1 //= CStringChecker.cpp - Checks calls to C string functions --------*- C++ -*-// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 // 10 // This defines CStringChecker, which is an assortment of checks on calls 11 // to functions in <string.h>. 12 // 13 //===----------------------------------------------------------------------===// 14 15 #include "ClangSACheckers.h" 16 #include "InterCheckerAPI.h" 17 #include "clang/Basic/CharInfo.h" 18 #include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" 19 #include "clang/StaticAnalyzer/Core/Checker.h" 20 #include "clang/StaticAnalyzer/Core/CheckerManager.h" 21 #include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" 22 #include "clang/StaticAnalyzer/Core/PathSensitive/ProgramStateTrait.h" 23 #include "llvm/ADT/STLExtras.h" 24 #include "llvm/ADT/SmallString.h" 25 #include "llvm/Support/raw_ostream.h" 26 27 using namespace clang; 28 using namespace ento; 29 30 namespace { 31 class CStringChecker : public Checker< eval::Call, 32 check::PreStmt<DeclStmt>, 33 check::LiveSymbols, 34 check::DeadSymbols, 35 check::RegionChanges 36 > { 37 mutable std::unique_ptr<BugType> BT_Null, BT_Bounds, BT_Overlap, 38 BT_NotCString, BT_AdditionOverflow; 39 40 mutable const char *CurrentFunctionDescription; 41 42 public: 43 /// The filter is used to filter out the diagnostics which are not enabled by 44 /// the user. 45 struct CStringChecksFilter { 46 DefaultBool CheckCStringNullArg; 47 DefaultBool CheckCStringOutOfBounds; 48 DefaultBool CheckCStringBufferOverlap; 49 DefaultBool CheckCStringNotNullTerm; 50 51 CheckName CheckNameCStringNullArg; 52 CheckName CheckNameCStringOutOfBounds; 53 CheckName CheckNameCStringBufferOverlap; 54 CheckName CheckNameCStringNotNullTerm; 55 }; 56 57 CStringChecksFilter Filter; 58 59 static void *getTag() { static int tag; return &tag; } 60 61 bool evalCall(const CallExpr *CE, CheckerContext &C) const; 62 void checkPreStmt(const DeclStmt *DS, CheckerContext &C) const; 63 void checkLiveSymbols(ProgramStateRef state, SymbolReaper &SR) const; 64 void checkDeadSymbols(SymbolReaper &SR, CheckerContext &C) const; 65 66 ProgramStateRef 67 checkRegionChanges(ProgramStateRef state, 68 const InvalidatedSymbols *, 69 ArrayRef<const MemRegion *> ExplicitRegions, 70 ArrayRef<const MemRegion *> Regions, 71 const LocationContext *LCtx, 72 const CallEvent *Call) const; 73 74 typedef void (CStringChecker::*FnCheck)(CheckerContext &, 75 const CallExpr *) const; 76 77 void evalMemcpy(CheckerContext &C, const CallExpr *CE) const; 78 void evalMempcpy(CheckerContext &C, const CallExpr *CE) const; 79 void evalMemmove(CheckerContext &C, const CallExpr *CE) const; 80 void evalBcopy(CheckerContext &C, const CallExpr *CE) const; 81 void evalCopyCommon(CheckerContext &C, const CallExpr *CE, 82 ProgramStateRef state, 83 const Expr *Size, 84 const Expr *Source, 85 const Expr *Dest, 86 bool Restricted = false, 87 bool IsMempcpy = false) const; 88 89 void evalMemcmp(CheckerContext &C, const CallExpr *CE) const; 90 91 void evalstrLength(CheckerContext &C, const CallExpr *CE) const; 92 void evalstrnLength(CheckerContext &C, const CallExpr *CE) const; 93 void evalstrLengthCommon(CheckerContext &C, 94 const CallExpr *CE, 95 bool IsStrnlen = false) const; 96 97 void evalStrcpy(CheckerContext &C, const CallExpr *CE) const; 98 void evalStrncpy(CheckerContext &C, const CallExpr *CE) const; 99 void evalStpcpy(CheckerContext &C, const CallExpr *CE) const; 100 void evalStrlcpy(CheckerContext &C, const CallExpr *CE) const; 101 void evalStrcpyCommon(CheckerContext &C, 102 const CallExpr *CE, 103 bool returnEnd, 104 bool isBounded, 105 bool isAppending, 106 bool canOverlap = false) const; 107 108 void evalStrcat(CheckerContext &C, const CallExpr *CE) const; 109 void evalStrncat(CheckerContext &C, const CallExpr *CE) const; 110 void evalStrlcat(CheckerContext &C, const CallExpr *CE) const; 111 112 void evalStrcmp(CheckerContext &C, const CallExpr *CE) const; 113 void evalStrncmp(CheckerContext &C, const CallExpr *CE) const; 114 void evalStrcasecmp(CheckerContext &C, const CallExpr *CE) const; 115 void evalStrncasecmp(CheckerContext &C, const CallExpr *CE) const; 116 void evalStrcmpCommon(CheckerContext &C, 117 const CallExpr *CE, 118 bool isBounded = false, 119 bool ignoreCase = false) const; 120 121 void evalStrsep(CheckerContext &C, const CallExpr *CE) const; 122 123 void evalStdCopy(CheckerContext &C, const CallExpr *CE) const; 124 void evalStdCopyBackward(CheckerContext &C, const CallExpr *CE) const; 125 void evalStdCopyCommon(CheckerContext &C, const CallExpr *CE) const; 126 void evalMemset(CheckerContext &C, const CallExpr *CE) const; 127 128 // Utility methods 129 std::pair<ProgramStateRef , ProgramStateRef > 130 static assumeZero(CheckerContext &C, 131 ProgramStateRef state, SVal V, QualType Ty); 132 133 static ProgramStateRef setCStringLength(ProgramStateRef state, 134 const MemRegion *MR, 135 SVal strLength); 136 static SVal getCStringLengthForRegion(CheckerContext &C, 137 ProgramStateRef &state, 138 const Expr *Ex, 139 const MemRegion *MR, 140 bool hypothetical); 141 SVal getCStringLength(CheckerContext &C, 142 ProgramStateRef &state, 143 const Expr *Ex, 144 SVal Buf, 145 bool hypothetical = false) const; 146 147 const StringLiteral *getCStringLiteral(CheckerContext &C, 148 ProgramStateRef &state, 149 const Expr *expr, 150 SVal val) const; 151 152 static ProgramStateRef InvalidateBuffer(CheckerContext &C, 153 ProgramStateRef state, 154 const Expr *Ex, SVal V, 155 bool IsSourceBuffer, 156 const Expr *Size); 157 158 static bool SummarizeRegion(raw_ostream &os, ASTContext &Ctx, 159 const MemRegion *MR); 160 161 // Re-usable checks 162 ProgramStateRef checkNonNull(CheckerContext &C, 163 ProgramStateRef state, 164 const Expr *S, 165 SVal l) const; 166 ProgramStateRef CheckLocation(CheckerContext &C, 167 ProgramStateRef state, 168 const Expr *S, 169 SVal l, 170 const char *message = nullptr) const; 171 ProgramStateRef CheckBufferAccess(CheckerContext &C, 172 ProgramStateRef state, 173 const Expr *Size, 174 const Expr *FirstBuf, 175 const Expr *SecondBuf, 176 const char *firstMessage = nullptr, 177 const char *secondMessage = nullptr, 178 bool WarnAboutSize = false) const; 179 180 ProgramStateRef CheckBufferAccess(CheckerContext &C, 181 ProgramStateRef state, 182 const Expr *Size, 183 const Expr *Buf, 184 const char *message = nullptr, 185 bool WarnAboutSize = false) const { 186 // This is a convenience override. 187 return CheckBufferAccess(C, state, Size, Buf, nullptr, message, nullptr, 188 WarnAboutSize); 189 } 190 ProgramStateRef CheckOverlap(CheckerContext &C, 191 ProgramStateRef state, 192 const Expr *Size, 193 const Expr *First, 194 const Expr *Second) const; 195 void emitOverlapBug(CheckerContext &C, 196 ProgramStateRef state, 197 const Stmt *First, 198 const Stmt *Second) const; 199 200 void emitNullArgBug(CheckerContext &C, ProgramStateRef State, const Stmt *S, 201 StringRef WarningMsg) const; 202 void emitOutOfBoundsBug(CheckerContext &C, ProgramStateRef State, 203 const Stmt *S, StringRef WarningMsg) const; 204 void emitNotCStringBug(CheckerContext &C, ProgramStateRef State, 205 const Stmt *S, StringRef WarningMsg) const; 206 void emitAdditionOverflowBug(CheckerContext &C, ProgramStateRef State) const; 207 208 ProgramStateRef checkAdditionOverflow(CheckerContext &C, 209 ProgramStateRef state, 210 NonLoc left, 211 NonLoc right) const; 212 213 // Return true if the destination buffer of the copy function may be in bound. 214 // Expects SVal of Size to be positive and unsigned. 215 // Expects SVal of FirstBuf to be a FieldRegion. 216 static bool IsFirstBufInBound(CheckerContext &C, 217 ProgramStateRef state, 218 const Expr *FirstBuf, 219 const Expr *Size); 220 }; 221 222 } //end anonymous namespace 223 224 REGISTER_MAP_WITH_PROGRAMSTATE(CStringLength, const MemRegion *, SVal) 225 226 //===----------------------------------------------------------------------===// 227 // Individual checks and utility methods. 228 //===----------------------------------------------------------------------===// 229 230 std::pair<ProgramStateRef , ProgramStateRef > 231 CStringChecker::assumeZero(CheckerContext &C, ProgramStateRef state, SVal V, 232 QualType Ty) { 233 Optional<DefinedSVal> val = V.getAs<DefinedSVal>(); 234 if (!val) 235 return std::pair<ProgramStateRef , ProgramStateRef >(state, state); 236 237 SValBuilder &svalBuilder = C.getSValBuilder(); 238 DefinedOrUnknownSVal zero = svalBuilder.makeZeroVal(Ty); 239 return state->assume(svalBuilder.evalEQ(state, *val, zero)); 240 } 241 242 ProgramStateRef CStringChecker::checkNonNull(CheckerContext &C, 243 ProgramStateRef state, 244 const Expr *S, SVal l) const { 245 // If a previous check has failed, propagate the failure. 246 if (!state) 247 return nullptr; 248 249 ProgramStateRef stateNull, stateNonNull; 250 std::tie(stateNull, stateNonNull) = assumeZero(C, state, l, S->getType()); 251 252 if (stateNull && !stateNonNull) { 253 if (Filter.CheckCStringNullArg) { 254 SmallString<80> buf; 255 llvm::raw_svector_ostream os(buf); 256 assert(CurrentFunctionDescription); 257 os << "Null pointer argument in call to " << CurrentFunctionDescription; 258 259 emitNullArgBug(C, stateNull, S, os.str()); 260 } 261 return nullptr; 262 } 263 264 // From here on, assume that the value is non-null. 265 assert(stateNonNull); 266 return stateNonNull; 267 } 268 269 // FIXME: This was originally copied from ArrayBoundChecker.cpp. Refactor? 270 ProgramStateRef CStringChecker::CheckLocation(CheckerContext &C, 271 ProgramStateRef state, 272 const Expr *S, SVal l, 273 const char *warningMsg) const { 274 // If a previous check has failed, propagate the failure. 275 if (!state) 276 return nullptr; 277 278 // Check for out of bound array element access. 279 const MemRegion *R = l.getAsRegion(); 280 if (!R) 281 return state; 282 283 const ElementRegion *ER = dyn_cast<ElementRegion>(R); 284 if (!ER) 285 return state; 286 287 if (ER->getValueType() != C.getASTContext().CharTy) 288 return state; 289 290 // Get the size of the array. 291 const SubRegion *superReg = cast<SubRegion>(ER->getSuperRegion()); 292 SValBuilder &svalBuilder = C.getSValBuilder(); 293 SVal Extent = 294 svalBuilder.convertToArrayIndex(superReg->getExtent(svalBuilder)); 295 DefinedOrUnknownSVal Size = Extent.castAs<DefinedOrUnknownSVal>(); 296 297 // Get the index of the accessed element. 298 DefinedOrUnknownSVal Idx = ER->getIndex().castAs<DefinedOrUnknownSVal>(); 299 300 ProgramStateRef StInBound = state->assumeInBound(Idx, Size, true); 301 ProgramStateRef StOutBound = state->assumeInBound(Idx, Size, false); 302 if (StOutBound && !StInBound) { 303 // These checks are either enabled by the CString out-of-bounds checker 304 // explicitly or the "basic" CStringNullArg checker support that Malloc 305 // checker enables. 306 assert(Filter.CheckCStringOutOfBounds || Filter.CheckCStringNullArg); 307 308 // Emit a bug report. 309 if (warningMsg) { 310 emitOutOfBoundsBug(C, StOutBound, S, warningMsg); 311 } else { 312 assert(CurrentFunctionDescription); 313 assert(CurrentFunctionDescription[0] != '\0'); 314 315 SmallString<80> buf; 316 llvm::raw_svector_ostream os(buf); 317 os << toUppercase(CurrentFunctionDescription[0]) 318 << &CurrentFunctionDescription[1] 319 << " accesses out-of-bound array element"; 320 emitOutOfBoundsBug(C, StOutBound, S, os.str()); 321 } 322 return nullptr; 323 } 324 325 // Array bound check succeeded. From this point forward the array bound 326 // should always succeed. 327 return StInBound; 328 } 329 330 ProgramStateRef CStringChecker::CheckBufferAccess(CheckerContext &C, 331 ProgramStateRef state, 332 const Expr *Size, 333 const Expr *FirstBuf, 334 const Expr *SecondBuf, 335 const char *firstMessage, 336 const char *secondMessage, 337 bool WarnAboutSize) const { 338 // If a previous check has failed, propagate the failure. 339 if (!state) 340 return nullptr; 341 342 SValBuilder &svalBuilder = C.getSValBuilder(); 343 ASTContext &Ctx = svalBuilder.getContext(); 344 const LocationContext *LCtx = C.getLocationContext(); 345 346 QualType sizeTy = Size->getType(); 347 QualType PtrTy = Ctx.getPointerType(Ctx.CharTy); 348 349 // Check that the first buffer is non-null. 350 SVal BufVal = C.getSVal(FirstBuf); 351 state = checkNonNull(C, state, FirstBuf, BufVal); 352 if (!state) 353 return nullptr; 354 355 // If out-of-bounds checking is turned off, skip the rest. 356 if (!Filter.CheckCStringOutOfBounds) 357 return state; 358 359 // Get the access length and make sure it is known. 360 // FIXME: This assumes the caller has already checked that the access length 361 // is positive. And that it's unsigned. 362 SVal LengthVal = C.getSVal(Size); 363 Optional<NonLoc> Length = LengthVal.getAs<NonLoc>(); 364 if (!Length) 365 return state; 366 367 // Compute the offset of the last element to be accessed: size-1. 368 NonLoc One = svalBuilder.makeIntVal(1, sizeTy).castAs<NonLoc>(); 369 SVal Offset = svalBuilder.evalBinOpNN(state, BO_Sub, *Length, One, sizeTy); 370 if (Offset.isUnknown()) 371 return nullptr; 372 NonLoc LastOffset = Offset.castAs<NonLoc>(); 373 374 // Check that the first buffer is sufficiently long. 375 SVal BufStart = svalBuilder.evalCast(BufVal, PtrTy, FirstBuf->getType()); 376 if (Optional<Loc> BufLoc = BufStart.getAs<Loc>()) { 377 const Expr *warningExpr = (WarnAboutSize ? Size : FirstBuf); 378 379 SVal BufEnd = svalBuilder.evalBinOpLN(state, BO_Add, *BufLoc, 380 LastOffset, PtrTy); 381 state = CheckLocation(C, state, warningExpr, BufEnd, firstMessage); 382 383 // If the buffer isn't large enough, abort. 384 if (!state) 385 return nullptr; 386 } 387 388 // If there's a second buffer, check it as well. 389 if (SecondBuf) { 390 BufVal = state->getSVal(SecondBuf, LCtx); 391 state = checkNonNull(C, state, SecondBuf, BufVal); 392 if (!state) 393 return nullptr; 394 395 BufStart = svalBuilder.evalCast(BufVal, PtrTy, SecondBuf->getType()); 396 if (Optional<Loc> BufLoc = BufStart.getAs<Loc>()) { 397 const Expr *warningExpr = (WarnAboutSize ? Size : SecondBuf); 398 399 SVal BufEnd = svalBuilder.evalBinOpLN(state, BO_Add, *BufLoc, 400 LastOffset, PtrTy); 401 state = CheckLocation(C, state, warningExpr, BufEnd, secondMessage); 402 } 403 } 404 405 // Large enough or not, return this state! 406 return state; 407 } 408 409 ProgramStateRef CStringChecker::CheckOverlap(CheckerContext &C, 410 ProgramStateRef state, 411 const Expr *Size, 412 const Expr *First, 413 const Expr *Second) const { 414 if (!Filter.CheckCStringBufferOverlap) 415 return state; 416 417 // Do a simple check for overlap: if the two arguments are from the same 418 // buffer, see if the end of the first is greater than the start of the second 419 // or vice versa. 420 421 // If a previous check has failed, propagate the failure. 422 if (!state) 423 return nullptr; 424 425 ProgramStateRef stateTrue, stateFalse; 426 427 // Get the buffer values and make sure they're known locations. 428 const LocationContext *LCtx = C.getLocationContext(); 429 SVal firstVal = state->getSVal(First, LCtx); 430 SVal secondVal = state->getSVal(Second, LCtx); 431 432 Optional<Loc> firstLoc = firstVal.getAs<Loc>(); 433 if (!firstLoc) 434 return state; 435 436 Optional<Loc> secondLoc = secondVal.getAs<Loc>(); 437 if (!secondLoc) 438 return state; 439 440 // Are the two values the same? 441 SValBuilder &svalBuilder = C.getSValBuilder(); 442 std::tie(stateTrue, stateFalse) = 443 state->assume(svalBuilder.evalEQ(state, *firstLoc, *secondLoc)); 444 445 if (stateTrue && !stateFalse) { 446 // If the values are known to be equal, that's automatically an overlap. 447 emitOverlapBug(C, stateTrue, First, Second); 448 return nullptr; 449 } 450 451 // assume the two expressions are not equal. 452 assert(stateFalse); 453 state = stateFalse; 454 455 // Which value comes first? 456 QualType cmpTy = svalBuilder.getConditionType(); 457 SVal reverse = svalBuilder.evalBinOpLL(state, BO_GT, 458 *firstLoc, *secondLoc, cmpTy); 459 Optional<DefinedOrUnknownSVal> reverseTest = 460 reverse.getAs<DefinedOrUnknownSVal>(); 461 if (!reverseTest) 462 return state; 463 464 std::tie(stateTrue, stateFalse) = state->assume(*reverseTest); 465 if (stateTrue) { 466 if (stateFalse) { 467 // If we don't know which one comes first, we can't perform this test. 468 return state; 469 } else { 470 // Switch the values so that firstVal is before secondVal. 471 std::swap(firstLoc, secondLoc); 472 473 // Switch the Exprs as well, so that they still correspond. 474 std::swap(First, Second); 475 } 476 } 477 478 // Get the length, and make sure it too is known. 479 SVal LengthVal = state->getSVal(Size, LCtx); 480 Optional<NonLoc> Length = LengthVal.getAs<NonLoc>(); 481 if (!Length) 482 return state; 483 484 // Convert the first buffer's start address to char*. 485 // Bail out if the cast fails. 486 ASTContext &Ctx = svalBuilder.getContext(); 487 QualType CharPtrTy = Ctx.getPointerType(Ctx.CharTy); 488 SVal FirstStart = svalBuilder.evalCast(*firstLoc, CharPtrTy, 489 First->getType()); 490 Optional<Loc> FirstStartLoc = FirstStart.getAs<Loc>(); 491 if (!FirstStartLoc) 492 return state; 493 494 // Compute the end of the first buffer. Bail out if THAT fails. 495 SVal FirstEnd = svalBuilder.evalBinOpLN(state, BO_Add, 496 *FirstStartLoc, *Length, CharPtrTy); 497 Optional<Loc> FirstEndLoc = FirstEnd.getAs<Loc>(); 498 if (!FirstEndLoc) 499 return state; 500 501 // Is the end of the first buffer past the start of the second buffer? 502 SVal Overlap = svalBuilder.evalBinOpLL(state, BO_GT, 503 *FirstEndLoc, *secondLoc, cmpTy); 504 Optional<DefinedOrUnknownSVal> OverlapTest = 505 Overlap.getAs<DefinedOrUnknownSVal>(); 506 if (!OverlapTest) 507 return state; 508 509 std::tie(stateTrue, stateFalse) = state->assume(*OverlapTest); 510 511 if (stateTrue && !stateFalse) { 512 // Overlap! 513 emitOverlapBug(C, stateTrue, First, Second); 514 return nullptr; 515 } 516 517 // assume the two expressions don't overlap. 518 assert(stateFalse); 519 return stateFalse; 520 } 521 522 void CStringChecker::emitOverlapBug(CheckerContext &C, ProgramStateRef state, 523 const Stmt *First, const Stmt *Second) const { 524 ExplodedNode *N = C.generateErrorNode(state); 525 if (!N) 526 return; 527 528 if (!BT_Overlap) 529 BT_Overlap.reset(new BugType(Filter.CheckNameCStringBufferOverlap, 530 categories::UnixAPI, "Improper arguments")); 531 532 // Generate a report for this bug. 533 auto report = llvm::make_unique<BugReport>( 534 *BT_Overlap, "Arguments must not be overlapping buffers", N); 535 report->addRange(First->getSourceRange()); 536 report->addRange(Second->getSourceRange()); 537 538 C.emitReport(std::move(report)); 539 } 540 541 void CStringChecker::emitNullArgBug(CheckerContext &C, ProgramStateRef State, 542 const Stmt *S, StringRef WarningMsg) const { 543 if (ExplodedNode *N = C.generateErrorNode(State)) { 544 if (!BT_Null) 545 BT_Null.reset(new BuiltinBug( 546 Filter.CheckNameCStringNullArg, categories::UnixAPI, 547 "Null pointer argument in call to byte string function")); 548 549 BuiltinBug *BT = static_cast<BuiltinBug *>(BT_Null.get()); 550 auto Report = llvm::make_unique<BugReport>(*BT, WarningMsg, N); 551 bugreporter::trackNullOrUndefValue(N, S, *Report); 552 C.emitReport(std::move(Report)); 553 } 554 } 555 556 void CStringChecker::emitOutOfBoundsBug(CheckerContext &C, 557 ProgramStateRef State, const Stmt *S, 558 StringRef WarningMsg) const { 559 if (ExplodedNode *N = C.generateErrorNode(State)) { 560 if (!BT_Bounds) 561 BT_Bounds.reset(new BuiltinBug( 562 Filter.CheckCStringOutOfBounds ? Filter.CheckNameCStringOutOfBounds 563 : Filter.CheckNameCStringNullArg, 564 "Out-of-bound array access", 565 "Byte string function accesses out-of-bound array element")); 566 567 BuiltinBug *BT = static_cast<BuiltinBug *>(BT_Bounds.get()); 568 569 // FIXME: It would be nice to eventually make this diagnostic more clear, 570 // e.g., by referencing the original declaration or by saying *why* this 571 // reference is outside the range. 572 auto Report = llvm::make_unique<BugReport>(*BT, WarningMsg, N); 573 Report->addRange(S->getSourceRange()); 574 C.emitReport(std::move(Report)); 575 } 576 } 577 578 void CStringChecker::emitNotCStringBug(CheckerContext &C, ProgramStateRef State, 579 const Stmt *S, 580 StringRef WarningMsg) const { 581 if (ExplodedNode *N = C.generateNonFatalErrorNode(State)) { 582 if (!BT_NotCString) 583 BT_NotCString.reset(new BuiltinBug( 584 Filter.CheckNameCStringNotNullTerm, categories::UnixAPI, 585 "Argument is not a null-terminated string.")); 586 587 auto Report = llvm::make_unique<BugReport>(*BT_NotCString, WarningMsg, N); 588 589 Report->addRange(S->getSourceRange()); 590 C.emitReport(std::move(Report)); 591 } 592 } 593 594 void CStringChecker::emitAdditionOverflowBug(CheckerContext &C, 595 ProgramStateRef State) const { 596 if (ExplodedNode *N = C.generateErrorNode(State)) { 597 if (!BT_NotCString) 598 BT_NotCString.reset( 599 new BuiltinBug(Filter.CheckNameCStringOutOfBounds, "API", 600 "Sum of expressions causes overflow.")); 601 602 // This isn't a great error message, but this should never occur in real 603 // code anyway -- you'd have to create a buffer longer than a size_t can 604 // represent, which is sort of a contradiction. 605 const char *WarningMsg = 606 "This expression will create a string whose length is too big to " 607 "be represented as a size_t"; 608 609 auto Report = llvm::make_unique<BugReport>(*BT_NotCString, WarningMsg, N); 610 C.emitReport(std::move(Report)); 611 } 612 } 613 614 ProgramStateRef CStringChecker::checkAdditionOverflow(CheckerContext &C, 615 ProgramStateRef state, 616 NonLoc left, 617 NonLoc right) const { 618 // If out-of-bounds checking is turned off, skip the rest. 619 if (!Filter.CheckCStringOutOfBounds) 620 return state; 621 622 // If a previous check has failed, propagate the failure. 623 if (!state) 624 return nullptr; 625 626 SValBuilder &svalBuilder = C.getSValBuilder(); 627 BasicValueFactory &BVF = svalBuilder.getBasicValueFactory(); 628 629 QualType sizeTy = svalBuilder.getContext().getSizeType(); 630 const llvm::APSInt &maxValInt = BVF.getMaxValue(sizeTy); 631 NonLoc maxVal = svalBuilder.makeIntVal(maxValInt); 632 633 SVal maxMinusRight; 634 if (right.getAs<nonloc::ConcreteInt>()) { 635 maxMinusRight = svalBuilder.evalBinOpNN(state, BO_Sub, maxVal, right, 636 sizeTy); 637 } else { 638 // Try switching the operands. (The order of these two assignments is 639 // important!) 640 maxMinusRight = svalBuilder.evalBinOpNN(state, BO_Sub, maxVal, left, 641 sizeTy); 642 left = right; 643 } 644 645 if (Optional<NonLoc> maxMinusRightNL = maxMinusRight.getAs<NonLoc>()) { 646 QualType cmpTy = svalBuilder.getConditionType(); 647 // If left > max - right, we have an overflow. 648 SVal willOverflow = svalBuilder.evalBinOpNN(state, BO_GT, left, 649 *maxMinusRightNL, cmpTy); 650 651 ProgramStateRef stateOverflow, stateOkay; 652 std::tie(stateOverflow, stateOkay) = 653 state->assume(willOverflow.castAs<DefinedOrUnknownSVal>()); 654 655 if (stateOverflow && !stateOkay) { 656 // We have an overflow. Emit a bug report. 657 emitAdditionOverflowBug(C, stateOverflow); 658 return nullptr; 659 } 660 661 // From now on, assume an overflow didn't occur. 662 assert(stateOkay); 663 state = stateOkay; 664 } 665 666 return state; 667 } 668 669 ProgramStateRef CStringChecker::setCStringLength(ProgramStateRef state, 670 const MemRegion *MR, 671 SVal strLength) { 672 assert(!strLength.isUndef() && "Attempt to set an undefined string length"); 673 674 MR = MR->StripCasts(); 675 676 switch (MR->getKind()) { 677 case MemRegion::StringRegionKind: 678 // FIXME: This can happen if we strcpy() into a string region. This is 679 // undefined [C99 6.4.5p6], but we should still warn about it. 680 return state; 681 682 case MemRegion::SymbolicRegionKind: 683 case MemRegion::AllocaRegionKind: 684 case MemRegion::VarRegionKind: 685 case MemRegion::FieldRegionKind: 686 case MemRegion::ObjCIvarRegionKind: 687 // These are the types we can currently track string lengths for. 688 break; 689 690 case MemRegion::ElementRegionKind: 691 // FIXME: Handle element regions by upper-bounding the parent region's 692 // string length. 693 return state; 694 695 default: 696 // Other regions (mostly non-data) can't have a reliable C string length. 697 // For now, just ignore the change. 698 // FIXME: These are rare but not impossible. We should output some kind of 699 // warning for things like strcpy((char[]){'a', 0}, "b"); 700 return state; 701 } 702 703 if (strLength.isUnknown()) 704 return state->remove<CStringLength>(MR); 705 706 return state->set<CStringLength>(MR, strLength); 707 } 708 709 SVal CStringChecker::getCStringLengthForRegion(CheckerContext &C, 710 ProgramStateRef &state, 711 const Expr *Ex, 712 const MemRegion *MR, 713 bool hypothetical) { 714 if (!hypothetical) { 715 // If there's a recorded length, go ahead and return it. 716 const SVal *Recorded = state->get<CStringLength>(MR); 717 if (Recorded) 718 return *Recorded; 719 } 720 721 // Otherwise, get a new symbol and update the state. 722 SValBuilder &svalBuilder = C.getSValBuilder(); 723 QualType sizeTy = svalBuilder.getContext().getSizeType(); 724 SVal strLength = svalBuilder.getMetadataSymbolVal(CStringChecker::getTag(), 725 MR, Ex, sizeTy, 726 C.getLocationContext(), 727 C.blockCount()); 728 729 if (!hypothetical) { 730 if (Optional<NonLoc> strLn = strLength.getAs<NonLoc>()) { 731 // In case of unbounded calls strlen etc bound the range to SIZE_MAX/4 732 BasicValueFactory &BVF = svalBuilder.getBasicValueFactory(); 733 const llvm::APSInt &maxValInt = BVF.getMaxValue(sizeTy); 734 llvm::APSInt fourInt = APSIntType(maxValInt).getValue(4); 735 const llvm::APSInt *maxLengthInt = BVF.evalAPSInt(BO_Div, maxValInt, 736 fourInt); 737 NonLoc maxLength = svalBuilder.makeIntVal(*maxLengthInt); 738 SVal evalLength = svalBuilder.evalBinOpNN(state, BO_LE, *strLn, 739 maxLength, sizeTy); 740 state = state->assume(evalLength.castAs<DefinedOrUnknownSVal>(), true); 741 } 742 state = state->set<CStringLength>(MR, strLength); 743 } 744 745 return strLength; 746 } 747 748 SVal CStringChecker::getCStringLength(CheckerContext &C, ProgramStateRef &state, 749 const Expr *Ex, SVal Buf, 750 bool hypothetical) const { 751 const MemRegion *MR = Buf.getAsRegion(); 752 if (!MR) { 753 // If we can't get a region, see if it's something we /know/ isn't a 754 // C string. In the context of locations, the only time we can issue such 755 // a warning is for labels. 756 if (Optional<loc::GotoLabel> Label = Buf.getAs<loc::GotoLabel>()) { 757 if (Filter.CheckCStringNotNullTerm) { 758 SmallString<120> buf; 759 llvm::raw_svector_ostream os(buf); 760 assert(CurrentFunctionDescription); 761 os << "Argument to " << CurrentFunctionDescription 762 << " is the address of the label '" << Label->getLabel()->getName() 763 << "', which is not a null-terminated string"; 764 765 emitNotCStringBug(C, state, Ex, os.str()); 766 } 767 return UndefinedVal(); 768 } 769 770 // If it's not a region and not a label, give up. 771 return UnknownVal(); 772 } 773 774 // If we have a region, strip casts from it and see if we can figure out 775 // its length. For anything we can't figure out, just return UnknownVal. 776 MR = MR->StripCasts(); 777 778 switch (MR->getKind()) { 779 case MemRegion::StringRegionKind: { 780 // Modifying the contents of string regions is undefined [C99 6.4.5p6], 781 // so we can assume that the byte length is the correct C string length. 782 SValBuilder &svalBuilder = C.getSValBuilder(); 783 QualType sizeTy = svalBuilder.getContext().getSizeType(); 784 const StringLiteral *strLit = cast<StringRegion>(MR)->getStringLiteral(); 785 return svalBuilder.makeIntVal(strLit->getByteLength(), sizeTy); 786 } 787 case MemRegion::SymbolicRegionKind: 788 case MemRegion::AllocaRegionKind: 789 case MemRegion::VarRegionKind: 790 case MemRegion::FieldRegionKind: 791 case MemRegion::ObjCIvarRegionKind: 792 return getCStringLengthForRegion(C, state, Ex, MR, hypothetical); 793 case MemRegion::CompoundLiteralRegionKind: 794 // FIXME: Can we track this? Is it necessary? 795 return UnknownVal(); 796 case MemRegion::ElementRegionKind: 797 // FIXME: How can we handle this? It's not good enough to subtract the 798 // offset from the base string length; consider "123\x00567" and &a[5]. 799 return UnknownVal(); 800 default: 801 // Other regions (mostly non-data) can't have a reliable C string length. 802 // In this case, an error is emitted and UndefinedVal is returned. 803 // The caller should always be prepared to handle this case. 804 if (Filter.CheckCStringNotNullTerm) { 805 SmallString<120> buf; 806 llvm::raw_svector_ostream os(buf); 807 808 assert(CurrentFunctionDescription); 809 os << "Argument to " << CurrentFunctionDescription << " is "; 810 811 if (SummarizeRegion(os, C.getASTContext(), MR)) 812 os << ", which is not a null-terminated string"; 813 else 814 os << "not a null-terminated string"; 815 816 emitNotCStringBug(C, state, Ex, os.str()); 817 } 818 return UndefinedVal(); 819 } 820 } 821 822 const StringLiteral *CStringChecker::getCStringLiteral(CheckerContext &C, 823 ProgramStateRef &state, const Expr *expr, SVal val) const { 824 825 // Get the memory region pointed to by the val. 826 const MemRegion *bufRegion = val.getAsRegion(); 827 if (!bufRegion) 828 return nullptr; 829 830 // Strip casts off the memory region. 831 bufRegion = bufRegion->StripCasts(); 832 833 // Cast the memory region to a string region. 834 const StringRegion *strRegion= dyn_cast<StringRegion>(bufRegion); 835 if (!strRegion) 836 return nullptr; 837 838 // Return the actual string in the string region. 839 return strRegion->getStringLiteral(); 840 } 841 842 bool CStringChecker::IsFirstBufInBound(CheckerContext &C, 843 ProgramStateRef state, 844 const Expr *FirstBuf, 845 const Expr *Size) { 846 // If we do not know that the buffer is long enough we return 'true'. 847 // Otherwise the parent region of this field region would also get 848 // invalidated, which would lead to warnings based on an unknown state. 849 850 // Originally copied from CheckBufferAccess and CheckLocation. 851 SValBuilder &svalBuilder = C.getSValBuilder(); 852 ASTContext &Ctx = svalBuilder.getContext(); 853 const LocationContext *LCtx = C.getLocationContext(); 854 855 QualType sizeTy = Size->getType(); 856 QualType PtrTy = Ctx.getPointerType(Ctx.CharTy); 857 SVal BufVal = state->getSVal(FirstBuf, LCtx); 858 859 SVal LengthVal = state->getSVal(Size, LCtx); 860 Optional<NonLoc> Length = LengthVal.getAs<NonLoc>(); 861 if (!Length) 862 return true; // cf top comment. 863 864 // Compute the offset of the last element to be accessed: size-1. 865 NonLoc One = svalBuilder.makeIntVal(1, sizeTy).castAs<NonLoc>(); 866 SVal Offset = svalBuilder.evalBinOpNN(state, BO_Sub, *Length, One, sizeTy); 867 if (Offset.isUnknown()) 868 return true; // cf top comment 869 NonLoc LastOffset = Offset.castAs<NonLoc>(); 870 871 // Check that the first buffer is sufficiently long. 872 SVal BufStart = svalBuilder.evalCast(BufVal, PtrTy, FirstBuf->getType()); 873 Optional<Loc> BufLoc = BufStart.getAs<Loc>(); 874 if (!BufLoc) 875 return true; // cf top comment. 876 877 SVal BufEnd = 878 svalBuilder.evalBinOpLN(state, BO_Add, *BufLoc, LastOffset, PtrTy); 879 880 // Check for out of bound array element access. 881 const MemRegion *R = BufEnd.getAsRegion(); 882 if (!R) 883 return true; // cf top comment. 884 885 const ElementRegion *ER = dyn_cast<ElementRegion>(R); 886 if (!ER) 887 return true; // cf top comment. 888 889 // FIXME: Does this crash when a non-standard definition 890 // of a library function is encountered? 891 assert(ER->getValueType() == C.getASTContext().CharTy && 892 "IsFirstBufInBound should only be called with char* ElementRegions"); 893 894 // Get the size of the array. 895 const SubRegion *superReg = cast<SubRegion>(ER->getSuperRegion()); 896 SVal Extent = 897 svalBuilder.convertToArrayIndex(superReg->getExtent(svalBuilder)); 898 DefinedOrUnknownSVal ExtentSize = Extent.castAs<DefinedOrUnknownSVal>(); 899 900 // Get the index of the accessed element. 901 DefinedOrUnknownSVal Idx = ER->getIndex().castAs<DefinedOrUnknownSVal>(); 902 903 ProgramStateRef StInBound = state->assumeInBound(Idx, ExtentSize, true); 904 905 return static_cast<bool>(StInBound); 906 } 907 908 ProgramStateRef CStringChecker::InvalidateBuffer(CheckerContext &C, 909 ProgramStateRef state, 910 const Expr *E, SVal V, 911 bool IsSourceBuffer, 912 const Expr *Size) { 913 Optional<Loc> L = V.getAs<Loc>(); 914 if (!L) 915 return state; 916 917 // FIXME: This is a simplified version of what's in CFRefCount.cpp -- it makes 918 // some assumptions about the value that CFRefCount can't. Even so, it should 919 // probably be refactored. 920 if (Optional<loc::MemRegionVal> MR = L->getAs<loc::MemRegionVal>()) { 921 const MemRegion *R = MR->getRegion()->StripCasts(); 922 923 // Are we dealing with an ElementRegion? If so, we should be invalidating 924 // the super-region. 925 if (const ElementRegion *ER = dyn_cast<ElementRegion>(R)) { 926 R = ER->getSuperRegion(); 927 // FIXME: What about layers of ElementRegions? 928 } 929 930 // Invalidate this region. 931 const LocationContext *LCtx = C.getPredecessor()->getLocationContext(); 932 933 bool CausesPointerEscape = false; 934 RegionAndSymbolInvalidationTraits ITraits; 935 // Invalidate and escape only indirect regions accessible through the source 936 // buffer. 937 if (IsSourceBuffer) { 938 ITraits.setTrait(R->getBaseRegion(), 939 RegionAndSymbolInvalidationTraits::TK_PreserveContents); 940 ITraits.setTrait(R, RegionAndSymbolInvalidationTraits::TK_SuppressEscape); 941 CausesPointerEscape = true; 942 } else { 943 const MemRegion::Kind& K = R->getKind(); 944 if (K == MemRegion::FieldRegionKind) 945 if (Size && IsFirstBufInBound(C, state, E, Size)) { 946 // If destination buffer is a field region and access is in bound, 947 // do not invalidate its super region. 948 ITraits.setTrait( 949 R, 950 RegionAndSymbolInvalidationTraits::TK_DoNotInvalidateSuperRegion); 951 } 952 } 953 954 return state->invalidateRegions(R, E, C.blockCount(), LCtx, 955 CausesPointerEscape, nullptr, nullptr, 956 &ITraits); 957 } 958 959 // If we have a non-region value by chance, just remove the binding. 960 // FIXME: is this necessary or correct? This handles the non-Region 961 // cases. Is it ever valid to store to these? 962 return state->killBinding(*L); 963 } 964 965 bool CStringChecker::SummarizeRegion(raw_ostream &os, ASTContext &Ctx, 966 const MemRegion *MR) { 967 const TypedValueRegion *TVR = dyn_cast<TypedValueRegion>(MR); 968 969 switch (MR->getKind()) { 970 case MemRegion::FunctionCodeRegionKind: { 971 const NamedDecl *FD = cast<FunctionCodeRegion>(MR)->getDecl(); 972 if (FD) 973 os << "the address of the function '" << *FD << '\''; 974 else 975 os << "the address of a function"; 976 return true; 977 } 978 case MemRegion::BlockCodeRegionKind: 979 os << "block text"; 980 return true; 981 case MemRegion::BlockDataRegionKind: 982 os << "a block"; 983 return true; 984 case MemRegion::CXXThisRegionKind: 985 case MemRegion::CXXTempObjectRegionKind: 986 os << "a C++ temp object of type " << TVR->getValueType().getAsString(); 987 return true; 988 case MemRegion::VarRegionKind: 989 os << "a variable of type" << TVR->getValueType().getAsString(); 990 return true; 991 case MemRegion::FieldRegionKind: 992 os << "a field of type " << TVR->getValueType().getAsString(); 993 return true; 994 case MemRegion::ObjCIvarRegionKind: 995 os << "an instance variable of type " << TVR->getValueType().getAsString(); 996 return true; 997 default: 998 return false; 999 } 1000 } 1001 1002 //===----------------------------------------------------------------------===// 1003 // evaluation of individual function calls. 1004 //===----------------------------------------------------------------------===// 1005 1006 void CStringChecker::evalCopyCommon(CheckerContext &C, 1007 const CallExpr *CE, 1008 ProgramStateRef state, 1009 const Expr *Size, const Expr *Dest, 1010 const Expr *Source, bool Restricted, 1011 bool IsMempcpy) const { 1012 CurrentFunctionDescription = "memory copy function"; 1013 1014 // See if the size argument is zero. 1015 const LocationContext *LCtx = C.getLocationContext(); 1016 SVal sizeVal = state->getSVal(Size, LCtx); 1017 QualType sizeTy = Size->getType(); 1018 1019 ProgramStateRef stateZeroSize, stateNonZeroSize; 1020 std::tie(stateZeroSize, stateNonZeroSize) = 1021 assumeZero(C, state, sizeVal, sizeTy); 1022 1023 // Get the value of the Dest. 1024 SVal destVal = state->getSVal(Dest, LCtx); 1025 1026 // If the size is zero, there won't be any actual memory access, so 1027 // just bind the return value to the destination buffer and return. 1028 if (stateZeroSize && !stateNonZeroSize) { 1029 stateZeroSize = stateZeroSize->BindExpr(CE, LCtx, destVal); 1030 C.addTransition(stateZeroSize); 1031 return; 1032 } 1033 1034 // If the size can be nonzero, we have to check the other arguments. 1035 if (stateNonZeroSize) { 1036 state = stateNonZeroSize; 1037 1038 // Ensure the destination is not null. If it is NULL there will be a 1039 // NULL pointer dereference. 1040 state = checkNonNull(C, state, Dest, destVal); 1041 if (!state) 1042 return; 1043 1044 // Get the value of the Src. 1045 SVal srcVal = state->getSVal(Source, LCtx); 1046 1047 // Ensure the source is not null. If it is NULL there will be a 1048 // NULL pointer dereference. 1049 state = checkNonNull(C, state, Source, srcVal); 1050 if (!state) 1051 return; 1052 1053 // Ensure the accesses are valid and that the buffers do not overlap. 1054 const char * const writeWarning = 1055 "Memory copy function overflows destination buffer"; 1056 state = CheckBufferAccess(C, state, Size, Dest, Source, 1057 writeWarning, /* sourceWarning = */ nullptr); 1058 if (Restricted) 1059 state = CheckOverlap(C, state, Size, Dest, Source); 1060 1061 if (!state) 1062 return; 1063 1064 // If this is mempcpy, get the byte after the last byte copied and 1065 // bind the expr. 1066 if (IsMempcpy) { 1067 // Get the byte after the last byte copied. 1068 SValBuilder &SvalBuilder = C.getSValBuilder(); 1069 ASTContext &Ctx = SvalBuilder.getContext(); 1070 QualType CharPtrTy = Ctx.getPointerType(Ctx.CharTy); 1071 SVal DestRegCharVal = 1072 SvalBuilder.evalCast(destVal, CharPtrTy, Dest->getType()); 1073 SVal lastElement = C.getSValBuilder().evalBinOp( 1074 state, BO_Add, DestRegCharVal, sizeVal, Dest->getType()); 1075 // If we don't know how much we copied, we can at least 1076 // conjure a return value for later. 1077 if (lastElement.isUnknown()) 1078 lastElement = C.getSValBuilder().conjureSymbolVal(nullptr, CE, LCtx, 1079 C.blockCount()); 1080 1081 // The byte after the last byte copied is the return value. 1082 state = state->BindExpr(CE, LCtx, lastElement); 1083 } else { 1084 // All other copies return the destination buffer. 1085 // (Well, bcopy() has a void return type, but this won't hurt.) 1086 state = state->BindExpr(CE, LCtx, destVal); 1087 } 1088 1089 // Invalidate the destination (regular invalidation without pointer-escaping 1090 // the address of the top-level region). 1091 // FIXME: Even if we can't perfectly model the copy, we should see if we 1092 // can use LazyCompoundVals to copy the source values into the destination. 1093 // This would probably remove any existing bindings past the end of the 1094 // copied region, but that's still an improvement over blank invalidation. 1095 state = InvalidateBuffer(C, state, Dest, C.getSVal(Dest), 1096 /*IsSourceBuffer*/false, Size); 1097 1098 // Invalidate the source (const-invalidation without const-pointer-escaping 1099 // the address of the top-level region). 1100 state = InvalidateBuffer(C, state, Source, C.getSVal(Source), 1101 /*IsSourceBuffer*/true, nullptr); 1102 1103 C.addTransition(state); 1104 } 1105 } 1106 1107 1108 void CStringChecker::evalMemcpy(CheckerContext &C, const CallExpr *CE) const { 1109 if (CE->getNumArgs() < 3) 1110 return; 1111 1112 // void *memcpy(void *restrict dst, const void *restrict src, size_t n); 1113 // The return value is the address of the destination buffer. 1114 const Expr *Dest = CE->getArg(0); 1115 ProgramStateRef state = C.getState(); 1116 1117 evalCopyCommon(C, CE, state, CE->getArg(2), Dest, CE->getArg(1), true); 1118 } 1119 1120 void CStringChecker::evalMempcpy(CheckerContext &C, const CallExpr *CE) const { 1121 if (CE->getNumArgs() < 3) 1122 return; 1123 1124 // void *mempcpy(void *restrict dst, const void *restrict src, size_t n); 1125 // The return value is a pointer to the byte following the last written byte. 1126 const Expr *Dest = CE->getArg(0); 1127 ProgramStateRef state = C.getState(); 1128 1129 evalCopyCommon(C, CE, state, CE->getArg(2), Dest, CE->getArg(1), true, true); 1130 } 1131 1132 void CStringChecker::evalMemmove(CheckerContext &C, const CallExpr *CE) const { 1133 if (CE->getNumArgs() < 3) 1134 return; 1135 1136 // void *memmove(void *dst, const void *src, size_t n); 1137 // The return value is the address of the destination buffer. 1138 const Expr *Dest = CE->getArg(0); 1139 ProgramStateRef state = C.getState(); 1140 1141 evalCopyCommon(C, CE, state, CE->getArg(2), Dest, CE->getArg(1)); 1142 } 1143 1144 void CStringChecker::evalBcopy(CheckerContext &C, const CallExpr *CE) const { 1145 if (CE->getNumArgs() < 3) 1146 return; 1147 1148 // void bcopy(const void *src, void *dst, size_t n); 1149 evalCopyCommon(C, CE, C.getState(), 1150 CE->getArg(2), CE->getArg(1), CE->getArg(0)); 1151 } 1152 1153 void CStringChecker::evalMemcmp(CheckerContext &C, const CallExpr *CE) const { 1154 if (CE->getNumArgs() < 3) 1155 return; 1156 1157 // int memcmp(const void *s1, const void *s2, size_t n); 1158 CurrentFunctionDescription = "memory comparison function"; 1159 1160 const Expr *Left = CE->getArg(0); 1161 const Expr *Right = CE->getArg(1); 1162 const Expr *Size = CE->getArg(2); 1163 1164 ProgramStateRef state = C.getState(); 1165 SValBuilder &svalBuilder = C.getSValBuilder(); 1166 1167 // See if the size argument is zero. 1168 const LocationContext *LCtx = C.getLocationContext(); 1169 SVal sizeVal = state->getSVal(Size, LCtx); 1170 QualType sizeTy = Size->getType(); 1171 1172 ProgramStateRef stateZeroSize, stateNonZeroSize; 1173 std::tie(stateZeroSize, stateNonZeroSize) = 1174 assumeZero(C, state, sizeVal, sizeTy); 1175 1176 // If the size can be zero, the result will be 0 in that case, and we don't 1177 // have to check either of the buffers. 1178 if (stateZeroSize) { 1179 state = stateZeroSize; 1180 state = state->BindExpr(CE, LCtx, 1181 svalBuilder.makeZeroVal(CE->getType())); 1182 C.addTransition(state); 1183 } 1184 1185 // If the size can be nonzero, we have to check the other arguments. 1186 if (stateNonZeroSize) { 1187 state = stateNonZeroSize; 1188 // If we know the two buffers are the same, we know the result is 0. 1189 // First, get the two buffers' addresses. Another checker will have already 1190 // made sure they're not undefined. 1191 DefinedOrUnknownSVal LV = 1192 state->getSVal(Left, LCtx).castAs<DefinedOrUnknownSVal>(); 1193 DefinedOrUnknownSVal RV = 1194 state->getSVal(Right, LCtx).castAs<DefinedOrUnknownSVal>(); 1195 1196 // See if they are the same. 1197 DefinedOrUnknownSVal SameBuf = svalBuilder.evalEQ(state, LV, RV); 1198 ProgramStateRef StSameBuf, StNotSameBuf; 1199 std::tie(StSameBuf, StNotSameBuf) = state->assume(SameBuf); 1200 1201 // If the two arguments might be the same buffer, we know the result is 0, 1202 // and we only need to check one size. 1203 if (StSameBuf) { 1204 state = StSameBuf; 1205 state = CheckBufferAccess(C, state, Size, Left); 1206 if (state) { 1207 state = StSameBuf->BindExpr(CE, LCtx, 1208 svalBuilder.makeZeroVal(CE->getType())); 1209 C.addTransition(state); 1210 } 1211 } 1212 1213 // If the two arguments might be different buffers, we have to check the 1214 // size of both of them. 1215 if (StNotSameBuf) { 1216 state = StNotSameBuf; 1217 state = CheckBufferAccess(C, state, Size, Left, Right); 1218 if (state) { 1219 // The return value is the comparison result, which we don't know. 1220 SVal CmpV = svalBuilder.conjureSymbolVal(nullptr, CE, LCtx, 1221 C.blockCount()); 1222 state = state->BindExpr(CE, LCtx, CmpV); 1223 C.addTransition(state); 1224 } 1225 } 1226 } 1227 } 1228 1229 void CStringChecker::evalstrLength(CheckerContext &C, 1230 const CallExpr *CE) const { 1231 if (CE->getNumArgs() < 1) 1232 return; 1233 1234 // size_t strlen(const char *s); 1235 evalstrLengthCommon(C, CE, /* IsStrnlen = */ false); 1236 } 1237 1238 void CStringChecker::evalstrnLength(CheckerContext &C, 1239 const CallExpr *CE) const { 1240 if (CE->getNumArgs() < 2) 1241 return; 1242 1243 // size_t strnlen(const char *s, size_t maxlen); 1244 evalstrLengthCommon(C, CE, /* IsStrnlen = */ true); 1245 } 1246 1247 void CStringChecker::evalstrLengthCommon(CheckerContext &C, const CallExpr *CE, 1248 bool IsStrnlen) const { 1249 CurrentFunctionDescription = "string length function"; 1250 ProgramStateRef state = C.getState(); 1251 const LocationContext *LCtx = C.getLocationContext(); 1252 1253 if (IsStrnlen) { 1254 const Expr *maxlenExpr = CE->getArg(1); 1255 SVal maxlenVal = state->getSVal(maxlenExpr, LCtx); 1256 1257 ProgramStateRef stateZeroSize, stateNonZeroSize; 1258 std::tie(stateZeroSize, stateNonZeroSize) = 1259 assumeZero(C, state, maxlenVal, maxlenExpr->getType()); 1260 1261 // If the size can be zero, the result will be 0 in that case, and we don't 1262 // have to check the string itself. 1263 if (stateZeroSize) { 1264 SVal zero = C.getSValBuilder().makeZeroVal(CE->getType()); 1265 stateZeroSize = stateZeroSize->BindExpr(CE, LCtx, zero); 1266 C.addTransition(stateZeroSize); 1267 } 1268 1269 // If the size is GUARANTEED to be zero, we're done! 1270 if (!stateNonZeroSize) 1271 return; 1272 1273 // Otherwise, record the assumption that the size is nonzero. 1274 state = stateNonZeroSize; 1275 } 1276 1277 // Check that the string argument is non-null. 1278 const Expr *Arg = CE->getArg(0); 1279 SVal ArgVal = state->getSVal(Arg, LCtx); 1280 1281 state = checkNonNull(C, state, Arg, ArgVal); 1282 1283 if (!state) 1284 return; 1285 1286 SVal strLength = getCStringLength(C, state, Arg, ArgVal); 1287 1288 // If the argument isn't a valid C string, there's no valid state to 1289 // transition to. 1290 if (strLength.isUndef()) 1291 return; 1292 1293 DefinedOrUnknownSVal result = UnknownVal(); 1294 1295 // If the check is for strnlen() then bind the return value to no more than 1296 // the maxlen value. 1297 if (IsStrnlen) { 1298 QualType cmpTy = C.getSValBuilder().getConditionType(); 1299 1300 // It's a little unfortunate to be getting this again, 1301 // but it's not that expensive... 1302 const Expr *maxlenExpr = CE->getArg(1); 1303 SVal maxlenVal = state->getSVal(maxlenExpr, LCtx); 1304 1305 Optional<NonLoc> strLengthNL = strLength.getAs<NonLoc>(); 1306 Optional<NonLoc> maxlenValNL = maxlenVal.getAs<NonLoc>(); 1307 1308 if (strLengthNL && maxlenValNL) { 1309 ProgramStateRef stateStringTooLong, stateStringNotTooLong; 1310 1311 // Check if the strLength is greater than the maxlen. 1312 std::tie(stateStringTooLong, stateStringNotTooLong) = state->assume( 1313 C.getSValBuilder() 1314 .evalBinOpNN(state, BO_GT, *strLengthNL, *maxlenValNL, cmpTy) 1315 .castAs<DefinedOrUnknownSVal>()); 1316 1317 if (stateStringTooLong && !stateStringNotTooLong) { 1318 // If the string is longer than maxlen, return maxlen. 1319 result = *maxlenValNL; 1320 } else if (stateStringNotTooLong && !stateStringTooLong) { 1321 // If the string is shorter than maxlen, return its length. 1322 result = *strLengthNL; 1323 } 1324 } 1325 1326 if (result.isUnknown()) { 1327 // If we don't have enough information for a comparison, there's 1328 // no guarantee the full string length will actually be returned. 1329 // All we know is the return value is the min of the string length 1330 // and the limit. This is better than nothing. 1331 result = C.getSValBuilder().conjureSymbolVal(nullptr, CE, LCtx, 1332 C.blockCount()); 1333 NonLoc resultNL = result.castAs<NonLoc>(); 1334 1335 if (strLengthNL) { 1336 state = state->assume(C.getSValBuilder().evalBinOpNN( 1337 state, BO_LE, resultNL, *strLengthNL, cmpTy) 1338 .castAs<DefinedOrUnknownSVal>(), true); 1339 } 1340 1341 if (maxlenValNL) { 1342 state = state->assume(C.getSValBuilder().evalBinOpNN( 1343 state, BO_LE, resultNL, *maxlenValNL, cmpTy) 1344 .castAs<DefinedOrUnknownSVal>(), true); 1345 } 1346 } 1347 1348 } else { 1349 // This is a plain strlen(), not strnlen(). 1350 result = strLength.castAs<DefinedOrUnknownSVal>(); 1351 1352 // If we don't know the length of the string, conjure a return 1353 // value, so it can be used in constraints, at least. 1354 if (result.isUnknown()) { 1355 result = C.getSValBuilder().conjureSymbolVal(nullptr, CE, LCtx, 1356 C.blockCount()); 1357 } 1358 } 1359 1360 // Bind the return value. 1361 assert(!result.isUnknown() && "Should have conjured a value by now"); 1362 state = state->BindExpr(CE, LCtx, result); 1363 C.addTransition(state); 1364 } 1365 1366 void CStringChecker::evalStrcpy(CheckerContext &C, const CallExpr *CE) const { 1367 if (CE->getNumArgs() < 2) 1368 return; 1369 1370 // char *strcpy(char *restrict dst, const char *restrict src); 1371 evalStrcpyCommon(C, CE, 1372 /* returnEnd = */ false, 1373 /* isBounded = */ false, 1374 /* isAppending = */ false); 1375 } 1376 1377 void CStringChecker::evalStrncpy(CheckerContext &C, const CallExpr *CE) const { 1378 if (CE->getNumArgs() < 3) 1379 return; 1380 1381 // char *strncpy(char *restrict dst, const char *restrict src, size_t n); 1382 evalStrcpyCommon(C, CE, 1383 /* returnEnd = */ false, 1384 /* isBounded = */ true, 1385 /* isAppending = */ false); 1386 } 1387 1388 void CStringChecker::evalStpcpy(CheckerContext &C, const CallExpr *CE) const { 1389 if (CE->getNumArgs() < 2) 1390 return; 1391 1392 // char *stpcpy(char *restrict dst, const char *restrict src); 1393 evalStrcpyCommon(C, CE, 1394 /* returnEnd = */ true, 1395 /* isBounded = */ false, 1396 /* isAppending = */ false); 1397 } 1398 1399 void CStringChecker::evalStrlcpy(CheckerContext &C, const CallExpr *CE) const { 1400 if (CE->getNumArgs() < 3) 1401 return; 1402 1403 // char *strlcpy(char *dst, const char *src, size_t n); 1404 evalStrcpyCommon(C, CE, 1405 /* returnEnd = */ true, 1406 /* isBounded = */ true, 1407 /* isAppending = */ false, 1408 /* canOverlap = */ true); 1409 } 1410 1411 void CStringChecker::evalStrcat(CheckerContext &C, const CallExpr *CE) const { 1412 if (CE->getNumArgs() < 2) 1413 return; 1414 1415 //char *strcat(char *restrict s1, const char *restrict s2); 1416 evalStrcpyCommon(C, CE, 1417 /* returnEnd = */ false, 1418 /* isBounded = */ false, 1419 /* isAppending = */ true); 1420 } 1421 1422 void CStringChecker::evalStrncat(CheckerContext &C, const CallExpr *CE) const { 1423 if (CE->getNumArgs() < 3) 1424 return; 1425 1426 //char *strncat(char *restrict s1, const char *restrict s2, size_t n); 1427 evalStrcpyCommon(C, CE, 1428 /* returnEnd = */ false, 1429 /* isBounded = */ true, 1430 /* isAppending = */ true); 1431 } 1432 1433 void CStringChecker::evalStrlcat(CheckerContext &C, const CallExpr *CE) const { 1434 if (CE->getNumArgs() < 3) 1435 return; 1436 1437 //char *strlcat(char *s1, const char *s2, size_t n); 1438 evalStrcpyCommon(C, CE, 1439 /* returnEnd = */ false, 1440 /* isBounded = */ true, 1441 /* isAppending = */ true, 1442 /* canOverlap = */ true); 1443 } 1444 1445 void CStringChecker::evalStrcpyCommon(CheckerContext &C, const CallExpr *CE, 1446 bool returnEnd, bool isBounded, 1447 bool isAppending, bool canOverlap) const { 1448 CurrentFunctionDescription = "string copy function"; 1449 ProgramStateRef state = C.getState(); 1450 const LocationContext *LCtx = C.getLocationContext(); 1451 1452 // Check that the destination is non-null. 1453 const Expr *Dst = CE->getArg(0); 1454 SVal DstVal = state->getSVal(Dst, LCtx); 1455 1456 state = checkNonNull(C, state, Dst, DstVal); 1457 if (!state) 1458 return; 1459 1460 // Check that the source is non-null. 1461 const Expr *srcExpr = CE->getArg(1); 1462 SVal srcVal = state->getSVal(srcExpr, LCtx); 1463 state = checkNonNull(C, state, srcExpr, srcVal); 1464 if (!state) 1465 return; 1466 1467 // Get the string length of the source. 1468 SVal strLength = getCStringLength(C, state, srcExpr, srcVal); 1469 1470 // If the source isn't a valid C string, give up. 1471 if (strLength.isUndef()) 1472 return; 1473 1474 SValBuilder &svalBuilder = C.getSValBuilder(); 1475 QualType cmpTy = svalBuilder.getConditionType(); 1476 QualType sizeTy = svalBuilder.getContext().getSizeType(); 1477 1478 // These two values allow checking two kinds of errors: 1479 // - actual overflows caused by a source that doesn't fit in the destination 1480 // - potential overflows caused by a bound that could exceed the destination 1481 SVal amountCopied = UnknownVal(); 1482 SVal maxLastElementIndex = UnknownVal(); 1483 const char *boundWarning = nullptr; 1484 1485 if (canOverlap) 1486 state = CheckOverlap(C, state, CE->getArg(2), Dst, srcExpr); 1487 1488 if (!state) 1489 return; 1490 1491 // If the function is strncpy, strncat, etc... it is bounded. 1492 if (isBounded) { 1493 // Get the max number of characters to copy. 1494 const Expr *lenExpr = CE->getArg(2); 1495 SVal lenVal = state->getSVal(lenExpr, LCtx); 1496 1497 // Protect against misdeclared strncpy(). 1498 lenVal = svalBuilder.evalCast(lenVal, sizeTy, lenExpr->getType()); 1499 1500 Optional<NonLoc> strLengthNL = strLength.getAs<NonLoc>(); 1501 Optional<NonLoc> lenValNL = lenVal.getAs<NonLoc>(); 1502 1503 // If we know both values, we might be able to figure out how much 1504 // we're copying. 1505 if (strLengthNL && lenValNL) { 1506 ProgramStateRef stateSourceTooLong, stateSourceNotTooLong; 1507 1508 // Check if the max number to copy is less than the length of the src. 1509 // If the bound is equal to the source length, strncpy won't null- 1510 // terminate the result! 1511 std::tie(stateSourceTooLong, stateSourceNotTooLong) = state->assume( 1512 svalBuilder.evalBinOpNN(state, BO_GE, *strLengthNL, *lenValNL, cmpTy) 1513 .castAs<DefinedOrUnknownSVal>()); 1514 1515 if (stateSourceTooLong && !stateSourceNotTooLong) { 1516 // Max number to copy is less than the length of the src, so the actual 1517 // strLength copied is the max number arg. 1518 state = stateSourceTooLong; 1519 amountCopied = lenVal; 1520 1521 } else if (!stateSourceTooLong && stateSourceNotTooLong) { 1522 // The source buffer entirely fits in the bound. 1523 state = stateSourceNotTooLong; 1524 amountCopied = strLength; 1525 } 1526 } 1527 1528 // We still want to know if the bound is known to be too large. 1529 if (lenValNL) { 1530 if (isAppending) { 1531 // For strncat, the check is strlen(dst) + lenVal < sizeof(dst) 1532 1533 // Get the string length of the destination. If the destination is 1534 // memory that can't have a string length, we shouldn't be copying 1535 // into it anyway. 1536 SVal dstStrLength = getCStringLength(C, state, Dst, DstVal); 1537 if (dstStrLength.isUndef()) 1538 return; 1539 1540 if (Optional<NonLoc> dstStrLengthNL = dstStrLength.getAs<NonLoc>()) { 1541 maxLastElementIndex = svalBuilder.evalBinOpNN(state, BO_Add, 1542 *lenValNL, 1543 *dstStrLengthNL, 1544 sizeTy); 1545 boundWarning = "Size argument is greater than the free space in the " 1546 "destination buffer"; 1547 } 1548 1549 } else { 1550 // For strncpy, this is just checking that lenVal <= sizeof(dst) 1551 // (Yes, strncpy and strncat differ in how they treat termination. 1552 // strncat ALWAYS terminates, but strncpy doesn't.) 1553 1554 // We need a special case for when the copy size is zero, in which 1555 // case strncpy will do no work at all. Our bounds check uses n-1 1556 // as the last element accessed, so n == 0 is problematic. 1557 ProgramStateRef StateZeroSize, StateNonZeroSize; 1558 std::tie(StateZeroSize, StateNonZeroSize) = 1559 assumeZero(C, state, *lenValNL, sizeTy); 1560 1561 // If the size is known to be zero, we're done. 1562 if (StateZeroSize && !StateNonZeroSize) { 1563 StateZeroSize = StateZeroSize->BindExpr(CE, LCtx, DstVal); 1564 C.addTransition(StateZeroSize); 1565 return; 1566 } 1567 1568 // Otherwise, go ahead and figure out the last element we'll touch. 1569 // We don't record the non-zero assumption here because we can't 1570 // be sure. We won't warn on a possible zero. 1571 NonLoc one = svalBuilder.makeIntVal(1, sizeTy).castAs<NonLoc>(); 1572 maxLastElementIndex = svalBuilder.evalBinOpNN(state, BO_Sub, *lenValNL, 1573 one, sizeTy); 1574 boundWarning = "Size argument is greater than the length of the " 1575 "destination buffer"; 1576 } 1577 } 1578 1579 // If we couldn't pin down the copy length, at least bound it. 1580 // FIXME: We should actually run this code path for append as well, but 1581 // right now it creates problems with constraints (since we can end up 1582 // trying to pass constraints from symbol to symbol). 1583 if (amountCopied.isUnknown() && !isAppending) { 1584 // Try to get a "hypothetical" string length symbol, which we can later 1585 // set as a real value if that turns out to be the case. 1586 amountCopied = getCStringLength(C, state, lenExpr, srcVal, true); 1587 assert(!amountCopied.isUndef()); 1588 1589 if (Optional<NonLoc> amountCopiedNL = amountCopied.getAs<NonLoc>()) { 1590 if (lenValNL) { 1591 // amountCopied <= lenVal 1592 SVal copiedLessThanBound = svalBuilder.evalBinOpNN(state, BO_LE, 1593 *amountCopiedNL, 1594 *lenValNL, 1595 cmpTy); 1596 state = state->assume( 1597 copiedLessThanBound.castAs<DefinedOrUnknownSVal>(), true); 1598 if (!state) 1599 return; 1600 } 1601 1602 if (strLengthNL) { 1603 // amountCopied <= strlen(source) 1604 SVal copiedLessThanSrc = svalBuilder.evalBinOpNN(state, BO_LE, 1605 *amountCopiedNL, 1606 *strLengthNL, 1607 cmpTy); 1608 state = state->assume( 1609 copiedLessThanSrc.castAs<DefinedOrUnknownSVal>(), true); 1610 if (!state) 1611 return; 1612 } 1613 } 1614 } 1615 1616 } else { 1617 // The function isn't bounded. The amount copied should match the length 1618 // of the source buffer. 1619 amountCopied = strLength; 1620 } 1621 1622 assert(state); 1623 1624 // This represents the number of characters copied into the destination 1625 // buffer. (It may not actually be the strlen if the destination buffer 1626 // is not terminated.) 1627 SVal finalStrLength = UnknownVal(); 1628 1629 // If this is an appending function (strcat, strncat...) then set the 1630 // string length to strlen(src) + strlen(dst) since the buffer will 1631 // ultimately contain both. 1632 if (isAppending) { 1633 // Get the string length of the destination. If the destination is memory 1634 // that can't have a string length, we shouldn't be copying into it anyway. 1635 SVal dstStrLength = getCStringLength(C, state, Dst, DstVal); 1636 if (dstStrLength.isUndef()) 1637 return; 1638 1639 Optional<NonLoc> srcStrLengthNL = amountCopied.getAs<NonLoc>(); 1640 Optional<NonLoc> dstStrLengthNL = dstStrLength.getAs<NonLoc>(); 1641 1642 // If we know both string lengths, we might know the final string length. 1643 if (srcStrLengthNL && dstStrLengthNL) { 1644 // Make sure the two lengths together don't overflow a size_t. 1645 state = checkAdditionOverflow(C, state, *srcStrLengthNL, *dstStrLengthNL); 1646 if (!state) 1647 return; 1648 1649 finalStrLength = svalBuilder.evalBinOpNN(state, BO_Add, *srcStrLengthNL, 1650 *dstStrLengthNL, sizeTy); 1651 } 1652 1653 // If we couldn't get a single value for the final string length, 1654 // we can at least bound it by the individual lengths. 1655 if (finalStrLength.isUnknown()) { 1656 // Try to get a "hypothetical" string length symbol, which we can later 1657 // set as a real value if that turns out to be the case. 1658 finalStrLength = getCStringLength(C, state, CE, DstVal, true); 1659 assert(!finalStrLength.isUndef()); 1660 1661 if (Optional<NonLoc> finalStrLengthNL = finalStrLength.getAs<NonLoc>()) { 1662 if (srcStrLengthNL) { 1663 // finalStrLength >= srcStrLength 1664 SVal sourceInResult = svalBuilder.evalBinOpNN(state, BO_GE, 1665 *finalStrLengthNL, 1666 *srcStrLengthNL, 1667 cmpTy); 1668 state = state->assume(sourceInResult.castAs<DefinedOrUnknownSVal>(), 1669 true); 1670 if (!state) 1671 return; 1672 } 1673 1674 if (dstStrLengthNL) { 1675 // finalStrLength >= dstStrLength 1676 SVal destInResult = svalBuilder.evalBinOpNN(state, BO_GE, 1677 *finalStrLengthNL, 1678 *dstStrLengthNL, 1679 cmpTy); 1680 state = 1681 state->assume(destInResult.castAs<DefinedOrUnknownSVal>(), true); 1682 if (!state) 1683 return; 1684 } 1685 } 1686 } 1687 1688 } else { 1689 // Otherwise, this is a copy-over function (strcpy, strncpy, ...), and 1690 // the final string length will match the input string length. 1691 finalStrLength = amountCopied; 1692 } 1693 1694 // The final result of the function will either be a pointer past the last 1695 // copied element, or a pointer to the start of the destination buffer. 1696 SVal Result = (returnEnd ? UnknownVal() : DstVal); 1697 1698 assert(state); 1699 1700 // If the destination is a MemRegion, try to check for a buffer overflow and 1701 // record the new string length. 1702 if (Optional<loc::MemRegionVal> dstRegVal = 1703 DstVal.getAs<loc::MemRegionVal>()) { 1704 QualType ptrTy = Dst->getType(); 1705 1706 // If we have an exact value on a bounded copy, use that to check for 1707 // overflows, rather than our estimate about how much is actually copied. 1708 if (boundWarning) { 1709 if (Optional<NonLoc> maxLastNL = maxLastElementIndex.getAs<NonLoc>()) { 1710 SVal maxLastElement = svalBuilder.evalBinOpLN(state, BO_Add, *dstRegVal, 1711 *maxLastNL, ptrTy); 1712 state = CheckLocation(C, state, CE->getArg(2), maxLastElement, 1713 boundWarning); 1714 if (!state) 1715 return; 1716 } 1717 } 1718 1719 // Then, if the final length is known... 1720 if (Optional<NonLoc> knownStrLength = finalStrLength.getAs<NonLoc>()) { 1721 SVal lastElement = svalBuilder.evalBinOpLN(state, BO_Add, *dstRegVal, 1722 *knownStrLength, ptrTy); 1723 1724 // ...and we haven't checked the bound, we'll check the actual copy. 1725 if (!boundWarning) { 1726 const char * const warningMsg = 1727 "String copy function overflows destination buffer"; 1728 state = CheckLocation(C, state, Dst, lastElement, warningMsg); 1729 if (!state) 1730 return; 1731 } 1732 1733 // If this is a stpcpy-style copy, the last element is the return value. 1734 if (returnEnd) 1735 Result = lastElement; 1736 } 1737 1738 // Invalidate the destination (regular invalidation without pointer-escaping 1739 // the address of the top-level region). This must happen before we set the 1740 // C string length because invalidation will clear the length. 1741 // FIXME: Even if we can't perfectly model the copy, we should see if we 1742 // can use LazyCompoundVals to copy the source values into the destination. 1743 // This would probably remove any existing bindings past the end of the 1744 // string, but that's still an improvement over blank invalidation. 1745 state = InvalidateBuffer(C, state, Dst, *dstRegVal, 1746 /*IsSourceBuffer*/false, nullptr); 1747 1748 // Invalidate the source (const-invalidation without const-pointer-escaping 1749 // the address of the top-level region). 1750 state = InvalidateBuffer(C, state, srcExpr, srcVal, /*IsSourceBuffer*/true, 1751 nullptr); 1752 1753 // Set the C string length of the destination, if we know it. 1754 if (isBounded && !isAppending) { 1755 // strncpy is annoying in that it doesn't guarantee to null-terminate 1756 // the result string. If the original string didn't fit entirely inside 1757 // the bound (including the null-terminator), we don't know how long the 1758 // result is. 1759 if (amountCopied != strLength) 1760 finalStrLength = UnknownVal(); 1761 } 1762 state = setCStringLength(state, dstRegVal->getRegion(), finalStrLength); 1763 } 1764 1765 assert(state); 1766 1767 // If this is a stpcpy-style copy, but we were unable to check for a buffer 1768 // overflow, we still need a result. Conjure a return value. 1769 if (returnEnd && Result.isUnknown()) { 1770 Result = svalBuilder.conjureSymbolVal(nullptr, CE, LCtx, C.blockCount()); 1771 } 1772 1773 // Set the return value. 1774 state = state->BindExpr(CE, LCtx, Result); 1775 C.addTransition(state); 1776 } 1777 1778 void CStringChecker::evalStrcmp(CheckerContext &C, const CallExpr *CE) const { 1779 if (CE->getNumArgs() < 2) 1780 return; 1781 1782 //int strcmp(const char *s1, const char *s2); 1783 evalStrcmpCommon(C, CE, /* isBounded = */ false, /* ignoreCase = */ false); 1784 } 1785 1786 void CStringChecker::evalStrncmp(CheckerContext &C, const CallExpr *CE) const { 1787 if (CE->getNumArgs() < 3) 1788 return; 1789 1790 //int strncmp(const char *s1, const char *s2, size_t n); 1791 evalStrcmpCommon(C, CE, /* isBounded = */ true, /* ignoreCase = */ false); 1792 } 1793 1794 void CStringChecker::evalStrcasecmp(CheckerContext &C, 1795 const CallExpr *CE) const { 1796 if (CE->getNumArgs() < 2) 1797 return; 1798 1799 //int strcasecmp(const char *s1, const char *s2); 1800 evalStrcmpCommon(C, CE, /* isBounded = */ false, /* ignoreCase = */ true); 1801 } 1802 1803 void CStringChecker::evalStrncasecmp(CheckerContext &C, 1804 const CallExpr *CE) const { 1805 if (CE->getNumArgs() < 3) 1806 return; 1807 1808 //int strncasecmp(const char *s1, const char *s2, size_t n); 1809 evalStrcmpCommon(C, CE, /* isBounded = */ true, /* ignoreCase = */ true); 1810 } 1811 1812 void CStringChecker::evalStrcmpCommon(CheckerContext &C, const CallExpr *CE, 1813 bool isBounded, bool ignoreCase) const { 1814 CurrentFunctionDescription = "string comparison function"; 1815 ProgramStateRef state = C.getState(); 1816 const LocationContext *LCtx = C.getLocationContext(); 1817 1818 // Check that the first string is non-null 1819 const Expr *s1 = CE->getArg(0); 1820 SVal s1Val = state->getSVal(s1, LCtx); 1821 state = checkNonNull(C, state, s1, s1Val); 1822 if (!state) 1823 return; 1824 1825 // Check that the second string is non-null. 1826 const Expr *s2 = CE->getArg(1); 1827 SVal s2Val = state->getSVal(s2, LCtx); 1828 state = checkNonNull(C, state, s2, s2Val); 1829 if (!state) 1830 return; 1831 1832 // Get the string length of the first string or give up. 1833 SVal s1Length = getCStringLength(C, state, s1, s1Val); 1834 if (s1Length.isUndef()) 1835 return; 1836 1837 // Get the string length of the second string or give up. 1838 SVal s2Length = getCStringLength(C, state, s2, s2Val); 1839 if (s2Length.isUndef()) 1840 return; 1841 1842 // If we know the two buffers are the same, we know the result is 0. 1843 // First, get the two buffers' addresses. Another checker will have already 1844 // made sure they're not undefined. 1845 DefinedOrUnknownSVal LV = s1Val.castAs<DefinedOrUnknownSVal>(); 1846 DefinedOrUnknownSVal RV = s2Val.castAs<DefinedOrUnknownSVal>(); 1847 1848 // See if they are the same. 1849 SValBuilder &svalBuilder = C.getSValBuilder(); 1850 DefinedOrUnknownSVal SameBuf = svalBuilder.evalEQ(state, LV, RV); 1851 ProgramStateRef StSameBuf, StNotSameBuf; 1852 std::tie(StSameBuf, StNotSameBuf) = state->assume(SameBuf); 1853 1854 // If the two arguments might be the same buffer, we know the result is 0, 1855 // and we only need to check one size. 1856 if (StSameBuf) { 1857 StSameBuf = StSameBuf->BindExpr(CE, LCtx, 1858 svalBuilder.makeZeroVal(CE->getType())); 1859 C.addTransition(StSameBuf); 1860 1861 // If the two arguments are GUARANTEED to be the same, we're done! 1862 if (!StNotSameBuf) 1863 return; 1864 } 1865 1866 assert(StNotSameBuf); 1867 state = StNotSameBuf; 1868 1869 // At this point we can go about comparing the two buffers. 1870 // For now, we only do this if they're both known string literals. 1871 1872 // Attempt to extract string literals from both expressions. 1873 const StringLiteral *s1StrLiteral = getCStringLiteral(C, state, s1, s1Val); 1874 const StringLiteral *s2StrLiteral = getCStringLiteral(C, state, s2, s2Val); 1875 bool canComputeResult = false; 1876 SVal resultVal = svalBuilder.conjureSymbolVal(nullptr, CE, LCtx, 1877 C.blockCount()); 1878 1879 if (s1StrLiteral && s2StrLiteral) { 1880 StringRef s1StrRef = s1StrLiteral->getString(); 1881 StringRef s2StrRef = s2StrLiteral->getString(); 1882 1883 if (isBounded) { 1884 // Get the max number of characters to compare. 1885 const Expr *lenExpr = CE->getArg(2); 1886 SVal lenVal = state->getSVal(lenExpr, LCtx); 1887 1888 // If the length is known, we can get the right substrings. 1889 if (const llvm::APSInt *len = svalBuilder.getKnownValue(state, lenVal)) { 1890 // Create substrings of each to compare the prefix. 1891 s1StrRef = s1StrRef.substr(0, (size_t)len->getZExtValue()); 1892 s2StrRef = s2StrRef.substr(0, (size_t)len->getZExtValue()); 1893 canComputeResult = true; 1894 } 1895 } else { 1896 // This is a normal, unbounded strcmp. 1897 canComputeResult = true; 1898 } 1899 1900 if (canComputeResult) { 1901 // Real strcmp stops at null characters. 1902 size_t s1Term = s1StrRef.find('\0'); 1903 if (s1Term != StringRef::npos) 1904 s1StrRef = s1StrRef.substr(0, s1Term); 1905 1906 size_t s2Term = s2StrRef.find('\0'); 1907 if (s2Term != StringRef::npos) 1908 s2StrRef = s2StrRef.substr(0, s2Term); 1909 1910 // Use StringRef's comparison methods to compute the actual result. 1911 int compareRes = ignoreCase ? s1StrRef.compare_lower(s2StrRef) 1912 : s1StrRef.compare(s2StrRef); 1913 1914 // The strcmp function returns an integer greater than, equal to, or less 1915 // than zero, [c11, p7.24.4.2]. 1916 if (compareRes == 0) { 1917 resultVal = svalBuilder.makeIntVal(compareRes, CE->getType()); 1918 } 1919 else { 1920 DefinedSVal zeroVal = svalBuilder.makeIntVal(0, CE->getType()); 1921 // Constrain strcmp's result range based on the result of StringRef's 1922 // comparison methods. 1923 BinaryOperatorKind op = (compareRes == 1) ? BO_GT : BO_LT; 1924 SVal compareWithZero = 1925 svalBuilder.evalBinOp(state, op, resultVal, zeroVal, 1926 svalBuilder.getConditionType()); 1927 DefinedSVal compareWithZeroVal = compareWithZero.castAs<DefinedSVal>(); 1928 state = state->assume(compareWithZeroVal, true); 1929 } 1930 } 1931 } 1932 1933 state = state->BindExpr(CE, LCtx, resultVal); 1934 1935 // Record this as a possible path. 1936 C.addTransition(state); 1937 } 1938 1939 void CStringChecker::evalStrsep(CheckerContext &C, const CallExpr *CE) const { 1940 //char *strsep(char **stringp, const char *delim); 1941 if (CE->getNumArgs() < 2) 1942 return; 1943 1944 // Sanity: does the search string parameter match the return type? 1945 const Expr *SearchStrPtr = CE->getArg(0); 1946 QualType CharPtrTy = SearchStrPtr->getType()->getPointeeType(); 1947 if (CharPtrTy.isNull() || 1948 CE->getType().getUnqualifiedType() != CharPtrTy.getUnqualifiedType()) 1949 return; 1950 1951 CurrentFunctionDescription = "strsep()"; 1952 ProgramStateRef State = C.getState(); 1953 const LocationContext *LCtx = C.getLocationContext(); 1954 1955 // Check that the search string pointer is non-null (though it may point to 1956 // a null string). 1957 SVal SearchStrVal = State->getSVal(SearchStrPtr, LCtx); 1958 State = checkNonNull(C, State, SearchStrPtr, SearchStrVal); 1959 if (!State) 1960 return; 1961 1962 // Check that the delimiter string is non-null. 1963 const Expr *DelimStr = CE->getArg(1); 1964 SVal DelimStrVal = State->getSVal(DelimStr, LCtx); 1965 State = checkNonNull(C, State, DelimStr, DelimStrVal); 1966 if (!State) 1967 return; 1968 1969 SValBuilder &SVB = C.getSValBuilder(); 1970 SVal Result; 1971 if (Optional<Loc> SearchStrLoc = SearchStrVal.getAs<Loc>()) { 1972 // Get the current value of the search string pointer, as a char*. 1973 Result = State->getSVal(*SearchStrLoc, CharPtrTy); 1974 1975 // Invalidate the search string, representing the change of one delimiter 1976 // character to NUL. 1977 State = InvalidateBuffer(C, State, SearchStrPtr, Result, 1978 /*IsSourceBuffer*/false, nullptr); 1979 1980 // Overwrite the search string pointer. The new value is either an address 1981 // further along in the same string, or NULL if there are no more tokens. 1982 State = State->bindLoc(*SearchStrLoc, 1983 SVB.conjureSymbolVal(getTag(), 1984 CE, 1985 LCtx, 1986 CharPtrTy, 1987 C.blockCount()), 1988 LCtx); 1989 } else { 1990 assert(SearchStrVal.isUnknown()); 1991 // Conjure a symbolic value. It's the best we can do. 1992 Result = SVB.conjureSymbolVal(nullptr, CE, LCtx, C.blockCount()); 1993 } 1994 1995 // Set the return value, and finish. 1996 State = State->BindExpr(CE, LCtx, Result); 1997 C.addTransition(State); 1998 } 1999 2000 // These should probably be moved into a C++ standard library checker. 2001 void CStringChecker::evalStdCopy(CheckerContext &C, const CallExpr *CE) const { 2002 evalStdCopyCommon(C, CE); 2003 } 2004 2005 void CStringChecker::evalStdCopyBackward(CheckerContext &C, 2006 const CallExpr *CE) const { 2007 evalStdCopyCommon(C, CE); 2008 } 2009 2010 void CStringChecker::evalStdCopyCommon(CheckerContext &C, 2011 const CallExpr *CE) const { 2012 if (CE->getNumArgs() < 3) 2013 return; 2014 2015 ProgramStateRef State = C.getState(); 2016 2017 const LocationContext *LCtx = C.getLocationContext(); 2018 2019 // template <class _InputIterator, class _OutputIterator> 2020 // _OutputIterator 2021 // copy(_InputIterator __first, _InputIterator __last, 2022 // _OutputIterator __result) 2023 2024 // Invalidate the destination buffer 2025 const Expr *Dst = CE->getArg(2); 2026 SVal DstVal = State->getSVal(Dst, LCtx); 2027 State = InvalidateBuffer(C, State, Dst, DstVal, /*IsSource=*/false, 2028 /*Size=*/nullptr); 2029 2030 SValBuilder &SVB = C.getSValBuilder(); 2031 2032 SVal ResultVal = SVB.conjureSymbolVal(nullptr, CE, LCtx, C.blockCount()); 2033 State = State->BindExpr(CE, LCtx, ResultVal); 2034 2035 C.addTransition(State); 2036 } 2037 2038 void CStringChecker::evalMemset(CheckerContext &C, const CallExpr *CE) const { 2039 if (CE->getNumArgs() != 3) 2040 return; 2041 2042 CurrentFunctionDescription = "memory set function"; 2043 2044 const Expr *Mem = CE->getArg(0); 2045 const Expr *Size = CE->getArg(2); 2046 ProgramStateRef State = C.getState(); 2047 2048 // See if the size argument is zero. 2049 const LocationContext *LCtx = C.getLocationContext(); 2050 SVal SizeVal = State->getSVal(Size, LCtx); 2051 QualType SizeTy = Size->getType(); 2052 2053 ProgramStateRef StateZeroSize, StateNonZeroSize; 2054 std::tie(StateZeroSize, StateNonZeroSize) = 2055 assumeZero(C, State, SizeVal, SizeTy); 2056 2057 // Get the value of the memory area. 2058 SVal MemVal = State->getSVal(Mem, LCtx); 2059 2060 // If the size is zero, there won't be any actual memory access, so 2061 // just bind the return value to the Mem buffer and return. 2062 if (StateZeroSize && !StateNonZeroSize) { 2063 StateZeroSize = StateZeroSize->BindExpr(CE, LCtx, MemVal); 2064 C.addTransition(StateZeroSize); 2065 return; 2066 } 2067 2068 // Ensure the memory area is not null. 2069 // If it is NULL there will be a NULL pointer dereference. 2070 State = checkNonNull(C, StateNonZeroSize, Mem, MemVal); 2071 if (!State) 2072 return; 2073 2074 State = CheckBufferAccess(C, State, Size, Mem); 2075 if (!State) 2076 return; 2077 State = InvalidateBuffer(C, State, Mem, C.getSVal(Mem), 2078 /*IsSourceBuffer*/false, Size); 2079 if (!State) 2080 return; 2081 2082 State = State->BindExpr(CE, LCtx, MemVal); 2083 C.addTransition(State); 2084 } 2085 2086 static bool isCPPStdLibraryFunction(const FunctionDecl *FD, StringRef Name) { 2087 IdentifierInfo *II = FD->getIdentifier(); 2088 if (!II) 2089 return false; 2090 2091 if (!AnalysisDeclContext::isInStdNamespace(FD)) 2092 return false; 2093 2094 if (II->getName().equals(Name)) 2095 return true; 2096 2097 return false; 2098 } 2099 //===----------------------------------------------------------------------===// 2100 // The driver method, and other Checker callbacks. 2101 //===----------------------------------------------------------------------===// 2102 2103 bool CStringChecker::evalCall(const CallExpr *CE, CheckerContext &C) const { 2104 const FunctionDecl *FDecl = C.getCalleeDecl(CE); 2105 2106 if (!FDecl) 2107 return false; 2108 2109 // FIXME: Poorly-factored string switches are slow. 2110 FnCheck evalFunction = nullptr; 2111 if (C.isCLibraryFunction(FDecl, "memcpy")) 2112 evalFunction = &CStringChecker::evalMemcpy; 2113 else if (C.isCLibraryFunction(FDecl, "mempcpy")) 2114 evalFunction = &CStringChecker::evalMempcpy; 2115 else if (C.isCLibraryFunction(FDecl, "memcmp")) 2116 evalFunction = &CStringChecker::evalMemcmp; 2117 else if (C.isCLibraryFunction(FDecl, "memmove")) 2118 evalFunction = &CStringChecker::evalMemmove; 2119 else if (C.isCLibraryFunction(FDecl, "memset")) 2120 evalFunction = &CStringChecker::evalMemset; 2121 else if (C.isCLibraryFunction(FDecl, "strcpy")) 2122 evalFunction = &CStringChecker::evalStrcpy; 2123 else if (C.isCLibraryFunction(FDecl, "strncpy")) 2124 evalFunction = &CStringChecker::evalStrncpy; 2125 else if (C.isCLibraryFunction(FDecl, "stpcpy")) 2126 evalFunction = &CStringChecker::evalStpcpy; 2127 else if (C.isCLibraryFunction(FDecl, "strlcpy")) 2128 evalFunction = &CStringChecker::evalStrlcpy; 2129 else if (C.isCLibraryFunction(FDecl, "strcat")) 2130 evalFunction = &CStringChecker::evalStrcat; 2131 else if (C.isCLibraryFunction(FDecl, "strncat")) 2132 evalFunction = &CStringChecker::evalStrncat; 2133 else if (C.isCLibraryFunction(FDecl, "strlcat")) 2134 evalFunction = &CStringChecker::evalStrlcat; 2135 else if (C.isCLibraryFunction(FDecl, "strlen")) 2136 evalFunction = &CStringChecker::evalstrLength; 2137 else if (C.isCLibraryFunction(FDecl, "strnlen")) 2138 evalFunction = &CStringChecker::evalstrnLength; 2139 else if (C.isCLibraryFunction(FDecl, "strcmp")) 2140 evalFunction = &CStringChecker::evalStrcmp; 2141 else if (C.isCLibraryFunction(FDecl, "strncmp")) 2142 evalFunction = &CStringChecker::evalStrncmp; 2143 else if (C.isCLibraryFunction(FDecl, "strcasecmp")) 2144 evalFunction = &CStringChecker::evalStrcasecmp; 2145 else if (C.isCLibraryFunction(FDecl, "strncasecmp")) 2146 evalFunction = &CStringChecker::evalStrncasecmp; 2147 else if (C.isCLibraryFunction(FDecl, "strsep")) 2148 evalFunction = &CStringChecker::evalStrsep; 2149 else if (C.isCLibraryFunction(FDecl, "bcopy")) 2150 evalFunction = &CStringChecker::evalBcopy; 2151 else if (C.isCLibraryFunction(FDecl, "bcmp")) 2152 evalFunction = &CStringChecker::evalMemcmp; 2153 else if (isCPPStdLibraryFunction(FDecl, "copy")) 2154 evalFunction = &CStringChecker::evalStdCopy; 2155 else if (isCPPStdLibraryFunction(FDecl, "copy_backward")) 2156 evalFunction = &CStringChecker::evalStdCopyBackward; 2157 2158 // If the callee isn't a string function, let another checker handle it. 2159 if (!evalFunction) 2160 return false; 2161 2162 // Check and evaluate the call. 2163 (this->*evalFunction)(C, CE); 2164 2165 // If the evaluate call resulted in no change, chain to the next eval call 2166 // handler. 2167 // Note, the custom CString evaluation calls assume that basic safety 2168 // properties are held. However, if the user chooses to turn off some of these 2169 // checks, we ignore the issues and leave the call evaluation to a generic 2170 // handler. 2171 return C.isDifferent(); 2172 } 2173 2174 void CStringChecker::checkPreStmt(const DeclStmt *DS, CheckerContext &C) const { 2175 // Record string length for char a[] = "abc"; 2176 ProgramStateRef state = C.getState(); 2177 2178 for (const auto *I : DS->decls()) { 2179 const VarDecl *D = dyn_cast<VarDecl>(I); 2180 if (!D) 2181 continue; 2182 2183 // FIXME: Handle array fields of structs. 2184 if (!D->getType()->isArrayType()) 2185 continue; 2186 2187 const Expr *Init = D->getInit(); 2188 if (!Init) 2189 continue; 2190 if (!isa<StringLiteral>(Init)) 2191 continue; 2192 2193 Loc VarLoc = state->getLValue(D, C.getLocationContext()); 2194 const MemRegion *MR = VarLoc.getAsRegion(); 2195 if (!MR) 2196 continue; 2197 2198 SVal StrVal = C.getSVal(Init); 2199 assert(StrVal.isValid() && "Initializer string is unknown or undefined"); 2200 DefinedOrUnknownSVal strLength = 2201 getCStringLength(C, state, Init, StrVal).castAs<DefinedOrUnknownSVal>(); 2202 2203 state = state->set<CStringLength>(MR, strLength); 2204 } 2205 2206 C.addTransition(state); 2207 } 2208 2209 ProgramStateRef 2210 CStringChecker::checkRegionChanges(ProgramStateRef state, 2211 const InvalidatedSymbols *, 2212 ArrayRef<const MemRegion *> ExplicitRegions, 2213 ArrayRef<const MemRegion *> Regions, 2214 const LocationContext *LCtx, 2215 const CallEvent *Call) const { 2216 CStringLengthTy Entries = state->get<CStringLength>(); 2217 if (Entries.isEmpty()) 2218 return state; 2219 2220 llvm::SmallPtrSet<const MemRegion *, 8> Invalidated; 2221 llvm::SmallPtrSet<const MemRegion *, 32> SuperRegions; 2222 2223 // First build sets for the changed regions and their super-regions. 2224 for (ArrayRef<const MemRegion *>::iterator 2225 I = Regions.begin(), E = Regions.end(); I != E; ++I) { 2226 const MemRegion *MR = *I; 2227 Invalidated.insert(MR); 2228 2229 SuperRegions.insert(MR); 2230 while (const SubRegion *SR = dyn_cast<SubRegion>(MR)) { 2231 MR = SR->getSuperRegion(); 2232 SuperRegions.insert(MR); 2233 } 2234 } 2235 2236 CStringLengthTy::Factory &F = state->get_context<CStringLength>(); 2237 2238 // Then loop over the entries in the current state. 2239 for (CStringLengthTy::iterator I = Entries.begin(), 2240 E = Entries.end(); I != E; ++I) { 2241 const MemRegion *MR = I.getKey(); 2242 2243 // Is this entry for a super-region of a changed region? 2244 if (SuperRegions.count(MR)) { 2245 Entries = F.remove(Entries, MR); 2246 continue; 2247 } 2248 2249 // Is this entry for a sub-region of a changed region? 2250 const MemRegion *Super = MR; 2251 while (const SubRegion *SR = dyn_cast<SubRegion>(Super)) { 2252 Super = SR->getSuperRegion(); 2253 if (Invalidated.count(Super)) { 2254 Entries = F.remove(Entries, MR); 2255 break; 2256 } 2257 } 2258 } 2259 2260 return state->set<CStringLength>(Entries); 2261 } 2262 2263 void CStringChecker::checkLiveSymbols(ProgramStateRef state, 2264 SymbolReaper &SR) const { 2265 // Mark all symbols in our string length map as valid. 2266 CStringLengthTy Entries = state->get<CStringLength>(); 2267 2268 for (CStringLengthTy::iterator I = Entries.begin(), E = Entries.end(); 2269 I != E; ++I) { 2270 SVal Len = I.getData(); 2271 2272 for (SymExpr::symbol_iterator si = Len.symbol_begin(), 2273 se = Len.symbol_end(); si != se; ++si) 2274 SR.markInUse(*si); 2275 } 2276 } 2277 2278 void CStringChecker::checkDeadSymbols(SymbolReaper &SR, 2279 CheckerContext &C) const { 2280 if (!SR.hasDeadSymbols()) 2281 return; 2282 2283 ProgramStateRef state = C.getState(); 2284 CStringLengthTy Entries = state->get<CStringLength>(); 2285 if (Entries.isEmpty()) 2286 return; 2287 2288 CStringLengthTy::Factory &F = state->get_context<CStringLength>(); 2289 for (CStringLengthTy::iterator I = Entries.begin(), E = Entries.end(); 2290 I != E; ++I) { 2291 SVal Len = I.getData(); 2292 if (SymbolRef Sym = Len.getAsSymbol()) { 2293 if (SR.isDead(Sym)) 2294 Entries = F.remove(Entries, I.getKey()); 2295 } 2296 } 2297 2298 state = state->set<CStringLength>(Entries); 2299 C.addTransition(state); 2300 } 2301 2302 #define REGISTER_CHECKER(name) \ 2303 void ento::register##name(CheckerManager &mgr) { \ 2304 CStringChecker *checker = mgr.registerChecker<CStringChecker>(); \ 2305 checker->Filter.Check##name = true; \ 2306 checker->Filter.CheckName##name = mgr.getCurrentCheckName(); \ 2307 } 2308 2309 REGISTER_CHECKER(CStringNullArg) 2310 REGISTER_CHECKER(CStringOutOfBounds) 2311 REGISTER_CHECKER(CStringBufferOverlap) 2312 REGISTER_CHECKER(CStringNotNullTerm) 2313 2314 void ento::registerCStringCheckerBasic(CheckerManager &Mgr) { 2315 registerCStringNullArg(Mgr); 2316 } 2317