//===- BugReporter.cpp - Generate PathDiagnostics for bugs ----------------===// // // The LLVM Compiler Infrastructure // // This file is distributed under the University of Illinois Open Source // License. See LICENSE.TXT for details. // //===----------------------------------------------------------------------===// // // This file defines BugReporter, a utility class for generating // PathDiagnostics. // //===----------------------------------------------------------------------===// #include "clang/StaticAnalyzer/Core/BugReporter/BugReporter.h" #include "clang/AST/Decl.h" #include "clang/AST/DeclBase.h" #include "clang/AST/DeclObjC.h" #include "clang/AST/Expr.h" #include "clang/AST/ExprCXX.h" #include "clang/AST/ParentMap.h" #include "clang/AST/Stmt.h" #include "clang/AST/StmtCXX.h" #include "clang/AST/StmtObjC.h" #include "clang/Analysis/AnalysisDeclContext.h" #include "clang/Analysis/CFG.h" #include "clang/Analysis/CFGStmtMap.h" #include "clang/Analysis/ProgramPoint.h" #include "clang/Basic/LLVM.h" #include "clang/Basic/SourceLocation.h" #include "clang/Basic/SourceManager.h" #include "clang/StaticAnalyzer/Core/AnalyzerOptions.h" #include "clang/StaticAnalyzer/Core/BugReporter/BugReporterVisitors.h" #include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" #include "clang/StaticAnalyzer/Core/BugReporter/PathDiagnostic.h" #include "clang/StaticAnalyzer/Core/Checker.h" #include "clang/StaticAnalyzer/Core/CheckerManager.h" #include "clang/StaticAnalyzer/Core/PathSensitive/ExplodedGraph.h" #include "clang/StaticAnalyzer/Core/PathSensitive/ExprEngine.h" #include "clang/StaticAnalyzer/Core/PathSensitive/MemRegion.h" #include "clang/StaticAnalyzer/Core/PathSensitive/ProgramState.h" #include "clang/StaticAnalyzer/Core/PathSensitive/SVals.h" #include "clang/StaticAnalyzer/Core/PathSensitive/SymbolManager.h" #include "llvm/ADT/ArrayRef.h" #include "llvm/ADT/DenseMap.h" #include "llvm/ADT/DenseSet.h" #include "llvm/ADT/FoldingSet.h" #include "llvm/ADT/None.h" #include "llvm/ADT/Optional.h" #include "llvm/ADT/STLExtras.h" #include "llvm/ADT/SmallPtrSet.h" #include "llvm/ADT/SmallString.h" #include "llvm/ADT/SmallVector.h" #include "llvm/ADT/Statistic.h" #include "llvm/ADT/StringRef.h" #include "llvm/ADT/iterator_range.h" #include "llvm/Support/Casting.h" #include "llvm/Support/Compiler.h" #include "llvm/Support/ErrorHandling.h" #include "llvm/Support/MemoryBuffer.h" #include "llvm/Support/raw_ostream.h" #include #include #include #include #include #include #include #include #include #include using namespace clang; using namespace ento; #define DEBUG_TYPE "BugReporter" STATISTIC(MaxBugClassSize, "The maximum number of bug reports in the same equivalence class"); STATISTIC(MaxValidBugClassSize, "The maximum number of bug reports in the same equivalence class " "where at least one report is valid (not suppressed)"); BugReporterVisitor::~BugReporterVisitor() = default; void BugReporterContext::anchor() {} //===----------------------------------------------------------------------===// // Helper routines for walking the ExplodedGraph and fetching statements. //===----------------------------------------------------------------------===// static const Stmt *GetPreviousStmt(const ExplodedNode *N) { for (N = N->getFirstPred(); N; N = N->getFirstPred()) if (const Stmt *S = PathDiagnosticLocation::getStmt(N)) return S; return nullptr; } static inline const Stmt* GetCurrentOrPreviousStmt(const ExplodedNode *N) { if (const Stmt *S = PathDiagnosticLocation::getStmt(N)) return S; return GetPreviousStmt(N); } //===----------------------------------------------------------------------===// // Diagnostic cleanup. //===----------------------------------------------------------------------===// static PathDiagnosticEventPiece * eventsDescribeSameCondition(PathDiagnosticEventPiece *X, PathDiagnosticEventPiece *Y) { // Prefer diagnostics that come from ConditionBRVisitor over // those that came from TrackConstraintBRVisitor, // unless the one from ConditionBRVisitor is // its generic fallback diagnostic. const void *tagPreferred = ConditionBRVisitor::getTag(); const void *tagLesser = TrackConstraintBRVisitor::getTag(); if (X->getLocation() != Y->getLocation()) return nullptr; if (X->getTag() == tagPreferred && Y->getTag() == tagLesser) return ConditionBRVisitor::isPieceMessageGeneric(X) ? Y : X; if (Y->getTag() == tagPreferred && X->getTag() == tagLesser) return ConditionBRVisitor::isPieceMessageGeneric(Y) ? X : Y; return nullptr; } /// An optimization pass over PathPieces that removes redundant diagnostics /// generated by both ConditionBRVisitor and TrackConstraintBRVisitor. Both /// BugReporterVisitors use different methods to generate diagnostics, with /// one capable of emitting diagnostics in some cases but not in others. This /// can lead to redundant diagnostic pieces at the same point in a path. static void removeRedundantMsgs(PathPieces &path) { unsigned N = path.size(); if (N < 2) return; // NOTE: this loop intentionally is not using an iterator. Instead, we // are streaming the path and modifying it in place. This is done by // grabbing the front, processing it, and if we decide to keep it append // it to the end of the path. The entire path is processed in this way. for (unsigned i = 0; i < N; ++i) { auto piece = std::move(path.front()); path.pop_front(); switch (piece->getKind()) { case PathDiagnosticPiece::Call: removeRedundantMsgs(cast(*piece).path); break; case PathDiagnosticPiece::Macro: removeRedundantMsgs(cast(*piece).subPieces); break; case PathDiagnosticPiece::ControlFlow: break; case PathDiagnosticPiece::Event: { if (i == N-1) break; if (auto *nextEvent = dyn_cast(path.front().get())) { auto *event = cast(piece.get()); // Check to see if we should keep one of the two pieces. If we // come up with a preference, record which piece to keep, and consume // another piece from the path. if (auto *pieceToKeep = eventsDescribeSameCondition(event, nextEvent)) { piece = std::move(pieceToKeep == event ? piece : path.front()); path.pop_front(); ++i; } } break; } case PathDiagnosticPiece::Note: break; } path.push_back(std::move(piece)); } } /// A map from PathDiagnosticPiece to the LocationContext of the inlined /// function call it represents. using LocationContextMap = llvm::DenseMap; /// Recursively scan through a path and prune out calls and macros pieces /// that aren't needed. Return true if afterwards the path contains /// "interesting stuff" which means it shouldn't be pruned from the parent path. static bool removeUnneededCalls(PathPieces &pieces, BugReport *R, LocationContextMap &LCM, bool IsInteresting = false) { bool containsSomethingInteresting = IsInteresting; const unsigned N = pieces.size(); for (unsigned i = 0 ; i < N ; ++i) { // Remove the front piece from the path. If it is still something we // want to keep once we are done, we will push it back on the end. auto piece = std::move(pieces.front()); pieces.pop_front(); switch (piece->getKind()) { case PathDiagnosticPiece::Call: { auto &call = cast(*piece); // Check if the location context is interesting. assert(LCM.count(&call.path)); if (!removeUnneededCalls(call.path, R, LCM, R->isInteresting(LCM[&call.path]))) continue; containsSomethingInteresting = true; break; } case PathDiagnosticPiece::Macro: { auto ¯o = cast(*piece); if (!removeUnneededCalls(macro.subPieces, R, LCM, IsInteresting)) continue; containsSomethingInteresting = true; break; } case PathDiagnosticPiece::Event: { auto &event = cast(*piece); // We never throw away an event, but we do throw it away wholesale // as part of a path if we throw the entire path away. containsSomethingInteresting |= !event.isPrunable(); break; } case PathDiagnosticPiece::ControlFlow: break; case PathDiagnosticPiece::Note: break; } pieces.push_back(std::move(piece)); } return containsSomethingInteresting; } /// Returns true if the given decl has been implicitly given a body, either by /// the analyzer or by the compiler proper. static bool hasImplicitBody(const Decl *D) { assert(D); return D->isImplicit() || !D->hasBody(); } /// Recursively scan through a path and make sure that all call pieces have /// valid locations. static void adjustCallLocations(PathPieces &Pieces, PathDiagnosticLocation *LastCallLocation = nullptr) { for (const auto &I : Pieces) { auto *Call = dyn_cast(I.get()); if (!Call) continue; if (LastCallLocation) { bool CallerIsImplicit = hasImplicitBody(Call->getCaller()); if (CallerIsImplicit || !Call->callEnter.asLocation().isValid()) Call->callEnter = *LastCallLocation; if (CallerIsImplicit || !Call->callReturn.asLocation().isValid()) Call->callReturn = *LastCallLocation; } // Recursively clean out the subclass. Keep this call around if // it contains any informative diagnostics. PathDiagnosticLocation *ThisCallLocation; if (Call->callEnterWithin.asLocation().isValid() && !hasImplicitBody(Call->getCallee())) ThisCallLocation = &Call->callEnterWithin; else ThisCallLocation = &Call->callEnter; assert(ThisCallLocation && "Outermost call has an invalid location"); adjustCallLocations(Call->path, ThisCallLocation); } } /// Remove edges in and out of C++ default initializer expressions. These are /// for fields that have in-class initializers, as opposed to being initialized /// explicitly in a constructor or braced list. static void removeEdgesToDefaultInitializers(PathPieces &Pieces) { for (PathPieces::iterator I = Pieces.begin(), E = Pieces.end(); I != E;) { if (auto *C = dyn_cast(I->get())) removeEdgesToDefaultInitializers(C->path); if (auto *M = dyn_cast(I->get())) removeEdgesToDefaultInitializers(M->subPieces); if (auto *CF = dyn_cast(I->get())) { const Stmt *Start = CF->getStartLocation().asStmt(); const Stmt *End = CF->getEndLocation().asStmt(); if (Start && isa(Start)) { I = Pieces.erase(I); continue; } else if (End && isa(End)) { PathPieces::iterator Next = std::next(I); if (Next != E) { if (auto *NextCF = dyn_cast(Next->get())) { NextCF->setStartLocation(CF->getStartLocation()); } } I = Pieces.erase(I); continue; } } I++; } } /// Remove all pieces with invalid locations as these cannot be serialized. /// We might have pieces with invalid locations as a result of inlining Body /// Farm generated functions. static void removePiecesWithInvalidLocations(PathPieces &Pieces) { for (PathPieces::iterator I = Pieces.begin(), E = Pieces.end(); I != E;) { if (auto *C = dyn_cast(I->get())) removePiecesWithInvalidLocations(C->path); if (auto *M = dyn_cast(I->get())) removePiecesWithInvalidLocations(M->subPieces); if (!(*I)->getLocation().isValid() || !(*I)->getLocation().asLocation().isValid()) { I = Pieces.erase(I); continue; } I++; } } //===----------------------------------------------------------------------===// // PathDiagnosticBuilder and its associated routines and helper objects. //===----------------------------------------------------------------------===// namespace { class PathDiagnosticBuilder : public BugReporterContext { BugReport *R; PathDiagnosticConsumer *PDC; public: const LocationContext *LC; PathDiagnosticBuilder(GRBugReporter &br, BugReport *r, InterExplodedGraphMap &Backmap, PathDiagnosticConsumer *pdc) : BugReporterContext(br, Backmap), R(r), PDC(pdc), LC(r->getErrorNode()->getLocationContext()) {} PathDiagnosticLocation ExecutionContinues(const ExplodedNode *N); PathDiagnosticLocation ExecutionContinues(llvm::raw_string_ostream &os, const ExplodedNode *N); BugReport *getBugReport() { return R; } Decl const &getCodeDecl() { return R->getErrorNode()->getCodeDecl(); } ParentMap& getParentMap() { return LC->getParentMap(); } const Stmt *getParent(const Stmt *S) { return getParentMap().getParent(S); } PathDiagnosticLocation getEnclosingStmtLocation(const Stmt *S); PathDiagnosticConsumer::PathGenerationScheme getGenerationScheme() const { return PDC ? PDC->getGenerationScheme() : PathDiagnosticConsumer::Minimal; } bool supportsLogicalOpControlFlow() const { return PDC ? PDC->supportsLogicalOpControlFlow() : true; } }; } // namespace PathDiagnosticLocation PathDiagnosticBuilder::ExecutionContinues(const ExplodedNode *N) { if (const Stmt *S = PathDiagnosticLocation::getNextStmt(N)) return PathDiagnosticLocation(S, getSourceManager(), LC); return PathDiagnosticLocation::createDeclEnd(N->getLocationContext(), getSourceManager()); } PathDiagnosticLocation PathDiagnosticBuilder::ExecutionContinues(llvm::raw_string_ostream &os, const ExplodedNode *N) { // Slow, but probably doesn't matter. if (os.str().empty()) os << ' '; const PathDiagnosticLocation &Loc = ExecutionContinues(N); if (Loc.asStmt()) os << "Execution continues on line " << getSourceManager().getExpansionLineNumber(Loc.asLocation()) << '.'; else { os << "Execution jumps to the end of the "; const Decl *D = N->getLocationContext()->getDecl(); if (isa(D)) os << "method"; else if (isa(D)) os << "function"; else { assert(isa(D)); os << "anonymous block"; } os << '.'; } return Loc; } static const Stmt *getEnclosingParent(const Stmt *S, const ParentMap &PM) { if (isa(S) && PM.isConsumedExpr(cast(S))) return PM.getParentIgnoreParens(S); const Stmt *Parent = PM.getParentIgnoreParens(S); if (!Parent) return nullptr; switch (Parent->getStmtClass()) { case Stmt::ForStmtClass: case Stmt::DoStmtClass: case Stmt::WhileStmtClass: case Stmt::ObjCForCollectionStmtClass: case Stmt::CXXForRangeStmtClass: return Parent; default: break; } return nullptr; } static PathDiagnosticLocation getEnclosingStmtLocation(const Stmt *S, SourceManager &SMgr, const ParentMap &P, const LocationContext *LC, bool allowNestedContexts) { if (!S) return {}; while (const Stmt *Parent = getEnclosingParent(S, P)) { switch (Parent->getStmtClass()) { case Stmt::BinaryOperatorClass: { const auto *B = cast(Parent); if (B->isLogicalOp()) return PathDiagnosticLocation(allowNestedContexts ? B : S, SMgr, LC); break; } case Stmt::CompoundStmtClass: case Stmt::StmtExprClass: return PathDiagnosticLocation(S, SMgr, LC); case Stmt::ChooseExprClass: // Similar to '?' if we are referring to condition, just have the edge // point to the entire choose expression. if (allowNestedContexts || cast(Parent)->getCond() == S) return PathDiagnosticLocation(Parent, SMgr, LC); else return PathDiagnosticLocation(S, SMgr, LC); case Stmt::BinaryConditionalOperatorClass: case Stmt::ConditionalOperatorClass: // For '?', if we are referring to condition, just have the edge point // to the entire '?' expression. if (allowNestedContexts || cast(Parent)->getCond() == S) return PathDiagnosticLocation(Parent, SMgr, LC); else return PathDiagnosticLocation(S, SMgr, LC); case Stmt::CXXForRangeStmtClass: if (cast(Parent)->getBody() == S) return PathDiagnosticLocation(S, SMgr, LC); break; case Stmt::DoStmtClass: return PathDiagnosticLocation(S, SMgr, LC); case Stmt::ForStmtClass: if (cast(Parent)->getBody() == S) return PathDiagnosticLocation(S, SMgr, LC); break; case Stmt::IfStmtClass: if (cast(Parent)->getCond() != S) return PathDiagnosticLocation(S, SMgr, LC); break; case Stmt::ObjCForCollectionStmtClass: if (cast(Parent)->getBody() == S) return PathDiagnosticLocation(S, SMgr, LC); break; case Stmt::WhileStmtClass: if (cast(Parent)->getCond() != S) return PathDiagnosticLocation(S, SMgr, LC); break; default: break; } S = Parent; } assert(S && "Cannot have null Stmt for PathDiagnosticLocation"); return PathDiagnosticLocation(S, SMgr, LC); } PathDiagnosticLocation PathDiagnosticBuilder::getEnclosingStmtLocation(const Stmt *S) { assert(S && "Null Stmt passed to getEnclosingStmtLocation"); return ::getEnclosingStmtLocation(S, getSourceManager(), getParentMap(), LC, /*allowNestedContexts=*/false); } //===----------------------------------------------------------------------===// // "Minimal" path diagnostic generation algorithm. //===----------------------------------------------------------------------===// using StackDiagPair = std::pair; using StackDiagVector = SmallVector; static void updateStackPiecesWithMessage(PathDiagnosticPiece &P, StackDiagVector &CallStack) { // If the piece contains a special message, add it to all the call // pieces on the active stack. if (auto *ep = dyn_cast(&P)) { if (ep->hasCallStackHint()) for (const auto &I : CallStack) { PathDiagnosticCallPiece *CP = I.first; const ExplodedNode *N = I.second; std::string stackMsg = ep->getCallStackMessage(N); // The last message on the path to final bug is the most important // one. Since we traverse the path backwards, do not add the message // if one has been previously added. if (!CP->hasCallStackMessage()) CP->setCallStackMessage(stackMsg); } } } static void CompactPathDiagnostic(PathPieces &path, const SourceManager& SM); std::shared_ptr generateDiagForSwitchOP( const ExplodedNode *N, const CFGBlock *Dst, const SourceManager &SM, const LocationContext *LC, PathDiagnosticBuilder &PDB, PathDiagnosticLocation &Start ) { // Figure out what case arm we took. std::string sbuf; llvm::raw_string_ostream os(sbuf); PathDiagnosticLocation End; if (const Stmt *S = Dst->getLabel()) { End = PathDiagnosticLocation(S, SM, LC); switch (S->getStmtClass()) { default: os << "No cases match in the switch statement. " "Control jumps to line " << End.asLocation().getExpansionLineNumber(); break; case Stmt::DefaultStmtClass: os << "Control jumps to the 'default' case at line " << End.asLocation().getExpansionLineNumber(); break; case Stmt::CaseStmtClass: { os << "Control jumps to 'case "; const auto *Case = cast(S); const Expr *LHS = Case->getLHS()->IgnoreParenCasts(); // Determine if it is an enum. bool GetRawInt = true; if (const auto *DR = dyn_cast(LHS)) { // FIXME: Maybe this should be an assertion. Are there cases // were it is not an EnumConstantDecl? const auto *D = dyn_cast(DR->getDecl()); if (D) { GetRawInt = false; os << *D; } } if (GetRawInt) os << LHS->EvaluateKnownConstInt(PDB.getASTContext()); os << ":' at line " << End.asLocation().getExpansionLineNumber(); break; } } } else { os << "'Default' branch taken. "; End = PDB.ExecutionContinues(os, N); } return std::make_shared(Start, End, os.str()); } std::shared_ptr generateDiagForGotoOP( const Stmt *S, PathDiagnosticBuilder &PDB, PathDiagnosticLocation &Start) { std::string sbuf; llvm::raw_string_ostream os(sbuf); const PathDiagnosticLocation &End = PDB.getEnclosingStmtLocation(S); os << "Control jumps to line " << End.asLocation().getExpansionLineNumber(); return std::make_shared(Start, End, os.str()); } std::shared_ptr generateDiagForBinaryOP( const ExplodedNode *N, const Stmt *T, const CFGBlock *Src, const CFGBlock *Dst, const SourceManager &SM, PathDiagnosticBuilder &PDB, const LocationContext *LC) { const auto *B = cast(T); std::string sbuf; llvm::raw_string_ostream os(sbuf); os << "Left side of '"; PathDiagnosticLocation Start, End; if (B->getOpcode() == BO_LAnd) { os << "&&" << "' is "; if (*(Src->succ_begin() + 1) == Dst) { os << "false"; End = PathDiagnosticLocation(B->getLHS(), SM, LC); Start = PathDiagnosticLocation::createOperatorLoc(B, SM); } else { os << "true"; Start = PathDiagnosticLocation(B->getLHS(), SM, LC); End = PDB.ExecutionContinues(N); } } else { assert(B->getOpcode() == BO_LOr); os << "||" << "' is "; if (*(Src->succ_begin() + 1) == Dst) { os << "false"; Start = PathDiagnosticLocation(B->getLHS(), SM, LC); End = PDB.ExecutionContinues(N); } else { os << "true"; End = PathDiagnosticLocation(B->getLHS(), SM, LC); Start = PathDiagnosticLocation::createOperatorLoc(B, SM); } } return std::make_shared(Start, End, os.str()); } void generateMinimalDiagForBlockEdge(const ExplodedNode *N, BlockEdge BE, const SourceManager &SM, PathDiagnosticBuilder &PDB, PathDiagnostic &PD) { const LocationContext *LC = N->getLocationContext(); const CFGBlock *Src = BE.getSrc(); const CFGBlock *Dst = BE.getDst(); const Stmt *T = Src->getTerminator(); if (!T) return; auto Start = PathDiagnosticLocation::createBegin(T, SM, LC); switch (T->getStmtClass()) { default: break; case Stmt::GotoStmtClass: case Stmt::IndirectGotoStmtClass: { if (const Stmt *S = PathDiagnosticLocation::getNextStmt(N)) PD.getActivePath().push_front(generateDiagForGotoOP(S, PDB, Start)); break; } case Stmt::SwitchStmtClass: { PD.getActivePath().push_front( generateDiagForSwitchOP(N, Dst, SM, LC, PDB, Start)); break; } case Stmt::BreakStmtClass: case Stmt::ContinueStmtClass: { std::string sbuf; llvm::raw_string_ostream os(sbuf); PathDiagnosticLocation End = PDB.ExecutionContinues(os, N); PD.getActivePath().push_front( std::make_shared(Start, End, os.str())); break; } // Determine control-flow for ternary '?'. case Stmt::BinaryConditionalOperatorClass: case Stmt::ConditionalOperatorClass: { std::string sbuf; llvm::raw_string_ostream os(sbuf); os << "'?' condition is "; if (*(Src->succ_begin() + 1) == Dst) os << "false"; else os << "true"; PathDiagnosticLocation End = PDB.ExecutionContinues(N); if (const Stmt *S = End.asStmt()) End = PDB.getEnclosingStmtLocation(S); PD.getActivePath().push_front( std::make_shared(Start, End, os.str())); break; } // Determine control-flow for short-circuited '&&' and '||'. case Stmt::BinaryOperatorClass: { if (!PDB.supportsLogicalOpControlFlow()) break; std::shared_ptr Diag = generateDiagForBinaryOP(N, T, Src, Dst, SM, PDB, LC); PD.getActivePath().push_front(Diag); break; } case Stmt::DoStmtClass: if (*(Src->succ_begin()) == Dst) { std::string sbuf; llvm::raw_string_ostream os(sbuf); os << "Loop condition is true. "; PathDiagnosticLocation End = PDB.ExecutionContinues(os, N); if (const Stmt *S = End.asStmt()) End = PDB.getEnclosingStmtLocation(S); PD.getActivePath().push_front( std::make_shared(Start, End, os.str())); } else { PathDiagnosticLocation End = PDB.ExecutionContinues(N); if (const Stmt *S = End.asStmt()) End = PDB.getEnclosingStmtLocation(S); PD.getActivePath().push_front( std::make_shared( Start, End, "Loop condition is false. Exiting loop")); } break; case Stmt::WhileStmtClass: case Stmt::ForStmtClass: if (*(Src->succ_begin() + 1) == Dst) { std::string sbuf; llvm::raw_string_ostream os(sbuf); os << "Loop condition is false. "; PathDiagnosticLocation End = PDB.ExecutionContinues(os, N); if (const Stmt *S = End.asStmt()) End = PDB.getEnclosingStmtLocation(S); PD.getActivePath().push_front( std::make_shared(Start, End, os.str())); } else { PathDiagnosticLocation End = PDB.ExecutionContinues(N); if (const Stmt *S = End.asStmt()) End = PDB.getEnclosingStmtLocation(S); PD.getActivePath().push_front( std::make_shared( Start, End, "Loop condition is true. Entering loop body")); } break; case Stmt::IfStmtClass: { PathDiagnosticLocation End = PDB.ExecutionContinues(N); if (const Stmt *S = End.asStmt()) End = PDB.getEnclosingStmtLocation(S); if (*(Src->succ_begin() + 1) == Dst) PD.getActivePath().push_front( std::make_shared( Start, End, "Taking false branch")); else PD.getActivePath().push_front( std::make_shared( Start, End, "Taking true branch")); break; } } } // Cone-of-influence: support the reverse propagation of "interesting" symbols // and values by tracing interesting calculations backwards through evaluated // expressions along a path. This is probably overly complicated, but the idea // is that if an expression computed an "interesting" value, the child // expressions are are also likely to be "interesting" as well (which then // propagates to the values they in turn compute). This reverse propagation // is needed to track interesting correlations across function call boundaries, // where formal arguments bind to actual arguments, etc. This is also needed // because the constraint solver sometimes simplifies certain symbolic values // into constants when appropriate, and this complicates reasoning about // interesting values. using InterestingExprs = llvm::DenseSet; static void reversePropagateIntererstingSymbols(BugReport &R, InterestingExprs &IE, const ProgramState *State, const Expr *Ex, const LocationContext *LCtx) { SVal V = State->getSVal(Ex, LCtx); if (!(R.isInteresting(V) || IE.count(Ex))) return; switch (Ex->getStmtClass()) { default: if (!isa(Ex)) break; // Fall through. case Stmt::BinaryOperatorClass: case Stmt::UnaryOperatorClass: { for (const Stmt *SubStmt : Ex->children()) { if (const auto *child = dyn_cast_or_null(SubStmt)) { IE.insert(child); SVal ChildV = State->getSVal(child, LCtx); R.markInteresting(ChildV); } } break; } } R.markInteresting(V); } static void reversePropagateInterestingSymbols(BugReport &R, InterestingExprs &IE, const ProgramState *State, const LocationContext *CalleeCtx, const LocationContext *CallerCtx) { // FIXME: Handle non-CallExpr-based CallEvents. const StackFrameContext *Callee = CalleeCtx->getStackFrame(); const Stmt *CallSite = Callee->getCallSite(); if (const auto *CE = dyn_cast_or_null(CallSite)) { if (const auto *FD = dyn_cast(CalleeCtx->getDecl())) { FunctionDecl::param_const_iterator PI = FD->param_begin(), PE = FD->param_end(); CallExpr::const_arg_iterator AI = CE->arg_begin(), AE = CE->arg_end(); for (; AI != AE && PI != PE; ++AI, ++PI) { if (const Expr *ArgE = *AI) { if (const ParmVarDecl *PD = *PI) { Loc LV = State->getLValue(PD, CalleeCtx); if (R.isInteresting(LV) || R.isInteresting(State->getRawSVal(LV))) IE.insert(ArgE); } } } } } } //===----------------------------------------------------------------------===// // Functions for determining if a loop was executed 0 times. //===----------------------------------------------------------------------===// static bool isLoop(const Stmt *Term) { switch (Term->getStmtClass()) { case Stmt::ForStmtClass: case Stmt::WhileStmtClass: case Stmt::ObjCForCollectionStmtClass: case Stmt::CXXForRangeStmtClass: return true; default: // Note that we intentionally do not include do..while here. return false; } } static bool isJumpToFalseBranch(const BlockEdge *BE) { const CFGBlock *Src = BE->getSrc(); assert(Src->succ_size() == 2); return (*(Src->succ_begin()+1) == BE->getDst()); } static bool isContainedByStmt(ParentMap &PM, const Stmt *S, const Stmt *SubS) { while (SubS) { if (SubS == S) return true; SubS = PM.getParent(SubS); } return false; } static const Stmt *getStmtBeforeCond(ParentMap &PM, const Stmt *Term, const ExplodedNode *N) { while (N) { Optional SP = N->getLocation().getAs(); if (SP) { const Stmt *S = SP->getStmt(); if (!isContainedByStmt(PM, Term, S)) return S; } N = N->getFirstPred(); } return nullptr; } static bool isInLoopBody(ParentMap &PM, const Stmt *S, const Stmt *Term) { const Stmt *LoopBody = nullptr; switch (Term->getStmtClass()) { case Stmt::CXXForRangeStmtClass: { const auto *FR = cast(Term); if (isContainedByStmt(PM, FR->getInc(), S)) return true; if (isContainedByStmt(PM, FR->getLoopVarStmt(), S)) return true; LoopBody = FR->getBody(); break; } case Stmt::ForStmtClass: { const auto *FS = cast(Term); if (isContainedByStmt(PM, FS->getInc(), S)) return true; LoopBody = FS->getBody(); break; } case Stmt::ObjCForCollectionStmtClass: { const auto *FC = cast(Term); LoopBody = FC->getBody(); break; } case Stmt::WhileStmtClass: LoopBody = cast(Term)->getBody(); break; default: return false; } return isContainedByStmt(PM, LoopBody, S); } /// Adds a sanitized control-flow diagnostic edge to a path. static void addEdgeToPath(PathPieces &path, PathDiagnosticLocation &PrevLoc, PathDiagnosticLocation NewLoc, const LocationContext *LC) { if (!NewLoc.isValid()) return; SourceLocation NewLocL = NewLoc.asLocation(); if (NewLocL.isInvalid()) return; if (!PrevLoc.isValid() || !PrevLoc.asLocation().isValid()) { PrevLoc = NewLoc; return; } // Ignore self-edges, which occur when there are multiple nodes at the same // statement. if (NewLoc.asStmt() && NewLoc.asStmt() == PrevLoc.asStmt()) return; path.push_front( std::make_shared(NewLoc, PrevLoc)); PrevLoc = NewLoc; } /// A customized wrapper for CFGBlock::getTerminatorCondition() /// which returns the element for ObjCForCollectionStmts. static const Stmt *getTerminatorCondition(const CFGBlock *B) { const Stmt *S = B->getTerminatorCondition(); if (const auto *FS = dyn_cast_or_null(S)) return FS->getElement(); return S; } static const char StrEnteringLoop[] = "Entering loop body"; static const char StrLoopBodyZero[] = "Loop body executed 0 times"; static const char StrLoopRangeEmpty[] = "Loop body skipped when range is empty"; static const char StrLoopCollectionEmpty[] = "Loop body skipped when collection is empty"; static std::unique_ptr findExecutedLines(SourceManager &SM, const ExplodedNode *N); /// Generate diagnostics for the node \p N, /// and write it into \p PD. /// \p AddPathEdges Whether diagnostic consumer can generate path arrows /// showing both row and column. static void generatePathDiagnosticsForNode(const ExplodedNode *N, PathDiagnostic &PD, PathDiagnosticLocation &PrevLoc, PathDiagnosticBuilder &PDB, LocationContextMap &LCM, StackDiagVector &CallStack, InterestingExprs &IE, bool AddPathEdges) { ProgramPoint P = N->getLocation(); const SourceManager& SM = PDB.getSourceManager(); // Have we encountered an entrance to a call? It may be // the case that we have not encountered a matching // call exit before this point. This means that the path // terminated within the call itself. if (auto CE = P.getAs()) { if (AddPathEdges) { // Add an edge to the start of the function. const StackFrameContext *CalleeLC = CE->getCalleeContext(); const Decl *D = CalleeLC->getDecl(); // Add the edge only when the callee has body. We jump to the beginning // of the *declaration*, however we expect it to be followed by the // body. This isn't the case for autosynthesized property accessors in // Objective-C. No need for a similar extra check for CallExit points // because the exit edge comes from a statement (i.e. return), // not from declaration. if (D->hasBody()) addEdgeToPath(PD.getActivePath(), PrevLoc, PathDiagnosticLocation::createBegin(D, SM), CalleeLC); } // Did we visit an entire call? bool VisitedEntireCall = PD.isWithinCall(); PD.popActivePath(); PathDiagnosticCallPiece *C; if (VisitedEntireCall) { C = cast(PD.getActivePath().front().get()); } else { const Decl *Caller = CE->getLocationContext()->getDecl(); C = PathDiagnosticCallPiece::construct(PD.getActivePath(), Caller); if (AddPathEdges) { // Since we just transferred the path over to the call piece, // reset the mapping from active to location context. assert(PD.getActivePath().size() == 1 && PD.getActivePath().front().get() == C); LCM[&PD.getActivePath()] = nullptr; } // Record the location context mapping for the path within // the call. assert(LCM[&C->path] == nullptr || LCM[&C->path] == CE->getCalleeContext()); LCM[&C->path] = CE->getCalleeContext(); // If this is the first item in the active path, record // the new mapping from active path to location context. const LocationContext *&NewLC = LCM[&PD.getActivePath()]; if (!NewLC) NewLC = N->getLocationContext(); PDB.LC = NewLC; } C->setCallee(*CE, SM); // Update the previous location in the active path. PrevLoc = C->getLocation(); if (!CallStack.empty()) { assert(CallStack.back().first == C); CallStack.pop_back(); } return; } if (AddPathEdges) { // Query the location context here and the previous location // as processing CallEnter may change the active path. PDB.LC = N->getLocationContext(); // Record the mapping from the active path to the location // context. assert(!LCM[&PD.getActivePath()] || LCM[&PD.getActivePath()] == PDB.LC); LCM[&PD.getActivePath()] = PDB.LC; } // Have we encountered an exit from a function call? if (Optional CE = P.getAs()) { // We are descending into a call (backwards). Construct // a new call piece to contain the path pieces for that call. auto C = PathDiagnosticCallPiece::construct(N, *CE, SM); // Record the mapping from call piece to LocationContext. LCM[&C->path] = CE->getCalleeContext(); if (AddPathEdges) { const Stmt *S = CE->getCalleeContext()->getCallSite(); // Propagate the interesting symbols accordingly. if (const auto *Ex = dyn_cast_or_null(S)) { reversePropagateIntererstingSymbols(*PDB.getBugReport(), IE, N->getState().get(), Ex, N->getLocationContext()); } // Add the edge to the return site. addEdgeToPath(PD.getActivePath(), PrevLoc, C->callReturn, PDB.LC); PrevLoc.invalidate(); } auto *P = C.get(); PD.getActivePath().push_front(std::move(C)); // Make the contents of the call the active path for now. PD.pushActivePath(&P->path); CallStack.push_back(StackDiagPair(P, N)); return; } if (auto PS = P.getAs()) { if (!AddPathEdges) return; // For expressions, make sure we propagate the // interesting symbols correctly. if (const Expr *Ex = PS->getStmtAs()) reversePropagateIntererstingSymbols(*PDB.getBugReport(), IE, N->getState().get(), Ex, N->getLocationContext()); // Add an edge. If this is an ObjCForCollectionStmt do // not add an edge here as it appears in the CFG both // as a terminator and as a terminator condition. if (!isa(PS->getStmt())) { PathDiagnosticLocation L = PathDiagnosticLocation(PS->getStmt(), SM, PDB.LC); addEdgeToPath(PD.getActivePath(), PrevLoc, L, PDB.LC); } } else if (auto BE = P.getAs()) { if (!AddPathEdges) { generateMinimalDiagForBlockEdge(N, *BE, SM, PDB, PD); return; } // Does this represent entering a call? If so, look at propagating // interesting symbols across call boundaries. if (const ExplodedNode *NextNode = N->getFirstPred()) { const LocationContext *CallerCtx = NextNode->getLocationContext(); const LocationContext *CalleeCtx = PDB.LC; if (CallerCtx != CalleeCtx && AddPathEdges) { reversePropagateInterestingSymbols(*PDB.getBugReport(), IE, N->getState().get(), CalleeCtx, CallerCtx); } } // Are we jumping to the head of a loop? Add a special diagnostic. if (const Stmt *Loop = BE->getSrc()->getLoopTarget()) { PathDiagnosticLocation L(Loop, SM, PDB.LC); const Stmt *Body = nullptr; if (const auto *FS = dyn_cast(Loop)) Body = FS->getBody(); else if (const auto *WS = dyn_cast(Loop)) Body = WS->getBody(); else if (const auto *OFS = dyn_cast(Loop)) { Body = OFS->getBody(); } else if (const auto *FRS = dyn_cast(Loop)) { Body = FRS->getBody(); } // do-while statements are explicitly excluded here auto p = std::make_shared( L, "Looping back to the head " "of the loop"); p->setPrunable(true); addEdgeToPath(PD.getActivePath(), PrevLoc, p->getLocation(), PDB.LC); PD.getActivePath().push_front(std::move(p)); if (const auto *CS = dyn_cast_or_null(Body)) { addEdgeToPath(PD.getActivePath(), PrevLoc, PathDiagnosticLocation::createEndBrace(CS, SM), PDB.LC); } } const CFGBlock *BSrc = BE->getSrc(); ParentMap &PM = PDB.getParentMap(); if (const Stmt *Term = BSrc->getTerminator()) { // Are we jumping past the loop body without ever executing the // loop (because the condition was false)? if (isLoop(Term)) { const Stmt *TermCond = getTerminatorCondition(BSrc); bool IsInLoopBody = isInLoopBody(PM, getStmtBeforeCond(PM, TermCond, N), Term); const char *str = nullptr; if (isJumpToFalseBranch(&*BE)) { if (!IsInLoopBody) { if (isa(Term)) { str = StrLoopCollectionEmpty; } else if (isa(Term)) { str = StrLoopRangeEmpty; } else { str = StrLoopBodyZero; } } } else { str = StrEnteringLoop; } if (str) { PathDiagnosticLocation L(TermCond ? TermCond : Term, SM, PDB.LC); auto PE = std::make_shared(L, str); PE->setPrunable(true); addEdgeToPath(PD.getActivePath(), PrevLoc, PE->getLocation(), PDB.LC); PD.getActivePath().push_front(std::move(PE)); } } else if (isa(Term) || isa(Term) || isa(Term)) { PathDiagnosticLocation L(Term, SM, PDB.LC); addEdgeToPath(PD.getActivePath(), PrevLoc, L, PDB.LC); } } } } static std::unique_ptr generateEmptyDiagnosticForReport(BugReport *R, SourceManager &SM) { BugType &BT = R->getBugType(); return llvm::make_unique( R->getBugType().getCheckName(), R->getDeclWithIssue(), R->getBugType().getName(), R->getDescription(), R->getShortDescription(/*Fallback=*/false), BT.getCategory(), R->getUniqueingLocation(), R->getUniqueingDecl(), findExecutedLines(SM, R->getErrorNode())); } static const Stmt *getStmtParent(const Stmt *S, const ParentMap &PM) { if (!S) return nullptr; while (true) { S = PM.getParentIgnoreParens(S); if (!S) break; if (isa(S) || isa(S) || isa(S)) continue; break; } return S; } static bool isConditionForTerminator(const Stmt *S, const Stmt *Cond) { switch (S->getStmtClass()) { case Stmt::BinaryOperatorClass: { const auto *BO = cast(S); if (!BO->isLogicalOp()) return false; return BO->getLHS() == Cond || BO->getRHS() == Cond; } case Stmt::IfStmtClass: return cast(S)->getCond() == Cond; case Stmt::ForStmtClass: return cast(S)->getCond() == Cond; case Stmt::WhileStmtClass: return cast(S)->getCond() == Cond; case Stmt::DoStmtClass: return cast(S)->getCond() == Cond; case Stmt::ChooseExprClass: return cast(S)->getCond() == Cond; case Stmt::IndirectGotoStmtClass: return cast(S)->getTarget() == Cond; case Stmt::SwitchStmtClass: return cast(S)->getCond() == Cond; case Stmt::BinaryConditionalOperatorClass: return cast(S)->getCond() == Cond; case Stmt::ConditionalOperatorClass: { const auto *CO = cast(S); return CO->getCond() == Cond || CO->getLHS() == Cond || CO->getRHS() == Cond; } case Stmt::ObjCForCollectionStmtClass: return cast(S)->getElement() == Cond; case Stmt::CXXForRangeStmtClass: { const auto *FRS = cast(S); return FRS->getCond() == Cond || FRS->getRangeInit() == Cond; } default: return false; } } static bool isIncrementOrInitInForLoop(const Stmt *S, const Stmt *FL) { if (const auto *FS = dyn_cast(FL)) return FS->getInc() == S || FS->getInit() == S; if (const auto *FRS = dyn_cast(FL)) return FRS->getInc() == S || FRS->getRangeStmt() == S || FRS->getLoopVarStmt() || FRS->getRangeInit() == S; return false; } using OptimizedCallsSet = llvm::DenseSet; /// Adds synthetic edges from top-level statements to their subexpressions. /// /// This avoids a "swoosh" effect, where an edge from a top-level statement A /// points to a sub-expression B.1 that's not at the start of B. In these cases, /// we'd like to see an edge from A to B, then another one from B to B.1. static void addContextEdges(PathPieces &pieces, SourceManager &SM, const ParentMap &PM, const LocationContext *LCtx) { PathPieces::iterator Prev = pieces.end(); for (PathPieces::iterator I = pieces.begin(), E = Prev; I != E; Prev = I, ++I) { auto *Piece = dyn_cast(I->get()); if (!Piece) continue; PathDiagnosticLocation SrcLoc = Piece->getStartLocation(); SmallVector SrcContexts; PathDiagnosticLocation NextSrcContext = SrcLoc; const Stmt *InnerStmt = nullptr; while (NextSrcContext.isValid() && NextSrcContext.asStmt() != InnerStmt) { SrcContexts.push_back(NextSrcContext); InnerStmt = NextSrcContext.asStmt(); NextSrcContext = getEnclosingStmtLocation(InnerStmt, SM, PM, LCtx, /*allowNested=*/true); } // Repeatedly split the edge as necessary. // This is important for nested logical expressions (||, &&, ?:) where we // want to show all the levels of context. while (true) { const Stmt *Dst = Piece->getEndLocation().getStmtOrNull(); // We are looking at an edge. Is the destination within a larger // expression? PathDiagnosticLocation DstContext = getEnclosingStmtLocation(Dst, SM, PM, LCtx, /*allowNested=*/true); if (!DstContext.isValid() || DstContext.asStmt() == Dst) break; // If the source is in the same context, we're already good. if (std::find(SrcContexts.begin(), SrcContexts.end(), DstContext) != SrcContexts.end()) break; // Update the subexpression node to point to the context edge. Piece->setStartLocation(DstContext); // Try to extend the previous edge if it's at the same level as the source // context. if (Prev != E) { auto *PrevPiece = dyn_cast(Prev->get()); if (PrevPiece) { if (const Stmt *PrevSrc = PrevPiece->getStartLocation().getStmtOrNull()) { const Stmt *PrevSrcParent = getStmtParent(PrevSrc, PM); if (PrevSrcParent == getStmtParent(DstContext.getStmtOrNull(), PM)) { PrevPiece->setEndLocation(DstContext); break; } } } } // Otherwise, split the current edge into a context edge and a // subexpression edge. Note that the context statement may itself have // context. auto P = std::make_shared(SrcLoc, DstContext); Piece = P.get(); I = pieces.insert(I, std::move(P)); } } } /// Move edges from a branch condition to a branch target /// when the condition is simple. /// /// This restructures some of the work of addContextEdges. That function /// creates edges this may destroy, but they work together to create a more /// aesthetically set of edges around branches. After the call to /// addContextEdges, we may have (1) an edge to the branch, (2) an edge from /// the branch to the branch condition, and (3) an edge from the branch /// condition to the branch target. We keep (1), but may wish to remove (2) /// and move the source of (3) to the branch if the branch condition is simple. static void simplifySimpleBranches(PathPieces &pieces) { for (PathPieces::iterator I = pieces.begin(), E = pieces.end(); I != E; ++I) { const auto *PieceI = dyn_cast(I->get()); if (!PieceI) continue; const Stmt *s1Start = PieceI->getStartLocation().getStmtOrNull(); const Stmt *s1End = PieceI->getEndLocation().getStmtOrNull(); if (!s1Start || !s1End) continue; PathPieces::iterator NextI = I; ++NextI; if (NextI == E) break; PathDiagnosticControlFlowPiece *PieceNextI = nullptr; while (true) { if (NextI == E) break; const auto *EV = dyn_cast(NextI->get()); if (EV) { StringRef S = EV->getString(); if (S == StrEnteringLoop || S == StrLoopBodyZero || S == StrLoopCollectionEmpty || S == StrLoopRangeEmpty) { ++NextI; continue; } break; } PieceNextI = dyn_cast(NextI->get()); break; } if (!PieceNextI) continue; const Stmt *s2Start = PieceNextI->getStartLocation().getStmtOrNull(); const Stmt *s2End = PieceNextI->getEndLocation().getStmtOrNull(); if (!s2Start || !s2End || s1End != s2Start) continue; // We only perform this transformation for specific branch kinds. // We don't want to do this for do..while, for example. if (!(isa(s1Start) || isa(s1Start) || isa(s1Start) || isa(s1Start) || isa(s1Start))) continue; // Is s1End the branch condition? if (!isConditionForTerminator(s1Start, s1End)) continue; // Perform the hoisting by eliminating (2) and changing the start // location of (3). PieceNextI->setStartLocation(PieceI->getStartLocation()); I = pieces.erase(I); } } /// Returns the number of bytes in the given (character-based) SourceRange. /// /// If the locations in the range are not on the same line, returns None. /// /// Note that this does not do a precise user-visible character or column count. static Optional getLengthOnSingleLine(SourceManager &SM, SourceRange Range) { SourceRange ExpansionRange(SM.getExpansionLoc(Range.getBegin()), SM.getExpansionRange(Range.getEnd()).getEnd()); FileID FID = SM.getFileID(ExpansionRange.getBegin()); if (FID != SM.getFileID(ExpansionRange.getEnd())) return None; bool Invalid; const llvm::MemoryBuffer *Buffer = SM.getBuffer(FID, &Invalid); if (Invalid) return None; unsigned BeginOffset = SM.getFileOffset(ExpansionRange.getBegin()); unsigned EndOffset = SM.getFileOffset(ExpansionRange.getEnd()); StringRef Snippet = Buffer->getBuffer().slice(BeginOffset, EndOffset); // We're searching the raw bytes of the buffer here, which might include // escaped newlines and such. That's okay; we're trying to decide whether the // SourceRange is covering a large or small amount of space in the user's // editor. if (Snippet.find_first_of("\r\n") != StringRef::npos) return None; // This isn't Unicode-aware, but it doesn't need to be. return Snippet.size(); } /// \sa getLengthOnSingleLine(SourceManager, SourceRange) static Optional getLengthOnSingleLine(SourceManager &SM, const Stmt *S) { return getLengthOnSingleLine(SM, S->getSourceRange()); } /// Eliminate two-edge cycles created by addContextEdges(). /// /// Once all the context edges are in place, there are plenty of cases where /// there's a single edge from a top-level statement to a subexpression, /// followed by a single path note, and then a reverse edge to get back out to /// the top level. If the statement is simple enough, the subexpression edges /// just add noise and make it harder to understand what's going on. /// /// This function only removes edges in pairs, because removing only one edge /// might leave other edges dangling. /// /// This will not remove edges in more complicated situations: /// - if there is more than one "hop" leading to or from a subexpression. /// - if there is an inlined call between the edges instead of a single event. /// - if the whole statement is large enough that having subexpression arrows /// might be helpful. static void removeContextCycles(PathPieces &Path, SourceManager &SM, ParentMap &PM) { for (PathPieces::iterator I = Path.begin(), E = Path.end(); I != E; ) { // Pattern match the current piece and its successor. const auto *PieceI = dyn_cast(I->get()); if (!PieceI) { ++I; continue; } const Stmt *s1Start = PieceI->getStartLocation().getStmtOrNull(); const Stmt *s1End = PieceI->getEndLocation().getStmtOrNull(); PathPieces::iterator NextI = I; ++NextI; if (NextI == E) break; const auto *PieceNextI = dyn_cast(NextI->get()); if (!PieceNextI) { if (isa(NextI->get())) { ++NextI; if (NextI == E) break; PieceNextI = dyn_cast(NextI->get()); } if (!PieceNextI) { ++I; continue; } } const Stmt *s2Start = PieceNextI->getStartLocation().getStmtOrNull(); const Stmt *s2End = PieceNextI->getEndLocation().getStmtOrNull(); if (s1Start && s2Start && s1Start == s2End && s2Start == s1End) { const size_t MAX_SHORT_LINE_LENGTH = 80; Optional s1Length = getLengthOnSingleLine(SM, s1Start); if (s1Length && *s1Length <= MAX_SHORT_LINE_LENGTH) { Optional s2Length = getLengthOnSingleLine(SM, s2Start); if (s2Length && *s2Length <= MAX_SHORT_LINE_LENGTH) { Path.erase(I); I = Path.erase(NextI); continue; } } } ++I; } } /// Return true if X is contained by Y. static bool lexicalContains(ParentMap &PM, const Stmt *X, const Stmt *Y) { while (X) { if (X == Y) return true; X = PM.getParent(X); } return false; } // Remove short edges on the same line less than 3 columns in difference. static void removePunyEdges(PathPieces &path, SourceManager &SM, ParentMap &PM) { bool erased = false; for (PathPieces::iterator I = path.begin(), E = path.end(); I != E; erased ? I : ++I) { erased = false; const auto *PieceI = dyn_cast(I->get()); if (!PieceI) continue; const Stmt *start = PieceI->getStartLocation().getStmtOrNull(); const Stmt *end = PieceI->getEndLocation().getStmtOrNull(); if (!start || !end) continue; const Stmt *endParent = PM.getParent(end); if (!endParent) continue; if (isConditionForTerminator(end, endParent)) continue; SourceLocation FirstLoc = start->getLocStart(); SourceLocation SecondLoc = end->getLocStart(); if (!SM.isWrittenInSameFile(FirstLoc, SecondLoc)) continue; if (SM.isBeforeInTranslationUnit(SecondLoc, FirstLoc)) std::swap(SecondLoc, FirstLoc); SourceRange EdgeRange(FirstLoc, SecondLoc); Optional ByteWidth = getLengthOnSingleLine(SM, EdgeRange); // If the statements are on different lines, continue. if (!ByteWidth) continue; const size_t MAX_PUNY_EDGE_LENGTH = 2; if (*ByteWidth <= MAX_PUNY_EDGE_LENGTH) { // FIXME: There are enough /bytes/ between the endpoints of the edge, but // there might not be enough /columns/. A proper user-visible column count // is probably too expensive, though. I = path.erase(I); erased = true; continue; } } } static void removeIdenticalEvents(PathPieces &path) { for (PathPieces::iterator I = path.begin(), E = path.end(); I != E; ++I) { const auto *PieceI = dyn_cast(I->get()); if (!PieceI) continue; PathPieces::iterator NextI = I; ++NextI; if (NextI == E) return; const auto *PieceNextI = dyn_cast(NextI->get()); if (!PieceNextI) continue; // Erase the second piece if it has the same exact message text. if (PieceI->getString() == PieceNextI->getString()) { path.erase(NextI); } } } static bool optimizeEdges(PathPieces &path, SourceManager &SM, OptimizedCallsSet &OCS, LocationContextMap &LCM) { bool hasChanges = false; const LocationContext *LC = LCM[&path]; assert(LC); ParentMap &PM = LC->getParentMap(); for (PathPieces::iterator I = path.begin(), E = path.end(); I != E; ) { // Optimize subpaths. if (auto *CallI = dyn_cast(I->get())) { // Record the fact that a call has been optimized so we only do the // effort once. if (!OCS.count(CallI)) { while (optimizeEdges(CallI->path, SM, OCS, LCM)) {} OCS.insert(CallI); } ++I; continue; } // Pattern match the current piece and its successor. auto *PieceI = dyn_cast(I->get()); if (!PieceI) { ++I; continue; } const Stmt *s1Start = PieceI->getStartLocation().getStmtOrNull(); const Stmt *s1End = PieceI->getEndLocation().getStmtOrNull(); const Stmt *level1 = getStmtParent(s1Start, PM); const Stmt *level2 = getStmtParent(s1End, PM); PathPieces::iterator NextI = I; ++NextI; if (NextI == E) break; const auto *PieceNextI = dyn_cast(NextI->get()); if (!PieceNextI) { ++I; continue; } const Stmt *s2Start = PieceNextI->getStartLocation().getStmtOrNull(); const Stmt *s2End = PieceNextI->getEndLocation().getStmtOrNull(); const Stmt *level3 = getStmtParent(s2Start, PM); const Stmt *level4 = getStmtParent(s2End, PM); // Rule I. // // If we have two consecutive control edges whose end/begin locations // are at the same level (e.g. statements or top-level expressions within // a compound statement, or siblings share a single ancestor expression), // then merge them if they have no interesting intermediate event. // // For example: // // (1.1 -> 1.2) -> (1.2 -> 1.3) becomes (1.1 -> 1.3) because the common // parent is '1'. Here 'x.y.z' represents the hierarchy of statements. // // NOTE: this will be limited later in cases where we add barriers // to prevent this optimization. if (level1 && level1 == level2 && level1 == level3 && level1 == level4) { PieceI->setEndLocation(PieceNextI->getEndLocation()); path.erase(NextI); hasChanges = true; continue; } // Rule II. // // Eliminate edges between subexpressions and parent expressions // when the subexpression is consumed. // // NOTE: this will be limited later in cases where we add barriers // to prevent this optimization. if (s1End && s1End == s2Start && level2) { bool removeEdge = false; // Remove edges into the increment or initialization of a // loop that have no interleaving event. This means that // they aren't interesting. if (isIncrementOrInitInForLoop(s1End, level2)) removeEdge = true; // Next only consider edges that are not anchored on // the condition of a terminator. This are intermediate edges // that we might want to trim. else if (!isConditionForTerminator(level2, s1End)) { // Trim edges on expressions that are consumed by // the parent expression. if (isa(s1End) && PM.isConsumedExpr(cast(s1End))) { removeEdge = true; } // Trim edges where a lexical containment doesn't exist. // For example: // // X -> Y -> Z // // If 'Z' lexically contains Y (it is an ancestor) and // 'X' does not lexically contain Y (it is a descendant OR // it has no lexical relationship at all) then trim. // // This can eliminate edges where we dive into a subexpression // and then pop back out, etc. else if (s1Start && s2End && lexicalContains(PM, s2Start, s2End) && !lexicalContains(PM, s1End, s1Start)) { removeEdge = true; } // Trim edges from a subexpression back to the top level if the // subexpression is on a different line. // // A.1 -> A -> B // becomes // A.1 -> B // // These edges just look ugly and don't usually add anything. else if (s1Start && s2End && lexicalContains(PM, s1Start, s1End)) { SourceRange EdgeRange(PieceI->getEndLocation().asLocation(), PieceI->getStartLocation().asLocation()); if (!getLengthOnSingleLine(SM, EdgeRange).hasValue()) removeEdge = true; } } if (removeEdge) { PieceI->setEndLocation(PieceNextI->getEndLocation()); path.erase(NextI); hasChanges = true; continue; } } // Optimize edges for ObjC fast-enumeration loops. // // (X -> collection) -> (collection -> element) // // becomes: // // (X -> element) if (s1End == s2Start) { const auto *FS = dyn_cast_or_null(level3); if (FS && FS->getCollection()->IgnoreParens() == s2Start && s2End == FS->getElement()) { PieceI->setEndLocation(PieceNextI->getEndLocation()); path.erase(NextI); hasChanges = true; continue; } } // No changes at this index? Move to the next one. ++I; } if (!hasChanges) { // Adjust edges into subexpressions to make them more uniform // and aesthetically pleasing. addContextEdges(path, SM, PM, LC); // Remove "cyclical" edges that include one or more context edges. removeContextCycles(path, SM, PM); // Hoist edges originating from branch conditions to branches // for simple branches. simplifySimpleBranches(path); // Remove any puny edges left over after primary optimization pass. removePunyEdges(path, SM, PM); // Remove identical events. removeIdenticalEvents(path); } return hasChanges; } /// Drop the very first edge in a path, which should be a function entry edge. /// /// If the first edge is not a function entry edge (say, because the first /// statement had an invalid source location), this function does nothing. // FIXME: We should just generate invalid edges anyway and have the optimizer // deal with them. static void dropFunctionEntryEdge(PathPieces &Path, LocationContextMap &LCM, SourceManager &SM) { const auto *FirstEdge = dyn_cast(Path.front().get()); if (!FirstEdge) return; const Decl *D = LCM[&Path]->getDecl(); PathDiagnosticLocation EntryLoc = PathDiagnosticLocation::createBegin(D, SM); if (FirstEdge->getStartLocation() != EntryLoc) return; Path.pop_front(); } using VisitorsDiagnosticsTy = llvm::DenseMap>>; /// This function is responsible for generating diagnostic pieces that are /// *not* provided by bug report visitors. /// These diagnostics may differ depending on the consumer's settings, /// and are therefore constructed separately for each consumer. /// /// There are two path diagnostics generation modes: with adding edges (used /// for plists) and without (used for HTML and text). /// When edges are added (\p ActiveScheme is Extensive), /// the path is modified to insert artificially generated /// edges. /// Otherwise, more detailed diagnostics is emitted for block edges, explaining /// the transitions in words. static std::unique_ptr generatePathDiagnosticForConsumer( PathDiagnosticConsumer::PathGenerationScheme ActiveScheme, PathDiagnosticBuilder &PDB, const ExplodedNode *ErrorNode, const VisitorsDiagnosticsTy &VisitorsDiagnostics) { bool GenerateDiagnostics = (ActiveScheme != PathDiagnosticConsumer::None); bool AddPathEdges = (ActiveScheme == PathDiagnosticConsumer::Extensive); SourceManager &SM = PDB.getSourceManager(); BugReport *R = PDB.getBugReport(); AnalyzerOptions &Opts = PDB.getBugReporter().getAnalyzerOptions(); StackDiagVector CallStack; InterestingExprs IE; LocationContextMap LCM; std::unique_ptr PD = generateEmptyDiagnosticForReport(R, SM); if (GenerateDiagnostics) { auto EndNotes = VisitorsDiagnostics.find(ErrorNode); std::shared_ptr LastPiece; if (EndNotes != VisitorsDiagnostics.end()) { assert(!EndNotes->second.empty()); LastPiece = EndNotes->second[0]; } else { LastPiece = BugReporterVisitor::getDefaultEndPath(PDB, ErrorNode, *R); } PD->setEndOfPath(LastPiece); } PathDiagnosticLocation PrevLoc = PD->getLocation(); const ExplodedNode *NextNode = ErrorNode->getFirstPred(); while (NextNode) { if (GenerateDiagnostics) generatePathDiagnosticsForNode( NextNode, *PD, PrevLoc, PDB, LCM, CallStack, IE, AddPathEdges); auto VisitorNotes = VisitorsDiagnostics.find(NextNode); NextNode = NextNode->getFirstPred(); if (!GenerateDiagnostics || VisitorNotes == VisitorsDiagnostics.end()) continue; // This is a workaround due to inability to put shared PathDiagnosticPiece // into a FoldingSet. std::set DeduplicationSet; // Add pieces from custom visitors. for (const auto &Note : VisitorNotes->second) { llvm::FoldingSetNodeID ID; Note->Profile(ID); auto P = DeduplicationSet.insert(ID); if (!P.second) continue; if (AddPathEdges) addEdgeToPath(PD->getActivePath(), PrevLoc, Note->getLocation(), PDB.LC); updateStackPiecesWithMessage(*Note, CallStack); PD->getActivePath().push_front(Note); } } if (AddPathEdges) { // Add an edge to the start of the function. // We'll prune it out later, but it helps make diagnostics more uniform. const StackFrameContext *CalleeLC = PDB.LC->getStackFrame(); const Decl *D = CalleeLC->getDecl(); addEdgeToPath(PD->getActivePath(), PrevLoc, PathDiagnosticLocation::createBegin(D, SM), CalleeLC); } if (!AddPathEdges && GenerateDiagnostics) CompactPathDiagnostic(PD->getMutablePieces(), SM); // Finally, prune the diagnostic path of uninteresting stuff. if (!PD->path.empty()) { if (R->shouldPrunePath() && Opts.shouldPrunePaths()) { bool stillHasNotes = removeUnneededCalls(PD->getMutablePieces(), R, LCM); assert(stillHasNotes); (void)stillHasNotes; } // Redirect all call pieces to have valid locations. adjustCallLocations(PD->getMutablePieces()); removePiecesWithInvalidLocations(PD->getMutablePieces()); if (AddPathEdges) { // Reduce the number of edges from a very conservative set // to an aesthetically pleasing subset that conveys the // necessary information. OptimizedCallsSet OCS; while (optimizeEdges(PD->getMutablePieces(), SM, OCS, LCM)) {} // Drop the very first function-entry edge. It's not really necessary // for top-level functions. dropFunctionEntryEdge(PD->getMutablePieces(), LCM, SM); } // Remove messages that are basically the same, and edges that may not // make sense. // We have to do this after edge optimization in the Extensive mode. removeRedundantMsgs(PD->getMutablePieces()); removeEdgesToDefaultInitializers(PD->getMutablePieces()); } return PD; } //===----------------------------------------------------------------------===// // Methods for BugType and subclasses. //===----------------------------------------------------------------------===// void BugType::anchor() {} void BugType::FlushReports(BugReporter &BR) {} void BuiltinBug::anchor() {} //===----------------------------------------------------------------------===// // Methods for BugReport and subclasses. //===----------------------------------------------------------------------===// void BugReport::NodeResolver::anchor() {} void BugReport::addVisitor(std::unique_ptr visitor) { if (!visitor) return; llvm::FoldingSetNodeID ID; visitor->Profile(ID); void *InsertPos = nullptr; if (CallbacksSet.FindNodeOrInsertPos(ID, InsertPos)) { return; } Callbacks.push_back(std::move(visitor)); } void BugReport::clearVisitors() { Callbacks.clear(); } BugReport::~BugReport() { while (!interestingSymbols.empty()) { popInterestingSymbolsAndRegions(); } } const Decl *BugReport::getDeclWithIssue() const { if (DeclWithIssue) return DeclWithIssue; const ExplodedNode *N = getErrorNode(); if (!N) return nullptr; const LocationContext *LC = N->getLocationContext(); return LC->getStackFrame()->getDecl(); } void BugReport::Profile(llvm::FoldingSetNodeID& hash) const { hash.AddPointer(&BT); hash.AddString(Description); PathDiagnosticLocation UL = getUniqueingLocation(); if (UL.isValid()) { UL.Profile(hash); } else if (Location.isValid()) { Location.Profile(hash); } else { assert(ErrorNode); hash.AddPointer(GetCurrentOrPreviousStmt(ErrorNode)); } for (SourceRange range : Ranges) { if (!range.isValid()) continue; hash.AddInteger(range.getBegin().getRawEncoding()); hash.AddInteger(range.getEnd().getRawEncoding()); } } void BugReport::markInteresting(SymbolRef sym) { if (!sym) return; getInterestingSymbols().insert(sym); if (const auto *meta = dyn_cast(sym)) getInterestingRegions().insert(meta->getRegion()); } void BugReport::markInteresting(const MemRegion *R) { if (!R) return; R = R->getBaseRegion(); getInterestingRegions().insert(R); if (const auto *SR = dyn_cast(R)) getInterestingSymbols().insert(SR->getSymbol()); } void BugReport::markInteresting(SVal V) { markInteresting(V.getAsRegion()); markInteresting(V.getAsSymbol()); } void BugReport::markInteresting(const LocationContext *LC) { if (!LC) return; InterestingLocationContexts.insert(LC); } bool BugReport::isInteresting(SVal V) { return isInteresting(V.getAsRegion()) || isInteresting(V.getAsSymbol()); } bool BugReport::isInteresting(SymbolRef sym) { if (!sym) return false; // We don't currently consider metadata symbols to be interesting // even if we know their region is interesting. Is that correct behavior? return getInterestingSymbols().count(sym); } bool BugReport::isInteresting(const MemRegion *R) { if (!R) return false; R = R->getBaseRegion(); bool b = getInterestingRegions().count(R); if (b) return true; if (const auto *SR = dyn_cast(R)) return getInterestingSymbols().count(SR->getSymbol()); return false; } bool BugReport::isInteresting(const LocationContext *LC) { if (!LC) return false; return InterestingLocationContexts.count(LC); } void BugReport::lazyInitializeInterestingSets() { if (interestingSymbols.empty()) { interestingSymbols.push_back(new Symbols()); interestingRegions.push_back(new Regions()); } } BugReport::Symbols &BugReport::getInterestingSymbols() { lazyInitializeInterestingSets(); return *interestingSymbols.back(); } BugReport::Regions &BugReport::getInterestingRegions() { lazyInitializeInterestingSets(); return *interestingRegions.back(); } void BugReport::pushInterestingSymbolsAndRegions() { interestingSymbols.push_back(new Symbols(getInterestingSymbols())); interestingRegions.push_back(new Regions(getInterestingRegions())); } void BugReport::popInterestingSymbolsAndRegions() { delete interestingSymbols.pop_back_val(); delete interestingRegions.pop_back_val(); } const Stmt *BugReport::getStmt() const { if (!ErrorNode) return nullptr; ProgramPoint ProgP = ErrorNode->getLocation(); const Stmt *S = nullptr; if (Optional BE = ProgP.getAs()) { CFGBlock &Exit = ProgP.getLocationContext()->getCFG()->getExit(); if (BE->getBlock() == &Exit) S = GetPreviousStmt(ErrorNode); } if (!S) S = PathDiagnosticLocation::getStmt(ErrorNode); return S; } llvm::iterator_range BugReport::getRanges() { // If no custom ranges, add the range of the statement corresponding to // the error node. if (Ranges.empty()) { if (const auto *E = dyn_cast_or_null(getStmt())) addRange(E->getSourceRange()); else return llvm::make_range(ranges_iterator(), ranges_iterator()); } // User-specified absence of range info. if (Ranges.size() == 1 && !Ranges.begin()->isValid()) return llvm::make_range(ranges_iterator(), ranges_iterator()); return llvm::make_range(Ranges.begin(), Ranges.end()); } PathDiagnosticLocation BugReport::getLocation(const SourceManager &SM) const { if (ErrorNode) { assert(!Location.isValid() && "Either Location or ErrorNode should be specified but not both."); return PathDiagnosticLocation::createEndOfPath(ErrorNode, SM); } assert(Location.isValid()); return Location; } //===----------------------------------------------------------------------===// // Methods for BugReporter and subclasses. //===----------------------------------------------------------------------===// BugReportEquivClass::~BugReportEquivClass() = default; GRBugReporter::~GRBugReporter() = default; BugReporterData::~BugReporterData() = default; ExplodedGraph &GRBugReporter::getGraph() { return Eng.getGraph(); } ProgramStateManager& GRBugReporter::getStateManager() { return Eng.getStateManager(); } BugReporter::~BugReporter() { FlushReports(); // Free the bug reports we are tracking. for (const auto I : EQClassesVector) delete I; } void BugReporter::FlushReports() { if (BugTypes.isEmpty()) return; // First flush the warnings for each BugType. This may end up creating new // warnings and new BugTypes. // FIXME: Only NSErrorChecker needs BugType's FlushReports. // Turn NSErrorChecker into a proper checker and remove this. SmallVector bugTypes(BugTypes.begin(), BugTypes.end()); for (const auto I : bugTypes) const_cast(I)->FlushReports(*this); // We need to flush reports in deterministic order to ensure the order // of the reports is consistent between runs. for (const auto EQ : EQClassesVector) FlushReport(*EQ); // BugReporter owns and deletes only BugTypes created implicitly through // EmitBasicReport. // FIXME: There are leaks from checkers that assume that the BugTypes they // create will be destroyed by the BugReporter. llvm::DeleteContainerSeconds(StrBugTypes); // Remove all references to the BugType objects. BugTypes = F.getEmptySet(); } //===----------------------------------------------------------------------===// // PathDiagnostics generation. //===----------------------------------------------------------------------===// namespace { /// A wrapper around a report graph, which contains only a single path, and its /// node maps. class ReportGraph { public: InterExplodedGraphMap BackMap; std::unique_ptr Graph; const ExplodedNode *ErrorNode; size_t Index; }; /// A wrapper around a trimmed graph and its node maps. class TrimmedGraph { InterExplodedGraphMap InverseMap; using PriorityMapTy = llvm::DenseMap; PriorityMapTy PriorityMap; using NodeIndexPair = std::pair; SmallVector ReportNodes; std::unique_ptr G; /// A helper class for sorting ExplodedNodes by priority. template class PriorityCompare { const PriorityMapTy &PriorityMap; public: PriorityCompare(const PriorityMapTy &M) : PriorityMap(M) {} bool operator()(const ExplodedNode *LHS, const ExplodedNode *RHS) const { PriorityMapTy::const_iterator LI = PriorityMap.find(LHS); PriorityMapTy::const_iterator RI = PriorityMap.find(RHS); PriorityMapTy::const_iterator E = PriorityMap.end(); if (LI == E) return Descending; if (RI == E) return !Descending; return Descending ? LI->second > RI->second : LI->second < RI->second; } bool operator()(const NodeIndexPair &LHS, const NodeIndexPair &RHS) const { return (*this)(LHS.first, RHS.first); } }; public: TrimmedGraph(const ExplodedGraph *OriginalGraph, ArrayRef Nodes); bool popNextReportGraph(ReportGraph &GraphWrapper); }; } // namespace TrimmedGraph::TrimmedGraph(const ExplodedGraph *OriginalGraph, ArrayRef Nodes) { // The trimmed graph is created in the body of the constructor to ensure // that the DenseMaps have been initialized already. InterExplodedGraphMap ForwardMap; G = OriginalGraph->trim(Nodes, &ForwardMap, &InverseMap); // Find the (first) error node in the trimmed graph. We just need to consult // the node map which maps from nodes in the original graph to nodes // in the new graph. llvm::SmallPtrSet RemainingNodes; for (unsigned i = 0, count = Nodes.size(); i < count; ++i) { if (const ExplodedNode *NewNode = ForwardMap.lookup(Nodes[i])) { ReportNodes.push_back(std::make_pair(NewNode, i)); RemainingNodes.insert(NewNode); } } assert(!RemainingNodes.empty() && "No error node found in the trimmed graph"); // Perform a forward BFS to find all the shortest paths. std::queue WS; assert(G->num_roots() == 1); WS.push(*G->roots_begin()); unsigned Priority = 0; while (!WS.empty()) { const ExplodedNode *Node = WS.front(); WS.pop(); PriorityMapTy::iterator PriorityEntry; bool IsNew; std::tie(PriorityEntry, IsNew) = PriorityMap.insert(std::make_pair(Node, Priority)); ++Priority; if (!IsNew) { assert(PriorityEntry->second <= Priority); continue; } if (RemainingNodes.erase(Node)) if (RemainingNodes.empty()) break; for (ExplodedNode::const_pred_iterator I = Node->succ_begin(), E = Node->succ_end(); I != E; ++I) WS.push(*I); } // Sort the error paths from longest to shortest. llvm::sort(ReportNodes.begin(), ReportNodes.end(), PriorityCompare(PriorityMap)); } bool TrimmedGraph::popNextReportGraph(ReportGraph &GraphWrapper) { if (ReportNodes.empty()) return false; const ExplodedNode *OrigN; std::tie(OrigN, GraphWrapper.Index) = ReportNodes.pop_back_val(); assert(PriorityMap.find(OrigN) != PriorityMap.end() && "error node not accessible from root"); // Create a new graph with a single path. This is the graph // that will be returned to the caller. auto GNew = llvm::make_unique(); GraphWrapper.BackMap.clear(); // Now walk from the error node up the BFS path, always taking the // predeccessor with the lowest number. ExplodedNode *Succ = nullptr; while (true) { // Create the equivalent node in the new graph with the same state // and location. ExplodedNode *NewN = GNew->createUncachedNode(OrigN->getLocation(), OrigN->getState(), OrigN->isSink()); // Store the mapping to the original node. InterExplodedGraphMap::const_iterator IMitr = InverseMap.find(OrigN); assert(IMitr != InverseMap.end() && "No mapping to original node."); GraphWrapper.BackMap[NewN] = IMitr->second; // Link up the new node with the previous node. if (Succ) Succ->addPredecessor(NewN, *GNew); else GraphWrapper.ErrorNode = NewN; Succ = NewN; // Are we at the final node? if (OrigN->pred_empty()) { GNew->addRoot(NewN); break; } // Find the next predeccessor node. We choose the node that is marked // with the lowest BFS number. OrigN = *std::min_element(OrigN->pred_begin(), OrigN->pred_end(), PriorityCompare(PriorityMap)); } GraphWrapper.Graph = std::move(GNew); return true; } /// CompactPathDiagnostic - This function postprocesses a PathDiagnostic object /// and collapses PathDiagosticPieces that are expanded by macros. static void CompactPathDiagnostic(PathPieces &path, const SourceManager& SM) { using MacroStackTy = std::vector< std::pair, SourceLocation>>; using PiecesTy = std::vector>; MacroStackTy MacroStack; PiecesTy Pieces; for (PathPieces::const_iterator I = path.begin(), E = path.end(); I != E; ++I) { const auto &piece = *I; // Recursively compact calls. if (auto *call = dyn_cast(&*piece)) { CompactPathDiagnostic(call->path, SM); } // Get the location of the PathDiagnosticPiece. const FullSourceLoc Loc = piece->getLocation().asLocation(); // Determine the instantiation location, which is the location we group // related PathDiagnosticPieces. SourceLocation InstantiationLoc = Loc.isMacroID() ? SM.getExpansionLoc(Loc) : SourceLocation(); if (Loc.isFileID()) { MacroStack.clear(); Pieces.push_back(piece); continue; } assert(Loc.isMacroID()); // Is the PathDiagnosticPiece within the same macro group? if (!MacroStack.empty() && InstantiationLoc == MacroStack.back().second) { MacroStack.back().first->subPieces.push_back(piece); continue; } // We aren't in the same group. Are we descending into a new macro // or are part of an old one? std::shared_ptr MacroGroup; SourceLocation ParentInstantiationLoc = InstantiationLoc.isMacroID() ? SM.getExpansionLoc(Loc) : SourceLocation(); // Walk the entire macro stack. while (!MacroStack.empty()) { if (InstantiationLoc == MacroStack.back().second) { MacroGroup = MacroStack.back().first; break; } if (ParentInstantiationLoc == MacroStack.back().second) { MacroGroup = MacroStack.back().first; break; } MacroStack.pop_back(); } if (!MacroGroup || ParentInstantiationLoc == MacroStack.back().second) { // Create a new macro group and add it to the stack. auto NewGroup = std::make_shared( PathDiagnosticLocation::createSingleLocation(piece->getLocation())); if (MacroGroup) MacroGroup->subPieces.push_back(NewGroup); else { assert(InstantiationLoc.isFileID()); Pieces.push_back(NewGroup); } MacroGroup = NewGroup; MacroStack.push_back(std::make_pair(MacroGroup, InstantiationLoc)); } // Finally, add the PathDiagnosticPiece to the group. MacroGroup->subPieces.push_back(piece); } // Now take the pieces and construct a new PathDiagnostic. path.clear(); path.insert(path.end(), Pieces.begin(), Pieces.end()); } /// Generate notes from all visitors. /// Notes associated with {@code ErrorNode} are generated using /// {@code getEndPath}, and the rest are generated with {@code VisitNode}. static std::unique_ptr generateVisitorsDiagnostics(BugReport *R, const ExplodedNode *ErrorNode, BugReporterContext &BRC) { auto Notes = llvm::make_unique(); BugReport::VisitorList visitors; // Run visitors on all nodes starting from the node *before* the last one. // The last node is reserved for notes generated with {@code getEndPath}. const ExplodedNode *NextNode = ErrorNode->getFirstPred(); while (NextNode) { // At each iteration, move all visitors from report to visitor list. for (BugReport::visitor_iterator I = R->visitor_begin(), E = R->visitor_end(); I != E; ++I) { visitors.push_back(std::move(*I)); } R->clearVisitors(); const ExplodedNode *Pred = NextNode->getFirstPred(); if (!Pred) { std::shared_ptr LastPiece; for (auto &V : visitors) { V->finalizeVisitor(BRC, ErrorNode, *R); if (auto Piece = V->getEndPath(BRC, ErrorNode, *R)) { assert(!LastPiece && "There can only be one final piece in a diagnostic."); LastPiece = std::move(Piece); (*Notes)[ErrorNode].push_back(LastPiece); } } break; } for (auto &V : visitors) { auto P = V->VisitNode(NextNode, Pred, BRC, *R); if (P) (*Notes)[NextNode].push_back(std::move(P)); } if (!R->isValid()) break; NextNode = Pred; } return Notes; } /// Find a non-invalidated report for a given equivalence class, /// and return together with a cache of visitors notes. /// If none found, return a nullptr paired with an empty cache. static std::pair> findValidReport( TrimmedGraph &TrimG, ReportGraph &ErrorGraph, ArrayRef &bugReports, AnalyzerOptions &Opts, GRBugReporter &Reporter) { while (TrimG.popNextReportGraph(ErrorGraph)) { // Find the BugReport with the original location. assert(ErrorGraph.Index < bugReports.size()); BugReport *R = bugReports[ErrorGraph.Index]; assert(R && "No original report found for sliced graph."); assert(R->isValid() && "Report selected by trimmed graph marked invalid."); const ExplodedNode *ErrorNode = ErrorGraph.ErrorNode; // Register refutation visitors first, if they mark the bug invalid no // further analysis is required R->addVisitor(llvm::make_unique()); // Register additional node visitors. R->addVisitor(llvm::make_unique()); R->addVisitor(llvm::make_unique()); R->addVisitor(llvm::make_unique()); BugReporterContext BRC(Reporter, ErrorGraph.BackMap); // Run all visitors on a given graph, once. std::unique_ptr visitorNotes = generateVisitorsDiagnostics(R, ErrorNode, BRC); if (R->isValid()) { if (Opts.shouldCrosscheckWithZ3()) { // If crosscheck is enabled, remove all visitors, add the refutation // visitor and check again R->clearVisitors(); R->addVisitor(llvm::make_unique()); // We don't overrite the notes inserted by other visitors because the // refutation manager does not add any new note to the path generateVisitorsDiagnostics(R, ErrorGraph.ErrorNode, BRC); } // Check if the bug is still valid if (R->isValid()) return std::make_pair(R, std::move(visitorNotes)); } } return std::make_pair(nullptr, llvm::make_unique()); } std::unique_ptr GRBugReporter::generatePathDiagnostics( ArrayRef consumers, ArrayRef &bugReports) { assert(!bugReports.empty()); auto Out = llvm::make_unique(); bool HasValid = false; SmallVector errorNodes; for (const auto I : bugReports) { if (I->isValid()) { HasValid = true; errorNodes.push_back(I->getErrorNode()); } else { // Keep the errorNodes list in sync with the bugReports list. errorNodes.push_back(nullptr); } } // If all the reports have been marked invalid by a previous path generation, // we're done. if (!HasValid) return Out; TrimmedGraph TrimG(&getGraph(), errorNodes); ReportGraph ErrorGraph; auto ReportInfo = findValidReport(TrimG, ErrorGraph, bugReports, getAnalyzerOptions(), *this); BugReport *R = ReportInfo.first; if (R && R->isValid()) { const ExplodedNode *ErrorNode = ErrorGraph.ErrorNode; for (PathDiagnosticConsumer *PC : consumers) { PathDiagnosticBuilder PDB(*this, R, ErrorGraph.BackMap, PC); std::unique_ptr PD = generatePathDiagnosticForConsumer( PC->getGenerationScheme(), PDB, ErrorNode, *ReportInfo.second); (*Out)[PC] = std::move(PD); } } return Out; } void BugReporter::Register(BugType *BT) { BugTypes = F.add(BugTypes, BT); } void BugReporter::emitReport(std::unique_ptr R) { if (const ExplodedNode *E = R->getErrorNode()) { // An error node must either be a sink or have a tag, otherwise // it could get reclaimed before the path diagnostic is created. assert((E->isSink() || E->getLocation().getTag()) && "Error node must either be a sink or have a tag"); const AnalysisDeclContext *DeclCtx = E->getLocationContext()->getAnalysisDeclContext(); // The source of autosynthesized body can be handcrafted AST or a model // file. The locations from handcrafted ASTs have no valid source locations // and have to be discarded. Locations from model files should be preserved // for processing and reporting. if (DeclCtx->isBodyAutosynthesized() && !DeclCtx->isBodyAutosynthesizedFromModelFile()) return; } bool ValidSourceLoc = R->getLocation(getSourceManager()).isValid(); assert(ValidSourceLoc); // If we mess up in a release build, we'd still prefer to just drop the bug // instead of trying to go on. if (!ValidSourceLoc) return; // Compute the bug report's hash to determine its equivalence class. llvm::FoldingSetNodeID ID; R->Profile(ID); // Lookup the equivance class. If there isn't one, create it. BugType& BT = R->getBugType(); Register(&BT); void *InsertPos; BugReportEquivClass* EQ = EQClasses.FindNodeOrInsertPos(ID, InsertPos); if (!EQ) { EQ = new BugReportEquivClass(std::move(R)); EQClasses.InsertNode(EQ, InsertPos); EQClassesVector.push_back(EQ); } else EQ->AddReport(std::move(R)); } //===----------------------------------------------------------------------===// // Emitting reports in equivalence classes. //===----------------------------------------------------------------------===// namespace { struct FRIEC_WLItem { const ExplodedNode *N; ExplodedNode::const_succ_iterator I, E; FRIEC_WLItem(const ExplodedNode *n) : N(n), I(N->succ_begin()), E(N->succ_end()) {} }; } // namespace static const CFGBlock *findBlockForNode(const ExplodedNode *N) { ProgramPoint P = N->getLocation(); if (auto BEP = P.getAs()) return BEP->getBlock(); // Find the node's current statement in the CFG. if (const Stmt *S = PathDiagnosticLocation::getStmt(N)) return N->getLocationContext()->getAnalysisDeclContext() ->getCFGStmtMap()->getBlock(S); return nullptr; } // Returns true if by simply looking at the block, we can be sure that it // results in a sink during analysis. This is useful to know when the analysis // was interrupted, and we try to figure out if it would sink eventually. // There may be many more reasons why a sink would appear during analysis // (eg. checkers may generate sinks arbitrarily), but here we only consider // sinks that would be obvious by looking at the CFG. static bool isImmediateSinkBlock(const CFGBlock *Blk) { if (Blk->hasNoReturnElement()) return true; // FIXME: Throw-expressions are currently generating sinks during analysis: // they're not supported yet, and also often used for actually terminating // the program. So we should treat them as sinks in this analysis as well, // at least for now, but once we have better support for exceptions, // we'd need to carefully handle the case when the throw is being // immediately caught. if (std::any_of(Blk->begin(), Blk->end(), [](const CFGElement &Elm) { if (Optional StmtElm = Elm.getAs()) if (isa(StmtElm->getStmt())) return true; return false; })) return true; return false; } // Returns true if by looking at the CFG surrounding the node's program // point, we can be sure that any analysis starting from this point would // eventually end with a sink. We scan the child CFG blocks in a depth-first // manner and see if all paths eventually end up in an immediate sink block. static bool isInevitablySinking(const ExplodedNode *N) { const CFG &Cfg = N->getCFG(); const CFGBlock *StartBlk = findBlockForNode(N); if (!StartBlk) return false; if (isImmediateSinkBlock(StartBlk)) return true; llvm::SmallVector DFSWorkList; llvm::SmallPtrSet Visited; DFSWorkList.push_back(StartBlk); while (!DFSWorkList.empty()) { const CFGBlock *Blk = DFSWorkList.back(); DFSWorkList.pop_back(); Visited.insert(Blk); for (const auto &Succ : Blk->succs()) { if (const CFGBlock *SuccBlk = Succ.getReachableBlock()) { if (SuccBlk == &Cfg.getExit()) { // If at least one path reaches the CFG exit, it means that control is // returned to the caller. For now, say that we are not sure what // happens next. If necessary, this can be improved to analyze // the parent StackFrameContext's call site in a similar manner. return false; } if (!isImmediateSinkBlock(SuccBlk) && !Visited.count(SuccBlk)) { // If the block has reachable child blocks that aren't no-return, // add them to the worklist. DFSWorkList.push_back(SuccBlk); } } } } // Nothing reached the exit. It can only mean one thing: there's no return. return true; } static BugReport * FindReportInEquivalenceClass(BugReportEquivClass& EQ, SmallVectorImpl &bugReports) { BugReportEquivClass::iterator I = EQ.begin(), E = EQ.end(); assert(I != E); BugType& BT = I->getBugType(); // If we don't need to suppress any of the nodes because they are // post-dominated by a sink, simply add all the nodes in the equivalence class // to 'Nodes'. Any of the reports will serve as a "representative" report. if (!BT.isSuppressOnSink()) { BugReport *R = &*I; for (auto &I : EQ) { const ExplodedNode *N = I.getErrorNode(); if (N) { R = &I; bugReports.push_back(R); } } return R; } // For bug reports that should be suppressed when all paths are post-dominated // by a sink node, iterate through the reports in the equivalence class // until we find one that isn't post-dominated (if one exists). We use a // DFS traversal of the ExplodedGraph to find a non-sink node. We could write // this as a recursive function, but we don't want to risk blowing out the // stack for very long paths. BugReport *exampleReport = nullptr; for (; I != E; ++I) { const ExplodedNode *errorNode = I->getErrorNode(); if (!errorNode) continue; if (errorNode->isSink()) { llvm_unreachable( "BugType::isSuppressSink() should not be 'true' for sink end nodes"); } // No successors? By definition this nodes isn't post-dominated by a sink. if (errorNode->succ_empty()) { bugReports.push_back(&*I); if (!exampleReport) exampleReport = &*I; continue; } // See if we are in a no-return CFG block. If so, treat this similarly // to being post-dominated by a sink. This works better when the analysis // is incomplete and we have never reached the no-return function call(s) // that we'd inevitably bump into on this path. if (isInevitablySinking(errorNode)) continue; // At this point we know that 'N' is not a sink and it has at least one // successor. Use a DFS worklist to find a non-sink end-of-path node. using WLItem = FRIEC_WLItem; using DFSWorkList = SmallVector; llvm::DenseMap Visited; DFSWorkList WL; WL.push_back(errorNode); Visited[errorNode] = 1; while (!WL.empty()) { WLItem &WI = WL.back(); assert(!WI.N->succ_empty()); for (; WI.I != WI.E; ++WI.I) { const ExplodedNode *Succ = *WI.I; // End-of-path node? if (Succ->succ_empty()) { // If we found an end-of-path node that is not a sink. if (!Succ->isSink()) { bugReports.push_back(&*I); if (!exampleReport) exampleReport = &*I; WL.clear(); break; } // Found a sink? Continue on to the next successor. continue; } // Mark the successor as visited. If it hasn't been explored, // enqueue it to the DFS worklist. unsigned &mark = Visited[Succ]; if (!mark) { mark = 1; WL.push_back(Succ); break; } } // The worklist may have been cleared at this point. First // check if it is empty before checking the last item. if (!WL.empty() && &WL.back() == &WI) WL.pop_back(); } } // ExampleReport will be NULL if all the nodes in the equivalence class // were post-dominated by sinks. return exampleReport; } void BugReporter::FlushReport(BugReportEquivClass& EQ) { SmallVector bugReports; BugReport *report = FindReportInEquivalenceClass(EQ, bugReports); if (!report) return; ArrayRef Consumers = getPathDiagnosticConsumers(); std::unique_ptr Diagnostics = generateDiagnosticForConsumerMap(report, Consumers, bugReports); for (auto &P : *Diagnostics) { PathDiagnosticConsumer *Consumer = P.first; std::unique_ptr &PD = P.second; // If the path is empty, generate a single step path with the location // of the issue. if (PD->path.empty()) { PathDiagnosticLocation L = report->getLocation(getSourceManager()); auto piece = llvm::make_unique( L, report->getDescription()); for (SourceRange Range : report->getRanges()) piece->addRange(Range); PD->setEndOfPath(std::move(piece)); } PathPieces &Pieces = PD->getMutablePieces(); if (getAnalyzerOptions().shouldDisplayNotesAsEvents()) { // For path diagnostic consumers that don't support extra notes, // we may optionally convert those to path notes. for (auto I = report->getNotes().rbegin(), E = report->getNotes().rend(); I != E; ++I) { PathDiagnosticNotePiece *Piece = I->get(); auto ConvertedPiece = std::make_shared( Piece->getLocation(), Piece->getString()); for (const auto &R: Piece->getRanges()) ConvertedPiece->addRange(R); Pieces.push_front(std::move(ConvertedPiece)); } } else { for (auto I = report->getNotes().rbegin(), E = report->getNotes().rend(); I != E; ++I) Pieces.push_front(*I); } // Get the meta data. const BugReport::ExtraTextList &Meta = report->getExtraText(); for (const auto &i : Meta) PD->addMeta(i); Consumer->HandlePathDiagnostic(std::move(PD)); } } /// Insert all lines participating in the function signature \p Signature /// into \p ExecutedLines. static void populateExecutedLinesWithFunctionSignature( const Decl *Signature, SourceManager &SM, std::unique_ptr &ExecutedLines) { SourceRange SignatureSourceRange; const Stmt* Body = Signature->getBody(); if (const auto FD = dyn_cast(Signature)) { SignatureSourceRange = FD->getSourceRange(); } else if (const auto OD = dyn_cast(Signature)) { SignatureSourceRange = OD->getSourceRange(); } else { return; } SourceLocation Start = SignatureSourceRange.getBegin(); SourceLocation End = Body ? Body->getSourceRange().getBegin() : SignatureSourceRange.getEnd(); unsigned StartLine = SM.getExpansionLineNumber(Start); unsigned EndLine = SM.getExpansionLineNumber(End); FileID FID = SM.getFileID(SM.getExpansionLoc(Start)); for (unsigned Line = StartLine; Line <= EndLine; Line++) ExecutedLines->operator[](FID.getHashValue()).insert(Line); } static void populateExecutedLinesWithStmt( const Stmt *S, SourceManager &SM, std::unique_ptr &ExecutedLines) { SourceLocation Loc = S->getSourceRange().getBegin(); SourceLocation ExpansionLoc = SM.getExpansionLoc(Loc); FileID FID = SM.getFileID(ExpansionLoc); unsigned LineNo = SM.getExpansionLineNumber(ExpansionLoc); ExecutedLines->operator[](FID.getHashValue()).insert(LineNo); } /// \return all executed lines including function signatures on the path /// starting from \p N. static std::unique_ptr findExecutedLines(SourceManager &SM, const ExplodedNode *N) { auto ExecutedLines = llvm::make_unique(); while (N) { if (N->getFirstPred() == nullptr) { // First node: show signature of the entrance point. const Decl *D = N->getLocationContext()->getDecl(); populateExecutedLinesWithFunctionSignature(D, SM, ExecutedLines); } else if (auto CE = N->getLocationAs()) { // Inlined function: show signature. const Decl* D = CE->getCalleeContext()->getDecl(); populateExecutedLinesWithFunctionSignature(D, SM, ExecutedLines); } else if (const Stmt *S = PathDiagnosticLocation::getStmt(N)) { populateExecutedLinesWithStmt(S, SM, ExecutedLines); // Show extra context for some parent kinds. const Stmt *P = N->getParentMap().getParent(S); // The path exploration can die before the node with the associated // return statement is generated, but we do want to show the whole // return. if (const auto *RS = dyn_cast_or_null(P)) { populateExecutedLinesWithStmt(RS, SM, ExecutedLines); P = N->getParentMap().getParent(RS); } if (P && (isa(P) || isa(P))) populateExecutedLinesWithStmt(P, SM, ExecutedLines); } N = N->getFirstPred(); } return ExecutedLines; } std::unique_ptr BugReporter::generateDiagnosticForConsumerMap( BugReport *report, ArrayRef consumers, ArrayRef bugReports) { if (!report->isPathSensitive()) { auto Out = llvm::make_unique(); for (auto *Consumer : consumers) (*Out)[Consumer] = generateEmptyDiagnosticForReport(report, getSourceManager()); return Out; } // Generate the full path sensitive diagnostic, using the generation scheme // specified by the PathDiagnosticConsumer. Note that we have to generate // path diagnostics even for consumers which do not support paths, because // the BugReporterVisitors may mark this bug as a false positive. assert(!bugReports.empty()); MaxBugClassSize.updateMax(bugReports.size()); std::unique_ptr Out = generatePathDiagnostics(consumers, bugReports); if (Out->empty()) return Out; MaxValidBugClassSize.updateMax(bugReports.size()); // Examine the report and see if the last piece is in a header. Reset the // report location to the last piece in the main source file. AnalyzerOptions &Opts = getAnalyzerOptions(); for (auto const &P : *Out) if (Opts.shouldReportIssuesInMainSourceFile() && !Opts.AnalyzeAll) P.second->resetDiagnosticLocationToMainFile(); return Out; } void BugReporter::EmitBasicReport(const Decl *DeclWithIssue, const CheckerBase *Checker, StringRef Name, StringRef Category, StringRef Str, PathDiagnosticLocation Loc, ArrayRef Ranges) { EmitBasicReport(DeclWithIssue, Checker->getCheckName(), Name, Category, Str, Loc, Ranges); } void BugReporter::EmitBasicReport(const Decl *DeclWithIssue, CheckName CheckName, StringRef name, StringRef category, StringRef str, PathDiagnosticLocation Loc, ArrayRef Ranges) { // 'BT' is owned by BugReporter. BugType *BT = getBugTypeForName(CheckName, name, category); auto R = llvm::make_unique(*BT, str, Loc); R->setDeclWithIssue(DeclWithIssue); for (ArrayRef::iterator I = Ranges.begin(), E = Ranges.end(); I != E; ++I) R->addRange(*I); emitReport(std::move(R)); } BugType *BugReporter::getBugTypeForName(CheckName CheckName, StringRef name, StringRef category) { SmallString<136> fullDesc; llvm::raw_svector_ostream(fullDesc) << CheckName.getName() << ":" << name << ":" << category; BugType *&BT = StrBugTypes[fullDesc]; if (!BT) BT = new BugType(CheckName, name, category); return BT; }