CStringChecker.cpp revision 5fc1d0c4532c55cc47ba6628f296bf5b86d2eaf0
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/StaticAnalyzer/Core/Checker.h" 18#include "clang/StaticAnalyzer/Core/CheckerManager.h" 19#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" 20#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" 21#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramStateTrait.h" 22#include "llvm/ADT/SmallString.h" 23#include "llvm/ADT/STLExtras.h" 24#include "llvm/ADT/StringSwitch.h" 25 26using namespace clang; 27using namespace ento; 28 29namespace { 30class CStringChecker : public Checker< eval::Call, 31 check::PreStmt<DeclStmt>, 32 check::LiveSymbols, 33 check::DeadSymbols, 34 check::RegionChanges 35 > { 36 mutable OwningPtr<BugType> BT_Null, 37 BT_Bounds, 38 BT_Overlap, 39 BT_NotCString, 40 BT_AdditionOverflow; 41 42 mutable const char *CurrentFunctionDescription; 43 44public: 45 /// The filter is used to filter out the diagnostics which are not enabled by 46 /// the user. 47 struct CStringChecksFilter { 48 DefaultBool CheckCStringNullArg; 49 DefaultBool CheckCStringOutOfBounds; 50 DefaultBool CheckCStringBufferOverlap; 51 DefaultBool CheckCStringNotNullTerm; 52 }; 53 54 CStringChecksFilter Filter; 55 56 static void *getTag() { static int tag; return &tag; } 57 58 bool evalCall(const CallExpr *CE, CheckerContext &C) const; 59 void checkPreStmt(const DeclStmt *DS, CheckerContext &C) const; 60 void checkLiveSymbols(ProgramStateRef state, SymbolReaper &SR) const; 61 void checkDeadSymbols(SymbolReaper &SR, CheckerContext &C) const; 62 bool wantsRegionChangeUpdate(ProgramStateRef state) const; 63 64 ProgramStateRef 65 checkRegionChanges(ProgramStateRef state, 66 const StoreManager::InvalidatedSymbols *, 67 ArrayRef<const MemRegion *> ExplicitRegions, 68 ArrayRef<const MemRegion *> Regions, 69 const CallEvent *Call) const; 70 71 typedef void (CStringChecker::*FnCheck)(CheckerContext &, 72 const CallExpr *) const; 73 74 void evalMemcpy(CheckerContext &C, const CallExpr *CE) const; 75 void evalMempcpy(CheckerContext &C, const CallExpr *CE) const; 76 void evalMemmove(CheckerContext &C, const CallExpr *CE) const; 77 void evalBcopy(CheckerContext &C, const CallExpr *CE) const; 78 void evalCopyCommon(CheckerContext &C, const CallExpr *CE, 79 ProgramStateRef state, 80 const Expr *Size, 81 const Expr *Source, 82 const Expr *Dest, 83 bool Restricted = false, 84 bool IsMempcpy = false) const; 85 86 void evalMemcmp(CheckerContext &C, const CallExpr *CE) const; 87 88 void evalstrLength(CheckerContext &C, const CallExpr *CE) const; 89 void evalstrnLength(CheckerContext &C, const CallExpr *CE) const; 90 void evalstrLengthCommon(CheckerContext &C, 91 const CallExpr *CE, 92 bool IsStrnlen = false) const; 93 94 void evalStrcpy(CheckerContext &C, const CallExpr *CE) const; 95 void evalStrncpy(CheckerContext &C, const CallExpr *CE) const; 96 void evalStpcpy(CheckerContext &C, const CallExpr *CE) const; 97 void evalStrcpyCommon(CheckerContext &C, 98 const CallExpr *CE, 99 bool returnEnd, 100 bool isBounded, 101 bool isAppending) const; 102 103 void evalStrcat(CheckerContext &C, const CallExpr *CE) const; 104 void evalStrncat(CheckerContext &C, const CallExpr *CE) const; 105 106 void evalStrcmp(CheckerContext &C, const CallExpr *CE) const; 107 void evalStrncmp(CheckerContext &C, const CallExpr *CE) const; 108 void evalStrcasecmp(CheckerContext &C, const CallExpr *CE) const; 109 void evalStrncasecmp(CheckerContext &C, const CallExpr *CE) const; 110 void evalStrcmpCommon(CheckerContext &C, 111 const CallExpr *CE, 112 bool isBounded = false, 113 bool ignoreCase = false) const; 114 115 // Utility methods 116 std::pair<ProgramStateRef , ProgramStateRef > 117 static assumeZero(CheckerContext &C, 118 ProgramStateRef state, SVal V, QualType Ty); 119 120 static ProgramStateRef setCStringLength(ProgramStateRef state, 121 const MemRegion *MR, 122 SVal strLength); 123 static SVal getCStringLengthForRegion(CheckerContext &C, 124 ProgramStateRef &state, 125 const Expr *Ex, 126 const MemRegion *MR, 127 bool hypothetical); 128 SVal getCStringLength(CheckerContext &C, 129 ProgramStateRef &state, 130 const Expr *Ex, 131 SVal Buf, 132 bool hypothetical = false) const; 133 134 const StringLiteral *getCStringLiteral(CheckerContext &C, 135 ProgramStateRef &state, 136 const Expr *expr, 137 SVal val) const; 138 139 static ProgramStateRef InvalidateBuffer(CheckerContext &C, 140 ProgramStateRef state, 141 const Expr *Ex, SVal V); 142 143 static bool SummarizeRegion(raw_ostream &os, ASTContext &Ctx, 144 const MemRegion *MR); 145 146 // Re-usable checks 147 ProgramStateRef checkNonNull(CheckerContext &C, 148 ProgramStateRef state, 149 const Expr *S, 150 SVal l) const; 151 ProgramStateRef CheckLocation(CheckerContext &C, 152 ProgramStateRef state, 153 const Expr *S, 154 SVal l, 155 const char *message = NULL) const; 156 ProgramStateRef CheckBufferAccess(CheckerContext &C, 157 ProgramStateRef state, 158 const Expr *Size, 159 const Expr *FirstBuf, 160 const Expr *SecondBuf, 161 const char *firstMessage = NULL, 162 const char *secondMessage = NULL, 163 bool WarnAboutSize = false) const; 164 165 ProgramStateRef CheckBufferAccess(CheckerContext &C, 166 ProgramStateRef state, 167 const Expr *Size, 168 const Expr *Buf, 169 const char *message = NULL, 170 bool WarnAboutSize = false) const { 171 // This is a convenience override. 172 return CheckBufferAccess(C, state, Size, Buf, NULL, message, NULL, 173 WarnAboutSize); 174 } 175 ProgramStateRef CheckOverlap(CheckerContext &C, 176 ProgramStateRef state, 177 const Expr *Size, 178 const Expr *First, 179 const Expr *Second) const; 180 void emitOverlapBug(CheckerContext &C, 181 ProgramStateRef state, 182 const Stmt *First, 183 const Stmt *Second) const; 184 185 ProgramStateRef checkAdditionOverflow(CheckerContext &C, 186 ProgramStateRef state, 187 NonLoc left, 188 NonLoc right) const; 189}; 190 191class CStringLength { 192public: 193 typedef llvm::ImmutableMap<const MemRegion *, SVal> EntryMap; 194}; 195} //end anonymous namespace 196 197namespace clang { 198namespace ento { 199 template <> 200 struct ProgramStateTrait<CStringLength> 201 : public ProgramStatePartialTrait<CStringLength::EntryMap> { 202 static void *GDMIndex() { return CStringChecker::getTag(); } 203 }; 204} 205} 206 207//===----------------------------------------------------------------------===// 208// Individual checks and utility methods. 209//===----------------------------------------------------------------------===// 210 211std::pair<ProgramStateRef , ProgramStateRef > 212CStringChecker::assumeZero(CheckerContext &C, ProgramStateRef state, SVal V, 213 QualType Ty) { 214 DefinedSVal *val = dyn_cast<DefinedSVal>(&V); 215 if (!val) 216 return std::pair<ProgramStateRef , ProgramStateRef >(state, state); 217 218 SValBuilder &svalBuilder = C.getSValBuilder(); 219 DefinedOrUnknownSVal zero = svalBuilder.makeZeroVal(Ty); 220 return state->assume(svalBuilder.evalEQ(state, *val, zero)); 221} 222 223ProgramStateRef CStringChecker::checkNonNull(CheckerContext &C, 224 ProgramStateRef state, 225 const Expr *S, SVal l) const { 226 // If a previous check has failed, propagate the failure. 227 if (!state) 228 return NULL; 229 230 ProgramStateRef stateNull, stateNonNull; 231 llvm::tie(stateNull, stateNonNull) = assumeZero(C, state, l, S->getType()); 232 233 if (stateNull && !stateNonNull) { 234 if (!Filter.CheckCStringNullArg) 235 return NULL; 236 237 ExplodedNode *N = C.generateSink(stateNull); 238 if (!N) 239 return NULL; 240 241 if (!BT_Null) 242 BT_Null.reset(new BuiltinBug("Unix API", 243 "Null pointer argument in call to byte string function")); 244 245 SmallString<80> buf; 246 llvm::raw_svector_ostream os(buf); 247 assert(CurrentFunctionDescription); 248 os << "Null pointer argument in call to " << CurrentFunctionDescription; 249 250 // Generate a report for this bug. 251 BuiltinBug *BT = static_cast<BuiltinBug*>(BT_Null.get()); 252 BugReport *report = new BugReport(*BT, os.str(), N); 253 254 report->addRange(S->getSourceRange()); 255 bugreporter::trackNullOrUndefValue(N, S, *report); 256 C.EmitReport(report); 257 return NULL; 258 } 259 260 // From here on, assume that the value is non-null. 261 assert(stateNonNull); 262 return stateNonNull; 263} 264 265// FIXME: This was originally copied from ArrayBoundChecker.cpp. Refactor? 266ProgramStateRef CStringChecker::CheckLocation(CheckerContext &C, 267 ProgramStateRef state, 268 const Expr *S, SVal l, 269 const char *warningMsg) const { 270 // If a previous check has failed, propagate the failure. 271 if (!state) 272 return NULL; 273 274 // Check for out of bound array element access. 275 const MemRegion *R = l.getAsRegion(); 276 if (!R) 277 return state; 278 279 const ElementRegion *ER = dyn_cast<ElementRegion>(R); 280 if (!ER) 281 return state; 282 283 assert(ER->getValueType() == C.getASTContext().CharTy && 284 "CheckLocation should only be called with char* ElementRegions"); 285 286 // Get the size of the array. 287 const SubRegion *superReg = cast<SubRegion>(ER->getSuperRegion()); 288 SValBuilder &svalBuilder = C.getSValBuilder(); 289 SVal Extent = 290 svalBuilder.convertToArrayIndex(superReg->getExtent(svalBuilder)); 291 DefinedOrUnknownSVal Size = cast<DefinedOrUnknownSVal>(Extent); 292 293 // Get the index of the accessed element. 294 DefinedOrUnknownSVal Idx = cast<DefinedOrUnknownSVal>(ER->getIndex()); 295 296 ProgramStateRef StInBound = state->assumeInBound(Idx, Size, true); 297 ProgramStateRef StOutBound = state->assumeInBound(Idx, Size, false); 298 if (StOutBound && !StInBound) { 299 ExplodedNode *N = C.generateSink(StOutBound); 300 if (!N) 301 return NULL; 302 303 if (!BT_Bounds) { 304 BT_Bounds.reset(new BuiltinBug("Out-of-bound array access", 305 "Byte string function accesses out-of-bound array element")); 306 } 307 BuiltinBug *BT = static_cast<BuiltinBug*>(BT_Bounds.get()); 308 309 // Generate a report for this bug. 310 BugReport *report; 311 if (warningMsg) { 312 report = new BugReport(*BT, warningMsg, N); 313 } else { 314 assert(CurrentFunctionDescription); 315 assert(CurrentFunctionDescription[0] != '\0'); 316 317 SmallString<80> buf; 318 llvm::raw_svector_ostream os(buf); 319 os << (char)toupper(CurrentFunctionDescription[0]) 320 << &CurrentFunctionDescription[1] 321 << " accesses out-of-bound array element"; 322 report = new BugReport(*BT, os.str(), N); 323 } 324 325 // FIXME: It would be nice to eventually make this diagnostic more clear, 326 // e.g., by referencing the original declaration or by saying *why* this 327 // reference is outside the range. 328 329 report->addRange(S->getSourceRange()); 330 C.EmitReport(report); 331 return NULL; 332 } 333 334 // Array bound check succeeded. From this point forward the array bound 335 // should always succeed. 336 return StInBound; 337} 338 339ProgramStateRef CStringChecker::CheckBufferAccess(CheckerContext &C, 340 ProgramStateRef state, 341 const Expr *Size, 342 const Expr *FirstBuf, 343 const Expr *SecondBuf, 344 const char *firstMessage, 345 const char *secondMessage, 346 bool WarnAboutSize) const { 347 // If a previous check has failed, propagate the failure. 348 if (!state) 349 return NULL; 350 351 SValBuilder &svalBuilder = C.getSValBuilder(); 352 ASTContext &Ctx = svalBuilder.getContext(); 353 const LocationContext *LCtx = C.getLocationContext(); 354 355 QualType sizeTy = Size->getType(); 356 QualType PtrTy = Ctx.getPointerType(Ctx.CharTy); 357 358 // Check that the first buffer is non-null. 359 SVal BufVal = state->getSVal(FirstBuf, LCtx); 360 state = checkNonNull(C, state, FirstBuf, BufVal); 361 if (!state) 362 return NULL; 363 364 // If out-of-bounds checking is turned off, skip the rest. 365 if (!Filter.CheckCStringOutOfBounds) 366 return state; 367 368 // Get the access length and make sure it is known. 369 // FIXME: This assumes the caller has already checked that the access length 370 // is positive. And that it's unsigned. 371 SVal LengthVal = state->getSVal(Size, LCtx); 372 NonLoc *Length = dyn_cast<NonLoc>(&LengthVal); 373 if (!Length) 374 return state; 375 376 // Compute the offset of the last element to be accessed: size-1. 377 NonLoc One = cast<NonLoc>(svalBuilder.makeIntVal(1, sizeTy)); 378 NonLoc LastOffset = cast<NonLoc>(svalBuilder.evalBinOpNN(state, BO_Sub, 379 *Length, One, sizeTy)); 380 381 // Check that the first buffer is sufficiently long. 382 SVal BufStart = svalBuilder.evalCast(BufVal, PtrTy, FirstBuf->getType()); 383 if (Loc *BufLoc = dyn_cast<Loc>(&BufStart)) { 384 const Expr *warningExpr = (WarnAboutSize ? Size : FirstBuf); 385 386 SVal BufEnd = svalBuilder.evalBinOpLN(state, BO_Add, *BufLoc, 387 LastOffset, PtrTy); 388 state = CheckLocation(C, state, warningExpr, BufEnd, firstMessage); 389 390 // If the buffer isn't large enough, abort. 391 if (!state) 392 return NULL; 393 } 394 395 // If there's a second buffer, check it as well. 396 if (SecondBuf) { 397 BufVal = state->getSVal(SecondBuf, LCtx); 398 state = checkNonNull(C, state, SecondBuf, BufVal); 399 if (!state) 400 return NULL; 401 402 BufStart = svalBuilder.evalCast(BufVal, PtrTy, SecondBuf->getType()); 403 if (Loc *BufLoc = dyn_cast<Loc>(&BufStart)) { 404 const Expr *warningExpr = (WarnAboutSize ? Size : SecondBuf); 405 406 SVal BufEnd = svalBuilder.evalBinOpLN(state, BO_Add, *BufLoc, 407 LastOffset, PtrTy); 408 state = CheckLocation(C, state, warningExpr, BufEnd, secondMessage); 409 } 410 } 411 412 // Large enough or not, return this state! 413 return state; 414} 415 416ProgramStateRef CStringChecker::CheckOverlap(CheckerContext &C, 417 ProgramStateRef state, 418 const Expr *Size, 419 const Expr *First, 420 const Expr *Second) const { 421 if (!Filter.CheckCStringBufferOverlap) 422 return state; 423 424 // Do a simple check for overlap: if the two arguments are from the same 425 // buffer, see if the end of the first is greater than the start of the second 426 // or vice versa. 427 428 // If a previous check has failed, propagate the failure. 429 if (!state) 430 return NULL; 431 432 ProgramStateRef stateTrue, stateFalse; 433 434 // Get the buffer values and make sure they're known locations. 435 const LocationContext *LCtx = C.getLocationContext(); 436 SVal firstVal = state->getSVal(First, LCtx); 437 SVal secondVal = state->getSVal(Second, LCtx); 438 439 Loc *firstLoc = dyn_cast<Loc>(&firstVal); 440 if (!firstLoc) 441 return state; 442 443 Loc *secondLoc = dyn_cast<Loc>(&secondVal); 444 if (!secondLoc) 445 return state; 446 447 // Are the two values the same? 448 SValBuilder &svalBuilder = C.getSValBuilder(); 449 llvm::tie(stateTrue, stateFalse) = 450 state->assume(svalBuilder.evalEQ(state, *firstLoc, *secondLoc)); 451 452 if (stateTrue && !stateFalse) { 453 // If the values are known to be equal, that's automatically an overlap. 454 emitOverlapBug(C, stateTrue, First, Second); 455 return NULL; 456 } 457 458 // assume the two expressions are not equal. 459 assert(stateFalse); 460 state = stateFalse; 461 462 // Which value comes first? 463 QualType cmpTy = svalBuilder.getConditionType(); 464 SVal reverse = svalBuilder.evalBinOpLL(state, BO_GT, 465 *firstLoc, *secondLoc, cmpTy); 466 DefinedOrUnknownSVal *reverseTest = dyn_cast<DefinedOrUnknownSVal>(&reverse); 467 if (!reverseTest) 468 return state; 469 470 llvm::tie(stateTrue, stateFalse) = state->assume(*reverseTest); 471 if (stateTrue) { 472 if (stateFalse) { 473 // If we don't know which one comes first, we can't perform this test. 474 return state; 475 } else { 476 // Switch the values so that firstVal is before secondVal. 477 Loc *tmpLoc = firstLoc; 478 firstLoc = secondLoc; 479 secondLoc = tmpLoc; 480 481 // Switch the Exprs as well, so that they still correspond. 482 const Expr *tmpExpr = First; 483 First = Second; 484 Second = tmpExpr; 485 } 486 } 487 488 // Get the length, and make sure it too is known. 489 SVal LengthVal = state->getSVal(Size, LCtx); 490 NonLoc *Length = dyn_cast<NonLoc>(&LengthVal); 491 if (!Length) 492 return state; 493 494 // Convert the first buffer's start address to char*. 495 // Bail out if the cast fails. 496 ASTContext &Ctx = svalBuilder.getContext(); 497 QualType CharPtrTy = Ctx.getPointerType(Ctx.CharTy); 498 SVal FirstStart = svalBuilder.evalCast(*firstLoc, CharPtrTy, 499 First->getType()); 500 Loc *FirstStartLoc = dyn_cast<Loc>(&FirstStart); 501 if (!FirstStartLoc) 502 return state; 503 504 // Compute the end of the first buffer. Bail out if THAT fails. 505 SVal FirstEnd = svalBuilder.evalBinOpLN(state, BO_Add, 506 *FirstStartLoc, *Length, CharPtrTy); 507 Loc *FirstEndLoc = dyn_cast<Loc>(&FirstEnd); 508 if (!FirstEndLoc) 509 return state; 510 511 // Is the end of the first buffer past the start of the second buffer? 512 SVal Overlap = svalBuilder.evalBinOpLL(state, BO_GT, 513 *FirstEndLoc, *secondLoc, cmpTy); 514 DefinedOrUnknownSVal *OverlapTest = dyn_cast<DefinedOrUnknownSVal>(&Overlap); 515 if (!OverlapTest) 516 return state; 517 518 llvm::tie(stateTrue, stateFalse) = state->assume(*OverlapTest); 519 520 if (stateTrue && !stateFalse) { 521 // Overlap! 522 emitOverlapBug(C, stateTrue, First, Second); 523 return NULL; 524 } 525 526 // assume the two expressions don't overlap. 527 assert(stateFalse); 528 return stateFalse; 529} 530 531void CStringChecker::emitOverlapBug(CheckerContext &C, ProgramStateRef state, 532 const Stmt *First, const Stmt *Second) const { 533 ExplodedNode *N = C.generateSink(state); 534 if (!N) 535 return; 536 537 if (!BT_Overlap) 538 BT_Overlap.reset(new BugType("Unix API", "Improper arguments")); 539 540 // Generate a report for this bug. 541 BugReport *report = 542 new BugReport(*BT_Overlap, 543 "Arguments must not be overlapping buffers", N); 544 report->addRange(First->getSourceRange()); 545 report->addRange(Second->getSourceRange()); 546 547 C.EmitReport(report); 548} 549 550ProgramStateRef CStringChecker::checkAdditionOverflow(CheckerContext &C, 551 ProgramStateRef state, 552 NonLoc left, 553 NonLoc right) const { 554 // If out-of-bounds checking is turned off, skip the rest. 555 if (!Filter.CheckCStringOutOfBounds) 556 return state; 557 558 // If a previous check has failed, propagate the failure. 559 if (!state) 560 return NULL; 561 562 SValBuilder &svalBuilder = C.getSValBuilder(); 563 BasicValueFactory &BVF = svalBuilder.getBasicValueFactory(); 564 565 QualType sizeTy = svalBuilder.getContext().getSizeType(); 566 const llvm::APSInt &maxValInt = BVF.getMaxValue(sizeTy); 567 NonLoc maxVal = svalBuilder.makeIntVal(maxValInt); 568 569 SVal maxMinusRight; 570 if (isa<nonloc::ConcreteInt>(right)) { 571 maxMinusRight = svalBuilder.evalBinOpNN(state, BO_Sub, maxVal, right, 572 sizeTy); 573 } else { 574 // Try switching the operands. (The order of these two assignments is 575 // important!) 576 maxMinusRight = svalBuilder.evalBinOpNN(state, BO_Sub, maxVal, left, 577 sizeTy); 578 left = right; 579 } 580 581 if (NonLoc *maxMinusRightNL = dyn_cast<NonLoc>(&maxMinusRight)) { 582 QualType cmpTy = svalBuilder.getConditionType(); 583 // If left > max - right, we have an overflow. 584 SVal willOverflow = svalBuilder.evalBinOpNN(state, BO_GT, left, 585 *maxMinusRightNL, cmpTy); 586 587 ProgramStateRef stateOverflow, stateOkay; 588 llvm::tie(stateOverflow, stateOkay) = 589 state->assume(cast<DefinedOrUnknownSVal>(willOverflow)); 590 591 if (stateOverflow && !stateOkay) { 592 // We have an overflow. Emit a bug report. 593 ExplodedNode *N = C.generateSink(stateOverflow); 594 if (!N) 595 return NULL; 596 597 if (!BT_AdditionOverflow) 598 BT_AdditionOverflow.reset(new BuiltinBug("API", 599 "Sum of expressions causes overflow")); 600 601 // This isn't a great error message, but this should never occur in real 602 // code anyway -- you'd have to create a buffer longer than a size_t can 603 // represent, which is sort of a contradiction. 604 const char *warning = 605 "This expression will create a string whose length is too big to " 606 "be represented as a size_t"; 607 608 // Generate a report for this bug. 609 BugReport *report = new BugReport(*BT_AdditionOverflow, warning, N); 610 C.EmitReport(report); 611 612 return NULL; 613 } 614 615 // From now on, assume an overflow didn't occur. 616 assert(stateOkay); 617 state = stateOkay; 618 } 619 620 return state; 621} 622 623ProgramStateRef CStringChecker::setCStringLength(ProgramStateRef state, 624 const MemRegion *MR, 625 SVal strLength) { 626 assert(!strLength.isUndef() && "Attempt to set an undefined string length"); 627 628 MR = MR->StripCasts(); 629 630 switch (MR->getKind()) { 631 case MemRegion::StringRegionKind: 632 // FIXME: This can happen if we strcpy() into a string region. This is 633 // undefined [C99 6.4.5p6], but we should still warn about it. 634 return state; 635 636 case MemRegion::SymbolicRegionKind: 637 case MemRegion::AllocaRegionKind: 638 case MemRegion::VarRegionKind: 639 case MemRegion::FieldRegionKind: 640 case MemRegion::ObjCIvarRegionKind: 641 // These are the types we can currently track string lengths for. 642 break; 643 644 case MemRegion::ElementRegionKind: 645 // FIXME: Handle element regions by upper-bounding the parent region's 646 // string length. 647 return state; 648 649 default: 650 // Other regions (mostly non-data) can't have a reliable C string length. 651 // For now, just ignore the change. 652 // FIXME: These are rare but not impossible. We should output some kind of 653 // warning for things like strcpy((char[]){'a', 0}, "b"); 654 return state; 655 } 656 657 if (strLength.isUnknown()) 658 return state->remove<CStringLength>(MR); 659 660 return state->set<CStringLength>(MR, strLength); 661} 662 663SVal CStringChecker::getCStringLengthForRegion(CheckerContext &C, 664 ProgramStateRef &state, 665 const Expr *Ex, 666 const MemRegion *MR, 667 bool hypothetical) { 668 if (!hypothetical) { 669 // If there's a recorded length, go ahead and return it. 670 const SVal *Recorded = state->get<CStringLength>(MR); 671 if (Recorded) 672 return *Recorded; 673 } 674 675 // Otherwise, get a new symbol and update the state. 676 SValBuilder &svalBuilder = C.getSValBuilder(); 677 QualType sizeTy = svalBuilder.getContext().getSizeType(); 678 SVal strLength = svalBuilder.getMetadataSymbolVal(CStringChecker::getTag(), 679 MR, Ex, sizeTy, 680 C.blockCount()); 681 682 if (!hypothetical) 683 state = state->set<CStringLength>(MR, strLength); 684 685 return strLength; 686} 687 688SVal CStringChecker::getCStringLength(CheckerContext &C, ProgramStateRef &state, 689 const Expr *Ex, SVal Buf, 690 bool hypothetical) const { 691 const MemRegion *MR = Buf.getAsRegion(); 692 if (!MR) { 693 // If we can't get a region, see if it's something we /know/ isn't a 694 // C string. In the context of locations, the only time we can issue such 695 // a warning is for labels. 696 if (loc::GotoLabel *Label = dyn_cast<loc::GotoLabel>(&Buf)) { 697 if (!Filter.CheckCStringNotNullTerm) 698 return UndefinedVal(); 699 700 if (ExplodedNode *N = C.addTransition(state)) { 701 if (!BT_NotCString) 702 BT_NotCString.reset(new BuiltinBug("Unix API", 703 "Argument is not a null-terminated string.")); 704 705 SmallString<120> buf; 706 llvm::raw_svector_ostream os(buf); 707 assert(CurrentFunctionDescription); 708 os << "Argument to " << CurrentFunctionDescription 709 << " is the address of the label '" << Label->getLabel()->getName() 710 << "', which is not a null-terminated string"; 711 712 // Generate a report for this bug. 713 BugReport *report = new BugReport(*BT_NotCString, 714 os.str(), N); 715 716 report->addRange(Ex->getSourceRange()); 717 C.EmitReport(report); 718 } 719 return UndefinedVal(); 720 721 } 722 723 // If it's not a region and not a label, give up. 724 return UnknownVal(); 725 } 726 727 // If we have a region, strip casts from it and see if we can figure out 728 // its length. For anything we can't figure out, just return UnknownVal. 729 MR = MR->StripCasts(); 730 731 switch (MR->getKind()) { 732 case MemRegion::StringRegionKind: { 733 // Modifying the contents of string regions is undefined [C99 6.4.5p6], 734 // so we can assume that the byte length is the correct C string length. 735 SValBuilder &svalBuilder = C.getSValBuilder(); 736 QualType sizeTy = svalBuilder.getContext().getSizeType(); 737 const StringLiteral *strLit = cast<StringRegion>(MR)->getStringLiteral(); 738 return svalBuilder.makeIntVal(strLit->getByteLength(), sizeTy); 739 } 740 case MemRegion::SymbolicRegionKind: 741 case MemRegion::AllocaRegionKind: 742 case MemRegion::VarRegionKind: 743 case MemRegion::FieldRegionKind: 744 case MemRegion::ObjCIvarRegionKind: 745 return getCStringLengthForRegion(C, state, Ex, MR, hypothetical); 746 case MemRegion::CompoundLiteralRegionKind: 747 // FIXME: Can we track this? Is it necessary? 748 return UnknownVal(); 749 case MemRegion::ElementRegionKind: 750 // FIXME: How can we handle this? It's not good enough to subtract the 751 // offset from the base string length; consider "123\x00567" and &a[5]. 752 return UnknownVal(); 753 default: 754 // Other regions (mostly non-data) can't have a reliable C string length. 755 // In this case, an error is emitted and UndefinedVal is returned. 756 // The caller should always be prepared to handle this case. 757 if (!Filter.CheckCStringNotNullTerm) 758 return UndefinedVal(); 759 760 if (ExplodedNode *N = C.addTransition(state)) { 761 if (!BT_NotCString) 762 BT_NotCString.reset(new BuiltinBug("Unix API", 763 "Argument is not a null-terminated string.")); 764 765 SmallString<120> buf; 766 llvm::raw_svector_ostream os(buf); 767 768 assert(CurrentFunctionDescription); 769 os << "Argument to " << CurrentFunctionDescription << " is "; 770 771 if (SummarizeRegion(os, C.getASTContext(), MR)) 772 os << ", which is not a null-terminated string"; 773 else 774 os << "not a null-terminated string"; 775 776 // Generate a report for this bug. 777 BugReport *report = new BugReport(*BT_NotCString, 778 os.str(), N); 779 780 report->addRange(Ex->getSourceRange()); 781 C.EmitReport(report); 782 } 783 784 return UndefinedVal(); 785 } 786} 787 788const StringLiteral *CStringChecker::getCStringLiteral(CheckerContext &C, 789 ProgramStateRef &state, const Expr *expr, SVal val) const { 790 791 // Get the memory region pointed to by the val. 792 const MemRegion *bufRegion = val.getAsRegion(); 793 if (!bufRegion) 794 return NULL; 795 796 // Strip casts off the memory region. 797 bufRegion = bufRegion->StripCasts(); 798 799 // Cast the memory region to a string region. 800 const StringRegion *strRegion= dyn_cast<StringRegion>(bufRegion); 801 if (!strRegion) 802 return NULL; 803 804 // Return the actual string in the string region. 805 return strRegion->getStringLiteral(); 806} 807 808ProgramStateRef CStringChecker::InvalidateBuffer(CheckerContext &C, 809 ProgramStateRef state, 810 const Expr *E, SVal V) { 811 Loc *L = dyn_cast<Loc>(&V); 812 if (!L) 813 return state; 814 815 // FIXME: This is a simplified version of what's in CFRefCount.cpp -- it makes 816 // some assumptions about the value that CFRefCount can't. Even so, it should 817 // probably be refactored. 818 if (loc::MemRegionVal* MR = dyn_cast<loc::MemRegionVal>(L)) { 819 const MemRegion *R = MR->getRegion()->StripCasts(); 820 821 // Are we dealing with an ElementRegion? If so, we should be invalidating 822 // the super-region. 823 if (const ElementRegion *ER = dyn_cast<ElementRegion>(R)) { 824 R = ER->getSuperRegion(); 825 // FIXME: What about layers of ElementRegions? 826 } 827 828 // Invalidate this region. 829 const LocationContext *LCtx = C.getPredecessor()->getLocationContext(); 830 return state->invalidateRegions(R, E, C.blockCount(), LCtx); 831 } 832 833 // If we have a non-region value by chance, just remove the binding. 834 // FIXME: is this necessary or correct? This handles the non-Region 835 // cases. Is it ever valid to store to these? 836 return state->killBinding(*L); 837} 838 839bool CStringChecker::SummarizeRegion(raw_ostream &os, ASTContext &Ctx, 840 const MemRegion *MR) { 841 const TypedValueRegion *TVR = dyn_cast<TypedValueRegion>(MR); 842 843 switch (MR->getKind()) { 844 case MemRegion::FunctionTextRegionKind: { 845 const NamedDecl *FD = cast<FunctionTextRegion>(MR)->getDecl(); 846 if (FD) 847 os << "the address of the function '" << *FD << '\''; 848 else 849 os << "the address of a function"; 850 return true; 851 } 852 case MemRegion::BlockTextRegionKind: 853 os << "block text"; 854 return true; 855 case MemRegion::BlockDataRegionKind: 856 os << "a block"; 857 return true; 858 case MemRegion::CXXThisRegionKind: 859 case MemRegion::CXXTempObjectRegionKind: 860 os << "a C++ temp object of type " << TVR->getValueType().getAsString(); 861 return true; 862 case MemRegion::VarRegionKind: 863 os << "a variable of type" << TVR->getValueType().getAsString(); 864 return true; 865 case MemRegion::FieldRegionKind: 866 os << "a field of type " << TVR->getValueType().getAsString(); 867 return true; 868 case MemRegion::ObjCIvarRegionKind: 869 os << "an instance variable of type " << TVR->getValueType().getAsString(); 870 return true; 871 default: 872 return false; 873 } 874} 875 876//===----------------------------------------------------------------------===// 877// evaluation of individual function calls. 878//===----------------------------------------------------------------------===// 879 880void CStringChecker::evalCopyCommon(CheckerContext &C, 881 const CallExpr *CE, 882 ProgramStateRef state, 883 const Expr *Size, const Expr *Dest, 884 const Expr *Source, bool Restricted, 885 bool IsMempcpy) const { 886 CurrentFunctionDescription = "memory copy function"; 887 888 // See if the size argument is zero. 889 const LocationContext *LCtx = C.getLocationContext(); 890 SVal sizeVal = state->getSVal(Size, LCtx); 891 QualType sizeTy = Size->getType(); 892 893 ProgramStateRef stateZeroSize, stateNonZeroSize; 894 llvm::tie(stateZeroSize, stateNonZeroSize) = 895 assumeZero(C, state, sizeVal, sizeTy); 896 897 // Get the value of the Dest. 898 SVal destVal = state->getSVal(Dest, LCtx); 899 900 // If the size is zero, there won't be any actual memory access, so 901 // just bind the return value to the destination buffer and return. 902 if (stateZeroSize && !stateNonZeroSize) { 903 stateZeroSize = stateZeroSize->BindExpr(CE, LCtx, destVal); 904 C.addTransition(stateZeroSize); 905 return; 906 } 907 908 // If the size can be nonzero, we have to check the other arguments. 909 if (stateNonZeroSize) { 910 state = stateNonZeroSize; 911 912 // Ensure the destination is not null. If it is NULL there will be a 913 // NULL pointer dereference. 914 state = checkNonNull(C, state, Dest, destVal); 915 if (!state) 916 return; 917 918 // Get the value of the Src. 919 SVal srcVal = state->getSVal(Source, LCtx); 920 921 // Ensure the source is not null. If it is NULL there will be a 922 // NULL pointer dereference. 923 state = checkNonNull(C, state, Source, srcVal); 924 if (!state) 925 return; 926 927 // Ensure the accesses are valid and that the buffers do not overlap. 928 const char * const writeWarning = 929 "Memory copy function overflows destination buffer"; 930 state = CheckBufferAccess(C, state, Size, Dest, Source, 931 writeWarning, /* sourceWarning = */ NULL); 932 if (Restricted) 933 state = CheckOverlap(C, state, Size, Dest, Source); 934 935 if (!state) 936 return; 937 938 // If this is mempcpy, get the byte after the last byte copied and 939 // bind the expr. 940 if (IsMempcpy) { 941 loc::MemRegionVal *destRegVal = dyn_cast<loc::MemRegionVal>(&destVal); 942 assert(destRegVal && "Destination should be a known MemRegionVal here"); 943 944 // Get the length to copy. 945 NonLoc *lenValNonLoc = dyn_cast<NonLoc>(&sizeVal); 946 947 if (lenValNonLoc) { 948 // Get the byte after the last byte copied. 949 SVal lastElement = C.getSValBuilder().evalBinOpLN(state, BO_Add, 950 *destRegVal, 951 *lenValNonLoc, 952 Dest->getType()); 953 954 // The byte after the last byte copied is the return value. 955 state = state->BindExpr(CE, LCtx, lastElement); 956 } else { 957 // If we don't know how much we copied, we can at least 958 // conjure a return value for later. 959 SVal result = C.getSValBuilder().conjureSymbolVal(0, CE, LCtx, 960 C.blockCount()); 961 state = state->BindExpr(CE, LCtx, result); 962 } 963 964 } else { 965 // All other copies return the destination buffer. 966 // (Well, bcopy() has a void return type, but this won't hurt.) 967 state = state->BindExpr(CE, LCtx, destVal); 968 } 969 970 // Invalidate the destination. 971 // FIXME: Even if we can't perfectly model the copy, we should see if we 972 // can use LazyCompoundVals to copy the source values into the destination. 973 // This would probably remove any existing bindings past the end of the 974 // copied region, but that's still an improvement over blank invalidation. 975 state = InvalidateBuffer(C, state, Dest, 976 state->getSVal(Dest, C.getLocationContext())); 977 C.addTransition(state); 978 } 979} 980 981 982void CStringChecker::evalMemcpy(CheckerContext &C, const CallExpr *CE) const { 983 if (CE->getNumArgs() < 3) 984 return; 985 986 // void *memcpy(void *restrict dst, const void *restrict src, size_t n); 987 // The return value is the address of the destination buffer. 988 const Expr *Dest = CE->getArg(0); 989 ProgramStateRef state = C.getState(); 990 991 evalCopyCommon(C, CE, state, CE->getArg(2), Dest, CE->getArg(1), true); 992} 993 994void CStringChecker::evalMempcpy(CheckerContext &C, const CallExpr *CE) const { 995 if (CE->getNumArgs() < 3) 996 return; 997 998 // void *mempcpy(void *restrict dst, const void *restrict src, size_t n); 999 // The return value is a pointer to the byte following the last written byte. 1000 const Expr *Dest = CE->getArg(0); 1001 ProgramStateRef state = C.getState(); 1002 1003 evalCopyCommon(C, CE, state, CE->getArg(2), Dest, CE->getArg(1), true, true); 1004} 1005 1006void CStringChecker::evalMemmove(CheckerContext &C, const CallExpr *CE) const { 1007 if (CE->getNumArgs() < 3) 1008 return; 1009 1010 // void *memmove(void *dst, const void *src, size_t n); 1011 // The return value is the address of the destination buffer. 1012 const Expr *Dest = CE->getArg(0); 1013 ProgramStateRef state = C.getState(); 1014 1015 evalCopyCommon(C, CE, state, CE->getArg(2), Dest, CE->getArg(1)); 1016} 1017 1018void CStringChecker::evalBcopy(CheckerContext &C, const CallExpr *CE) const { 1019 if (CE->getNumArgs() < 3) 1020 return; 1021 1022 // void bcopy(const void *src, void *dst, size_t n); 1023 evalCopyCommon(C, CE, C.getState(), 1024 CE->getArg(2), CE->getArg(1), CE->getArg(0)); 1025} 1026 1027void CStringChecker::evalMemcmp(CheckerContext &C, const CallExpr *CE) const { 1028 if (CE->getNumArgs() < 3) 1029 return; 1030 1031 // int memcmp(const void *s1, const void *s2, size_t n); 1032 CurrentFunctionDescription = "memory comparison function"; 1033 1034 const Expr *Left = CE->getArg(0); 1035 const Expr *Right = CE->getArg(1); 1036 const Expr *Size = CE->getArg(2); 1037 1038 ProgramStateRef state = C.getState(); 1039 SValBuilder &svalBuilder = C.getSValBuilder(); 1040 1041 // See if the size argument is zero. 1042 const LocationContext *LCtx = C.getLocationContext(); 1043 SVal sizeVal = state->getSVal(Size, LCtx); 1044 QualType sizeTy = Size->getType(); 1045 1046 ProgramStateRef stateZeroSize, stateNonZeroSize; 1047 llvm::tie(stateZeroSize, stateNonZeroSize) = 1048 assumeZero(C, state, sizeVal, sizeTy); 1049 1050 // If the size can be zero, the result will be 0 in that case, and we don't 1051 // have to check either of the buffers. 1052 if (stateZeroSize) { 1053 state = stateZeroSize; 1054 state = state->BindExpr(CE, LCtx, 1055 svalBuilder.makeZeroVal(CE->getType())); 1056 C.addTransition(state); 1057 } 1058 1059 // If the size can be nonzero, we have to check the other arguments. 1060 if (stateNonZeroSize) { 1061 state = stateNonZeroSize; 1062 // If we know the two buffers are the same, we know the result is 0. 1063 // First, get the two buffers' addresses. Another checker will have already 1064 // made sure they're not undefined. 1065 DefinedOrUnknownSVal LV = 1066 cast<DefinedOrUnknownSVal>(state->getSVal(Left, LCtx)); 1067 DefinedOrUnknownSVal RV = 1068 cast<DefinedOrUnknownSVal>(state->getSVal(Right, LCtx)); 1069 1070 // See if they are the same. 1071 DefinedOrUnknownSVal SameBuf = svalBuilder.evalEQ(state, LV, RV); 1072 ProgramStateRef StSameBuf, StNotSameBuf; 1073 llvm::tie(StSameBuf, StNotSameBuf) = state->assume(SameBuf); 1074 1075 // If the two arguments might be the same buffer, we know the result is 0, 1076 // and we only need to check one size. 1077 if (StSameBuf) { 1078 state = StSameBuf; 1079 state = CheckBufferAccess(C, state, Size, Left); 1080 if (state) { 1081 state = StSameBuf->BindExpr(CE, LCtx, 1082 svalBuilder.makeZeroVal(CE->getType())); 1083 C.addTransition(state); 1084 } 1085 } 1086 1087 // If the two arguments might be different buffers, we have to check the 1088 // size of both of them. 1089 if (StNotSameBuf) { 1090 state = StNotSameBuf; 1091 state = CheckBufferAccess(C, state, Size, Left, Right); 1092 if (state) { 1093 // The return value is the comparison result, which we don't know. 1094 SVal CmpV = svalBuilder.conjureSymbolVal(0, CE, LCtx, C.blockCount()); 1095 state = state->BindExpr(CE, LCtx, CmpV); 1096 C.addTransition(state); 1097 } 1098 } 1099 } 1100} 1101 1102void CStringChecker::evalstrLength(CheckerContext &C, 1103 const CallExpr *CE) const { 1104 if (CE->getNumArgs() < 1) 1105 return; 1106 1107 // size_t strlen(const char *s); 1108 evalstrLengthCommon(C, CE, /* IsStrnlen = */ false); 1109} 1110 1111void CStringChecker::evalstrnLength(CheckerContext &C, 1112 const CallExpr *CE) const { 1113 if (CE->getNumArgs() < 2) 1114 return; 1115 1116 // size_t strnlen(const char *s, size_t maxlen); 1117 evalstrLengthCommon(C, CE, /* IsStrnlen = */ true); 1118} 1119 1120void CStringChecker::evalstrLengthCommon(CheckerContext &C, const CallExpr *CE, 1121 bool IsStrnlen) const { 1122 CurrentFunctionDescription = "string length function"; 1123 ProgramStateRef state = C.getState(); 1124 const LocationContext *LCtx = C.getLocationContext(); 1125 1126 if (IsStrnlen) { 1127 const Expr *maxlenExpr = CE->getArg(1); 1128 SVal maxlenVal = state->getSVal(maxlenExpr, LCtx); 1129 1130 ProgramStateRef stateZeroSize, stateNonZeroSize; 1131 llvm::tie(stateZeroSize, stateNonZeroSize) = 1132 assumeZero(C, state, maxlenVal, maxlenExpr->getType()); 1133 1134 // If the size can be zero, the result will be 0 in that case, and we don't 1135 // have to check the string itself. 1136 if (stateZeroSize) { 1137 SVal zero = C.getSValBuilder().makeZeroVal(CE->getType()); 1138 stateZeroSize = stateZeroSize->BindExpr(CE, LCtx, zero); 1139 C.addTransition(stateZeroSize); 1140 } 1141 1142 // If the size is GUARANTEED to be zero, we're done! 1143 if (!stateNonZeroSize) 1144 return; 1145 1146 // Otherwise, record the assumption that the size is nonzero. 1147 state = stateNonZeroSize; 1148 } 1149 1150 // Check that the string argument is non-null. 1151 const Expr *Arg = CE->getArg(0); 1152 SVal ArgVal = state->getSVal(Arg, LCtx); 1153 1154 state = checkNonNull(C, state, Arg, ArgVal); 1155 1156 if (!state) 1157 return; 1158 1159 SVal strLength = getCStringLength(C, state, Arg, ArgVal); 1160 1161 // If the argument isn't a valid C string, there's no valid state to 1162 // transition to. 1163 if (strLength.isUndef()) 1164 return; 1165 1166 DefinedOrUnknownSVal result = UnknownVal(); 1167 1168 // If the check is for strnlen() then bind the return value to no more than 1169 // the maxlen value. 1170 if (IsStrnlen) { 1171 QualType cmpTy = C.getSValBuilder().getConditionType(); 1172 1173 // It's a little unfortunate to be getting this again, 1174 // but it's not that expensive... 1175 const Expr *maxlenExpr = CE->getArg(1); 1176 SVal maxlenVal = state->getSVal(maxlenExpr, LCtx); 1177 1178 NonLoc *strLengthNL = dyn_cast<NonLoc>(&strLength); 1179 NonLoc *maxlenValNL = dyn_cast<NonLoc>(&maxlenVal); 1180 1181 if (strLengthNL && maxlenValNL) { 1182 ProgramStateRef stateStringTooLong, stateStringNotTooLong; 1183 1184 // Check if the strLength is greater than the maxlen. 1185 llvm::tie(stateStringTooLong, stateStringNotTooLong) = 1186 state->assume(cast<DefinedOrUnknownSVal> 1187 (C.getSValBuilder().evalBinOpNN(state, BO_GT, 1188 *strLengthNL, 1189 *maxlenValNL, 1190 cmpTy))); 1191 1192 if (stateStringTooLong && !stateStringNotTooLong) { 1193 // If the string is longer than maxlen, return maxlen. 1194 result = *maxlenValNL; 1195 } else if (stateStringNotTooLong && !stateStringTooLong) { 1196 // If the string is shorter than maxlen, return its length. 1197 result = *strLengthNL; 1198 } 1199 } 1200 1201 if (result.isUnknown()) { 1202 // If we don't have enough information for a comparison, there's 1203 // no guarantee the full string length will actually be returned. 1204 // All we know is the return value is the min of the string length 1205 // and the limit. This is better than nothing. 1206 result = C.getSValBuilder().conjureSymbolVal(0, CE, LCtx, C.blockCount()); 1207 NonLoc *resultNL = cast<NonLoc>(&result); 1208 1209 if (strLengthNL) { 1210 state = state->assume(cast<DefinedOrUnknownSVal> 1211 (C.getSValBuilder().evalBinOpNN(state, BO_LE, 1212 *resultNL, 1213 *strLengthNL, 1214 cmpTy)), true); 1215 } 1216 1217 if (maxlenValNL) { 1218 state = state->assume(cast<DefinedOrUnknownSVal> 1219 (C.getSValBuilder().evalBinOpNN(state, BO_LE, 1220 *resultNL, 1221 *maxlenValNL, 1222 cmpTy)), true); 1223 } 1224 } 1225 1226 } else { 1227 // This is a plain strlen(), not strnlen(). 1228 result = cast<DefinedOrUnknownSVal>(strLength); 1229 1230 // If we don't know the length of the string, conjure a return 1231 // value, so it can be used in constraints, at least. 1232 if (result.isUnknown()) { 1233 result = C.getSValBuilder().conjureSymbolVal(0, CE, LCtx, C.blockCount()); 1234 } 1235 } 1236 1237 // Bind the return value. 1238 assert(!result.isUnknown() && "Should have conjured a value by now"); 1239 state = state->BindExpr(CE, LCtx, result); 1240 C.addTransition(state); 1241} 1242 1243void CStringChecker::evalStrcpy(CheckerContext &C, const CallExpr *CE) const { 1244 if (CE->getNumArgs() < 2) 1245 return; 1246 1247 // char *strcpy(char *restrict dst, const char *restrict src); 1248 evalStrcpyCommon(C, CE, 1249 /* returnEnd = */ false, 1250 /* isBounded = */ false, 1251 /* isAppending = */ false); 1252} 1253 1254void CStringChecker::evalStrncpy(CheckerContext &C, const CallExpr *CE) const { 1255 if (CE->getNumArgs() < 3) 1256 return; 1257 1258 // char *strncpy(char *restrict dst, const char *restrict src, size_t n); 1259 evalStrcpyCommon(C, CE, 1260 /* returnEnd = */ false, 1261 /* isBounded = */ true, 1262 /* isAppending = */ false); 1263} 1264 1265void CStringChecker::evalStpcpy(CheckerContext &C, const CallExpr *CE) const { 1266 if (CE->getNumArgs() < 2) 1267 return; 1268 1269 // char *stpcpy(char *restrict dst, const char *restrict src); 1270 evalStrcpyCommon(C, CE, 1271 /* returnEnd = */ true, 1272 /* isBounded = */ false, 1273 /* isAppending = */ false); 1274} 1275 1276void CStringChecker::evalStrcat(CheckerContext &C, const CallExpr *CE) const { 1277 if (CE->getNumArgs() < 2) 1278 return; 1279 1280 //char *strcat(char *restrict s1, const char *restrict s2); 1281 evalStrcpyCommon(C, CE, 1282 /* returnEnd = */ false, 1283 /* isBounded = */ false, 1284 /* isAppending = */ true); 1285} 1286 1287void CStringChecker::evalStrncat(CheckerContext &C, const CallExpr *CE) const { 1288 if (CE->getNumArgs() < 3) 1289 return; 1290 1291 //char *strncat(char *restrict s1, const char *restrict s2, size_t n); 1292 evalStrcpyCommon(C, CE, 1293 /* returnEnd = */ false, 1294 /* isBounded = */ true, 1295 /* isAppending = */ true); 1296} 1297 1298void CStringChecker::evalStrcpyCommon(CheckerContext &C, const CallExpr *CE, 1299 bool returnEnd, bool isBounded, 1300 bool isAppending) const { 1301 CurrentFunctionDescription = "string copy function"; 1302 ProgramStateRef state = C.getState(); 1303 const LocationContext *LCtx = C.getLocationContext(); 1304 1305 // Check that the destination is non-null. 1306 const Expr *Dst = CE->getArg(0); 1307 SVal DstVal = state->getSVal(Dst, LCtx); 1308 1309 state = checkNonNull(C, state, Dst, DstVal); 1310 if (!state) 1311 return; 1312 1313 // Check that the source is non-null. 1314 const Expr *srcExpr = CE->getArg(1); 1315 SVal srcVal = state->getSVal(srcExpr, LCtx); 1316 state = checkNonNull(C, state, srcExpr, srcVal); 1317 if (!state) 1318 return; 1319 1320 // Get the string length of the source. 1321 SVal strLength = getCStringLength(C, state, srcExpr, srcVal); 1322 1323 // If the source isn't a valid C string, give up. 1324 if (strLength.isUndef()) 1325 return; 1326 1327 SValBuilder &svalBuilder = C.getSValBuilder(); 1328 QualType cmpTy = svalBuilder.getConditionType(); 1329 QualType sizeTy = svalBuilder.getContext().getSizeType(); 1330 1331 // These two values allow checking two kinds of errors: 1332 // - actual overflows caused by a source that doesn't fit in the destination 1333 // - potential overflows caused by a bound that could exceed the destination 1334 SVal amountCopied = UnknownVal(); 1335 SVal maxLastElementIndex = UnknownVal(); 1336 const char *boundWarning = NULL; 1337 1338 // If the function is strncpy, strncat, etc... it is bounded. 1339 if (isBounded) { 1340 // Get the max number of characters to copy. 1341 const Expr *lenExpr = CE->getArg(2); 1342 SVal lenVal = state->getSVal(lenExpr, LCtx); 1343 1344 // Protect against misdeclared strncpy(). 1345 lenVal = svalBuilder.evalCast(lenVal, sizeTy, lenExpr->getType()); 1346 1347 NonLoc *strLengthNL = dyn_cast<NonLoc>(&strLength); 1348 NonLoc *lenValNL = dyn_cast<NonLoc>(&lenVal); 1349 1350 // If we know both values, we might be able to figure out how much 1351 // we're copying. 1352 if (strLengthNL && lenValNL) { 1353 ProgramStateRef stateSourceTooLong, stateSourceNotTooLong; 1354 1355 // Check if the max number to copy is less than the length of the src. 1356 // If the bound is equal to the source length, strncpy won't null- 1357 // terminate the result! 1358 llvm::tie(stateSourceTooLong, stateSourceNotTooLong) = 1359 state->assume(cast<DefinedOrUnknownSVal> 1360 (svalBuilder.evalBinOpNN(state, BO_GE, *strLengthNL, 1361 *lenValNL, cmpTy))); 1362 1363 if (stateSourceTooLong && !stateSourceNotTooLong) { 1364 // Max number to copy is less than the length of the src, so the actual 1365 // strLength copied is the max number arg. 1366 state = stateSourceTooLong; 1367 amountCopied = lenVal; 1368 1369 } else if (!stateSourceTooLong && stateSourceNotTooLong) { 1370 // The source buffer entirely fits in the bound. 1371 state = stateSourceNotTooLong; 1372 amountCopied = strLength; 1373 } 1374 } 1375 1376 // We still want to know if the bound is known to be too large. 1377 if (lenValNL) { 1378 if (isAppending) { 1379 // For strncat, the check is strlen(dst) + lenVal < sizeof(dst) 1380 1381 // Get the string length of the destination. If the destination is 1382 // memory that can't have a string length, we shouldn't be copying 1383 // into it anyway. 1384 SVal dstStrLength = getCStringLength(C, state, Dst, DstVal); 1385 if (dstStrLength.isUndef()) 1386 return; 1387 1388 if (NonLoc *dstStrLengthNL = dyn_cast<NonLoc>(&dstStrLength)) { 1389 maxLastElementIndex = svalBuilder.evalBinOpNN(state, BO_Add, 1390 *lenValNL, 1391 *dstStrLengthNL, 1392 sizeTy); 1393 boundWarning = "Size argument is greater than the free space in the " 1394 "destination buffer"; 1395 } 1396 1397 } else { 1398 // For strncpy, this is just checking that lenVal <= sizeof(dst) 1399 // (Yes, strncpy and strncat differ in how they treat termination. 1400 // strncat ALWAYS terminates, but strncpy doesn't.) 1401 1402 // We need a special case for when the copy size is zero, in which 1403 // case strncpy will do no work at all. Our bounds check uses n-1 1404 // as the last element accessed, so n == 0 is problematic. 1405 ProgramStateRef StateZeroSize, StateNonZeroSize; 1406 llvm::tie(StateZeroSize, StateNonZeroSize) = 1407 assumeZero(C, state, *lenValNL, sizeTy); 1408 1409 // If the size is known to be zero, we're done. 1410 if (StateZeroSize && !StateNonZeroSize) { 1411 StateZeroSize = StateZeroSize->BindExpr(CE, LCtx, DstVal); 1412 C.addTransition(StateZeroSize); 1413 return; 1414 } 1415 1416 // Otherwise, go ahead and figure out the last element we'll touch. 1417 // We don't record the non-zero assumption here because we can't 1418 // be sure. We won't warn on a possible zero. 1419 NonLoc one = cast<NonLoc>(svalBuilder.makeIntVal(1, sizeTy)); 1420 maxLastElementIndex = svalBuilder.evalBinOpNN(state, BO_Sub, *lenValNL, 1421 one, sizeTy); 1422 boundWarning = "Size argument is greater than the length of the " 1423 "destination buffer"; 1424 } 1425 } 1426 1427 // If we couldn't pin down the copy length, at least bound it. 1428 // FIXME: We should actually run this code path for append as well, but 1429 // right now it creates problems with constraints (since we can end up 1430 // trying to pass constraints from symbol to symbol). 1431 if (amountCopied.isUnknown() && !isAppending) { 1432 // Try to get a "hypothetical" string length symbol, which we can later 1433 // set as a real value if that turns out to be the case. 1434 amountCopied = getCStringLength(C, state, lenExpr, srcVal, true); 1435 assert(!amountCopied.isUndef()); 1436 1437 if (NonLoc *amountCopiedNL = dyn_cast<NonLoc>(&amountCopied)) { 1438 if (lenValNL) { 1439 // amountCopied <= lenVal 1440 SVal copiedLessThanBound = svalBuilder.evalBinOpNN(state, BO_LE, 1441 *amountCopiedNL, 1442 *lenValNL, 1443 cmpTy); 1444 state = state->assume(cast<DefinedOrUnknownSVal>(copiedLessThanBound), 1445 true); 1446 if (!state) 1447 return; 1448 } 1449 1450 if (strLengthNL) { 1451 // amountCopied <= strlen(source) 1452 SVal copiedLessThanSrc = svalBuilder.evalBinOpNN(state, BO_LE, 1453 *amountCopiedNL, 1454 *strLengthNL, 1455 cmpTy); 1456 state = state->assume(cast<DefinedOrUnknownSVal>(copiedLessThanSrc), 1457 true); 1458 if (!state) 1459 return; 1460 } 1461 } 1462 } 1463 1464 } else { 1465 // The function isn't bounded. The amount copied should match the length 1466 // of the source buffer. 1467 amountCopied = strLength; 1468 } 1469 1470 assert(state); 1471 1472 // This represents the number of characters copied into the destination 1473 // buffer. (It may not actually be the strlen if the destination buffer 1474 // is not terminated.) 1475 SVal finalStrLength = UnknownVal(); 1476 1477 // If this is an appending function (strcat, strncat...) then set the 1478 // string length to strlen(src) + strlen(dst) since the buffer will 1479 // ultimately contain both. 1480 if (isAppending) { 1481 // Get the string length of the destination. If the destination is memory 1482 // that can't have a string length, we shouldn't be copying into it anyway. 1483 SVal dstStrLength = getCStringLength(C, state, Dst, DstVal); 1484 if (dstStrLength.isUndef()) 1485 return; 1486 1487 NonLoc *srcStrLengthNL = dyn_cast<NonLoc>(&amountCopied); 1488 NonLoc *dstStrLengthNL = dyn_cast<NonLoc>(&dstStrLength); 1489 1490 // If we know both string lengths, we might know the final string length. 1491 if (srcStrLengthNL && dstStrLengthNL) { 1492 // Make sure the two lengths together don't overflow a size_t. 1493 state = checkAdditionOverflow(C, state, *srcStrLengthNL, *dstStrLengthNL); 1494 if (!state) 1495 return; 1496 1497 finalStrLength = svalBuilder.evalBinOpNN(state, BO_Add, *srcStrLengthNL, 1498 *dstStrLengthNL, sizeTy); 1499 } 1500 1501 // If we couldn't get a single value for the final string length, 1502 // we can at least bound it by the individual lengths. 1503 if (finalStrLength.isUnknown()) { 1504 // Try to get a "hypothetical" string length symbol, which we can later 1505 // set as a real value if that turns out to be the case. 1506 finalStrLength = getCStringLength(C, state, CE, DstVal, true); 1507 assert(!finalStrLength.isUndef()); 1508 1509 if (NonLoc *finalStrLengthNL = dyn_cast<NonLoc>(&finalStrLength)) { 1510 if (srcStrLengthNL) { 1511 // finalStrLength >= srcStrLength 1512 SVal sourceInResult = svalBuilder.evalBinOpNN(state, BO_GE, 1513 *finalStrLengthNL, 1514 *srcStrLengthNL, 1515 cmpTy); 1516 state = state->assume(cast<DefinedOrUnknownSVal>(sourceInResult), 1517 true); 1518 if (!state) 1519 return; 1520 } 1521 1522 if (dstStrLengthNL) { 1523 // finalStrLength >= dstStrLength 1524 SVal destInResult = svalBuilder.evalBinOpNN(state, BO_GE, 1525 *finalStrLengthNL, 1526 *dstStrLengthNL, 1527 cmpTy); 1528 state = state->assume(cast<DefinedOrUnknownSVal>(destInResult), 1529 true); 1530 if (!state) 1531 return; 1532 } 1533 } 1534 } 1535 1536 } else { 1537 // Otherwise, this is a copy-over function (strcpy, strncpy, ...), and 1538 // the final string length will match the input string length. 1539 finalStrLength = amountCopied; 1540 } 1541 1542 // The final result of the function will either be a pointer past the last 1543 // copied element, or a pointer to the start of the destination buffer. 1544 SVal Result = (returnEnd ? UnknownVal() : DstVal); 1545 1546 assert(state); 1547 1548 // If the destination is a MemRegion, try to check for a buffer overflow and 1549 // record the new string length. 1550 if (loc::MemRegionVal *dstRegVal = dyn_cast<loc::MemRegionVal>(&DstVal)) { 1551 QualType ptrTy = Dst->getType(); 1552 1553 // If we have an exact value on a bounded copy, use that to check for 1554 // overflows, rather than our estimate about how much is actually copied. 1555 if (boundWarning) { 1556 if (NonLoc *maxLastNL = dyn_cast<NonLoc>(&maxLastElementIndex)) { 1557 SVal maxLastElement = svalBuilder.evalBinOpLN(state, BO_Add, *dstRegVal, 1558 *maxLastNL, ptrTy); 1559 state = CheckLocation(C, state, CE->getArg(2), maxLastElement, 1560 boundWarning); 1561 if (!state) 1562 return; 1563 } 1564 } 1565 1566 // Then, if the final length is known... 1567 if (NonLoc *knownStrLength = dyn_cast<NonLoc>(&finalStrLength)) { 1568 SVal lastElement = svalBuilder.evalBinOpLN(state, BO_Add, *dstRegVal, 1569 *knownStrLength, ptrTy); 1570 1571 // ...and we haven't checked the bound, we'll check the actual copy. 1572 if (!boundWarning) { 1573 const char * const warningMsg = 1574 "String copy function overflows destination buffer"; 1575 state = CheckLocation(C, state, Dst, lastElement, warningMsg); 1576 if (!state) 1577 return; 1578 } 1579 1580 // If this is a stpcpy-style copy, the last element is the return value. 1581 if (returnEnd) 1582 Result = lastElement; 1583 } 1584 1585 // Invalidate the destination. This must happen before we set the C string 1586 // length because invalidation will clear the length. 1587 // FIXME: Even if we can't perfectly model the copy, we should see if we 1588 // can use LazyCompoundVals to copy the source values into the destination. 1589 // This would probably remove any existing bindings past the end of the 1590 // string, but that's still an improvement over blank invalidation. 1591 state = InvalidateBuffer(C, state, Dst, *dstRegVal); 1592 1593 // Set the C string length of the destination, if we know it. 1594 if (isBounded && !isAppending) { 1595 // strncpy is annoying in that it doesn't guarantee to null-terminate 1596 // the result string. If the original string didn't fit entirely inside 1597 // the bound (including the null-terminator), we don't know how long the 1598 // result is. 1599 if (amountCopied != strLength) 1600 finalStrLength = UnknownVal(); 1601 } 1602 state = setCStringLength(state, dstRegVal->getRegion(), finalStrLength); 1603 } 1604 1605 assert(state); 1606 1607 // If this is a stpcpy-style copy, but we were unable to check for a buffer 1608 // overflow, we still need a result. Conjure a return value. 1609 if (returnEnd && Result.isUnknown()) { 1610 Result = svalBuilder.conjureSymbolVal(0, CE, LCtx, C.blockCount()); 1611 } 1612 1613 // Set the return value. 1614 state = state->BindExpr(CE, LCtx, Result); 1615 C.addTransition(state); 1616} 1617 1618void CStringChecker::evalStrcmp(CheckerContext &C, const CallExpr *CE) const { 1619 if (CE->getNumArgs() < 2) 1620 return; 1621 1622 //int strcmp(const char *s1, const char *s2); 1623 evalStrcmpCommon(C, CE, /* isBounded = */ false, /* ignoreCase = */ false); 1624} 1625 1626void CStringChecker::evalStrncmp(CheckerContext &C, const CallExpr *CE) const { 1627 if (CE->getNumArgs() < 3) 1628 return; 1629 1630 //int strncmp(const char *s1, const char *s2, size_t n); 1631 evalStrcmpCommon(C, CE, /* isBounded = */ true, /* ignoreCase = */ false); 1632} 1633 1634void CStringChecker::evalStrcasecmp(CheckerContext &C, 1635 const CallExpr *CE) const { 1636 if (CE->getNumArgs() < 2) 1637 return; 1638 1639 //int strcasecmp(const char *s1, const char *s2); 1640 evalStrcmpCommon(C, CE, /* isBounded = */ false, /* ignoreCase = */ true); 1641} 1642 1643void CStringChecker::evalStrncasecmp(CheckerContext &C, 1644 const CallExpr *CE) const { 1645 if (CE->getNumArgs() < 3) 1646 return; 1647 1648 //int strncasecmp(const char *s1, const char *s2, size_t n); 1649 evalStrcmpCommon(C, CE, /* isBounded = */ true, /* ignoreCase = */ true); 1650} 1651 1652void CStringChecker::evalStrcmpCommon(CheckerContext &C, const CallExpr *CE, 1653 bool isBounded, bool ignoreCase) const { 1654 CurrentFunctionDescription = "string comparison function"; 1655 ProgramStateRef state = C.getState(); 1656 const LocationContext *LCtx = C.getLocationContext(); 1657 1658 // Check that the first string is non-null 1659 const Expr *s1 = CE->getArg(0); 1660 SVal s1Val = state->getSVal(s1, LCtx); 1661 state = checkNonNull(C, state, s1, s1Val); 1662 if (!state) 1663 return; 1664 1665 // Check that the second string is non-null. 1666 const Expr *s2 = CE->getArg(1); 1667 SVal s2Val = state->getSVal(s2, LCtx); 1668 state = checkNonNull(C, state, s2, s2Val); 1669 if (!state) 1670 return; 1671 1672 // Get the string length of the first string or give up. 1673 SVal s1Length = getCStringLength(C, state, s1, s1Val); 1674 if (s1Length.isUndef()) 1675 return; 1676 1677 // Get the string length of the second string or give up. 1678 SVal s2Length = getCStringLength(C, state, s2, s2Val); 1679 if (s2Length.isUndef()) 1680 return; 1681 1682 // If we know the two buffers are the same, we know the result is 0. 1683 // First, get the two buffers' addresses. Another checker will have already 1684 // made sure they're not undefined. 1685 DefinedOrUnknownSVal LV = cast<DefinedOrUnknownSVal>(s1Val); 1686 DefinedOrUnknownSVal RV = cast<DefinedOrUnknownSVal>(s2Val); 1687 1688 // See if they are the same. 1689 SValBuilder &svalBuilder = C.getSValBuilder(); 1690 DefinedOrUnknownSVal SameBuf = svalBuilder.evalEQ(state, LV, RV); 1691 ProgramStateRef StSameBuf, StNotSameBuf; 1692 llvm::tie(StSameBuf, StNotSameBuf) = state->assume(SameBuf); 1693 1694 // If the two arguments might be the same buffer, we know the result is 0, 1695 // and we only need to check one size. 1696 if (StSameBuf) { 1697 StSameBuf = StSameBuf->BindExpr(CE, LCtx, 1698 svalBuilder.makeZeroVal(CE->getType())); 1699 C.addTransition(StSameBuf); 1700 1701 // If the two arguments are GUARANTEED to be the same, we're done! 1702 if (!StNotSameBuf) 1703 return; 1704 } 1705 1706 assert(StNotSameBuf); 1707 state = StNotSameBuf; 1708 1709 // At this point we can go about comparing the two buffers. 1710 // For now, we only do this if they're both known string literals. 1711 1712 // Attempt to extract string literals from both expressions. 1713 const StringLiteral *s1StrLiteral = getCStringLiteral(C, state, s1, s1Val); 1714 const StringLiteral *s2StrLiteral = getCStringLiteral(C, state, s2, s2Val); 1715 bool canComputeResult = false; 1716 1717 if (s1StrLiteral && s2StrLiteral) { 1718 StringRef s1StrRef = s1StrLiteral->getString(); 1719 StringRef s2StrRef = s2StrLiteral->getString(); 1720 1721 if (isBounded) { 1722 // Get the max number of characters to compare. 1723 const Expr *lenExpr = CE->getArg(2); 1724 SVal lenVal = state->getSVal(lenExpr, LCtx); 1725 1726 // If the length is known, we can get the right substrings. 1727 if (const llvm::APSInt *len = svalBuilder.getKnownValue(state, lenVal)) { 1728 // Create substrings of each to compare the prefix. 1729 s1StrRef = s1StrRef.substr(0, (size_t)len->getZExtValue()); 1730 s2StrRef = s2StrRef.substr(0, (size_t)len->getZExtValue()); 1731 canComputeResult = true; 1732 } 1733 } else { 1734 // This is a normal, unbounded strcmp. 1735 canComputeResult = true; 1736 } 1737 1738 if (canComputeResult) { 1739 // Real strcmp stops at null characters. 1740 size_t s1Term = s1StrRef.find('\0'); 1741 if (s1Term != StringRef::npos) 1742 s1StrRef = s1StrRef.substr(0, s1Term); 1743 1744 size_t s2Term = s2StrRef.find('\0'); 1745 if (s2Term != StringRef::npos) 1746 s2StrRef = s2StrRef.substr(0, s2Term); 1747 1748 // Use StringRef's comparison methods to compute the actual result. 1749 int result; 1750 1751 if (ignoreCase) { 1752 // Compare string 1 to string 2 the same way strcasecmp() does. 1753 result = s1StrRef.compare_lower(s2StrRef); 1754 } else { 1755 // Compare string 1 to string 2 the same way strcmp() does. 1756 result = s1StrRef.compare(s2StrRef); 1757 } 1758 1759 // Build the SVal of the comparison and bind the return value. 1760 SVal resultVal = svalBuilder.makeIntVal(result, CE->getType()); 1761 state = state->BindExpr(CE, LCtx, resultVal); 1762 } 1763 } 1764 1765 if (!canComputeResult) { 1766 // Conjure a symbolic value. It's the best we can do. 1767 SVal resultVal = svalBuilder.conjureSymbolVal(0, CE, LCtx, C.blockCount()); 1768 state = state->BindExpr(CE, LCtx, resultVal); 1769 } 1770 1771 // Record this as a possible path. 1772 C.addTransition(state); 1773} 1774 1775//===----------------------------------------------------------------------===// 1776// The driver method, and other Checker callbacks. 1777//===----------------------------------------------------------------------===// 1778 1779bool CStringChecker::evalCall(const CallExpr *CE, CheckerContext &C) const { 1780 const FunctionDecl *FDecl = C.getCalleeDecl(CE); 1781 1782 if (!FDecl) 1783 return false; 1784 1785 FnCheck evalFunction = 0; 1786 if (C.isCLibraryFunction(FDecl, "memcpy")) 1787 evalFunction = &CStringChecker::evalMemcpy; 1788 else if (C.isCLibraryFunction(FDecl, "mempcpy")) 1789 evalFunction = &CStringChecker::evalMempcpy; 1790 else if (C.isCLibraryFunction(FDecl, "memcmp")) 1791 evalFunction = &CStringChecker::evalMemcmp; 1792 else if (C.isCLibraryFunction(FDecl, "memmove")) 1793 evalFunction = &CStringChecker::evalMemmove; 1794 else if (C.isCLibraryFunction(FDecl, "strcpy")) 1795 evalFunction = &CStringChecker::evalStrcpy; 1796 else if (C.isCLibraryFunction(FDecl, "strncpy")) 1797 evalFunction = &CStringChecker::evalStrncpy; 1798 else if (C.isCLibraryFunction(FDecl, "stpcpy")) 1799 evalFunction = &CStringChecker::evalStpcpy; 1800 else if (C.isCLibraryFunction(FDecl, "strcat")) 1801 evalFunction = &CStringChecker::evalStrcat; 1802 else if (C.isCLibraryFunction(FDecl, "strncat")) 1803 evalFunction = &CStringChecker::evalStrncat; 1804 else if (C.isCLibraryFunction(FDecl, "strlen")) 1805 evalFunction = &CStringChecker::evalstrLength; 1806 else if (C.isCLibraryFunction(FDecl, "strnlen")) 1807 evalFunction = &CStringChecker::evalstrnLength; 1808 else if (C.isCLibraryFunction(FDecl, "strcmp")) 1809 evalFunction = &CStringChecker::evalStrcmp; 1810 else if (C.isCLibraryFunction(FDecl, "strncmp")) 1811 evalFunction = &CStringChecker::evalStrncmp; 1812 else if (C.isCLibraryFunction(FDecl, "strcasecmp")) 1813 evalFunction = &CStringChecker::evalStrcasecmp; 1814 else if (C.isCLibraryFunction(FDecl, "strncasecmp")) 1815 evalFunction = &CStringChecker::evalStrncasecmp; 1816 else if (C.isCLibraryFunction(FDecl, "bcopy")) 1817 evalFunction = &CStringChecker::evalBcopy; 1818 else if (C.isCLibraryFunction(FDecl, "bcmp")) 1819 evalFunction = &CStringChecker::evalMemcmp; 1820 1821 // If the callee isn't a string function, let another checker handle it. 1822 if (!evalFunction) 1823 return false; 1824 1825 // Make sure each function sets its own description. 1826 // (But don't bother in a release build.) 1827 assert(!(CurrentFunctionDescription = NULL)); 1828 1829 // Check and evaluate the call. 1830 (this->*evalFunction)(C, CE); 1831 1832 // If the evaluate call resulted in no change, chain to the next eval call 1833 // handler. 1834 // Note, the custom CString evaluation calls assume that basic safety 1835 // properties are held. However, if the user chooses to turn off some of these 1836 // checks, we ignore the issues and leave the call evaluation to a generic 1837 // handler. 1838 if (!C.isDifferent()) 1839 return false; 1840 1841 return true; 1842} 1843 1844void CStringChecker::checkPreStmt(const DeclStmt *DS, CheckerContext &C) const { 1845 // Record string length for char a[] = "abc"; 1846 ProgramStateRef state = C.getState(); 1847 1848 for (DeclStmt::const_decl_iterator I = DS->decl_begin(), E = DS->decl_end(); 1849 I != E; ++I) { 1850 const VarDecl *D = dyn_cast<VarDecl>(*I); 1851 if (!D) 1852 continue; 1853 1854 // FIXME: Handle array fields of structs. 1855 if (!D->getType()->isArrayType()) 1856 continue; 1857 1858 const Expr *Init = D->getInit(); 1859 if (!Init) 1860 continue; 1861 if (!isa<StringLiteral>(Init)) 1862 continue; 1863 1864 Loc VarLoc = state->getLValue(D, C.getLocationContext()); 1865 const MemRegion *MR = VarLoc.getAsRegion(); 1866 if (!MR) 1867 continue; 1868 1869 SVal StrVal = state->getSVal(Init, C.getLocationContext()); 1870 assert(StrVal.isValid() && "Initializer string is unknown or undefined"); 1871 DefinedOrUnknownSVal strLength 1872 = cast<DefinedOrUnknownSVal>(getCStringLength(C, state, Init, StrVal)); 1873 1874 state = state->set<CStringLength>(MR, strLength); 1875 } 1876 1877 C.addTransition(state); 1878} 1879 1880bool CStringChecker::wantsRegionChangeUpdate(ProgramStateRef state) const { 1881 CStringLength::EntryMap Entries = state->get<CStringLength>(); 1882 return !Entries.isEmpty(); 1883} 1884 1885ProgramStateRef 1886CStringChecker::checkRegionChanges(ProgramStateRef state, 1887 const StoreManager::InvalidatedSymbols *, 1888 ArrayRef<const MemRegion *> ExplicitRegions, 1889 ArrayRef<const MemRegion *> Regions, 1890 const CallEvent *Call) const { 1891 CStringLength::EntryMap Entries = state->get<CStringLength>(); 1892 if (Entries.isEmpty()) 1893 return state; 1894 1895 llvm::SmallPtrSet<const MemRegion *, 8> Invalidated; 1896 llvm::SmallPtrSet<const MemRegion *, 32> SuperRegions; 1897 1898 // First build sets for the changed regions and their super-regions. 1899 for (ArrayRef<const MemRegion *>::iterator 1900 I = Regions.begin(), E = Regions.end(); I != E; ++I) { 1901 const MemRegion *MR = *I; 1902 Invalidated.insert(MR); 1903 1904 SuperRegions.insert(MR); 1905 while (const SubRegion *SR = dyn_cast<SubRegion>(MR)) { 1906 MR = SR->getSuperRegion(); 1907 SuperRegions.insert(MR); 1908 } 1909 } 1910 1911 CStringLength::EntryMap::Factory &F = state->get_context<CStringLength>(); 1912 1913 // Then loop over the entries in the current state. 1914 for (CStringLength::EntryMap::iterator I = Entries.begin(), 1915 E = Entries.end(); I != E; ++I) { 1916 const MemRegion *MR = I.getKey(); 1917 1918 // Is this entry for a super-region of a changed region? 1919 if (SuperRegions.count(MR)) { 1920 Entries = F.remove(Entries, MR); 1921 continue; 1922 } 1923 1924 // Is this entry for a sub-region of a changed region? 1925 const MemRegion *Super = MR; 1926 while (const SubRegion *SR = dyn_cast<SubRegion>(Super)) { 1927 Super = SR->getSuperRegion(); 1928 if (Invalidated.count(Super)) { 1929 Entries = F.remove(Entries, MR); 1930 break; 1931 } 1932 } 1933 } 1934 1935 return state->set<CStringLength>(Entries); 1936} 1937 1938void CStringChecker::checkLiveSymbols(ProgramStateRef state, 1939 SymbolReaper &SR) const { 1940 // Mark all symbols in our string length map as valid. 1941 CStringLength::EntryMap Entries = state->get<CStringLength>(); 1942 1943 for (CStringLength::EntryMap::iterator I = Entries.begin(), E = Entries.end(); 1944 I != E; ++I) { 1945 SVal Len = I.getData(); 1946 1947 for (SymExpr::symbol_iterator si = Len.symbol_begin(), 1948 se = Len.symbol_end(); si != se; ++si) 1949 SR.markInUse(*si); 1950 } 1951} 1952 1953void CStringChecker::checkDeadSymbols(SymbolReaper &SR, 1954 CheckerContext &C) const { 1955 if (!SR.hasDeadSymbols()) 1956 return; 1957 1958 ProgramStateRef state = C.getState(); 1959 CStringLength::EntryMap Entries = state->get<CStringLength>(); 1960 if (Entries.isEmpty()) 1961 return; 1962 1963 CStringLength::EntryMap::Factory &F = state->get_context<CStringLength>(); 1964 for (CStringLength::EntryMap::iterator I = Entries.begin(), E = Entries.end(); 1965 I != E; ++I) { 1966 SVal Len = I.getData(); 1967 if (SymbolRef Sym = Len.getAsSymbol()) { 1968 if (SR.isDead(Sym)) 1969 Entries = F.remove(Entries, I.getKey()); 1970 } 1971 } 1972 1973 state = state->set<CStringLength>(Entries); 1974 C.addTransition(state); 1975} 1976 1977#define REGISTER_CHECKER(name) \ 1978void ento::register##name(CheckerManager &mgr) {\ 1979 static CStringChecker *TheChecker = 0; \ 1980 if (TheChecker == 0) \ 1981 TheChecker = mgr.registerChecker<CStringChecker>(); \ 1982 TheChecker->Filter.Check##name = true; \ 1983} 1984 1985REGISTER_CHECKER(CStringNullArg) 1986REGISTER_CHECKER(CStringOutOfBounds) 1987REGISTER_CHECKER(CStringBufferOverlap) 1988REGISTER_CHECKER(CStringNotNullTerm) 1989 1990void ento::registerCStringCheckerBasic(CheckerManager &Mgr) { 1991 registerCStringNullArg(Mgr); 1992} 1993