getAs<NonLoc>();
868 if (!Length)
869 return true; // cf top comment.
870
871 // Compute the offset of the last element to be accessed: size-1.
872 NonLoc One = svalBuilder.makeIntVal(1, sizeTy).castAs<NonLoc>();
873 SVal Offset = svalBuilder.evalBinOpNN(state, BO_Sub, *Length, One, sizeTy);
874 if (Offset.isUnknown())
875 return true; // cf top comment
876 NonLoc LastOffset = Offset.castAs<NonLoc>();
877
878 // Check that the first buffer is sufficiently long.
879 SVal BufStart = svalBuilder.evalCast(BufVal, PtrTy, FirstBuf->getType());
880 Optional<Loc> BufLoc = BufStart.getAs<Loc>();
881 if (!BufLoc)
882 return true; // cf top comment.
883
884 SVal BufEnd =
885 svalBuilder.evalBinOpLN(state, BO_Add, *BufLoc, LastOffset, PtrTy);
886
887 // Check for out of bound array element access.
888 const MemRegion *R = BufEnd.getAsRegion();
889 if (!R)
890 return true; // cf top comment.
891
892 const ElementRegion *ER = dyn_cast<ElementRegion>(R);
893 if (!ER)
894 return true; // cf top comment.
895
896 // FIXME: Does this crash when a non-standard definition
897 // of a library function is encountered?
898 assert(ER->getValueType() == C.getASTContext().CharTy &&
899 "IsFirstBufInBound should only be called with char* ElementRegions");
900
901 // Get the size of the array.
902 const SubRegion *superReg = cast<SubRegion>(ER->getSuperRegion());
903 SVal Extent =
904 svalBuilder.convertToArrayIndex(superReg->getExtent(svalBuilder));
905 DefinedOrUnknownSVal ExtentSize = Extent.castAs<DefinedOrUnknownSVal>();
906
907 // Get the index of the accessed element.
908 DefinedOrUnknownSVal Idx = ER->getIndex().castAs<DefinedOrUnknownSVal>();
909
910 ProgramStateRef StInBound = state->assumeInBound(Idx, ExtentSize, true);
911
912 return static_cast<bool>(StInBound);
913 }
914
InvalidateBuffer(CheckerContext & C,ProgramStateRef state,const Expr * E,SVal V,bool IsSourceBuffer,const Expr * Size)915 ProgramStateRef CStringChecker::InvalidateBuffer(CheckerContext &C,
916 ProgramStateRef state,
917 const Expr *E, SVal V,
918 bool IsSourceBuffer,
919 const Expr *Size) {
920 Optional<Loc> L = V.getAs<Loc>();
921 if (!L)
922 return state;
923
924 // FIXME: This is a simplified version of what's in CFRefCount.cpp -- it makes
925 // some assumptions about the value that CFRefCount can't. Even so, it should
926 // probably be refactored.
927 if (Optional<loc::MemRegionVal> MR = L->getAs<loc::MemRegionVal>()) {
928 const MemRegion *R = MR->getRegion()->StripCasts();
929
930 // Are we dealing with an ElementRegion? If so, we should be invalidating
931 // the super-region.
932 if (const ElementRegion *ER = dyn_cast<ElementRegion>(R)) {
933 R = ER->getSuperRegion();
934 // FIXME: What about layers of ElementRegions?
935 }
936
937 // Invalidate this region.
938 const LocationContext *LCtx = C.getPredecessor()->getLocationContext();
939
940 bool CausesPointerEscape = false;
941 RegionAndSymbolInvalidationTraits ITraits;
942 // Invalidate and escape only indirect regions accessible through the source
943 // buffer.
944 if (IsSourceBuffer) {
945 ITraits.setTrait(R->getBaseRegion(),
946 RegionAndSymbolInvalidationTraits::TK_PreserveContents);
947 ITraits.setTrait(R, RegionAndSymbolInvalidationTraits::TK_SuppressEscape);
948 CausesPointerEscape = true;
949 } else {
950 const MemRegion::Kind& K = R->getKind();
951 if (K == MemRegion::FieldRegionKind)
952 if (Size && IsFirstBufInBound(C, state, E, Size)) {
953 // If destination buffer is a field region and access is in bound,
954 // do not invalidate its super region.
955 ITraits.setTrait(
956 R,
957 RegionAndSymbolInvalidationTraits::TK_DoNotInvalidateSuperRegion);
958 }
959 }
960
961 return state->invalidateRegions(R, E, C.blockCount(), LCtx,
962 CausesPointerEscape, nullptr, nullptr,
963 &ITraits);
964 }
965
966 // If we have a non-region value by chance, just remove the binding.
967 // FIXME: is this necessary or correct? This handles the non-Region
968 // cases. Is it ever valid to store to these?
969 return state->killBinding(*L);
970 }
971
SummarizeRegion(raw_ostream & os,ASTContext & Ctx,const MemRegion * MR)972 bool CStringChecker::SummarizeRegion(raw_ostream &os, ASTContext &Ctx,
973 const MemRegion *MR) {
974 const TypedValueRegion *TVR = dyn_cast<TypedValueRegion>(MR);
975
976 switch (MR->getKind()) {
977 case MemRegion::FunctionCodeRegionKind: {
978 const NamedDecl *FD = cast<FunctionCodeRegion>(MR)->getDecl();
979 if (FD)
980 os << "the address of the function '" << *FD << '\'';
981 else
982 os << "the address of a function";
983 return true;
984 }
985 case MemRegion::BlockCodeRegionKind:
986 os << "block text";
987 return true;
988 case MemRegion::BlockDataRegionKind:
989 os << "a block";
990 return true;
991 case MemRegion::CXXThisRegionKind:
992 case MemRegion::CXXTempObjectRegionKind:
993 os << "a C++ temp object of type " << TVR->getValueType().getAsString();
994 return true;
995 case MemRegion::VarRegionKind:
996 os << "a variable of type" << TVR->getValueType().getAsString();
997 return true;
998 case MemRegion::FieldRegionKind:
999 os << "a field of type " << TVR->getValueType().getAsString();
1000 return true;
1001 case MemRegion::ObjCIvarRegionKind:
1002 os << "an instance variable of type " << TVR->getValueType().getAsString();
1003 return true;
1004 default:
1005 return false;
1006 }
1007 }
1008
memsetAux(const Expr * DstBuffer,SVal CharVal,const Expr * Size,CheckerContext & C,ProgramStateRef & State)1009 bool CStringChecker::memsetAux(const Expr *DstBuffer, SVal CharVal,
1010 const Expr *Size, CheckerContext &C,
1011 ProgramStateRef &State) {
1012 SVal MemVal = C.getSVal(DstBuffer);
1013 SVal SizeVal = C.getSVal(Size);
1014 const MemRegion *MR = MemVal.getAsRegion();
1015 if (!MR)
1016 return false;
1017
1018 // We're about to model memset by producing a "default binding" in the Store.
1019 // Our current implementation - RegionStore - doesn't support default bindings
1020 // that don't cover the whole base region. So we should first get the offset
1021 // and the base region to figure out whether the offset of buffer is 0.
1022 RegionOffset Offset = MR->getAsOffset();
1023 const MemRegion *BR = Offset.getRegion();
1024
1025 Optional<NonLoc> SizeNL = SizeVal.getAs<NonLoc>();
1026 if (!SizeNL)
1027 return false;
1028
1029 SValBuilder &svalBuilder = C.getSValBuilder();
1030 ASTContext &Ctx = C.getASTContext();
1031
1032 // void *memset(void *dest, int ch, size_t count);
1033 // For now we can only handle the case of offset is 0 and concrete char value.
1034 if (Offset.isValid() && !Offset.hasSymbolicOffset() &&
1035 Offset.getOffset() == 0) {
1036 // Get the base region's extent.
1037 auto *SubReg = cast<SubRegion>(BR);
1038 DefinedOrUnknownSVal Extent = SubReg->getExtent(svalBuilder);
1039
1040 ProgramStateRef StateWholeReg, StateNotWholeReg;
1041 std::tie(StateWholeReg, StateNotWholeReg) =
1042 State->assume(svalBuilder.evalEQ(State, Extent, *SizeNL));
1043
1044 // With the semantic of 'memset()', we should convert the CharVal to
1045 // unsigned char.
1046 CharVal = svalBuilder.evalCast(CharVal, Ctx.UnsignedCharTy, Ctx.IntTy);
1047
1048 ProgramStateRef StateNullChar, StateNonNullChar;
1049 std::tie(StateNullChar, StateNonNullChar) =
1050 assumeZero(C, State, CharVal, Ctx.UnsignedCharTy);
1051
1052 if (StateWholeReg && !StateNotWholeReg && StateNullChar &&
1053 !StateNonNullChar) {
1054 // If the 'memset()' acts on the whole region of destination buffer and
1055 // the value of the second argument of 'memset()' is zero, bind the second
1056 // argument's value to the destination buffer with 'default binding'.
1057 // FIXME: Since there is no perfect way to bind the non-zero character, we
1058 // can only deal with zero value here. In the future, we need to deal with
1059 // the binding of non-zero value in the case of whole region.
1060 State = State->bindDefaultZero(svalBuilder.makeLoc(BR),
1061 C.getLocationContext());
1062 } else {
1063 // If the destination buffer's extent is not equal to the value of
1064 // third argument, just invalidate buffer.
1065 State = InvalidateBuffer(C, State, DstBuffer, MemVal,
1066 /*IsSourceBuffer*/ false, Size);
1067 }
1068
1069 if (StateNullChar && !StateNonNullChar) {
1070 // If the value of the second argument of 'memset()' is zero, set the
1071 // string length of destination buffer to 0 directly.
1072 State = setCStringLength(State, MR,
1073 svalBuilder.makeZeroVal(Ctx.getSizeType()));
1074 } else if (!StateNullChar && StateNonNullChar) {
1075 SVal NewStrLen = svalBuilder.getMetadataSymbolVal(
1076 CStringChecker::getTag(), MR, DstBuffer, Ctx.getSizeType(),
1077 C.getLocationContext(), C.blockCount());
1078
1079 // If the value of second argument is not zero, then the string length
1080 // is at least the size argument.
1081 SVal NewStrLenGESize = svalBuilder.evalBinOp(
1082 State, BO_GE, NewStrLen, SizeVal, svalBuilder.getConditionType());
1083
1084 State = setCStringLength(
1085 State->assume(NewStrLenGESize.castAs<DefinedOrUnknownSVal>(), true),
1086 MR, NewStrLen);
1087 }
1088 } else {
1089 // If the offset is not zero and char value is not concrete, we can do
1090 // nothing but invalidate the buffer.
1091 State = InvalidateBuffer(C, State, DstBuffer, MemVal,
1092 /*IsSourceBuffer*/ false, Size);
1093 }
1094 return true;
1095 }
1096
1097 //===----------------------------------------------------------------------===//
1098 // evaluation of individual function calls.
1099 //===----------------------------------------------------------------------===//
1100
evalCopyCommon(CheckerContext & C,const CallExpr * CE,ProgramStateRef state,const Expr * Size,const Expr * Dest,const Expr * Source,bool Restricted,bool IsMempcpy) const1101 void CStringChecker::evalCopyCommon(CheckerContext &C,
1102 const CallExpr *CE,
1103 ProgramStateRef state,
1104 const Expr *Size, const Expr *Dest,
1105 const Expr *Source, bool Restricted,
1106 bool IsMempcpy) const {
1107 CurrentFunctionDescription = "memory copy function";
1108
1109 // See if the size argument is zero.
1110 const LocationContext *LCtx = C.getLocationContext();
1111 SVal sizeVal = state->getSVal(Size, LCtx);
1112 QualType sizeTy = Size->getType();
1113
1114 ProgramStateRef stateZeroSize, stateNonZeroSize;
1115 std::tie(stateZeroSize, stateNonZeroSize) =
1116 assumeZero(C, state, sizeVal, sizeTy);
1117
1118 // Get the value of the Dest.
1119 SVal destVal = state->getSVal(Dest, LCtx);
1120
1121 // If the size is zero, there won't be any actual memory access, so
1122 // just bind the return value to the destination buffer and return.
1123 if (stateZeroSize && !stateNonZeroSize) {
1124 stateZeroSize = stateZeroSize->BindExpr(CE, LCtx, destVal);
1125 C.addTransition(stateZeroSize);
1126 return;
1127 }
1128
1129 // If the size can be nonzero, we have to check the other arguments.
1130 if (stateNonZeroSize) {
1131 state = stateNonZeroSize;
1132
1133 // Ensure the destination is not null. If it is NULL there will be a
1134 // NULL pointer dereference.
1135 state = checkNonNull(C, state, Dest, destVal);
1136 if (!state)
1137 return;
1138
1139 // Get the value of the Src.
1140 SVal srcVal = state->getSVal(Source, LCtx);
1141
1142 // Ensure the source is not null. If it is NULL there will be a
1143 // NULL pointer dereference.
1144 state = checkNonNull(C, state, Source, srcVal);
1145 if (!state)
1146 return;
1147
1148 // Ensure the accesses are valid and that the buffers do not overlap.
1149 const char * const writeWarning =
1150 "Memory copy function overflows destination buffer";
1151 state = CheckBufferAccess(C, state, Size, Dest, Source,
1152 writeWarning, /* sourceWarning = */ nullptr);
1153 if (Restricted)
1154 state = CheckOverlap(C, state, Size, Dest, Source);
1155
1156 if (!state)
1157 return;
1158
1159 // If this is mempcpy, get the byte after the last byte copied and
1160 // bind the expr.
1161 if (IsMempcpy) {
1162 // Get the byte after the last byte copied.
1163 SValBuilder &SvalBuilder = C.getSValBuilder();
1164 ASTContext &Ctx = SvalBuilder.getContext();
1165 QualType CharPtrTy = Ctx.getPointerType(Ctx.CharTy);
1166 SVal DestRegCharVal =
1167 SvalBuilder.evalCast(destVal, CharPtrTy, Dest->getType());
1168 SVal lastElement = C.getSValBuilder().evalBinOp(
1169 state, BO_Add, DestRegCharVal, sizeVal, Dest->getType());
1170 // If we don't know how much we copied, we can at least
1171 // conjure a return value for later.
1172 if (lastElement.isUnknown())
1173 lastElement = C.getSValBuilder().conjureSymbolVal(nullptr, CE, LCtx,
1174 C.blockCount());
1175
1176 // The byte after the last byte copied is the return value.
1177 state = state->BindExpr(CE, LCtx, lastElement);
1178 } else {
1179 // All other copies return the destination buffer.
1180 // (Well, bcopy() has a void return type, but this won't hurt.)
1181 state = state->BindExpr(CE, LCtx, destVal);
1182 }
1183
1184 // Invalidate the destination (regular invalidation without pointer-escaping
1185 // the address of the top-level region).
1186 // FIXME: Even if we can't perfectly model the copy, we should see if we
1187 // can use LazyCompoundVals to copy the source values into the destination.
1188 // This would probably remove any existing bindings past the end of the
1189 // copied region, but that's still an improvement over blank invalidation.
1190 state = InvalidateBuffer(C, state, Dest, C.getSVal(Dest),
1191 /*IsSourceBuffer*/false, Size);
1192
1193 // Invalidate the source (const-invalidation without const-pointer-escaping
1194 // the address of the top-level region).
1195 state = InvalidateBuffer(C, state, Source, C.getSVal(Source),
1196 /*IsSourceBuffer*/true, nullptr);
1197
1198 C.addTransition(state);
1199 }
1200 }
1201
1202
evalMemcpy(CheckerContext & C,const CallExpr * CE) const1203 void CStringChecker::evalMemcpy(CheckerContext &C, const CallExpr *CE) const {
1204 if (CE->getNumArgs() < 3)
1205 return;
1206
1207 // void *memcpy(void *restrict dst, const void *restrict src, size_t n);
1208 // The return value is the address of the destination buffer.
1209 const Expr *Dest = CE->getArg(0);
1210 ProgramStateRef state = C.getState();
1211
1212 evalCopyCommon(C, CE, state, CE->getArg(2), Dest, CE->getArg(1), true);
1213 }
1214
evalMempcpy(CheckerContext & C,const CallExpr * CE) const1215 void CStringChecker::evalMempcpy(CheckerContext &C, const CallExpr *CE) const {
1216 if (CE->getNumArgs() < 3)
1217 return;
1218
1219 // void *mempcpy(void *restrict dst, const void *restrict src, size_t n);
1220 // The return value is a pointer to the byte following the last written byte.
1221 const Expr *Dest = CE->getArg(0);
1222 ProgramStateRef state = C.getState();
1223
1224 evalCopyCommon(C, CE, state, CE->getArg(2), Dest, CE->getArg(1), true, true);
1225 }
1226
evalMemmove(CheckerContext & C,const CallExpr * CE) const1227 void CStringChecker::evalMemmove(CheckerContext &C, const CallExpr *CE) const {
1228 if (CE->getNumArgs() < 3)
1229 return;
1230
1231 // void *memmove(void *dst, const void *src, size_t n);
1232 // The return value is the address of the destination buffer.
1233 const Expr *Dest = CE->getArg(0);
1234 ProgramStateRef state = C.getState();
1235
1236 evalCopyCommon(C, CE, state, CE->getArg(2), Dest, CE->getArg(1));
1237 }
1238
evalBcopy(CheckerContext & C,const CallExpr * CE) const1239 void CStringChecker::evalBcopy(CheckerContext &C, const CallExpr *CE) const {
1240 if (CE->getNumArgs() < 3)
1241 return;
1242
1243 // void bcopy(const void *src, void *dst, size_t n);
1244 evalCopyCommon(C, CE, C.getState(),
1245 CE->getArg(2), CE->getArg(1), CE->getArg(0));
1246 }
1247
evalMemcmp(CheckerContext & C,const CallExpr * CE) const1248 void CStringChecker::evalMemcmp(CheckerContext &C, const CallExpr *CE) const {
1249 if (CE->getNumArgs() < 3)
1250 return;
1251
1252 // int memcmp(const void *s1, const void *s2, size_t n);
1253 CurrentFunctionDescription = "memory comparison function";
1254
1255 const Expr *Left = CE->getArg(0);
1256 const Expr *Right = CE->getArg(1);
1257 const Expr *Size = CE->getArg(2);
1258
1259 ProgramStateRef state = C.getState();
1260 SValBuilder &svalBuilder = C.getSValBuilder();
1261
1262 // See if the size argument is zero.
1263 const LocationContext *LCtx = C.getLocationContext();
1264 SVal sizeVal = state->getSVal(Size, LCtx);
1265 QualType sizeTy = Size->getType();
1266
1267 ProgramStateRef stateZeroSize, stateNonZeroSize;
1268 std::tie(stateZeroSize, stateNonZeroSize) =
1269 assumeZero(C, state, sizeVal, sizeTy);
1270
1271 // If the size can be zero, the result will be 0 in that case, and we don't
1272 // have to check either of the buffers.
1273 if (stateZeroSize) {
1274 state = stateZeroSize;
1275 state = state->BindExpr(CE, LCtx,
1276 svalBuilder.makeZeroVal(CE->getType()));
1277 C.addTransition(state);
1278 }
1279
1280 // If the size can be nonzero, we have to check the other arguments.
1281 if (stateNonZeroSize) {
1282 state = stateNonZeroSize;
1283 // If we know the two buffers are the same, we know the result is 0.
1284 // First, get the two buffers' addresses. Another checker will have already
1285 // made sure they're not undefined.
1286 DefinedOrUnknownSVal LV =
1287 state->getSVal(Left, LCtx).castAs<DefinedOrUnknownSVal>();
1288 DefinedOrUnknownSVal RV =
1289 state->getSVal(Right, LCtx).castAs<DefinedOrUnknownSVal>();
1290
1291 // See if they are the same.
1292 DefinedOrUnknownSVal SameBuf = svalBuilder.evalEQ(state, LV, RV);
1293 ProgramStateRef StSameBuf, StNotSameBuf;
1294 std::tie(StSameBuf, StNotSameBuf) = state->assume(SameBuf);
1295
1296 // If the two arguments might be the same buffer, we know the result is 0,
1297 // and we only need to check one size.
1298 if (StSameBuf) {
1299 state = StSameBuf;
1300 state = CheckBufferAccess(C, state, Size, Left);
1301 if (state) {
1302 state = StSameBuf->BindExpr(CE, LCtx,
1303 svalBuilder.makeZeroVal(CE->getType()));
1304 C.addTransition(state);
1305 }
1306 }
1307
1308 // If the two arguments might be different buffers, we have to check the
1309 // size of both of them.
1310 if (StNotSameBuf) {
1311 state = StNotSameBuf;
1312 state = CheckBufferAccess(C, state, Size, Left, Right);
1313 if (state) {
1314 // The return value is the comparison result, which we don't know.
1315 SVal CmpV = svalBuilder.conjureSymbolVal(nullptr, CE, LCtx,
1316 C.blockCount());
1317 state = state->BindExpr(CE, LCtx, CmpV);
1318 C.addTransition(state);
1319 }
1320 }
1321 }
1322 }
1323
evalstrLength(CheckerContext & C,const CallExpr * CE) const1324 void CStringChecker::evalstrLength(CheckerContext &C,
1325 const CallExpr *CE) const {
1326 if (CE->getNumArgs() < 1)
1327 return;
1328
1329 // size_t strlen(const char *s);
1330 evalstrLengthCommon(C, CE, /* IsStrnlen = */ false);
1331 }
1332
evalstrnLength(CheckerContext & C,const CallExpr * CE) const1333 void CStringChecker::evalstrnLength(CheckerContext &C,
1334 const CallExpr *CE) const {
1335 if (CE->getNumArgs() < 2)
1336 return;
1337
1338 // size_t strnlen(const char *s, size_t maxlen);
1339 evalstrLengthCommon(C, CE, /* IsStrnlen = */ true);
1340 }
1341
evalstrLengthCommon(CheckerContext & C,const CallExpr * CE,bool IsStrnlen) const1342 void CStringChecker::evalstrLengthCommon(CheckerContext &C, const CallExpr *CE,
1343 bool IsStrnlen) const {
1344 CurrentFunctionDescription = "string length function";
1345 ProgramStateRef state = C.getState();
1346 const LocationContext *LCtx = C.getLocationContext();
1347
1348 if (IsStrnlen) {
1349 const Expr *maxlenExpr = CE->getArg(1);
1350 SVal maxlenVal = state->getSVal(maxlenExpr, LCtx);
1351
1352 ProgramStateRef stateZeroSize, stateNonZeroSize;
1353 std::tie(stateZeroSize, stateNonZeroSize) =
1354 assumeZero(C, state, maxlenVal, maxlenExpr->getType());
1355
1356 // If the size can be zero, the result will be 0 in that case, and we don't
1357 // have to check the string itself.
1358 if (stateZeroSize) {
1359 SVal zero = C.getSValBuilder().makeZeroVal(CE->getType());
1360 stateZeroSize = stateZeroSize->BindExpr(CE, LCtx, zero);
1361 C.addTransition(stateZeroSize);
1362 }
1363
1364 // If the size is GUARANTEED to be zero, we're done!
1365 if (!stateNonZeroSize)
1366 return;
1367
1368 // Otherwise, record the assumption that the size is nonzero.
1369 state = stateNonZeroSize;
1370 }
1371
1372 // Check that the string argument is non-null.
1373 const Expr *Arg = CE->getArg(0);
1374 SVal ArgVal = state->getSVal(Arg, LCtx);
1375
1376 state = checkNonNull(C, state, Arg, ArgVal);
1377
1378 if (!state)
1379 return;
1380
1381 SVal strLength = getCStringLength(C, state, Arg, ArgVal);
1382
1383 // If the argument isn't a valid C string, there's no valid state to
1384 // transition to.
1385 if (strLength.isUndef())
1386 return;
1387
1388 DefinedOrUnknownSVal result = UnknownVal();
1389
1390 // If the check is for strnlen() then bind the return value to no more than
1391 // the maxlen value.
1392 if (IsStrnlen) {
1393 QualType cmpTy = C.getSValBuilder().getConditionType();
1394
1395 // It's a little unfortunate to be getting this again,
1396 // but it's not that expensive...
1397 const Expr *maxlenExpr = CE->getArg(1);
1398 SVal maxlenVal = state->getSVal(maxlenExpr, LCtx);
1399
1400 Optional<NonLoc> strLengthNL = strLength.getAs<NonLoc>();
1401 Optional<NonLoc> maxlenValNL = maxlenVal.getAs<NonLoc>();
1402
1403 if (strLengthNL && maxlenValNL) {
1404 ProgramStateRef stateStringTooLong, stateStringNotTooLong;
1405
1406 // Check if the strLength is greater than the maxlen.
1407 std::tie(stateStringTooLong, stateStringNotTooLong) = state->assume(
1408 C.getSValBuilder()
1409 .evalBinOpNN(state, BO_GT, *strLengthNL, *maxlenValNL, cmpTy)
1410 .castAs<DefinedOrUnknownSVal>());
1411
1412 if (stateStringTooLong && !stateStringNotTooLong) {
1413 // If the string is longer than maxlen, return maxlen.
1414 result = *maxlenValNL;
1415 } else if (stateStringNotTooLong && !stateStringTooLong) {
1416 // If the string is shorter than maxlen, return its length.
1417 result = *strLengthNL;
1418 }
1419 }
1420
1421 if (result.isUnknown()) {
1422 // If we don't have enough information for a comparison, there's
1423 // no guarantee the full string length will actually be returned.
1424 // All we know is the return value is the min of the string length
1425 // and the limit. This is better than nothing.
1426 result = C.getSValBuilder().conjureSymbolVal(nullptr, CE, LCtx,
1427 C.blockCount());
1428 NonLoc resultNL = result.castAs<NonLoc>();
1429
1430 if (strLengthNL) {
1431 state = state->assume(C.getSValBuilder().evalBinOpNN(
1432 state, BO_LE, resultNL, *strLengthNL, cmpTy)
1433 .castAs<DefinedOrUnknownSVal>(), true);
1434 }
1435
1436 if (maxlenValNL) {
1437 state = state->assume(C.getSValBuilder().evalBinOpNN(
1438 state, BO_LE, resultNL, *maxlenValNL, cmpTy)
1439 .castAs<DefinedOrUnknownSVal>(), true);
1440 }
1441 }
1442
1443 } else {
1444 // This is a plain strlen(), not strnlen().
1445 result = strLength.castAs<DefinedOrUnknownSVal>();
1446
1447 // If we don't know the length of the string, conjure a return
1448 // value, so it can be used in constraints, at least.
1449 if (result.isUnknown()) {
1450 result = C.getSValBuilder().conjureSymbolVal(nullptr, CE, LCtx,
1451 C.blockCount());
1452 }
1453 }
1454
1455 // Bind the return value.
1456 assert(!result.isUnknown() && "Should have conjured a value by now");
1457 state = state->BindExpr(CE, LCtx, result);
1458 C.addTransition(state);
1459 }
1460
evalStrcpy(CheckerContext & C,const CallExpr * CE) const1461 void CStringChecker::evalStrcpy(CheckerContext &C, const CallExpr *CE) const {
1462 if (CE->getNumArgs() < 2)
1463 return;
1464
1465 // char *strcpy(char *restrict dst, const char *restrict src);
1466 evalStrcpyCommon(C, CE,
1467 /* returnEnd = */ false,
1468 /* isBounded = */ false,
1469 /* isAppending = */ false);
1470 }
1471
evalStrncpy(CheckerContext & C,const CallExpr * CE) const1472 void CStringChecker::evalStrncpy(CheckerContext &C, const CallExpr *CE) const {
1473 if (CE->getNumArgs() < 3)
1474 return;
1475
1476 // char *strncpy(char *restrict dst, const char *restrict src, size_t n);
1477 evalStrcpyCommon(C, CE,
1478 /* returnEnd = */ false,
1479 /* isBounded = */ true,
1480 /* isAppending = */ false);
1481 }
1482
evalStpcpy(CheckerContext & C,const CallExpr * CE) const1483 void CStringChecker::evalStpcpy(CheckerContext &C, const CallExpr *CE) const {
1484 if (CE->getNumArgs() < 2)
1485 return;
1486
1487 // char *stpcpy(char *restrict dst, const char *restrict src);
1488 evalStrcpyCommon(C, CE,
1489 /* returnEnd = */ true,
1490 /* isBounded = */ false,
1491 /* isAppending = */ false);
1492 }
1493
evalStrlcpy(CheckerContext & C,const CallExpr * CE) const1494 void CStringChecker::evalStrlcpy(CheckerContext &C, const CallExpr *CE) const {
1495 if (CE->getNumArgs() < 3)
1496 return;
1497
1498 // char *strlcpy(char *dst, const char *src, size_t n);
1499 evalStrcpyCommon(C, CE,
1500 /* returnEnd = */ true,
1501 /* isBounded = */ true,
1502 /* isAppending = */ false,
1503 /* returnPtr = */ false);
1504 }
1505
evalStrcat(CheckerContext & C,const CallExpr * CE) const1506 void CStringChecker::evalStrcat(CheckerContext &C, const CallExpr *CE) const {
1507 if (CE->getNumArgs() < 2)
1508 return;
1509
1510 //char *strcat(char *restrict s1, const char *restrict s2);
1511 evalStrcpyCommon(C, CE,
1512 /* returnEnd = */ false,
1513 /* isBounded = */ false,
1514 /* isAppending = */ true);
1515 }
1516
evalStrncat(CheckerContext & C,const CallExpr * CE) const1517 void CStringChecker::evalStrncat(CheckerContext &C, const CallExpr *CE) const {
1518 if (CE->getNumArgs() < 3)
1519 return;
1520
1521 //char *strncat(char *restrict s1, const char *restrict s2, size_t n);
1522 evalStrcpyCommon(C, CE,
1523 /* returnEnd = */ false,
1524 /* isBounded = */ true,
1525 /* isAppending = */ true);
1526 }
1527
evalStrlcat(CheckerContext & C,const CallExpr * CE) const1528 void CStringChecker::evalStrlcat(CheckerContext &C, const CallExpr *CE) const {
1529 if (CE->getNumArgs() < 3)
1530 return;
1531
1532 //char *strlcat(char *s1, const char *s2, size_t n);
1533 evalStrcpyCommon(C, CE,
1534 /* returnEnd = */ false,
1535 /* isBounded = */ true,
1536 /* isAppending = */ true,
1537 /* returnPtr = */ false);
1538 }
1539
evalStrcpyCommon(CheckerContext & C,const CallExpr * CE,bool returnEnd,bool isBounded,bool isAppending,bool returnPtr) const1540 void CStringChecker::evalStrcpyCommon(CheckerContext &C, const CallExpr *CE,
1541 bool returnEnd, bool isBounded,
1542 bool isAppending, bool returnPtr) const {
1543 CurrentFunctionDescription = "string copy function";
1544 ProgramStateRef state = C.getState();
1545 const LocationContext *LCtx = C.getLocationContext();
1546
1547 // Check that the destination is non-null.
1548 const Expr *Dst = CE->getArg(0);
1549 SVal DstVal = state->getSVal(Dst, LCtx);
1550
1551 state = checkNonNull(C, state, Dst, DstVal);
1552 if (!state)
1553 return;
1554
1555 // Check that the source is non-null.
1556 const Expr *srcExpr = CE->getArg(1);
1557 SVal srcVal = state->getSVal(srcExpr, LCtx);
1558 state = checkNonNull(C, state, srcExpr, srcVal);
1559 if (!state)
1560 return;
1561
1562 // Get the string length of the source.
1563 SVal strLength = getCStringLength(C, state, srcExpr, srcVal);
1564
1565 // If the source isn't a valid C string, give up.
1566 if (strLength.isUndef())
1567 return;
1568
1569 SValBuilder &svalBuilder = C.getSValBuilder();
1570 QualType cmpTy = svalBuilder.getConditionType();
1571 QualType sizeTy = svalBuilder.getContext().getSizeType();
1572
1573 // These two values allow checking two kinds of errors:
1574 // - actual overflows caused by a source that doesn't fit in the destination
1575 // - potential overflows caused by a bound that could exceed the destination
1576 SVal amountCopied = UnknownVal();
1577 SVal maxLastElementIndex = UnknownVal();
1578 const char *boundWarning = nullptr;
1579
1580 state = CheckOverlap(C, state, isBounded ? CE->getArg(2) : CE->getArg(1), Dst, srcExpr);
1581
1582 if (!state)
1583 return;
1584
1585 // If the function is strncpy, strncat, etc... it is bounded.
1586 if (isBounded) {
1587 // Get the max number of characters to copy.
1588 const Expr *lenExpr = CE->getArg(2);
1589 SVal lenVal = state->getSVal(lenExpr, LCtx);
1590
1591 // Protect against misdeclared strncpy().
1592 lenVal = svalBuilder.evalCast(lenVal, sizeTy, lenExpr->getType());
1593
1594 Optional<NonLoc> strLengthNL = strLength.getAs<NonLoc>();
1595 Optional<NonLoc> lenValNL = lenVal.getAs<NonLoc>();
1596
1597 // If we know both values, we might be able to figure out how much
1598 // we're copying.
1599 if (strLengthNL && lenValNL) {
1600 ProgramStateRef stateSourceTooLong, stateSourceNotTooLong;
1601
1602 // Check if the max number to copy is less than the length of the src.
1603 // If the bound is equal to the source length, strncpy won't null-
1604 // terminate the result!
1605 std::tie(stateSourceTooLong, stateSourceNotTooLong) = state->assume(
1606 svalBuilder.evalBinOpNN(state, BO_GE, *strLengthNL, *lenValNL, cmpTy)
1607 .castAs<DefinedOrUnknownSVal>());
1608
1609 if (stateSourceTooLong && !stateSourceNotTooLong) {
1610 // Max number to copy is less than the length of the src, so the actual
1611 // strLength copied is the max number arg.
1612 state = stateSourceTooLong;
1613 amountCopied = lenVal;
1614
1615 } else if (!stateSourceTooLong && stateSourceNotTooLong) {
1616 // The source buffer entirely fits in the bound.
1617 state = stateSourceNotTooLong;
1618 amountCopied = strLength;
1619 }
1620 }
1621
1622 // We still want to know if the bound is known to be too large.
1623 if (lenValNL) {
1624 if (isAppending) {
1625 // For strncat, the check is strlen(dst) + lenVal < sizeof(dst)
1626
1627 // Get the string length of the destination. If the destination is
1628 // memory that can't have a string length, we shouldn't be copying
1629 // into it anyway.
1630 SVal dstStrLength = getCStringLength(C, state, Dst, DstVal);
1631 if (dstStrLength.isUndef())
1632 return;
1633
1634 if (Optional<NonLoc> dstStrLengthNL = dstStrLength.getAs<NonLoc>()) {
1635 maxLastElementIndex = svalBuilder.evalBinOpNN(state, BO_Add,
1636 *lenValNL,
1637 *dstStrLengthNL,
1638 sizeTy);
1639 boundWarning = "Size argument is greater than the free space in the "
1640 "destination buffer";
1641 }
1642
1643 } else {
1644 // For strncpy, this is just checking that lenVal <= sizeof(dst)
1645 // (Yes, strncpy and strncat differ in how they treat termination.
1646 // strncat ALWAYS terminates, but strncpy doesn't.)
1647
1648 // We need a special case for when the copy size is zero, in which
1649 // case strncpy will do no work at all. Our bounds check uses n-1
1650 // as the last element accessed, so n == 0 is problematic.
1651 ProgramStateRef StateZeroSize, StateNonZeroSize;
1652 std::tie(StateZeroSize, StateNonZeroSize) =
1653 assumeZero(C, state, *lenValNL, sizeTy);
1654
1655 // If the size is known to be zero, we're done.
1656 if (StateZeroSize && !StateNonZeroSize) {
1657 if (returnPtr) {
1658 StateZeroSize = StateZeroSize->BindExpr(CE, LCtx, DstVal);
1659 } else {
1660 StateZeroSize = StateZeroSize->BindExpr(CE, LCtx, *lenValNL);
1661 }
1662 C.addTransition(StateZeroSize);
1663 return;
1664 }
1665
1666 // Otherwise, go ahead and figure out the last element we'll touch.
1667 // We don't record the non-zero assumption here because we can't
1668 // be sure. We won't warn on a possible zero.
1669 NonLoc one = svalBuilder.makeIntVal(1, sizeTy).castAs<NonLoc>();
1670 maxLastElementIndex = svalBuilder.evalBinOpNN(state, BO_Sub, *lenValNL,
1671 one, sizeTy);
1672 boundWarning = "Size argument is greater than the length of the "
1673 "destination buffer";
1674 }
1675 }
1676
1677 // If we couldn't pin down the copy length, at least bound it.
1678 // FIXME: We should actually run this code path for append as well, but
1679 // right now it creates problems with constraints (since we can end up
1680 // trying to pass constraints from symbol to symbol).
1681 if (amountCopied.isUnknown() && !isAppending) {
1682 // Try to get a "hypothetical" string length symbol, which we can later
1683 // set as a real value if that turns out to be the case.
1684 amountCopied = getCStringLength(C, state, lenExpr, srcVal, true);
1685 assert(!amountCopied.isUndef());
1686
1687 if (Optional<NonLoc> amountCopiedNL = amountCopied.getAs<NonLoc>()) {
1688 if (lenValNL) {
1689 // amountCopied <= lenVal
1690 SVal copiedLessThanBound = svalBuilder.evalBinOpNN(state, BO_LE,
1691 *amountCopiedNL,
1692 *lenValNL,
1693 cmpTy);
1694 state = state->assume(
1695 copiedLessThanBound.castAs<DefinedOrUnknownSVal>(), true);
1696 if (!state)
1697 return;
1698 }
1699
1700 if (strLengthNL) {
1701 // amountCopied <= strlen(source)
1702 SVal copiedLessThanSrc = svalBuilder.evalBinOpNN(state, BO_LE,
1703 *amountCopiedNL,
1704 *strLengthNL,
1705 cmpTy);
1706 state = state->assume(
1707 copiedLessThanSrc.castAs<DefinedOrUnknownSVal>(), true);
1708 if (!state)
1709 return;
1710 }
1711 }
1712 }
1713
1714 } else {
1715 // The function isn't bounded. The amount copied should match the length
1716 // of the source buffer.
1717 amountCopied = strLength;
1718 }
1719
1720 assert(state);
1721
1722 // This represents the number of characters copied into the destination
1723 // buffer. (It may not actually be the strlen if the destination buffer
1724 // is not terminated.)
1725 SVal finalStrLength = UnknownVal();
1726
1727 // If this is an appending function (strcat, strncat...) then set the
1728 // string length to strlen(src) + strlen(dst) since the buffer will
1729 // ultimately contain both.
1730 if (isAppending) {
1731 // Get the string length of the destination. If the destination is memory
1732 // that can't have a string length, we shouldn't be copying into it anyway.
1733 SVal dstStrLength = getCStringLength(C, state, Dst, DstVal);
1734 if (dstStrLength.isUndef())
1735 return;
1736
1737 Optional<NonLoc> srcStrLengthNL = amountCopied.getAs<NonLoc>();
1738 Optional<NonLoc> dstStrLengthNL = dstStrLength.getAs<NonLoc>();
1739
1740 // If we know both string lengths, we might know the final string length.
1741 if (srcStrLengthNL && dstStrLengthNL) {
1742 // Make sure the two lengths together don't overflow a size_t.
1743 state = checkAdditionOverflow(C, state, *srcStrLengthNL, *dstStrLengthNL);
1744 if (!state)
1745 return;
1746
1747 finalStrLength = svalBuilder.evalBinOpNN(state, BO_Add, *srcStrLengthNL,
1748 *dstStrLengthNL, sizeTy);
1749 }
1750
1751 // If we couldn't get a single value for the final string length,
1752 // we can at least bound it by the individual lengths.
1753 if (finalStrLength.isUnknown()) {
1754 // Try to get a "hypothetical" string length symbol, which we can later
1755 // set as a real value if that turns out to be the case.
1756 finalStrLength = getCStringLength(C, state, CE, DstVal, true);
1757 assert(!finalStrLength.isUndef());
1758
1759 if (Optional<NonLoc> finalStrLengthNL = finalStrLength.getAs<NonLoc>()) {
1760 if (srcStrLengthNL) {
1761 // finalStrLength >= srcStrLength
1762 SVal sourceInResult = svalBuilder.evalBinOpNN(state, BO_GE,
1763 *finalStrLengthNL,
1764 *srcStrLengthNL,
1765 cmpTy);
1766 state = state->assume(sourceInResult.castAs<DefinedOrUnknownSVal>(),
1767 true);
1768 if (!state)
1769 return;
1770 }
1771
1772 if (dstStrLengthNL) {
1773 // finalStrLength >= dstStrLength
1774 SVal destInResult = svalBuilder.evalBinOpNN(state, BO_GE,
1775 *finalStrLengthNL,
1776 *dstStrLengthNL,
1777 cmpTy);
1778 state =
1779 state->assume(destInResult.castAs<DefinedOrUnknownSVal>(), true);
1780 if (!state)
1781 return;
1782 }
1783 }
1784 }
1785
1786 } else {
1787 // Otherwise, this is a copy-over function (strcpy, strncpy, ...), and
1788 // the final string length will match the input string length.
1789 finalStrLength = amountCopied;
1790 }
1791
1792 SVal Result;
1793
1794 if (returnPtr) {
1795 // The final result of the function will either be a pointer past the last
1796 // copied element, or a pointer to the start of the destination buffer.
1797 Result = (returnEnd ? UnknownVal() : DstVal);
1798 } else {
1799 Result = finalStrLength;
1800 }
1801
1802 assert(state);
1803
1804 // If the destination is a MemRegion, try to check for a buffer overflow and
1805 // record the new string length.
1806 if (Optional<loc::MemRegionVal> dstRegVal =
1807 DstVal.getAs<loc::MemRegionVal>()) {
1808 QualType ptrTy = Dst->getType();
1809
1810 // If we have an exact value on a bounded copy, use that to check for
1811 // overflows, rather than our estimate about how much is actually copied.
1812 if (boundWarning) {
1813 if (Optional<NonLoc> maxLastNL = maxLastElementIndex.getAs<NonLoc>()) {
1814 SVal maxLastElement = svalBuilder.evalBinOpLN(state, BO_Add, *dstRegVal,
1815 *maxLastNL, ptrTy);
1816 state = CheckLocation(C, state, CE->getArg(2), maxLastElement,
1817 boundWarning);
1818 if (!state)
1819 return;
1820 }
1821 }
1822
1823 // Then, if the final length is known...
1824 if (Optional<NonLoc> knownStrLength = finalStrLength.getAs<NonLoc>()) {
1825 SVal lastElement = svalBuilder.evalBinOpLN(state, BO_Add, *dstRegVal,
1826 *knownStrLength, ptrTy);
1827
1828 // ...and we haven't checked the bound, we'll check the actual copy.
1829 if (!boundWarning) {
1830 const char * const warningMsg =
1831 "String copy function overflows destination buffer";
1832 state = CheckLocation(C, state, Dst, lastElement, warningMsg);
1833 if (!state)
1834 return;
1835 }
1836
1837 // If this is a stpcpy-style copy, the last element is the return value.
1838 if (returnPtr && returnEnd)
1839 Result = lastElement;
1840 }
1841
1842 // Invalidate the destination (regular invalidation without pointer-escaping
1843 // the address of the top-level region). This must happen before we set the
1844 // C string length because invalidation will clear the length.
1845 // FIXME: Even if we can't perfectly model the copy, we should see if we
1846 // can use LazyCompoundVals to copy the source values into the destination.
1847 // This would probably remove any existing bindings past the end of the
1848 // string, but that's still an improvement over blank invalidation.
1849 state = InvalidateBuffer(C, state, Dst, *dstRegVal,
1850 /*IsSourceBuffer*/false, nullptr);
1851
1852 // Invalidate the source (const-invalidation without const-pointer-escaping
1853 // the address of the top-level region).
1854 state = InvalidateBuffer(C, state, srcExpr, srcVal, /*IsSourceBuffer*/true,
1855 nullptr);
1856
1857 // Set the C string length of the destination, if we know it.
1858 if (isBounded && !isAppending) {
1859 // strncpy is annoying in that it doesn't guarantee to null-terminate
1860 // the result string. If the original string didn't fit entirely inside
1861 // the bound (including the null-terminator), we don't know how long the
1862 // result is.
1863 if (amountCopied != strLength)
1864 finalStrLength = UnknownVal();
1865 }
1866 state = setCStringLength(state, dstRegVal->getRegion(), finalStrLength);
1867 }
1868
1869 assert(state);
1870
1871 if (returnPtr) {
1872 // If this is a stpcpy-style copy, but we were unable to check for a buffer
1873 // overflow, we still need a result. Conjure a return value.
1874 if (returnEnd && Result.isUnknown()) {
1875 Result = svalBuilder.conjureSymbolVal(nullptr, CE, LCtx, C.blockCount());
1876 }
1877 }
1878 // Set the return value.
1879 state = state->BindExpr(CE, LCtx, Result);
1880 C.addTransition(state);
1881 }
1882
evalStrcmp(CheckerContext & C,const CallExpr * CE) const1883 void CStringChecker::evalStrcmp(CheckerContext &C, const CallExpr *CE) const {
1884 if (CE->getNumArgs() < 2)
1885 return;
1886
1887 //int strcmp(const char *s1, const char *s2);
1888 evalStrcmpCommon(C, CE, /* isBounded = */ false, /* ignoreCase = */ false);
1889 }
1890
evalStrncmp(CheckerContext & C,const CallExpr * CE) const1891 void CStringChecker::evalStrncmp(CheckerContext &C, const CallExpr *CE) const {
1892 if (CE->getNumArgs() < 3)
1893 return;
1894
1895 //int strncmp(const char *s1, const char *s2, size_t n);
1896 evalStrcmpCommon(C, CE, /* isBounded = */ true, /* ignoreCase = */ false);
1897 }
1898
evalStrcasecmp(CheckerContext & C,const CallExpr * CE) const1899 void CStringChecker::evalStrcasecmp(CheckerContext &C,
1900 const CallExpr *CE) const {
1901 if (CE->getNumArgs() < 2)
1902 return;
1903
1904 //int strcasecmp(const char *s1, const char *s2);
1905 evalStrcmpCommon(C, CE, /* isBounded = */ false, /* ignoreCase = */ true);
1906 }
1907
evalStrncasecmp(CheckerContext & C,const CallExpr * CE) const1908 void CStringChecker::evalStrncasecmp(CheckerContext &C,
1909 const CallExpr *CE) const {
1910 if (CE->getNumArgs() < 3)
1911 return;
1912
1913 //int strncasecmp(const char *s1, const char *s2, size_t n);
1914 evalStrcmpCommon(C, CE, /* isBounded = */ true, /* ignoreCase = */ true);
1915 }
1916
evalStrcmpCommon(CheckerContext & C,const CallExpr * CE,bool isBounded,bool ignoreCase) const1917 void CStringChecker::evalStrcmpCommon(CheckerContext &C, const CallExpr *CE,
1918 bool isBounded, bool ignoreCase) const {
1919 CurrentFunctionDescription = "string comparison function";
1920 ProgramStateRef state = C.getState();
1921 const LocationContext *LCtx = C.getLocationContext();
1922
1923 // Check that the first string is non-null
1924 const Expr *s1 = CE->getArg(0);
1925 SVal s1Val = state->getSVal(s1, LCtx);
1926 state = checkNonNull(C, state, s1, s1Val);
1927 if (!state)
1928 return;
1929
1930 // Check that the second string is non-null.
1931 const Expr *s2 = CE->getArg(1);
1932 SVal s2Val = state->getSVal(s2, LCtx);
1933 state = checkNonNull(C, state, s2, s2Val);
1934 if (!state)
1935 return;
1936
1937 // Get the string length of the first string or give up.
1938 SVal s1Length = getCStringLength(C, state, s1, s1Val);
1939 if (s1Length.isUndef())
1940 return;
1941
1942 // Get the string length of the second string or give up.
1943 SVal s2Length = getCStringLength(C, state, s2, s2Val);
1944 if (s2Length.isUndef())
1945 return;
1946
1947 // If we know the two buffers are the same, we know the result is 0.
1948 // First, get the two buffers' addresses. Another checker will have already
1949 // made sure they're not undefined.
1950 DefinedOrUnknownSVal LV = s1Val.castAs<DefinedOrUnknownSVal>();
1951 DefinedOrUnknownSVal RV = s2Val.castAs<DefinedOrUnknownSVal>();
1952
1953 // See if they are the same.
1954 SValBuilder &svalBuilder = C.getSValBuilder();
1955 DefinedOrUnknownSVal SameBuf = svalBuilder.evalEQ(state, LV, RV);
1956 ProgramStateRef StSameBuf, StNotSameBuf;
1957 std::tie(StSameBuf, StNotSameBuf) = state->assume(SameBuf);
1958
1959 // If the two arguments might be the same buffer, we know the result is 0,
1960 // and we only need to check one size.
1961 if (StSameBuf) {
1962 StSameBuf = StSameBuf->BindExpr(CE, LCtx,
1963 svalBuilder.makeZeroVal(CE->getType()));
1964 C.addTransition(StSameBuf);
1965
1966 // If the two arguments are GUARANTEED to be the same, we're done!
1967 if (!StNotSameBuf)
1968 return;
1969 }
1970
1971 assert(StNotSameBuf);
1972 state = StNotSameBuf;
1973
1974 // At this point we can go about comparing the two buffers.
1975 // For now, we only do this if they're both known string literals.
1976
1977 // Attempt to extract string literals from both expressions.
1978 const StringLiteral *s1StrLiteral = getCStringLiteral(C, state, s1, s1Val);
1979 const StringLiteral *s2StrLiteral = getCStringLiteral(C, state, s2, s2Val);
1980 bool canComputeResult = false;
1981 SVal resultVal = svalBuilder.conjureSymbolVal(nullptr, CE, LCtx,
1982 C.blockCount());
1983
1984 if (s1StrLiteral && s2StrLiteral) {
1985 StringRef s1StrRef = s1StrLiteral->getString();
1986 StringRef s2StrRef = s2StrLiteral->getString();
1987
1988 if (isBounded) {
1989 // Get the max number of characters to compare.
1990 const Expr *lenExpr = CE->getArg(2);
1991 SVal lenVal = state->getSVal(lenExpr, LCtx);
1992
1993 // If the length is known, we can get the right substrings.
1994 if (const llvm::APSInt *len = svalBuilder.getKnownValue(state, lenVal)) {
1995 // Create substrings of each to compare the prefix.
1996 s1StrRef = s1StrRef.substr(0, (size_t)len->getZExtValue());
1997 s2StrRef = s2StrRef.substr(0, (size_t)len->getZExtValue());
1998 canComputeResult = true;
1999 }
2000 } else {
2001 // This is a normal, unbounded strcmp.
2002 canComputeResult = true;
2003 }
2004
2005 if (canComputeResult) {
2006 // Real strcmp stops at null characters.
2007 size_t s1Term = s1StrRef.find('\0');
2008 if (s1Term != StringRef::npos)
2009 s1StrRef = s1StrRef.substr(0, s1Term);
2010
2011 size_t s2Term = s2StrRef.find('\0');
2012 if (s2Term != StringRef::npos)
2013 s2StrRef = s2StrRef.substr(0, s2Term);
2014
2015 // Use StringRef's comparison methods to compute the actual result.
2016 int compareRes = ignoreCase ? s1StrRef.compare_lower(s2StrRef)
2017 : s1StrRef.compare(s2StrRef);
2018
2019 // The strcmp function returns an integer greater than, equal to, or less
2020 // than zero, [c11, p7.24.4.2].
2021 if (compareRes == 0) {
2022 resultVal = svalBuilder.makeIntVal(compareRes, CE->getType());
2023 }
2024 else {
2025 DefinedSVal zeroVal = svalBuilder.makeIntVal(0, CE->getType());
2026 // Constrain strcmp's result range based on the result of StringRef's
2027 // comparison methods.
2028 BinaryOperatorKind op = (compareRes == 1) ? BO_GT : BO_LT;
2029 SVal compareWithZero =
2030 svalBuilder.evalBinOp(state, op, resultVal, zeroVal,
2031 svalBuilder.getConditionType());
2032 DefinedSVal compareWithZeroVal = compareWithZero.castAs<DefinedSVal>();
2033 state = state->assume(compareWithZeroVal, true);
2034 }
2035 }
2036 }
2037
2038 state = state->BindExpr(CE, LCtx, resultVal);
2039
2040 // Record this as a possible path.
2041 C.addTransition(state);
2042 }
2043
evalStrsep(CheckerContext & C,const CallExpr * CE) const2044 void CStringChecker::evalStrsep(CheckerContext &C, const CallExpr *CE) const {
2045 //char *strsep(char **stringp, const char *delim);
2046 if (CE->getNumArgs() < 2)
2047 return;
2048
2049 // Sanity: does the search string parameter match the return type?
2050 const Expr *SearchStrPtr = CE->getArg(0);
2051 QualType CharPtrTy = SearchStrPtr->getType()->getPointeeType();
2052 if (CharPtrTy.isNull() ||
2053 CE->getType().getUnqualifiedType() != CharPtrTy.getUnqualifiedType())
2054 return;
2055
2056 CurrentFunctionDescription = "strsep()";
2057 ProgramStateRef State = C.getState();
2058 const LocationContext *LCtx = C.getLocationContext();
2059
2060 // Check that the search string pointer is non-null (though it may point to
2061 // a null string).
2062 SVal SearchStrVal = State->getSVal(SearchStrPtr, LCtx);
2063 State = checkNonNull(C, State, SearchStrPtr, SearchStrVal);
2064 if (!State)
2065 return;
2066
2067 // Check that the delimiter string is non-null.
2068 const Expr *DelimStr = CE->getArg(1);
2069 SVal DelimStrVal = State->getSVal(DelimStr, LCtx);
2070 State = checkNonNull(C, State, DelimStr, DelimStrVal);
2071 if (!State)
2072 return;
2073
2074 SValBuilder &SVB = C.getSValBuilder();
2075 SVal Result;
2076 if (Optional<Loc> SearchStrLoc = SearchStrVal.getAs<Loc>()) {
2077 // Get the current value of the search string pointer, as a char*.
2078 Result = State->getSVal(*SearchStrLoc, CharPtrTy);
2079
2080 // Invalidate the search string, representing the change of one delimiter
2081 // character to NUL.
2082 State = InvalidateBuffer(C, State, SearchStrPtr, Result,
2083 /*IsSourceBuffer*/false, nullptr);
2084
2085 // Overwrite the search string pointer. The new value is either an address
2086 // further along in the same string, or NULL if there are no more tokens.
2087 State = State->bindLoc(*SearchStrLoc,
2088 SVB.conjureSymbolVal(getTag(),
2089 CE,
2090 LCtx,
2091 CharPtrTy,
2092 C.blockCount()),
2093 LCtx);
2094 } else {
2095 assert(SearchStrVal.isUnknown());
2096 // Conjure a symbolic value. It's the best we can do.
2097 Result = SVB.conjureSymbolVal(nullptr, CE, LCtx, C.blockCount());
2098 }
2099
2100 // Set the return value, and finish.
2101 State = State->BindExpr(CE, LCtx, Result);
2102 C.addTransition(State);
2103 }
2104
2105 // These should probably be moved into a C++ standard library checker.
evalStdCopy(CheckerContext & C,const CallExpr * CE) const2106 void CStringChecker::evalStdCopy(CheckerContext &C, const CallExpr *CE) const {
2107 evalStdCopyCommon(C, CE);
2108 }
2109
evalStdCopyBackward(CheckerContext & C,const CallExpr * CE) const2110 void CStringChecker::evalStdCopyBackward(CheckerContext &C,
2111 const CallExpr *CE) const {
2112 evalStdCopyCommon(C, CE);
2113 }
2114
evalStdCopyCommon(CheckerContext & C,const CallExpr * CE) const2115 void CStringChecker::evalStdCopyCommon(CheckerContext &C,
2116 const CallExpr *CE) const {
2117 if (CE->getNumArgs() < 3)
2118 return;
2119
2120 ProgramStateRef State = C.getState();
2121
2122 const LocationContext *LCtx = C.getLocationContext();
2123
2124 // template <class _InputIterator, class _OutputIterator>
2125 // _OutputIterator
2126 // copy(_InputIterator __first, _InputIterator __last,
2127 // _OutputIterator __result)
2128
2129 // Invalidate the destination buffer
2130 const Expr *Dst = CE->getArg(2);
2131 SVal DstVal = State->getSVal(Dst, LCtx);
2132 State = InvalidateBuffer(C, State, Dst, DstVal, /*IsSource=*/false,
2133 /*Size=*/nullptr);
2134
2135 SValBuilder &SVB = C.getSValBuilder();
2136
2137 SVal ResultVal = SVB.conjureSymbolVal(nullptr, CE, LCtx, C.blockCount());
2138 State = State->BindExpr(CE, LCtx, ResultVal);
2139
2140 C.addTransition(State);
2141 }
2142
evalMemset(CheckerContext & C,const CallExpr * CE) const2143 void CStringChecker::evalMemset(CheckerContext &C, const CallExpr *CE) const {
2144 if (CE->getNumArgs() != 3)
2145 return;
2146
2147 CurrentFunctionDescription = "memory set function";
2148
2149 const Expr *Mem = CE->getArg(0);
2150 const Expr *CharE = CE->getArg(1);
2151 const Expr *Size = CE->getArg(2);
2152 ProgramStateRef State = C.getState();
2153
2154 // See if the size argument is zero.
2155 const LocationContext *LCtx = C.getLocationContext();
2156 SVal SizeVal = State->getSVal(Size, LCtx);
2157 QualType SizeTy = Size->getType();
2158
2159 ProgramStateRef StateZeroSize, StateNonZeroSize;
2160 std::tie(StateZeroSize, StateNonZeroSize) =
2161 assumeZero(C, State, SizeVal, SizeTy);
2162
2163 // Get the value of the memory area.
2164 SVal MemVal = State->getSVal(Mem, LCtx);
2165
2166 // If the size is zero, there won't be any actual memory access, so
2167 // just bind the return value to the Mem buffer and return.
2168 if (StateZeroSize && !StateNonZeroSize) {
2169 StateZeroSize = StateZeroSize->BindExpr(CE, LCtx, MemVal);
2170 C.addTransition(StateZeroSize);
2171 return;
2172 }
2173
2174 // Ensure the memory area is not null.
2175 // If it is NULL there will be a NULL pointer dereference.
2176 State = checkNonNull(C, StateNonZeroSize, Mem, MemVal);
2177 if (!State)
2178 return;
2179
2180 State = CheckBufferAccess(C, State, Size, Mem);
2181 if (!State)
2182 return;
2183
2184 // According to the values of the arguments, bind the value of the second
2185 // argument to the destination buffer and set string length, or just
2186 // invalidate the destination buffer.
2187 if (!memsetAux(Mem, C.getSVal(CharE), Size, C, State))
2188 return;
2189
2190 State = State->BindExpr(CE, LCtx, MemVal);
2191 C.addTransition(State);
2192 }
2193
evalBzero(CheckerContext & C,const CallExpr * CE) const2194 void CStringChecker::evalBzero(CheckerContext &C, const CallExpr *CE) const {
2195 if (CE->getNumArgs() != 2)
2196 return;
2197
2198 CurrentFunctionDescription = "memory clearance function";
2199
2200 const Expr *Mem = CE->getArg(0);
2201 const Expr *Size = CE->getArg(1);
2202 SVal Zero = C.getSValBuilder().makeZeroVal(C.getASTContext().IntTy);
2203
2204 ProgramStateRef State = C.getState();
2205
2206 // See if the size argument is zero.
2207 SVal SizeVal = C.getSVal(Size);
2208 QualType SizeTy = Size->getType();
2209
2210 ProgramStateRef StateZeroSize, StateNonZeroSize;
2211 std::tie(StateZeroSize, StateNonZeroSize) =
2212 assumeZero(C, State, SizeVal, SizeTy);
2213
2214 // If the size is zero, there won't be any actual memory access,
2215 // In this case we just return.
2216 if (StateZeroSize && !StateNonZeroSize) {
2217 C.addTransition(StateZeroSize);
2218 return;
2219 }
2220
2221 // Get the value of the memory area.
2222 SVal MemVal = C.getSVal(Mem);
2223
2224 // Ensure the memory area is not null.
2225 // If it is NULL there will be a NULL pointer dereference.
2226 State = checkNonNull(C, StateNonZeroSize, Mem, MemVal);
2227 if (!State)
2228 return;
2229
2230 State = CheckBufferAccess(C, State, Size, Mem);
2231 if (!State)
2232 return;
2233
2234 if (!memsetAux(Mem, Zero, Size, C, State))
2235 return;
2236
2237 C.addTransition(State);
2238 }
2239
isCPPStdLibraryFunction(const FunctionDecl * FD,StringRef Name)2240 static bool isCPPStdLibraryFunction(const FunctionDecl *FD, StringRef Name) {
2241 IdentifierInfo *II = FD->getIdentifier();
2242 if (!II)
2243 return false;
2244
2245 if (!AnalysisDeclContext::isInStdNamespace(FD))
2246 return false;
2247
2248 if (II->getName().equals(Name))
2249 return true;
2250
2251 return false;
2252 }
2253 //===----------------------------------------------------------------------===//
2254 // The driver method, and other Checker callbacks.
2255 //===----------------------------------------------------------------------===//
2256
identifyCall(const CallExpr * CE,CheckerContext & C)2257 static CStringChecker::FnCheck identifyCall(const CallExpr *CE,
2258 CheckerContext &C) {
2259 const FunctionDecl *FDecl = C.getCalleeDecl(CE);
2260 if (!FDecl)
2261 return nullptr;
2262
2263 // Pro-actively check that argument types are safe to do arithmetic upon.
2264 // We do not want to crash if someone accidentally passes a structure
2265 // into, say, a C++ overload of any of these functions.
2266 if (isCPPStdLibraryFunction(FDecl, "copy")) {
2267 if (CE->getNumArgs() < 3 || !CE->getArg(2)->getType()->isPointerType())
2268 return nullptr;
2269 return &CStringChecker::evalStdCopy;
2270 } else if (isCPPStdLibraryFunction(FDecl, "copy_backward")) {
2271 if (CE->getNumArgs() < 3 || !CE->getArg(2)->getType()->isPointerType())
2272 return nullptr;
2273 return &CStringChecker::evalStdCopyBackward;
2274 } else {
2275 // An umbrella check for all C library functions.
2276 for (auto I: CE->arguments()) {
2277 QualType T = I->getType();
2278 if (!T->isIntegralOrEnumerationType() && !T->isPointerType())
2279 return nullptr;
2280 }
2281 }
2282
2283 // FIXME: Poorly-factored string switches are slow.
2284 if (C.isCLibraryFunction(FDecl, "memcpy"))
2285 return &CStringChecker::evalMemcpy;
2286 else if (C.isCLibraryFunction(FDecl, "mempcpy"))
2287 return &CStringChecker::evalMempcpy;
2288 else if (C.isCLibraryFunction(FDecl, "memcmp"))
2289 return &CStringChecker::evalMemcmp;
2290 else if (C.isCLibraryFunction(FDecl, "memmove"))
2291 return &CStringChecker::evalMemmove;
2292 else if (C.isCLibraryFunction(FDecl, "memset") ||
2293 C.isCLibraryFunction(FDecl, "explicit_memset"))
2294 return &CStringChecker::evalMemset;
2295 else if (C.isCLibraryFunction(FDecl, "strcpy"))
2296 return &CStringChecker::evalStrcpy;
2297 else if (C.isCLibraryFunction(FDecl, "strncpy"))
2298 return &CStringChecker::evalStrncpy;
2299 else if (C.isCLibraryFunction(FDecl, "stpcpy"))
2300 return &CStringChecker::evalStpcpy;
2301 else if (C.isCLibraryFunction(FDecl, "strlcpy"))
2302 return &CStringChecker::evalStrlcpy;
2303 else if (C.isCLibraryFunction(FDecl, "strcat"))
2304 return &CStringChecker::evalStrcat;
2305 else if (C.isCLibraryFunction(FDecl, "strncat"))
2306 return &CStringChecker::evalStrncat;
2307 else if (C.isCLibraryFunction(FDecl, "strlcat"))
2308 return &CStringChecker::evalStrlcat;
2309 else if (C.isCLibraryFunction(FDecl, "strlen"))
2310 return &CStringChecker::evalstrLength;
2311 else if (C.isCLibraryFunction(FDecl, "strnlen"))
2312 return &CStringChecker::evalstrnLength;
2313 else if (C.isCLibraryFunction(FDecl, "strcmp"))
2314 return &CStringChecker::evalStrcmp;
2315 else if (C.isCLibraryFunction(FDecl, "strncmp"))
2316 return &CStringChecker::evalStrncmp;
2317 else if (C.isCLibraryFunction(FDecl, "strcasecmp"))
2318 return &CStringChecker::evalStrcasecmp;
2319 else if (C.isCLibraryFunction(FDecl, "strncasecmp"))
2320 return &CStringChecker::evalStrncasecmp;
2321 else if (C.isCLibraryFunction(FDecl, "strsep"))
2322 return &CStringChecker::evalStrsep;
2323 else if (C.isCLibraryFunction(FDecl, "bcopy"))
2324 return &CStringChecker::evalBcopy;
2325 else if (C.isCLibraryFunction(FDecl, "bcmp"))
2326 return &CStringChecker::evalMemcmp;
2327 else if (C.isCLibraryFunction(FDecl, "bzero") ||
2328 C.isCLibraryFunction(FDecl, "explicit_bzero"))
2329 return &CStringChecker::evalBzero;
2330
2331 return nullptr;
2332 }
2333
evalCall(const CallExpr * CE,CheckerContext & C) const2334 bool CStringChecker::evalCall(const CallExpr *CE, CheckerContext &C) const {
2335
2336 FnCheck evalFunction = identifyCall(CE, C);
2337
2338 // If the callee isn't a string function, let another checker handle it.
2339 if (!evalFunction)
2340 return false;
2341
2342 // Check and evaluate the call.
2343 (this->*evalFunction)(C, CE);
2344
2345 // If the evaluate call resulted in no change, chain to the next eval call
2346 // handler.
2347 // Note, the custom CString evaluation calls assume that basic safety
2348 // properties are held. However, if the user chooses to turn off some of these
2349 // checks, we ignore the issues and leave the call evaluation to a generic
2350 // handler.
2351 return C.isDifferent();
2352 }
2353
checkPreStmt(const DeclStmt * DS,CheckerContext & C) const2354 void CStringChecker::checkPreStmt(const DeclStmt *DS, CheckerContext &C) const {
2355 // Record string length for char a[] = "abc";
2356 ProgramStateRef state = C.getState();
2357
2358 for (const auto *I : DS->decls()) {
2359 const VarDecl *D = dyn_cast<VarDecl>(I);
2360 if (!D)
2361 continue;
2362
2363 // FIXME: Handle array fields of structs.
2364 if (!D->getType()->isArrayType())
2365 continue;
2366
2367 const Expr *Init = D->getInit();
2368 if (!Init)
2369 continue;
2370 if (!isa<StringLiteral>(Init))
2371 continue;
2372
2373 Loc VarLoc = state->getLValue(D, C.getLocationContext());
2374 const MemRegion *MR = VarLoc.getAsRegion();
2375 if (!MR)
2376 continue;
2377
2378 SVal StrVal = C.getSVal(Init);
2379 assert(StrVal.isValid() && "Initializer string is unknown or undefined");
2380 DefinedOrUnknownSVal strLength =
2381 getCStringLength(C, state, Init, StrVal).castAs<DefinedOrUnknownSVal>();
2382
2383 state = state->set<CStringLength>(MR, strLength);
2384 }
2385
2386 C.addTransition(state);
2387 }
2388
2389 ProgramStateRef
checkRegionChanges(ProgramStateRef state,const InvalidatedSymbols *,ArrayRef<const MemRegion * > ExplicitRegions,ArrayRef<const MemRegion * > Regions,const LocationContext * LCtx,const CallEvent * Call) const2390 CStringChecker::checkRegionChanges(ProgramStateRef state,
2391 const InvalidatedSymbols *,
2392 ArrayRef<const MemRegion *> ExplicitRegions,
2393 ArrayRef<const MemRegion *> Regions,
2394 const LocationContext *LCtx,
2395 const CallEvent *Call) const {
2396 CStringLengthTy Entries = state->get<CStringLength>();
2397 if (Entries.isEmpty())
2398 return state;
2399
2400 llvm::SmallPtrSet<const MemRegion *, 8> Invalidated;
2401 llvm::SmallPtrSet<const MemRegion *, 32> SuperRegions;
2402
2403 // First build sets for the changed regions and their super-regions.
2404 for (ArrayRef<const MemRegion *>::iterator
2405 I = Regions.begin(), E = Regions.end(); I != E; ++I) {
2406 const MemRegion *MR = *I;
2407 Invalidated.insert(MR);
2408
2409 SuperRegions.insert(MR);
2410 while (const SubRegion *SR = dyn_cast<SubRegion>(MR)) {
2411 MR = SR->getSuperRegion();
2412 SuperRegions.insert(MR);
2413 }
2414 }
2415
2416 CStringLengthTy::Factory &F = state->get_context<CStringLength>();
2417
2418 // Then loop over the entries in the current state.
2419 for (CStringLengthTy::iterator I = Entries.begin(),
2420 E = Entries.end(); I != E; ++I) {
2421 const MemRegion *MR = I.getKey();
2422
2423 // Is this entry for a super-region of a changed region?
2424 if (SuperRegions.count(MR)) {
2425 Entries = F.remove(Entries, MR);
2426 continue;
2427 }
2428
2429 // Is this entry for a sub-region of a changed region?
2430 const MemRegion *Super = MR;
2431 while (const SubRegion *SR = dyn_cast<SubRegion>(Super)) {
2432 Super = SR->getSuperRegion();
2433 if (Invalidated.count(Super)) {
2434 Entries = F.remove(Entries, MR);
2435 break;
2436 }
2437 }
2438 }
2439
2440 return state->set<CStringLength>(Entries);
2441 }
2442
checkLiveSymbols(ProgramStateRef state,SymbolReaper & SR) const2443 void CStringChecker::checkLiveSymbols(ProgramStateRef state,
2444 SymbolReaper &SR) const {
2445 // Mark all symbols in our string length map as valid.
2446 CStringLengthTy Entries = state->get<CStringLength>();
2447
2448 for (CStringLengthTy::iterator I = Entries.begin(), E = Entries.end();
2449 I != E; ++I) {
2450 SVal Len = I.getData();
2451
2452 for (SymExpr::symbol_iterator si = Len.symbol_begin(),
2453 se = Len.symbol_end(); si != se; ++si)
2454 SR.markInUse(*si);
2455 }
2456 }
2457
checkDeadSymbols(SymbolReaper & SR,CheckerContext & C) const2458 void CStringChecker::checkDeadSymbols(SymbolReaper &SR,
2459 CheckerContext &C) const {
2460 ProgramStateRef state = C.getState();
2461 CStringLengthTy Entries = state->get<CStringLength>();
2462 if (Entries.isEmpty())
2463 return;
2464
2465 CStringLengthTy::Factory &F = state->get_context<CStringLength>();
2466 for (CStringLengthTy::iterator I = Entries.begin(), E = Entries.end();
2467 I != E; ++I) {
2468 SVal Len = I.getData();
2469 if (SymbolRef Sym = Len.getAsSymbol()) {
2470 if (SR.isDead(Sym))
2471 Entries = F.remove(Entries, I.getKey());
2472 }
2473 }
2474
2475 state = state->set<CStringLength>(Entries);
2476 C.addTransition(state);
2477 }
2478
2479 #define REGISTER_CHECKER(name) \
2480 void ento::register##name(CheckerManager &mgr) { \
2481 CStringChecker *checker = mgr.registerChecker<CStringChecker>(); \
2482 checker->Filter.Check##name = true; \
2483 checker->Filter.CheckName##name = mgr.getCurrentCheckName(); \
2484 }
2485
2486 REGISTER_CHECKER(CStringNullArg)
REGISTER_CHECKER(CStringOutOfBounds)2487 REGISTER_CHECKER(CStringOutOfBounds)
2488 REGISTER_CHECKER(CStringBufferOverlap)
2489 REGISTER_CHECKER(CStringNotNullTerm)
2490
2491 void ento::registerCStringCheckerBasic(CheckerManager &Mgr) {
2492 Mgr.registerChecker<CStringChecker>();
2493 }
2494