1 //===- BugReporterVisitors.cpp - Helpers for reporting bugs ---------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 // This file defines a set of BugReporter "visitors" which can be used to 10 // enhance the diagnostics reported for a bug. 11 // 12 //===----------------------------------------------------------------------===// 13 14 #include "clang/StaticAnalyzer/Core/BugReporter/BugReporterVisitors.h" 15 #include "clang/AST/ASTContext.h" 16 #include "clang/AST/Decl.h" 17 #include "clang/AST/DeclBase.h" 18 #include "clang/AST/DeclCXX.h" 19 #include "clang/AST/Expr.h" 20 #include "clang/AST/ExprCXX.h" 21 #include "clang/AST/ExprObjC.h" 22 #include "clang/AST/Stmt.h" 23 #include "clang/AST/Type.h" 24 #include "clang/ASTMatchers/ASTMatchFinder.h" 25 #include "clang/Analysis/Analyses/Dominators.h" 26 #include "clang/Analysis/AnalysisDeclContext.h" 27 #include "clang/Analysis/CFG.h" 28 #include "clang/Analysis/CFGStmtMap.h" 29 #include "clang/Analysis/PathDiagnostic.h" 30 #include "clang/Analysis/ProgramPoint.h" 31 #include "clang/Basic/IdentifierTable.h" 32 #include "clang/Basic/LLVM.h" 33 #include "clang/Basic/SourceLocation.h" 34 #include "clang/Basic/SourceManager.h" 35 #include "clang/Lex/Lexer.h" 36 #include "clang/StaticAnalyzer/Core/AnalyzerOptions.h" 37 #include "clang/StaticAnalyzer/Core/BugReporter/BugReporter.h" 38 #include "clang/StaticAnalyzer/Core/PathSensitive/AnalysisManager.h" 39 #include "clang/StaticAnalyzer/Core/PathSensitive/CallEvent.h" 40 #include "clang/StaticAnalyzer/Core/PathSensitive/ExplodedGraph.h" 41 #include "clang/StaticAnalyzer/Core/PathSensitive/ExprEngine.h" 42 #include "clang/StaticAnalyzer/Core/PathSensitive/MemRegion.h" 43 #include "clang/StaticAnalyzer/Core/PathSensitive/ProgramState.h" 44 #include "clang/StaticAnalyzer/Core/PathSensitive/ProgramState_Fwd.h" 45 #include "clang/StaticAnalyzer/Core/PathSensitive/SMTConv.h" 46 #include "clang/StaticAnalyzer/Core/PathSensitive/SValBuilder.h" 47 #include "clang/StaticAnalyzer/Core/PathSensitive/SVals.h" 48 #include "llvm/ADT/ArrayRef.h" 49 #include "llvm/ADT/None.h" 50 #include "llvm/ADT/Optional.h" 51 #include "llvm/ADT/STLExtras.h" 52 #include "llvm/ADT/SmallPtrSet.h" 53 #include "llvm/ADT/SmallString.h" 54 #include "llvm/ADT/SmallVector.h" 55 #include "llvm/ADT/StringExtras.h" 56 #include "llvm/ADT/StringRef.h" 57 #include "llvm/Support/Casting.h" 58 #include "llvm/Support/ErrorHandling.h" 59 #include "llvm/Support/raw_ostream.h" 60 #include <cassert> 61 #include <deque> 62 #include <memory> 63 #include <string> 64 #include <utility> 65 66 using namespace clang; 67 using namespace ento; 68 using namespace bugreporter; 69 70 //===----------------------------------------------------------------------===// 71 // Utility functions. 72 //===----------------------------------------------------------------------===// 73 74 static const Expr *peelOffPointerArithmetic(const BinaryOperator *B) { 75 if (B->isAdditiveOp() && B->getType()->isPointerType()) { 76 if (B->getLHS()->getType()->isPointerType()) { 77 return B->getLHS(); 78 } else if (B->getRHS()->getType()->isPointerType()) { 79 return B->getRHS(); 80 } 81 } 82 return nullptr; 83 } 84 85 /// Given that expression S represents a pointer that would be dereferenced, 86 /// try to find a sub-expression from which the pointer came from. 87 /// This is used for tracking down origins of a null or undefined value: 88 /// "this is null because that is null because that is null" etc. 89 /// We wipe away field and element offsets because they merely add offsets. 90 /// We also wipe away all casts except lvalue-to-rvalue casts, because the 91 /// latter represent an actual pointer dereference; however, we remove 92 /// the final lvalue-to-rvalue cast before returning from this function 93 /// because it demonstrates more clearly from where the pointer rvalue was 94 /// loaded. Examples: 95 /// x->y.z ==> x (lvalue) 96 /// foo()->y.z ==> foo() (rvalue) 97 const Expr *bugreporter::getDerefExpr(const Stmt *S) { 98 const auto *E = dyn_cast<Expr>(S); 99 if (!E) 100 return nullptr; 101 102 while (true) { 103 if (const auto *CE = dyn_cast<CastExpr>(E)) { 104 if (CE->getCastKind() == CK_LValueToRValue) { 105 // This cast represents the load we're looking for. 106 break; 107 } 108 E = CE->getSubExpr(); 109 } else if (const auto *B = dyn_cast<BinaryOperator>(E)) { 110 // Pointer arithmetic: '*(x + 2)' -> 'x') etc. 111 if (const Expr *Inner = peelOffPointerArithmetic(B)) { 112 E = Inner; 113 } else { 114 // Probably more arithmetic can be pattern-matched here, 115 // but for now give up. 116 break; 117 } 118 } else if (const auto *U = dyn_cast<UnaryOperator>(E)) { 119 if (U->getOpcode() == UO_Deref || U->getOpcode() == UO_AddrOf || 120 (U->isIncrementDecrementOp() && U->getType()->isPointerType())) { 121 // Operators '*' and '&' don't actually mean anything. 122 // We look at casts instead. 123 E = U->getSubExpr(); 124 } else { 125 // Probably more arithmetic can be pattern-matched here, 126 // but for now give up. 127 break; 128 } 129 } 130 // Pattern match for a few useful cases: a[0], p->f, *p etc. 131 else if (const auto *ME = dyn_cast<MemberExpr>(E)) { 132 E = ME->getBase(); 133 } else if (const auto *IvarRef = dyn_cast<ObjCIvarRefExpr>(E)) { 134 E = IvarRef->getBase(); 135 } else if (const auto *AE = dyn_cast<ArraySubscriptExpr>(E)) { 136 E = AE->getBase(); 137 } else if (const auto *PE = dyn_cast<ParenExpr>(E)) { 138 E = PE->getSubExpr(); 139 } else if (const auto *FE = dyn_cast<FullExpr>(E)) { 140 E = FE->getSubExpr(); 141 } else { 142 // Other arbitrary stuff. 143 break; 144 } 145 } 146 147 // Special case: remove the final lvalue-to-rvalue cast, but do not recurse 148 // deeper into the sub-expression. This way we return the lvalue from which 149 // our pointer rvalue was loaded. 150 if (const auto *CE = dyn_cast<ImplicitCastExpr>(E)) 151 if (CE->getCastKind() == CK_LValueToRValue) 152 E = CE->getSubExpr(); 153 154 return E; 155 } 156 157 static const MemRegion * 158 getLocationRegionIfReference(const Expr *E, const ExplodedNode *N, 159 bool LookingForReference = true) { 160 if (const auto *DR = dyn_cast<DeclRefExpr>(E)) { 161 if (const auto *VD = dyn_cast<VarDecl>(DR->getDecl())) { 162 if (LookingForReference && !VD->getType()->isReferenceType()) 163 return nullptr; 164 return N->getState() 165 ->getLValue(VD, N->getLocationContext()) 166 .getAsRegion(); 167 } 168 } 169 170 // FIXME: This does not handle other kinds of null references, 171 // for example, references from FieldRegions: 172 // struct Wrapper { int &ref; }; 173 // Wrapper w = { *(int *)0 }; 174 // w.ref = 1; 175 176 return nullptr; 177 } 178 179 /// Comparing internal representations of symbolic values (via 180 /// SVal::operator==()) is a valid way to check if the value was updated, 181 /// unless it's a LazyCompoundVal that may have a different internal 182 /// representation every time it is loaded from the state. In this function we 183 /// do an approximate comparison for lazy compound values, checking that they 184 /// are the immediate snapshots of the tracked region's bindings within the 185 /// node's respective states but not really checking that these snapshots 186 /// actually contain the same set of bindings. 187 static bool hasVisibleUpdate(const ExplodedNode *LeftNode, SVal LeftVal, 188 const ExplodedNode *RightNode, SVal RightVal) { 189 if (LeftVal == RightVal) 190 return true; 191 192 const auto LLCV = LeftVal.getAs<nonloc::LazyCompoundVal>(); 193 if (!LLCV) 194 return false; 195 196 const auto RLCV = RightVal.getAs<nonloc::LazyCompoundVal>(); 197 if (!RLCV) 198 return false; 199 200 return LLCV->getRegion() == RLCV->getRegion() && 201 LLCV->getStore() == LeftNode->getState()->getStore() && 202 RLCV->getStore() == RightNode->getState()->getStore(); 203 } 204 205 static Optional<SVal> getSValForVar(const Expr *CondVarExpr, 206 const ExplodedNode *N) { 207 ProgramStateRef State = N->getState(); 208 const LocationContext *LCtx = N->getLocationContext(); 209 210 assert(CondVarExpr); 211 CondVarExpr = CondVarExpr->IgnoreImpCasts(); 212 213 // The declaration of the value may rely on a pointer so take its l-value. 214 // FIXME: As seen in VisitCommonDeclRefExpr, sometimes DeclRefExpr may 215 // evaluate to a FieldRegion when it refers to a declaration of a lambda 216 // capture variable. We most likely need to duplicate that logic here. 217 if (const auto *DRE = dyn_cast<DeclRefExpr>(CondVarExpr)) 218 if (const auto *VD = dyn_cast<VarDecl>(DRE->getDecl())) 219 return State->getSVal(State->getLValue(VD, LCtx)); 220 221 if (const auto *ME = dyn_cast<MemberExpr>(CondVarExpr)) 222 if (const auto *FD = dyn_cast<FieldDecl>(ME->getMemberDecl())) 223 if (auto FieldL = State->getSVal(ME, LCtx).getAs<Loc>()) 224 return State->getRawSVal(*FieldL, FD->getType()); 225 226 return None; 227 } 228 229 static Optional<const llvm::APSInt *> 230 getConcreteIntegerValue(const Expr *CondVarExpr, const ExplodedNode *N) { 231 232 if (Optional<SVal> V = getSValForVar(CondVarExpr, N)) 233 if (auto CI = V->getAs<nonloc::ConcreteInt>()) 234 return &CI->getValue(); 235 return None; 236 } 237 238 static bool isVarAnInterestingCondition(const Expr *CondVarExpr, 239 const ExplodedNode *N, 240 const PathSensitiveBugReport *B) { 241 // Even if this condition is marked as interesting, it isn't *that* 242 // interesting if it didn't happen in a nested stackframe, the user could just 243 // follow the arrows. 244 if (!B->getErrorNode()->getStackFrame()->isParentOf(N->getStackFrame())) 245 return false; 246 247 if (Optional<SVal> V = getSValForVar(CondVarExpr, N)) 248 if (Optional<bugreporter::TrackingKind> K = B->getInterestingnessKind(*V)) 249 return *K == bugreporter::TrackingKind::Condition; 250 251 return false; 252 } 253 254 static bool isInterestingExpr(const Expr *E, const ExplodedNode *N, 255 const PathSensitiveBugReport *B) { 256 if (Optional<SVal> V = getSValForVar(E, N)) 257 return B->getInterestingnessKind(*V).hasValue(); 258 return false; 259 } 260 261 /// \return name of the macro inside the location \p Loc. 262 static StringRef getMacroName(SourceLocation Loc, 263 BugReporterContext &BRC) { 264 return Lexer::getImmediateMacroName( 265 Loc, 266 BRC.getSourceManager(), 267 BRC.getASTContext().getLangOpts()); 268 } 269 270 /// \return Whether given spelling location corresponds to an expansion 271 /// of a function-like macro. 272 static bool isFunctionMacroExpansion(SourceLocation Loc, 273 const SourceManager &SM) { 274 if (!Loc.isMacroID()) 275 return false; 276 while (SM.isMacroArgExpansion(Loc)) 277 Loc = SM.getImmediateExpansionRange(Loc).getBegin(); 278 std::pair<FileID, unsigned> TLInfo = SM.getDecomposedLoc(Loc); 279 SrcMgr::SLocEntry SE = SM.getSLocEntry(TLInfo.first); 280 const SrcMgr::ExpansionInfo &EInfo = SE.getExpansion(); 281 return EInfo.isFunctionMacroExpansion(); 282 } 283 284 /// \return Whether \c RegionOfInterest was modified at \p N, 285 /// where \p ValueAfter is \c RegionOfInterest's value at the end of the 286 /// stack frame. 287 static bool wasRegionOfInterestModifiedAt(const SubRegion *RegionOfInterest, 288 const ExplodedNode *N, 289 SVal ValueAfter) { 290 ProgramStateRef State = N->getState(); 291 ProgramStateManager &Mgr = N->getState()->getStateManager(); 292 293 if (!N->getLocationAs<PostStore>() && !N->getLocationAs<PostInitializer>() && 294 !N->getLocationAs<PostStmt>()) 295 return false; 296 297 // Writing into region of interest. 298 if (auto PS = N->getLocationAs<PostStmt>()) 299 if (auto *BO = PS->getStmtAs<BinaryOperator>()) 300 if (BO->isAssignmentOp() && RegionOfInterest->isSubRegionOf( 301 N->getSVal(BO->getLHS()).getAsRegion())) 302 return true; 303 304 // SVal after the state is possibly different. 305 SVal ValueAtN = N->getState()->getSVal(RegionOfInterest); 306 if (!Mgr.getSValBuilder() 307 .areEqual(State, ValueAtN, ValueAfter) 308 .isConstrainedTrue() && 309 (!ValueAtN.isUndef() || !ValueAfter.isUndef())) 310 return true; 311 312 return false; 313 } 314 315 //===----------------------------------------------------------------------===// 316 // Implementation of BugReporterVisitor. 317 //===----------------------------------------------------------------------===// 318 319 PathDiagnosticPieceRef BugReporterVisitor::getEndPath(BugReporterContext &, 320 const ExplodedNode *, 321 PathSensitiveBugReport &) { 322 return nullptr; 323 } 324 325 void BugReporterVisitor::finalizeVisitor(BugReporterContext &, 326 const ExplodedNode *, 327 PathSensitiveBugReport &) {} 328 329 PathDiagnosticPieceRef 330 BugReporterVisitor::getDefaultEndPath(const BugReporterContext &BRC, 331 const ExplodedNode *EndPathNode, 332 const PathSensitiveBugReport &BR) { 333 PathDiagnosticLocation L = BR.getLocation(); 334 const auto &Ranges = BR.getRanges(); 335 336 // Only add the statement itself as a range if we didn't specify any 337 // special ranges for this report. 338 auto P = std::make_shared<PathDiagnosticEventPiece>( 339 L, BR.getDescription(), Ranges.begin() == Ranges.end()); 340 for (SourceRange Range : Ranges) 341 P->addRange(Range); 342 343 return P; 344 } 345 346 //===----------------------------------------------------------------------===// 347 // Implementation of NoStoreFuncVisitor. 348 //===----------------------------------------------------------------------===// 349 350 namespace { 351 352 /// Put a diagnostic on return statement of all inlined functions 353 /// for which the region of interest \p RegionOfInterest was passed into, 354 /// but not written inside, and it has caused an undefined read or a null 355 /// pointer dereference outside. 356 class NoStoreFuncVisitor final : public BugReporterVisitor { 357 const SubRegion *RegionOfInterest; 358 MemRegionManager &MmrMgr; 359 const SourceManager &SM; 360 const PrintingPolicy &PP; 361 bugreporter::TrackingKind TKind; 362 363 /// Recursion limit for dereferencing fields when looking for the 364 /// region of interest. 365 /// The limit of two indicates that we will dereference fields only once. 366 static const unsigned DEREFERENCE_LIMIT = 2; 367 368 /// Frames writing into \c RegionOfInterest. 369 /// This visitor generates a note only if a function does not write into 370 /// a region of interest. This information is not immediately available 371 /// by looking at the node associated with the exit from the function 372 /// (usually the return statement). To avoid recomputing the same information 373 /// many times (going up the path for each node and checking whether the 374 /// region was written into) we instead lazily compute the 375 /// stack frames along the path which write into the region of interest. 376 llvm::SmallPtrSet<const StackFrameContext *, 32> FramesModifyingRegion; 377 llvm::SmallPtrSet<const StackFrameContext *, 32> FramesModifyingCalculated; 378 379 using RegionVector = SmallVector<const MemRegion *, 5>; 380 381 public: 382 NoStoreFuncVisitor(const SubRegion *R, bugreporter::TrackingKind TKind) 383 : RegionOfInterest(R), MmrMgr(R->getMemRegionManager()), 384 SM(MmrMgr.getContext().getSourceManager()), 385 PP(MmrMgr.getContext().getPrintingPolicy()), TKind(TKind) {} 386 387 void Profile(llvm::FoldingSetNodeID &ID) const override { 388 static int Tag = 0; 389 ID.AddPointer(&Tag); 390 ID.AddPointer(RegionOfInterest); 391 } 392 393 void *getTag() const { 394 static int Tag = 0; 395 return static_cast<void *>(&Tag); 396 } 397 398 PathDiagnosticPieceRef VisitNode(const ExplodedNode *N, 399 BugReporterContext &BR, 400 PathSensitiveBugReport &R) override; 401 402 private: 403 /// Attempts to find the region of interest in a given record decl, 404 /// by either following the base classes or fields. 405 /// Dereferences fields up to a given recursion limit. 406 /// Note that \p Vec is passed by value, leading to quadratic copying cost, 407 /// but it's OK in practice since its length is limited to DEREFERENCE_LIMIT. 408 /// \return A chain fields leading to the region of interest or None. 409 const Optional<RegionVector> 410 findRegionOfInterestInRecord(const RecordDecl *RD, ProgramStateRef State, 411 const MemRegion *R, const RegionVector &Vec = {}, 412 int depth = 0); 413 414 /// Check and lazily calculate whether the region of interest is 415 /// modified in the stack frame to which \p N belongs. 416 /// The calculation is cached in FramesModifyingRegion. 417 bool isRegionOfInterestModifiedInFrame(const ExplodedNode *N) { 418 const LocationContext *Ctx = N->getLocationContext(); 419 const StackFrameContext *SCtx = Ctx->getStackFrame(); 420 if (!FramesModifyingCalculated.count(SCtx)) 421 findModifyingFrames(N); 422 return FramesModifyingRegion.count(SCtx); 423 } 424 425 /// Write to \c FramesModifyingRegion all stack frames along 426 /// the path in the current stack frame which modify \c RegionOfInterest. 427 void findModifyingFrames(const ExplodedNode *N); 428 429 /// Consume the information on the no-store stack frame in order to 430 /// either emit a note or suppress the report enirely. 431 /// \return Diagnostics piece for region not modified in the current function, 432 /// if it decides to emit one. 433 PathDiagnosticPieceRef 434 maybeEmitNote(PathSensitiveBugReport &R, const CallEvent &Call, 435 const ExplodedNode *N, const RegionVector &FieldChain, 436 const MemRegion *MatchedRegion, StringRef FirstElement, 437 bool FirstIsReferenceType, unsigned IndirectionLevel); 438 439 /// Pretty-print region \p MatchedRegion to \p os. 440 /// \return Whether printing succeeded. 441 bool prettyPrintRegionName(StringRef FirstElement, bool FirstIsReferenceType, 442 const MemRegion *MatchedRegion, 443 const RegionVector &FieldChain, 444 int IndirectionLevel, 445 llvm::raw_svector_ostream &os); 446 447 /// Print first item in the chain, return new separator. 448 static StringRef prettyPrintFirstElement(StringRef FirstElement, 449 bool MoreItemsExpected, 450 int IndirectionLevel, 451 llvm::raw_svector_ostream &os); 452 }; 453 454 } // end of anonymous namespace 455 456 /// \return Whether the method declaration \p Parent 457 /// syntactically has a binary operation writing into the ivar \p Ivar. 458 static bool potentiallyWritesIntoIvar(const Decl *Parent, 459 const ObjCIvarDecl *Ivar) { 460 using namespace ast_matchers; 461 const char *IvarBind = "Ivar"; 462 if (!Parent || !Parent->hasBody()) 463 return false; 464 StatementMatcher WriteIntoIvarM = binaryOperator( 465 hasOperatorName("="), 466 hasLHS(ignoringParenImpCasts( 467 objcIvarRefExpr(hasDeclaration(equalsNode(Ivar))).bind(IvarBind)))); 468 StatementMatcher ParentM = stmt(hasDescendant(WriteIntoIvarM)); 469 auto Matches = match(ParentM, *Parent->getBody(), Parent->getASTContext()); 470 for (BoundNodes &Match : Matches) { 471 auto IvarRef = Match.getNodeAs<ObjCIvarRefExpr>(IvarBind); 472 if (IvarRef->isFreeIvar()) 473 return true; 474 475 const Expr *Base = IvarRef->getBase(); 476 if (const auto *ICE = dyn_cast<ImplicitCastExpr>(Base)) 477 Base = ICE->getSubExpr(); 478 479 if (const auto *DRE = dyn_cast<DeclRefExpr>(Base)) 480 if (const auto *ID = dyn_cast<ImplicitParamDecl>(DRE->getDecl())) 481 if (ID->getParameterKind() == ImplicitParamDecl::ObjCSelf) 482 return true; 483 484 return false; 485 } 486 return false; 487 } 488 489 /// Get parameters associated with runtime definition in order 490 /// to get the correct parameter name. 491 static ArrayRef<ParmVarDecl *> getCallParameters(CallEventRef<> Call) { 492 // Use runtime definition, if available. 493 RuntimeDefinition RD = Call->getRuntimeDefinition(); 494 if (const auto *FD = dyn_cast_or_null<FunctionDecl>(RD.getDecl())) 495 return FD->parameters(); 496 if (const auto *MD = dyn_cast_or_null<ObjCMethodDecl>(RD.getDecl())) 497 return MD->parameters(); 498 499 return Call->parameters(); 500 } 501 502 /// \return whether \p Ty points to a const type, or is a const reference. 503 static bool isPointerToConst(QualType Ty) { 504 return !Ty->getPointeeType().isNull() && 505 Ty->getPointeeType().getCanonicalType().isConstQualified(); 506 } 507 508 /// Attempts to find the region of interest in a given CXX decl, 509 /// by either following the base classes or fields. 510 /// Dereferences fields up to a given recursion limit. 511 /// Note that \p Vec is passed by value, leading to quadratic copying cost, 512 /// but it's OK in practice since its length is limited to DEREFERENCE_LIMIT. 513 /// \return A chain fields leading to the region of interest or None. 514 const Optional<NoStoreFuncVisitor::RegionVector> 515 NoStoreFuncVisitor::findRegionOfInterestInRecord( 516 const RecordDecl *RD, ProgramStateRef State, const MemRegion *R, 517 const NoStoreFuncVisitor::RegionVector &Vec /* = {} */, 518 int depth /* = 0 */) { 519 520 if (depth == DEREFERENCE_LIMIT) // Limit the recursion depth. 521 return None; 522 523 if (const auto *RDX = dyn_cast<CXXRecordDecl>(RD)) 524 if (!RDX->hasDefinition()) 525 return None; 526 527 // Recursively examine the base classes. 528 // Note that following base classes does not increase the recursion depth. 529 if (const auto *RDX = dyn_cast<CXXRecordDecl>(RD)) 530 for (const auto &II : RDX->bases()) 531 if (const RecordDecl *RRD = II.getType()->getAsRecordDecl()) 532 if (Optional<RegionVector> Out = 533 findRegionOfInterestInRecord(RRD, State, R, Vec, depth)) 534 return Out; 535 536 for (const FieldDecl *I : RD->fields()) { 537 QualType FT = I->getType(); 538 const FieldRegion *FR = MmrMgr.getFieldRegion(I, cast<SubRegion>(R)); 539 const SVal V = State->getSVal(FR); 540 const MemRegion *VR = V.getAsRegion(); 541 542 RegionVector VecF = Vec; 543 VecF.push_back(FR); 544 545 if (RegionOfInterest == VR) 546 return VecF; 547 548 if (const RecordDecl *RRD = FT->getAsRecordDecl()) 549 if (auto Out = 550 findRegionOfInterestInRecord(RRD, State, FR, VecF, depth + 1)) 551 return Out; 552 553 QualType PT = FT->getPointeeType(); 554 if (PT.isNull() || PT->isVoidType() || !VR) 555 continue; 556 557 if (const RecordDecl *RRD = PT->getAsRecordDecl()) 558 if (Optional<RegionVector> Out = 559 findRegionOfInterestInRecord(RRD, State, VR, VecF, depth + 1)) 560 return Out; 561 } 562 563 return None; 564 } 565 566 PathDiagnosticPieceRef 567 NoStoreFuncVisitor::VisitNode(const ExplodedNode *N, BugReporterContext &BR, 568 PathSensitiveBugReport &R) { 569 570 const LocationContext *Ctx = N->getLocationContext(); 571 const StackFrameContext *SCtx = Ctx->getStackFrame(); 572 ProgramStateRef State = N->getState(); 573 auto CallExitLoc = N->getLocationAs<CallExitBegin>(); 574 575 // No diagnostic if region was modified inside the frame. 576 if (!CallExitLoc || isRegionOfInterestModifiedInFrame(N)) 577 return nullptr; 578 579 CallEventRef<> Call = 580 BR.getStateManager().getCallEventManager().getCaller(SCtx, State); 581 582 // Region of interest corresponds to an IVar, exiting a method 583 // which could have written into that IVar, but did not. 584 if (const auto *MC = dyn_cast<ObjCMethodCall>(Call)) { 585 if (const auto *IvarR = dyn_cast<ObjCIvarRegion>(RegionOfInterest)) { 586 const MemRegion *SelfRegion = MC->getReceiverSVal().getAsRegion(); 587 if (RegionOfInterest->isSubRegionOf(SelfRegion) && 588 potentiallyWritesIntoIvar(Call->getRuntimeDefinition().getDecl(), 589 IvarR->getDecl())) 590 return maybeEmitNote(R, *Call, N, {}, SelfRegion, "self", 591 /*FirstIsReferenceType=*/false, 1); 592 } 593 } 594 595 if (const auto *CCall = dyn_cast<CXXConstructorCall>(Call)) { 596 const MemRegion *ThisR = CCall->getCXXThisVal().getAsRegion(); 597 if (RegionOfInterest->isSubRegionOf(ThisR) && 598 !CCall->getDecl()->isImplicit()) 599 return maybeEmitNote(R, *Call, N, {}, ThisR, "this", 600 /*FirstIsReferenceType=*/false, 1); 601 602 // Do not generate diagnostics for not modified parameters in 603 // constructors. 604 return nullptr; 605 } 606 607 ArrayRef<ParmVarDecl *> parameters = getCallParameters(Call); 608 for (unsigned I = 0; I < Call->getNumArgs() && I < parameters.size(); ++I) { 609 const ParmVarDecl *PVD = parameters[I]; 610 SVal V = Call->getArgSVal(I); 611 bool ParamIsReferenceType = PVD->getType()->isReferenceType(); 612 std::string ParamName = PVD->getNameAsString(); 613 614 int IndirectionLevel = 1; 615 QualType T = PVD->getType(); 616 while (const MemRegion *MR = V.getAsRegion()) { 617 if (RegionOfInterest->isSubRegionOf(MR) && !isPointerToConst(T)) 618 return maybeEmitNote(R, *Call, N, {}, MR, ParamName, 619 ParamIsReferenceType, IndirectionLevel); 620 621 QualType PT = T->getPointeeType(); 622 if (PT.isNull() || PT->isVoidType()) 623 break; 624 625 if (const RecordDecl *RD = PT->getAsRecordDecl()) 626 if (Optional<RegionVector> P = 627 findRegionOfInterestInRecord(RD, State, MR)) 628 return maybeEmitNote(R, *Call, N, *P, RegionOfInterest, ParamName, 629 ParamIsReferenceType, IndirectionLevel); 630 631 V = State->getSVal(MR, PT); 632 T = PT; 633 IndirectionLevel++; 634 } 635 } 636 637 return nullptr; 638 } 639 640 void NoStoreFuncVisitor::findModifyingFrames(const ExplodedNode *N) { 641 assert(N->getLocationAs<CallExitBegin>()); 642 ProgramStateRef LastReturnState = N->getState(); 643 SVal ValueAtReturn = LastReturnState->getSVal(RegionOfInterest); 644 const LocationContext *Ctx = N->getLocationContext(); 645 const StackFrameContext *OriginalSCtx = Ctx->getStackFrame(); 646 647 do { 648 ProgramStateRef State = N->getState(); 649 auto CallExitLoc = N->getLocationAs<CallExitBegin>(); 650 if (CallExitLoc) { 651 LastReturnState = State; 652 ValueAtReturn = LastReturnState->getSVal(RegionOfInterest); 653 } 654 655 FramesModifyingCalculated.insert(N->getLocationContext()->getStackFrame()); 656 657 if (wasRegionOfInterestModifiedAt(RegionOfInterest, N, ValueAtReturn)) { 658 const StackFrameContext *SCtx = N->getStackFrame(); 659 while (!SCtx->inTopFrame()) { 660 auto p = FramesModifyingRegion.insert(SCtx); 661 if (!p.second) 662 break; // Frame and all its parents already inserted. 663 SCtx = SCtx->getParent()->getStackFrame(); 664 } 665 } 666 667 // Stop calculation at the call to the current function. 668 if (auto CE = N->getLocationAs<CallEnter>()) 669 if (CE->getCalleeContext() == OriginalSCtx) 670 break; 671 672 N = N->getFirstPred(); 673 } while (N); 674 } 675 676 static llvm::StringLiteral WillBeUsedForACondition = 677 ", which participates in a condition later"; 678 679 PathDiagnosticPieceRef NoStoreFuncVisitor::maybeEmitNote( 680 PathSensitiveBugReport &R, const CallEvent &Call, const ExplodedNode *N, 681 const RegionVector &FieldChain, const MemRegion *MatchedRegion, 682 StringRef FirstElement, bool FirstIsReferenceType, 683 unsigned IndirectionLevel) { 684 // Optimistically suppress uninitialized value bugs that result 685 // from system headers having a chance to initialize the value 686 // but failing to do so. It's too unlikely a system header's fault. 687 // It's much more likely a situation in which the function has a failure 688 // mode that the user decided not to check. If we want to hunt such 689 // omitted checks, we should provide an explicit function-specific note 690 // describing the precondition under which the function isn't supposed to 691 // initialize its out-parameter, and additionally check that such 692 // precondition can actually be fulfilled on the current path. 693 if (Call.isInSystemHeader()) { 694 // We make an exception for system header functions that have no branches. 695 // Such functions unconditionally fail to initialize the variable. 696 // If they call other functions that have more paths within them, 697 // this suppression would still apply when we visit these inner functions. 698 // One common example of a standard function that doesn't ever initialize 699 // its out parameter is operator placement new; it's up to the follow-up 700 // constructor (if any) to initialize the memory. 701 if (!N->getStackFrame()->getCFG()->isLinear()) 702 R.markInvalid(getTag(), nullptr); 703 return nullptr; 704 } 705 706 PathDiagnosticLocation L = 707 PathDiagnosticLocation::create(N->getLocation(), SM); 708 709 // For now this shouldn't trigger, but once it does (as we add more 710 // functions to the body farm), we'll need to decide if these reports 711 // are worth suppressing as well. 712 if (!L.hasValidLocation()) 713 return nullptr; 714 715 SmallString<256> sbuf; 716 llvm::raw_svector_ostream os(sbuf); 717 os << "Returning without writing to '"; 718 719 // Do not generate the note if failed to pretty-print. 720 if (!prettyPrintRegionName(FirstElement, FirstIsReferenceType, MatchedRegion, 721 FieldChain, IndirectionLevel, os)) 722 return nullptr; 723 724 os << "'"; 725 if (TKind == bugreporter::TrackingKind::Condition) 726 os << WillBeUsedForACondition; 727 return std::make_shared<PathDiagnosticEventPiece>(L, os.str()); 728 } 729 730 bool NoStoreFuncVisitor::prettyPrintRegionName(StringRef FirstElement, 731 bool FirstIsReferenceType, 732 const MemRegion *MatchedRegion, 733 const RegionVector &FieldChain, 734 int IndirectionLevel, 735 llvm::raw_svector_ostream &os) { 736 737 if (FirstIsReferenceType) 738 IndirectionLevel--; 739 740 RegionVector RegionSequence; 741 742 // Add the regions in the reverse order, then reverse the resulting array. 743 assert(RegionOfInterest->isSubRegionOf(MatchedRegion)); 744 const MemRegion *R = RegionOfInterest; 745 while (R != MatchedRegion) { 746 RegionSequence.push_back(R); 747 R = cast<SubRegion>(R)->getSuperRegion(); 748 } 749 std::reverse(RegionSequence.begin(), RegionSequence.end()); 750 RegionSequence.append(FieldChain.begin(), FieldChain.end()); 751 752 StringRef Sep; 753 for (const MemRegion *R : RegionSequence) { 754 755 // Just keep going up to the base region. 756 // Element regions may appear due to casts. 757 if (isa<CXXBaseObjectRegion>(R) || isa<CXXTempObjectRegion>(R)) 758 continue; 759 760 if (Sep.empty()) 761 Sep = prettyPrintFirstElement(FirstElement, 762 /*MoreItemsExpected=*/true, 763 IndirectionLevel, os); 764 765 os << Sep; 766 767 // Can only reasonably pretty-print DeclRegions. 768 if (!isa<DeclRegion>(R)) 769 return false; 770 771 const auto *DR = cast<DeclRegion>(R); 772 Sep = DR->getValueType()->isAnyPointerType() ? "->" : "."; 773 DR->getDecl()->getDeclName().print(os, PP); 774 } 775 776 if (Sep.empty()) 777 prettyPrintFirstElement(FirstElement, 778 /*MoreItemsExpected=*/false, IndirectionLevel, os); 779 return true; 780 } 781 782 StringRef NoStoreFuncVisitor::prettyPrintFirstElement( 783 StringRef FirstElement, bool MoreItemsExpected, int IndirectionLevel, 784 llvm::raw_svector_ostream &os) { 785 StringRef Out = "."; 786 787 if (IndirectionLevel > 0 && MoreItemsExpected) { 788 IndirectionLevel--; 789 Out = "->"; 790 } 791 792 if (IndirectionLevel > 0 && MoreItemsExpected) 793 os << "("; 794 795 for (int i = 0; i < IndirectionLevel; i++) 796 os << "*"; 797 os << FirstElement; 798 799 if (IndirectionLevel > 0 && MoreItemsExpected) 800 os << ")"; 801 802 return Out; 803 } 804 805 //===----------------------------------------------------------------------===// 806 // Implementation of MacroNullReturnSuppressionVisitor. 807 //===----------------------------------------------------------------------===// 808 809 namespace { 810 811 /// Suppress null-pointer-dereference bugs where dereferenced null was returned 812 /// the macro. 813 class MacroNullReturnSuppressionVisitor final : public BugReporterVisitor { 814 const SubRegion *RegionOfInterest; 815 const SVal ValueAtDereference; 816 817 // Do not invalidate the reports where the value was modified 818 // after it got assigned to from the macro. 819 bool WasModified = false; 820 821 public: 822 MacroNullReturnSuppressionVisitor(const SubRegion *R, const SVal V) 823 : RegionOfInterest(R), ValueAtDereference(V) {} 824 825 PathDiagnosticPieceRef VisitNode(const ExplodedNode *N, 826 BugReporterContext &BRC, 827 PathSensitiveBugReport &BR) override { 828 if (WasModified) 829 return nullptr; 830 831 auto BugPoint = BR.getErrorNode()->getLocation().getAs<StmtPoint>(); 832 if (!BugPoint) 833 return nullptr; 834 835 const SourceManager &SMgr = BRC.getSourceManager(); 836 if (auto Loc = matchAssignment(N)) { 837 if (isFunctionMacroExpansion(*Loc, SMgr)) { 838 std::string MacroName = std::string(getMacroName(*Loc, BRC)); 839 SourceLocation BugLoc = BugPoint->getStmt()->getBeginLoc(); 840 if (!BugLoc.isMacroID() || getMacroName(BugLoc, BRC) != MacroName) 841 BR.markInvalid(getTag(), MacroName.c_str()); 842 } 843 } 844 845 if (wasRegionOfInterestModifiedAt(RegionOfInterest, N, ValueAtDereference)) 846 WasModified = true; 847 848 return nullptr; 849 } 850 851 static void addMacroVisitorIfNecessary( 852 const ExplodedNode *N, const MemRegion *R, 853 bool EnableNullFPSuppression, PathSensitiveBugReport &BR, 854 const SVal V) { 855 AnalyzerOptions &Options = N->getState()->getAnalysisManager().options; 856 if (EnableNullFPSuppression && Options.ShouldSuppressNullReturnPaths && 857 V.getAs<Loc>()) 858 BR.addVisitor<MacroNullReturnSuppressionVisitor>(R->getAs<SubRegion>(), 859 V); 860 } 861 862 void* getTag() const { 863 static int Tag = 0; 864 return static_cast<void *>(&Tag); 865 } 866 867 void Profile(llvm::FoldingSetNodeID &ID) const override { 868 ID.AddPointer(getTag()); 869 } 870 871 private: 872 /// \return Source location of right hand side of an assignment 873 /// into \c RegionOfInterest, empty optional if none found. 874 Optional<SourceLocation> matchAssignment(const ExplodedNode *N) { 875 const Stmt *S = N->getStmtForDiagnostics(); 876 ProgramStateRef State = N->getState(); 877 auto *LCtx = N->getLocationContext(); 878 if (!S) 879 return None; 880 881 if (const auto *DS = dyn_cast<DeclStmt>(S)) { 882 if (const auto *VD = dyn_cast<VarDecl>(DS->getSingleDecl())) 883 if (const Expr *RHS = VD->getInit()) 884 if (RegionOfInterest->isSubRegionOf( 885 State->getLValue(VD, LCtx).getAsRegion())) 886 return RHS->getBeginLoc(); 887 } else if (const auto *BO = dyn_cast<BinaryOperator>(S)) { 888 const MemRegion *R = N->getSVal(BO->getLHS()).getAsRegion(); 889 const Expr *RHS = BO->getRHS(); 890 if (BO->isAssignmentOp() && RegionOfInterest->isSubRegionOf(R)) { 891 return RHS->getBeginLoc(); 892 } 893 } 894 return None; 895 } 896 }; 897 898 } // end of anonymous namespace 899 900 namespace { 901 902 /// Emits an extra note at the return statement of an interesting stack frame. 903 /// 904 /// The returned value is marked as an interesting value, and if it's null, 905 /// adds a visitor to track where it became null. 906 /// 907 /// This visitor is intended to be used when another visitor discovers that an 908 /// interesting value comes from an inlined function call. 909 class ReturnVisitor : public TrackingBugReporterVisitor { 910 const StackFrameContext *CalleeSFC; 911 enum { 912 Initial, 913 MaybeUnsuppress, 914 Satisfied 915 } Mode = Initial; 916 917 bool EnableNullFPSuppression; 918 bool ShouldInvalidate = true; 919 AnalyzerOptions& Options; 920 bugreporter::TrackingKind TKind; 921 922 public: 923 ReturnVisitor(TrackerRef ParentTracker, const StackFrameContext *Frame, 924 bool Suppressed, AnalyzerOptions &Options, 925 bugreporter::TrackingKind TKind) 926 : TrackingBugReporterVisitor(ParentTracker), CalleeSFC(Frame), 927 EnableNullFPSuppression(Suppressed), Options(Options), TKind(TKind) {} 928 929 static void *getTag() { 930 static int Tag = 0; 931 return static_cast<void *>(&Tag); 932 } 933 934 void Profile(llvm::FoldingSetNodeID &ID) const override { 935 ID.AddPointer(ReturnVisitor::getTag()); 936 ID.AddPointer(CalleeSFC); 937 ID.AddBoolean(EnableNullFPSuppression); 938 } 939 940 /// Adds a ReturnVisitor if the given statement represents a call that was 941 /// inlined. 942 /// 943 /// This will search back through the ExplodedGraph, starting from the given 944 /// node, looking for when the given statement was processed. If it turns out 945 /// the statement is a call that was inlined, we add the visitor to the 946 /// bug report, so it can print a note later. 947 static void addVisitorIfNecessary(TrackerRef ParentTracker, 948 const ExplodedNode *Node, const Stmt *S, 949 PathSensitiveBugReport &BR, 950 bool InEnableNullFPSuppression, 951 bugreporter::TrackingKind TKind) { 952 if (!CallEvent::isCallStmt(S)) 953 return; 954 955 // First, find when we processed the statement. 956 // If we work with a 'CXXNewExpr' that is going to be purged away before 957 // its call take place. We would catch that purge in the last condition 958 // as a 'StmtPoint' so we have to bypass it. 959 const bool BypassCXXNewExprEval = isa<CXXNewExpr>(S); 960 961 // This is moving forward when we enter into another context. 962 const StackFrameContext *CurrentSFC = Node->getStackFrame(); 963 964 do { 965 // If that is satisfied we found our statement as an inlined call. 966 if (Optional<CallExitEnd> CEE = Node->getLocationAs<CallExitEnd>()) 967 if (CEE->getCalleeContext()->getCallSite() == S) 968 break; 969 970 // Try to move forward to the end of the call-chain. 971 Node = Node->getFirstPred(); 972 if (!Node) 973 break; 974 975 const StackFrameContext *PredSFC = Node->getStackFrame(); 976 977 // If that is satisfied we found our statement. 978 // FIXME: This code currently bypasses the call site for the 979 // conservatively evaluated allocator. 980 if (!BypassCXXNewExprEval) 981 if (Optional<StmtPoint> SP = Node->getLocationAs<StmtPoint>()) 982 // See if we do not enter into another context. 983 if (SP->getStmt() == S && CurrentSFC == PredSFC) 984 break; 985 986 CurrentSFC = PredSFC; 987 } while (Node->getStackFrame() == CurrentSFC); 988 989 // Next, step over any post-statement checks. 990 while (Node && Node->getLocation().getAs<PostStmt>()) 991 Node = Node->getFirstPred(); 992 if (!Node) 993 return; 994 995 // Finally, see if we inlined the call. 996 Optional<CallExitEnd> CEE = Node->getLocationAs<CallExitEnd>(); 997 if (!CEE) 998 return; 999 1000 const StackFrameContext *CalleeContext = CEE->getCalleeContext(); 1001 if (CalleeContext->getCallSite() != S) 1002 return; 1003 1004 // Check the return value. 1005 ProgramStateRef State = Node->getState(); 1006 SVal RetVal = Node->getSVal(S); 1007 1008 // Handle cases where a reference is returned and then immediately used. 1009 if (cast<Expr>(S)->isGLValue()) 1010 if (Optional<Loc> LValue = RetVal.getAs<Loc>()) 1011 RetVal = State->getSVal(*LValue); 1012 1013 // See if the return value is NULL. If so, suppress the report. 1014 AnalyzerOptions &Options = State->getAnalysisManager().options; 1015 1016 bool EnableNullFPSuppression = false; 1017 if (InEnableNullFPSuppression && Options.ShouldSuppressNullReturnPaths) 1018 if (Optional<Loc> RetLoc = RetVal.getAs<Loc>()) 1019 EnableNullFPSuppression = State->isNull(*RetLoc).isConstrainedTrue(); 1020 1021 BR.addVisitor<ReturnVisitor>(ParentTracker, CalleeContext, 1022 EnableNullFPSuppression, Options, TKind); 1023 } 1024 1025 PathDiagnosticPieceRef visitNodeInitial(const ExplodedNode *N, 1026 BugReporterContext &BRC, 1027 PathSensitiveBugReport &BR) { 1028 // Only print a message at the interesting return statement. 1029 if (N->getLocationContext() != CalleeSFC) 1030 return nullptr; 1031 1032 Optional<StmtPoint> SP = N->getLocationAs<StmtPoint>(); 1033 if (!SP) 1034 return nullptr; 1035 1036 const auto *Ret = dyn_cast<ReturnStmt>(SP->getStmt()); 1037 if (!Ret) 1038 return nullptr; 1039 1040 // Okay, we're at the right return statement, but do we have the return 1041 // value available? 1042 ProgramStateRef State = N->getState(); 1043 SVal V = State->getSVal(Ret, CalleeSFC); 1044 if (V.isUnknownOrUndef()) 1045 return nullptr; 1046 1047 // Don't print any more notes after this one. 1048 Mode = Satisfied; 1049 1050 const Expr *RetE = Ret->getRetValue(); 1051 assert(RetE && "Tracking a return value for a void function"); 1052 1053 // Handle cases where a reference is returned and then immediately used. 1054 Optional<Loc> LValue; 1055 if (RetE->isGLValue()) { 1056 if ((LValue = V.getAs<Loc>())) { 1057 SVal RValue = State->getRawSVal(*LValue, RetE->getType()); 1058 if (RValue.getAs<DefinedSVal>()) 1059 V = RValue; 1060 } 1061 } 1062 1063 // Ignore aggregate rvalues. 1064 if (V.getAs<nonloc::LazyCompoundVal>() || 1065 V.getAs<nonloc::CompoundVal>()) 1066 return nullptr; 1067 1068 RetE = RetE->IgnoreParenCasts(); 1069 1070 // Let's track the return value. 1071 getParentTracker().track(RetE, N, {TKind, EnableNullFPSuppression}); 1072 1073 // Build an appropriate message based on the return value. 1074 SmallString<64> Msg; 1075 llvm::raw_svector_ostream Out(Msg); 1076 1077 bool WouldEventBeMeaningless = false; 1078 1079 if (State->isNull(V).isConstrainedTrue()) { 1080 if (V.getAs<Loc>()) { 1081 1082 // If we have counter-suppression enabled, make sure we keep visiting 1083 // future nodes. We want to emit a path note as well, in case 1084 // the report is resurrected as valid later on. 1085 if (EnableNullFPSuppression && 1086 Options.ShouldAvoidSuppressingNullArgumentPaths) 1087 Mode = MaybeUnsuppress; 1088 1089 if (RetE->getType()->isObjCObjectPointerType()) { 1090 Out << "Returning nil"; 1091 } else { 1092 Out << "Returning null pointer"; 1093 } 1094 } else { 1095 Out << "Returning zero"; 1096 } 1097 1098 } else { 1099 if (auto CI = V.getAs<nonloc::ConcreteInt>()) { 1100 Out << "Returning the value " << CI->getValue(); 1101 } else { 1102 // There is nothing interesting about returning a value, when it is 1103 // plain value without any constraints, and the function is guaranteed 1104 // to return that every time. We could use CFG::isLinear() here, but 1105 // constexpr branches are obvious to the compiler, not necesserily to 1106 // the programmer. 1107 if (N->getCFG().size() == 3) 1108 WouldEventBeMeaningless = true; 1109 1110 if (V.getAs<Loc>()) 1111 Out << "Returning pointer"; 1112 else 1113 Out << "Returning value"; 1114 } 1115 } 1116 1117 if (LValue) { 1118 if (const MemRegion *MR = LValue->getAsRegion()) { 1119 if (MR->canPrintPretty()) { 1120 Out << " (reference to "; 1121 MR->printPretty(Out); 1122 Out << ")"; 1123 } 1124 } 1125 } else { 1126 // FIXME: We should have a more generalized location printing mechanism. 1127 if (const auto *DR = dyn_cast<DeclRefExpr>(RetE)) 1128 if (const auto *DD = dyn_cast<DeclaratorDecl>(DR->getDecl())) 1129 Out << " (loaded from '" << *DD << "')"; 1130 } 1131 1132 PathDiagnosticLocation L(Ret, BRC.getSourceManager(), CalleeSFC); 1133 if (!L.isValid() || !L.asLocation().isValid()) 1134 return nullptr; 1135 1136 if (TKind == bugreporter::TrackingKind::Condition) 1137 Out << WillBeUsedForACondition; 1138 1139 auto EventPiece = std::make_shared<PathDiagnosticEventPiece>(L, Out.str()); 1140 1141 // If we determined that the note is meaningless, make it prunable, and 1142 // don't mark the stackframe interesting. 1143 if (WouldEventBeMeaningless) 1144 EventPiece->setPrunable(true); 1145 else 1146 BR.markInteresting(CalleeSFC); 1147 1148 return EventPiece; 1149 } 1150 1151 PathDiagnosticPieceRef visitNodeMaybeUnsuppress(const ExplodedNode *N, 1152 BugReporterContext &BRC, 1153 PathSensitiveBugReport &BR) { 1154 assert(Options.ShouldAvoidSuppressingNullArgumentPaths); 1155 1156 // Are we at the entry node for this call? 1157 Optional<CallEnter> CE = N->getLocationAs<CallEnter>(); 1158 if (!CE) 1159 return nullptr; 1160 1161 if (CE->getCalleeContext() != CalleeSFC) 1162 return nullptr; 1163 1164 Mode = Satisfied; 1165 1166 // Don't automatically suppress a report if one of the arguments is 1167 // known to be a null pointer. Instead, start tracking /that/ null 1168 // value back to its origin. 1169 ProgramStateManager &StateMgr = BRC.getStateManager(); 1170 CallEventManager &CallMgr = StateMgr.getCallEventManager(); 1171 1172 ProgramStateRef State = N->getState(); 1173 CallEventRef<> Call = CallMgr.getCaller(CalleeSFC, State); 1174 for (unsigned I = 0, E = Call->getNumArgs(); I != E; ++I) { 1175 Optional<Loc> ArgV = Call->getArgSVal(I).getAs<Loc>(); 1176 if (!ArgV) 1177 continue; 1178 1179 const Expr *ArgE = Call->getArgExpr(I); 1180 if (!ArgE) 1181 continue; 1182 1183 // Is it possible for this argument to be non-null? 1184 if (!State->isNull(*ArgV).isConstrainedTrue()) 1185 continue; 1186 1187 if (getParentTracker() 1188 .track(ArgE, N, {TKind, EnableNullFPSuppression}) 1189 .FoundSomethingToTrack) 1190 ShouldInvalidate = false; 1191 1192 // If we /can't/ track the null pointer, we should err on the side of 1193 // false negatives, and continue towards marking this report invalid. 1194 // (We will still look at the other arguments, though.) 1195 } 1196 1197 return nullptr; 1198 } 1199 1200 PathDiagnosticPieceRef VisitNode(const ExplodedNode *N, 1201 BugReporterContext &BRC, 1202 PathSensitiveBugReport &BR) override { 1203 switch (Mode) { 1204 case Initial: 1205 return visitNodeInitial(N, BRC, BR); 1206 case MaybeUnsuppress: 1207 return visitNodeMaybeUnsuppress(N, BRC, BR); 1208 case Satisfied: 1209 return nullptr; 1210 } 1211 1212 llvm_unreachable("Invalid visit mode!"); 1213 } 1214 1215 void finalizeVisitor(BugReporterContext &, const ExplodedNode *, 1216 PathSensitiveBugReport &BR) override { 1217 if (EnableNullFPSuppression && ShouldInvalidate) 1218 BR.markInvalid(ReturnVisitor::getTag(), CalleeSFC); 1219 } 1220 }; 1221 1222 } // end of anonymous namespace 1223 1224 //===----------------------------------------------------------------------===// 1225 // StoreSiteFinder 1226 //===----------------------------------------------------------------------===// 1227 1228 /// Finds last store into the given region, 1229 /// which is different from a given symbolic value. 1230 class StoreSiteFinder final : public TrackingBugReporterVisitor { 1231 const MemRegion *R; 1232 SVal V; 1233 bool Satisfied = false; 1234 1235 /// If the visitor is tracking the value directly responsible for the 1236 /// bug, we are going to employ false positive suppression. 1237 bool EnableNullFPSuppression; 1238 1239 using TrackingKind = bugreporter::TrackingKind; 1240 TrackingKind TKind; 1241 const StackFrameContext *OriginSFC; 1242 1243 public: 1244 /// \param V We're searching for the store where \c R received this value. 1245 /// \param R The region we're tracking. 1246 /// \param TKind May limit the amount of notes added to the bug report. 1247 /// \param OriginSFC Only adds notes when the last store happened in a 1248 /// different stackframe to this one. Disregarded if the tracking kind 1249 /// is thorough. 1250 /// This is useful, because for non-tracked regions, notes about 1251 /// changes to its value in a nested stackframe could be pruned, and 1252 /// this visitor can prevent that without polluting the bugpath too 1253 /// much. 1254 StoreSiteFinder(bugreporter::TrackerRef ParentTracker, KnownSVal V, 1255 const MemRegion *R, bool InEnableNullFPSuppression, 1256 TrackingKind TKind, 1257 const StackFrameContext *OriginSFC = nullptr) 1258 : TrackingBugReporterVisitor(ParentTracker), R(R), V(V), 1259 EnableNullFPSuppression(InEnableNullFPSuppression), TKind(TKind), 1260 OriginSFC(OriginSFC) { 1261 assert(R); 1262 } 1263 1264 void Profile(llvm::FoldingSetNodeID &ID) const override; 1265 1266 PathDiagnosticPieceRef VisitNode(const ExplodedNode *N, 1267 BugReporterContext &BRC, 1268 PathSensitiveBugReport &BR) override; 1269 }; 1270 1271 void StoreSiteFinder::Profile(llvm::FoldingSetNodeID &ID) const { 1272 static int tag = 0; 1273 ID.AddPointer(&tag); 1274 ID.AddPointer(R); 1275 ID.Add(V); 1276 ID.AddInteger(static_cast<int>(TKind)); 1277 ID.AddBoolean(EnableNullFPSuppression); 1278 } 1279 1280 /// Returns true if \p N represents the DeclStmt declaring and initializing 1281 /// \p VR. 1282 static bool isInitializationOfVar(const ExplodedNode *N, const VarRegion *VR) { 1283 Optional<PostStmt> P = N->getLocationAs<PostStmt>(); 1284 if (!P) 1285 return false; 1286 1287 const DeclStmt *DS = P->getStmtAs<DeclStmt>(); 1288 if (!DS) 1289 return false; 1290 1291 if (DS->getSingleDecl() != VR->getDecl()) 1292 return false; 1293 1294 const MemSpaceRegion *VarSpace = VR->getMemorySpace(); 1295 const auto *FrameSpace = dyn_cast<StackSpaceRegion>(VarSpace); 1296 if (!FrameSpace) { 1297 // If we ever directly evaluate global DeclStmts, this assertion will be 1298 // invalid, but this still seems preferable to silently accepting an 1299 // initialization that may be for a path-sensitive variable. 1300 assert(VR->getDecl()->isStaticLocal() && "non-static stackless VarRegion"); 1301 return true; 1302 } 1303 1304 assert(VR->getDecl()->hasLocalStorage()); 1305 const LocationContext *LCtx = N->getLocationContext(); 1306 return FrameSpace->getStackFrame() == LCtx->getStackFrame(); 1307 } 1308 1309 /// Show diagnostics for initializing or declaring a region \p R with a bad value. 1310 static void showBRDiagnostics(const char *action, llvm::raw_svector_ostream &os, 1311 const MemRegion *NewR, SVal V, 1312 const MemRegion *OldR, const DeclStmt *DS) { 1313 if (NewR->canPrintPretty()) { 1314 NewR->printPretty(os); 1315 os << " "; 1316 } 1317 1318 if (V.getAs<loc::ConcreteInt>()) { 1319 bool b = false; 1320 if (NewR->isBoundable()) { 1321 if (const auto *TR = dyn_cast<TypedValueRegion>(NewR)) { 1322 if (TR->getValueType()->isObjCObjectPointerType()) { 1323 os << action << "nil"; 1324 b = true; 1325 } 1326 } 1327 } 1328 if (!b) 1329 os << action << "a null pointer value"; 1330 1331 } else if (auto CVal = V.getAs<nonloc::ConcreteInt>()) { 1332 os << action << CVal->getValue(); 1333 } else if (OldR && OldR->canPrintPretty()) { 1334 os << action << "the value of "; 1335 OldR->printPretty(os); 1336 } else if (DS) { 1337 if (V.isUndef()) { 1338 if (isa<VarRegion>(NewR)) { 1339 const auto *VD = cast<VarDecl>(DS->getSingleDecl()); 1340 if (VD->getInit()) { 1341 os << (NewR->canPrintPretty() ? "initialized" : "Initializing") 1342 << " to a garbage value"; 1343 } else { 1344 os << (NewR->canPrintPretty() ? "declared" : "Declaring") 1345 << " without an initial value"; 1346 } 1347 } 1348 } else { 1349 os << (NewR->canPrintPretty() ? "initialized" : "Initialized") << " here"; 1350 } 1351 } 1352 } 1353 1354 /// Display diagnostics for passing bad region as a parameter. 1355 static void showBRParamDiagnostics(llvm::raw_svector_ostream &os, 1356 const VarRegion *VR, SVal V, 1357 const MemRegion *ValueR) { 1358 const auto *Param = cast<ParmVarDecl>(VR->getDecl()); 1359 1360 os << "Passing "; 1361 1362 if (V.getAs<loc::ConcreteInt>()) { 1363 if (Param->getType()->isObjCObjectPointerType()) 1364 os << "nil object reference"; 1365 else 1366 os << "null pointer value"; 1367 } else if (V.isUndef()) { 1368 os << "uninitialized value"; 1369 } else if (auto CI = V.getAs<nonloc::ConcreteInt>()) { 1370 os << "the value " << CI->getValue(); 1371 } else if (ValueR && ValueR->canPrintPretty()) { 1372 ValueR->printPretty(os); 1373 } else { 1374 os << "value"; 1375 } 1376 1377 // Printed parameter indexes are 1-based, not 0-based. 1378 unsigned Idx = Param->getFunctionScopeIndex() + 1; 1379 os << " via " << Idx << llvm::getOrdinalSuffix(Idx) << " parameter"; 1380 if (VR->canPrintPretty()) { 1381 os << " "; 1382 VR->printPretty(os); 1383 } 1384 } 1385 1386 /// Show default diagnostics for storing bad region. 1387 static void showBRDefaultDiagnostics(llvm::raw_svector_ostream &os, 1388 const MemRegion *NewR, SVal V, 1389 const MemRegion *OldR) { 1390 if (V.getAs<loc::ConcreteInt>()) { 1391 bool b = false; 1392 if (NewR->isBoundable()) { 1393 if (const auto *TR = dyn_cast<TypedValueRegion>(NewR)) { 1394 if (TR->getValueType()->isObjCObjectPointerType()) { 1395 os << "nil object reference stored"; 1396 b = true; 1397 } 1398 } 1399 } 1400 if (!b) { 1401 if (NewR->canPrintPretty()) 1402 os << "Null pointer value stored"; 1403 else 1404 os << "Storing null pointer value"; 1405 } 1406 1407 } else if (V.isUndef()) { 1408 if (NewR->canPrintPretty()) 1409 os << "Uninitialized value stored"; 1410 else 1411 os << "Storing uninitialized value"; 1412 1413 } else if (auto CV = V.getAs<nonloc::ConcreteInt>()) { 1414 if (NewR->canPrintPretty()) 1415 os << "The value " << CV->getValue() << " is assigned"; 1416 else 1417 os << "Assigning " << CV->getValue(); 1418 1419 } else if (OldR && OldR->canPrintPretty()) { 1420 if (NewR->canPrintPretty()) { 1421 os << "The value of "; 1422 OldR->printPretty(os); 1423 os << " is assigned"; 1424 } else { 1425 os << "Assigning the value of "; 1426 OldR->printPretty(os); 1427 } 1428 1429 } else { 1430 if (NewR->canPrintPretty()) 1431 os << "Value assigned"; 1432 else 1433 os << "Assigning value"; 1434 } 1435 1436 if (NewR->canPrintPretty()) { 1437 os << " to "; 1438 NewR->printPretty(os); 1439 } 1440 } 1441 1442 PathDiagnosticPieceRef StoreSiteFinder::VisitNode(const ExplodedNode *Succ, 1443 BugReporterContext &BRC, 1444 PathSensitiveBugReport &BR) { 1445 if (Satisfied) 1446 return nullptr; 1447 1448 const ExplodedNode *StoreSite = nullptr; 1449 const ExplodedNode *Pred = Succ->getFirstPred(); 1450 const Expr *InitE = nullptr; 1451 bool IsParam = false; 1452 1453 // First see if we reached the declaration of the region. 1454 if (const auto *VR = dyn_cast<VarRegion>(R)) { 1455 if (isInitializationOfVar(Pred, VR)) { 1456 StoreSite = Pred; 1457 InitE = VR->getDecl()->getInit(); 1458 } 1459 } 1460 1461 // If this is a post initializer expression, initializing the region, we 1462 // should track the initializer expression. 1463 if (Optional<PostInitializer> PIP = Pred->getLocationAs<PostInitializer>()) { 1464 const MemRegion *FieldReg = (const MemRegion *)PIP->getLocationValue(); 1465 if (FieldReg == R) { 1466 StoreSite = Pred; 1467 InitE = PIP->getInitializer()->getInit(); 1468 } 1469 } 1470 1471 // Otherwise, see if this is the store site: 1472 // (1) Succ has this binding and Pred does not, i.e. this is 1473 // where the binding first occurred. 1474 // (2) Succ has this binding and is a PostStore node for this region, i.e. 1475 // the same binding was re-assigned here. 1476 if (!StoreSite) { 1477 if (Succ->getState()->getSVal(R) != V) 1478 return nullptr; 1479 1480 if (hasVisibleUpdate(Pred, Pred->getState()->getSVal(R), Succ, V)) { 1481 Optional<PostStore> PS = Succ->getLocationAs<PostStore>(); 1482 if (!PS || PS->getLocationValue() != R) 1483 return nullptr; 1484 } 1485 1486 StoreSite = Succ; 1487 1488 // If this is an assignment expression, we can track the value 1489 // being assigned. 1490 if (Optional<PostStmt> P = Succ->getLocationAs<PostStmt>()) 1491 if (const BinaryOperator *BO = P->getStmtAs<BinaryOperator>()) 1492 if (BO->isAssignmentOp()) 1493 InitE = BO->getRHS(); 1494 1495 // If this is a call entry, the variable should be a parameter. 1496 // FIXME: Handle CXXThisRegion as well. (This is not a priority because 1497 // 'this' should never be NULL, but this visitor isn't just for NULL and 1498 // UndefinedVal.) 1499 if (Optional<CallEnter> CE = Succ->getLocationAs<CallEnter>()) { 1500 if (const auto *VR = dyn_cast<VarRegion>(R)) { 1501 1502 if (const auto *Param = dyn_cast<ParmVarDecl>(VR->getDecl())) { 1503 ProgramStateManager &StateMgr = BRC.getStateManager(); 1504 CallEventManager &CallMgr = StateMgr.getCallEventManager(); 1505 1506 CallEventRef<> Call = CallMgr.getCaller(CE->getCalleeContext(), 1507 Succ->getState()); 1508 InitE = Call->getArgExpr(Param->getFunctionScopeIndex()); 1509 } else { 1510 // Handle Objective-C 'self'. 1511 assert(isa<ImplicitParamDecl>(VR->getDecl())); 1512 InitE = cast<ObjCMessageExpr>(CE->getCalleeContext()->getCallSite()) 1513 ->getInstanceReceiver()->IgnoreParenCasts(); 1514 } 1515 IsParam = true; 1516 } 1517 } 1518 1519 // If this is a CXXTempObjectRegion, the Expr responsible for its creation 1520 // is wrapped inside of it. 1521 if (const auto *TmpR = dyn_cast<CXXTempObjectRegion>(R)) 1522 InitE = TmpR->getExpr(); 1523 } 1524 1525 if (!StoreSite) 1526 return nullptr; 1527 1528 Satisfied = true; 1529 1530 // If we have an expression that provided the value, try to track where it 1531 // came from. 1532 if (InitE) { 1533 if (!IsParam) 1534 InitE = InitE->IgnoreParenCasts(); 1535 1536 getParentTracker().track(InitE, StoreSite, 1537 {TKind, EnableNullFPSuppression}); 1538 } 1539 1540 // Let's try to find the region where the value came from. 1541 const MemRegion *OldRegion = nullptr; 1542 1543 // If we have init expression, it might be simply a reference 1544 // to a variable, so we can use it. 1545 if (InitE) { 1546 // That region might still be not exactly what we are looking for. 1547 // In situations like `int &ref = val;`, we can't say that 1548 // `ref` is initialized with `val`, rather refers to `val`. 1549 // 1550 // In order, to mitigate situations like this, we check if the last 1551 // stored value in that region is the value that we track. 1552 // 1553 // TODO: support other situations better. 1554 if (const MemRegion *Candidate = 1555 getLocationRegionIfReference(InitE, Succ, false)) { 1556 const StoreManager &SM = BRC.getStateManager().getStoreManager(); 1557 1558 // Here we traverse the graph up to find the last node where the 1559 // candidate region is still in the store. 1560 for (const ExplodedNode *N = StoreSite; N; N = N->getFirstPred()) { 1561 if (SM.includedInBindings(N->getState()->getStore(), Candidate)) { 1562 // And if it was bound to the target value, we can use it. 1563 if (N->getState()->getSVal(Candidate) == V) { 1564 OldRegion = Candidate; 1565 } 1566 break; 1567 } 1568 } 1569 } 1570 } 1571 1572 // Otherwise, if the current region does indeed contain the value 1573 // we are looking for, we can look for a region where this value 1574 // was before. 1575 // 1576 // It can be useful for situations like: 1577 // new = identity(old) 1578 // where the analyzer knows that 'identity' returns the value of its 1579 // first argument. 1580 // 1581 // NOTE: If the region R is not a simple var region, it can contain 1582 // V in one of its subregions. 1583 if (!OldRegion && StoreSite->getState()->getSVal(R) == V) { 1584 // Let's go up the graph to find the node where the region is 1585 // bound to V. 1586 const ExplodedNode *NodeWithoutBinding = StoreSite->getFirstPred(); 1587 for (; 1588 NodeWithoutBinding && NodeWithoutBinding->getState()->getSVal(R) == V; 1589 NodeWithoutBinding = NodeWithoutBinding->getFirstPred()) { 1590 } 1591 1592 if (NodeWithoutBinding) { 1593 // Let's try to find a unique binding for the value in that node. 1594 // We want to use this to find unique bindings because of the following 1595 // situations: 1596 // b = a; 1597 // c = identity(b); 1598 // 1599 // Telling the user that the value of 'a' is assigned to 'c', while 1600 // correct, can be confusing. 1601 StoreManager::FindUniqueBinding FB(V.getAsLocSymbol()); 1602 BRC.getStateManager().iterBindings(NodeWithoutBinding->getState(), FB); 1603 if (FB) 1604 OldRegion = FB.getRegion(); 1605 } 1606 } 1607 1608 if (TKind == TrackingKind::Condition && 1609 !OriginSFC->isParentOf(StoreSite->getStackFrame())) 1610 return nullptr; 1611 1612 // Okay, we've found the binding. Emit an appropriate message. 1613 SmallString<256> sbuf; 1614 llvm::raw_svector_ostream os(sbuf); 1615 1616 if (Optional<PostStmt> PS = StoreSite->getLocationAs<PostStmt>()) { 1617 const Stmt *S = PS->getStmt(); 1618 const char *action = nullptr; 1619 const auto *DS = dyn_cast<DeclStmt>(S); 1620 const auto *VR = dyn_cast<VarRegion>(R); 1621 1622 if (DS) { 1623 action = R->canPrintPretty() ? "initialized to " : 1624 "Initializing to "; 1625 } else if (isa<BlockExpr>(S)) { 1626 action = R->canPrintPretty() ? "captured by block as " : 1627 "Captured by block as "; 1628 if (VR) { 1629 // See if we can get the BlockVarRegion. 1630 ProgramStateRef State = StoreSite->getState(); 1631 SVal V = StoreSite->getSVal(S); 1632 if (const auto *BDR = 1633 dyn_cast_or_null<BlockDataRegion>(V.getAsRegion())) { 1634 if (const VarRegion *OriginalR = BDR->getOriginalRegion(VR)) { 1635 if (auto KV = State->getSVal(OriginalR).getAs<KnownSVal>()) 1636 getParentTracker().track( 1637 *KV, OriginalR, {TKind, EnableNullFPSuppression}, OriginSFC); 1638 } 1639 } 1640 } 1641 } 1642 if (action) 1643 showBRDiagnostics(action, os, R, V, OldRegion, DS); 1644 1645 } else if (StoreSite->getLocation().getAs<CallEnter>()) { 1646 if (const auto *VR = dyn_cast<VarRegion>(R)) 1647 showBRParamDiagnostics(os, VR, V, OldRegion); 1648 } 1649 1650 if (os.str().empty()) 1651 showBRDefaultDiagnostics(os, R, V, OldRegion); 1652 1653 if (TKind == bugreporter::TrackingKind::Condition) 1654 os << WillBeUsedForACondition; 1655 1656 // Construct a new PathDiagnosticPiece. 1657 ProgramPoint P = StoreSite->getLocation(); 1658 PathDiagnosticLocation L; 1659 if (P.getAs<CallEnter>() && InitE) 1660 L = PathDiagnosticLocation(InitE, BRC.getSourceManager(), 1661 P.getLocationContext()); 1662 1663 if (!L.isValid() || !L.asLocation().isValid()) 1664 L = PathDiagnosticLocation::create(P, BRC.getSourceManager()); 1665 1666 if (!L.isValid() || !L.asLocation().isValid()) 1667 return nullptr; 1668 1669 return std::make_shared<PathDiagnosticEventPiece>(L, os.str()); 1670 } 1671 1672 //===----------------------------------------------------------------------===// 1673 // Implementation of TrackConstraintBRVisitor. 1674 //===----------------------------------------------------------------------===// 1675 1676 void TrackConstraintBRVisitor::Profile(llvm::FoldingSetNodeID &ID) const { 1677 static int tag = 0; 1678 ID.AddPointer(&tag); 1679 ID.AddBoolean(Assumption); 1680 ID.Add(Constraint); 1681 } 1682 1683 /// Return the tag associated with this visitor. This tag will be used 1684 /// to make all PathDiagnosticPieces created by this visitor. 1685 const char *TrackConstraintBRVisitor::getTag() { 1686 return "TrackConstraintBRVisitor"; 1687 } 1688 1689 bool TrackConstraintBRVisitor::isUnderconstrained(const ExplodedNode *N) const { 1690 if (IsZeroCheck) 1691 return N->getState()->isNull(Constraint).isUnderconstrained(); 1692 return (bool)N->getState()->assume(Constraint, !Assumption); 1693 } 1694 1695 PathDiagnosticPieceRef TrackConstraintBRVisitor::VisitNode( 1696 const ExplodedNode *N, BugReporterContext &BRC, PathSensitiveBugReport &) { 1697 const ExplodedNode *PrevN = N->getFirstPred(); 1698 if (IsSatisfied) 1699 return nullptr; 1700 1701 // Start tracking after we see the first state in which the value is 1702 // constrained. 1703 if (!IsTrackingTurnedOn) 1704 if (!isUnderconstrained(N)) 1705 IsTrackingTurnedOn = true; 1706 if (!IsTrackingTurnedOn) 1707 return nullptr; 1708 1709 // Check if in the previous state it was feasible for this constraint 1710 // to *not* be true. 1711 if (isUnderconstrained(PrevN)) { 1712 IsSatisfied = true; 1713 1714 // As a sanity check, make sure that the negation of the constraint 1715 // was infeasible in the current state. If it is feasible, we somehow 1716 // missed the transition point. 1717 assert(!isUnderconstrained(N)); 1718 1719 // We found the transition point for the constraint. We now need to 1720 // pretty-print the constraint. (work-in-progress) 1721 SmallString<64> sbuf; 1722 llvm::raw_svector_ostream os(sbuf); 1723 1724 if (Constraint.getAs<Loc>()) { 1725 os << "Assuming pointer value is "; 1726 os << (Assumption ? "non-null" : "null"); 1727 } 1728 1729 if (os.str().empty()) 1730 return nullptr; 1731 1732 // Construct a new PathDiagnosticPiece. 1733 ProgramPoint P = N->getLocation(); 1734 PathDiagnosticLocation L = 1735 PathDiagnosticLocation::create(P, BRC.getSourceManager()); 1736 if (!L.isValid()) 1737 return nullptr; 1738 1739 auto X = std::make_shared<PathDiagnosticEventPiece>(L, os.str()); 1740 X->setTag(getTag()); 1741 return std::move(X); 1742 } 1743 1744 return nullptr; 1745 } 1746 1747 //===----------------------------------------------------------------------===// 1748 // Implementation of SuppressInlineDefensiveChecksVisitor. 1749 //===----------------------------------------------------------------------===// 1750 1751 SuppressInlineDefensiveChecksVisitor:: 1752 SuppressInlineDefensiveChecksVisitor(DefinedSVal Value, const ExplodedNode *N) 1753 : V(Value) { 1754 // Check if the visitor is disabled. 1755 AnalyzerOptions &Options = N->getState()->getAnalysisManager().options; 1756 if (!Options.ShouldSuppressInlinedDefensiveChecks) 1757 IsSatisfied = true; 1758 } 1759 1760 void SuppressInlineDefensiveChecksVisitor::Profile( 1761 llvm::FoldingSetNodeID &ID) const { 1762 static int id = 0; 1763 ID.AddPointer(&id); 1764 ID.Add(V); 1765 } 1766 1767 const char *SuppressInlineDefensiveChecksVisitor::getTag() { 1768 return "IDCVisitor"; 1769 } 1770 1771 PathDiagnosticPieceRef 1772 SuppressInlineDefensiveChecksVisitor::VisitNode(const ExplodedNode *Succ, 1773 BugReporterContext &BRC, 1774 PathSensitiveBugReport &BR) { 1775 const ExplodedNode *Pred = Succ->getFirstPred(); 1776 if (IsSatisfied) 1777 return nullptr; 1778 1779 // Start tracking after we see the first state in which the value is null. 1780 if (!IsTrackingTurnedOn) 1781 if (Succ->getState()->isNull(V).isConstrainedTrue()) 1782 IsTrackingTurnedOn = true; 1783 if (!IsTrackingTurnedOn) 1784 return nullptr; 1785 1786 // Check if in the previous state it was feasible for this value 1787 // to *not* be null. 1788 if (!Pred->getState()->isNull(V).isConstrainedTrue() && 1789 Succ->getState()->isNull(V).isConstrainedTrue()) { 1790 IsSatisfied = true; 1791 1792 // Check if this is inlined defensive checks. 1793 const LocationContext *CurLC = Succ->getLocationContext(); 1794 const LocationContext *ReportLC = BR.getErrorNode()->getLocationContext(); 1795 if (CurLC != ReportLC && !CurLC->isParentOf(ReportLC)) { 1796 BR.markInvalid("Suppress IDC", CurLC); 1797 return nullptr; 1798 } 1799 1800 // Treat defensive checks in function-like macros as if they were an inlined 1801 // defensive check. If the bug location is not in a macro and the 1802 // terminator for the current location is in a macro then suppress the 1803 // warning. 1804 auto BugPoint = BR.getErrorNode()->getLocation().getAs<StmtPoint>(); 1805 1806 if (!BugPoint) 1807 return nullptr; 1808 1809 ProgramPoint CurPoint = Succ->getLocation(); 1810 const Stmt *CurTerminatorStmt = nullptr; 1811 if (auto BE = CurPoint.getAs<BlockEdge>()) { 1812 CurTerminatorStmt = BE->getSrc()->getTerminator().getStmt(); 1813 } else if (auto SP = CurPoint.getAs<StmtPoint>()) { 1814 const Stmt *CurStmt = SP->getStmt(); 1815 if (!CurStmt->getBeginLoc().isMacroID()) 1816 return nullptr; 1817 1818 CFGStmtMap *Map = CurLC->getAnalysisDeclContext()->getCFGStmtMap(); 1819 CurTerminatorStmt = Map->getBlock(CurStmt)->getTerminatorStmt(); 1820 } else { 1821 return nullptr; 1822 } 1823 1824 if (!CurTerminatorStmt) 1825 return nullptr; 1826 1827 SourceLocation TerminatorLoc = CurTerminatorStmt->getBeginLoc(); 1828 if (TerminatorLoc.isMacroID()) { 1829 SourceLocation BugLoc = BugPoint->getStmt()->getBeginLoc(); 1830 1831 // Suppress reports unless we are in that same macro. 1832 if (!BugLoc.isMacroID() || 1833 getMacroName(BugLoc, BRC) != getMacroName(TerminatorLoc, BRC)) { 1834 BR.markInvalid("Suppress Macro IDC", CurLC); 1835 } 1836 return nullptr; 1837 } 1838 } 1839 return nullptr; 1840 } 1841 1842 //===----------------------------------------------------------------------===// 1843 // TrackControlDependencyCondBRVisitor. 1844 //===----------------------------------------------------------------------===// 1845 1846 namespace { 1847 /// Tracks the expressions that are a control dependency of the node that was 1848 /// supplied to the constructor. 1849 /// For example: 1850 /// 1851 /// cond = 1; 1852 /// if (cond) 1853 /// 10 / 0; 1854 /// 1855 /// An error is emitted at line 3. This visitor realizes that the branch 1856 /// on line 2 is a control dependency of line 3, and tracks it's condition via 1857 /// trackExpressionValue(). 1858 class TrackControlDependencyCondBRVisitor final 1859 : public TrackingBugReporterVisitor { 1860 const ExplodedNode *Origin; 1861 ControlDependencyCalculator ControlDeps; 1862 llvm::SmallSet<const CFGBlock *, 32> VisitedBlocks; 1863 1864 public: 1865 TrackControlDependencyCondBRVisitor(TrackerRef ParentTracker, 1866 const ExplodedNode *O) 1867 : TrackingBugReporterVisitor(ParentTracker), Origin(O), 1868 ControlDeps(&O->getCFG()) {} 1869 1870 void Profile(llvm::FoldingSetNodeID &ID) const override { 1871 static int x = 0; 1872 ID.AddPointer(&x); 1873 } 1874 1875 PathDiagnosticPieceRef VisitNode(const ExplodedNode *N, 1876 BugReporterContext &BRC, 1877 PathSensitiveBugReport &BR) override; 1878 }; 1879 } // end of anonymous namespace 1880 1881 static std::shared_ptr<PathDiagnosticEventPiece> 1882 constructDebugPieceForTrackedCondition(const Expr *Cond, 1883 const ExplodedNode *N, 1884 BugReporterContext &BRC) { 1885 1886 if (BRC.getAnalyzerOptions().AnalysisDiagOpt == PD_NONE || 1887 !BRC.getAnalyzerOptions().ShouldTrackConditionsDebug) 1888 return nullptr; 1889 1890 std::string ConditionText = std::string(Lexer::getSourceText( 1891 CharSourceRange::getTokenRange(Cond->getSourceRange()), 1892 BRC.getSourceManager(), BRC.getASTContext().getLangOpts())); 1893 1894 return std::make_shared<PathDiagnosticEventPiece>( 1895 PathDiagnosticLocation::createBegin( 1896 Cond, BRC.getSourceManager(), N->getLocationContext()), 1897 (Twine() + "Tracking condition '" + ConditionText + "'").str()); 1898 } 1899 1900 static bool isAssertlikeBlock(const CFGBlock *B, ASTContext &Context) { 1901 if (B->succ_size() != 2) 1902 return false; 1903 1904 const CFGBlock *Then = B->succ_begin()->getReachableBlock(); 1905 const CFGBlock *Else = (B->succ_begin() + 1)->getReachableBlock(); 1906 1907 if (!Then || !Else) 1908 return false; 1909 1910 if (Then->isInevitablySinking() != Else->isInevitablySinking()) 1911 return true; 1912 1913 // For the following condition the following CFG would be built: 1914 // 1915 // -------------> 1916 // / \ 1917 // [B1] -> [B2] -> [B3] -> [sink] 1918 // assert(A && B || C); \ \ 1919 // -----------> [go on with the execution] 1920 // 1921 // It so happens that CFGBlock::getTerminatorCondition returns 'A' for block 1922 // B1, 'A && B' for B2, and 'A && B || C' for B3. Let's check whether we 1923 // reached the end of the condition! 1924 if (const Stmt *ElseCond = Else->getTerminatorCondition()) 1925 if (const auto *BinOp = dyn_cast<BinaryOperator>(ElseCond)) 1926 if (BinOp->isLogicalOp()) 1927 return isAssertlikeBlock(Else, Context); 1928 1929 return false; 1930 } 1931 1932 PathDiagnosticPieceRef 1933 TrackControlDependencyCondBRVisitor::VisitNode(const ExplodedNode *N, 1934 BugReporterContext &BRC, 1935 PathSensitiveBugReport &BR) { 1936 // We can only reason about control dependencies within the same stack frame. 1937 if (Origin->getStackFrame() != N->getStackFrame()) 1938 return nullptr; 1939 1940 CFGBlock *NB = const_cast<CFGBlock *>(N->getCFGBlock()); 1941 1942 // Skip if we already inspected this block. 1943 if (!VisitedBlocks.insert(NB).second) 1944 return nullptr; 1945 1946 CFGBlock *OriginB = const_cast<CFGBlock *>(Origin->getCFGBlock()); 1947 1948 // TODO: Cache CFGBlocks for each ExplodedNode. 1949 if (!OriginB || !NB) 1950 return nullptr; 1951 1952 if (isAssertlikeBlock(NB, BRC.getASTContext())) 1953 return nullptr; 1954 1955 if (ControlDeps.isControlDependent(OriginB, NB)) { 1956 // We don't really want to explain for range loops. Evidence suggests that 1957 // the only thing that leads to is the addition of calls to operator!=. 1958 if (llvm::isa_and_nonnull<CXXForRangeStmt>(NB->getTerminatorStmt())) 1959 return nullptr; 1960 1961 if (const Expr *Condition = NB->getLastCondition()) { 1962 // Keeping track of the already tracked conditions on a visitor level 1963 // isn't sufficient, because a new visitor is created for each tracked 1964 // expression, hence the BugReport level set. 1965 if (BR.addTrackedCondition(N)) { 1966 getParentTracker().track(Condition, N, 1967 {bugreporter::TrackingKind::Condition, 1968 /*EnableNullFPSuppression=*/false}); 1969 return constructDebugPieceForTrackedCondition(Condition, N, BRC); 1970 } 1971 } 1972 } 1973 1974 return nullptr; 1975 } 1976 1977 //===----------------------------------------------------------------------===// 1978 // Implementation of trackExpressionValue. 1979 //===----------------------------------------------------------------------===// 1980 1981 /// \return A subexpression of @c Ex which represents the 1982 /// expression-of-interest. 1983 static const Expr *peelOffOuterExpr(const Expr *Ex, 1984 const ExplodedNode *N) { 1985 Ex = Ex->IgnoreParenCasts(); 1986 if (const auto *FE = dyn_cast<FullExpr>(Ex)) 1987 return peelOffOuterExpr(FE->getSubExpr(), N); 1988 if (const auto *OVE = dyn_cast<OpaqueValueExpr>(Ex)) 1989 return peelOffOuterExpr(OVE->getSourceExpr(), N); 1990 if (const auto *POE = dyn_cast<PseudoObjectExpr>(Ex)) { 1991 const auto *PropRef = dyn_cast<ObjCPropertyRefExpr>(POE->getSyntacticForm()); 1992 if (PropRef && PropRef->isMessagingGetter()) { 1993 const Expr *GetterMessageSend = 1994 POE->getSemanticExpr(POE->getNumSemanticExprs() - 1); 1995 assert(isa<ObjCMessageExpr>(GetterMessageSend->IgnoreParenCasts())); 1996 return peelOffOuterExpr(GetterMessageSend, N); 1997 } 1998 } 1999 2000 // Peel off the ternary operator. 2001 if (const auto *CO = dyn_cast<ConditionalOperator>(Ex)) { 2002 // Find a node where the branching occurred and find out which branch 2003 // we took (true/false) by looking at the ExplodedGraph. 2004 const ExplodedNode *NI = N; 2005 do { 2006 ProgramPoint ProgPoint = NI->getLocation(); 2007 if (Optional<BlockEdge> BE = ProgPoint.getAs<BlockEdge>()) { 2008 const CFGBlock *srcBlk = BE->getSrc(); 2009 if (const Stmt *term = srcBlk->getTerminatorStmt()) { 2010 if (term == CO) { 2011 bool TookTrueBranch = (*(srcBlk->succ_begin()) == BE->getDst()); 2012 if (TookTrueBranch) 2013 return peelOffOuterExpr(CO->getTrueExpr(), N); 2014 else 2015 return peelOffOuterExpr(CO->getFalseExpr(), N); 2016 } 2017 } 2018 } 2019 NI = NI->getFirstPred(); 2020 } while (NI); 2021 } 2022 2023 if (auto *BO = dyn_cast<BinaryOperator>(Ex)) 2024 if (const Expr *SubEx = peelOffPointerArithmetic(BO)) 2025 return peelOffOuterExpr(SubEx, N); 2026 2027 if (auto *UO = dyn_cast<UnaryOperator>(Ex)) { 2028 if (UO->getOpcode() == UO_LNot) 2029 return peelOffOuterExpr(UO->getSubExpr(), N); 2030 2031 // FIXME: There's a hack in our Store implementation that always computes 2032 // field offsets around null pointers as if they are always equal to 0. 2033 // The idea here is to report accesses to fields as null dereferences 2034 // even though the pointer value that's being dereferenced is actually 2035 // the offset of the field rather than exactly 0. 2036 // See the FIXME in StoreManager's getLValueFieldOrIvar() method. 2037 // This code interacts heavily with this hack; otherwise the value 2038 // would not be null at all for most fields, so we'd be unable to track it. 2039 if (UO->getOpcode() == UO_AddrOf && UO->getSubExpr()->isLValue()) 2040 if (const Expr *DerefEx = bugreporter::getDerefExpr(UO->getSubExpr())) 2041 return peelOffOuterExpr(DerefEx, N); 2042 } 2043 2044 return Ex; 2045 } 2046 2047 /// Find the ExplodedNode where the lvalue (the value of 'Ex') 2048 /// was computed. 2049 static const ExplodedNode* findNodeForExpression(const ExplodedNode *N, 2050 const Expr *Inner) { 2051 while (N) { 2052 if (N->getStmtForDiagnostics() == Inner) 2053 return N; 2054 N = N->getFirstPred(); 2055 } 2056 return N; 2057 } 2058 2059 //===----------------------------------------------------------------------===// 2060 // Tracker implementation 2061 //===----------------------------------------------------------------------===// 2062 2063 class DefaultExpressionHandler final : public ExpressionHandler { 2064 public: 2065 using ExpressionHandler::ExpressionHandler; 2066 2067 Tracker::Result handle(const Expr *Inner, const ExplodedNode *InputNode, 2068 const ExplodedNode *LVNode, 2069 TrackingOptions Opts) override { 2070 ProgramStateRef LVState = LVNode->getState(); 2071 const StackFrameContext *SFC = LVNode->getStackFrame(); 2072 PathSensitiveBugReport &Report = getParentTracker().getReport(); 2073 Tracker::Result Result; 2074 2075 // We only track expressions if we believe that they are important. Chances 2076 // are good that control dependencies to the tracking point are also 2077 // important because of this, let's explain why we believe control reached 2078 // this point. 2079 // TODO: Shouldn't we track control dependencies of every bug location, 2080 // rather than only tracked expressions? 2081 if (LVState->getAnalysisManager() 2082 .getAnalyzerOptions() 2083 .ShouldTrackConditions) { 2084 Report.addVisitor<TrackControlDependencyCondBRVisitor>( 2085 &getParentTracker(), InputNode); 2086 Result.FoundSomethingToTrack = true; 2087 } 2088 2089 // The message send could be nil due to the receiver being nil. 2090 // At this point in the path, the receiver should be live since we are at 2091 // the message send expr. If it is nil, start tracking it. 2092 if (const Expr *Receiver = 2093 NilReceiverBRVisitor::getNilReceiver(Inner, LVNode)) 2094 Result.combineWith(getParentTracker().track(Receiver, LVNode, Opts)); 2095 2096 // Track the index if this is an array subscript. 2097 if (const auto *Arr = dyn_cast<ArraySubscriptExpr>(Inner)) 2098 Result.combineWith(getParentTracker().track( 2099 Arr->getIdx(), LVNode, 2100 {Opts.Kind, /*EnableNullFPSuppression*/ false})); 2101 2102 // See if the expression we're interested refers to a variable. 2103 // If so, we can track both its contents and constraints on its value. 2104 if (ExplodedGraph::isInterestingLValueExpr(Inner)) { 2105 SVal LVal = LVNode->getSVal(Inner); 2106 2107 const MemRegion *RR = getLocationRegionIfReference(Inner, LVNode); 2108 bool LVIsNull = LVState->isNull(LVal).isConstrainedTrue(); 2109 2110 // If this is a C++ reference to a null pointer, we are tracking the 2111 // pointer. In addition, we should find the store at which the reference 2112 // got initialized. 2113 if (RR && !LVIsNull) 2114 Result.combineWith(getParentTracker().track(LVal, RR, Opts, SFC)); 2115 2116 // In case of C++ references, we want to differentiate between a null 2117 // reference and reference to null pointer. 2118 // If the LVal is null, check if we are dealing with null reference. 2119 // For those, we want to track the location of the reference. 2120 const MemRegion *R = 2121 (RR && LVIsNull) ? RR : LVNode->getSVal(Inner).getAsRegion(); 2122 2123 if (R) { 2124 2125 // Mark both the variable region and its contents as interesting. 2126 SVal V = LVState->getRawSVal(loc::MemRegionVal(R)); 2127 Report.addVisitor<NoStoreFuncVisitor>(cast<SubRegion>(R), Opts.Kind); 2128 2129 // When we got here, we do have something to track, and we will 2130 // interrupt. 2131 Result.FoundSomethingToTrack = true; 2132 Result.WasInterrupted = true; 2133 2134 MacroNullReturnSuppressionVisitor::addMacroVisitorIfNecessary( 2135 LVNode, R, Opts.EnableNullFPSuppression, Report, V); 2136 2137 Report.markInteresting(V, Opts.Kind); 2138 Report.addVisitor<UndefOrNullArgVisitor>(R); 2139 2140 // If the contents are symbolic and null, find out when they became 2141 // null. 2142 if (V.getAsLocSymbol(/*IncludeBaseRegions=*/true)) 2143 if (LVState->isNull(V).isConstrainedTrue()) 2144 Report.addVisitor<TrackConstraintBRVisitor>(V.castAs<DefinedSVal>(), 2145 false); 2146 2147 // Add visitor, which will suppress inline defensive checks. 2148 if (auto DV = V.getAs<DefinedSVal>()) 2149 if (!DV->isZeroConstant() && Opts.EnableNullFPSuppression) 2150 // Note that LVNode may be too late (i.e., too far from the 2151 // InputNode) because the lvalue may have been computed before the 2152 // inlined call was evaluated. InputNode may as well be too early 2153 // here, because the symbol is already dead; this, however, is fine 2154 // because we can still find the node in which it collapsed to null 2155 // previously. 2156 Report.addVisitor<SuppressInlineDefensiveChecksVisitor>(*DV, 2157 InputNode); 2158 2159 getParentTracker().track(V, R, Opts, SFC); 2160 2161 return Result; 2162 } 2163 } 2164 2165 // If the expression is not an "lvalue expression", we can still 2166 // track the constraints on its contents. 2167 SVal V = LVState->getSValAsScalarOrLoc(Inner, LVNode->getLocationContext()); 2168 2169 ReturnVisitor::addVisitorIfNecessary(&getParentTracker(), LVNode, Inner, 2170 Report, Opts.EnableNullFPSuppression, 2171 Opts.Kind); 2172 2173 // Is it a symbolic value? 2174 if (auto L = V.getAs<loc::MemRegionVal>()) { 2175 // FIXME: this is a hack for fixing a later crash when attempting to 2176 // dereference a void* pointer. 2177 // We should not try to dereference pointers at all when we don't care 2178 // what is written inside the pointer. 2179 bool CanDereference = true; 2180 if (const auto *SR = L->getRegionAs<SymbolicRegion>()) { 2181 if (SR->getSymbol()->getType()->getPointeeType()->isVoidType()) 2182 CanDereference = false; 2183 } else if (L->getRegionAs<AllocaRegion>()) 2184 CanDereference = false; 2185 2186 // At this point we are dealing with the region's LValue. 2187 // However, if the rvalue is a symbolic region, we should track it as 2188 // well. Try to use the correct type when looking up the value. 2189 SVal RVal; 2190 if (ExplodedGraph::isInterestingLValueExpr(Inner)) 2191 RVal = LVState->getRawSVal(L.getValue(), Inner->getType()); 2192 else if (CanDereference) 2193 RVal = LVState->getSVal(L->getRegion()); 2194 2195 if (CanDereference) { 2196 Report.addVisitor<UndefOrNullArgVisitor>(L->getRegion()); 2197 Result.FoundSomethingToTrack = true; 2198 2199 if (auto KV = RVal.getAs<KnownSVal>()) 2200 Result.combineWith( 2201 getParentTracker().track(*KV, L->getRegion(), Opts, SFC)); 2202 } 2203 2204 const MemRegion *RegionRVal = RVal.getAsRegion(); 2205 if (isa_and_nonnull<SymbolicRegion>(RegionRVal)) { 2206 Report.markInteresting(RegionRVal, Opts.Kind); 2207 Report.addVisitor<TrackConstraintBRVisitor>( 2208 loc::MemRegionVal(RegionRVal), 2209 /*assumption=*/false); 2210 Result.FoundSomethingToTrack = true; 2211 } 2212 } 2213 2214 return Result; 2215 } 2216 }; 2217 2218 /// Attempts to add visitors to track an RValue expression back to its point of 2219 /// origin. 2220 class PRValueHandler final : public ExpressionHandler { 2221 public: 2222 using ExpressionHandler::ExpressionHandler; 2223 2224 Tracker::Result handle(const Expr *E, const ExplodedNode *InputNode, 2225 const ExplodedNode *ExprNode, 2226 TrackingOptions Opts) override { 2227 if (!E->isPRValue()) 2228 return {}; 2229 2230 const ExplodedNode *RVNode = findNodeForExpression(ExprNode, E); 2231 if (!RVNode) 2232 return {}; 2233 2234 ProgramStateRef RVState = RVNode->getState(); 2235 SVal V = RVState->getSValAsScalarOrLoc(E, RVNode->getLocationContext()); 2236 const auto *BO = dyn_cast<BinaryOperator>(E); 2237 2238 if (!BO || !BO->isMultiplicativeOp() || !V.isZeroConstant()) 2239 return {}; 2240 2241 SVal RHSV = RVState->getSVal(BO->getRHS(), RVNode->getLocationContext()); 2242 SVal LHSV = RVState->getSVal(BO->getLHS(), RVNode->getLocationContext()); 2243 2244 // Track both LHS and RHS of a multiplication. 2245 Tracker::Result CombinedResult; 2246 Tracker &Parent = getParentTracker(); 2247 2248 const auto track = [&CombinedResult, &Parent, ExprNode, Opts](Expr *Inner) { 2249 CombinedResult.combineWith(Parent.track(Inner, ExprNode, Opts)); 2250 }; 2251 2252 if (BO->getOpcode() == BO_Mul) { 2253 if (LHSV.isZeroConstant()) 2254 track(BO->getLHS()); 2255 if (RHSV.isZeroConstant()) 2256 track(BO->getRHS()); 2257 } else { // Track only the LHS of a division or a modulo. 2258 if (LHSV.isZeroConstant()) 2259 track(BO->getLHS()); 2260 } 2261 2262 return CombinedResult; 2263 } 2264 }; 2265 2266 Tracker::Tracker(PathSensitiveBugReport &Report) : Report(Report) { 2267 addHighPriorityHandler<DefaultExpressionHandler>(); 2268 addLowPriorityHandler<PRValueHandler>(); 2269 // TODO: split trackExpressionValue and FindLastStoreBRVisitor into handlers 2270 // and add them here. 2271 } 2272 2273 Tracker::Result Tracker::track(const Expr *E, const ExplodedNode *N, 2274 TrackingOptions Opts) { 2275 if (!E || !N) 2276 return {}; 2277 2278 const Expr *Inner = peelOffOuterExpr(E, N); 2279 const ExplodedNode *LVNode = findNodeForExpression(N, Inner); 2280 if (!LVNode) 2281 return {}; 2282 2283 Result CombinedResult; 2284 // Iterate through the handlers in the order according to their priorities. 2285 for (ExpressionHandlerPtr &Handler : ExpressionHandlers) { 2286 CombinedResult.combineWith(Handler->handle(Inner, N, LVNode, Opts)); 2287 if (CombinedResult.WasInterrupted) 2288 break; 2289 } 2290 2291 return CombinedResult; 2292 } 2293 2294 Tracker::Result Tracker::track(SVal V, const MemRegion *R, TrackingOptions Opts, 2295 const StackFrameContext *Origin) { 2296 if (auto KV = V.getAs<KnownSVal>()) { 2297 Report.addVisitor<StoreSiteFinder>( 2298 this, *KV, R, Opts.EnableNullFPSuppression, Opts.Kind, Origin); 2299 return {true}; 2300 } 2301 return {}; 2302 } 2303 2304 PathDiagnosticPieceRef Tracker::handle(StoreInfo SI, TrackingOptions Opts) { 2305 // Iterate through the handlers in the order according to their priorities. 2306 for (StoreHandlerPtr &Handler : StoreHandlers) { 2307 if (PathDiagnosticPieceRef Result = Handler->handle(SI, Opts)) 2308 // If the handler produced a non-null piece, return it. 2309 // There is no need in asking other handlers. 2310 return Result; 2311 } 2312 return {}; 2313 } 2314 2315 bool bugreporter::trackExpressionValue(const ExplodedNode *InputNode, 2316 const Expr *E, 2317 2318 PathSensitiveBugReport &Report, 2319 TrackingOptions Opts) { 2320 return Tracker::create(Report) 2321 ->track(E, InputNode, Opts) 2322 .FoundSomethingToTrack; 2323 } 2324 2325 void bugreporter::trackStoredValue(KnownSVal V, const MemRegion *R, 2326 PathSensitiveBugReport &Report, 2327 TrackingOptions Opts, 2328 const StackFrameContext *Origin) { 2329 Tracker::create(Report)->track(V, R, Opts, Origin); 2330 } 2331 2332 //===----------------------------------------------------------------------===// 2333 // Implementation of NulReceiverBRVisitor. 2334 //===----------------------------------------------------------------------===// 2335 2336 const Expr *NilReceiverBRVisitor::getNilReceiver(const Stmt *S, 2337 const ExplodedNode *N) { 2338 const auto *ME = dyn_cast<ObjCMessageExpr>(S); 2339 if (!ME) 2340 return nullptr; 2341 if (const Expr *Receiver = ME->getInstanceReceiver()) { 2342 ProgramStateRef state = N->getState(); 2343 SVal V = N->getSVal(Receiver); 2344 if (state->isNull(V).isConstrainedTrue()) 2345 return Receiver; 2346 } 2347 return nullptr; 2348 } 2349 2350 PathDiagnosticPieceRef 2351 NilReceiverBRVisitor::VisitNode(const ExplodedNode *N, BugReporterContext &BRC, 2352 PathSensitiveBugReport &BR) { 2353 Optional<PreStmt> P = N->getLocationAs<PreStmt>(); 2354 if (!P) 2355 return nullptr; 2356 2357 const Stmt *S = P->getStmt(); 2358 const Expr *Receiver = getNilReceiver(S, N); 2359 if (!Receiver) 2360 return nullptr; 2361 2362 llvm::SmallString<256> Buf; 2363 llvm::raw_svector_ostream OS(Buf); 2364 2365 if (const auto *ME = dyn_cast<ObjCMessageExpr>(S)) { 2366 OS << "'"; 2367 ME->getSelector().print(OS); 2368 OS << "' not called"; 2369 } 2370 else { 2371 OS << "No method is called"; 2372 } 2373 OS << " because the receiver is nil"; 2374 2375 // The receiver was nil, and hence the method was skipped. 2376 // Register a BugReporterVisitor to issue a message telling us how 2377 // the receiver was null. 2378 bugreporter::trackExpressionValue(N, Receiver, BR, 2379 {bugreporter::TrackingKind::Thorough, 2380 /*EnableNullFPSuppression*/ false}); 2381 // Issue a message saying that the method was skipped. 2382 PathDiagnosticLocation L(Receiver, BRC.getSourceManager(), 2383 N->getLocationContext()); 2384 return std::make_shared<PathDiagnosticEventPiece>(L, OS.str()); 2385 } 2386 2387 //===----------------------------------------------------------------------===// 2388 // Visitor that tries to report interesting diagnostics from conditions. 2389 //===----------------------------------------------------------------------===// 2390 2391 /// Return the tag associated with this visitor. This tag will be used 2392 /// to make all PathDiagnosticPieces created by this visitor. 2393 const char *ConditionBRVisitor::getTag() { return "ConditionBRVisitor"; } 2394 2395 PathDiagnosticPieceRef 2396 ConditionBRVisitor::VisitNode(const ExplodedNode *N, BugReporterContext &BRC, 2397 PathSensitiveBugReport &BR) { 2398 auto piece = VisitNodeImpl(N, BRC, BR); 2399 if (piece) { 2400 piece->setTag(getTag()); 2401 if (auto *ev = dyn_cast<PathDiagnosticEventPiece>(piece.get())) 2402 ev->setPrunable(true, /* override */ false); 2403 } 2404 return piece; 2405 } 2406 2407 PathDiagnosticPieceRef 2408 ConditionBRVisitor::VisitNodeImpl(const ExplodedNode *N, 2409 BugReporterContext &BRC, 2410 PathSensitiveBugReport &BR) { 2411 ProgramPoint ProgPoint = N->getLocation(); 2412 const std::pair<const ProgramPointTag *, const ProgramPointTag *> &Tags = 2413 ExprEngine::geteagerlyAssumeBinOpBifurcationTags(); 2414 2415 // If an assumption was made on a branch, it should be caught 2416 // here by looking at the state transition. 2417 if (Optional<BlockEdge> BE = ProgPoint.getAs<BlockEdge>()) { 2418 const CFGBlock *SrcBlock = BE->getSrc(); 2419 if (const Stmt *Term = SrcBlock->getTerminatorStmt()) { 2420 // If the tag of the previous node is 'Eagerly Assume...' the current 2421 // 'BlockEdge' has the same constraint information. We do not want to 2422 // report the value as it is just an assumption on the predecessor node 2423 // which will be caught in the next VisitNode() iteration as a 'PostStmt'. 2424 const ProgramPointTag *PreviousNodeTag = 2425 N->getFirstPred()->getLocation().getTag(); 2426 if (PreviousNodeTag == Tags.first || PreviousNodeTag == Tags.second) 2427 return nullptr; 2428 2429 return VisitTerminator(Term, N, SrcBlock, BE->getDst(), BR, BRC); 2430 } 2431 return nullptr; 2432 } 2433 2434 if (Optional<PostStmt> PS = ProgPoint.getAs<PostStmt>()) { 2435 const ProgramPointTag *CurrentNodeTag = PS->getTag(); 2436 if (CurrentNodeTag != Tags.first && CurrentNodeTag != Tags.second) 2437 return nullptr; 2438 2439 bool TookTrue = CurrentNodeTag == Tags.first; 2440 return VisitTrueTest(cast<Expr>(PS->getStmt()), BRC, BR, N, TookTrue); 2441 } 2442 2443 return nullptr; 2444 } 2445 2446 PathDiagnosticPieceRef ConditionBRVisitor::VisitTerminator( 2447 const Stmt *Term, const ExplodedNode *N, const CFGBlock *srcBlk, 2448 const CFGBlock *dstBlk, PathSensitiveBugReport &R, 2449 BugReporterContext &BRC) { 2450 const Expr *Cond = nullptr; 2451 2452 // In the code below, Term is a CFG terminator and Cond is a branch condition 2453 // expression upon which the decision is made on this terminator. 2454 // 2455 // For example, in "if (x == 0)", the "if (x == 0)" statement is a terminator, 2456 // and "x == 0" is the respective condition. 2457 // 2458 // Another example: in "if (x && y)", we've got two terminators and two 2459 // conditions due to short-circuit nature of operator "&&": 2460 // 1. The "if (x && y)" statement is a terminator, 2461 // and "y" is the respective condition. 2462 // 2. Also "x && ..." is another terminator, 2463 // and "x" is its condition. 2464 2465 switch (Term->getStmtClass()) { 2466 // FIXME: Stmt::SwitchStmtClass is worth handling, however it is a bit 2467 // more tricky because there are more than two branches to account for. 2468 default: 2469 return nullptr; 2470 case Stmt::IfStmtClass: 2471 Cond = cast<IfStmt>(Term)->getCond(); 2472 break; 2473 case Stmt::ConditionalOperatorClass: 2474 Cond = cast<ConditionalOperator>(Term)->getCond(); 2475 break; 2476 case Stmt::BinaryOperatorClass: 2477 // When we encounter a logical operator (&& or ||) as a CFG terminator, 2478 // then the condition is actually its LHS; otherwise, we'd encounter 2479 // the parent, such as if-statement, as a terminator. 2480 const auto *BO = cast<BinaryOperator>(Term); 2481 assert(BO->isLogicalOp() && 2482 "CFG terminator is not a short-circuit operator!"); 2483 Cond = BO->getLHS(); 2484 break; 2485 } 2486 2487 Cond = Cond->IgnoreParens(); 2488 2489 // However, when we encounter a logical operator as a branch condition, 2490 // then the condition is actually its RHS, because LHS would be 2491 // the condition for the logical operator terminator. 2492 while (const auto *InnerBO = dyn_cast<BinaryOperator>(Cond)) { 2493 if (!InnerBO->isLogicalOp()) 2494 break; 2495 Cond = InnerBO->getRHS()->IgnoreParens(); 2496 } 2497 2498 assert(Cond); 2499 assert(srcBlk->succ_size() == 2); 2500 const bool TookTrue = *(srcBlk->succ_begin()) == dstBlk; 2501 return VisitTrueTest(Cond, BRC, R, N, TookTrue); 2502 } 2503 2504 PathDiagnosticPieceRef 2505 ConditionBRVisitor::VisitTrueTest(const Expr *Cond, BugReporterContext &BRC, 2506 PathSensitiveBugReport &R, 2507 const ExplodedNode *N, bool TookTrue) { 2508 ProgramStateRef CurrentState = N->getState(); 2509 ProgramStateRef PrevState = N->getFirstPred()->getState(); 2510 const LocationContext *LCtx = N->getLocationContext(); 2511 2512 // If the constraint information is changed between the current and the 2513 // previous program state we assuming the newly seen constraint information. 2514 // If we cannot evaluate the condition (and the constraints are the same) 2515 // the analyzer has no information about the value and just assuming it. 2516 bool IsAssuming = 2517 !BRC.getStateManager().haveEqualConstraints(CurrentState, PrevState) || 2518 CurrentState->getSVal(Cond, LCtx).isUnknownOrUndef(); 2519 2520 // These will be modified in code below, but we need to preserve the original 2521 // values in case we want to throw the generic message. 2522 const Expr *CondTmp = Cond; 2523 bool TookTrueTmp = TookTrue; 2524 2525 while (true) { 2526 CondTmp = CondTmp->IgnoreParenCasts(); 2527 switch (CondTmp->getStmtClass()) { 2528 default: 2529 break; 2530 case Stmt::BinaryOperatorClass: 2531 if (auto P = VisitTrueTest(Cond, cast<BinaryOperator>(CondTmp), 2532 BRC, R, N, TookTrueTmp, IsAssuming)) 2533 return P; 2534 break; 2535 case Stmt::DeclRefExprClass: 2536 if (auto P = VisitTrueTest(Cond, cast<DeclRefExpr>(CondTmp), 2537 BRC, R, N, TookTrueTmp, IsAssuming)) 2538 return P; 2539 break; 2540 case Stmt::MemberExprClass: 2541 if (auto P = VisitTrueTest(Cond, cast<MemberExpr>(CondTmp), 2542 BRC, R, N, TookTrueTmp, IsAssuming)) 2543 return P; 2544 break; 2545 case Stmt::UnaryOperatorClass: { 2546 const auto *UO = cast<UnaryOperator>(CondTmp); 2547 if (UO->getOpcode() == UO_LNot) { 2548 TookTrueTmp = !TookTrueTmp; 2549 CondTmp = UO->getSubExpr(); 2550 continue; 2551 } 2552 break; 2553 } 2554 } 2555 break; 2556 } 2557 2558 // Condition too complex to explain? Just say something so that the user 2559 // knew we've made some path decision at this point. 2560 // If it is too complex and we know the evaluation of the condition do not 2561 // repeat the note from 'BugReporter.cpp' 2562 if (!IsAssuming) 2563 return nullptr; 2564 2565 PathDiagnosticLocation Loc(Cond, BRC.getSourceManager(), LCtx); 2566 if (!Loc.isValid() || !Loc.asLocation().isValid()) 2567 return nullptr; 2568 2569 return std::make_shared<PathDiagnosticEventPiece>( 2570 Loc, TookTrue ? GenericTrueMessage : GenericFalseMessage); 2571 } 2572 2573 bool ConditionBRVisitor::patternMatch(const Expr *Ex, 2574 const Expr *ParentEx, 2575 raw_ostream &Out, 2576 BugReporterContext &BRC, 2577 PathSensitiveBugReport &report, 2578 const ExplodedNode *N, 2579 Optional<bool> &prunable, 2580 bool IsSameFieldName) { 2581 const Expr *OriginalExpr = Ex; 2582 Ex = Ex->IgnoreParenCasts(); 2583 2584 if (isa<GNUNullExpr>(Ex) || isa<ObjCBoolLiteralExpr>(Ex) || 2585 isa<CXXBoolLiteralExpr>(Ex) || isa<IntegerLiteral>(Ex) || 2586 isa<FloatingLiteral>(Ex)) { 2587 // Use heuristics to determine if the expression is a macro 2588 // expanding to a literal and if so, use the macro's name. 2589 SourceLocation BeginLoc = OriginalExpr->getBeginLoc(); 2590 SourceLocation EndLoc = OriginalExpr->getEndLoc(); 2591 if (BeginLoc.isMacroID() && EndLoc.isMacroID()) { 2592 const SourceManager &SM = BRC.getSourceManager(); 2593 const LangOptions &LO = BRC.getASTContext().getLangOpts(); 2594 if (Lexer::isAtStartOfMacroExpansion(BeginLoc, SM, LO) && 2595 Lexer::isAtEndOfMacroExpansion(EndLoc, SM, LO)) { 2596 CharSourceRange R = Lexer::getAsCharRange({BeginLoc, EndLoc}, SM, LO); 2597 Out << Lexer::getSourceText(R, SM, LO); 2598 return false; 2599 } 2600 } 2601 } 2602 2603 if (const auto *DR = dyn_cast<DeclRefExpr>(Ex)) { 2604 const bool quotes = isa<VarDecl>(DR->getDecl()); 2605 if (quotes) { 2606 Out << '\''; 2607 const LocationContext *LCtx = N->getLocationContext(); 2608 const ProgramState *state = N->getState().get(); 2609 if (const MemRegion *R = state->getLValue(cast<VarDecl>(DR->getDecl()), 2610 LCtx).getAsRegion()) { 2611 if (report.isInteresting(R)) 2612 prunable = false; 2613 else { 2614 const ProgramState *state = N->getState().get(); 2615 SVal V = state->getSVal(R); 2616 if (report.isInteresting(V)) 2617 prunable = false; 2618 } 2619 } 2620 } 2621 Out << DR->getDecl()->getDeclName().getAsString(); 2622 if (quotes) 2623 Out << '\''; 2624 return quotes; 2625 } 2626 2627 if (const auto *IL = dyn_cast<IntegerLiteral>(Ex)) { 2628 QualType OriginalTy = OriginalExpr->getType(); 2629 if (OriginalTy->isPointerType()) { 2630 if (IL->getValue() == 0) { 2631 Out << "null"; 2632 return false; 2633 } 2634 } 2635 else if (OriginalTy->isObjCObjectPointerType()) { 2636 if (IL->getValue() == 0) { 2637 Out << "nil"; 2638 return false; 2639 } 2640 } 2641 2642 Out << IL->getValue(); 2643 return false; 2644 } 2645 2646 if (const auto *ME = dyn_cast<MemberExpr>(Ex)) { 2647 if (!IsSameFieldName) 2648 Out << "field '" << ME->getMemberDecl()->getName() << '\''; 2649 else 2650 Out << '\'' 2651 << Lexer::getSourceText( 2652 CharSourceRange::getTokenRange(Ex->getSourceRange()), 2653 BRC.getSourceManager(), BRC.getASTContext().getLangOpts(), 0) 2654 << '\''; 2655 } 2656 2657 return false; 2658 } 2659 2660 PathDiagnosticPieceRef ConditionBRVisitor::VisitTrueTest( 2661 const Expr *Cond, const BinaryOperator *BExpr, BugReporterContext &BRC, 2662 PathSensitiveBugReport &R, const ExplodedNode *N, bool TookTrue, 2663 bool IsAssuming) { 2664 bool shouldInvert = false; 2665 Optional<bool> shouldPrune; 2666 2667 // Check if the field name of the MemberExprs is ambiguous. Example: 2668 // " 'a.d' is equal to 'h.d' " in 'test/Analysis/null-deref-path-notes.cpp'. 2669 bool IsSameFieldName = false; 2670 const auto *LhsME = dyn_cast<MemberExpr>(BExpr->getLHS()->IgnoreParenCasts()); 2671 const auto *RhsME = dyn_cast<MemberExpr>(BExpr->getRHS()->IgnoreParenCasts()); 2672 2673 if (LhsME && RhsME) 2674 IsSameFieldName = 2675 LhsME->getMemberDecl()->getName() == RhsME->getMemberDecl()->getName(); 2676 2677 SmallString<128> LhsString, RhsString; 2678 { 2679 llvm::raw_svector_ostream OutLHS(LhsString), OutRHS(RhsString); 2680 const bool isVarLHS = patternMatch(BExpr->getLHS(), BExpr, OutLHS, BRC, R, 2681 N, shouldPrune, IsSameFieldName); 2682 const bool isVarRHS = patternMatch(BExpr->getRHS(), BExpr, OutRHS, BRC, R, 2683 N, shouldPrune, IsSameFieldName); 2684 2685 shouldInvert = !isVarLHS && isVarRHS; 2686 } 2687 2688 BinaryOperator::Opcode Op = BExpr->getOpcode(); 2689 2690 if (BinaryOperator::isAssignmentOp(Op)) { 2691 // For assignment operators, all that we care about is that the LHS 2692 // evaluates to "true" or "false". 2693 return VisitConditionVariable(LhsString, BExpr->getLHS(), BRC, R, N, 2694 TookTrue); 2695 } 2696 2697 // For non-assignment operations, we require that we can understand 2698 // both the LHS and RHS. 2699 if (LhsString.empty() || RhsString.empty() || 2700 !BinaryOperator::isComparisonOp(Op) || Op == BO_Cmp) 2701 return nullptr; 2702 2703 // Should we invert the strings if the LHS is not a variable name? 2704 SmallString<256> buf; 2705 llvm::raw_svector_ostream Out(buf); 2706 Out << (IsAssuming ? "Assuming " : "") 2707 << (shouldInvert ? RhsString : LhsString) << " is "; 2708 2709 // Do we need to invert the opcode? 2710 if (shouldInvert) 2711 switch (Op) { 2712 default: break; 2713 case BO_LT: Op = BO_GT; break; 2714 case BO_GT: Op = BO_LT; break; 2715 case BO_LE: Op = BO_GE; break; 2716 case BO_GE: Op = BO_LE; break; 2717 } 2718 2719 if (!TookTrue) 2720 switch (Op) { 2721 case BO_EQ: Op = BO_NE; break; 2722 case BO_NE: Op = BO_EQ; break; 2723 case BO_LT: Op = BO_GE; break; 2724 case BO_GT: Op = BO_LE; break; 2725 case BO_LE: Op = BO_GT; break; 2726 case BO_GE: Op = BO_LT; break; 2727 default: 2728 return nullptr; 2729 } 2730 2731 switch (Op) { 2732 case BO_EQ: 2733 Out << "equal to "; 2734 break; 2735 case BO_NE: 2736 Out << "not equal to "; 2737 break; 2738 default: 2739 Out << BinaryOperator::getOpcodeStr(Op) << ' '; 2740 break; 2741 } 2742 2743 Out << (shouldInvert ? LhsString : RhsString); 2744 const LocationContext *LCtx = N->getLocationContext(); 2745 const SourceManager &SM = BRC.getSourceManager(); 2746 2747 if (isVarAnInterestingCondition(BExpr->getLHS(), N, &R) || 2748 isVarAnInterestingCondition(BExpr->getRHS(), N, &R)) 2749 Out << WillBeUsedForACondition; 2750 2751 // Convert 'field ...' to 'Field ...' if it is a MemberExpr. 2752 std::string Message = std::string(Out.str()); 2753 Message[0] = toupper(Message[0]); 2754 2755 // If we know the value create a pop-up note to the value part of 'BExpr'. 2756 if (!IsAssuming) { 2757 PathDiagnosticLocation Loc; 2758 if (!shouldInvert) { 2759 if (LhsME && LhsME->getMemberLoc().isValid()) 2760 Loc = PathDiagnosticLocation(LhsME->getMemberLoc(), SM); 2761 else 2762 Loc = PathDiagnosticLocation(BExpr->getLHS(), SM, LCtx); 2763 } else { 2764 if (RhsME && RhsME->getMemberLoc().isValid()) 2765 Loc = PathDiagnosticLocation(RhsME->getMemberLoc(), SM); 2766 else 2767 Loc = PathDiagnosticLocation(BExpr->getRHS(), SM, LCtx); 2768 } 2769 2770 return std::make_shared<PathDiagnosticPopUpPiece>(Loc, Message); 2771 } 2772 2773 PathDiagnosticLocation Loc(Cond, SM, LCtx); 2774 auto event = std::make_shared<PathDiagnosticEventPiece>(Loc, Message); 2775 if (shouldPrune.hasValue()) 2776 event->setPrunable(shouldPrune.getValue()); 2777 return event; 2778 } 2779 2780 PathDiagnosticPieceRef ConditionBRVisitor::VisitConditionVariable( 2781 StringRef LhsString, const Expr *CondVarExpr, BugReporterContext &BRC, 2782 PathSensitiveBugReport &report, const ExplodedNode *N, bool TookTrue) { 2783 // FIXME: If there's already a constraint tracker for this variable, 2784 // we shouldn't emit anything here (c.f. the double note in 2785 // test/Analysis/inlining/path-notes.c) 2786 SmallString<256> buf; 2787 llvm::raw_svector_ostream Out(buf); 2788 Out << "Assuming " << LhsString << " is "; 2789 2790 if (!printValue(CondVarExpr, Out, N, TookTrue, /*IsAssuming=*/true)) 2791 return nullptr; 2792 2793 const LocationContext *LCtx = N->getLocationContext(); 2794 PathDiagnosticLocation Loc(CondVarExpr, BRC.getSourceManager(), LCtx); 2795 2796 if (isVarAnInterestingCondition(CondVarExpr, N, &report)) 2797 Out << WillBeUsedForACondition; 2798 2799 auto event = std::make_shared<PathDiagnosticEventPiece>(Loc, Out.str()); 2800 2801 if (isInterestingExpr(CondVarExpr, N, &report)) 2802 event->setPrunable(false); 2803 2804 return event; 2805 } 2806 2807 PathDiagnosticPieceRef ConditionBRVisitor::VisitTrueTest( 2808 const Expr *Cond, const DeclRefExpr *DRE, BugReporterContext &BRC, 2809 PathSensitiveBugReport &report, const ExplodedNode *N, bool TookTrue, 2810 bool IsAssuming) { 2811 const auto *VD = dyn_cast<VarDecl>(DRE->getDecl()); 2812 if (!VD) 2813 return nullptr; 2814 2815 SmallString<256> Buf; 2816 llvm::raw_svector_ostream Out(Buf); 2817 2818 Out << (IsAssuming ? "Assuming '" : "'") << VD->getDeclName() << "' is "; 2819 2820 if (!printValue(DRE, Out, N, TookTrue, IsAssuming)) 2821 return nullptr; 2822 2823 const LocationContext *LCtx = N->getLocationContext(); 2824 2825 if (isVarAnInterestingCondition(DRE, N, &report)) 2826 Out << WillBeUsedForACondition; 2827 2828 // If we know the value create a pop-up note to the 'DRE'. 2829 if (!IsAssuming) { 2830 PathDiagnosticLocation Loc(DRE, BRC.getSourceManager(), LCtx); 2831 return std::make_shared<PathDiagnosticPopUpPiece>(Loc, Out.str()); 2832 } 2833 2834 PathDiagnosticLocation Loc(Cond, BRC.getSourceManager(), LCtx); 2835 auto event = std::make_shared<PathDiagnosticEventPiece>(Loc, Out.str()); 2836 2837 if (isInterestingExpr(DRE, N, &report)) 2838 event->setPrunable(false); 2839 2840 return std::move(event); 2841 } 2842 2843 PathDiagnosticPieceRef ConditionBRVisitor::VisitTrueTest( 2844 const Expr *Cond, const MemberExpr *ME, BugReporterContext &BRC, 2845 PathSensitiveBugReport &report, const ExplodedNode *N, bool TookTrue, 2846 bool IsAssuming) { 2847 SmallString<256> Buf; 2848 llvm::raw_svector_ostream Out(Buf); 2849 2850 Out << (IsAssuming ? "Assuming field '" : "Field '") 2851 << ME->getMemberDecl()->getName() << "' is "; 2852 2853 if (!printValue(ME, Out, N, TookTrue, IsAssuming)) 2854 return nullptr; 2855 2856 const LocationContext *LCtx = N->getLocationContext(); 2857 PathDiagnosticLocation Loc; 2858 2859 // If we know the value create a pop-up note to the member of the MemberExpr. 2860 if (!IsAssuming && ME->getMemberLoc().isValid()) 2861 Loc = PathDiagnosticLocation(ME->getMemberLoc(), BRC.getSourceManager()); 2862 else 2863 Loc = PathDiagnosticLocation(Cond, BRC.getSourceManager(), LCtx); 2864 2865 if (!Loc.isValid() || !Loc.asLocation().isValid()) 2866 return nullptr; 2867 2868 if (isVarAnInterestingCondition(ME, N, &report)) 2869 Out << WillBeUsedForACondition; 2870 2871 // If we know the value create a pop-up note. 2872 if (!IsAssuming) 2873 return std::make_shared<PathDiagnosticPopUpPiece>(Loc, Out.str()); 2874 2875 auto event = std::make_shared<PathDiagnosticEventPiece>(Loc, Out.str()); 2876 if (isInterestingExpr(ME, N, &report)) 2877 event->setPrunable(false); 2878 return event; 2879 } 2880 2881 bool ConditionBRVisitor::printValue(const Expr *CondVarExpr, raw_ostream &Out, 2882 const ExplodedNode *N, bool TookTrue, 2883 bool IsAssuming) { 2884 QualType Ty = CondVarExpr->getType(); 2885 2886 if (Ty->isPointerType()) { 2887 Out << (TookTrue ? "non-null" : "null"); 2888 return true; 2889 } 2890 2891 if (Ty->isObjCObjectPointerType()) { 2892 Out << (TookTrue ? "non-nil" : "nil"); 2893 return true; 2894 } 2895 2896 if (!Ty->isIntegralOrEnumerationType()) 2897 return false; 2898 2899 Optional<const llvm::APSInt *> IntValue; 2900 if (!IsAssuming) 2901 IntValue = getConcreteIntegerValue(CondVarExpr, N); 2902 2903 if (IsAssuming || !IntValue.hasValue()) { 2904 if (Ty->isBooleanType()) 2905 Out << (TookTrue ? "true" : "false"); 2906 else 2907 Out << (TookTrue ? "not equal to 0" : "0"); 2908 } else { 2909 if (Ty->isBooleanType()) 2910 Out << (IntValue.getValue()->getBoolValue() ? "true" : "false"); 2911 else 2912 Out << *IntValue.getValue(); 2913 } 2914 2915 return true; 2916 } 2917 2918 constexpr llvm::StringLiteral ConditionBRVisitor::GenericTrueMessage; 2919 constexpr llvm::StringLiteral ConditionBRVisitor::GenericFalseMessage; 2920 2921 bool ConditionBRVisitor::isPieceMessageGeneric( 2922 const PathDiagnosticPiece *Piece) { 2923 return Piece->getString() == GenericTrueMessage || 2924 Piece->getString() == GenericFalseMessage; 2925 } 2926 2927 //===----------------------------------------------------------------------===// 2928 // Implementation of LikelyFalsePositiveSuppressionBRVisitor. 2929 //===----------------------------------------------------------------------===// 2930 2931 void LikelyFalsePositiveSuppressionBRVisitor::finalizeVisitor( 2932 BugReporterContext &BRC, const ExplodedNode *N, 2933 PathSensitiveBugReport &BR) { 2934 // Here we suppress false positives coming from system headers. This list is 2935 // based on known issues. 2936 const AnalyzerOptions &Options = BRC.getAnalyzerOptions(); 2937 const Decl *D = N->getLocationContext()->getDecl(); 2938 2939 if (AnalysisDeclContext::isInStdNamespace(D)) { 2940 // Skip reports within the 'std' namespace. Although these can sometimes be 2941 // the user's fault, we currently don't report them very well, and 2942 // Note that this will not help for any other data structure libraries, like 2943 // TR1, Boost, or llvm/ADT. 2944 if (Options.ShouldSuppressFromCXXStandardLibrary) { 2945 BR.markInvalid(getTag(), nullptr); 2946 return; 2947 } else { 2948 // If the complete 'std' suppression is not enabled, suppress reports 2949 // from the 'std' namespace that are known to produce false positives. 2950 2951 // The analyzer issues a false use-after-free when std::list::pop_front 2952 // or std::list::pop_back are called multiple times because we cannot 2953 // reason about the internal invariants of the data structure. 2954 if (const auto *MD = dyn_cast<CXXMethodDecl>(D)) { 2955 const CXXRecordDecl *CD = MD->getParent(); 2956 if (CD->getName() == "list") { 2957 BR.markInvalid(getTag(), nullptr); 2958 return; 2959 } 2960 } 2961 2962 // The analyzer issues a false positive when the constructor of 2963 // std::__independent_bits_engine from algorithms is used. 2964 if (const auto *MD = dyn_cast<CXXConstructorDecl>(D)) { 2965 const CXXRecordDecl *CD = MD->getParent(); 2966 if (CD->getName() == "__independent_bits_engine") { 2967 BR.markInvalid(getTag(), nullptr); 2968 return; 2969 } 2970 } 2971 2972 for (const LocationContext *LCtx = N->getLocationContext(); LCtx; 2973 LCtx = LCtx->getParent()) { 2974 const auto *MD = dyn_cast<CXXMethodDecl>(LCtx->getDecl()); 2975 if (!MD) 2976 continue; 2977 2978 const CXXRecordDecl *CD = MD->getParent(); 2979 // The analyzer issues a false positive on 2980 // std::basic_string<uint8_t> v; v.push_back(1); 2981 // and 2982 // std::u16string s; s += u'a'; 2983 // because we cannot reason about the internal invariants of the 2984 // data structure. 2985 if (CD->getName() == "basic_string") { 2986 BR.markInvalid(getTag(), nullptr); 2987 return; 2988 } 2989 2990 // The analyzer issues a false positive on 2991 // std::shared_ptr<int> p(new int(1)); p = nullptr; 2992 // because it does not reason properly about temporary destructors. 2993 if (CD->getName() == "shared_ptr") { 2994 BR.markInvalid(getTag(), nullptr); 2995 return; 2996 } 2997 } 2998 } 2999 } 3000 3001 // Skip reports within the sys/queue.h macros as we do not have the ability to 3002 // reason about data structure shapes. 3003 const SourceManager &SM = BRC.getSourceManager(); 3004 FullSourceLoc Loc = BR.getLocation().asLocation(); 3005 while (Loc.isMacroID()) { 3006 Loc = Loc.getSpellingLoc(); 3007 if (SM.getFilename(Loc).endswith("sys/queue.h")) { 3008 BR.markInvalid(getTag(), nullptr); 3009 return; 3010 } 3011 } 3012 } 3013 3014 //===----------------------------------------------------------------------===// 3015 // Implementation of UndefOrNullArgVisitor. 3016 //===----------------------------------------------------------------------===// 3017 3018 PathDiagnosticPieceRef 3019 UndefOrNullArgVisitor::VisitNode(const ExplodedNode *N, BugReporterContext &BRC, 3020 PathSensitiveBugReport &BR) { 3021 ProgramStateRef State = N->getState(); 3022 ProgramPoint ProgLoc = N->getLocation(); 3023 3024 // We are only interested in visiting CallEnter nodes. 3025 Optional<CallEnter> CEnter = ProgLoc.getAs<CallEnter>(); 3026 if (!CEnter) 3027 return nullptr; 3028 3029 // Check if one of the arguments is the region the visitor is tracking. 3030 CallEventManager &CEMgr = BRC.getStateManager().getCallEventManager(); 3031 CallEventRef<> Call = CEMgr.getCaller(CEnter->getCalleeContext(), State); 3032 unsigned Idx = 0; 3033 ArrayRef<ParmVarDecl *> parms = Call->parameters(); 3034 3035 for (const auto ParamDecl : parms) { 3036 const MemRegion *ArgReg = Call->getArgSVal(Idx).getAsRegion(); 3037 ++Idx; 3038 3039 // Are we tracking the argument or its subregion? 3040 if ( !ArgReg || !R->isSubRegionOf(ArgReg->StripCasts())) 3041 continue; 3042 3043 // Check the function parameter type. 3044 assert(ParamDecl && "Formal parameter has no decl?"); 3045 QualType T = ParamDecl->getType(); 3046 3047 if (!(T->isAnyPointerType() || T->isReferenceType())) { 3048 // Function can only change the value passed in by address. 3049 continue; 3050 } 3051 3052 // If it is a const pointer value, the function does not intend to 3053 // change the value. 3054 if (T->getPointeeType().isConstQualified()) 3055 continue; 3056 3057 // Mark the call site (LocationContext) as interesting if the value of the 3058 // argument is undefined or '0'/'NULL'. 3059 SVal BoundVal = State->getSVal(R); 3060 if (BoundVal.isUndef() || BoundVal.isZeroConstant()) { 3061 BR.markInteresting(CEnter->getCalleeContext()); 3062 return nullptr; 3063 } 3064 } 3065 return nullptr; 3066 } 3067 3068 //===----------------------------------------------------------------------===// 3069 // Implementation of FalsePositiveRefutationBRVisitor. 3070 //===----------------------------------------------------------------------===// 3071 3072 FalsePositiveRefutationBRVisitor::FalsePositiveRefutationBRVisitor() 3073 : Constraints(ConstraintMap::Factory().getEmptyMap()) {} 3074 3075 void FalsePositiveRefutationBRVisitor::finalizeVisitor( 3076 BugReporterContext &BRC, const ExplodedNode *EndPathNode, 3077 PathSensitiveBugReport &BR) { 3078 // Collect new constraints 3079 addConstraints(EndPathNode, /*OverwriteConstraintsOnExistingSyms=*/true); 3080 3081 // Create a refutation manager 3082 llvm::SMTSolverRef RefutationSolver = llvm::CreateZ3Solver(); 3083 ASTContext &Ctx = BRC.getASTContext(); 3084 3085 // Add constraints to the solver 3086 for (const auto &I : Constraints) { 3087 const SymbolRef Sym = I.first; 3088 auto RangeIt = I.second.begin(); 3089 3090 llvm::SMTExprRef SMTConstraints = SMTConv::getRangeExpr( 3091 RefutationSolver, Ctx, Sym, RangeIt->From(), RangeIt->To(), 3092 /*InRange=*/true); 3093 while ((++RangeIt) != I.second.end()) { 3094 SMTConstraints = RefutationSolver->mkOr( 3095 SMTConstraints, SMTConv::getRangeExpr(RefutationSolver, Ctx, Sym, 3096 RangeIt->From(), RangeIt->To(), 3097 /*InRange=*/true)); 3098 } 3099 3100 RefutationSolver->addConstraint(SMTConstraints); 3101 } 3102 3103 // And check for satisfiability 3104 Optional<bool> IsSAT = RefutationSolver->check(); 3105 if (!IsSAT.hasValue()) 3106 return; 3107 3108 if (!IsSAT.getValue()) 3109 BR.markInvalid("Infeasible constraints", EndPathNode->getLocationContext()); 3110 } 3111 3112 void FalsePositiveRefutationBRVisitor::addConstraints( 3113 const ExplodedNode *N, bool OverwriteConstraintsOnExistingSyms) { 3114 // Collect new constraints 3115 ConstraintMap NewCs = getConstraintMap(N->getState()); 3116 ConstraintMap::Factory &CF = N->getState()->get_context<ConstraintMap>(); 3117 3118 // Add constraints if we don't have them yet 3119 for (auto const &C : NewCs) { 3120 const SymbolRef &Sym = C.first; 3121 if (!Constraints.contains(Sym)) { 3122 // This symbol is new, just add the constraint. 3123 Constraints = CF.add(Constraints, Sym, C.second); 3124 } else if (OverwriteConstraintsOnExistingSyms) { 3125 // Overwrite the associated constraint of the Symbol. 3126 Constraints = CF.remove(Constraints, Sym); 3127 Constraints = CF.add(Constraints, Sym, C.second); 3128 } 3129 } 3130 } 3131 3132 PathDiagnosticPieceRef FalsePositiveRefutationBRVisitor::VisitNode( 3133 const ExplodedNode *N, BugReporterContext &, PathSensitiveBugReport &) { 3134 addConstraints(N, /*OverwriteConstraintsOnExistingSyms=*/false); 3135 return nullptr; 3136 } 3137 3138 void FalsePositiveRefutationBRVisitor::Profile( 3139 llvm::FoldingSetNodeID &ID) const { 3140 static int Tag = 0; 3141 ID.AddPointer(&Tag); 3142 } 3143 3144 //===----------------------------------------------------------------------===// 3145 // Implementation of TagVisitor. 3146 //===----------------------------------------------------------------------===// 3147 3148 int NoteTag::Kind = 0; 3149 3150 void TagVisitor::Profile(llvm::FoldingSetNodeID &ID) const { 3151 static int Tag = 0; 3152 ID.AddPointer(&Tag); 3153 } 3154 3155 PathDiagnosticPieceRef TagVisitor::VisitNode(const ExplodedNode *N, 3156 BugReporterContext &BRC, 3157 PathSensitiveBugReport &R) { 3158 ProgramPoint PP = N->getLocation(); 3159 const NoteTag *T = dyn_cast_or_null<NoteTag>(PP.getTag()); 3160 if (!T) 3161 return nullptr; 3162 3163 if (Optional<std::string> Msg = T->generateMessage(BRC, R)) { 3164 PathDiagnosticLocation Loc = 3165 PathDiagnosticLocation::create(PP, BRC.getSourceManager()); 3166 auto Piece = std::make_shared<PathDiagnosticEventPiece>(Loc, *Msg); 3167 Piece->setPrunable(T->isPrunable()); 3168 return Piece; 3169 } 3170 3171 return nullptr; 3172 } 3173