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