1 //= CStringChecker.cpp - Checks calls to C string functions --------*- C++ -*-//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This defines CStringChecker, which is an assortment of checks on calls
10 // to functions in <string.h>.
11 //
12 //===----------------------------------------------------------------------===//
13
14 #include "InterCheckerAPI.h"
15 #include "clang/Basic/CharInfo.h"
16 #include "clang/StaticAnalyzer/Checkers/BuiltinCheckerRegistration.h"
17 #include "clang/StaticAnalyzer/Core/BugReporter/BugType.h"
18 #include "clang/StaticAnalyzer/Core/Checker.h"
19 #include "clang/StaticAnalyzer/Core/CheckerManager.h"
20 #include "clang/StaticAnalyzer/Core/PathSensitive/CallEvent.h"
21 #include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h"
22 #include "clang/StaticAnalyzer/Core/PathSensitive/DynamicExtent.h"
23 #include "clang/StaticAnalyzer/Core/PathSensitive/ProgramStateTrait.h"
24 #include "llvm/ADT/STLExtras.h"
25 #include "llvm/ADT/SmallString.h"
26 #include "llvm/ADT/StringExtras.h"
27 #include "llvm/Support/raw_ostream.h"
28
29 using namespace clang;
30 using namespace ento;
31
32 namespace {
33 struct AnyArgExpr {
34 // FIXME: Remove constructor in C++17 to turn it into an aggregate.
AnyArgExpr__anona731634b0111::AnyArgExpr35 AnyArgExpr(const Expr *Expression, unsigned ArgumentIndex)
36 : Expression{Expression}, ArgumentIndex{ArgumentIndex} {}
37 const Expr *Expression;
38 unsigned ArgumentIndex;
39 };
40
41 struct SourceArgExpr : AnyArgExpr {
42 using AnyArgExpr::AnyArgExpr; // FIXME: Remove using in C++17.
43 };
44
45 struct DestinationArgExpr : AnyArgExpr {
46 using AnyArgExpr::AnyArgExpr; // FIXME: Same.
47 };
48
49 struct SizeArgExpr : AnyArgExpr {
50 using AnyArgExpr::AnyArgExpr; // FIXME: Same.
51 };
52
53 using ErrorMessage = SmallString<128>;
54 enum class AccessKind { write, read };
55
createOutOfBoundErrorMsg(StringRef FunctionDescription,AccessKind Access)56 static ErrorMessage createOutOfBoundErrorMsg(StringRef FunctionDescription,
57 AccessKind Access) {
58 ErrorMessage Message;
59 llvm::raw_svector_ostream Os(Message);
60
61 // Function classification like: Memory copy function
62 Os << toUppercase(FunctionDescription.front())
63 << &FunctionDescription.data()[1];
64
65 if (Access == AccessKind::write) {
66 Os << " overflows the destination buffer";
67 } else { // read access
68 Os << " accesses out-of-bound array element";
69 }
70
71 return Message;
72 }
73
74 enum class ConcatFnKind { none = 0, strcat = 1, strlcat = 2 };
75 class CStringChecker : public Checker< eval::Call,
76 check::PreStmt<DeclStmt>,
77 check::LiveSymbols,
78 check::DeadSymbols,
79 check::RegionChanges
80 > {
81 mutable std::unique_ptr<BugType> BT_Null, BT_Bounds, BT_Overlap,
82 BT_NotCString, BT_AdditionOverflow;
83
84 mutable const char *CurrentFunctionDescription;
85
86 public:
87 /// The filter is used to filter out the diagnostics which are not enabled by
88 /// the user.
89 struct CStringChecksFilter {
90 DefaultBool CheckCStringNullArg;
91 DefaultBool CheckCStringOutOfBounds;
92 DefaultBool CheckCStringBufferOverlap;
93 DefaultBool CheckCStringNotNullTerm;
94
95 CheckerNameRef CheckNameCStringNullArg;
96 CheckerNameRef CheckNameCStringOutOfBounds;
97 CheckerNameRef CheckNameCStringBufferOverlap;
98 CheckerNameRef CheckNameCStringNotNullTerm;
99 };
100
101 CStringChecksFilter Filter;
102
getTag()103 static void *getTag() { static int tag; return &tag; }
104
105 bool evalCall(const CallEvent &Call, CheckerContext &C) const;
106 void checkPreStmt(const DeclStmt *DS, CheckerContext &C) const;
107 void checkLiveSymbols(ProgramStateRef state, SymbolReaper &SR) const;
108 void checkDeadSymbols(SymbolReaper &SR, CheckerContext &C) const;
109
110 ProgramStateRef
111 checkRegionChanges(ProgramStateRef state,
112 const InvalidatedSymbols *,
113 ArrayRef<const MemRegion *> ExplicitRegions,
114 ArrayRef<const MemRegion *> Regions,
115 const LocationContext *LCtx,
116 const CallEvent *Call) const;
117
118 typedef void (CStringChecker::*FnCheck)(CheckerContext &,
119 const CallExpr *) const;
120 CallDescriptionMap<FnCheck> Callbacks = {
121 {{CDF_MaybeBuiltin, "memcpy", 3}, &CStringChecker::evalMemcpy},
122 {{CDF_MaybeBuiltin, "mempcpy", 3}, &CStringChecker::evalMempcpy},
123 {{CDF_MaybeBuiltin, "memcmp", 3}, &CStringChecker::evalMemcmp},
124 {{CDF_MaybeBuiltin, "memmove", 3}, &CStringChecker::evalMemmove},
125 {{CDF_MaybeBuiltin, "memset", 3}, &CStringChecker::evalMemset},
126 {{CDF_MaybeBuiltin, "explicit_memset", 3}, &CStringChecker::evalMemset},
127 {{CDF_MaybeBuiltin, "strcpy", 2}, &CStringChecker::evalStrcpy},
128 {{CDF_MaybeBuiltin, "strncpy", 3}, &CStringChecker::evalStrncpy},
129 {{CDF_MaybeBuiltin, "stpcpy", 2}, &CStringChecker::evalStpcpy},
130 {{CDF_MaybeBuiltin, "strlcpy", 3}, &CStringChecker::evalStrlcpy},
131 {{CDF_MaybeBuiltin, "strcat", 2}, &CStringChecker::evalStrcat},
132 {{CDF_MaybeBuiltin, "strncat", 3}, &CStringChecker::evalStrncat},
133 {{CDF_MaybeBuiltin, "strlcat", 3}, &CStringChecker::evalStrlcat},
134 {{CDF_MaybeBuiltin, "strlen", 1}, &CStringChecker::evalstrLength},
135 {{CDF_MaybeBuiltin, "strnlen", 2}, &CStringChecker::evalstrnLength},
136 {{CDF_MaybeBuiltin, "strcmp", 2}, &CStringChecker::evalStrcmp},
137 {{CDF_MaybeBuiltin, "strncmp", 3}, &CStringChecker::evalStrncmp},
138 {{CDF_MaybeBuiltin, "strcasecmp", 2}, &CStringChecker::evalStrcasecmp},
139 {{CDF_MaybeBuiltin, "strncasecmp", 3}, &CStringChecker::evalStrncasecmp},
140 {{CDF_MaybeBuiltin, "strsep", 2}, &CStringChecker::evalStrsep},
141 {{CDF_MaybeBuiltin, "bcopy", 3}, &CStringChecker::evalBcopy},
142 {{CDF_MaybeBuiltin, "bcmp", 3}, &CStringChecker::evalMemcmp},
143 {{CDF_MaybeBuiltin, "bzero", 2}, &CStringChecker::evalBzero},
144 {{CDF_MaybeBuiltin, "explicit_bzero", 2}, &CStringChecker::evalBzero},
145 };
146
147 // These require a bit of special handling.
148 CallDescription StdCopy{{"std", "copy"}, 3},
149 StdCopyBackward{{"std", "copy_backward"}, 3};
150
151 FnCheck identifyCall(const CallEvent &Call, CheckerContext &C) const;
152 void evalMemcpy(CheckerContext &C, const CallExpr *CE) const;
153 void evalMempcpy(CheckerContext &C, const CallExpr *CE) const;
154 void evalMemmove(CheckerContext &C, const CallExpr *CE) const;
155 void evalBcopy(CheckerContext &C, const CallExpr *CE) const;
156 void evalCopyCommon(CheckerContext &C, const CallExpr *CE,
157 ProgramStateRef state, SizeArgExpr Size,
158 DestinationArgExpr Dest, SourceArgExpr Source,
159 bool Restricted, bool IsMempcpy) const;
160
161 void evalMemcmp(CheckerContext &C, const CallExpr *CE) const;
162
163 void evalstrLength(CheckerContext &C, const CallExpr *CE) const;
164 void evalstrnLength(CheckerContext &C, const CallExpr *CE) const;
165 void evalstrLengthCommon(CheckerContext &C,
166 const CallExpr *CE,
167 bool IsStrnlen = false) const;
168
169 void evalStrcpy(CheckerContext &C, const CallExpr *CE) const;
170 void evalStrncpy(CheckerContext &C, const CallExpr *CE) const;
171 void evalStpcpy(CheckerContext &C, const CallExpr *CE) const;
172 void evalStrlcpy(CheckerContext &C, const CallExpr *CE) const;
173 void evalStrcpyCommon(CheckerContext &C, const CallExpr *CE, bool ReturnEnd,
174 bool IsBounded, ConcatFnKind appendK,
175 bool returnPtr = true) const;
176
177 void evalStrcat(CheckerContext &C, const CallExpr *CE) const;
178 void evalStrncat(CheckerContext &C, const CallExpr *CE) const;
179 void evalStrlcat(CheckerContext &C, const CallExpr *CE) const;
180
181 void evalStrcmp(CheckerContext &C, const CallExpr *CE) const;
182 void evalStrncmp(CheckerContext &C, const CallExpr *CE) const;
183 void evalStrcasecmp(CheckerContext &C, const CallExpr *CE) const;
184 void evalStrncasecmp(CheckerContext &C, const CallExpr *CE) const;
185 void evalStrcmpCommon(CheckerContext &C,
186 const CallExpr *CE,
187 bool IsBounded = false,
188 bool IgnoreCase = false) const;
189
190 void evalStrsep(CheckerContext &C, const CallExpr *CE) const;
191
192 void evalStdCopy(CheckerContext &C, const CallExpr *CE) const;
193 void evalStdCopyBackward(CheckerContext &C, const CallExpr *CE) const;
194 void evalStdCopyCommon(CheckerContext &C, const CallExpr *CE) const;
195 void evalMemset(CheckerContext &C, const CallExpr *CE) const;
196 void evalBzero(CheckerContext &C, const CallExpr *CE) const;
197
198 // Utility methods
199 std::pair<ProgramStateRef , ProgramStateRef >
200 static assumeZero(CheckerContext &C,
201 ProgramStateRef state, SVal V, QualType Ty);
202
203 static ProgramStateRef setCStringLength(ProgramStateRef state,
204 const MemRegion *MR,
205 SVal strLength);
206 static SVal getCStringLengthForRegion(CheckerContext &C,
207 ProgramStateRef &state,
208 const Expr *Ex,
209 const MemRegion *MR,
210 bool hypothetical);
211 SVal getCStringLength(CheckerContext &C,
212 ProgramStateRef &state,
213 const Expr *Ex,
214 SVal Buf,
215 bool hypothetical = false) const;
216
217 const StringLiteral *getCStringLiteral(CheckerContext &C,
218 ProgramStateRef &state,
219 const Expr *expr,
220 SVal val) const;
221
222 static ProgramStateRef InvalidateBuffer(CheckerContext &C,
223 ProgramStateRef state,
224 const Expr *Ex, SVal V,
225 bool IsSourceBuffer,
226 const Expr *Size);
227
228 static bool SummarizeRegion(raw_ostream &os, ASTContext &Ctx,
229 const MemRegion *MR);
230
231 static bool memsetAux(const Expr *DstBuffer, SVal CharE,
232 const Expr *Size, CheckerContext &C,
233 ProgramStateRef &State);
234
235 // Re-usable checks
236 ProgramStateRef checkNonNull(CheckerContext &C, ProgramStateRef State,
237 AnyArgExpr Arg, SVal l) const;
238 ProgramStateRef CheckLocation(CheckerContext &C, ProgramStateRef state,
239 AnyArgExpr Buffer, SVal Element,
240 AccessKind Access) const;
241 ProgramStateRef CheckBufferAccess(CheckerContext &C, ProgramStateRef State,
242 AnyArgExpr Buffer, SizeArgExpr Size,
243 AccessKind Access) const;
244 ProgramStateRef CheckOverlap(CheckerContext &C, ProgramStateRef state,
245 SizeArgExpr Size, AnyArgExpr First,
246 AnyArgExpr Second) const;
247 void emitOverlapBug(CheckerContext &C,
248 ProgramStateRef state,
249 const Stmt *First,
250 const Stmt *Second) const;
251
252 void emitNullArgBug(CheckerContext &C, ProgramStateRef State, const Stmt *S,
253 StringRef WarningMsg) const;
254 void emitOutOfBoundsBug(CheckerContext &C, ProgramStateRef State,
255 const Stmt *S, StringRef WarningMsg) const;
256 void emitNotCStringBug(CheckerContext &C, ProgramStateRef State,
257 const Stmt *S, StringRef WarningMsg) const;
258 void emitAdditionOverflowBug(CheckerContext &C, ProgramStateRef State) const;
259
260 ProgramStateRef checkAdditionOverflow(CheckerContext &C,
261 ProgramStateRef state,
262 NonLoc left,
263 NonLoc right) const;
264
265 // Return true if the destination buffer of the copy function may be in bound.
266 // Expects SVal of Size to be positive and unsigned.
267 // Expects SVal of FirstBuf to be a FieldRegion.
268 static bool IsFirstBufInBound(CheckerContext &C,
269 ProgramStateRef state,
270 const Expr *FirstBuf,
271 const Expr *Size);
272 };
273
274 } //end anonymous namespace
275
REGISTER_MAP_WITH_PROGRAMSTATE(CStringLength,const MemRegion *,SVal)276 REGISTER_MAP_WITH_PROGRAMSTATE(CStringLength, const MemRegion *, SVal)
277
278 //===----------------------------------------------------------------------===//
279 // Individual checks and utility methods.
280 //===----------------------------------------------------------------------===//
281
282 std::pair<ProgramStateRef , ProgramStateRef >
283 CStringChecker::assumeZero(CheckerContext &C, ProgramStateRef state, SVal V,
284 QualType Ty) {
285 Optional<DefinedSVal> val = V.getAs<DefinedSVal>();
286 if (!val)
287 return std::pair<ProgramStateRef , ProgramStateRef >(state, state);
288
289 SValBuilder &svalBuilder = C.getSValBuilder();
290 DefinedOrUnknownSVal zero = svalBuilder.makeZeroVal(Ty);
291 return state->assume(svalBuilder.evalEQ(state, *val, zero));
292 }
293
checkNonNull(CheckerContext & C,ProgramStateRef State,AnyArgExpr Arg,SVal l) const294 ProgramStateRef CStringChecker::checkNonNull(CheckerContext &C,
295 ProgramStateRef State,
296 AnyArgExpr Arg, SVal l) const {
297 // If a previous check has failed, propagate the failure.
298 if (!State)
299 return nullptr;
300
301 ProgramStateRef stateNull, stateNonNull;
302 std::tie(stateNull, stateNonNull) =
303 assumeZero(C, State, l, Arg.Expression->getType());
304
305 if (stateNull && !stateNonNull) {
306 if (Filter.CheckCStringNullArg) {
307 SmallString<80> buf;
308 llvm::raw_svector_ostream OS(buf);
309 assert(CurrentFunctionDescription);
310 OS << "Null pointer passed as " << (Arg.ArgumentIndex + 1)
311 << llvm::getOrdinalSuffix(Arg.ArgumentIndex + 1) << " argument to "
312 << CurrentFunctionDescription;
313
314 emitNullArgBug(C, stateNull, Arg.Expression, OS.str());
315 }
316 return nullptr;
317 }
318
319 // From here on, assume that the value is non-null.
320 assert(stateNonNull);
321 return stateNonNull;
322 }
323
324 // FIXME: This was originally copied from ArrayBoundChecker.cpp. Refactor?
CheckLocation(CheckerContext & C,ProgramStateRef state,AnyArgExpr Buffer,SVal Element,AccessKind Access) const325 ProgramStateRef CStringChecker::CheckLocation(CheckerContext &C,
326 ProgramStateRef state,
327 AnyArgExpr Buffer, SVal Element,
328 AccessKind Access) const {
329
330 // If a previous check has failed, propagate the failure.
331 if (!state)
332 return nullptr;
333
334 // Check for out of bound array element access.
335 const MemRegion *R = Element.getAsRegion();
336 if (!R)
337 return state;
338
339 const auto *ER = dyn_cast<ElementRegion>(R);
340 if (!ER)
341 return state;
342
343 if (ER->getValueType() != C.getASTContext().CharTy)
344 return state;
345
346 // Get the size of the array.
347 const auto *superReg = cast<SubRegion>(ER->getSuperRegion());
348 DefinedOrUnknownSVal Size =
349 getDynamicExtent(state, superReg, C.getSValBuilder());
350
351 // Get the index of the accessed element.
352 DefinedOrUnknownSVal Idx = ER->getIndex().castAs<DefinedOrUnknownSVal>();
353
354 ProgramStateRef StInBound = state->assumeInBound(Idx, Size, true);
355 ProgramStateRef StOutBound = state->assumeInBound(Idx, Size, false);
356 if (StOutBound && !StInBound) {
357 // These checks are either enabled by the CString out-of-bounds checker
358 // explicitly or implicitly by the Malloc checker.
359 // In the latter case we only do modeling but do not emit warning.
360 if (!Filter.CheckCStringOutOfBounds)
361 return nullptr;
362
363 // Emit a bug report.
364 ErrorMessage Message =
365 createOutOfBoundErrorMsg(CurrentFunctionDescription, Access);
366 emitOutOfBoundsBug(C, StOutBound, Buffer.Expression, Message);
367 return nullptr;
368 }
369
370 // Array bound check succeeded. From this point forward the array bound
371 // should always succeed.
372 return StInBound;
373 }
374
CheckBufferAccess(CheckerContext & C,ProgramStateRef State,AnyArgExpr Buffer,SizeArgExpr Size,AccessKind Access) const375 ProgramStateRef CStringChecker::CheckBufferAccess(CheckerContext &C,
376 ProgramStateRef State,
377 AnyArgExpr Buffer,
378 SizeArgExpr Size,
379 AccessKind Access) const {
380 // If a previous check has failed, propagate the failure.
381 if (!State)
382 return nullptr;
383
384 SValBuilder &svalBuilder = C.getSValBuilder();
385 ASTContext &Ctx = svalBuilder.getContext();
386
387 QualType SizeTy = Size.Expression->getType();
388 QualType PtrTy = Ctx.getPointerType(Ctx.CharTy);
389
390 // Check that the first buffer is non-null.
391 SVal BufVal = C.getSVal(Buffer.Expression);
392 State = checkNonNull(C, State, Buffer, BufVal);
393 if (!State)
394 return nullptr;
395
396 // If out-of-bounds checking is turned off, skip the rest.
397 if (!Filter.CheckCStringOutOfBounds)
398 return State;
399
400 // Get the access length and make sure it is known.
401 // FIXME: This assumes the caller has already checked that the access length
402 // is positive. And that it's unsigned.
403 SVal LengthVal = C.getSVal(Size.Expression);
404 Optional<NonLoc> Length = LengthVal.getAs<NonLoc>();
405 if (!Length)
406 return State;
407
408 // Compute the offset of the last element to be accessed: size-1.
409 NonLoc One = svalBuilder.makeIntVal(1, SizeTy).castAs<NonLoc>();
410 SVal Offset = svalBuilder.evalBinOpNN(State, BO_Sub, *Length, One, SizeTy);
411 if (Offset.isUnknown())
412 return nullptr;
413 NonLoc LastOffset = Offset.castAs<NonLoc>();
414
415 // Check that the first buffer is sufficiently long.
416 SVal BufStart =
417 svalBuilder.evalCast(BufVal, PtrTy, Buffer.Expression->getType());
418 if (Optional<Loc> BufLoc = BufStart.getAs<Loc>()) {
419
420 SVal BufEnd =
421 svalBuilder.evalBinOpLN(State, BO_Add, *BufLoc, LastOffset, PtrTy);
422
423 State = CheckLocation(C, State, Buffer, BufEnd, Access);
424
425 // If the buffer isn't large enough, abort.
426 if (!State)
427 return nullptr;
428 }
429
430 // Large enough or not, return this state!
431 return State;
432 }
433
CheckOverlap(CheckerContext & C,ProgramStateRef state,SizeArgExpr Size,AnyArgExpr First,AnyArgExpr Second) const434 ProgramStateRef CStringChecker::CheckOverlap(CheckerContext &C,
435 ProgramStateRef state,
436 SizeArgExpr Size, AnyArgExpr First,
437 AnyArgExpr Second) const {
438 if (!Filter.CheckCStringBufferOverlap)
439 return state;
440
441 // Do a simple check for overlap: if the two arguments are from the same
442 // buffer, see if the end of the first is greater than the start of the second
443 // or vice versa.
444
445 // If a previous check has failed, propagate the failure.
446 if (!state)
447 return nullptr;
448
449 ProgramStateRef stateTrue, stateFalse;
450
451 // Get the buffer values and make sure they're known locations.
452 const LocationContext *LCtx = C.getLocationContext();
453 SVal firstVal = state->getSVal(First.Expression, LCtx);
454 SVal secondVal = state->getSVal(Second.Expression, LCtx);
455
456 Optional<Loc> firstLoc = firstVal.getAs<Loc>();
457 if (!firstLoc)
458 return state;
459
460 Optional<Loc> secondLoc = secondVal.getAs<Loc>();
461 if (!secondLoc)
462 return state;
463
464 // Are the two values the same?
465 SValBuilder &svalBuilder = C.getSValBuilder();
466 std::tie(stateTrue, stateFalse) =
467 state->assume(svalBuilder.evalEQ(state, *firstLoc, *secondLoc));
468
469 if (stateTrue && !stateFalse) {
470 // If the values are known to be equal, that's automatically an overlap.
471 emitOverlapBug(C, stateTrue, First.Expression, Second.Expression);
472 return nullptr;
473 }
474
475 // assume the two expressions are not equal.
476 assert(stateFalse);
477 state = stateFalse;
478
479 // Which value comes first?
480 QualType cmpTy = svalBuilder.getConditionType();
481 SVal reverse =
482 svalBuilder.evalBinOpLL(state, BO_GT, *firstLoc, *secondLoc, cmpTy);
483 Optional<DefinedOrUnknownSVal> reverseTest =
484 reverse.getAs<DefinedOrUnknownSVal>();
485 if (!reverseTest)
486 return state;
487
488 std::tie(stateTrue, stateFalse) = state->assume(*reverseTest);
489 if (stateTrue) {
490 if (stateFalse) {
491 // If we don't know which one comes first, we can't perform this test.
492 return state;
493 } else {
494 // Switch the values so that firstVal is before secondVal.
495 std::swap(firstLoc, secondLoc);
496
497 // Switch the Exprs as well, so that they still correspond.
498 std::swap(First, Second);
499 }
500 }
501
502 // Get the length, and make sure it too is known.
503 SVal LengthVal = state->getSVal(Size.Expression, LCtx);
504 Optional<NonLoc> Length = LengthVal.getAs<NonLoc>();
505 if (!Length)
506 return state;
507
508 // Convert the first buffer's start address to char*.
509 // Bail out if the cast fails.
510 ASTContext &Ctx = svalBuilder.getContext();
511 QualType CharPtrTy = Ctx.getPointerType(Ctx.CharTy);
512 SVal FirstStart =
513 svalBuilder.evalCast(*firstLoc, CharPtrTy, First.Expression->getType());
514 Optional<Loc> FirstStartLoc = FirstStart.getAs<Loc>();
515 if (!FirstStartLoc)
516 return state;
517
518 // Compute the end of the first buffer. Bail out if THAT fails.
519 SVal FirstEnd = svalBuilder.evalBinOpLN(state, BO_Add, *FirstStartLoc,
520 *Length, CharPtrTy);
521 Optional<Loc> FirstEndLoc = FirstEnd.getAs<Loc>();
522 if (!FirstEndLoc)
523 return state;
524
525 // Is the end of the first buffer past the start of the second buffer?
526 SVal Overlap =
527 svalBuilder.evalBinOpLL(state, BO_GT, *FirstEndLoc, *secondLoc, cmpTy);
528 Optional<DefinedOrUnknownSVal> OverlapTest =
529 Overlap.getAs<DefinedOrUnknownSVal>();
530 if (!OverlapTest)
531 return state;
532
533 std::tie(stateTrue, stateFalse) = state->assume(*OverlapTest);
534
535 if (stateTrue && !stateFalse) {
536 // Overlap!
537 emitOverlapBug(C, stateTrue, First.Expression, Second.Expression);
538 return nullptr;
539 }
540
541 // assume the two expressions don't overlap.
542 assert(stateFalse);
543 return stateFalse;
544 }
545
emitOverlapBug(CheckerContext & C,ProgramStateRef state,const Stmt * First,const Stmt * Second) const546 void CStringChecker::emitOverlapBug(CheckerContext &C, ProgramStateRef state,
547 const Stmt *First, const Stmt *Second) const {
548 ExplodedNode *N = C.generateErrorNode(state);
549 if (!N)
550 return;
551
552 if (!BT_Overlap)
553 BT_Overlap.reset(new BugType(Filter.CheckNameCStringBufferOverlap,
554 categories::UnixAPI, "Improper arguments"));
555
556 // Generate a report for this bug.
557 auto report = std::make_unique<PathSensitiveBugReport>(
558 *BT_Overlap, "Arguments must not be overlapping buffers", N);
559 report->addRange(First->getSourceRange());
560 report->addRange(Second->getSourceRange());
561
562 C.emitReport(std::move(report));
563 }
564
emitNullArgBug(CheckerContext & C,ProgramStateRef State,const Stmt * S,StringRef WarningMsg) const565 void CStringChecker::emitNullArgBug(CheckerContext &C, ProgramStateRef State,
566 const Stmt *S, StringRef WarningMsg) const {
567 if (ExplodedNode *N = C.generateErrorNode(State)) {
568 if (!BT_Null)
569 BT_Null.reset(new BuiltinBug(
570 Filter.CheckNameCStringNullArg, categories::UnixAPI,
571 "Null pointer argument in call to byte string function"));
572
573 BuiltinBug *BT = static_cast<BuiltinBug *>(BT_Null.get());
574 auto Report = std::make_unique<PathSensitiveBugReport>(*BT, WarningMsg, N);
575 Report->addRange(S->getSourceRange());
576 if (const auto *Ex = dyn_cast<Expr>(S))
577 bugreporter::trackExpressionValue(N, Ex, *Report);
578 C.emitReport(std::move(Report));
579 }
580 }
581
emitOutOfBoundsBug(CheckerContext & C,ProgramStateRef State,const Stmt * S,StringRef WarningMsg) const582 void CStringChecker::emitOutOfBoundsBug(CheckerContext &C,
583 ProgramStateRef State, const Stmt *S,
584 StringRef WarningMsg) const {
585 if (ExplodedNode *N = C.generateErrorNode(State)) {
586 if (!BT_Bounds)
587 BT_Bounds.reset(new BuiltinBug(
588 Filter.CheckCStringOutOfBounds ? Filter.CheckNameCStringOutOfBounds
589 : Filter.CheckNameCStringNullArg,
590 "Out-of-bound array access",
591 "Byte string function accesses out-of-bound array element"));
592
593 BuiltinBug *BT = static_cast<BuiltinBug *>(BT_Bounds.get());
594
595 // FIXME: It would be nice to eventually make this diagnostic more clear,
596 // e.g., by referencing the original declaration or by saying *why* this
597 // reference is outside the range.
598 auto Report = std::make_unique<PathSensitiveBugReport>(*BT, WarningMsg, N);
599 Report->addRange(S->getSourceRange());
600 C.emitReport(std::move(Report));
601 }
602 }
603
emitNotCStringBug(CheckerContext & C,ProgramStateRef State,const Stmt * S,StringRef WarningMsg) const604 void CStringChecker::emitNotCStringBug(CheckerContext &C, ProgramStateRef State,
605 const Stmt *S,
606 StringRef WarningMsg) const {
607 if (ExplodedNode *N = C.generateNonFatalErrorNode(State)) {
608 if (!BT_NotCString)
609 BT_NotCString.reset(new BuiltinBug(
610 Filter.CheckNameCStringNotNullTerm, categories::UnixAPI,
611 "Argument is not a null-terminated string."));
612
613 auto Report =
614 std::make_unique<PathSensitiveBugReport>(*BT_NotCString, WarningMsg, N);
615
616 Report->addRange(S->getSourceRange());
617 C.emitReport(std::move(Report));
618 }
619 }
620
emitAdditionOverflowBug(CheckerContext & C,ProgramStateRef State) const621 void CStringChecker::emitAdditionOverflowBug(CheckerContext &C,
622 ProgramStateRef State) const {
623 if (ExplodedNode *N = C.generateErrorNode(State)) {
624 if (!BT_NotCString)
625 BT_NotCString.reset(
626 new BuiltinBug(Filter.CheckNameCStringOutOfBounds, "API",
627 "Sum of expressions causes overflow."));
628
629 // This isn't a great error message, but this should never occur in real
630 // code anyway -- you'd have to create a buffer longer than a size_t can
631 // represent, which is sort of a contradiction.
632 const char *WarningMsg =
633 "This expression will create a string whose length is too big to "
634 "be represented as a size_t";
635
636 auto Report =
637 std::make_unique<PathSensitiveBugReport>(*BT_NotCString, WarningMsg, N);
638 C.emitReport(std::move(Report));
639 }
640 }
641
checkAdditionOverflow(CheckerContext & C,ProgramStateRef state,NonLoc left,NonLoc right) const642 ProgramStateRef CStringChecker::checkAdditionOverflow(CheckerContext &C,
643 ProgramStateRef state,
644 NonLoc left,
645 NonLoc right) const {
646 // If out-of-bounds checking is turned off, skip the rest.
647 if (!Filter.CheckCStringOutOfBounds)
648 return state;
649
650 // If a previous check has failed, propagate the failure.
651 if (!state)
652 return nullptr;
653
654 SValBuilder &svalBuilder = C.getSValBuilder();
655 BasicValueFactory &BVF = svalBuilder.getBasicValueFactory();
656
657 QualType sizeTy = svalBuilder.getContext().getSizeType();
658 const llvm::APSInt &maxValInt = BVF.getMaxValue(sizeTy);
659 NonLoc maxVal = svalBuilder.makeIntVal(maxValInt);
660
661 SVal maxMinusRight;
662 if (right.getAs<nonloc::ConcreteInt>()) {
663 maxMinusRight = svalBuilder.evalBinOpNN(state, BO_Sub, maxVal, right,
664 sizeTy);
665 } else {
666 // Try switching the operands. (The order of these two assignments is
667 // important!)
668 maxMinusRight = svalBuilder.evalBinOpNN(state, BO_Sub, maxVal, left,
669 sizeTy);
670 left = right;
671 }
672
673 if (Optional<NonLoc> maxMinusRightNL = maxMinusRight.getAs<NonLoc>()) {
674 QualType cmpTy = svalBuilder.getConditionType();
675 // If left > max - right, we have an overflow.
676 SVal willOverflow = svalBuilder.evalBinOpNN(state, BO_GT, left,
677 *maxMinusRightNL, cmpTy);
678
679 ProgramStateRef stateOverflow, stateOkay;
680 std::tie(stateOverflow, stateOkay) =
681 state->assume(willOverflow.castAs<DefinedOrUnknownSVal>());
682
683 if (stateOverflow && !stateOkay) {
684 // We have an overflow. Emit a bug report.
685 emitAdditionOverflowBug(C, stateOverflow);
686 return nullptr;
687 }
688
689 // From now on, assume an overflow didn't occur.
690 assert(stateOkay);
691 state = stateOkay;
692 }
693
694 return state;
695 }
696
setCStringLength(ProgramStateRef state,const MemRegion * MR,SVal strLength)697 ProgramStateRef CStringChecker::setCStringLength(ProgramStateRef state,
698 const MemRegion *MR,
699 SVal strLength) {
700 assert(!strLength.isUndef() && "Attempt to set an undefined string length");
701
702 MR = MR->StripCasts();
703
704 switch (MR->getKind()) {
705 case MemRegion::StringRegionKind:
706 // FIXME: This can happen if we strcpy() into a string region. This is
707 // undefined [C99 6.4.5p6], but we should still warn about it.
708 return state;
709
710 case MemRegion::SymbolicRegionKind:
711 case MemRegion::AllocaRegionKind:
712 case MemRegion::NonParamVarRegionKind:
713 case MemRegion::ParamVarRegionKind:
714 case MemRegion::FieldRegionKind:
715 case MemRegion::ObjCIvarRegionKind:
716 // These are the types we can currently track string lengths for.
717 break;
718
719 case MemRegion::ElementRegionKind:
720 // FIXME: Handle element regions by upper-bounding the parent region's
721 // string length.
722 return state;
723
724 default:
725 // Other regions (mostly non-data) can't have a reliable C string length.
726 // For now, just ignore the change.
727 // FIXME: These are rare but not impossible. We should output some kind of
728 // warning for things like strcpy((char[]){'a', 0}, "b");
729 return state;
730 }
731
732 if (strLength.isUnknown())
733 return state->remove<CStringLength>(MR);
734
735 return state->set<CStringLength>(MR, strLength);
736 }
737
getCStringLengthForRegion(CheckerContext & C,ProgramStateRef & state,const Expr * Ex,const MemRegion * MR,bool hypothetical)738 SVal CStringChecker::getCStringLengthForRegion(CheckerContext &C,
739 ProgramStateRef &state,
740 const Expr *Ex,
741 const MemRegion *MR,
742 bool hypothetical) {
743 if (!hypothetical) {
744 // If there's a recorded length, go ahead and return it.
745 const SVal *Recorded = state->get<CStringLength>(MR);
746 if (Recorded)
747 return *Recorded;
748 }
749
750 // Otherwise, get a new symbol and update the state.
751 SValBuilder &svalBuilder = C.getSValBuilder();
752 QualType sizeTy = svalBuilder.getContext().getSizeType();
753 SVal strLength = svalBuilder.getMetadataSymbolVal(CStringChecker::getTag(),
754 MR, Ex, sizeTy,
755 C.getLocationContext(),
756 C.blockCount());
757
758 if (!hypothetical) {
759 if (Optional<NonLoc> strLn = strLength.getAs<NonLoc>()) {
760 // In case of unbounded calls strlen etc bound the range to SIZE_MAX/4
761 BasicValueFactory &BVF = svalBuilder.getBasicValueFactory();
762 const llvm::APSInt &maxValInt = BVF.getMaxValue(sizeTy);
763 llvm::APSInt fourInt = APSIntType(maxValInt).getValue(4);
764 const llvm::APSInt *maxLengthInt = BVF.evalAPSInt(BO_Div, maxValInt,
765 fourInt);
766 NonLoc maxLength = svalBuilder.makeIntVal(*maxLengthInt);
767 SVal evalLength = svalBuilder.evalBinOpNN(state, BO_LE, *strLn,
768 maxLength, sizeTy);
769 state = state->assume(evalLength.castAs<DefinedOrUnknownSVal>(), true);
770 }
771 state = state->set<CStringLength>(MR, strLength);
772 }
773
774 return strLength;
775 }
776
getCStringLength(CheckerContext & C,ProgramStateRef & state,const Expr * Ex,SVal Buf,bool hypothetical) const777 SVal CStringChecker::getCStringLength(CheckerContext &C, ProgramStateRef &state,
778 const Expr *Ex, SVal Buf,
779 bool hypothetical) const {
780 const MemRegion *MR = Buf.getAsRegion();
781 if (!MR) {
782 // If we can't get a region, see if it's something we /know/ isn't a
783 // C string. In the context of locations, the only time we can issue such
784 // a warning is for labels.
785 if (Optional<loc::GotoLabel> Label = Buf.getAs<loc::GotoLabel>()) {
786 if (Filter.CheckCStringNotNullTerm) {
787 SmallString<120> buf;
788 llvm::raw_svector_ostream os(buf);
789 assert(CurrentFunctionDescription);
790 os << "Argument to " << CurrentFunctionDescription
791 << " is the address of the label '" << Label->getLabel()->getName()
792 << "', which is not a null-terminated string";
793
794 emitNotCStringBug(C, state, Ex, os.str());
795 }
796 return UndefinedVal();
797 }
798
799 // If it's not a region and not a label, give up.
800 return UnknownVal();
801 }
802
803 // If we have a region, strip casts from it and see if we can figure out
804 // its length. For anything we can't figure out, just return UnknownVal.
805 MR = MR->StripCasts();
806
807 switch (MR->getKind()) {
808 case MemRegion::StringRegionKind: {
809 // Modifying the contents of string regions is undefined [C99 6.4.5p6],
810 // so we can assume that the byte length is the correct C string length.
811 SValBuilder &svalBuilder = C.getSValBuilder();
812 QualType sizeTy = svalBuilder.getContext().getSizeType();
813 const StringLiteral *strLit = cast<StringRegion>(MR)->getStringLiteral();
814 return svalBuilder.makeIntVal(strLit->getByteLength(), sizeTy);
815 }
816 case MemRegion::SymbolicRegionKind:
817 case MemRegion::AllocaRegionKind:
818 case MemRegion::NonParamVarRegionKind:
819 case MemRegion::ParamVarRegionKind:
820 case MemRegion::FieldRegionKind:
821 case MemRegion::ObjCIvarRegionKind:
822 return getCStringLengthForRegion(C, state, Ex, MR, hypothetical);
823 case MemRegion::CompoundLiteralRegionKind:
824 // FIXME: Can we track this? Is it necessary?
825 return UnknownVal();
826 case MemRegion::ElementRegionKind:
827 // FIXME: How can we handle this? It's not good enough to subtract the
828 // offset from the base string length; consider "123\x00567" and &a[5].
829 return UnknownVal();
830 default:
831 // Other regions (mostly non-data) can't have a reliable C string length.
832 // In this case, an error is emitted and UndefinedVal is returned.
833 // The caller should always be prepared to handle this case.
834 if (Filter.CheckCStringNotNullTerm) {
835 SmallString<120> buf;
836 llvm::raw_svector_ostream os(buf);
837
838 assert(CurrentFunctionDescription);
839 os << "Argument to " << CurrentFunctionDescription << " is ";
840
841 if (SummarizeRegion(os, C.getASTContext(), MR))
842 os << ", which is not a null-terminated string";
843 else
844 os << "not a null-terminated string";
845
846 emitNotCStringBug(C, state, Ex, os.str());
847 }
848 return UndefinedVal();
849 }
850 }
851
getCStringLiteral(CheckerContext & C,ProgramStateRef & state,const Expr * expr,SVal val) const852 const StringLiteral *CStringChecker::getCStringLiteral(CheckerContext &C,
853 ProgramStateRef &state, const Expr *expr, SVal val) const {
854
855 // Get the memory region pointed to by the val.
856 const MemRegion *bufRegion = val.getAsRegion();
857 if (!bufRegion)
858 return nullptr;
859
860 // Strip casts off the memory region.
861 bufRegion = bufRegion->StripCasts();
862
863 // Cast the memory region to a string region.
864 const StringRegion *strRegion= dyn_cast<StringRegion>(bufRegion);
865 if (!strRegion)
866 return nullptr;
867
868 // Return the actual string in the string region.
869 return strRegion->getStringLiteral();
870 }
871
IsFirstBufInBound(CheckerContext & C,ProgramStateRef state,const Expr * FirstBuf,const Expr * Size)872 bool CStringChecker::IsFirstBufInBound(CheckerContext &C,
873 ProgramStateRef state,
874 const Expr *FirstBuf,
875 const Expr *Size) {
876 // If we do not know that the buffer is long enough we return 'true'.
877 // Otherwise the parent region of this field region would also get
878 // invalidated, which would lead to warnings based on an unknown state.
879
880 // Originally copied from CheckBufferAccess and CheckLocation.
881 SValBuilder &svalBuilder = C.getSValBuilder();
882 ASTContext &Ctx = svalBuilder.getContext();
883 const LocationContext *LCtx = C.getLocationContext();
884
885 QualType sizeTy = Size->getType();
886 QualType PtrTy = Ctx.getPointerType(Ctx.CharTy);
887 SVal BufVal = state->getSVal(FirstBuf, LCtx);
888
889 SVal LengthVal = state->getSVal(Size, LCtx);
890 Optional<NonLoc> Length = LengthVal.getAs<NonLoc>();
891 if (!Length)
892 return true; // cf top comment.
893
894 // Compute the offset of the last element to be accessed: size-1.
895 NonLoc One = svalBuilder.makeIntVal(1, sizeTy).castAs<NonLoc>();
896 SVal Offset = svalBuilder.evalBinOpNN(state, BO_Sub, *Length, One, sizeTy);
897 if (Offset.isUnknown())
898 return true; // cf top comment
899 NonLoc LastOffset = Offset.castAs<NonLoc>();
900
901 // Check that the first buffer is sufficiently long.
902 SVal BufStart = svalBuilder.evalCast(BufVal, PtrTy, FirstBuf->getType());
903 Optional<Loc> BufLoc = BufStart.getAs<Loc>();
904 if (!BufLoc)
905 return true; // cf top comment.
906
907 SVal BufEnd =
908 svalBuilder.evalBinOpLN(state, BO_Add, *BufLoc, LastOffset, PtrTy);
909
910 // Check for out of bound array element access.
911 const MemRegion *R = BufEnd.getAsRegion();
912 if (!R)
913 return true; // cf top comment.
914
915 const ElementRegion *ER = dyn_cast<ElementRegion>(R);
916 if (!ER)
917 return true; // cf top comment.
918
919 // FIXME: Does this crash when a non-standard definition
920 // of a library function is encountered?
921 assert(ER->getValueType() == C.getASTContext().CharTy &&
922 "IsFirstBufInBound should only be called with char* ElementRegions");
923
924 // Get the size of the array.
925 const SubRegion *superReg = cast<SubRegion>(ER->getSuperRegion());
926 DefinedOrUnknownSVal SizeDV = getDynamicExtent(state, superReg, svalBuilder);
927
928 // Get the index of the accessed element.
929 DefinedOrUnknownSVal Idx = ER->getIndex().castAs<DefinedOrUnknownSVal>();
930
931 ProgramStateRef StInBound = state->assumeInBound(Idx, SizeDV, true);
932
933 return static_cast<bool>(StInBound);
934 }
935
InvalidateBuffer(CheckerContext & C,ProgramStateRef state,const Expr * E,SVal V,bool IsSourceBuffer,const Expr * Size)936 ProgramStateRef CStringChecker::InvalidateBuffer(CheckerContext &C,
937 ProgramStateRef state,
938 const Expr *E, SVal V,
939 bool IsSourceBuffer,
940 const Expr *Size) {
941 Optional<Loc> L = V.getAs<Loc>();
942 if (!L)
943 return state;
944
945 // FIXME: This is a simplified version of what's in CFRefCount.cpp -- it makes
946 // some assumptions about the value that CFRefCount can't. Even so, it should
947 // probably be refactored.
948 if (Optional<loc::MemRegionVal> MR = L->getAs<loc::MemRegionVal>()) {
949 const MemRegion *R = MR->getRegion()->StripCasts();
950
951 // Are we dealing with an ElementRegion? If so, we should be invalidating
952 // the super-region.
953 if (const ElementRegion *ER = dyn_cast<ElementRegion>(R)) {
954 R = ER->getSuperRegion();
955 // FIXME: What about layers of ElementRegions?
956 }
957
958 // Invalidate this region.
959 const LocationContext *LCtx = C.getPredecessor()->getLocationContext();
960
961 bool CausesPointerEscape = false;
962 RegionAndSymbolInvalidationTraits ITraits;
963 // Invalidate and escape only indirect regions accessible through the source
964 // buffer.
965 if (IsSourceBuffer) {
966 ITraits.setTrait(R->getBaseRegion(),
967 RegionAndSymbolInvalidationTraits::TK_PreserveContents);
968 ITraits.setTrait(R, RegionAndSymbolInvalidationTraits::TK_SuppressEscape);
969 CausesPointerEscape = true;
970 } else {
971 const MemRegion::Kind& K = R->getKind();
972 if (K == MemRegion::FieldRegionKind)
973 if (Size && IsFirstBufInBound(C, state, E, Size)) {
974 // If destination buffer is a field region and access is in bound,
975 // do not invalidate its super region.
976 ITraits.setTrait(
977 R,
978 RegionAndSymbolInvalidationTraits::TK_DoNotInvalidateSuperRegion);
979 }
980 }
981
982 return state->invalidateRegions(R, E, C.blockCount(), LCtx,
983 CausesPointerEscape, nullptr, nullptr,
984 &ITraits);
985 }
986
987 // If we have a non-region value by chance, just remove the binding.
988 // FIXME: is this necessary or correct? This handles the non-Region
989 // cases. Is it ever valid to store to these?
990 return state->killBinding(*L);
991 }
992
SummarizeRegion(raw_ostream & os,ASTContext & Ctx,const MemRegion * MR)993 bool CStringChecker::SummarizeRegion(raw_ostream &os, ASTContext &Ctx,
994 const MemRegion *MR) {
995 switch (MR->getKind()) {
996 case MemRegion::FunctionCodeRegionKind: {
997 if (const auto *FD = cast<FunctionCodeRegion>(MR)->getDecl())
998 os << "the address of the function '" << *FD << '\'';
999 else
1000 os << "the address of a function";
1001 return true;
1002 }
1003 case MemRegion::BlockCodeRegionKind:
1004 os << "block text";
1005 return true;
1006 case MemRegion::BlockDataRegionKind:
1007 os << "a block";
1008 return true;
1009 case MemRegion::CXXThisRegionKind:
1010 case MemRegion::CXXTempObjectRegionKind:
1011 os << "a C++ temp object of type "
1012 << cast<TypedValueRegion>(MR)->getValueType().getAsString();
1013 return true;
1014 case MemRegion::NonParamVarRegionKind:
1015 os << "a variable of type"
1016 << cast<TypedValueRegion>(MR)->getValueType().getAsString();
1017 return true;
1018 case MemRegion::ParamVarRegionKind:
1019 os << "a parameter of type"
1020 << cast<TypedValueRegion>(MR)->getValueType().getAsString();
1021 return true;
1022 case MemRegion::FieldRegionKind:
1023 os << "a field of type "
1024 << cast<TypedValueRegion>(MR)->getValueType().getAsString();
1025 return true;
1026 case MemRegion::ObjCIvarRegionKind:
1027 os << "an instance variable of type "
1028 << cast<TypedValueRegion>(MR)->getValueType().getAsString();
1029 return true;
1030 default:
1031 return false;
1032 }
1033 }
1034
memsetAux(const Expr * DstBuffer,SVal CharVal,const Expr * Size,CheckerContext & C,ProgramStateRef & State)1035 bool CStringChecker::memsetAux(const Expr *DstBuffer, SVal CharVal,
1036 const Expr *Size, CheckerContext &C,
1037 ProgramStateRef &State) {
1038 SVal MemVal = C.getSVal(DstBuffer);
1039 SVal SizeVal = C.getSVal(Size);
1040 const MemRegion *MR = MemVal.getAsRegion();
1041 if (!MR)
1042 return false;
1043
1044 // We're about to model memset by producing a "default binding" in the Store.
1045 // Our current implementation - RegionStore - doesn't support default bindings
1046 // that don't cover the whole base region. So we should first get the offset
1047 // and the base region to figure out whether the offset of buffer is 0.
1048 RegionOffset Offset = MR->getAsOffset();
1049 const MemRegion *BR = Offset.getRegion();
1050
1051 Optional<NonLoc> SizeNL = SizeVal.getAs<NonLoc>();
1052 if (!SizeNL)
1053 return false;
1054
1055 SValBuilder &svalBuilder = C.getSValBuilder();
1056 ASTContext &Ctx = C.getASTContext();
1057
1058 // void *memset(void *dest, int ch, size_t count);
1059 // For now we can only handle the case of offset is 0 and concrete char value.
1060 if (Offset.isValid() && !Offset.hasSymbolicOffset() &&
1061 Offset.getOffset() == 0) {
1062 // Get the base region's size.
1063 DefinedOrUnknownSVal SizeDV = getDynamicExtent(State, BR, svalBuilder);
1064
1065 ProgramStateRef StateWholeReg, StateNotWholeReg;
1066 std::tie(StateWholeReg, StateNotWholeReg) =
1067 State->assume(svalBuilder.evalEQ(State, SizeDV, *SizeNL));
1068
1069 // With the semantic of 'memset()', we should convert the CharVal to
1070 // unsigned char.
1071 CharVal = svalBuilder.evalCast(CharVal, Ctx.UnsignedCharTy, Ctx.IntTy);
1072
1073 ProgramStateRef StateNullChar, StateNonNullChar;
1074 std::tie(StateNullChar, StateNonNullChar) =
1075 assumeZero(C, State, CharVal, Ctx.UnsignedCharTy);
1076
1077 if (StateWholeReg && !StateNotWholeReg && StateNullChar &&
1078 !StateNonNullChar) {
1079 // If the 'memset()' acts on the whole region of destination buffer and
1080 // the value of the second argument of 'memset()' is zero, bind the second
1081 // argument's value to the destination buffer with 'default binding'.
1082 // FIXME: Since there is no perfect way to bind the non-zero character, we
1083 // can only deal with zero value here. In the future, we need to deal with
1084 // the binding of non-zero value in the case of whole region.
1085 State = State->bindDefaultZero(svalBuilder.makeLoc(BR),
1086 C.getLocationContext());
1087 } else {
1088 // If the destination buffer's extent is not equal to the value of
1089 // third argument, just invalidate buffer.
1090 State = InvalidateBuffer(C, State, DstBuffer, MemVal,
1091 /*IsSourceBuffer*/ false, Size);
1092 }
1093
1094 if (StateNullChar && !StateNonNullChar) {
1095 // If the value of the second argument of 'memset()' is zero, set the
1096 // string length of destination buffer to 0 directly.
1097 State = setCStringLength(State, MR,
1098 svalBuilder.makeZeroVal(Ctx.getSizeType()));
1099 } else if (!StateNullChar && StateNonNullChar) {
1100 SVal NewStrLen = svalBuilder.getMetadataSymbolVal(
1101 CStringChecker::getTag(), MR, DstBuffer, Ctx.getSizeType(),
1102 C.getLocationContext(), C.blockCount());
1103
1104 // If the value of second argument is not zero, then the string length
1105 // is at least the size argument.
1106 SVal NewStrLenGESize = svalBuilder.evalBinOp(
1107 State, BO_GE, NewStrLen, SizeVal, svalBuilder.getConditionType());
1108
1109 State = setCStringLength(
1110 State->assume(NewStrLenGESize.castAs<DefinedOrUnknownSVal>(), true),
1111 MR, NewStrLen);
1112 }
1113 } else {
1114 // If the offset is not zero and char value is not concrete, we can do
1115 // nothing but invalidate the buffer.
1116 State = InvalidateBuffer(C, State, DstBuffer, MemVal,
1117 /*IsSourceBuffer*/ false, Size);
1118 }
1119 return true;
1120 }
1121
1122 //===----------------------------------------------------------------------===//
1123 // evaluation of individual function calls.
1124 //===----------------------------------------------------------------------===//
1125
evalCopyCommon(CheckerContext & C,const CallExpr * CE,ProgramStateRef state,SizeArgExpr Size,DestinationArgExpr Dest,SourceArgExpr Source,bool Restricted,bool IsMempcpy) const1126 void CStringChecker::evalCopyCommon(CheckerContext &C, const CallExpr *CE,
1127 ProgramStateRef state, SizeArgExpr Size,
1128 DestinationArgExpr Dest,
1129 SourceArgExpr Source, bool Restricted,
1130 bool IsMempcpy) const {
1131 CurrentFunctionDescription = "memory copy function";
1132
1133 // See if the size argument is zero.
1134 const LocationContext *LCtx = C.getLocationContext();
1135 SVal sizeVal = state->getSVal(Size.Expression, LCtx);
1136 QualType sizeTy = Size.Expression->getType();
1137
1138 ProgramStateRef stateZeroSize, stateNonZeroSize;
1139 std::tie(stateZeroSize, stateNonZeroSize) =
1140 assumeZero(C, state, sizeVal, sizeTy);
1141
1142 // Get the value of the Dest.
1143 SVal destVal = state->getSVal(Dest.Expression, LCtx);
1144
1145 // If the size is zero, there won't be any actual memory access, so
1146 // just bind the return value to the destination buffer and return.
1147 if (stateZeroSize && !stateNonZeroSize) {
1148 stateZeroSize = stateZeroSize->BindExpr(CE, LCtx, destVal);
1149 C.addTransition(stateZeroSize);
1150 return;
1151 }
1152
1153 // If the size can be nonzero, we have to check the other arguments.
1154 if (stateNonZeroSize) {
1155 state = stateNonZeroSize;
1156
1157 // Ensure the destination is not null. If it is NULL there will be a
1158 // NULL pointer dereference.
1159 state = checkNonNull(C, state, Dest, destVal);
1160 if (!state)
1161 return;
1162
1163 // Get the value of the Src.
1164 SVal srcVal = state->getSVal(Source.Expression, LCtx);
1165
1166 // Ensure the source is not null. If it is NULL there will be a
1167 // NULL pointer dereference.
1168 state = checkNonNull(C, state, Source, srcVal);
1169 if (!state)
1170 return;
1171
1172 // Ensure the accesses are valid and that the buffers do not overlap.
1173 state = CheckBufferAccess(C, state, Dest, Size, AccessKind::write);
1174 state = CheckBufferAccess(C, state, Source, Size, AccessKind::read);
1175
1176 if (Restricted)
1177 state = CheckOverlap(C, state, Size, Dest, Source);
1178
1179 if (!state)
1180 return;
1181
1182 // If this is mempcpy, get the byte after the last byte copied and
1183 // bind the expr.
1184 if (IsMempcpy) {
1185 // Get the byte after the last byte copied.
1186 SValBuilder &SvalBuilder = C.getSValBuilder();
1187 ASTContext &Ctx = SvalBuilder.getContext();
1188 QualType CharPtrTy = Ctx.getPointerType(Ctx.CharTy);
1189 SVal DestRegCharVal =
1190 SvalBuilder.evalCast(destVal, CharPtrTy, Dest.Expression->getType());
1191 SVal lastElement = C.getSValBuilder().evalBinOp(
1192 state, BO_Add, DestRegCharVal, sizeVal, Dest.Expression->getType());
1193 // If we don't know how much we copied, we can at least
1194 // conjure a return value for later.
1195 if (lastElement.isUnknown())
1196 lastElement = C.getSValBuilder().conjureSymbolVal(nullptr, CE, LCtx,
1197 C.blockCount());
1198
1199 // The byte after the last byte copied is the return value.
1200 state = state->BindExpr(CE, LCtx, lastElement);
1201 } else {
1202 // All other copies return the destination buffer.
1203 // (Well, bcopy() has a void return type, but this won't hurt.)
1204 state = state->BindExpr(CE, LCtx, destVal);
1205 }
1206
1207 // Invalidate the destination (regular invalidation without pointer-escaping
1208 // the address of the top-level region).
1209 // FIXME: Even if we can't perfectly model the copy, we should see if we
1210 // can use LazyCompoundVals to copy the source values into the destination.
1211 // This would probably remove any existing bindings past the end of the
1212 // copied region, but that's still an improvement over blank invalidation.
1213 state =
1214 InvalidateBuffer(C, state, Dest.Expression, C.getSVal(Dest.Expression),
1215 /*IsSourceBuffer*/ false, Size.Expression);
1216
1217 // Invalidate the source (const-invalidation without const-pointer-escaping
1218 // the address of the top-level region).
1219 state = InvalidateBuffer(C, state, Source.Expression,
1220 C.getSVal(Source.Expression),
1221 /*IsSourceBuffer*/ true, nullptr);
1222
1223 C.addTransition(state);
1224 }
1225 }
1226
evalMemcpy(CheckerContext & C,const CallExpr * CE) const1227 void CStringChecker::evalMemcpy(CheckerContext &C, const CallExpr *CE) const {
1228 // void *memcpy(void *restrict dst, const void *restrict src, size_t n);
1229 // The return value is the address of the destination buffer.
1230 DestinationArgExpr Dest = {CE->getArg(0), 0};
1231 SourceArgExpr Src = {CE->getArg(1), 1};
1232 SizeArgExpr Size = {CE->getArg(2), 2};
1233
1234 ProgramStateRef State = C.getState();
1235
1236 constexpr bool IsRestricted = true;
1237 constexpr bool IsMempcpy = false;
1238 evalCopyCommon(C, CE, State, Size, Dest, Src, IsRestricted, IsMempcpy);
1239 }
1240
evalMempcpy(CheckerContext & C,const CallExpr * CE) const1241 void CStringChecker::evalMempcpy(CheckerContext &C, const CallExpr *CE) const {
1242 // void *mempcpy(void *restrict dst, const void *restrict src, size_t n);
1243 // The return value is a pointer to the byte following the last written byte.
1244 DestinationArgExpr Dest = {CE->getArg(0), 0};
1245 SourceArgExpr Src = {CE->getArg(1), 1};
1246 SizeArgExpr Size = {CE->getArg(2), 2};
1247
1248 constexpr bool IsRestricted = true;
1249 constexpr bool IsMempcpy = true;
1250 evalCopyCommon(C, CE, C.getState(), Size, Dest, Src, IsRestricted, IsMempcpy);
1251 }
1252
evalMemmove(CheckerContext & C,const CallExpr * CE) const1253 void CStringChecker::evalMemmove(CheckerContext &C, const CallExpr *CE) const {
1254 // void *memmove(void *dst, const void *src, size_t n);
1255 // The return value is the address of the destination buffer.
1256 DestinationArgExpr Dest = {CE->getArg(0), 0};
1257 SourceArgExpr Src = {CE->getArg(1), 1};
1258 SizeArgExpr Size = {CE->getArg(2), 2};
1259
1260 constexpr bool IsRestricted = false;
1261 constexpr bool IsMempcpy = false;
1262 evalCopyCommon(C, CE, C.getState(), Size, Dest, Src, IsRestricted, IsMempcpy);
1263 }
1264
evalBcopy(CheckerContext & C,const CallExpr * CE) const1265 void CStringChecker::evalBcopy(CheckerContext &C, const CallExpr *CE) const {
1266 // void bcopy(const void *src, void *dst, size_t n);
1267 SourceArgExpr Src(CE->getArg(0), 0);
1268 DestinationArgExpr Dest = {CE->getArg(1), 1};
1269 SizeArgExpr Size = {CE->getArg(2), 2};
1270
1271 constexpr bool IsRestricted = false;
1272 constexpr bool IsMempcpy = false;
1273 evalCopyCommon(C, CE, C.getState(), Size, Dest, Src, IsRestricted, IsMempcpy);
1274 }
1275
evalMemcmp(CheckerContext & C,const CallExpr * CE) const1276 void CStringChecker::evalMemcmp(CheckerContext &C, const CallExpr *CE) const {
1277 // int memcmp(const void *s1, const void *s2, size_t n);
1278 CurrentFunctionDescription = "memory comparison function";
1279
1280 AnyArgExpr Left = {CE->getArg(0), 0};
1281 AnyArgExpr Right = {CE->getArg(1), 1};
1282 SizeArgExpr Size = {CE->getArg(2), 2};
1283
1284 ProgramStateRef State = C.getState();
1285 SValBuilder &Builder = C.getSValBuilder();
1286 const LocationContext *LCtx = C.getLocationContext();
1287
1288 // See if the size argument is zero.
1289 SVal sizeVal = State->getSVal(Size.Expression, LCtx);
1290 QualType sizeTy = Size.Expression->getType();
1291
1292 ProgramStateRef stateZeroSize, stateNonZeroSize;
1293 std::tie(stateZeroSize, stateNonZeroSize) =
1294 assumeZero(C, State, sizeVal, sizeTy);
1295
1296 // If the size can be zero, the result will be 0 in that case, and we don't
1297 // have to check either of the buffers.
1298 if (stateZeroSize) {
1299 State = stateZeroSize;
1300 State = State->BindExpr(CE, LCtx, Builder.makeZeroVal(CE->getType()));
1301 C.addTransition(State);
1302 }
1303
1304 // If the size can be nonzero, we have to check the other arguments.
1305 if (stateNonZeroSize) {
1306 State = stateNonZeroSize;
1307 // If we know the two buffers are the same, we know the result is 0.
1308 // First, get the two buffers' addresses. Another checker will have already
1309 // made sure they're not undefined.
1310 DefinedOrUnknownSVal LV =
1311 State->getSVal(Left.Expression, LCtx).castAs<DefinedOrUnknownSVal>();
1312 DefinedOrUnknownSVal RV =
1313 State->getSVal(Right.Expression, LCtx).castAs<DefinedOrUnknownSVal>();
1314
1315 // See if they are the same.
1316 ProgramStateRef SameBuffer, NotSameBuffer;
1317 std::tie(SameBuffer, NotSameBuffer) =
1318 State->assume(Builder.evalEQ(State, LV, RV));
1319
1320 // If the two arguments are the same buffer, we know the result is 0,
1321 // and we only need to check one size.
1322 if (SameBuffer && !NotSameBuffer) {
1323 State = SameBuffer;
1324 State = CheckBufferAccess(C, State, Left, Size, AccessKind::read);
1325 if (State) {
1326 State =
1327 SameBuffer->BindExpr(CE, LCtx, Builder.makeZeroVal(CE->getType()));
1328 C.addTransition(State);
1329 }
1330 return;
1331 }
1332
1333 // If the two arguments might be different buffers, we have to check
1334 // the size of both of them.
1335 assert(NotSameBuffer);
1336 State = CheckBufferAccess(C, State, Right, Size, AccessKind::read);
1337 State = CheckBufferAccess(C, State, Left, Size, AccessKind::read);
1338 if (State) {
1339 // The return value is the comparison result, which we don't know.
1340 SVal CmpV = Builder.conjureSymbolVal(nullptr, CE, LCtx, C.blockCount());
1341 State = State->BindExpr(CE, LCtx, CmpV);
1342 C.addTransition(State);
1343 }
1344 }
1345 }
1346
evalstrLength(CheckerContext & C,const CallExpr * CE) const1347 void CStringChecker::evalstrLength(CheckerContext &C,
1348 const CallExpr *CE) const {
1349 // size_t strlen(const char *s);
1350 evalstrLengthCommon(C, CE, /* IsStrnlen = */ false);
1351 }
1352
evalstrnLength(CheckerContext & C,const CallExpr * CE) const1353 void CStringChecker::evalstrnLength(CheckerContext &C,
1354 const CallExpr *CE) const {
1355 // size_t strnlen(const char *s, size_t maxlen);
1356 evalstrLengthCommon(C, CE, /* IsStrnlen = */ true);
1357 }
1358
evalstrLengthCommon(CheckerContext & C,const CallExpr * CE,bool IsStrnlen) const1359 void CStringChecker::evalstrLengthCommon(CheckerContext &C, const CallExpr *CE,
1360 bool IsStrnlen) const {
1361 CurrentFunctionDescription = "string length function";
1362 ProgramStateRef state = C.getState();
1363 const LocationContext *LCtx = C.getLocationContext();
1364
1365 if (IsStrnlen) {
1366 const Expr *maxlenExpr = CE->getArg(1);
1367 SVal maxlenVal = state->getSVal(maxlenExpr, LCtx);
1368
1369 ProgramStateRef stateZeroSize, stateNonZeroSize;
1370 std::tie(stateZeroSize, stateNonZeroSize) =
1371 assumeZero(C, state, maxlenVal, maxlenExpr->getType());
1372
1373 // If the size can be zero, the result will be 0 in that case, and we don't
1374 // have to check the string itself.
1375 if (stateZeroSize) {
1376 SVal zero = C.getSValBuilder().makeZeroVal(CE->getType());
1377 stateZeroSize = stateZeroSize->BindExpr(CE, LCtx, zero);
1378 C.addTransition(stateZeroSize);
1379 }
1380
1381 // If the size is GUARANTEED to be zero, we're done!
1382 if (!stateNonZeroSize)
1383 return;
1384
1385 // Otherwise, record the assumption that the size is nonzero.
1386 state = stateNonZeroSize;
1387 }
1388
1389 // Check that the string argument is non-null.
1390 AnyArgExpr Arg = {CE->getArg(0), 0};
1391 SVal ArgVal = state->getSVal(Arg.Expression, LCtx);
1392 state = checkNonNull(C, state, Arg, ArgVal);
1393
1394 if (!state)
1395 return;
1396
1397 SVal strLength = getCStringLength(C, state, Arg.Expression, ArgVal);
1398
1399 // If the argument isn't a valid C string, there's no valid state to
1400 // transition to.
1401 if (strLength.isUndef())
1402 return;
1403
1404 DefinedOrUnknownSVal result = UnknownVal();
1405
1406 // If the check is for strnlen() then bind the return value to no more than
1407 // the maxlen value.
1408 if (IsStrnlen) {
1409 QualType cmpTy = C.getSValBuilder().getConditionType();
1410
1411 // It's a little unfortunate to be getting this again,
1412 // but it's not that expensive...
1413 const Expr *maxlenExpr = CE->getArg(1);
1414 SVal maxlenVal = state->getSVal(maxlenExpr, LCtx);
1415
1416 Optional<NonLoc> strLengthNL = strLength.getAs<NonLoc>();
1417 Optional<NonLoc> maxlenValNL = maxlenVal.getAs<NonLoc>();
1418
1419 if (strLengthNL && maxlenValNL) {
1420 ProgramStateRef stateStringTooLong, stateStringNotTooLong;
1421
1422 // Check if the strLength is greater than the maxlen.
1423 std::tie(stateStringTooLong, stateStringNotTooLong) = state->assume(
1424 C.getSValBuilder()
1425 .evalBinOpNN(state, BO_GT, *strLengthNL, *maxlenValNL, cmpTy)
1426 .castAs<DefinedOrUnknownSVal>());
1427
1428 if (stateStringTooLong && !stateStringNotTooLong) {
1429 // If the string is longer than maxlen, return maxlen.
1430 result = *maxlenValNL;
1431 } else if (stateStringNotTooLong && !stateStringTooLong) {
1432 // If the string is shorter than maxlen, return its length.
1433 result = *strLengthNL;
1434 }
1435 }
1436
1437 if (result.isUnknown()) {
1438 // If we don't have enough information for a comparison, there's
1439 // no guarantee the full string length will actually be returned.
1440 // All we know is the return value is the min of the string length
1441 // and the limit. This is better than nothing.
1442 result = C.getSValBuilder().conjureSymbolVal(nullptr, CE, LCtx,
1443 C.blockCount());
1444 NonLoc resultNL = result.castAs<NonLoc>();
1445
1446 if (strLengthNL) {
1447 state = state->assume(C.getSValBuilder().evalBinOpNN(
1448 state, BO_LE, resultNL, *strLengthNL, cmpTy)
1449 .castAs<DefinedOrUnknownSVal>(), true);
1450 }
1451
1452 if (maxlenValNL) {
1453 state = state->assume(C.getSValBuilder().evalBinOpNN(
1454 state, BO_LE, resultNL, *maxlenValNL, cmpTy)
1455 .castAs<DefinedOrUnknownSVal>(), true);
1456 }
1457 }
1458
1459 } else {
1460 // This is a plain strlen(), not strnlen().
1461 result = strLength.castAs<DefinedOrUnknownSVal>();
1462
1463 // If we don't know the length of the string, conjure a return
1464 // value, so it can be used in constraints, at least.
1465 if (result.isUnknown()) {
1466 result = C.getSValBuilder().conjureSymbolVal(nullptr, CE, LCtx,
1467 C.blockCount());
1468 }
1469 }
1470
1471 // Bind the return value.
1472 assert(!result.isUnknown() && "Should have conjured a value by now");
1473 state = state->BindExpr(CE, LCtx, result);
1474 C.addTransition(state);
1475 }
1476
evalStrcpy(CheckerContext & C,const CallExpr * CE) const1477 void CStringChecker::evalStrcpy(CheckerContext &C, const CallExpr *CE) const {
1478 // char *strcpy(char *restrict dst, const char *restrict src);
1479 evalStrcpyCommon(C, CE,
1480 /* ReturnEnd = */ false,
1481 /* IsBounded = */ false,
1482 /* appendK = */ ConcatFnKind::none);
1483 }
1484
evalStrncpy(CheckerContext & C,const CallExpr * CE) const1485 void CStringChecker::evalStrncpy(CheckerContext &C, const CallExpr *CE) const {
1486 // char *strncpy(char *restrict dst, const char *restrict src, size_t n);
1487 evalStrcpyCommon(C, CE,
1488 /* ReturnEnd = */ false,
1489 /* IsBounded = */ true,
1490 /* appendK = */ ConcatFnKind::none);
1491 }
1492
evalStpcpy(CheckerContext & C,const CallExpr * CE) const1493 void CStringChecker::evalStpcpy(CheckerContext &C, const CallExpr *CE) const {
1494 // char *stpcpy(char *restrict dst, const char *restrict src);
1495 evalStrcpyCommon(C, CE,
1496 /* ReturnEnd = */ true,
1497 /* IsBounded = */ false,
1498 /* appendK = */ ConcatFnKind::none);
1499 }
1500
evalStrlcpy(CheckerContext & C,const CallExpr * CE) const1501 void CStringChecker::evalStrlcpy(CheckerContext &C, const CallExpr *CE) const {
1502 // size_t strlcpy(char *dest, const char *src, size_t size);
1503 evalStrcpyCommon(C, CE,
1504 /* ReturnEnd = */ true,
1505 /* IsBounded = */ true,
1506 /* appendK = */ ConcatFnKind::none,
1507 /* returnPtr = */ false);
1508 }
1509
evalStrcat(CheckerContext & C,const CallExpr * CE) const1510 void CStringChecker::evalStrcat(CheckerContext &C, const CallExpr *CE) const {
1511 // char *strcat(char *restrict s1, const char *restrict s2);
1512 evalStrcpyCommon(C, CE,
1513 /* ReturnEnd = */ false,
1514 /* IsBounded = */ false,
1515 /* appendK = */ ConcatFnKind::strcat);
1516 }
1517
evalStrncat(CheckerContext & C,const CallExpr * CE) const1518 void CStringChecker::evalStrncat(CheckerContext &C, const CallExpr *CE) const {
1519 //char *strncat(char *restrict s1, const char *restrict s2, size_t n);
1520 evalStrcpyCommon(C, CE,
1521 /* ReturnEnd = */ false,
1522 /* IsBounded = */ true,
1523 /* appendK = */ ConcatFnKind::strcat);
1524 }
1525
evalStrlcat(CheckerContext & C,const CallExpr * CE) const1526 void CStringChecker::evalStrlcat(CheckerContext &C, const CallExpr *CE) const {
1527 // size_t strlcat(char *dst, const char *src, size_t size);
1528 // It will append at most size - strlen(dst) - 1 bytes,
1529 // NULL-terminating the result.
1530 evalStrcpyCommon(C, CE,
1531 /* ReturnEnd = */ false,
1532 /* IsBounded = */ true,
1533 /* appendK = */ ConcatFnKind::strlcat,
1534 /* returnPtr = */ false);
1535 }
1536
evalStrcpyCommon(CheckerContext & C,const CallExpr * CE,bool ReturnEnd,bool IsBounded,ConcatFnKind appendK,bool returnPtr) const1537 void CStringChecker::evalStrcpyCommon(CheckerContext &C, const CallExpr *CE,
1538 bool ReturnEnd, bool IsBounded,
1539 ConcatFnKind appendK,
1540 bool returnPtr) const {
1541 if (appendK == ConcatFnKind::none)
1542 CurrentFunctionDescription = "string copy function";
1543 else
1544 CurrentFunctionDescription = "string concatenation function";
1545
1546 ProgramStateRef state = C.getState();
1547 const LocationContext *LCtx = C.getLocationContext();
1548
1549 // Check that the destination is non-null.
1550 DestinationArgExpr Dst = {CE->getArg(0), 0};
1551 SVal DstVal = state->getSVal(Dst.Expression, LCtx);
1552 state = checkNonNull(C, state, Dst, DstVal);
1553 if (!state)
1554 return;
1555
1556 // Check that the source is non-null.
1557 SourceArgExpr srcExpr = {CE->getArg(1), 1};
1558 SVal srcVal = state->getSVal(srcExpr.Expression, LCtx);
1559 state = checkNonNull(C, state, srcExpr, srcVal);
1560 if (!state)
1561 return;
1562
1563 // Get the string length of the source.
1564 SVal strLength = getCStringLength(C, state, srcExpr.Expression, srcVal);
1565 Optional<NonLoc> strLengthNL = strLength.getAs<NonLoc>();
1566
1567 // Get the string length of the destination buffer.
1568 SVal dstStrLength = getCStringLength(C, state, Dst.Expression, DstVal);
1569 Optional<NonLoc> dstStrLengthNL = dstStrLength.getAs<NonLoc>();
1570
1571 // If the source isn't a valid C string, give up.
1572 if (strLength.isUndef())
1573 return;
1574
1575 SValBuilder &svalBuilder = C.getSValBuilder();
1576 QualType cmpTy = svalBuilder.getConditionType();
1577 QualType sizeTy = svalBuilder.getContext().getSizeType();
1578
1579 // These two values allow checking two kinds of errors:
1580 // - actual overflows caused by a source that doesn't fit in the destination
1581 // - potential overflows caused by a bound that could exceed the destination
1582 SVal amountCopied = UnknownVal();
1583 SVal maxLastElementIndex = UnknownVal();
1584 const char *boundWarning = nullptr;
1585
1586 // FIXME: Why do we choose the srcExpr if the access has no size?
1587 // Note that the 3rd argument of the call would be the size parameter.
1588 SizeArgExpr SrcExprAsSizeDummy = {srcExpr.Expression, srcExpr.ArgumentIndex};
1589 state = CheckOverlap(
1590 C, state,
1591 (IsBounded ? SizeArgExpr{CE->getArg(2), 2} : SrcExprAsSizeDummy), Dst,
1592 srcExpr);
1593
1594 if (!state)
1595 return;
1596
1597 // If the function is strncpy, strncat, etc... it is bounded.
1598 if (IsBounded) {
1599 // Get the max number of characters to copy.
1600 SizeArgExpr lenExpr = {CE->getArg(2), 2};
1601 SVal lenVal = state->getSVal(lenExpr.Expression, LCtx);
1602
1603 // Protect against misdeclared strncpy().
1604 lenVal =
1605 svalBuilder.evalCast(lenVal, sizeTy, lenExpr.Expression->getType());
1606
1607 Optional<NonLoc> lenValNL = lenVal.getAs<NonLoc>();
1608
1609 // If we know both values, we might be able to figure out how much
1610 // we're copying.
1611 if (strLengthNL && lenValNL) {
1612 switch (appendK) {
1613 case ConcatFnKind::none:
1614 case ConcatFnKind::strcat: {
1615 ProgramStateRef stateSourceTooLong, stateSourceNotTooLong;
1616 // Check if the max number to copy is less than the length of the src.
1617 // If the bound is equal to the source length, strncpy won't null-
1618 // terminate the result!
1619 std::tie(stateSourceTooLong, stateSourceNotTooLong) = state->assume(
1620 svalBuilder
1621 .evalBinOpNN(state, BO_GE, *strLengthNL, *lenValNL, cmpTy)
1622 .castAs<DefinedOrUnknownSVal>());
1623
1624 if (stateSourceTooLong && !stateSourceNotTooLong) {
1625 // Max number to copy is less than the length of the src, so the
1626 // actual strLength copied is the max number arg.
1627 state = stateSourceTooLong;
1628 amountCopied = lenVal;
1629
1630 } else if (!stateSourceTooLong && stateSourceNotTooLong) {
1631 // The source buffer entirely fits in the bound.
1632 state = stateSourceNotTooLong;
1633 amountCopied = strLength;
1634 }
1635 break;
1636 }
1637 case ConcatFnKind::strlcat:
1638 if (!dstStrLengthNL)
1639 return;
1640
1641 // amountCopied = min (size - dstLen - 1 , srcLen)
1642 SVal freeSpace = svalBuilder.evalBinOpNN(state, BO_Sub, *lenValNL,
1643 *dstStrLengthNL, sizeTy);
1644 if (!freeSpace.getAs<NonLoc>())
1645 return;
1646 freeSpace =
1647 svalBuilder.evalBinOp(state, BO_Sub, freeSpace,
1648 svalBuilder.makeIntVal(1, sizeTy), sizeTy);
1649 Optional<NonLoc> freeSpaceNL = freeSpace.getAs<NonLoc>();
1650
1651 // While unlikely, it is possible that the subtraction is
1652 // too complex to compute, let's check whether it succeeded.
1653 if (!freeSpaceNL)
1654 return;
1655 SVal hasEnoughSpace = svalBuilder.evalBinOpNN(
1656 state, BO_LE, *strLengthNL, *freeSpaceNL, cmpTy);
1657
1658 ProgramStateRef TrueState, FalseState;
1659 std::tie(TrueState, FalseState) =
1660 state->assume(hasEnoughSpace.castAs<DefinedOrUnknownSVal>());
1661
1662 // srcStrLength <= size - dstStrLength -1
1663 if (TrueState && !FalseState) {
1664 amountCopied = strLength;
1665 }
1666
1667 // srcStrLength > size - dstStrLength -1
1668 if (!TrueState && FalseState) {
1669 amountCopied = freeSpace;
1670 }
1671
1672 if (TrueState && FalseState)
1673 amountCopied = UnknownVal();
1674 break;
1675 }
1676 }
1677 // We still want to know if the bound is known to be too large.
1678 if (lenValNL) {
1679 switch (appendK) {
1680 case ConcatFnKind::strcat:
1681 // For strncat, the check is strlen(dst) + lenVal < sizeof(dst)
1682
1683 // Get the string length of the destination. If the destination is
1684 // memory that can't have a string length, we shouldn't be copying
1685 // into it anyway.
1686 if (dstStrLength.isUndef())
1687 return;
1688
1689 if (dstStrLengthNL) {
1690 maxLastElementIndex = svalBuilder.evalBinOpNN(
1691 state, BO_Add, *lenValNL, *dstStrLengthNL, sizeTy);
1692
1693 boundWarning = "Size argument is greater than the free space in the "
1694 "destination buffer";
1695 }
1696 break;
1697 case ConcatFnKind::none:
1698 case ConcatFnKind::strlcat:
1699 // For strncpy and strlcat, this is just checking
1700 // that lenVal <= sizeof(dst).
1701 // (Yes, strncpy and strncat differ in how they treat termination.
1702 // strncat ALWAYS terminates, but strncpy doesn't.)
1703
1704 // We need a special case for when the copy size is zero, in which
1705 // case strncpy will do no work at all. Our bounds check uses n-1
1706 // as the last element accessed, so n == 0 is problematic.
1707 ProgramStateRef StateZeroSize, StateNonZeroSize;
1708 std::tie(StateZeroSize, StateNonZeroSize) =
1709 assumeZero(C, state, *lenValNL, sizeTy);
1710
1711 // If the size is known to be zero, we're done.
1712 if (StateZeroSize && !StateNonZeroSize) {
1713 if (returnPtr) {
1714 StateZeroSize = StateZeroSize->BindExpr(CE, LCtx, DstVal);
1715 } else {
1716 if (appendK == ConcatFnKind::none) {
1717 // strlcpy returns strlen(src)
1718 StateZeroSize = StateZeroSize->BindExpr(CE, LCtx, strLength);
1719 } else {
1720 // strlcat returns strlen(src) + strlen(dst)
1721 SVal retSize = svalBuilder.evalBinOp(
1722 state, BO_Add, strLength, dstStrLength, sizeTy);
1723 StateZeroSize = StateZeroSize->BindExpr(CE, LCtx, retSize);
1724 }
1725 }
1726 C.addTransition(StateZeroSize);
1727 return;
1728 }
1729
1730 // Otherwise, go ahead and figure out the last element we'll touch.
1731 // We don't record the non-zero assumption here because we can't
1732 // be sure. We won't warn on a possible zero.
1733 NonLoc one = svalBuilder.makeIntVal(1, sizeTy).castAs<NonLoc>();
1734 maxLastElementIndex =
1735 svalBuilder.evalBinOpNN(state, BO_Sub, *lenValNL, one, sizeTy);
1736 boundWarning = "Size argument is greater than the length of the "
1737 "destination buffer";
1738 break;
1739 }
1740 }
1741 } else {
1742 // The function isn't bounded. The amount copied should match the length
1743 // of the source buffer.
1744 amountCopied = strLength;
1745 }
1746
1747 assert(state);
1748
1749 // This represents the number of characters copied into the destination
1750 // buffer. (It may not actually be the strlen if the destination buffer
1751 // is not terminated.)
1752 SVal finalStrLength = UnknownVal();
1753 SVal strlRetVal = UnknownVal();
1754
1755 if (appendK == ConcatFnKind::none && !returnPtr) {
1756 // strlcpy returns the sizeof(src)
1757 strlRetVal = strLength;
1758 }
1759
1760 // If this is an appending function (strcat, strncat...) then set the
1761 // string length to strlen(src) + strlen(dst) since the buffer will
1762 // ultimately contain both.
1763 if (appendK != ConcatFnKind::none) {
1764 // Get the string length of the destination. If the destination is memory
1765 // that can't have a string length, we shouldn't be copying into it anyway.
1766 if (dstStrLength.isUndef())
1767 return;
1768
1769 if (appendK == ConcatFnKind::strlcat && dstStrLengthNL && strLengthNL) {
1770 strlRetVal = svalBuilder.evalBinOpNN(state, BO_Add, *strLengthNL,
1771 *dstStrLengthNL, sizeTy);
1772 }
1773
1774 Optional<NonLoc> amountCopiedNL = amountCopied.getAs<NonLoc>();
1775
1776 // If we know both string lengths, we might know the final string length.
1777 if (amountCopiedNL && dstStrLengthNL) {
1778 // Make sure the two lengths together don't overflow a size_t.
1779 state = checkAdditionOverflow(C, state, *amountCopiedNL, *dstStrLengthNL);
1780 if (!state)
1781 return;
1782
1783 finalStrLength = svalBuilder.evalBinOpNN(state, BO_Add, *amountCopiedNL,
1784 *dstStrLengthNL, sizeTy);
1785 }
1786
1787 // If we couldn't get a single value for the final string length,
1788 // we can at least bound it by the individual lengths.
1789 if (finalStrLength.isUnknown()) {
1790 // Try to get a "hypothetical" string length symbol, which we can later
1791 // set as a real value if that turns out to be the case.
1792 finalStrLength = getCStringLength(C, state, CE, DstVal, true);
1793 assert(!finalStrLength.isUndef());
1794
1795 if (Optional<NonLoc> finalStrLengthNL = finalStrLength.getAs<NonLoc>()) {
1796 if (amountCopiedNL && appendK == ConcatFnKind::none) {
1797 // we overwrite dst string with the src
1798 // finalStrLength >= srcStrLength
1799 SVal sourceInResult = svalBuilder.evalBinOpNN(
1800 state, BO_GE, *finalStrLengthNL, *amountCopiedNL, cmpTy);
1801 state = state->assume(sourceInResult.castAs<DefinedOrUnknownSVal>(),
1802 true);
1803 if (!state)
1804 return;
1805 }
1806
1807 if (dstStrLengthNL && appendK != ConcatFnKind::none) {
1808 // we extend the dst string with the src
1809 // finalStrLength >= dstStrLength
1810 SVal destInResult = svalBuilder.evalBinOpNN(state, BO_GE,
1811 *finalStrLengthNL,
1812 *dstStrLengthNL,
1813 cmpTy);
1814 state =
1815 state->assume(destInResult.castAs<DefinedOrUnknownSVal>(), true);
1816 if (!state)
1817 return;
1818 }
1819 }
1820 }
1821
1822 } else {
1823 // Otherwise, this is a copy-over function (strcpy, strncpy, ...), and
1824 // the final string length will match the input string length.
1825 finalStrLength = amountCopied;
1826 }
1827
1828 SVal Result;
1829
1830 if (returnPtr) {
1831 // The final result of the function will either be a pointer past the last
1832 // copied element, or a pointer to the start of the destination buffer.
1833 Result = (ReturnEnd ? UnknownVal() : DstVal);
1834 } else {
1835 if (appendK == ConcatFnKind::strlcat || appendK == ConcatFnKind::none)
1836 //strlcpy, strlcat
1837 Result = strlRetVal;
1838 else
1839 Result = finalStrLength;
1840 }
1841
1842 assert(state);
1843
1844 // If the destination is a MemRegion, try to check for a buffer overflow and
1845 // record the new string length.
1846 if (Optional<loc::MemRegionVal> dstRegVal =
1847 DstVal.getAs<loc::MemRegionVal>()) {
1848 QualType ptrTy = Dst.Expression->getType();
1849
1850 // If we have an exact value on a bounded copy, use that to check for
1851 // overflows, rather than our estimate about how much is actually copied.
1852 if (Optional<NonLoc> maxLastNL = maxLastElementIndex.getAs<NonLoc>()) {
1853 SVal maxLastElement =
1854 svalBuilder.evalBinOpLN(state, BO_Add, *dstRegVal, *maxLastNL, ptrTy);
1855
1856 state = CheckLocation(C, state, Dst, maxLastElement, AccessKind::write);
1857 if (!state)
1858 return;
1859 }
1860
1861 // Then, if the final length is known...
1862 if (Optional<NonLoc> knownStrLength = finalStrLength.getAs<NonLoc>()) {
1863 SVal lastElement = svalBuilder.evalBinOpLN(state, BO_Add, *dstRegVal,
1864 *knownStrLength, ptrTy);
1865
1866 // ...and we haven't checked the bound, we'll check the actual copy.
1867 if (!boundWarning) {
1868 state = CheckLocation(C, state, Dst, lastElement, AccessKind::write);
1869 if (!state)
1870 return;
1871 }
1872
1873 // If this is a stpcpy-style copy, the last element is the return value.
1874 if (returnPtr && ReturnEnd)
1875 Result = lastElement;
1876 }
1877
1878 // Invalidate the destination (regular invalidation without pointer-escaping
1879 // the address of the top-level region). This must happen before we set the
1880 // C string length because invalidation will clear the length.
1881 // FIXME: Even if we can't perfectly model the copy, we should see if we
1882 // can use LazyCompoundVals to copy the source values into the destination.
1883 // This would probably remove any existing bindings past the end of the
1884 // string, but that's still an improvement over blank invalidation.
1885 state = InvalidateBuffer(C, state, Dst.Expression, *dstRegVal,
1886 /*IsSourceBuffer*/ false, nullptr);
1887
1888 // Invalidate the source (const-invalidation without const-pointer-escaping
1889 // the address of the top-level region).
1890 state = InvalidateBuffer(C, state, srcExpr.Expression, srcVal,
1891 /*IsSourceBuffer*/ true, nullptr);
1892
1893 // Set the C string length of the destination, if we know it.
1894 if (IsBounded && (appendK == ConcatFnKind::none)) {
1895 // strncpy is annoying in that it doesn't guarantee to null-terminate
1896 // the result string. If the original string didn't fit entirely inside
1897 // the bound (including the null-terminator), we don't know how long the
1898 // result is.
1899 if (amountCopied != strLength)
1900 finalStrLength = UnknownVal();
1901 }
1902 state = setCStringLength(state, dstRegVal->getRegion(), finalStrLength);
1903 }
1904
1905 assert(state);
1906
1907 if (returnPtr) {
1908 // If this is a stpcpy-style copy, but we were unable to check for a buffer
1909 // overflow, we still need a result. Conjure a return value.
1910 if (ReturnEnd && Result.isUnknown()) {
1911 Result = svalBuilder.conjureSymbolVal(nullptr, CE, LCtx, C.blockCount());
1912 }
1913 }
1914 // Set the return value.
1915 state = state->BindExpr(CE, LCtx, Result);
1916 C.addTransition(state);
1917 }
1918
evalStrcmp(CheckerContext & C,const CallExpr * CE) const1919 void CStringChecker::evalStrcmp(CheckerContext &C, const CallExpr *CE) const {
1920 //int strcmp(const char *s1, const char *s2);
1921 evalStrcmpCommon(C, CE, /* IsBounded = */ false, /* IgnoreCase = */ false);
1922 }
1923
evalStrncmp(CheckerContext & C,const CallExpr * CE) const1924 void CStringChecker::evalStrncmp(CheckerContext &C, const CallExpr *CE) const {
1925 //int strncmp(const char *s1, const char *s2, size_t n);
1926 evalStrcmpCommon(C, CE, /* IsBounded = */ true, /* IgnoreCase = */ false);
1927 }
1928
evalStrcasecmp(CheckerContext & C,const CallExpr * CE) const1929 void CStringChecker::evalStrcasecmp(CheckerContext &C,
1930 const CallExpr *CE) const {
1931 //int strcasecmp(const char *s1, const char *s2);
1932 evalStrcmpCommon(C, CE, /* IsBounded = */ false, /* IgnoreCase = */ true);
1933 }
1934
evalStrncasecmp(CheckerContext & C,const CallExpr * CE) const1935 void CStringChecker::evalStrncasecmp(CheckerContext &C,
1936 const CallExpr *CE) const {
1937 //int strncasecmp(const char *s1, const char *s2, size_t n);
1938 evalStrcmpCommon(C, CE, /* IsBounded = */ true, /* IgnoreCase = */ true);
1939 }
1940
evalStrcmpCommon(CheckerContext & C,const CallExpr * CE,bool IsBounded,bool IgnoreCase) const1941 void CStringChecker::evalStrcmpCommon(CheckerContext &C, const CallExpr *CE,
1942 bool IsBounded, bool IgnoreCase) const {
1943 CurrentFunctionDescription = "string comparison function";
1944 ProgramStateRef state = C.getState();
1945 const LocationContext *LCtx = C.getLocationContext();
1946
1947 // Check that the first string is non-null
1948 AnyArgExpr Left = {CE->getArg(0), 0};
1949 SVal LeftVal = state->getSVal(Left.Expression, LCtx);
1950 state = checkNonNull(C, state, Left, LeftVal);
1951 if (!state)
1952 return;
1953
1954 // Check that the second string is non-null.
1955 AnyArgExpr Right = {CE->getArg(1), 1};
1956 SVal RightVal = state->getSVal(Right.Expression, LCtx);
1957 state = checkNonNull(C, state, Right, RightVal);
1958 if (!state)
1959 return;
1960
1961 // Get the string length of the first string or give up.
1962 SVal LeftLength = getCStringLength(C, state, Left.Expression, LeftVal);
1963 if (LeftLength.isUndef())
1964 return;
1965
1966 // Get the string length of the second string or give up.
1967 SVal RightLength = getCStringLength(C, state, Right.Expression, RightVal);
1968 if (RightLength.isUndef())
1969 return;
1970
1971 // If we know the two buffers are the same, we know the result is 0.
1972 // First, get the two buffers' addresses. Another checker will have already
1973 // made sure they're not undefined.
1974 DefinedOrUnknownSVal LV = LeftVal.castAs<DefinedOrUnknownSVal>();
1975 DefinedOrUnknownSVal RV = RightVal.castAs<DefinedOrUnknownSVal>();
1976
1977 // See if they are the same.
1978 SValBuilder &svalBuilder = C.getSValBuilder();
1979 DefinedOrUnknownSVal SameBuf = svalBuilder.evalEQ(state, LV, RV);
1980 ProgramStateRef StSameBuf, StNotSameBuf;
1981 std::tie(StSameBuf, StNotSameBuf) = state->assume(SameBuf);
1982
1983 // If the two arguments might be the same buffer, we know the result is 0,
1984 // and we only need to check one size.
1985 if (StSameBuf) {
1986 StSameBuf = StSameBuf->BindExpr(CE, LCtx,
1987 svalBuilder.makeZeroVal(CE->getType()));
1988 C.addTransition(StSameBuf);
1989
1990 // If the two arguments are GUARANTEED to be the same, we're done!
1991 if (!StNotSameBuf)
1992 return;
1993 }
1994
1995 assert(StNotSameBuf);
1996 state = StNotSameBuf;
1997
1998 // At this point we can go about comparing the two buffers.
1999 // For now, we only do this if they're both known string literals.
2000
2001 // Attempt to extract string literals from both expressions.
2002 const StringLiteral *LeftStrLiteral =
2003 getCStringLiteral(C, state, Left.Expression, LeftVal);
2004 const StringLiteral *RightStrLiteral =
2005 getCStringLiteral(C, state, Right.Expression, RightVal);
2006 bool canComputeResult = false;
2007 SVal resultVal = svalBuilder.conjureSymbolVal(nullptr, CE, LCtx,
2008 C.blockCount());
2009
2010 if (LeftStrLiteral && RightStrLiteral) {
2011 StringRef LeftStrRef = LeftStrLiteral->getString();
2012 StringRef RightStrRef = RightStrLiteral->getString();
2013
2014 if (IsBounded) {
2015 // Get the max number of characters to compare.
2016 const Expr *lenExpr = CE->getArg(2);
2017 SVal lenVal = state->getSVal(lenExpr, LCtx);
2018
2019 // If the length is known, we can get the right substrings.
2020 if (const llvm::APSInt *len = svalBuilder.getKnownValue(state, lenVal)) {
2021 // Create substrings of each to compare the prefix.
2022 LeftStrRef = LeftStrRef.substr(0, (size_t)len->getZExtValue());
2023 RightStrRef = RightStrRef.substr(0, (size_t)len->getZExtValue());
2024 canComputeResult = true;
2025 }
2026 } else {
2027 // This is a normal, unbounded strcmp.
2028 canComputeResult = true;
2029 }
2030
2031 if (canComputeResult) {
2032 // Real strcmp stops at null characters.
2033 size_t s1Term = LeftStrRef.find('\0');
2034 if (s1Term != StringRef::npos)
2035 LeftStrRef = LeftStrRef.substr(0, s1Term);
2036
2037 size_t s2Term = RightStrRef.find('\0');
2038 if (s2Term != StringRef::npos)
2039 RightStrRef = RightStrRef.substr(0, s2Term);
2040
2041 // Use StringRef's comparison methods to compute the actual result.
2042 int compareRes = IgnoreCase ? LeftStrRef.compare_insensitive(RightStrRef)
2043 : LeftStrRef.compare(RightStrRef);
2044
2045 // The strcmp function returns an integer greater than, equal to, or less
2046 // than zero, [c11, p7.24.4.2].
2047 if (compareRes == 0) {
2048 resultVal = svalBuilder.makeIntVal(compareRes, CE->getType());
2049 }
2050 else {
2051 DefinedSVal zeroVal = svalBuilder.makeIntVal(0, CE->getType());
2052 // Constrain strcmp's result range based on the result of StringRef's
2053 // comparison methods.
2054 BinaryOperatorKind op = (compareRes == 1) ? BO_GT : BO_LT;
2055 SVal compareWithZero =
2056 svalBuilder.evalBinOp(state, op, resultVal, zeroVal,
2057 svalBuilder.getConditionType());
2058 DefinedSVal compareWithZeroVal = compareWithZero.castAs<DefinedSVal>();
2059 state = state->assume(compareWithZeroVal, true);
2060 }
2061 }
2062 }
2063
2064 state = state->BindExpr(CE, LCtx, resultVal);
2065
2066 // Record this as a possible path.
2067 C.addTransition(state);
2068 }
2069
evalStrsep(CheckerContext & C,const CallExpr * CE) const2070 void CStringChecker::evalStrsep(CheckerContext &C, const CallExpr *CE) const {
2071 //char *strsep(char **stringp, const char *delim);
2072 // Sanity: does the search string parameter match the return type?
2073 SourceArgExpr SearchStrPtr = {CE->getArg(0), 0};
2074
2075 QualType CharPtrTy = SearchStrPtr.Expression->getType()->getPointeeType();
2076 if (CharPtrTy.isNull() ||
2077 CE->getType().getUnqualifiedType() != CharPtrTy.getUnqualifiedType())
2078 return;
2079
2080 CurrentFunctionDescription = "strsep()";
2081 ProgramStateRef State = C.getState();
2082 const LocationContext *LCtx = C.getLocationContext();
2083
2084 // Check that the search string pointer is non-null (though it may point to
2085 // a null string).
2086 SVal SearchStrVal = State->getSVal(SearchStrPtr.Expression, LCtx);
2087 State = checkNonNull(C, State, SearchStrPtr, SearchStrVal);
2088 if (!State)
2089 return;
2090
2091 // Check that the delimiter string is non-null.
2092 AnyArgExpr DelimStr = {CE->getArg(1), 1};
2093 SVal DelimStrVal = State->getSVal(DelimStr.Expression, LCtx);
2094 State = checkNonNull(C, State, DelimStr, DelimStrVal);
2095 if (!State)
2096 return;
2097
2098 SValBuilder &SVB = C.getSValBuilder();
2099 SVal Result;
2100 if (Optional<Loc> SearchStrLoc = SearchStrVal.getAs<Loc>()) {
2101 // Get the current value of the search string pointer, as a char*.
2102 Result = State->getSVal(*SearchStrLoc, CharPtrTy);
2103
2104 // Invalidate the search string, representing the change of one delimiter
2105 // character to NUL.
2106 State = InvalidateBuffer(C, State, SearchStrPtr.Expression, Result,
2107 /*IsSourceBuffer*/ false, nullptr);
2108
2109 // Overwrite the search string pointer. The new value is either an address
2110 // further along in the same string, or NULL if there are no more tokens.
2111 State = State->bindLoc(*SearchStrLoc,
2112 SVB.conjureSymbolVal(getTag(),
2113 CE,
2114 LCtx,
2115 CharPtrTy,
2116 C.blockCount()),
2117 LCtx);
2118 } else {
2119 assert(SearchStrVal.isUnknown());
2120 // Conjure a symbolic value. It's the best we can do.
2121 Result = SVB.conjureSymbolVal(nullptr, CE, LCtx, C.blockCount());
2122 }
2123
2124 // Set the return value, and finish.
2125 State = State->BindExpr(CE, LCtx, Result);
2126 C.addTransition(State);
2127 }
2128
2129 // These should probably be moved into a C++ standard library checker.
evalStdCopy(CheckerContext & C,const CallExpr * CE) const2130 void CStringChecker::evalStdCopy(CheckerContext &C, const CallExpr *CE) const {
2131 evalStdCopyCommon(C, CE);
2132 }
2133
evalStdCopyBackward(CheckerContext & C,const CallExpr * CE) const2134 void CStringChecker::evalStdCopyBackward(CheckerContext &C,
2135 const CallExpr *CE) const {
2136 evalStdCopyCommon(C, CE);
2137 }
2138
evalStdCopyCommon(CheckerContext & C,const CallExpr * CE) const2139 void CStringChecker::evalStdCopyCommon(CheckerContext &C,
2140 const CallExpr *CE) const {
2141 if (!CE->getArg(2)->getType()->isPointerType())
2142 return;
2143
2144 ProgramStateRef State = C.getState();
2145
2146 const LocationContext *LCtx = C.getLocationContext();
2147
2148 // template <class _InputIterator, class _OutputIterator>
2149 // _OutputIterator
2150 // copy(_InputIterator __first, _InputIterator __last,
2151 // _OutputIterator __result)
2152
2153 // Invalidate the destination buffer
2154 const Expr *Dst = CE->getArg(2);
2155 SVal DstVal = State->getSVal(Dst, LCtx);
2156 State = InvalidateBuffer(C, State, Dst, DstVal, /*IsSource=*/false,
2157 /*Size=*/nullptr);
2158
2159 SValBuilder &SVB = C.getSValBuilder();
2160
2161 SVal ResultVal = SVB.conjureSymbolVal(nullptr, CE, LCtx, C.blockCount());
2162 State = State->BindExpr(CE, LCtx, ResultVal);
2163
2164 C.addTransition(State);
2165 }
2166
evalMemset(CheckerContext & C,const CallExpr * CE) const2167 void CStringChecker::evalMemset(CheckerContext &C, const CallExpr *CE) const {
2168 // void *memset(void *s, int c, size_t n);
2169 CurrentFunctionDescription = "memory set function";
2170
2171 DestinationArgExpr Buffer = {CE->getArg(0), 0};
2172 AnyArgExpr CharE = {CE->getArg(1), 1};
2173 SizeArgExpr Size = {CE->getArg(2), 2};
2174
2175 ProgramStateRef State = C.getState();
2176
2177 // See if the size argument is zero.
2178 const LocationContext *LCtx = C.getLocationContext();
2179 SVal SizeVal = C.getSVal(Size.Expression);
2180 QualType SizeTy = Size.Expression->getType();
2181
2182 ProgramStateRef ZeroSize, NonZeroSize;
2183 std::tie(ZeroSize, NonZeroSize) = assumeZero(C, State, SizeVal, SizeTy);
2184
2185 // Get the value of the memory area.
2186 SVal BufferPtrVal = C.getSVal(Buffer.Expression);
2187
2188 // If the size is zero, there won't be any actual memory access, so
2189 // just bind the return value to the buffer and return.
2190 if (ZeroSize && !NonZeroSize) {
2191 ZeroSize = ZeroSize->BindExpr(CE, LCtx, BufferPtrVal);
2192 C.addTransition(ZeroSize);
2193 return;
2194 }
2195
2196 // Ensure the memory area is not null.
2197 // If it is NULL there will be a NULL pointer dereference.
2198 State = checkNonNull(C, NonZeroSize, Buffer, BufferPtrVal);
2199 if (!State)
2200 return;
2201
2202 State = CheckBufferAccess(C, State, Buffer, Size, AccessKind::write);
2203 if (!State)
2204 return;
2205
2206 // According to the values of the arguments, bind the value of the second
2207 // argument to the destination buffer and set string length, or just
2208 // invalidate the destination buffer.
2209 if (!memsetAux(Buffer.Expression, C.getSVal(CharE.Expression),
2210 Size.Expression, C, State))
2211 return;
2212
2213 State = State->BindExpr(CE, LCtx, BufferPtrVal);
2214 C.addTransition(State);
2215 }
2216
evalBzero(CheckerContext & C,const CallExpr * CE) const2217 void CStringChecker::evalBzero(CheckerContext &C, const CallExpr *CE) const {
2218 CurrentFunctionDescription = "memory clearance function";
2219
2220 DestinationArgExpr Buffer = {CE->getArg(0), 0};
2221 SizeArgExpr Size = {CE->getArg(1), 1};
2222 SVal Zero = C.getSValBuilder().makeZeroVal(C.getASTContext().IntTy);
2223
2224 ProgramStateRef State = C.getState();
2225
2226 // See if the size argument is zero.
2227 SVal SizeVal = C.getSVal(Size.Expression);
2228 QualType SizeTy = Size.Expression->getType();
2229
2230 ProgramStateRef StateZeroSize, StateNonZeroSize;
2231 std::tie(StateZeroSize, StateNonZeroSize) =
2232 assumeZero(C, State, SizeVal, SizeTy);
2233
2234 // If the size is zero, there won't be any actual memory access,
2235 // In this case we just return.
2236 if (StateZeroSize && !StateNonZeroSize) {
2237 C.addTransition(StateZeroSize);
2238 return;
2239 }
2240
2241 // Get the value of the memory area.
2242 SVal MemVal = C.getSVal(Buffer.Expression);
2243
2244 // Ensure the memory area is not null.
2245 // If it is NULL there will be a NULL pointer dereference.
2246 State = checkNonNull(C, StateNonZeroSize, Buffer, MemVal);
2247 if (!State)
2248 return;
2249
2250 State = CheckBufferAccess(C, State, Buffer, Size, AccessKind::write);
2251 if (!State)
2252 return;
2253
2254 if (!memsetAux(Buffer.Expression, Zero, Size.Expression, C, State))
2255 return;
2256
2257 C.addTransition(State);
2258 }
2259
2260 //===----------------------------------------------------------------------===//
2261 // The driver method, and other Checker callbacks.
2262 //===----------------------------------------------------------------------===//
2263
identifyCall(const CallEvent & Call,CheckerContext & C) const2264 CStringChecker::FnCheck CStringChecker::identifyCall(const CallEvent &Call,
2265 CheckerContext &C) const {
2266 const auto *CE = dyn_cast_or_null<CallExpr>(Call.getOriginExpr());
2267 if (!CE)
2268 return nullptr;
2269
2270 const FunctionDecl *FD = dyn_cast_or_null<FunctionDecl>(Call.getDecl());
2271 if (!FD)
2272 return nullptr;
2273
2274 if (Call.isCalled(StdCopy)) {
2275 return &CStringChecker::evalStdCopy;
2276 } else if (Call.isCalled(StdCopyBackward)) {
2277 return &CStringChecker::evalStdCopyBackward;
2278 }
2279
2280 // Pro-actively check that argument types are safe to do arithmetic upon.
2281 // We do not want to crash if someone accidentally passes a structure
2282 // into, say, a C++ overload of any of these functions. We could not check
2283 // that for std::copy because they may have arguments of other types.
2284 for (auto I : CE->arguments()) {
2285 QualType T = I->getType();
2286 if (!T->isIntegralOrEnumerationType() && !T->isPointerType())
2287 return nullptr;
2288 }
2289
2290 const FnCheck *Callback = Callbacks.lookup(Call);
2291 if (Callback)
2292 return *Callback;
2293
2294 return nullptr;
2295 }
2296
evalCall(const CallEvent & Call,CheckerContext & C) const2297 bool CStringChecker::evalCall(const CallEvent &Call, CheckerContext &C) const {
2298 FnCheck Callback = identifyCall(Call, C);
2299
2300 // If the callee isn't a string function, let another checker handle it.
2301 if (!Callback)
2302 return false;
2303
2304 // Check and evaluate the call.
2305 const auto *CE = cast<CallExpr>(Call.getOriginExpr());
2306 (this->*Callback)(C, CE);
2307
2308 // If the evaluate call resulted in no change, chain to the next eval call
2309 // handler.
2310 // Note, the custom CString evaluation calls assume that basic safety
2311 // properties are held. However, if the user chooses to turn off some of these
2312 // checks, we ignore the issues and leave the call evaluation to a generic
2313 // handler.
2314 return C.isDifferent();
2315 }
2316
checkPreStmt(const DeclStmt * DS,CheckerContext & C) const2317 void CStringChecker::checkPreStmt(const DeclStmt *DS, CheckerContext &C) const {
2318 // Record string length for char a[] = "abc";
2319 ProgramStateRef state = C.getState();
2320
2321 for (const auto *I : DS->decls()) {
2322 const VarDecl *D = dyn_cast<VarDecl>(I);
2323 if (!D)
2324 continue;
2325
2326 // FIXME: Handle array fields of structs.
2327 if (!D->getType()->isArrayType())
2328 continue;
2329
2330 const Expr *Init = D->getInit();
2331 if (!Init)
2332 continue;
2333 if (!isa<StringLiteral>(Init))
2334 continue;
2335
2336 Loc VarLoc = state->getLValue(D, C.getLocationContext());
2337 const MemRegion *MR = VarLoc.getAsRegion();
2338 if (!MR)
2339 continue;
2340
2341 SVal StrVal = C.getSVal(Init);
2342 assert(StrVal.isValid() && "Initializer string is unknown or undefined");
2343 DefinedOrUnknownSVal strLength =
2344 getCStringLength(C, state, Init, StrVal).castAs<DefinedOrUnknownSVal>();
2345
2346 state = state->set<CStringLength>(MR, strLength);
2347 }
2348
2349 C.addTransition(state);
2350 }
2351
2352 ProgramStateRef
checkRegionChanges(ProgramStateRef state,const InvalidatedSymbols *,ArrayRef<const MemRegion * > ExplicitRegions,ArrayRef<const MemRegion * > Regions,const LocationContext * LCtx,const CallEvent * Call) const2353 CStringChecker::checkRegionChanges(ProgramStateRef state,
2354 const InvalidatedSymbols *,
2355 ArrayRef<const MemRegion *> ExplicitRegions,
2356 ArrayRef<const MemRegion *> Regions,
2357 const LocationContext *LCtx,
2358 const CallEvent *Call) const {
2359 CStringLengthTy Entries = state->get<CStringLength>();
2360 if (Entries.isEmpty())
2361 return state;
2362
2363 llvm::SmallPtrSet<const MemRegion *, 8> Invalidated;
2364 llvm::SmallPtrSet<const MemRegion *, 32> SuperRegions;
2365
2366 // First build sets for the changed regions and their super-regions.
2367 for (ArrayRef<const MemRegion *>::iterator
2368 I = Regions.begin(), E = Regions.end(); I != E; ++I) {
2369 const MemRegion *MR = *I;
2370 Invalidated.insert(MR);
2371
2372 SuperRegions.insert(MR);
2373 while (const SubRegion *SR = dyn_cast<SubRegion>(MR)) {
2374 MR = SR->getSuperRegion();
2375 SuperRegions.insert(MR);
2376 }
2377 }
2378
2379 CStringLengthTy::Factory &F = state->get_context<CStringLength>();
2380
2381 // Then loop over the entries in the current state.
2382 for (CStringLengthTy::iterator I = Entries.begin(),
2383 E = Entries.end(); I != E; ++I) {
2384 const MemRegion *MR = I.getKey();
2385
2386 // Is this entry for a super-region of a changed region?
2387 if (SuperRegions.count(MR)) {
2388 Entries = F.remove(Entries, MR);
2389 continue;
2390 }
2391
2392 // Is this entry for a sub-region of a changed region?
2393 const MemRegion *Super = MR;
2394 while (const SubRegion *SR = dyn_cast<SubRegion>(Super)) {
2395 Super = SR->getSuperRegion();
2396 if (Invalidated.count(Super)) {
2397 Entries = F.remove(Entries, MR);
2398 break;
2399 }
2400 }
2401 }
2402
2403 return state->set<CStringLength>(Entries);
2404 }
2405
checkLiveSymbols(ProgramStateRef state,SymbolReaper & SR) const2406 void CStringChecker::checkLiveSymbols(ProgramStateRef state,
2407 SymbolReaper &SR) const {
2408 // Mark all symbols in our string length map as valid.
2409 CStringLengthTy Entries = state->get<CStringLength>();
2410
2411 for (CStringLengthTy::iterator I = Entries.begin(), E = Entries.end();
2412 I != E; ++I) {
2413 SVal Len = I.getData();
2414
2415 for (SymExpr::symbol_iterator si = Len.symbol_begin(),
2416 se = Len.symbol_end(); si != se; ++si)
2417 SR.markInUse(*si);
2418 }
2419 }
2420
checkDeadSymbols(SymbolReaper & SR,CheckerContext & C) const2421 void CStringChecker::checkDeadSymbols(SymbolReaper &SR,
2422 CheckerContext &C) const {
2423 ProgramStateRef state = C.getState();
2424 CStringLengthTy Entries = state->get<CStringLength>();
2425 if (Entries.isEmpty())
2426 return;
2427
2428 CStringLengthTy::Factory &F = state->get_context<CStringLength>();
2429 for (CStringLengthTy::iterator I = Entries.begin(), E = Entries.end();
2430 I != E; ++I) {
2431 SVal Len = I.getData();
2432 if (SymbolRef Sym = Len.getAsSymbol()) {
2433 if (SR.isDead(Sym))
2434 Entries = F.remove(Entries, I.getKey());
2435 }
2436 }
2437
2438 state = state->set<CStringLength>(Entries);
2439 C.addTransition(state);
2440 }
2441
registerCStringModeling(CheckerManager & Mgr)2442 void ento::registerCStringModeling(CheckerManager &Mgr) {
2443 Mgr.registerChecker<CStringChecker>();
2444 }
2445
shouldRegisterCStringModeling(const CheckerManager & mgr)2446 bool ento::shouldRegisterCStringModeling(const CheckerManager &mgr) {
2447 return true;
2448 }
2449
2450 #define REGISTER_CHECKER(name) \
2451 void ento::register##name(CheckerManager &mgr) { \
2452 CStringChecker *checker = mgr.getChecker<CStringChecker>(); \
2453 checker->Filter.Check##name = true; \
2454 checker->Filter.CheckName##name = mgr.getCurrentCheckerName(); \
2455 } \
2456 \
2457 bool ento::shouldRegister##name(const CheckerManager &mgr) { return true; }
2458
2459 REGISTER_CHECKER(CStringNullArg)
2460 REGISTER_CHECKER(CStringOutOfBounds)
2461 REGISTER_CHECKER(CStringBufferOverlap)
2462 REGISTER_CHECKER(CStringNotNullTerm)
2463