diff options
author | Zancanaro; Carlo <czan8762@plang3.cs.usyd.edu.au> | 2012-09-24 09:58:17 +1000 |
---|---|---|
committer | Zancanaro; Carlo <czan8762@plang3.cs.usyd.edu.au> | 2012-09-24 09:58:17 +1000 |
commit | 222e2a7620e6520ffaf4fc4e69d79c18da31542e (patch) | |
tree | 7bfbc05bfa3b41c8f9d2e56d53a0bc3e310df239 /clang/lib/StaticAnalyzer/Checkers | |
parent | 3d206f03985b50beacae843d880bccdc91a9f424 (diff) |
Add the clang library to the repo (with some of my changes, too).
Diffstat (limited to 'clang/lib/StaticAnalyzer/Checkers')
76 files changed, 21121 insertions, 0 deletions
diff --git a/clang/lib/StaticAnalyzer/Checkers/.#ArrayBoundCheckerV2_flymake.cpp b/clang/lib/StaticAnalyzer/Checkers/.#ArrayBoundCheckerV2_flymake.cpp new file mode 120000 index 0000000..235903b --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/.#ArrayBoundCheckerV2_flymake.cpp @@ -0,0 +1 @@ +carlo@pc-4w14-0.cs.usyd.edu.au.1585:1347012043
\ No newline at end of file diff --git a/clang/lib/StaticAnalyzer/Checkers/.#CMakeLists.txt b/clang/lib/StaticAnalyzer/Checkers/.#CMakeLists.txt new file mode 120000 index 0000000..235903b --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/.#CMakeLists.txt @@ -0,0 +1 @@ +carlo@pc-4w14-0.cs.usyd.edu.au.1585:1347012043
\ No newline at end of file diff --git a/clang/lib/StaticAnalyzer/Checkers/.#Checkers.td b/clang/lib/StaticAnalyzer/Checkers/.#Checkers.td new file mode 120000 index 0000000..235903b --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/.#Checkers.td @@ -0,0 +1 @@ +carlo@pc-4w14-0.cs.usyd.edu.au.1585:1347012043
\ No newline at end of file diff --git a/clang/lib/StaticAnalyzer/Checkers/.#DebugCheckers.cpp b/clang/lib/StaticAnalyzer/Checkers/.#DebugCheckers.cpp new file mode 120000 index 0000000..235903b --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/.#DebugCheckers.cpp @@ -0,0 +1 @@ +carlo@pc-4w14-0.cs.usyd.edu.au.1585:1347012043
\ No newline at end of file diff --git a/clang/lib/StaticAnalyzer/Checkers/.#DebugCheckers_flymake.cpp b/clang/lib/StaticAnalyzer/Checkers/.#DebugCheckers_flymake.cpp new file mode 120000 index 0000000..235903b --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/.#DebugCheckers_flymake.cpp @@ -0,0 +1 @@ +carlo@pc-4w14-0.cs.usyd.edu.au.1585:1347012043
\ No newline at end of file diff --git a/clang/lib/StaticAnalyzer/Checkers/.#DivZeroChecker_flymake.cpp b/clang/lib/StaticAnalyzer/Checkers/.#DivZeroChecker_flymake.cpp new file mode 120000 index 0000000..235903b --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/.#DivZeroChecker_flymake.cpp @@ -0,0 +1 @@ +carlo@pc-4w14-0.cs.usyd.edu.au.1585:1347012043
\ No newline at end of file diff --git a/clang/lib/StaticAnalyzer/Checkers/.#IntervalTest.cpp b/clang/lib/StaticAnalyzer/Checkers/.#IntervalTest.cpp new file mode 120000 index 0000000..235903b --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/.#IntervalTest.cpp @@ -0,0 +1 @@ +carlo@pc-4w14-0.cs.usyd.edu.au.1585:1347012043
\ No newline at end of file diff --git a/clang/lib/StaticAnalyzer/Checkers/.#IntervalTest_flymake.cpp b/clang/lib/StaticAnalyzer/Checkers/.#IntervalTest_flymake.cpp new file mode 120000 index 0000000..235903b --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/.#IntervalTest_flymake.cpp @@ -0,0 +1 @@ +carlo@pc-4w14-0.cs.usyd.edu.au.1585:1347012043
\ No newline at end of file diff --git a/clang/lib/StaticAnalyzer/Checkers/AdjustedReturnValueChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/AdjustedReturnValueChecker.cpp new file mode 100644 index 0000000..84ea8c7 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/AdjustedReturnValueChecker.cpp @@ -0,0 +1,92 @@ +//== AdjustedReturnValueChecker.cpp -----------------------------*- C++ -*--==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines AdjustedReturnValueChecker, a simple check to see if the +// return value of a function call is different than the one the caller thinks +// it is. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugReporter.h" + +using namespace clang; +using namespace ento; + +namespace { +class AdjustedReturnValueChecker : + public Checker< check::PostStmt<CallExpr> > { +public: + void checkPostStmt(const CallExpr *CE, CheckerContext &C) const; +}; +} + +void AdjustedReturnValueChecker::checkPostStmt(const CallExpr *CE, + CheckerContext &C) const { + + // Get the result type of the call. + QualType expectedResultTy = CE->getType(); + + // Fetch the signature of the called function. + ProgramStateRef state = C.getState(); + const LocationContext *LCtx = C.getLocationContext(); + + SVal V = state->getSVal(CE, LCtx); + + if (V.isUnknown()) + return; + + // Casting to void? Discard the value. + if (expectedResultTy->isVoidType()) { + C.addTransition(state->BindExpr(CE, LCtx, UnknownVal())); + return; + } + + const MemRegion *callee = state->getSVal(CE->getCallee(), LCtx).getAsRegion(); + if (!callee) + return; + + QualType actualResultTy; + + if (const FunctionTextRegion *FT = dyn_cast<FunctionTextRegion>(callee)) { + const FunctionDecl *FD = FT->getDecl(); + actualResultTy = FD->getResultType(); + } + else if (const BlockDataRegion *BD = dyn_cast<BlockDataRegion>(callee)) { + const BlockTextRegion *BR = BD->getCodeRegion(); + const BlockPointerType *BT=BR->getLocationType()->getAs<BlockPointerType>(); + const FunctionType *FT = BT->getPointeeType()->getAs<FunctionType>(); + actualResultTy = FT->getResultType(); + } + + // Can this happen? + if (actualResultTy.isNull()) + return; + + // For now, ignore references. + if (actualResultTy->getAs<ReferenceType>()) + return; + + + // Are they the same? + if (expectedResultTy != actualResultTy) { + // FIXME: Do more checking and actual emit an error. At least performing + // the cast avoids some assertion failures elsewhere. + SValBuilder &svalBuilder = C.getSValBuilder(); + V = svalBuilder.evalCast(V, expectedResultTy, actualResultTy); + C.addTransition(state->BindExpr(CE, LCtx, V)); + } +} + +void ento::registerAdjustedReturnValueChecker(CheckerManager &mgr) { + mgr.registerChecker<AdjustedReturnValueChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/AnalyzerStatsChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/AnalyzerStatsChecker.cpp new file mode 100644 index 0000000..aa6f97b --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/AnalyzerStatsChecker.cpp @@ -0,0 +1,140 @@ +//==--AnalyzerStatsChecker.cpp - Analyzer visitation statistics --*- C++ -*-==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// This file reports various statistics about analyzer visitation. +//===----------------------------------------------------------------------===// +#define DEBUG_TYPE "StatsChecker" + +#include "ClangSACheckers.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/BugReporter/BugReporter.h" + +#include "clang/AST/DeclObjC.h" +#include "clang/Basic/SourceManager.h" +#include "llvm/ADT/SmallPtrSet.h" +#include "llvm/ADT/SmallString.h" +#include "llvm/ADT/Statistic.h" + +using namespace clang; +using namespace ento; + +STATISTIC(NumBlocks, + "The # of blocks in top level functions"); +STATISTIC(NumBlocksUnreachable, + "The # of unreachable blocks in analyzing top level functions"); + +namespace { +class AnalyzerStatsChecker : public Checker<check::EndAnalysis> { +public: + void checkEndAnalysis(ExplodedGraph &G, BugReporter &B,ExprEngine &Eng) const; +}; +} + +void AnalyzerStatsChecker::checkEndAnalysis(ExplodedGraph &G, + BugReporter &B, + ExprEngine &Eng) const { + const CFG *C = 0; + const SourceManager &SM = B.getSourceManager(); + llvm::SmallPtrSet<const CFGBlock*, 256> reachable; + + // Root node should have the location context of the top most function. + const ExplodedNode *GraphRoot = *G.roots_begin(); + const LocationContext *LC = GraphRoot->getLocation().getLocationContext(); + + const Decl *D = LC->getDecl(); + + // Iterate over the exploded graph. + for (ExplodedGraph::node_iterator I = G.nodes_begin(); + I != G.nodes_end(); ++I) { + const ProgramPoint &P = I->getLocation(); + + // Only check the coverage in the top level function (optimization). + if (D != P.getLocationContext()->getDecl()) + continue; + + if (const BlockEntrance *BE = dyn_cast<BlockEntrance>(&P)) { + const CFGBlock *CB = BE->getBlock(); + reachable.insert(CB); + } + } + + // Get the CFG and the Decl of this block. + C = LC->getCFG(); + + unsigned total = 0, unreachable = 0; + + // Find CFGBlocks that were not covered by any node + for (CFG::const_iterator I = C->begin(); I != C->end(); ++I) { + const CFGBlock *CB = *I; + ++total; + // Check if the block is unreachable + if (!reachable.count(CB)) { + ++unreachable; + } + } + + // We never 'reach' the entry block, so correct the unreachable count + unreachable--; + // There is no BlockEntrance corresponding to the exit block as well, so + // assume it is reached as well. + unreachable--; + + // Generate the warning string + SmallString<128> buf; + llvm::raw_svector_ostream output(buf); + PresumedLoc Loc = SM.getPresumedLoc(D->getLocation()); + if (!Loc.isValid()) + return; + + if (isa<FunctionDecl>(D) || isa<ObjCMethodDecl>(D)) { + const NamedDecl *ND = cast<NamedDecl>(D); + output << *ND; + } + else if (isa<BlockDecl>(D)) { + output << "block(line:" << Loc.getLine() << ":col:" << Loc.getColumn(); + } + + NumBlocksUnreachable += unreachable; + NumBlocks += total; + std::string NameOfRootFunction = output.str(); + + output << " -> Total CFGBlocks: " << total << " | Unreachable CFGBlocks: " + << unreachable << " | Exhausted Block: " + << (Eng.wasBlocksExhausted() ? "yes" : "no") + << " | Empty WorkList: " + << (Eng.hasEmptyWorkList() ? "yes" : "no"); + + B.EmitBasicReport(D, "Analyzer Statistics", "Internal Statistics", + output.str(), PathDiagnosticLocation(D, SM)); + + // Emit warning for each block we bailed out on. + typedef CoreEngine::BlocksExhausted::const_iterator ExhaustedIterator; + const CoreEngine &CE = Eng.getCoreEngine(); + for (ExhaustedIterator I = CE.blocks_exhausted_begin(), + E = CE.blocks_exhausted_end(); I != E; ++I) { + const BlockEdge &BE = I->first; + const CFGBlock *Exit = BE.getDst(); + const CFGElement &CE = Exit->front(); + if (const CFGStmt *CS = dyn_cast<CFGStmt>(&CE)) { + SmallString<128> bufI; + llvm::raw_svector_ostream outputI(bufI); + outputI << "(" << NameOfRootFunction << ")" << + ": The analyzer generated a sink at this point"; + B.EmitBasicReport(D, "Sink Point", "Internal Statistics", outputI.str(), + PathDiagnosticLocation::createBegin(CS->getStmt(), + SM, LC)); + } + } +} + +void ento::registerAnalyzerStatsChecker(CheckerManager &mgr) { + mgr.registerChecker<AnalyzerStatsChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/ArrayBoundChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/ArrayBoundChecker.cpp new file mode 100644 index 0000000..b2ad184 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/ArrayBoundChecker.cpp @@ -0,0 +1,92 @@ +//== ArrayBoundChecker.cpp ------------------------------*- C++ -*--==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines ArrayBoundChecker, which is a path-sensitive check +// which looks for an out-of-bound array element access. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ExprEngine.h" + +using namespace clang; +using namespace ento; + +namespace { +class ArrayBoundChecker : + public Checker<check::Location> { + mutable OwningPtr<BuiltinBug> BT; +public: + void checkLocation(SVal l, bool isLoad, const Stmt* S, + CheckerContext &C) const; +}; +} + +void ArrayBoundChecker::checkLocation(SVal l, bool isLoad, const Stmt* LoadS, + CheckerContext &C) const { + // Check for out of bound array element access. + const MemRegion *R = l.getAsRegion(); + if (!R) + return; + + const ElementRegion *ER = dyn_cast<ElementRegion>(R); + if (!ER) + return; + + // Get the index of the accessed element. + DefinedOrUnknownSVal Idx = cast<DefinedOrUnknownSVal>(ER->getIndex()); + + // Zero index is always in bound, this also passes ElementRegions created for + // pointer casts. + if (Idx.isZeroConstant()) + return; + + ProgramStateRef state = C.getState(); + + // Get the size of the array. + DefinedOrUnknownSVal NumElements + = C.getStoreManager().getSizeInElements(state, ER->getSuperRegion(), + ER->getValueType()); + + ProgramStateRef StInBound = state->assumeInBound(Idx, NumElements, true); + ProgramStateRef StOutBound = state->assumeInBound(Idx, NumElements, false); + if (StOutBound && !StInBound) { + ExplodedNode *N = C.generateSink(StOutBound); + if (!N) + return; + + if (!BT) + BT.reset(new BuiltinBug("Out-of-bound array access", + "Access out-of-bound array element (buffer overflow)")); + + // FIXME: It would be nice to eventually make this diagnostic more clear, + // e.g., by referencing the original declaration or by saying *why* this + // reference is outside the range. + + // Generate a report for this bug. + BugReport *report = + new BugReport(*BT, BT->getDescription(), N); + + report->addRange(LoadS->getSourceRange()); + C.EmitReport(report); + return; + } + + // Array bound check succeeded. From this point forward the array bound + // should always succeed. + C.addTransition(StInBound); +} + +void ento::registerArrayBoundChecker(CheckerManager &mgr) { + mgr.registerChecker<ArrayBoundChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/ArrayBoundCheckerV2.cpp b/clang/lib/StaticAnalyzer/Checkers/ArrayBoundCheckerV2.cpp new file mode 100644 index 0000000..c6efe94 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/ArrayBoundCheckerV2.cpp @@ -0,0 +1,318 @@ +//== ArrayBoundCheckerV2.cpp ------------------------------------*- C++ -*--==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines ArrayBoundCheckerV2, which is a path-sensitive check +// which looks for an out-of-bound array element access. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ExprEngine.h" +#include "clang/AST/CharUnits.h" +#include "llvm/ADT/SmallString.h" +#include "llvm/ADT/STLExtras.h" + +using namespace clang; +using namespace ento; + +namespace { +class ArrayBoundCheckerV2 : + public Checker<check::Location> { + mutable OwningPtr<BuiltinBug> BT; + + enum OOB_Kind { OOB_Precedes, OOB_Excedes, OOB_Tainted }; + + void reportOOB(CheckerContext &C, ProgramStateRef errorState, + OOB_Kind kind) const; + +public: + void checkLocation(SVal l, bool isLoad, const Stmt*S, + CheckerContext &C) const; +}; + +// FIXME: Eventually replace RegionRawOffset with this class. +class RegionRawOffsetV2 { +private: + const SubRegion *baseRegion; + SVal byteOffset; + + RegionRawOffsetV2() + : baseRegion(0), byteOffset(UnknownVal()) {} + +public: + RegionRawOffsetV2(const SubRegion* base, SVal offset) + : baseRegion(base), byteOffset(offset) {} + + NonLoc getByteOffset() const { return cast<NonLoc>(byteOffset); } + const SubRegion *getRegion() const { return baseRegion; } + + static RegionRawOffsetV2 computeOffset(ProgramStateRef state, + SValBuilder &svalBuilder, + SVal location); + + void dump() const; + void dumpToStream(raw_ostream &os) const; +}; +} + +static SVal computeExtentBegin(SValBuilder &svalBuilder, + const MemRegion *region) { + while (true) + switch (region->getKind()) { + default: + return svalBuilder.makeZeroArrayIndex(); + case MemRegion::SymbolicRegionKind: + // FIXME: improve this later by tracking symbolic lower bounds + // for symbolic regions. + return UnknownVal(); + case MemRegion::ElementRegionKind: + region = cast<SubRegion>(region)->getSuperRegion(); + continue; + } +} + +void ArrayBoundCheckerV2::checkLocation(SVal location, bool isLoad, + const Stmt* LoadS, + CheckerContext &checkerContext) const { + + // NOTE: Instead of using ProgramState::assumeInBound(), we are prototyping + // some new logic here that reasons directly about memory region extents. + // Once that logic is more mature, we can bring it back to assumeInBound() + // for all clients to use. + // + // The algorithm we are using here for bounds checking is to see if the + // memory access is within the extent of the base region. Since we + // have some flexibility in defining the base region, we can achieve + // various levels of conservatism in our buffer overflow checking. + ProgramStateRef state = checkerContext.getState(); + ProgramStateRef originalState = state; + + SValBuilder &svalBuilder = checkerContext.getSValBuilder(); + const RegionRawOffsetV2 &rawOffset = + RegionRawOffsetV2::computeOffset(state, svalBuilder, location); + + if (!rawOffset.getRegion()) + return; + + // CHECK LOWER BOUND: Is byteOffset < extent begin? + // If so, we are doing a load/store + // before the first valid offset in the memory region. + + SVal extentBegin = computeExtentBegin(svalBuilder, rawOffset.getRegion()); + + if (isa<NonLoc>(extentBegin)) { + SVal lowerBound + = svalBuilder.evalBinOpNN(state, BO_LT, rawOffset.getByteOffset(), + cast<NonLoc>(extentBegin), + svalBuilder.getConditionType()); + + NonLoc *lowerBoundToCheck = dyn_cast<NonLoc>(&lowerBound); + if (!lowerBoundToCheck) + return; + + ProgramStateRef state_precedesLowerBound, state_withinLowerBound; + llvm::tie(state_precedesLowerBound, state_withinLowerBound) = + state->assume(*lowerBoundToCheck); + + // Are we constrained enough to definitely precede the lower bound? + if (state_precedesLowerBound && !state_withinLowerBound) { + reportOOB(checkerContext, state_precedesLowerBound, OOB_Precedes); + return; + } + + // Otherwise, assume the constraint of the lower bound. + assert(state_withinLowerBound); + state = state_withinLowerBound; + } + + do { + // CHECK UPPER BOUND: Is byteOffset >= extent(baseRegion)? If so, + // we are doing a load/store after the last valid offset. + DefinedOrUnknownSVal extentVal = + rawOffset.getRegion()->getExtent(svalBuilder); + if (!isa<NonLoc>(extentVal)) + break; + + SVal upperbound + = svalBuilder.evalBinOpNN(state, BO_GE, rawOffset.getByteOffset(), + cast<NonLoc>(extentVal), + svalBuilder.getConditionType()); + + NonLoc *upperboundToCheck = dyn_cast<NonLoc>(&upperbound); + if (!upperboundToCheck) + break; + + ProgramStateRef state_exceedsUpperBound, state_withinUpperBound; + llvm::tie(state_exceedsUpperBound, state_withinUpperBound) = + state->assume(*upperboundToCheck); + + // If we are under constrained and the index variables are tainted, report. + if (state_exceedsUpperBound && state_withinUpperBound) { + if (state->isTainted(rawOffset.getByteOffset())) + reportOOB(checkerContext, state_exceedsUpperBound, OOB_Tainted); + return; + } + + // If we are constrained enough to definitely exceed the upper bound, report. + if (state_exceedsUpperBound) { + assert(!state_withinUpperBound); + reportOOB(checkerContext, state_exceedsUpperBound, OOB_Excedes); + return; + } + + assert(state_withinUpperBound); + state = state_withinUpperBound; + } + while (false); + + if (state != originalState) + checkerContext.addTransition(state); +} + +void ArrayBoundCheckerV2::reportOOB(CheckerContext &checkerContext, + ProgramStateRef errorState, + OOB_Kind kind) const { + + ExplodedNode *errorNode = checkerContext.generateSink(errorState); + if (!errorNode) + return; + + if (!BT) + BT.reset(new BuiltinBug("Out-of-bound access")); + + // FIXME: This diagnostics are preliminary. We should get far better + // diagnostics for explaining buffer overruns. + + SmallString<256> buf; + llvm::raw_svector_ostream os(buf); + os << "Out of bound memory access "; + switch (kind) { + case OOB_Precedes: + os << "(accessed memory precedes memory block)"; + break; + case OOB_Excedes: + os << "(access exceeds upper limit of memory block)"; + break; + case OOB_Tainted: + os << "(index is tainted)"; + break; + } + + checkerContext.EmitReport(new BugReport(*BT, os.str(), errorNode)); +} + +void RegionRawOffsetV2::dump() const { + dumpToStream(llvm::errs()); +} + +void RegionRawOffsetV2::dumpToStream(raw_ostream &os) const { + os << "raw_offset_v2{" << getRegion() << ',' << getByteOffset() << '}'; +} + +// FIXME: Merge with the implementation of the same method in Store.cpp +static bool IsCompleteType(ASTContext &Ctx, QualType Ty) { + if (const RecordType *RT = Ty->getAs<RecordType>()) { + const RecordDecl *D = RT->getDecl(); + if (!D->getDefinition()) + return false; + } + + return true; +} + + +// Lazily computes a value to be used by 'computeOffset'. If 'val' +// is unknown or undefined, we lazily substitute '0'. Otherwise, +// return 'val'. +static inline SVal getValue(SVal val, SValBuilder &svalBuilder) { + return isa<UndefinedVal>(val) ? svalBuilder.makeArrayIndex(0) : val; +} + +// Scale a base value by a scaling factor, and return the scaled +// value as an SVal. Used by 'computeOffset'. +static inline SVal scaleValue(ProgramStateRef state, + NonLoc baseVal, CharUnits scaling, + SValBuilder &sb) { + return sb.evalBinOpNN(state, BO_Mul, baseVal, + sb.makeArrayIndex(scaling.getQuantity()), + sb.getArrayIndexType()); +} + +// Add an SVal to another, treating unknown and undefined values as +// summing to UnknownVal. Used by 'computeOffset'. +static SVal addValue(ProgramStateRef state, SVal x, SVal y, + SValBuilder &svalBuilder) { + // We treat UnknownVals and UndefinedVals the same here because we + // only care about computing offsets. + if (x.isUnknownOrUndef() || y.isUnknownOrUndef()) + return UnknownVal(); + + return svalBuilder.evalBinOpNN(state, BO_Add, + cast<NonLoc>(x), cast<NonLoc>(y), + svalBuilder.getArrayIndexType()); +} + +/// Compute a raw byte offset from a base region. Used for array bounds +/// checking. +RegionRawOffsetV2 RegionRawOffsetV2::computeOffset(ProgramStateRef state, + SValBuilder &svalBuilder, + SVal location) +{ + const MemRegion *region = location.getAsRegion(); + SVal offset = UndefinedVal(); + + while (region) { + switch (region->getKind()) { + default: { + if (const SubRegion *subReg = dyn_cast<SubRegion>(region)) { + offset = getValue(offset, svalBuilder); + if (!offset.isUnknownOrUndef()) + return RegionRawOffsetV2(subReg, offset); + } + return RegionRawOffsetV2(); + } + case MemRegion::ElementRegionKind: { + const ElementRegion *elemReg = cast<ElementRegion>(region); + SVal index = elemReg->getIndex(); + if (!isa<NonLoc>(index)) + return RegionRawOffsetV2(); + QualType elemType = elemReg->getElementType(); + // If the element is an incomplete type, go no further. + ASTContext &astContext = svalBuilder.getContext(); + if (!IsCompleteType(astContext, elemType)) + return RegionRawOffsetV2(); + + // Update the offset. + offset = addValue(state, + getValue(offset, svalBuilder), + scaleValue(state, + cast<NonLoc>(index), + astContext.getTypeSizeInChars(elemType), + svalBuilder), + svalBuilder); + + if (offset.isUnknownOrUndef()) + return RegionRawOffsetV2(); + + region = elemReg->getSuperRegion(); + continue; + } + } + } + return RegionRawOffsetV2(); +} + + +void ento::registerArrayBoundCheckerV2(CheckerManager &mgr) { + mgr.registerChecker<ArrayBoundCheckerV2>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/AttrNonNullChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/AttrNonNullChecker.cpp new file mode 100644 index 0000000..ab66e98 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/AttrNonNullChecker.cpp @@ -0,0 +1,134 @@ +//===--- AttrNonNullChecker.h - Undefined arguments checker ----*- C++ -*--===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This defines AttrNonNullChecker, a builtin check in ExprEngine that +// performs checks for arguments declared to have nonnull attribute. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" + +using namespace clang; +using namespace ento; + +namespace { +class AttrNonNullChecker + : public Checker< check::PreStmt<CallExpr> > { + mutable OwningPtr<BugType> BT; +public: + + void checkPreStmt(const CallExpr *CE, CheckerContext &C) const; +}; +} // end anonymous namespace + +void AttrNonNullChecker::checkPreStmt(const CallExpr *CE, + CheckerContext &C) const { + ProgramStateRef state = C.getState(); + const LocationContext *LCtx = C.getLocationContext(); + + // Check if the callee has a 'nonnull' attribute. + SVal X = state->getSVal(CE->getCallee(), LCtx); + + const FunctionDecl *FD = X.getAsFunctionDecl(); + if (!FD) + return; + + const NonNullAttr* Att = FD->getAttr<NonNullAttr>(); + if (!Att) + return; + + // Iterate through the arguments of CE and check them for null. + unsigned idx = 0; + + for (CallExpr::const_arg_iterator I=CE->arg_begin(), E=CE->arg_end(); I!=E; + ++I, ++idx) { + + if (!Att->isNonNull(idx)) + continue; + + SVal V = state->getSVal(*I, LCtx); + DefinedSVal *DV = dyn_cast<DefinedSVal>(&V); + + // If the value is unknown or undefined, we can't perform this check. + if (!DV) + continue; + + if (!isa<Loc>(*DV)) { + // If the argument is a union type, we want to handle a potential + // transparent_unoin GCC extension. + QualType T = (*I)->getType(); + const RecordType *UT = T->getAsUnionType(); + if (!UT || !UT->getDecl()->hasAttr<TransparentUnionAttr>()) + continue; + if (nonloc::CompoundVal *CSV = dyn_cast<nonloc::CompoundVal>(DV)) { + nonloc::CompoundVal::iterator CSV_I = CSV->begin(); + assert(CSV_I != CSV->end()); + V = *CSV_I; + DV = dyn_cast<DefinedSVal>(&V); + assert(++CSV_I == CSV->end()); + if (!DV) + continue; + } + else { + // FIXME: Handle LazyCompoundVals? + continue; + } + } + + ConstraintManager &CM = C.getConstraintManager(); + ProgramStateRef stateNotNull, stateNull; + llvm::tie(stateNotNull, stateNull) = CM.assumeDual(state, *DV); + + if (stateNull && !stateNotNull) { + // Generate an error node. Check for a null node in case + // we cache out. + if (ExplodedNode *errorNode = C.generateSink(stateNull)) { + + // Lazily allocate the BugType object if it hasn't already been + // created. Ownership is transferred to the BugReporter object once + // the BugReport is passed to 'EmitWarning'. + if (!BT) + BT.reset(new BugType("Argument with 'nonnull' attribute passed null", + "API")); + + BugReport *R = + new BugReport(*BT, "Null pointer passed as an argument to a " + "'nonnull' parameter", errorNode); + + // Highlight the range of the argument that was null. + const Expr *arg = *I; + R->addRange(arg->getSourceRange()); + R->addVisitor(bugreporter::getTrackNullOrUndefValueVisitor(errorNode, + arg, R)); + // Emit the bug report. + C.EmitReport(R); + } + + // Always return. Either we cached out or we just emitted an error. + return; + } + + // If a pointer value passed the check we should assume that it is + // indeed not null from this point forward. + assert(stateNotNull); + state = stateNotNull; + } + + // If we reach here all of the arguments passed the nonnull check. + // If 'state' has been updated generated a new node. + C.addTransition(state); +} + +void ento::registerAttrNonNullChecker(CheckerManager &mgr) { + mgr.registerChecker<AttrNonNullChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/BasicObjCFoundationChecks.cpp b/clang/lib/StaticAnalyzer/Checkers/BasicObjCFoundationChecks.cpp new file mode 100644 index 0000000..6dd0a8c --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/BasicObjCFoundationChecks.cpp @@ -0,0 +1,672 @@ +//== BasicObjCFoundationChecks.cpp - Simple Apple-Foundation checks -*- C++ -*-- +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines BasicObjCFoundationChecks, a class that encapsulates +// a set of simple checks to run on Objective-C code using Apple's Foundation +// classes. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/Analysis/DomainSpecific/CocoaConventions.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ExplodedGraph.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ExprEngine.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ObjCMessage.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramState.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/MemRegion.h" +#include "clang/AST/DeclObjC.h" +#include "clang/AST/Expr.h" +#include "clang/AST/ExprObjC.h" +#include "clang/AST/ASTContext.h" +#include "llvm/ADT/SmallString.h" + +using namespace clang; +using namespace ento; + +namespace { +class APIMisuse : public BugType { +public: + APIMisuse(const char* name) : BugType(name, "API Misuse (Apple)") {} +}; +} // end anonymous namespace + +//===----------------------------------------------------------------------===// +// Utility functions. +//===----------------------------------------------------------------------===// + +static const char* GetReceiverNameType(const ObjCMessage &msg) { + if (const ObjCInterfaceDecl *ID = msg.getReceiverInterface()) + return ID->getIdentifier()->getNameStart(); + return 0; +} + +static bool isReceiverClassOrSuperclass(const ObjCInterfaceDecl *ID, + StringRef ClassName) { + if (ID->getIdentifier()->getName() == ClassName) + return true; + + if (const ObjCInterfaceDecl *Super = ID->getSuperClass()) + return isReceiverClassOrSuperclass(Super, ClassName); + + return false; +} + +static inline bool isNil(SVal X) { + return isa<loc::ConcreteInt>(X); +} + +//===----------------------------------------------------------------------===// +// NilArgChecker - Check for prohibited nil arguments to ObjC method calls. +//===----------------------------------------------------------------------===// + +namespace { + class NilArgChecker : public Checker<check::PreObjCMessage> { + mutable OwningPtr<APIMisuse> BT; + + void WarnNilArg(CheckerContext &C, + const ObjCMessage &msg, unsigned Arg) const; + + public: + void checkPreObjCMessage(ObjCMessage msg, CheckerContext &C) const; + }; +} + +void NilArgChecker::WarnNilArg(CheckerContext &C, + const ObjCMessage &msg, + unsigned int Arg) const +{ + if (!BT) + BT.reset(new APIMisuse("nil argument")); + + if (ExplodedNode *N = C.generateSink()) { + SmallString<128> sbuf; + llvm::raw_svector_ostream os(sbuf); + os << "Argument to '" << GetReceiverNameType(msg) << "' method '" + << msg.getSelector().getAsString() << "' cannot be nil"; + + BugReport *R = new BugReport(*BT, os.str(), N); + R->addRange(msg.getArgSourceRange(Arg)); + C.EmitReport(R); + } +} + +void NilArgChecker::checkPreObjCMessage(ObjCMessage msg, + CheckerContext &C) const { + const ObjCInterfaceDecl *ID = msg.getReceiverInterface(); + if (!ID) + return; + + if (isReceiverClassOrSuperclass(ID, "NSString")) { + Selector S = msg.getSelector(); + + if (S.isUnarySelector()) + return; + + // FIXME: This is going to be really slow doing these checks with + // lexical comparisons. + + std::string NameStr = S.getAsString(); + StringRef Name(NameStr); + assert(!Name.empty()); + + // FIXME: Checking for initWithFormat: will not work in most cases + // yet because [NSString alloc] returns id, not NSString*. We will + // need support for tracking expected-type information in the analyzer + // to find these errors. + if (Name == "caseInsensitiveCompare:" || + Name == "compare:" || + Name == "compare:options:" || + Name == "compare:options:range:" || + Name == "compare:options:range:locale:" || + Name == "componentsSeparatedByCharactersInSet:" || + Name == "initWithFormat:") { + if (isNil(msg.getArgSVal(0, C.getLocationContext(), C.getState()))) + WarnNilArg(C, msg, 0); + } + } +} + +//===----------------------------------------------------------------------===// +// Error reporting. +//===----------------------------------------------------------------------===// + +namespace { +class CFNumberCreateChecker : public Checker< check::PreStmt<CallExpr> > { + mutable OwningPtr<APIMisuse> BT; + mutable IdentifierInfo* II; +public: + CFNumberCreateChecker() : II(0) {} + + void checkPreStmt(const CallExpr *CE, CheckerContext &C) const; + +private: + void EmitError(const TypedRegion* R, const Expr *Ex, + uint64_t SourceSize, uint64_t TargetSize, uint64_t NumberKind); +}; +} // end anonymous namespace + +enum CFNumberType { + kCFNumberSInt8Type = 1, + kCFNumberSInt16Type = 2, + kCFNumberSInt32Type = 3, + kCFNumberSInt64Type = 4, + kCFNumberFloat32Type = 5, + kCFNumberFloat64Type = 6, + kCFNumberCharType = 7, + kCFNumberShortType = 8, + kCFNumberIntType = 9, + kCFNumberLongType = 10, + kCFNumberLongLongType = 11, + kCFNumberFloatType = 12, + kCFNumberDoubleType = 13, + kCFNumberCFIndexType = 14, + kCFNumberNSIntegerType = 15, + kCFNumberCGFloatType = 16 +}; + +namespace { + template<typename T> + class Optional { + bool IsKnown; + T Val; + public: + Optional() : IsKnown(false), Val(0) {} + Optional(const T& val) : IsKnown(true), Val(val) {} + + bool isKnown() const { return IsKnown; } + + const T& getValue() const { + assert (isKnown()); + return Val; + } + + operator const T&() const { + return getValue(); + } + }; +} + +static Optional<uint64_t> GetCFNumberSize(ASTContext &Ctx, uint64_t i) { + static const unsigned char FixedSize[] = { 8, 16, 32, 64, 32, 64 }; + + if (i < kCFNumberCharType) + return FixedSize[i-1]; + + QualType T; + + switch (i) { + case kCFNumberCharType: T = Ctx.CharTy; break; + case kCFNumberShortType: T = Ctx.ShortTy; break; + case kCFNumberIntType: T = Ctx.IntTy; break; + case kCFNumberLongType: T = Ctx.LongTy; break; + case kCFNumberLongLongType: T = Ctx.LongLongTy; break; + case kCFNumberFloatType: T = Ctx.FloatTy; break; + case kCFNumberDoubleType: T = Ctx.DoubleTy; break; + case kCFNumberCFIndexType: + case kCFNumberNSIntegerType: + case kCFNumberCGFloatType: + // FIXME: We need a way to map from names to Type*. + default: + return Optional<uint64_t>(); + } + + return Ctx.getTypeSize(T); +} + +#if 0 +static const char* GetCFNumberTypeStr(uint64_t i) { + static const char* Names[] = { + "kCFNumberSInt8Type", + "kCFNumberSInt16Type", + "kCFNumberSInt32Type", + "kCFNumberSInt64Type", + "kCFNumberFloat32Type", + "kCFNumberFloat64Type", + "kCFNumberCharType", + "kCFNumberShortType", + "kCFNumberIntType", + "kCFNumberLongType", + "kCFNumberLongLongType", + "kCFNumberFloatType", + "kCFNumberDoubleType", + "kCFNumberCFIndexType", + "kCFNumberNSIntegerType", + "kCFNumberCGFloatType" + }; + + return i <= kCFNumberCGFloatType ? Names[i-1] : "Invalid CFNumberType"; +} +#endif + +void CFNumberCreateChecker::checkPreStmt(const CallExpr *CE, + CheckerContext &C) const { + ProgramStateRef state = C.getState(); + const FunctionDecl *FD = C.getCalleeDecl(CE); + if (!FD) + return; + + ASTContext &Ctx = C.getASTContext(); + if (!II) + II = &Ctx.Idents.get("CFNumberCreate"); + + if (FD->getIdentifier() != II || CE->getNumArgs() != 3) + return; + + // Get the value of the "theType" argument. + const LocationContext *LCtx = C.getLocationContext(); + SVal TheTypeVal = state->getSVal(CE->getArg(1), LCtx); + + // FIXME: We really should allow ranges of valid theType values, and + // bifurcate the state appropriately. + nonloc::ConcreteInt* V = dyn_cast<nonloc::ConcreteInt>(&TheTypeVal); + if (!V) + return; + + uint64_t NumberKind = V->getValue().getLimitedValue(); + Optional<uint64_t> TargetSize = GetCFNumberSize(Ctx, NumberKind); + + // FIXME: In some cases we can emit an error. + if (!TargetSize.isKnown()) + return; + + // Look at the value of the integer being passed by reference. Essentially + // we want to catch cases where the value passed in is not equal to the + // size of the type being created. + SVal TheValueExpr = state->getSVal(CE->getArg(2), LCtx); + + // FIXME: Eventually we should handle arbitrary locations. We can do this + // by having an enhanced memory model that does low-level typing. + loc::MemRegionVal* LV = dyn_cast<loc::MemRegionVal>(&TheValueExpr); + if (!LV) + return; + + const TypedValueRegion* R = dyn_cast<TypedValueRegion>(LV->stripCasts()); + if (!R) + return; + + QualType T = Ctx.getCanonicalType(R->getValueType()); + + // FIXME: If the pointee isn't an integer type, should we flag a warning? + // People can do weird stuff with pointers. + + if (!T->isIntegerType()) + return; + + uint64_t SourceSize = Ctx.getTypeSize(T); + + // CHECK: is SourceSize == TargetSize + if (SourceSize == TargetSize) + return; + + // Generate an error. Only generate a sink if 'SourceSize < TargetSize'; + // otherwise generate a regular node. + // + // FIXME: We can actually create an abstract "CFNumber" object that has + // the bits initialized to the provided values. + // + if (ExplodedNode *N = SourceSize < TargetSize ? C.generateSink() + : C.addTransition()) { + SmallString<128> sbuf; + llvm::raw_svector_ostream os(sbuf); + + os << (SourceSize == 8 ? "An " : "A ") + << SourceSize << " bit integer is used to initialize a CFNumber " + "object that represents " + << (TargetSize == 8 ? "an " : "a ") + << TargetSize << " bit integer. "; + + if (SourceSize < TargetSize) + os << (TargetSize - SourceSize) + << " bits of the CFNumber value will be garbage." ; + else + os << (SourceSize - TargetSize) + << " bits of the input integer will be lost."; + + if (!BT) + BT.reset(new APIMisuse("Bad use of CFNumberCreate")); + + BugReport *report = new BugReport(*BT, os.str(), N); + report->addRange(CE->getArg(2)->getSourceRange()); + C.EmitReport(report); + } +} + +//===----------------------------------------------------------------------===// +// CFRetain/CFRelease checking for null arguments. +//===----------------------------------------------------------------------===// + +namespace { +class CFRetainReleaseChecker : public Checker< check::PreStmt<CallExpr> > { + mutable OwningPtr<APIMisuse> BT; + mutable IdentifierInfo *Retain, *Release; +public: + CFRetainReleaseChecker(): Retain(0), Release(0) {} + void checkPreStmt(const CallExpr *CE, CheckerContext &C) const; +}; +} // end anonymous namespace + + +void CFRetainReleaseChecker::checkPreStmt(const CallExpr *CE, + CheckerContext &C) const { + // If the CallExpr doesn't have exactly 1 argument just give up checking. + if (CE->getNumArgs() != 1) + return; + + ProgramStateRef state = C.getState(); + const FunctionDecl *FD = C.getCalleeDecl(CE); + if (!FD) + return; + + if (!BT) { + ASTContext &Ctx = C.getASTContext(); + Retain = &Ctx.Idents.get("CFRetain"); + Release = &Ctx.Idents.get("CFRelease"); + BT.reset(new APIMisuse("null passed to CFRetain/CFRelease")); + } + + // Check if we called CFRetain/CFRelease. + const IdentifierInfo *FuncII = FD->getIdentifier(); + if (!(FuncII == Retain || FuncII == Release)) + return; + + // FIXME: The rest of this just checks that the argument is non-null. + // It should probably be refactored and combined with AttrNonNullChecker. + + // Get the argument's value. + const Expr *Arg = CE->getArg(0); + SVal ArgVal = state->getSVal(Arg, C.getLocationContext()); + DefinedSVal *DefArgVal = dyn_cast<DefinedSVal>(&ArgVal); + if (!DefArgVal) + return; + + // Get a NULL value. + SValBuilder &svalBuilder = C.getSValBuilder(); + DefinedSVal zero = cast<DefinedSVal>(svalBuilder.makeZeroVal(Arg->getType())); + + // Make an expression asserting that they're equal. + DefinedOrUnknownSVal ArgIsNull = svalBuilder.evalEQ(state, zero, *DefArgVal); + + // Are they equal? + ProgramStateRef stateTrue, stateFalse; + llvm::tie(stateTrue, stateFalse) = state->assume(ArgIsNull); + + if (stateTrue && !stateFalse) { + ExplodedNode *N = C.generateSink(stateTrue); + if (!N) + return; + + const char *description = (FuncII == Retain) + ? "Null pointer argument in call to CFRetain" + : "Null pointer argument in call to CFRelease"; + + BugReport *report = new BugReport(*BT, description, N); + report->addRange(Arg->getSourceRange()); + report->addVisitor(bugreporter::getTrackNullOrUndefValueVisitor(N, Arg, + report)); + C.EmitReport(report); + return; + } + + // From here on, we know the argument is non-null. + C.addTransition(stateFalse); +} + +//===----------------------------------------------------------------------===// +// Check for sending 'retain', 'release', or 'autorelease' directly to a Class. +//===----------------------------------------------------------------------===// + +namespace { +class ClassReleaseChecker : public Checker<check::PreObjCMessage> { + mutable Selector releaseS; + mutable Selector retainS; + mutable Selector autoreleaseS; + mutable Selector drainS; + mutable OwningPtr<BugType> BT; + +public: + void checkPreObjCMessage(ObjCMessage msg, CheckerContext &C) const; +}; +} + +void ClassReleaseChecker::checkPreObjCMessage(ObjCMessage msg, + CheckerContext &C) const { + + if (!BT) { + BT.reset(new APIMisuse("message incorrectly sent to class instead of class " + "instance")); + + ASTContext &Ctx = C.getASTContext(); + releaseS = GetNullarySelector("release", Ctx); + retainS = GetNullarySelector("retain", Ctx); + autoreleaseS = GetNullarySelector("autorelease", Ctx); + drainS = GetNullarySelector("drain", Ctx); + } + + if (msg.isInstanceMessage()) + return; + const ObjCInterfaceDecl *Class = msg.getReceiverInterface(); + assert(Class); + + Selector S = msg.getSelector(); + if (!(S == releaseS || S == retainS || S == autoreleaseS || S == drainS)) + return; + + if (ExplodedNode *N = C.addTransition()) { + SmallString<200> buf; + llvm::raw_svector_ostream os(buf); + + os << "The '" << S.getAsString() << "' message should be sent to instances " + "of class '" << Class->getName() + << "' and not the class directly"; + + BugReport *report = new BugReport(*BT, os.str(), N); + report->addRange(msg.getSourceRange()); + C.EmitReport(report); + } +} + +//===----------------------------------------------------------------------===// +// Check for passing non-Objective-C types to variadic methods that expect +// only Objective-C types. +//===----------------------------------------------------------------------===// + +namespace { +class VariadicMethodTypeChecker : public Checker<check::PreObjCMessage> { + mutable Selector arrayWithObjectsS; + mutable Selector dictionaryWithObjectsAndKeysS; + mutable Selector setWithObjectsS; + mutable Selector orderedSetWithObjectsS; + mutable Selector initWithObjectsS; + mutable Selector initWithObjectsAndKeysS; + mutable OwningPtr<BugType> BT; + + bool isVariadicMessage(const ObjCMessage &msg) const; + +public: + void checkPreObjCMessage(ObjCMessage msg, CheckerContext &C) const; +}; +} + +/// isVariadicMessage - Returns whether the given message is a variadic message, +/// where all arguments must be Objective-C types. +bool +VariadicMethodTypeChecker::isVariadicMessage(const ObjCMessage &msg) const { + const ObjCMethodDecl *MD = msg.getMethodDecl(); + + if (!MD || !MD->isVariadic() || isa<ObjCProtocolDecl>(MD->getDeclContext())) + return false; + + Selector S = msg.getSelector(); + + if (msg.isInstanceMessage()) { + // FIXME: Ideally we'd look at the receiver interface here, but that's not + // useful for init, because alloc returns 'id'. In theory, this could lead + // to false positives, for example if there existed a class that had an + // initWithObjects: implementation that does accept non-Objective-C pointer + // types, but the chance of that happening is pretty small compared to the + // gains that this analysis gives. + const ObjCInterfaceDecl *Class = MD->getClassInterface(); + + // -[NSArray initWithObjects:] + if (isReceiverClassOrSuperclass(Class, "NSArray") && + S == initWithObjectsS) + return true; + + // -[NSDictionary initWithObjectsAndKeys:] + if (isReceiverClassOrSuperclass(Class, "NSDictionary") && + S == initWithObjectsAndKeysS) + return true; + + // -[NSSet initWithObjects:] + if (isReceiverClassOrSuperclass(Class, "NSSet") && + S == initWithObjectsS) + return true; + + // -[NSOrderedSet initWithObjects:] + if (isReceiverClassOrSuperclass(Class, "NSOrderedSet") && + S == initWithObjectsS) + return true; + } else { + const ObjCInterfaceDecl *Class = msg.getReceiverInterface(); + + // -[NSArray arrayWithObjects:] + if (isReceiverClassOrSuperclass(Class, "NSArray") && + S == arrayWithObjectsS) + return true; + + // -[NSDictionary dictionaryWithObjectsAndKeys:] + if (isReceiverClassOrSuperclass(Class, "NSDictionary") && + S == dictionaryWithObjectsAndKeysS) + return true; + + // -[NSSet setWithObjects:] + if (isReceiverClassOrSuperclass(Class, "NSSet") && + S == setWithObjectsS) + return true; + + // -[NSOrderedSet orderedSetWithObjects:] + if (isReceiverClassOrSuperclass(Class, "NSOrderedSet") && + S == orderedSetWithObjectsS) + return true; + } + + return false; +} + +void VariadicMethodTypeChecker::checkPreObjCMessage(ObjCMessage msg, + CheckerContext &C) const { + if (!BT) { + BT.reset(new APIMisuse("Arguments passed to variadic method aren't all " + "Objective-C pointer types")); + + ASTContext &Ctx = C.getASTContext(); + arrayWithObjectsS = GetUnarySelector("arrayWithObjects", Ctx); + dictionaryWithObjectsAndKeysS = + GetUnarySelector("dictionaryWithObjectsAndKeys", Ctx); + setWithObjectsS = GetUnarySelector("setWithObjects", Ctx); + orderedSetWithObjectsS = GetUnarySelector("orderedSetWithObjects", Ctx); + + initWithObjectsS = GetUnarySelector("initWithObjects", Ctx); + initWithObjectsAndKeysS = GetUnarySelector("initWithObjectsAndKeys", Ctx); + } + + if (!isVariadicMessage(msg)) + return; + + // We are not interested in the selector arguments since they have + // well-defined types, so the compiler will issue a warning for them. + unsigned variadicArgsBegin = msg.getSelector().getNumArgs(); + + // We're not interested in the last argument since it has to be nil or the + // compiler would have issued a warning for it elsewhere. + unsigned variadicArgsEnd = msg.getNumArgs() - 1; + + if (variadicArgsEnd <= variadicArgsBegin) + return; + + // Verify that all arguments have Objective-C types. + llvm::Optional<ExplodedNode*> errorNode; + ProgramStateRef state = C.getState(); + + for (unsigned I = variadicArgsBegin; I != variadicArgsEnd; ++I) { + QualType ArgTy = msg.getArgType(I); + if (ArgTy->isObjCObjectPointerType()) + continue; + + // Block pointers are treaded as Objective-C pointers. + if (ArgTy->isBlockPointerType()) + continue; + + // Ignore pointer constants. + if (isa<loc::ConcreteInt>(msg.getArgSVal(I, C.getLocationContext(), + state))) + continue; + + // Ignore pointer types annotated with 'NSObject' attribute. + if (C.getASTContext().isObjCNSObjectType(ArgTy)) + continue; + + // Ignore CF references, which can be toll-free bridged. + if (coreFoundation::isCFObjectRef(ArgTy)) + continue; + + // Generate only one error node to use for all bug reports. + if (!errorNode.hasValue()) { + errorNode = C.addTransition(); + } + + if (!errorNode.getValue()) + continue; + + SmallString<128> sbuf; + llvm::raw_svector_ostream os(sbuf); + + if (const char *TypeName = GetReceiverNameType(msg)) + os << "Argument to '" << TypeName << "' method '"; + else + os << "Argument to method '"; + + os << msg.getSelector().getAsString() + << "' should be an Objective-C pointer type, not '" + << ArgTy.getAsString() << "'"; + + BugReport *R = new BugReport(*BT, os.str(), + errorNode.getValue()); + R->addRange(msg.getArgSourceRange(I)); + C.EmitReport(R); + } +} + +//===----------------------------------------------------------------------===// +// Check registration. +//===----------------------------------------------------------------------===// + +void ento::registerNilArgChecker(CheckerManager &mgr) { + mgr.registerChecker<NilArgChecker>(); +} + +void ento::registerCFNumberCreateChecker(CheckerManager &mgr) { + mgr.registerChecker<CFNumberCreateChecker>(); +} + +void ento::registerCFRetainReleaseChecker(CheckerManager &mgr) { + mgr.registerChecker<CFRetainReleaseChecker>(); +} + +void ento::registerClassReleaseChecker(CheckerManager &mgr) { + mgr.registerChecker<ClassReleaseChecker>(); +} + +void ento::registerVariadicMethodTypeChecker(CheckerManager &mgr) { + mgr.registerChecker<VariadicMethodTypeChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/BoolAssignmentChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/BoolAssignmentChecker.cpp new file mode 100644 index 0000000..a4fc396 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/BoolAssignmentChecker.cpp @@ -0,0 +1,157 @@ +//== BoolAssignmentChecker.cpp - Boolean assignment checker -----*- C++ -*--==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This defines BoolAssignmentChecker, a builtin check in ExprEngine that +// performs checks for assignment of non-Boolean values to Boolean variables. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" + +using namespace clang; +using namespace ento; + +namespace { + class BoolAssignmentChecker : public Checker< check::Bind > { + mutable llvm::OwningPtr<BuiltinBug> BT; + void emitReport(ProgramStateRef state, CheckerContext &C) const; + public: + void checkBind(SVal loc, SVal val, const Stmt *S, CheckerContext &C) const; + }; +} // end anonymous namespace + +void BoolAssignmentChecker::emitReport(ProgramStateRef state, + CheckerContext &C) const { + if (ExplodedNode *N = C.addTransition(state)) { + if (!BT) + BT.reset(new BuiltinBug("Assignment of a non-Boolean value")); + C.EmitReport(new BugReport(*BT, BT->getDescription(), N)); + } +} + +static bool isBooleanType(QualType Ty) { + if (Ty->isBooleanType()) // C++ or C99 + return true; + + if (const TypedefType *TT = Ty->getAs<TypedefType>()) + return TT->getDecl()->getName() == "BOOL" || // Objective-C + TT->getDecl()->getName() == "_Bool" || // stdbool.h < C99 + TT->getDecl()->getName() == "Boolean"; // MacTypes.h + + return false; +} + +void BoolAssignmentChecker::checkBind(SVal loc, SVal val, const Stmt *S, + CheckerContext &C) const { + + // We are only interested in stores into Booleans. + const TypedValueRegion *TR = + dyn_cast_or_null<TypedValueRegion>(loc.getAsRegion()); + + if (!TR) + return; + + QualType valTy = TR->getValueType(); + + if (!isBooleanType(valTy)) + return; + + // Get the value of the right-hand side. We only care about values + // that are defined (UnknownVals and UndefinedVals are handled by other + // checkers). + const DefinedSVal *DV = dyn_cast<DefinedSVal>(&val); + if (!DV) + return; + + // Check if the assigned value meets our criteria for correctness. It must + // be a value that is either 0 or 1. One way to check this is to see if + // the value is possibly < 0 (for a negative value) or greater than 1. + ProgramStateRef state = C.getState(); + SValBuilder &svalBuilder = C.getSValBuilder(); + ConstraintManager &CM = C.getConstraintManager(); + + // First, ensure that the value is >= 0. + DefinedSVal zeroVal = svalBuilder.makeIntVal(0, valTy); + SVal greaterThanOrEqualToZeroVal = + svalBuilder.evalBinOp(state, BO_GE, *DV, zeroVal, + svalBuilder.getConditionType()); + + DefinedSVal *greaterThanEqualToZero = + dyn_cast<DefinedSVal>(&greaterThanOrEqualToZeroVal); + + if (!greaterThanEqualToZero) { + // The SValBuilder cannot construct a valid SVal for this condition. + // This means we cannot properly reason about it. + return; + } + + ProgramStateRef stateLT, stateGE; + llvm::tie(stateGE, stateLT) = CM.assumeDual(state, *greaterThanEqualToZero); + + // Is it possible for the value to be less than zero? + if (stateLT) { + // It is possible for the value to be less than zero. We only + // want to emit a warning, however, if that value is fully constrained. + // If it it possible for the value to be >= 0, then essentially the + // value is underconstrained and there is nothing left to be done. + if (!stateGE) + emitReport(stateLT, C); + + // In either case, we are done. + return; + } + + // If we reach here, it must be the case that the value is constrained + // to only be >= 0. + assert(stateGE == state); + + // At this point we know that the value is >= 0. + // Now check to ensure that the value is <= 1. + DefinedSVal OneVal = svalBuilder.makeIntVal(1, valTy); + SVal lessThanEqToOneVal = + svalBuilder.evalBinOp(state, BO_LE, *DV, OneVal, + svalBuilder.getConditionType()); + + DefinedSVal *lessThanEqToOne = + dyn_cast<DefinedSVal>(&lessThanEqToOneVal); + + if (!lessThanEqToOne) { + // The SValBuilder cannot construct a valid SVal for this condition. + // This means we cannot properly reason about it. + return; + } + + ProgramStateRef stateGT, stateLE; + llvm::tie(stateLE, stateGT) = CM.assumeDual(state, *lessThanEqToOne); + + // Is it possible for the value to be greater than one? + if (stateGT) { + // It is possible for the value to be greater than one. We only + // want to emit a warning, however, if that value is fully constrained. + // If it is possible for the value to be <= 1, then essentially the + // value is underconstrained and there is nothing left to be done. + if (!stateLE) + emitReport(stateGT, C); + + // In either case, we are done. + return; + } + + // If we reach here, it must be the case that the value is constrained + // to only be <= 1. + assert(stateLE == state); +} + +void ento::registerBoolAssignmentChecker(CheckerManager &mgr) { + mgr.registerChecker<BoolAssignmentChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/BuiltinFunctionChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/BuiltinFunctionChecker.cpp new file mode 100644 index 0000000..509bc79 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/BuiltinFunctionChecker.cpp @@ -0,0 +1,82 @@ +//=== BuiltinFunctionChecker.cpp --------------------------------*- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This checker evaluates clang builtin functions. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/Basic/Builtins.h" + +using namespace clang; +using namespace ento; + +namespace { + +class BuiltinFunctionChecker : public Checker<eval::Call> { +public: + bool evalCall(const CallExpr *CE, CheckerContext &C) const; +}; + +} + +bool BuiltinFunctionChecker::evalCall(const CallExpr *CE, + CheckerContext &C) const { + ProgramStateRef state = C.getState(); + const FunctionDecl *FD = C.getCalleeDecl(CE); + const LocationContext *LCtx = C.getLocationContext(); + if (!FD) + return false; + + unsigned id = FD->getBuiltinID(); + + if (!id) + return false; + + switch (id) { + case Builtin::BI__builtin_expect: { + // For __builtin_expect, just return the value of the subexpression. + assert (CE->arg_begin() != CE->arg_end()); + SVal X = state->getSVal(*(CE->arg_begin()), LCtx); + C.addTransition(state->BindExpr(CE, LCtx, X)); + return true; + } + + case Builtin::BI__builtin_alloca: { + // FIXME: Refactor into StoreManager itself? + MemRegionManager& RM = C.getStoreManager().getRegionManager(); + const AllocaRegion* R = + RM.getAllocaRegion(CE, C.getCurrentBlockCount(), C.getLocationContext()); + + // Set the extent of the region in bytes. This enables us to use the + // SVal of the argument directly. If we save the extent in bits, we + // cannot represent values like symbol*8. + DefinedOrUnknownSVal Size = + cast<DefinedOrUnknownSVal>(state->getSVal(*(CE->arg_begin()), LCtx)); + + SValBuilder& svalBuilder = C.getSValBuilder(); + DefinedOrUnknownSVal Extent = R->getExtent(svalBuilder); + DefinedOrUnknownSVal extentMatchesSizeArg = + svalBuilder.evalEQ(state, Extent, Size); + state = state->assume(extentMatchesSizeArg, true); + + C.addTransition(state->BindExpr(CE, LCtx, loc::MemRegionVal(R))); + return true; + } + } + + return false; +} + +void ento::registerBuiltinFunctionChecker(CheckerManager &mgr) { + mgr.registerChecker<BuiltinFunctionChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/CMakeLists.txt b/clang/lib/StaticAnalyzer/Checkers/CMakeLists.txt new file mode 100644 index 0000000..75d5448 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/CMakeLists.txt @@ -0,0 +1,81 @@ +clang_tablegen(Checkers.inc -gen-clang-sa-checkers + -I ${CMAKE_CURRENT_SOURCE_DIR}/../../../include + SOURCE Checkers.td + TARGET ClangSACheckers) + +set(LLVM_USED_LIBS clangBasic clangAST clangStaticAnalyzerCore) + +add_clang_library(clangStaticAnalyzerCheckers + AdjustedReturnValueChecker.cpp + AnalyzerStatsChecker.cpp + ArrayBoundChecker.cpp + ArrayBoundCheckerV2.cpp + AttrNonNullChecker.cpp + BasicObjCFoundationChecks.cpp + BoolAssignmentChecker.cpp + BuiltinFunctionChecker.cpp + CStringChecker.cpp + CStringSyntaxChecker.cpp + CallAndMessageChecker.cpp + CastSizeChecker.cpp + CastToStructChecker.cpp + CheckObjCDealloc.cpp + CheckObjCInstMethSignature.cpp + CheckSecuritySyntaxOnly.cpp + CheckSizeofPointer.cpp + CheckerDocumentation.cpp + ChrootChecker.cpp + ClangCheckers.cpp + CommonBugCategories.cpp + DeadStoresChecker.cpp + DebugCheckers.cpp + DereferenceChecker.cpp + DivZeroChecker.cpp + FixedAddressChecker.cpp + GenericTaintChecker.cpp + IdempotentOperationChecker.cpp + IteratorsChecker.cpp + IntervalTest.cpp + LLVMConventionsChecker.cpp + MacOSKeychainAPIChecker.cpp + MacOSXAPIChecker.cpp + MallocChecker.cpp + MallocOverflowSecurityChecker.cpp + MallocSizeofChecker.cpp + NSAutoreleasePoolChecker.cpp + NSErrorChecker.cpp + NoReturnFunctionChecker.cpp + OSAtomicChecker.cpp + ObjCAtSyncChecker.cpp + ObjCContainersASTChecker.cpp + ObjCContainersChecker.cpp + ObjCSelfInitChecker.cpp + ObjCUnusedIVarsChecker.cpp + PointerArithChecker.cpp + PointerSubChecker.cpp + PthreadLockChecker.cpp + RetainCountChecker.cpp + ReturnPointerRangeChecker.cpp + ReturnUndefChecker.cpp + StackAddrEscapeChecker.cpp + StreamChecker.cpp + TaintTesterChecker.cpp + UndefBranchChecker.cpp + UndefCapturedBlockVarChecker.cpp + UndefResultChecker.cpp + UndefinedArraySubscriptChecker.cpp + UndefinedAssignmentChecker.cpp + UnixAPIChecker.cpp + UnreachableCodeChecker.cpp + VLASizeChecker.cpp + VirtualCallChecker.cpp + ) + +add_dependencies(clangStaticAnalyzerCheckers + clangStaticAnalyzerCore + ClangAttrClasses + ClangAttrList + ClangDeclNodes + ClangStmtNodes + ClangSACheckers + ) diff --git a/clang/lib/StaticAnalyzer/Checkers/CStringChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/CStringChecker.cpp new file mode 100644 index 0000000..9eb7edf --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/CStringChecker.cpp @@ -0,0 +1,1981 @@ +//= CStringChecker.cpp - Checks calls to C string functions --------*- C++ -*-// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This defines CStringChecker, which is an assortment of checks on calls +// to functions in <string.h>. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "InterCheckerAPI.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramStateTrait.h" +#include "llvm/ADT/SmallString.h" +#include "llvm/ADT/STLExtras.h" +#include "llvm/ADT/StringSwitch.h" + +using namespace clang; +using namespace ento; + +namespace { +class CStringChecker : public Checker< eval::Call, + check::PreStmt<DeclStmt>, + check::LiveSymbols, + check::DeadSymbols, + check::RegionChanges + > { + mutable OwningPtr<BugType> BT_Null, + BT_Bounds, + BT_Overlap, + BT_NotCString, + BT_AdditionOverflow; + + mutable const char *CurrentFunctionDescription; + +public: + /// The filter is used to filter out the diagnostics which are not enabled by + /// the user. + struct CStringChecksFilter { + DefaultBool CheckCStringNullArg; + DefaultBool CheckCStringOutOfBounds; + DefaultBool CheckCStringBufferOverlap; + DefaultBool CheckCStringNotNullTerm; + }; + + CStringChecksFilter Filter; + + static void *getTag() { static int tag; return &tag; } + + bool evalCall(const CallExpr *CE, CheckerContext &C) const; + void checkPreStmt(const DeclStmt *DS, CheckerContext &C) const; + void checkLiveSymbols(ProgramStateRef state, SymbolReaper &SR) const; + void checkDeadSymbols(SymbolReaper &SR, CheckerContext &C) const; + bool wantsRegionChangeUpdate(ProgramStateRef state) const; + + ProgramStateRef + checkRegionChanges(ProgramStateRef state, + const StoreManager::InvalidatedSymbols *, + ArrayRef<const MemRegion *> ExplicitRegions, + ArrayRef<const MemRegion *> Regions, + const CallOrObjCMessage *Call) const; + + typedef void (CStringChecker::*FnCheck)(CheckerContext &, + const CallExpr *) const; + + void evalMemcpy(CheckerContext &C, const CallExpr *CE) const; + void evalMempcpy(CheckerContext &C, const CallExpr *CE) const; + void evalMemmove(CheckerContext &C, const CallExpr *CE) const; + void evalBcopy(CheckerContext &C, const CallExpr *CE) const; + void evalCopyCommon(CheckerContext &C, const CallExpr *CE, + ProgramStateRef state, + const Expr *Size, + const Expr *Source, + const Expr *Dest, + bool Restricted = false, + bool IsMempcpy = false) const; + + void evalMemcmp(CheckerContext &C, const CallExpr *CE) const; + + void evalstrLength(CheckerContext &C, const CallExpr *CE) const; + void evalstrnLength(CheckerContext &C, const CallExpr *CE) const; + void evalstrLengthCommon(CheckerContext &C, + const CallExpr *CE, + bool IsStrnlen = false) const; + + void evalStrcpy(CheckerContext &C, const CallExpr *CE) const; + void evalStrncpy(CheckerContext &C, const CallExpr *CE) const; + void evalStpcpy(CheckerContext &C, const CallExpr *CE) const; + void evalStrcpyCommon(CheckerContext &C, + const CallExpr *CE, + bool returnEnd, + bool isBounded, + bool isAppending) const; + + void evalStrcat(CheckerContext &C, const CallExpr *CE) const; + void evalStrncat(CheckerContext &C, const CallExpr *CE) const; + + void evalStrcmp(CheckerContext &C, const CallExpr *CE) const; + void evalStrncmp(CheckerContext &C, const CallExpr *CE) const; + void evalStrcasecmp(CheckerContext &C, const CallExpr *CE) const; + void evalStrncasecmp(CheckerContext &C, const CallExpr *CE) const; + void evalStrcmpCommon(CheckerContext &C, + const CallExpr *CE, + bool isBounded = false, + bool ignoreCase = false) const; + + // Utility methods + std::pair<ProgramStateRef , ProgramStateRef > + static assumeZero(CheckerContext &C, + ProgramStateRef state, SVal V, QualType Ty); + + static ProgramStateRef setCStringLength(ProgramStateRef state, + const MemRegion *MR, + SVal strLength); + static SVal getCStringLengthForRegion(CheckerContext &C, + ProgramStateRef &state, + const Expr *Ex, + const MemRegion *MR, + bool hypothetical); + SVal getCStringLength(CheckerContext &C, + ProgramStateRef &state, + const Expr *Ex, + SVal Buf, + bool hypothetical = false) const; + + const StringLiteral *getCStringLiteral(CheckerContext &C, + ProgramStateRef &state, + const Expr *expr, + SVal val) const; + + static ProgramStateRef InvalidateBuffer(CheckerContext &C, + ProgramStateRef state, + const Expr *Ex, SVal V); + + static bool SummarizeRegion(raw_ostream &os, ASTContext &Ctx, + const MemRegion *MR); + + // Re-usable checks + ProgramStateRef checkNonNull(CheckerContext &C, + ProgramStateRef state, + const Expr *S, + SVal l) const; + ProgramStateRef CheckLocation(CheckerContext &C, + ProgramStateRef state, + const Expr *S, + SVal l, + const char *message = NULL) const; + ProgramStateRef CheckBufferAccess(CheckerContext &C, + ProgramStateRef state, + const Expr *Size, + const Expr *FirstBuf, + const Expr *SecondBuf, + const char *firstMessage = NULL, + const char *secondMessage = NULL, + bool WarnAboutSize = false) const; + + ProgramStateRef CheckBufferAccess(CheckerContext &C, + ProgramStateRef state, + const Expr *Size, + const Expr *Buf, + const char *message = NULL, + bool WarnAboutSize = false) const { + // This is a convenience override. + return CheckBufferAccess(C, state, Size, Buf, NULL, message, NULL, + WarnAboutSize); + } + ProgramStateRef CheckOverlap(CheckerContext &C, + ProgramStateRef state, + const Expr *Size, + const Expr *First, + const Expr *Second) const; + void emitOverlapBug(CheckerContext &C, + ProgramStateRef state, + const Stmt *First, + const Stmt *Second) const; + + ProgramStateRef checkAdditionOverflow(CheckerContext &C, + ProgramStateRef state, + NonLoc left, + NonLoc right) const; +}; + +class CStringLength { +public: + typedef llvm::ImmutableMap<const MemRegion *, SVal> EntryMap; +}; +} //end anonymous namespace + +namespace clang { +namespace ento { + template <> + struct ProgramStateTrait<CStringLength> + : public ProgramStatePartialTrait<CStringLength::EntryMap> { + static void *GDMIndex() { return CStringChecker::getTag(); } + }; +} +} + +//===----------------------------------------------------------------------===// +// Individual checks and utility methods. +//===----------------------------------------------------------------------===// + +std::pair<ProgramStateRef , ProgramStateRef > +CStringChecker::assumeZero(CheckerContext &C, ProgramStateRef state, SVal V, + QualType Ty) { + DefinedSVal *val = dyn_cast<DefinedSVal>(&V); + if (!val) + return std::pair<ProgramStateRef , ProgramStateRef >(state, state); + + SValBuilder &svalBuilder = C.getSValBuilder(); + DefinedOrUnknownSVal zero = svalBuilder.makeZeroVal(Ty); + return state->assume(svalBuilder.evalEQ(state, *val, zero)); +} + +ProgramStateRef CStringChecker::checkNonNull(CheckerContext &C, + ProgramStateRef state, + const Expr *S, SVal l) const { + // If a previous check has failed, propagate the failure. + if (!state) + return NULL; + + ProgramStateRef stateNull, stateNonNull; + llvm::tie(stateNull, stateNonNull) = assumeZero(C, state, l, S->getType()); + + if (stateNull && !stateNonNull) { + if (!Filter.CheckCStringNullArg) + return NULL; + + ExplodedNode *N = C.generateSink(stateNull); + if (!N) + return NULL; + + if (!BT_Null) + BT_Null.reset(new BuiltinBug("Unix API", + "Null pointer argument in call to byte string function")); + + SmallString<80> buf; + llvm::raw_svector_ostream os(buf); + assert(CurrentFunctionDescription); + os << "Null pointer argument in call to " << CurrentFunctionDescription; + + // Generate a report for this bug. + BuiltinBug *BT = static_cast<BuiltinBug*>(BT_Null.get()); + BugReport *report = new BugReport(*BT, os.str(), N); + + report->addRange(S->getSourceRange()); + report->addVisitor(bugreporter::getTrackNullOrUndefValueVisitor(N, S, + report)); + C.EmitReport(report); + return NULL; + } + + // From here on, assume that the value is non-null. + assert(stateNonNull); + return stateNonNull; +} + +// FIXME: This was originally copied from ArrayBoundChecker.cpp. Refactor? +ProgramStateRef CStringChecker::CheckLocation(CheckerContext &C, + ProgramStateRef state, + const Expr *S, SVal l, + const char *warningMsg) const { + // If a previous check has failed, propagate the failure. + if (!state) + return NULL; + + // Check for out of bound array element access. + const MemRegion *R = l.getAsRegion(); + if (!R) + return state; + + const ElementRegion *ER = dyn_cast<ElementRegion>(R); + if (!ER) + return state; + + assert(ER->getValueType() == C.getASTContext().CharTy && + "CheckLocation should only be called with char* ElementRegions"); + + // Get the size of the array. + const SubRegion *superReg = cast<SubRegion>(ER->getSuperRegion()); + SValBuilder &svalBuilder = C.getSValBuilder(); + SVal Extent = + svalBuilder.convertToArrayIndex(superReg->getExtent(svalBuilder)); + DefinedOrUnknownSVal Size = cast<DefinedOrUnknownSVal>(Extent); + + // Get the index of the accessed element. + DefinedOrUnknownSVal Idx = cast<DefinedOrUnknownSVal>(ER->getIndex()); + + ProgramStateRef StInBound = state->assumeInBound(Idx, Size, true); + ProgramStateRef StOutBound = state->assumeInBound(Idx, Size, false); + if (StOutBound && !StInBound) { + ExplodedNode *N = C.generateSink(StOutBound); + if (!N) + return NULL; + + if (!BT_Bounds) { + BT_Bounds.reset(new BuiltinBug("Out-of-bound array access", + "Byte string function accesses out-of-bound array element")); + } + BuiltinBug *BT = static_cast<BuiltinBug*>(BT_Bounds.get()); + + // Generate a report for this bug. + BugReport *report; + if (warningMsg) { + report = new BugReport(*BT, warningMsg, N); + } else { + assert(CurrentFunctionDescription); + assert(CurrentFunctionDescription[0] != '\0'); + + SmallString<80> buf; + llvm::raw_svector_ostream os(buf); + os << (char)toupper(CurrentFunctionDescription[0]) + << &CurrentFunctionDescription[1] + << " accesses out-of-bound array element"; + report = new BugReport(*BT, os.str(), N); + } + + // FIXME: It would be nice to eventually make this diagnostic more clear, + // e.g., by referencing the original declaration or by saying *why* this + // reference is outside the range. + + report->addRange(S->getSourceRange()); + C.EmitReport(report); + return NULL; + } + + // Array bound check succeeded. From this point forward the array bound + // should always succeed. + return StInBound; +} + +ProgramStateRef CStringChecker::CheckBufferAccess(CheckerContext &C, + ProgramStateRef state, + const Expr *Size, + const Expr *FirstBuf, + const Expr *SecondBuf, + const char *firstMessage, + const char *secondMessage, + bool WarnAboutSize) const { + // If a previous check has failed, propagate the failure. + if (!state) + return NULL; + + SValBuilder &svalBuilder = C.getSValBuilder(); + ASTContext &Ctx = svalBuilder.getContext(); + const LocationContext *LCtx = C.getLocationContext(); + + QualType sizeTy = Size->getType(); + QualType PtrTy = Ctx.getPointerType(Ctx.CharTy); + + // Check that the first buffer is non-null. + SVal BufVal = state->getSVal(FirstBuf, LCtx); + state = checkNonNull(C, state, FirstBuf, BufVal); + if (!state) + return NULL; + + // If out-of-bounds checking is turned off, skip the rest. + if (!Filter.CheckCStringOutOfBounds) + return state; + + // Get the access length and make sure it is known. + // FIXME: This assumes the caller has already checked that the access length + // is positive. And that it's unsigned. + SVal LengthVal = state->getSVal(Size, LCtx); + NonLoc *Length = dyn_cast<NonLoc>(&LengthVal); + if (!Length) + return state; + + // Compute the offset of the last element to be accessed: size-1. + NonLoc One = cast<NonLoc>(svalBuilder.makeIntVal(1, sizeTy)); + NonLoc LastOffset = cast<NonLoc>(svalBuilder.evalBinOpNN(state, BO_Sub, + *Length, One, sizeTy)); + + // Check that the first buffer is sufficiently long. + SVal BufStart = svalBuilder.evalCast(BufVal, PtrTy, FirstBuf->getType()); + if (Loc *BufLoc = dyn_cast<Loc>(&BufStart)) { + const Expr *warningExpr = (WarnAboutSize ? Size : FirstBuf); + + SVal BufEnd = svalBuilder.evalBinOpLN(state, BO_Add, *BufLoc, + LastOffset, PtrTy); + state = CheckLocation(C, state, warningExpr, BufEnd, firstMessage); + + // If the buffer isn't large enough, abort. + if (!state) + return NULL; + } + + // If there's a second buffer, check it as well. + if (SecondBuf) { + BufVal = state->getSVal(SecondBuf, LCtx); + state = checkNonNull(C, state, SecondBuf, BufVal); + if (!state) + return NULL; + + BufStart = svalBuilder.evalCast(BufVal, PtrTy, SecondBuf->getType()); + if (Loc *BufLoc = dyn_cast<Loc>(&BufStart)) { + const Expr *warningExpr = (WarnAboutSize ? Size : SecondBuf); + + SVal BufEnd = svalBuilder.evalBinOpLN(state, BO_Add, *BufLoc, + LastOffset, PtrTy); + state = CheckLocation(C, state, warningExpr, BufEnd, secondMessage); + } + } + + // Large enough or not, return this state! + return state; +} + +ProgramStateRef CStringChecker::CheckOverlap(CheckerContext &C, + ProgramStateRef state, + const Expr *Size, + const Expr *First, + const Expr *Second) const { + if (!Filter.CheckCStringBufferOverlap) + return state; + + // Do a simple check for overlap: if the two arguments are from the same + // buffer, see if the end of the first is greater than the start of the second + // or vice versa. + + // If a previous check has failed, propagate the failure. + if (!state) + return NULL; + + ProgramStateRef stateTrue, stateFalse; + + // Get the buffer values and make sure they're known locations. + const LocationContext *LCtx = C.getLocationContext(); + SVal firstVal = state->getSVal(First, LCtx); + SVal secondVal = state->getSVal(Second, LCtx); + + Loc *firstLoc = dyn_cast<Loc>(&firstVal); + if (!firstLoc) + return state; + + Loc *secondLoc = dyn_cast<Loc>(&secondVal); + if (!secondLoc) + return state; + + // Are the two values the same? + SValBuilder &svalBuilder = C.getSValBuilder(); + llvm::tie(stateTrue, stateFalse) = + state->assume(svalBuilder.evalEQ(state, *firstLoc, *secondLoc)); + + if (stateTrue && !stateFalse) { + // If the values are known to be equal, that's automatically an overlap. + emitOverlapBug(C, stateTrue, First, Second); + return NULL; + } + + // assume the two expressions are not equal. + assert(stateFalse); + state = stateFalse; + + // Which value comes first? + QualType cmpTy = svalBuilder.getConditionType(); + SVal reverse = svalBuilder.evalBinOpLL(state, BO_GT, + *firstLoc, *secondLoc, cmpTy); + DefinedOrUnknownSVal *reverseTest = dyn_cast<DefinedOrUnknownSVal>(&reverse); + if (!reverseTest) + return state; + + llvm::tie(stateTrue, stateFalse) = state->assume(*reverseTest); + if (stateTrue) { + if (stateFalse) { + // If we don't know which one comes first, we can't perform this test. + return state; + } else { + // Switch the values so that firstVal is before secondVal. + Loc *tmpLoc = firstLoc; + firstLoc = secondLoc; + secondLoc = tmpLoc; + + // Switch the Exprs as well, so that they still correspond. + const Expr *tmpExpr = First; + First = Second; + Second = tmpExpr; + } + } + + // Get the length, and make sure it too is known. + SVal LengthVal = state->getSVal(Size, LCtx); + NonLoc *Length = dyn_cast<NonLoc>(&LengthVal); + if (!Length) + return state; + + // Convert the first buffer's start address to char*. + // Bail out if the cast fails. + ASTContext &Ctx = svalBuilder.getContext(); + QualType CharPtrTy = Ctx.getPointerType(Ctx.CharTy); + SVal FirstStart = svalBuilder.evalCast(*firstLoc, CharPtrTy, + First->getType()); + Loc *FirstStartLoc = dyn_cast<Loc>(&FirstStart); + if (!FirstStartLoc) + return state; + + // Compute the end of the first buffer. Bail out if THAT fails. + SVal FirstEnd = svalBuilder.evalBinOpLN(state, BO_Add, + *FirstStartLoc, *Length, CharPtrTy); + Loc *FirstEndLoc = dyn_cast<Loc>(&FirstEnd); + if (!FirstEndLoc) + return state; + + // Is the end of the first buffer past the start of the second buffer? + SVal Overlap = svalBuilder.evalBinOpLL(state, BO_GT, + *FirstEndLoc, *secondLoc, cmpTy); + DefinedOrUnknownSVal *OverlapTest = dyn_cast<DefinedOrUnknownSVal>(&Overlap); + if (!OverlapTest) + return state; + + llvm::tie(stateTrue, stateFalse) = state->assume(*OverlapTest); + + if (stateTrue && !stateFalse) { + // Overlap! + emitOverlapBug(C, stateTrue, First, Second); + return NULL; + } + + // assume the two expressions don't overlap. + assert(stateFalse); + return stateFalse; +} + +void CStringChecker::emitOverlapBug(CheckerContext &C, ProgramStateRef state, + const Stmt *First, const Stmt *Second) const { + ExplodedNode *N = C.generateSink(state); + if (!N) + return; + + if (!BT_Overlap) + BT_Overlap.reset(new BugType("Unix API", "Improper arguments")); + + // Generate a report for this bug. + BugReport *report = + new BugReport(*BT_Overlap, + "Arguments must not be overlapping buffers", N); + report->addRange(First->getSourceRange()); + report->addRange(Second->getSourceRange()); + + C.EmitReport(report); +} + +ProgramStateRef CStringChecker::checkAdditionOverflow(CheckerContext &C, + ProgramStateRef state, + NonLoc left, + NonLoc right) const { + // If out-of-bounds checking is turned off, skip the rest. + if (!Filter.CheckCStringOutOfBounds) + return state; + + // If a previous check has failed, propagate the failure. + if (!state) + return NULL; + + SValBuilder &svalBuilder = C.getSValBuilder(); + BasicValueFactory &BVF = svalBuilder.getBasicValueFactory(); + + QualType sizeTy = svalBuilder.getContext().getSizeType(); + const llvm::APSInt &maxValInt = BVF.getMaxValue(sizeTy); + NonLoc maxVal = svalBuilder.makeIntVal(maxValInt); + + SVal maxMinusRight; + if (isa<nonloc::ConcreteInt>(right)) { + maxMinusRight = svalBuilder.evalBinOpNN(state, BO_Sub, maxVal, right, + sizeTy); + } else { + // Try switching the operands. (The order of these two assignments is + // important!) + maxMinusRight = svalBuilder.evalBinOpNN(state, BO_Sub, maxVal, left, + sizeTy); + left = right; + } + + if (NonLoc *maxMinusRightNL = dyn_cast<NonLoc>(&maxMinusRight)) { + QualType cmpTy = svalBuilder.getConditionType(); + // If left > max - right, we have an overflow. + SVal willOverflow = svalBuilder.evalBinOpNN(state, BO_GT, left, + *maxMinusRightNL, cmpTy); + + ProgramStateRef stateOverflow, stateOkay; + llvm::tie(stateOverflow, stateOkay) = + state->assume(cast<DefinedOrUnknownSVal>(willOverflow)); + + if (stateOverflow && !stateOkay) { + // We have an overflow. Emit a bug report. + ExplodedNode *N = C.generateSink(stateOverflow); + if (!N) + return NULL; + + if (!BT_AdditionOverflow) + BT_AdditionOverflow.reset(new BuiltinBug("API", + "Sum of expressions causes overflow")); + + // This isn't a great error message, but this should never occur in real + // code anyway -- you'd have to create a buffer longer than a size_t can + // represent, which is sort of a contradiction. + const char *warning = + "This expression will create a string whose length is too big to " + "be represented as a size_t"; + + // Generate a report for this bug. + BugReport *report = new BugReport(*BT_AdditionOverflow, warning, N); + C.EmitReport(report); + + return NULL; + } + + // From now on, assume an overflow didn't occur. + assert(stateOkay); + state = stateOkay; + } + + return state; +} + +ProgramStateRef CStringChecker::setCStringLength(ProgramStateRef state, + const MemRegion *MR, + SVal strLength) { + assert(!strLength.isUndef() && "Attempt to set an undefined string length"); + + MR = MR->StripCasts(); + + switch (MR->getKind()) { + case MemRegion::StringRegionKind: + // FIXME: This can happen if we strcpy() into a string region. This is + // undefined [C99 6.4.5p6], but we should still warn about it. + return state; + + case MemRegion::SymbolicRegionKind: + case MemRegion::AllocaRegionKind: + case MemRegion::VarRegionKind: + case MemRegion::FieldRegionKind: + case MemRegion::ObjCIvarRegionKind: + // These are the types we can currently track string lengths for. + break; + + case MemRegion::ElementRegionKind: + // FIXME: Handle element regions by upper-bounding the parent region's + // string length. + return state; + + default: + // Other regions (mostly non-data) can't have a reliable C string length. + // For now, just ignore the change. + // FIXME: These are rare but not impossible. We should output some kind of + // warning for things like strcpy((char[]){'a', 0}, "b"); + return state; + } + + if (strLength.isUnknown()) + return state->remove<CStringLength>(MR); + + return state->set<CStringLength>(MR, strLength); +} + +SVal CStringChecker::getCStringLengthForRegion(CheckerContext &C, + ProgramStateRef &state, + const Expr *Ex, + const MemRegion *MR, + bool hypothetical) { + if (!hypothetical) { + // If there's a recorded length, go ahead and return it. + const SVal *Recorded = state->get<CStringLength>(MR); + if (Recorded) + return *Recorded; + } + + // Otherwise, get a new symbol and update the state. + unsigned Count = C.getCurrentBlockCount(); + SValBuilder &svalBuilder = C.getSValBuilder(); + QualType sizeTy = svalBuilder.getContext().getSizeType(); + SVal strLength = svalBuilder.getMetadataSymbolVal(CStringChecker::getTag(), + MR, Ex, sizeTy, Count); + + if (!hypothetical) + state = state->set<CStringLength>(MR, strLength); + + return strLength; +} + +SVal CStringChecker::getCStringLength(CheckerContext &C, ProgramStateRef &state, + const Expr *Ex, SVal Buf, + bool hypothetical) const { + const MemRegion *MR = Buf.getAsRegion(); + if (!MR) { + // If we can't get a region, see if it's something we /know/ isn't a + // C string. In the context of locations, the only time we can issue such + // a warning is for labels. + if (loc::GotoLabel *Label = dyn_cast<loc::GotoLabel>(&Buf)) { + if (!Filter.CheckCStringNotNullTerm) + return UndefinedVal(); + + if (ExplodedNode *N = C.addTransition(state)) { + if (!BT_NotCString) + BT_NotCString.reset(new BuiltinBug("Unix API", + "Argument is not a null-terminated string.")); + + SmallString<120> buf; + llvm::raw_svector_ostream os(buf); + assert(CurrentFunctionDescription); + os << "Argument to " << CurrentFunctionDescription + << " is the address of the label '" << Label->getLabel()->getName() + << "', which is not a null-terminated string"; + + // Generate a report for this bug. + BugReport *report = new BugReport(*BT_NotCString, + os.str(), N); + + report->addRange(Ex->getSourceRange()); + C.EmitReport(report); + } + return UndefinedVal(); + + } + + // If it's not a region and not a label, give up. + return UnknownVal(); + } + + // If we have a region, strip casts from it and see if we can figure out + // its length. For anything we can't figure out, just return UnknownVal. + MR = MR->StripCasts(); + + switch (MR->getKind()) { + case MemRegion::StringRegionKind: { + // Modifying the contents of string regions is undefined [C99 6.4.5p6], + // so we can assume that the byte length is the correct C string length. + SValBuilder &svalBuilder = C.getSValBuilder(); + QualType sizeTy = svalBuilder.getContext().getSizeType(); + const StringLiteral *strLit = cast<StringRegion>(MR)->getStringLiteral(); + return svalBuilder.makeIntVal(strLit->getByteLength(), sizeTy); + } + case MemRegion::SymbolicRegionKind: + case MemRegion::AllocaRegionKind: + case MemRegion::VarRegionKind: + case MemRegion::FieldRegionKind: + case MemRegion::ObjCIvarRegionKind: + return getCStringLengthForRegion(C, state, Ex, MR, hypothetical); + case MemRegion::CompoundLiteralRegionKind: + // FIXME: Can we track this? Is it necessary? + return UnknownVal(); + case MemRegion::ElementRegionKind: + // FIXME: How can we handle this? It's not good enough to subtract the + // offset from the base string length; consider "123\x00567" and &a[5]. + return UnknownVal(); + default: + // Other regions (mostly non-data) can't have a reliable C string length. + // In this case, an error is emitted and UndefinedVal is returned. + // The caller should always be prepared to handle this case. + if (!Filter.CheckCStringNotNullTerm) + return UndefinedVal(); + + if (ExplodedNode *N = C.addTransition(state)) { + if (!BT_NotCString) + BT_NotCString.reset(new BuiltinBug("Unix API", + "Argument is not a null-terminated string.")); + + SmallString<120> buf; + llvm::raw_svector_ostream os(buf); + + assert(CurrentFunctionDescription); + os << "Argument to " << CurrentFunctionDescription << " is "; + + if (SummarizeRegion(os, C.getASTContext(), MR)) + os << ", which is not a null-terminated string"; + else + os << "not a null-terminated string"; + + // Generate a report for this bug. + BugReport *report = new BugReport(*BT_NotCString, + os.str(), N); + + report->addRange(Ex->getSourceRange()); + C.EmitReport(report); + } + + return UndefinedVal(); + } +} + +const StringLiteral *CStringChecker::getCStringLiteral(CheckerContext &C, + ProgramStateRef &state, const Expr *expr, SVal val) const { + + // Get the memory region pointed to by the val. + const MemRegion *bufRegion = val.getAsRegion(); + if (!bufRegion) + return NULL; + + // Strip casts off the memory region. + bufRegion = bufRegion->StripCasts(); + + // Cast the memory region to a string region. + const StringRegion *strRegion= dyn_cast<StringRegion>(bufRegion); + if (!strRegion) + return NULL; + + // Return the actual string in the string region. + return strRegion->getStringLiteral(); +} + +ProgramStateRef CStringChecker::InvalidateBuffer(CheckerContext &C, + ProgramStateRef state, + const Expr *E, SVal V) { + Loc *L = dyn_cast<Loc>(&V); + if (!L) + return state; + + // FIXME: This is a simplified version of what's in CFRefCount.cpp -- it makes + // some assumptions about the value that CFRefCount can't. Even so, it should + // probably be refactored. + if (loc::MemRegionVal* MR = dyn_cast<loc::MemRegionVal>(L)) { + const MemRegion *R = MR->getRegion()->StripCasts(); + + // Are we dealing with an ElementRegion? If so, we should be invalidating + // the super-region. + if (const ElementRegion *ER = dyn_cast<ElementRegion>(R)) { + R = ER->getSuperRegion(); + // FIXME: What about layers of ElementRegions? + } + + // Invalidate this region. + unsigned Count = C.getCurrentBlockCount(); + const LocationContext *LCtx = C.getPredecessor()->getLocationContext(); + return state->invalidateRegions(R, E, Count, LCtx); + } + + // If we have a non-region value by chance, just remove the binding. + // FIXME: is this necessary or correct? This handles the non-Region + // cases. Is it ever valid to store to these? + return state->unbindLoc(*L); +} + +bool CStringChecker::SummarizeRegion(raw_ostream &os, ASTContext &Ctx, + const MemRegion *MR) { + const TypedValueRegion *TVR = dyn_cast<TypedValueRegion>(MR); + + switch (MR->getKind()) { + case MemRegion::FunctionTextRegionKind: { + const FunctionDecl *FD = cast<FunctionTextRegion>(MR)->getDecl(); + if (FD) + os << "the address of the function '" << *FD << '\''; + else + os << "the address of a function"; + return true; + } + case MemRegion::BlockTextRegionKind: + os << "block text"; + return true; + case MemRegion::BlockDataRegionKind: + os << "a block"; + return true; + case MemRegion::CXXThisRegionKind: + case MemRegion::CXXTempObjectRegionKind: + os << "a C++ temp object of type " << TVR->getValueType().getAsString(); + return true; + case MemRegion::VarRegionKind: + os << "a variable of type" << TVR->getValueType().getAsString(); + return true; + case MemRegion::FieldRegionKind: + os << "a field of type " << TVR->getValueType().getAsString(); + return true; + case MemRegion::ObjCIvarRegionKind: + os << "an instance variable of type " << TVR->getValueType().getAsString(); + return true; + default: + return false; + } +} + +//===----------------------------------------------------------------------===// +// evaluation of individual function calls. +//===----------------------------------------------------------------------===// + +void CStringChecker::evalCopyCommon(CheckerContext &C, + const CallExpr *CE, + ProgramStateRef state, + const Expr *Size, const Expr *Dest, + const Expr *Source, bool Restricted, + bool IsMempcpy) const { + CurrentFunctionDescription = "memory copy function"; + + // See if the size argument is zero. + const LocationContext *LCtx = C.getLocationContext(); + SVal sizeVal = state->getSVal(Size, LCtx); + QualType sizeTy = Size->getType(); + + ProgramStateRef stateZeroSize, stateNonZeroSize; + llvm::tie(stateZeroSize, stateNonZeroSize) = + assumeZero(C, state, sizeVal, sizeTy); + + // Get the value of the Dest. + SVal destVal = state->getSVal(Dest, LCtx); + + // If the size is zero, there won't be any actual memory access, so + // just bind the return value to the destination buffer and return. + if (stateZeroSize) { + stateZeroSize = stateZeroSize->BindExpr(CE, LCtx, destVal); + C.addTransition(stateZeroSize); + } + + // If the size can be nonzero, we have to check the other arguments. + if (stateNonZeroSize) { + state = stateNonZeroSize; + + // Ensure the destination is not null. If it is NULL there will be a + // NULL pointer dereference. + state = checkNonNull(C, state, Dest, destVal); + if (!state) + return; + + // Get the value of the Src. + SVal srcVal = state->getSVal(Source, LCtx); + + // Ensure the source is not null. If it is NULL there will be a + // NULL pointer dereference. + state = checkNonNull(C, state, Source, srcVal); + if (!state) + return; + + // Ensure the accesses are valid and that the buffers do not overlap. + const char * const writeWarning = + "Memory copy function overflows destination buffer"; + state = CheckBufferAccess(C, state, Size, Dest, Source, + writeWarning, /* sourceWarning = */ NULL); + if (Restricted) + state = CheckOverlap(C, state, Size, Dest, Source); + + if (!state) + return; + + // If this is mempcpy, get the byte after the last byte copied and + // bind the expr. + if (IsMempcpy) { + loc::MemRegionVal *destRegVal = dyn_cast<loc::MemRegionVal>(&destVal); + assert(destRegVal && "Destination should be a known MemRegionVal here"); + + // Get the length to copy. + NonLoc *lenValNonLoc = dyn_cast<NonLoc>(&sizeVal); + + if (lenValNonLoc) { + // Get the byte after the last byte copied. + SVal lastElement = C.getSValBuilder().evalBinOpLN(state, BO_Add, + *destRegVal, + *lenValNonLoc, + Dest->getType()); + + // The byte after the last byte copied is the return value. + state = state->BindExpr(CE, LCtx, lastElement); + } else { + // If we don't know how much we copied, we can at least + // conjure a return value for later. + unsigned Count = C.getCurrentBlockCount(); + SVal result = + C.getSValBuilder().getConjuredSymbolVal(NULL, CE, LCtx, Count); + state = state->BindExpr(CE, LCtx, result); + } + + } else { + // All other copies return the destination buffer. + // (Well, bcopy() has a void return type, but this won't hurt.) + state = state->BindExpr(CE, LCtx, destVal); + } + + // Invalidate the destination. + // FIXME: Even if we can't perfectly model the copy, we should see if we + // can use LazyCompoundVals to copy the source values into the destination. + // This would probably remove any existing bindings past the end of the + // copied region, but that's still an improvement over blank invalidation. + state = InvalidateBuffer(C, state, Dest, + state->getSVal(Dest, C.getLocationContext())); + C.addTransition(state); + } +} + + +void CStringChecker::evalMemcpy(CheckerContext &C, const CallExpr *CE) const { + if (CE->getNumArgs() < 3) + return; + + // void *memcpy(void *restrict dst, const void *restrict src, size_t n); + // The return value is the address of the destination buffer. + const Expr *Dest = CE->getArg(0); + ProgramStateRef state = C.getState(); + + evalCopyCommon(C, CE, state, CE->getArg(2), Dest, CE->getArg(1), true); +} + +void CStringChecker::evalMempcpy(CheckerContext &C, const CallExpr *CE) const { + if (CE->getNumArgs() < 3) + return; + + // void *mempcpy(void *restrict dst, const void *restrict src, size_t n); + // The return value is a pointer to the byte following the last written byte. + const Expr *Dest = CE->getArg(0); + ProgramStateRef state = C.getState(); + + evalCopyCommon(C, CE, state, CE->getArg(2), Dest, CE->getArg(1), true, true); +} + +void CStringChecker::evalMemmove(CheckerContext &C, const CallExpr *CE) const { + if (CE->getNumArgs() < 3) + return; + + // void *memmove(void *dst, const void *src, size_t n); + // The return value is the address of the destination buffer. + const Expr *Dest = CE->getArg(0); + ProgramStateRef state = C.getState(); + + evalCopyCommon(C, CE, state, CE->getArg(2), Dest, CE->getArg(1)); +} + +void CStringChecker::evalBcopy(CheckerContext &C, const CallExpr *CE) const { + if (CE->getNumArgs() < 3) + return; + + // void bcopy(const void *src, void *dst, size_t n); + evalCopyCommon(C, CE, C.getState(), + CE->getArg(2), CE->getArg(1), CE->getArg(0)); +} + +void CStringChecker::evalMemcmp(CheckerContext &C, const CallExpr *CE) const { + if (CE->getNumArgs() < 3) + return; + + // int memcmp(const void *s1, const void *s2, size_t n); + CurrentFunctionDescription = "memory comparison function"; + + const Expr *Left = CE->getArg(0); + const Expr *Right = CE->getArg(1); + const Expr *Size = CE->getArg(2); + + ProgramStateRef state = C.getState(); + SValBuilder &svalBuilder = C.getSValBuilder(); + + // See if the size argument is zero. + const LocationContext *LCtx = C.getLocationContext(); + SVal sizeVal = state->getSVal(Size, LCtx); + QualType sizeTy = Size->getType(); + + ProgramStateRef stateZeroSize, stateNonZeroSize; + llvm::tie(stateZeroSize, stateNonZeroSize) = + assumeZero(C, state, sizeVal, sizeTy); + + // If the size can be zero, the result will be 0 in that case, and we don't + // have to check either of the buffers. + if (stateZeroSize) { + state = stateZeroSize; + state = state->BindExpr(CE, LCtx, + svalBuilder.makeZeroVal(CE->getType())); + C.addTransition(state); + } + + // If the size can be nonzero, we have to check the other arguments. + if (stateNonZeroSize) { + state = stateNonZeroSize; + // If we know the two buffers are the same, we know the result is 0. + // First, get the two buffers' addresses. Another checker will have already + // made sure they're not undefined. + DefinedOrUnknownSVal LV = + cast<DefinedOrUnknownSVal>(state->getSVal(Left, LCtx)); + DefinedOrUnknownSVal RV = + cast<DefinedOrUnknownSVal>(state->getSVal(Right, LCtx)); + + // See if they are the same. + DefinedOrUnknownSVal SameBuf = svalBuilder.evalEQ(state, LV, RV); + ProgramStateRef StSameBuf, StNotSameBuf; + llvm::tie(StSameBuf, StNotSameBuf) = state->assume(SameBuf); + + // If the two arguments might be the same buffer, we know the result is 0, + // and we only need to check one size. + if (StSameBuf) { + state = StSameBuf; + state = CheckBufferAccess(C, state, Size, Left); + if (state) { + state = StSameBuf->BindExpr(CE, LCtx, + svalBuilder.makeZeroVal(CE->getType())); + C.addTransition(state); + } + } + + // If the two arguments might be different buffers, we have to check the + // size of both of them. + if (StNotSameBuf) { + state = StNotSameBuf; + state = CheckBufferAccess(C, state, Size, Left, Right); + if (state) { + // The return value is the comparison result, which we don't know. + unsigned Count = C.getCurrentBlockCount(); + SVal CmpV = svalBuilder.getConjuredSymbolVal(NULL, CE, LCtx, Count); + state = state->BindExpr(CE, LCtx, CmpV); + C.addTransition(state); + } + } + } +} + +void CStringChecker::evalstrLength(CheckerContext &C, + const CallExpr *CE) const { + if (CE->getNumArgs() < 1) + return; + + // size_t strlen(const char *s); + evalstrLengthCommon(C, CE, /* IsStrnlen = */ false); +} + +void CStringChecker::evalstrnLength(CheckerContext &C, + const CallExpr *CE) const { + if (CE->getNumArgs() < 2) + return; + + // size_t strnlen(const char *s, size_t maxlen); + evalstrLengthCommon(C, CE, /* IsStrnlen = */ true); +} + +void CStringChecker::evalstrLengthCommon(CheckerContext &C, const CallExpr *CE, + bool IsStrnlen) const { + CurrentFunctionDescription = "string length function"; + ProgramStateRef state = C.getState(); + const LocationContext *LCtx = C.getLocationContext(); + + if (IsStrnlen) { + const Expr *maxlenExpr = CE->getArg(1); + SVal maxlenVal = state->getSVal(maxlenExpr, LCtx); + + ProgramStateRef stateZeroSize, stateNonZeroSize; + llvm::tie(stateZeroSize, stateNonZeroSize) = + assumeZero(C, state, maxlenVal, maxlenExpr->getType()); + + // If the size can be zero, the result will be 0 in that case, and we don't + // have to check the string itself. + if (stateZeroSize) { + SVal zero = C.getSValBuilder().makeZeroVal(CE->getType()); + stateZeroSize = stateZeroSize->BindExpr(CE, LCtx, zero); + C.addTransition(stateZeroSize); + } + + // If the size is GUARANTEED to be zero, we're done! + if (!stateNonZeroSize) + return; + + // Otherwise, record the assumption that the size is nonzero. + state = stateNonZeroSize; + } + + // Check that the string argument is non-null. + const Expr *Arg = CE->getArg(0); + SVal ArgVal = state->getSVal(Arg, LCtx); + + state = checkNonNull(C, state, Arg, ArgVal); + + if (!state) + return; + + SVal strLength = getCStringLength(C, state, Arg, ArgVal); + + // If the argument isn't a valid C string, there's no valid state to + // transition to. + if (strLength.isUndef()) + return; + + DefinedOrUnknownSVal result = UnknownVal(); + + // If the check is for strnlen() then bind the return value to no more than + // the maxlen value. + if (IsStrnlen) { + QualType cmpTy = C.getSValBuilder().getConditionType(); + + // It's a little unfortunate to be getting this again, + // but it's not that expensive... + const Expr *maxlenExpr = CE->getArg(1); + SVal maxlenVal = state->getSVal(maxlenExpr, LCtx); + + NonLoc *strLengthNL = dyn_cast<NonLoc>(&strLength); + NonLoc *maxlenValNL = dyn_cast<NonLoc>(&maxlenVal); + + if (strLengthNL && maxlenValNL) { + ProgramStateRef stateStringTooLong, stateStringNotTooLong; + + // Check if the strLength is greater than the maxlen. + llvm::tie(stateStringTooLong, stateStringNotTooLong) = + state->assume(cast<DefinedOrUnknownSVal> + (C.getSValBuilder().evalBinOpNN(state, BO_GT, + *strLengthNL, + *maxlenValNL, + cmpTy))); + + if (stateStringTooLong && !stateStringNotTooLong) { + // If the string is longer than maxlen, return maxlen. + result = *maxlenValNL; + } else if (stateStringNotTooLong && !stateStringTooLong) { + // If the string is shorter than maxlen, return its length. + result = *strLengthNL; + } + } + + if (result.isUnknown()) { + // If we don't have enough information for a comparison, there's + // no guarantee the full string length will actually be returned. + // All we know is the return value is the min of the string length + // and the limit. This is better than nothing. + unsigned Count = C.getCurrentBlockCount(); + result = C.getSValBuilder().getConjuredSymbolVal(NULL, CE, LCtx, Count); + NonLoc *resultNL = cast<NonLoc>(&result); + + if (strLengthNL) { + state = state->assume(cast<DefinedOrUnknownSVal> + (C.getSValBuilder().evalBinOpNN(state, BO_LE, + *resultNL, + *strLengthNL, + cmpTy)), true); + } + + if (maxlenValNL) { + state = state->assume(cast<DefinedOrUnknownSVal> + (C.getSValBuilder().evalBinOpNN(state, BO_LE, + *resultNL, + *maxlenValNL, + cmpTy)), true); + } + } + + } else { + // This is a plain strlen(), not strnlen(). + result = cast<DefinedOrUnknownSVal>(strLength); + + // If we don't know the length of the string, conjure a return + // value, so it can be used in constraints, at least. + if (result.isUnknown()) { + unsigned Count = C.getCurrentBlockCount(); + result = C.getSValBuilder().getConjuredSymbolVal(NULL, CE, LCtx, Count); + } + } + + // Bind the return value. + assert(!result.isUnknown() && "Should have conjured a value by now"); + state = state->BindExpr(CE, LCtx, result); + C.addTransition(state); +} + +void CStringChecker::evalStrcpy(CheckerContext &C, const CallExpr *CE) const { + if (CE->getNumArgs() < 2) + return; + + // char *strcpy(char *restrict dst, const char *restrict src); + evalStrcpyCommon(C, CE, + /* returnEnd = */ false, + /* isBounded = */ false, + /* isAppending = */ false); +} + +void CStringChecker::evalStrncpy(CheckerContext &C, const CallExpr *CE) const { + if (CE->getNumArgs() < 3) + return; + + // char *strncpy(char *restrict dst, const char *restrict src, size_t n); + evalStrcpyCommon(C, CE, + /* returnEnd = */ false, + /* isBounded = */ true, + /* isAppending = */ false); +} + +void CStringChecker::evalStpcpy(CheckerContext &C, const CallExpr *CE) const { + if (CE->getNumArgs() < 2) + return; + + // char *stpcpy(char *restrict dst, const char *restrict src); + evalStrcpyCommon(C, CE, + /* returnEnd = */ true, + /* isBounded = */ false, + /* isAppending = */ false); +} + +void CStringChecker::evalStrcat(CheckerContext &C, const CallExpr *CE) const { + if (CE->getNumArgs() < 2) + return; + + //char *strcat(char *restrict s1, const char *restrict s2); + evalStrcpyCommon(C, CE, + /* returnEnd = */ false, + /* isBounded = */ false, + /* isAppending = */ true); +} + +void CStringChecker::evalStrncat(CheckerContext &C, const CallExpr *CE) const { + if (CE->getNumArgs() < 3) + return; + + //char *strncat(char *restrict s1, const char *restrict s2, size_t n); + evalStrcpyCommon(C, CE, + /* returnEnd = */ false, + /* isBounded = */ true, + /* isAppending = */ true); +} + +void CStringChecker::evalStrcpyCommon(CheckerContext &C, const CallExpr *CE, + bool returnEnd, bool isBounded, + bool isAppending) const { + CurrentFunctionDescription = "string copy function"; + ProgramStateRef state = C.getState(); + const LocationContext *LCtx = C.getLocationContext(); + + // Check that the destination is non-null. + const Expr *Dst = CE->getArg(0); + SVal DstVal = state->getSVal(Dst, LCtx); + + state = checkNonNull(C, state, Dst, DstVal); + if (!state) + return; + + // Check that the source is non-null. + const Expr *srcExpr = CE->getArg(1); + SVal srcVal = state->getSVal(srcExpr, LCtx); + state = checkNonNull(C, state, srcExpr, srcVal); + if (!state) + return; + + // Get the string length of the source. + SVal strLength = getCStringLength(C, state, srcExpr, srcVal); + + // If the source isn't a valid C string, give up. + if (strLength.isUndef()) + return; + + SValBuilder &svalBuilder = C.getSValBuilder(); + QualType cmpTy = svalBuilder.getConditionType(); + QualType sizeTy = svalBuilder.getContext().getSizeType(); + + // These two values allow checking two kinds of errors: + // - actual overflows caused by a source that doesn't fit in the destination + // - potential overflows caused by a bound that could exceed the destination + SVal amountCopied = UnknownVal(); + SVal maxLastElementIndex = UnknownVal(); + const char *boundWarning = NULL; + + // If the function is strncpy, strncat, etc... it is bounded. + if (isBounded) { + // Get the max number of characters to copy. + const Expr *lenExpr = CE->getArg(2); + SVal lenVal = state->getSVal(lenExpr, LCtx); + + // Protect against misdeclared strncpy(). + lenVal = svalBuilder.evalCast(lenVal, sizeTy, lenExpr->getType()); + + NonLoc *strLengthNL = dyn_cast<NonLoc>(&strLength); + NonLoc *lenValNL = dyn_cast<NonLoc>(&lenVal); + + // If we know both values, we might be able to figure out how much + // we're copying. + if (strLengthNL && lenValNL) { + ProgramStateRef stateSourceTooLong, stateSourceNotTooLong; + + // Check if the max number to copy is less than the length of the src. + // If the bound is equal to the source length, strncpy won't null- + // terminate the result! + llvm::tie(stateSourceTooLong, stateSourceNotTooLong) = + state->assume(cast<DefinedOrUnknownSVal> + (svalBuilder.evalBinOpNN(state, BO_GE, *strLengthNL, + *lenValNL, cmpTy))); + + if (stateSourceTooLong && !stateSourceNotTooLong) { + // Max number to copy is less than the length of the src, so the actual + // strLength copied is the max number arg. + state = stateSourceTooLong; + amountCopied = lenVal; + + } else if (!stateSourceTooLong && stateSourceNotTooLong) { + // The source buffer entirely fits in the bound. + state = stateSourceNotTooLong; + amountCopied = strLength; + } + } + + // We still want to know if the bound is known to be too large. + if (lenValNL) { + if (isAppending) { + // For strncat, the check is strlen(dst) + lenVal < sizeof(dst) + + // Get the string length of the destination. If the destination is + // memory that can't have a string length, we shouldn't be copying + // into it anyway. + SVal dstStrLength = getCStringLength(C, state, Dst, DstVal); + if (dstStrLength.isUndef()) + return; + + if (NonLoc *dstStrLengthNL = dyn_cast<NonLoc>(&dstStrLength)) { + maxLastElementIndex = svalBuilder.evalBinOpNN(state, BO_Add, + *lenValNL, + *dstStrLengthNL, + sizeTy); + boundWarning = "Size argument is greater than the free space in the " + "destination buffer"; + } + + } else { + // For strncpy, this is just checking that lenVal <= sizeof(dst) + // (Yes, strncpy and strncat differ in how they treat termination. + // strncat ALWAYS terminates, but strncpy doesn't.) + NonLoc one = cast<NonLoc>(svalBuilder.makeIntVal(1, sizeTy)); + maxLastElementIndex = svalBuilder.evalBinOpNN(state, BO_Sub, *lenValNL, + one, sizeTy); + boundWarning = "Size argument is greater than the length of the " + "destination buffer"; + } + } + + // If we couldn't pin down the copy length, at least bound it. + // FIXME: We should actually run this code path for append as well, but + // right now it creates problems with constraints (since we can end up + // trying to pass constraints from symbol to symbol). + if (amountCopied.isUnknown() && !isAppending) { + // Try to get a "hypothetical" string length symbol, which we can later + // set as a real value if that turns out to be the case. + amountCopied = getCStringLength(C, state, lenExpr, srcVal, true); + assert(!amountCopied.isUndef()); + + if (NonLoc *amountCopiedNL = dyn_cast<NonLoc>(&amountCopied)) { + if (lenValNL) { + // amountCopied <= lenVal + SVal copiedLessThanBound = svalBuilder.evalBinOpNN(state, BO_LE, + *amountCopiedNL, + *lenValNL, + cmpTy); + state = state->assume(cast<DefinedOrUnknownSVal>(copiedLessThanBound), + true); + if (!state) + return; + } + + if (strLengthNL) { + // amountCopied <= strlen(source) + SVal copiedLessThanSrc = svalBuilder.evalBinOpNN(state, BO_LE, + *amountCopiedNL, + *strLengthNL, + cmpTy); + state = state->assume(cast<DefinedOrUnknownSVal>(copiedLessThanSrc), + true); + if (!state) + return; + } + } + } + + } else { + // The function isn't bounded. The amount copied should match the length + // of the source buffer. + amountCopied = strLength; + } + + assert(state); + + // This represents the number of characters copied into the destination + // buffer. (It may not actually be the strlen if the destination buffer + // is not terminated.) + SVal finalStrLength = UnknownVal(); + + // If this is an appending function (strcat, strncat...) then set the + // string length to strlen(src) + strlen(dst) since the buffer will + // ultimately contain both. + if (isAppending) { + // Get the string length of the destination. If the destination is memory + // that can't have a string length, we shouldn't be copying into it anyway. + SVal dstStrLength = getCStringLength(C, state, Dst, DstVal); + if (dstStrLength.isUndef()) + return; + + NonLoc *srcStrLengthNL = dyn_cast<NonLoc>(&amountCopied); + NonLoc *dstStrLengthNL = dyn_cast<NonLoc>(&dstStrLength); + + // If we know both string lengths, we might know the final string length. + if (srcStrLengthNL && dstStrLengthNL) { + // Make sure the two lengths together don't overflow a size_t. + state = checkAdditionOverflow(C, state, *srcStrLengthNL, *dstStrLengthNL); + if (!state) + return; + + finalStrLength = svalBuilder.evalBinOpNN(state, BO_Add, *srcStrLengthNL, + *dstStrLengthNL, sizeTy); + } + + // If we couldn't get a single value for the final string length, + // we can at least bound it by the individual lengths. + if (finalStrLength.isUnknown()) { + // Try to get a "hypothetical" string length symbol, which we can later + // set as a real value if that turns out to be the case. + finalStrLength = getCStringLength(C, state, CE, DstVal, true); + assert(!finalStrLength.isUndef()); + + if (NonLoc *finalStrLengthNL = dyn_cast<NonLoc>(&finalStrLength)) { + if (srcStrLengthNL) { + // finalStrLength >= srcStrLength + SVal sourceInResult = svalBuilder.evalBinOpNN(state, BO_GE, + *finalStrLengthNL, + *srcStrLengthNL, + cmpTy); + state = state->assume(cast<DefinedOrUnknownSVal>(sourceInResult), + true); + if (!state) + return; + } + + if (dstStrLengthNL) { + // finalStrLength >= dstStrLength + SVal destInResult = svalBuilder.evalBinOpNN(state, BO_GE, + *finalStrLengthNL, + *dstStrLengthNL, + cmpTy); + state = state->assume(cast<DefinedOrUnknownSVal>(destInResult), + true); + if (!state) + return; + } + } + } + + } else { + // Otherwise, this is a copy-over function (strcpy, strncpy, ...), and + // the final string length will match the input string length. + finalStrLength = amountCopied; + } + + // The final result of the function will either be a pointer past the last + // copied element, or a pointer to the start of the destination buffer. + SVal Result = (returnEnd ? UnknownVal() : DstVal); + + assert(state); + + // If the destination is a MemRegion, try to check for a buffer overflow and + // record the new string length. + if (loc::MemRegionVal *dstRegVal = dyn_cast<loc::MemRegionVal>(&DstVal)) { + QualType ptrTy = Dst->getType(); + + // If we have an exact value on a bounded copy, use that to check for + // overflows, rather than our estimate about how much is actually copied. + if (boundWarning) { + if (NonLoc *maxLastNL = dyn_cast<NonLoc>(&maxLastElementIndex)) { + SVal maxLastElement = svalBuilder.evalBinOpLN(state, BO_Add, *dstRegVal, + *maxLastNL, ptrTy); + state = CheckLocation(C, state, CE->getArg(2), maxLastElement, + boundWarning); + if (!state) + return; + } + } + + // Then, if the final length is known... + if (NonLoc *knownStrLength = dyn_cast<NonLoc>(&finalStrLength)) { + SVal lastElement = svalBuilder.evalBinOpLN(state, BO_Add, *dstRegVal, + *knownStrLength, ptrTy); + + // ...and we haven't checked the bound, we'll check the actual copy. + if (!boundWarning) { + const char * const warningMsg = + "String copy function overflows destination buffer"; + state = CheckLocation(C, state, Dst, lastElement, warningMsg); + if (!state) + return; + } + + // If this is a stpcpy-style copy, the last element is the return value. + if (returnEnd) + Result = lastElement; + } + + // Invalidate the destination. This must happen before we set the C string + // length because invalidation will clear the length. + // FIXME: Even if we can't perfectly model the copy, we should see if we + // can use LazyCompoundVals to copy the source values into the destination. + // This would probably remove any existing bindings past the end of the + // string, but that's still an improvement over blank invalidation. + state = InvalidateBuffer(C, state, Dst, *dstRegVal); + + // Set the C string length of the destination, if we know it. + if (isBounded && !isAppending) { + // strncpy is annoying in that it doesn't guarantee to null-terminate + // the result string. If the original string didn't fit entirely inside + // the bound (including the null-terminator), we don't know how long the + // result is. + if (amountCopied != strLength) + finalStrLength = UnknownVal(); + } + state = setCStringLength(state, dstRegVal->getRegion(), finalStrLength); + } + + assert(state); + + // If this is a stpcpy-style copy, but we were unable to check for a buffer + // overflow, we still need a result. Conjure a return value. + if (returnEnd && Result.isUnknown()) { + unsigned Count = C.getCurrentBlockCount(); + Result = svalBuilder.getConjuredSymbolVal(NULL, CE, LCtx, Count); + } + + // Set the return value. + state = state->BindExpr(CE, LCtx, Result); + C.addTransition(state); +} + +void CStringChecker::evalStrcmp(CheckerContext &C, const CallExpr *CE) const { + if (CE->getNumArgs() < 2) + return; + + //int strcmp(const char *s1, const char *s2); + evalStrcmpCommon(C, CE, /* isBounded = */ false, /* ignoreCase = */ false); +} + +void CStringChecker::evalStrncmp(CheckerContext &C, const CallExpr *CE) const { + if (CE->getNumArgs() < 3) + return; + + //int strncmp(const char *s1, const char *s2, size_t n); + evalStrcmpCommon(C, CE, /* isBounded = */ true, /* ignoreCase = */ false); +} + +void CStringChecker::evalStrcasecmp(CheckerContext &C, + const CallExpr *CE) const { + if (CE->getNumArgs() < 2) + return; + + //int strcasecmp(const char *s1, const char *s2); + evalStrcmpCommon(C, CE, /* isBounded = */ false, /* ignoreCase = */ true); +} + +void CStringChecker::evalStrncasecmp(CheckerContext &C, + const CallExpr *CE) const { + if (CE->getNumArgs() < 3) + return; + + //int strncasecmp(const char *s1, const char *s2, size_t n); + evalStrcmpCommon(C, CE, /* isBounded = */ true, /* ignoreCase = */ true); +} + +void CStringChecker::evalStrcmpCommon(CheckerContext &C, const CallExpr *CE, + bool isBounded, bool ignoreCase) const { + CurrentFunctionDescription = "string comparison function"; + ProgramStateRef state = C.getState(); + const LocationContext *LCtx = C.getLocationContext(); + + // Check that the first string is non-null + const Expr *s1 = CE->getArg(0); + SVal s1Val = state->getSVal(s1, LCtx); + state = checkNonNull(C, state, s1, s1Val); + if (!state) + return; + + // Check that the second string is non-null. + const Expr *s2 = CE->getArg(1); + SVal s2Val = state->getSVal(s2, LCtx); + state = checkNonNull(C, state, s2, s2Val); + if (!state) + return; + + // Get the string length of the first string or give up. + SVal s1Length = getCStringLength(C, state, s1, s1Val); + if (s1Length.isUndef()) + return; + + // Get the string length of the second string or give up. + SVal s2Length = getCStringLength(C, state, s2, s2Val); + if (s2Length.isUndef()) + return; + + // If we know the two buffers are the same, we know the result is 0. + // First, get the two buffers' addresses. Another checker will have already + // made sure they're not undefined. + DefinedOrUnknownSVal LV = cast<DefinedOrUnknownSVal>(s1Val); + DefinedOrUnknownSVal RV = cast<DefinedOrUnknownSVal>(s2Val); + + // See if they are the same. + SValBuilder &svalBuilder = C.getSValBuilder(); + DefinedOrUnknownSVal SameBuf = svalBuilder.evalEQ(state, LV, RV); + ProgramStateRef StSameBuf, StNotSameBuf; + llvm::tie(StSameBuf, StNotSameBuf) = state->assume(SameBuf); + + // If the two arguments might be the same buffer, we know the result is 0, + // and we only need to check one size. + if (StSameBuf) { + StSameBuf = StSameBuf->BindExpr(CE, LCtx, + svalBuilder.makeZeroVal(CE->getType())); + C.addTransition(StSameBuf); + + // If the two arguments are GUARANTEED to be the same, we're done! + if (!StNotSameBuf) + return; + } + + assert(StNotSameBuf); + state = StNotSameBuf; + + // At this point we can go about comparing the two buffers. + // For now, we only do this if they're both known string literals. + + // Attempt to extract string literals from both expressions. + const StringLiteral *s1StrLiteral = getCStringLiteral(C, state, s1, s1Val); + const StringLiteral *s2StrLiteral = getCStringLiteral(C, state, s2, s2Val); + bool canComputeResult = false; + + if (s1StrLiteral && s2StrLiteral) { + StringRef s1StrRef = s1StrLiteral->getString(); + StringRef s2StrRef = s2StrLiteral->getString(); + + if (isBounded) { + // Get the max number of characters to compare. + const Expr *lenExpr = CE->getArg(2); + SVal lenVal = state->getSVal(lenExpr, LCtx); + + // If the length is known, we can get the right substrings. + if (const llvm::APSInt *len = svalBuilder.getKnownValue(state, lenVal)) { + // Create substrings of each to compare the prefix. + s1StrRef = s1StrRef.substr(0, (size_t)len->getZExtValue()); + s2StrRef = s2StrRef.substr(0, (size_t)len->getZExtValue()); + canComputeResult = true; + } + } else { + // This is a normal, unbounded strcmp. + canComputeResult = true; + } + + if (canComputeResult) { + // Real strcmp stops at null characters. + size_t s1Term = s1StrRef.find('\0'); + if (s1Term != StringRef::npos) + s1StrRef = s1StrRef.substr(0, s1Term); + + size_t s2Term = s2StrRef.find('\0'); + if (s2Term != StringRef::npos) + s2StrRef = s2StrRef.substr(0, s2Term); + + // Use StringRef's comparison methods to compute the actual result. + int result; + + if (ignoreCase) { + // Compare string 1 to string 2 the same way strcasecmp() does. + result = s1StrRef.compare_lower(s2StrRef); + } else { + // Compare string 1 to string 2 the same way strcmp() does. + result = s1StrRef.compare(s2StrRef); + } + + // Build the SVal of the comparison and bind the return value. + SVal resultVal = svalBuilder.makeIntVal(result, CE->getType()); + state = state->BindExpr(CE, LCtx, resultVal); + } + } + + if (!canComputeResult) { + // Conjure a symbolic value. It's the best we can do. + unsigned Count = C.getCurrentBlockCount(); + SVal resultVal = svalBuilder.getConjuredSymbolVal(NULL, CE, LCtx, Count); + state = state->BindExpr(CE, LCtx, resultVal); + } + + // Record this as a possible path. + C.addTransition(state); +} + +//===----------------------------------------------------------------------===// +// The driver method, and other Checker callbacks. +//===----------------------------------------------------------------------===// + +bool CStringChecker::evalCall(const CallExpr *CE, CheckerContext &C) const { + const FunctionDecl *FDecl = C.getCalleeDecl(CE); + + if (!FDecl) + return false; + + FnCheck evalFunction = 0; + if (C.isCLibraryFunction(FDecl, "memcpy")) + evalFunction = &CStringChecker::evalMemcpy; + else if (C.isCLibraryFunction(FDecl, "mempcpy")) + evalFunction = &CStringChecker::evalMempcpy; + else if (C.isCLibraryFunction(FDecl, "memcmp")) + evalFunction = &CStringChecker::evalMemcmp; + else if (C.isCLibraryFunction(FDecl, "memmove")) + evalFunction = &CStringChecker::evalMemmove; + else if (C.isCLibraryFunction(FDecl, "strcpy")) + evalFunction = &CStringChecker::evalStrcpy; + else if (C.isCLibraryFunction(FDecl, "strncpy")) + evalFunction = &CStringChecker::evalStrncpy; + else if (C.isCLibraryFunction(FDecl, "stpcpy")) + evalFunction = &CStringChecker::evalStpcpy; + else if (C.isCLibraryFunction(FDecl, "strcat")) + evalFunction = &CStringChecker::evalStrcat; + else if (C.isCLibraryFunction(FDecl, "strncat")) + evalFunction = &CStringChecker::evalStrncat; + else if (C.isCLibraryFunction(FDecl, "strlen")) + evalFunction = &CStringChecker::evalstrLength; + else if (C.isCLibraryFunction(FDecl, "strnlen")) + evalFunction = &CStringChecker::evalstrnLength; + else if (C.isCLibraryFunction(FDecl, "strcmp")) + evalFunction = &CStringChecker::evalStrcmp; + else if (C.isCLibraryFunction(FDecl, "strncmp")) + evalFunction = &CStringChecker::evalStrncmp; + else if (C.isCLibraryFunction(FDecl, "strcasecmp")) + evalFunction = &CStringChecker::evalStrcasecmp; + else if (C.isCLibraryFunction(FDecl, "strncasecmp")) + evalFunction = &CStringChecker::evalStrncasecmp; + else if (C.isCLibraryFunction(FDecl, "bcopy")) + evalFunction = &CStringChecker::evalBcopy; + else if (C.isCLibraryFunction(FDecl, "bcmp")) + evalFunction = &CStringChecker::evalMemcmp; + + // If the callee isn't a string function, let another checker handle it. + if (!evalFunction) + return false; + + // Make sure each function sets its own description. + // (But don't bother in a release build.) + assert(!(CurrentFunctionDescription = NULL)); + + // Check and evaluate the call. + (this->*evalFunction)(C, CE); + + // If the evaluate call resulted in no change, chain to the next eval call + // handler. + // Note, the custom CString evaluation calls assume that basic safety + // properties are held. However, if the user chooses to turn off some of these + // checks, we ignore the issues and leave the call evaluation to a generic + // handler. + if (!C.isDifferent()) + return false; + + return true; +} + +void CStringChecker::checkPreStmt(const DeclStmt *DS, CheckerContext &C) const { + // Record string length for char a[] = "abc"; + ProgramStateRef state = C.getState(); + + for (DeclStmt::const_decl_iterator I = DS->decl_begin(), E = DS->decl_end(); + I != E; ++I) { + const VarDecl *D = dyn_cast<VarDecl>(*I); + if (!D) + continue; + + // FIXME: Handle array fields of structs. + if (!D->getType()->isArrayType()) + continue; + + const Expr *Init = D->getInit(); + if (!Init) + continue; + if (!isa<StringLiteral>(Init)) + continue; + + Loc VarLoc = state->getLValue(D, C.getLocationContext()); + const MemRegion *MR = VarLoc.getAsRegion(); + if (!MR) + continue; + + SVal StrVal = state->getSVal(Init, C.getLocationContext()); + assert(StrVal.isValid() && "Initializer string is unknown or undefined"); + DefinedOrUnknownSVal strLength + = cast<DefinedOrUnknownSVal>(getCStringLength(C, state, Init, StrVal)); + + state = state->set<CStringLength>(MR, strLength); + } + + C.addTransition(state); +} + +bool CStringChecker::wantsRegionChangeUpdate(ProgramStateRef state) const { + CStringLength::EntryMap Entries = state->get<CStringLength>(); + return !Entries.isEmpty(); +} + +ProgramStateRef +CStringChecker::checkRegionChanges(ProgramStateRef state, + const StoreManager::InvalidatedSymbols *, + ArrayRef<const MemRegion *> ExplicitRegions, + ArrayRef<const MemRegion *> Regions, + const CallOrObjCMessage *Call) const { + CStringLength::EntryMap Entries = state->get<CStringLength>(); + if (Entries.isEmpty()) + return state; + + llvm::SmallPtrSet<const MemRegion *, 8> Invalidated; + llvm::SmallPtrSet<const MemRegion *, 32> SuperRegions; + + // First build sets for the changed regions and their super-regions. + for (ArrayRef<const MemRegion *>::iterator + I = Regions.begin(), E = Regions.end(); I != E; ++I) { + const MemRegion *MR = *I; + Invalidated.insert(MR); + + SuperRegions.insert(MR); + while (const SubRegion *SR = dyn_cast<SubRegion>(MR)) { + MR = SR->getSuperRegion(); + SuperRegions.insert(MR); + } + } + + CStringLength::EntryMap::Factory &F = state->get_context<CStringLength>(); + + // Then loop over the entries in the current state. + for (CStringLength::EntryMap::iterator I = Entries.begin(), + E = Entries.end(); I != E; ++I) { + const MemRegion *MR = I.getKey(); + + // Is this entry for a super-region of a changed region? + if (SuperRegions.count(MR)) { + Entries = F.remove(Entries, MR); + continue; + } + + // Is this entry for a sub-region of a changed region? + const MemRegion *Super = MR; + while (const SubRegion *SR = dyn_cast<SubRegion>(Super)) { + Super = SR->getSuperRegion(); + if (Invalidated.count(Super)) { + Entries = F.remove(Entries, MR); + break; + } + } + } + + return state->set<CStringLength>(Entries); +} + +void CStringChecker::checkLiveSymbols(ProgramStateRef state, + SymbolReaper &SR) const { + // Mark all symbols in our string length map as valid. + CStringLength::EntryMap Entries = state->get<CStringLength>(); + + for (CStringLength::EntryMap::iterator I = Entries.begin(), E = Entries.end(); + I != E; ++I) { + SVal Len = I.getData(); + + for (SymExpr::symbol_iterator si = Len.symbol_begin(), + se = Len.symbol_end(); si != se; ++si) + SR.markInUse(*si); + } +} + +void CStringChecker::checkDeadSymbols(SymbolReaper &SR, + CheckerContext &C) const { + if (!SR.hasDeadSymbols()) + return; + + ProgramStateRef state = C.getState(); + CStringLength::EntryMap Entries = state->get<CStringLength>(); + if (Entries.isEmpty()) + return; + + CStringLength::EntryMap::Factory &F = state->get_context<CStringLength>(); + for (CStringLength::EntryMap::iterator I = Entries.begin(), E = Entries.end(); + I != E; ++I) { + SVal Len = I.getData(); + if (SymbolRef Sym = Len.getAsSymbol()) { + if (SR.isDead(Sym)) + Entries = F.remove(Entries, I.getKey()); + } + } + + state = state->set<CStringLength>(Entries); + C.addTransition(state); +} + +#define REGISTER_CHECKER(name) \ +void ento::register##name(CheckerManager &mgr) {\ + static CStringChecker *TheChecker = 0; \ + if (TheChecker == 0) \ + TheChecker = mgr.registerChecker<CStringChecker>(); \ + TheChecker->Filter.Check##name = true; \ +} + +REGISTER_CHECKER(CStringNullArg) +REGISTER_CHECKER(CStringOutOfBounds) +REGISTER_CHECKER(CStringBufferOverlap) +REGISTER_CHECKER(CStringNotNullTerm) + +void ento::registerCStringCheckerBasic(CheckerManager &Mgr) { + registerCStringNullArg(Mgr); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/CStringSyntaxChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/CStringSyntaxChecker.cpp new file mode 100644 index 0000000..befc935 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/CStringSyntaxChecker.cpp @@ -0,0 +1,191 @@ +//== CStringSyntaxChecker.cpp - CoreFoundation containers API *- C++ -*-==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// An AST checker that looks for common pitfalls when using C string APIs. +// - Identifies erroneous patterns in the last argument to strncat - the number +// of bytes to copy. +// +//===----------------------------------------------------------------------===// +#include "ClangSACheckers.h" +#include "clang/Analysis/AnalysisContext.h" +#include "clang/AST/Expr.h" +#include "clang/AST/OperationKinds.h" +#include "clang/AST/StmtVisitor.h" +#include "clang/Basic/TargetInfo.h" +#include "clang/Basic/TypeTraits.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugReporter.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/AnalysisManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "llvm/ADT/SmallString.h" +#include "llvm/Support/raw_ostream.h" + +using namespace clang; +using namespace ento; + +namespace { +class WalkAST: public StmtVisitor<WalkAST> { + BugReporter &BR; + AnalysisDeclContext* AC; + ASTContext &ASTC; + + /// Check if two expressions refer to the same declaration. + inline bool sameDecl(const Expr *A1, const Expr *A2) { + if (const DeclRefExpr *D1 = dyn_cast<DeclRefExpr>(A1->IgnoreParenCasts())) + if (const DeclRefExpr *D2 = dyn_cast<DeclRefExpr>(A2->IgnoreParenCasts())) + return D1->getDecl() == D2->getDecl(); + return false; + } + + /// Check if the expression E is a sizeof(WithArg). + inline bool isSizeof(const Expr *E, const Expr *WithArg) { + if (const UnaryExprOrTypeTraitExpr *UE = + dyn_cast<UnaryExprOrTypeTraitExpr>(E)) + if (UE->getKind() == UETT_SizeOf) + return sameDecl(UE->getArgumentExpr(), WithArg); + return false; + } + + /// Check if the expression E is a strlen(WithArg). + inline bool isStrlen(const Expr *E, const Expr *WithArg) { + if (const CallExpr *CE = dyn_cast<CallExpr>(E)) { + const FunctionDecl *FD = CE->getDirectCallee(); + if (!FD) + return false; + return (CheckerContext::isCLibraryFunction(FD, "strlen", ASTC) + && sameDecl(CE->getArg(0), WithArg)); + } + return false; + } + + /// Check if the expression is an integer literal with value 1. + inline bool isOne(const Expr *E) { + if (const IntegerLiteral *IL = dyn_cast<IntegerLiteral>(E)) + return (IL->getValue().isIntN(1)); + return false; + } + + inline StringRef getPrintableName(const Expr *E) { + if (const DeclRefExpr *D = dyn_cast<DeclRefExpr>(E->IgnoreParenCasts())) + return D->getDecl()->getName(); + return StringRef(); + } + + /// Identify erroneous patterns in the last argument to strncat - the number + /// of bytes to copy. + bool containsBadStrncatPattern(const CallExpr *CE); + +public: + WalkAST(BugReporter &br, AnalysisDeclContext* ac) : + BR(br), AC(ac), ASTC(AC->getASTContext()) { + } + + // Statement visitor methods. + void VisitChildren(Stmt *S); + void VisitStmt(Stmt *S) { + VisitChildren(S); + } + void VisitCallExpr(CallExpr *CE); +}; +} // end anonymous namespace + +// The correct size argument should look like following: +// strncat(dst, src, sizeof(dst) - strlen(dest) - 1); +// We look for the following anti-patterns: +// - strncat(dst, src, sizeof(dst) - strlen(dst)); +// - strncat(dst, src, sizeof(dst) - 1); +// - strncat(dst, src, sizeof(dst)); +bool WalkAST::containsBadStrncatPattern(const CallExpr *CE) { + const Expr *DstArg = CE->getArg(0); + const Expr *SrcArg = CE->getArg(1); + const Expr *LenArg = CE->getArg(2); + + // Identify wrong size expressions, which are commonly used instead. + if (const BinaryOperator *BE = + dyn_cast<BinaryOperator>(LenArg->IgnoreParenCasts())) { + // - sizeof(dst) - strlen(dst) + if (BE->getOpcode() == BO_Sub) { + const Expr *L = BE->getLHS(); + const Expr *R = BE->getRHS(); + if (isSizeof(L, DstArg) && isStrlen(R, DstArg)) + return true; + + // - sizeof(dst) - 1 + if (isSizeof(L, DstArg) && isOne(R->IgnoreParenCasts())) + return true; + } + } + // - sizeof(dst) + if (isSizeof(LenArg, DstArg)) + return true; + + // - sizeof(src) + if (isSizeof(LenArg, SrcArg)) + return true; + return false; +} + +void WalkAST::VisitCallExpr(CallExpr *CE) { + const FunctionDecl *FD = CE->getDirectCallee(); + if (!FD) + return; + + if (CheckerContext::isCLibraryFunction(FD, "strncat", ASTC)) { + if (containsBadStrncatPattern(CE)) { + const Expr *DstArg = CE->getArg(0); + const Expr *LenArg = CE->getArg(2); + SourceRange R = LenArg->getSourceRange(); + PathDiagnosticLocation Loc = + PathDiagnosticLocation::createBegin(LenArg, BR.getSourceManager(), AC); + + StringRef DstName = getPrintableName(DstArg); + + SmallString<256> S; + llvm::raw_svector_ostream os(S); + os << "Potential buffer overflow. "; + if (!DstName.empty()) { + os << "Replace with 'sizeof(" << DstName << ") " + "- strlen(" << DstName <<") - 1'"; + os << " or u"; + } else + os << "U"; + os << "se a safer 'strlcat' API"; + + BR.EmitBasicReport(FD, "Anti-pattern in the argument", "C String API", + os.str(), Loc, &R, 1); + } + } + + // Recurse and check children. + VisitChildren(CE); +} + +void WalkAST::VisitChildren(Stmt *S) { + for (Stmt::child_iterator I = S->child_begin(), E = S->child_end(); I != E; + ++I) + if (Stmt *child = *I) + Visit(child); +} + +namespace { +class CStringSyntaxChecker: public Checker<check::ASTCodeBody> { +public: + + void checkASTCodeBody(const Decl *D, AnalysisManager& Mgr, + BugReporter &BR) const { + WalkAST walker(BR, Mgr.getAnalysisDeclContext(D)); + walker.Visit(D->getBody()); + } +}; +} + +void ento::registerCStringSyntaxChecker(CheckerManager &mgr) { + mgr.registerChecker<CStringSyntaxChecker>(); +} + diff --git a/clang/lib/StaticAnalyzer/Checkers/CallAndMessageChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/CallAndMessageChecker.cpp new file mode 100644 index 0000000..f601431 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/CallAndMessageChecker.cpp @@ -0,0 +1,385 @@ +//===--- CallAndMessageChecker.cpp ------------------------------*- C++ -*--==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This defines CallAndMessageChecker, a builtin checker that checks for various +// errors of call and objc message expressions. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ObjCMessage.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/AST/ParentMap.h" +#include "clang/Basic/TargetInfo.h" +#include "llvm/ADT/SmallString.h" + +using namespace clang; +using namespace ento; + +namespace { +class CallAndMessageChecker + : public Checker< check::PreStmt<CallExpr>, check::PreObjCMessage > { + mutable OwningPtr<BugType> BT_call_null; + mutable OwningPtr<BugType> BT_call_undef; + mutable OwningPtr<BugType> BT_call_arg; + mutable OwningPtr<BugType> BT_msg_undef; + mutable OwningPtr<BugType> BT_objc_prop_undef; + mutable OwningPtr<BugType> BT_msg_arg; + mutable OwningPtr<BugType> BT_msg_ret; +public: + + void checkPreStmt(const CallExpr *CE, CheckerContext &C) const; + void checkPreObjCMessage(ObjCMessage msg, CheckerContext &C) const; + +private: + static void PreVisitProcessArgs(CheckerContext &C,CallOrObjCMessage callOrMsg, + const char *BT_desc, OwningPtr<BugType> &BT); + static bool PreVisitProcessArg(CheckerContext &C, SVal V,SourceRange argRange, + const Expr *argEx, + const bool checkUninitFields, + const char *BT_desc, + OwningPtr<BugType> &BT); + + static void EmitBadCall(BugType *BT, CheckerContext &C, const CallExpr *CE); + void emitNilReceiverBug(CheckerContext &C, const ObjCMessage &msg, + ExplodedNode *N) const; + + void HandleNilReceiver(CheckerContext &C, + ProgramStateRef state, + ObjCMessage msg) const; + + static void LazyInit_BT(const char *desc, OwningPtr<BugType> &BT) { + if (!BT) + BT.reset(new BuiltinBug(desc)); + } +}; +} // end anonymous namespace + +void CallAndMessageChecker::EmitBadCall(BugType *BT, CheckerContext &C, + const CallExpr *CE) { + ExplodedNode *N = C.generateSink(); + if (!N) + return; + + BugReport *R = new BugReport(*BT, BT->getName(), N); + R->addVisitor(bugreporter::getTrackNullOrUndefValueVisitor(N, + bugreporter::GetCalleeExpr(N), R)); + C.EmitReport(R); +} + +void CallAndMessageChecker::PreVisitProcessArgs(CheckerContext &C, + CallOrObjCMessage callOrMsg, + const char *BT_desc, + OwningPtr<BugType> &BT) { + // Don't check for uninitialized field values in arguments if the + // caller has a body that is available and we have the chance to inline it. + // This is a hack, but is a reasonable compromise betweens sometimes warning + // and sometimes not depending on if we decide to inline a function. + const Decl *D = callOrMsg.getDecl(); + const bool checkUninitFields = + !(C.getAnalysisManager().shouldInlineCall() && + (D && D->getBody())); + + for (unsigned i = 0, e = callOrMsg.getNumArgs(); i != e; ++i) + if (PreVisitProcessArg(C, callOrMsg.getArgSVal(i), + callOrMsg.getArgSourceRange(i), callOrMsg.getArg(i), + checkUninitFields, + BT_desc, BT)) + return; +} + +bool CallAndMessageChecker::PreVisitProcessArg(CheckerContext &C, + SVal V, SourceRange argRange, + const Expr *argEx, + const bool checkUninitFields, + const char *BT_desc, + OwningPtr<BugType> &BT) { + if (V.isUndef()) { + if (ExplodedNode *N = C.generateSink()) { + LazyInit_BT(BT_desc, BT); + + // Generate a report for this bug. + BugReport *R = new BugReport(*BT, BT->getName(), N); + R->addRange(argRange); + if (argEx) + R->addVisitor(bugreporter::getTrackNullOrUndefValueVisitor(N, argEx, + R)); + C.EmitReport(R); + } + return true; + } + + if (!checkUninitFields) + return false; + + if (const nonloc::LazyCompoundVal *LV = + dyn_cast<nonloc::LazyCompoundVal>(&V)) { + + class FindUninitializedField { + public: + SmallVector<const FieldDecl *, 10> FieldChain; + private: + ASTContext &C; + StoreManager &StoreMgr; + MemRegionManager &MrMgr; + Store store; + public: + FindUninitializedField(ASTContext &c, StoreManager &storeMgr, + MemRegionManager &mrMgr, Store s) + : C(c), StoreMgr(storeMgr), MrMgr(mrMgr), store(s) {} + + bool Find(const TypedValueRegion *R) { + QualType T = R->getValueType(); + if (const RecordType *RT = T->getAsStructureType()) { + const RecordDecl *RD = RT->getDecl()->getDefinition(); + assert(RD && "Referred record has no definition"); + for (RecordDecl::field_iterator I = + RD->field_begin(), E = RD->field_end(); I!=E; ++I) { + const FieldRegion *FR = MrMgr.getFieldRegion(*I, R); + FieldChain.push_back(*I); + T = (*I)->getType(); + if (T->getAsStructureType()) { + if (Find(FR)) + return true; + } + else { + const SVal &V = StoreMgr.getBinding(store, loc::MemRegionVal(FR)); + if (V.isUndef()) + return true; + } + FieldChain.pop_back(); + } + } + + return false; + } + }; + + const LazyCompoundValData *D = LV->getCVData(); + FindUninitializedField F(C.getASTContext(), + C.getState()->getStateManager().getStoreManager(), + C.getSValBuilder().getRegionManager(), + D->getStore()); + + if (F.Find(D->getRegion())) { + if (ExplodedNode *N = C.generateSink()) { + LazyInit_BT(BT_desc, BT); + SmallString<512> Str; + llvm::raw_svector_ostream os(Str); + os << "Passed-by-value struct argument contains uninitialized data"; + + if (F.FieldChain.size() == 1) + os << " (e.g., field: '" << *F.FieldChain[0] << "')"; + else { + os << " (e.g., via the field chain: '"; + bool first = true; + for (SmallVectorImpl<const FieldDecl *>::iterator + DI = F.FieldChain.begin(), DE = F.FieldChain.end(); DI!=DE;++DI){ + if (first) + first = false; + else + os << '.'; + os << **DI; + } + os << "')"; + } + + // Generate a report for this bug. + BugReport *R = new BugReport(*BT, os.str(), N); + R->addRange(argRange); + + // FIXME: enhance track back for uninitialized value for arbitrary + // memregions + C.EmitReport(R); + } + return true; + } + } + + return false; +} + +void CallAndMessageChecker::checkPreStmt(const CallExpr *CE, + CheckerContext &C) const{ + + const Expr *Callee = CE->getCallee()->IgnoreParens(); + const LocationContext *LCtx = C.getLocationContext(); + SVal L = C.getState()->getSVal(Callee, LCtx); + + if (L.isUndef()) { + if (!BT_call_undef) + BT_call_undef.reset(new BuiltinBug("Called function pointer is an " + "uninitalized pointer value")); + EmitBadCall(BT_call_undef.get(), C, CE); + return; + } + + if (isa<loc::ConcreteInt>(L)) { + if (!BT_call_null) + BT_call_null.reset( + new BuiltinBug("Called function pointer is null (null dereference)")); + EmitBadCall(BT_call_null.get(), C, CE); + } + + PreVisitProcessArgs(C, CallOrObjCMessage(CE, C.getState(), LCtx), + "Function call argument is an uninitialized value", + BT_call_arg); +} + +void CallAndMessageChecker::checkPreObjCMessage(ObjCMessage msg, + CheckerContext &C) const { + + ProgramStateRef state = C.getState(); + const LocationContext *LCtx = C.getLocationContext(); + + // FIXME: Handle 'super'? + if (const Expr *receiver = msg.getInstanceReceiver()) { + SVal recVal = state->getSVal(receiver, LCtx); + if (recVal.isUndef()) { + if (ExplodedNode *N = C.generateSink()) { + BugType *BT = 0; + if (msg.isPureMessageExpr()) { + if (!BT_msg_undef) + BT_msg_undef.reset(new BuiltinBug("Receiver in message expression " + "is an uninitialized value")); + BT = BT_msg_undef.get(); + } + else { + if (!BT_objc_prop_undef) + BT_objc_prop_undef.reset(new BuiltinBug("Property access on an " + "uninitialized object pointer")); + BT = BT_objc_prop_undef.get(); + } + BugReport *R = + new BugReport(*BT, BT->getName(), N); + R->addRange(receiver->getSourceRange()); + R->addVisitor(bugreporter::getTrackNullOrUndefValueVisitor(N, + receiver, + R)); + C.EmitReport(R); + } + return; + } else { + // Bifurcate the state into nil and non-nil ones. + DefinedOrUnknownSVal receiverVal = cast<DefinedOrUnknownSVal>(recVal); + + ProgramStateRef notNilState, nilState; + llvm::tie(notNilState, nilState) = state->assume(receiverVal); + + // Handle receiver must be nil. + if (nilState && !notNilState) { + HandleNilReceiver(C, state, msg); + return; + } + } + } + + const char *bugDesc = msg.isPropertySetter() ? + "Argument for property setter is an uninitialized value" + : "Argument in message expression is an uninitialized value"; + // Check for any arguments that are uninitialized/undefined. + PreVisitProcessArgs(C, CallOrObjCMessage(msg, state, LCtx), + bugDesc, BT_msg_arg); +} + +void CallAndMessageChecker::emitNilReceiverBug(CheckerContext &C, + const ObjCMessage &msg, + ExplodedNode *N) const { + + if (!BT_msg_ret) + BT_msg_ret.reset( + new BuiltinBug("Receiver in message expression is " + "'nil' and returns a garbage value")); + + SmallString<200> buf; + llvm::raw_svector_ostream os(buf); + os << "The receiver of message '" << msg.getSelector().getAsString() + << "' is nil and returns a value of type '" + << msg.getType(C.getASTContext()).getAsString() << "' that will be garbage"; + + BugReport *report = new BugReport(*BT_msg_ret, os.str(), N); + if (const Expr *receiver = msg.getInstanceReceiver()) { + report->addRange(receiver->getSourceRange()); + report->addVisitor(bugreporter::getTrackNullOrUndefValueVisitor(N, + receiver, + report)); + } + C.EmitReport(report); +} + +static bool supportsNilWithFloatRet(const llvm::Triple &triple) { + return (triple.getVendor() == llvm::Triple::Apple && + (triple.getOS() == llvm::Triple::IOS || + !triple.isMacOSXVersionLT(10,5))); +} + +void CallAndMessageChecker::HandleNilReceiver(CheckerContext &C, + ProgramStateRef state, + ObjCMessage msg) const { + ASTContext &Ctx = C.getASTContext(); + + // Check the return type of the message expression. A message to nil will + // return different values depending on the return type and the architecture. + QualType RetTy = msg.getType(Ctx); + CanQualType CanRetTy = Ctx.getCanonicalType(RetTy); + const LocationContext *LCtx = C.getLocationContext(); + + if (CanRetTy->isStructureOrClassType()) { + // Structure returns are safe since the compiler zeroes them out. + SVal V = C.getSValBuilder().makeZeroVal(msg.getType(Ctx)); + C.addTransition(state->BindExpr(msg.getMessageExpr(), LCtx, V)); + return; + } + + // Other cases: check if sizeof(return type) > sizeof(void*) + if (CanRetTy != Ctx.VoidTy && C.getLocationContext()->getParentMap() + .isConsumedExpr(msg.getMessageExpr())) { + // Compute: sizeof(void *) and sizeof(return type) + const uint64_t voidPtrSize = Ctx.getTypeSize(Ctx.VoidPtrTy); + const uint64_t returnTypeSize = Ctx.getTypeSize(CanRetTy); + + if (voidPtrSize < returnTypeSize && + !(supportsNilWithFloatRet(Ctx.getTargetInfo().getTriple()) && + (Ctx.FloatTy == CanRetTy || + Ctx.DoubleTy == CanRetTy || + Ctx.LongDoubleTy == CanRetTy || + Ctx.LongLongTy == CanRetTy || + Ctx.UnsignedLongLongTy == CanRetTy))) { + if (ExplodedNode *N = C.generateSink(state)) + emitNilReceiverBug(C, msg, N); + return; + } + + // Handle the safe cases where the return value is 0 if the + // receiver is nil. + // + // FIXME: For now take the conservative approach that we only + // return null values if we *know* that the receiver is nil. + // This is because we can have surprises like: + // + // ... = [[NSScreens screens] objectAtIndex:0]; + // + // What can happen is that [... screens] could return nil, but + // it most likely isn't nil. We should assume the semantics + // of this case unless we have *a lot* more knowledge. + // + SVal V = C.getSValBuilder().makeZeroVal(msg.getType(Ctx)); + C.addTransition(state->BindExpr(msg.getMessageExpr(), LCtx, V)); + return; + } + + C.addTransition(state); +} + +void ento::registerCallAndMessageChecker(CheckerManager &mgr) { + mgr.registerChecker<CallAndMessageChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/CastSizeChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/CastSizeChecker.cpp new file mode 100644 index 0000000..2e184fb --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/CastSizeChecker.cpp @@ -0,0 +1,86 @@ +//=== CastSizeChecker.cpp ---------------------------------------*- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// CastSizeChecker checks when casting a malloc'ed symbolic region to type T, +// whether the size of the symbolic region is a multiple of the size of T. +// +//===----------------------------------------------------------------------===// +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/AST/CharUnits.h" + +using namespace clang; +using namespace ento; + +namespace { +class CastSizeChecker : public Checker< check::PreStmt<CastExpr> > { + mutable OwningPtr<BuiltinBug> BT; +public: + void checkPreStmt(const CastExpr *CE, CheckerContext &C) const; +}; +} + +void CastSizeChecker::checkPreStmt(const CastExpr *CE,CheckerContext &C) const { + const Expr *E = CE->getSubExpr(); + ASTContext &Ctx = C.getASTContext(); + QualType ToTy = Ctx.getCanonicalType(CE->getType()); + const PointerType *ToPTy = dyn_cast<PointerType>(ToTy.getTypePtr()); + + if (!ToPTy) + return; + + QualType ToPointeeTy = ToPTy->getPointeeType(); + + // Only perform the check if 'ToPointeeTy' is a complete type. + if (ToPointeeTy->isIncompleteType()) + return; + + ProgramStateRef state = C.getState(); + const MemRegion *R = state->getSVal(E, C.getLocationContext()).getAsRegion(); + if (R == 0) + return; + + const SymbolicRegion *SR = dyn_cast<SymbolicRegion>(R); + if (SR == 0) + return; + + SValBuilder &svalBuilder = C.getSValBuilder(); + SVal extent = SR->getExtent(svalBuilder); + const llvm::APSInt *extentInt = svalBuilder.getKnownValue(state, extent); + if (!extentInt) + return; + + CharUnits regionSize = CharUnits::fromQuantity(extentInt->getSExtValue()); + CharUnits typeSize = C.getASTContext().getTypeSizeInChars(ToPointeeTy); + + // Ignore void, and a few other un-sizeable types. + if (typeSize.isZero()) + return; + + if (regionSize % typeSize != 0) { + if (ExplodedNode *errorNode = C.generateSink()) { + if (!BT) + BT.reset(new BuiltinBug("Cast region with wrong size.", + "Cast a region whose size is not a multiple of the" + " destination type size.")); + BugReport *R = new BugReport(*BT, BT->getDescription(), + errorNode); + R->addRange(CE->getSourceRange()); + C.EmitReport(R); + } + } +} + + +void ento::registerCastSizeChecker(CheckerManager &mgr) { + mgr.registerChecker<CastSizeChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/CastToStructChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/CastToStructChecker.cpp new file mode 100644 index 0000000..1407638 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/CastToStructChecker.cpp @@ -0,0 +1,74 @@ +//=== CastToStructChecker.cpp - Fixed address usage checker ----*- C++ -*--===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This files defines CastToStructChecker, a builtin checker that checks for +// cast from non-struct pointer to struct pointer. +// This check corresponds to CWE-588. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" + +using namespace clang; +using namespace ento; + +namespace { +class CastToStructChecker : public Checker< check::PreStmt<CastExpr> > { + mutable OwningPtr<BuiltinBug> BT; + +public: + void checkPreStmt(const CastExpr *CE, CheckerContext &C) const; +}; +} + +void CastToStructChecker::checkPreStmt(const CastExpr *CE, + CheckerContext &C) const { + const Expr *E = CE->getSubExpr(); + ASTContext &Ctx = C.getASTContext(); + QualType OrigTy = Ctx.getCanonicalType(E->getType()); + QualType ToTy = Ctx.getCanonicalType(CE->getType()); + + const PointerType *OrigPTy = dyn_cast<PointerType>(OrigTy.getTypePtr()); + const PointerType *ToPTy = dyn_cast<PointerType>(ToTy.getTypePtr()); + + if (!ToPTy || !OrigPTy) + return; + + QualType OrigPointeeTy = OrigPTy->getPointeeType(); + QualType ToPointeeTy = ToPTy->getPointeeType(); + + if (!ToPointeeTy->isStructureOrClassType()) + return; + + // We allow cast from void*. + if (OrigPointeeTy->isVoidType()) + return; + + // Now the cast-to-type is struct pointer, the original type is not void*. + if (!OrigPointeeTy->isRecordType()) { + if (ExplodedNode *N = C.addTransition()) { + if (!BT) + BT.reset(new BuiltinBug("Cast from non-struct type to struct type", + "Casting a non-structure type to a structure type " + "and accessing a field can lead to memory access " + "errors or data corruption.")); + BugReport *R = new BugReport(*BT,BT->getDescription(), N); + R->addRange(CE->getSourceRange()); + C.EmitReport(R); + } + } +} + +void ento::registerCastToStructChecker(CheckerManager &mgr) { + mgr.registerChecker<CastToStructChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/CheckObjCDealloc.cpp b/clang/lib/StaticAnalyzer/Checkers/CheckObjCDealloc.cpp new file mode 100644 index 0000000..133204a --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/CheckObjCDealloc.cpp @@ -0,0 +1,291 @@ +//==- CheckObjCDealloc.cpp - Check ObjC -dealloc implementation --*- C++ -*-==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines a CheckObjCDealloc, a checker that +// analyzes an Objective-C class's implementation to determine if it +// correctly implements -dealloc. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/AnalysisManager.h" +#include "clang/StaticAnalyzer/Core/BugReporter/PathDiagnostic.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugReporter.h" +#include "clang/AST/ExprObjC.h" +#include "clang/AST/Expr.h" +#include "clang/AST/DeclObjC.h" +#include "clang/Basic/LangOptions.h" +#include "llvm/Support/raw_ostream.h" + +using namespace clang; +using namespace ento; + +static bool scan_dealloc(Stmt *S, Selector Dealloc) { + + if (ObjCMessageExpr *ME = dyn_cast<ObjCMessageExpr>(S)) + if (ME->getSelector() == Dealloc) { + switch (ME->getReceiverKind()) { + case ObjCMessageExpr::Instance: return false; + case ObjCMessageExpr::SuperInstance: return true; + case ObjCMessageExpr::Class: break; + case ObjCMessageExpr::SuperClass: break; + } + } + + // Recurse to children. + + for (Stmt::child_iterator I = S->child_begin(), E= S->child_end(); I!=E; ++I) + if (*I && scan_dealloc(*I, Dealloc)) + return true; + + return false; +} + +static bool scan_ivar_release(Stmt *S, ObjCIvarDecl *ID, + const ObjCPropertyDecl *PD, + Selector Release, + IdentifierInfo* SelfII, + ASTContext &Ctx) { + + // [mMyIvar release] + if (ObjCMessageExpr *ME = dyn_cast<ObjCMessageExpr>(S)) + if (ME->getSelector() == Release) + if (ME->getInstanceReceiver()) + if (Expr *Receiver = ME->getInstanceReceiver()->IgnoreParenCasts()) + if (ObjCIvarRefExpr *E = dyn_cast<ObjCIvarRefExpr>(Receiver)) + if (E->getDecl() == ID) + return true; + + // [self setMyIvar:nil]; + if (ObjCMessageExpr *ME = dyn_cast<ObjCMessageExpr>(S)) + if (ME->getInstanceReceiver()) + if (Expr *Receiver = ME->getInstanceReceiver()->IgnoreParenCasts()) + if (DeclRefExpr *E = dyn_cast<DeclRefExpr>(Receiver)) + if (E->getDecl()->getIdentifier() == SelfII) + if (ME->getMethodDecl() == PD->getSetterMethodDecl() && + ME->getNumArgs() == 1 && + ME->getArg(0)->isNullPointerConstant(Ctx, + Expr::NPC_ValueDependentIsNull)) + return true; + + // self.myIvar = nil; + if (BinaryOperator* BO = dyn_cast<BinaryOperator>(S)) + if (BO->isAssignmentOp()) + if (ObjCPropertyRefExpr *PRE = + dyn_cast<ObjCPropertyRefExpr>(BO->getLHS()->IgnoreParenCasts())) + if (PRE->isExplicitProperty() && PRE->getExplicitProperty() == PD) + if (BO->getRHS()->isNullPointerConstant(Ctx, + Expr::NPC_ValueDependentIsNull)) { + // This is only a 'release' if the property kind is not + // 'assign'. + return PD->getSetterKind() != ObjCPropertyDecl::Assign;; + } + + // Recurse to children. + for (Stmt::child_iterator I = S->child_begin(), E= S->child_end(); I!=E; ++I) + if (*I && scan_ivar_release(*I, ID, PD, Release, SelfII, Ctx)) + return true; + + return false; +} + +static void checkObjCDealloc(const ObjCImplementationDecl *D, + const LangOptions& LOpts, BugReporter& BR) { + + assert (LOpts.getGC() != LangOptions::GCOnly); + + ASTContext &Ctx = BR.getContext(); + const ObjCInterfaceDecl *ID = D->getClassInterface(); + + // Does the class contain any ivars that are pointers (or id<...>)? + // If not, skip the check entirely. + // NOTE: This is motivated by PR 2517: + // http://llvm.org/bugs/show_bug.cgi?id=2517 + + bool containsPointerIvar = false; + + for (ObjCInterfaceDecl::ivar_iterator I=ID->ivar_begin(), E=ID->ivar_end(); + I!=E; ++I) { + + ObjCIvarDecl *ID = *I; + QualType T = ID->getType(); + + if (!T->isObjCObjectPointerType() || + ID->getAttr<IBOutletAttr>() || // Skip IBOutlets. + ID->getAttr<IBOutletCollectionAttr>()) // Skip IBOutletCollections. + continue; + + containsPointerIvar = true; + break; + } + + if (!containsPointerIvar) + return; + + // Determine if the class subclasses NSObject. + IdentifierInfo* NSObjectII = &Ctx.Idents.get("NSObject"); + IdentifierInfo* SenTestCaseII = &Ctx.Idents.get("SenTestCase"); + + + for ( ; ID ; ID = ID->getSuperClass()) { + IdentifierInfo *II = ID->getIdentifier(); + + if (II == NSObjectII) + break; + + // FIXME: For now, ignore classes that subclass SenTestCase, as these don't + // need to implement -dealloc. They implement tear down in another way, + // which we should try and catch later. + // http://llvm.org/bugs/show_bug.cgi?id=3187 + if (II == SenTestCaseII) + return; + } + + if (!ID) + return; + + // Get the "dealloc" selector. + IdentifierInfo* II = &Ctx.Idents.get("dealloc"); + Selector S = Ctx.Selectors.getSelector(0, &II); + ObjCMethodDecl *MD = 0; + + // Scan the instance methods for "dealloc". + for (ObjCImplementationDecl::instmeth_iterator I = D->instmeth_begin(), + E = D->instmeth_end(); I!=E; ++I) { + + if ((*I)->getSelector() == S) { + MD = *I; + break; + } + } + + PathDiagnosticLocation DLoc = + PathDiagnosticLocation::createBegin(D, BR.getSourceManager()); + + if (!MD) { // No dealloc found. + + const char* name = LOpts.getGC() == LangOptions::NonGC + ? "missing -dealloc" + : "missing -dealloc (Hybrid MM, non-GC)"; + + std::string buf; + llvm::raw_string_ostream os(buf); + os << "Objective-C class '" << *D << "' lacks a 'dealloc' instance method"; + + BR.EmitBasicReport(D, name, categories::CoreFoundationObjectiveC, + os.str(), DLoc); + return; + } + + // dealloc found. Scan for missing [super dealloc]. + if (MD->getBody() && !scan_dealloc(MD->getBody(), S)) { + + const char* name = LOpts.getGC() == LangOptions::NonGC + ? "missing [super dealloc]" + : "missing [super dealloc] (Hybrid MM, non-GC)"; + + std::string buf; + llvm::raw_string_ostream os(buf); + os << "The 'dealloc' instance method in Objective-C class '" << *D + << "' does not send a 'dealloc' message to its super class" + " (missing [super dealloc])"; + + BR.EmitBasicReport(MD, name, categories::CoreFoundationObjectiveC, + os.str(), DLoc); + return; + } + + // Get the "release" selector. + IdentifierInfo* RII = &Ctx.Idents.get("release"); + Selector RS = Ctx.Selectors.getSelector(0, &RII); + + // Get the "self" identifier + IdentifierInfo* SelfII = &Ctx.Idents.get("self"); + + // Scan for missing and extra releases of ivars used by implementations + // of synthesized properties + for (ObjCImplementationDecl::propimpl_iterator I = D->propimpl_begin(), + E = D->propimpl_end(); I!=E; ++I) { + + // We can only check the synthesized properties + if ((*I)->getPropertyImplementation() != ObjCPropertyImplDecl::Synthesize) + continue; + + ObjCIvarDecl *ID = (*I)->getPropertyIvarDecl(); + if (!ID) + continue; + + QualType T = ID->getType(); + if (!T->isObjCObjectPointerType()) // Skip non-pointer ivars + continue; + + const ObjCPropertyDecl *PD = (*I)->getPropertyDecl(); + if (!PD) + continue; + + // ivars cannot be set via read-only properties, so we'll skip them + if (PD->isReadOnly()) + continue; + + // ivar must be released if and only if the kind of setter was not 'assign' + bool requiresRelease = PD->getSetterKind() != ObjCPropertyDecl::Assign; + if (scan_ivar_release(MD->getBody(), ID, PD, RS, SelfII, Ctx) + != requiresRelease) { + const char *name = 0; + std::string buf; + llvm::raw_string_ostream os(buf); + + if (requiresRelease) { + name = LOpts.getGC() == LangOptions::NonGC + ? "missing ivar release (leak)" + : "missing ivar release (Hybrid MM, non-GC)"; + + os << "The '" << *ID + << "' instance variable was retained by a synthesized property but " + "wasn't released in 'dealloc'"; + } else { + name = LOpts.getGC() == LangOptions::NonGC + ? "extra ivar release (use-after-release)" + : "extra ivar release (Hybrid MM, non-GC)"; + + os << "The '" << *ID + << "' instance variable was not retained by a synthesized property " + "but was released in 'dealloc'"; + } + + PathDiagnosticLocation SDLoc = + PathDiagnosticLocation::createBegin((*I), BR.getSourceManager()); + + BR.EmitBasicReport(MD, name, categories::CoreFoundationObjectiveC, + os.str(), SDLoc); + } + } +} + +//===----------------------------------------------------------------------===// +// ObjCDeallocChecker +//===----------------------------------------------------------------------===// + +namespace { +class ObjCDeallocChecker : public Checker< + check::ASTDecl<ObjCImplementationDecl> > { +public: + void checkASTDecl(const ObjCImplementationDecl *D, AnalysisManager& mgr, + BugReporter &BR) const { + if (mgr.getLangOpts().getGC() == LangOptions::GCOnly) + return; + checkObjCDealloc(cast<ObjCImplementationDecl>(D), mgr.getLangOpts(), BR); + } +}; +} + +void ento::registerObjCDeallocChecker(CheckerManager &mgr) { + mgr.registerChecker<ObjCDeallocChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/CheckObjCInstMethSignature.cpp b/clang/lib/StaticAnalyzer/Checkers/CheckObjCInstMethSignature.cpp new file mode 100644 index 0000000..6df47b1 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/CheckObjCInstMethSignature.cpp @@ -0,0 +1,146 @@ +//=- CheckObjCInstMethodRetTy.cpp - Check ObjC method signatures -*- C++ -*-==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines a CheckObjCInstMethSignature, a flow-insenstive check +// that determines if an Objective-C class interface incorrectly redefines +// the method signature in a subclass. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/BugReporter/PathDiagnostic.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugReporter.h" +#include "clang/AST/DeclObjC.h" +#include "clang/AST/Type.h" +#include "clang/AST/ASTContext.h" + +#include "llvm/ADT/DenseMap.h" +#include "llvm/Support/raw_ostream.h" + +using namespace clang; +using namespace ento; + +static bool AreTypesCompatible(QualType Derived, QualType Ancestor, + ASTContext &C) { + + // Right now don't compare the compatibility of pointers. That involves + // looking at subtyping relationships. FIXME: Future patch. + if (Derived->isAnyPointerType() && Ancestor->isAnyPointerType()) + return true; + + return C.typesAreCompatible(Derived, Ancestor); +} + +static void CompareReturnTypes(const ObjCMethodDecl *MethDerived, + const ObjCMethodDecl *MethAncestor, + BugReporter &BR, ASTContext &Ctx, + const ObjCImplementationDecl *ID) { + + QualType ResDerived = MethDerived->getResultType(); + QualType ResAncestor = MethAncestor->getResultType(); + + if (!AreTypesCompatible(ResDerived, ResAncestor, Ctx)) { + std::string sbuf; + llvm::raw_string_ostream os(sbuf); + + os << "The Objective-C class '" + << *MethDerived->getClassInterface() + << "', which is derived from class '" + << *MethAncestor->getClassInterface() + << "', defines the instance method '" + << MethDerived->getSelector().getAsString() + << "' whose return type is '" + << ResDerived.getAsString() + << "'. A method with the same name (same selector) is also defined in " + "class '" + << *MethAncestor->getClassInterface() + << "' and has a return type of '" + << ResAncestor.getAsString() + << "'. These two types are incompatible, and may result in undefined " + "behavior for clients of these classes."; + + PathDiagnosticLocation MethDLoc = + PathDiagnosticLocation::createBegin(MethDerived, + BR.getSourceManager()); + + BR.EmitBasicReport(MethDerived, + "Incompatible instance method return type", + categories::CoreFoundationObjectiveC, + os.str(), MethDLoc); + } +} + +static void CheckObjCInstMethSignature(const ObjCImplementationDecl *ID, + BugReporter& BR) { + + const ObjCInterfaceDecl *D = ID->getClassInterface(); + const ObjCInterfaceDecl *C = D->getSuperClass(); + + if (!C) + return; + + ASTContext &Ctx = BR.getContext(); + + // Build a DenseMap of the methods for quick querying. + typedef llvm::DenseMap<Selector,ObjCMethodDecl*> MapTy; + MapTy IMeths; + unsigned NumMethods = 0; + + for (ObjCImplementationDecl::instmeth_iterator I=ID->instmeth_begin(), + E=ID->instmeth_end(); I!=E; ++I) { + + ObjCMethodDecl *M = *I; + IMeths[M->getSelector()] = M; + ++NumMethods; + } + + // Now recurse the class hierarchy chain looking for methods with the + // same signatures. + while (C && NumMethods) { + for (ObjCInterfaceDecl::instmeth_iterator I=C->instmeth_begin(), + E=C->instmeth_end(); I!=E; ++I) { + + ObjCMethodDecl *M = *I; + Selector S = M->getSelector(); + + MapTy::iterator MI = IMeths.find(S); + + if (MI == IMeths.end() || MI->second == 0) + continue; + + --NumMethods; + ObjCMethodDecl *MethDerived = MI->second; + MI->second = 0; + + CompareReturnTypes(MethDerived, M, BR, Ctx, ID); + } + + C = C->getSuperClass(); + } +} + +//===----------------------------------------------------------------------===// +// ObjCMethSigsChecker +//===----------------------------------------------------------------------===// + +namespace { +class ObjCMethSigsChecker : public Checker< + check::ASTDecl<ObjCImplementationDecl> > { +public: + void checkASTDecl(const ObjCImplementationDecl *D, AnalysisManager& mgr, + BugReporter &BR) const { + CheckObjCInstMethSignature(D, BR); + } +}; +} + +void ento::registerObjCMethSigsChecker(CheckerManager &mgr) { + mgr.registerChecker<ObjCMethSigsChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/CheckSecuritySyntaxOnly.cpp b/clang/lib/StaticAnalyzer/Checkers/CheckSecuritySyntaxOnly.cpp new file mode 100644 index 0000000..dde9071 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/CheckSecuritySyntaxOnly.cpp @@ -0,0 +1,786 @@ +//==- CheckSecuritySyntaxOnly.cpp - Basic security checks --------*- C++ -*-==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines a set of flow-insensitive security checks. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/Analysis/AnalysisContext.h" +#include "clang/AST/StmtVisitor.h" +#include "clang/Basic/TargetInfo.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugReporter.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/AnalysisManager.h" +#include "llvm/ADT/SmallString.h" +#include "llvm/ADT/StringSwitch.h" +#include "llvm/Support/raw_ostream.h" + +using namespace clang; +using namespace ento; + +static bool isArc4RandomAvailable(const ASTContext &Ctx) { + const llvm::Triple &T = Ctx.getTargetInfo().getTriple(); + return T.getVendor() == llvm::Triple::Apple || + T.getOS() == llvm::Triple::FreeBSD || + T.getOS() == llvm::Triple::NetBSD || + T.getOS() == llvm::Triple::OpenBSD || + T.getOS() == llvm::Triple::DragonFly; +} + +namespace { +struct DefaultBool { + bool val; + DefaultBool() : val(false) {} + operator bool() const { return val; } + DefaultBool &operator=(bool b) { val = b; return *this; } +}; + +struct ChecksFilter { + DefaultBool check_gets; + DefaultBool check_getpw; + DefaultBool check_mktemp; + DefaultBool check_mkstemp; + DefaultBool check_strcpy; + DefaultBool check_rand; + DefaultBool check_vfork; + DefaultBool check_FloatLoopCounter; + DefaultBool check_UncheckedReturn; +}; + +class WalkAST : public StmtVisitor<WalkAST> { + BugReporter &BR; + AnalysisDeclContext* AC; + enum { num_setids = 6 }; + IdentifierInfo *II_setid[num_setids]; + + const bool CheckRand; + const ChecksFilter &filter; + +public: + WalkAST(BugReporter &br, AnalysisDeclContext* ac, + const ChecksFilter &f) + : BR(br), AC(ac), II_setid(), + CheckRand(isArc4RandomAvailable(BR.getContext())), + filter(f) {} + + // Statement visitor methods. + void VisitCallExpr(CallExpr *CE); + void VisitForStmt(ForStmt *S); + void VisitCompoundStmt (CompoundStmt *S); + void VisitStmt(Stmt *S) { VisitChildren(S); } + + void VisitChildren(Stmt *S); + + // Helpers. + bool checkCall_strCommon(const CallExpr *CE, const FunctionDecl *FD); + + typedef void (WalkAST::*FnCheck)(const CallExpr *, + const FunctionDecl *); + + // Checker-specific methods. + void checkLoopConditionForFloat(const ForStmt *FS); + void checkCall_gets(const CallExpr *CE, const FunctionDecl *FD); + void checkCall_getpw(const CallExpr *CE, const FunctionDecl *FD); + void checkCall_mktemp(const CallExpr *CE, const FunctionDecl *FD); + void checkCall_mkstemp(const CallExpr *CE, const FunctionDecl *FD); + void checkCall_strcpy(const CallExpr *CE, const FunctionDecl *FD); + void checkCall_strcat(const CallExpr *CE, const FunctionDecl *FD); + void checkCall_rand(const CallExpr *CE, const FunctionDecl *FD); + void checkCall_random(const CallExpr *CE, const FunctionDecl *FD); + void checkCall_vfork(const CallExpr *CE, const FunctionDecl *FD); + void checkUncheckedReturnValue(CallExpr *CE); +}; +} // end anonymous namespace + +//===----------------------------------------------------------------------===// +// AST walking. +//===----------------------------------------------------------------------===// + +void WalkAST::VisitChildren(Stmt *S) { + for (Stmt::child_iterator I = S->child_begin(), E = S->child_end(); I!=E; ++I) + if (Stmt *child = *I) + Visit(child); +} + +void WalkAST::VisitCallExpr(CallExpr *CE) { + // Get the callee. + const FunctionDecl *FD = CE->getDirectCallee(); + + if (!FD) + return; + + // Get the name of the callee. If it's a builtin, strip off the prefix. + IdentifierInfo *II = FD->getIdentifier(); + if (!II) // if no identifier, not a simple C function + return; + StringRef Name = II->getName(); + if (Name.startswith("__builtin_")) + Name = Name.substr(10); + + // Set the evaluation function by switching on the callee name. + FnCheck evalFunction = llvm::StringSwitch<FnCheck>(Name) + .Case("gets", &WalkAST::checkCall_gets) + .Case("getpw", &WalkAST::checkCall_getpw) + .Case("mktemp", &WalkAST::checkCall_mktemp) + .Case("mkstemp", &WalkAST::checkCall_mkstemp) + .Case("mkdtemp", &WalkAST::checkCall_mkstemp) + .Case("mkstemps", &WalkAST::checkCall_mkstemp) + .Cases("strcpy", "__strcpy_chk", &WalkAST::checkCall_strcpy) + .Cases("strcat", "__strcat_chk", &WalkAST::checkCall_strcat) + .Case("drand48", &WalkAST::checkCall_rand) + .Case("erand48", &WalkAST::checkCall_rand) + .Case("jrand48", &WalkAST::checkCall_rand) + .Case("lrand48", &WalkAST::checkCall_rand) + .Case("mrand48", &WalkAST::checkCall_rand) + .Case("nrand48", &WalkAST::checkCall_rand) + .Case("lcong48", &WalkAST::checkCall_rand) + .Case("rand", &WalkAST::checkCall_rand) + .Case("rand_r", &WalkAST::checkCall_rand) + .Case("random", &WalkAST::checkCall_random) + .Case("vfork", &WalkAST::checkCall_vfork) + .Default(NULL); + + // If the callee isn't defined, it is not of security concern. + // Check and evaluate the call. + if (evalFunction) + (this->*evalFunction)(CE, FD); + + // Recurse and check children. + VisitChildren(CE); +} + +void WalkAST::VisitCompoundStmt(CompoundStmt *S) { + for (Stmt::child_iterator I = S->child_begin(), E = S->child_end(); I!=E; ++I) + if (Stmt *child = *I) { + if (CallExpr *CE = dyn_cast<CallExpr>(child)) + checkUncheckedReturnValue(CE); + Visit(child); + } +} + +void WalkAST::VisitForStmt(ForStmt *FS) { + checkLoopConditionForFloat(FS); + + // Recurse and check children. + VisitChildren(FS); +} + +//===----------------------------------------------------------------------===// +// Check: floating poing variable used as loop counter. +// Originally: <rdar://problem/6336718> +// Implements: CERT security coding advisory FLP-30. +//===----------------------------------------------------------------------===// + +static const DeclRefExpr* +getIncrementedVar(const Expr *expr, const VarDecl *x, const VarDecl *y) { + expr = expr->IgnoreParenCasts(); + + if (const BinaryOperator *B = dyn_cast<BinaryOperator>(expr)) { + if (!(B->isAssignmentOp() || B->isCompoundAssignmentOp() || + B->getOpcode() == BO_Comma)) + return NULL; + + if (const DeclRefExpr *lhs = getIncrementedVar(B->getLHS(), x, y)) + return lhs; + + if (const DeclRefExpr *rhs = getIncrementedVar(B->getRHS(), x, y)) + return rhs; + + return NULL; + } + + if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(expr)) { + const NamedDecl *ND = DR->getDecl(); + return ND == x || ND == y ? DR : NULL; + } + + if (const UnaryOperator *U = dyn_cast<UnaryOperator>(expr)) + return U->isIncrementDecrementOp() + ? getIncrementedVar(U->getSubExpr(), x, y) : NULL; + + return NULL; +} + +/// CheckLoopConditionForFloat - This check looks for 'for' statements that +/// use a floating point variable as a loop counter. +/// CERT: FLP30-C, FLP30-CPP. +/// +void WalkAST::checkLoopConditionForFloat(const ForStmt *FS) { + if (!filter.check_FloatLoopCounter) + return; + + // Does the loop have a condition? + const Expr *condition = FS->getCond(); + + if (!condition) + return; + + // Does the loop have an increment? + const Expr *increment = FS->getInc(); + + if (!increment) + return; + + // Strip away '()' and casts. + condition = condition->IgnoreParenCasts(); + increment = increment->IgnoreParenCasts(); + + // Is the loop condition a comparison? + const BinaryOperator *B = dyn_cast<BinaryOperator>(condition); + + if (!B) + return; + + // Is this a comparison? + if (!(B->isRelationalOp() || B->isEqualityOp())) + return; + + // Are we comparing variables? + const DeclRefExpr *drLHS = + dyn_cast<DeclRefExpr>(B->getLHS()->IgnoreParenLValueCasts()); + const DeclRefExpr *drRHS = + dyn_cast<DeclRefExpr>(B->getRHS()->IgnoreParenLValueCasts()); + + // Does at least one of the variables have a floating point type? + drLHS = drLHS && drLHS->getType()->isRealFloatingType() ? drLHS : NULL; + drRHS = drRHS && drRHS->getType()->isRealFloatingType() ? drRHS : NULL; + + if (!drLHS && !drRHS) + return; + + const VarDecl *vdLHS = drLHS ? dyn_cast<VarDecl>(drLHS->getDecl()) : NULL; + const VarDecl *vdRHS = drRHS ? dyn_cast<VarDecl>(drRHS->getDecl()) : NULL; + + if (!vdLHS && !vdRHS) + return; + + // Does either variable appear in increment? + const DeclRefExpr *drInc = getIncrementedVar(increment, vdLHS, vdRHS); + + if (!drInc) + return; + + // Emit the error. First figure out which DeclRefExpr in the condition + // referenced the compared variable. + const DeclRefExpr *drCond = vdLHS == drInc->getDecl() ? drLHS : drRHS; + + SmallVector<SourceRange, 2> ranges; + SmallString<256> sbuf; + llvm::raw_svector_ostream os(sbuf); + + os << "Variable '" << drCond->getDecl()->getName() + << "' with floating point type '" << drCond->getType().getAsString() + << "' should not be used as a loop counter"; + + ranges.push_back(drCond->getSourceRange()); + ranges.push_back(drInc->getSourceRange()); + + const char *bugType = "Floating point variable used as loop counter"; + + PathDiagnosticLocation FSLoc = + PathDiagnosticLocation::createBegin(FS, BR.getSourceManager(), AC); + BR.EmitBasicReport(AC->getDecl(), + bugType, "Security", os.str(), + FSLoc, ranges.data(), ranges.size()); +} + +//===----------------------------------------------------------------------===// +// Check: Any use of 'gets' is insecure. +// Originally: <rdar://problem/6335715> +// Implements (part of): 300-BSI (buildsecurityin.us-cert.gov) +// CWE-242: Use of Inherently Dangerous Function +//===----------------------------------------------------------------------===// + +void WalkAST::checkCall_gets(const CallExpr *CE, const FunctionDecl *FD) { + if (!filter.check_gets) + return; + + const FunctionProtoType *FPT + = dyn_cast<FunctionProtoType>(FD->getType().IgnoreParens()); + if (!FPT) + return; + + // Verify that the function takes a single argument. + if (FPT->getNumArgs() != 1) + return; + + // Is the argument a 'char*'? + const PointerType *PT = dyn_cast<PointerType>(FPT->getArgType(0)); + if (!PT) + return; + + if (PT->getPointeeType().getUnqualifiedType() != BR.getContext().CharTy) + return; + + // Issue a warning. + SourceRange R = CE->getCallee()->getSourceRange(); + PathDiagnosticLocation CELoc = + PathDiagnosticLocation::createBegin(CE, BR.getSourceManager(), AC); + BR.EmitBasicReport(AC->getDecl(), + "Potential buffer overflow in call to 'gets'", + "Security", + "Call to function 'gets' is extremely insecure as it can " + "always result in a buffer overflow", + CELoc, &R, 1); +} + +//===----------------------------------------------------------------------===// +// Check: Any use of 'getpwd' is insecure. +// CWE-477: Use of Obsolete Functions +//===----------------------------------------------------------------------===// + +void WalkAST::checkCall_getpw(const CallExpr *CE, const FunctionDecl *FD) { + if (!filter.check_getpw) + return; + + const FunctionProtoType *FPT + = dyn_cast<FunctionProtoType>(FD->getType().IgnoreParens()); + if (!FPT) + return; + + // Verify that the function takes two arguments. + if (FPT->getNumArgs() != 2) + return; + + // Verify the first argument type is integer. + if (!FPT->getArgType(0)->isIntegerType()) + return; + + // Verify the second argument type is char*. + const PointerType *PT = dyn_cast<PointerType>(FPT->getArgType(1)); + if (!PT) + return; + + if (PT->getPointeeType().getUnqualifiedType() != BR.getContext().CharTy) + return; + + // Issue a warning. + SourceRange R = CE->getCallee()->getSourceRange(); + PathDiagnosticLocation CELoc = + PathDiagnosticLocation::createBegin(CE, BR.getSourceManager(), AC); + BR.EmitBasicReport(AC->getDecl(), + "Potential buffer overflow in call to 'getpw'", + "Security", + "The getpw() function is dangerous as it may overflow the " + "provided buffer. It is obsoleted by getpwuid().", + CELoc, &R, 1); +} + +//===----------------------------------------------------------------------===// +// Check: Any use of 'mktemp' is insecure. It is obsoleted by mkstemp(). +// CWE-377: Insecure Temporary File +//===----------------------------------------------------------------------===// + +void WalkAST::checkCall_mktemp(const CallExpr *CE, const FunctionDecl *FD) { + if (!filter.check_mktemp) { + // Fall back to the security check of looking for enough 'X's in the + // format string, since that is a less severe warning. + checkCall_mkstemp(CE, FD); + return; + } + + const FunctionProtoType *FPT + = dyn_cast<FunctionProtoType>(FD->getType().IgnoreParens()); + if(!FPT) + return; + + // Verify that the function takes a single argument. + if (FPT->getNumArgs() != 1) + return; + + // Verify that the argument is Pointer Type. + const PointerType *PT = dyn_cast<PointerType>(FPT->getArgType(0)); + if (!PT) + return; + + // Verify that the argument is a 'char*'. + if (PT->getPointeeType().getUnqualifiedType() != BR.getContext().CharTy) + return; + + // Issue a waring. + SourceRange R = CE->getCallee()->getSourceRange(); + PathDiagnosticLocation CELoc = + PathDiagnosticLocation::createBegin(CE, BR.getSourceManager(), AC); + BR.EmitBasicReport(AC->getDecl(), + "Potential insecure temporary file in call 'mktemp'", + "Security", + "Call to function 'mktemp' is insecure as it always " + "creates or uses insecure temporary file. Use 'mkstemp' " + "instead", + CELoc, &R, 1); +} + + +//===----------------------------------------------------------------------===// +// Check: Use of 'mkstemp', 'mktemp', 'mkdtemp' should contain at least 6 X's. +//===----------------------------------------------------------------------===// + +void WalkAST::checkCall_mkstemp(const CallExpr *CE, const FunctionDecl *FD) { + if (!filter.check_mkstemp) + return; + + StringRef Name = FD->getIdentifier()->getName(); + std::pair<signed, signed> ArgSuffix = + llvm::StringSwitch<std::pair<signed, signed> >(Name) + .Case("mktemp", std::make_pair(0,-1)) + .Case("mkstemp", std::make_pair(0,-1)) + .Case("mkdtemp", std::make_pair(0,-1)) + .Case("mkstemps", std::make_pair(0,1)) + .Default(std::make_pair(-1, -1)); + + assert(ArgSuffix.first >= 0 && "Unsupported function"); + + // Check if the number of arguments is consistent with out expectations. + unsigned numArgs = CE->getNumArgs(); + if ((signed) numArgs <= ArgSuffix.first) + return; + + const StringLiteral *strArg = + dyn_cast<StringLiteral>(CE->getArg((unsigned)ArgSuffix.first) + ->IgnoreParenImpCasts()); + + // Currently we only handle string literals. It is possible to do better, + // either by looking at references to const variables, or by doing real + // flow analysis. + if (!strArg || strArg->getCharByteWidth() != 1) + return; + + // Count the number of X's, taking into account a possible cutoff suffix. + StringRef str = strArg->getString(); + unsigned numX = 0; + unsigned n = str.size(); + + // Take into account the suffix. + unsigned suffix = 0; + if (ArgSuffix.second >= 0) { + const Expr *suffixEx = CE->getArg((unsigned)ArgSuffix.second); + llvm::APSInt Result; + if (!suffixEx->EvaluateAsInt(Result, BR.getContext())) + return; + // FIXME: Issue a warning. + if (Result.isNegative()) + return; + suffix = (unsigned) Result.getZExtValue(); + n = (n > suffix) ? n - suffix : 0; + } + + for (unsigned i = 0; i < n; ++i) + if (str[i] == 'X') ++numX; + + if (numX >= 6) + return; + + // Issue a warning. + SourceRange R = strArg->getSourceRange(); + PathDiagnosticLocation CELoc = + PathDiagnosticLocation::createBegin(CE, BR.getSourceManager(), AC); + SmallString<512> buf; + llvm::raw_svector_ostream out(buf); + out << "Call to '" << Name << "' should have at least 6 'X's in the" + " format string to be secure (" << numX << " 'X'"; + if (numX != 1) + out << 's'; + out << " seen"; + if (suffix) { + out << ", " << suffix << " character"; + if (suffix > 1) + out << 's'; + out << " used as a suffix"; + } + out << ')'; + BR.EmitBasicReport(AC->getDecl(), + "Insecure temporary file creation", "Security", + out.str(), CELoc, &R, 1); +} + +//===----------------------------------------------------------------------===// +// Check: Any use of 'strcpy' is insecure. +// +// CWE-119: Improper Restriction of Operations within +// the Bounds of a Memory Buffer +//===----------------------------------------------------------------------===// +void WalkAST::checkCall_strcpy(const CallExpr *CE, const FunctionDecl *FD) { + if (!filter.check_strcpy) + return; + + if (!checkCall_strCommon(CE, FD)) + return; + + // Issue a warning. + SourceRange R = CE->getCallee()->getSourceRange(); + PathDiagnosticLocation CELoc = + PathDiagnosticLocation::createBegin(CE, BR.getSourceManager(), AC); + BR.EmitBasicReport(AC->getDecl(), + "Potential insecure memory buffer bounds restriction in " + "call 'strcpy'", + "Security", + "Call to function 'strcpy' is insecure as it does not " + "provide bounding of the memory buffer. Replace " + "unbounded copy functions with analogous functions that " + "support length arguments such as 'strlcpy'. CWE-119.", + CELoc, &R, 1); +} + +//===----------------------------------------------------------------------===// +// Check: Any use of 'strcat' is insecure. +// +// CWE-119: Improper Restriction of Operations within +// the Bounds of a Memory Buffer +//===----------------------------------------------------------------------===// +void WalkAST::checkCall_strcat(const CallExpr *CE, const FunctionDecl *FD) { + if (!filter.check_strcpy) + return; + + if (!checkCall_strCommon(CE, FD)) + return; + + // Issue a warning. + SourceRange R = CE->getCallee()->getSourceRange(); + PathDiagnosticLocation CELoc = + PathDiagnosticLocation::createBegin(CE, BR.getSourceManager(), AC); + BR.EmitBasicReport(AC->getDecl(), + "Potential insecure memory buffer bounds restriction in " + "call 'strcat'", + "Security", + "Call to function 'strcat' is insecure as it does not " + "provide bounding of the memory buffer. Replace " + "unbounded copy functions with analogous functions that " + "support length arguments such as 'strlcat'. CWE-119.", + CELoc, &R, 1); +} + +//===----------------------------------------------------------------------===// +// Common check for str* functions with no bounds parameters. +//===----------------------------------------------------------------------===// +bool WalkAST::checkCall_strCommon(const CallExpr *CE, const FunctionDecl *FD) { + const FunctionProtoType *FPT + = dyn_cast<FunctionProtoType>(FD->getType().IgnoreParens()); + if (!FPT) + return false; + + // Verify the function takes two arguments, three in the _chk version. + int numArgs = FPT->getNumArgs(); + if (numArgs != 2 && numArgs != 3) + return false; + + // Verify the type for both arguments. + for (int i = 0; i < 2; i++) { + // Verify that the arguments are pointers. + const PointerType *PT = dyn_cast<PointerType>(FPT->getArgType(i)); + if (!PT) + return false; + + // Verify that the argument is a 'char*'. + if (PT->getPointeeType().getUnqualifiedType() != BR.getContext().CharTy) + return false; + } + + return true; +} + +//===----------------------------------------------------------------------===// +// Check: Linear congruent random number generators should not be used +// Originally: <rdar://problem/63371000> +// CWE-338: Use of cryptographically weak prng +//===----------------------------------------------------------------------===// + +void WalkAST::checkCall_rand(const CallExpr *CE, const FunctionDecl *FD) { + if (!filter.check_rand || !CheckRand) + return; + + const FunctionProtoType *FTP + = dyn_cast<FunctionProtoType>(FD->getType().IgnoreParens()); + if (!FTP) + return; + + if (FTP->getNumArgs() == 1) { + // Is the argument an 'unsigned short *'? + // (Actually any integer type is allowed.) + const PointerType *PT = dyn_cast<PointerType>(FTP->getArgType(0)); + if (!PT) + return; + + if (! PT->getPointeeType()->isIntegerType()) + return; + } + else if (FTP->getNumArgs() != 0) + return; + + // Issue a warning. + SmallString<256> buf1; + llvm::raw_svector_ostream os1(buf1); + os1 << '\'' << *FD << "' is a poor random number generator"; + + SmallString<256> buf2; + llvm::raw_svector_ostream os2(buf2); + os2 << "Function '" << *FD + << "' is obsolete because it implements a poor random number generator." + << " Use 'arc4random' instead"; + + SourceRange R = CE->getCallee()->getSourceRange(); + PathDiagnosticLocation CELoc = + PathDiagnosticLocation::createBegin(CE, BR.getSourceManager(), AC); + BR.EmitBasicReport(AC->getDecl(), os1.str(), "Security", os2.str(), + CELoc, &R, 1); +} + +//===----------------------------------------------------------------------===// +// Check: 'random' should not be used +// Originally: <rdar://problem/63371000> +//===----------------------------------------------------------------------===// + +void WalkAST::checkCall_random(const CallExpr *CE, const FunctionDecl *FD) { + if (!CheckRand || !filter.check_rand) + return; + + const FunctionProtoType *FTP + = dyn_cast<FunctionProtoType>(FD->getType().IgnoreParens()); + if (!FTP) + return; + + // Verify that the function takes no argument. + if (FTP->getNumArgs() != 0) + return; + + // Issue a warning. + SourceRange R = CE->getCallee()->getSourceRange(); + PathDiagnosticLocation CELoc = + PathDiagnosticLocation::createBegin(CE, BR.getSourceManager(), AC); + BR.EmitBasicReport(AC->getDecl(), + "'random' is not a secure random number generator", + "Security", + "The 'random' function produces a sequence of values that " + "an adversary may be able to predict. Use 'arc4random' " + "instead", CELoc, &R, 1); +} + +//===----------------------------------------------------------------------===// +// Check: 'vfork' should not be used. +// POS33-C: Do not use vfork(). +//===----------------------------------------------------------------------===// + +void WalkAST::checkCall_vfork(const CallExpr *CE, const FunctionDecl *FD) { + if (!filter.check_vfork) + return; + + // All calls to vfork() are insecure, issue a warning. + SourceRange R = CE->getCallee()->getSourceRange(); + PathDiagnosticLocation CELoc = + PathDiagnosticLocation::createBegin(CE, BR.getSourceManager(), AC); + BR.EmitBasicReport(AC->getDecl(), + "Potential insecure implementation-specific behavior in " + "call 'vfork'", + "Security", + "Call to function 'vfork' is insecure as it can lead to " + "denial of service situations in the parent process. " + "Replace calls to vfork with calls to the safer " + "'posix_spawn' function", + CELoc, &R, 1); +} + +//===----------------------------------------------------------------------===// +// Check: Should check whether privileges are dropped successfully. +// Originally: <rdar://problem/6337132> +//===----------------------------------------------------------------------===// + +void WalkAST::checkUncheckedReturnValue(CallExpr *CE) { + if (!filter.check_UncheckedReturn) + return; + + const FunctionDecl *FD = CE->getDirectCallee(); + if (!FD) + return; + + if (II_setid[0] == NULL) { + static const char * const identifiers[num_setids] = { + "setuid", "setgid", "seteuid", "setegid", + "setreuid", "setregid" + }; + + for (size_t i = 0; i < num_setids; i++) + II_setid[i] = &BR.getContext().Idents.get(identifiers[i]); + } + + const IdentifierInfo *id = FD->getIdentifier(); + size_t identifierid; + + for (identifierid = 0; identifierid < num_setids; identifierid++) + if (id == II_setid[identifierid]) + break; + + if (identifierid >= num_setids) + return; + + const FunctionProtoType *FTP + = dyn_cast<FunctionProtoType>(FD->getType().IgnoreParens()); + if (!FTP) + return; + + // Verify that the function takes one or two arguments (depending on + // the function). + if (FTP->getNumArgs() != (identifierid < 4 ? 1 : 2)) + return; + + // The arguments must be integers. + for (unsigned i = 0; i < FTP->getNumArgs(); i++) + if (! FTP->getArgType(i)->isIntegerType()) + return; + + // Issue a warning. + SmallString<256> buf1; + llvm::raw_svector_ostream os1(buf1); + os1 << "Return value is not checked in call to '" << *FD << '\''; + + SmallString<256> buf2; + llvm::raw_svector_ostream os2(buf2); + os2 << "The return value from the call to '" << *FD + << "' is not checked. If an error occurs in '" << *FD + << "', the following code may execute with unexpected privileges"; + + SourceRange R = CE->getCallee()->getSourceRange(); + PathDiagnosticLocation CELoc = + PathDiagnosticLocation::createBegin(CE, BR.getSourceManager(), AC); + BR.EmitBasicReport(AC->getDecl(), os1.str(), "Security", os2.str(), + CELoc, &R, 1); +} + +//===----------------------------------------------------------------------===// +// SecuritySyntaxChecker +//===----------------------------------------------------------------------===// + +namespace { +class SecuritySyntaxChecker : public Checker<check::ASTCodeBody> { +public: + ChecksFilter filter; + + void checkASTCodeBody(const Decl *D, AnalysisManager& mgr, + BugReporter &BR) const { + WalkAST walker(BR, mgr.getAnalysisDeclContext(D), filter); + walker.Visit(D->getBody()); + } +}; +} + +#define REGISTER_CHECKER(name) \ +void ento::register##name(CheckerManager &mgr) {\ + mgr.registerChecker<SecuritySyntaxChecker>()->filter.check_##name = true;\ +} + +REGISTER_CHECKER(gets) +REGISTER_CHECKER(getpw) +REGISTER_CHECKER(mkstemp) +REGISTER_CHECKER(mktemp) +REGISTER_CHECKER(strcpy) +REGISTER_CHECKER(rand) +REGISTER_CHECKER(vfork) +REGISTER_CHECKER(FloatLoopCounter) +REGISTER_CHECKER(UncheckedReturn) + + diff --git a/clang/lib/StaticAnalyzer/Checkers/CheckSizeofPointer.cpp b/clang/lib/StaticAnalyzer/Checkers/CheckSizeofPointer.cpp new file mode 100644 index 0000000..cc7fd37 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/CheckSizeofPointer.cpp @@ -0,0 +1,92 @@ +//==- CheckSizeofPointer.cpp - Check for sizeof on pointers ------*- C++ -*-==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines a check for unintended use of sizeof() on pointer +// expressions. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/AST/StmtVisitor.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugReporter.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/AnalysisManager.h" + +using namespace clang; +using namespace ento; + +namespace { +class WalkAST : public StmtVisitor<WalkAST> { + BugReporter &BR; + AnalysisDeclContext* AC; + +public: + WalkAST(BugReporter &br, AnalysisDeclContext* ac) : BR(br), AC(ac) {} + void VisitUnaryExprOrTypeTraitExpr(UnaryExprOrTypeTraitExpr *E); + void VisitStmt(Stmt *S) { VisitChildren(S); } + void VisitChildren(Stmt *S); +}; +} + +void WalkAST::VisitChildren(Stmt *S) { + for (Stmt::child_iterator I = S->child_begin(), E = S->child_end(); I!=E; ++I) + if (Stmt *child = *I) + Visit(child); +} + +// CWE-467: Use of sizeof() on a Pointer Type +void WalkAST::VisitUnaryExprOrTypeTraitExpr(UnaryExprOrTypeTraitExpr *E) { + if (E->getKind() != UETT_SizeOf) + return; + + // If an explicit type is used in the code, usually the coder knows what he is + // doing. + if (E->isArgumentType()) + return; + + QualType T = E->getTypeOfArgument(); + if (T->isPointerType()) { + + // Many false positives have the form 'sizeof *p'. This is reasonable + // because people know what they are doing when they intentionally + // dereference the pointer. + Expr *ArgEx = E->getArgumentExpr(); + if (!isa<DeclRefExpr>(ArgEx->IgnoreParens())) + return; + + SourceRange R = ArgEx->getSourceRange(); + PathDiagnosticLocation ELoc = + PathDiagnosticLocation::createBegin(E, BR.getSourceManager(), AC); + BR.EmitBasicReport(AC->getDecl(), + "Potential unintended use of sizeof() on pointer type", + "Logic", + "The code calls sizeof() on a pointer type. " + "This can produce an unexpected result.", + ELoc, &R, 1); + } +} + +//===----------------------------------------------------------------------===// +// SizeofPointerChecker +//===----------------------------------------------------------------------===// + +namespace { +class SizeofPointerChecker : public Checker<check::ASTCodeBody> { +public: + void checkASTCodeBody(const Decl *D, AnalysisManager& mgr, + BugReporter &BR) const { + WalkAST walker(BR, mgr.getAnalysisDeclContext(D)); + walker.Visit(D->getBody()); + } +}; +} + +void ento::registerSizeofPointerChecker(CheckerManager &mgr) { + mgr.registerChecker<SizeofPointerChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/CheckerDocumentation.cpp b/clang/lib/StaticAnalyzer/Checkers/CheckerDocumentation.cpp new file mode 100644 index 0000000..843502f --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/CheckerDocumentation.cpp @@ -0,0 +1,233 @@ +//= CheckerDocumentation.cpp - Documentation checker ---------------*- C++ -*-// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This checker lists all the checker callbacks and provides documentation for +// checker writers. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramStateTrait.h" + +using namespace clang; +using namespace ento; + +// All checkers should be placed into anonymous namespace. +// We place the CheckerDocumentation inside ento namespace to make the +// it visible in doxygen. +namespace ento { + +/// This checker documents the callback functions checkers can use to implement +/// the custom handling of the specific events during path exploration as well +/// as reporting bugs. Most of the callbacks are targeted at path-sensitive +/// checking. +/// +/// \sa CheckerContext +class CheckerDocumentation : public Checker< check::PreStmt<DeclStmt>, + check::PostStmt<CallExpr>, + check::PreObjCMessage, + check::PostObjCMessage, + check::BranchCondition, + check::Location, + check::Bind, + check::DeadSymbols, + check::EndPath, + check::EndAnalysis, + check::EndOfTranslationUnit, + eval::Call, + eval::Assume, + check::LiveSymbols, + check::RegionChanges, + check::Event<ImplicitNullDerefEvent>, + check::ASTDecl<FunctionDecl> > { +public: + + /// \brief Pre-visit the Statement. + /// + /// The method will be called before the analyzer core processes the + /// statement. The notification is performed for every explored CFGElement, + /// which does not include the control flow statements such as IfStmt. The + /// callback can be specialized to be called with any subclass of Stmt. + /// + /// See checkBranchCondition() callback for performing custom processing of + /// the branching statements. + /// + /// check::PreStmt<DeclStmt> + void checkPreStmt(const DeclStmt *DS, CheckerContext &C) const {} + + /// \brief Post-visit the Statement. + /// + /// The method will be called after the analyzer core processes the + /// statement. The notification is performed for every explored CFGElement, + /// which does not include the control flow statements such as IfStmt. The + /// callback can be specialized to be called with any subclass of Stmt. + /// + /// check::PostStmt<DeclStmt> + void checkPostStmt(const CallExpr *DS, CheckerContext &C) const; + + /// \brief Pre-visit the Objective C messages. + void checkPreObjCMessage(const ObjCMessage &Msg, CheckerContext &C) const {} + + /// \brief Post-visit the Objective C messages. + void checkPostObjCMessage(const ObjCMessage &Msg, CheckerContext &C) const {} + + /// \brief Pre-visit of the condition statement of a branch (such as IfStmt). + void checkBranchCondition(const Stmt *Condition, CheckerContext &Ctx) const {} + + /// \brief Called on a load from and a store to a location. + /// + /// The method will be called each time a location (pointer) value is + /// accessed. + /// \param Loc The value of the location (pointer). + /// \param IsLoad The flag specifying if the location is a store or a load. + /// \param S The load is performed while processing the statement. + /// + /// check::Location + void checkLocation(SVal Loc, bool IsLoad, const Stmt *S, + CheckerContext &C) const {} + + /// \brief Called on binding of a value to a location. + /// + /// \param Loc The value of the location (pointer). + /// \param Val The value which will be stored at the location Loc. + /// \param S The bind is performed while processing the statement S. + /// + /// check::Bind + void checkBind(SVal Loc, SVal Val, const Stmt *S, CheckerContext &C) const {} + + + /// \brief Called whenever a symbol becomes dead. + /// + /// This callback should be used by the checkers to aggressively clean + /// up/reduce the checker state, which is important for reducing the overall + /// memory usage. Specifically, if a checker keeps symbol specific information + /// in the sate, it can and should be dropped after the symbol becomes dead. + /// In addition, reporting a bug as soon as the checker becomes dead leads to + /// more precise diagnostics. (For example, one should report that a malloced + /// variable is not freed right after it goes out of scope.) + /// + /// \param SR The SymbolReaper object can be queried to determine which + /// symbols are dead. + /// + /// check::DeadSymbols + void checkDeadSymbols(SymbolReaper &SR, CheckerContext &C) const {} + + /// \brief Called when an end of path is reached in the ExplodedGraph. + /// + /// This callback should be used to check if the allocated resources are freed. + /// + /// check::EndPath + void checkEndPath(CheckerContext &Ctx) const {} + + /// \brief Called after all the paths in the ExplodedGraph reach end of path + /// - the symbolic execution graph is fully explored. + /// + /// This callback should be used in cases when a checker needs to have a + /// global view of the information generated on all paths. For example, to + /// compare execution summary/result several paths. + /// See IdempotentOperationChecker for a usage example. + /// + /// check::EndAnalysis + void checkEndAnalysis(ExplodedGraph &G, + BugReporter &BR, + ExprEngine &Eng) const {} + + /// \brief Called after analysis of a TranslationUnit is complete. + /// + /// check::EndOfTranslationUnit + void checkEndOfTranslationUnit(const TranslationUnitDecl *TU, + AnalysisManager &Mgr, + BugReporter &BR) const {} + + + /// \brief Evaluates function call. + /// + /// The analysis core threats all function calls in the same way. However, some + /// functions have special meaning, which should be reflected in the program + /// state. This callback allows a checker to provide domain specific knowledge + /// about the particular functions it knows about. + /// + /// \returns true if the call has been successfully evaluated + /// and false otherwise. Note, that only one checker can evaluate a call. If + /// more then one checker claim that they can evaluate the same call the + /// first one wins. + /// + /// eval::Call + bool evalCall(const CallExpr *CE, CheckerContext &C) const { return true; } + + /// \brief Handles assumptions on symbolic values. + /// + /// This method is called when a symbolic expression is assumed to be true or + /// false. For example, the assumptions are performed when evaluating a + /// condition at a branch. The callback allows checkers track the assumptions + /// performed on the symbols of interest and change the state accordingly. + /// + /// eval::Assume + ProgramStateRef evalAssume(ProgramStateRef State, + SVal Cond, + bool Assumption) const { return State; } + + /// Allows modifying SymbolReaper object. For example, checkers can explicitly + /// register symbols of interest as live. These symbols will not be marked + /// dead and removed. + /// + /// check::LiveSymbols + void checkLiveSymbols(ProgramStateRef State, SymbolReaper &SR) const {} + + + bool wantsRegionChangeUpdate(ProgramStateRef St) const { return true; } + + /// check::RegionChanges + /// Allows tracking regions which get invalidated. + /// \param state The current program state. + /// \param invalidated A set of all symbols potentially touched by the change. + /// \param ExplicitRegions The regions explicitly requested for invalidation. + /// For example, in the case of a function call, these would be arguments. + /// \param Regions The transitive closure of accessible regions, + /// i.e. all regions that may have been touched by this change. + /// \param The call expression wrapper if the regions are invalidated by a + /// call, 0 otherwise. + /// Note, in order to be notified, the checker should also implement + /// wantsRegionChangeUpdate callback. + ProgramStateRef + checkRegionChanges(ProgramStateRef State, + const StoreManager::InvalidatedSymbols *, + ArrayRef<const MemRegion *> ExplicitRegions, + ArrayRef<const MemRegion *> Regions, + const CallOrObjCMessage *Call) const { + return State; + } + + /// check::Event<ImplicitNullDerefEvent> + void checkEvent(ImplicitNullDerefEvent Event) const {} + + /// \brief Check every declaration in the AST. + /// + /// An AST traversal callback, which should only be used when the checker is + /// not path sensitive. It will be called for every Declaration in the AST and + /// can be specialized to only be called on subclasses of Decl, for example, + /// FunctionDecl. + /// + /// check::ASTDecl<FunctionDecl> + void checkASTDecl(const FunctionDecl *D, + AnalysisManager &Mgr, + BugReporter &BR) const {} + +}; + +void CheckerDocumentation::checkPostStmt(const CallExpr *DS, + CheckerContext &C) const { + return; +} + +} // end namespace diff --git a/clang/lib/StaticAnalyzer/Checkers/Checkers.td b/clang/lib/StaticAnalyzer/Checkers/Checkers.td new file mode 100644 index 0000000..f577a89 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/Checkers.td @@ -0,0 +1,491 @@ +//===--- Checkers.td - Static Analyzer Checkers -===-----------------------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// + +include "clang/StaticAnalyzer/Checkers/CheckerBase.td" + +//===----------------------------------------------------------------------===// +// Packages. +//===----------------------------------------------------------------------===// + +def Experimental : Package<"experimental">; + +def Core : Package<"core">; +def CoreBuiltin : Package<"builtin">, InPackage<Core>; +def CoreUninitialized : Package<"uninitialized">, InPackage<Core>; +def CoreExperimental : Package<"core">, InPackage<Experimental>, Hidden; + +def Cplusplus : Package<"cplusplus">; +def CplusplusExperimental : Package<"cplusplus">, InPackage<Experimental>, Hidden; + +def DeadCode : Package<"deadcode">; +def DeadCodeExperimental : Package<"deadcode">, InPackage<Experimental>, Hidden; + +def Security : Package <"security">; +def InsecureAPI : Package<"insecureAPI">, InPackage<Security>; +def SecurityExperimental : Package<"security">, InPackage<Experimental>, Hidden; +def Taint : Package<"taint">, InPackage<SecurityExperimental>, Hidden; + +def Unix : Package<"unix">; +def UnixExperimental : Package<"unix">, InPackage<Experimental>, Hidden; +def CString : Package<"cstring">, InPackage<Unix>, Hidden; +def CStringExperimental : Package<"cstring">, InPackage<UnixExperimental>, Hidden; + +def OSX : Package<"osx">; +def OSXExperimental : Package<"osx">, InPackage<Experimental>, Hidden; +def Cocoa : Package<"cocoa">, InPackage<OSX>; +def CocoaExperimental : Package<"cocoa">, InPackage<OSXExperimental>, Hidden; +def CoreFoundation : Package<"coreFoundation">, InPackage<OSX>; +def Containers : Package<"containers">, InPackage<CoreFoundation>; + +def LLVM : Package<"llvm">; +def Debug : Package<"debug">; + +//===----------------------------------------------------------------------===// +// Core Checkers. +//===----------------------------------------------------------------------===// + +let ParentPackage = Core in { + +def DereferenceChecker : Checker<"NullDereference">, + HelpText<"Check for dereferences of null pointers">, + DescFile<"DereferenceChecker.cpp">; + +def CallAndMessageChecker : Checker<"CallAndMessage">, + HelpText<"Check for logical errors for function calls and Objective-C message expressions (e.g., uninitialized arguments, null function pointers)">, + DescFile<"CallAndMessageChecker.cpp">; + +def AdjustedReturnValueChecker : Checker<"AdjustedReturnValue">, + HelpText<"Check to see if the return value of a function call is different than the caller expects (e.g., from calls through function pointers)">, + DescFile<"AdjustedReturnValueChecker.cpp">; + +def AttrNonNullChecker : Checker<"AttributeNonNull">, + HelpText<"Check for null pointers passed as arguments to a function whose arguments are marked with the 'nonnull' attribute">, + DescFile<"AttrNonNullChecker.cpp">; + +def VLASizeChecker : Checker<"VLASize">, + HelpText<"Check for declarations of VLA of undefined or zero size">, + DescFile<"VLASizeChecker.cpp">; + +def DivZeroChecker : Checker<"DivideZero">, + HelpText<"Check for division by zero">, + DescFile<"DivZeroChecker.cpp">; + +def UndefResultChecker : Checker<"UndefinedBinaryOperatorResult">, + HelpText<"Check for undefined results of binary operators">, + DescFile<"UndefResultChecker.cpp">; + +def StackAddrEscapeChecker : Checker<"StackAddressEscape">, + HelpText<"Check that addresses to stack memory do not escape the function">, + DescFile<"StackAddrEscapeChecker.cpp">; + +} // end "core" + +let ParentPackage = CoreExperimental in { + +def BoolAssignmentChecker : Checker<"BoolAssignment">, + HelpText<"Warn about assigning non-{0,1} values to Boolean variables">, + DescFile<"BoolAssignmentChecker.cpp">; + +def CastSizeChecker : Checker<"CastSize">, + HelpText<"Check when casting a malloc'ed type T, whether the size is a multiple of the size of T">, + DescFile<"CastSizeChecker.cpp">; + +def CastToStructChecker : Checker<"CastToStruct">, + HelpText<"Check for cast from non-struct pointer to struct pointer">, + DescFile<"CastToStructChecker.cpp">; + +def FixedAddressChecker : Checker<"FixedAddr">, + HelpText<"Check for assignment of a fixed address to a pointer">, + DescFile<"FixedAddressChecker.cpp">; + +def PointerArithChecker : Checker<"PointerArithm">, + HelpText<"Check for pointer arithmetic on locations other than array elements">, + DescFile<"PointerArithChecker">; + +def PointerSubChecker : Checker<"PointerSub">, + HelpText<"Check for pointer subtractions on two pointers pointing to different memory chunks">, + DescFile<"PointerSubChecker">; + +def SizeofPointerChecker : Checker<"SizeofPtr">, + HelpText<"Warn about unintended use of sizeof() on pointer expressions">, + DescFile<"CheckSizeofPointer.cpp">; + +} // end "core.experimental" + +//===----------------------------------------------------------------------===// +// Evaluate "builtin" functions. +//===----------------------------------------------------------------------===// + +let ParentPackage = CoreBuiltin in { + +def NoReturnFunctionChecker : Checker<"NoReturnFunctions">, + HelpText<"Evaluate \"panic\" functions that are known to not return to the caller">, + DescFile<"NoReturnFunctionChecker.cpp">; + +def BuiltinFunctionChecker : Checker<"BuiltinFunctions">, + HelpText<"Evaluate compiler builtin functions (e.g., alloca())">, + DescFile<"BuiltinFunctionChecker.cpp">; + +} // end "core.builtin" + +//===----------------------------------------------------------------------===// +// Uninitialized values checkers. +//===----------------------------------------------------------------------===// + +let ParentPackage = CoreUninitialized in { + +def UndefinedArraySubscriptChecker : Checker<"ArraySubscript">, + HelpText<"Check for uninitialized values used as array subscripts">, + DescFile<"UndefinedArraySubscriptChecker.cpp">; + +def UndefinedAssignmentChecker : Checker<"Assign">, + HelpText<"Check for assigning uninitialized values">, + DescFile<"UndefinedAssignmentChecker.cpp">; + +def UndefBranchChecker : Checker<"Branch">, + HelpText<"Check for uninitialized values used as branch conditions">, + DescFile<"UndefBranchChecker.cpp">; + +def UndefCapturedBlockVarChecker : Checker<"CapturedBlockVariable">, + HelpText<"Check for blocks that capture uninitialized values">, + DescFile<"UndefCapturedBlockVarChecker.cpp">; + +def ReturnUndefChecker : Checker<"UndefReturn">, + HelpText<"Check for uninitialized values being returned to the caller">, + DescFile<"ReturnUndefChecker.cpp">; + +} // end "core.uninitialized" + +//===----------------------------------------------------------------------===// +// C++ checkers. +//===----------------------------------------------------------------------===// + +let ParentPackage = CplusplusExperimental in { + +def IteratorsChecker : Checker<"Iterators">, + HelpText<"Check improper uses of STL vector iterators">, + DescFile<"IteratorsChecker.cpp">; + +def VirtualCallChecker : Checker<"VirtualCall">, + HelpText<"Check virtual function calls during construction or destruction">, + DescFile<"VirtualCallChecker.cpp">; + +} // end: "cplusplus.experimental" + +//===----------------------------------------------------------------------===// +// Deadcode checkers. +//===----------------------------------------------------------------------===// + +let ParentPackage = DeadCode in { + +def DeadStoresChecker : Checker<"DeadStores">, + HelpText<"Check for values stored to variables that are never read afterwards">, + DescFile<"DeadStoresChecker.cpp">; +} // end DeadCode + +let ParentPackage = DeadCodeExperimental in { + +def IdempotentOperationChecker : Checker<"IdempotentOperations">, + HelpText<"Warn about idempotent operations">, + DescFile<"IdempotentOperationChecker.cpp">; + +def UnreachableCodeChecker : Checker<"UnreachableCode">, + HelpText<"Check unreachable code">, + DescFile<"UnreachableCodeChecker.cpp">; + +} // end "deadcode.experimental" + +//===----------------------------------------------------------------------===// +// Security checkers. +//===----------------------------------------------------------------------===// + +let ParentPackage = InsecureAPI in { + def gets : Checker<"gets">, + HelpText<"Warn on uses of the 'gets' function">, + DescFile<"CheckSecuritySyntaxOnly.cpp">; + def getpw : Checker<"getpw">, + HelpText<"Warn on uses of the 'getpw' function">, + DescFile<"CheckSecuritySyntaxOnly.cpp">; + def mktemp : Checker<"mktemp">, + HelpText<"Warn on uses of the 'mktemp' function">, + DescFile<"CheckSecuritySyntaxOnly.cpp">; + def mkstemp : Checker<"mkstemp">, + HelpText<"Warn when 'mkstemp' is passed fewer than 6 X's in the format string">, + DescFile<"CheckSecuritySyntaxOnly.cpp">; + def rand : Checker<"rand">, + HelpText<"Warn on uses of the 'rand', 'random', and related functions">, + DescFile<"CheckSecuritySyntaxOnly.cpp">; + def strcpy : Checker<"strcpy">, + HelpText<"Warn on uses of the 'strcpy' and 'strcat' functions">, + DescFile<"CheckSecuritySyntaxOnly.cpp">; + def vfork : Checker<"vfork">, + HelpText<"Warn on uses of the 'vfork' function">, + DescFile<"CheckSecuritySyntaxOnly.cpp">; + def UncheckedReturn : Checker<"UncheckedReturn">, + HelpText<"Warn on uses of functions whose return values must be always checked">, + DescFile<"CheckSecuritySyntaxOnly.cpp">; +} +let ParentPackage = Security in { + def FloatLoopCounter : Checker<"FloatLoopCounter">, + HelpText<"Warn on using a floating point value as a loop counter (CERT: FLP30-C, FLP30-CPP)">, + DescFile<"CheckSecuritySyntaxOnly.cpp">; +} + +let ParentPackage = SecurityExperimental in { + +def ArrayBoundChecker : Checker<"ArrayBound">, + HelpText<"Warn about buffer overflows (older checker)">, + DescFile<"ArrayBoundChecker.cpp">; + +def ArrayBoundCheckerV2 : Checker<"ArrayBoundV2">, + HelpText<"Warn about buffer overflows (newer checker)">, + DescFile<"ArrayBoundCheckerV2.cpp">; + +def ReturnPointerRangeChecker : Checker<"ReturnPtrRange">, + HelpText<"Check for an out-of-bound pointer being returned to callers">, + DescFile<"ReturnPointerRangeChecker.cpp">; + +def MallocOverflowSecurityChecker : Checker<"MallocOverflow">, + HelpText<"Check for overflows in the arguments to malloc()">, + DescFile<"MallocOverflowSecurityChecker.cpp">; + +} // end "security.experimental" + +//===----------------------------------------------------------------------===// +// Taint checkers. +//===----------------------------------------------------------------------===// + +let ParentPackage = Taint in { + +def GenericTaintChecker : Checker<"TaintPropagation">, + HelpText<"Generate taint information used by other checkers">, + DescFile<"GenericTaintChecker.cpp">; + +} // end "experimental.security.taint" + +//===----------------------------------------------------------------------===// +// Unix API checkers. +//===----------------------------------------------------------------------===// + +let ParentPackage = Unix in { + +def UnixAPIChecker : Checker<"API">, + HelpText<"Check calls to various UNIX/Posix functions">, + DescFile<"UnixAPIChecker.cpp">; + +def MallocPessimistic : Checker<"Malloc">, + HelpText<"Check for memory leaks, double free, and use-after-free problems.">, + DescFile<"MallocChecker.cpp">; + +} // end "unix" + +let ParentPackage = UnixExperimental in { + +def ChrootChecker : Checker<"Chroot">, + HelpText<"Check improper use of chroot">, + DescFile<"ChrootChecker.cpp">; + +def MallocOptimistic : Checker<"MallocWithAnnotations">, + HelpText<"Check for memory leaks, double free, and use-after-free problems. Assumes that all user-defined functions which might free a pointer are annotated.">, + DescFile<"MallocChecker.cpp">; + +def MallocSizeofChecker : Checker<"MallocSizeof">, + HelpText<"Check for dubious malloc arguments involving sizeof">, + DescFile<"MallocSizeofChecker.cpp">; + +def PthreadLockChecker : Checker<"PthreadLock">, + HelpText<"Simple lock -> unlock checker">, + DescFile<"PthreadLockChecker.cpp">; + +def StreamChecker : Checker<"Stream">, + HelpText<"Check stream handling functions">, + DescFile<"StreamChecker.cpp">; + +} // end "unix.experimental" + +let ParentPackage = CString in { + +def CStringNullArg : Checker<"NullArg">, + HelpText<"Check for null pointers being passed as arguments to C string functions">, + DescFile<"CStringChecker.cpp">; + +def CStringSyntaxChecker : Checker<"BadSizeArg">, + HelpText<"Check the size argument passed into C string functions for common erroneous patterns">, + DescFile<"CStringSyntaxChecker.cpp">; +} + +let ParentPackage = CStringExperimental in { + +def CStringOutOfBounds : Checker<"OutOfBounds">, + HelpText<"Check for out-of-bounds access in string functions">, + DescFile<"CStringChecker.cpp">; + +def CStringBufferOverlap : Checker<"BufferOverlap">, + HelpText<"Checks for overlap in two buffer arguments">, + DescFile<"CStringChecker.cpp">; + +def CStringNotNullTerm : Checker<"NotNullTerminated">, + HelpText<"Check for arguments which are not null-terminating strings">, + DescFile<"CStringChecker.cpp">; +} + +//===----------------------------------------------------------------------===// +// Mac OS X, Cocoa, and Core Foundation checkers. +//===----------------------------------------------------------------------===// + +let ParentPackage = OSX in { + +def MacOSXAPIChecker : Checker<"API">, + InPackage<OSX>, + HelpText<"Check for proper uses of various Mac OS X APIs">, + DescFile<"MacOSXAPIChecker.cpp">; + +def OSAtomicChecker : Checker<"AtomicCAS">, + InPackage<OSX>, + HelpText<"Evaluate calls to OSAtomic functions">, + DescFile<"OSAtomicChecker.cpp">; + +def MacOSKeychainAPIChecker : Checker<"SecKeychainAPI">, + InPackage<OSX>, + HelpText<"Check for proper uses of Secure Keychain APIs">, + DescFile<"MacOSKeychainAPIChecker.cpp">; + +} // end "macosx" + +let ParentPackage = Cocoa in { + +def ObjCAtSyncChecker : Checker<"AtSync">, + HelpText<"Check for null pointers used as mutexes for @synchronized">, + DescFile<"ObjCAtSyncChecker.cpp">; + +def NilArgChecker : Checker<"NilArg">, + HelpText<"Check for prohibited nil arguments to ObjC method calls">, + DescFile<"BasicObjCFoundationChecks.cpp">; + +def ClassReleaseChecker : Checker<"ClassRelease">, + HelpText<"Check for sending 'retain', 'release', or 'autorelease' directly to a Class">, + DescFile<"BasicObjCFoundationChecks.cpp">; + +def VariadicMethodTypeChecker : Checker<"VariadicMethodTypes">, + HelpText<"Check for passing non-Objective-C types to variadic methods that expect " + "only Objective-C types">, + DescFile<"BasicObjCFoundationChecks.cpp">; + +def NSAutoreleasePoolChecker : Checker<"NSAutoreleasePool">, + HelpText<"Warn for suboptimal uses of NSAutoreleasePool in Objective-C GC mode">, + DescFile<"NSAutoreleasePoolChecker.cpp">; + +def ObjCMethSigsChecker : Checker<"IncompatibleMethodTypes">, + HelpText<"Warn about Objective-C method signatures with type incompatibilities">, + DescFile<"CheckObjCInstMethSignature.cpp">; + +def ObjCUnusedIvarsChecker : Checker<"UnusedIvars">, + HelpText<"Warn about private ivars that are never used">, + DescFile<"ObjCUnusedIVarsChecker.cpp">; + +def ObjCSelfInitChecker : Checker<"SelfInit">, + HelpText<"Check that 'self' is properly initialized inside an initializer method">, + DescFile<"ObjCSelfInitChecker.cpp">; + +def NSErrorChecker : Checker<"NSError">, + HelpText<"Check usage of NSError** parameters">, + DescFile<"NSErrorChecker.cpp">; + +def RetainCountChecker : Checker<"RetainCount">, + HelpText<"Check for leaks and improper reference count management">, + DescFile<"RetainCountChecker.cpp">; + +} // end "cocoa" + +let ParentPackage = CocoaExperimental in { + +def ObjCDeallocChecker : Checker<"Dealloc">, + HelpText<"Warn about Objective-C classes that lack a correct implementation of -dealloc">, + DescFile<"CheckObjCDealloc.cpp">; + +} // end "cocoa.experimental" + +let ParentPackage = CoreFoundation in { + +def CFNumberCreateChecker : Checker<"CFNumber">, + HelpText<"Check for proper uses of CFNumberCreate">, + DescFile<"BasicObjCFoundationChecks.cpp">; + +def CFRetainReleaseChecker : Checker<"CFRetainRelease">, + HelpText<"Check for null arguments to CFRetain/CFRelease">, + DescFile<"BasicObjCFoundationChecks.cpp">; + +def CFErrorChecker : Checker<"CFError">, + HelpText<"Check usage of CFErrorRef* parameters">, + DescFile<"NSErrorChecker.cpp">; +} + +let ParentPackage = Containers in { +def ObjCContainersASTChecker : Checker<"PointerSizedValues">, + HelpText<"Warns if 'CFArray', 'CFDictionary', 'CFSet' are created with non-pointer-size values">, + DescFile<"ObjCContainersASTChecker.cpp">; + +def ObjCContainersChecker : Checker<"OutOfBounds">, + HelpText<"Checks for index out-of-bounds when using 'CFArray' API">, + DescFile<"ObjCContainersChecker.cpp">; + +} +//===----------------------------------------------------------------------===// +// Checkers for LLVM development. +//===----------------------------------------------------------------------===// + +def LLVMConventionsChecker : Checker<"Conventions">, + InPackage<LLVM>, + HelpText<"Check code for LLVM codebase conventions">, + DescFile<"LLVMConventionsChecker.cpp">; + +//===----------------------------------------------------------------------===// +// Debugging checkers (for analyzer development). +//===----------------------------------------------------------------------===// + +let ParentPackage = Debug in { + +def DominatorsTreeDumper : Checker<"DumpDominators">, + HelpText<"Print the dominance tree for a given CFG">, + DescFile<"DebugCheckers.cpp">; + +def LiveVariablesDumper : Checker<"DumpLiveVars">, + HelpText<"Print results of live variable analysis">, + DescFile<"DebugCheckers.cpp">; + +def CFGViewer : Checker<"ViewCFG">, + HelpText<"View Control-Flow Graphs using GraphViz">, + DescFile<"DebugCheckers.cpp">; + +def CFGDumper : Checker<"DumpCFG">, + HelpText<"Display Control-Flow Graphs">, + DescFile<"DebugCheckers.cpp">; + +def CallGraphViewer : Checker<"ViewCallGraph">, + HelpText<"View Call Graph using GraphViz">, + DescFile<"DebugCheckers.cpp">; + +def CallGraphDumper : Checker<"DumpCallGraph">, + HelpText<"Display Call Graph">, + DescFile<"DebugCheckers.cpp">; + +def AnalyzerStatsChecker : Checker<"Stats">, + HelpText<"Emit warnings with analyzer statistics">, + DescFile<"AnalyzerStatsChecker.cpp">; + +def TaintTesterChecker : Checker<"TaintTest">, + HelpText<"Mark tainted symbols as such.">, + DescFile<"TaintTesterChecker.cpp">; + +def IntervalTestChecker : Checker<"IntervalTest">, + HelpText<"Test the IntervalAnalysis stuff">, + DescFile<"IntervalTest.cpp">; + +} // end "debug" + diff --git a/clang/lib/StaticAnalyzer/Checkers/ChrootChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/ChrootChecker.cpp new file mode 100644 index 0000000..30d0609 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/ChrootChecker.cpp @@ -0,0 +1,158 @@ +//===- Chrootchecker.cpp -------- Basic security checks ----------*- C++ -*-==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines chroot checker, which checks improper use of chroot. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramState.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramStateTrait.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/SymbolManager.h" +#include "llvm/ADT/ImmutableMap.h" +using namespace clang; +using namespace ento; + +namespace { + +// enum value that represent the jail state +enum Kind { NO_CHROOT, ROOT_CHANGED, JAIL_ENTERED }; + +bool isRootChanged(intptr_t k) { return k == ROOT_CHANGED; } +//bool isJailEntered(intptr_t k) { return k == JAIL_ENTERED; } + +// This checker checks improper use of chroot. +// The state transition: +// NO_CHROOT ---chroot(path)--> ROOT_CHANGED ---chdir(/) --> JAIL_ENTERED +// | | +// ROOT_CHANGED<--chdir(..)-- JAIL_ENTERED<--chdir(..)-- +// | | +// bug<--foo()-- JAIL_ENTERED<--foo()-- +class ChrootChecker : public Checker<eval::Call, check::PreStmt<CallExpr> > { + mutable IdentifierInfo *II_chroot, *II_chdir; + // This bug refers to possibly break out of a chroot() jail. + mutable OwningPtr<BuiltinBug> BT_BreakJail; + +public: + ChrootChecker() : II_chroot(0), II_chdir(0) {} + + static void *getTag() { + static int x; + return &x; + } + + bool evalCall(const CallExpr *CE, CheckerContext &C) const; + void checkPreStmt(const CallExpr *CE, CheckerContext &C) const; + +private: + void Chroot(CheckerContext &C, const CallExpr *CE) const; + void Chdir(CheckerContext &C, const CallExpr *CE) const; +}; + +} // end anonymous namespace + +bool ChrootChecker::evalCall(const CallExpr *CE, CheckerContext &C) const { + const FunctionDecl *FD = C.getCalleeDecl(CE); + if (!FD) + return false; + + ASTContext &Ctx = C.getASTContext(); + if (!II_chroot) + II_chroot = &Ctx.Idents.get("chroot"); + if (!II_chdir) + II_chdir = &Ctx.Idents.get("chdir"); + + if (FD->getIdentifier() == II_chroot) { + Chroot(C, CE); + return true; + } + if (FD->getIdentifier() == II_chdir) { + Chdir(C, CE); + return true; + } + + return false; +} + +void ChrootChecker::Chroot(CheckerContext &C, const CallExpr *CE) const { + ProgramStateRef state = C.getState(); + ProgramStateManager &Mgr = state->getStateManager(); + + // Once encouter a chroot(), set the enum value ROOT_CHANGED directly in + // the GDM. + state = Mgr.addGDM(state, ChrootChecker::getTag(), (void*) ROOT_CHANGED); + C.addTransition(state); +} + +void ChrootChecker::Chdir(CheckerContext &C, const CallExpr *CE) const { + ProgramStateRef state = C.getState(); + ProgramStateManager &Mgr = state->getStateManager(); + + // If there are no jail state in the GDM, just return. + const void *k = state->FindGDM(ChrootChecker::getTag()); + if (!k) + return; + + // After chdir("/"), enter the jail, set the enum value JAIL_ENTERED. + const Expr *ArgExpr = CE->getArg(0); + SVal ArgVal = state->getSVal(ArgExpr, C.getLocationContext()); + + if (const MemRegion *R = ArgVal.getAsRegion()) { + R = R->StripCasts(); + if (const StringRegion* StrRegion= dyn_cast<StringRegion>(R)) { + const StringLiteral* Str = StrRegion->getStringLiteral(); + if (Str->getString() == "/") + state = Mgr.addGDM(state, ChrootChecker::getTag(), + (void*) JAIL_ENTERED); + } + } + + C.addTransition(state); +} + +// Check the jail state before any function call except chroot and chdir(). +void ChrootChecker::checkPreStmt(const CallExpr *CE, CheckerContext &C) const { + const FunctionDecl *FD = C.getCalleeDecl(CE); + if (!FD) + return; + + ASTContext &Ctx = C.getASTContext(); + if (!II_chroot) + II_chroot = &Ctx.Idents.get("chroot"); + if (!II_chdir) + II_chdir = &Ctx.Idents.get("chdir"); + + // Ingnore chroot and chdir. + if (FD->getIdentifier() == II_chroot || FD->getIdentifier() == II_chdir) + return; + + // If jail state is ROOT_CHANGED, generate BugReport. + void *const* k = C.getState()->FindGDM(ChrootChecker::getTag()); + if (k) + if (isRootChanged((intptr_t) *k)) + if (ExplodedNode *N = C.addTransition()) { + if (!BT_BreakJail) + BT_BreakJail.reset(new BuiltinBug("Break out of jail", + "No call of chdir(\"/\") immediately " + "after chroot")); + BugReport *R = new BugReport(*BT_BreakJail, + BT_BreakJail->getDescription(), N); + C.EmitReport(R); + } + + return; +} + +void ento::registerChrootChecker(CheckerManager &mgr) { + mgr.registerChecker<ChrootChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/ClangCheckers.cpp b/clang/lib/StaticAnalyzer/Checkers/ClangCheckers.cpp new file mode 100644 index 0000000..77a5a72 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/ClangCheckers.cpp @@ -0,0 +1,32 @@ +//===--- ClangCheckers.h - Provides builtin checkers ------------*- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// + +#include "clang/StaticAnalyzer/Checkers/ClangCheckers.h" +#include "clang/StaticAnalyzer/Core/CheckerRegistry.h" + +// FIXME: This is only necessary as long as there are checker registration +// functions that do additional work besides mgr.registerChecker<CLASS>(). +// The only checkers that currently do this are: +// - NSAutoreleasePoolChecker +// - NSErrorChecker +// - ObjCAtSyncChecker +// It's probably worth including this information in Checkers.td to minimize +// boilerplate code. +#include "ClangSACheckers.h" + +using namespace clang; +using namespace ento; + +void ento::registerBuiltinCheckers(CheckerRegistry ®istry) { +#define GET_CHECKERS +#define CHECKER(FULLNAME,CLASS,DESCFILE,HELPTEXT,GROUPINDEX,HIDDEN) \ + registry.addChecker(register##CLASS, FULLNAME, HELPTEXT); +#include "Checkers.inc" +#undef GET_CHECKERS +} diff --git a/clang/lib/StaticAnalyzer/Checkers/ClangSACheckers.h b/clang/lib/StaticAnalyzer/Checkers/ClangSACheckers.h new file mode 100644 index 0000000..230baa7 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/ClangSACheckers.h @@ -0,0 +1,37 @@ +//===--- ClangSACheckers.h - Registration functions for Checkers *- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// Declares the registation functions for the checkers defined in +// libclangStaticAnalyzerCheckers. +// +//===----------------------------------------------------------------------===// + +#include "clang/StaticAnalyzer/Checkers/CommonBugCategories.h" + +#ifndef LLVM_CLANG_SA_LIB_CHECKERS_CLANGSACHECKERS_H +#define LLVM_CLANG_SA_LIB_CHECKERS_CLANGSACHECKERS_H + +namespace clang { + +namespace ento { +class CheckerManager; +class CheckerRegistry; + +#define GET_CHECKERS +#define CHECKER(FULLNAME,CLASS,CXXFILE,HELPTEXT,GROUPINDEX,HIDDEN) \ + void register##CLASS(CheckerManager &mgr); +#include "Checkers.inc" +#undef CHECKER +#undef GET_CHECKERS + +} // end ento namespace + +} // end clang namespace + +#endif diff --git a/clang/lib/StaticAnalyzer/Checkers/CommonBugCategories.cpp b/clang/lib/StaticAnalyzer/Checkers/CommonBugCategories.cpp new file mode 100644 index 0000000..e2a8ea6 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/CommonBugCategories.cpp @@ -0,0 +1,18 @@ +//=--- CommonBugCategories.cpp - Provides common issue categories -*- C++ -*-=// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// + +// Common strings used for the "category" of many static analyzer issues. +namespace clang { namespace ento { namespace categories { + +const char *CoreFoundationObjectiveC = "Core Foundation/Objective-C"; +const char *MemoryCoreFoundationObjectiveC = + "Memory (Core Foundation/Objective-C)"; +const char *UnixAPI = "Unix API"; +}}} + diff --git a/clang/lib/StaticAnalyzer/Checkers/DeadStoresChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/DeadStoresChecker.cpp new file mode 100644 index 0000000..510e8cd --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/DeadStoresChecker.cpp @@ -0,0 +1,386 @@ +//==- DeadStoresChecker.cpp - Check for stores to dead variables -*- C++ -*-==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines a DeadStores, a flow-sensitive checker that looks for +// stores to variables that are no longer live. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/Analysis/Analyses/LiveVariables.h" +#include "clang/Analysis/Visitors/CFGRecStmtVisitor.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugReporter.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ExprEngine.h" +#include "clang/Analysis/Visitors/CFGRecStmtDeclVisitor.h" +#include "clang/Basic/Diagnostic.h" +#include "clang/AST/ASTContext.h" +#include "clang/AST/ParentMap.h" +#include "llvm/ADT/SmallPtrSet.h" +#include "llvm/ADT/SmallString.h" + +using namespace clang; +using namespace ento; + +namespace { + +// FIXME: Eventually migrate into its own file, and have it managed by +// AnalysisManager. +class ReachableCode { + const CFG &cfg; + llvm::BitVector reachable; +public: + ReachableCode(const CFG &cfg) + : cfg(cfg), reachable(cfg.getNumBlockIDs(), false) {} + + void computeReachableBlocks(); + + bool isReachable(const CFGBlock *block) const { + return reachable[block->getBlockID()]; + } +}; +} + +void ReachableCode::computeReachableBlocks() { + if (!cfg.getNumBlockIDs()) + return; + + SmallVector<const CFGBlock*, 10> worklist; + worklist.push_back(&cfg.getEntry()); + + while (!worklist.empty()) { + const CFGBlock *block = worklist.back(); + worklist.pop_back(); + llvm::BitVector::reference isReachable = reachable[block->getBlockID()]; + if (isReachable) + continue; + isReachable = true; + for (CFGBlock::const_succ_iterator i = block->succ_begin(), + e = block->succ_end(); i != e; ++i) + if (const CFGBlock *succ = *i) + worklist.push_back(succ); + } +} + +static const Expr *LookThroughTransitiveAssignments(const Expr *Ex) { + while (Ex) { + const BinaryOperator *BO = + dyn_cast<BinaryOperator>(Ex->IgnoreParenCasts()); + if (!BO) + break; + if (BO->getOpcode() == BO_Assign) { + Ex = BO->getRHS(); + continue; + } + break; + } + return Ex; +} + +namespace { +class DeadStoreObs : public LiveVariables::Observer { + const CFG &cfg; + ASTContext &Ctx; + BugReporter& BR; + AnalysisDeclContext* AC; + ParentMap& Parents; + llvm::SmallPtrSet<const VarDecl*, 20> Escaped; + OwningPtr<ReachableCode> reachableCode; + const CFGBlock *currentBlock; + + enum DeadStoreKind { Standard, Enclosing, DeadIncrement, DeadInit }; + +public: + DeadStoreObs(const CFG &cfg, ASTContext &ctx, + BugReporter& br, AnalysisDeclContext* ac, ParentMap& parents, + llvm::SmallPtrSet<const VarDecl*, 20> &escaped) + : cfg(cfg), Ctx(ctx), BR(br), AC(ac), Parents(parents), + Escaped(escaped), currentBlock(0) {} + + virtual ~DeadStoreObs() {} + + void Report(const VarDecl *V, DeadStoreKind dsk, + PathDiagnosticLocation L, SourceRange R) { + if (Escaped.count(V)) + return; + + // Compute reachable blocks within the CFG for trivial cases + // where a bogus dead store can be reported because itself is unreachable. + if (!reachableCode.get()) { + reachableCode.reset(new ReachableCode(cfg)); + reachableCode->computeReachableBlocks(); + } + + if (!reachableCode->isReachable(currentBlock)) + return; + + SmallString<64> buf; + llvm::raw_svector_ostream os(buf); + const char *BugType = 0; + + switch (dsk) { + case DeadInit: + BugType = "Dead initialization"; + os << "Value stored to '" << *V + << "' during its initialization is never read"; + break; + + case DeadIncrement: + BugType = "Dead increment"; + case Standard: + if (!BugType) BugType = "Dead assignment"; + os << "Value stored to '" << *V << "' is never read"; + break; + + case Enclosing: + // Don't report issues in this case, e.g.: "if (x = foo())", + // where 'x' is unused later. We have yet to see a case where + // this is a real bug. + return; + } + + BR.EmitBasicReport(AC->getDecl(), BugType, "Dead store", os.str(), L, R); + } + + void CheckVarDecl(const VarDecl *VD, const Expr *Ex, const Expr *Val, + DeadStoreKind dsk, + const LiveVariables::LivenessValues &Live) { + + if (!VD->hasLocalStorage()) + return; + // Reference types confuse the dead stores checker. Skip them + // for now. + if (VD->getType()->getAs<ReferenceType>()) + return; + + if (!Live.isLive(VD) && + !(VD->getAttr<UnusedAttr>() || VD->getAttr<BlocksAttr>())) { + + PathDiagnosticLocation ExLoc = + PathDiagnosticLocation::createBegin(Ex, BR.getSourceManager(), AC); + Report(VD, dsk, ExLoc, Val->getSourceRange()); + } + } + + void CheckDeclRef(const DeclRefExpr *DR, const Expr *Val, DeadStoreKind dsk, + const LiveVariables::LivenessValues& Live) { + if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) + CheckVarDecl(VD, DR, Val, dsk, Live); + } + + bool isIncrement(VarDecl *VD, const BinaryOperator* B) { + if (B->isCompoundAssignmentOp()) + return true; + + const Expr *RHS = B->getRHS()->IgnoreParenCasts(); + const BinaryOperator* BRHS = dyn_cast<BinaryOperator>(RHS); + + if (!BRHS) + return false; + + const DeclRefExpr *DR; + + if ((DR = dyn_cast<DeclRefExpr>(BRHS->getLHS()->IgnoreParenCasts()))) + if (DR->getDecl() == VD) + return true; + + if ((DR = dyn_cast<DeclRefExpr>(BRHS->getRHS()->IgnoreParenCasts()))) + if (DR->getDecl() == VD) + return true; + + return false; + } + + virtual void observeStmt(const Stmt *S, const CFGBlock *block, + const LiveVariables::LivenessValues &Live) { + + currentBlock = block; + + // Skip statements in macros. + if (S->getLocStart().isMacroID()) + return; + + // Only cover dead stores from regular assignments. ++/-- dead stores + // have never flagged a real bug. + if (const BinaryOperator* B = dyn_cast<BinaryOperator>(S)) { + if (!B->isAssignmentOp()) return; // Skip non-assignments. + + if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(B->getLHS())) + if (VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { + // Special case: check for assigning null to a pointer. + // This is a common form of defensive programming. + const Expr *RHS = LookThroughTransitiveAssignments(B->getRHS()); + + QualType T = VD->getType(); + if (T->isPointerType() || T->isObjCObjectPointerType()) { + if (RHS->isNullPointerConstant(Ctx, Expr::NPC_ValueDependentIsNull)) + return; + } + + RHS = RHS->IgnoreParenCasts(); + // Special case: self-assignments. These are often used to shut up + // "unused variable" compiler warnings. + if (const DeclRefExpr *RhsDR = dyn_cast<DeclRefExpr>(RHS)) + if (VD == dyn_cast<VarDecl>(RhsDR->getDecl())) + return; + + // Otherwise, issue a warning. + DeadStoreKind dsk = Parents.isConsumedExpr(B) + ? Enclosing + : (isIncrement(VD,B) ? DeadIncrement : Standard); + + CheckVarDecl(VD, DR, B->getRHS(), dsk, Live); + } + } + else if (const UnaryOperator* U = dyn_cast<UnaryOperator>(S)) { + if (!U->isIncrementOp() || U->isPrefix()) + return; + + const Stmt *parent = Parents.getParentIgnoreParenCasts(U); + if (!parent || !isa<ReturnStmt>(parent)) + return; + + const Expr *Ex = U->getSubExpr()->IgnoreParenCasts(); + + if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Ex)) + CheckDeclRef(DR, U, DeadIncrement, Live); + } + else if (const DeclStmt *DS = dyn_cast<DeclStmt>(S)) + // Iterate through the decls. Warn if any initializers are complex + // expressions that are not live (never used). + for (DeclStmt::const_decl_iterator DI=DS->decl_begin(), DE=DS->decl_end(); + DI != DE; ++DI) { + + VarDecl *V = dyn_cast<VarDecl>(*DI); + + if (!V) + continue; + + if (V->hasLocalStorage()) { + // Reference types confuse the dead stores checker. Skip them + // for now. + if (V->getType()->getAs<ReferenceType>()) + return; + + if (const Expr *E = V->getInit()) { + while (const ExprWithCleanups *exprClean = + dyn_cast<ExprWithCleanups>(E)) + E = exprClean->getSubExpr(); + + // Look through transitive assignments, e.g.: + // int x = y = 0; + E = LookThroughTransitiveAssignments(E); + + // Don't warn on C++ objects (yet) until we can show that their + // constructors/destructors don't have side effects. + if (isa<CXXConstructExpr>(E)) + return; + + // A dead initialization is a variable that is dead after it + // is initialized. We don't flag warnings for those variables + // marked 'unused'. + if (!Live.isLive(V) && V->getAttr<UnusedAttr>() == 0) { + // Special case: check for initializations with constants. + // + // e.g. : int x = 0; + // + // If x is EVER assigned a new value later, don't issue + // a warning. This is because such initialization can be + // due to defensive programming. + if (E->isEvaluatable(Ctx)) + return; + + if (const DeclRefExpr *DRE = + dyn_cast<DeclRefExpr>(E->IgnoreParenCasts())) + if (const VarDecl *VD = dyn_cast<VarDecl>(DRE->getDecl())) { + // Special case: check for initialization from constant + // variables. + // + // e.g. extern const int MyConstant; + // int x = MyConstant; + // + if (VD->hasGlobalStorage() && + VD->getType().isConstQualified()) + return; + // Special case: check for initialization from scalar + // parameters. This is often a form of defensive + // programming. Non-scalars are still an error since + // because it more likely represents an actual algorithmic + // bug. + if (isa<ParmVarDecl>(VD) && VD->getType()->isScalarType()) + return; + } + + PathDiagnosticLocation Loc = + PathDiagnosticLocation::create(V, BR.getSourceManager()); + Report(V, DeadInit, Loc, E->getSourceRange()); + } + } + } + } + } +}; + +} // end anonymous namespace + +//===----------------------------------------------------------------------===// +// Driver function to invoke the Dead-Stores checker on a CFG. +//===----------------------------------------------------------------------===// + +namespace { +class FindEscaped : public CFGRecStmtDeclVisitor<FindEscaped>{ + CFG *cfg; +public: + FindEscaped(CFG *c) : cfg(c) {} + + CFG& getCFG() { return *cfg; } + + llvm::SmallPtrSet<const VarDecl*, 20> Escaped; + + void VisitUnaryOperator(UnaryOperator* U) { + // Check for '&'. Any VarDecl whose value has its address-taken we + // treat as escaped. + Expr *E = U->getSubExpr()->IgnoreParenCasts(); + if (U->getOpcode() == UO_AddrOf) + if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(E)) + if (VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { + Escaped.insert(VD); + return; + } + Visit(E); + } +}; +} // end anonymous namespace + + +//===----------------------------------------------------------------------===// +// DeadStoresChecker +//===----------------------------------------------------------------------===// + +namespace { +class DeadStoresChecker : public Checker<check::ASTCodeBody> { +public: + void checkASTCodeBody(const Decl *D, AnalysisManager& mgr, + BugReporter &BR) const { + if (LiveVariables *L = mgr.getAnalysis<LiveVariables>(D)) { + CFG &cfg = *mgr.getCFG(D); + AnalysisDeclContext *AC = mgr.getAnalysisDeclContext(D); + ParentMap &pmap = mgr.getParentMap(D); + FindEscaped FS(&cfg); + FS.getCFG().VisitBlockStmts(FS); + DeadStoreObs A(cfg, BR.getContext(), BR, AC, pmap, FS.Escaped); + L->runOnAllBlocks(A); + } + } +}; +} + +void ento::registerDeadStoresChecker(CheckerManager &mgr) { + mgr.registerChecker<DeadStoresChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/DebugCheckers.cpp b/clang/lib/StaticAnalyzer/Checkers/DebugCheckers.cpp new file mode 100644 index 0000000..34053cd --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/DebugCheckers.cpp @@ -0,0 +1,146 @@ +//==- DebugCheckers.cpp - Debugging Checkers ---------------------*- C++ -*-==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines a checkers that display debugging information. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/AnalysisManager.h" +#include "clang/Analysis/Analyses/LiveVariables.h" +#include "clang/Analysis/Analyses/Dominators.h" +#include "clang/Analysis/CallGraph.h" +#include "llvm/Support/Process.h" + +using namespace clang; +using namespace ento; + +//===----------------------------------------------------------------------===// +// DominatorsTreeDumper +//===----------------------------------------------------------------------===// + +namespace { +class DominatorsTreeDumper : public Checker<check::ASTCodeBody> { +public: + void checkASTCodeBody(const Decl *D, AnalysisManager& mgr, + BugReporter &BR) const { + if (AnalysisDeclContext *AC = mgr.getAnalysisDeclContext(D)) { + DominatorTree dom; + dom.buildDominatorTree(*AC); + dom.dump(); + } + } +}; +} + +void ento::registerDominatorsTreeDumper(CheckerManager &mgr) { + mgr.registerChecker<DominatorsTreeDumper>(); +} + +//===----------------------------------------------------------------------===// +// LiveVariablesDumper +//===----------------------------------------------------------------------===// + +namespace { +class LiveVariablesDumper : public Checker<check::ASTCodeBody> { +public: + void checkASTCodeBody(const Decl *D, AnalysisManager& mgr, + BugReporter &BR) const { + if (LiveVariables* L = mgr.getAnalysis<LiveVariables>(D)) { + L->dumpBlockLiveness(mgr.getSourceManager()); + } + } +}; +} + +void ento::registerLiveVariablesDumper(CheckerManager &mgr) { + mgr.registerChecker<LiveVariablesDumper>(); +} + +//===----------------------------------------------------------------------===// +// CFGViewer +//===----------------------------------------------------------------------===// + +namespace { +class CFGViewer : public Checker<check::ASTCodeBody> { +public: + void checkASTCodeBody(const Decl *D, AnalysisManager& mgr, + BugReporter &BR) const { + if (CFG *cfg = mgr.getCFG(D)) { + cfg->viewCFG(mgr.getLangOpts()); + } + } +}; +} + +void ento::registerCFGViewer(CheckerManager &mgr) { + mgr.registerChecker<CFGViewer>(); +} + +//===----------------------------------------------------------------------===// +// CFGDumper +//===----------------------------------------------------------------------===// + +namespace { +class CFGDumper : public Checker<check::ASTCodeBody> { +public: + void checkASTCodeBody(const Decl *D, AnalysisManager& mgr, + BugReporter &BR) const { + if (CFG *cfg = mgr.getCFG(D)) { + cfg->dump(mgr.getLangOpts(), + llvm::sys::Process::StandardErrHasColors()); + } + } +}; +} + +void ento::registerCFGDumper(CheckerManager &mgr) { + mgr.registerChecker<CFGDumper>(); +} + +//===----------------------------------------------------------------------===// +// CallGraphViewer +//===----------------------------------------------------------------------===// + +namespace { +class CallGraphViewer : public Checker< check::ASTDecl<TranslationUnitDecl> > { +public: + void checkASTDecl(const TranslationUnitDecl *TU, AnalysisManager& mgr, + BugReporter &BR) const { + CallGraph CG; + CG.addToCallGraph(const_cast<TranslationUnitDecl*>(TU)); + CG.viewGraph(); + } +}; +} + +void ento::registerCallGraphViewer(CheckerManager &mgr) { + mgr.registerChecker<CallGraphViewer>(); +} + +//===----------------------------------------------------------------------===// +// CallGraphDumper +//===----------------------------------------------------------------------===// + +namespace { +class CallGraphDumper : public Checker< check::ASTDecl<TranslationUnitDecl> > { +public: + void checkASTDecl(const TranslationUnitDecl *TU, AnalysisManager& mgr, + BugReporter &BR) const { + CallGraph CG; + CG.addToCallGraph(const_cast<TranslationUnitDecl*>(TU)); + CG.dump(); + } +}; +} + +void ento::registerCallGraphDumper(CheckerManager &mgr) { + mgr.registerChecker<CallGraphDumper>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/DereferenceChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/DereferenceChecker.cpp new file mode 100644 index 0000000..81a2745 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/DereferenceChecker.cpp @@ -0,0 +1,216 @@ +//== NullDerefChecker.cpp - Null dereference checker ------------*- C++ -*--==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This defines NullDerefChecker, a builtin check in ExprEngine that performs +// checks for null pointers at loads and stores. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/AST/ExprObjC.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "llvm/ADT/SmallString.h" + +using namespace clang; +using namespace ento; + +namespace { +class DereferenceChecker + : public Checker< check::Location, + EventDispatcher<ImplicitNullDerefEvent> > { + mutable OwningPtr<BuiltinBug> BT_null; + mutable OwningPtr<BuiltinBug> BT_undef; + +public: + void checkLocation(SVal location, bool isLoad, const Stmt* S, + CheckerContext &C) const; + + static const MemRegion *AddDerefSource(raw_ostream &os, + SmallVectorImpl<SourceRange> &Ranges, + const Expr *Ex, const ProgramState *state, + const LocationContext *LCtx, + bool loadedFrom = false); +}; +} // end anonymous namespace + +const MemRegion * +DereferenceChecker::AddDerefSource(raw_ostream &os, + SmallVectorImpl<SourceRange> &Ranges, + const Expr *Ex, + const ProgramState *state, + const LocationContext *LCtx, + bool loadedFrom) { + Ex = Ex->IgnoreParenLValueCasts(); + const MemRegion *sourceR = 0; + switch (Ex->getStmtClass()) { + default: + break; + case Stmt::DeclRefExprClass: { + const DeclRefExpr *DR = cast<DeclRefExpr>(Ex); + if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { + os << " (" << (loadedFrom ? "loaded from" : "from") + << " variable '" << VD->getName() << "')"; + Ranges.push_back(DR->getSourceRange()); + sourceR = state->getLValue(VD, LCtx).getAsRegion(); + } + break; + } + case Stmt::MemberExprClass: { + const MemberExpr *ME = cast<MemberExpr>(Ex); + os << " (" << (loadedFrom ? "loaded from" : "via") + << " field '" << ME->getMemberNameInfo() << "')"; + SourceLocation L = ME->getMemberLoc(); + Ranges.push_back(SourceRange(L, L)); + break; + } + } + return sourceR; +} + +void DereferenceChecker::checkLocation(SVal l, bool isLoad, const Stmt* S, + CheckerContext &C) const { + // Check for dereference of an undefined value. + if (l.isUndef()) { + if (ExplodedNode *N = C.generateSink()) { + if (!BT_undef) + BT_undef.reset(new BuiltinBug("Dereference of undefined pointer value")); + + BugReport *report = + new BugReport(*BT_undef, BT_undef->getDescription(), N); + report->addVisitor(bugreporter::getTrackNullOrUndefValueVisitor(N, + bugreporter::GetDerefExpr(N), report)); + C.EmitReport(report); + } + return; + } + + DefinedOrUnknownSVal location = cast<DefinedOrUnknownSVal>(l); + + // Check for null dereferences. + if (!isa<Loc>(location)) + return; + + ProgramStateRef state = C.getState(); + const LocationContext *LCtx = C.getLocationContext(); + ProgramStateRef notNullState, nullState; + llvm::tie(notNullState, nullState) = state->assume(location); + + // The explicit NULL case. + if (nullState) { + if (!notNullState) { + // Generate an error node. + ExplodedNode *N = C.generateSink(nullState); + if (!N) + return; + + // We know that 'location' cannot be non-null. This is what + // we call an "explicit" null dereference. + if (!BT_null) + BT_null.reset(new BuiltinBug("Dereference of null pointer")); + + SmallString<100> buf; + SmallVector<SourceRange, 2> Ranges; + + // Walk through lvalue casts to get the original expression + // that syntactically caused the load. + if (const Expr *expr = dyn_cast<Expr>(S)) + S = expr->IgnoreParenLValueCasts(); + + const MemRegion *sourceR = 0; + + switch (S->getStmtClass()) { + case Stmt::ArraySubscriptExprClass: { + llvm::raw_svector_ostream os(buf); + os << "Array access"; + const ArraySubscriptExpr *AE = cast<ArraySubscriptExpr>(S); + sourceR = + AddDerefSource(os, Ranges, AE->getBase()->IgnoreParenCasts(), + state.getPtr(), LCtx); + os << " results in a null pointer dereference"; + break; + } + case Stmt::UnaryOperatorClass: { + llvm::raw_svector_ostream os(buf); + os << "Dereference of null pointer"; + const UnaryOperator *U = cast<UnaryOperator>(S); + sourceR = + AddDerefSource(os, Ranges, U->getSubExpr()->IgnoreParens(), + state.getPtr(), LCtx, true); + break; + } + case Stmt::MemberExprClass: { + const MemberExpr *M = cast<MemberExpr>(S); + if (M->isArrow()) { + llvm::raw_svector_ostream os(buf); + os << "Access to field '" << M->getMemberNameInfo() + << "' results in a dereference of a null pointer"; + sourceR = + AddDerefSource(os, Ranges, M->getBase()->IgnoreParenCasts(), + state.getPtr(), LCtx, true); + } + break; + } + case Stmt::ObjCIvarRefExprClass: { + const ObjCIvarRefExpr *IV = cast<ObjCIvarRefExpr>(S); + if (const DeclRefExpr *DR = + dyn_cast<DeclRefExpr>(IV->getBase()->IgnoreParenCasts())) { + if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { + llvm::raw_svector_ostream os(buf); + os << "Instance variable access (via '" << VD->getName() + << "') results in a null pointer dereference"; + } + } + Ranges.push_back(IV->getSourceRange()); + break; + } + default: + break; + } + + BugReport *report = + new BugReport(*BT_null, + buf.empty() ? BT_null->getDescription():buf.str(), + N); + + report->addVisitor(bugreporter::getTrackNullOrUndefValueVisitor(N, + bugreporter::GetDerefExpr(N), report)); + + for (SmallVectorImpl<SourceRange>::iterator + I = Ranges.begin(), E = Ranges.end(); I!=E; ++I) + report->addRange(*I); + + if (sourceR) { + report->markInteresting(sourceR); + report->markInteresting(state->getRawSVal(loc::MemRegionVal(sourceR))); + } + + C.EmitReport(report); + return; + } + else { + // Otherwise, we have the case where the location could either be + // null or not-null. Record the error node as an "implicit" null + // dereference. + if (ExplodedNode *N = C.generateSink(nullState)) { + ImplicitNullDerefEvent event = { l, isLoad, N, &C.getBugReporter() }; + dispatchEvent(event); + } + } + } + + // From this point forward, we know that the location is not null. + C.addTransition(notNullState); +} + +void ento::registerDereferenceChecker(CheckerManager &mgr) { + mgr.registerChecker<DereferenceChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/DivZeroChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/DivZeroChecker.cpp new file mode 100644 index 0000000..2627f0c --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/DivZeroChecker.cpp @@ -0,0 +1,96 @@ +//== DivZeroChecker.cpp - Division by zero checker --------------*- C++ -*--==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This defines DivZeroChecker, a builtin check in ExprEngine that performs +// checks for division by zeros. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" + +using namespace clang; +using namespace ento; + +namespace { +class DivZeroChecker : public Checker< check::PreStmt<BinaryOperator> > { + mutable OwningPtr<BuiltinBug> BT; + void reportBug(const char *Msg, + ProgramStateRef StateZero, + CheckerContext &C) const ; +public: + void checkPreStmt(const BinaryOperator *B, CheckerContext &C) const; +}; +} // end anonymous namespace + +void DivZeroChecker::reportBug(const char *Msg, + ProgramStateRef StateZero, + CheckerContext &C) const { + if (ExplodedNode *N = C.generateSink(StateZero)) { + if (!BT) + BT.reset(new BuiltinBug("Division by zero")); + + BugReport *R = + new BugReport(*BT, Msg, N); + + R->addVisitor(bugreporter::getTrackNullOrUndefValueVisitor(N, + bugreporter::GetDenomExpr(N), R)); + C.EmitReport(R); + } +} + +void DivZeroChecker::checkPreStmt(const BinaryOperator *B, + CheckerContext &C) const { + BinaryOperator::Opcode Op = B->getOpcode(); + if (Op != BO_Div && + Op != BO_Rem && + Op != BO_DivAssign && + Op != BO_RemAssign) + return; + + if (!B->getRHS()->getType()->isIntegerType() || + !B->getRHS()->getType()->isScalarType()) + return; + + SVal Denom = C.getState()->getSVal(B->getRHS(), C.getLocationContext()); + const DefinedSVal *DV = dyn_cast<DefinedSVal>(&Denom); + + // Divide-by-undefined handled in the generic checking for uses of + // undefined values. + if (!DV) + return; + + // Check for divide by zero. + ConstraintManager &CM = C.getConstraintManager(); + ProgramStateRef stateNotZero, stateZero; + llvm::tie(stateNotZero, stateZero) = CM.assumeDual(C.getState(), *DV); + + if (!stateNotZero) { + assert(stateZero); + reportBug("Division by zero", stateZero, C); + return; + } + + bool TaintedD = C.getState()->isTainted(*DV); + if ((stateNotZero && stateZero && TaintedD)) { + reportBug("Division by a tainted value, possibly zero", stateZero, C); + return; + } + + // If we get here, then the denom should not be zero. We abandon the implicit + // zero denom case for now. + C.addTransition(stateNotZero); +} + +void ento::registerDivZeroChecker(CheckerManager &mgr) { + mgr.registerChecker<DivZeroChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/FixedAddressChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/FixedAddressChecker.cpp new file mode 100644 index 0000000..a1f2f3b --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/FixedAddressChecker.cpp @@ -0,0 +1,67 @@ +//=== FixedAddressChecker.cpp - Fixed address usage checker ----*- C++ -*--===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This files defines FixedAddressChecker, a builtin checker that checks for +// assignment of a fixed address to a pointer. +// This check corresponds to CWE-587. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" + +using namespace clang; +using namespace ento; + +namespace { +class FixedAddressChecker + : public Checker< check::PreStmt<BinaryOperator> > { + mutable OwningPtr<BuiltinBug> BT; + +public: + void checkPreStmt(const BinaryOperator *B, CheckerContext &C) const; +}; +} + +void FixedAddressChecker::checkPreStmt(const BinaryOperator *B, + CheckerContext &C) const { + // Using a fixed address is not portable because that address will probably + // not be valid in all environments or platforms. + + if (B->getOpcode() != BO_Assign) + return; + + QualType T = B->getType(); + if (!T->isPointerType()) + return; + + ProgramStateRef state = C.getState(); + SVal RV = state->getSVal(B->getRHS(), C.getLocationContext()); + + if (!RV.isConstant() || RV.isZeroConstant()) + return; + + if (ExplodedNode *N = C.addTransition()) { + if (!BT) + BT.reset(new BuiltinBug("Use fixed address", + "Using a fixed address is not portable because that " + "address will probably not be valid in all " + "environments or platforms.")); + BugReport *R = new BugReport(*BT, BT->getDescription(), N); + R->addRange(B->getRHS()->getSourceRange()); + C.EmitReport(R); + } +} + +void ento::registerFixedAddressChecker(CheckerManager &mgr) { + mgr.registerChecker<FixedAddressChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/GenericTaintChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/GenericTaintChecker.cpp new file mode 100644 index 0000000..135b81d --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/GenericTaintChecker.cpp @@ -0,0 +1,740 @@ +//== GenericTaintChecker.cpp ----------------------------------- -*- C++ -*--=// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This checker defines the attack surface for generic taint propagation. +// +// The taint information produced by it might be useful to other checkers. For +// example, checkers should report errors which involve tainted data more +// aggressively, even if the involved symbols are under constrained. +// +//===----------------------------------------------------------------------===// +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramStateTrait.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/Basic/Builtins.h" +#include <climits> + +using namespace clang; +using namespace ento; + +namespace { +class GenericTaintChecker : public Checker< check::PostStmt<CallExpr>, + check::PreStmt<CallExpr> > { +public: + static void *getTag() { static int Tag; return &Tag; } + + void checkPostStmt(const CallExpr *CE, CheckerContext &C) const; + void checkPostStmt(const DeclRefExpr *DRE, CheckerContext &C) const; + + void checkPreStmt(const CallExpr *CE, CheckerContext &C) const; + +private: + static const unsigned InvalidArgIndex = UINT_MAX; + /// Denotes the return vale. + static const unsigned ReturnValueIndex = UINT_MAX - 1; + + mutable OwningPtr<BugType> BT; + inline void initBugType() const { + if (!BT) + BT.reset(new BugType("Use of Untrusted Data", "Untrusted Data")); + } + + /// \brief Catch taint related bugs. Check if tainted data is passed to a + /// system call etc. + bool checkPre(const CallExpr *CE, CheckerContext &C) const; + + /// \brief Add taint sources on a pre-visit. + void addSourcesPre(const CallExpr *CE, CheckerContext &C) const; + + /// \brief Propagate taint generated at pre-visit. + bool propagateFromPre(const CallExpr *CE, CheckerContext &C) const; + + /// \brief Add taint sources on a post visit. + void addSourcesPost(const CallExpr *CE, CheckerContext &C) const; + + /// Check if the region the expression evaluates to is the standard input, + /// and thus, is tainted. + static bool isStdin(const Expr *E, CheckerContext &C); + + /// \brief Given a pointer argument, get the symbol of the value it contains + /// (points to). + static SymbolRef getPointedToSymbol(CheckerContext &C, const Expr *Arg); + + /// Functions defining the attack surface. + typedef ProgramStateRef (GenericTaintChecker::*FnCheck)(const CallExpr *, + CheckerContext &C) const; + ProgramStateRef postScanf(const CallExpr *CE, CheckerContext &C) const; + ProgramStateRef postSocket(const CallExpr *CE, CheckerContext &C) const; + ProgramStateRef postRetTaint(const CallExpr *CE, CheckerContext &C) const; + + /// Taint the scanned input if the file is tainted. + ProgramStateRef preFscanf(const CallExpr *CE, CheckerContext &C) const; + + /// Check for CWE-134: Uncontrolled Format String. + static const char MsgUncontrolledFormatString[]; + bool checkUncontrolledFormatString(const CallExpr *CE, + CheckerContext &C) const; + + /// Check for: + /// CERT/STR02-C. "Sanitize data passed to complex subsystems" + /// CWE-78, "Failure to Sanitize Data into an OS Command" + static const char MsgSanitizeSystemArgs[]; + bool checkSystemCall(const CallExpr *CE, StringRef Name, + CheckerContext &C) const; + + /// Check if tainted data is used as a buffer size ins strn.. functions, + /// and allocators. + static const char MsgTaintedBufferSize[]; + bool checkTaintedBufferSize(const CallExpr *CE, const FunctionDecl *FDecl, + CheckerContext &C) const; + + /// Generate a report if the expression is tainted or points to tainted data. + bool generateReportIfTainted(const Expr *E, const char Msg[], + CheckerContext &C) const; + + + typedef llvm::SmallVector<unsigned, 2> ArgVector; + + /// \brief A struct used to specify taint propagation rules for a function. + /// + /// If any of the possible taint source arguments is tainted, all of the + /// destination arguments should also be tainted. Use InvalidArgIndex in the + /// src list to specify that all of the arguments can introduce taint. Use + /// InvalidArgIndex in the dst arguments to signify that all the non-const + /// pointer and reference arguments might be tainted on return. If + /// ReturnValueIndex is added to the dst list, the return value will be + /// tainted. + struct TaintPropagationRule { + /// List of arguments which can be taint sources and should be checked. + ArgVector SrcArgs; + /// List of arguments which should be tainted on function return. + ArgVector DstArgs; + // TODO: Check if using other data structures would be more optimal. + + TaintPropagationRule() {} + + TaintPropagationRule(unsigned SArg, + unsigned DArg, bool TaintRet = false) { + SrcArgs.push_back(SArg); + DstArgs.push_back(DArg); + if (TaintRet) + DstArgs.push_back(ReturnValueIndex); + } + + TaintPropagationRule(unsigned SArg1, unsigned SArg2, + unsigned DArg, bool TaintRet = false) { + SrcArgs.push_back(SArg1); + SrcArgs.push_back(SArg2); + DstArgs.push_back(DArg); + if (TaintRet) + DstArgs.push_back(ReturnValueIndex); + } + + /// Get the propagation rule for a given function. + static TaintPropagationRule + getTaintPropagationRule(const FunctionDecl *FDecl, + StringRef Name, + CheckerContext &C); + + inline void addSrcArg(unsigned A) { SrcArgs.push_back(A); } + inline void addDstArg(unsigned A) { DstArgs.push_back(A); } + + inline bool isNull() const { return SrcArgs.empty(); } + + inline bool isDestinationArgument(unsigned ArgNum) const { + return (std::find(DstArgs.begin(), + DstArgs.end(), ArgNum) != DstArgs.end()); + } + + static inline bool isTaintedOrPointsToTainted(const Expr *E, + ProgramStateRef State, + CheckerContext &C) { + return (State->isTainted(E, C.getLocationContext()) || isStdin(E, C) || + (E->getType().getTypePtr()->isPointerType() && + State->isTainted(getPointedToSymbol(C, E)))); + } + + /// \brief Pre-process a function which propagates taint according to the + /// taint rule. + ProgramStateRef process(const CallExpr *CE, CheckerContext &C) const; + + }; +}; + +const unsigned GenericTaintChecker::ReturnValueIndex; +const unsigned GenericTaintChecker::InvalidArgIndex; + +const char GenericTaintChecker::MsgUncontrolledFormatString[] = + "Untrusted data is used as a format string " + "(CWE-134: Uncontrolled Format String)"; + +const char GenericTaintChecker::MsgSanitizeSystemArgs[] = + "Untrusted data is passed to a system call " + "(CERT/STR02-C. Sanitize data passed to complex subsystems)"; + +const char GenericTaintChecker::MsgTaintedBufferSize[] = + "Untrusted data is used to specify the buffer size " + "(CERT/STR31-C. Guarantee that storage for strings has sufficient space for " + "character data and the null terminator)"; + +} // end of anonymous namespace + +/// A set which is used to pass information from call pre-visit instruction +/// to the call post-visit. The values are unsigned integers, which are either +/// ReturnValueIndex, or indexes of the pointer/reference argument, which +/// points to data, which should be tainted on return. +namespace { struct TaintArgsOnPostVisit{}; } +namespace clang { namespace ento { +template<> struct ProgramStateTrait<TaintArgsOnPostVisit> + : public ProgramStatePartialTrait<llvm::ImmutableSet<unsigned> > { + static void *GDMIndex() { return GenericTaintChecker::getTag(); } +}; +}} + +GenericTaintChecker::TaintPropagationRule +GenericTaintChecker::TaintPropagationRule::getTaintPropagationRule( + const FunctionDecl *FDecl, + StringRef Name, + CheckerContext &C) { + // TODO: Currently, we might loose precision here: we always mark a return + // value as tainted even if it's just a pointer, pointing to tainted data. + + // Check for exact name match for functions without builtin substitutes. + TaintPropagationRule Rule = llvm::StringSwitch<TaintPropagationRule>(Name) + .Case("atoi", TaintPropagationRule(0, ReturnValueIndex)) + .Case("atol", TaintPropagationRule(0, ReturnValueIndex)) + .Case("atoll", TaintPropagationRule(0, ReturnValueIndex)) + .Case("getc", TaintPropagationRule(0, ReturnValueIndex)) + .Case("fgetc", TaintPropagationRule(0, ReturnValueIndex)) + .Case("getc_unlocked", TaintPropagationRule(0, ReturnValueIndex)) + .Case("getw", TaintPropagationRule(0, ReturnValueIndex)) + .Case("toupper", TaintPropagationRule(0, ReturnValueIndex)) + .Case("tolower", TaintPropagationRule(0, ReturnValueIndex)) + .Case("strchr", TaintPropagationRule(0, ReturnValueIndex)) + .Case("strrchr", TaintPropagationRule(0, ReturnValueIndex)) + .Case("read", TaintPropagationRule(0, 2, 1, true)) + .Case("pread", TaintPropagationRule(InvalidArgIndex, 1, true)) + .Case("gets", TaintPropagationRule(InvalidArgIndex, 0, true)) + .Case("fgets", TaintPropagationRule(2, 0, true)) + .Case("getline", TaintPropagationRule(2, 0)) + .Case("getdelim", TaintPropagationRule(3, 0)) + .Case("fgetln", TaintPropagationRule(0, ReturnValueIndex)) + .Default(TaintPropagationRule()); + + if (!Rule.isNull()) + return Rule; + + // Check if it's one of the memory setting/copying functions. + // This check is specialized but faster then calling isCLibraryFunction. + unsigned BId = 0; + if ( (BId = FDecl->getMemoryFunctionKind()) ) + switch(BId) { + case Builtin::BImemcpy: + case Builtin::BImemmove: + case Builtin::BIstrncpy: + case Builtin::BIstrncat: + return TaintPropagationRule(1, 2, 0, true); + case Builtin::BIstrlcpy: + case Builtin::BIstrlcat: + return TaintPropagationRule(1, 2, 0, false); + case Builtin::BIstrndup: + return TaintPropagationRule(0, 1, ReturnValueIndex); + + default: + break; + }; + + // Process all other functions which could be defined as builtins. + if (Rule.isNull()) { + if (C.isCLibraryFunction(FDecl, "snprintf") || + C.isCLibraryFunction(FDecl, "sprintf")) + return TaintPropagationRule(InvalidArgIndex, 0, true); + else if (C.isCLibraryFunction(FDecl, "strcpy") || + C.isCLibraryFunction(FDecl, "stpcpy") || + C.isCLibraryFunction(FDecl, "strcat")) + return TaintPropagationRule(1, 0, true); + else if (C.isCLibraryFunction(FDecl, "bcopy")) + return TaintPropagationRule(0, 2, 1, false); + else if (C.isCLibraryFunction(FDecl, "strdup") || + C.isCLibraryFunction(FDecl, "strdupa")) + return TaintPropagationRule(0, ReturnValueIndex); + else if (C.isCLibraryFunction(FDecl, "wcsdup")) + return TaintPropagationRule(0, ReturnValueIndex); + } + + // Skipping the following functions, since they might be used for cleansing + // or smart memory copy: + // - memccpy - copying untill hitting a special character. + + return TaintPropagationRule(); +} + +void GenericTaintChecker::checkPreStmt(const CallExpr *CE, + CheckerContext &C) const { + // Check for errors first. + if (checkPre(CE, C)) + return; + + // Add taint second. + addSourcesPre(CE, C); +} + +void GenericTaintChecker::checkPostStmt(const CallExpr *CE, + CheckerContext &C) const { + if (propagateFromPre(CE, C)) + return; + addSourcesPost(CE, C); +} + +void GenericTaintChecker::addSourcesPre(const CallExpr *CE, + CheckerContext &C) const { + ProgramStateRef State = 0; + const FunctionDecl *FDecl = C.getCalleeDecl(CE); + StringRef Name = C.getCalleeName(FDecl); + if (Name.empty()) + return; + + // First, try generating a propagation rule for this function. + TaintPropagationRule Rule = + TaintPropagationRule::getTaintPropagationRule(FDecl, Name, C); + if (!Rule.isNull()) { + State = Rule.process(CE, C); + if (!State) + return; + C.addTransition(State); + return; + } + + // Otherwise, check if we have custom pre-processing implemented. + FnCheck evalFunction = llvm::StringSwitch<FnCheck>(Name) + .Case("fscanf", &GenericTaintChecker::preFscanf) + .Default(0); + // Check and evaluate the call. + if (evalFunction) + State = (this->*evalFunction)(CE, C); + if (!State) + return; + C.addTransition(State); + +} + +bool GenericTaintChecker::propagateFromPre(const CallExpr *CE, + CheckerContext &C) const { + ProgramStateRef State = C.getState(); + + // Depending on what was tainted at pre-visit, we determined a set of + // arguments which should be tainted after the function returns. These are + // stored in the state as TaintArgsOnPostVisit set. + llvm::ImmutableSet<unsigned> TaintArgs = State->get<TaintArgsOnPostVisit>(); + if (TaintArgs.isEmpty()) + return false; + + for (llvm::ImmutableSet<unsigned>::iterator + I = TaintArgs.begin(), E = TaintArgs.end(); I != E; ++I) { + unsigned ArgNum = *I; + + // Special handling for the tainted return value. + if (ArgNum == ReturnValueIndex) { + State = State->addTaint(CE, C.getLocationContext()); + continue; + } + + // The arguments are pointer arguments. The data they are pointing at is + // tainted after the call. + if (CE->getNumArgs() < (ArgNum + 1)) + return false; + const Expr* Arg = CE->getArg(ArgNum); + SymbolRef Sym = getPointedToSymbol(C, Arg); + if (Sym) + State = State->addTaint(Sym); + } + + // Clear up the taint info from the state. + State = State->remove<TaintArgsOnPostVisit>(); + + if (State != C.getState()) { + C.addTransition(State); + return true; + } + return false; +} + +void GenericTaintChecker::addSourcesPost(const CallExpr *CE, + CheckerContext &C) const { + // Define the attack surface. + // Set the evaluation function by switching on the callee name. + StringRef Name = C.getCalleeName(CE); + if (Name.empty()) + return; + FnCheck evalFunction = llvm::StringSwitch<FnCheck>(Name) + .Case("scanf", &GenericTaintChecker::postScanf) + // TODO: Add support for vfscanf & family. + .Case("getchar", &GenericTaintChecker::postRetTaint) + .Case("getchar_unlocked", &GenericTaintChecker::postRetTaint) + .Case("getenv", &GenericTaintChecker::postRetTaint) + .Case("fopen", &GenericTaintChecker::postRetTaint) + .Case("fdopen", &GenericTaintChecker::postRetTaint) + .Case("freopen", &GenericTaintChecker::postRetTaint) + .Case("getch", &GenericTaintChecker::postRetTaint) + .Case("wgetch", &GenericTaintChecker::postRetTaint) + .Case("socket", &GenericTaintChecker::postSocket) + .Default(0); + + // If the callee isn't defined, it is not of security concern. + // Check and evaluate the call. + ProgramStateRef State = 0; + if (evalFunction) + State = (this->*evalFunction)(CE, C); + if (!State) + return; + + C.addTransition(State); +} + +bool GenericTaintChecker::checkPre(const CallExpr *CE, CheckerContext &C) const{ + + if (checkUncontrolledFormatString(CE, C)) + return true; + + const FunctionDecl *FDecl = C.getCalleeDecl(CE); + StringRef Name = C.getCalleeName(FDecl); + if (Name.empty()) + return false; + + if (checkSystemCall(CE, Name, C)) + return true; + + if (checkTaintedBufferSize(CE, FDecl, C)) + return true; + + return false; +} + +SymbolRef GenericTaintChecker::getPointedToSymbol(CheckerContext &C, + const Expr* Arg) { + ProgramStateRef State = C.getState(); + SVal AddrVal = State->getSVal(Arg->IgnoreParens(), C.getLocationContext()); + if (AddrVal.isUnknownOrUndef()) + return 0; + + Loc *AddrLoc = dyn_cast<Loc>(&AddrVal); + if (!AddrLoc) + return 0; + + const PointerType *ArgTy = + dyn_cast<PointerType>(Arg->getType().getCanonicalType().getTypePtr()); + SVal Val = State->getSVal(*AddrLoc, + ArgTy ? ArgTy->getPointeeType(): QualType()); + return Val.getAsSymbol(); +} + +ProgramStateRef +GenericTaintChecker::TaintPropagationRule::process(const CallExpr *CE, + CheckerContext &C) const { + ProgramStateRef State = C.getState(); + + // Check for taint in arguments. + bool IsTainted = false; + for (ArgVector::const_iterator I = SrcArgs.begin(), + E = SrcArgs.end(); I != E; ++I) { + unsigned ArgNum = *I; + + if (ArgNum == InvalidArgIndex) { + // Check if any of the arguments is tainted, but skip the + // destination arguments. + for (unsigned int i = 0; i < CE->getNumArgs(); ++i) { + if (isDestinationArgument(i)) + continue; + if ((IsTainted = isTaintedOrPointsToTainted(CE->getArg(i), State, C))) + break; + } + break; + } + + if (CE->getNumArgs() < (ArgNum + 1)) + return State; + if ((IsTainted = isTaintedOrPointsToTainted(CE->getArg(ArgNum), State, C))) + break; + } + if (!IsTainted) + return State; + + // Mark the arguments which should be tainted after the function returns. + for (ArgVector::const_iterator I = DstArgs.begin(), + E = DstArgs.end(); I != E; ++I) { + unsigned ArgNum = *I; + + // Should we mark all arguments as tainted? + if (ArgNum == InvalidArgIndex) { + // For all pointer and references that were passed in: + // If they are not pointing to const data, mark data as tainted. + // TODO: So far we are just going one level down; ideally we'd need to + // recurse here. + for (unsigned int i = 0; i < CE->getNumArgs(); ++i) { + const Expr *Arg = CE->getArg(i); + // Process pointer argument. + const Type *ArgTy = Arg->getType().getTypePtr(); + QualType PType = ArgTy->getPointeeType(); + if ((!PType.isNull() && !PType.isConstQualified()) + || (ArgTy->isReferenceType() && !Arg->getType().isConstQualified())) + State = State->add<TaintArgsOnPostVisit>(i); + } + continue; + } + + // Should mark the return value? + if (ArgNum == ReturnValueIndex) { + State = State->add<TaintArgsOnPostVisit>(ReturnValueIndex); + continue; + } + + // Mark the given argument. + assert(ArgNum < CE->getNumArgs()); + State = State->add<TaintArgsOnPostVisit>(ArgNum); + } + + return State; +} + + +// If argument 0 (file descriptor) is tainted, all arguments except for arg 0 +// and arg 1 should get taint. +ProgramStateRef GenericTaintChecker::preFscanf(const CallExpr *CE, + CheckerContext &C) const { + assert(CE->getNumArgs() >= 2); + ProgramStateRef State = C.getState(); + + // Check is the file descriptor is tainted. + if (State->isTainted(CE->getArg(0), C.getLocationContext()) || + isStdin(CE->getArg(0), C)) { + // All arguments except for the first two should get taint. + for (unsigned int i = 2; i < CE->getNumArgs(); ++i) + State = State->add<TaintArgsOnPostVisit>(i); + return State; + } + + return 0; +} + + +// If argument 0(protocol domain) is network, the return value should get taint. +ProgramStateRef GenericTaintChecker::postSocket(const CallExpr *CE, + CheckerContext &C) const { + ProgramStateRef State = C.getState(); + if (CE->getNumArgs() < 3) + return State; + + SourceLocation DomLoc = CE->getArg(0)->getExprLoc(); + StringRef DomName = C.getMacroNameOrSpelling(DomLoc); + // White list the internal communication protocols. + if (DomName.equals("AF_SYSTEM") || DomName.equals("AF_LOCAL") || + DomName.equals("AF_UNIX") || DomName.equals("AF_RESERVED_36")) + return State; + State = State->addTaint(CE, C.getLocationContext()); + return State; +} + +ProgramStateRef GenericTaintChecker::postScanf(const CallExpr *CE, + CheckerContext &C) const { + ProgramStateRef State = C.getState(); + if (CE->getNumArgs() < 2) + return State; + + SVal x = State->getSVal(CE->getArg(1), C.getLocationContext()); + // All arguments except for the very first one should get taint. + for (unsigned int i = 1; i < CE->getNumArgs(); ++i) { + // The arguments are pointer arguments. The data they are pointing at is + // tainted after the call. + const Expr* Arg = CE->getArg(i); + SymbolRef Sym = getPointedToSymbol(C, Arg); + if (Sym) + State = State->addTaint(Sym); + } + return State; +} + +ProgramStateRef GenericTaintChecker::postRetTaint(const CallExpr *CE, + CheckerContext &C) const { + return C.getState()->addTaint(CE, C.getLocationContext()); +} + +bool GenericTaintChecker::isStdin(const Expr *E, CheckerContext &C) { + ProgramStateRef State = C.getState(); + SVal Val = State->getSVal(E, C.getLocationContext()); + + // stdin is a pointer, so it would be a region. + const MemRegion *MemReg = Val.getAsRegion(); + + // The region should be symbolic, we do not know it's value. + const SymbolicRegion *SymReg = dyn_cast_or_null<SymbolicRegion>(MemReg); + if (!SymReg) + return false; + + // Get it's symbol and find the declaration region it's pointing to. + const SymbolRegionValue *Sm =dyn_cast<SymbolRegionValue>(SymReg->getSymbol()); + if (!Sm) + return false; + const DeclRegion *DeclReg = dyn_cast_or_null<DeclRegion>(Sm->getRegion()); + if (!DeclReg) + return false; + + // This region corresponds to a declaration, find out if it's a global/extern + // variable named stdin with the proper type. + if (const VarDecl *D = dyn_cast_or_null<VarDecl>(DeclReg->getDecl())) { + D = D->getCanonicalDecl(); + if ((D->getName().find("stdin") != StringRef::npos) && D->isExternC()) + if (const PointerType * PtrTy = + dyn_cast<PointerType>(D->getType().getTypePtr())) + if (PtrTy->getPointeeType() == C.getASTContext().getFILEType()) + return true; + } + return false; +} + +static bool getPrintfFormatArgumentNum(const CallExpr *CE, + const CheckerContext &C, + unsigned int &ArgNum) { + // Find if the function contains a format string argument. + // Handles: fprintf, printf, sprintf, snprintf, vfprintf, vprintf, vsprintf, + // vsnprintf, syslog, custom annotated functions. + const FunctionDecl *FDecl = C.getCalleeDecl(CE); + if (!FDecl) + return false; + for (specific_attr_iterator<FormatAttr> + i = FDecl->specific_attr_begin<FormatAttr>(), + e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { + + const FormatAttr *Format = *i; + ArgNum = Format->getFormatIdx() - 1; + if ((Format->getType() == "printf") && CE->getNumArgs() > ArgNum) + return true; + } + + // Or if a function is named setproctitle (this is a heuristic). + if (C.getCalleeName(CE).find("setproctitle") != StringRef::npos) { + ArgNum = 0; + return true; + } + + return false; +} + +bool GenericTaintChecker::generateReportIfTainted(const Expr *E, + const char Msg[], + CheckerContext &C) const { + assert(E); + + // Check for taint. + ProgramStateRef State = C.getState(); + if (!State->isTainted(getPointedToSymbol(C, E)) && + !State->isTainted(E, C.getLocationContext())) + return false; + + // Generate diagnostic. + if (ExplodedNode *N = C.addTransition()) { + initBugType(); + BugReport *report = new BugReport(*BT, Msg, N); + report->addRange(E->getSourceRange()); + C.EmitReport(report); + return true; + } + return false; +} + +bool GenericTaintChecker::checkUncontrolledFormatString(const CallExpr *CE, + CheckerContext &C) const{ + // Check if the function contains a format string argument. + unsigned int ArgNum = 0; + if (!getPrintfFormatArgumentNum(CE, C, ArgNum)) + return false; + + // If either the format string content or the pointer itself are tainted, warn. + if (generateReportIfTainted(CE->getArg(ArgNum), + MsgUncontrolledFormatString, C)) + return true; + return false; +} + +bool GenericTaintChecker::checkSystemCall(const CallExpr *CE, + StringRef Name, + CheckerContext &C) const { + // TODO: It might make sense to run this check on demand. In some cases, + // we should check if the environment has been cleansed here. We also might + // need to know if the user was reset before these calls(seteuid). + unsigned ArgNum = llvm::StringSwitch<unsigned>(Name) + .Case("system", 0) + .Case("popen", 0) + .Case("execl", 0) + .Case("execle", 0) + .Case("execlp", 0) + .Case("execv", 0) + .Case("execvp", 0) + .Case("execvP", 0) + .Case("execve", 0) + .Case("dlopen", 0) + .Default(UINT_MAX); + + if (ArgNum == UINT_MAX || CE->getNumArgs() < (ArgNum + 1)) + return false; + + if (generateReportIfTainted(CE->getArg(ArgNum), + MsgSanitizeSystemArgs, C)) + return true; + + return false; +} + +// TODO: Should this check be a part of the CString checker? +// If yes, should taint be a global setting? +bool GenericTaintChecker::checkTaintedBufferSize(const CallExpr *CE, + const FunctionDecl *FDecl, + CheckerContext &C) const { + // If the function has a buffer size argument, set ArgNum. + unsigned ArgNum = InvalidArgIndex; + unsigned BId = 0; + if ( (BId = FDecl->getMemoryFunctionKind()) ) + switch(BId) { + case Builtin::BImemcpy: + case Builtin::BImemmove: + case Builtin::BIstrncpy: + ArgNum = 2; + break; + case Builtin::BIstrndup: + ArgNum = 1; + break; + default: + break; + }; + + if (ArgNum == InvalidArgIndex) { + if (C.isCLibraryFunction(FDecl, "malloc") || + C.isCLibraryFunction(FDecl, "calloc") || + C.isCLibraryFunction(FDecl, "alloca")) + ArgNum = 0; + else if (C.isCLibraryFunction(FDecl, "memccpy")) + ArgNum = 3; + else if (C.isCLibraryFunction(FDecl, "realloc")) + ArgNum = 1; + else if (C.isCLibraryFunction(FDecl, "bcopy")) + ArgNum = 2; + } + + if (ArgNum != InvalidArgIndex && CE->getNumArgs() > ArgNum && + generateReportIfTainted(CE->getArg(ArgNum), MsgTaintedBufferSize, C)) + return true; + + return false; +} + +void ento::registerGenericTaintChecker(CheckerManager &mgr) { + mgr.registerChecker<GenericTaintChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/IdempotentOperationChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/IdempotentOperationChecker.cpp new file mode 100644 index 0000000..c08f163 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/IdempotentOperationChecker.cpp @@ -0,0 +1,747 @@ +//==- IdempotentOperationChecker.cpp - Idempotent Operations ----*- C++ -*-==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines a set of path-sensitive checks for idempotent and/or +// tautological operations. Each potential operation is checked along all paths +// to see if every path results in a pointless operation. +// +-------------------------------------------+ +// |Table of idempotent/tautological operations| +// +-------------------------------------------+ +//+--------------------------------------------------------------------------+ +//|Operator | x op x | x op 1 | 1 op x | x op 0 | 0 op x | x op ~0 | ~0 op x | +//+--------------------------------------------------------------------------+ +// +, += | | | | x | x | | +// -, -= | | | | x | -x | | +// *, *= | | x | x | 0 | 0 | | +// /, /= | 1 | x | | N/A | 0 | | +// &, &= | x | | | 0 | 0 | x | x +// |, |= | x | | | x | x | ~0 | ~0 +// ^, ^= | 0 | | | x | x | | +// <<, <<= | | | | x | 0 | | +// >>, >>= | | | | x | 0 | | +// || | 1 | 1 | 1 | x | x | 1 | 1 +// && | 1 | x | x | 0 | 0 | x | x +// = | x | | | | | | +// == | 1 | | | | | | +// >= | 1 | | | | | | +// <= | 1 | | | | | | +// > | 0 | | | | | | +// < | 0 | | | | | | +// != | 0 | | | | | | +//===----------------------------------------------------------------------===// +// +// Things TODO: +// - Improved error messages +// - Handle mixed assumptions (which assumptions can belong together?) +// - Finer grained false positive control (levels) +// - Handling ~0 values + +#include "ClangSACheckers.h" +#include "clang/Analysis/CFGStmtMap.h" +#include "clang/Analysis/Analyses/PseudoConstantAnalysis.h" +#include "clang/Analysis/Analyses/CFGReachabilityAnalysis.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugReporter.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerHelpers.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CoreEngine.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/SVals.h" +#include "clang/AST/Stmt.h" +#include "llvm/ADT/DenseMap.h" +#include "llvm/ADT/SmallSet.h" +#include "llvm/ADT/SmallString.h" +#include "llvm/ADT/BitVector.h" +#include "llvm/Support/ErrorHandling.h" + +using namespace clang; +using namespace ento; + +namespace { +class IdempotentOperationChecker + : public Checker<check::PreStmt<BinaryOperator>, + check::PostStmt<BinaryOperator>, + check::EndAnalysis> { +public: + void checkPreStmt(const BinaryOperator *B, CheckerContext &C) const; + void checkPostStmt(const BinaryOperator *B, CheckerContext &C) const; + void checkEndAnalysis(ExplodedGraph &G, BugReporter &B,ExprEngine &Eng) const; + +private: + // Our assumption about a particular operation. + enum Assumption { Possible = 0, Impossible, Equal, LHSis1, RHSis1, LHSis0, + RHSis0 }; + + static void UpdateAssumption(Assumption &A, const Assumption &New); + + // False positive reduction methods + static bool isSelfAssign(const Expr *LHS, const Expr *RHS); + static bool isUnused(const Expr *E, AnalysisDeclContext *AC); + static bool isTruncationExtensionAssignment(const Expr *LHS, + const Expr *RHS); + static bool pathWasCompletelyAnalyzed(AnalysisDeclContext *AC, + const CFGBlock *CB, + const CoreEngine &CE); + static bool CanVary(const Expr *Ex, + AnalysisDeclContext *AC); + static bool isConstantOrPseudoConstant(const DeclRefExpr *DR, + AnalysisDeclContext *AC); + static bool containsNonLocalVarDecl(const Stmt *S); + + // Hash table and related data structures + struct BinaryOperatorData { + BinaryOperatorData() : assumption(Possible) {} + + Assumption assumption; + ExplodedNodeSet explodedNodes; // Set of ExplodedNodes that refer to a + // BinaryOperator + }; + typedef llvm::DenseMap<const BinaryOperator *, BinaryOperatorData> + AssumptionMap; + mutable AssumptionMap hash; +}; +} + +void IdempotentOperationChecker::checkPreStmt(const BinaryOperator *B, + CheckerContext &C) const { + // Find or create an entry in the hash for this BinaryOperator instance. + // If we haven't done a lookup before, it will get default initialized to + // 'Possible'. At this stage we do not store the ExplodedNode, as it has not + // been created yet. + BinaryOperatorData &Data = hash[B]; + Assumption &A = Data.assumption; + AnalysisDeclContext *AC = C.getCurrentAnalysisDeclContext(); + + // If we already have visited this node on a path that does not contain an + // idempotent operation, return immediately. + if (A == Impossible) + return; + + // Retrieve both sides of the operator and determine if they can vary (which + // may mean this is a false positive. + const Expr *LHS = B->getLHS(); + const Expr *RHS = B->getRHS(); + + // At this stage we can calculate whether each side contains a false positive + // that applies to all operators. We only need to calculate this the first + // time. + bool LHSContainsFalsePositive = false, RHSContainsFalsePositive = false; + if (A == Possible) { + // An expression contains a false positive if it can't vary, or if it + // contains a known false positive VarDecl. + LHSContainsFalsePositive = !CanVary(LHS, AC) + || containsNonLocalVarDecl(LHS); + RHSContainsFalsePositive = !CanVary(RHS, AC) + || containsNonLocalVarDecl(RHS); + } + + ProgramStateRef state = C.getState(); + const LocationContext *LCtx = C.getLocationContext(); + SVal LHSVal = state->getSVal(LHS, LCtx); + SVal RHSVal = state->getSVal(RHS, LCtx); + + // If either value is unknown, we can't be 100% sure of all paths. + if (LHSVal.isUnknownOrUndef() || RHSVal.isUnknownOrUndef()) { + A = Impossible; + return; + } + BinaryOperator::Opcode Op = B->getOpcode(); + + // Dereference the LHS SVal if this is an assign operation + switch (Op) { + default: + break; + + // Fall through intentional + case BO_AddAssign: + case BO_SubAssign: + case BO_MulAssign: + case BO_DivAssign: + case BO_AndAssign: + case BO_OrAssign: + case BO_XorAssign: + case BO_ShlAssign: + case BO_ShrAssign: + case BO_Assign: + // Assign statements have one extra level of indirection + if (!isa<Loc>(LHSVal)) { + A = Impossible; + return; + } + LHSVal = state->getSVal(cast<Loc>(LHSVal), LHS->getType()); + } + + + // We now check for various cases which result in an idempotent operation. + + // x op x + switch (Op) { + default: + break; // We don't care about any other operators. + + // Fall through intentional + case BO_Assign: + // x Assign x can be used to silence unused variable warnings intentionally. + // If this is a self assignment and the variable is referenced elsewhere, + // and the assignment is not a truncation or extension, then it is a false + // positive. + if (isSelfAssign(LHS, RHS)) { + if (!isUnused(LHS, AC) && !isTruncationExtensionAssignment(LHS, RHS)) { + UpdateAssumption(A, Equal); + return; + } + else { + A = Impossible; + return; + } + } + + case BO_SubAssign: + case BO_DivAssign: + case BO_AndAssign: + case BO_OrAssign: + case BO_XorAssign: + case BO_Sub: + case BO_Div: + case BO_And: + case BO_Or: + case BO_Xor: + case BO_LOr: + case BO_LAnd: + case BO_EQ: + case BO_NE: + if (LHSVal != RHSVal || LHSContainsFalsePositive + || RHSContainsFalsePositive) + break; + UpdateAssumption(A, Equal); + return; + } + + // x op 1 + switch (Op) { + default: + break; // We don't care about any other operators. + + // Fall through intentional + case BO_MulAssign: + case BO_DivAssign: + case BO_Mul: + case BO_Div: + case BO_LOr: + case BO_LAnd: + if (!RHSVal.isConstant(1) || RHSContainsFalsePositive) + break; + UpdateAssumption(A, RHSis1); + return; + } + + // 1 op x + switch (Op) { + default: + break; // We don't care about any other operators. + + // Fall through intentional + case BO_MulAssign: + case BO_Mul: + case BO_LOr: + case BO_LAnd: + if (!LHSVal.isConstant(1) || LHSContainsFalsePositive) + break; + UpdateAssumption(A, LHSis1); + return; + } + + // x op 0 + switch (Op) { + default: + break; // We don't care about any other operators. + + // Fall through intentional + case BO_AddAssign: + case BO_SubAssign: + case BO_MulAssign: + case BO_AndAssign: + case BO_OrAssign: + case BO_XorAssign: + case BO_Add: + case BO_Sub: + case BO_Mul: + case BO_And: + case BO_Or: + case BO_Xor: + case BO_Shl: + case BO_Shr: + case BO_LOr: + case BO_LAnd: + if (!RHSVal.isConstant(0) || RHSContainsFalsePositive) + break; + UpdateAssumption(A, RHSis0); + return; + } + + // 0 op x + switch (Op) { + default: + break; // We don't care about any other operators. + + // Fall through intentional + //case BO_AddAssign: // Common false positive + case BO_SubAssign: // Check only if unsigned + case BO_MulAssign: + case BO_DivAssign: + case BO_AndAssign: + //case BO_OrAssign: // Common false positive + //case BO_XorAssign: // Common false positive + case BO_ShlAssign: + case BO_ShrAssign: + case BO_Add: + case BO_Sub: + case BO_Mul: + case BO_Div: + case BO_And: + case BO_Or: + case BO_Xor: + case BO_Shl: + case BO_Shr: + case BO_LOr: + case BO_LAnd: + if (!LHSVal.isConstant(0) || LHSContainsFalsePositive) + break; + UpdateAssumption(A, LHSis0); + return; + } + + // If we get to this point, there has been a valid use of this operation. + A = Impossible; +} + +// At the post visit stage, the predecessor ExplodedNode will be the +// BinaryOperator that was just created. We use this hook to collect the +// ExplodedNode. +void IdempotentOperationChecker::checkPostStmt(const BinaryOperator *B, + CheckerContext &C) const { + // Add the ExplodedNode we just visited + BinaryOperatorData &Data = hash[B]; + + const Stmt *predStmt + = cast<StmtPoint>(C.getPredecessor()->getLocation()).getStmt(); + + // Ignore implicit calls to setters. + if (!isa<BinaryOperator>(predStmt)) + return; + + Data.explodedNodes.Add(C.getPredecessor()); +} + +void IdempotentOperationChecker::checkEndAnalysis(ExplodedGraph &G, + BugReporter &BR, + ExprEngine &Eng) const { + BugType *BT = new BugType("Idempotent operation", "Dead code"); + // Iterate over the hash to see if we have any paths with definite + // idempotent operations. + for (AssumptionMap::const_iterator i = hash.begin(); i != hash.end(); ++i) { + // Unpack the hash contents + const BinaryOperatorData &Data = i->second; + const Assumption &A = Data.assumption; + const ExplodedNodeSet &ES = Data.explodedNodes; + + // If there are no nodes accosted with the expression, nothing to report. + // FIXME: This is possible because the checker does part of processing in + // checkPreStmt and part in checkPostStmt. + if (ES.begin() == ES.end()) + continue; + + const BinaryOperator *B = i->first; + + if (A == Impossible) + continue; + + // If the analyzer did not finish, check to see if we can still emit this + // warning + if (Eng.hasWorkRemaining()) { + // If we can trace back + AnalysisDeclContext *AC = (*ES.begin())->getLocationContext() + ->getAnalysisDeclContext(); + if (!pathWasCompletelyAnalyzed(AC, + AC->getCFGStmtMap()->getBlock(B), + Eng.getCoreEngine())) + continue; + } + + // Select the error message and SourceRanges to report. + SmallString<128> buf; + llvm::raw_svector_ostream os(buf); + bool LHSRelevant = false, RHSRelevant = false; + switch (A) { + case Equal: + LHSRelevant = true; + RHSRelevant = true; + if (B->getOpcode() == BO_Assign) + os << "Assigned value is always the same as the existing value"; + else + os << "Both operands to '" << B->getOpcodeStr() + << "' always have the same value"; + break; + case LHSis1: + LHSRelevant = true; + os << "The left operand to '" << B->getOpcodeStr() << "' is always 1"; + break; + case RHSis1: + RHSRelevant = true; + os << "The right operand to '" << B->getOpcodeStr() << "' is always 1"; + break; + case LHSis0: + LHSRelevant = true; + os << "The left operand to '" << B->getOpcodeStr() << "' is always 0"; + break; + case RHSis0: + RHSRelevant = true; + os << "The right operand to '" << B->getOpcodeStr() << "' is always 0"; + break; + case Possible: + llvm_unreachable("Operation was never marked with an assumption"); + case Impossible: + llvm_unreachable(0); + } + + // Add a report for each ExplodedNode + for (ExplodedNodeSet::iterator I = ES.begin(), E = ES.end(); I != E; ++I) { + BugReport *report = new BugReport(*BT, os.str(), *I); + + // Add source ranges and visitor hooks + if (LHSRelevant) { + const Expr *LHS = i->first->getLHS(); + report->addRange(LHS->getSourceRange()); + FindLastStoreBRVisitor::registerStatementVarDecls(*report, LHS); + } + if (RHSRelevant) { + const Expr *RHS = i->first->getRHS(); + report->addRange(i->first->getRHS()->getSourceRange()); + FindLastStoreBRVisitor::registerStatementVarDecls(*report, RHS); + } + + BR.EmitReport(report); + } + } + + hash.clear(); +} + +// Updates the current assumption given the new assumption +inline void IdempotentOperationChecker::UpdateAssumption(Assumption &A, + const Assumption &New) { +// If the assumption is the same, there is nothing to do + if (A == New) + return; + + switch (A) { + // If we don't currently have an assumption, set it + case Possible: + A = New; + return; + + // If we have determined that a valid state happened, ignore the new + // assumption. + case Impossible: + return; + + // Any other case means that we had a different assumption last time. We don't + // currently support mixing assumptions for diagnostic reasons, so we set + // our assumption to be impossible. + default: + A = Impossible; + return; + } +} + +// Check for a statement where a variable is self assigned to possibly avoid an +// unused variable warning. +bool IdempotentOperationChecker::isSelfAssign(const Expr *LHS, const Expr *RHS) { + LHS = LHS->IgnoreParenCasts(); + RHS = RHS->IgnoreParenCasts(); + + const DeclRefExpr *LHS_DR = dyn_cast<DeclRefExpr>(LHS); + if (!LHS_DR) + return false; + + const VarDecl *VD = dyn_cast<VarDecl>(LHS_DR->getDecl()); + if (!VD) + return false; + + const DeclRefExpr *RHS_DR = dyn_cast<DeclRefExpr>(RHS); + if (!RHS_DR) + return false; + + if (VD != RHS_DR->getDecl()) + return false; + + return true; +} + +// Returns true if the Expr points to a VarDecl that is not read anywhere +// outside of self-assignments. +bool IdempotentOperationChecker::isUnused(const Expr *E, + AnalysisDeclContext *AC) { + if (!E) + return false; + + const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(E->IgnoreParenCasts()); + if (!DR) + return false; + + const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl()); + if (!VD) + return false; + + if (AC->getPseudoConstantAnalysis()->wasReferenced(VD)) + return false; + + return true; +} + +// Check for self casts truncating/extending a variable +bool IdempotentOperationChecker::isTruncationExtensionAssignment( + const Expr *LHS, + const Expr *RHS) { + + const DeclRefExpr *LHS_DR = dyn_cast<DeclRefExpr>(LHS->IgnoreParenCasts()); + if (!LHS_DR) + return false; + + const VarDecl *VD = dyn_cast<VarDecl>(LHS_DR->getDecl()); + if (!VD) + return false; + + const DeclRefExpr *RHS_DR = dyn_cast<DeclRefExpr>(RHS->IgnoreParenCasts()); + if (!RHS_DR) + return false; + + if (VD != RHS_DR->getDecl()) + return false; + + return dyn_cast<DeclRefExpr>(RHS->IgnoreParenLValueCasts()) == NULL; +} + +// Returns false if a path to this block was not completely analyzed, or true +// otherwise. +bool +IdempotentOperationChecker::pathWasCompletelyAnalyzed(AnalysisDeclContext *AC, + const CFGBlock *CB, + const CoreEngine &CE) { + + CFGReverseBlockReachabilityAnalysis *CRA = AC->getCFGReachablityAnalysis(); + + // Test for reachability from any aborted blocks to this block + typedef CoreEngine::BlocksExhausted::const_iterator ExhaustedIterator; + for (ExhaustedIterator I = CE.blocks_exhausted_begin(), + E = CE.blocks_exhausted_end(); I != E; ++I) { + const BlockEdge &BE = I->first; + + // The destination block on the BlockEdge is the first block that was not + // analyzed. If we can reach this block from the aborted block, then this + // block was not completely analyzed. + // + // Also explicitly check if the current block is the destination block. + // While technically reachable, it means we aborted the analysis on + // a path that included that block. + const CFGBlock *destBlock = BE.getDst(); + if (destBlock == CB || CRA->isReachable(destBlock, CB)) + return false; + } + + // Test for reachability from blocks we just gave up on. + typedef CoreEngine::BlocksAborted::const_iterator AbortedIterator; + for (AbortedIterator I = CE.blocks_aborted_begin(), + E = CE.blocks_aborted_end(); I != E; ++I) { + const CFGBlock *destBlock = I->first; + if (destBlock == CB || CRA->isReachable(destBlock, CB)) + return false; + } + + // For the items still on the worklist, see if they are in blocks that + // can eventually reach 'CB'. + class VisitWL : public WorkList::Visitor { + const CFGStmtMap *CBM; + const CFGBlock *TargetBlock; + CFGReverseBlockReachabilityAnalysis &CRA; + public: + VisitWL(const CFGStmtMap *cbm, const CFGBlock *targetBlock, + CFGReverseBlockReachabilityAnalysis &cra) + : CBM(cbm), TargetBlock(targetBlock), CRA(cra) {} + virtual bool visit(const WorkListUnit &U) { + ProgramPoint P = U.getNode()->getLocation(); + const CFGBlock *B = 0; + if (StmtPoint *SP = dyn_cast<StmtPoint>(&P)) { + B = CBM->getBlock(SP->getStmt()); + } + else if (BlockEdge *BE = dyn_cast<BlockEdge>(&P)) { + B = BE->getDst(); + } + else if (BlockEntrance *BEnt = dyn_cast<BlockEntrance>(&P)) { + B = BEnt->getBlock(); + } + else if (BlockExit *BExit = dyn_cast<BlockExit>(&P)) { + B = BExit->getBlock(); + } + if (!B) + return true; + + return B == TargetBlock || CRA.isReachable(B, TargetBlock); + } + }; + VisitWL visitWL(AC->getCFGStmtMap(), CB, *CRA); + // Were there any items in the worklist that could potentially reach + // this block? + if (CE.getWorkList()->visitItemsInWorkList(visitWL)) + return false; + + // Verify that this block is reachable from the entry block + if (!CRA->isReachable(&AC->getCFG()->getEntry(), CB)) + return false; + + // If we get to this point, there is no connection to the entry block or an + // aborted block. This path is unreachable and we can report the error. + return true; +} + +// Recursive function that determines whether an expression contains any element +// that varies. This could be due to a compile-time constant like sizeof. An +// expression may also involve a variable that behaves like a constant. The +// function returns true if the expression varies, and false otherwise. +bool IdempotentOperationChecker::CanVary(const Expr *Ex, + AnalysisDeclContext *AC) { + // Parentheses and casts are irrelevant here + Ex = Ex->IgnoreParenCasts(); + + if (Ex->getLocStart().isMacroID()) + return false; + + switch (Ex->getStmtClass()) { + // Trivially true cases + case Stmt::ArraySubscriptExprClass: + case Stmt::MemberExprClass: + case Stmt::StmtExprClass: + case Stmt::CallExprClass: + case Stmt::VAArgExprClass: + case Stmt::ShuffleVectorExprClass: + return true; + default: + return true; + + // Trivially false cases + case Stmt::IntegerLiteralClass: + case Stmt::CharacterLiteralClass: + case Stmt::FloatingLiteralClass: + case Stmt::PredefinedExprClass: + case Stmt::ImaginaryLiteralClass: + case Stmt::StringLiteralClass: + case Stmt::OffsetOfExprClass: + case Stmt::CompoundLiteralExprClass: + case Stmt::AddrLabelExprClass: + case Stmt::BinaryTypeTraitExprClass: + case Stmt::GNUNullExprClass: + case Stmt::InitListExprClass: + case Stmt::DesignatedInitExprClass: + case Stmt::BlockExprClass: + return false; + + // Cases requiring custom logic + case Stmt::UnaryExprOrTypeTraitExprClass: { + const UnaryExprOrTypeTraitExpr *SE = + cast<const UnaryExprOrTypeTraitExpr>(Ex); + if (SE->getKind() != UETT_SizeOf) + return false; + return SE->getTypeOfArgument()->isVariableArrayType(); + } + case Stmt::DeclRefExprClass: + // Check for constants/pseudoconstants + return !isConstantOrPseudoConstant(cast<DeclRefExpr>(Ex), AC); + + // The next cases require recursion for subexpressions + case Stmt::BinaryOperatorClass: { + const BinaryOperator *B = cast<const BinaryOperator>(Ex); + + // Exclude cases involving pointer arithmetic. These are usually + // false positives. + if (B->getOpcode() == BO_Sub || B->getOpcode() == BO_Add) + if (B->getLHS()->getType()->getAs<PointerType>()) + return false; + + return CanVary(B->getRHS(), AC) + || CanVary(B->getLHS(), AC); + } + case Stmt::UnaryOperatorClass: { + const UnaryOperator *U = cast<const UnaryOperator>(Ex); + // Handle trivial case first + switch (U->getOpcode()) { + case UO_Extension: + return false; + default: + return CanVary(U->getSubExpr(), AC); + } + } + case Stmt::ChooseExprClass: + return CanVary(cast<const ChooseExpr>(Ex)->getChosenSubExpr( + AC->getASTContext()), AC); + case Stmt::ConditionalOperatorClass: + case Stmt::BinaryConditionalOperatorClass: + return CanVary(cast<AbstractConditionalOperator>(Ex)->getCond(), AC); + } +} + +// Returns true if a DeclRefExpr is or behaves like a constant. +bool IdempotentOperationChecker::isConstantOrPseudoConstant( + const DeclRefExpr *DR, + AnalysisDeclContext *AC) { + // Check if the type of the Decl is const-qualified + if (DR->getType().isConstQualified()) + return true; + + // Check for an enum + if (isa<EnumConstantDecl>(DR->getDecl())) + return true; + + const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl()); + if (!VD) + return true; + + // Check if the Decl behaves like a constant. This check also takes care of + // static variables, which can only change between function calls if they are + // modified in the AST. + PseudoConstantAnalysis *PCA = AC->getPseudoConstantAnalysis(); + if (PCA->isPseudoConstant(VD)) + return true; + + return false; +} + +// Recursively find any substatements containing VarDecl's with storage other +// than local +bool IdempotentOperationChecker::containsNonLocalVarDecl(const Stmt *S) { + const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(S); + + if (DR) + if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) + if (!VD->hasLocalStorage()) + return true; + + for (Stmt::const_child_iterator I = S->child_begin(); I != S->child_end(); + ++I) + if (const Stmt *child = *I) + if (containsNonLocalVarDecl(child)) + return true; + + return false; +} + + +void ento::registerIdempotentOperationChecker(CheckerManager &mgr) { + mgr.registerChecker<IdempotentOperationChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/InterCheckerAPI.h b/clang/lib/StaticAnalyzer/Checkers/InterCheckerAPI.h new file mode 100644 index 0000000..e35557f --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/InterCheckerAPI.h @@ -0,0 +1,22 @@ +//==--- InterCheckerAPI.h ---------------------------------------*- C++ -*-==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// This file allows introduction of checker dependencies. It contains APIs for +// inter-checker communications. +//===----------------------------------------------------------------------===// + +#ifndef INTERCHECKERAPI_H_ +#define INTERCHECKERAPI_H_ +namespace clang { +namespace ento { + +/// Register the checker which evaluates CString API calls. +void registerCStringCheckerBasic(CheckerManager &Mgr); + +}} +#endif /* INTERCHECKERAPI_H_ */ diff --git a/clang/lib/StaticAnalyzer/Checkers/IntervalTest.cpp b/clang/lib/StaticAnalyzer/Checkers/IntervalTest.cpp new file mode 100644 index 0000000..2f3e155 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/IntervalTest.cpp @@ -0,0 +1,25 @@ +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/AnalysisManager.h" +#include "clang/Analysis/Analyses/Interval.h" +#include "clang/Analysis/CallGraph.h" +#include "llvm/Support/Process.h" + +using namespace clang; +using namespace ento; + +namespace { +class IntervalTest: public Checker<check::ASTCodeBody> { +public: + void checkASTCodeBody(const Decl *D, AnalysisManager& mgr, + BugReporter &BR) const { + if (IntervalAnalysis *a = mgr.getAnalysis<IntervalAnalysis>(D)) { + a->runOnAllBlocks(); + } + } +}; +} + +void ento::registerIntervalTestChecker(CheckerManager &mgr) { + mgr.registerChecker<IntervalTest>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/IteratorsChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/IteratorsChecker.cpp new file mode 100644 index 0000000..b0bac33 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/IteratorsChecker.cpp @@ -0,0 +1,603 @@ +//=== IteratorsChecker.cpp - Check for Invalidated Iterators ------*- C++ -*---- +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This defines IteratorsChecker, a number of small checks for conditions +// leading to invalid iterators being used. +// FIXME: Currently only supports 'vector' and 'deque' +// +//===----------------------------------------------------------------------===// + +#include "clang/AST/DeclTemplate.h" +#include "clang/Basic/SourceManager.h" +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramStateTrait.h" +#include "clang/AST/DeclCXX.h" +#include "clang/AST/ExprCXX.h" +#include "clang/AST/Type.h" +#include "clang/AST/PrettyPrinter.h" +#include "llvm/ADT/SmallPtrSet.h" +#include "llvm/ADT/StringSwitch.h" + + +using namespace clang; +using namespace ento; + +// This is the state associated with each iterator which includes both the +// kind of state and the instance used to initialize it. +// FIXME: add location where invalidated for better error reporting. +namespace { +class RefState { + enum Kind { BeginValid, EndValid, Invalid, Undefined, Unknown } K; + const void *VR; + +public: + RefState(Kind k, const void *vr) : K(k), VR(vr) {} + + bool isValid() const { return K == BeginValid || K == EndValid; } + bool isInvalid() const { return K == Invalid; } + bool isUndefined() const { return K == Undefined; } + bool isUnknown() const { return K == Unknown; } + const MemRegion *getMemRegion() const { + if (K == BeginValid || K == EndValid) + return(const MemRegion *)VR; + return 0; + } + const MemberExpr *getMemberExpr() const { + if (K == Invalid) + return(const MemberExpr *)VR; + return 0; + } + + bool operator==(const RefState &X) const { + return K == X.K && VR == X.VR; + } + + static RefState getBeginValid(const MemRegion *vr) { + assert(vr); + return RefState(BeginValid, vr); + } + static RefState getEndValid(const MemRegion *vr) { + assert(vr); + return RefState(EndValid, vr); + } + static RefState getInvalid( const MemberExpr *ME ) { + return RefState(Invalid, ME); + } + static RefState getUndefined( void ) { + return RefState(Undefined, 0); + } + static RefState getUnknown( void ) { + return RefState(Unknown, 0); + } + + void Profile(llvm::FoldingSetNodeID &ID) const { + ID.AddInteger(K); + ID.AddPointer(VR); + } +}; + +enum RefKind { NoKind, VectorKind, VectorIteratorKind }; + +class IteratorsChecker : + public Checker<check::PreStmt<CXXOperatorCallExpr>, + check::PreStmt<DeclStmt>, + check::PreStmt<CXXMemberCallExpr>, + check::PreStmt<CallExpr> > + { + // Used when parsing iterators and vectors and deques. + BuiltinBug *BT_Invalid, *BT_Undefined, *BT_Incompatible; + +public: + IteratorsChecker() : + BT_Invalid(0), BT_Undefined(0), BT_Incompatible(0) + {} + static void *getTag() { static int tag; return &tag; } + + // Checker entry points. + void checkPreStmt(const CXXOperatorCallExpr *OCE, + CheckerContext &C) const; + + void checkPreStmt(const DeclStmt *DS, + CheckerContext &C) const; + + void checkPreStmt(const CXXMemberCallExpr *MCE, + CheckerContext &C) const; + + void checkPreStmt(const CallExpr *CE, + CheckerContext &C) const; + +private: + ProgramStateRef handleAssign(ProgramStateRef state, + const Expr *lexp, + const Expr *rexp, + const LocationContext *LC) const; + + ProgramStateRef handleAssign(ProgramStateRef state, + const MemRegion *MR, + const Expr *rexp, + const LocationContext *LC) const; + + ProgramStateRef invalidateIterators(ProgramStateRef state, + const MemRegion *MR, + const MemberExpr *ME) const; + + void checkExpr(CheckerContext &C, const Expr *E) const; + + void checkArgs(CheckerContext &C, const CallExpr *CE) const; + + const MemRegion *getRegion(ProgramStateRef state, + const Expr *E, + const LocationContext *LC) const; + + const DeclRefExpr *getDeclRefExpr(const Expr *E) const; +}; + +class IteratorState { +public: + typedef llvm::ImmutableMap<const MemRegion *, RefState> EntryMap; +}; +} //end anonymous namespace + +namespace clang { + namespace ento { + template <> + struct ProgramStateTrait<IteratorState> + : public ProgramStatePartialTrait<IteratorState::EntryMap> { + static void *GDMIndex() { return IteratorsChecker::getTag(); } + }; + } +} + +void ento::registerIteratorsChecker(CheckerManager &mgr) { + mgr.registerChecker<IteratorsChecker>(); +} + +// =============================================== +// Utility functions used by visitor functions +// =============================================== + +// check a templated type for std::vector or std::deque +static RefKind getTemplateKind(const NamedDecl *td) { + const DeclContext *dc = td->getDeclContext(); + const NamespaceDecl *nameSpace = dyn_cast<NamespaceDecl>(dc); + if (!nameSpace || !isa<TranslationUnitDecl>(nameSpace->getDeclContext()) + || nameSpace->getName() != "std") + return NoKind; + + StringRef name = td->getName(); + return llvm::StringSwitch<RefKind>(name) + .Cases("vector", "deque", VectorKind) + .Default(NoKind); +} + +static RefKind getTemplateKind(const DeclContext *dc) { + if (const ClassTemplateSpecializationDecl *td = + dyn_cast<ClassTemplateSpecializationDecl>(dc)) + return getTemplateKind(cast<NamedDecl>(td)); + return NoKind; +} + +static RefKind getTemplateKind(const TypedefType *tdt) { + const TypedefNameDecl *td = tdt->getDecl(); + RefKind parentKind = getTemplateKind(td->getDeclContext()); + if (parentKind == VectorKind) { + return llvm::StringSwitch<RefKind>(td->getName()) + .Cases("iterator", + "const_iterator", + "reverse_iterator", VectorIteratorKind) + .Default(NoKind); + } + return NoKind; +} + +static RefKind getTemplateKind(const TemplateSpecializationType *tsp) { + const TemplateName &tname = tsp->getTemplateName(); + TemplateDecl *td = tname.getAsTemplateDecl(); + if (!td) + return NoKind; + return getTemplateKind(td); +} + +static RefKind getTemplateKind(QualType T) { + if (const TemplateSpecializationType *tsp = + T->getAs<TemplateSpecializationType>()) { + return getTemplateKind(tsp); + } + if (const ElaboratedType *ET = dyn_cast<ElaboratedType>(T)) { + QualType namedType = ET->getNamedType(); + if (const TypedefType *tdt = namedType->getAs<TypedefType>()) + return getTemplateKind(tdt); + if (const TemplateSpecializationType *tsp = + namedType->getAs<TemplateSpecializationType>()) { + return getTemplateKind(tsp); + } + } + return NoKind; +} + +// Iterate through our map and invalidate any iterators that were +// initialized fromt the specified instance MemRegion. +ProgramStateRef IteratorsChecker::invalidateIterators(ProgramStateRef state, + const MemRegion *MR, const MemberExpr *ME) const { + IteratorState::EntryMap Map = state->get<IteratorState>(); + if (Map.isEmpty()) + return state; + + // Loop over the entries in the current state. + // The key doesn't change, so the map iterators won't change. + for (IteratorState::EntryMap::iterator I = Map.begin(), E = Map.end(); + I != E; ++I) { + RefState RS = I.getData(); + if (RS.getMemRegion() == MR) + state = state->set<IteratorState>(I.getKey(), RefState::getInvalid(ME)); + } + + return state; +} + +// Handle assigning to an iterator where we don't have the LValue MemRegion. +ProgramStateRef IteratorsChecker::handleAssign(ProgramStateRef state, + const Expr *lexp, const Expr *rexp, const LocationContext *LC) const { + // Skip the cast if present. + if (const MaterializeTemporaryExpr *M + = dyn_cast<MaterializeTemporaryExpr>(lexp)) + lexp = M->GetTemporaryExpr(); + if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(lexp)) + lexp = ICE->getSubExpr(); + SVal sv = state->getSVal(lexp, LC); + const MemRegion *MR = sv.getAsRegion(); + if (!MR) + return state; + RefKind kind = getTemplateKind(lexp->getType()); + + // If assigning to a vector, invalidate any iterators currently associated. + if (kind == VectorKind) + return invalidateIterators(state, MR, 0); + + // Make sure that we are assigning to an iterator. + if (getTemplateKind(lexp->getType()) != VectorIteratorKind) + return state; + return handleAssign(state, MR, rexp, LC); +} + +// handle assigning to an iterator +ProgramStateRef IteratorsChecker::handleAssign(ProgramStateRef state, + const MemRegion *MR, const Expr *rexp, const LocationContext *LC) const { + // Assume unknown until we find something definite. + state = state->set<IteratorState>(MR, RefState::getUnknown()); + if (const MaterializeTemporaryExpr *M + = dyn_cast<MaterializeTemporaryExpr>(rexp)) + rexp = M->GetTemporaryExpr(); + if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(rexp)) + rexp = ICE->getSubExpr(); + // Need to handle three cases: MemberCall, copy, copy with addition. + if (const CallExpr *CE = dyn_cast<CallExpr>(rexp)) { + // Handle MemberCall. + if (const MemberExpr *ME = dyn_cast<MemberExpr>(CE->getCallee())) { + const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ME->getBase()); + if (!DRE) + return state; + // Verify that the type is std::vector<T>. + if (getTemplateKind(DRE->getType()) != VectorKind) + return state; + // Now get the MemRegion associated with the instance. + const VarDecl *VD = dyn_cast<VarDecl>(DRE->getDecl()); + if (!VD) + return state; + const MemRegion *IMR = state->getRegion(VD, LC); + if (!IMR) + return state; + // Finally, see if it is one of the calls that will create + // a valid iterator and mark it if so, else mark as Unknown. + StringRef mName = ME->getMemberDecl()->getName(); + + if (llvm::StringSwitch<bool>(mName) + .Cases("begin", "insert", "erase", true).Default(false)) { + return state->set<IteratorState>(MR, RefState::getBeginValid(IMR)); + } + if (mName == "end") + return state->set<IteratorState>(MR, RefState::getEndValid(IMR)); + + return state->set<IteratorState>(MR, RefState::getUnknown()); + } + } + // Handle straight copy from another iterator. + if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(rexp)) { + if (getTemplateKind(DRE->getType()) != VectorIteratorKind) + return state; + // Now get the MemRegion associated with the instance. + const VarDecl *VD = dyn_cast<VarDecl>(DRE->getDecl()); + if (!VD) + return state; + const MemRegion *IMR = state->getRegion(VD, LC); + if (!IMR) + return state; + // Get the RefState of the iterator being copied. + const RefState *RS = state->get<IteratorState>(IMR); + if (!RS) + return state; + // Use it to set the state of the LValue. + return state->set<IteratorState>(MR, *RS); + } + // If we have operator+ or operator- ... + if (const CXXOperatorCallExpr *OCE = dyn_cast<CXXOperatorCallExpr>(rexp)) { + OverloadedOperatorKind Kind = OCE->getOperator(); + if (Kind == OO_Plus || Kind == OO_Minus) { + // Check left side of tree for a valid value. + state = handleAssign( state, MR, OCE->getArg(0), LC); + const RefState *RS = state->get<IteratorState>(MR); + // If found, return it. + if (!RS->isUnknown()) + return state; + // Otherwise return what we find in the right side. + return handleAssign(state, MR, OCE->getArg(1), LC); + } + } + // Fall through if nothing matched. + return state; +} + +// Iterate through the arguments looking for an Invalid or Undefined iterator. +void IteratorsChecker::checkArgs(CheckerContext &C, const CallExpr *CE) const { + for (CallExpr::const_arg_iterator I = CE->arg_begin(), E = CE->arg_end(); + I != E; ++I) { + checkExpr(C, *I); + } +} + +// Get the DeclRefExpr associated with the expression. +const DeclRefExpr *IteratorsChecker::getDeclRefExpr(const Expr *E) const { + // If it is a CXXConstructExpr, need to get the subexpression. + if (const CXXConstructExpr *CE = dyn_cast<CXXConstructExpr>(E)) { + if (CE->getNumArgs()== 1) { + CXXConstructorDecl *CD = CE->getConstructor(); + if (CD->isTrivial()) + E = CE->getArg(0); + } + } + if (const MaterializeTemporaryExpr *M = dyn_cast<MaterializeTemporaryExpr>(E)) + E = M->GetTemporaryExpr(); + if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) + E = ICE->getSubExpr(); + // If it isn't one of our types, don't do anything. + if (getTemplateKind(E->getType()) != VectorIteratorKind) + return NULL; + return dyn_cast<DeclRefExpr>(E); +} + +// Get the MemRegion associated with the expresssion. +const MemRegion *IteratorsChecker::getRegion(ProgramStateRef state, + const Expr *E, const LocationContext *LC) const { + const DeclRefExpr *DRE = getDeclRefExpr(E); + if (!DRE) + return NULL; + const VarDecl *VD = dyn_cast<VarDecl>(DRE->getDecl()); + if (!VD) + return NULL; + // return the MemRegion associated with the iterator + return state->getRegion(VD, LC); +} + +// Check the expression and if it is an iterator, generate a diagnostic +// if the iterator is not valid. +// FIXME: this method can generate new nodes, and subsequent logic should +// use those nodes. We also cannot create multiple nodes at one ProgramPoint +// with the same tag. +void IteratorsChecker::checkExpr(CheckerContext &C, const Expr *E) const { + ProgramStateRef state = C.getState(); + const MemRegion *MR = getRegion(state, E, C.getLocationContext()); + if (!MR) + return; + + // Get the state associated with the iterator. + const RefState *RS = state->get<IteratorState>(MR); + if (!RS) + return; + if (RS->isInvalid()) { + if (ExplodedNode *N = C.addTransition()) { + if (!BT_Invalid) + // FIXME: We are eluding constness here. + const_cast<IteratorsChecker*>(this)->BT_Invalid = new BuiltinBug(""); + + std::string msg; + const MemberExpr *ME = RS->getMemberExpr(); + if (ME) { + std::string name = ME->getMemberNameInfo().getAsString(); + msg = "Attempt to use an iterator made invalid by call to '" + + name + "'"; + } + else { + msg = "Attempt to use an iterator made invalid by copying another " + "container to its container"; + } + + BugReport *R = new BugReport(*BT_Invalid, msg, N); + R->addRange(getDeclRefExpr(E)->getSourceRange()); + C.EmitReport(R); + } + } + else if (RS->isUndefined()) { + if (ExplodedNode *N = C.addTransition()) { + if (!BT_Undefined) + // FIXME: We are eluding constness here. + const_cast<IteratorsChecker*>(this)->BT_Undefined = + new BuiltinBug("Use of iterator that is not defined"); + + BugReport *R = new BugReport(*BT_Undefined, + BT_Undefined->getDescription(), N); + R->addRange(getDeclRefExpr(E)->getSourceRange()); + C.EmitReport(R); + } + } +} + +// =============================================== +// Path analysis visitor functions +// =============================================== + +// For a generic Call, just check the args for bad iterators. +void IteratorsChecker::checkPreStmt(const CallExpr *CE, + CheckerContext &C) const{ + + // FIXME: These checks are to currently work around a bug + // in CheckerManager. + if (isa<CXXOperatorCallExpr>(CE)) + return; + if (isa<CXXMemberCallExpr>(CE)) + return; + + checkArgs(C, CE); +} + +// Handle operator calls. First, if it is operator=, check the argument, +// and handle assigning and set target state appropriately. Otherwise, for +// other operators, check the args for bad iterators and handle comparisons. +void IteratorsChecker::checkPreStmt(const CXXOperatorCallExpr *OCE, + CheckerContext &C) const +{ + const LocationContext *LC = C.getLocationContext(); + ProgramStateRef state = C.getState(); + OverloadedOperatorKind Kind = OCE->getOperator(); + if (Kind == OO_Equal) { + checkExpr(C, OCE->getArg(1)); + state = handleAssign(state, OCE->getArg(0), OCE->getArg(1), LC); + C.addTransition(state); + return; + } + else { + checkArgs(C, OCE); + // If it is a compare and both are iterators, ensure that they are for + // the same container. + if (Kind == OO_EqualEqual || Kind == OO_ExclaimEqual || + Kind == OO_Less || Kind == OO_LessEqual || + Kind == OO_Greater || Kind == OO_GreaterEqual) { + const MemRegion *MR0, *MR1; + MR0 = getRegion(state, OCE->getArg(0), LC); + if (!MR0) + return; + MR1 = getRegion(state, OCE->getArg(1), LC); + if (!MR1) + return; + const RefState *RS0, *RS1; + RS0 = state->get<IteratorState>(MR0); + if (!RS0) + return; + RS1 = state->get<IteratorState>(MR1); + if (!RS1) + return; + if (RS0->getMemRegion() != RS1->getMemRegion()) { + if (ExplodedNode *N = C.addTransition()) { + if (!BT_Incompatible) + const_cast<IteratorsChecker*>(this)->BT_Incompatible = + new BuiltinBug( + "Cannot compare iterators from different containers"); + + BugReport *R = new BugReport(*BT_Incompatible, + BT_Incompatible->getDescription(), N); + R->addRange(OCE->getSourceRange()); + C.EmitReport(R); + } + } + } + } +} + +// Need to handle DeclStmts to pick up initializing of iterators and to mark +// uninitialized ones as Undefined. +void IteratorsChecker::checkPreStmt(const DeclStmt *DS, + CheckerContext &C) const { + const Decl *D = *DS->decl_begin(); + const VarDecl *VD = dyn_cast<VarDecl>(D); + // Only care about iterators. + if (getTemplateKind(VD->getType()) != VectorIteratorKind) + return; + + // Get the MemRegion associated with the iterator and mark it as Undefined. + ProgramStateRef state = C.getState(); + Loc VarLoc = state->getLValue(VD, C.getLocationContext()); + const MemRegion *MR = VarLoc.getAsRegion(); + if (!MR) + return; + state = state->set<IteratorState>(MR, RefState::getUndefined()); + + // if there is an initializer, handle marking Valid if a proper initializer + const Expr *InitEx = VD->getInit(); + if (InitEx) { + // FIXME: This is too syntactic. Since 'InitEx' will be analyzed first + // it should resolve to an SVal that we can check for validity + // *semantically* instead of walking through the AST. + if (const CXXConstructExpr *CE = dyn_cast<CXXConstructExpr>(InitEx)) { + if (CE->getNumArgs() == 1) { + const Expr *E = CE->getArg(0); + if (const MaterializeTemporaryExpr *M + = dyn_cast<MaterializeTemporaryExpr>(E)) + E = M->GetTemporaryExpr(); + if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) + InitEx = ICE->getSubExpr(); + state = handleAssign(state, MR, InitEx, C.getLocationContext()); + } + } + } + C.addTransition(state); +} + + +namespace { struct CalledReserved {}; } +namespace clang { namespace ento { +template<> struct ProgramStateTrait<CalledReserved> + : public ProgramStatePartialTrait<llvm::ImmutableSet<const MemRegion*> > { + static void *GDMIndex() { static int index = 0; return &index; } +}; +}} + +// on a member call, first check the args for any bad iterators +// then, check to see if it is a call to a function that will invalidate +// the iterators +void IteratorsChecker::checkPreStmt(const CXXMemberCallExpr *MCE, + CheckerContext &C) const { + // Check the arguments. + checkArgs(C, MCE); + const MemberExpr *ME = dyn_cast<MemberExpr>(MCE->getCallee()); + if (!ME) + return; + // Make sure we have the right kind of container. + const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ME->getBase()); + if (!DRE || getTemplateKind(DRE->getType()) != VectorKind) + return; + SVal tsv = C.getState()->getSVal(DRE, C.getLocationContext()); + // Get the MemRegion associated with the container instance. + const MemRegion *MR = tsv.getAsRegion(); + if (!MR) + return; + // If we are calling a function that invalidates iterators, mark them + // appropriately by finding matching instances. + ProgramStateRef state = C.getState(); + StringRef mName = ME->getMemberDecl()->getName(); + if (llvm::StringSwitch<bool>(mName) + .Cases("insert", "reserve", "push_back", true) + .Cases("erase", "pop_back", "clear", "resize", true) + .Default(false)) { + // If there was a 'reserve' call, assume iterators are good. + if (!state->contains<CalledReserved>(MR)) + state = invalidateIterators(state, MR, ME); + } + // Keep track of instances that have called 'reserve' + // note: do this after we invalidate any iterators by calling + // 'reserve' itself. + if (mName == "reserve") + state = state->add<CalledReserved>(MR); + + if (state != C.getState()) + C.addTransition(state); +} + diff --git a/clang/lib/StaticAnalyzer/Checkers/LLVMConventionsChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/LLVMConventionsChecker.cpp new file mode 100644 index 0000000..757a4ce --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/LLVMConventionsChecker.cpp @@ -0,0 +1,314 @@ +//=== LLVMConventionsChecker.cpp - Check LLVM codebase conventions ---*- C++ -*- +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This defines LLVMConventionsChecker, a bunch of small little checks +// for checking specific coding conventions in the LLVM/Clang codebase. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugReporter.h" +#include "clang/AST/DeclTemplate.h" +#include "clang/AST/StmtVisitor.h" +#include "llvm/ADT/SmallString.h" + +using namespace clang; +using namespace ento; + +//===----------------------------------------------------------------------===// +// Generic type checking routines. +//===----------------------------------------------------------------------===// + +static bool IsLLVMStringRef(QualType T) { + const RecordType *RT = T->getAs<RecordType>(); + if (!RT) + return false; + + return StringRef(QualType(RT, 0).getAsString()) == + "class StringRef"; +} + +/// Check whether the declaration is semantically inside the top-level +/// namespace named by ns. +static bool InNamespace(const Decl *D, StringRef NS) { + const NamespaceDecl *ND = dyn_cast<NamespaceDecl>(D->getDeclContext()); + if (!ND) + return false; + const IdentifierInfo *II = ND->getIdentifier(); + if (!II || !II->getName().equals(NS)) + return false; + return isa<TranslationUnitDecl>(ND->getDeclContext()); +} + +static bool IsStdString(QualType T) { + if (const ElaboratedType *QT = T->getAs<ElaboratedType>()) + T = QT->getNamedType(); + + const TypedefType *TT = T->getAs<TypedefType>(); + if (!TT) + return false; + + const TypedefNameDecl *TD = TT->getDecl(); + + if (!InNamespace(TD, "std")) + return false; + + return TD->getName() == "string"; +} + +static bool IsClangType(const RecordDecl *RD) { + return RD->getName() == "Type" && InNamespace(RD, "clang"); +} + +static bool IsClangDecl(const RecordDecl *RD) { + return RD->getName() == "Decl" && InNamespace(RD, "clang"); +} + +static bool IsClangStmt(const RecordDecl *RD) { + return RD->getName() == "Stmt" && InNamespace(RD, "clang"); +} + +static bool IsClangAttr(const RecordDecl *RD) { + return RD->getName() == "Attr" && InNamespace(RD, "clang"); +} + +static bool IsStdVector(QualType T) { + const TemplateSpecializationType *TS = T->getAs<TemplateSpecializationType>(); + if (!TS) + return false; + + TemplateName TM = TS->getTemplateName(); + TemplateDecl *TD = TM.getAsTemplateDecl(); + + if (!TD || !InNamespace(TD, "std")) + return false; + + return TD->getName() == "vector"; +} + +static bool IsSmallVector(QualType T) { + const TemplateSpecializationType *TS = T->getAs<TemplateSpecializationType>(); + if (!TS) + return false; + + TemplateName TM = TS->getTemplateName(); + TemplateDecl *TD = TM.getAsTemplateDecl(); + + if (!TD || !InNamespace(TD, "llvm")) + return false; + + return TD->getName() == "SmallVector"; +} + +//===----------------------------------------------------------------------===// +// CHECK: a StringRef should not be bound to a temporary std::string whose +// lifetime is shorter than the StringRef's. +//===----------------------------------------------------------------------===// + +namespace { +class StringRefCheckerVisitor : public StmtVisitor<StringRefCheckerVisitor> { + BugReporter &BR; + const Decl *DeclWithIssue; +public: + StringRefCheckerVisitor(const Decl *declWithIssue, BugReporter &br) + : BR(br), DeclWithIssue(declWithIssue) {} + void VisitChildren(Stmt *S) { + for (Stmt::child_iterator I = S->child_begin(), E = S->child_end() ; + I != E; ++I) + if (Stmt *child = *I) + Visit(child); + } + void VisitStmt(Stmt *S) { VisitChildren(S); } + void VisitDeclStmt(DeclStmt *DS); +private: + void VisitVarDecl(VarDecl *VD); +}; +} // end anonymous namespace + +static void CheckStringRefAssignedTemporary(const Decl *D, BugReporter &BR) { + StringRefCheckerVisitor walker(D, BR); + walker.Visit(D->getBody()); +} + +void StringRefCheckerVisitor::VisitDeclStmt(DeclStmt *S) { + VisitChildren(S); + + for (DeclStmt::decl_iterator I = S->decl_begin(), E = S->decl_end();I!=E; ++I) + if (VarDecl *VD = dyn_cast<VarDecl>(*I)) + VisitVarDecl(VD); +} + +void StringRefCheckerVisitor::VisitVarDecl(VarDecl *VD) { + Expr *Init = VD->getInit(); + if (!Init) + return; + + // Pattern match for: + // StringRef x = call() (where call returns std::string) + if (!IsLLVMStringRef(VD->getType())) + return; + ExprWithCleanups *Ex1 = dyn_cast<ExprWithCleanups>(Init); + if (!Ex1) + return; + CXXConstructExpr *Ex2 = dyn_cast<CXXConstructExpr>(Ex1->getSubExpr()); + if (!Ex2 || Ex2->getNumArgs() != 1) + return; + ImplicitCastExpr *Ex3 = dyn_cast<ImplicitCastExpr>(Ex2->getArg(0)); + if (!Ex3) + return; + CXXConstructExpr *Ex4 = dyn_cast<CXXConstructExpr>(Ex3->getSubExpr()); + if (!Ex4 || Ex4->getNumArgs() != 1) + return; + ImplicitCastExpr *Ex5 = dyn_cast<ImplicitCastExpr>(Ex4->getArg(0)); + if (!Ex5) + return; + CXXBindTemporaryExpr *Ex6 = dyn_cast<CXXBindTemporaryExpr>(Ex5->getSubExpr()); + if (!Ex6 || !IsStdString(Ex6->getType())) + return; + + // Okay, badness! Report an error. + const char *desc = "StringRef should not be bound to temporary " + "std::string that it outlives"; + PathDiagnosticLocation VDLoc = + PathDiagnosticLocation::createBegin(VD, BR.getSourceManager()); + BR.EmitBasicReport(DeclWithIssue, desc, "LLVM Conventions", desc, + VDLoc, Init->getSourceRange()); +} + +//===----------------------------------------------------------------------===// +// CHECK: Clang AST nodes should not have fields that can allocate +// memory. +//===----------------------------------------------------------------------===// + +static bool AllocatesMemory(QualType T) { + return IsStdVector(T) || IsStdString(T) || IsSmallVector(T); +} + +// This type checking could be sped up via dynamic programming. +static bool IsPartOfAST(const CXXRecordDecl *R) { + if (IsClangStmt(R) || IsClangType(R) || IsClangDecl(R) || IsClangAttr(R)) + return true; + + for (CXXRecordDecl::base_class_const_iterator I = R->bases_begin(), + E = R->bases_end(); I!=E; ++I) { + CXXBaseSpecifier BS = *I; + QualType T = BS.getType(); + if (const RecordType *baseT = T->getAs<RecordType>()) { + CXXRecordDecl *baseD = cast<CXXRecordDecl>(baseT->getDecl()); + if (IsPartOfAST(baseD)) + return true; + } + } + + return false; +} + +namespace { +class ASTFieldVisitor { + SmallVector<FieldDecl*, 10> FieldChain; + const CXXRecordDecl *Root; + BugReporter &BR; +public: + ASTFieldVisitor(const CXXRecordDecl *root, BugReporter &br) + : Root(root), BR(br) {} + + void Visit(FieldDecl *D); + void ReportError(QualType T); +}; +} // end anonymous namespace + +static void CheckASTMemory(const CXXRecordDecl *R, BugReporter &BR) { + if (!IsPartOfAST(R)) + return; + + for (RecordDecl::field_iterator I = R->field_begin(), E = R->field_end(); + I != E; ++I) { + ASTFieldVisitor walker(R, BR); + walker.Visit(*I); + } +} + +void ASTFieldVisitor::Visit(FieldDecl *D) { + FieldChain.push_back(D); + + QualType T = D->getType(); + + if (AllocatesMemory(T)) + ReportError(T); + + if (const RecordType *RT = T->getAs<RecordType>()) { + const RecordDecl *RD = RT->getDecl()->getDefinition(); + for (RecordDecl::field_iterator I = RD->field_begin(), E = RD->field_end(); + I != E; ++I) + Visit(*I); + } + + FieldChain.pop_back(); +} + +void ASTFieldVisitor::ReportError(QualType T) { + SmallString<1024> buf; + llvm::raw_svector_ostream os(buf); + + os << "AST class '" << Root->getName() << "' has a field '" + << FieldChain.front()->getName() << "' that allocates heap memory"; + if (FieldChain.size() > 1) { + os << " via the following chain: "; + bool isFirst = true; + for (SmallVectorImpl<FieldDecl*>::iterator I=FieldChain.begin(), + E=FieldChain.end(); I!=E; ++I) { + if (!isFirst) + os << '.'; + else + isFirst = false; + os << (*I)->getName(); + } + } + os << " (type " << FieldChain.back()->getType().getAsString() << ")"; + os.flush(); + + // Note that this will fire for every translation unit that uses this + // class. This is suboptimal, but at least scan-build will merge + // duplicate HTML reports. In the future we need a unified way of merging + // duplicate reports across translation units. For C++ classes we cannot + // just report warnings when we see an out-of-line method definition for a + // class, as that heuristic doesn't always work (the complete definition of + // the class may be in the header file, for example). + PathDiagnosticLocation L = PathDiagnosticLocation::createBegin( + FieldChain.front(), BR.getSourceManager()); + BR.EmitBasicReport(Root, "AST node allocates heap memory", "LLVM Conventions", + os.str(), L); +} + +//===----------------------------------------------------------------------===// +// LLVMConventionsChecker +//===----------------------------------------------------------------------===// + +namespace { +class LLVMConventionsChecker : public Checker< + check::ASTDecl<CXXRecordDecl>, + check::ASTCodeBody > { +public: + void checkASTDecl(const CXXRecordDecl *R, AnalysisManager& mgr, + BugReporter &BR) const { + if (R->isCompleteDefinition()) + CheckASTMemory(R, BR); + } + + void checkASTCodeBody(const Decl *D, AnalysisManager& mgr, + BugReporter &BR) const { + CheckStringRefAssignedTemporary(D, BR); + } +}; +} + +void ento::registerLLVMConventionsChecker(CheckerManager &mgr) { + mgr.registerChecker<LLVMConventionsChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/MacOSKeychainAPIChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/MacOSKeychainAPIChecker.cpp new file mode 100644 index 0000000..cb976e0 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/MacOSKeychainAPIChecker.cpp @@ -0,0 +1,681 @@ +//==--- MacOSKeychainAPIChecker.cpp ------------------------------*- C++ -*-==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// This checker flags misuses of KeyChainAPI. In particular, the password data +// allocated/returned by SecKeychainItemCopyContent, +// SecKeychainFindGenericPassword, SecKeychainFindInternetPassword functions has +// to be freed using a call to SecKeychainItemFreeContent. +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramState.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramStateTrait.h" +#include "llvm/ADT/SmallString.h" + +using namespace clang; +using namespace ento; + +namespace { +class MacOSKeychainAPIChecker : public Checker<check::PreStmt<CallExpr>, + check::PreStmt<ReturnStmt>, + check::PostStmt<CallExpr>, + check::EndPath, + check::DeadSymbols> { + mutable OwningPtr<BugType> BT; + +public: + /// AllocationState is a part of the checker specific state together with the + /// MemRegion corresponding to the allocated data. + struct AllocationState { + /// The index of the allocator function. + unsigned int AllocatorIdx; + SymbolRef Region; + + AllocationState(const Expr *E, unsigned int Idx, SymbolRef R) : + AllocatorIdx(Idx), + Region(R) {} + + bool operator==(const AllocationState &X) const { + return (AllocatorIdx == X.AllocatorIdx && + Region == X.Region); + } + + void Profile(llvm::FoldingSetNodeID &ID) const { + ID.AddInteger(AllocatorIdx); + ID.AddPointer(Region); + } + }; + + void checkPreStmt(const CallExpr *S, CheckerContext &C) const; + void checkPreStmt(const ReturnStmt *S, CheckerContext &C) const; + void checkPostStmt(const CallExpr *S, CheckerContext &C) const; + void checkDeadSymbols(SymbolReaper &SR, CheckerContext &C) const; + void checkEndPath(CheckerContext &C) const; + +private: + typedef std::pair<SymbolRef, const AllocationState*> AllocationPair; + typedef llvm::SmallVector<AllocationPair, 2> AllocationPairVec; + + enum APIKind { + /// Denotes functions tracked by this checker. + ValidAPI = 0, + /// The functions commonly/mistakenly used in place of the given API. + ErrorAPI = 1, + /// The functions which may allocate the data. These are tracked to reduce + /// the false alarm rate. + PossibleAPI = 2 + }; + /// Stores the information about the allocator and deallocator functions - + /// these are the functions the checker is tracking. + struct ADFunctionInfo { + const char* Name; + unsigned int Param; + unsigned int DeallocatorIdx; + APIKind Kind; + }; + static const unsigned InvalidIdx = 100000; + static const unsigned FunctionsToTrackSize = 8; + static const ADFunctionInfo FunctionsToTrack[FunctionsToTrackSize]; + /// The value, which represents no error return value for allocator functions. + static const unsigned NoErr = 0; + + /// Given the function name, returns the index of the allocator/deallocator + /// function. + static unsigned getTrackedFunctionIndex(StringRef Name, bool IsAllocator); + + inline void initBugType() const { + if (!BT) + BT.reset(new BugType("Improper use of SecKeychain API", "Mac OS API")); + } + + void generateDeallocatorMismatchReport(const AllocationPair &AP, + const Expr *ArgExpr, + CheckerContext &C) const; + + /// Find the allocation site for Sym on the path leading to the node N. + const Stmt *getAllocationSite(const ExplodedNode *N, SymbolRef Sym, + CheckerContext &C) const; + + BugReport *generateAllocatedDataNotReleasedReport(const AllocationPair &AP, + ExplodedNode *N, + CheckerContext &C) const; + + /// Check if RetSym evaluates to an error value in the current state. + bool definitelyReturnedError(SymbolRef RetSym, + ProgramStateRef State, + SValBuilder &Builder, + bool noError = false) const; + + /// Check if RetSym evaluates to a NoErr value in the current state. + bool definitelyDidnotReturnError(SymbolRef RetSym, + ProgramStateRef State, + SValBuilder &Builder) const { + return definitelyReturnedError(RetSym, State, Builder, true); + } + + /// Mark an AllocationPair interesting for diagnostic reporting. + void markInteresting(BugReport *R, const AllocationPair &AP) const { + R->markInteresting(AP.first); + R->markInteresting(AP.second->Region); + } + + /// The bug visitor which allows us to print extra diagnostics along the + /// BugReport path. For example, showing the allocation site of the leaked + /// region. + class SecKeychainBugVisitor + : public BugReporterVisitorImpl<SecKeychainBugVisitor> { + protected: + // The allocated region symbol tracked by the main analysis. + SymbolRef Sym; + + public: + SecKeychainBugVisitor(SymbolRef S) : Sym(S) {} + virtual ~SecKeychainBugVisitor() {} + + void Profile(llvm::FoldingSetNodeID &ID) const { + static int X = 0; + ID.AddPointer(&X); + ID.AddPointer(Sym); + } + + PathDiagnosticPiece *VisitNode(const ExplodedNode *N, + const ExplodedNode *PrevN, + BugReporterContext &BRC, + BugReport &BR); + }; +}; +} + +/// ProgramState traits to store the currently allocated (and not yet freed) +/// symbols. This is a map from the allocated content symbol to the +/// corresponding AllocationState. +typedef llvm::ImmutableMap<SymbolRef, + MacOSKeychainAPIChecker::AllocationState> AllocatedSetTy; + +namespace { struct AllocatedData {}; } +namespace clang { namespace ento { +template<> struct ProgramStateTrait<AllocatedData> + : public ProgramStatePartialTrait<AllocatedSetTy > { + static void *GDMIndex() { static int index = 0; return &index; } +}; +}} + +static bool isEnclosingFunctionParam(const Expr *E) { + E = E->IgnoreParenCasts(); + if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) { + const ValueDecl *VD = DRE->getDecl(); + if (isa<ImplicitParamDecl>(VD) || isa<ParmVarDecl>(VD)) + return true; + } + return false; +} + +const MacOSKeychainAPIChecker::ADFunctionInfo + MacOSKeychainAPIChecker::FunctionsToTrack[FunctionsToTrackSize] = { + {"SecKeychainItemCopyContent", 4, 3, ValidAPI}, // 0 + {"SecKeychainFindGenericPassword", 6, 3, ValidAPI}, // 1 + {"SecKeychainFindInternetPassword", 13, 3, ValidAPI}, // 2 + {"SecKeychainItemFreeContent", 1, InvalidIdx, ValidAPI}, // 3 + {"SecKeychainItemCopyAttributesAndData", 5, 5, ValidAPI}, // 4 + {"SecKeychainItemFreeAttributesAndData", 1, InvalidIdx, ValidAPI}, // 5 + {"free", 0, InvalidIdx, ErrorAPI}, // 6 + {"CFStringCreateWithBytesNoCopy", 1, InvalidIdx, PossibleAPI}, // 7 +}; + +unsigned MacOSKeychainAPIChecker::getTrackedFunctionIndex(StringRef Name, + bool IsAllocator) { + for (unsigned I = 0; I < FunctionsToTrackSize; ++I) { + ADFunctionInfo FI = FunctionsToTrack[I]; + if (FI.Name != Name) + continue; + // Make sure the function is of the right type (allocator vs deallocator). + if (IsAllocator && (FI.DeallocatorIdx == InvalidIdx)) + return InvalidIdx; + if (!IsAllocator && (FI.DeallocatorIdx != InvalidIdx)) + return InvalidIdx; + + return I; + } + // The function is not tracked. + return InvalidIdx; +} + +static bool isBadDeallocationArgument(const MemRegion *Arg) { + if (!Arg) + return false; + if (isa<AllocaRegion>(Arg) || + isa<BlockDataRegion>(Arg) || + isa<TypedRegion>(Arg)) { + return true; + } + return false; +} + +/// Given the address expression, retrieve the value it's pointing to. Assume +/// that value is itself an address, and return the corresponding symbol. +static SymbolRef getAsPointeeSymbol(const Expr *Expr, + CheckerContext &C) { + ProgramStateRef State = C.getState(); + SVal ArgV = State->getSVal(Expr, C.getLocationContext()); + + if (const loc::MemRegionVal *X = dyn_cast<loc::MemRegionVal>(&ArgV)) { + StoreManager& SM = C.getStoreManager(); + SymbolRef sym = SM.getBinding(State->getStore(), *X).getAsLocSymbol(); + if (sym) + return sym; + } + return 0; +} + +// When checking for error code, we need to consider the following cases: +// 1) noErr / [0] +// 2) someErr / [1, inf] +// 3) unknown +// If noError, returns true iff (1). +// If !noError, returns true iff (2). +bool MacOSKeychainAPIChecker::definitelyReturnedError(SymbolRef RetSym, + ProgramStateRef State, + SValBuilder &Builder, + bool noError) const { + DefinedOrUnknownSVal NoErrVal = Builder.makeIntVal(NoErr, + Builder.getSymbolManager().getType(RetSym)); + DefinedOrUnknownSVal NoErr = Builder.evalEQ(State, NoErrVal, + nonloc::SymbolVal(RetSym)); + ProgramStateRef ErrState = State->assume(NoErr, noError); + if (ErrState == State) { + return true; + } + + return false; +} + +// Report deallocator mismatch. Remove the region from tracking - reporting a +// missing free error after this one is redundant. +void MacOSKeychainAPIChecker:: + generateDeallocatorMismatchReport(const AllocationPair &AP, + const Expr *ArgExpr, + CheckerContext &C) const { + ProgramStateRef State = C.getState(); + State = State->remove<AllocatedData>(AP.first); + ExplodedNode *N = C.addTransition(State); + + if (!N) + return; + initBugType(); + SmallString<80> sbuf; + llvm::raw_svector_ostream os(sbuf); + unsigned int PDeallocIdx = + FunctionsToTrack[AP.second->AllocatorIdx].DeallocatorIdx; + + os << "Deallocator doesn't match the allocator: '" + << FunctionsToTrack[PDeallocIdx].Name << "' should be used."; + BugReport *Report = new BugReport(*BT, os.str(), N); + Report->addVisitor(new SecKeychainBugVisitor(AP.first)); + Report->addRange(ArgExpr->getSourceRange()); + markInteresting(Report, AP); + C.EmitReport(Report); +} + +void MacOSKeychainAPIChecker::checkPreStmt(const CallExpr *CE, + CheckerContext &C) const { + unsigned idx = InvalidIdx; + ProgramStateRef State = C.getState(); + + StringRef funName = C.getCalleeName(CE); + if (funName.empty()) + return; + + // If it is a call to an allocator function, it could be a double allocation. + idx = getTrackedFunctionIndex(funName, true); + if (idx != InvalidIdx) { + const Expr *ArgExpr = CE->getArg(FunctionsToTrack[idx].Param); + if (SymbolRef V = getAsPointeeSymbol(ArgExpr, C)) + if (const AllocationState *AS = State->get<AllocatedData>(V)) { + if (!definitelyReturnedError(AS->Region, State, C.getSValBuilder())) { + // Remove the value from the state. The new symbol will be added for + // tracking when the second allocator is processed in checkPostStmt(). + State = State->remove<AllocatedData>(V); + ExplodedNode *N = C.addTransition(State); + if (!N) + return; + initBugType(); + SmallString<128> sbuf; + llvm::raw_svector_ostream os(sbuf); + unsigned int DIdx = FunctionsToTrack[AS->AllocatorIdx].DeallocatorIdx; + os << "Allocated data should be released before another call to " + << "the allocator: missing a call to '" + << FunctionsToTrack[DIdx].Name + << "'."; + BugReport *Report = new BugReport(*BT, os.str(), N); + Report->addVisitor(new SecKeychainBugVisitor(V)); + Report->addRange(ArgExpr->getSourceRange()); + Report->markInteresting(AS->Region); + C.EmitReport(Report); + } + } + return; + } + + // Is it a call to one of deallocator functions? + idx = getTrackedFunctionIndex(funName, false); + if (idx == InvalidIdx) + return; + + // Check the argument to the deallocator. + const Expr *ArgExpr = CE->getArg(FunctionsToTrack[idx].Param); + SVal ArgSVal = State->getSVal(ArgExpr, C.getLocationContext()); + + // Undef is reported by another checker. + if (ArgSVal.isUndef()) + return; + + SymbolRef ArgSM = ArgSVal.getAsLocSymbol(); + + // If the argument is coming from the heap, globals, or unknown, do not + // report it. + bool RegionArgIsBad = false; + if (!ArgSM) { + if (!isBadDeallocationArgument(ArgSVal.getAsRegion())) + return; + RegionArgIsBad = true; + } + + // Is the argument to the call being tracked? + const AllocationState *AS = State->get<AllocatedData>(ArgSM); + if (!AS && FunctionsToTrack[idx].Kind != ValidAPI) { + return; + } + // If trying to free data which has not been allocated yet, report as a bug. + // TODO: We might want a more precise diagnostic for double free + // (that would involve tracking all the freed symbols in the checker state). + if (!AS || RegionArgIsBad) { + // It is possible that this is a false positive - the argument might + // have entered as an enclosing function parameter. + if (isEnclosingFunctionParam(ArgExpr)) + return; + + ExplodedNode *N = C.addTransition(State); + if (!N) + return; + initBugType(); + BugReport *Report = new BugReport(*BT, + "Trying to free data which has not been allocated.", N); + Report->addRange(ArgExpr->getSourceRange()); + if (AS) + Report->markInteresting(AS->Region); + C.EmitReport(Report); + return; + } + + // Process functions which might deallocate. + if (FunctionsToTrack[idx].Kind == PossibleAPI) { + + if (funName == "CFStringCreateWithBytesNoCopy") { + const Expr *DeallocatorExpr = CE->getArg(5)->IgnoreParenCasts(); + // NULL ~ default deallocator, so warn. + if (DeallocatorExpr->isNullPointerConstant(C.getASTContext(), + Expr::NPC_ValueDependentIsNotNull)) { + const AllocationPair AP = std::make_pair(ArgSM, AS); + generateDeallocatorMismatchReport(AP, ArgExpr, C); + return; + } + // One of the default allocators, so warn. + if (const DeclRefExpr *DE = dyn_cast<DeclRefExpr>(DeallocatorExpr)) { + StringRef DeallocatorName = DE->getFoundDecl()->getName(); + if (DeallocatorName == "kCFAllocatorDefault" || + DeallocatorName == "kCFAllocatorSystemDefault" || + DeallocatorName == "kCFAllocatorMalloc") { + const AllocationPair AP = std::make_pair(ArgSM, AS); + generateDeallocatorMismatchReport(AP, ArgExpr, C); + return; + } + // If kCFAllocatorNull, which does not deallocate, we still have to + // find the deallocator. Otherwise, assume that the user had written a + // custom deallocator which does the right thing. + if (DE->getFoundDecl()->getName() != "kCFAllocatorNull") { + State = State->remove<AllocatedData>(ArgSM); + C.addTransition(State); + return; + } + } + } + return; + } + + // The call is deallocating a value we previously allocated, so remove it + // from the next state. + State = State->remove<AllocatedData>(ArgSM); + + // Check if the proper deallocator is used. + unsigned int PDeallocIdx = FunctionsToTrack[AS->AllocatorIdx].DeallocatorIdx; + if (PDeallocIdx != idx || (FunctionsToTrack[idx].Kind == ErrorAPI)) { + const AllocationPair AP = std::make_pair(ArgSM, AS); + generateDeallocatorMismatchReport(AP, ArgExpr, C); + return; + } + + // If the buffer can be null and the return status can be an error, + // report a bad call to free. + if (State->assume(cast<DefinedSVal>(ArgSVal), false) && + !definitelyDidnotReturnError(AS->Region, State, C.getSValBuilder())) { + ExplodedNode *N = C.addTransition(State); + if (!N) + return; + initBugType(); + BugReport *Report = new BugReport(*BT, + "Only call free if a valid (non-NULL) buffer was returned.", N); + Report->addVisitor(new SecKeychainBugVisitor(ArgSM)); + Report->addRange(ArgExpr->getSourceRange()); + Report->markInteresting(AS->Region); + C.EmitReport(Report); + return; + } + + C.addTransition(State); +} + +void MacOSKeychainAPIChecker::checkPostStmt(const CallExpr *CE, + CheckerContext &C) const { + ProgramStateRef State = C.getState(); + StringRef funName = C.getCalleeName(CE); + + // If a value has been allocated, add it to the set for tracking. + unsigned idx = getTrackedFunctionIndex(funName, true); + if (idx == InvalidIdx) + return; + + const Expr *ArgExpr = CE->getArg(FunctionsToTrack[idx].Param); + // If the argument entered as an enclosing function parameter, skip it to + // avoid false positives. + if (isEnclosingFunctionParam(ArgExpr) && + C.getLocationContext()->getParent() == 0) + return; + + if (SymbolRef V = getAsPointeeSymbol(ArgExpr, C)) { + // If the argument points to something that's not a symbolic region, it + // can be: + // - unknown (cannot reason about it) + // - undefined (already reported by other checker) + // - constant (null - should not be tracked, + // other constant will generate a compiler warning) + // - goto (should be reported by other checker) + + // The call return value symbol should stay alive for as long as the + // allocated value symbol, since our diagnostics depend on the value + // returned by the call. Ex: Data should only be freed if noErr was + // returned during allocation.) + SymbolRef RetStatusSymbol = + State->getSVal(CE, C.getLocationContext()).getAsSymbol(); + C.getSymbolManager().addSymbolDependency(V, RetStatusSymbol); + + // Track the allocated value in the checker state. + State = State->set<AllocatedData>(V, AllocationState(ArgExpr, idx, + RetStatusSymbol)); + assert(State); + C.addTransition(State); + } +} + +void MacOSKeychainAPIChecker::checkPreStmt(const ReturnStmt *S, + CheckerContext &C) const { + const Expr *retExpr = S->getRetValue(); + if (!retExpr) + return; + + // If inside inlined call, skip it. + const LocationContext *LC = C.getLocationContext(); + if (LC->getParent() != 0) + return; + + // Check if the value is escaping through the return. + ProgramStateRef state = C.getState(); + SymbolRef sym = state->getSVal(retExpr, LC).getAsLocSymbol(); + if (!sym) + return; + state = state->remove<AllocatedData>(sym); + + // Proceed from the new state. + C.addTransition(state); +} + +// TODO: This logic is the same as in Malloc checker. +const Stmt * +MacOSKeychainAPIChecker::getAllocationSite(const ExplodedNode *N, + SymbolRef Sym, + CheckerContext &C) const { + const LocationContext *LeakContext = N->getLocationContext(); + // Walk the ExplodedGraph backwards and find the first node that referred to + // the tracked symbol. + const ExplodedNode *AllocNode = N; + + while (N) { + if (!N->getState()->get<AllocatedData>(Sym)) + break; + // Allocation node, is the last node in the current context in which the + // symbol was tracked. + if (N->getLocationContext() == LeakContext) + AllocNode = N; + N = N->pred_empty() ? NULL : *(N->pred_begin()); + } + + ProgramPoint P = AllocNode->getLocation(); + if (!isa<StmtPoint>(P)) + return 0; + return cast<clang::PostStmt>(P).getStmt(); +} + +BugReport *MacOSKeychainAPIChecker:: + generateAllocatedDataNotReleasedReport(const AllocationPair &AP, + ExplodedNode *N, + CheckerContext &C) const { + const ADFunctionInfo &FI = FunctionsToTrack[AP.second->AllocatorIdx]; + initBugType(); + SmallString<70> sbuf; + llvm::raw_svector_ostream os(sbuf); + os << "Allocated data is not released: missing a call to '" + << FunctionsToTrack[FI.DeallocatorIdx].Name << "'."; + + // Most bug reports are cached at the location where they occurred. + // With leaks, we want to unique them by the location where they were + // allocated, and only report a single path. + PathDiagnosticLocation LocUsedForUniqueing; + if (const Stmt *AllocStmt = getAllocationSite(N, AP.first, C)) + LocUsedForUniqueing = PathDiagnosticLocation::createBegin(AllocStmt, + C.getSourceManager(), N->getLocationContext()); + + BugReport *Report = new BugReport(*BT, os.str(), N, LocUsedForUniqueing); + Report->addVisitor(new SecKeychainBugVisitor(AP.first)); + markInteresting(Report, AP); + return Report; +} + +void MacOSKeychainAPIChecker::checkDeadSymbols(SymbolReaper &SR, + CheckerContext &C) const { + ProgramStateRef State = C.getState(); + AllocatedSetTy ASet = State->get<AllocatedData>(); + if (ASet.isEmpty()) + return; + + bool Changed = false; + AllocationPairVec Errors; + for (AllocatedSetTy::iterator I = ASet.begin(), E = ASet.end(); I != E; ++I) { + if (SR.isLive(I->first)) + continue; + + Changed = true; + State = State->remove<AllocatedData>(I->first); + // If the allocated symbol is null or if the allocation call might have + // returned an error, do not report. + if (State->getSymVal(I->first) || + definitelyReturnedError(I->second.Region, State, C.getSValBuilder())) + continue; + Errors.push_back(std::make_pair(I->first, &I->second)); + } + if (!Changed) { + // Generate the new, cleaned up state. + C.addTransition(State); + return; + } + + static SimpleProgramPointTag Tag("MacOSKeychainAPIChecker : DeadSymbolsLeak"); + ExplodedNode *N = C.addTransition(C.getState(), C.getPredecessor(), &Tag); + + // Generate the error reports. + for (AllocationPairVec::iterator I = Errors.begin(), E = Errors.end(); + I != E; ++I) { + C.EmitReport(generateAllocatedDataNotReleasedReport(*I, N, C)); + } + + // Generate the new, cleaned up state. + C.addTransition(State, N); +} + +// TODO: Remove this after we ensure that checkDeadSymbols are always called. +void MacOSKeychainAPIChecker::checkEndPath(CheckerContext &C) const { + ProgramStateRef state = C.getState(); + + // If inside inlined call, skip it. + if (C.getLocationContext()->getParent() != 0) + return; + + AllocatedSetTy AS = state->get<AllocatedData>(); + if (AS.isEmpty()) + return; + + // Anything which has been allocated but not freed (nor escaped) will be + // found here, so report it. + bool Changed = false; + AllocationPairVec Errors; + for (AllocatedSetTy::iterator I = AS.begin(), E = AS.end(); I != E; ++I ) { + Changed = true; + state = state->remove<AllocatedData>(I->first); + // If the allocated symbol is null or if error code was returned at + // allocation, do not report. + if (state->getSymVal(I.getKey()) || + definitelyReturnedError(I->second.Region, state, + C.getSValBuilder())) { + continue; + } + Errors.push_back(std::make_pair(I->first, &I->second)); + } + + // If no change, do not generate a new state. + if (!Changed) { + C.addTransition(state); + return; + } + + static SimpleProgramPointTag Tag("MacOSKeychainAPIChecker : EndPathLeak"); + ExplodedNode *N = C.addTransition(C.getState(), C.getPredecessor(), &Tag); + + // Generate the error reports. + for (AllocationPairVec::iterator I = Errors.begin(), E = Errors.end(); + I != E; ++I) { + C.EmitReport(generateAllocatedDataNotReleasedReport(*I, N, C)); + } + + C.addTransition(state, N); +} + + +PathDiagnosticPiece *MacOSKeychainAPIChecker::SecKeychainBugVisitor::VisitNode( + const ExplodedNode *N, + const ExplodedNode *PrevN, + BugReporterContext &BRC, + BugReport &BR) { + const AllocationState *AS = N->getState()->get<AllocatedData>(Sym); + if (!AS) + return 0; + const AllocationState *ASPrev = PrevN->getState()->get<AllocatedData>(Sym); + if (ASPrev) + return 0; + + // (!ASPrev && AS) ~ We started tracking symbol in node N, it must be the + // allocation site. + const CallExpr *CE = cast<CallExpr>(cast<StmtPoint>(N->getLocation()) + .getStmt()); + const FunctionDecl *funDecl = CE->getDirectCallee(); + assert(funDecl && "We do not support indirect function calls as of now."); + StringRef funName = funDecl->getName(); + + // Get the expression of the corresponding argument. + unsigned Idx = getTrackedFunctionIndex(funName, true); + assert(Idx != InvalidIdx && "This should be a call to an allocator."); + const Expr *ArgExpr = CE->getArg(FunctionsToTrack[Idx].Param); + PathDiagnosticLocation Pos(ArgExpr, BRC.getSourceManager(), + N->getLocationContext()); + return new PathDiagnosticEventPiece(Pos, "Data is allocated here."); +} + +void ento::registerMacOSKeychainAPIChecker(CheckerManager &mgr) { + mgr.registerChecker<MacOSKeychainAPIChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/MacOSXAPIChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/MacOSXAPIChecker.cpp new file mode 100644 index 0000000..cfdb55d --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/MacOSXAPIChecker.cpp @@ -0,0 +1,116 @@ +// MacOSXAPIChecker.h - Checks proper use of various MacOS X APIs --*- C++ -*-// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This defines MacOSXAPIChecker, which is an assortment of checks on calls +// to various, widely used Mac OS X functions. +// +// FIXME: What's currently in BasicObjCFoundationChecks.cpp should be migrated +// to here, using the new Checker interface. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramStateTrait.h" +#include "clang/Basic/TargetInfo.h" +#include "llvm/ADT/SmallString.h" +#include "llvm/ADT/StringSwitch.h" +#include "llvm/Support/raw_ostream.h" + +using namespace clang; +using namespace ento; + +namespace { +class MacOSXAPIChecker : public Checker< check::PreStmt<CallExpr> > { + mutable OwningPtr<BugType> BT_dispatchOnce; + +public: + void checkPreStmt(const CallExpr *CE, CheckerContext &C) const; + + void CheckDispatchOnce(CheckerContext &C, const CallExpr *CE, + StringRef FName) const; + + typedef void (MacOSXAPIChecker::*SubChecker)(CheckerContext &, + const CallExpr *, + StringRef FName) const; +}; +} //end anonymous namespace + +//===----------------------------------------------------------------------===// +// dispatch_once and dispatch_once_f +//===----------------------------------------------------------------------===// + +void MacOSXAPIChecker::CheckDispatchOnce(CheckerContext &C, const CallExpr *CE, + StringRef FName) const { + if (CE->getNumArgs() < 1) + return; + + // Check if the first argument is stack allocated. If so, issue a warning + // because that's likely to be bad news. + ProgramStateRef state = C.getState(); + const MemRegion *R = + state->getSVal(CE->getArg(0), C.getLocationContext()).getAsRegion(); + if (!R || !isa<StackSpaceRegion>(R->getMemorySpace())) + return; + + ExplodedNode *N = C.generateSink(state); + if (!N) + return; + + if (!BT_dispatchOnce) + BT_dispatchOnce.reset(new BugType("Improper use of 'dispatch_once'", + "Mac OS X API")); + + SmallString<256> S; + llvm::raw_svector_ostream os(S); + os << "Call to '" << FName << "' uses"; + if (const VarRegion *VR = dyn_cast<VarRegion>(R)) + os << " the local variable '" << VR->getDecl()->getName() << '\''; + else + os << " stack allocated memory"; + os << " for the predicate value. Using such transient memory for " + "the predicate is potentially dangerous."; + if (isa<VarRegion>(R) && isa<StackLocalsSpaceRegion>(R->getMemorySpace())) + os << " Perhaps you intended to declare the variable as 'static'?"; + + BugReport *report = new BugReport(*BT_dispatchOnce, os.str(), N); + report->addRange(CE->getArg(0)->getSourceRange()); + C.EmitReport(report); +} + +//===----------------------------------------------------------------------===// +// Central dispatch function. +//===----------------------------------------------------------------------===// + +void MacOSXAPIChecker::checkPreStmt(const CallExpr *CE, + CheckerContext &C) const { + StringRef Name = C.getCalleeName(CE); + if (Name.empty()) + return; + + SubChecker SC = + llvm::StringSwitch<SubChecker>(Name) + .Cases("dispatch_once", "dispatch_once_f", + &MacOSXAPIChecker::CheckDispatchOnce) + .Default(NULL); + + if (SC) + (this->*SC)(C, CE, Name); +} + +//===----------------------------------------------------------------------===// +// Registration. +//===----------------------------------------------------------------------===// + +void ento::registerMacOSXAPIChecker(CheckerManager &mgr) { + mgr.registerChecker<MacOSXAPIChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/Makefile b/clang/lib/StaticAnalyzer/Checkers/Makefile new file mode 100644 index 0000000..2582908 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/Makefile @@ -0,0 +1,24 @@ +##===- clang/lib/Checker/Makefile --------------------------*- Makefile -*-===## +# +# The LLVM Compiler Infrastructure +# +# This file is distributed under the University of Illinois Open Source +# License. See LICENSE.TXT for details. +# +##===----------------------------------------------------------------------===## +# +# This implements analyses built on top of source-level CFGs. +# +##===----------------------------------------------------------------------===## + +CLANG_LEVEL := ../../.. +LIBRARYNAME := clangStaticAnalyzerCheckers + +BUILT_SOURCES = Checkers.inc +TABLEGEN_INC_FILES_COMMON = 1 + +include $(CLANG_LEVEL)/Makefile + +$(ObjDir)/Checkers.inc.tmp : Checkers.td $(PROJ_SRC_DIR)/$(CLANG_LEVEL)/include/clang/StaticAnalyzer/Checkers/CheckerBase.td $(CLANG_TBLGEN) $(ObjDir)/.dir + $(Echo) "Building Clang SA Checkers tables with tblgen" + $(Verb) $(ClangTableGen) -gen-clang-sa-checkers -I $(PROJ_SRC_DIR)/$(CLANG_LEVEL)/include -o $(call SYSPATH, $@) $< diff --git a/clang/lib/StaticAnalyzer/Checkers/MallocChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/MallocChecker.cpp new file mode 100644 index 0000000..8bce88a --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/MallocChecker.cpp @@ -0,0 +1,1463 @@ +//=== MallocChecker.cpp - A malloc/free checker -------------------*- C++ -*--// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines malloc/free checker, which checks for potential memory +// leaks, double free, and use-after-free problems. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "InterCheckerAPI.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ObjCMessage.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramState.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramStateTrait.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/SymbolManager.h" +#include "clang/Basic/SourceManager.h" +#include "llvm/ADT/ImmutableMap.h" +#include "llvm/ADT/SmallString.h" +#include "llvm/ADT/STLExtras.h" +#include <climits> + +using namespace clang; +using namespace ento; + +namespace { + +class RefState { + enum Kind { AllocateUnchecked, AllocateFailed, Released, Escaped, + Relinquished } K; + const Stmt *S; + +public: + RefState(Kind k, const Stmt *s) : K(k), S(s) {} + + bool isAllocated() const { return K == AllocateUnchecked; } + bool isReleased() const { return K == Released; } + + const Stmt *getStmt() const { return S; } + + bool operator==(const RefState &X) const { + return K == X.K && S == X.S; + } + + static RefState getAllocateUnchecked(const Stmt *s) { + return RefState(AllocateUnchecked, s); + } + static RefState getAllocateFailed() { + return RefState(AllocateFailed, 0); + } + static RefState getReleased(const Stmt *s) { return RefState(Released, s); } + static RefState getEscaped(const Stmt *s) { return RefState(Escaped, s); } + static RefState getRelinquished(const Stmt *s) { + return RefState(Relinquished, s); + } + + void Profile(llvm::FoldingSetNodeID &ID) const { + ID.AddInteger(K); + ID.AddPointer(S); + } +}; + +struct ReallocPair { + SymbolRef ReallocatedSym; + bool IsFreeOnFailure; + ReallocPair(SymbolRef S, bool F) : ReallocatedSym(S), IsFreeOnFailure(F) {} + void Profile(llvm::FoldingSetNodeID &ID) const { + ID.AddInteger(IsFreeOnFailure); + ID.AddPointer(ReallocatedSym); + } + bool operator==(const ReallocPair &X) const { + return ReallocatedSym == X.ReallocatedSym && + IsFreeOnFailure == X.IsFreeOnFailure; + } +}; + +typedef std::pair<const Stmt*, const MemRegion*> LeakInfo; + +class MallocChecker : public Checker<check::DeadSymbols, + check::EndPath, + check::PreStmt<ReturnStmt>, + check::PreStmt<CallExpr>, + check::PostStmt<CallExpr>, + check::PostStmt<BlockExpr>, + check::Location, + check::Bind, + eval::Assume, + check::RegionChanges> +{ + mutable OwningPtr<BugType> BT_DoubleFree; + mutable OwningPtr<BugType> BT_Leak; + mutable OwningPtr<BugType> BT_UseFree; + mutable OwningPtr<BugType> BT_BadFree; + mutable IdentifierInfo *II_malloc, *II_free, *II_realloc, *II_calloc, + *II_valloc, *II_reallocf, *II_strndup, *II_strdup; + +public: + MallocChecker() : II_malloc(0), II_free(0), II_realloc(0), II_calloc(0), + II_valloc(0), II_reallocf(0), II_strndup(0), II_strdup(0) {} + + /// In pessimistic mode, the checker assumes that it does not know which + /// functions might free the memory. + struct ChecksFilter { + DefaultBool CMallocPessimistic; + DefaultBool CMallocOptimistic; + }; + + ChecksFilter Filter; + + void checkPreStmt(const CallExpr *S, CheckerContext &C) const; + void checkPostStmt(const CallExpr *CE, CheckerContext &C) const; + void checkPostStmt(const BlockExpr *BE, CheckerContext &C) const; + void checkDeadSymbols(SymbolReaper &SymReaper, CheckerContext &C) const; + void checkEndPath(CheckerContext &C) const; + void checkPreStmt(const ReturnStmt *S, CheckerContext &C) const; + ProgramStateRef evalAssume(ProgramStateRef state, SVal Cond, + bool Assumption) const; + void checkLocation(SVal l, bool isLoad, const Stmt *S, + CheckerContext &C) const; + void checkBind(SVal location, SVal val, const Stmt*S, + CheckerContext &C) const; + ProgramStateRef + checkRegionChanges(ProgramStateRef state, + const StoreManager::InvalidatedSymbols *invalidated, + ArrayRef<const MemRegion *> ExplicitRegions, + ArrayRef<const MemRegion *> Regions, + const CallOrObjCMessage *Call) const; + bool wantsRegionChangeUpdate(ProgramStateRef state) const { + return true; + } + +private: + void initIdentifierInfo(ASTContext &C) const; + + /// Check if this is one of the functions which can allocate/reallocate memory + /// pointed to by one of its arguments. + bool isMemFunction(const FunctionDecl *FD, ASTContext &C) const; + + static ProgramStateRef MallocMemReturnsAttr(CheckerContext &C, + const CallExpr *CE, + const OwnershipAttr* Att); + static ProgramStateRef MallocMemAux(CheckerContext &C, const CallExpr *CE, + const Expr *SizeEx, SVal Init, + ProgramStateRef state) { + return MallocMemAux(C, CE, + state->getSVal(SizeEx, C.getLocationContext()), + Init, state); + } + + static ProgramStateRef MallocMemAux(CheckerContext &C, const CallExpr *CE, + SVal SizeEx, SVal Init, + ProgramStateRef state); + + /// Update the RefState to reflect the new memory allocation. + static ProgramStateRef MallocUpdateRefState(CheckerContext &C, + const CallExpr *CE, + ProgramStateRef state); + + ProgramStateRef FreeMemAttr(CheckerContext &C, const CallExpr *CE, + const OwnershipAttr* Att) const; + ProgramStateRef FreeMemAux(CheckerContext &C, const CallExpr *CE, + ProgramStateRef state, unsigned Num, + bool Hold) const; + + ProgramStateRef ReallocMem(CheckerContext &C, const CallExpr *CE, + bool FreesMemOnFailure) const; + static ProgramStateRef CallocMem(CheckerContext &C, const CallExpr *CE); + + bool checkEscape(SymbolRef Sym, const Stmt *S, CheckerContext &C) const; + bool checkUseAfterFree(SymbolRef Sym, CheckerContext &C, + const Stmt *S = 0) const; + + /// Check if the function is not known to us. So, for example, we could + /// conservatively assume it can free/reallocate it's pointer arguments. + bool doesNotFreeMemory(const CallOrObjCMessage *Call, + ProgramStateRef State) const; + + static bool SummarizeValue(raw_ostream &os, SVal V); + static bool SummarizeRegion(raw_ostream &os, const MemRegion *MR); + void ReportBadFree(CheckerContext &C, SVal ArgVal, SourceRange range) const; + + /// Find the location of the allocation for Sym on the path leading to the + /// exploded node N. + LeakInfo getAllocationSite(const ExplodedNode *N, SymbolRef Sym, + CheckerContext &C) const; + + void reportLeak(SymbolRef Sym, ExplodedNode *N, CheckerContext &C) const; + + /// The bug visitor which allows us to print extra diagnostics along the + /// BugReport path. For example, showing the allocation site of the leaked + /// region. + class MallocBugVisitor : public BugReporterVisitorImpl<MallocBugVisitor> { + protected: + enum NotificationMode { + Normal, + ReallocationFailed + }; + + // The allocated region symbol tracked by the main analysis. + SymbolRef Sym; + + // The mode we are in, i.e. what kind of diagnostics will be emitted. + NotificationMode Mode; + + // A symbol from when the primary region should have been reallocated. + SymbolRef FailedReallocSymbol; + + public: + MallocBugVisitor(SymbolRef S) + : Sym(S), Mode(Normal), FailedReallocSymbol(0) {} + + virtual ~MallocBugVisitor() {} + + void Profile(llvm::FoldingSetNodeID &ID) const { + static int X = 0; + ID.AddPointer(&X); + ID.AddPointer(Sym); + } + + inline bool isAllocated(const RefState *S, const RefState *SPrev, + const Stmt *Stmt) { + // Did not track -> allocated. Other state (released) -> allocated. + return (Stmt && isa<CallExpr>(Stmt) && + (S && S->isAllocated()) && (!SPrev || !SPrev->isAllocated())); + } + + inline bool isReleased(const RefState *S, const RefState *SPrev, + const Stmt *Stmt) { + // Did not track -> released. Other state (allocated) -> released. + return (Stmt && isa<CallExpr>(Stmt) && + (S && S->isReleased()) && (!SPrev || !SPrev->isReleased())); + } + + inline bool isReallocFailedCheck(const RefState *S, const RefState *SPrev, + const Stmt *Stmt) { + // If the expression is not a call, and the state change is + // released -> allocated, it must be the realloc return value + // check. If we have to handle more cases here, it might be cleaner just + // to track this extra bit in the state itself. + return ((!Stmt || !isa<CallExpr>(Stmt)) && + (S && S->isAllocated()) && (SPrev && !SPrev->isAllocated())); + } + + PathDiagnosticPiece *VisitNode(const ExplodedNode *N, + const ExplodedNode *PrevN, + BugReporterContext &BRC, + BugReport &BR); + private: + class StackHintGeneratorForReallocationFailed + : public StackHintGeneratorForSymbol { + public: + StackHintGeneratorForReallocationFailed(SymbolRef S, StringRef M) + : StackHintGeneratorForSymbol(S, M) {} + + virtual std::string getMessageForArg(const Expr *ArgE, unsigned ArgIndex) { + SmallString<200> buf; + llvm::raw_svector_ostream os(buf); + + os << "Reallocation of "; + // Printed parameters start at 1, not 0. + printOrdinal(++ArgIndex, os); + os << " parameter failed"; + + return os.str(); + } + + virtual std::string getMessageForReturn(const CallExpr *CallExpr) { + return "Reallocation of returned value failed"; + } + }; + }; +}; +} // end anonymous namespace + +typedef llvm::ImmutableMap<SymbolRef, RefState> RegionStateTy; +typedef llvm::ImmutableMap<SymbolRef, ReallocPair > ReallocMap; +class RegionState {}; +class ReallocPairs {}; +namespace clang { +namespace ento { + template <> + struct ProgramStateTrait<RegionState> + : public ProgramStatePartialTrait<RegionStateTy> { + static void *GDMIndex() { static int x; return &x; } + }; + + template <> + struct ProgramStateTrait<ReallocPairs> + : public ProgramStatePartialTrait<ReallocMap> { + static void *GDMIndex() { static int x; return &x; } + }; +} +} + +namespace { +class StopTrackingCallback : public SymbolVisitor { + ProgramStateRef state; +public: + StopTrackingCallback(ProgramStateRef st) : state(st) {} + ProgramStateRef getState() const { return state; } + + bool VisitSymbol(SymbolRef sym) { + state = state->remove<RegionState>(sym); + return true; + } +}; +} // end anonymous namespace + +void MallocChecker::initIdentifierInfo(ASTContext &Ctx) const { + if (!II_malloc) + II_malloc = &Ctx.Idents.get("malloc"); + if (!II_free) + II_free = &Ctx.Idents.get("free"); + if (!II_realloc) + II_realloc = &Ctx.Idents.get("realloc"); + if (!II_reallocf) + II_reallocf = &Ctx.Idents.get("reallocf"); + if (!II_calloc) + II_calloc = &Ctx.Idents.get("calloc"); + if (!II_valloc) + II_valloc = &Ctx.Idents.get("valloc"); + if (!II_strdup) + II_strdup = &Ctx.Idents.get("strdup"); + if (!II_strndup) + II_strndup = &Ctx.Idents.get("strndup"); +} + +bool MallocChecker::isMemFunction(const FunctionDecl *FD, ASTContext &C) const { + if (!FD) + return false; + IdentifierInfo *FunI = FD->getIdentifier(); + if (!FunI) + return false; + + initIdentifierInfo(C); + + if (FunI == II_malloc || FunI == II_free || FunI == II_realloc || + FunI == II_reallocf || FunI == II_calloc || FunI == II_valloc || + FunI == II_strdup || FunI == II_strndup) + return true; + + if (Filter.CMallocOptimistic && FD->hasAttrs() && + FD->specific_attr_begin<OwnershipAttr>() != + FD->specific_attr_end<OwnershipAttr>()) + return true; + + + return false; +} + +void MallocChecker::checkPostStmt(const CallExpr *CE, CheckerContext &C) const { + const FunctionDecl *FD = C.getCalleeDecl(CE); + if (!FD) + return; + + initIdentifierInfo(C.getASTContext()); + IdentifierInfo *FunI = FD->getIdentifier(); + if (!FunI) + return; + + ProgramStateRef State = C.getState(); + if (FunI == II_malloc || FunI == II_valloc) { + if (CE->getNumArgs() < 1) + return; + State = MallocMemAux(C, CE, CE->getArg(0), UndefinedVal(), State); + } else if (FunI == II_realloc) { + State = ReallocMem(C, CE, false); + } else if (FunI == II_reallocf) { + State = ReallocMem(C, CE, true); + } else if (FunI == II_calloc) { + State = CallocMem(C, CE); + } else if (FunI == II_free) { + State = FreeMemAux(C, CE, C.getState(), 0, false); + } else if (FunI == II_strdup) { + State = MallocUpdateRefState(C, CE, State); + } else if (FunI == II_strndup) { + State = MallocUpdateRefState(C, CE, State); + } else if (Filter.CMallocOptimistic) { + // Check all the attributes, if there are any. + // There can be multiple of these attributes. + if (FD->hasAttrs()) + for (specific_attr_iterator<OwnershipAttr> + i = FD->specific_attr_begin<OwnershipAttr>(), + e = FD->specific_attr_end<OwnershipAttr>(); + i != e; ++i) { + switch ((*i)->getOwnKind()) { + case OwnershipAttr::Returns: + State = MallocMemReturnsAttr(C, CE, *i); + break; + case OwnershipAttr::Takes: + case OwnershipAttr::Holds: + State = FreeMemAttr(C, CE, *i); + break; + } + } + } + C.addTransition(State); +} + +ProgramStateRef MallocChecker::MallocMemReturnsAttr(CheckerContext &C, + const CallExpr *CE, + const OwnershipAttr* Att) { + if (Att->getModule() != "malloc") + return 0; + + OwnershipAttr::args_iterator I = Att->args_begin(), E = Att->args_end(); + if (I != E) { + return MallocMemAux(C, CE, CE->getArg(*I), UndefinedVal(), C.getState()); + } + return MallocMemAux(C, CE, UnknownVal(), UndefinedVal(), C.getState()); +} + +ProgramStateRef MallocChecker::MallocMemAux(CheckerContext &C, + const CallExpr *CE, + SVal Size, SVal Init, + ProgramStateRef state) { + // Get the return value. + SVal retVal = state->getSVal(CE, C.getLocationContext()); + + // We expect the malloc functions to return a pointer. + if (!isa<Loc>(retVal)) + return 0; + + // Fill the region with the initialization value. + state = state->bindDefault(retVal, Init); + + // Set the region's extent equal to the Size parameter. + const SymbolicRegion *R = + dyn_cast_or_null<SymbolicRegion>(retVal.getAsRegion()); + if (!R) + return 0; + if (isa<DefinedOrUnknownSVal>(Size)) { + SValBuilder &svalBuilder = C.getSValBuilder(); + DefinedOrUnknownSVal Extent = R->getExtent(svalBuilder); + DefinedOrUnknownSVal DefinedSize = cast<DefinedOrUnknownSVal>(Size); + DefinedOrUnknownSVal extentMatchesSize = + svalBuilder.evalEQ(state, Extent, DefinedSize); + + state = state->assume(extentMatchesSize, true); + assert(state); + } + + return MallocUpdateRefState(C, CE, state); +} + +ProgramStateRef MallocChecker::MallocUpdateRefState(CheckerContext &C, + const CallExpr *CE, + ProgramStateRef state) { + // Get the return value. + SVal retVal = state->getSVal(CE, C.getLocationContext()); + + // We expect the malloc functions to return a pointer. + if (!isa<Loc>(retVal)) + return 0; + + SymbolRef Sym = retVal.getAsLocSymbol(); + assert(Sym); + + // Set the symbol's state to Allocated. + return state->set<RegionState>(Sym, RefState::getAllocateUnchecked(CE)); + +} + +ProgramStateRef MallocChecker::FreeMemAttr(CheckerContext &C, + const CallExpr *CE, + const OwnershipAttr* Att) const { + if (Att->getModule() != "malloc") + return 0; + + ProgramStateRef State = C.getState(); + + for (OwnershipAttr::args_iterator I = Att->args_begin(), E = Att->args_end(); + I != E; ++I) { + ProgramStateRef StateI = FreeMemAux(C, CE, State, *I, + Att->getOwnKind() == OwnershipAttr::Holds); + if (StateI) + State = StateI; + } + return State; +} + +ProgramStateRef MallocChecker::FreeMemAux(CheckerContext &C, + const CallExpr *CE, + ProgramStateRef state, + unsigned Num, + bool Hold) const { + if (CE->getNumArgs() < (Num + 1)) + return 0; + + const Expr *ArgExpr = CE->getArg(Num); + SVal ArgVal = state->getSVal(ArgExpr, C.getLocationContext()); + if (!isa<DefinedOrUnknownSVal>(ArgVal)) + return 0; + DefinedOrUnknownSVal location = cast<DefinedOrUnknownSVal>(ArgVal); + + // Check for null dereferences. + if (!isa<Loc>(location)) + return 0; + + // The explicit NULL case, no operation is performed. + ProgramStateRef notNullState, nullState; + llvm::tie(notNullState, nullState) = state->assume(location); + if (nullState && !notNullState) + return 0; + + // Unknown values could easily be okay + // Undefined values are handled elsewhere + if (ArgVal.isUnknownOrUndef()) + return 0; + + const MemRegion *R = ArgVal.getAsRegion(); + + // Nonlocs can't be freed, of course. + // Non-region locations (labels and fixed addresses) also shouldn't be freed. + if (!R) { + ReportBadFree(C, ArgVal, ArgExpr->getSourceRange()); + return 0; + } + + R = R->StripCasts(); + + // Blocks might show up as heap data, but should not be free()d + if (isa<BlockDataRegion>(R)) { + ReportBadFree(C, ArgVal, ArgExpr->getSourceRange()); + return 0; + } + + const MemSpaceRegion *MS = R->getMemorySpace(); + + // Parameters, locals, statics, and globals shouldn't be freed. + if (!(isa<UnknownSpaceRegion>(MS) || isa<HeapSpaceRegion>(MS))) { + // FIXME: at the time this code was written, malloc() regions were + // represented by conjured symbols, which are all in UnknownSpaceRegion. + // This means that there isn't actually anything from HeapSpaceRegion + // that should be freed, even though we allow it here. + // Of course, free() can work on memory allocated outside the current + // function, so UnknownSpaceRegion is always a possibility. + // False negatives are better than false positives. + + ReportBadFree(C, ArgVal, ArgExpr->getSourceRange()); + return 0; + } + + const SymbolicRegion *SR = dyn_cast<SymbolicRegion>(R); + // Various cases could lead to non-symbol values here. + // For now, ignore them. + if (!SR) + return 0; + + SymbolRef Sym = SR->getSymbol(); + const RefState *RS = state->get<RegionState>(Sym); + + // If the symbol has not been tracked, return. This is possible when free() is + // called on a pointer that does not get its pointee directly from malloc(). + // Full support of this requires inter-procedural analysis. + if (!RS) + return 0; + + // Check double free. + if (RS->isReleased()) { + if (ExplodedNode *N = C.generateSink()) { + if (!BT_DoubleFree) + BT_DoubleFree.reset( + new BugType("Double free", "Memory Error")); + BugReport *R = new BugReport(*BT_DoubleFree, + "Attempt to free released memory", N); + R->addRange(ArgExpr->getSourceRange()); + R->markInteresting(Sym); + R->addVisitor(new MallocBugVisitor(Sym)); + C.EmitReport(R); + } + return 0; + } + + // Normal free. + if (Hold) + return state->set<RegionState>(Sym, RefState::getRelinquished(CE)); + return state->set<RegionState>(Sym, RefState::getReleased(CE)); +} + +bool MallocChecker::SummarizeValue(raw_ostream &os, SVal V) { + if (nonloc::ConcreteInt *IntVal = dyn_cast<nonloc::ConcreteInt>(&V)) + os << "an integer (" << IntVal->getValue() << ")"; + else if (loc::ConcreteInt *ConstAddr = dyn_cast<loc::ConcreteInt>(&V)) + os << "a constant address (" << ConstAddr->getValue() << ")"; + else if (loc::GotoLabel *Label = dyn_cast<loc::GotoLabel>(&V)) + os << "the address of the label '" << Label->getLabel()->getName() << "'"; + else + return false; + + return true; +} + +bool MallocChecker::SummarizeRegion(raw_ostream &os, + const MemRegion *MR) { + switch (MR->getKind()) { + case MemRegion::FunctionTextRegionKind: { + const FunctionDecl *FD = cast<FunctionTextRegion>(MR)->getDecl(); + if (FD) + os << "the address of the function '" << *FD << '\''; + else + os << "the address of a function"; + return true; + } + case MemRegion::BlockTextRegionKind: + os << "block text"; + return true; + case MemRegion::BlockDataRegionKind: + // FIXME: where the block came from? + os << "a block"; + return true; + default: { + const MemSpaceRegion *MS = MR->getMemorySpace(); + + if (isa<StackLocalsSpaceRegion>(MS)) { + const VarRegion *VR = dyn_cast<VarRegion>(MR); + const VarDecl *VD; + if (VR) + VD = VR->getDecl(); + else + VD = NULL; + + if (VD) + os << "the address of the local variable '" << VD->getName() << "'"; + else + os << "the address of a local stack variable"; + return true; + } + + if (isa<StackArgumentsSpaceRegion>(MS)) { + const VarRegion *VR = dyn_cast<VarRegion>(MR); + const VarDecl *VD; + if (VR) + VD = VR->getDecl(); + else + VD = NULL; + + if (VD) + os << "the address of the parameter '" << VD->getName() << "'"; + else + os << "the address of a parameter"; + return true; + } + + if (isa<GlobalsSpaceRegion>(MS)) { + const VarRegion *VR = dyn_cast<VarRegion>(MR); + const VarDecl *VD; + if (VR) + VD = VR->getDecl(); + else + VD = NULL; + + if (VD) { + if (VD->isStaticLocal()) + os << "the address of the static variable '" << VD->getName() << "'"; + else + os << "the address of the global variable '" << VD->getName() << "'"; + } else + os << "the address of a global variable"; + return true; + } + + return false; + } + } +} + +void MallocChecker::ReportBadFree(CheckerContext &C, SVal ArgVal, + SourceRange range) const { + if (ExplodedNode *N = C.generateSink()) { + if (!BT_BadFree) + BT_BadFree.reset(new BugType("Bad free", "Memory Error")); + + SmallString<100> buf; + llvm::raw_svector_ostream os(buf); + + const MemRegion *MR = ArgVal.getAsRegion(); + if (MR) { + while (const ElementRegion *ER = dyn_cast<ElementRegion>(MR)) + MR = ER->getSuperRegion(); + + // Special case for alloca() + if (isa<AllocaRegion>(MR)) + os << "Argument to free() was allocated by alloca(), not malloc()"; + else { + os << "Argument to free() is "; + if (SummarizeRegion(os, MR)) + os << ", which is not memory allocated by malloc()"; + else + os << "not memory allocated by malloc()"; + } + } else { + os << "Argument to free() is "; + if (SummarizeValue(os, ArgVal)) + os << ", which is not memory allocated by malloc()"; + else + os << "not memory allocated by malloc()"; + } + + BugReport *R = new BugReport(*BT_BadFree, os.str(), N); + R->markInteresting(MR); + R->addRange(range); + C.EmitReport(R); + } +} + +ProgramStateRef MallocChecker::ReallocMem(CheckerContext &C, + const CallExpr *CE, + bool FreesOnFail) const { + if (CE->getNumArgs() < 2) + return 0; + + ProgramStateRef state = C.getState(); + const Expr *arg0Expr = CE->getArg(0); + const LocationContext *LCtx = C.getLocationContext(); + SVal Arg0Val = state->getSVal(arg0Expr, LCtx); + if (!isa<DefinedOrUnknownSVal>(Arg0Val)) + return 0; + DefinedOrUnknownSVal arg0Val = cast<DefinedOrUnknownSVal>(Arg0Val); + + SValBuilder &svalBuilder = C.getSValBuilder(); + + DefinedOrUnknownSVal PtrEQ = + svalBuilder.evalEQ(state, arg0Val, svalBuilder.makeNull()); + + // Get the size argument. If there is no size arg then give up. + const Expr *Arg1 = CE->getArg(1); + if (!Arg1) + return 0; + + // Get the value of the size argument. + SVal Arg1ValG = state->getSVal(Arg1, LCtx); + if (!isa<DefinedOrUnknownSVal>(Arg1ValG)) + return 0; + DefinedOrUnknownSVal Arg1Val = cast<DefinedOrUnknownSVal>(Arg1ValG); + + // Compare the size argument to 0. + DefinedOrUnknownSVal SizeZero = + svalBuilder.evalEQ(state, Arg1Val, + svalBuilder.makeIntValWithPtrWidth(0, false)); + + ProgramStateRef StatePtrIsNull, StatePtrNotNull; + llvm::tie(StatePtrIsNull, StatePtrNotNull) = state->assume(PtrEQ); + ProgramStateRef StateSizeIsZero, StateSizeNotZero; + llvm::tie(StateSizeIsZero, StateSizeNotZero) = state->assume(SizeZero); + // We only assume exceptional states if they are definitely true; if the + // state is under-constrained, assume regular realloc behavior. + bool PrtIsNull = StatePtrIsNull && !StatePtrNotNull; + bool SizeIsZero = StateSizeIsZero && !StateSizeNotZero; + + // If the ptr is NULL and the size is not 0, the call is equivalent to + // malloc(size). + if ( PrtIsNull && !SizeIsZero) { + ProgramStateRef stateMalloc = MallocMemAux(C, CE, CE->getArg(1), + UndefinedVal(), StatePtrIsNull); + return stateMalloc; + } + + if (PrtIsNull && SizeIsZero) + return 0; + + // Get the from and to pointer symbols as in toPtr = realloc(fromPtr, size). + assert(!PrtIsNull); + SymbolRef FromPtr = arg0Val.getAsSymbol(); + SVal RetVal = state->getSVal(CE, LCtx); + SymbolRef ToPtr = RetVal.getAsSymbol(); + if (!FromPtr || !ToPtr) + return 0; + + // If the size is 0, free the memory. + if (SizeIsZero) + if (ProgramStateRef stateFree = FreeMemAux(C, CE, StateSizeIsZero,0,false)){ + // The semantics of the return value are: + // If size was equal to 0, either NULL or a pointer suitable to be passed + // to free() is returned. + stateFree = stateFree->set<ReallocPairs>(ToPtr, + ReallocPair(FromPtr, FreesOnFail)); + C.getSymbolManager().addSymbolDependency(ToPtr, FromPtr); + return stateFree; + } + + // Default behavior. + if (ProgramStateRef stateFree = FreeMemAux(C, CE, state, 0, false)) { + // FIXME: We should copy the content of the original buffer. + ProgramStateRef stateRealloc = MallocMemAux(C, CE, CE->getArg(1), + UnknownVal(), stateFree); + if (!stateRealloc) + return 0; + stateRealloc = stateRealloc->set<ReallocPairs>(ToPtr, + ReallocPair(FromPtr, FreesOnFail)); + C.getSymbolManager().addSymbolDependency(ToPtr, FromPtr); + return stateRealloc; + } + return 0; +} + +ProgramStateRef MallocChecker::CallocMem(CheckerContext &C, const CallExpr *CE){ + if (CE->getNumArgs() < 2) + return 0; + + ProgramStateRef state = C.getState(); + SValBuilder &svalBuilder = C.getSValBuilder(); + const LocationContext *LCtx = C.getLocationContext(); + SVal count = state->getSVal(CE->getArg(0), LCtx); + SVal elementSize = state->getSVal(CE->getArg(1), LCtx); + SVal TotalSize = svalBuilder.evalBinOp(state, BO_Mul, count, elementSize, + svalBuilder.getContext().getSizeType()); + SVal zeroVal = svalBuilder.makeZeroVal(svalBuilder.getContext().CharTy); + + return MallocMemAux(C, CE, TotalSize, zeroVal, state); +} + +LeakInfo +MallocChecker::getAllocationSite(const ExplodedNode *N, SymbolRef Sym, + CheckerContext &C) const { + const LocationContext *LeakContext = N->getLocationContext(); + // Walk the ExplodedGraph backwards and find the first node that referred to + // the tracked symbol. + const ExplodedNode *AllocNode = N; + const MemRegion *ReferenceRegion = 0; + + while (N) { + ProgramStateRef State = N->getState(); + if (!State->get<RegionState>(Sym)) + break; + + // Find the most recent expression bound to the symbol in the current + // context. + if (!ReferenceRegion) { + if (const MemRegion *MR = C.getLocationRegionIfPostStore(N)) { + SVal Val = State->getSVal(MR); + if (Val.getAsLocSymbol() == Sym) + ReferenceRegion = MR; + } + } + + // Allocation node, is the last node in the current context in which the + // symbol was tracked. + if (N->getLocationContext() == LeakContext) + AllocNode = N; + N = N->pred_empty() ? NULL : *(N->pred_begin()); + } + + ProgramPoint P = AllocNode->getLocation(); + const Stmt *AllocationStmt = 0; + if (isa<StmtPoint>(P)) + AllocationStmt = cast<StmtPoint>(P).getStmt(); + + return LeakInfo(AllocationStmt, ReferenceRegion); +} + +void MallocChecker::reportLeak(SymbolRef Sym, ExplodedNode *N, + CheckerContext &C) const { + assert(N); + if (!BT_Leak) { + BT_Leak.reset(new BugType("Memory leak", "Memory Error")); + // Leaks should not be reported if they are post-dominated by a sink: + // (1) Sinks are higher importance bugs. + // (2) NoReturnFunctionChecker uses sink nodes to represent paths ending + // with __noreturn functions such as assert() or exit(). We choose not + // to report leaks on such paths. + BT_Leak->setSuppressOnSink(true); + } + + // Most bug reports are cached at the location where they occurred. + // With leaks, we want to unique them by the location where they were + // allocated, and only report a single path. + PathDiagnosticLocation LocUsedForUniqueing; + const Stmt *AllocStmt = 0; + const MemRegion *Region = 0; + llvm::tie(AllocStmt, Region) = getAllocationSite(N, Sym, C); + if (AllocStmt) + LocUsedForUniqueing = PathDiagnosticLocation::createBegin(AllocStmt, + C.getSourceManager(), N->getLocationContext()); + + SmallString<200> buf; + llvm::raw_svector_ostream os(buf); + os << "Memory is never released; potential leak"; + if (Region) { + os << " of memory pointed to by '"; + Region->dumpPretty(os); + os <<'\''; + } + + BugReport *R = new BugReport(*BT_Leak, os.str(), N, LocUsedForUniqueing); + R->markInteresting(Sym); + R->addVisitor(new MallocBugVisitor(Sym)); + C.EmitReport(R); +} + +void MallocChecker::checkDeadSymbols(SymbolReaper &SymReaper, + CheckerContext &C) const +{ + if (!SymReaper.hasDeadSymbols()) + return; + + ProgramStateRef state = C.getState(); + RegionStateTy RS = state->get<RegionState>(); + RegionStateTy::Factory &F = state->get_context<RegionState>(); + + bool generateReport = false; + llvm::SmallVector<SymbolRef, 2> Errors; + for (RegionStateTy::iterator I = RS.begin(), E = RS.end(); I != E; ++I) { + if (SymReaper.isDead(I->first)) { + if (I->second.isAllocated()) { + generateReport = true; + Errors.push_back(I->first); + } + // Remove the dead symbol from the map. + RS = F.remove(RS, I->first); + + } + } + + // Cleanup the Realloc Pairs Map. + ReallocMap RP = state->get<ReallocPairs>(); + for (ReallocMap::iterator I = RP.begin(), E = RP.end(); I != E; ++I) { + if (SymReaper.isDead(I->first) || + SymReaper.isDead(I->second.ReallocatedSym)) { + state = state->remove<ReallocPairs>(I->first); + } + } + + // Generate leak node. + static SimpleProgramPointTag Tag("MallocChecker : DeadSymbolsLeak"); + ExplodedNode *N = C.addTransition(C.getState(), C.getPredecessor(), &Tag); + + if (generateReport) { + for (llvm::SmallVector<SymbolRef, 2>::iterator + I = Errors.begin(), E = Errors.end(); I != E; ++I) { + reportLeak(*I, N, C); + } + } + C.addTransition(state->set<RegionState>(RS), N); +} + +void MallocChecker::checkEndPath(CheckerContext &C) const { + ProgramStateRef state = C.getState(); + RegionStateTy M = state->get<RegionState>(); + + // If inside inlined call, skip it. + if (C.getLocationContext()->getParent() != 0) + return; + + for (RegionStateTy::iterator I = M.begin(), E = M.end(); I != E; ++I) { + RefState RS = I->second; + if (RS.isAllocated()) { + ExplodedNode *N = C.addTransition(state); + if (N) + reportLeak(I->first, N, C); + } + } +} + +bool MallocChecker::checkEscape(SymbolRef Sym, const Stmt *S, + CheckerContext &C) const { + ProgramStateRef state = C.getState(); + const RefState *RS = state->get<RegionState>(Sym); + if (!RS) + return false; + + if (RS->isAllocated()) { + state = state->set<RegionState>(Sym, RefState::getEscaped(S)); + C.addTransition(state); + return true; + } + return false; +} + +void MallocChecker::checkPreStmt(const CallExpr *CE, CheckerContext &C) const { + if (isMemFunction(C.getCalleeDecl(CE), C.getASTContext())) + return; + + // Check use after free, when a freed pointer is passed to a call. + ProgramStateRef State = C.getState(); + for (CallExpr::const_arg_iterator I = CE->arg_begin(), + E = CE->arg_end(); I != E; ++I) { + const Expr *A = *I; + if (A->getType().getTypePtr()->isAnyPointerType()) { + SymbolRef Sym = State->getSVal(A, C.getLocationContext()).getAsSymbol(); + if (!Sym) + continue; + if (checkUseAfterFree(Sym, C, A)) + return; + } + } +} + +void MallocChecker::checkPreStmt(const ReturnStmt *S, CheckerContext &C) const { + const Expr *E = S->getRetValue(); + if (!E) + return; + + // Check if we are returning a symbol. + SVal RetVal = C.getState()->getSVal(E, C.getLocationContext()); + SymbolRef Sym = RetVal.getAsSymbol(); + if (!Sym) + // If we are returning a field of the allocated struct or an array element, + // the callee could still free the memory. + // TODO: This logic should be a part of generic symbol escape callback. + if (const MemRegion *MR = RetVal.getAsRegion()) + if (isa<FieldRegion>(MR) || isa<ElementRegion>(MR)) + if (const SymbolicRegion *BMR = + dyn_cast<SymbolicRegion>(MR->getBaseRegion())) + Sym = BMR->getSymbol(); + if (!Sym) + return; + + // Check if we are returning freed memory. + if (checkUseAfterFree(Sym, C, E)) + return; + + // If this function body is not inlined, check if the symbol is escaping. + if (C.getLocationContext()->getParent() == 0) + checkEscape(Sym, E, C); +} + +// TODO: Blocks should be either inlined or should call invalidate regions +// upon invocation. After that's in place, special casing here will not be +// needed. +void MallocChecker::checkPostStmt(const BlockExpr *BE, + CheckerContext &C) const { + + // Scan the BlockDecRefExprs for any object the retain count checker + // may be tracking. + if (!BE->getBlockDecl()->hasCaptures()) + return; + + ProgramStateRef state = C.getState(); + const BlockDataRegion *R = + cast<BlockDataRegion>(state->getSVal(BE, + C.getLocationContext()).getAsRegion()); + + BlockDataRegion::referenced_vars_iterator I = R->referenced_vars_begin(), + E = R->referenced_vars_end(); + + if (I == E) + return; + + SmallVector<const MemRegion*, 10> Regions; + const LocationContext *LC = C.getLocationContext(); + MemRegionManager &MemMgr = C.getSValBuilder().getRegionManager(); + + for ( ; I != E; ++I) { + const VarRegion *VR = *I; + if (VR->getSuperRegion() == R) { + VR = MemMgr.getVarRegion(VR->getDecl(), LC); + } + Regions.push_back(VR); + } + + state = + state->scanReachableSymbols<StopTrackingCallback>(Regions.data(), + Regions.data() + Regions.size()).getState(); + C.addTransition(state); +} + +bool MallocChecker::checkUseAfterFree(SymbolRef Sym, CheckerContext &C, + const Stmt *S) const { + assert(Sym); + const RefState *RS = C.getState()->get<RegionState>(Sym); + if (RS && RS->isReleased()) { + if (ExplodedNode *N = C.generateSink()) { + if (!BT_UseFree) + BT_UseFree.reset(new BugType("Use-after-free", "Memory Error")); + + BugReport *R = new BugReport(*BT_UseFree, + "Use of memory after it is freed",N); + if (S) + R->addRange(S->getSourceRange()); + R->markInteresting(Sym); + R->addVisitor(new MallocBugVisitor(Sym)); + C.EmitReport(R); + return true; + } + } + return false; +} + +// Check if the location is a freed symbolic region. +void MallocChecker::checkLocation(SVal l, bool isLoad, const Stmt *S, + CheckerContext &C) const { + SymbolRef Sym = l.getLocSymbolInBase(); + if (Sym) + checkUseAfterFree(Sym, C); +} + +//===----------------------------------------------------------------------===// +// Check various ways a symbol can be invalidated. +// TODO: This logic (the next 3 functions) is copied/similar to the +// RetainRelease checker. We might want to factor this out. +//===----------------------------------------------------------------------===// + +// Stop tracking symbols when a value escapes as a result of checkBind. +// A value escapes in three possible cases: +// (1) we are binding to something that is not a memory region. +// (2) we are binding to a memregion that does not have stack storage +// (3) we are binding to a memregion with stack storage that the store +// does not understand. +void MallocChecker::checkBind(SVal loc, SVal val, const Stmt *S, + CheckerContext &C) const { + // Are we storing to something that causes the value to "escape"? + bool escapes = true; + ProgramStateRef state = C.getState(); + + if (loc::MemRegionVal *regionLoc = dyn_cast<loc::MemRegionVal>(&loc)) { + escapes = !regionLoc->getRegion()->hasStackStorage(); + + if (!escapes) { + // To test (3), generate a new state with the binding added. If it is + // the same state, then it escapes (since the store cannot represent + // the binding). + escapes = (state == (state->bindLoc(*regionLoc, val))); + } + if (!escapes) { + // Case 4: We do not currently model what happens when a symbol is + // assigned to a struct field, so be conservative here and let the symbol + // go. TODO: This could definitely be improved upon. + escapes = !isa<VarRegion>(regionLoc->getRegion()); + } + } + + // If our store can represent the binding and we aren't storing to something + // that doesn't have local storage then just return and have the simulation + // state continue as is. + if (!escapes) + return; + + // Otherwise, find all symbols referenced by 'val' that we are tracking + // and stop tracking them. + state = state->scanReachableSymbols<StopTrackingCallback>(val).getState(); + C.addTransition(state); +} + +// If a symbolic region is assumed to NULL (or another constant), stop tracking +// it - assuming that allocation failed on this path. +ProgramStateRef MallocChecker::evalAssume(ProgramStateRef state, + SVal Cond, + bool Assumption) const { + RegionStateTy RS = state->get<RegionState>(); + for (RegionStateTy::iterator I = RS.begin(), E = RS.end(); I != E; ++I) { + // If the symbol is assumed to NULL or another constant, this will + // return an APSInt*. + if (state->getSymVal(I.getKey())) + state = state->remove<RegionState>(I.getKey()); + } + + // Realloc returns 0 when reallocation fails, which means that we should + // restore the state of the pointer being reallocated. + ReallocMap RP = state->get<ReallocPairs>(); + for (ReallocMap::iterator I = RP.begin(), E = RP.end(); I != E; ++I) { + // If the symbol is assumed to NULL or another constant, this will + // return an APSInt*. + if (state->getSymVal(I.getKey())) { + SymbolRef ReallocSym = I.getData().ReallocatedSym; + const RefState *RS = state->get<RegionState>(ReallocSym); + if (RS) { + if (RS->isReleased() && ! I.getData().IsFreeOnFailure) + state = state->set<RegionState>(ReallocSym, + RefState::getAllocateUnchecked(RS->getStmt())); + } + state = state->remove<ReallocPairs>(I.getKey()); + } + } + + return state; +} + +// Check if the function is known to us. So, for example, we could +// conservatively assume it can free/reallocate it's pointer arguments. +// (We assume that the pointers cannot escape through calls to system +// functions not handled by this checker.) +bool MallocChecker::doesNotFreeMemory(const CallOrObjCMessage *Call, + ProgramStateRef State) const { + if (!Call) + return false; + + // For now, assume that any C++ call can free memory. + // TODO: If we want to be more optimistic here, we'll need to make sure that + // regions escape to C++ containers. They seem to do that even now, but for + // mysterious reasons. + if (Call->isCXXCall()) + return false; + + const Decl *D = Call->getDecl(); + if (!D) + return false; + + ASTContext &ASTC = State->getStateManager().getContext(); + + // If it's one of the allocation functions we can reason about, we model + // its behavior explicitly. + if (isa<FunctionDecl>(D) && isMemFunction(cast<FunctionDecl>(D), ASTC)) { + return true; + } + + // If it's not a system call, assume it frees memory. + SourceManager &SM = ASTC.getSourceManager(); + if (!SM.isInSystemHeader(D->getLocation())) + return false; + + // Process C/ObjC functions. + if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(D)) { + // White list the system functions whose arguments escape. + const IdentifierInfo *II = FD->getIdentifier(); + if (!II) + return true; + StringRef FName = II->getName(); + + // White list thread local storage. + if (FName.equals("pthread_setspecific")) + return false; + + // White list the 'XXXNoCopy' ObjC functions. + if (FName.endswith("NoCopy")) { + // Look for the deallocator argument. We know that the memory ownership + // is not transfered only if the deallocator argument is + // 'kCFAllocatorNull'. + for (unsigned i = 1; i < Call->getNumArgs(); ++i) { + const Expr *ArgE = Call->getArg(i)->IgnoreParenCasts(); + if (const DeclRefExpr *DE = dyn_cast<DeclRefExpr>(ArgE)) { + StringRef DeallocatorName = DE->getFoundDecl()->getName(); + if (DeallocatorName == "kCFAllocatorNull") + return true; + } + } + return false; + } + + // PR12101 + // Many CoreFoundation and CoreGraphics might allow a tracked object + // to escape. + if (Call->isCFCGAllowingEscape(FName)) + return false; + + // Associating streams with malloced buffers. The pointer can escape if + // 'closefn' is specified (and if that function does free memory). + // Currently, we do not inspect the 'closefn' function (PR12101). + if (FName == "funopen") + if (Call->getNumArgs() >= 4 && !Call->getArgSVal(4).isConstant(0)) + return false; + + // Do not warn on pointers passed to 'setbuf' when used with std streams, + // these leaks might be intentional when setting the buffer for stdio. + // http://stackoverflow.com/questions/2671151/who-frees-setvbuf-buffer + if (FName == "setbuf" || FName =="setbuffer" || + FName == "setlinebuf" || FName == "setvbuf") { + if (Call->getNumArgs() >= 1) + if (const DeclRefExpr *Arg = + dyn_cast<DeclRefExpr>(Call->getArg(0)->IgnoreParenCasts())) + if (const VarDecl *D = dyn_cast<VarDecl>(Arg->getDecl())) + if (D->getCanonicalDecl()->getName().find("std") + != StringRef::npos) + return false; + } + + // A bunch of other functions, which take ownership of a pointer (See retain + // release checker). Not all the parameters here are invalidated, but the + // Malloc checker cannot differentiate between them. The right way of doing + // this would be to implement a pointer escapes callback. + if (FName == "CVPixelBufferCreateWithBytes" || + FName == "CGBitmapContextCreateWithData" || + FName == "CVPixelBufferCreateWithPlanarBytes" || + FName == "OSAtomicEnqueue") { + return false; + } + + // Whitelist NSXXInsertXX, for example NSMapInsertIfAbsent, since they can + // be deallocated by NSMapRemove. + if (FName.startswith("NS") && (FName.find("Insert") != StringRef::npos)) + return false; + + // Otherwise, assume that the function does not free memory. + // Most system calls, do not free the memory. + return true; + + // Process ObjC functions. + } else if (const ObjCMethodDecl * ObjCD = dyn_cast<ObjCMethodDecl>(D)) { + Selector S = ObjCD->getSelector(); + + // White list the ObjC functions which do free memory. + // - Anything containing 'freeWhenDone' param set to 1. + // Ex: dataWithBytesNoCopy:length:freeWhenDone. + for (unsigned i = 1; i < S.getNumArgs(); ++i) { + if (S.getNameForSlot(i).equals("freeWhenDone")) { + if (Call->getArgSVal(i).isConstant(1)) + return false; + else + return true; + } + } + + // If the first selector ends with NoCopy, assume that the ownership is + // transfered as well. + // Ex: [NSData dataWithBytesNoCopy:bytes length:10]; + if (S.getNameForSlot(0).endswith("NoCopy")) { + return false; + } + + // Otherwise, assume that the function does not free memory. + // Most system calls, do not free the memory. + return true; + } + + // Otherwise, assume that the function can free memory. + return false; + +} + +// If the symbol we are tracking is invalidated, but not explicitly (ex: the &p +// escapes, when we are tracking p), do not track the symbol as we cannot reason +// about it anymore. +ProgramStateRef +MallocChecker::checkRegionChanges(ProgramStateRef State, + const StoreManager::InvalidatedSymbols *invalidated, + ArrayRef<const MemRegion *> ExplicitRegions, + ArrayRef<const MemRegion *> Regions, + const CallOrObjCMessage *Call) const { + if (!invalidated || invalidated->empty()) + return State; + llvm::SmallPtrSet<SymbolRef, 8> WhitelistedSymbols; + + // If it's a call which might free or reallocate memory, we assume that all + // regions (explicit and implicit) escaped. + + // Otherwise, whitelist explicit pointers; we still can track them. + if (!Call || doesNotFreeMemory(Call, State)) { + for (ArrayRef<const MemRegion *>::iterator I = ExplicitRegions.begin(), + E = ExplicitRegions.end(); I != E; ++I) { + if (const SymbolicRegion *R = (*I)->StripCasts()->getAs<SymbolicRegion>()) + WhitelistedSymbols.insert(R->getSymbol()); + } + } + + for (StoreManager::InvalidatedSymbols::const_iterator I=invalidated->begin(), + E = invalidated->end(); I!=E; ++I) { + SymbolRef sym = *I; + if (WhitelistedSymbols.count(sym)) + continue; + // The symbol escaped. + if (const RefState *RS = State->get<RegionState>(sym)) + State = State->set<RegionState>(sym, RefState::getEscaped(RS->getStmt())); + } + return State; +} + +static SymbolRef findFailedReallocSymbol(ProgramStateRef currState, + ProgramStateRef prevState) { + ReallocMap currMap = currState->get<ReallocPairs>(); + ReallocMap prevMap = prevState->get<ReallocPairs>(); + + for (ReallocMap::iterator I = prevMap.begin(), E = prevMap.end(); + I != E; ++I) { + SymbolRef sym = I.getKey(); + if (!currMap.lookup(sym)) + return sym; + } + + return NULL; +} + +PathDiagnosticPiece * +MallocChecker::MallocBugVisitor::VisitNode(const ExplodedNode *N, + const ExplodedNode *PrevN, + BugReporterContext &BRC, + BugReport &BR) { + ProgramStateRef state = N->getState(); + ProgramStateRef statePrev = PrevN->getState(); + + const RefState *RS = state->get<RegionState>(Sym); + const RefState *RSPrev = statePrev->get<RegionState>(Sym); + if (!RS && !RSPrev) + return 0; + + const Stmt *S = 0; + const char *Msg = 0; + StackHintGeneratorForSymbol *StackHint = 0; + + // Retrieve the associated statement. + ProgramPoint ProgLoc = N->getLocation(); + if (isa<StmtPoint>(ProgLoc)) + S = cast<StmtPoint>(ProgLoc).getStmt(); + // If an assumption was made on a branch, it should be caught + // here by looking at the state transition. + if (isa<BlockEdge>(ProgLoc)) { + const CFGBlock *srcBlk = cast<BlockEdge>(ProgLoc).getSrc(); + S = srcBlk->getTerminator(); + } + if (!S) + return 0; + + // Find out if this is an interesting point and what is the kind. + if (Mode == Normal) { + if (isAllocated(RS, RSPrev, S)) { + Msg = "Memory is allocated"; + StackHint = new StackHintGeneratorForSymbol(Sym, + "Returned allocated memory"); + } else if (isReleased(RS, RSPrev, S)) { + Msg = "Memory is released"; + StackHint = new StackHintGeneratorForSymbol(Sym, + "Returned released memory"); + } else if (isReallocFailedCheck(RS, RSPrev, S)) { + Mode = ReallocationFailed; + Msg = "Reallocation failed"; + StackHint = new StackHintGeneratorForReallocationFailed(Sym, + "Reallocation failed"); + + if (SymbolRef sym = findFailedReallocSymbol(state, statePrev)) { + // Is it possible to fail two reallocs WITHOUT testing in between? + assert((!FailedReallocSymbol || FailedReallocSymbol == sym) && + "We only support one failed realloc at a time."); + BR.markInteresting(sym); + FailedReallocSymbol = sym; + } + } + + // We are in a special mode if a reallocation failed later in the path. + } else if (Mode == ReallocationFailed) { + assert(FailedReallocSymbol && "No symbol to look for."); + + // Is this is the first appearance of the reallocated symbol? + if (!statePrev->get<RegionState>(FailedReallocSymbol)) { + // If we ever hit this assert, that means BugReporter has decided to skip + // node pairs or visit them out of order. + assert(state->get<RegionState>(FailedReallocSymbol) && + "Missed the reallocation point"); + + // We're at the reallocation point. + Msg = "Attempt to reallocate memory"; + StackHint = new StackHintGeneratorForSymbol(Sym, + "Returned reallocated memory"); + FailedReallocSymbol = NULL; + Mode = Normal; + } + } + + if (!Msg) + return 0; + assert(StackHint); + + // Generate the extra diagnostic. + PathDiagnosticLocation Pos(S, BRC.getSourceManager(), + N->getLocationContext()); + return new PathDiagnosticEventPiece(Pos, Msg, true, StackHint); +} + + +#define REGISTER_CHECKER(name) \ +void ento::register##name(CheckerManager &mgr) {\ + registerCStringCheckerBasic(mgr); \ + mgr.registerChecker<MallocChecker>()->Filter.C##name = true;\ +} + +REGISTER_CHECKER(MallocPessimistic) +REGISTER_CHECKER(MallocOptimistic) diff --git a/clang/lib/StaticAnalyzer/Checkers/MallocOverflowSecurityChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/MallocOverflowSecurityChecker.cpp new file mode 100644 index 0000000..daec418 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/MallocOverflowSecurityChecker.cpp @@ -0,0 +1,267 @@ +// MallocOverflowSecurityChecker.cpp - Check for malloc overflows -*- C++ -*-=// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This checker detects a common memory allocation security flaw. +// Suppose 'unsigned int n' comes from an untrusted source. If the +// code looks like 'malloc (n * 4)', and an attacker can make 'n' be +// say MAX_UINT/4+2, then instead of allocating the correct 'n' 4-byte +// elements, this will actually allocate only two because of overflow. +// Then when the rest of the program attempts to store values past the +// second element, these values will actually overwrite other items in +// the heap, probably allowing the attacker to execute arbitrary code. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/AST/EvaluatedExprVisitor.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/AnalysisManager.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugReporter.h" +#include "llvm/ADT/SmallVector.h" + +using namespace clang; +using namespace ento; + +namespace { +struct MallocOverflowCheck { + const BinaryOperator *mulop; + const Expr *variable; + + MallocOverflowCheck (const BinaryOperator *m, const Expr *v) + : mulop(m), variable (v) + {} +}; + +class MallocOverflowSecurityChecker : public Checker<check::ASTCodeBody> { +public: + void checkASTCodeBody(const Decl *D, AnalysisManager &mgr, + BugReporter &BR) const; + + void CheckMallocArgument( + llvm::SmallVectorImpl<MallocOverflowCheck> &PossibleMallocOverflows, + const Expr *TheArgument, ASTContext &Context) const; + + void OutputPossibleOverflows( + llvm::SmallVectorImpl<MallocOverflowCheck> &PossibleMallocOverflows, + const Decl *D, BugReporter &BR, AnalysisManager &mgr) const; + +}; +} // end anonymous namespace + +void MallocOverflowSecurityChecker::CheckMallocArgument( + llvm::SmallVectorImpl<MallocOverflowCheck> &PossibleMallocOverflows, + const Expr *TheArgument, + ASTContext &Context) const { + + /* Look for a linear combination with a single variable, and at least + one multiplication. + Reject anything that applies to the variable: an explicit cast, + conditional expression, an operation that could reduce the range + of the result, or anything too complicated :-). */ + const Expr * e = TheArgument; + const BinaryOperator * mulop = NULL; + + for (;;) { + e = e->IgnoreParenImpCasts(); + if (isa<BinaryOperator>(e)) { + const BinaryOperator * binop = dyn_cast<BinaryOperator>(e); + BinaryOperatorKind opc = binop->getOpcode(); + // TODO: ignore multiplications by 1, reject if multiplied by 0. + if (mulop == NULL && opc == BO_Mul) + mulop = binop; + if (opc != BO_Mul && opc != BO_Add && opc != BO_Sub && opc != BO_Shl) + return; + + const Expr *lhs = binop->getLHS(); + const Expr *rhs = binop->getRHS(); + if (rhs->isEvaluatable(Context)) + e = lhs; + else if ((opc == BO_Add || opc == BO_Mul) + && lhs->isEvaluatable(Context)) + e = rhs; + else + return; + } + else if (isa<DeclRefExpr>(e) || isa<MemberExpr>(e)) + break; + else + return; + } + + if (mulop == NULL) + return; + + // We've found the right structure of malloc argument, now save + // the data so when the body of the function is completely available + // we can check for comparisons. + + // TODO: Could push this into the innermost scope where 'e' is + // defined, rather than the whole function. + PossibleMallocOverflows.push_back(MallocOverflowCheck(mulop, e)); +} + +namespace { +// A worker class for OutputPossibleOverflows. +class CheckOverflowOps : + public EvaluatedExprVisitor<CheckOverflowOps> { +public: + typedef llvm::SmallVectorImpl<MallocOverflowCheck> theVecType; + +private: + theVecType &toScanFor; + ASTContext &Context; + + bool isIntZeroExpr(const Expr *E) const { + if (!E->getType()->isIntegralOrEnumerationType()) + return false; + llvm::APSInt Result; + if (E->EvaluateAsInt(Result, Context)) + return Result == 0; + return false; + } + + void CheckExpr(const Expr *E_p) { + const Expr *E = E_p->IgnoreParenImpCasts(); + + theVecType::iterator i = toScanFor.end(); + theVecType::iterator e = toScanFor.begin(); + + if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(E)) { + const Decl * EdreD = DR->getDecl(); + while (i != e) { + --i; + if (const DeclRefExpr *DR_i = dyn_cast<DeclRefExpr>(i->variable)) { + if (DR_i->getDecl() == EdreD) + i = toScanFor.erase(i); + } + } + } + else if (isa<MemberExpr>(E)) { + // No points-to analysis, just look at the member + const Decl * EmeMD = dyn_cast<MemberExpr>(E)->getMemberDecl(); + while (i != e) { + --i; + if (isa<MemberExpr>(i->variable)) { + if (dyn_cast<MemberExpr>(i->variable)->getMemberDecl() == EmeMD) + i = toScanFor.erase (i); + } + } + } + } + + public: + void VisitBinaryOperator(BinaryOperator *E) { + if (E->isComparisonOp()) { + const Expr * lhs = E->getLHS(); + const Expr * rhs = E->getRHS(); + // Ignore comparisons against zero, since they generally don't + // protect against an overflow. + if (!isIntZeroExpr(lhs) && ! isIntZeroExpr(rhs)) { + CheckExpr(lhs); + CheckExpr(rhs); + } + } + EvaluatedExprVisitor<CheckOverflowOps>::VisitBinaryOperator(E); + } + + /* We specifically ignore loop conditions, because they're typically + not error checks. */ + void VisitWhileStmt(WhileStmt *S) { + return this->Visit(S->getBody()); + } + void VisitForStmt(ForStmt *S) { + return this->Visit(S->getBody()); + } + void VisitDoStmt(DoStmt *S) { + return this->Visit(S->getBody()); + } + + CheckOverflowOps(theVecType &v, ASTContext &ctx) + : EvaluatedExprVisitor<CheckOverflowOps>(ctx), + toScanFor(v), Context(ctx) + { } + }; +} + +// OutputPossibleOverflows - We've found a possible overflow earlier, +// now check whether Body might contain a comparison which might be +// preventing the overflow. +// This doesn't do flow analysis, range analysis, or points-to analysis; it's +// just a dumb "is there a comparison" scan. The aim here is to +// detect the most blatent cases of overflow and educate the +// programmer. +void MallocOverflowSecurityChecker::OutputPossibleOverflows( + llvm::SmallVectorImpl<MallocOverflowCheck> &PossibleMallocOverflows, + const Decl *D, BugReporter &BR, AnalysisManager &mgr) const { + // By far the most common case: nothing to check. + if (PossibleMallocOverflows.empty()) + return; + + // Delete any possible overflows which have a comparison. + CheckOverflowOps c(PossibleMallocOverflows, BR.getContext()); + c.Visit(mgr.getAnalysisDeclContext(D)->getBody()); + + // Output warnings for all overflows that are left. + for (CheckOverflowOps::theVecType::iterator + i = PossibleMallocOverflows.begin(), + e = PossibleMallocOverflows.end(); + i != e; + ++i) { + SourceRange R = i->mulop->getSourceRange(); + BR.EmitBasicReport(D, "malloc() size overflow", categories::UnixAPI, + "the computation of the size of the memory allocation may overflow", + PathDiagnosticLocation::createOperatorLoc(i->mulop, + BR.getSourceManager()), &R, 1); + } +} + +void MallocOverflowSecurityChecker::checkASTCodeBody(const Decl *D, + AnalysisManager &mgr, + BugReporter &BR) const { + + CFG *cfg = mgr.getCFG(D); + if (!cfg) + return; + + // A list of variables referenced in possibly overflowing malloc operands. + llvm::SmallVector<MallocOverflowCheck, 2> PossibleMallocOverflows; + + for (CFG::iterator it = cfg->begin(), ei = cfg->end(); it != ei; ++it) { + CFGBlock *block = *it; + for (CFGBlock::iterator bi = block->begin(), be = block->end(); + bi != be; ++bi) { + if (const CFGStmt *CS = bi->getAs<CFGStmt>()) { + if (const CallExpr *TheCall = dyn_cast<CallExpr>(CS->getStmt())) { + // Get the callee. + const FunctionDecl *FD = TheCall->getDirectCallee(); + + if (!FD) + return; + + // Get the name of the callee. If it's a builtin, strip off the prefix. + IdentifierInfo *FnInfo = FD->getIdentifier(); + if (!FnInfo) + return; + + if (FnInfo->isStr ("malloc") || FnInfo->isStr ("_MALLOC")) { + if (TheCall->getNumArgs() == 1) + CheckMallocArgument(PossibleMallocOverflows, TheCall->getArg(0), + mgr.getASTContext()); + } + } + } + } + } + + OutputPossibleOverflows(PossibleMallocOverflows, D, BR, mgr); +} + +void ento::registerMallocOverflowSecurityChecker(CheckerManager &mgr) { + mgr.registerChecker<MallocOverflowSecurityChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/MallocSizeofChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/MallocSizeofChecker.cpp new file mode 100644 index 0000000..08a9da1 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/MallocSizeofChecker.cpp @@ -0,0 +1,211 @@ +// MallocSizeofChecker.cpp - Check for dubious malloc arguments ---*- C++ -*-=// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// Reports inconsistencies between the casted type of the return value of a +// malloc/calloc/realloc call and the operand of any sizeof expressions +// contained within its argument(s). +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugReporter.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/AnalysisManager.h" +#include "clang/AST/StmtVisitor.h" +#include "clang/AST/TypeLoc.h" +#include "llvm/ADT/SmallString.h" + +using namespace clang; +using namespace ento; + +namespace { + +typedef std::pair<const TypeSourceInfo *, const CallExpr *> TypeCallPair; +typedef llvm::PointerUnion<const Stmt *, const VarDecl *> ExprParent; + +class CastedAllocFinder + : public ConstStmtVisitor<CastedAllocFinder, TypeCallPair> { + IdentifierInfo *II_malloc, *II_calloc, *II_realloc; + +public: + struct CallRecord { + ExprParent CastedExprParent; + const Expr *CastedExpr; + const TypeSourceInfo *ExplicitCastType; + const CallExpr *AllocCall; + + CallRecord(ExprParent CastedExprParent, const Expr *CastedExpr, + const TypeSourceInfo *ExplicitCastType, + const CallExpr *AllocCall) + : CastedExprParent(CastedExprParent), CastedExpr(CastedExpr), + ExplicitCastType(ExplicitCastType), AllocCall(AllocCall) {} + }; + + typedef std::vector<CallRecord> CallVec; + CallVec Calls; + + CastedAllocFinder(ASTContext *Ctx) : + II_malloc(&Ctx->Idents.get("malloc")), + II_calloc(&Ctx->Idents.get("calloc")), + II_realloc(&Ctx->Idents.get("realloc")) {} + + void VisitChild(ExprParent Parent, const Stmt *S) { + TypeCallPair AllocCall = Visit(S); + if (AllocCall.second && AllocCall.second != S) + Calls.push_back(CallRecord(Parent, cast<Expr>(S), AllocCall.first, + AllocCall.second)); + } + + void VisitChildren(const Stmt *S) { + for (Stmt::const_child_iterator I = S->child_begin(), E = S->child_end(); + I!=E; ++I) + if (const Stmt *child = *I) + VisitChild(S, child); + } + + TypeCallPair VisitCastExpr(const CastExpr *E) { + return Visit(E->getSubExpr()); + } + + TypeCallPair VisitExplicitCastExpr(const ExplicitCastExpr *E) { + return TypeCallPair(E->getTypeInfoAsWritten(), + Visit(E->getSubExpr()).second); + } + + TypeCallPair VisitParenExpr(const ParenExpr *E) { + return Visit(E->getSubExpr()); + } + + TypeCallPair VisitStmt(const Stmt *S) { + VisitChildren(S); + return TypeCallPair(); + } + + TypeCallPair VisitCallExpr(const CallExpr *E) { + VisitChildren(E); + const FunctionDecl *FD = E->getDirectCallee(); + if (FD) { + IdentifierInfo *II = FD->getIdentifier(); + if (II == II_malloc || II == II_calloc || II == II_realloc) + return TypeCallPair((const TypeSourceInfo *)0, E); + } + return TypeCallPair(); + } + + TypeCallPair VisitDeclStmt(const DeclStmt *S) { + for (DeclStmt::const_decl_iterator I = S->decl_begin(), E = S->decl_end(); + I!=E; ++I) + if (const VarDecl *VD = dyn_cast<VarDecl>(*I)) + if (const Expr *Init = VD->getInit()) + VisitChild(VD, Init); + return TypeCallPair(); + } +}; + +class SizeofFinder : public ConstStmtVisitor<SizeofFinder> { +public: + std::vector<const UnaryExprOrTypeTraitExpr *> Sizeofs; + + void VisitBinMul(const BinaryOperator *E) { + Visit(E->getLHS()); + Visit(E->getRHS()); + } + + void VisitBinAdd(const BinaryOperator *E) { + Visit(E->getLHS()); + Visit(E->getRHS()); + } + + void VisitImplicitCastExpr(const ImplicitCastExpr *E) { + return Visit(E->getSubExpr()); + } + + void VisitParenExpr(const ParenExpr *E) { + return Visit(E->getSubExpr()); + } + + void VisitUnaryExprOrTypeTraitExpr(const UnaryExprOrTypeTraitExpr *E) { + if (E->getKind() != UETT_SizeOf) + return; + + Sizeofs.push_back(E); + } +}; + +class MallocSizeofChecker : public Checker<check::ASTCodeBody> { +public: + void checkASTCodeBody(const Decl *D, AnalysisManager& mgr, + BugReporter &BR) const { + AnalysisDeclContext *ADC = mgr.getAnalysisDeclContext(D); + CastedAllocFinder Finder(&BR.getContext()); + Finder.Visit(D->getBody()); + for (CastedAllocFinder::CallVec::iterator i = Finder.Calls.begin(), + e = Finder.Calls.end(); i != e; ++i) { + QualType CastedType = i->CastedExpr->getType(); + if (!CastedType->isPointerType()) + continue; + QualType PointeeType = CastedType->getAs<PointerType>()->getPointeeType(); + if (PointeeType->isVoidType()) + continue; + + for (CallExpr::const_arg_iterator ai = i->AllocCall->arg_begin(), + ae = i->AllocCall->arg_end(); ai != ae; ++ai) { + if (!(*ai)->getType()->isIntegerType()) + continue; + + SizeofFinder SFinder; + SFinder.Visit(*ai); + if (SFinder.Sizeofs.size() != 1) + continue; + + QualType SizeofType = SFinder.Sizeofs[0]->getTypeOfArgument(); + if (!BR.getContext().hasSameUnqualifiedType(PointeeType, SizeofType)) { + const TypeSourceInfo *TSI = 0; + if (i->CastedExprParent.is<const VarDecl *>()) { + TSI = + i->CastedExprParent.get<const VarDecl *>()->getTypeSourceInfo(); + } else { + TSI = i->ExplicitCastType; + } + + SmallString<64> buf; + llvm::raw_svector_ostream OS(buf); + + OS << "Result of '" + << i->AllocCall->getDirectCallee()->getIdentifier()->getName() + << "' is converted to type '" + << CastedType.getAsString() << "', whose pointee type '" + << PointeeType.getAsString() << "' is incompatible with " + << "sizeof operand type '" << SizeofType.getAsString() << "'"; + llvm::SmallVector<SourceRange, 4> Ranges; + Ranges.push_back(i->AllocCall->getCallee()->getSourceRange()); + Ranges.push_back(SFinder.Sizeofs[0]->getSourceRange()); + if (TSI) + Ranges.push_back(TSI->getTypeLoc().getSourceRange()); + + PathDiagnosticLocation L = + PathDiagnosticLocation::createBegin(i->AllocCall->getCallee(), + BR.getSourceManager(), ADC); + + BR.EmitBasicReport(D, "allocator sizeof operand mismatch", + categories::UnixAPI, + OS.str(), + L, Ranges.data(), Ranges.size()); + } + } + } + } +}; + +} + +void ento::registerMallocSizeofChecker(CheckerManager &mgr) { + mgr.registerChecker<MallocSizeofChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/NSAutoreleasePoolChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/NSAutoreleasePoolChecker.cpp new file mode 100644 index 0000000..4989ba8 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/NSAutoreleasePoolChecker.cpp @@ -0,0 +1,89 @@ +//=- NSAutoreleasePoolChecker.cpp --------------------------------*- C++ -*-==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines a NSAutoreleasePoolChecker, a small checker that warns +// about subpar uses of NSAutoreleasePool. Note that while the check itself +// (in its current form) could be written as a flow-insensitive check, in +// can be potentially enhanced in the future with flow-sensitive information. +// It is also a good example of the CheckerVisitor interface. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugReporter.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ExprEngine.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ObjCMessage.h" +#include "clang/AST/DeclObjC.h" +#include "clang/AST/Decl.h" + +using namespace clang; +using namespace ento; + +namespace { +class NSAutoreleasePoolChecker + : public Checker<check::PreObjCMessage> { + mutable OwningPtr<BugType> BT; + mutable Selector releaseS; + +public: + void checkPreObjCMessage(ObjCMessage msg, CheckerContext &C) const; +}; + +} // end anonymous namespace + +void NSAutoreleasePoolChecker::checkPreObjCMessage(ObjCMessage msg, + CheckerContext &C) const { + + const Expr *receiver = msg.getInstanceReceiver(); + if (!receiver) + return; + + // FIXME: Enhance with value-tracking information instead of consulting + // the type of the expression. + const ObjCObjectPointerType* PT = + receiver->getType()->getAs<ObjCObjectPointerType>(); + + if (!PT) + return; + const ObjCInterfaceDecl *OD = PT->getInterfaceDecl(); + if (!OD) + return; + if (!OD->getIdentifier()->getName().equals("NSAutoreleasePool")) + return; + + if (releaseS.isNull()) + releaseS = GetNullarySelector("release", C.getASTContext()); + // Sending 'release' message? + if (msg.getSelector() != releaseS) + return; + + if (!BT) + BT.reset(new BugType("Use -drain instead of -release", + "API Upgrade (Apple)")); + + ExplodedNode *N = C.addTransition(); + if (!N) { + assert(0); + return; + } + + BugReport *Report = new BugReport(*BT, "Use -drain instead of -release when " + "using NSAutoreleasePool and garbage collection", N); + Report->addRange(msg.getSourceRange()); + C.EmitReport(Report); +} + +void ento::registerNSAutoreleasePoolChecker(CheckerManager &mgr) { + if (mgr.getLangOpts().getGC() != LangOptions::NonGC) + mgr.registerChecker<NSAutoreleasePoolChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/NSErrorChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/NSErrorChecker.cpp new file mode 100644 index 0000000..f826573 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/NSErrorChecker.cpp @@ -0,0 +1,334 @@ +//=- NSErrorChecker.cpp - Coding conventions for uses of NSError -*- C++ -*-==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines a CheckNSError, a flow-insenstive check +// that determines if an Objective-C class interface correctly returns +// a non-void return type. +// +// File under feature request PR 2600. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramStateTrait.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/AST/DeclObjC.h" +#include "clang/AST/Decl.h" +#include "llvm/ADT/SmallVector.h" + +using namespace clang; +using namespace ento; + +static bool IsNSError(QualType T, IdentifierInfo *II); +static bool IsCFError(QualType T, IdentifierInfo *II); + +//===----------------------------------------------------------------------===// +// NSErrorMethodChecker +//===----------------------------------------------------------------------===// + +namespace { +class NSErrorMethodChecker + : public Checker< check::ASTDecl<ObjCMethodDecl> > { + mutable IdentifierInfo *II; + +public: + NSErrorMethodChecker() : II(0) { } + + void checkASTDecl(const ObjCMethodDecl *D, + AnalysisManager &mgr, BugReporter &BR) const; +}; +} + +void NSErrorMethodChecker::checkASTDecl(const ObjCMethodDecl *D, + AnalysisManager &mgr, + BugReporter &BR) const { + if (!D->isThisDeclarationADefinition()) + return; + if (!D->getResultType()->isVoidType()) + return; + + if (!II) + II = &D->getASTContext().Idents.get("NSError"); + + bool hasNSError = false; + for (ObjCMethodDecl::param_const_iterator + I = D->param_begin(), E = D->param_end(); I != E; ++I) { + if (IsNSError((*I)->getType(), II)) { + hasNSError = true; + break; + } + } + + if (hasNSError) { + const char *err = "Method accepting NSError** " + "should have a non-void return value to indicate whether or not an " + "error occurred"; + PathDiagnosticLocation L = + PathDiagnosticLocation::create(D, BR.getSourceManager()); + BR.EmitBasicReport(D, "Bad return type when passing NSError**", + "Coding conventions (Apple)", err, L); + } +} + +//===----------------------------------------------------------------------===// +// CFErrorFunctionChecker +//===----------------------------------------------------------------------===// + +namespace { +class CFErrorFunctionChecker + : public Checker< check::ASTDecl<FunctionDecl> > { + mutable IdentifierInfo *II; + +public: + CFErrorFunctionChecker() : II(0) { } + + void checkASTDecl(const FunctionDecl *D, + AnalysisManager &mgr, BugReporter &BR) const; +}; +} + +void CFErrorFunctionChecker::checkASTDecl(const FunctionDecl *D, + AnalysisManager &mgr, + BugReporter &BR) const { + if (!D->doesThisDeclarationHaveABody()) + return; + if (!D->getResultType()->isVoidType()) + return; + + if (!II) + II = &D->getASTContext().Idents.get("CFErrorRef"); + + bool hasCFError = false; + for (FunctionDecl::param_const_iterator + I = D->param_begin(), E = D->param_end(); I != E; ++I) { + if (IsCFError((*I)->getType(), II)) { + hasCFError = true; + break; + } + } + + if (hasCFError) { + const char *err = "Function accepting CFErrorRef* " + "should have a non-void return value to indicate whether or not an " + "error occurred"; + PathDiagnosticLocation L = + PathDiagnosticLocation::create(D, BR.getSourceManager()); + BR.EmitBasicReport(D, "Bad return type when passing CFErrorRef*", + "Coding conventions (Apple)", err, L); + } +} + +//===----------------------------------------------------------------------===// +// NSOrCFErrorDerefChecker +//===----------------------------------------------------------------------===// + +namespace { + +class NSErrorDerefBug : public BugType { +public: + NSErrorDerefBug() : BugType("NSError** null dereference", + "Coding conventions (Apple)") {} +}; + +class CFErrorDerefBug : public BugType { +public: + CFErrorDerefBug() : BugType("CFErrorRef* null dereference", + "Coding conventions (Apple)") {} +}; + +} + +namespace { +class NSOrCFErrorDerefChecker + : public Checker< check::Location, + check::Event<ImplicitNullDerefEvent> > { + mutable IdentifierInfo *NSErrorII, *CFErrorII; +public: + bool ShouldCheckNSError, ShouldCheckCFError; + NSOrCFErrorDerefChecker() : NSErrorII(0), CFErrorII(0), + ShouldCheckNSError(0), ShouldCheckCFError(0) { } + + void checkLocation(SVal loc, bool isLoad, const Stmt *S, + CheckerContext &C) const; + void checkEvent(ImplicitNullDerefEvent event) const; +}; +} + +namespace { struct NSErrorOut {}; } +namespace { struct CFErrorOut {}; } + +typedef llvm::ImmutableMap<SymbolRef, unsigned> ErrorOutFlag; + +namespace clang { +namespace ento { + template <> + struct ProgramStateTrait<NSErrorOut> : public ProgramStatePartialTrait<ErrorOutFlag> { + static void *GDMIndex() { static int index = 0; return &index; } + }; + template <> + struct ProgramStateTrait<CFErrorOut> : public ProgramStatePartialTrait<ErrorOutFlag> { + static void *GDMIndex() { static int index = 0; return &index; } + }; +} +} + +template <typename T> +static bool hasFlag(SVal val, ProgramStateRef state) { + if (SymbolRef sym = val.getAsSymbol()) + if (const unsigned *attachedFlags = state->get<T>(sym)) + return *attachedFlags; + return false; +} + +template <typename T> +static void setFlag(ProgramStateRef state, SVal val, CheckerContext &C) { + // We tag the symbol that the SVal wraps. + if (SymbolRef sym = val.getAsSymbol()) + C.addTransition(state->set<T>(sym, true)); +} + +static QualType parameterTypeFromSVal(SVal val, CheckerContext &C) { + const StackFrameContext * + SFC = C.getLocationContext()->getCurrentStackFrame(); + if (const loc::MemRegionVal* X = dyn_cast<loc::MemRegionVal>(&val)) { + const MemRegion* R = X->getRegion(); + if (const VarRegion *VR = R->getAs<VarRegion>()) + if (const StackArgumentsSpaceRegion * + stackReg = dyn_cast<StackArgumentsSpaceRegion>(VR->getMemorySpace())) + if (stackReg->getStackFrame() == SFC) + return VR->getValueType(); + } + + return QualType(); +} + +void NSOrCFErrorDerefChecker::checkLocation(SVal loc, bool isLoad, + const Stmt *S, + CheckerContext &C) const { + if (!isLoad) + return; + if (loc.isUndef() || !isa<Loc>(loc)) + return; + + ASTContext &Ctx = C.getASTContext(); + ProgramStateRef state = C.getState(); + + // If we are loading from NSError**/CFErrorRef* parameter, mark the resulting + // SVal so that we can later check it when handling the + // ImplicitNullDerefEvent event. + // FIXME: Cumbersome! Maybe add hook at construction of SVals at start of + // function ? + + QualType parmT = parameterTypeFromSVal(loc, C); + if (parmT.isNull()) + return; + + if (!NSErrorII) + NSErrorII = &Ctx.Idents.get("NSError"); + if (!CFErrorII) + CFErrorII = &Ctx.Idents.get("CFErrorRef"); + + if (ShouldCheckNSError && IsNSError(parmT, NSErrorII)) { + setFlag<NSErrorOut>(state, state->getSVal(cast<Loc>(loc)), C); + return; + } + + if (ShouldCheckCFError && IsCFError(parmT, CFErrorII)) { + setFlag<CFErrorOut>(state, state->getSVal(cast<Loc>(loc)), C); + return; + } +} + +void NSOrCFErrorDerefChecker::checkEvent(ImplicitNullDerefEvent event) const { + if (event.IsLoad) + return; + + SVal loc = event.Location; + ProgramStateRef state = event.SinkNode->getState(); + BugReporter &BR = *event.BR; + + bool isNSError = hasFlag<NSErrorOut>(loc, state); + bool isCFError = false; + if (!isNSError) + isCFError = hasFlag<CFErrorOut>(loc, state); + + if (!(isNSError || isCFError)) + return; + + // Storing to possible null NSError/CFErrorRef out parameter. + + // Emit an error. + std::string err; + llvm::raw_string_ostream os(err); + os << "Potential null dereference. According to coding standards "; + + if (isNSError) + os << "in 'Creating and Returning NSError Objects' the parameter '"; + else + os << "documented in CoreFoundation/CFError.h the parameter '"; + + os << "' may be null."; + + BugType *bug = 0; + if (isNSError) + bug = new NSErrorDerefBug(); + else + bug = new CFErrorDerefBug(); + BugReport *report = new BugReport(*bug, os.str(), + event.SinkNode); + BR.EmitReport(report); +} + +static bool IsNSError(QualType T, IdentifierInfo *II) { + + const PointerType* PPT = T->getAs<PointerType>(); + if (!PPT) + return false; + + const ObjCObjectPointerType* PT = + PPT->getPointeeType()->getAs<ObjCObjectPointerType>(); + + if (!PT) + return false; + + const ObjCInterfaceDecl *ID = PT->getInterfaceDecl(); + + // FIXME: Can ID ever be NULL? + if (ID) + return II == ID->getIdentifier(); + + return false; +} + +static bool IsCFError(QualType T, IdentifierInfo *II) { + const PointerType* PPT = T->getAs<PointerType>(); + if (!PPT) return false; + + const TypedefType* TT = PPT->getPointeeType()->getAs<TypedefType>(); + if (!TT) return false; + + return TT->getDecl()->getIdentifier() == II; +} + +void ento::registerNSErrorChecker(CheckerManager &mgr) { + mgr.registerChecker<NSErrorMethodChecker>(); + NSOrCFErrorDerefChecker * + checker = mgr.registerChecker<NSOrCFErrorDerefChecker>(); + checker->ShouldCheckNSError = true; +} + +void ento::registerCFErrorChecker(CheckerManager &mgr) { + mgr.registerChecker<CFErrorFunctionChecker>(); + NSOrCFErrorDerefChecker * + checker = mgr.registerChecker<NSOrCFErrorDerefChecker>(); + checker->ShouldCheckCFError = true; +} diff --git a/clang/lib/StaticAnalyzer/Checkers/NoReturnFunctionChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/NoReturnFunctionChecker.cpp new file mode 100644 index 0000000..c2d7c09 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/NoReturnFunctionChecker.cpp @@ -0,0 +1,146 @@ +//=== NoReturnFunctionChecker.cpp -------------------------------*- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This defines NoReturnFunctionChecker, which evaluates functions that do not +// return to the caller. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ObjCMessage.h" +#include "llvm/ADT/StringSwitch.h" +#include <cstdarg> + +using namespace clang; +using namespace ento; + +namespace { + +class NoReturnFunctionChecker : public Checker< check::PostStmt<CallExpr>, + check::PostObjCMessage > { +public: + void checkPostStmt(const CallExpr *CE, CheckerContext &C) const; + void checkPostObjCMessage(const ObjCMessage &msg, CheckerContext &C) const; +}; + +} + +void NoReturnFunctionChecker::checkPostStmt(const CallExpr *CE, + CheckerContext &C) const { + ProgramStateRef state = C.getState(); + const Expr *Callee = CE->getCallee(); + + bool BuildSinks = getFunctionExtInfo(Callee->getType()).getNoReturn(); + + if (!BuildSinks) { + SVal L = state->getSVal(Callee, C.getLocationContext()); + const FunctionDecl *FD = L.getAsFunctionDecl(); + if (!FD) + return; + + if (FD->getAttr<AnalyzerNoReturnAttr>()) + BuildSinks = true; + else if (const IdentifierInfo *II = FD->getIdentifier()) { + // HACK: Some functions are not marked noreturn, and don't return. + // Here are a few hardwired ones. If this takes too long, we can + // potentially cache these results. + BuildSinks + = llvm::StringSwitch<bool>(StringRef(II->getName())) + .Case("exit", true) + .Case("panic", true) + .Case("error", true) + .Case("Assert", true) + // FIXME: This is just a wrapper around throwing an exception. + // Eventually inter-procedural analysis should handle this easily. + .Case("ziperr", true) + .Case("assfail", true) + .Case("db_error", true) + .Case("__assert", true) + .Case("__assert_rtn", true) + .Case("__assert_fail", true) + .Case("dtrace_assfail", true) + .Case("yy_fatal_error", true) + .Case("_XCAssertionFailureHandler", true) + .Case("_DTAssertionFailureHandler", true) + .Case("_TSAssertionFailureHandler", true) + .Default(false); + } + } + + if (BuildSinks) + C.generateSink(); +} + +static bool END_WITH_NULL isMultiArgSelector(const Selector *Sel, ...) { + va_list argp; + va_start(argp, Sel); + + unsigned Slot = 0; + const char *Arg; + while ((Arg = va_arg(argp, const char *))) { + if (!Sel->getNameForSlot(Slot).equals(Arg)) + break; // still need to va_end! + ++Slot; + } + + va_end(argp); + + // We only succeeded if we made it to the end of the argument list. + return (Arg == NULL); +} + +void NoReturnFunctionChecker::checkPostObjCMessage(const ObjCMessage &Msg, + CheckerContext &C) const { + // HACK: This entire check is to handle two messages in the Cocoa frameworks: + // -[NSAssertionHandler + // handleFailureInMethod:object:file:lineNumber:description:] + // -[NSAssertionHandler + // handleFailureInFunction:file:lineNumber:description:] + // Eventually these should be annotated with __attribute__((noreturn)). + // Because ObjC messages use dynamic dispatch, it is not generally safe to + // assume certain methods can't return. In cases where it is definitely valid, + // see if you can mark the methods noreturn or analyzer_noreturn instead of + // adding more explicit checks to this method. + + if (!Msg.isInstanceMessage()) + return; + + const ObjCInterfaceDecl *Receiver = Msg.getReceiverInterface(); + if (!Receiver) + return; + if (!Receiver->getIdentifier()->isStr("NSAssertionHandler")) + return; + + Selector Sel = Msg.getSelector(); + switch (Sel.getNumArgs()) { + default: + return; + case 4: + if (!isMultiArgSelector(&Sel, "handleFailureInFunction", "file", + "lineNumber", "description", NULL)) + return; + break; + case 5: + if (!isMultiArgSelector(&Sel, "handleFailureInMethod", "object", "file", + "lineNumber", "description", NULL)) + return; + break; + } + + // If we got here, it's one of the messages we care about. + C.generateSink(); +} + + +void ento::registerNoReturnFunctionChecker(CheckerManager &mgr) { + mgr.registerChecker<NoReturnFunctionChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/OSAtomicChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/OSAtomicChecker.cpp new file mode 100644 index 0000000..7b724d2 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/OSAtomicChecker.cpp @@ -0,0 +1,218 @@ +//=== OSAtomicChecker.cpp - OSAtomic functions evaluator --------*- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This checker evaluates OSAtomic functions. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/Basic/Builtins.h" + +using namespace clang; +using namespace ento; + +namespace { + +class OSAtomicChecker : public Checker<eval::InlineCall> { +public: + bool inlineCall(const CallExpr *CE, ExprEngine &Eng, + ExplodedNode *Pred, ExplodedNodeSet &Dst) const; + +private: + bool evalOSAtomicCompareAndSwap(const CallExpr *CE, + ExprEngine &Eng, + ExplodedNode *Pred, + ExplodedNodeSet &Dst) const; +}; +} + +static StringRef getCalleeName(ProgramStateRef State, + const CallExpr *CE, + const LocationContext *LCtx) { + const Expr *Callee = CE->getCallee(); + SVal L = State->getSVal(Callee, LCtx); + const FunctionDecl *funDecl = L.getAsFunctionDecl(); + if (!funDecl) + return StringRef(); + IdentifierInfo *funI = funDecl->getIdentifier(); + if (!funI) + return StringRef(); + return funI->getName(); +} + +bool OSAtomicChecker::inlineCall(const CallExpr *CE, + ExprEngine &Eng, + ExplodedNode *Pred, + ExplodedNodeSet &Dst) const { + StringRef FName = getCalleeName(Pred->getState(), + CE, Pred->getLocationContext()); + if (FName.empty()) + return false; + + // Check for compare and swap. + if (FName.startswith("OSAtomicCompareAndSwap") || + FName.startswith("objc_atomicCompareAndSwap")) + return evalOSAtomicCompareAndSwap(CE, Eng, Pred, Dst); + + // FIXME: Other atomics. + return false; +} + +bool OSAtomicChecker::evalOSAtomicCompareAndSwap(const CallExpr *CE, + ExprEngine &Eng, + ExplodedNode *Pred, + ExplodedNodeSet &Dst) const { + // Not enough arguments to match OSAtomicCompareAndSwap? + if (CE->getNumArgs() != 3) + return false; + + ASTContext &Ctx = Eng.getContext(); + const Expr *oldValueExpr = CE->getArg(0); + QualType oldValueType = Ctx.getCanonicalType(oldValueExpr->getType()); + + const Expr *newValueExpr = CE->getArg(1); + QualType newValueType = Ctx.getCanonicalType(newValueExpr->getType()); + + // Do the types of 'oldValue' and 'newValue' match? + if (oldValueType != newValueType) + return false; + + const Expr *theValueExpr = CE->getArg(2); + const PointerType *theValueType=theValueExpr->getType()->getAs<PointerType>(); + + // theValueType not a pointer? + if (!theValueType) + return false; + + QualType theValueTypePointee = + Ctx.getCanonicalType(theValueType->getPointeeType()).getUnqualifiedType(); + + // The pointee must match newValueType and oldValueType. + if (theValueTypePointee != newValueType) + return false; + + static SimpleProgramPointTag OSAtomicLoadTag("OSAtomicChecker : Load"); + static SimpleProgramPointTag OSAtomicStoreTag("OSAtomicChecker : Store"); + + // Load 'theValue'. + ProgramStateRef state = Pred->getState(); + const LocationContext *LCtx = Pred->getLocationContext(); + ExplodedNodeSet Tmp; + SVal location = state->getSVal(theValueExpr, LCtx); + // Here we should use the value type of the region as the load type, because + // we are simulating the semantics of the function, not the semantics of + // passing argument. So the type of theValue expr is not we are loading. + // But usually the type of the varregion is not the type we want either, + // we still need to do a CastRetrievedVal in store manager. So actually this + // LoadTy specifying can be omitted. But we put it here to emphasize the + // semantics. + QualType LoadTy; + if (const TypedValueRegion *TR = + dyn_cast_or_null<TypedValueRegion>(location.getAsRegion())) { + LoadTy = TR->getValueType(); + } + Eng.evalLoad(Tmp, CE, theValueExpr, Pred, + state, location, &OSAtomicLoadTag, LoadTy); + + if (Tmp.empty()) { + // If no nodes were generated, other checkers must have generated sinks. + // We return an empty Dst. + return true; + } + + for (ExplodedNodeSet::iterator I = Tmp.begin(), E = Tmp.end(); + I != E; ++I) { + + ExplodedNode *N = *I; + ProgramStateRef stateLoad = N->getState(); + + // Use direct bindings from the environment since we are forcing a load + // from a location that the Environment would typically not be used + // to bind a value. + SVal theValueVal_untested = stateLoad->getSVal(theValueExpr, LCtx, true); + + SVal oldValueVal_untested = stateLoad->getSVal(oldValueExpr, LCtx); + + // FIXME: Issue an error. + if (theValueVal_untested.isUndef() || oldValueVal_untested.isUndef()) { + return false; + } + + DefinedOrUnknownSVal theValueVal = + cast<DefinedOrUnknownSVal>(theValueVal_untested); + DefinedOrUnknownSVal oldValueVal = + cast<DefinedOrUnknownSVal>(oldValueVal_untested); + + SValBuilder &svalBuilder = Eng.getSValBuilder(); + + // Perform the comparison. + DefinedOrUnknownSVal Cmp = + svalBuilder.evalEQ(stateLoad,theValueVal,oldValueVal); + + ProgramStateRef stateEqual = stateLoad->assume(Cmp, true); + + // Were they equal? + if (stateEqual) { + // Perform the store. + ExplodedNodeSet TmpStore; + SVal val = stateEqual->getSVal(newValueExpr, LCtx); + + // Handle implicit value casts. + if (const TypedValueRegion *R = + dyn_cast_or_null<TypedValueRegion>(location.getAsRegion())) { + val = svalBuilder.evalCast(val,R->getValueType(), newValueExpr->getType()); + } + + Eng.evalStore(TmpStore, CE, theValueExpr, N, + stateEqual, location, val, &OSAtomicStoreTag); + + if (TmpStore.empty()) { + // If no nodes were generated, other checkers must have generated sinks. + // We return an empty Dst. + return true; + } + + StmtNodeBuilder B(TmpStore, Dst, Eng.getBuilderContext()); + // Now bind the result of the comparison. + for (ExplodedNodeSet::iterator I2 = TmpStore.begin(), + E2 = TmpStore.end(); I2 != E2; ++I2) { + ExplodedNode *predNew = *I2; + ProgramStateRef stateNew = predNew->getState(); + // Check for 'void' return type if we have a bogus function prototype. + SVal Res = UnknownVal(); + QualType T = CE->getType(); + if (!T->isVoidType()) + Res = Eng.getSValBuilder().makeTruthVal(true, T); + B.generateNode(CE, predNew, stateNew->BindExpr(CE, LCtx, Res), + false, this); + } + } + + // Were they not equal? + if (ProgramStateRef stateNotEqual = stateLoad->assume(Cmp, false)) { + // Check for 'void' return type if we have a bogus function prototype. + SVal Res = UnknownVal(); + QualType T = CE->getType(); + if (!T->isVoidType()) + Res = Eng.getSValBuilder().makeTruthVal(false, CE->getType()); + StmtNodeBuilder B(N, Dst, Eng.getBuilderContext()); + B.generateNode(CE, N, stateNotEqual->BindExpr(CE, LCtx, Res), + false, this); + } + } + + return true; +} + +void ento::registerOSAtomicChecker(CheckerManager &mgr) { + mgr.registerChecker<OSAtomicChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/ObjCAtSyncChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/ObjCAtSyncChecker.cpp new file mode 100644 index 0000000..777e9ea --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/ObjCAtSyncChecker.cpp @@ -0,0 +1,96 @@ +//== ObjCAtSyncChecker.cpp - nil mutex checker for @synchronized -*- C++ -*--=// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This defines ObjCAtSyncChecker, a builtin check that checks for null pointers +// used as mutexes for @synchronized. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/AST/StmtObjC.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/StaticAnalyzer/Checkers/DereferenceChecker.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ExprEngine.h" + +using namespace clang; +using namespace ento; + +namespace { +class ObjCAtSyncChecker + : public Checker< check::PreStmt<ObjCAtSynchronizedStmt> > { + mutable OwningPtr<BuiltinBug> BT_null; + mutable OwningPtr<BuiltinBug> BT_undef; + +public: + void checkPreStmt(const ObjCAtSynchronizedStmt *S, CheckerContext &C) const; +}; +} // end anonymous namespace + +void ObjCAtSyncChecker::checkPreStmt(const ObjCAtSynchronizedStmt *S, + CheckerContext &C) const { + + const Expr *Ex = S->getSynchExpr(); + ProgramStateRef state = C.getState(); + SVal V = state->getSVal(Ex, C.getLocationContext()); + + // Uninitialized value used for the mutex? + if (isa<UndefinedVal>(V)) { + if (ExplodedNode *N = C.generateSink()) { + if (!BT_undef) + BT_undef.reset(new BuiltinBug("Uninitialized value used as mutex " + "for @synchronized")); + BugReport *report = + new BugReport(*BT_undef, BT_undef->getDescription(), N); + report->addVisitor(bugreporter::getTrackNullOrUndefValueVisitor(N, Ex, + report)); + C.EmitReport(report); + } + return; + } + + if (V.isUnknown()) + return; + + // Check for null mutexes. + ProgramStateRef notNullState, nullState; + llvm::tie(notNullState, nullState) = state->assume(cast<DefinedSVal>(V)); + + if (nullState) { + if (!notNullState) { + // Generate an error node. This isn't a sink since + // a null mutex just means no synchronization occurs. + if (ExplodedNode *N = C.addTransition(nullState)) { + if (!BT_null) + BT_null.reset(new BuiltinBug("Nil value used as mutex for @synchronized() " + "(no synchronization will occur)")); + BugReport *report = + new BugReport(*BT_null, BT_null->getDescription(), N); + report->addVisitor(bugreporter::getTrackNullOrUndefValueVisitor(N, Ex, + report)); + + C.EmitReport(report); + return; + } + } + // Don't add a transition for 'nullState'. If the value is + // under-constrained to be null or non-null, assume it is non-null + // afterwards. + } + + if (notNullState) + C.addTransition(notNullState); +} + +void ento::registerObjCAtSyncChecker(CheckerManager &mgr) { + if (mgr.getLangOpts().ObjC2) + mgr.registerChecker<ObjCAtSyncChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/ObjCContainersASTChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/ObjCContainersASTChecker.cpp new file mode 100644 index 0000000..f2929c0 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/ObjCContainersASTChecker.cpp @@ -0,0 +1,174 @@ +//== ObjCContainersASTChecker.cpp - CoreFoundation containers API *- C++ -*-==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// An AST checker that looks for common pitfalls when using 'CFArray', +// 'CFDictionary', 'CFSet' APIs. +// +//===----------------------------------------------------------------------===// +#include "ClangSACheckers.h" +#include "clang/Analysis/AnalysisContext.h" +#include "clang/AST/StmtVisitor.h" +#include "clang/Basic/TargetInfo.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugReporter.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/AnalysisManager.h" +#include "llvm/ADT/SmallString.h" +#include "llvm/Support/raw_ostream.h" + +using namespace clang; +using namespace ento; + +namespace { +class WalkAST : public StmtVisitor<WalkAST> { + BugReporter &BR; + AnalysisDeclContext* AC; + ASTContext &ASTC; + uint64_t PtrWidth; + + static const unsigned InvalidArgIndex = UINT_MAX; + + /// Check if the type has pointer size (very conservative). + inline bool isPointerSize(const Type *T) { + if (!T) + return true; + if (T->isIncompleteType()) + return true; + return (ASTC.getTypeSize(T) == PtrWidth); + } + + /// Check if the type is a pointer/array to pointer sized values. + inline bool hasPointerToPointerSizedType(const Expr *E) { + QualType T = E->getType(); + + // The type could be either a pointer or array. + const Type *TP = T.getTypePtr(); + QualType PointeeT = TP->getPointeeType(); + if (!PointeeT.isNull()) { + // If the type is a pointer to an array, check the size of the array + // elements. To avoid false positives coming from assumption that the + // values x and &x are equal when x is an array. + if (const Type *TElem = PointeeT->getArrayElementTypeNoTypeQual()) + if (isPointerSize(TElem)) + return true; + + // Else, check the pointee size. + return isPointerSize(PointeeT.getTypePtr()); + } + + if (const Type *TElem = TP->getArrayElementTypeNoTypeQual()) + return isPointerSize(TElem); + + // The type must be an array/pointer type. + + // This could be a null constant, which is allowed. + if (E->isNullPointerConstant(ASTC, Expr::NPC_ValueDependentIsNull)) + return true; + return false; + } + +public: + WalkAST(BugReporter &br, AnalysisDeclContext* ac) + : BR(br), AC(ac), ASTC(AC->getASTContext()), + PtrWidth(ASTC.getTargetInfo().getPointerWidth(0)) {} + + // Statement visitor methods. + void VisitChildren(Stmt *S); + void VisitStmt(Stmt *S) { VisitChildren(S); } + void VisitCallExpr(CallExpr *CE); +}; +} // end anonymous namespace + +static StringRef getCalleeName(CallExpr *CE) { + const FunctionDecl *FD = CE->getDirectCallee(); + if (!FD) + return StringRef(); + + IdentifierInfo *II = FD->getIdentifier(); + if (!II) // if no identifier, not a simple C function + return StringRef(); + + return II->getName(); +} + +void WalkAST::VisitCallExpr(CallExpr *CE) { + StringRef Name = getCalleeName(CE); + if (Name.empty()) + return; + + const Expr *Arg = 0; + unsigned ArgNum = InvalidArgIndex; + + if (Name.equals("CFArrayCreate") || Name.equals("CFSetCreate")) { + ArgNum = 1; + Arg = CE->getArg(ArgNum)->IgnoreParenCasts(); + if (hasPointerToPointerSizedType(Arg)) + return; + } + + if (Arg == 0 && Name.equals("CFDictionaryCreate")) { + // Check first argument. + ArgNum = 1; + Arg = CE->getArg(ArgNum)->IgnoreParenCasts(); + if (hasPointerToPointerSizedType(Arg)) { + // Check second argument. + ArgNum = 2; + Arg = CE->getArg(ArgNum)->IgnoreParenCasts(); + if (hasPointerToPointerSizedType(Arg)) + // Both are good, return. + return; + } + } + + if (ArgNum != InvalidArgIndex) { + assert(ArgNum == 1 || ArgNum == 2); + + SmallString<256> BufName; + llvm::raw_svector_ostream OsName(BufName); + assert(ArgNum == 1 || ArgNum == 2); + OsName << " Invalid use of '" << Name << "'" ; + + SmallString<256> Buf; + llvm::raw_svector_ostream Os(Buf); + Os << " The "<< ((ArgNum == 1) ? "first" : "second") << " argument to '" + << Name << "' must be a C array of pointer-sized values, not '" + << Arg->getType().getAsString() << "'"; + + SourceRange R = Arg->getSourceRange(); + PathDiagnosticLocation CELoc = + PathDiagnosticLocation::createBegin(CE, BR.getSourceManager(), AC); + BR.EmitBasicReport(AC->getDecl(), + OsName.str(), categories::CoreFoundationObjectiveC, + Os.str(), CELoc, &R, 1); + } + + // Recurse and check children. + VisitChildren(CE); +} + +void WalkAST::VisitChildren(Stmt *S) { + for (Stmt::child_iterator I = S->child_begin(), E = S->child_end(); I!=E; ++I) + if (Stmt *child = *I) + Visit(child); +} + +namespace { +class ObjCContainersASTChecker : public Checker<check::ASTCodeBody> { +public: + + void checkASTCodeBody(const Decl *D, AnalysisManager& Mgr, + BugReporter &BR) const { + WalkAST walker(BR, Mgr.getAnalysisDeclContext(D)); + walker.Visit(D->getBody()); + } +}; +} + +void ento::registerObjCContainersASTChecker(CheckerManager &mgr) { + mgr.registerChecker<ObjCContainersASTChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/ObjCContainersChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/ObjCContainersChecker.cpp new file mode 100644 index 0000000..f4655b6 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/ObjCContainersChecker.cpp @@ -0,0 +1,159 @@ +//== ObjCContainersChecker.cpp - Path sensitive checker for CFArray *- C++ -*=// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// Performs path sensitive checks of Core Foundation static containers like +// CFArray. +// 1) Check for buffer overflows: +// In CFArrayGetArrayAtIndex( myArray, index), if the index is outside the +// index space of theArray (0 to N-1 inclusive (where N is the count of +// theArray), the behavior is undefined. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramStateTrait.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ObjCMessage.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/AST/ParentMap.h" + +using namespace clang; +using namespace ento; + +namespace { +class ObjCContainersChecker : public Checker< check::PreStmt<CallExpr>, + check::PostStmt<CallExpr> > { + mutable OwningPtr<BugType> BT; + inline void initBugType() const { + if (!BT) + BT.reset(new BugType("CFArray API", + categories::CoreFoundationObjectiveC)); + } + + inline SymbolRef getArraySym(const Expr *E, CheckerContext &C) const { + SVal ArrayRef = C.getState()->getSVal(E, C.getLocationContext()); + SymbolRef ArraySym = ArrayRef.getAsSymbol(); + return ArraySym; + } + + void addSizeInfo(const Expr *Array, const Expr *Size, + CheckerContext &C) const; + +public: + /// A tag to id this checker. + static void *getTag() { static int Tag; return &Tag; } + + void checkPostStmt(const CallExpr *CE, CheckerContext &C) const; + void checkPreStmt(const CallExpr *CE, CheckerContext &C) const; +}; +} // end anonymous namespace + +// ProgramState trait - a map from array symbol to it's state. +typedef llvm::ImmutableMap<SymbolRef, DefinedSVal> ArraySizeM; + +namespace { struct ArraySizeMap {}; } +namespace clang { namespace ento { +template<> struct ProgramStateTrait<ArraySizeMap> + : public ProgramStatePartialTrait<ArraySizeM > { + static void *GDMIndex() { return ObjCContainersChecker::getTag(); } +}; +}} + +void ObjCContainersChecker::addSizeInfo(const Expr *Array, const Expr *Size, + CheckerContext &C) const { + ProgramStateRef State = C.getState(); + SVal SizeV = State->getSVal(Size, C.getLocationContext()); + // Undefined is reported by another checker. + if (SizeV.isUnknownOrUndef()) + return; + + // Get the ArrayRef symbol. + SVal ArrayRef = State->getSVal(Array, C.getLocationContext()); + SymbolRef ArraySym = ArrayRef.getAsSymbol(); + if (!ArraySym) + return; + + C.addTransition(State->set<ArraySizeMap>(ArraySym, cast<DefinedSVal>(SizeV))); + return; +} + +void ObjCContainersChecker::checkPostStmt(const CallExpr *CE, + CheckerContext &C) const { + StringRef Name = C.getCalleeName(CE); + if (Name.empty() || CE->getNumArgs() < 1) + return; + + // Add array size information to the state. + if (Name.equals("CFArrayCreate")) { + if (CE->getNumArgs() < 3) + return; + // Note, we can visit the Create method in the post-visit because + // the CFIndex parameter is passed in by value and will not be invalidated + // by the call. + addSizeInfo(CE, CE->getArg(2), C); + return; + } + + if (Name.equals("CFArrayGetCount")) { + addSizeInfo(CE->getArg(0), CE, C); + return; + } +} + +void ObjCContainersChecker::checkPreStmt(const CallExpr *CE, + CheckerContext &C) const { + StringRef Name = C.getCalleeName(CE); + if (Name.empty() || CE->getNumArgs() < 2) + return; + + // Check the array access. + if (Name.equals("CFArrayGetValueAtIndex")) { + ProgramStateRef State = C.getState(); + // Retrieve the size. + // Find out if we saw this array symbol before and have information about it. + const Expr *ArrayExpr = CE->getArg(0); + SymbolRef ArraySym = getArraySym(ArrayExpr, C); + if (!ArraySym) + return; + + const DefinedSVal *Size = State->get<ArraySizeMap>(ArraySym); + + if (!Size) + return; + + // Get the index. + const Expr *IdxExpr = CE->getArg(1); + SVal IdxVal = State->getSVal(IdxExpr, C.getLocationContext()); + if (IdxVal.isUnknownOrUndef()) + return; + DefinedSVal Idx = cast<DefinedSVal>(IdxVal); + + // Now, check if 'Idx in [0, Size-1]'. + const QualType T = IdxExpr->getType(); + ProgramStateRef StInBound = State->assumeInBound(Idx, *Size, true, T); + ProgramStateRef StOutBound = State->assumeInBound(Idx, *Size, false, T); + if (StOutBound && !StInBound) { + ExplodedNode *N = C.generateSink(StOutBound); + if (!N) + return; + initBugType(); + BugReport *R = new BugReport(*BT, "Index is out of bounds", N); + R->addRange(IdxExpr->getSourceRange()); + C.EmitReport(R); + return; + } + } +} + +/// Register checker. +void ento::registerObjCContainersChecker(CheckerManager &mgr) { + mgr.registerChecker<ObjCContainersChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/ObjCSelfInitChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/ObjCSelfInitChecker.cpp new file mode 100644 index 0000000..97b58cf --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/ObjCSelfInitChecker.cpp @@ -0,0 +1,396 @@ +//== ObjCSelfInitChecker.cpp - Checker for 'self' initialization -*- C++ -*--=// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This defines ObjCSelfInitChecker, a builtin check that checks for uses of +// 'self' before proper initialization. +// +//===----------------------------------------------------------------------===// + +// This checks initialization methods to verify that they assign 'self' to the +// result of an initialization call (e.g. [super init], or [self initWith..]) +// before using 'self' or any instance variable. +// +// To perform the required checking, values are tagged with flags that indicate +// 1) if the object is the one pointed to by 'self', and 2) if the object +// is the result of an initializer (e.g. [super init]). +// +// Uses of an object that is true for 1) but not 2) trigger a diagnostic. +// The uses that are currently checked are: +// - Using instance variables. +// - Returning the object. +// +// Note that we don't check for an invalid 'self' that is the receiver of an +// obj-c message expression to cut down false positives where logging functions +// get information from self (like its class) or doing "invalidation" on self +// when the initialization fails. +// +// Because the object that 'self' points to gets invalidated when a call +// receives a reference to 'self', the checker keeps track and passes the flags +// for 1) and 2) to the new object that 'self' points to after the call. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramStateTrait.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ObjCMessage.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/AST/ParentMap.h" + +using namespace clang; +using namespace ento; + +static bool shouldRunOnFunctionOrMethod(const NamedDecl *ND); +static bool isInitializationMethod(const ObjCMethodDecl *MD); +static bool isInitMessage(const ObjCMessage &msg); +static bool isSelfVar(SVal location, CheckerContext &C); + +namespace { +class ObjCSelfInitChecker : public Checker< check::PreObjCMessage, + check::PostObjCMessage, + check::PostStmt<ObjCIvarRefExpr>, + check::PreStmt<ReturnStmt>, + check::PreStmt<CallExpr>, + check::PostStmt<CallExpr>, + check::Location > { +public: + void checkPreObjCMessage(ObjCMessage msg, CheckerContext &C) const; + void checkPostObjCMessage(ObjCMessage msg, CheckerContext &C) const; + void checkPostStmt(const ObjCIvarRefExpr *E, CheckerContext &C) const; + void checkPreStmt(const ReturnStmt *S, CheckerContext &C) const; + void checkPreStmt(const CallExpr *CE, CheckerContext &C) const; + void checkPostStmt(const CallExpr *CE, CheckerContext &C) const; + void checkLocation(SVal location, bool isLoad, const Stmt *S, + CheckerContext &C) const; + + void checkPreStmt(const CallOrObjCMessage &CE, CheckerContext &C) const; + void checkPostStmt(const CallOrObjCMessage &CE, CheckerContext &C) const; + +}; +} // end anonymous namespace + +namespace { + +class InitSelfBug : public BugType { + const std::string desc; +public: + InitSelfBug() : BugType("Missing \"self = [(super or self) init...]\"", + categories::CoreFoundationObjectiveC) {} +}; + +} // end anonymous namespace + +namespace { +enum SelfFlagEnum { + /// \brief No flag set. + SelfFlag_None = 0x0, + /// \brief Value came from 'self'. + SelfFlag_Self = 0x1, + /// \brief Value came from the result of an initializer (e.g. [super init]). + SelfFlag_InitRes = 0x2 +}; +} + +typedef llvm::ImmutableMap<SymbolRef, unsigned> SelfFlag; +namespace { struct CalledInit {}; } +namespace { struct PreCallSelfFlags {}; } + +namespace clang { +namespace ento { + template<> + struct ProgramStateTrait<SelfFlag> : public ProgramStatePartialTrait<SelfFlag> { + static void *GDMIndex() { static int index = 0; return &index; } + }; + template <> + struct ProgramStateTrait<CalledInit> : public ProgramStatePartialTrait<bool> { + static void *GDMIndex() { static int index = 0; return &index; } + }; + + /// \brief A call receiving a reference to 'self' invalidates the object that + /// 'self' contains. This keeps the "self flags" assigned to the 'self' + /// object before the call so we can assign them to the new object that 'self' + /// points to after the call. + template <> + struct ProgramStateTrait<PreCallSelfFlags> : public ProgramStatePartialTrait<unsigned> { + static void *GDMIndex() { static int index = 0; return &index; } + }; +} +} + +static SelfFlagEnum getSelfFlags(SVal val, ProgramStateRef state) { + if (SymbolRef sym = val.getAsSymbol()) + if (const unsigned *attachedFlags = state->get<SelfFlag>(sym)) + return (SelfFlagEnum)*attachedFlags; + return SelfFlag_None; +} + +static SelfFlagEnum getSelfFlags(SVal val, CheckerContext &C) { + return getSelfFlags(val, C.getState()); +} + +static void addSelfFlag(ProgramStateRef state, SVal val, + SelfFlagEnum flag, CheckerContext &C) { + // We tag the symbol that the SVal wraps. + if (SymbolRef sym = val.getAsSymbol()) + C.addTransition(state->set<SelfFlag>(sym, getSelfFlags(val, C) | flag)); +} + +static bool hasSelfFlag(SVal val, SelfFlagEnum flag, CheckerContext &C) { + return getSelfFlags(val, C) & flag; +} + +/// \brief Returns true of the value of the expression is the object that 'self' +/// points to and is an object that did not come from the result of calling +/// an initializer. +static bool isInvalidSelf(const Expr *E, CheckerContext &C) { + SVal exprVal = C.getState()->getSVal(E, C.getLocationContext()); + if (!hasSelfFlag(exprVal, SelfFlag_Self, C)) + return false; // value did not come from 'self'. + if (hasSelfFlag(exprVal, SelfFlag_InitRes, C)) + return false; // 'self' is properly initialized. + + return true; +} + +static void checkForInvalidSelf(const Expr *E, CheckerContext &C, + const char *errorStr) { + if (!E) + return; + + if (!C.getState()->get<CalledInit>()) + return; + + if (!isInvalidSelf(E, C)) + return; + + // Generate an error node. + ExplodedNode *N = C.generateSink(); + if (!N) + return; + + BugReport *report = + new BugReport(*new InitSelfBug(), errorStr, N); + C.EmitReport(report); +} + +void ObjCSelfInitChecker::checkPostObjCMessage(ObjCMessage msg, + CheckerContext &C) const { + // When encountering a message that does initialization (init rule), + // tag the return value so that we know later on that if self has this value + // then it is properly initialized. + + // FIXME: A callback should disable checkers at the start of functions. + if (!shouldRunOnFunctionOrMethod(dyn_cast<NamedDecl>( + C.getCurrentAnalysisDeclContext()->getDecl()))) + return; + + if (isInitMessage(msg)) { + // Tag the return value as the result of an initializer. + ProgramStateRef state = C.getState(); + + // FIXME this really should be context sensitive, where we record + // the current stack frame (for IPA). Also, we need to clean this + // value out when we return from this method. + state = state->set<CalledInit>(true); + + SVal V = state->getSVal(msg.getMessageExpr(), C.getLocationContext()); + addSelfFlag(state, V, SelfFlag_InitRes, C); + return; + } + + CallOrObjCMessage MsgWrapper(msg, C.getState(), C.getLocationContext()); + checkPostStmt(MsgWrapper, C); + + // We don't check for an invalid 'self' in an obj-c message expression to cut + // down false positives where logging functions get information from self + // (like its class) or doing "invalidation" on self when the initialization + // fails. +} + +void ObjCSelfInitChecker::checkPostStmt(const ObjCIvarRefExpr *E, + CheckerContext &C) const { + // FIXME: A callback should disable checkers at the start of functions. + if (!shouldRunOnFunctionOrMethod(dyn_cast<NamedDecl>( + C.getCurrentAnalysisDeclContext()->getDecl()))) + return; + + checkForInvalidSelf(E->getBase(), C, + "Instance variable used while 'self' is not set to the result of " + "'[(super or self) init...]'"); +} + +void ObjCSelfInitChecker::checkPreStmt(const ReturnStmt *S, + CheckerContext &C) const { + // FIXME: A callback should disable checkers at the start of functions. + if (!shouldRunOnFunctionOrMethod(dyn_cast<NamedDecl>( + C.getCurrentAnalysisDeclContext()->getDecl()))) + return; + + checkForInvalidSelf(S->getRetValue(), C, + "Returning 'self' while it is not set to the result of " + "'[(super or self) init...]'"); +} + +// When a call receives a reference to 'self', [Pre/Post]VisitGenericCall pass +// the SelfFlags from the object 'self' point to before the call, to the new +// object after the call. This is to avoid invalidation of 'self' by logging +// functions. +// Another common pattern in classes with multiple initializers is to put the +// subclass's common initialization bits into a static function that receives +// the value of 'self', e.g: +// @code +// if (!(self = [super init])) +// return nil; +// if (!(self = _commonInit(self))) +// return nil; +// @endcode +// Until we can use inter-procedural analysis, in such a call, transfer the +// SelfFlags to the result of the call. + +void ObjCSelfInitChecker::checkPreStmt(const CallExpr *CE, + CheckerContext &C) const { + CallOrObjCMessage CEWrapper(CE, C.getState(), C.getLocationContext()); + checkPreStmt(CEWrapper, C); +} + +void ObjCSelfInitChecker::checkPostStmt(const CallExpr *CE, + CheckerContext &C) const { + CallOrObjCMessage CEWrapper(CE, C.getState(), C.getLocationContext()); + checkPostStmt(CEWrapper, C); +} + +void ObjCSelfInitChecker::checkPreObjCMessage(ObjCMessage Msg, + CheckerContext &C) const { + CallOrObjCMessage MsgWrapper(Msg, C.getState(), C.getLocationContext()); + checkPreStmt(MsgWrapper, C); +} + +void ObjCSelfInitChecker::checkPreStmt(const CallOrObjCMessage &CE, + CheckerContext &C) const { + ProgramStateRef state = C.getState(); + unsigned NumArgs = CE.getNumArgs(); + // If we passed 'self' as and argument to the call, record it in the state + // to be propagated after the call. + // Note, we could have just given up, but try to be more optimistic here and + // assume that the functions are going to continue initialization or will not + // modify self. + for (unsigned i = 0; i < NumArgs; ++i) { + SVal argV = CE.getArgSVal(i); + if (isSelfVar(argV, C)) { + unsigned selfFlags = getSelfFlags(state->getSVal(cast<Loc>(argV)), C); + C.addTransition(state->set<PreCallSelfFlags>(selfFlags)); + return; + } else if (hasSelfFlag(argV, SelfFlag_Self, C)) { + unsigned selfFlags = getSelfFlags(argV, C); + C.addTransition(state->set<PreCallSelfFlags>(selfFlags)); + return; + } + } +} + +void ObjCSelfInitChecker::checkPostStmt(const CallOrObjCMessage &CE, + CheckerContext &C) const { + ProgramStateRef state = C.getState(); + unsigned NumArgs = CE.getNumArgs(); + for (unsigned i = 0; i < NumArgs; ++i) { + SVal argV = CE.getArgSVal(i); + if (isSelfVar(argV, C)) { + // If the address of 'self' is being passed to the call, assume that the + // 'self' after the call will have the same flags. + // EX: log(&self) + SelfFlagEnum prevFlags = (SelfFlagEnum)state->get<PreCallSelfFlags>(); + state = state->remove<PreCallSelfFlags>(); + addSelfFlag(state, state->getSVal(cast<Loc>(argV)), prevFlags, C); + return; + } else if (hasSelfFlag(argV, SelfFlag_Self, C)) { + // If 'self' is passed to the call by value, assume that the function + // returns 'self'. So assign the flags, which were set on 'self' to the + // return value. + // EX: self = performMoreInitialization(self) + SelfFlagEnum prevFlags = (SelfFlagEnum)state->get<PreCallSelfFlags>(); + state = state->remove<PreCallSelfFlags>(); + const Expr *CallExpr = CE.getOriginExpr(); + if (CallExpr) + addSelfFlag(state, state->getSVal(CallExpr, C.getLocationContext()), + prevFlags, C); + return; + } + } +} + +void ObjCSelfInitChecker::checkLocation(SVal location, bool isLoad, + const Stmt *S, + CheckerContext &C) const { + // Tag the result of a load from 'self' so that we can easily know that the + // value is the object that 'self' points to. + ProgramStateRef state = C.getState(); + if (isSelfVar(location, C)) + addSelfFlag(state, state->getSVal(cast<Loc>(location)), SelfFlag_Self, C); +} + +// FIXME: A callback should disable checkers at the start of functions. +static bool shouldRunOnFunctionOrMethod(const NamedDecl *ND) { + if (!ND) + return false; + + const ObjCMethodDecl *MD = dyn_cast<ObjCMethodDecl>(ND); + if (!MD) + return false; + if (!isInitializationMethod(MD)) + return false; + + // self = [super init] applies only to NSObject subclasses. + // For instance, NSProxy doesn't implement -init. + ASTContext &Ctx = MD->getASTContext(); + IdentifierInfo* NSObjectII = &Ctx.Idents.get("NSObject"); + ObjCInterfaceDecl *ID = MD->getClassInterface()->getSuperClass(); + for ( ; ID ; ID = ID->getSuperClass()) { + IdentifierInfo *II = ID->getIdentifier(); + + if (II == NSObjectII) + break; + } + if (!ID) + return false; + + return true; +} + +/// \brief Returns true if the location is 'self'. +static bool isSelfVar(SVal location, CheckerContext &C) { + AnalysisDeclContext *analCtx = C.getCurrentAnalysisDeclContext(); + if (!analCtx->getSelfDecl()) + return false; + if (!isa<loc::MemRegionVal>(location)) + return false; + + loc::MemRegionVal MRV = cast<loc::MemRegionVal>(location); + if (const DeclRegion *DR = dyn_cast<DeclRegion>(MRV.stripCasts())) + return (DR->getDecl() == analCtx->getSelfDecl()); + + return false; +} + +static bool isInitializationMethod(const ObjCMethodDecl *MD) { + return MD->getMethodFamily() == OMF_init; +} + +static bool isInitMessage(const ObjCMessage &msg) { + return msg.getMethodFamily() == OMF_init; +} + +//===----------------------------------------------------------------------===// +// Registration. +//===----------------------------------------------------------------------===// + +void ento::registerObjCSelfInitChecker(CheckerManager &mgr) { + mgr.registerChecker<ObjCSelfInitChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/ObjCUnusedIVarsChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/ObjCUnusedIVarsChecker.cpp new file mode 100644 index 0000000..4718dc7 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/ObjCUnusedIVarsChecker.cpp @@ -0,0 +1,186 @@ +//==- ObjCUnusedIVarsChecker.cpp - Check for unused ivars --------*- C++ -*-==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines a CheckObjCUnusedIvars, a checker that +// analyzes an Objective-C class's interface/implementation to determine if it +// has any ivars that are never accessed. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/BugReporter/PathDiagnostic.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugReporter.h" +#include "clang/AST/ExprObjC.h" +#include "clang/AST/Expr.h" +#include "clang/AST/DeclObjC.h" +#include "clang/Basic/LangOptions.h" +#include "clang/Basic/SourceManager.h" + +using namespace clang; +using namespace ento; + +enum IVarState { Unused, Used }; +typedef llvm::DenseMap<const ObjCIvarDecl*,IVarState> IvarUsageMap; + +static void Scan(IvarUsageMap& M, const Stmt *S) { + if (!S) + return; + + if (const ObjCIvarRefExpr *Ex = dyn_cast<ObjCIvarRefExpr>(S)) { + const ObjCIvarDecl *D = Ex->getDecl(); + IvarUsageMap::iterator I = M.find(D); + if (I != M.end()) + I->second = Used; + return; + } + + // Blocks can reference an instance variable of a class. + if (const BlockExpr *BE = dyn_cast<BlockExpr>(S)) { + Scan(M, BE->getBody()); + return; + } + + for (Stmt::const_child_iterator I=S->child_begin(),E=S->child_end(); I!=E;++I) + Scan(M, *I); +} + +static void Scan(IvarUsageMap& M, const ObjCPropertyImplDecl *D) { + if (!D) + return; + + const ObjCIvarDecl *ID = D->getPropertyIvarDecl(); + + if (!ID) + return; + + IvarUsageMap::iterator I = M.find(ID); + if (I != M.end()) + I->second = Used; +} + +static void Scan(IvarUsageMap& M, const ObjCContainerDecl *D) { + // Scan the methods for accesses. + for (ObjCContainerDecl::instmeth_iterator I = D->instmeth_begin(), + E = D->instmeth_end(); I!=E; ++I) + Scan(M, (*I)->getBody()); + + if (const ObjCImplementationDecl *ID = dyn_cast<ObjCImplementationDecl>(D)) { + // Scan for @synthesized property methods that act as setters/getters + // to an ivar. + for (ObjCImplementationDecl::propimpl_iterator I = ID->propimpl_begin(), + E = ID->propimpl_end(); I!=E; ++I) + Scan(M, *I); + + // Scan the associated categories as well. + for (const ObjCCategoryDecl *CD = + ID->getClassInterface()->getCategoryList(); CD ; + CD = CD->getNextClassCategory()) { + if (const ObjCCategoryImplDecl *CID = CD->getImplementation()) + Scan(M, CID); + } + } +} + +static void Scan(IvarUsageMap &M, const DeclContext *C, const FileID FID, + SourceManager &SM) { + for (DeclContext::decl_iterator I=C->decls_begin(), E=C->decls_end(); + I!=E; ++I) + if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(*I)) { + SourceLocation L = FD->getLocStart(); + if (SM.getFileID(L) == FID) + Scan(M, FD->getBody()); + } +} + +static void checkObjCUnusedIvar(const ObjCImplementationDecl *D, + BugReporter &BR) { + + const ObjCInterfaceDecl *ID = D->getClassInterface(); + IvarUsageMap M; + + // Iterate over the ivars. + for (ObjCInterfaceDecl::ivar_iterator I=ID->ivar_begin(), + E=ID->ivar_end(); I!=E; ++I) { + + const ObjCIvarDecl *ID = *I; + + // Ignore ivars that... + // (a) aren't private + // (b) explicitly marked unused + // (c) are iboutlets + // (d) are unnamed bitfields + if (ID->getAccessControl() != ObjCIvarDecl::Private || + ID->getAttr<UnusedAttr>() || ID->getAttr<IBOutletAttr>() || + ID->getAttr<IBOutletCollectionAttr>() || + ID->isUnnamedBitfield()) + continue; + + M[ID] = Unused; + } + + if (M.empty()) + return; + + // Now scan the implementation declaration. + Scan(M, D); + + // Any potentially unused ivars? + bool hasUnused = false; + for (IvarUsageMap::iterator I = M.begin(), E = M.end(); I!=E; ++I) + if (I->second == Unused) { + hasUnused = true; + break; + } + + if (!hasUnused) + return; + + // We found some potentially unused ivars. Scan the entire translation unit + // for functions inside the @implementation that reference these ivars. + // FIXME: In the future hopefully we can just use the lexical DeclContext + // to go from the ObjCImplementationDecl to the lexically "nested" + // C functions. + SourceManager &SM = BR.getSourceManager(); + Scan(M, D->getDeclContext(), SM.getFileID(D->getLocation()), SM); + + // Find ivars that are unused. + for (IvarUsageMap::iterator I = M.begin(), E = M.end(); I!=E; ++I) + if (I->second == Unused) { + std::string sbuf; + llvm::raw_string_ostream os(sbuf); + os << "Instance variable '" << *I->first << "' in class '" << *ID + << "' is never used by the methods in its @implementation " + "(although it may be used by category methods)."; + + PathDiagnosticLocation L = + PathDiagnosticLocation::create(I->first, BR.getSourceManager()); + BR.EmitBasicReport(D, "Unused instance variable", "Optimization", + os.str(), L); + } +} + +//===----------------------------------------------------------------------===// +// ObjCUnusedIvarsChecker +//===----------------------------------------------------------------------===// + +namespace { +class ObjCUnusedIvarsChecker : public Checker< + check::ASTDecl<ObjCImplementationDecl> > { +public: + void checkASTDecl(const ObjCImplementationDecl *D, AnalysisManager& mgr, + BugReporter &BR) const { + checkObjCUnusedIvar(D, BR); + } +}; +} + +void ento::registerObjCUnusedIvarsChecker(CheckerManager &mgr) { + mgr.registerChecker<ObjCUnusedIvarsChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/PointerArithChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/PointerArithChecker.cpp new file mode 100644 index 0000000..fe4845b --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/PointerArithChecker.cpp @@ -0,0 +1,69 @@ +//=== PointerArithChecker.cpp - Pointer arithmetic checker -----*- C++ -*--===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This files defines PointerArithChecker, a builtin checker that checks for +// pointer arithmetic on locations other than array elements. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" + +using namespace clang; +using namespace ento; + +namespace { +class PointerArithChecker + : public Checker< check::PreStmt<BinaryOperator> > { + mutable OwningPtr<BuiltinBug> BT; + +public: + void checkPreStmt(const BinaryOperator *B, CheckerContext &C) const; +}; +} + +void PointerArithChecker::checkPreStmt(const BinaryOperator *B, + CheckerContext &C) const { + if (B->getOpcode() != BO_Sub && B->getOpcode() != BO_Add) + return; + + ProgramStateRef state = C.getState(); + const LocationContext *LCtx = C.getLocationContext(); + SVal LV = state->getSVal(B->getLHS(), LCtx); + SVal RV = state->getSVal(B->getRHS(), LCtx); + + const MemRegion *LR = LV.getAsRegion(); + + if (!LR || !RV.isConstant()) + return; + + // If pointer arithmetic is done on variables of non-array type, this often + // means behavior rely on memory organization, which is dangerous. + if (isa<VarRegion>(LR) || isa<CodeTextRegion>(LR) || + isa<CompoundLiteralRegion>(LR)) { + + if (ExplodedNode *N = C.addTransition()) { + if (!BT) + BT.reset(new BuiltinBug("Dangerous pointer arithmetic", + "Pointer arithmetic done on non-array variables " + "means reliance on memory layout, which is " + "dangerous.")); + BugReport *R = new BugReport(*BT, BT->getDescription(), N); + R->addRange(B->getSourceRange()); + C.EmitReport(R); + } + } +} + +void ento::registerPointerArithChecker(CheckerManager &mgr) { + mgr.registerChecker<PointerArithChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/PointerSubChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/PointerSubChecker.cpp new file mode 100644 index 0000000..fa5c6a3 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/PointerSubChecker.cpp @@ -0,0 +1,76 @@ +//=== PointerSubChecker.cpp - Pointer subtraction checker ------*- C++ -*--===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This files defines PointerSubChecker, a builtin checker that checks for +// pointer subtractions on two pointers pointing to different memory chunks. +// This check corresponds to CWE-469. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" + +using namespace clang; +using namespace ento; + +namespace { +class PointerSubChecker + : public Checker< check::PreStmt<BinaryOperator> > { + mutable OwningPtr<BuiltinBug> BT; + +public: + void checkPreStmt(const BinaryOperator *B, CheckerContext &C) const; +}; +} + +void PointerSubChecker::checkPreStmt(const BinaryOperator *B, + CheckerContext &C) const { + // When doing pointer subtraction, if the two pointers do not point to the + // same memory chunk, emit a warning. + if (B->getOpcode() != BO_Sub) + return; + + ProgramStateRef state = C.getState(); + const LocationContext *LCtx = C.getLocationContext(); + SVal LV = state->getSVal(B->getLHS(), LCtx); + SVal RV = state->getSVal(B->getRHS(), LCtx); + + const MemRegion *LR = LV.getAsRegion(); + const MemRegion *RR = RV.getAsRegion(); + + if (!(LR && RR)) + return; + + const MemRegion *BaseLR = LR->getBaseRegion(); + const MemRegion *BaseRR = RR->getBaseRegion(); + + if (BaseLR == BaseRR) + return; + + // Allow arithmetic on different symbolic regions. + if (isa<SymbolicRegion>(BaseLR) || isa<SymbolicRegion>(BaseRR)) + return; + + if (ExplodedNode *N = C.addTransition()) { + if (!BT) + BT.reset(new BuiltinBug("Pointer subtraction", + "Subtraction of two pointers that do not point to " + "the same memory chunk may cause incorrect result.")); + BugReport *R = new BugReport(*BT, BT->getDescription(), N); + R->addRange(B->getSourceRange()); + C.EmitReport(R); + } +} + +void ento::registerPointerSubChecker(CheckerManager &mgr) { + mgr.registerChecker<PointerSubChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/PthreadLockChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/PthreadLockChecker.cpp new file mode 100644 index 0000000..2d018ef --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/PthreadLockChecker.cpp @@ -0,0 +1,198 @@ +//===--- PthreadLockChecker.cpp - Check for locking problems ---*- C++ -*--===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This defines PthreadLockChecker, a simple lock -> unlock checker. +// Also handles XNU locks, which behave similarly enough to share code. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramStateTrait.h" +#include "llvm/ADT/ImmutableList.h" + +using namespace clang; +using namespace ento; + +namespace { +class PthreadLockChecker : public Checker< check::PostStmt<CallExpr> > { + mutable OwningPtr<BugType> BT_doublelock; + mutable OwningPtr<BugType> BT_lor; + enum LockingSemantics { + NotApplicable = 0, + PthreadSemantics, + XNUSemantics + }; +public: + void checkPostStmt(const CallExpr *CE, CheckerContext &C) const; + + void AcquireLock(CheckerContext &C, const CallExpr *CE, SVal lock, + bool isTryLock, enum LockingSemantics semantics) const; + + void ReleaseLock(CheckerContext &C, const CallExpr *CE, SVal lock) const; +}; +} // end anonymous namespace + +// GDM Entry for tracking lock state. +namespace { class LockSet {}; } +namespace clang { +namespace ento { +template <> struct ProgramStateTrait<LockSet> : + public ProgramStatePartialTrait<llvm::ImmutableList<const MemRegion*> > { + static void *GDMIndex() { static int x = 0; return &x; } +}; +} // end of ento (ProgramState) namespace +} // end clang namespace + + +void PthreadLockChecker::checkPostStmt(const CallExpr *CE, + CheckerContext &C) const { + ProgramStateRef state = C.getState(); + const LocationContext *LCtx = C.getLocationContext(); + StringRef FName = C.getCalleeName(CE); + if (FName.empty()) + return; + + if (CE->getNumArgs() != 1) + return; + + if (FName == "pthread_mutex_lock" || + FName == "pthread_rwlock_rdlock" || + FName == "pthread_rwlock_wrlock") + AcquireLock(C, CE, state->getSVal(CE->getArg(0), LCtx), + false, PthreadSemantics); + else if (FName == "lck_mtx_lock" || + FName == "lck_rw_lock_exclusive" || + FName == "lck_rw_lock_shared") + AcquireLock(C, CE, state->getSVal(CE->getArg(0), LCtx), + false, XNUSemantics); + else if (FName == "pthread_mutex_trylock" || + FName == "pthread_rwlock_tryrdlock" || + FName == "pthread_rwlock_tryrwlock") + AcquireLock(C, CE, state->getSVal(CE->getArg(0), LCtx), + true, PthreadSemantics); + else if (FName == "lck_mtx_try_lock" || + FName == "lck_rw_try_lock_exclusive" || + FName == "lck_rw_try_lock_shared") + AcquireLock(C, CE, state->getSVal(CE->getArg(0), LCtx), + true, XNUSemantics); + else if (FName == "pthread_mutex_unlock" || + FName == "pthread_rwlock_unlock" || + FName == "lck_mtx_unlock" || + FName == "lck_rw_done") + ReleaseLock(C, CE, state->getSVal(CE->getArg(0), LCtx)); +} + +void PthreadLockChecker::AcquireLock(CheckerContext &C, const CallExpr *CE, + SVal lock, bool isTryLock, + enum LockingSemantics semantics) const { + + const MemRegion *lockR = lock.getAsRegion(); + if (!lockR) + return; + + ProgramStateRef state = C.getState(); + + SVal X = state->getSVal(CE, C.getLocationContext()); + if (X.isUnknownOrUndef()) + return; + + DefinedSVal retVal = cast<DefinedSVal>(X); + + if (state->contains<LockSet>(lockR)) { + if (!BT_doublelock) + BT_doublelock.reset(new BugType("Double locking", "Lock checker")); + ExplodedNode *N = C.generateSink(); + if (!N) + return; + BugReport *report = new BugReport(*BT_doublelock, + "This lock has already " + "been acquired", N); + report->addRange(CE->getArg(0)->getSourceRange()); + C.EmitReport(report); + return; + } + + ProgramStateRef lockSucc = state; + if (isTryLock) { + // Bifurcate the state, and allow a mode where the lock acquisition fails. + ProgramStateRef lockFail; + switch (semantics) { + case PthreadSemantics: + llvm::tie(lockFail, lockSucc) = state->assume(retVal); + break; + case XNUSemantics: + llvm::tie(lockSucc, lockFail) = state->assume(retVal); + break; + default: + llvm_unreachable("Unknown tryLock locking semantics"); + } + assert(lockFail && lockSucc); + C.addTransition(lockFail); + + } else if (semantics == PthreadSemantics) { + // Assume that the return value was 0. + lockSucc = state->assume(retVal, false); + assert(lockSucc); + + } else { + // XNU locking semantics return void on non-try locks + assert((semantics == XNUSemantics) && "Unknown locking semantics"); + lockSucc = state; + } + + // Record that the lock was acquired. + lockSucc = lockSucc->add<LockSet>(lockR); + C.addTransition(lockSucc); +} + +void PthreadLockChecker::ReleaseLock(CheckerContext &C, const CallExpr *CE, + SVal lock) const { + + const MemRegion *lockR = lock.getAsRegion(); + if (!lockR) + return; + + ProgramStateRef state = C.getState(); + llvm::ImmutableList<const MemRegion*> LS = state->get<LockSet>(); + + // FIXME: Better analysis requires IPA for wrappers. + // FIXME: check for double unlocks + if (LS.isEmpty()) + return; + + const MemRegion *firstLockR = LS.getHead(); + if (firstLockR != lockR) { + if (!BT_lor) + BT_lor.reset(new BugType("Lock order reversal", "Lock checker")); + ExplodedNode *N = C.generateSink(); + if (!N) + return; + BugReport *report = new BugReport(*BT_lor, + "This was not the most " + "recently acquired lock. " + "Possible lock order " + "reversal", N); + report->addRange(CE->getArg(0)->getSourceRange()); + C.EmitReport(report); + return; + } + + // Record that the lock was released. + state = state->set<LockSet>(LS.getTail()); + C.addTransition(state); +} + + +void ento::registerPthreadLockChecker(CheckerManager &mgr) { + mgr.registerChecker<PthreadLockChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/RetainCountChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/RetainCountChecker.cpp new file mode 100644 index 0000000..b569e41 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/RetainCountChecker.cpp @@ -0,0 +1,3702 @@ +//==-- RetainCountChecker.cpp - Checks for leaks and other issues -*- C++ -*--// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines the methods for RetainCountChecker, which implements +// a reference count checker for Core Foundation and Cocoa on (Mac OS X). +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/AST/DeclObjC.h" +#include "clang/AST/DeclCXX.h" +#include "clang/Basic/LangOptions.h" +#include "clang/Basic/SourceManager.h" +#include "clang/Analysis/DomainSpecific/CocoaConventions.h" +#include "clang/AST/ParentMap.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/StaticAnalyzer/Core/BugReporter/PathDiagnostic.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramStateTrait.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/SymbolManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ObjCMessage.h" +#include "llvm/ADT/DenseMap.h" +#include "llvm/ADT/FoldingSet.h" +#include "llvm/ADT/ImmutableList.h" +#include "llvm/ADT/ImmutableMap.h" +#include "llvm/ADT/SmallString.h" +#include "llvm/ADT/STLExtras.h" +#include "llvm/ADT/StringExtras.h" +#include <cstdarg> + +using namespace clang; +using namespace ento; +using llvm::StrInStrNoCase; + +namespace { +/// Wrapper around different kinds of node builder, so that helper functions +/// can have a common interface. +class GenericNodeBuilderRefCount { + CheckerContext *C; + const ProgramPointTag *tag; +public: + GenericNodeBuilderRefCount(CheckerContext &c, + const ProgramPointTag *t = 0) + : C(&c), tag(t){} + + ExplodedNode *MakeNode(ProgramStateRef state, ExplodedNode *Pred, + bool MarkAsSink = false) { + return C->addTransition(state, Pred, tag, MarkAsSink); + } +}; +} // end anonymous namespace + +//===----------------------------------------------------------------------===// +// Primitives used for constructing summaries for function/method calls. +//===----------------------------------------------------------------------===// + +/// ArgEffect is used to summarize a function/method call's effect on a +/// particular argument. +enum ArgEffect { DoNothing, Autorelease, Dealloc, DecRef, DecRefMsg, + DecRefBridgedTransfered, + IncRefMsg, IncRef, MakeCollectable, MayEscape, + NewAutoreleasePool, SelfOwn, StopTracking }; + +namespace llvm { +template <> struct FoldingSetTrait<ArgEffect> { +static inline void Profile(const ArgEffect X, FoldingSetNodeID& ID) { + ID.AddInteger((unsigned) X); +} +}; +} // end llvm namespace + +/// ArgEffects summarizes the effects of a function/method call on all of +/// its arguments. +typedef llvm::ImmutableMap<unsigned,ArgEffect> ArgEffects; + +namespace { + +/// RetEffect is used to summarize a function/method call's behavior with +/// respect to its return value. +class RetEffect { +public: + enum Kind { NoRet, OwnedSymbol, OwnedAllocatedSymbol, + NotOwnedSymbol, GCNotOwnedSymbol, ARCNotOwnedSymbol, + OwnedWhenTrackedReceiver }; + + enum ObjKind { CF, ObjC, AnyObj }; + +private: + Kind K; + ObjKind O; + + RetEffect(Kind k, ObjKind o = AnyObj) : K(k), O(o) {} + +public: + Kind getKind() const { return K; } + + ObjKind getObjKind() const { return O; } + + bool isOwned() const { + return K == OwnedSymbol || K == OwnedAllocatedSymbol || + K == OwnedWhenTrackedReceiver; + } + + bool operator==(const RetEffect &Other) const { + return K == Other.K && O == Other.O; + } + + static RetEffect MakeOwnedWhenTrackedReceiver() { + return RetEffect(OwnedWhenTrackedReceiver, ObjC); + } + + static RetEffect MakeOwned(ObjKind o, bool isAllocated = false) { + return RetEffect(isAllocated ? OwnedAllocatedSymbol : OwnedSymbol, o); + } + static RetEffect MakeNotOwned(ObjKind o) { + return RetEffect(NotOwnedSymbol, o); + } + static RetEffect MakeGCNotOwned() { + return RetEffect(GCNotOwnedSymbol, ObjC); + } + static RetEffect MakeARCNotOwned() { + return RetEffect(ARCNotOwnedSymbol, ObjC); + } + static RetEffect MakeNoRet() { + return RetEffect(NoRet); + } + + void Profile(llvm::FoldingSetNodeID& ID) const { + ID.AddInteger((unsigned) K); + ID.AddInteger((unsigned) O); + } +}; + +//===----------------------------------------------------------------------===// +// Reference-counting logic (typestate + counts). +//===----------------------------------------------------------------------===// + +class RefVal { +public: + enum Kind { + Owned = 0, // Owning reference. + NotOwned, // Reference is not owned by still valid (not freed). + Released, // Object has been released. + ReturnedOwned, // Returned object passes ownership to caller. + ReturnedNotOwned, // Return object does not pass ownership to caller. + ERROR_START, + ErrorDeallocNotOwned, // -dealloc called on non-owned object. + ErrorDeallocGC, // Calling -dealloc with GC enabled. + ErrorUseAfterRelease, // Object used after released. + ErrorReleaseNotOwned, // Release of an object that was not owned. + ERROR_LEAK_START, + ErrorLeak, // A memory leak due to excessive reference counts. + ErrorLeakReturned, // A memory leak due to the returning method not having + // the correct naming conventions. + ErrorGCLeakReturned, + ErrorOverAutorelease, + ErrorReturnedNotOwned + }; + +private: + Kind kind; + RetEffect::ObjKind okind; + unsigned Cnt; + unsigned ACnt; + QualType T; + + RefVal(Kind k, RetEffect::ObjKind o, unsigned cnt, unsigned acnt, QualType t) + : kind(k), okind(o), Cnt(cnt), ACnt(acnt), T(t) {} + +public: + Kind getKind() const { return kind; } + + RetEffect::ObjKind getObjKind() const { return okind; } + + unsigned getCount() const { return Cnt; } + unsigned getAutoreleaseCount() const { return ACnt; } + unsigned getCombinedCounts() const { return Cnt + ACnt; } + void clearCounts() { Cnt = 0; ACnt = 0; } + void setCount(unsigned i) { Cnt = i; } + void setAutoreleaseCount(unsigned i) { ACnt = i; } + + QualType getType() const { return T; } + + bool isOwned() const { + return getKind() == Owned; + } + + bool isNotOwned() const { + return getKind() == NotOwned; + } + + bool isReturnedOwned() const { + return getKind() == ReturnedOwned; + } + + bool isReturnedNotOwned() const { + return getKind() == ReturnedNotOwned; + } + + static RefVal makeOwned(RetEffect::ObjKind o, QualType t, + unsigned Count = 1) { + return RefVal(Owned, o, Count, 0, t); + } + + static RefVal makeNotOwned(RetEffect::ObjKind o, QualType t, + unsigned Count = 0) { + return RefVal(NotOwned, o, Count, 0, t); + } + + // Comparison, profiling, and pretty-printing. + + bool operator==(const RefVal& X) const { + return kind == X.kind && Cnt == X.Cnt && T == X.T && ACnt == X.ACnt; + } + + RefVal operator-(size_t i) const { + return RefVal(getKind(), getObjKind(), getCount() - i, + getAutoreleaseCount(), getType()); + } + + RefVal operator+(size_t i) const { + return RefVal(getKind(), getObjKind(), getCount() + i, + getAutoreleaseCount(), getType()); + } + + RefVal operator^(Kind k) const { + return RefVal(k, getObjKind(), getCount(), getAutoreleaseCount(), + getType()); + } + + RefVal autorelease() const { + return RefVal(getKind(), getObjKind(), getCount(), getAutoreleaseCount()+1, + getType()); + } + + void Profile(llvm::FoldingSetNodeID& ID) const { + ID.AddInteger((unsigned) kind); + ID.AddInteger(Cnt); + ID.AddInteger(ACnt); + ID.Add(T); + } + + void print(raw_ostream &Out) const; +}; + +void RefVal::print(raw_ostream &Out) const { + if (!T.isNull()) + Out << "Tracked " << T.getAsString() << '/'; + + switch (getKind()) { + default: llvm_unreachable("Invalid RefVal kind"); + case Owned: { + Out << "Owned"; + unsigned cnt = getCount(); + if (cnt) Out << " (+ " << cnt << ")"; + break; + } + + case NotOwned: { + Out << "NotOwned"; + unsigned cnt = getCount(); + if (cnt) Out << " (+ " << cnt << ")"; + break; + } + + case ReturnedOwned: { + Out << "ReturnedOwned"; + unsigned cnt = getCount(); + if (cnt) Out << " (+ " << cnt << ")"; + break; + } + + case ReturnedNotOwned: { + Out << "ReturnedNotOwned"; + unsigned cnt = getCount(); + if (cnt) Out << " (+ " << cnt << ")"; + break; + } + + case Released: + Out << "Released"; + break; + + case ErrorDeallocGC: + Out << "-dealloc (GC)"; + break; + + case ErrorDeallocNotOwned: + Out << "-dealloc (not-owned)"; + break; + + case ErrorLeak: + Out << "Leaked"; + break; + + case ErrorLeakReturned: + Out << "Leaked (Bad naming)"; + break; + + case ErrorGCLeakReturned: + Out << "Leaked (GC-ed at return)"; + break; + + case ErrorUseAfterRelease: + Out << "Use-After-Release [ERROR]"; + break; + + case ErrorReleaseNotOwned: + Out << "Release of Not-Owned [ERROR]"; + break; + + case RefVal::ErrorOverAutorelease: + Out << "Over autoreleased"; + break; + + case RefVal::ErrorReturnedNotOwned: + Out << "Non-owned object returned instead of owned"; + break; + } + + if (ACnt) { + Out << " [ARC +" << ACnt << ']'; + } +} +} //end anonymous namespace + +//===----------------------------------------------------------------------===// +// RefBindings - State used to track object reference counts. +//===----------------------------------------------------------------------===// + +typedef llvm::ImmutableMap<SymbolRef, RefVal> RefBindings; + +namespace clang { +namespace ento { +template<> +struct ProgramStateTrait<RefBindings> + : public ProgramStatePartialTrait<RefBindings> { + static void *GDMIndex() { + static int RefBIndex = 0; + return &RefBIndex; + } +}; +} +} + +//===----------------------------------------------------------------------===// +// Function/Method behavior summaries. +//===----------------------------------------------------------------------===// + +namespace { +class RetainSummary { + /// Args - a map of (index, ArgEffect) pairs, where index + /// specifies the argument (starting from 0). This can be sparsely + /// populated; arguments with no entry in Args use 'DefaultArgEffect'. + ArgEffects Args; + + /// DefaultArgEffect - The default ArgEffect to apply to arguments that + /// do not have an entry in Args. + ArgEffect DefaultArgEffect; + + /// Receiver - If this summary applies to an Objective-C message expression, + /// this is the effect applied to the state of the receiver. + ArgEffect Receiver; + + /// Ret - The effect on the return value. Used to indicate if the + /// function/method call returns a new tracked symbol. + RetEffect Ret; + +public: + RetainSummary(ArgEffects A, RetEffect R, ArgEffect defaultEff, + ArgEffect ReceiverEff) + : Args(A), DefaultArgEffect(defaultEff), Receiver(ReceiverEff), Ret(R) {} + + /// getArg - Return the argument effect on the argument specified by + /// idx (starting from 0). + ArgEffect getArg(unsigned idx) const { + if (const ArgEffect *AE = Args.lookup(idx)) + return *AE; + + return DefaultArgEffect; + } + + void addArg(ArgEffects::Factory &af, unsigned idx, ArgEffect e) { + Args = af.add(Args, idx, e); + } + + /// setDefaultArgEffect - Set the default argument effect. + void setDefaultArgEffect(ArgEffect E) { + DefaultArgEffect = E; + } + + /// getRetEffect - Returns the effect on the return value of the call. + RetEffect getRetEffect() const { return Ret; } + + /// setRetEffect - Set the effect of the return value of the call. + void setRetEffect(RetEffect E) { Ret = E; } + + + /// Sets the effect on the receiver of the message. + void setReceiverEffect(ArgEffect e) { Receiver = e; } + + /// getReceiverEffect - Returns the effect on the receiver of the call. + /// This is only meaningful if the summary applies to an ObjCMessageExpr*. + ArgEffect getReceiverEffect() const { return Receiver; } + + /// Test if two retain summaries are identical. Note that merely equivalent + /// summaries are not necessarily identical (for example, if an explicit + /// argument effect matches the default effect). + bool operator==(const RetainSummary &Other) const { + return Args == Other.Args && DefaultArgEffect == Other.DefaultArgEffect && + Receiver == Other.Receiver && Ret == Other.Ret; + } + + /// Profile this summary for inclusion in a FoldingSet. + void Profile(llvm::FoldingSetNodeID& ID) const { + ID.Add(Args); + ID.Add(DefaultArgEffect); + ID.Add(Receiver); + ID.Add(Ret); + } + + /// A retain summary is simple if it has no ArgEffects other than the default. + bool isSimple() const { + return Args.isEmpty(); + } +}; +} // end anonymous namespace + +//===----------------------------------------------------------------------===// +// Data structures for constructing summaries. +//===----------------------------------------------------------------------===// + +namespace { +class ObjCSummaryKey { + IdentifierInfo* II; + Selector S; +public: + ObjCSummaryKey(IdentifierInfo* ii, Selector s) + : II(ii), S(s) {} + + ObjCSummaryKey(const ObjCInterfaceDecl *d, Selector s) + : II(d ? d->getIdentifier() : 0), S(s) {} + + ObjCSummaryKey(const ObjCInterfaceDecl *d, IdentifierInfo *ii, Selector s) + : II(d ? d->getIdentifier() : ii), S(s) {} + + ObjCSummaryKey(Selector s) + : II(0), S(s) {} + + IdentifierInfo *getIdentifier() const { return II; } + Selector getSelector() const { return S; } +}; +} + +namespace llvm { +template <> struct DenseMapInfo<ObjCSummaryKey> { + static inline ObjCSummaryKey getEmptyKey() { + return ObjCSummaryKey(DenseMapInfo<IdentifierInfo*>::getEmptyKey(), + DenseMapInfo<Selector>::getEmptyKey()); + } + + static inline ObjCSummaryKey getTombstoneKey() { + return ObjCSummaryKey(DenseMapInfo<IdentifierInfo*>::getTombstoneKey(), + DenseMapInfo<Selector>::getTombstoneKey()); + } + + static unsigned getHashValue(const ObjCSummaryKey &V) { + return (DenseMapInfo<IdentifierInfo*>::getHashValue(V.getIdentifier()) + & 0x88888888) + | (DenseMapInfo<Selector>::getHashValue(V.getSelector()) + & 0x55555555); + } + + static bool isEqual(const ObjCSummaryKey& LHS, const ObjCSummaryKey& RHS) { + return DenseMapInfo<IdentifierInfo*>::isEqual(LHS.getIdentifier(), + RHS.getIdentifier()) && + DenseMapInfo<Selector>::isEqual(LHS.getSelector(), + RHS.getSelector()); + } + +}; +template <> +struct isPodLike<ObjCSummaryKey> { static const bool value = true; }; +} // end llvm namespace + +namespace { +class ObjCSummaryCache { + typedef llvm::DenseMap<ObjCSummaryKey, const RetainSummary *> MapTy; + MapTy M; +public: + ObjCSummaryCache() {} + + const RetainSummary * find(const ObjCInterfaceDecl *D, IdentifierInfo *ClsName, + Selector S) { + // Lookup the method using the decl for the class @interface. If we + // have no decl, lookup using the class name. + return D ? find(D, S) : find(ClsName, S); + } + + const RetainSummary * find(const ObjCInterfaceDecl *D, Selector S) { + // Do a lookup with the (D,S) pair. If we find a match return + // the iterator. + ObjCSummaryKey K(D, S); + MapTy::iterator I = M.find(K); + + if (I != M.end() || !D) + return I->second; + + // Walk the super chain. If we find a hit with a parent, we'll end + // up returning that summary. We actually allow that key (null,S), as + // we cache summaries for the null ObjCInterfaceDecl* to allow us to + // generate initial summaries without having to worry about NSObject + // being declared. + // FIXME: We may change this at some point. + for (ObjCInterfaceDecl *C=D->getSuperClass() ;; C=C->getSuperClass()) { + if ((I = M.find(ObjCSummaryKey(C, S))) != M.end()) + break; + + if (!C) + return NULL; + } + + // Cache the summary with original key to make the next lookup faster + // and return the iterator. + const RetainSummary *Summ = I->second; + M[K] = Summ; + return Summ; + } + + const RetainSummary *find(IdentifierInfo* II, Selector S) { + // FIXME: Class method lookup. Right now we dont' have a good way + // of going between IdentifierInfo* and the class hierarchy. + MapTy::iterator I = M.find(ObjCSummaryKey(II, S)); + + if (I == M.end()) + I = M.find(ObjCSummaryKey(S)); + + return I == M.end() ? NULL : I->second; + } + + const RetainSummary *& operator[](ObjCSummaryKey K) { + return M[K]; + } + + const RetainSummary *& operator[](Selector S) { + return M[ ObjCSummaryKey(S) ]; + } +}; +} // end anonymous namespace + +//===----------------------------------------------------------------------===// +// Data structures for managing collections of summaries. +//===----------------------------------------------------------------------===// + +namespace { +class RetainSummaryManager { + + //==-----------------------------------------------------------------==// + // Typedefs. + //==-----------------------------------------------------------------==// + + typedef llvm::DenseMap<const FunctionDecl*, const RetainSummary *> + FuncSummariesTy; + + typedef ObjCSummaryCache ObjCMethodSummariesTy; + + typedef llvm::FoldingSetNodeWrapper<RetainSummary> CachedSummaryNode; + + //==-----------------------------------------------------------------==// + // Data. + //==-----------------------------------------------------------------==// + + /// Ctx - The ASTContext object for the analyzed ASTs. + ASTContext &Ctx; + + /// GCEnabled - Records whether or not the analyzed code runs in GC mode. + const bool GCEnabled; + + /// Records whether or not the analyzed code runs in ARC mode. + const bool ARCEnabled; + + /// FuncSummaries - A map from FunctionDecls to summaries. + FuncSummariesTy FuncSummaries; + + /// ObjCClassMethodSummaries - A map from selectors (for instance methods) + /// to summaries. + ObjCMethodSummariesTy ObjCClassMethodSummaries; + + /// ObjCMethodSummaries - A map from selectors to summaries. + ObjCMethodSummariesTy ObjCMethodSummaries; + + /// BPAlloc - A BumpPtrAllocator used for allocating summaries, ArgEffects, + /// and all other data used by the checker. + llvm::BumpPtrAllocator BPAlloc; + + /// AF - A factory for ArgEffects objects. + ArgEffects::Factory AF; + + /// ScratchArgs - A holding buffer for construct ArgEffects. + ArgEffects ScratchArgs; + + /// ObjCAllocRetE - Default return effect for methods returning Objective-C + /// objects. + RetEffect ObjCAllocRetE; + + /// ObjCInitRetE - Default return effect for init methods returning + /// Objective-C objects. + RetEffect ObjCInitRetE; + + /// SimpleSummaries - Used for uniquing summaries that don't have special + /// effects. + llvm::FoldingSet<CachedSummaryNode> SimpleSummaries; + + //==-----------------------------------------------------------------==// + // Methods. + //==-----------------------------------------------------------------==// + + /// getArgEffects - Returns a persistent ArgEffects object based on the + /// data in ScratchArgs. + ArgEffects getArgEffects(); + + enum UnaryFuncKind { cfretain, cfrelease, cfmakecollectable }; + +public: + RetEffect getObjAllocRetEffect() const { return ObjCAllocRetE; } + + const RetainSummary *getUnarySummary(const FunctionType* FT, + UnaryFuncKind func); + + const RetainSummary *getCFSummaryCreateRule(const FunctionDecl *FD); + const RetainSummary *getCFSummaryGetRule(const FunctionDecl *FD); + const RetainSummary *getCFCreateGetRuleSummary(const FunctionDecl *FD); + + const RetainSummary *getPersistentSummary(const RetainSummary &OldSumm); + + const RetainSummary *getPersistentSummary(RetEffect RetEff, + ArgEffect ReceiverEff = DoNothing, + ArgEffect DefaultEff = MayEscape) { + RetainSummary Summ(getArgEffects(), RetEff, DefaultEff, ReceiverEff); + return getPersistentSummary(Summ); + } + + const RetainSummary *getDefaultSummary() { + return getPersistentSummary(RetEffect::MakeNoRet(), + DoNothing, MayEscape); + } + + const RetainSummary *getPersistentStopSummary() { + return getPersistentSummary(RetEffect::MakeNoRet(), + StopTracking, StopTracking); + } + + void InitializeClassMethodSummaries(); + void InitializeMethodSummaries(); +private: + void addNSObjectClsMethSummary(Selector S, const RetainSummary *Summ) { + ObjCClassMethodSummaries[S] = Summ; + } + + void addNSObjectMethSummary(Selector S, const RetainSummary *Summ) { + ObjCMethodSummaries[S] = Summ; + } + + void addClassMethSummary(const char* Cls, const char* name, + const RetainSummary *Summ, bool isNullary = true) { + IdentifierInfo* ClsII = &Ctx.Idents.get(Cls); + Selector S = isNullary ? GetNullarySelector(name, Ctx) + : GetUnarySelector(name, Ctx); + ObjCClassMethodSummaries[ObjCSummaryKey(ClsII, S)] = Summ; + } + + void addInstMethSummary(const char* Cls, const char* nullaryName, + const RetainSummary *Summ) { + IdentifierInfo* ClsII = &Ctx.Idents.get(Cls); + Selector S = GetNullarySelector(nullaryName, Ctx); + ObjCMethodSummaries[ObjCSummaryKey(ClsII, S)] = Summ; + } + + Selector generateSelector(va_list argp) { + SmallVector<IdentifierInfo*, 10> II; + + while (const char* s = va_arg(argp, const char*)) + II.push_back(&Ctx.Idents.get(s)); + + return Ctx.Selectors.getSelector(II.size(), &II[0]); + } + + void addMethodSummary(IdentifierInfo *ClsII, ObjCMethodSummariesTy& Summaries, + const RetainSummary * Summ, va_list argp) { + Selector S = generateSelector(argp); + Summaries[ObjCSummaryKey(ClsII, S)] = Summ; + } + + void addInstMethSummary(const char* Cls, const RetainSummary * Summ, ...) { + va_list argp; + va_start(argp, Summ); + addMethodSummary(&Ctx.Idents.get(Cls), ObjCMethodSummaries, Summ, argp); + va_end(argp); + } + + void addClsMethSummary(const char* Cls, const RetainSummary * Summ, ...) { + va_list argp; + va_start(argp, Summ); + addMethodSummary(&Ctx.Idents.get(Cls),ObjCClassMethodSummaries, Summ, argp); + va_end(argp); + } + + void addClsMethSummary(IdentifierInfo *II, const RetainSummary * Summ, ...) { + va_list argp; + va_start(argp, Summ); + addMethodSummary(II, ObjCClassMethodSummaries, Summ, argp); + va_end(argp); + } + +public: + + RetainSummaryManager(ASTContext &ctx, bool gcenabled, bool usesARC) + : Ctx(ctx), + GCEnabled(gcenabled), + ARCEnabled(usesARC), + AF(BPAlloc), ScratchArgs(AF.getEmptyMap()), + ObjCAllocRetE(gcenabled + ? RetEffect::MakeGCNotOwned() + : (usesARC ? RetEffect::MakeARCNotOwned() + : RetEffect::MakeOwned(RetEffect::ObjC, true))), + ObjCInitRetE(gcenabled + ? RetEffect::MakeGCNotOwned() + : (usesARC ? RetEffect::MakeARCNotOwned() + : RetEffect::MakeOwnedWhenTrackedReceiver())) { + InitializeClassMethodSummaries(); + InitializeMethodSummaries(); + } + + const RetainSummary *getSummary(const FunctionDecl *FD); + + const RetainSummary *getMethodSummary(Selector S, IdentifierInfo *ClsName, + const ObjCInterfaceDecl *ID, + const ObjCMethodDecl *MD, + QualType RetTy, + ObjCMethodSummariesTy &CachedSummaries); + + const RetainSummary *getInstanceMethodSummary(const ObjCMessage &msg, + ProgramStateRef state, + const LocationContext *LC); + + const RetainSummary *getInstanceMethodSummary(const ObjCMessage &msg, + const ObjCInterfaceDecl *ID) { + return getMethodSummary(msg.getSelector(), 0, ID, msg.getMethodDecl(), + msg.getType(Ctx), ObjCMethodSummaries); + } + + const RetainSummary *getClassMethodSummary(const ObjCMessage &msg) { + const ObjCInterfaceDecl *Class = 0; + if (!msg.isInstanceMessage()) + Class = msg.getReceiverInterface(); + + return getMethodSummary(msg.getSelector(), Class->getIdentifier(), + Class, msg.getMethodDecl(), msg.getType(Ctx), + ObjCClassMethodSummaries); + } + + /// getMethodSummary - This version of getMethodSummary is used to query + /// the summary for the current method being analyzed. + const RetainSummary *getMethodSummary(const ObjCMethodDecl *MD) { + // FIXME: Eventually this should be unneeded. + const ObjCInterfaceDecl *ID = MD->getClassInterface(); + Selector S = MD->getSelector(); + IdentifierInfo *ClsName = ID->getIdentifier(); + QualType ResultTy = MD->getResultType(); + + ObjCMethodSummariesTy *CachedSummaries; + if (MD->isInstanceMethod()) + CachedSummaries = &ObjCMethodSummaries; + else + CachedSummaries = &ObjCClassMethodSummaries; + + return getMethodSummary(S, ClsName, ID, MD, ResultTy, *CachedSummaries); + } + + const RetainSummary *getStandardMethodSummary(const ObjCMethodDecl *MD, + Selector S, QualType RetTy); + + void updateSummaryFromAnnotations(const RetainSummary *&Summ, + const ObjCMethodDecl *MD); + + void updateSummaryFromAnnotations(const RetainSummary *&Summ, + const FunctionDecl *FD); + + bool isGCEnabled() const { return GCEnabled; } + + bool isARCEnabled() const { return ARCEnabled; } + + bool isARCorGCEnabled() const { return GCEnabled || ARCEnabled; } +}; + +// Used to avoid allocating long-term (BPAlloc'd) memory for default retain +// summaries. If a function or method looks like it has a default summary, but +// it has annotations, the annotations are added to the stack-based template +// and then copied into managed memory. +class RetainSummaryTemplate { + RetainSummaryManager &Manager; + const RetainSummary *&RealSummary; + RetainSummary ScratchSummary; + bool Accessed; +public: + RetainSummaryTemplate(const RetainSummary *&real, const RetainSummary &base, + RetainSummaryManager &mgr) + : Manager(mgr), RealSummary(real), ScratchSummary(real ? *real : base), + Accessed(false) {} + + ~RetainSummaryTemplate() { + if (Accessed) + RealSummary = Manager.getPersistentSummary(ScratchSummary); + } + + RetainSummary &operator*() { + Accessed = true; + return ScratchSummary; + } + + RetainSummary *operator->() { + Accessed = true; + return &ScratchSummary; + } +}; + +} // end anonymous namespace + +//===----------------------------------------------------------------------===// +// Implementation of checker data structures. +//===----------------------------------------------------------------------===// + +ArgEffects RetainSummaryManager::getArgEffects() { + ArgEffects AE = ScratchArgs; + ScratchArgs = AF.getEmptyMap(); + return AE; +} + +const RetainSummary * +RetainSummaryManager::getPersistentSummary(const RetainSummary &OldSumm) { + // Unique "simple" summaries -- those without ArgEffects. + if (OldSumm.isSimple()) { + llvm::FoldingSetNodeID ID; + OldSumm.Profile(ID); + + void *Pos; + CachedSummaryNode *N = SimpleSummaries.FindNodeOrInsertPos(ID, Pos); + + if (!N) { + N = (CachedSummaryNode *) BPAlloc.Allocate<CachedSummaryNode>(); + new (N) CachedSummaryNode(OldSumm); + SimpleSummaries.InsertNode(N, Pos); + } + + return &N->getValue(); + } + + RetainSummary *Summ = (RetainSummary *) BPAlloc.Allocate<RetainSummary>(); + new (Summ) RetainSummary(OldSumm); + return Summ; +} + +//===----------------------------------------------------------------------===// +// Summary creation for functions (largely uses of Core Foundation). +//===----------------------------------------------------------------------===// + +static bool isRetain(const FunctionDecl *FD, StringRef FName) { + return FName.endswith("Retain"); +} + +static bool isRelease(const FunctionDecl *FD, StringRef FName) { + return FName.endswith("Release"); +} + +static bool isMakeCollectable(const FunctionDecl *FD, StringRef FName) { + // FIXME: Remove FunctionDecl parameter. + // FIXME: Is it really okay if MakeCollectable isn't a suffix? + return FName.find("MakeCollectable") != StringRef::npos; +} + +const RetainSummary * RetainSummaryManager::getSummary(const FunctionDecl *FD) { + // Look up a summary in our cache of FunctionDecls -> Summaries. + FuncSummariesTy::iterator I = FuncSummaries.find(FD); + if (I != FuncSummaries.end()) + return I->second; + + // No summary? Generate one. + const RetainSummary *S = 0; + + do { + // We generate "stop" summaries for implicitly defined functions. + if (FD->isImplicit()) { + S = getPersistentStopSummary(); + break; + } + // For C++ methods, generate an implicit "stop" summary as well. We + // can relax this once we have a clear policy for C++ methods and + // ownership attributes. + if (isa<CXXMethodDecl>(FD)) { + S = getPersistentStopSummary(); + break; + } + + // [PR 3337] Use 'getAs<FunctionType>' to strip away any typedefs on the + // function's type. + const FunctionType* FT = FD->getType()->getAs<FunctionType>(); + const IdentifierInfo *II = FD->getIdentifier(); + if (!II) + break; + + StringRef FName = II->getName(); + + // Strip away preceding '_'. Doing this here will effect all the checks + // down below. + FName = FName.substr(FName.find_first_not_of('_')); + + // Inspect the result type. + QualType RetTy = FT->getResultType(); + + // FIXME: This should all be refactored into a chain of "summary lookup" + // filters. + assert(ScratchArgs.isEmpty()); + + if (FName == "pthread_create") { + // Part of: <rdar://problem/7299394>. This will be addressed + // better with IPA. + S = getPersistentStopSummary(); + } else if (FName == "NSMakeCollectable") { + // Handle: id NSMakeCollectable(CFTypeRef) + S = (RetTy->isObjCIdType()) + ? getUnarySummary(FT, cfmakecollectable) + : getPersistentStopSummary(); + } else if (FName == "IOBSDNameMatching" || + FName == "IOServiceMatching" || + FName == "IOServiceNameMatching" || + FName == "IORegistryEntryIDMatching" || + FName == "IOOpenFirmwarePathMatching") { + // Part of <rdar://problem/6961230>. (IOKit) + // This should be addressed using a API table. + S = getPersistentSummary(RetEffect::MakeOwned(RetEffect::CF, true), + DoNothing, DoNothing); + } else if (FName == "IOServiceGetMatchingService" || + FName == "IOServiceGetMatchingServices") { + // FIXES: <rdar://problem/6326900> + // This should be addressed using a API table. This strcmp is also + // a little gross, but there is no need to super optimize here. + ScratchArgs = AF.add(ScratchArgs, 1, DecRef); + S = getPersistentSummary(RetEffect::MakeNoRet(), DoNothing, DoNothing); + } else if (FName == "IOServiceAddNotification" || + FName == "IOServiceAddMatchingNotification") { + // Part of <rdar://problem/6961230>. (IOKit) + // This should be addressed using a API table. + ScratchArgs = AF.add(ScratchArgs, 2, DecRef); + S = getPersistentSummary(RetEffect::MakeNoRet(), DoNothing, DoNothing); + } else if (FName == "CVPixelBufferCreateWithBytes") { + // FIXES: <rdar://problem/7283567> + // Eventually this can be improved by recognizing that the pixel + // buffer passed to CVPixelBufferCreateWithBytes is released via + // a callback and doing full IPA to make sure this is done correctly. + // FIXME: This function has an out parameter that returns an + // allocated object. + ScratchArgs = AF.add(ScratchArgs, 7, StopTracking); + S = getPersistentSummary(RetEffect::MakeNoRet(), DoNothing, DoNothing); + } else if (FName == "CGBitmapContextCreateWithData") { + // FIXES: <rdar://problem/7358899> + // Eventually this can be improved by recognizing that 'releaseInfo' + // passed to CGBitmapContextCreateWithData is released via + // a callback and doing full IPA to make sure this is done correctly. + ScratchArgs = AF.add(ScratchArgs, 8, StopTracking); + S = getPersistentSummary(RetEffect::MakeOwned(RetEffect::CF, true), + DoNothing, DoNothing); + } else if (FName == "CVPixelBufferCreateWithPlanarBytes") { + // FIXES: <rdar://problem/7283567> + // Eventually this can be improved by recognizing that the pixel + // buffer passed to CVPixelBufferCreateWithPlanarBytes is released + // via a callback and doing full IPA to make sure this is done + // correctly. + ScratchArgs = AF.add(ScratchArgs, 12, StopTracking); + S = getPersistentSummary(RetEffect::MakeNoRet(), DoNothing, DoNothing); + } else if (FName == "dispatch_set_context") { + // <rdar://problem/11059275> - The analyzer currently doesn't have + // a good way to reason about the finalizer function for libdispatch. + // If we pass a context object that is memory managed, stop tracking it. + // FIXME: this hack should possibly go away once we can handle + // libdispatch finalizers. + ScratchArgs = AF.add(ScratchArgs, 1, StopTracking); + S = getPersistentSummary(RetEffect::MakeNoRet(), DoNothing, DoNothing); + } else if (FName.startswith("NS") && + (FName.find("Insert") != StringRef::npos)) { + // Whitelist NSXXInsertXX, for example NSMapInsertIfAbsent, since they can + // be deallocated by NSMapRemove. (radar://11152419) + ScratchArgs = AF.add(ScratchArgs, 1, StopTracking); + ScratchArgs = AF.add(ScratchArgs, 2, StopTracking); + S = getPersistentSummary(RetEffect::MakeNoRet(), DoNothing, DoNothing); + } + + // Did we get a summary? + if (S) + break; + + // Enable this code once the semantics of NSDeallocateObject are resolved + // for GC. <rdar://problem/6619988> +#if 0 + // Handle: NSDeallocateObject(id anObject); + // This method does allow 'nil' (although we don't check it now). + if (strcmp(FName, "NSDeallocateObject") == 0) { + return RetTy == Ctx.VoidTy + ? getPersistentSummary(RetEffect::MakeNoRet(), DoNothing, Dealloc) + : getPersistentStopSummary(); + } +#endif + + if (RetTy->isPointerType()) { + // For CoreFoundation ('CF') types. + if (cocoa::isRefType(RetTy, "CF", FName)) { + if (isRetain(FD, FName)) + S = getUnarySummary(FT, cfretain); + else if (isMakeCollectable(FD, FName)) + S = getUnarySummary(FT, cfmakecollectable); + else + S = getCFCreateGetRuleSummary(FD); + + break; + } + + // For CoreGraphics ('CG') types. + if (cocoa::isRefType(RetTy, "CG", FName)) { + if (isRetain(FD, FName)) + S = getUnarySummary(FT, cfretain); + else + S = getCFCreateGetRuleSummary(FD); + + break; + } + + // For the Disk Arbitration API (DiskArbitration/DADisk.h) + if (cocoa::isRefType(RetTy, "DADisk") || + cocoa::isRefType(RetTy, "DADissenter") || + cocoa::isRefType(RetTy, "DASessionRef")) { + S = getCFCreateGetRuleSummary(FD); + break; + } + + break; + } + + // Check for release functions, the only kind of functions that we care + // about that don't return a pointer type. + if (FName[0] == 'C' && (FName[1] == 'F' || FName[1] == 'G')) { + // Test for 'CGCF'. + FName = FName.substr(FName.startswith("CGCF") ? 4 : 2); + + if (isRelease(FD, FName)) + S = getUnarySummary(FT, cfrelease); + else { + assert (ScratchArgs.isEmpty()); + // Remaining CoreFoundation and CoreGraphics functions. + // We use to assume that they all strictly followed the ownership idiom + // and that ownership cannot be transferred. While this is technically + // correct, many methods allow a tracked object to escape. For example: + // + // CFMutableDictionaryRef x = CFDictionaryCreateMutable(...); + // CFDictionaryAddValue(y, key, x); + // CFRelease(x); + // ... it is okay to use 'x' since 'y' has a reference to it + // + // We handle this and similar cases with the follow heuristic. If the + // function name contains "InsertValue", "SetValue", "AddValue", + // "AppendValue", or "SetAttribute", then we assume that arguments may + // "escape." This means that something else holds on to the object, + // allowing it be used even after its local retain count drops to 0. + ArgEffect E = (StrInStrNoCase(FName, "InsertValue") != StringRef::npos|| + StrInStrNoCase(FName, "AddValue") != StringRef::npos || + StrInStrNoCase(FName, "SetValue") != StringRef::npos || + StrInStrNoCase(FName, "AppendValue") != StringRef::npos|| + StrInStrNoCase(FName, "SetAttribute") != StringRef::npos) + ? MayEscape : DoNothing; + + S = getPersistentSummary(RetEffect::MakeNoRet(), DoNothing, E); + } + } + } + while (0); + + // Annotations override defaults. + updateSummaryFromAnnotations(S, FD); + + FuncSummaries[FD] = S; + return S; +} + +const RetainSummary * +RetainSummaryManager::getCFCreateGetRuleSummary(const FunctionDecl *FD) { + if (coreFoundation::followsCreateRule(FD)) + return getCFSummaryCreateRule(FD); + + return getCFSummaryGetRule(FD); +} + +const RetainSummary * +RetainSummaryManager::getUnarySummary(const FunctionType* FT, + UnaryFuncKind func) { + + // Sanity check that this is *really* a unary function. This can + // happen if people do weird things. + const FunctionProtoType* FTP = dyn_cast<FunctionProtoType>(FT); + if (!FTP || FTP->getNumArgs() != 1) + return getPersistentStopSummary(); + + assert (ScratchArgs.isEmpty()); + + ArgEffect Effect; + switch (func) { + case cfretain: Effect = IncRef; break; + case cfrelease: Effect = DecRef; break; + case cfmakecollectable: Effect = MakeCollectable; break; + } + + ScratchArgs = AF.add(ScratchArgs, 0, Effect); + return getPersistentSummary(RetEffect::MakeNoRet(), DoNothing, DoNothing); +} + +const RetainSummary * +RetainSummaryManager::getCFSummaryCreateRule(const FunctionDecl *FD) { + assert (ScratchArgs.isEmpty()); + + return getPersistentSummary(RetEffect::MakeOwned(RetEffect::CF, true)); +} + +const RetainSummary * +RetainSummaryManager::getCFSummaryGetRule(const FunctionDecl *FD) { + assert (ScratchArgs.isEmpty()); + return getPersistentSummary(RetEffect::MakeNotOwned(RetEffect::CF), + DoNothing, DoNothing); +} + +//===----------------------------------------------------------------------===// +// Summary creation for Selectors. +//===----------------------------------------------------------------------===// + +void +RetainSummaryManager::updateSummaryFromAnnotations(const RetainSummary *&Summ, + const FunctionDecl *FD) { + if (!FD) + return; + + RetainSummaryTemplate Template(Summ, *getDefaultSummary(), *this); + + // Effects on the parameters. + unsigned parm_idx = 0; + for (FunctionDecl::param_const_iterator pi = FD->param_begin(), + pe = FD->param_end(); pi != pe; ++pi, ++parm_idx) { + const ParmVarDecl *pd = *pi; + if (pd->getAttr<NSConsumedAttr>()) { + if (!GCEnabled) { + Template->addArg(AF, parm_idx, DecRef); + } + } else if (pd->getAttr<CFConsumedAttr>()) { + Template->addArg(AF, parm_idx, DecRef); + } + } + + QualType RetTy = FD->getResultType(); + + // Determine if there is a special return effect for this method. + if (cocoa::isCocoaObjectRef(RetTy)) { + if (FD->getAttr<NSReturnsRetainedAttr>()) { + Template->setRetEffect(ObjCAllocRetE); + } + else if (FD->getAttr<CFReturnsRetainedAttr>()) { + Template->setRetEffect(RetEffect::MakeOwned(RetEffect::CF, true)); + } + else if (FD->getAttr<NSReturnsNotRetainedAttr>()) { + Template->setRetEffect(RetEffect::MakeNotOwned(RetEffect::ObjC)); + } + else if (FD->getAttr<CFReturnsNotRetainedAttr>()) { + Template->setRetEffect(RetEffect::MakeNotOwned(RetEffect::CF)); + } + } else if (RetTy->getAs<PointerType>()) { + if (FD->getAttr<CFReturnsRetainedAttr>()) { + Template->setRetEffect(RetEffect::MakeOwned(RetEffect::CF, true)); + } + else if (FD->getAttr<CFReturnsNotRetainedAttr>()) { + Template->setRetEffect(RetEffect::MakeNotOwned(RetEffect::CF)); + } + } +} + +void +RetainSummaryManager::updateSummaryFromAnnotations(const RetainSummary *&Summ, + const ObjCMethodDecl *MD) { + if (!MD) + return; + + RetainSummaryTemplate Template(Summ, *getDefaultSummary(), *this); + bool isTrackedLoc = false; + + // Effects on the receiver. + if (MD->getAttr<NSConsumesSelfAttr>()) { + if (!GCEnabled) + Template->setReceiverEffect(DecRefMsg); + } + + // Effects on the parameters. + unsigned parm_idx = 0; + for (ObjCMethodDecl::param_const_iterator + pi=MD->param_begin(), pe=MD->param_end(); + pi != pe; ++pi, ++parm_idx) { + const ParmVarDecl *pd = *pi; + if (pd->getAttr<NSConsumedAttr>()) { + if (!GCEnabled) + Template->addArg(AF, parm_idx, DecRef); + } + else if(pd->getAttr<CFConsumedAttr>()) { + Template->addArg(AF, parm_idx, DecRef); + } + } + + // Determine if there is a special return effect for this method. + if (cocoa::isCocoaObjectRef(MD->getResultType())) { + if (MD->getAttr<NSReturnsRetainedAttr>()) { + Template->setRetEffect(ObjCAllocRetE); + return; + } + if (MD->getAttr<NSReturnsNotRetainedAttr>()) { + Template->setRetEffect(RetEffect::MakeNotOwned(RetEffect::ObjC)); + return; + } + + isTrackedLoc = true; + } else { + isTrackedLoc = MD->getResultType()->getAs<PointerType>() != NULL; + } + + if (isTrackedLoc) { + if (MD->getAttr<CFReturnsRetainedAttr>()) + Template->setRetEffect(RetEffect::MakeOwned(RetEffect::CF, true)); + else if (MD->getAttr<CFReturnsNotRetainedAttr>()) + Template->setRetEffect(RetEffect::MakeNotOwned(RetEffect::CF)); + } +} + +const RetainSummary * +RetainSummaryManager::getStandardMethodSummary(const ObjCMethodDecl *MD, + Selector S, QualType RetTy) { + + if (MD) { + // Scan the method decl for 'void*' arguments. These should be treated + // as 'StopTracking' because they are often used with delegates. + // Delegates are a frequent form of false positives with the retain + // count checker. + unsigned i = 0; + for (ObjCMethodDecl::param_const_iterator I = MD->param_begin(), + E = MD->param_end(); I != E; ++I, ++i) + if (const ParmVarDecl *PD = *I) { + QualType Ty = Ctx.getCanonicalType(PD->getType()); + if (Ty.getLocalUnqualifiedType() == Ctx.VoidPtrTy) + ScratchArgs = AF.add(ScratchArgs, i, StopTracking); + } + } + + // Any special effects? + ArgEffect ReceiverEff = DoNothing; + RetEffect ResultEff = RetEffect::MakeNoRet(); + + // Check the method family, and apply any default annotations. + switch (MD ? MD->getMethodFamily() : S.getMethodFamily()) { + case OMF_None: + case OMF_performSelector: + // Assume all Objective-C methods follow Cocoa Memory Management rules. + // FIXME: Does the non-threaded performSelector family really belong here? + // The selector could be, say, @selector(copy). + if (cocoa::isCocoaObjectRef(RetTy)) + ResultEff = RetEffect::MakeNotOwned(RetEffect::ObjC); + else if (coreFoundation::isCFObjectRef(RetTy)) { + // ObjCMethodDecl currently doesn't consider CF objects as valid return + // values for alloc, new, copy, or mutableCopy, so we have to + // double-check with the selector. This is ugly, but there aren't that + // many Objective-C methods that return CF objects, right? + if (MD) { + switch (S.getMethodFamily()) { + case OMF_alloc: + case OMF_new: + case OMF_copy: + case OMF_mutableCopy: + ResultEff = RetEffect::MakeOwned(RetEffect::CF, true); + break; + default: + ResultEff = RetEffect::MakeNotOwned(RetEffect::CF); + break; + } + } else { + ResultEff = RetEffect::MakeNotOwned(RetEffect::CF); + } + } + break; + case OMF_init: + ResultEff = ObjCInitRetE; + ReceiverEff = DecRefMsg; + break; + case OMF_alloc: + case OMF_new: + case OMF_copy: + case OMF_mutableCopy: + if (cocoa::isCocoaObjectRef(RetTy)) + ResultEff = ObjCAllocRetE; + else if (coreFoundation::isCFObjectRef(RetTy)) + ResultEff = RetEffect::MakeOwned(RetEffect::CF, true); + break; + case OMF_autorelease: + ReceiverEff = Autorelease; + break; + case OMF_retain: + ReceiverEff = IncRefMsg; + break; + case OMF_release: + ReceiverEff = DecRefMsg; + break; + case OMF_dealloc: + ReceiverEff = Dealloc; + break; + case OMF_self: + // -self is handled specially by the ExprEngine to propagate the receiver. + break; + case OMF_retainCount: + case OMF_finalize: + // These methods don't return objects. + break; + } + + // If one of the arguments in the selector has the keyword 'delegate' we + // should stop tracking the reference count for the receiver. This is + // because the reference count is quite possibly handled by a delegate + // method. + if (S.isKeywordSelector()) { + const std::string &str = S.getAsString(); + assert(!str.empty()); + if (StrInStrNoCase(str, "delegate:") != StringRef::npos) + ReceiverEff = StopTracking; + } + + if (ScratchArgs.isEmpty() && ReceiverEff == DoNothing && + ResultEff.getKind() == RetEffect::NoRet) + return getDefaultSummary(); + + return getPersistentSummary(ResultEff, ReceiverEff, MayEscape); +} + +const RetainSummary * +RetainSummaryManager::getInstanceMethodSummary(const ObjCMessage &msg, + ProgramStateRef state, + const LocationContext *LC) { + + // We need the type-information of the tracked receiver object + // Retrieve it from the state. + const Expr *Receiver = msg.getInstanceReceiver(); + const ObjCInterfaceDecl *ID = 0; + + // FIXME: Is this really working as expected? There are cases where + // we just use the 'ID' from the message expression. + SVal receiverV; + + if (Receiver) { + receiverV = state->getSValAsScalarOrLoc(Receiver, LC); + + // FIXME: Eventually replace the use of state->get<RefBindings> with + // a generic API for reasoning about the Objective-C types of symbolic + // objects. + if (SymbolRef Sym = receiverV.getAsLocSymbol()) + if (const RefVal *T = state->get<RefBindings>(Sym)) + if (const ObjCObjectPointerType* PT = + T->getType()->getAs<ObjCObjectPointerType>()) + ID = PT->getInterfaceDecl(); + + // FIXME: this is a hack. This may or may not be the actual method + // that is called. + if (!ID) { + if (const ObjCObjectPointerType *PT = + Receiver->getType()->getAs<ObjCObjectPointerType>()) + ID = PT->getInterfaceDecl(); + } + } else { + // FIXME: Hack for 'super'. + ID = msg.getReceiverInterface(); + } + + // FIXME: The receiver could be a reference to a class, meaning that + // we should use the class method. + return getInstanceMethodSummary(msg, ID); +} + +const RetainSummary * +RetainSummaryManager::getMethodSummary(Selector S, IdentifierInfo *ClsName, + const ObjCInterfaceDecl *ID, + const ObjCMethodDecl *MD, QualType RetTy, + ObjCMethodSummariesTy &CachedSummaries) { + + // Look up a summary in our summary cache. + const RetainSummary *Summ = CachedSummaries.find(ID, ClsName, S); + + if (!Summ) { + Summ = getStandardMethodSummary(MD, S, RetTy); + + // Annotations override defaults. + updateSummaryFromAnnotations(Summ, MD); + + // Memoize the summary. + CachedSummaries[ObjCSummaryKey(ID, ClsName, S)] = Summ; + } + + return Summ; +} + +void RetainSummaryManager::InitializeClassMethodSummaries() { + assert(ScratchArgs.isEmpty()); + // Create the [NSAssertionHandler currentHander] summary. + addClassMethSummary("NSAssertionHandler", "currentHandler", + getPersistentSummary(RetEffect::MakeNotOwned(RetEffect::ObjC))); + + // Create the [NSAutoreleasePool addObject:] summary. + ScratchArgs = AF.add(ScratchArgs, 0, Autorelease); + addClassMethSummary("NSAutoreleasePool", "addObject", + getPersistentSummary(RetEffect::MakeNoRet(), + DoNothing, Autorelease)); + + // Create the summaries for [NSObject performSelector...]. We treat + // these as 'stop tracking' for the arguments because they are often + // used for delegates that can release the object. When we have better + // inter-procedural analysis we can potentially do something better. This + // workaround is to remove false positives. + const RetainSummary *Summ = + getPersistentSummary(RetEffect::MakeNoRet(), DoNothing, StopTracking); + IdentifierInfo *NSObjectII = &Ctx.Idents.get("NSObject"); + addClsMethSummary(NSObjectII, Summ, "performSelector", "withObject", + "afterDelay", NULL); + addClsMethSummary(NSObjectII, Summ, "performSelector", "withObject", + "afterDelay", "inModes", NULL); + addClsMethSummary(NSObjectII, Summ, "performSelectorOnMainThread", + "withObject", "waitUntilDone", NULL); + addClsMethSummary(NSObjectII, Summ, "performSelectorOnMainThread", + "withObject", "waitUntilDone", "modes", NULL); + addClsMethSummary(NSObjectII, Summ, "performSelector", "onThread", + "withObject", "waitUntilDone", NULL); + addClsMethSummary(NSObjectII, Summ, "performSelector", "onThread", + "withObject", "waitUntilDone", "modes", NULL); + addClsMethSummary(NSObjectII, Summ, "performSelectorInBackground", + "withObject", NULL); +} + +void RetainSummaryManager::InitializeMethodSummaries() { + + assert (ScratchArgs.isEmpty()); + + // Create the "init" selector. It just acts as a pass-through for the + // receiver. + const RetainSummary *InitSumm = getPersistentSummary(ObjCInitRetE, DecRefMsg); + addNSObjectMethSummary(GetNullarySelector("init", Ctx), InitSumm); + + // awakeAfterUsingCoder: behaves basically like an 'init' method. It + // claims the receiver and returns a retained object. + addNSObjectMethSummary(GetUnarySelector("awakeAfterUsingCoder", Ctx), + InitSumm); + + // The next methods are allocators. + const RetainSummary *AllocSumm = getPersistentSummary(ObjCAllocRetE); + const RetainSummary *CFAllocSumm = + getPersistentSummary(RetEffect::MakeOwned(RetEffect::CF, true)); + + // Create the "retain" selector. + RetEffect NoRet = RetEffect::MakeNoRet(); + const RetainSummary *Summ = getPersistentSummary(NoRet, IncRefMsg); + addNSObjectMethSummary(GetNullarySelector("retain", Ctx), Summ); + + // Create the "release" selector. + Summ = getPersistentSummary(NoRet, DecRefMsg); + addNSObjectMethSummary(GetNullarySelector("release", Ctx), Summ); + + // Create the "drain" selector. + Summ = getPersistentSummary(NoRet, isGCEnabled() ? DoNothing : DecRef); + addNSObjectMethSummary(GetNullarySelector("drain", Ctx), Summ); + + // Create the -dealloc summary. + Summ = getPersistentSummary(NoRet, Dealloc); + addNSObjectMethSummary(GetNullarySelector("dealloc", Ctx), Summ); + + // Create the "autorelease" selector. + Summ = getPersistentSummary(NoRet, Autorelease); + addNSObjectMethSummary(GetNullarySelector("autorelease", Ctx), Summ); + + // Specially handle NSAutoreleasePool. + addInstMethSummary("NSAutoreleasePool", "init", + getPersistentSummary(NoRet, NewAutoreleasePool)); + + // For NSWindow, allocated objects are (initially) self-owned. + // FIXME: For now we opt for false negatives with NSWindow, as these objects + // self-own themselves. However, they only do this once they are displayed. + // Thus, we need to track an NSWindow's display status. + // This is tracked in <rdar://problem/6062711>. + // See also http://llvm.org/bugs/show_bug.cgi?id=3714. + const RetainSummary *NoTrackYet = getPersistentSummary(RetEffect::MakeNoRet(), + StopTracking, + StopTracking); + + addClassMethSummary("NSWindow", "alloc", NoTrackYet); + +#if 0 + addInstMethSummary("NSWindow", NoTrackYet, "initWithContentRect", + "styleMask", "backing", "defer", NULL); + + addInstMethSummary("NSWindow", NoTrackYet, "initWithContentRect", + "styleMask", "backing", "defer", "screen", NULL); +#endif + + // For NSPanel (which subclasses NSWindow), allocated objects are not + // self-owned. + // FIXME: For now we don't track NSPanels. object for the same reason + // as for NSWindow objects. + addClassMethSummary("NSPanel", "alloc", NoTrackYet); + +#if 0 + addInstMethSummary("NSPanel", NoTrackYet, "initWithContentRect", + "styleMask", "backing", "defer", NULL); + + addInstMethSummary("NSPanel", NoTrackYet, "initWithContentRect", + "styleMask", "backing", "defer", "screen", NULL); +#endif + + // Don't track allocated autorelease pools yet, as it is okay to prematurely + // exit a method. + addClassMethSummary("NSAutoreleasePool", "alloc", NoTrackYet); + addClassMethSummary("NSAutoreleasePool", "allocWithZone", NoTrackYet, false); + + // Create summaries QCRenderer/QCView -createSnapShotImageOfType: + addInstMethSummary("QCRenderer", AllocSumm, + "createSnapshotImageOfType", NULL); + addInstMethSummary("QCView", AllocSumm, + "createSnapshotImageOfType", NULL); + + // Create summaries for CIContext, 'createCGImage' and + // 'createCGLayerWithSize'. These objects are CF objects, and are not + // automatically garbage collected. + addInstMethSummary("CIContext", CFAllocSumm, + "createCGImage", "fromRect", NULL); + addInstMethSummary("CIContext", CFAllocSumm, + "createCGImage", "fromRect", "format", "colorSpace", NULL); + addInstMethSummary("CIContext", CFAllocSumm, "createCGLayerWithSize", + "info", NULL); +} + +//===----------------------------------------------------------------------===// +// AutoreleaseBindings - State used to track objects in autorelease pools. +//===----------------------------------------------------------------------===// + +typedef llvm::ImmutableMap<SymbolRef, unsigned> ARCounts; +typedef llvm::ImmutableMap<SymbolRef, ARCounts> ARPoolContents; +typedef llvm::ImmutableList<SymbolRef> ARStack; + +static int AutoRCIndex = 0; +static int AutoRBIndex = 0; + +namespace { class AutoreleasePoolContents {}; } +namespace { class AutoreleaseStack {}; } + +namespace clang { +namespace ento { +template<> struct ProgramStateTrait<AutoreleaseStack> + : public ProgramStatePartialTrait<ARStack> { + static inline void *GDMIndex() { return &AutoRBIndex; } +}; + +template<> struct ProgramStateTrait<AutoreleasePoolContents> + : public ProgramStatePartialTrait<ARPoolContents> { + static inline void *GDMIndex() { return &AutoRCIndex; } +}; +} // end GR namespace +} // end clang namespace + +static SymbolRef GetCurrentAutoreleasePool(ProgramStateRef state) { + ARStack stack = state->get<AutoreleaseStack>(); + return stack.isEmpty() ? SymbolRef() : stack.getHead(); +} + +static ProgramStateRef +SendAutorelease(ProgramStateRef state, + ARCounts::Factory &F, + SymbolRef sym) { + SymbolRef pool = GetCurrentAutoreleasePool(state); + const ARCounts *cnts = state->get<AutoreleasePoolContents>(pool); + ARCounts newCnts(0); + + if (cnts) { + const unsigned *cnt = (*cnts).lookup(sym); + newCnts = F.add(*cnts, sym, cnt ? *cnt + 1 : 1); + } + else + newCnts = F.add(F.getEmptyMap(), sym, 1); + + return state->set<AutoreleasePoolContents>(pool, newCnts); +} + +//===----------------------------------------------------------------------===// +// Error reporting. +//===----------------------------------------------------------------------===// +namespace { + typedef llvm::DenseMap<const ExplodedNode *, const RetainSummary *> + SummaryLogTy; + + //===-------------===// + // Bug Descriptions. // + //===-------------===// + + class CFRefBug : public BugType { + protected: + CFRefBug(StringRef name) + : BugType(name, categories::MemoryCoreFoundationObjectiveC) {} + public: + + // FIXME: Eventually remove. + virtual const char *getDescription() const = 0; + + virtual bool isLeak() const { return false; } + }; + + class UseAfterRelease : public CFRefBug { + public: + UseAfterRelease() : CFRefBug("Use-after-release") {} + + const char *getDescription() const { + return "Reference-counted object is used after it is released"; + } + }; + + class BadRelease : public CFRefBug { + public: + BadRelease() : CFRefBug("Bad release") {} + + const char *getDescription() const { + return "Incorrect decrement of the reference count of an object that is " + "not owned at this point by the caller"; + } + }; + + class DeallocGC : public CFRefBug { + public: + DeallocGC() + : CFRefBug("-dealloc called while using garbage collection") {} + + const char *getDescription() const { + return "-dealloc called while using garbage collection"; + } + }; + + class DeallocNotOwned : public CFRefBug { + public: + DeallocNotOwned() + : CFRefBug("-dealloc sent to non-exclusively owned object") {} + + const char *getDescription() const { + return "-dealloc sent to object that may be referenced elsewhere"; + } + }; + + class OverAutorelease : public CFRefBug { + public: + OverAutorelease() + : CFRefBug("Object sent -autorelease too many times") {} + + const char *getDescription() const { + return "Object sent -autorelease too many times"; + } + }; + + class ReturnedNotOwnedForOwned : public CFRefBug { + public: + ReturnedNotOwnedForOwned() + : CFRefBug("Method should return an owned object") {} + + const char *getDescription() const { + return "Object with a +0 retain count returned to caller where a +1 " + "(owning) retain count is expected"; + } + }; + + class Leak : public CFRefBug { + const bool isReturn; + protected: + Leak(StringRef name, bool isRet) + : CFRefBug(name), isReturn(isRet) { + // Leaks should not be reported if they are post-dominated by a sink. + setSuppressOnSink(true); + } + public: + + const char *getDescription() const { return ""; } + + bool isLeak() const { return true; } + }; + + class LeakAtReturn : public Leak { + public: + LeakAtReturn(StringRef name) + : Leak(name, true) {} + }; + + class LeakWithinFunction : public Leak { + public: + LeakWithinFunction(StringRef name) + : Leak(name, false) {} + }; + + //===---------===// + // Bug Reports. // + //===---------===// + + class CFRefReportVisitor : public BugReporterVisitorImpl<CFRefReportVisitor> { + protected: + SymbolRef Sym; + const SummaryLogTy &SummaryLog; + bool GCEnabled; + + public: + CFRefReportVisitor(SymbolRef sym, bool gcEnabled, const SummaryLogTy &log) + : Sym(sym), SummaryLog(log), GCEnabled(gcEnabled) {} + + virtual void Profile(llvm::FoldingSetNodeID &ID) const { + static int x = 0; + ID.AddPointer(&x); + ID.AddPointer(Sym); + } + + virtual PathDiagnosticPiece *VisitNode(const ExplodedNode *N, + const ExplodedNode *PrevN, + BugReporterContext &BRC, + BugReport &BR); + + virtual PathDiagnosticPiece *getEndPath(BugReporterContext &BRC, + const ExplodedNode *N, + BugReport &BR); + }; + + class CFRefLeakReportVisitor : public CFRefReportVisitor { + public: + CFRefLeakReportVisitor(SymbolRef sym, bool GCEnabled, + const SummaryLogTy &log) + : CFRefReportVisitor(sym, GCEnabled, log) {} + + PathDiagnosticPiece *getEndPath(BugReporterContext &BRC, + const ExplodedNode *N, + BugReport &BR); + + virtual BugReporterVisitor *clone() const { + // The curiously-recurring template pattern only works for one level of + // subclassing. Rather than make a new template base for + // CFRefReportVisitor, we simply override clone() to do the right thing. + // This could be trouble someday if BugReporterVisitorImpl is ever + // used for something else besides a convenient implementation of clone(). + return new CFRefLeakReportVisitor(*this); + } + }; + + class CFRefReport : public BugReport { + void addGCModeDescription(const LangOptions &LOpts, bool GCEnabled); + + public: + CFRefReport(CFRefBug &D, const LangOptions &LOpts, bool GCEnabled, + const SummaryLogTy &Log, ExplodedNode *n, SymbolRef sym, + bool registerVisitor = true) + : BugReport(D, D.getDescription(), n) { + if (registerVisitor) + addVisitor(new CFRefReportVisitor(sym, GCEnabled, Log)); + addGCModeDescription(LOpts, GCEnabled); + } + + CFRefReport(CFRefBug &D, const LangOptions &LOpts, bool GCEnabled, + const SummaryLogTy &Log, ExplodedNode *n, SymbolRef sym, + StringRef endText) + : BugReport(D, D.getDescription(), endText, n) { + addVisitor(new CFRefReportVisitor(sym, GCEnabled, Log)); + addGCModeDescription(LOpts, GCEnabled); + } + + virtual std::pair<ranges_iterator, ranges_iterator> getRanges() { + const CFRefBug& BugTy = static_cast<CFRefBug&>(getBugType()); + if (!BugTy.isLeak()) + return BugReport::getRanges(); + else + return std::make_pair(ranges_iterator(), ranges_iterator()); + } + }; + + class CFRefLeakReport : public CFRefReport { + const MemRegion* AllocBinding; + + public: + CFRefLeakReport(CFRefBug &D, const LangOptions &LOpts, bool GCEnabled, + const SummaryLogTy &Log, ExplodedNode *n, SymbolRef sym, + CheckerContext &Ctx); + + PathDiagnosticLocation getLocation(const SourceManager &SM) const { + assert(Location.isValid()); + return Location; + } + }; +} // end anonymous namespace + +void CFRefReport::addGCModeDescription(const LangOptions &LOpts, + bool GCEnabled) { + const char *GCModeDescription = 0; + + switch (LOpts.getGC()) { + case LangOptions::GCOnly: + assert(GCEnabled); + GCModeDescription = "Code is compiled to only use garbage collection"; + break; + + case LangOptions::NonGC: + assert(!GCEnabled); + GCModeDescription = "Code is compiled to use reference counts"; + break; + + case LangOptions::HybridGC: + if (GCEnabled) { + GCModeDescription = "Code is compiled to use either garbage collection " + "(GC) or reference counts (non-GC). The bug occurs " + "with GC enabled"; + break; + } else { + GCModeDescription = "Code is compiled to use either garbage collection " + "(GC) or reference counts (non-GC). The bug occurs " + "in non-GC mode"; + break; + } + } + + assert(GCModeDescription && "invalid/unknown GC mode"); + addExtraText(GCModeDescription); +} + +// FIXME: This should be a method on SmallVector. +static inline bool contains(const SmallVectorImpl<ArgEffect>& V, + ArgEffect X) { + for (SmallVectorImpl<ArgEffect>::const_iterator I=V.begin(), E=V.end(); + I!=E; ++I) + if (*I == X) return true; + + return false; +} + +static bool isPropertyAccess(const Stmt *S, ParentMap &PM) { + unsigned maxDepth = 4; + while (S && maxDepth) { + if (const PseudoObjectExpr *PO = dyn_cast<PseudoObjectExpr>(S)) { + if (!isa<ObjCMessageExpr>(PO->getSyntacticForm())) + return true; + return false; + } + S = PM.getParent(S); + --maxDepth; + } + return false; +} + +PathDiagnosticPiece *CFRefReportVisitor::VisitNode(const ExplodedNode *N, + const ExplodedNode *PrevN, + BugReporterContext &BRC, + BugReport &BR) { + + if (!isa<StmtPoint>(N->getLocation())) + return NULL; + + // Check if the type state has changed. + ProgramStateRef PrevSt = PrevN->getState(); + ProgramStateRef CurrSt = N->getState(); + const LocationContext *LCtx = N->getLocationContext(); + + const RefVal* CurrT = CurrSt->get<RefBindings>(Sym); + if (!CurrT) return NULL; + + const RefVal &CurrV = *CurrT; + const RefVal *PrevT = PrevSt->get<RefBindings>(Sym); + + // Create a string buffer to constain all the useful things we want + // to tell the user. + std::string sbuf; + llvm::raw_string_ostream os(sbuf); + + // This is the allocation site since the previous node had no bindings + // for this symbol. + if (!PrevT) { + const Stmt *S = cast<StmtPoint>(N->getLocation()).getStmt(); + + if (isa<ObjCArrayLiteral>(S)) { + os << "NSArray literal is an object with a +0 retain count"; + } + else if (isa<ObjCDictionaryLiteral>(S)) { + os << "NSDictionary literal is an object with a +0 retain count"; + } + else { + if (const CallExpr *CE = dyn_cast<CallExpr>(S)) { + // Get the name of the callee (if it is available). + SVal X = CurrSt->getSValAsScalarOrLoc(CE->getCallee(), LCtx); + if (const FunctionDecl *FD = X.getAsFunctionDecl()) + os << "Call to function '" << *FD << '\''; + else + os << "function call"; + } + else { + assert(isa<ObjCMessageExpr>(S)); + // The message expression may have between written directly or as + // a property access. Lazily determine which case we are looking at. + os << (isPropertyAccess(S, N->getParentMap()) ? "Property" : "Method"); + } + + if (CurrV.getObjKind() == RetEffect::CF) { + os << " returns a Core Foundation object with a "; + } + else { + assert (CurrV.getObjKind() == RetEffect::ObjC); + os << " returns an Objective-C object with a "; + } + + if (CurrV.isOwned()) { + os << "+1 retain count"; + + if (GCEnabled) { + assert(CurrV.getObjKind() == RetEffect::CF); + os << ". " + "Core Foundation objects are not automatically garbage collected."; + } + } + else { + assert (CurrV.isNotOwned()); + os << "+0 retain count"; + } + } + + PathDiagnosticLocation Pos(S, BRC.getSourceManager(), + N->getLocationContext()); + return new PathDiagnosticEventPiece(Pos, os.str()); + } + + // Gather up the effects that were performed on the object at this + // program point + SmallVector<ArgEffect, 2> AEffects; + + const ExplodedNode *OrigNode = BRC.getNodeResolver().getOriginalNode(N); + if (const RetainSummary *Summ = SummaryLog.lookup(OrigNode)) { + // We only have summaries attached to nodes after evaluating CallExpr and + // ObjCMessageExprs. + const Stmt *S = cast<StmtPoint>(N->getLocation()).getStmt(); + + if (const CallExpr *CE = dyn_cast<CallExpr>(S)) { + // Iterate through the parameter expressions and see if the symbol + // was ever passed as an argument. + unsigned i = 0; + + for (CallExpr::const_arg_iterator AI=CE->arg_begin(), AE=CE->arg_end(); + AI!=AE; ++AI, ++i) { + + // Retrieve the value of the argument. Is it the symbol + // we are interested in? + if (CurrSt->getSValAsScalarOrLoc(*AI, LCtx).getAsLocSymbol() != Sym) + continue; + + // We have an argument. Get the effect! + AEffects.push_back(Summ->getArg(i)); + } + } + else if (const ObjCMessageExpr *ME = dyn_cast<ObjCMessageExpr>(S)) { + if (const Expr *receiver = ME->getInstanceReceiver()) + if (CurrSt->getSValAsScalarOrLoc(receiver, LCtx) + .getAsLocSymbol() == Sym) { + // The symbol we are tracking is the receiver. + AEffects.push_back(Summ->getReceiverEffect()); + } + } + } + + do { + // Get the previous type state. + RefVal PrevV = *PrevT; + + // Specially handle -dealloc. + if (!GCEnabled && contains(AEffects, Dealloc)) { + // Determine if the object's reference count was pushed to zero. + assert(!(PrevV == CurrV) && "The typestate *must* have changed."); + // We may not have transitioned to 'release' if we hit an error. + // This case is handled elsewhere. + if (CurrV.getKind() == RefVal::Released) { + assert(CurrV.getCombinedCounts() == 0); + os << "Object released by directly sending the '-dealloc' message"; + break; + } + } + + // Specially handle CFMakeCollectable and friends. + if (contains(AEffects, MakeCollectable)) { + // Get the name of the function. + const Stmt *S = cast<StmtPoint>(N->getLocation()).getStmt(); + SVal X = + CurrSt->getSValAsScalarOrLoc(cast<CallExpr>(S)->getCallee(), LCtx); + const FunctionDecl *FD = X.getAsFunctionDecl(); + + if (GCEnabled) { + // Determine if the object's reference count was pushed to zero. + assert(!(PrevV == CurrV) && "The typestate *must* have changed."); + + os << "In GC mode a call to '" << *FD + << "' decrements an object's retain count and registers the " + "object with the garbage collector. "; + + if (CurrV.getKind() == RefVal::Released) { + assert(CurrV.getCount() == 0); + os << "Since it now has a 0 retain count the object can be " + "automatically collected by the garbage collector."; + } + else + os << "An object must have a 0 retain count to be garbage collected. " + "After this call its retain count is +" << CurrV.getCount() + << '.'; + } + else + os << "When GC is not enabled a call to '" << *FD + << "' has no effect on its argument."; + + // Nothing more to say. + break; + } + + // Determine if the typestate has changed. + if (!(PrevV == CurrV)) + switch (CurrV.getKind()) { + case RefVal::Owned: + case RefVal::NotOwned: + + if (PrevV.getCount() == CurrV.getCount()) { + // Did an autorelease message get sent? + if (PrevV.getAutoreleaseCount() == CurrV.getAutoreleaseCount()) + return 0; + + assert(PrevV.getAutoreleaseCount() < CurrV.getAutoreleaseCount()); + os << "Object sent -autorelease message"; + break; + } + + if (PrevV.getCount() > CurrV.getCount()) + os << "Reference count decremented."; + else + os << "Reference count incremented."; + + if (unsigned Count = CurrV.getCount()) + os << " The object now has a +" << Count << " retain count."; + + if (PrevV.getKind() == RefVal::Released) { + assert(GCEnabled && CurrV.getCount() > 0); + os << " The object is not eligible for garbage collection until " + "the retain count reaches 0 again."; + } + + break; + + case RefVal::Released: + os << "Object released."; + break; + + case RefVal::ReturnedOwned: + // Autoreleases can be applied after marking a node ReturnedOwned. + if (CurrV.getAutoreleaseCount()) + return NULL; + + os << "Object returned to caller as an owning reference (single " + "retain count transferred to caller)"; + break; + + case RefVal::ReturnedNotOwned: + os << "Object returned to caller with a +0 retain count"; + break; + + default: + return NULL; + } + + // Emit any remaining diagnostics for the argument effects (if any). + for (SmallVectorImpl<ArgEffect>::iterator I=AEffects.begin(), + E=AEffects.end(); I != E; ++I) { + + // A bunch of things have alternate behavior under GC. + if (GCEnabled) + switch (*I) { + default: break; + case Autorelease: + os << "In GC mode an 'autorelease' has no effect."; + continue; + case IncRefMsg: + os << "In GC mode the 'retain' message has no effect."; + continue; + case DecRefMsg: + os << "In GC mode the 'release' message has no effect."; + continue; + } + } + } while (0); + + if (os.str().empty()) + return 0; // We have nothing to say! + + const Stmt *S = cast<StmtPoint>(N->getLocation()).getStmt(); + PathDiagnosticLocation Pos(S, BRC.getSourceManager(), + N->getLocationContext()); + PathDiagnosticPiece *P = new PathDiagnosticEventPiece(Pos, os.str()); + + // Add the range by scanning the children of the statement for any bindings + // to Sym. + for (Stmt::const_child_iterator I = S->child_begin(), E = S->child_end(); + I!=E; ++I) + if (const Expr *Exp = dyn_cast_or_null<Expr>(*I)) + if (CurrSt->getSValAsScalarOrLoc(Exp, LCtx).getAsLocSymbol() == Sym) { + P->addRange(Exp->getSourceRange()); + break; + } + + return P; +} + +// Find the first node in the current function context that referred to the +// tracked symbol and the memory location that value was stored to. Note, the +// value is only reported if the allocation occurred in the same function as +// the leak. +static std::pair<const ExplodedNode*,const MemRegion*> +GetAllocationSite(ProgramStateManager& StateMgr, const ExplodedNode *N, + SymbolRef Sym) { + const ExplodedNode *Last = N; + const MemRegion* FirstBinding = 0; + const LocationContext *LeakContext = N->getLocationContext(); + + while (N) { + ProgramStateRef St = N->getState(); + RefBindings B = St->get<RefBindings>(); + + if (!B.lookup(Sym)) + break; + + StoreManager::FindUniqueBinding FB(Sym); + StateMgr.iterBindings(St, FB); + if (FB) FirstBinding = FB.getRegion(); + + // Allocation node, is the last node in the current context in which the + // symbol was tracked. + if (N->getLocationContext() == LeakContext) + Last = N; + + N = N->pred_empty() ? NULL : *(N->pred_begin()); + } + + // If allocation happened in a function different from the leak node context, + // do not report the binding. + if (N->getLocationContext() != LeakContext) { + FirstBinding = 0; + } + + return std::make_pair(Last, FirstBinding); +} + +PathDiagnosticPiece* +CFRefReportVisitor::getEndPath(BugReporterContext &BRC, + const ExplodedNode *EndN, + BugReport &BR) { + BR.markInteresting(Sym); + return BugReporterVisitor::getDefaultEndPath(BRC, EndN, BR); +} + +PathDiagnosticPiece* +CFRefLeakReportVisitor::getEndPath(BugReporterContext &BRC, + const ExplodedNode *EndN, + BugReport &BR) { + + // Tell the BugReporterContext to report cases when the tracked symbol is + // assigned to different variables, etc. + BR.markInteresting(Sym); + + // We are reporting a leak. Walk up the graph to get to the first node where + // the symbol appeared, and also get the first VarDecl that tracked object + // is stored to. + const ExplodedNode *AllocNode = 0; + const MemRegion* FirstBinding = 0; + + llvm::tie(AllocNode, FirstBinding) = + GetAllocationSite(BRC.getStateManager(), EndN, Sym); + + SourceManager& SM = BRC.getSourceManager(); + + // Compute an actual location for the leak. Sometimes a leak doesn't + // occur at an actual statement (e.g., transition between blocks; end + // of function) so we need to walk the graph and compute a real location. + const ExplodedNode *LeakN = EndN; + PathDiagnosticLocation L = PathDiagnosticLocation::createEndOfPath(LeakN, SM); + + std::string sbuf; + llvm::raw_string_ostream os(sbuf); + + os << "Object leaked: "; + + if (FirstBinding) { + os << "object allocated and stored into '" + << FirstBinding->getString() << '\''; + } + else + os << "allocated object"; + + // Get the retain count. + const RefVal* RV = EndN->getState()->get<RefBindings>(Sym); + + if (RV->getKind() == RefVal::ErrorLeakReturned) { + // FIXME: Per comments in rdar://6320065, "create" only applies to CF + // objects. Only "copy", "alloc", "retain" and "new" transfer ownership + // to the caller for NS objects. + const Decl *D = &EndN->getCodeDecl(); + if (const ObjCMethodDecl *MD = dyn_cast<ObjCMethodDecl>(D)) { + os << " is returned from a method whose name ('" + << MD->getSelector().getAsString() + << "') does not start with 'copy', 'mutableCopy', 'alloc' or 'new'." + " This violates the naming convention rules" + " given in the Memory Management Guide for Cocoa"; + } + else { + const FunctionDecl *FD = cast<FunctionDecl>(D); + os << " is returned from a function whose name ('" + << *FD + << "') does not contain 'Copy' or 'Create'. This violates the naming" + " convention rules given in the Memory Management Guide for Core" + " Foundation"; + } + } + else if (RV->getKind() == RefVal::ErrorGCLeakReturned) { + ObjCMethodDecl &MD = cast<ObjCMethodDecl>(EndN->getCodeDecl()); + os << " and returned from method '" << MD.getSelector().getAsString() + << "' is potentially leaked when using garbage collection. Callers " + "of this method do not expect a returned object with a +1 retain " + "count since they expect the object to be managed by the garbage " + "collector"; + } + else + os << " is not referenced later in this execution path and has a retain " + "count of +" << RV->getCount(); + + return new PathDiagnosticEventPiece(L, os.str()); +} + +CFRefLeakReport::CFRefLeakReport(CFRefBug &D, const LangOptions &LOpts, + bool GCEnabled, const SummaryLogTy &Log, + ExplodedNode *n, SymbolRef sym, + CheckerContext &Ctx) +: CFRefReport(D, LOpts, GCEnabled, Log, n, sym, false) { + + // Most bug reports are cached at the location where they occurred. + // With leaks, we want to unique them by the location where they were + // allocated, and only report a single path. To do this, we need to find + // the allocation site of a piece of tracked memory, which we do via a + // call to GetAllocationSite. This will walk the ExplodedGraph backwards. + // Note that this is *not* the trimmed graph; we are guaranteed, however, + // that all ancestor nodes that represent the allocation site have the + // same SourceLocation. + const ExplodedNode *AllocNode = 0; + + const SourceManager& SMgr = Ctx.getSourceManager(); + + llvm::tie(AllocNode, AllocBinding) = // Set AllocBinding. + GetAllocationSite(Ctx.getStateManager(), getErrorNode(), sym); + + // Get the SourceLocation for the allocation site. + ProgramPoint P = AllocNode->getLocation(); + const Stmt *AllocStmt = cast<PostStmt>(P).getStmt(); + Location = PathDiagnosticLocation::createBegin(AllocStmt, SMgr, + n->getLocationContext()); + // Fill in the description of the bug. + Description.clear(); + llvm::raw_string_ostream os(Description); + os << "Potential leak "; + if (GCEnabled) + os << "(when using garbage collection) "; + os << "of an object"; + + // FIXME: AllocBinding doesn't get populated for RegionStore yet. + if (AllocBinding) + os << " stored into '" << AllocBinding->getString() << '\''; + + addVisitor(new CFRefLeakReportVisitor(sym, GCEnabled, Log)); +} + +//===----------------------------------------------------------------------===// +// Main checker logic. +//===----------------------------------------------------------------------===// + +namespace { +class RetainCountChecker + : public Checker< check::Bind, + check::DeadSymbols, + check::EndAnalysis, + check::EndPath, + check::PostStmt<BlockExpr>, + check::PostStmt<CastExpr>, + check::PostStmt<CallExpr>, + check::PostStmt<CXXConstructExpr>, + check::PostStmt<ObjCArrayLiteral>, + check::PostStmt<ObjCDictionaryLiteral>, + check::PostObjCMessage, + check::PreStmt<ReturnStmt>, + check::RegionChanges, + eval::Assume, + eval::Call > { + mutable OwningPtr<CFRefBug> useAfterRelease, releaseNotOwned; + mutable OwningPtr<CFRefBug> deallocGC, deallocNotOwned; + mutable OwningPtr<CFRefBug> overAutorelease, returnNotOwnedForOwned; + mutable OwningPtr<CFRefBug> leakWithinFunction, leakAtReturn; + mutable OwningPtr<CFRefBug> leakWithinFunctionGC, leakAtReturnGC; + + typedef llvm::DenseMap<SymbolRef, const SimpleProgramPointTag *> SymbolTagMap; + + // This map is only used to ensure proper deletion of any allocated tags. + mutable SymbolTagMap DeadSymbolTags; + + mutable OwningPtr<RetainSummaryManager> Summaries; + mutable OwningPtr<RetainSummaryManager> SummariesGC; + + mutable ARCounts::Factory ARCountFactory; + + mutable SummaryLogTy SummaryLog; + mutable bool ShouldResetSummaryLog; + +public: + RetainCountChecker() : ShouldResetSummaryLog(false) {} + + virtual ~RetainCountChecker() { + DeleteContainerSeconds(DeadSymbolTags); + } + + void checkEndAnalysis(ExplodedGraph &G, BugReporter &BR, + ExprEngine &Eng) const { + // FIXME: This is a hack to make sure the summary log gets cleared between + // analyses of different code bodies. + // + // Why is this necessary? Because a checker's lifetime is tied to a + // translation unit, but an ExplodedGraph's lifetime is just a code body. + // Once in a blue moon, a new ExplodedNode will have the same address as an + // old one with an associated summary, and the bug report visitor gets very + // confused. (To make things worse, the summary lifetime is currently also + // tied to a code body, so we get a crash instead of incorrect results.) + // + // Why is this a bad solution? Because if the lifetime of the ExplodedGraph + // changes, things will start going wrong again. Really the lifetime of this + // log needs to be tied to either the specific nodes in it or the entire + // ExplodedGraph, not to a specific part of the code being analyzed. + // + // (Also, having stateful local data means that the same checker can't be + // used from multiple threads, but a lot of checkers have incorrect + // assumptions about that anyway. So that wasn't a priority at the time of + // this fix.) + // + // This happens at the end of analysis, but bug reports are emitted /after/ + // this point. So we can't just clear the summary log now. Instead, we mark + // that the next time we access the summary log, it should be cleared. + + // If we never reset the summary log during /this/ code body analysis, + // there were no new summaries. There might still have been summaries from + // the /last/ analysis, so clear them out to make sure the bug report + // visitors don't get confused. + if (ShouldResetSummaryLog) + SummaryLog.clear(); + + ShouldResetSummaryLog = !SummaryLog.empty(); + } + + CFRefBug *getLeakWithinFunctionBug(const LangOptions &LOpts, + bool GCEnabled) const { + if (GCEnabled) { + if (!leakWithinFunctionGC) + leakWithinFunctionGC.reset(new LeakWithinFunction("Leak of object when " + "using garbage " + "collection")); + return leakWithinFunctionGC.get(); + } else { + if (!leakWithinFunction) { + if (LOpts.getGC() == LangOptions::HybridGC) { + leakWithinFunction.reset(new LeakWithinFunction("Leak of object when " + "not using garbage " + "collection (GC) in " + "dual GC/non-GC " + "code")); + } else { + leakWithinFunction.reset(new LeakWithinFunction("Leak")); + } + } + return leakWithinFunction.get(); + } + } + + CFRefBug *getLeakAtReturnBug(const LangOptions &LOpts, bool GCEnabled) const { + if (GCEnabled) { + if (!leakAtReturnGC) + leakAtReturnGC.reset(new LeakAtReturn("Leak of returned object when " + "using garbage collection")); + return leakAtReturnGC.get(); + } else { + if (!leakAtReturn) { + if (LOpts.getGC() == LangOptions::HybridGC) { + leakAtReturn.reset(new LeakAtReturn("Leak of returned object when " + "not using garbage collection " + "(GC) in dual GC/non-GC code")); + } else { + leakAtReturn.reset(new LeakAtReturn("Leak of returned object")); + } + } + return leakAtReturn.get(); + } + } + + RetainSummaryManager &getSummaryManager(ASTContext &Ctx, + bool GCEnabled) const { + // FIXME: We don't support ARC being turned on and off during one analysis. + // (nor, for that matter, do we support changing ASTContexts) + bool ARCEnabled = (bool)Ctx.getLangOpts().ObjCAutoRefCount; + if (GCEnabled) { + if (!SummariesGC) + SummariesGC.reset(new RetainSummaryManager(Ctx, true, ARCEnabled)); + else + assert(SummariesGC->isARCEnabled() == ARCEnabled); + return *SummariesGC; + } else { + if (!Summaries) + Summaries.reset(new RetainSummaryManager(Ctx, false, ARCEnabled)); + else + assert(Summaries->isARCEnabled() == ARCEnabled); + return *Summaries; + } + } + + RetainSummaryManager &getSummaryManager(CheckerContext &C) const { + return getSummaryManager(C.getASTContext(), C.isObjCGCEnabled()); + } + + void printState(raw_ostream &Out, ProgramStateRef State, + const char *NL, const char *Sep) const; + + void checkBind(SVal loc, SVal val, const Stmt *S, CheckerContext &C) const; + void checkPostStmt(const BlockExpr *BE, CheckerContext &C) const; + void checkPostStmt(const CastExpr *CE, CheckerContext &C) const; + + void checkPostStmt(const CallExpr *CE, CheckerContext &C) const; + void checkPostStmt(const CXXConstructExpr *CE, CheckerContext &C) const; + void checkPostStmt(const ObjCArrayLiteral *AL, CheckerContext &C) const; + void checkPostStmt(const ObjCDictionaryLiteral *DL, CheckerContext &C) const; + void checkPostObjCMessage(const ObjCMessage &Msg, CheckerContext &C) const; + + void checkSummary(const RetainSummary &Summ, const CallOrObjCMessage &Call, + CheckerContext &C) const; + + bool evalCall(const CallExpr *CE, CheckerContext &C) const; + + ProgramStateRef evalAssume(ProgramStateRef state, SVal Cond, + bool Assumption) const; + + ProgramStateRef + checkRegionChanges(ProgramStateRef state, + const StoreManager::InvalidatedSymbols *invalidated, + ArrayRef<const MemRegion *> ExplicitRegions, + ArrayRef<const MemRegion *> Regions, + const CallOrObjCMessage *Call) const; + + bool wantsRegionChangeUpdate(ProgramStateRef state) const { + return true; + } + + void checkPreStmt(const ReturnStmt *S, CheckerContext &C) const; + void checkReturnWithRetEffect(const ReturnStmt *S, CheckerContext &C, + ExplodedNode *Pred, RetEffect RE, RefVal X, + SymbolRef Sym, ProgramStateRef state) const; + + void checkDeadSymbols(SymbolReaper &SymReaper, CheckerContext &C) const; + void checkEndPath(CheckerContext &C) const; + + ProgramStateRef updateSymbol(ProgramStateRef state, SymbolRef sym, + RefVal V, ArgEffect E, RefVal::Kind &hasErr, + CheckerContext &C) const; + + void processNonLeakError(ProgramStateRef St, SourceRange ErrorRange, + RefVal::Kind ErrorKind, SymbolRef Sym, + CheckerContext &C) const; + + void processObjCLiterals(CheckerContext &C, const Expr *Ex) const; + + const ProgramPointTag *getDeadSymbolTag(SymbolRef sym) const; + + ProgramStateRef handleSymbolDeath(ProgramStateRef state, + SymbolRef sid, RefVal V, + SmallVectorImpl<SymbolRef> &Leaked) const; + + std::pair<ExplodedNode *, ProgramStateRef > + handleAutoreleaseCounts(ProgramStateRef state, + GenericNodeBuilderRefCount Bd, ExplodedNode *Pred, + CheckerContext &Ctx, SymbolRef Sym, RefVal V) const; + + ExplodedNode *processLeaks(ProgramStateRef state, + SmallVectorImpl<SymbolRef> &Leaked, + GenericNodeBuilderRefCount &Builder, + CheckerContext &Ctx, + ExplodedNode *Pred = 0) const; +}; +} // end anonymous namespace + +namespace { +class StopTrackingCallback : public SymbolVisitor { + ProgramStateRef state; +public: + StopTrackingCallback(ProgramStateRef st) : state(st) {} + ProgramStateRef getState() const { return state; } + + bool VisitSymbol(SymbolRef sym) { + state = state->remove<RefBindings>(sym); + return true; + } +}; +} // end anonymous namespace + +//===----------------------------------------------------------------------===// +// Handle statements that may have an effect on refcounts. +//===----------------------------------------------------------------------===// + +void RetainCountChecker::checkPostStmt(const BlockExpr *BE, + CheckerContext &C) const { + + // Scan the BlockDecRefExprs for any object the retain count checker + // may be tracking. + if (!BE->getBlockDecl()->hasCaptures()) + return; + + ProgramStateRef state = C.getState(); + const BlockDataRegion *R = + cast<BlockDataRegion>(state->getSVal(BE, + C.getLocationContext()).getAsRegion()); + + BlockDataRegion::referenced_vars_iterator I = R->referenced_vars_begin(), + E = R->referenced_vars_end(); + + if (I == E) + return; + + // FIXME: For now we invalidate the tracking of all symbols passed to blocks + // via captured variables, even though captured variables result in a copy + // and in implicit increment/decrement of a retain count. + SmallVector<const MemRegion*, 10> Regions; + const LocationContext *LC = C.getLocationContext(); + MemRegionManager &MemMgr = C.getSValBuilder().getRegionManager(); + + for ( ; I != E; ++I) { + const VarRegion *VR = *I; + if (VR->getSuperRegion() == R) { + VR = MemMgr.getVarRegion(VR->getDecl(), LC); + } + Regions.push_back(VR); + } + + state = + state->scanReachableSymbols<StopTrackingCallback>(Regions.data(), + Regions.data() + Regions.size()).getState(); + C.addTransition(state); +} + +void RetainCountChecker::checkPostStmt(const CastExpr *CE, + CheckerContext &C) const { + const ObjCBridgedCastExpr *BE = dyn_cast<ObjCBridgedCastExpr>(CE); + if (!BE) + return; + + ArgEffect AE = IncRef; + + switch (BE->getBridgeKind()) { + case clang::OBC_Bridge: + // Do nothing. + return; + case clang::OBC_BridgeRetained: + AE = IncRef; + break; + case clang::OBC_BridgeTransfer: + AE = DecRefBridgedTransfered; + break; + } + + ProgramStateRef state = C.getState(); + SymbolRef Sym = state->getSVal(CE, C.getLocationContext()).getAsLocSymbol(); + if (!Sym) + return; + const RefVal* T = state->get<RefBindings>(Sym); + if (!T) + return; + + RefVal::Kind hasErr = (RefVal::Kind) 0; + state = updateSymbol(state, Sym, *T, AE, hasErr, C); + + if (hasErr) { + // FIXME: If we get an error during a bridge cast, should we report it? + // Should we assert that there is no error? + return; + } + + C.addTransition(state); +} + +void RetainCountChecker::checkPostStmt(const CallExpr *CE, + CheckerContext &C) const { + if (C.wasInlined) + return; + + // Get the callee. + ProgramStateRef state = C.getState(); + const Expr *Callee = CE->getCallee(); + SVal L = state->getSVal(Callee, C.getLocationContext()); + + RetainSummaryManager &Summaries = getSummaryManager(C); + const RetainSummary *Summ = 0; + + // FIXME: Better support for blocks. For now we stop tracking anything + // that is passed to blocks. + // FIXME: Need to handle variables that are "captured" by the block. + if (dyn_cast_or_null<BlockDataRegion>(L.getAsRegion())) { + Summ = Summaries.getPersistentStopSummary(); + } else if (const FunctionDecl *FD = L.getAsFunctionDecl()) { + Summ = Summaries.getSummary(FD); + } else if (const CXXMemberCallExpr *me = dyn_cast<CXXMemberCallExpr>(CE)) { + if (const CXXMethodDecl *MD = me->getMethodDecl()) + Summ = Summaries.getSummary(MD); + } + + if (!Summ) + Summ = Summaries.getDefaultSummary(); + + checkSummary(*Summ, CallOrObjCMessage(CE, state, C.getLocationContext()), C); +} + +void RetainCountChecker::checkPostStmt(const CXXConstructExpr *CE, + CheckerContext &C) const { + const CXXConstructorDecl *Ctor = CE->getConstructor(); + if (!Ctor) + return; + + RetainSummaryManager &Summaries = getSummaryManager(C); + const RetainSummary *Summ = Summaries.getSummary(Ctor); + + // If we didn't get a summary, this constructor doesn't affect retain counts. + if (!Summ) + return; + + ProgramStateRef state = C.getState(); + checkSummary(*Summ, CallOrObjCMessage(CE, state, C.getLocationContext()), C); +} + +void RetainCountChecker::processObjCLiterals(CheckerContext &C, + const Expr *Ex) const { + ProgramStateRef state = C.getState(); + const ExplodedNode *pred = C.getPredecessor(); + for (Stmt::const_child_iterator it = Ex->child_begin(), et = Ex->child_end() ; + it != et ; ++it) { + const Stmt *child = *it; + SVal V = state->getSVal(child, pred->getLocationContext()); + if (SymbolRef sym = V.getAsSymbol()) + if (const RefVal* T = state->get<RefBindings>(sym)) { + RefVal::Kind hasErr = (RefVal::Kind) 0; + state = updateSymbol(state, sym, *T, MayEscape, hasErr, C); + if (hasErr) { + processNonLeakError(state, child->getSourceRange(), hasErr, sym, C); + return; + } + } + } + + // Return the object as autoreleased. + // RetEffect RE = RetEffect::MakeNotOwned(RetEffect::ObjC); + if (SymbolRef sym = + state->getSVal(Ex, pred->getLocationContext()).getAsSymbol()) { + QualType ResultTy = Ex->getType(); + state = state->set<RefBindings>(sym, RefVal::makeNotOwned(RetEffect::ObjC, + ResultTy)); + } + + C.addTransition(state); +} + +void RetainCountChecker::checkPostStmt(const ObjCArrayLiteral *AL, + CheckerContext &C) const { + // Apply the 'MayEscape' to all values. + processObjCLiterals(C, AL); +} + +void RetainCountChecker::checkPostStmt(const ObjCDictionaryLiteral *DL, + CheckerContext &C) const { + // Apply the 'MayEscape' to all keys and values. + processObjCLiterals(C, DL); +} + +void RetainCountChecker::checkPostObjCMessage(const ObjCMessage &Msg, + CheckerContext &C) const { + ProgramStateRef state = C.getState(); + + RetainSummaryManager &Summaries = getSummaryManager(C); + + const RetainSummary *Summ; + if (Msg.isInstanceMessage()) { + const LocationContext *LC = C.getLocationContext(); + Summ = Summaries.getInstanceMethodSummary(Msg, state, LC); + } else { + Summ = Summaries.getClassMethodSummary(Msg); + } + + // If we didn't get a summary, this message doesn't affect retain counts. + if (!Summ) + return; + + checkSummary(*Summ, CallOrObjCMessage(Msg, state, C.getLocationContext()), C); +} + +/// GetReturnType - Used to get the return type of a message expression or +/// function call with the intention of affixing that type to a tracked symbol. +/// While the the return type can be queried directly from RetEx, when +/// invoking class methods we augment to the return type to be that of +/// a pointer to the class (as opposed it just being id). +// FIXME: We may be able to do this with related result types instead. +// This function is probably overestimating. +static QualType GetReturnType(const Expr *RetE, ASTContext &Ctx) { + QualType RetTy = RetE->getType(); + // If RetE is not a message expression just return its type. + // If RetE is a message expression, return its types if it is something + /// more specific than id. + if (const ObjCMessageExpr *ME = dyn_cast<ObjCMessageExpr>(RetE)) + if (const ObjCObjectPointerType *PT = RetTy->getAs<ObjCObjectPointerType>()) + if (PT->isObjCQualifiedIdType() || PT->isObjCIdType() || + PT->isObjCClassType()) { + // At this point we know the return type of the message expression is + // id, id<...>, or Class. If we have an ObjCInterfaceDecl, we know this + // is a call to a class method whose type we can resolve. In such + // cases, promote the return type to XXX* (where XXX is the class). + const ObjCInterfaceDecl *D = ME->getReceiverInterface(); + return !D ? RetTy : + Ctx.getObjCObjectPointerType(Ctx.getObjCInterfaceType(D)); + } + + return RetTy; +} + +void RetainCountChecker::checkSummary(const RetainSummary &Summ, + const CallOrObjCMessage &CallOrMsg, + CheckerContext &C) const { + ProgramStateRef state = C.getState(); + + // Evaluate the effect of the arguments. + RefVal::Kind hasErr = (RefVal::Kind) 0; + SourceRange ErrorRange; + SymbolRef ErrorSym = 0; + + for (unsigned idx = 0, e = CallOrMsg.getNumArgs(); idx != e; ++idx) { + SVal V = CallOrMsg.getArgSVal(idx); + + if (SymbolRef Sym = V.getAsLocSymbol()) { + if (RefBindings::data_type *T = state->get<RefBindings>(Sym)) { + state = updateSymbol(state, Sym, *T, Summ.getArg(idx), hasErr, C); + if (hasErr) { + ErrorRange = CallOrMsg.getArgSourceRange(idx); + ErrorSym = Sym; + break; + } + } + } + } + + // Evaluate the effect on the message receiver. + bool ReceiverIsTracked = false; + if (!hasErr && CallOrMsg.isObjCMessage()) { + const LocationContext *LC = C.getLocationContext(); + SVal Receiver = CallOrMsg.getInstanceMessageReceiver(LC); + if (SymbolRef Sym = Receiver.getAsLocSymbol()) { + if (const RefVal *T = state->get<RefBindings>(Sym)) { + ReceiverIsTracked = true; + state = updateSymbol(state, Sym, *T, Summ.getReceiverEffect(), + hasErr, C); + if (hasErr) { + ErrorRange = CallOrMsg.getReceiverSourceRange(); + ErrorSym = Sym; + } + } + } + } + + // Process any errors. + if (hasErr) { + processNonLeakError(state, ErrorRange, hasErr, ErrorSym, C); + return; + } + + // Consult the summary for the return value. + RetEffect RE = Summ.getRetEffect(); + + if (RE.getKind() == RetEffect::OwnedWhenTrackedReceiver) { + if (ReceiverIsTracked) + RE = getSummaryManager(C).getObjAllocRetEffect(); + else + RE = RetEffect::MakeNoRet(); + } + + switch (RE.getKind()) { + default: + llvm_unreachable("Unhandled RetEffect."); + + case RetEffect::NoRet: + // No work necessary. + break; + + case RetEffect::OwnedAllocatedSymbol: + case RetEffect::OwnedSymbol: { + SymbolRef Sym = state->getSVal(CallOrMsg.getOriginExpr(), + C.getLocationContext()).getAsSymbol(); + if (!Sym) + break; + + // Use the result type from callOrMsg as it automatically adjusts + // for methods/functions that return references. + QualType ResultTy = CallOrMsg.getResultType(C.getASTContext()); + state = state->set<RefBindings>(Sym, RefVal::makeOwned(RE.getObjKind(), + ResultTy)); + + // FIXME: Add a flag to the checker where allocations are assumed to + // *not* fail. (The code below is out-of-date, though.) +#if 0 + if (RE.getKind() == RetEffect::OwnedAllocatedSymbol) { + bool isFeasible; + state = state.assume(loc::SymbolVal(Sym), true, isFeasible); + assert(isFeasible && "Cannot assume fresh symbol is non-null."); + } +#endif + + break; + } + + case RetEffect::GCNotOwnedSymbol: + case RetEffect::ARCNotOwnedSymbol: + case RetEffect::NotOwnedSymbol: { + const Expr *Ex = CallOrMsg.getOriginExpr(); + SymbolRef Sym = state->getSVal(Ex, C.getLocationContext()).getAsSymbol(); + if (!Sym) + break; + + // Use GetReturnType in order to give [NSFoo alloc] the type NSFoo *. + QualType ResultTy = GetReturnType(Ex, C.getASTContext()); + state = state->set<RefBindings>(Sym, RefVal::makeNotOwned(RE.getObjKind(), + ResultTy)); + break; + } + } + + // This check is actually necessary; otherwise the statement builder thinks + // we've hit a previously-found path. + // Normally addTransition takes care of this, but we want the node pointer. + ExplodedNode *NewNode; + if (state == C.getState()) { + NewNode = C.getPredecessor(); + } else { + NewNode = C.addTransition(state); + } + + // Annotate the node with summary we used. + if (NewNode) { + // FIXME: This is ugly. See checkEndAnalysis for why it's necessary. + if (ShouldResetSummaryLog) { + SummaryLog.clear(); + ShouldResetSummaryLog = false; + } + SummaryLog[NewNode] = &Summ; + } +} + + +ProgramStateRef +RetainCountChecker::updateSymbol(ProgramStateRef state, SymbolRef sym, + RefVal V, ArgEffect E, RefVal::Kind &hasErr, + CheckerContext &C) const { + // In GC mode [... release] and [... retain] do nothing. + // In ARC mode they shouldn't exist at all, but we just ignore them. + bool IgnoreRetainMsg = C.isObjCGCEnabled(); + if (!IgnoreRetainMsg) + IgnoreRetainMsg = (bool)C.getASTContext().getLangOpts().ObjCAutoRefCount; + + switch (E) { + default: break; + case IncRefMsg: E = IgnoreRetainMsg ? DoNothing : IncRef; break; + case DecRefMsg: E = IgnoreRetainMsg ? DoNothing : DecRef; break; + case MakeCollectable: E = C.isObjCGCEnabled() ? DecRef : DoNothing; break; + case NewAutoreleasePool: E = C.isObjCGCEnabled() ? DoNothing : + NewAutoreleasePool; break; + } + + // Handle all use-after-releases. + if (!C.isObjCGCEnabled() && V.getKind() == RefVal::Released) { + V = V ^ RefVal::ErrorUseAfterRelease; + hasErr = V.getKind(); + return state->set<RefBindings>(sym, V); + } + + switch (E) { + case DecRefMsg: + case IncRefMsg: + case MakeCollectable: + llvm_unreachable("DecRefMsg/IncRefMsg/MakeCollectable already converted"); + + case Dealloc: + // Any use of -dealloc in GC is *bad*. + if (C.isObjCGCEnabled()) { + V = V ^ RefVal::ErrorDeallocGC; + hasErr = V.getKind(); + break; + } + + switch (V.getKind()) { + default: + llvm_unreachable("Invalid RefVal state for an explicit dealloc."); + case RefVal::Owned: + // The object immediately transitions to the released state. + V = V ^ RefVal::Released; + V.clearCounts(); + return state->set<RefBindings>(sym, V); + case RefVal::NotOwned: + V = V ^ RefVal::ErrorDeallocNotOwned; + hasErr = V.getKind(); + break; + } + break; + + case NewAutoreleasePool: + assert(!C.isObjCGCEnabled()); + return state->add<AutoreleaseStack>(sym); + + case MayEscape: + if (V.getKind() == RefVal::Owned) { + V = V ^ RefVal::NotOwned; + break; + } + + // Fall-through. + + case DoNothing: + return state; + + case Autorelease: + if (C.isObjCGCEnabled()) + return state; + + // Update the autorelease counts. + state = SendAutorelease(state, ARCountFactory, sym); + V = V.autorelease(); + break; + + case StopTracking: + return state->remove<RefBindings>(sym); + + case IncRef: + switch (V.getKind()) { + default: + llvm_unreachable("Invalid RefVal state for a retain."); + case RefVal::Owned: + case RefVal::NotOwned: + V = V + 1; + break; + case RefVal::Released: + // Non-GC cases are handled above. + assert(C.isObjCGCEnabled()); + V = (V ^ RefVal::Owned) + 1; + break; + } + break; + + case SelfOwn: + V = V ^ RefVal::NotOwned; + // Fall-through. + case DecRef: + case DecRefBridgedTransfered: + switch (V.getKind()) { + default: + // case 'RefVal::Released' handled above. + llvm_unreachable("Invalid RefVal state for a release."); + + case RefVal::Owned: + assert(V.getCount() > 0); + if (V.getCount() == 1) + V = V ^ (E == DecRefBridgedTransfered ? + RefVal::NotOwned : RefVal::Released); + V = V - 1; + break; + + case RefVal::NotOwned: + if (V.getCount() > 0) + V = V - 1; + else { + V = V ^ RefVal::ErrorReleaseNotOwned; + hasErr = V.getKind(); + } + break; + + case RefVal::Released: + // Non-GC cases are handled above. + assert(C.isObjCGCEnabled()); + V = V ^ RefVal::ErrorUseAfterRelease; + hasErr = V.getKind(); + break; + } + break; + } + return state->set<RefBindings>(sym, V); +} + +void RetainCountChecker::processNonLeakError(ProgramStateRef St, + SourceRange ErrorRange, + RefVal::Kind ErrorKind, + SymbolRef Sym, + CheckerContext &C) const { + ExplodedNode *N = C.generateSink(St); + if (!N) + return; + + CFRefBug *BT; + switch (ErrorKind) { + default: + llvm_unreachable("Unhandled error."); + case RefVal::ErrorUseAfterRelease: + if (!useAfterRelease) + useAfterRelease.reset(new UseAfterRelease()); + BT = &*useAfterRelease; + break; + case RefVal::ErrorReleaseNotOwned: + if (!releaseNotOwned) + releaseNotOwned.reset(new BadRelease()); + BT = &*releaseNotOwned; + break; + case RefVal::ErrorDeallocGC: + if (!deallocGC) + deallocGC.reset(new DeallocGC()); + BT = &*deallocGC; + break; + case RefVal::ErrorDeallocNotOwned: + if (!deallocNotOwned) + deallocNotOwned.reset(new DeallocNotOwned()); + BT = &*deallocNotOwned; + break; + } + + assert(BT); + CFRefReport *report = new CFRefReport(*BT, C.getASTContext().getLangOpts(), + C.isObjCGCEnabled(), SummaryLog, + N, Sym); + report->addRange(ErrorRange); + C.EmitReport(report); +} + +//===----------------------------------------------------------------------===// +// Handle the return values of retain-count-related functions. +//===----------------------------------------------------------------------===// + +bool RetainCountChecker::evalCall(const CallExpr *CE, CheckerContext &C) const { + // Get the callee. We're only interested in simple C functions. + ProgramStateRef state = C.getState(); + const FunctionDecl *FD = C.getCalleeDecl(CE); + if (!FD) + return false; + + IdentifierInfo *II = FD->getIdentifier(); + if (!II) + return false; + + // For now, we're only handling the functions that return aliases of their + // arguments: CFRetain and CFMakeCollectable (and their families). + // Eventually we should add other functions we can model entirely, + // such as CFRelease, which don't invalidate their arguments or globals. + if (CE->getNumArgs() != 1) + return false; + + // Get the name of the function. + StringRef FName = II->getName(); + FName = FName.substr(FName.find_first_not_of('_')); + + // See if it's one of the specific functions we know how to eval. + bool canEval = false; + + QualType ResultTy = CE->getCallReturnType(); + if (ResultTy->isObjCIdType()) { + // Handle: id NSMakeCollectable(CFTypeRef) + canEval = II->isStr("NSMakeCollectable"); + } else if (ResultTy->isPointerType()) { + // Handle: (CF|CG)Retain + // CFMakeCollectable + // It's okay to be a little sloppy here (CGMakeCollectable doesn't exist). + if (cocoa::isRefType(ResultTy, "CF", FName) || + cocoa::isRefType(ResultTy, "CG", FName)) { + canEval = isRetain(FD, FName) || isMakeCollectable(FD, FName); + } + } + + if (!canEval) + return false; + + // Bind the return value. + const LocationContext *LCtx = C.getLocationContext(); + SVal RetVal = state->getSVal(CE->getArg(0), LCtx); + if (RetVal.isUnknown()) { + // If the receiver is unknown, conjure a return value. + SValBuilder &SVB = C.getSValBuilder(); + unsigned Count = C.getCurrentBlockCount(); + SVal RetVal = SVB.getConjuredSymbolVal(0, CE, LCtx, ResultTy, Count); + } + state = state->BindExpr(CE, LCtx, RetVal, false); + + // FIXME: This should not be necessary, but otherwise the argument seems to be + // considered alive during the next statement. + if (const MemRegion *ArgRegion = RetVal.getAsRegion()) { + // Save the refcount status of the argument. + SymbolRef Sym = RetVal.getAsLocSymbol(); + RefBindings::data_type *Binding = 0; + if (Sym) + Binding = state->get<RefBindings>(Sym); + + // Invalidate the argument region. + unsigned Count = C.getCurrentBlockCount(); + state = state->invalidateRegions(ArgRegion, CE, Count, LCtx); + + // Restore the refcount status of the argument. + if (Binding) + state = state->set<RefBindings>(Sym, *Binding); + } + + C.addTransition(state); + return true; +} + +//===----------------------------------------------------------------------===// +// Handle return statements. +//===----------------------------------------------------------------------===// + +// Return true if the current LocationContext has no caller context. +static bool inTopFrame(CheckerContext &C) { + const LocationContext *LC = C.getLocationContext(); + return LC->getParent() == 0; +} + +void RetainCountChecker::checkPreStmt(const ReturnStmt *S, + CheckerContext &C) const { + + // Only adjust the reference count if this is the top-level call frame, + // and not the result of inlining. In the future, we should do + // better checking even for inlined calls, and see if they match + // with their expected semantics (e.g., the method should return a retained + // object, etc.). + if (!inTopFrame(C)) + return; + + const Expr *RetE = S->getRetValue(); + if (!RetE) + return; + + ProgramStateRef state = C.getState(); + SymbolRef Sym = + state->getSValAsScalarOrLoc(RetE, C.getLocationContext()).getAsLocSymbol(); + if (!Sym) + return; + + // Get the reference count binding (if any). + const RefVal *T = state->get<RefBindings>(Sym); + if (!T) + return; + + // Change the reference count. + RefVal X = *T; + + switch (X.getKind()) { + case RefVal::Owned: { + unsigned cnt = X.getCount(); + assert(cnt > 0); + X.setCount(cnt - 1); + X = X ^ RefVal::ReturnedOwned; + break; + } + + case RefVal::NotOwned: { + unsigned cnt = X.getCount(); + if (cnt) { + X.setCount(cnt - 1); + X = X ^ RefVal::ReturnedOwned; + } + else { + X = X ^ RefVal::ReturnedNotOwned; + } + break; + } + + default: + return; + } + + // Update the binding. + state = state->set<RefBindings>(Sym, X); + ExplodedNode *Pred = C.addTransition(state); + + // At this point we have updated the state properly. + // Everything after this is merely checking to see if the return value has + // been over- or under-retained. + + // Did we cache out? + if (!Pred) + return; + + // Update the autorelease counts. + static SimpleProgramPointTag + AutoreleaseTag("RetainCountChecker : Autorelease"); + GenericNodeBuilderRefCount Bd(C, &AutoreleaseTag); + llvm::tie(Pred, state) = handleAutoreleaseCounts(state, Bd, Pred, C, Sym, X); + + // Did we cache out? + if (!Pred) + return; + + // Get the updated binding. + T = state->get<RefBindings>(Sym); + assert(T); + X = *T; + + // Consult the summary of the enclosing method. + RetainSummaryManager &Summaries = getSummaryManager(C); + const Decl *CD = &Pred->getCodeDecl(); + + if (const ObjCMethodDecl *MD = dyn_cast<ObjCMethodDecl>(CD)) { + // Unlike regular functions, /all/ ObjC methods are assumed to always + // follow Cocoa retain-count conventions, not just those with special + // names or attributes. + const RetainSummary *Summ = Summaries.getMethodSummary(MD); + RetEffect RE = Summ ? Summ->getRetEffect() : RetEffect::MakeNoRet(); + checkReturnWithRetEffect(S, C, Pred, RE, X, Sym, state); + } + + if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(CD)) { + if (!isa<CXXMethodDecl>(FD)) + if (const RetainSummary *Summ = Summaries.getSummary(FD)) + checkReturnWithRetEffect(S, C, Pred, Summ->getRetEffect(), X, + Sym, state); + } +} + +void RetainCountChecker::checkReturnWithRetEffect(const ReturnStmt *S, + CheckerContext &C, + ExplodedNode *Pred, + RetEffect RE, RefVal X, + SymbolRef Sym, + ProgramStateRef state) const { + // Any leaks or other errors? + if (X.isReturnedOwned() && X.getCount() == 0) { + if (RE.getKind() != RetEffect::NoRet) { + bool hasError = false; + if (C.isObjCGCEnabled() && RE.getObjKind() == RetEffect::ObjC) { + // Things are more complicated with garbage collection. If the + // returned object is suppose to be an Objective-C object, we have + // a leak (as the caller expects a GC'ed object) because no + // method should return ownership unless it returns a CF object. + hasError = true; + X = X ^ RefVal::ErrorGCLeakReturned; + } + else if (!RE.isOwned()) { + // Either we are using GC and the returned object is a CF type + // or we aren't using GC. In either case, we expect that the + // enclosing method is expected to return ownership. + hasError = true; + X = X ^ RefVal::ErrorLeakReturned; + } + + if (hasError) { + // Generate an error node. + state = state->set<RefBindings>(Sym, X); + + static SimpleProgramPointTag + ReturnOwnLeakTag("RetainCountChecker : ReturnsOwnLeak"); + ExplodedNode *N = C.addTransition(state, Pred, &ReturnOwnLeakTag); + if (N) { + const LangOptions &LOpts = C.getASTContext().getLangOpts(); + bool GCEnabled = C.isObjCGCEnabled(); + CFRefReport *report = + new CFRefLeakReport(*getLeakAtReturnBug(LOpts, GCEnabled), + LOpts, GCEnabled, SummaryLog, + N, Sym, C); + C.EmitReport(report); + } + } + } + } else if (X.isReturnedNotOwned()) { + if (RE.isOwned()) { + // Trying to return a not owned object to a caller expecting an + // owned object. + state = state->set<RefBindings>(Sym, X ^ RefVal::ErrorReturnedNotOwned); + + static SimpleProgramPointTag + ReturnNotOwnedTag("RetainCountChecker : ReturnNotOwnedForOwned"); + ExplodedNode *N = C.addTransition(state, Pred, &ReturnNotOwnedTag); + if (N) { + if (!returnNotOwnedForOwned) + returnNotOwnedForOwned.reset(new ReturnedNotOwnedForOwned()); + + CFRefReport *report = + new CFRefReport(*returnNotOwnedForOwned, + C.getASTContext().getLangOpts(), + C.isObjCGCEnabled(), SummaryLog, N, Sym); + C.EmitReport(report); + } + } + } +} + +//===----------------------------------------------------------------------===// +// Check various ways a symbol can be invalidated. +//===----------------------------------------------------------------------===// + +void RetainCountChecker::checkBind(SVal loc, SVal val, const Stmt *S, + CheckerContext &C) const { + // Are we storing to something that causes the value to "escape"? + bool escapes = true; + + // A value escapes in three possible cases (this may change): + // + // (1) we are binding to something that is not a memory region. + // (2) we are binding to a memregion that does not have stack storage + // (3) we are binding to a memregion with stack storage that the store + // does not understand. + ProgramStateRef state = C.getState(); + + if (loc::MemRegionVal *regionLoc = dyn_cast<loc::MemRegionVal>(&loc)) { + escapes = !regionLoc->getRegion()->hasStackStorage(); + + if (!escapes) { + // To test (3), generate a new state with the binding added. If it is + // the same state, then it escapes (since the store cannot represent + // the binding). + escapes = (state == (state->bindLoc(*regionLoc, val))); + } + if (!escapes) { + // Case 4: We do not currently model what happens when a symbol is + // assigned to a struct field, so be conservative here and let the symbol + // go. TODO: This could definitely be improved upon. + escapes = !isa<VarRegion>(regionLoc->getRegion()); + } + } + + // If our store can represent the binding and we aren't storing to something + // that doesn't have local storage then just return and have the simulation + // state continue as is. + if (!escapes) + return; + + // Otherwise, find all symbols referenced by 'val' that we are tracking + // and stop tracking them. + state = state->scanReachableSymbols<StopTrackingCallback>(val).getState(); + C.addTransition(state); +} + +ProgramStateRef RetainCountChecker::evalAssume(ProgramStateRef state, + SVal Cond, + bool Assumption) const { + + // FIXME: We may add to the interface of evalAssume the list of symbols + // whose assumptions have changed. For now we just iterate through the + // bindings and check if any of the tracked symbols are NULL. This isn't + // too bad since the number of symbols we will track in practice are + // probably small and evalAssume is only called at branches and a few + // other places. + RefBindings B = state->get<RefBindings>(); + + if (B.isEmpty()) + return state; + + bool changed = false; + RefBindings::Factory &RefBFactory = state->get_context<RefBindings>(); + + for (RefBindings::iterator I = B.begin(), E = B.end(); I != E; ++I) { + // Check if the symbol is null (or equal to any constant). + // If this is the case, stop tracking the symbol. + if (state->getSymVal(I.getKey())) { + changed = true; + B = RefBFactory.remove(B, I.getKey()); + } + } + + if (changed) + state = state->set<RefBindings>(B); + + return state; +} + +ProgramStateRef +RetainCountChecker::checkRegionChanges(ProgramStateRef state, + const StoreManager::InvalidatedSymbols *invalidated, + ArrayRef<const MemRegion *> ExplicitRegions, + ArrayRef<const MemRegion *> Regions, + const CallOrObjCMessage *Call) const { + if (!invalidated) + return state; + + llvm::SmallPtrSet<SymbolRef, 8> WhitelistedSymbols; + for (ArrayRef<const MemRegion *>::iterator I = ExplicitRegions.begin(), + E = ExplicitRegions.end(); I != E; ++I) { + if (const SymbolicRegion *SR = (*I)->StripCasts()->getAs<SymbolicRegion>()) + WhitelistedSymbols.insert(SR->getSymbol()); + } + + for (StoreManager::InvalidatedSymbols::const_iterator I=invalidated->begin(), + E = invalidated->end(); I!=E; ++I) { + SymbolRef sym = *I; + if (WhitelistedSymbols.count(sym)) + continue; + // Remove any existing reference-count binding. + state = state->remove<RefBindings>(sym); + } + return state; +} + +//===----------------------------------------------------------------------===// +// Handle dead symbols and end-of-path. +//===----------------------------------------------------------------------===// + +std::pair<ExplodedNode *, ProgramStateRef > +RetainCountChecker::handleAutoreleaseCounts(ProgramStateRef state, + GenericNodeBuilderRefCount Bd, + ExplodedNode *Pred, + CheckerContext &Ctx, + SymbolRef Sym, RefVal V) const { + unsigned ACnt = V.getAutoreleaseCount(); + + // No autorelease counts? Nothing to be done. + if (!ACnt) + return std::make_pair(Pred, state); + + assert(!Ctx.isObjCGCEnabled() && "Autorelease counts in GC mode?"); + unsigned Cnt = V.getCount(); + + // FIXME: Handle sending 'autorelease' to already released object. + + if (V.getKind() == RefVal::ReturnedOwned) + ++Cnt; + + if (ACnt <= Cnt) { + if (ACnt == Cnt) { + V.clearCounts(); + if (V.getKind() == RefVal::ReturnedOwned) + V = V ^ RefVal::ReturnedNotOwned; + else + V = V ^ RefVal::NotOwned; + } else { + V.setCount(Cnt - ACnt); + V.setAutoreleaseCount(0); + } + state = state->set<RefBindings>(Sym, V); + ExplodedNode *N = Bd.MakeNode(state, Pred); + if (N == 0) + state = 0; + return std::make_pair(N, state); + } + + // Woah! More autorelease counts then retain counts left. + // Emit hard error. + V = V ^ RefVal::ErrorOverAutorelease; + state = state->set<RefBindings>(Sym, V); + + if (ExplodedNode *N = Bd.MakeNode(state, Pred, true)) { + SmallString<128> sbuf; + llvm::raw_svector_ostream os(sbuf); + os << "Object over-autoreleased: object was sent -autorelease "; + if (V.getAutoreleaseCount() > 1) + os << V.getAutoreleaseCount() << " times "; + os << "but the object has a +" << V.getCount() << " retain count"; + + if (!overAutorelease) + overAutorelease.reset(new OverAutorelease()); + + const LangOptions &LOpts = Ctx.getASTContext().getLangOpts(); + CFRefReport *report = + new CFRefReport(*overAutorelease, LOpts, /* GCEnabled = */ false, + SummaryLog, N, Sym, os.str()); + Ctx.EmitReport(report); + } + + return std::make_pair((ExplodedNode *)0, (ProgramStateRef )0); +} + +ProgramStateRef +RetainCountChecker::handleSymbolDeath(ProgramStateRef state, + SymbolRef sid, RefVal V, + SmallVectorImpl<SymbolRef> &Leaked) const { + bool hasLeak = false; + if (V.isOwned()) + hasLeak = true; + else if (V.isNotOwned() || V.isReturnedOwned()) + hasLeak = (V.getCount() > 0); + + if (!hasLeak) + return state->remove<RefBindings>(sid); + + Leaked.push_back(sid); + return state->set<RefBindings>(sid, V ^ RefVal::ErrorLeak); +} + +ExplodedNode * +RetainCountChecker::processLeaks(ProgramStateRef state, + SmallVectorImpl<SymbolRef> &Leaked, + GenericNodeBuilderRefCount &Builder, + CheckerContext &Ctx, + ExplodedNode *Pred) const { + if (Leaked.empty()) + return Pred; + + // Generate an intermediate node representing the leak point. + ExplodedNode *N = Builder.MakeNode(state, Pred); + + if (N) { + for (SmallVectorImpl<SymbolRef>::iterator + I = Leaked.begin(), E = Leaked.end(); I != E; ++I) { + + const LangOptions &LOpts = Ctx.getASTContext().getLangOpts(); + bool GCEnabled = Ctx.isObjCGCEnabled(); + CFRefBug *BT = Pred ? getLeakWithinFunctionBug(LOpts, GCEnabled) + : getLeakAtReturnBug(LOpts, GCEnabled); + assert(BT && "BugType not initialized."); + + CFRefLeakReport *report = new CFRefLeakReport(*BT, LOpts, GCEnabled, + SummaryLog, N, *I, Ctx); + Ctx.EmitReport(report); + } + } + + return N; +} + +void RetainCountChecker::checkEndPath(CheckerContext &Ctx) const { + ProgramStateRef state = Ctx.getState(); + GenericNodeBuilderRefCount Bd(Ctx); + RefBindings B = state->get<RefBindings>(); + ExplodedNode *Pred = Ctx.getPredecessor(); + + for (RefBindings::iterator I = B.begin(), E = B.end(); I != E; ++I) { + llvm::tie(Pred, state) = handleAutoreleaseCounts(state, Bd, Pred, Ctx, + I->first, I->second); + if (!state) + return; + } + + // If the current LocationContext has a parent, don't check for leaks. + // We will do that later. + // FIXME: we should instead check for imblances of the retain/releases, + // and suggest annotations. + if (Ctx.getLocationContext()->getParent()) + return; + + B = state->get<RefBindings>(); + SmallVector<SymbolRef, 10> Leaked; + + for (RefBindings::iterator I = B.begin(), E = B.end(); I != E; ++I) + state = handleSymbolDeath(state, I->first, I->second, Leaked); + + processLeaks(state, Leaked, Bd, Ctx, Pred); +} + +const ProgramPointTag * +RetainCountChecker::getDeadSymbolTag(SymbolRef sym) const { + const SimpleProgramPointTag *&tag = DeadSymbolTags[sym]; + if (!tag) { + SmallString<64> buf; + llvm::raw_svector_ostream out(buf); + out << "RetainCountChecker : Dead Symbol : "; + sym->dumpToStream(out); + tag = new SimpleProgramPointTag(out.str()); + } + return tag; +} + +void RetainCountChecker::checkDeadSymbols(SymbolReaper &SymReaper, + CheckerContext &C) const { + ExplodedNode *Pred = C.getPredecessor(); + + ProgramStateRef state = C.getState(); + RefBindings B = state->get<RefBindings>(); + + // Update counts from autorelease pools + for (SymbolReaper::dead_iterator I = SymReaper.dead_begin(), + E = SymReaper.dead_end(); I != E; ++I) { + SymbolRef Sym = *I; + if (const RefVal *T = B.lookup(Sym)){ + // Use the symbol as the tag. + // FIXME: This might not be as unique as we would like. + GenericNodeBuilderRefCount Bd(C, getDeadSymbolTag(Sym)); + llvm::tie(Pred, state) = handleAutoreleaseCounts(state, Bd, Pred, C, + Sym, *T); + if (!state) + return; + } + } + + B = state->get<RefBindings>(); + SmallVector<SymbolRef, 10> Leaked; + + for (SymbolReaper::dead_iterator I = SymReaper.dead_begin(), + E = SymReaper.dead_end(); I != E; ++I) { + if (const RefVal *T = B.lookup(*I)) + state = handleSymbolDeath(state, *I, *T, Leaked); + } + + { + GenericNodeBuilderRefCount Bd(C, this); + Pred = processLeaks(state, Leaked, Bd, C, Pred); + } + + // Did we cache out? + if (!Pred) + return; + + // Now generate a new node that nukes the old bindings. + RefBindings::Factory &F = state->get_context<RefBindings>(); + + for (SymbolReaper::dead_iterator I = SymReaper.dead_begin(), + E = SymReaper.dead_end(); I != E; ++I) + B = F.remove(B, *I); + + state = state->set<RefBindings>(B); + C.addTransition(state, Pred); +} + +//===----------------------------------------------------------------------===// +// Debug printing of refcount bindings and autorelease pools. +//===----------------------------------------------------------------------===// + +static void PrintPool(raw_ostream &Out, SymbolRef Sym, + ProgramStateRef State) { + Out << ' '; + if (Sym) + Sym->dumpToStream(Out); + else + Out << "<pool>"; + Out << ":{"; + + // Get the contents of the pool. + if (const ARCounts *Cnts = State->get<AutoreleasePoolContents>(Sym)) + for (ARCounts::iterator I = Cnts->begin(), E = Cnts->end(); I != E; ++I) + Out << '(' << I.getKey() << ',' << I.getData() << ')'; + + Out << '}'; +} + +static bool UsesAutorelease(ProgramStateRef state) { + // A state uses autorelease if it allocated an autorelease pool or if it has + // objects in the caller's autorelease pool. + return !state->get<AutoreleaseStack>().isEmpty() || + state->get<AutoreleasePoolContents>(SymbolRef()); +} + +void RetainCountChecker::printState(raw_ostream &Out, ProgramStateRef State, + const char *NL, const char *Sep) const { + + RefBindings B = State->get<RefBindings>(); + + if (!B.isEmpty()) + Out << Sep << NL; + + for (RefBindings::iterator I = B.begin(), E = B.end(); I != E; ++I) { + Out << I->first << " : "; + I->second.print(Out); + Out << NL; + } + + // Print the autorelease stack. + if (UsesAutorelease(State)) { + Out << Sep << NL << "AR pool stack:"; + ARStack Stack = State->get<AutoreleaseStack>(); + + PrintPool(Out, SymbolRef(), State); // Print the caller's pool. + for (ARStack::iterator I = Stack.begin(), E = Stack.end(); I != E; ++I) + PrintPool(Out, *I, State); + + Out << NL; + } +} + +//===----------------------------------------------------------------------===// +// Checker registration. +//===----------------------------------------------------------------------===// + +void ento::registerRetainCountChecker(CheckerManager &Mgr) { + Mgr.registerChecker<RetainCountChecker>(); +} + diff --git a/clang/lib/StaticAnalyzer/Checkers/ReturnPointerRangeChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/ReturnPointerRangeChecker.cpp new file mode 100644 index 0000000..6e56593 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/ReturnPointerRangeChecker.cpp @@ -0,0 +1,91 @@ +//== ReturnPointerRangeChecker.cpp ------------------------------*- C++ -*--==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines ReturnPointerRangeChecker, which is a path-sensitive check +// which looks for an out-of-bound pointer being returned to callers. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ExprEngine.h" + +using namespace clang; +using namespace ento; + +namespace { +class ReturnPointerRangeChecker : + public Checker< check::PreStmt<ReturnStmt> > { + mutable OwningPtr<BuiltinBug> BT; +public: + void checkPreStmt(const ReturnStmt *RS, CheckerContext &C) const; +}; +} + +void ReturnPointerRangeChecker::checkPreStmt(const ReturnStmt *RS, + CheckerContext &C) const { + ProgramStateRef state = C.getState(); + + const Expr *RetE = RS->getRetValue(); + if (!RetE) + return; + + SVal V = state->getSVal(RetE, C.getLocationContext()); + const MemRegion *R = V.getAsRegion(); + + const ElementRegion *ER = dyn_cast_or_null<ElementRegion>(R); + if (!ER) + return; + + DefinedOrUnknownSVal Idx = cast<DefinedOrUnknownSVal>(ER->getIndex()); + // Zero index is always in bound, this also passes ElementRegions created for + // pointer casts. + if (Idx.isZeroConstant()) + return; + // FIXME: All of this out-of-bounds checking should eventually be refactored + // into a common place. + + DefinedOrUnknownSVal NumElements + = C.getStoreManager().getSizeInElements(state, ER->getSuperRegion(), + ER->getValueType()); + + ProgramStateRef StInBound = state->assumeInBound(Idx, NumElements, true); + ProgramStateRef StOutBound = state->assumeInBound(Idx, NumElements, false); + if (StOutBound && !StInBound) { + ExplodedNode *N = C.generateSink(StOutBound); + + if (!N) + return; + + // FIXME: This bug correspond to CWE-466. Eventually we should have bug + // types explicitly reference such exploit categories (when applicable). + if (!BT) + BT.reset(new BuiltinBug("Return of pointer value outside of expected range", + "Returned pointer value points outside the original object " + "(potential buffer overflow)")); + + // FIXME: It would be nice to eventually make this diagnostic more clear, + // e.g., by referencing the original declaration or by saying *why* this + // reference is outside the range. + + // Generate a report for this bug. + BugReport *report = + new BugReport(*BT, BT->getDescription(), N); + + report->addRange(RetE->getSourceRange()); + C.EmitReport(report); + } +} + +void ento::registerReturnPointerRangeChecker(CheckerManager &mgr) { + mgr.registerChecker<ReturnPointerRangeChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/ReturnUndefChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/ReturnUndefChecker.cpp new file mode 100644 index 0000000..7b1f0b1 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/ReturnUndefChecker.cpp @@ -0,0 +1,65 @@ +//== ReturnUndefChecker.cpp -------------------------------------*- C++ -*--==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines ReturnUndefChecker, which is a path-sensitive +// check which looks for undefined or garbage values being returned to the +// caller. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" + +using namespace clang; +using namespace ento; + +namespace { +class ReturnUndefChecker : + public Checker< check::PreStmt<ReturnStmt> > { + mutable OwningPtr<BuiltinBug> BT; +public: + void checkPreStmt(const ReturnStmt *RS, CheckerContext &C) const; +}; +} + +void ReturnUndefChecker::checkPreStmt(const ReturnStmt *RS, + CheckerContext &C) const { + + const Expr *RetE = RS->getRetValue(); + if (!RetE) + return; + + if (!C.getState()->getSVal(RetE, C.getLocationContext()).isUndef()) + return; + + ExplodedNode *N = C.generateSink(); + + if (!N) + return; + + if (!BT) + BT.reset(new BuiltinBug("Garbage return value", + "Undefined or garbage value returned to caller")); + + BugReport *report = + new BugReport(*BT, BT->getDescription(), N); + + report->addRange(RetE->getSourceRange()); + report->addVisitor(bugreporter::getTrackNullOrUndefValueVisitor(N, RetE, + report)); + + C.EmitReport(report); +} + +void ento::registerReturnUndefChecker(CheckerManager &mgr) { + mgr.registerChecker<ReturnUndefChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/StackAddrEscapeChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/StackAddrEscapeChecker.cpp new file mode 100644 index 0000000..54cf569 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/StackAddrEscapeChecker.cpp @@ -0,0 +1,230 @@ +//=== StackAddrEscapeChecker.cpp ----------------------------------*- C++ -*--// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines stack address leak checker, which checks if an invalid +// stack address is stored into a global or heap location. See CERT DCL30-C. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramState.h" +#include "clang/Basic/SourceManager.h" +#include "llvm/ADT/SmallString.h" +using namespace clang; +using namespace ento; + +namespace { +class StackAddrEscapeChecker : public Checker< check::PreStmt<ReturnStmt>, + check::EndPath > { + mutable OwningPtr<BuiltinBug> BT_stackleak; + mutable OwningPtr<BuiltinBug> BT_returnstack; + +public: + void checkPreStmt(const ReturnStmt *RS, CheckerContext &C) const; + void checkEndPath(CheckerContext &Ctx) const; +private: + void EmitStackError(CheckerContext &C, const MemRegion *R, + const Expr *RetE) const; + static SourceRange GenName(raw_ostream &os, const MemRegion *R, + SourceManager &SM); +}; +} + +SourceRange StackAddrEscapeChecker::GenName(raw_ostream &os, + const MemRegion *R, + SourceManager &SM) { + // Get the base region, stripping away fields and elements. + R = R->getBaseRegion(); + SourceRange range; + os << "Address of "; + + // Check if the region is a compound literal. + if (const CompoundLiteralRegion* CR = dyn_cast<CompoundLiteralRegion>(R)) { + const CompoundLiteralExpr *CL = CR->getLiteralExpr(); + os << "stack memory associated with a compound literal " + "declared on line " + << SM.getExpansionLineNumber(CL->getLocStart()) + << " returned to caller"; + range = CL->getSourceRange(); + } + else if (const AllocaRegion* AR = dyn_cast<AllocaRegion>(R)) { + const Expr *ARE = AR->getExpr(); + SourceLocation L = ARE->getLocStart(); + range = ARE->getSourceRange(); + os << "stack memory allocated by call to alloca() on line " + << SM.getExpansionLineNumber(L); + } + else if (const BlockDataRegion *BR = dyn_cast<BlockDataRegion>(R)) { + const BlockDecl *BD = BR->getCodeRegion()->getDecl(); + SourceLocation L = BD->getLocStart(); + range = BD->getSourceRange(); + os << "stack-allocated block declared on line " + << SM.getExpansionLineNumber(L); + } + else if (const VarRegion *VR = dyn_cast<VarRegion>(R)) { + os << "stack memory associated with local variable '" + << VR->getString() << '\''; + range = VR->getDecl()->getSourceRange(); + } + else if (const CXXTempObjectRegion *TOR = dyn_cast<CXXTempObjectRegion>(R)) { + os << "stack memory associated with temporary object of type '" + << TOR->getValueType().getAsString() << '\''; + range = TOR->getExpr()->getSourceRange(); + } + else { + llvm_unreachable("Invalid region in ReturnStackAddressChecker."); + } + + return range; +} + +void StackAddrEscapeChecker::EmitStackError(CheckerContext &C, const MemRegion *R, + const Expr *RetE) const { + ExplodedNode *N = C.generateSink(); + + if (!N) + return; + + if (!BT_returnstack) + BT_returnstack.reset( + new BuiltinBug("Return of address to stack-allocated memory")); + + // Generate a report for this bug. + SmallString<512> buf; + llvm::raw_svector_ostream os(buf); + SourceRange range = GenName(os, R, C.getSourceManager()); + os << " returned to caller"; + BugReport *report = new BugReport(*BT_returnstack, os.str(), N); + report->addRange(RetE->getSourceRange()); + if (range.isValid()) + report->addRange(range); + + C.EmitReport(report); +} + +void StackAddrEscapeChecker::checkPreStmt(const ReturnStmt *RS, + CheckerContext &C) const { + + const Expr *RetE = RS->getRetValue(); + if (!RetE) + return; + + SVal V = C.getState()->getSVal(RetE, C.getLocationContext()); + const MemRegion *R = V.getAsRegion(); + + if (!R) + return; + + const StackSpaceRegion *SS = + dyn_cast_or_null<StackSpaceRegion>(R->getMemorySpace()); + + if (!SS) + return; + + // Return stack memory in an ancestor stack frame is fine. + const StackFrameContext *SFC = SS->getStackFrame(); + if (SFC != C.getLocationContext()->getCurrentStackFrame()) + return; + + // Automatic reference counting automatically copies blocks. + if (C.getASTContext().getLangOpts().ObjCAutoRefCount && + isa<BlockDataRegion>(R)) + return; + + EmitStackError(C, R, RetE); +} + +void StackAddrEscapeChecker::checkEndPath(CheckerContext &Ctx) const { + ProgramStateRef state = Ctx.getState(); + + // Iterate over all bindings to global variables and see if it contains + // a memory region in the stack space. + class CallBack : public StoreManager::BindingsHandler { + private: + CheckerContext &Ctx; + const StackFrameContext *CurSFC; + public: + SmallVector<std::pair<const MemRegion*, const MemRegion*>, 10> V; + + CallBack(CheckerContext &CC) : + Ctx(CC), + CurSFC(CC.getLocationContext()->getCurrentStackFrame()) + {} + + bool HandleBinding(StoreManager &SMgr, Store store, + const MemRegion *region, SVal val) { + + if (!isa<GlobalsSpaceRegion>(region->getMemorySpace())) + return true; + + const MemRegion *vR = val.getAsRegion(); + if (!vR) + return true; + + // Under automated retain release, it is okay to assign a block + // directly to a global variable. + if (Ctx.getASTContext().getLangOpts().ObjCAutoRefCount && + isa<BlockDataRegion>(vR)) + return true; + + if (const StackSpaceRegion *SSR = + dyn_cast<StackSpaceRegion>(vR->getMemorySpace())) { + // If the global variable holds a location in the current stack frame, + // record the binding to emit a warning. + if (SSR->getStackFrame() == CurSFC) + V.push_back(std::make_pair(region, vR)); + } + + return true; + } + }; + + CallBack cb(Ctx); + state->getStateManager().getStoreManager().iterBindings(state->getStore(),cb); + + if (cb.V.empty()) + return; + + // Generate an error node. + ExplodedNode *N = Ctx.addTransition(state); + if (!N) + return; + + if (!BT_stackleak) + BT_stackleak.reset( + new BuiltinBug("Stack address stored into global variable", + "Stack address was saved into a global variable. " + "This is dangerous because the address will become " + "invalid after returning from the function")); + + for (unsigned i = 0, e = cb.V.size(); i != e; ++i) { + // Generate a report for this bug. + SmallString<512> buf; + llvm::raw_svector_ostream os(buf); + SourceRange range = GenName(os, cb.V[i].second, + Ctx.getSourceManager()); + os << " is still referred to by the global variable '"; + const VarRegion *VR = cast<VarRegion>(cb.V[i].first->getBaseRegion()); + os << *VR->getDecl() + << "' upon returning to the caller. This will be a dangling reference"; + BugReport *report = new BugReport(*BT_stackleak, os.str(), N); + if (range.isValid()) + report->addRange(range); + + Ctx.EmitReport(report); + } +} + +void ento::registerStackAddrEscapeChecker(CheckerManager &mgr) { + mgr.registerChecker<StackAddrEscapeChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/StreamChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/StreamChecker.cpp new file mode 100644 index 0000000..3745d4a --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/StreamChecker.cpp @@ -0,0 +1,475 @@ +//===-- StreamChecker.cpp -----------------------------------------*- C++ -*--// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines checkers that model and check stream handling functions. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramState.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramStateTrait.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/SymbolManager.h" +#include "llvm/ADT/ImmutableMap.h" + +using namespace clang; +using namespace ento; + +namespace { + +struct StreamState { + enum Kind { Opened, Closed, OpenFailed, Escaped } K; + const Stmt *S; + + StreamState(Kind k, const Stmt *s) : K(k), S(s) {} + + bool isOpened() const { return K == Opened; } + bool isClosed() const { return K == Closed; } + //bool isOpenFailed() const { return K == OpenFailed; } + //bool isEscaped() const { return K == Escaped; } + + bool operator==(const StreamState &X) const { + return K == X.K && S == X.S; + } + + static StreamState getOpened(const Stmt *s) { return StreamState(Opened, s); } + static StreamState getClosed(const Stmt *s) { return StreamState(Closed, s); } + static StreamState getOpenFailed(const Stmt *s) { + return StreamState(OpenFailed, s); + } + static StreamState getEscaped(const Stmt *s) { + return StreamState(Escaped, s); + } + + void Profile(llvm::FoldingSetNodeID &ID) const { + ID.AddInteger(K); + ID.AddPointer(S); + } +}; + +class StreamChecker : public Checker<eval::Call, + check::DeadSymbols, + check::EndPath, + check::PreStmt<ReturnStmt> > { + mutable IdentifierInfo *II_fopen, *II_tmpfile, *II_fclose, *II_fread, + *II_fwrite, + *II_fseek, *II_ftell, *II_rewind, *II_fgetpos, *II_fsetpos, + *II_clearerr, *II_feof, *II_ferror, *II_fileno; + mutable OwningPtr<BuiltinBug> BT_nullfp, BT_illegalwhence, + BT_doubleclose, BT_ResourceLeak; + +public: + StreamChecker() + : II_fopen(0), II_tmpfile(0) ,II_fclose(0), II_fread(0), II_fwrite(0), + II_fseek(0), II_ftell(0), II_rewind(0), II_fgetpos(0), II_fsetpos(0), + II_clearerr(0), II_feof(0), II_ferror(0), II_fileno(0) {} + + bool evalCall(const CallExpr *CE, CheckerContext &C) const; + void checkDeadSymbols(SymbolReaper &SymReaper, CheckerContext &C) const; + void checkEndPath(CheckerContext &Ctx) const; + void checkPreStmt(const ReturnStmt *S, CheckerContext &C) const; + +private: + void Fopen(CheckerContext &C, const CallExpr *CE) const; + void Tmpfile(CheckerContext &C, const CallExpr *CE) const; + void Fclose(CheckerContext &C, const CallExpr *CE) const; + void Fread(CheckerContext &C, const CallExpr *CE) const; + void Fwrite(CheckerContext &C, const CallExpr *CE) const; + void Fseek(CheckerContext &C, const CallExpr *CE) const; + void Ftell(CheckerContext &C, const CallExpr *CE) const; + void Rewind(CheckerContext &C, const CallExpr *CE) const; + void Fgetpos(CheckerContext &C, const CallExpr *CE) const; + void Fsetpos(CheckerContext &C, const CallExpr *CE) const; + void Clearerr(CheckerContext &C, const CallExpr *CE) const; + void Feof(CheckerContext &C, const CallExpr *CE) const; + void Ferror(CheckerContext &C, const CallExpr *CE) const; + void Fileno(CheckerContext &C, const CallExpr *CE) const; + + void OpenFileAux(CheckerContext &C, const CallExpr *CE) const; + + ProgramStateRef CheckNullStream(SVal SV, ProgramStateRef state, + CheckerContext &C) const; + ProgramStateRef CheckDoubleClose(const CallExpr *CE, ProgramStateRef state, + CheckerContext &C) const; +}; + +} // end anonymous namespace + +namespace clang { +namespace ento { + template <> + struct ProgramStateTrait<StreamState> + : public ProgramStatePartialTrait<llvm::ImmutableMap<SymbolRef, StreamState> > { + static void *GDMIndex() { static int x; return &x; } + }; +} +} + +bool StreamChecker::evalCall(const CallExpr *CE, CheckerContext &C) const { + const FunctionDecl *FD = C.getCalleeDecl(CE); + if (!FD) + return false; + + ASTContext &Ctx = C.getASTContext(); + if (!II_fopen) + II_fopen = &Ctx.Idents.get("fopen"); + if (!II_tmpfile) + II_tmpfile = &Ctx.Idents.get("tmpfile"); + if (!II_fclose) + II_fclose = &Ctx.Idents.get("fclose"); + if (!II_fread) + II_fread = &Ctx.Idents.get("fread"); + if (!II_fwrite) + II_fwrite = &Ctx.Idents.get("fwrite"); + if (!II_fseek) + II_fseek = &Ctx.Idents.get("fseek"); + if (!II_ftell) + II_ftell = &Ctx.Idents.get("ftell"); + if (!II_rewind) + II_rewind = &Ctx.Idents.get("rewind"); + if (!II_fgetpos) + II_fgetpos = &Ctx.Idents.get("fgetpos"); + if (!II_fsetpos) + II_fsetpos = &Ctx.Idents.get("fsetpos"); + if (!II_clearerr) + II_clearerr = &Ctx.Idents.get("clearerr"); + if (!II_feof) + II_feof = &Ctx.Idents.get("feof"); + if (!II_ferror) + II_ferror = &Ctx.Idents.get("ferror"); + if (!II_fileno) + II_fileno = &Ctx.Idents.get("fileno"); + + if (FD->getIdentifier() == II_fopen) { + Fopen(C, CE); + return true; + } + if (FD->getIdentifier() == II_tmpfile) { + Tmpfile(C, CE); + return true; + } + if (FD->getIdentifier() == II_fclose) { + Fclose(C, CE); + return true; + } + if (FD->getIdentifier() == II_fread) { + Fread(C, CE); + return true; + } + if (FD->getIdentifier() == II_fwrite) { + Fwrite(C, CE); + return true; + } + if (FD->getIdentifier() == II_fseek) { + Fseek(C, CE); + return true; + } + if (FD->getIdentifier() == II_ftell) { + Ftell(C, CE); + return true; + } + if (FD->getIdentifier() == II_rewind) { + Rewind(C, CE); + return true; + } + if (FD->getIdentifier() == II_fgetpos) { + Fgetpos(C, CE); + return true; + } + if (FD->getIdentifier() == II_fsetpos) { + Fsetpos(C, CE); + return true; + } + if (FD->getIdentifier() == II_clearerr) { + Clearerr(C, CE); + return true; + } + if (FD->getIdentifier() == II_feof) { + Feof(C, CE); + return true; + } + if (FD->getIdentifier() == II_ferror) { + Ferror(C, CE); + return true; + } + if (FD->getIdentifier() == II_fileno) { + Fileno(C, CE); + return true; + } + + return false; +} + +void StreamChecker::Fopen(CheckerContext &C, const CallExpr *CE) const { + OpenFileAux(C, CE); +} + +void StreamChecker::Tmpfile(CheckerContext &C, const CallExpr *CE) const { + OpenFileAux(C, CE); +} + +void StreamChecker::OpenFileAux(CheckerContext &C, const CallExpr *CE) const { + ProgramStateRef state = C.getState(); + unsigned Count = C.getCurrentBlockCount(); + SValBuilder &svalBuilder = C.getSValBuilder(); + const LocationContext *LCtx = C.getPredecessor()->getLocationContext(); + DefinedSVal RetVal = + cast<DefinedSVal>(svalBuilder.getConjuredSymbolVal(0, CE, LCtx, Count)); + state = state->BindExpr(CE, C.getLocationContext(), RetVal); + + ConstraintManager &CM = C.getConstraintManager(); + // Bifurcate the state into two: one with a valid FILE* pointer, the other + // with a NULL. + ProgramStateRef stateNotNull, stateNull; + llvm::tie(stateNotNull, stateNull) = CM.assumeDual(state, RetVal); + + if (SymbolRef Sym = RetVal.getAsSymbol()) { + // if RetVal is not NULL, set the symbol's state to Opened. + stateNotNull = + stateNotNull->set<StreamState>(Sym,StreamState::getOpened(CE)); + stateNull = + stateNull->set<StreamState>(Sym, StreamState::getOpenFailed(CE)); + + C.addTransition(stateNotNull); + C.addTransition(stateNull); + } +} + +void StreamChecker::Fclose(CheckerContext &C, const CallExpr *CE) const { + ProgramStateRef state = CheckDoubleClose(CE, C.getState(), C); + if (state) + C.addTransition(state); +} + +void StreamChecker::Fread(CheckerContext &C, const CallExpr *CE) const { + ProgramStateRef state = C.getState(); + if (!CheckNullStream(state->getSVal(CE->getArg(3), C.getLocationContext()), + state, C)) + return; +} + +void StreamChecker::Fwrite(CheckerContext &C, const CallExpr *CE) const { + ProgramStateRef state = C.getState(); + if (!CheckNullStream(state->getSVal(CE->getArg(3), C.getLocationContext()), + state, C)) + return; +} + +void StreamChecker::Fseek(CheckerContext &C, const CallExpr *CE) const { + ProgramStateRef state = C.getState(); + if (!(state = CheckNullStream(state->getSVal(CE->getArg(0), + C.getLocationContext()), state, C))) + return; + // Check the legality of the 'whence' argument of 'fseek'. + SVal Whence = state->getSVal(CE->getArg(2), C.getLocationContext()); + const nonloc::ConcreteInt *CI = dyn_cast<nonloc::ConcreteInt>(&Whence); + + if (!CI) + return; + + int64_t x = CI->getValue().getSExtValue(); + if (x >= 0 && x <= 2) + return; + + if (ExplodedNode *N = C.addTransition(state)) { + if (!BT_illegalwhence) + BT_illegalwhence.reset(new BuiltinBug("Illegal whence argument", + "The whence argument to fseek() should be " + "SEEK_SET, SEEK_END, or SEEK_CUR.")); + BugReport *R = new BugReport(*BT_illegalwhence, + BT_illegalwhence->getDescription(), N); + C.EmitReport(R); + } +} + +void StreamChecker::Ftell(CheckerContext &C, const CallExpr *CE) const { + ProgramStateRef state = C.getState(); + if (!CheckNullStream(state->getSVal(CE->getArg(0), C.getLocationContext()), + state, C)) + return; +} + +void StreamChecker::Rewind(CheckerContext &C, const CallExpr *CE) const { + ProgramStateRef state = C.getState(); + if (!CheckNullStream(state->getSVal(CE->getArg(0), C.getLocationContext()), + state, C)) + return; +} + +void StreamChecker::Fgetpos(CheckerContext &C, const CallExpr *CE) const { + ProgramStateRef state = C.getState(); + if (!CheckNullStream(state->getSVal(CE->getArg(0), C.getLocationContext()), + state, C)) + return; +} + +void StreamChecker::Fsetpos(CheckerContext &C, const CallExpr *CE) const { + ProgramStateRef state = C.getState(); + if (!CheckNullStream(state->getSVal(CE->getArg(0), C.getLocationContext()), + state, C)) + return; +} + +void StreamChecker::Clearerr(CheckerContext &C, const CallExpr *CE) const { + ProgramStateRef state = C.getState(); + if (!CheckNullStream(state->getSVal(CE->getArg(0), C.getLocationContext()), + state, C)) + return; +} + +void StreamChecker::Feof(CheckerContext &C, const CallExpr *CE) const { + ProgramStateRef state = C.getState(); + if (!CheckNullStream(state->getSVal(CE->getArg(0), C.getLocationContext()), + state, C)) + return; +} + +void StreamChecker::Ferror(CheckerContext &C, const CallExpr *CE) const { + ProgramStateRef state = C.getState(); + if (!CheckNullStream(state->getSVal(CE->getArg(0), C.getLocationContext()), + state, C)) + return; +} + +void StreamChecker::Fileno(CheckerContext &C, const CallExpr *CE) const { + ProgramStateRef state = C.getState(); + if (!CheckNullStream(state->getSVal(CE->getArg(0), C.getLocationContext()), + state, C)) + return; +} + +ProgramStateRef StreamChecker::CheckNullStream(SVal SV, ProgramStateRef state, + CheckerContext &C) const { + const DefinedSVal *DV = dyn_cast<DefinedSVal>(&SV); + if (!DV) + return 0; + + ConstraintManager &CM = C.getConstraintManager(); + ProgramStateRef stateNotNull, stateNull; + llvm::tie(stateNotNull, stateNull) = CM.assumeDual(state, *DV); + + if (!stateNotNull && stateNull) { + if (ExplodedNode *N = C.generateSink(stateNull)) { + if (!BT_nullfp) + BT_nullfp.reset(new BuiltinBug("NULL stream pointer", + "Stream pointer might be NULL.")); + BugReport *R =new BugReport(*BT_nullfp, BT_nullfp->getDescription(), N); + C.EmitReport(R); + } + return 0; + } + return stateNotNull; +} + +ProgramStateRef StreamChecker::CheckDoubleClose(const CallExpr *CE, + ProgramStateRef state, + CheckerContext &C) const { + SymbolRef Sym = + state->getSVal(CE->getArg(0), C.getLocationContext()).getAsSymbol(); + if (!Sym) + return state; + + const StreamState *SS = state->get<StreamState>(Sym); + + // If the file stream is not tracked, return. + if (!SS) + return state; + + // Check: Double close a File Descriptor could cause undefined behaviour. + // Conforming to man-pages + if (SS->isClosed()) { + ExplodedNode *N = C.generateSink(); + if (N) { + if (!BT_doubleclose) + BT_doubleclose.reset(new BuiltinBug("Double fclose", + "Try to close a file Descriptor already" + " closed. Cause undefined behaviour.")); + BugReport *R = new BugReport(*BT_doubleclose, + BT_doubleclose->getDescription(), N); + C.EmitReport(R); + } + return NULL; + } + + // Close the File Descriptor. + return state->set<StreamState>(Sym, StreamState::getClosed(CE)); +} + +void StreamChecker::checkDeadSymbols(SymbolReaper &SymReaper, + CheckerContext &C) const { + for (SymbolReaper::dead_iterator I = SymReaper.dead_begin(), + E = SymReaper.dead_end(); I != E; ++I) { + SymbolRef Sym = *I; + ProgramStateRef state = C.getState(); + const StreamState *SS = state->get<StreamState>(Sym); + if (!SS) + return; + + if (SS->isOpened()) { + ExplodedNode *N = C.generateSink(); + if (N) { + if (!BT_ResourceLeak) + BT_ResourceLeak.reset(new BuiltinBug("Resource Leak", + "Opened File never closed. Potential Resource leak.")); + BugReport *R = new BugReport(*BT_ResourceLeak, + BT_ResourceLeak->getDescription(), N); + C.EmitReport(R); + } + } + } +} + +void StreamChecker::checkEndPath(CheckerContext &Ctx) const { + ProgramStateRef state = Ctx.getState(); + typedef llvm::ImmutableMap<SymbolRef, StreamState> SymMap; + SymMap M = state->get<StreamState>(); + + for (SymMap::iterator I = M.begin(), E = M.end(); I != E; ++I) { + StreamState SS = I->second; + if (SS.isOpened()) { + ExplodedNode *N = Ctx.addTransition(state); + if (N) { + if (!BT_ResourceLeak) + BT_ResourceLeak.reset(new BuiltinBug("Resource Leak", + "Opened File never closed. Potential Resource leak.")); + BugReport *R = new BugReport(*BT_ResourceLeak, + BT_ResourceLeak->getDescription(), N); + Ctx.EmitReport(R); + } + } + } +} + +void StreamChecker::checkPreStmt(const ReturnStmt *S, CheckerContext &C) const { + const Expr *RetE = S->getRetValue(); + if (!RetE) + return; + + ProgramStateRef state = C.getState(); + SymbolRef Sym = state->getSVal(RetE, C.getLocationContext()).getAsSymbol(); + + if (!Sym) + return; + + const StreamState *SS = state->get<StreamState>(Sym); + if(!SS) + return; + + if (SS->isOpened()) + state = state->set<StreamState>(Sym, StreamState::getEscaped(S)); + + C.addTransition(state); +} + +void ento::registerStreamChecker(CheckerManager &mgr) { + mgr.registerChecker<StreamChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/TaintTesterChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/TaintTesterChecker.cpp new file mode 100644 index 0000000..1133682 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/TaintTesterChecker.cpp @@ -0,0 +1,62 @@ +//== TaintTesterChecker.cpp ----------------------------------- -*- C++ -*--=// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This checker can be used for testing how taint data is propagated. +// +//===----------------------------------------------------------------------===// +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" + +using namespace clang; +using namespace ento; + +namespace { +class TaintTesterChecker : public Checker< check::PostStmt<Expr> > { + + mutable OwningPtr<BugType> BT; + void initBugType() const; + + /// Given a pointer argument, get the symbol of the value it contains + /// (points to). + SymbolRef getPointedToSymbol(CheckerContext &C, + const Expr* Arg, + bool IssueWarning = true) const; + +public: + void checkPostStmt(const Expr *E, CheckerContext &C) const; +}; +} + +inline void TaintTesterChecker::initBugType() const { + if (!BT) + BT.reset(new BugType("Tainted data", "General")); +} + +void TaintTesterChecker::checkPostStmt(const Expr *E, + CheckerContext &C) const { + ProgramStateRef State = C.getState(); + if (!State) + return; + + if (State->isTainted(E, C.getLocationContext())) { + if (ExplodedNode *N = C.addTransition()) { + initBugType(); + BugReport *report = new BugReport(*BT, "tainted",N); + report->addRange(E->getSourceRange()); + C.EmitReport(report); + } + } +} + +void ento::registerTaintTesterChecker(CheckerManager &mgr) { + mgr.registerChecker<TaintTesterChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/UndefBranchChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/UndefBranchChecker.cpp new file mode 100644 index 0000000..a30f6d5 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/UndefBranchChecker.cpp @@ -0,0 +1,112 @@ +//=== UndefBranchChecker.cpp -----------------------------------*- C++ -*--===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines UndefBranchChecker, which checks for undefined branch +// condition. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" + +using namespace clang; +using namespace ento; + +namespace { + +class UndefBranchChecker : public Checker<check::BranchCondition> { + mutable OwningPtr<BuiltinBug> BT; + + struct FindUndefExpr { + ProgramStateRef St; + const LocationContext *LCtx; + + FindUndefExpr(ProgramStateRef S, const LocationContext *L) + : St(S), LCtx(L) {} + + const Expr *FindExpr(const Expr *Ex) { + if (!MatchesCriteria(Ex)) + return 0; + + for (Stmt::const_child_iterator I = Ex->child_begin(), + E = Ex->child_end();I!=E;++I) + if (const Expr *ExI = dyn_cast_or_null<Expr>(*I)) { + const Expr *E2 = FindExpr(ExI); + if (E2) return E2; + } + + return Ex; + } + + bool MatchesCriteria(const Expr *Ex) { + return St->getSVal(Ex, LCtx).isUndef(); + } + }; + +public: + void checkBranchCondition(const Stmt *Condition, CheckerContext &Ctx) const; +}; + +} + +void UndefBranchChecker::checkBranchCondition(const Stmt *Condition, + CheckerContext &Ctx) const { + SVal X = Ctx.getState()->getSVal(Condition, Ctx.getLocationContext()); + if (X.isUndef()) { + // Generate a sink node, which implicitly marks both outgoing branches as + // infeasible. + ExplodedNode *N = Ctx.generateSink(); + if (N) { + if (!BT) + BT.reset( + new BuiltinBug("Branch condition evaluates to a garbage value")); + + // What's going on here: we want to highlight the subexpression of the + // condition that is the most likely source of the "uninitialized + // branch condition." We do a recursive walk of the condition's + // subexpressions and roughly look for the most nested subexpression + // that binds to Undefined. We then highlight that expression's range. + + // Get the predecessor node and check if is a PostStmt with the Stmt + // being the terminator condition. We want to inspect the state + // of that node instead because it will contain main information about + // the subexpressions. + + // Note: any predecessor will do. They should have identical state, + // since all the BlockEdge did was act as an error sink since the value + // had to already be undefined. + assert (!N->pred_empty()); + const Expr *Ex = cast<Expr>(Condition); + ExplodedNode *PrevN = *N->pred_begin(); + ProgramPoint P = PrevN->getLocation(); + ProgramStateRef St = N->getState(); + + if (PostStmt *PS = dyn_cast<PostStmt>(&P)) + if (PS->getStmt() == Ex) + St = PrevN->getState(); + + FindUndefExpr FindIt(St, Ctx.getLocationContext()); + Ex = FindIt.FindExpr(Ex); + + // Emit the bug report. + BugReport *R = new BugReport(*BT, BT->getDescription(), N); + R->addVisitor(bugreporter::getTrackNullOrUndefValueVisitor(N, Ex, R)); + R->addRange(Ex->getSourceRange()); + + Ctx.EmitReport(R); + } + } +} + +void ento::registerUndefBranchChecker(CheckerManager &mgr) { + mgr.registerChecker<UndefBranchChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/UndefCapturedBlockVarChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/UndefCapturedBlockVarChecker.cpp new file mode 100644 index 0000000..d57767e --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/UndefCapturedBlockVarChecker.cpp @@ -0,0 +1,105 @@ +// UndefCapturedBlockVarChecker.cpp - Uninitialized captured vars -*- C++ -*-=// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This checker detects blocks that capture uninitialized values. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ExprEngine.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "llvm/ADT/SmallString.h" +#include "llvm/Support/raw_ostream.h" + +using namespace clang; +using namespace ento; + +namespace { +class UndefCapturedBlockVarChecker + : public Checker< check::PostStmt<BlockExpr> > { + mutable OwningPtr<BugType> BT; + +public: + void checkPostStmt(const BlockExpr *BE, CheckerContext &C) const; +}; +} // end anonymous namespace + +static const DeclRefExpr *FindBlockDeclRefExpr(const Stmt *S, + const VarDecl *VD) { + if (const DeclRefExpr *BR = dyn_cast<DeclRefExpr>(S)) + if (BR->getDecl() == VD) + return BR; + + for (Stmt::const_child_iterator I = S->child_begin(), E = S->child_end(); + I!=E; ++I) + if (const Stmt *child = *I) { + const DeclRefExpr *BR = FindBlockDeclRefExpr(child, VD); + if (BR) + return BR; + } + + return NULL; +} + +void +UndefCapturedBlockVarChecker::checkPostStmt(const BlockExpr *BE, + CheckerContext &C) const { + if (!BE->getBlockDecl()->hasCaptures()) + return; + + ProgramStateRef state = C.getState(); + const BlockDataRegion *R = + cast<BlockDataRegion>(state->getSVal(BE, + C.getLocationContext()).getAsRegion()); + + BlockDataRegion::referenced_vars_iterator I = R->referenced_vars_begin(), + E = R->referenced_vars_end(); + + for (; I != E; ++I) { + // This VarRegion is the region associated with the block; we need + // the one associated with the encompassing context. + const VarRegion *VR = *I; + const VarDecl *VD = VR->getDecl(); + + if (VD->getAttr<BlocksAttr>() || !VD->hasLocalStorage()) + continue; + + // Get the VarRegion associated with VD in the local stack frame. + const LocationContext *LC = C.getLocationContext(); + VR = C.getSValBuilder().getRegionManager().getVarRegion(VD, LC); + SVal VRVal = state->getSVal(VR); + + if (VRVal.isUndef()) + if (ExplodedNode *N = C.generateSink()) { + if (!BT) + BT.reset(new BuiltinBug("uninitialized variable captured by block")); + + // Generate a bug report. + SmallString<128> buf; + llvm::raw_svector_ostream os(buf); + + os << "Variable '" << VD->getName() + << "' is uninitialized when captured by block"; + + BugReport *R = new BugReport(*BT, os.str(), N); + if (const Expr *Ex = FindBlockDeclRefExpr(BE->getBody(), VD)) + R->addRange(Ex->getSourceRange()); + R->addVisitor(new FindLastStoreBRVisitor(VRVal, VR)); + // need location of block + C.EmitReport(R); + } + } +} + +void ento::registerUndefCapturedBlockVarChecker(CheckerManager &mgr) { + mgr.registerChecker<UndefCapturedBlockVarChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/UndefResultChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/UndefResultChecker.cpp new file mode 100644 index 0000000..c3c9ed7 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/UndefResultChecker.cpp @@ -0,0 +1,91 @@ +//=== UndefResultChecker.cpp ------------------------------------*- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This defines UndefResultChecker, a builtin check in ExprEngine that +// performs checks for undefined results of non-assignment binary operators. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ExprEngine.h" +#include "llvm/ADT/SmallString.h" + +using namespace clang; +using namespace ento; + +namespace { +class UndefResultChecker + : public Checker< check::PostStmt<BinaryOperator> > { + + mutable OwningPtr<BugType> BT; + +public: + void checkPostStmt(const BinaryOperator *B, CheckerContext &C) const; +}; +} // end anonymous namespace + +void UndefResultChecker::checkPostStmt(const BinaryOperator *B, + CheckerContext &C) const { + ProgramStateRef state = C.getState(); + const LocationContext *LCtx = C.getLocationContext(); + if (state->getSVal(B, LCtx).isUndef()) { + // Generate an error node. + ExplodedNode *N = C.generateSink(); + if (!N) + return; + + if (!BT) + BT.reset(new BuiltinBug("Result of operation is garbage or undefined")); + + SmallString<256> sbuf; + llvm::raw_svector_ostream OS(sbuf); + const Expr *Ex = NULL; + bool isLeft = true; + + if (state->getSVal(B->getLHS(), LCtx).isUndef()) { + Ex = B->getLHS()->IgnoreParenCasts(); + isLeft = true; + } + else if (state->getSVal(B->getRHS(), LCtx).isUndef()) { + Ex = B->getRHS()->IgnoreParenCasts(); + isLeft = false; + } + + if (Ex) { + OS << "The " << (isLeft ? "left" : "right") + << " operand of '" + << BinaryOperator::getOpcodeStr(B->getOpcode()) + << "' is a garbage value"; + } + else { + // Neither operand was undefined, but the result is undefined. + OS << "The result of the '" + << BinaryOperator::getOpcodeStr(B->getOpcode()) + << "' expression is undefined"; + } + BugReport *report = new BugReport(*BT, OS.str(), N); + if (Ex) { + report->addRange(Ex->getSourceRange()); + report->addVisitor(bugreporter::getTrackNullOrUndefValueVisitor(N, Ex, + report)); + } + else + report->addVisitor(bugreporter::getTrackNullOrUndefValueVisitor(N, B, + report)); + C.EmitReport(report); + } +} + +void ento::registerUndefResultChecker(CheckerManager &mgr) { + mgr.registerChecker<UndefResultChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/UndefinedArraySubscriptChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/UndefinedArraySubscriptChecker.cpp new file mode 100644 index 0000000..0297c4e --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/UndefinedArraySubscriptChecker.cpp @@ -0,0 +1,55 @@ +//===--- UndefinedArraySubscriptChecker.h ----------------------*- C++ -*--===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This defines UndefinedArraySubscriptChecker, a builtin check in ExprEngine +// that performs checks for undefined array subscripts. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" + +using namespace clang; +using namespace ento; + +namespace { +class UndefinedArraySubscriptChecker + : public Checker< check::PreStmt<ArraySubscriptExpr> > { + mutable OwningPtr<BugType> BT; + +public: + void checkPreStmt(const ArraySubscriptExpr *A, CheckerContext &C) const; +}; +} // end anonymous namespace + +void +UndefinedArraySubscriptChecker::checkPreStmt(const ArraySubscriptExpr *A, + CheckerContext &C) const { + if (C.getState()->getSVal(A->getIdx(), C.getLocationContext()).isUndef()) { + if (ExplodedNode *N = C.generateSink()) { + if (!BT) + BT.reset(new BuiltinBug("Array subscript is undefined")); + + // Generate a report for this bug. + BugReport *R = new BugReport(*BT, BT->getName(), N); + R->addRange(A->getIdx()->getSourceRange()); + R->addVisitor(bugreporter::getTrackNullOrUndefValueVisitor(N, + A->getIdx(), + R)); + C.EmitReport(R); + } + } +} + +void ento::registerUndefinedArraySubscriptChecker(CheckerManager &mgr) { + mgr.registerChecker<UndefinedArraySubscriptChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/UndefinedAssignmentChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/UndefinedAssignmentChecker.cpp new file mode 100644 index 0000000..78f7fa6 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/UndefinedAssignmentChecker.cpp @@ -0,0 +1,88 @@ +//===--- UndefinedAssignmentChecker.h ---------------------------*- C++ -*--==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This defines UndefinedAssignmentChecker, a builtin check in ExprEngine that +// checks for assigning undefined values. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" + +using namespace clang; +using namespace ento; + +namespace { +class UndefinedAssignmentChecker + : public Checker<check::Bind> { + mutable OwningPtr<BugType> BT; + +public: + void checkBind(SVal location, SVal val, const Stmt *S, + CheckerContext &C) const; +}; +} + +void UndefinedAssignmentChecker::checkBind(SVal location, SVal val, + const Stmt *StoreE, + CheckerContext &C) const { + if (!val.isUndef()) + return; + + ExplodedNode *N = C.generateSink(); + + if (!N) + return; + + const char *str = "Assigned value is garbage or undefined"; + + if (!BT) + BT.reset(new BuiltinBug(str)); + + // Generate a report for this bug. + const Expr *ex = 0; + + while (StoreE) { + if (const BinaryOperator *B = dyn_cast<BinaryOperator>(StoreE)) { + if (B->isCompoundAssignmentOp()) { + ProgramStateRef state = C.getState(); + if (state->getSVal(B->getLHS(), C.getLocationContext()).isUndef()) { + str = "The left expression of the compound assignment is an " + "uninitialized value. The computed value will also be garbage"; + ex = B->getLHS(); + break; + } + } + + ex = B->getRHS(); + break; + } + + if (const DeclStmt *DS = dyn_cast<DeclStmt>(StoreE)) { + const VarDecl *VD = dyn_cast<VarDecl>(DS->getSingleDecl()); + ex = VD->getInit(); + } + + break; + } + + BugReport *R = new BugReport(*BT, str, N); + if (ex) { + R->addRange(ex->getSourceRange()); + R->addVisitor(bugreporter::getTrackNullOrUndefValueVisitor(N, ex, R)); + } + C.EmitReport(R); +} + +void ento::registerUndefinedAssignmentChecker(CheckerManager &mgr) { + mgr.registerChecker<UndefinedAssignmentChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/UnixAPIChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/UnixAPIChecker.cpp new file mode 100644 index 0000000..60e665f --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/UnixAPIChecker.cpp @@ -0,0 +1,353 @@ +//= UnixAPIChecker.h - Checks preconditions for various Unix APIs --*- C++ -*-// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This defines UnixAPIChecker, which is an assortment of checks on calls +// to various, widely used UNIX/Posix functions. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/Basic/TargetInfo.h" +#include "llvm/ADT/Optional.h" +#include "llvm/ADT/SmallString.h" +#include "llvm/ADT/STLExtras.h" +#include "llvm/ADT/StringSwitch.h" +#include <fcntl.h> + +using namespace clang; +using namespace ento; +using llvm::Optional; + +namespace { +class UnixAPIChecker : public Checker< check::PreStmt<CallExpr> > { + mutable OwningPtr<BugType> BT_open, BT_pthreadOnce, BT_mallocZero; + mutable Optional<uint64_t> Val_O_CREAT; + +public: + void checkPreStmt(const CallExpr *CE, CheckerContext &C) const; + + void CheckOpen(CheckerContext &C, const CallExpr *CE) const; + void CheckPthreadOnce(CheckerContext &C, const CallExpr *CE) const; + void CheckCallocZero(CheckerContext &C, const CallExpr *CE) const; + void CheckMallocZero(CheckerContext &C, const CallExpr *CE) const; + void CheckReallocZero(CheckerContext &C, const CallExpr *CE) const; + void CheckAllocaZero(CheckerContext &C, const CallExpr *CE) const; + void CheckVallocZero(CheckerContext &C, const CallExpr *CE) const; + + typedef void (UnixAPIChecker::*SubChecker)(CheckerContext &, + const CallExpr *) const; +private: + bool ReportZeroByteAllocation(CheckerContext &C, + ProgramStateRef falseState, + const Expr *arg, + const char *fn_name) const; + void BasicAllocationCheck(CheckerContext &C, + const CallExpr *CE, + const unsigned numArgs, + const unsigned sizeArg, + const char *fn) const; +}; +} //end anonymous namespace + +//===----------------------------------------------------------------------===// +// Utility functions. +//===----------------------------------------------------------------------===// + +static inline void LazyInitialize(OwningPtr<BugType> &BT, + const char *name) { + if (BT) + return; + BT.reset(new BugType(name, categories::UnixAPI)); +} + +//===----------------------------------------------------------------------===// +// "open" (man 2 open) +//===----------------------------------------------------------------------===// + +void UnixAPIChecker::CheckOpen(CheckerContext &C, const CallExpr *CE) const { + // The definition of O_CREAT is platform specific. We need a better way + // of querying this information from the checking environment. + if (!Val_O_CREAT.hasValue()) { + if (C.getASTContext().getTargetInfo().getTriple().getVendor() + == llvm::Triple::Apple) + Val_O_CREAT = 0x0200; + else { + // FIXME: We need a more general way of getting the O_CREAT value. + // We could possibly grovel through the preprocessor state, but + // that would require passing the Preprocessor object to the ExprEngine. + return; + } + } + + // Look at the 'oflags' argument for the O_CREAT flag. + ProgramStateRef state = C.getState(); + + if (CE->getNumArgs() < 2) { + // The frontend should issue a warning for this case, so this is a sanity + // check. + return; + } + + // Now check if oflags has O_CREAT set. + const Expr *oflagsEx = CE->getArg(1); + const SVal V = state->getSVal(oflagsEx, C.getLocationContext()); + if (!isa<NonLoc>(V)) { + // The case where 'V' can be a location can only be due to a bad header, + // so in this case bail out. + return; + } + NonLoc oflags = cast<NonLoc>(V); + NonLoc ocreateFlag = + cast<NonLoc>(C.getSValBuilder().makeIntVal(Val_O_CREAT.getValue(), + oflagsEx->getType())); + SVal maskedFlagsUC = C.getSValBuilder().evalBinOpNN(state, BO_And, + oflags, ocreateFlag, + oflagsEx->getType()); + if (maskedFlagsUC.isUnknownOrUndef()) + return; + DefinedSVal maskedFlags = cast<DefinedSVal>(maskedFlagsUC); + + // Check if maskedFlags is non-zero. + ProgramStateRef trueState, falseState; + llvm::tie(trueState, falseState) = state->assume(maskedFlags); + + // Only emit an error if the value of 'maskedFlags' is properly + // constrained; + if (!(trueState && !falseState)) + return; + + if (CE->getNumArgs() < 3) { + ExplodedNode *N = C.generateSink(trueState); + if (!N) + return; + + LazyInitialize(BT_open, "Improper use of 'open'"); + + BugReport *report = + new BugReport(*BT_open, + "Call to 'open' requires a third argument when " + "the 'O_CREAT' flag is set", N); + report->addRange(oflagsEx->getSourceRange()); + C.EmitReport(report); + } +} + +//===----------------------------------------------------------------------===// +// pthread_once +//===----------------------------------------------------------------------===// + +void UnixAPIChecker::CheckPthreadOnce(CheckerContext &C, + const CallExpr *CE) const { + + // This is similar to 'CheckDispatchOnce' in the MacOSXAPIChecker. + // They can possibly be refactored. + + if (CE->getNumArgs() < 1) + return; + + // Check if the first argument is stack allocated. If so, issue a warning + // because that's likely to be bad news. + ProgramStateRef state = C.getState(); + const MemRegion *R = + state->getSVal(CE->getArg(0), C.getLocationContext()).getAsRegion(); + if (!R || !isa<StackSpaceRegion>(R->getMemorySpace())) + return; + + ExplodedNode *N = C.generateSink(state); + if (!N) + return; + + SmallString<256> S; + llvm::raw_svector_ostream os(S); + os << "Call to 'pthread_once' uses"; + if (const VarRegion *VR = dyn_cast<VarRegion>(R)) + os << " the local variable '" << VR->getDecl()->getName() << '\''; + else + os << " stack allocated memory"; + os << " for the \"control\" value. Using such transient memory for " + "the control value is potentially dangerous."; + if (isa<VarRegion>(R) && isa<StackLocalsSpaceRegion>(R->getMemorySpace())) + os << " Perhaps you intended to declare the variable as 'static'?"; + + LazyInitialize(BT_pthreadOnce, "Improper use of 'pthread_once'"); + + BugReport *report = new BugReport(*BT_pthreadOnce, os.str(), N); + report->addRange(CE->getArg(0)->getSourceRange()); + C.EmitReport(report); +} + +//===----------------------------------------------------------------------===// +// "calloc", "malloc", "realloc", "alloca" and "valloc" with allocation size 0 +//===----------------------------------------------------------------------===// +// FIXME: Eventually these should be rolled into the MallocChecker, but right now +// they're more basic and valuable for widespread use. + +// Returns true if we try to do a zero byte allocation, false otherwise. +// Fills in trueState and falseState. +static bool IsZeroByteAllocation(ProgramStateRef state, + const SVal argVal, + ProgramStateRef *trueState, + ProgramStateRef *falseState) { + llvm::tie(*trueState, *falseState) = + state->assume(cast<DefinedSVal>(argVal)); + + return (*falseState && !*trueState); +} + +// Generates an error report, indicating that the function whose name is given +// will perform a zero byte allocation. +// Returns false if an error occured, true otherwise. +bool UnixAPIChecker::ReportZeroByteAllocation(CheckerContext &C, + ProgramStateRef falseState, + const Expr *arg, + const char *fn_name) const { + ExplodedNode *N = C.generateSink(falseState); + if (!N) + return false; + + LazyInitialize(BT_mallocZero, + "Undefined allocation of 0 bytes (CERT MEM04-C; CWE-131)"); + + SmallString<256> S; + llvm::raw_svector_ostream os(S); + os << "Call to '" << fn_name << "' has an allocation size of 0 bytes"; + BugReport *report = new BugReport(*BT_mallocZero, os.str(), N); + + report->addRange(arg->getSourceRange()); + report->addVisitor(bugreporter::getTrackNullOrUndefValueVisitor(N, arg, + report)); + C.EmitReport(report); + + return true; +} + +// Does a basic check for 0-sized allocations suitable for most of the below +// functions (modulo "calloc") +void UnixAPIChecker::BasicAllocationCheck(CheckerContext &C, + const CallExpr *CE, + const unsigned numArgs, + const unsigned sizeArg, + const char *fn) const { + // Sanity check for the correct number of arguments + if (CE->getNumArgs() != numArgs) + return; + + // Check if the allocation size is 0. + ProgramStateRef state = C.getState(); + ProgramStateRef trueState = NULL, falseState = NULL; + const Expr *arg = CE->getArg(sizeArg); + SVal argVal = state->getSVal(arg, C.getLocationContext()); + + if (argVal.isUnknownOrUndef()) + return; + + // Is the value perfectly constrained to zero? + if (IsZeroByteAllocation(state, argVal, &trueState, &falseState)) { + (void) ReportZeroByteAllocation(C, falseState, arg, fn); + return; + } + // Assume the the value is non-zero going forward. + assert(trueState); + if (trueState != state) + C.addTransition(trueState); +} + +void UnixAPIChecker::CheckCallocZero(CheckerContext &C, + const CallExpr *CE) const { + unsigned int nArgs = CE->getNumArgs(); + if (nArgs != 2) + return; + + ProgramStateRef state = C.getState(); + ProgramStateRef trueState = NULL, falseState = NULL; + + unsigned int i; + for (i = 0; i < nArgs; i++) { + const Expr *arg = CE->getArg(i); + SVal argVal = state->getSVal(arg, C.getLocationContext()); + if (argVal.isUnknownOrUndef()) { + if (i == 0) + continue; + else + return; + } + + if (IsZeroByteAllocation(state, argVal, &trueState, &falseState)) { + if (ReportZeroByteAllocation(C, falseState, arg, "calloc")) + return; + else if (i == 0) + continue; + else + return; + } + } + + // Assume the the value is non-zero going forward. + assert(trueState); + if (trueState != state) + C.addTransition(trueState); +} + +void UnixAPIChecker::CheckMallocZero(CheckerContext &C, + const CallExpr *CE) const { + BasicAllocationCheck(C, CE, 1, 0, "malloc"); +} + +void UnixAPIChecker::CheckReallocZero(CheckerContext &C, + const CallExpr *CE) const { + BasicAllocationCheck(C, CE, 2, 1, "realloc"); +} + +void UnixAPIChecker::CheckAllocaZero(CheckerContext &C, + const CallExpr *CE) const { + BasicAllocationCheck(C, CE, 1, 0, "alloca"); +} + +void UnixAPIChecker::CheckVallocZero(CheckerContext &C, + const CallExpr *CE) const { + BasicAllocationCheck(C, CE, 1, 0, "valloc"); +} + + +//===----------------------------------------------------------------------===// +// Central dispatch function. +//===----------------------------------------------------------------------===// + +void UnixAPIChecker::checkPreStmt(const CallExpr *CE, + CheckerContext &C) const { + StringRef FName = C.getCalleeName(CE); + if (FName.empty()) + return; + + SubChecker SC = + llvm::StringSwitch<SubChecker>(FName) + .Case("open", &UnixAPIChecker::CheckOpen) + .Case("pthread_once", &UnixAPIChecker::CheckPthreadOnce) + .Case("calloc", &UnixAPIChecker::CheckCallocZero) + .Case("malloc", &UnixAPIChecker::CheckMallocZero) + .Case("realloc", &UnixAPIChecker::CheckReallocZero) + .Cases("alloca", "__builtin_alloca", &UnixAPIChecker::CheckAllocaZero) + .Case("valloc", &UnixAPIChecker::CheckVallocZero) + .Default(NULL); + + if (SC) + (this->*SC)(C, CE); +} + +//===----------------------------------------------------------------------===// +// Registration. +//===----------------------------------------------------------------------===// + +void ento::registerUnixAPIChecker(CheckerManager &mgr) { + mgr.registerChecker<UnixAPIChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/UnreachableCodeChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/UnreachableCodeChecker.cpp new file mode 100644 index 0000000..5a13ed0 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/UnreachableCodeChecker.cpp @@ -0,0 +1,247 @@ +//==- UnreachableCodeChecker.cpp - Generalized dead code checker -*- C++ -*-==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// This file implements a generalized unreachable code checker using a +// path-sensitive analysis. We mark any path visited, and then walk the CFG as a +// post-analysis to determine what was never visited. +// +// A similar flow-sensitive only check exists in Analysis/ReachableCode.cpp +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/ExplodedGraph.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/SVals.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerHelpers.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugReporter.h" +#include "clang/AST/ParentMap.h" +#include "clang/Basic/Builtins.h" +#include "clang/Basic/SourceManager.h" +#include "llvm/ADT/SmallSet.h" + +// The number of CFGBlock pointers we want to reserve memory for. This is used +// once for each function we analyze. +#define DEFAULT_CFGBLOCKS 256 + +using namespace clang; +using namespace ento; + +namespace { +class UnreachableCodeChecker : public Checker<check::EndAnalysis> { +public: + void checkEndAnalysis(ExplodedGraph &G, BugReporter &B, + ExprEngine &Eng) const; +private: + typedef llvm::SmallSet<unsigned, DEFAULT_CFGBLOCKS> CFGBlocksSet; + + static inline const Stmt *getUnreachableStmt(const CFGBlock *CB); + static void FindUnreachableEntryPoints(const CFGBlock *CB, + CFGBlocksSet &reachable, + CFGBlocksSet &visited); + static bool isInvalidPath(const CFGBlock *CB, const ParentMap &PM); + static inline bool isEmptyCFGBlock(const CFGBlock *CB); +}; +} + +void UnreachableCodeChecker::checkEndAnalysis(ExplodedGraph &G, + BugReporter &B, + ExprEngine &Eng) const { + CFGBlocksSet reachable, visited; + + if (Eng.hasWorkRemaining()) + return; + + const Decl *D = 0; + CFG *C = 0; + ParentMap *PM = 0; + const LocationContext *LC = 0; + // Iterate over ExplodedGraph + for (ExplodedGraph::node_iterator I = G.nodes_begin(), E = G.nodes_end(); + I != E; ++I) { + const ProgramPoint &P = I->getLocation(); + LC = P.getLocationContext(); + + if (!D) + D = LC->getAnalysisDeclContext()->getDecl(); + // Save the CFG if we don't have it already + if (!C) + C = LC->getAnalysisDeclContext()->getUnoptimizedCFG(); + if (!PM) + PM = &LC->getParentMap(); + + if (const BlockEntrance *BE = dyn_cast<BlockEntrance>(&P)) { + const CFGBlock *CB = BE->getBlock(); + reachable.insert(CB->getBlockID()); + } + } + + // Bail out if we didn't get the CFG or the ParentMap. + if (!D || !C || !PM) + return; + + // Don't do anything for template instantiations. Proving that code + // in a template instantiation is unreachable means proving that it is + // unreachable in all instantiations. + if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(D)) + if (FD->isTemplateInstantiation()) + return; + + // Find CFGBlocks that were not covered by any node + for (CFG::const_iterator I = C->begin(), E = C->end(); I != E; ++I) { + const CFGBlock *CB = *I; + // Check if the block is unreachable + if (reachable.count(CB->getBlockID())) + continue; + + // Check if the block is empty (an artificial block) + if (isEmptyCFGBlock(CB)) + continue; + + // Find the entry points for this block + if (!visited.count(CB->getBlockID())) + FindUnreachableEntryPoints(CB, reachable, visited); + + // This block may have been pruned; check if we still want to report it + if (reachable.count(CB->getBlockID())) + continue; + + // Check for false positives + if (CB->size() > 0 && isInvalidPath(CB, *PM)) + continue; + + // It is good practice to always have a "default" label in a "switch", even + // if we should never get there. It can be used to detect errors, for + // instance. Unreachable code directly under a "default" label is therefore + // likely to be a false positive. + if (const Stmt *label = CB->getLabel()) + if (label->getStmtClass() == Stmt::DefaultStmtClass) + continue; + + // Special case for __builtin_unreachable. + // FIXME: This should be extended to include other unreachable markers, + // such as llvm_unreachable. + if (!CB->empty()) { + bool foundUnreachable = false; + for (CFGBlock::const_iterator ci = CB->begin(), ce = CB->end(); + ci != ce; ++ci) { + if (const CFGStmt *S = (*ci).getAs<CFGStmt>()) + if (const CallExpr *CE = dyn_cast<CallExpr>(S->getStmt())) { + if (CE->isBuiltinCall() == Builtin::BI__builtin_unreachable) { + foundUnreachable = true; + break; + } + } + } + if (foundUnreachable) + continue; + } + + // We found a block that wasn't covered - find the statement to report + SourceRange SR; + PathDiagnosticLocation DL; + SourceLocation SL; + if (const Stmt *S = getUnreachableStmt(CB)) { + SR = S->getSourceRange(); + DL = PathDiagnosticLocation::createBegin(S, B.getSourceManager(), LC); + SL = DL.asLocation(); + if (SR.isInvalid() || !SL.isValid()) + continue; + } + else + continue; + + // Check if the SourceLocation is in a system header + const SourceManager &SM = B.getSourceManager(); + if (SM.isInSystemHeader(SL) || SM.isInExternCSystemHeader(SL)) + continue; + + B.EmitBasicReport(D, "Unreachable code", "Dead code", + "This statement is never executed", DL, SR); + } +} + +// Recursively finds the entry point(s) for this dead CFGBlock. +void UnreachableCodeChecker::FindUnreachableEntryPoints(const CFGBlock *CB, + CFGBlocksSet &reachable, + CFGBlocksSet &visited) { + visited.insert(CB->getBlockID()); + + for (CFGBlock::const_pred_iterator I = CB->pred_begin(), E = CB->pred_end(); + I != E; ++I) { + if (!reachable.count((*I)->getBlockID())) { + // If we find an unreachable predecessor, mark this block as reachable so + // we don't report this block + reachable.insert(CB->getBlockID()); + if (!visited.count((*I)->getBlockID())) + // If we haven't previously visited the unreachable predecessor, recurse + FindUnreachableEntryPoints(*I, reachable, visited); + } + } +} + +// Find the Stmt* in a CFGBlock for reporting a warning +const Stmt *UnreachableCodeChecker::getUnreachableStmt(const CFGBlock *CB) { + for (CFGBlock::const_iterator I = CB->begin(), E = CB->end(); I != E; ++I) { + if (const CFGStmt *S = I->getAs<CFGStmt>()) + return S->getStmt(); + } + if (const Stmt *S = CB->getTerminator()) + return S; + else + return 0; +} + +// Determines if the path to this CFGBlock contained an element that infers this +// block is a false positive. We assume that FindUnreachableEntryPoints has +// already marked only the entry points to any dead code, so we need only to +// find the condition that led to this block (the predecessor of this block.) +// There will never be more than one predecessor. +bool UnreachableCodeChecker::isInvalidPath(const CFGBlock *CB, + const ParentMap &PM) { + // We only expect a predecessor size of 0 or 1. If it is >1, then an external + // condition has broken our assumption (for example, a sink being placed by + // another check). In these cases, we choose not to report. + if (CB->pred_size() > 1) + return true; + + // If there are no predecessors, then this block is trivially unreachable + if (CB->pred_size() == 0) + return false; + + const CFGBlock *pred = *CB->pred_begin(); + + // Get the predecessor block's terminator conditon + const Stmt *cond = pred->getTerminatorCondition(); + + //assert(cond && "CFGBlock's predecessor has a terminator condition"); + // The previous assertion is invalid in some cases (eg do/while). Leaving + // reporting of these situations on at the moment to help triage these cases. + if (!cond) + return false; + + // Run each of the checks on the conditions + if (containsMacro(cond) || containsEnum(cond) + || containsStaticLocal(cond) || containsBuiltinOffsetOf(cond) + || containsStmt<UnaryExprOrTypeTraitExpr>(cond)) + return true; + + return false; +} + +// Returns true if the given CFGBlock is empty +bool UnreachableCodeChecker::isEmptyCFGBlock(const CFGBlock *CB) { + return CB->getLabel() == 0 // No labels + && CB->size() == 0 // No statements + && CB->getTerminator() == 0; // No terminator +} + +void ento::registerUnreachableCodeChecker(CheckerManager &mgr) { + mgr.registerChecker<UnreachableCodeChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/VLASizeChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/VLASizeChecker.cpp new file mode 100644 index 0000000..38c9cc1 --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/VLASizeChecker.cpp @@ -0,0 +1,162 @@ +//=== VLASizeChecker.cpp - Undefined dereference checker --------*- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This defines VLASizeChecker, a builtin check in ExprEngine that +// performs checks for declaration of VLA of undefined or zero size. +// In addition, VLASizeChecker is responsible for defining the extent +// of the MemRegion that represents a VLA. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/CheckerManager.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" +#include "clang/AST/CharUnits.h" +#include "llvm/ADT/SmallString.h" +#include "llvm/ADT/STLExtras.h" + +using namespace clang; +using namespace ento; + +namespace { +class VLASizeChecker : public Checker< check::PreStmt<DeclStmt> > { + mutable OwningPtr<BugType> BT; + enum VLASize_Kind { VLA_Garbage, VLA_Zero, VLA_Tainted }; + + void reportBug(VLASize_Kind Kind, + const Expr *SizeE, + ProgramStateRef State, + CheckerContext &C) const; +public: + void checkPreStmt(const DeclStmt *DS, CheckerContext &C) const; +}; +} // end anonymous namespace + +void VLASizeChecker::reportBug(VLASize_Kind Kind, + const Expr *SizeE, + ProgramStateRef State, + CheckerContext &C) const { + // Generate an error node. + ExplodedNode *N = C.generateSink(State); + if (!N) + return; + + if (!BT) + BT.reset(new BuiltinBug("Dangerous variable-length array (VLA) declaration")); + + SmallString<256> buf; + llvm::raw_svector_ostream os(buf); + os << "Declared variable-length array (VLA) "; + switch (Kind) { + case VLA_Garbage: + os << "uses a garbage value as its size"; + break; + case VLA_Zero: + os << "has zero size"; + break; + case VLA_Tainted: + os << "has tainted size"; + break; + } + + BugReport *report = new BugReport(*BT, os.str(), N); + report->addRange(SizeE->getSourceRange()); + report->addVisitor(bugreporter::getTrackNullOrUndefValueVisitor(N, SizeE, + report)); + C.EmitReport(report); + return; +} + +void VLASizeChecker::checkPreStmt(const DeclStmt *DS, CheckerContext &C) const { + if (!DS->isSingleDecl()) + return; + + const VarDecl *VD = dyn_cast<VarDecl>(DS->getSingleDecl()); + if (!VD) + return; + + ASTContext &Ctx = C.getASTContext(); + const VariableArrayType *VLA = Ctx.getAsVariableArrayType(VD->getType()); + if (!VLA) + return; + + // FIXME: Handle multi-dimensional VLAs. + const Expr *SE = VLA->getSizeExpr(); + ProgramStateRef state = C.getState(); + SVal sizeV = state->getSVal(SE, C.getLocationContext()); + + if (sizeV.isUndef()) { + reportBug(VLA_Garbage, SE, state, C); + return; + } + + // See if the size value is known. It can't be undefined because we would have + // warned about that already. + if (sizeV.isUnknown()) + return; + + // Check if the size is tainted. + if (state->isTainted(sizeV)) { + reportBug(VLA_Tainted, SE, 0, C); + return; + } + + // Check if the size is zero. + DefinedSVal sizeD = cast<DefinedSVal>(sizeV); + + ProgramStateRef stateNotZero, stateZero; + llvm::tie(stateNotZero, stateZero) = state->assume(sizeD); + + if (stateZero && !stateNotZero) { + reportBug(VLA_Zero, SE, stateZero, C); + return; + } + + // From this point on, assume that the size is not zero. + state = stateNotZero; + + // VLASizeChecker is responsible for defining the extent of the array being + // declared. We do this by multiplying the array length by the element size, + // then matching that with the array region's extent symbol. + + // Convert the array length to size_t. + SValBuilder &svalBuilder = C.getSValBuilder(); + QualType SizeTy = Ctx.getSizeType(); + NonLoc ArrayLength = cast<NonLoc>(svalBuilder.evalCast(sizeD, SizeTy, + SE->getType())); + + // Get the element size. + CharUnits EleSize = Ctx.getTypeSizeInChars(VLA->getElementType()); + SVal EleSizeVal = svalBuilder.makeIntVal(EleSize.getQuantity(), SizeTy); + + // Multiply the array length by the element size. + SVal ArraySizeVal = svalBuilder.evalBinOpNN(state, BO_Mul, ArrayLength, + cast<NonLoc>(EleSizeVal), SizeTy); + + // Finally, assume that the array's extent matches the given size. + const LocationContext *LC = C.getLocationContext(); + DefinedOrUnknownSVal Extent = + state->getRegion(VD, LC)->getExtent(svalBuilder); + DefinedOrUnknownSVal ArraySize = cast<DefinedOrUnknownSVal>(ArraySizeVal); + DefinedOrUnknownSVal sizeIsKnown = + svalBuilder.evalEQ(state, Extent, ArraySize); + state = state->assume(sizeIsKnown, true); + + // Assume should not fail at this point. + assert(state); + + // Remember our assumptions! + C.addTransition(state); +} + +void ento::registerVLASizeChecker(CheckerManager &mgr) { + mgr.registerChecker<VLASizeChecker>(); +} diff --git a/clang/lib/StaticAnalyzer/Checkers/VirtualCallChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/VirtualCallChecker.cpp new file mode 100644 index 0000000..f7c7c0c --- /dev/null +++ b/clang/lib/StaticAnalyzer/Checkers/VirtualCallChecker.cpp @@ -0,0 +1,241 @@ +//=======- VirtualCallChecker.cpp --------------------------------*- C++ -*-==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines a checker that checks virtual function calls during +// construction or destruction of C++ objects. +// +//===----------------------------------------------------------------------===// + +#include "ClangSACheckers.h" +#include "clang/AST/DeclCXX.h" +#include "clang/AST/StmtVisitor.h" +#include "llvm/Support/SaveAndRestore.h" +#include "clang/StaticAnalyzer/Core/PathSensitive/AnalysisManager.h" +#include "clang/StaticAnalyzer/Core/Checker.h" +#include "clang/StaticAnalyzer/Core/BugReporter/BugReporter.h" +#include "llvm/ADT/SmallString.h" + +using namespace clang; +using namespace ento; + +namespace { + +class WalkAST : public StmtVisitor<WalkAST> { + BugReporter &BR; + AnalysisDeclContext *AC; + + typedef const CallExpr * WorkListUnit; + typedef SmallVector<WorkListUnit, 20> DFSWorkList; + + /// A vector representing the worklist which has a chain of CallExprs. + DFSWorkList WList; + + // PreVisited : A CallExpr to this FunctionDecl is in the worklist, but the + // body has not been visited yet. + // PostVisited : A CallExpr to this FunctionDecl is in the worklist, and the + // body has been visited. + enum Kind { NotVisited, + PreVisited, /**< A CallExpr to this FunctionDecl is in the + worklist, but the body has not yet been + visited. */ + PostVisited /**< A CallExpr to this FunctionDecl is in the + worklist, and the body has been visited. */ + } K; + + /// A DenseMap that records visited states of FunctionDecls. + llvm::DenseMap<const FunctionDecl *, Kind> VisitedFunctions; + + /// The CallExpr whose body is currently being visited. This is used for + /// generating bug reports. This is null while visiting the body of a + /// constructor or destructor. + const CallExpr *visitingCallExpr; + +public: + WalkAST(BugReporter &br, AnalysisDeclContext *ac) + : BR(br), + AC(ac), + visitingCallExpr(0) {} + + bool hasWork() const { return !WList.empty(); } + + /// This method adds a CallExpr to the worklist and marks the callee as + /// being PreVisited. + void Enqueue(WorkListUnit WLUnit) { + const FunctionDecl *FD = WLUnit->getDirectCallee(); + if (!FD || !FD->getBody()) + return; + Kind &K = VisitedFunctions[FD]; + if (K != NotVisited) + return; + K = PreVisited; + WList.push_back(WLUnit); + } + + /// This method returns an item from the worklist without removing it. + WorkListUnit Dequeue() { + assert(!WList.empty()); + return WList.back(); + } + + void Execute() { + while (hasWork()) { + WorkListUnit WLUnit = Dequeue(); + const FunctionDecl *FD = WLUnit->getDirectCallee(); + assert(FD && FD->getBody()); + + if (VisitedFunctions[FD] == PreVisited) { + // If the callee is PreVisited, walk its body. + // Visit the body. + SaveAndRestore<const CallExpr *> SaveCall(visitingCallExpr, WLUnit); + Visit(FD->getBody()); + + // Mark the function as being PostVisited to indicate we have + // scanned the body. + VisitedFunctions[FD] = PostVisited; + continue; + } + + // Otherwise, the callee is PostVisited. + // Remove it from the worklist. + assert(VisitedFunctions[FD] == PostVisited); + WList.pop_back(); + } + } + + // Stmt visitor methods. + void VisitCallExpr(CallExpr *CE); + void VisitCXXMemberCallExpr(CallExpr *CE); + void VisitStmt(Stmt *S) { VisitChildren(S); } + void VisitChildren(Stmt *S); + + void ReportVirtualCall(const CallExpr *CE, bool isPure); + +}; +} // end anonymous namespace + +//===----------------------------------------------------------------------===// +// AST walking. +//===----------------------------------------------------------------------===// + +void WalkAST::VisitChildren(Stmt *S) { + for (Stmt::child_iterator I = S->child_begin(), E = S->child_end(); I!=E; ++I) + if (Stmt *child = *I) + Visit(child); +} + +void WalkAST::VisitCallExpr(CallExpr *CE) { + VisitChildren(CE); + Enqueue(CE); +} + +void WalkAST::VisitCXXMemberCallExpr(CallExpr *CE) { + VisitChildren(CE); + bool callIsNonVirtual = false; + + // Several situations to elide for checking. + if (MemberExpr *CME = dyn_cast<MemberExpr>(CE->getCallee())) { + // If the member access is fully qualified (i.e., X::F), then treat + // this as a non-virtual call and do not warn. + if (CME->getQualifier()) + callIsNonVirtual = true; + + // Elide analyzing the call entirely if the base pointer is not 'this'. + if (Expr *base = CME->getBase()->IgnoreImpCasts()) + if (!isa<CXXThisExpr>(base)) + return; + } + + // Get the callee. + const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(CE->getDirectCallee()); + if (MD && MD->isVirtual() && !callIsNonVirtual) + ReportVirtualCall(CE, MD->isPure()); + + Enqueue(CE); +} + +void WalkAST::ReportVirtualCall(const CallExpr *CE, bool isPure) { + SmallString<100> buf; + llvm::raw_svector_ostream os(buf); + + os << "Call Path : "; + // Name of current visiting CallExpr. + os << *CE->getDirectCallee(); + + // Name of the CallExpr whose body is current walking. + if (visitingCallExpr) + os << " <-- " << *visitingCallExpr->getDirectCallee(); + // Names of FunctionDecls in worklist with state PostVisited. + for (SmallVectorImpl<const CallExpr *>::iterator I = WList.end(), + E = WList.begin(); I != E; --I) { + const FunctionDecl *FD = (*(I-1))->getDirectCallee(); + assert(FD); + if (VisitedFunctions[FD] == PostVisited) + os << " <-- " << *FD; + } + + PathDiagnosticLocation CELoc = + PathDiagnosticLocation::createBegin(CE, BR.getSourceManager(), AC); + SourceRange R = CE->getCallee()->getSourceRange(); + + if (isPure) { + os << "\n" << "Call pure virtual functions during construction or " + << "destruction may leads undefined behaviour"; + BR.EmitBasicReport(AC->getDecl(), + "Call pure virtual function during construction or " + "Destruction", + "Cplusplus", + os.str(), CELoc, &R, 1); + return; + } + else { + os << "\n" << "Call virtual functions during construction or " + << "destruction will never go to a more derived class"; + BR.EmitBasicReport(AC->getDecl(), + "Call virtual function during construction or " + "Destruction", + "Cplusplus", + os.str(), CELoc, &R, 1); + return; + } +} + +//===----------------------------------------------------------------------===// +// VirtualCallChecker +//===----------------------------------------------------------------------===// + +namespace { +class VirtualCallChecker : public Checker<check::ASTDecl<CXXRecordDecl> > { +public: + void checkASTDecl(const CXXRecordDecl *RD, AnalysisManager& mgr, + BugReporter &BR) const { + WalkAST walker(BR, mgr.getAnalysisDeclContext(RD)); + + // Check the constructors. + for (CXXRecordDecl::ctor_iterator I = RD->ctor_begin(), E = RD->ctor_end(); + I != E; ++I) { + if (!I->isCopyOrMoveConstructor()) + if (Stmt *Body = I->getBody()) { + walker.Visit(Body); + walker.Execute(); + } + } + + // Check the destructor. + if (CXXDestructorDecl *DD = RD->getDestructor()) + if (Stmt *Body = DD->getBody()) { + walker.Visit(Body); + walker.Execute(); + } + } +}; +} + +void ento::registerVirtualCallChecker(CheckerManager &mgr) { + mgr.registerChecker<VirtualCallChecker>(); +} |