1 //===- BugReporterVisitors.cpp - Helpers for reporting bugs ---------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 //  This file defines a set of BugReporter "visitors" which can be used to
10 //  enhance the diagnostics reported for a bug.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "clang/StaticAnalyzer/Core/BugReporter/BugReporterVisitors.h"
15 #include "clang/AST/ASTContext.h"
16 #include "clang/AST/Decl.h"
17 #include "clang/AST/DeclBase.h"
18 #include "clang/AST/DeclCXX.h"
19 #include "clang/AST/Expr.h"
20 #include "clang/AST/ExprCXX.h"
21 #include "clang/AST/ExprObjC.h"
22 #include "clang/AST/Stmt.h"
23 #include "clang/AST/Type.h"
24 #include "clang/ASTMatchers/ASTMatchFinder.h"
25 #include "clang/Analysis/Analyses/Dominators.h"
26 #include "clang/Analysis/AnalysisDeclContext.h"
27 #include "clang/Analysis/CFG.h"
28 #include "clang/Analysis/CFGStmtMap.h"
29 #include "clang/Analysis/PathDiagnostic.h"
30 #include "clang/Analysis/ProgramPoint.h"
31 #include "clang/Basic/IdentifierTable.h"
32 #include "clang/Basic/LLVM.h"
33 #include "clang/Basic/SourceLocation.h"
34 #include "clang/Basic/SourceManager.h"
35 #include "clang/Lex/Lexer.h"
36 #include "clang/StaticAnalyzer/Core/AnalyzerOptions.h"
37 #include "clang/StaticAnalyzer/Core/BugReporter/BugReporter.h"
38 #include "clang/StaticAnalyzer/Core/PathSensitive/AnalysisManager.h"
39 #include "clang/StaticAnalyzer/Core/PathSensitive/CallEvent.h"
40 #include "clang/StaticAnalyzer/Core/PathSensitive/ExplodedGraph.h"
41 #include "clang/StaticAnalyzer/Core/PathSensitive/ExprEngine.h"
42 #include "clang/StaticAnalyzer/Core/PathSensitive/MemRegion.h"
43 #include "clang/StaticAnalyzer/Core/PathSensitive/ProgramState.h"
44 #include "clang/StaticAnalyzer/Core/PathSensitive/ProgramState_Fwd.h"
45 #include "clang/StaticAnalyzer/Core/PathSensitive/SMTConv.h"
46 #include "clang/StaticAnalyzer/Core/PathSensitive/SValBuilder.h"
47 #include "clang/StaticAnalyzer/Core/PathSensitive/SVals.h"
48 #include "llvm/ADT/ArrayRef.h"
49 #include "llvm/ADT/None.h"
50 #include "llvm/ADT/Optional.h"
51 #include "llvm/ADT/STLExtras.h"
52 #include "llvm/ADT/SmallPtrSet.h"
53 #include "llvm/ADT/SmallString.h"
54 #include "llvm/ADT/SmallVector.h"
55 #include "llvm/ADT/StringExtras.h"
56 #include "llvm/ADT/StringRef.h"
57 #include "llvm/Support/Casting.h"
58 #include "llvm/Support/ErrorHandling.h"
59 #include "llvm/Support/raw_ostream.h"
60 #include <cassert>
61 #include <deque>
62 #include <memory>
63 #include <string>
64 #include <utility>
65 
66 using namespace clang;
67 using namespace ento;
68 
69 //===----------------------------------------------------------------------===//
70 // Utility functions.
71 //===----------------------------------------------------------------------===//
72 
peelOffPointerArithmetic(const BinaryOperator * B)73 static const Expr *peelOffPointerArithmetic(const BinaryOperator *B) {
74   if (B->isAdditiveOp() && B->getType()->isPointerType()) {
75     if (B->getLHS()->getType()->isPointerType()) {
76       return B->getLHS();
77     } else if (B->getRHS()->getType()->isPointerType()) {
78       return B->getRHS();
79     }
80   }
81   return nullptr;
82 }
83 
84 /// Given that expression S represents a pointer that would be dereferenced,
85 /// try to find a sub-expression from which the pointer came from.
86 /// This is used for tracking down origins of a null or undefined value:
87 /// "this is null because that is null because that is null" etc.
88 /// We wipe away field and element offsets because they merely add offsets.
89 /// We also wipe away all casts except lvalue-to-rvalue casts, because the
90 /// latter represent an actual pointer dereference; however, we remove
91 /// the final lvalue-to-rvalue cast before returning from this function
92 /// because it demonstrates more clearly from where the pointer rvalue was
93 /// loaded. Examples:
94 ///   x->y.z      ==>  x (lvalue)
95 ///   foo()->y.z  ==>  foo() (rvalue)
getDerefExpr(const Stmt * S)96 const Expr *bugreporter::getDerefExpr(const Stmt *S) {
97   const auto *E = dyn_cast<Expr>(S);
98   if (!E)
99     return nullptr;
100 
101   while (true) {
102     if (const auto *CE = dyn_cast<CastExpr>(E)) {
103       if (CE->getCastKind() == CK_LValueToRValue) {
104         // This cast represents the load we're looking for.
105         break;
106       }
107       E = CE->getSubExpr();
108     } else if (const auto *B = dyn_cast<BinaryOperator>(E)) {
109       // Pointer arithmetic: '*(x + 2)' -> 'x') etc.
110       if (const Expr *Inner = peelOffPointerArithmetic(B)) {
111         E = Inner;
112       } else {
113         // Probably more arithmetic can be pattern-matched here,
114         // but for now give up.
115         break;
116       }
117     } else if (const auto *U = dyn_cast<UnaryOperator>(E)) {
118       if (U->getOpcode() == UO_Deref || U->getOpcode() == UO_AddrOf ||
119           (U->isIncrementDecrementOp() && U->getType()->isPointerType())) {
120         // Operators '*' and '&' don't actually mean anything.
121         // We look at casts instead.
122         E = U->getSubExpr();
123       } else {
124         // Probably more arithmetic can be pattern-matched here,
125         // but for now give up.
126         break;
127       }
128     }
129     // Pattern match for a few useful cases: a[0], p->f, *p etc.
130     else if (const auto *ME = dyn_cast<MemberExpr>(E)) {
131       E = ME->getBase();
132     } else if (const auto *IvarRef = dyn_cast<ObjCIvarRefExpr>(E)) {
133       E = IvarRef->getBase();
134     } else if (const auto *AE = dyn_cast<ArraySubscriptExpr>(E)) {
135       E = AE->getBase();
136     } else if (const auto *PE = dyn_cast<ParenExpr>(E)) {
137       E = PE->getSubExpr();
138     } else if (const auto *FE = dyn_cast<FullExpr>(E)) {
139       E = FE->getSubExpr();
140     } else {
141       // Other arbitrary stuff.
142       break;
143     }
144   }
145 
146   // Special case: remove the final lvalue-to-rvalue cast, but do not recurse
147   // deeper into the sub-expression. This way we return the lvalue from which
148   // our pointer rvalue was loaded.
149   if (const auto *CE = dyn_cast<ImplicitCastExpr>(E))
150     if (CE->getCastKind() == CK_LValueToRValue)
151       E = CE->getSubExpr();
152 
153   return E;
154 }
155 
156 /// Comparing internal representations of symbolic values (via
157 /// SVal::operator==()) is a valid way to check if the value was updated,
158 /// unless it's a LazyCompoundVal that may have a different internal
159 /// representation every time it is loaded from the state. In this function we
160 /// do an approximate comparison for lazy compound values, checking that they
161 /// are the immediate snapshots of the tracked region's bindings within the
162 /// node's respective states but not really checking that these snapshots
163 /// actually contain the same set of bindings.
hasVisibleUpdate(const ExplodedNode * LeftNode,SVal LeftVal,const ExplodedNode * RightNode,SVal RightVal)164 static bool hasVisibleUpdate(const ExplodedNode *LeftNode, SVal LeftVal,
165                              const ExplodedNode *RightNode, SVal RightVal) {
166   if (LeftVal == RightVal)
167     return true;
168 
169   const auto LLCV = LeftVal.getAs<nonloc::LazyCompoundVal>();
170   if (!LLCV)
171     return false;
172 
173   const auto RLCV = RightVal.getAs<nonloc::LazyCompoundVal>();
174   if (!RLCV)
175     return false;
176 
177   return LLCV->getRegion() == RLCV->getRegion() &&
178     LLCV->getStore() == LeftNode->getState()->getStore() &&
179     RLCV->getStore() == RightNode->getState()->getStore();
180 }
181 
getSValForVar(const Expr * CondVarExpr,const ExplodedNode * N)182 static Optional<SVal> getSValForVar(const Expr *CondVarExpr,
183                                     const ExplodedNode *N) {
184   ProgramStateRef State = N->getState();
185   const LocationContext *LCtx = N->getLocationContext();
186 
187   assert(CondVarExpr);
188   CondVarExpr = CondVarExpr->IgnoreImpCasts();
189 
190   // The declaration of the value may rely on a pointer so take its l-value.
191   // FIXME: As seen in VisitCommonDeclRefExpr, sometimes DeclRefExpr may
192   // evaluate to a FieldRegion when it refers to a declaration of a lambda
193   // capture variable. We most likely need to duplicate that logic here.
194   if (const auto *DRE = dyn_cast<DeclRefExpr>(CondVarExpr))
195     if (const auto *VD = dyn_cast<VarDecl>(DRE->getDecl()))
196       return State->getSVal(State->getLValue(VD, LCtx));
197 
198   if (const auto *ME = dyn_cast<MemberExpr>(CondVarExpr))
199     if (const auto *FD = dyn_cast<FieldDecl>(ME->getMemberDecl()))
200       if (auto FieldL = State->getSVal(ME, LCtx).getAs<Loc>())
201         return State->getRawSVal(*FieldL, FD->getType());
202 
203   return None;
204 }
205 
206 static Optional<const llvm::APSInt *>
getConcreteIntegerValue(const Expr * CondVarExpr,const ExplodedNode * N)207 getConcreteIntegerValue(const Expr *CondVarExpr, const ExplodedNode *N) {
208 
209   if (Optional<SVal> V = getSValForVar(CondVarExpr, N))
210     if (auto CI = V->getAs<nonloc::ConcreteInt>())
211       return &CI->getValue();
212   return None;
213 }
214 
isVarAnInterestingCondition(const Expr * CondVarExpr,const ExplodedNode * N,const PathSensitiveBugReport * B)215 static bool isVarAnInterestingCondition(const Expr *CondVarExpr,
216                                         const ExplodedNode *N,
217                                         const PathSensitiveBugReport *B) {
218   // Even if this condition is marked as interesting, it isn't *that*
219   // interesting if it didn't happen in a nested stackframe, the user could just
220   // follow the arrows.
221   if (!B->getErrorNode()->getStackFrame()->isParentOf(N->getStackFrame()))
222     return false;
223 
224   if (Optional<SVal> V = getSValForVar(CondVarExpr, N))
225     if (Optional<bugreporter::TrackingKind> K = B->getInterestingnessKind(*V))
226       return *K == bugreporter::TrackingKind::Condition;
227 
228   return false;
229 }
230 
isInterestingExpr(const Expr * E,const ExplodedNode * N,const PathSensitiveBugReport * B)231 static bool isInterestingExpr(const Expr *E, const ExplodedNode *N,
232                               const PathSensitiveBugReport *B) {
233   if (Optional<SVal> V = getSValForVar(E, N))
234     return B->getInterestingnessKind(*V).hasValue();
235   return false;
236 }
237 
238 /// \return name of the macro inside the location \p Loc.
getMacroName(SourceLocation Loc,BugReporterContext & BRC)239 static StringRef getMacroName(SourceLocation Loc,
240     BugReporterContext &BRC) {
241   return Lexer::getImmediateMacroName(
242       Loc,
243       BRC.getSourceManager(),
244       BRC.getASTContext().getLangOpts());
245 }
246 
247 /// \return Whether given spelling location corresponds to an expansion
248 /// of a function-like macro.
isFunctionMacroExpansion(SourceLocation Loc,const SourceManager & SM)249 static bool isFunctionMacroExpansion(SourceLocation Loc,
250                                 const SourceManager &SM) {
251   if (!Loc.isMacroID())
252     return false;
253   while (SM.isMacroArgExpansion(Loc))
254     Loc = SM.getImmediateExpansionRange(Loc).getBegin();
255   std::pair<FileID, unsigned> TLInfo = SM.getDecomposedLoc(Loc);
256   SrcMgr::SLocEntry SE = SM.getSLocEntry(TLInfo.first);
257   const SrcMgr::ExpansionInfo &EInfo = SE.getExpansion();
258   return EInfo.isFunctionMacroExpansion();
259 }
260 
261 /// \return Whether \c RegionOfInterest was modified at \p N,
262 /// where \p ValueAfter is \c RegionOfInterest's value at the end of the
263 /// stack frame.
wasRegionOfInterestModifiedAt(const SubRegion * RegionOfInterest,const ExplodedNode * N,SVal ValueAfter)264 static bool wasRegionOfInterestModifiedAt(const SubRegion *RegionOfInterest,
265                                           const ExplodedNode *N,
266                                           SVal ValueAfter) {
267   ProgramStateRef State = N->getState();
268   ProgramStateManager &Mgr = N->getState()->getStateManager();
269 
270   if (!N->getLocationAs<PostStore>() && !N->getLocationAs<PostInitializer>() &&
271       !N->getLocationAs<PostStmt>())
272     return false;
273 
274   // Writing into region of interest.
275   if (auto PS = N->getLocationAs<PostStmt>())
276     if (auto *BO = PS->getStmtAs<BinaryOperator>())
277       if (BO->isAssignmentOp() && RegionOfInterest->isSubRegionOf(
278                                       N->getSVal(BO->getLHS()).getAsRegion()))
279         return true;
280 
281   // SVal after the state is possibly different.
282   SVal ValueAtN = N->getState()->getSVal(RegionOfInterest);
283   if (!Mgr.getSValBuilder()
284            .areEqual(State, ValueAtN, ValueAfter)
285            .isConstrainedTrue() &&
286       (!ValueAtN.isUndef() || !ValueAfter.isUndef()))
287     return true;
288 
289   return false;
290 }
291 
292 //===----------------------------------------------------------------------===//
293 // Implementation of BugReporterVisitor.
294 //===----------------------------------------------------------------------===//
295 
getEndPath(BugReporterContext &,const ExplodedNode *,PathSensitiveBugReport &)296 PathDiagnosticPieceRef BugReporterVisitor::getEndPath(BugReporterContext &,
297                                                       const ExplodedNode *,
298                                                       PathSensitiveBugReport &) {
299   return nullptr;
300 }
301 
finalizeVisitor(BugReporterContext &,const ExplodedNode *,PathSensitiveBugReport &)302 void BugReporterVisitor::finalizeVisitor(BugReporterContext &,
303                                          const ExplodedNode *,
304                                          PathSensitiveBugReport &) {}
305 
306 PathDiagnosticPieceRef
getDefaultEndPath(const BugReporterContext & BRC,const ExplodedNode * EndPathNode,const PathSensitiveBugReport & BR)307 BugReporterVisitor::getDefaultEndPath(const BugReporterContext &BRC,
308                                       const ExplodedNode *EndPathNode,
309                                       const PathSensitiveBugReport &BR) {
310   PathDiagnosticLocation L = BR.getLocation();
311   const auto &Ranges = BR.getRanges();
312 
313   // Only add the statement itself as a range if we didn't specify any
314   // special ranges for this report.
315   auto P = std::make_shared<PathDiagnosticEventPiece>(
316       L, BR.getDescription(), Ranges.begin() == Ranges.end());
317   for (SourceRange Range : Ranges)
318     P->addRange(Range);
319 
320   return P;
321 }
322 
323 //===----------------------------------------------------------------------===//
324 // Implementation of NoStoreFuncVisitor.
325 //===----------------------------------------------------------------------===//
326 
327 namespace {
328 
329 /// Put a diagnostic on return statement of all inlined functions
330 /// for which  the region of interest \p RegionOfInterest was passed into,
331 /// but not written inside, and it has caused an undefined read or a null
332 /// pointer dereference outside.
333 class NoStoreFuncVisitor final : public BugReporterVisitor {
334   const SubRegion *RegionOfInterest;
335   MemRegionManager &MmrMgr;
336   const SourceManager &SM;
337   const PrintingPolicy &PP;
338   bugreporter::TrackingKind TKind;
339 
340   /// Recursion limit for dereferencing fields when looking for the
341   /// region of interest.
342   /// The limit of two indicates that we will dereference fields only once.
343   static const unsigned DEREFERENCE_LIMIT = 2;
344 
345   /// Frames writing into \c RegionOfInterest.
346   /// This visitor generates a note only if a function does not write into
347   /// a region of interest. This information is not immediately available
348   /// by looking at the node associated with the exit from the function
349   /// (usually the return statement). To avoid recomputing the same information
350   /// many times (going up the path for each node and checking whether the
351   /// region was written into) we instead lazily compute the
352   /// stack frames along the path which write into the region of interest.
353   llvm::SmallPtrSet<const StackFrameContext *, 32> FramesModifyingRegion;
354   llvm::SmallPtrSet<const StackFrameContext *, 32> FramesModifyingCalculated;
355 
356   using RegionVector = SmallVector<const MemRegion *, 5>;
357 
358 public:
NoStoreFuncVisitor(const SubRegion * R,bugreporter::TrackingKind TKind)359   NoStoreFuncVisitor(const SubRegion *R, bugreporter::TrackingKind TKind)
360       : RegionOfInterest(R), MmrMgr(R->getMemRegionManager()),
361         SM(MmrMgr.getContext().getSourceManager()),
362         PP(MmrMgr.getContext().getPrintingPolicy()), TKind(TKind) {}
363 
Profile(llvm::FoldingSetNodeID & ID) const364   void Profile(llvm::FoldingSetNodeID &ID) const override {
365     static int Tag = 0;
366     ID.AddPointer(&Tag);
367     ID.AddPointer(RegionOfInterest);
368   }
369 
getTag() const370   void *getTag() const {
371     static int Tag = 0;
372     return static_cast<void *>(&Tag);
373   }
374 
375   PathDiagnosticPieceRef VisitNode(const ExplodedNode *N,
376                                    BugReporterContext &BR,
377                                    PathSensitiveBugReport &R) override;
378 
379 private:
380   /// Attempts to find the region of interest in a given record decl,
381   /// by either following the base classes or fields.
382   /// Dereferences fields up to a given recursion limit.
383   /// Note that \p Vec is passed by value, leading to quadratic copying cost,
384   /// but it's OK in practice since its length is limited to DEREFERENCE_LIMIT.
385   /// \return A chain fields leading to the region of interest or None.
386   const Optional<RegionVector>
387   findRegionOfInterestInRecord(const RecordDecl *RD, ProgramStateRef State,
388                                const MemRegion *R, const RegionVector &Vec = {},
389                                int depth = 0);
390 
391   /// Check and lazily calculate whether the region of interest is
392   /// modified in the stack frame to which \p N belongs.
393   /// The calculation is cached in FramesModifyingRegion.
isRegionOfInterestModifiedInFrame(const ExplodedNode * N)394   bool isRegionOfInterestModifiedInFrame(const ExplodedNode *N) {
395     const LocationContext *Ctx = N->getLocationContext();
396     const StackFrameContext *SCtx = Ctx->getStackFrame();
397     if (!FramesModifyingCalculated.count(SCtx))
398       findModifyingFrames(N);
399     return FramesModifyingRegion.count(SCtx);
400   }
401 
402   /// Write to \c FramesModifyingRegion all stack frames along
403   /// the path in the current stack frame which modify \c RegionOfInterest.
404   void findModifyingFrames(const ExplodedNode *N);
405 
406   /// Consume the information on the no-store stack frame in order to
407   /// either emit a note or suppress the report enirely.
408   /// \return Diagnostics piece for region not modified in the current function,
409   /// if it decides to emit one.
410   PathDiagnosticPieceRef
411   maybeEmitNote(PathSensitiveBugReport &R, const CallEvent &Call,
412                 const ExplodedNode *N, const RegionVector &FieldChain,
413                 const MemRegion *MatchedRegion, StringRef FirstElement,
414                 bool FirstIsReferenceType, unsigned IndirectionLevel);
415 
416   /// Pretty-print region \p MatchedRegion to \p os.
417   /// \return Whether printing succeeded.
418   bool prettyPrintRegionName(StringRef FirstElement, bool FirstIsReferenceType,
419                              const MemRegion *MatchedRegion,
420                              const RegionVector &FieldChain,
421                              int IndirectionLevel,
422                              llvm::raw_svector_ostream &os);
423 
424   /// Print first item in the chain, return new separator.
425   static StringRef prettyPrintFirstElement(StringRef FirstElement,
426                                            bool MoreItemsExpected,
427                                            int IndirectionLevel,
428                                            llvm::raw_svector_ostream &os);
429 };
430 
431 } // end of anonymous namespace
432 
433 /// \return Whether the method declaration \p Parent
434 /// syntactically has a binary operation writing into the ivar \p Ivar.
potentiallyWritesIntoIvar(const Decl * Parent,const ObjCIvarDecl * Ivar)435 static bool potentiallyWritesIntoIvar(const Decl *Parent,
436                                       const ObjCIvarDecl *Ivar) {
437   using namespace ast_matchers;
438   const char *IvarBind = "Ivar";
439   if (!Parent || !Parent->hasBody())
440     return false;
441   StatementMatcher WriteIntoIvarM = binaryOperator(
442       hasOperatorName("="),
443       hasLHS(ignoringParenImpCasts(
444           objcIvarRefExpr(hasDeclaration(equalsNode(Ivar))).bind(IvarBind))));
445   StatementMatcher ParentM = stmt(hasDescendant(WriteIntoIvarM));
446   auto Matches = match(ParentM, *Parent->getBody(), Parent->getASTContext());
447   for (BoundNodes &Match : Matches) {
448     auto IvarRef = Match.getNodeAs<ObjCIvarRefExpr>(IvarBind);
449     if (IvarRef->isFreeIvar())
450       return true;
451 
452     const Expr *Base = IvarRef->getBase();
453     if (const auto *ICE = dyn_cast<ImplicitCastExpr>(Base))
454       Base = ICE->getSubExpr();
455 
456     if (const auto *DRE = dyn_cast<DeclRefExpr>(Base))
457       if (const auto *ID = dyn_cast<ImplicitParamDecl>(DRE->getDecl()))
458         if (ID->getParameterKind() == ImplicitParamDecl::ObjCSelf)
459           return true;
460 
461     return false;
462   }
463   return false;
464 }
465 
466 /// Get parameters associated with runtime definition in order
467 /// to get the correct parameter name.
getCallParameters(CallEventRef<> Call)468 static ArrayRef<ParmVarDecl *> getCallParameters(CallEventRef<> Call) {
469   // Use runtime definition, if available.
470   RuntimeDefinition RD = Call->getRuntimeDefinition();
471   if (const auto *FD = dyn_cast_or_null<FunctionDecl>(RD.getDecl()))
472     return FD->parameters();
473   if (const auto *MD = dyn_cast_or_null<ObjCMethodDecl>(RD.getDecl()))
474     return MD->parameters();
475 
476   return Call->parameters();
477 }
478 
479 /// \return whether \p Ty points to a const type, or is a const reference.
isPointerToConst(QualType Ty)480 static bool isPointerToConst(QualType Ty) {
481   return !Ty->getPointeeType().isNull() &&
482          Ty->getPointeeType().getCanonicalType().isConstQualified();
483 }
484 
485 /// Attempts to find the region of interest in a given CXX decl,
486 /// by either following the base classes or fields.
487 /// Dereferences fields up to a given recursion limit.
488 /// Note that \p Vec is passed by value, leading to quadratic copying cost,
489 /// but it's OK in practice since its length is limited to DEREFERENCE_LIMIT.
490 /// \return A chain fields leading to the region of interest or None.
491 const Optional<NoStoreFuncVisitor::RegionVector>
findRegionOfInterestInRecord(const RecordDecl * RD,ProgramStateRef State,const MemRegion * R,const NoStoreFuncVisitor::RegionVector & Vec,int depth)492 NoStoreFuncVisitor::findRegionOfInterestInRecord(
493     const RecordDecl *RD, ProgramStateRef State, const MemRegion *R,
494     const NoStoreFuncVisitor::RegionVector &Vec /* = {} */,
495     int depth /* = 0 */) {
496 
497   if (depth == DEREFERENCE_LIMIT) // Limit the recursion depth.
498     return None;
499 
500   if (const auto *RDX = dyn_cast<CXXRecordDecl>(RD))
501     if (!RDX->hasDefinition())
502       return None;
503 
504   // Recursively examine the base classes.
505   // Note that following base classes does not increase the recursion depth.
506   if (const auto *RDX = dyn_cast<CXXRecordDecl>(RD))
507     for (const auto &II : RDX->bases())
508       if (const RecordDecl *RRD = II.getType()->getAsRecordDecl())
509         if (Optional<RegionVector> Out =
510                 findRegionOfInterestInRecord(RRD, State, R, Vec, depth))
511           return Out;
512 
513   for (const FieldDecl *I : RD->fields()) {
514     QualType FT = I->getType();
515     const FieldRegion *FR = MmrMgr.getFieldRegion(I, cast<SubRegion>(R));
516     const SVal V = State->getSVal(FR);
517     const MemRegion *VR = V.getAsRegion();
518 
519     RegionVector VecF = Vec;
520     VecF.push_back(FR);
521 
522     if (RegionOfInterest == VR)
523       return VecF;
524 
525     if (const RecordDecl *RRD = FT->getAsRecordDecl())
526       if (auto Out =
527               findRegionOfInterestInRecord(RRD, State, FR, VecF, depth + 1))
528         return Out;
529 
530     QualType PT = FT->getPointeeType();
531     if (PT.isNull() || PT->isVoidType() || !VR)
532       continue;
533 
534     if (const RecordDecl *RRD = PT->getAsRecordDecl())
535       if (Optional<RegionVector> Out =
536               findRegionOfInterestInRecord(RRD, State, VR, VecF, depth + 1))
537         return Out;
538   }
539 
540   return None;
541 }
542 
543 PathDiagnosticPieceRef
VisitNode(const ExplodedNode * N,BugReporterContext & BR,PathSensitiveBugReport & R)544 NoStoreFuncVisitor::VisitNode(const ExplodedNode *N, BugReporterContext &BR,
545                               PathSensitiveBugReport &R) {
546 
547   const LocationContext *Ctx = N->getLocationContext();
548   const StackFrameContext *SCtx = Ctx->getStackFrame();
549   ProgramStateRef State = N->getState();
550   auto CallExitLoc = N->getLocationAs<CallExitBegin>();
551 
552   // No diagnostic if region was modified inside the frame.
553   if (!CallExitLoc || isRegionOfInterestModifiedInFrame(N))
554     return nullptr;
555 
556   CallEventRef<> Call =
557       BR.getStateManager().getCallEventManager().getCaller(SCtx, State);
558 
559   // Region of interest corresponds to an IVar, exiting a method
560   // which could have written into that IVar, but did not.
561   if (const auto *MC = dyn_cast<ObjCMethodCall>(Call)) {
562     if (const auto *IvarR = dyn_cast<ObjCIvarRegion>(RegionOfInterest)) {
563       const MemRegion *SelfRegion = MC->getReceiverSVal().getAsRegion();
564       if (RegionOfInterest->isSubRegionOf(SelfRegion) &&
565           potentiallyWritesIntoIvar(Call->getRuntimeDefinition().getDecl(),
566                                     IvarR->getDecl()))
567         return maybeEmitNote(R, *Call, N, {}, SelfRegion, "self",
568                              /*FirstIsReferenceType=*/false, 1);
569     }
570   }
571 
572   if (const auto *CCall = dyn_cast<CXXConstructorCall>(Call)) {
573     const MemRegion *ThisR = CCall->getCXXThisVal().getAsRegion();
574     if (RegionOfInterest->isSubRegionOf(ThisR) &&
575         !CCall->getDecl()->isImplicit())
576       return maybeEmitNote(R, *Call, N, {}, ThisR, "this",
577                            /*FirstIsReferenceType=*/false, 1);
578 
579     // Do not generate diagnostics for not modified parameters in
580     // constructors.
581     return nullptr;
582   }
583 
584   ArrayRef<ParmVarDecl *> parameters = getCallParameters(Call);
585   for (unsigned I = 0; I < Call->getNumArgs() && I < parameters.size(); ++I) {
586     const ParmVarDecl *PVD = parameters[I];
587     SVal V = Call->getArgSVal(I);
588     bool ParamIsReferenceType = PVD->getType()->isReferenceType();
589     std::string ParamName = PVD->getNameAsString();
590 
591     int IndirectionLevel = 1;
592     QualType T = PVD->getType();
593     while (const MemRegion *MR = V.getAsRegion()) {
594       if (RegionOfInterest->isSubRegionOf(MR) && !isPointerToConst(T))
595         return maybeEmitNote(R, *Call, N, {}, MR, ParamName,
596                              ParamIsReferenceType, IndirectionLevel);
597 
598       QualType PT = T->getPointeeType();
599       if (PT.isNull() || PT->isVoidType())
600         break;
601 
602       if (const RecordDecl *RD = PT->getAsRecordDecl())
603         if (Optional<RegionVector> P =
604                 findRegionOfInterestInRecord(RD, State, MR))
605           return maybeEmitNote(R, *Call, N, *P, RegionOfInterest, ParamName,
606                                ParamIsReferenceType, IndirectionLevel);
607 
608       V = State->getSVal(MR, PT);
609       T = PT;
610       IndirectionLevel++;
611     }
612   }
613 
614   return nullptr;
615 }
616 
findModifyingFrames(const ExplodedNode * N)617 void NoStoreFuncVisitor::findModifyingFrames(const ExplodedNode *N) {
618   assert(N->getLocationAs<CallExitBegin>());
619   ProgramStateRef LastReturnState = N->getState();
620   SVal ValueAtReturn = LastReturnState->getSVal(RegionOfInterest);
621   const LocationContext *Ctx = N->getLocationContext();
622   const StackFrameContext *OriginalSCtx = Ctx->getStackFrame();
623 
624   do {
625     ProgramStateRef State = N->getState();
626     auto CallExitLoc = N->getLocationAs<CallExitBegin>();
627     if (CallExitLoc) {
628       LastReturnState = State;
629       ValueAtReturn = LastReturnState->getSVal(RegionOfInterest);
630     }
631 
632     FramesModifyingCalculated.insert(N->getLocationContext()->getStackFrame());
633 
634     if (wasRegionOfInterestModifiedAt(RegionOfInterest, N, ValueAtReturn)) {
635       const StackFrameContext *SCtx = N->getStackFrame();
636       while (!SCtx->inTopFrame()) {
637         auto p = FramesModifyingRegion.insert(SCtx);
638         if (!p.second)
639           break; // Frame and all its parents already inserted.
640         SCtx = SCtx->getParent()->getStackFrame();
641       }
642     }
643 
644     // Stop calculation at the call to the current function.
645     if (auto CE = N->getLocationAs<CallEnter>())
646       if (CE->getCalleeContext() == OriginalSCtx)
647         break;
648 
649     N = N->getFirstPred();
650   } while (N);
651 }
652 
653 static llvm::StringLiteral WillBeUsedForACondition =
654     ", which participates in a condition later";
655 
maybeEmitNote(PathSensitiveBugReport & R,const CallEvent & Call,const ExplodedNode * N,const RegionVector & FieldChain,const MemRegion * MatchedRegion,StringRef FirstElement,bool FirstIsReferenceType,unsigned IndirectionLevel)656 PathDiagnosticPieceRef NoStoreFuncVisitor::maybeEmitNote(
657     PathSensitiveBugReport &R, const CallEvent &Call, const ExplodedNode *N,
658     const RegionVector &FieldChain, const MemRegion *MatchedRegion,
659     StringRef FirstElement, bool FirstIsReferenceType,
660     unsigned IndirectionLevel) {
661   // Optimistically suppress uninitialized value bugs that result
662   // from system headers having a chance to initialize the value
663   // but failing to do so. It's too unlikely a system header's fault.
664   // It's much more likely a situation in which the function has a failure
665   // mode that the user decided not to check. If we want to hunt such
666   // omitted checks, we should provide an explicit function-specific note
667   // describing the precondition under which the function isn't supposed to
668   // initialize its out-parameter, and additionally check that such
669   // precondition can actually be fulfilled on the current path.
670   if (Call.isInSystemHeader()) {
671     // We make an exception for system header functions that have no branches.
672     // Such functions unconditionally fail to initialize the variable.
673     // If they call other functions that have more paths within them,
674     // this suppression would still apply when we visit these inner functions.
675     // One common example of a standard function that doesn't ever initialize
676     // its out parameter is operator placement new; it's up to the follow-up
677     // constructor (if any) to initialize the memory.
678     if (!N->getStackFrame()->getCFG()->isLinear())
679       R.markInvalid(getTag(), nullptr);
680     return nullptr;
681   }
682 
683   PathDiagnosticLocation L =
684       PathDiagnosticLocation::create(N->getLocation(), SM);
685 
686   // For now this shouldn't trigger, but once it does (as we add more
687   // functions to the body farm), we'll need to decide if these reports
688   // are worth suppressing as well.
689   if (!L.hasValidLocation())
690     return nullptr;
691 
692   SmallString<256> sbuf;
693   llvm::raw_svector_ostream os(sbuf);
694   os << "Returning without writing to '";
695 
696   // Do not generate the note if failed to pretty-print.
697   if (!prettyPrintRegionName(FirstElement, FirstIsReferenceType, MatchedRegion,
698                              FieldChain, IndirectionLevel, os))
699     return nullptr;
700 
701   os << "'";
702   if (TKind == bugreporter::TrackingKind::Condition)
703     os << WillBeUsedForACondition;
704   return std::make_shared<PathDiagnosticEventPiece>(L, os.str());
705 }
706 
prettyPrintRegionName(StringRef FirstElement,bool FirstIsReferenceType,const MemRegion * MatchedRegion,const RegionVector & FieldChain,int IndirectionLevel,llvm::raw_svector_ostream & os)707 bool NoStoreFuncVisitor::prettyPrintRegionName(StringRef FirstElement,
708                                                bool FirstIsReferenceType,
709                                                const MemRegion *MatchedRegion,
710                                                const RegionVector &FieldChain,
711                                                int IndirectionLevel,
712                                                llvm::raw_svector_ostream &os) {
713 
714   if (FirstIsReferenceType)
715     IndirectionLevel--;
716 
717   RegionVector RegionSequence;
718 
719   // Add the regions in the reverse order, then reverse the resulting array.
720   assert(RegionOfInterest->isSubRegionOf(MatchedRegion));
721   const MemRegion *R = RegionOfInterest;
722   while (R != MatchedRegion) {
723     RegionSequence.push_back(R);
724     R = cast<SubRegion>(R)->getSuperRegion();
725   }
726   std::reverse(RegionSequence.begin(), RegionSequence.end());
727   RegionSequence.append(FieldChain.begin(), FieldChain.end());
728 
729   StringRef Sep;
730   for (const MemRegion *R : RegionSequence) {
731 
732     // Just keep going up to the base region.
733     // Element regions may appear due to casts.
734     if (isa<CXXBaseObjectRegion>(R) || isa<CXXTempObjectRegion>(R))
735       continue;
736 
737     if (Sep.empty())
738       Sep = prettyPrintFirstElement(FirstElement,
739                                     /*MoreItemsExpected=*/true,
740                                     IndirectionLevel, os);
741 
742     os << Sep;
743 
744     // Can only reasonably pretty-print DeclRegions.
745     if (!isa<DeclRegion>(R))
746       return false;
747 
748     const auto *DR = cast<DeclRegion>(R);
749     Sep = DR->getValueType()->isAnyPointerType() ? "->" : ".";
750     DR->getDecl()->getDeclName().print(os, PP);
751   }
752 
753   if (Sep.empty())
754     prettyPrintFirstElement(FirstElement,
755                             /*MoreItemsExpected=*/false, IndirectionLevel, os);
756   return true;
757 }
758 
prettyPrintFirstElement(StringRef FirstElement,bool MoreItemsExpected,int IndirectionLevel,llvm::raw_svector_ostream & os)759 StringRef NoStoreFuncVisitor::prettyPrintFirstElement(
760     StringRef FirstElement, bool MoreItemsExpected, int IndirectionLevel,
761     llvm::raw_svector_ostream &os) {
762   StringRef Out = ".";
763 
764   if (IndirectionLevel > 0 && MoreItemsExpected) {
765     IndirectionLevel--;
766     Out = "->";
767   }
768 
769   if (IndirectionLevel > 0 && MoreItemsExpected)
770     os << "(";
771 
772   for (int i = 0; i < IndirectionLevel; i++)
773     os << "*";
774   os << FirstElement;
775 
776   if (IndirectionLevel > 0 && MoreItemsExpected)
777     os << ")";
778 
779   return Out;
780 }
781 
782 //===----------------------------------------------------------------------===//
783 // Implementation of MacroNullReturnSuppressionVisitor.
784 //===----------------------------------------------------------------------===//
785 
786 namespace {
787 
788 /// Suppress null-pointer-dereference bugs where dereferenced null was returned
789 /// the macro.
790 class MacroNullReturnSuppressionVisitor final : public BugReporterVisitor {
791   const SubRegion *RegionOfInterest;
792   const SVal ValueAtDereference;
793 
794   // Do not invalidate the reports where the value was modified
795   // after it got assigned to from the macro.
796   bool WasModified = false;
797 
798 public:
MacroNullReturnSuppressionVisitor(const SubRegion * R,const SVal V)799   MacroNullReturnSuppressionVisitor(const SubRegion *R, const SVal V)
800       : RegionOfInterest(R), ValueAtDereference(V) {}
801 
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)802   PathDiagnosticPieceRef VisitNode(const ExplodedNode *N,
803                                    BugReporterContext &BRC,
804                                    PathSensitiveBugReport &BR) override {
805     if (WasModified)
806       return nullptr;
807 
808     auto BugPoint = BR.getErrorNode()->getLocation().getAs<StmtPoint>();
809     if (!BugPoint)
810       return nullptr;
811 
812     const SourceManager &SMgr = BRC.getSourceManager();
813     if (auto Loc = matchAssignment(N)) {
814       if (isFunctionMacroExpansion(*Loc, SMgr)) {
815         std::string MacroName = std::string(getMacroName(*Loc, BRC));
816         SourceLocation BugLoc = BugPoint->getStmt()->getBeginLoc();
817         if (!BugLoc.isMacroID() || getMacroName(BugLoc, BRC) != MacroName)
818           BR.markInvalid(getTag(), MacroName.c_str());
819       }
820     }
821 
822     if (wasRegionOfInterestModifiedAt(RegionOfInterest, N, ValueAtDereference))
823       WasModified = true;
824 
825     return nullptr;
826   }
827 
addMacroVisitorIfNecessary(const ExplodedNode * N,const MemRegion * R,bool EnableNullFPSuppression,PathSensitiveBugReport & BR,const SVal V)828   static void addMacroVisitorIfNecessary(
829         const ExplodedNode *N, const MemRegion *R,
830         bool EnableNullFPSuppression, PathSensitiveBugReport &BR,
831         const SVal V) {
832     AnalyzerOptions &Options = N->getState()->getAnalysisManager().options;
833     if (EnableNullFPSuppression &&
834         Options.ShouldSuppressNullReturnPaths && V.getAs<Loc>())
835       BR.addVisitor(std::make_unique<MacroNullReturnSuppressionVisitor>(
836               R->getAs<SubRegion>(), V));
837   }
838 
getTag() const839   void* getTag() const {
840     static int Tag = 0;
841     return static_cast<void *>(&Tag);
842   }
843 
Profile(llvm::FoldingSetNodeID & ID) const844   void Profile(llvm::FoldingSetNodeID &ID) const override {
845     ID.AddPointer(getTag());
846   }
847 
848 private:
849   /// \return Source location of right hand side of an assignment
850   /// into \c RegionOfInterest, empty optional if none found.
matchAssignment(const ExplodedNode * N)851   Optional<SourceLocation> matchAssignment(const ExplodedNode *N) {
852     const Stmt *S = N->getStmtForDiagnostics();
853     ProgramStateRef State = N->getState();
854     auto *LCtx = N->getLocationContext();
855     if (!S)
856       return None;
857 
858     if (const auto *DS = dyn_cast<DeclStmt>(S)) {
859       if (const auto *VD = dyn_cast<VarDecl>(DS->getSingleDecl()))
860         if (const Expr *RHS = VD->getInit())
861           if (RegionOfInterest->isSubRegionOf(
862                   State->getLValue(VD, LCtx).getAsRegion()))
863             return RHS->getBeginLoc();
864     } else if (const auto *BO = dyn_cast<BinaryOperator>(S)) {
865       const MemRegion *R = N->getSVal(BO->getLHS()).getAsRegion();
866       const Expr *RHS = BO->getRHS();
867       if (BO->isAssignmentOp() && RegionOfInterest->isSubRegionOf(R)) {
868         return RHS->getBeginLoc();
869       }
870     }
871     return None;
872   }
873 };
874 
875 } // end of anonymous namespace
876 
877 namespace {
878 
879 /// Emits an extra note at the return statement of an interesting stack frame.
880 ///
881 /// The returned value is marked as an interesting value, and if it's null,
882 /// adds a visitor to track where it became null.
883 ///
884 /// This visitor is intended to be used when another visitor discovers that an
885 /// interesting value comes from an inlined function call.
886 class ReturnVisitor : public BugReporterVisitor {
887   const StackFrameContext *CalleeSFC;
888   enum {
889     Initial,
890     MaybeUnsuppress,
891     Satisfied
892   } Mode = Initial;
893 
894   bool EnableNullFPSuppression;
895   bool ShouldInvalidate = true;
896   AnalyzerOptions& Options;
897   bugreporter::TrackingKind TKind;
898 
899 public:
ReturnVisitor(const StackFrameContext * Frame,bool Suppressed,AnalyzerOptions & Options,bugreporter::TrackingKind TKind)900   ReturnVisitor(const StackFrameContext *Frame, bool Suppressed,
901                 AnalyzerOptions &Options, bugreporter::TrackingKind TKind)
902       : CalleeSFC(Frame), EnableNullFPSuppression(Suppressed),
903         Options(Options), TKind(TKind) {}
904 
getTag()905   static void *getTag() {
906     static int Tag = 0;
907     return static_cast<void *>(&Tag);
908   }
909 
Profile(llvm::FoldingSetNodeID & ID) const910   void Profile(llvm::FoldingSetNodeID &ID) const override {
911     ID.AddPointer(ReturnVisitor::getTag());
912     ID.AddPointer(CalleeSFC);
913     ID.AddBoolean(EnableNullFPSuppression);
914   }
915 
916   /// Adds a ReturnVisitor if the given statement represents a call that was
917   /// inlined.
918   ///
919   /// This will search back through the ExplodedGraph, starting from the given
920   /// node, looking for when the given statement was processed. If it turns out
921   /// the statement is a call that was inlined, we add the visitor to the
922   /// bug report, so it can print a note later.
addVisitorIfNecessary(const ExplodedNode * Node,const Stmt * S,PathSensitiveBugReport & BR,bool InEnableNullFPSuppression,bugreporter::TrackingKind TKind)923   static void addVisitorIfNecessary(const ExplodedNode *Node, const Stmt *S,
924                                     PathSensitiveBugReport &BR,
925                                     bool InEnableNullFPSuppression,
926                                     bugreporter::TrackingKind TKind) {
927     if (!CallEvent::isCallStmt(S))
928       return;
929 
930     // First, find when we processed the statement.
931     // If we work with a 'CXXNewExpr' that is going to be purged away before
932     // its call take place. We would catch that purge in the last condition
933     // as a 'StmtPoint' so we have to bypass it.
934     const bool BypassCXXNewExprEval = isa<CXXNewExpr>(S);
935 
936     // This is moving forward when we enter into another context.
937     const StackFrameContext *CurrentSFC = Node->getStackFrame();
938 
939     do {
940       // If that is satisfied we found our statement as an inlined call.
941       if (Optional<CallExitEnd> CEE = Node->getLocationAs<CallExitEnd>())
942         if (CEE->getCalleeContext()->getCallSite() == S)
943           break;
944 
945       // Try to move forward to the end of the call-chain.
946       Node = Node->getFirstPred();
947       if (!Node)
948         break;
949 
950       const StackFrameContext *PredSFC = Node->getStackFrame();
951 
952       // If that is satisfied we found our statement.
953       // FIXME: This code currently bypasses the call site for the
954       //        conservatively evaluated allocator.
955       if (!BypassCXXNewExprEval)
956         if (Optional<StmtPoint> SP = Node->getLocationAs<StmtPoint>())
957           // See if we do not enter into another context.
958           if (SP->getStmt() == S && CurrentSFC == PredSFC)
959             break;
960 
961       CurrentSFC = PredSFC;
962     } while (Node->getStackFrame() == CurrentSFC);
963 
964     // Next, step over any post-statement checks.
965     while (Node && Node->getLocation().getAs<PostStmt>())
966       Node = Node->getFirstPred();
967     if (!Node)
968       return;
969 
970     // Finally, see if we inlined the call.
971     Optional<CallExitEnd> CEE = Node->getLocationAs<CallExitEnd>();
972     if (!CEE)
973       return;
974 
975     const StackFrameContext *CalleeContext = CEE->getCalleeContext();
976     if (CalleeContext->getCallSite() != S)
977       return;
978 
979     // Check the return value.
980     ProgramStateRef State = Node->getState();
981     SVal RetVal = Node->getSVal(S);
982 
983     // Handle cases where a reference is returned and then immediately used.
984     if (cast<Expr>(S)->isGLValue())
985       if (Optional<Loc> LValue = RetVal.getAs<Loc>())
986         RetVal = State->getSVal(*LValue);
987 
988     // See if the return value is NULL. If so, suppress the report.
989     AnalyzerOptions &Options = State->getAnalysisManager().options;
990 
991     bool EnableNullFPSuppression = false;
992     if (InEnableNullFPSuppression &&
993         Options.ShouldSuppressNullReturnPaths)
994       if (Optional<Loc> RetLoc = RetVal.getAs<Loc>())
995         EnableNullFPSuppression = State->isNull(*RetLoc).isConstrainedTrue();
996 
997     BR.addVisitor(std::make_unique<ReturnVisitor>(CalleeContext,
998                                                    EnableNullFPSuppression,
999                                                    Options, TKind));
1000   }
1001 
visitNodeInitial(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)1002   PathDiagnosticPieceRef visitNodeInitial(const ExplodedNode *N,
1003                                           BugReporterContext &BRC,
1004                                           PathSensitiveBugReport &BR) {
1005     // Only print a message at the interesting return statement.
1006     if (N->getLocationContext() != CalleeSFC)
1007       return nullptr;
1008 
1009     Optional<StmtPoint> SP = N->getLocationAs<StmtPoint>();
1010     if (!SP)
1011       return nullptr;
1012 
1013     const auto *Ret = dyn_cast<ReturnStmt>(SP->getStmt());
1014     if (!Ret)
1015       return nullptr;
1016 
1017     // Okay, we're at the right return statement, but do we have the return
1018     // value available?
1019     ProgramStateRef State = N->getState();
1020     SVal V = State->getSVal(Ret, CalleeSFC);
1021     if (V.isUnknownOrUndef())
1022       return nullptr;
1023 
1024     // Don't print any more notes after this one.
1025     Mode = Satisfied;
1026 
1027     const Expr *RetE = Ret->getRetValue();
1028     assert(RetE && "Tracking a return value for a void function");
1029 
1030     // Handle cases where a reference is returned and then immediately used.
1031     Optional<Loc> LValue;
1032     if (RetE->isGLValue()) {
1033       if ((LValue = V.getAs<Loc>())) {
1034         SVal RValue = State->getRawSVal(*LValue, RetE->getType());
1035         if (RValue.getAs<DefinedSVal>())
1036           V = RValue;
1037       }
1038     }
1039 
1040     // Ignore aggregate rvalues.
1041     if (V.getAs<nonloc::LazyCompoundVal>() ||
1042         V.getAs<nonloc::CompoundVal>())
1043       return nullptr;
1044 
1045     RetE = RetE->IgnoreParenCasts();
1046 
1047     // Let's track the return value.
1048     bugreporter::trackExpressionValue(
1049         N, RetE, BR, TKind, EnableNullFPSuppression);
1050 
1051     // Build an appropriate message based on the return value.
1052     SmallString<64> Msg;
1053     llvm::raw_svector_ostream Out(Msg);
1054 
1055     bool WouldEventBeMeaningless = false;
1056 
1057     if (State->isNull(V).isConstrainedTrue()) {
1058       if (V.getAs<Loc>()) {
1059 
1060         // If we have counter-suppression enabled, make sure we keep visiting
1061         // future nodes. We want to emit a path note as well, in case
1062         // the report is resurrected as valid later on.
1063         if (EnableNullFPSuppression &&
1064             Options.ShouldAvoidSuppressingNullArgumentPaths)
1065           Mode = MaybeUnsuppress;
1066 
1067         if (RetE->getType()->isObjCObjectPointerType()) {
1068           Out << "Returning nil";
1069         } else {
1070           Out << "Returning null pointer";
1071         }
1072       } else {
1073         Out << "Returning zero";
1074       }
1075 
1076     } else {
1077       if (auto CI = V.getAs<nonloc::ConcreteInt>()) {
1078         Out << "Returning the value " << CI->getValue();
1079       } else {
1080         // There is nothing interesting about returning a value, when it is
1081         // plain value without any constraints, and the function is guaranteed
1082         // to return that every time. We could use CFG::isLinear() here, but
1083         // constexpr branches are obvious to the compiler, not necesserily to
1084         // the programmer.
1085         if (N->getCFG().size() == 3)
1086           WouldEventBeMeaningless = true;
1087 
1088         if (V.getAs<Loc>())
1089           Out << "Returning pointer";
1090         else
1091           Out << "Returning value";
1092       }
1093     }
1094 
1095     if (LValue) {
1096       if (const MemRegion *MR = LValue->getAsRegion()) {
1097         if (MR->canPrintPretty()) {
1098           Out << " (reference to ";
1099           MR->printPretty(Out);
1100           Out << ")";
1101         }
1102       }
1103     } else {
1104       // FIXME: We should have a more generalized location printing mechanism.
1105       if (const auto *DR = dyn_cast<DeclRefExpr>(RetE))
1106         if (const auto *DD = dyn_cast<DeclaratorDecl>(DR->getDecl()))
1107           Out << " (loaded from '" << *DD << "')";
1108     }
1109 
1110     PathDiagnosticLocation L(Ret, BRC.getSourceManager(), CalleeSFC);
1111     if (!L.isValid() || !L.asLocation().isValid())
1112       return nullptr;
1113 
1114     if (TKind == bugreporter::TrackingKind::Condition)
1115       Out << WillBeUsedForACondition;
1116 
1117     auto EventPiece = std::make_shared<PathDiagnosticEventPiece>(L, Out.str());
1118 
1119     // If we determined that the note is meaningless, make it prunable, and
1120     // don't mark the stackframe interesting.
1121     if (WouldEventBeMeaningless)
1122       EventPiece->setPrunable(true);
1123     else
1124       BR.markInteresting(CalleeSFC);
1125 
1126     return EventPiece;
1127   }
1128 
visitNodeMaybeUnsuppress(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)1129   PathDiagnosticPieceRef visitNodeMaybeUnsuppress(const ExplodedNode *N,
1130                                                   BugReporterContext &BRC,
1131                                                   PathSensitiveBugReport &BR) {
1132     assert(Options.ShouldAvoidSuppressingNullArgumentPaths);
1133 
1134     // Are we at the entry node for this call?
1135     Optional<CallEnter> CE = N->getLocationAs<CallEnter>();
1136     if (!CE)
1137       return nullptr;
1138 
1139     if (CE->getCalleeContext() != CalleeSFC)
1140       return nullptr;
1141 
1142     Mode = Satisfied;
1143 
1144     // Don't automatically suppress a report if one of the arguments is
1145     // known to be a null pointer. Instead, start tracking /that/ null
1146     // value back to its origin.
1147     ProgramStateManager &StateMgr = BRC.getStateManager();
1148     CallEventManager &CallMgr = StateMgr.getCallEventManager();
1149 
1150     ProgramStateRef State = N->getState();
1151     CallEventRef<> Call = CallMgr.getCaller(CalleeSFC, State);
1152     for (unsigned I = 0, E = Call->getNumArgs(); I != E; ++I) {
1153       Optional<Loc> ArgV = Call->getArgSVal(I).getAs<Loc>();
1154       if (!ArgV)
1155         continue;
1156 
1157       const Expr *ArgE = Call->getArgExpr(I);
1158       if (!ArgE)
1159         continue;
1160 
1161       // Is it possible for this argument to be non-null?
1162       if (!State->isNull(*ArgV).isConstrainedTrue())
1163         continue;
1164 
1165       if (trackExpressionValue(N, ArgE, BR, TKind, EnableNullFPSuppression))
1166         ShouldInvalidate = false;
1167 
1168       // If we /can't/ track the null pointer, we should err on the side of
1169       // false negatives, and continue towards marking this report invalid.
1170       // (We will still look at the other arguments, though.)
1171     }
1172 
1173     return nullptr;
1174   }
1175 
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)1176   PathDiagnosticPieceRef VisitNode(const ExplodedNode *N,
1177                                    BugReporterContext &BRC,
1178                                    PathSensitiveBugReport &BR) override {
1179     switch (Mode) {
1180     case Initial:
1181       return visitNodeInitial(N, BRC, BR);
1182     case MaybeUnsuppress:
1183       return visitNodeMaybeUnsuppress(N, BRC, BR);
1184     case Satisfied:
1185       return nullptr;
1186     }
1187 
1188     llvm_unreachable("Invalid visit mode!");
1189   }
1190 
finalizeVisitor(BugReporterContext &,const ExplodedNode *,PathSensitiveBugReport & BR)1191   void finalizeVisitor(BugReporterContext &, const ExplodedNode *,
1192                        PathSensitiveBugReport &BR) override {
1193     if (EnableNullFPSuppression && ShouldInvalidate)
1194       BR.markInvalid(ReturnVisitor::getTag(), CalleeSFC);
1195   }
1196 };
1197 
1198 } // end of anonymous namespace
1199 
1200 //===----------------------------------------------------------------------===//
1201 // Implementation of FindLastStoreBRVisitor.
1202 //===----------------------------------------------------------------------===//
1203 
Profile(llvm::FoldingSetNodeID & ID) const1204 void FindLastStoreBRVisitor::Profile(llvm::FoldingSetNodeID &ID) const {
1205   static int tag = 0;
1206   ID.AddPointer(&tag);
1207   ID.AddPointer(R);
1208   ID.Add(V);
1209   ID.AddInteger(static_cast<int>(TKind));
1210   ID.AddBoolean(EnableNullFPSuppression);
1211 }
1212 
1213 /// Returns true if \p N represents the DeclStmt declaring and initializing
1214 /// \p VR.
isInitializationOfVar(const ExplodedNode * N,const VarRegion * VR)1215 static bool isInitializationOfVar(const ExplodedNode *N, const VarRegion *VR) {
1216   Optional<PostStmt> P = N->getLocationAs<PostStmt>();
1217   if (!P)
1218     return false;
1219 
1220   const DeclStmt *DS = P->getStmtAs<DeclStmt>();
1221   if (!DS)
1222     return false;
1223 
1224   if (DS->getSingleDecl() != VR->getDecl())
1225     return false;
1226 
1227   const MemSpaceRegion *VarSpace = VR->getMemorySpace();
1228   const auto *FrameSpace = dyn_cast<StackSpaceRegion>(VarSpace);
1229   if (!FrameSpace) {
1230     // If we ever directly evaluate global DeclStmts, this assertion will be
1231     // invalid, but this still seems preferable to silently accepting an
1232     // initialization that may be for a path-sensitive variable.
1233     assert(VR->getDecl()->isStaticLocal() && "non-static stackless VarRegion");
1234     return true;
1235   }
1236 
1237   assert(VR->getDecl()->hasLocalStorage());
1238   const LocationContext *LCtx = N->getLocationContext();
1239   return FrameSpace->getStackFrame() == LCtx->getStackFrame();
1240 }
1241 
1242 /// Show diagnostics for initializing or declaring a region \p R with a bad value.
showBRDiagnostics(const char * action,llvm::raw_svector_ostream & os,const MemRegion * R,SVal V,const DeclStmt * DS)1243 static void showBRDiagnostics(const char *action, llvm::raw_svector_ostream &os,
1244                               const MemRegion *R, SVal V, const DeclStmt *DS) {
1245   if (R->canPrintPretty()) {
1246     R->printPretty(os);
1247     os << " ";
1248   }
1249 
1250   if (V.getAs<loc::ConcreteInt>()) {
1251     bool b = false;
1252     if (R->isBoundable()) {
1253       if (const auto *TR = dyn_cast<TypedValueRegion>(R)) {
1254         if (TR->getValueType()->isObjCObjectPointerType()) {
1255           os << action << "nil";
1256           b = true;
1257         }
1258       }
1259     }
1260     if (!b)
1261       os << action << "a null pointer value";
1262 
1263   } else if (auto CVal = V.getAs<nonloc::ConcreteInt>()) {
1264     os << action << CVal->getValue();
1265   } else if (DS) {
1266     if (V.isUndef()) {
1267       if (isa<VarRegion>(R)) {
1268         const auto *VD = cast<VarDecl>(DS->getSingleDecl());
1269         if (VD->getInit()) {
1270           os << (R->canPrintPretty() ? "initialized" : "Initializing")
1271             << " to a garbage value";
1272         } else {
1273           os << (R->canPrintPretty() ? "declared" : "Declaring")
1274             << " without an initial value";
1275         }
1276       }
1277     } else {
1278       os << (R->canPrintPretty() ? "initialized" : "Initialized")
1279         << " here";
1280     }
1281   }
1282 }
1283 
1284 /// Display diagnostics for passing bad region as a parameter.
showBRParamDiagnostics(llvm::raw_svector_ostream & os,const VarRegion * VR,SVal V)1285 static void showBRParamDiagnostics(llvm::raw_svector_ostream& os,
1286     const VarRegion *VR,
1287     SVal V) {
1288   const auto *Param = cast<ParmVarDecl>(VR->getDecl());
1289 
1290   os << "Passing ";
1291 
1292   if (V.getAs<loc::ConcreteInt>()) {
1293     if (Param->getType()->isObjCObjectPointerType())
1294       os << "nil object reference";
1295     else
1296       os << "null pointer value";
1297   } else if (V.isUndef()) {
1298     os << "uninitialized value";
1299   } else if (auto CI = V.getAs<nonloc::ConcreteInt>()) {
1300     os << "the value " << CI->getValue();
1301   } else {
1302     os << "value";
1303   }
1304 
1305   // Printed parameter indexes are 1-based, not 0-based.
1306   unsigned Idx = Param->getFunctionScopeIndex() + 1;
1307   os << " via " << Idx << llvm::getOrdinalSuffix(Idx) << " parameter";
1308   if (VR->canPrintPretty()) {
1309     os << " ";
1310     VR->printPretty(os);
1311   }
1312 }
1313 
1314 /// Show default diagnostics for storing bad region.
showBRDefaultDiagnostics(llvm::raw_svector_ostream & os,const MemRegion * R,SVal V)1315 static void showBRDefaultDiagnostics(llvm::raw_svector_ostream &os,
1316                                      const MemRegion *R, SVal V) {
1317   if (V.getAs<loc::ConcreteInt>()) {
1318     bool b = false;
1319     if (R->isBoundable()) {
1320       if (const auto *TR = dyn_cast<TypedValueRegion>(R)) {
1321         if (TR->getValueType()->isObjCObjectPointerType()) {
1322           os << "nil object reference stored";
1323           b = true;
1324         }
1325       }
1326     }
1327     if (!b) {
1328       if (R->canPrintPretty())
1329         os << "Null pointer value stored";
1330       else
1331         os << "Storing null pointer value";
1332     }
1333 
1334   } else if (V.isUndef()) {
1335     if (R->canPrintPretty())
1336       os << "Uninitialized value stored";
1337     else
1338       os << "Storing uninitialized value";
1339 
1340   } else if (auto CV = V.getAs<nonloc::ConcreteInt>()) {
1341     if (R->canPrintPretty())
1342       os << "The value " << CV->getValue() << " is assigned";
1343     else
1344       os << "Assigning " << CV->getValue();
1345 
1346   } else {
1347     if (R->canPrintPretty())
1348       os << "Value assigned";
1349     else
1350       os << "Assigning value";
1351   }
1352 
1353   if (R->canPrintPretty()) {
1354     os << " to ";
1355     R->printPretty(os);
1356   }
1357 }
1358 
1359 PathDiagnosticPieceRef
VisitNode(const ExplodedNode * Succ,BugReporterContext & BRC,PathSensitiveBugReport & BR)1360 FindLastStoreBRVisitor::VisitNode(const ExplodedNode *Succ,
1361                                   BugReporterContext &BRC,
1362                                   PathSensitiveBugReport &BR) {
1363   if (Satisfied)
1364     return nullptr;
1365 
1366   const ExplodedNode *StoreSite = nullptr;
1367   const ExplodedNode *Pred = Succ->getFirstPred();
1368   const Expr *InitE = nullptr;
1369   bool IsParam = false;
1370 
1371   // First see if we reached the declaration of the region.
1372   if (const auto *VR = dyn_cast<VarRegion>(R)) {
1373     if (isInitializationOfVar(Pred, VR)) {
1374       StoreSite = Pred;
1375       InitE = VR->getDecl()->getInit();
1376     }
1377   }
1378 
1379   // If this is a post initializer expression, initializing the region, we
1380   // should track the initializer expression.
1381   if (Optional<PostInitializer> PIP = Pred->getLocationAs<PostInitializer>()) {
1382     const MemRegion *FieldReg = (const MemRegion *)PIP->getLocationValue();
1383     if (FieldReg == R) {
1384       StoreSite = Pred;
1385       InitE = PIP->getInitializer()->getInit();
1386     }
1387   }
1388 
1389   // Otherwise, see if this is the store site:
1390   // (1) Succ has this binding and Pred does not, i.e. this is
1391   //     where the binding first occurred.
1392   // (2) Succ has this binding and is a PostStore node for this region, i.e.
1393   //     the same binding was re-assigned here.
1394   if (!StoreSite) {
1395     if (Succ->getState()->getSVal(R) != V)
1396       return nullptr;
1397 
1398     if (hasVisibleUpdate(Pred, Pred->getState()->getSVal(R), Succ, V)) {
1399       Optional<PostStore> PS = Succ->getLocationAs<PostStore>();
1400       if (!PS || PS->getLocationValue() != R)
1401         return nullptr;
1402     }
1403 
1404     StoreSite = Succ;
1405 
1406     // If this is an assignment expression, we can track the value
1407     // being assigned.
1408     if (Optional<PostStmt> P = Succ->getLocationAs<PostStmt>())
1409       if (const BinaryOperator *BO = P->getStmtAs<BinaryOperator>())
1410         if (BO->isAssignmentOp())
1411           InitE = BO->getRHS();
1412 
1413     // If this is a call entry, the variable should be a parameter.
1414     // FIXME: Handle CXXThisRegion as well. (This is not a priority because
1415     // 'this' should never be NULL, but this visitor isn't just for NULL and
1416     // UndefinedVal.)
1417     if (Optional<CallEnter> CE = Succ->getLocationAs<CallEnter>()) {
1418       if (const auto *VR = dyn_cast<VarRegion>(R)) {
1419 
1420         if (const auto *Param = dyn_cast<ParmVarDecl>(VR->getDecl())) {
1421           ProgramStateManager &StateMgr = BRC.getStateManager();
1422           CallEventManager &CallMgr = StateMgr.getCallEventManager();
1423 
1424           CallEventRef<> Call = CallMgr.getCaller(CE->getCalleeContext(),
1425                                                   Succ->getState());
1426           InitE = Call->getArgExpr(Param->getFunctionScopeIndex());
1427         } else {
1428           // Handle Objective-C 'self'.
1429           assert(isa<ImplicitParamDecl>(VR->getDecl()));
1430           InitE = cast<ObjCMessageExpr>(CE->getCalleeContext()->getCallSite())
1431                       ->getInstanceReceiver()->IgnoreParenCasts();
1432         }
1433         IsParam = true;
1434       }
1435     }
1436 
1437     // If this is a CXXTempObjectRegion, the Expr responsible for its creation
1438     // is wrapped inside of it.
1439     if (const auto *TmpR = dyn_cast<CXXTempObjectRegion>(R))
1440       InitE = TmpR->getExpr();
1441   }
1442 
1443   if (!StoreSite)
1444     return nullptr;
1445 
1446   Satisfied = true;
1447 
1448   // If we have an expression that provided the value, try to track where it
1449   // came from.
1450   if (InitE) {
1451     if (!IsParam)
1452       InitE = InitE->IgnoreParenCasts();
1453 
1454     bugreporter::trackExpressionValue(
1455         StoreSite, InitE, BR, TKind, EnableNullFPSuppression);
1456   }
1457 
1458   if (TKind == TrackingKind::Condition &&
1459       !OriginSFC->isParentOf(StoreSite->getStackFrame()))
1460     return nullptr;
1461 
1462   // Okay, we've found the binding. Emit an appropriate message.
1463   SmallString<256> sbuf;
1464   llvm::raw_svector_ostream os(sbuf);
1465 
1466   if (Optional<PostStmt> PS = StoreSite->getLocationAs<PostStmt>()) {
1467     const Stmt *S = PS->getStmt();
1468     const char *action = nullptr;
1469     const auto *DS = dyn_cast<DeclStmt>(S);
1470     const auto *VR = dyn_cast<VarRegion>(R);
1471 
1472     if (DS) {
1473       action = R->canPrintPretty() ? "initialized to " :
1474                                      "Initializing to ";
1475     } else if (isa<BlockExpr>(S)) {
1476       action = R->canPrintPretty() ? "captured by block as " :
1477                                      "Captured by block as ";
1478       if (VR) {
1479         // See if we can get the BlockVarRegion.
1480         ProgramStateRef State = StoreSite->getState();
1481         SVal V = StoreSite->getSVal(S);
1482         if (const auto *BDR =
1483               dyn_cast_or_null<BlockDataRegion>(V.getAsRegion())) {
1484           if (const VarRegion *OriginalR = BDR->getOriginalRegion(VR)) {
1485             if (auto KV = State->getSVal(OriginalR).getAs<KnownSVal>())
1486               BR.addVisitor(std::make_unique<FindLastStoreBRVisitor>(
1487                   *KV, OriginalR, EnableNullFPSuppression, TKind, OriginSFC));
1488           }
1489         }
1490       }
1491     }
1492     if (action)
1493       showBRDiagnostics(action, os, R, V, DS);
1494 
1495   } else if (StoreSite->getLocation().getAs<CallEnter>()) {
1496     if (const auto *VR = dyn_cast<VarRegion>(R))
1497       showBRParamDiagnostics(os, VR, V);
1498   }
1499 
1500   if (os.str().empty())
1501     showBRDefaultDiagnostics(os, R, V);
1502 
1503   if (TKind == bugreporter::TrackingKind::Condition)
1504     os << WillBeUsedForACondition;
1505 
1506   // Construct a new PathDiagnosticPiece.
1507   ProgramPoint P = StoreSite->getLocation();
1508   PathDiagnosticLocation L;
1509   if (P.getAs<CallEnter>() && InitE)
1510     L = PathDiagnosticLocation(InitE, BRC.getSourceManager(),
1511                                P.getLocationContext());
1512 
1513   if (!L.isValid() || !L.asLocation().isValid())
1514     L = PathDiagnosticLocation::create(P, BRC.getSourceManager());
1515 
1516   if (!L.isValid() || !L.asLocation().isValid())
1517     return nullptr;
1518 
1519   return std::make_shared<PathDiagnosticEventPiece>(L, os.str());
1520 }
1521 
1522 //===----------------------------------------------------------------------===//
1523 // Implementation of TrackConstraintBRVisitor.
1524 //===----------------------------------------------------------------------===//
1525 
Profile(llvm::FoldingSetNodeID & ID) const1526 void TrackConstraintBRVisitor::Profile(llvm::FoldingSetNodeID &ID) const {
1527   static int tag = 0;
1528   ID.AddPointer(&tag);
1529   ID.AddBoolean(Assumption);
1530   ID.Add(Constraint);
1531 }
1532 
1533 /// Return the tag associated with this visitor.  This tag will be used
1534 /// to make all PathDiagnosticPieces created by this visitor.
getTag()1535 const char *TrackConstraintBRVisitor::getTag() {
1536   return "TrackConstraintBRVisitor";
1537 }
1538 
isUnderconstrained(const ExplodedNode * N) const1539 bool TrackConstraintBRVisitor::isUnderconstrained(const ExplodedNode *N) const {
1540   if (IsZeroCheck)
1541     return N->getState()->isNull(Constraint).isUnderconstrained();
1542   return (bool)N->getState()->assume(Constraint, !Assumption);
1543 }
1544 
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport &)1545 PathDiagnosticPieceRef TrackConstraintBRVisitor::VisitNode(
1546     const ExplodedNode *N, BugReporterContext &BRC, PathSensitiveBugReport &) {
1547   const ExplodedNode *PrevN = N->getFirstPred();
1548   if (IsSatisfied)
1549     return nullptr;
1550 
1551   // Start tracking after we see the first state in which the value is
1552   // constrained.
1553   if (!IsTrackingTurnedOn)
1554     if (!isUnderconstrained(N))
1555       IsTrackingTurnedOn = true;
1556   if (!IsTrackingTurnedOn)
1557     return nullptr;
1558 
1559   // Check if in the previous state it was feasible for this constraint
1560   // to *not* be true.
1561   if (isUnderconstrained(PrevN)) {
1562     IsSatisfied = true;
1563 
1564     // As a sanity check, make sure that the negation of the constraint
1565     // was infeasible in the current state.  If it is feasible, we somehow
1566     // missed the transition point.
1567     assert(!isUnderconstrained(N));
1568 
1569     // We found the transition point for the constraint.  We now need to
1570     // pretty-print the constraint. (work-in-progress)
1571     SmallString<64> sbuf;
1572     llvm::raw_svector_ostream os(sbuf);
1573 
1574     if (Constraint.getAs<Loc>()) {
1575       os << "Assuming pointer value is ";
1576       os << (Assumption ? "non-null" : "null");
1577     }
1578 
1579     if (os.str().empty())
1580       return nullptr;
1581 
1582     // Construct a new PathDiagnosticPiece.
1583     ProgramPoint P = N->getLocation();
1584     PathDiagnosticLocation L =
1585       PathDiagnosticLocation::create(P, BRC.getSourceManager());
1586     if (!L.isValid())
1587       return nullptr;
1588 
1589     auto X = std::make_shared<PathDiagnosticEventPiece>(L, os.str());
1590     X->setTag(getTag());
1591     return std::move(X);
1592   }
1593 
1594   return nullptr;
1595 }
1596 
1597 //===----------------------------------------------------------------------===//
1598 // Implementation of SuppressInlineDefensiveChecksVisitor.
1599 //===----------------------------------------------------------------------===//
1600 
1601 SuppressInlineDefensiveChecksVisitor::
SuppressInlineDefensiveChecksVisitor(DefinedSVal Value,const ExplodedNode * N)1602 SuppressInlineDefensiveChecksVisitor(DefinedSVal Value, const ExplodedNode *N)
1603     : V(Value) {
1604   // Check if the visitor is disabled.
1605   AnalyzerOptions &Options = N->getState()->getAnalysisManager().options;
1606   if (!Options.ShouldSuppressInlinedDefensiveChecks)
1607     IsSatisfied = true;
1608 }
1609 
Profile(llvm::FoldingSetNodeID & ID) const1610 void SuppressInlineDefensiveChecksVisitor::Profile(
1611     llvm::FoldingSetNodeID &ID) const {
1612   static int id = 0;
1613   ID.AddPointer(&id);
1614   ID.Add(V);
1615 }
1616 
getTag()1617 const char *SuppressInlineDefensiveChecksVisitor::getTag() {
1618   return "IDCVisitor";
1619 }
1620 
1621 PathDiagnosticPieceRef
VisitNode(const ExplodedNode * Succ,BugReporterContext & BRC,PathSensitiveBugReport & BR)1622 SuppressInlineDefensiveChecksVisitor::VisitNode(const ExplodedNode *Succ,
1623                                                 BugReporterContext &BRC,
1624                                                 PathSensitiveBugReport &BR) {
1625   const ExplodedNode *Pred = Succ->getFirstPred();
1626   if (IsSatisfied)
1627     return nullptr;
1628 
1629   // Start tracking after we see the first state in which the value is null.
1630   if (!IsTrackingTurnedOn)
1631     if (Succ->getState()->isNull(V).isConstrainedTrue())
1632       IsTrackingTurnedOn = true;
1633   if (!IsTrackingTurnedOn)
1634     return nullptr;
1635 
1636   // Check if in the previous state it was feasible for this value
1637   // to *not* be null.
1638   if (!Pred->getState()->isNull(V).isConstrainedTrue() &&
1639       Succ->getState()->isNull(V).isConstrainedTrue()) {
1640     IsSatisfied = true;
1641 
1642     // Check if this is inlined defensive checks.
1643     const LocationContext *CurLC = Succ->getLocationContext();
1644     const LocationContext *ReportLC = BR.getErrorNode()->getLocationContext();
1645     if (CurLC != ReportLC && !CurLC->isParentOf(ReportLC)) {
1646       BR.markInvalid("Suppress IDC", CurLC);
1647       return nullptr;
1648     }
1649 
1650     // Treat defensive checks in function-like macros as if they were an inlined
1651     // defensive check. If the bug location is not in a macro and the
1652     // terminator for the current location is in a macro then suppress the
1653     // warning.
1654     auto BugPoint = BR.getErrorNode()->getLocation().getAs<StmtPoint>();
1655 
1656     if (!BugPoint)
1657       return nullptr;
1658 
1659     ProgramPoint CurPoint = Succ->getLocation();
1660     const Stmt *CurTerminatorStmt = nullptr;
1661     if (auto BE = CurPoint.getAs<BlockEdge>()) {
1662       CurTerminatorStmt = BE->getSrc()->getTerminator().getStmt();
1663     } else if (auto SP = CurPoint.getAs<StmtPoint>()) {
1664       const Stmt *CurStmt = SP->getStmt();
1665       if (!CurStmt->getBeginLoc().isMacroID())
1666         return nullptr;
1667 
1668       CFGStmtMap *Map = CurLC->getAnalysisDeclContext()->getCFGStmtMap();
1669       CurTerminatorStmt = Map->getBlock(CurStmt)->getTerminatorStmt();
1670     } else {
1671       return nullptr;
1672     }
1673 
1674     if (!CurTerminatorStmt)
1675       return nullptr;
1676 
1677     SourceLocation TerminatorLoc = CurTerminatorStmt->getBeginLoc();
1678     if (TerminatorLoc.isMacroID()) {
1679       SourceLocation BugLoc = BugPoint->getStmt()->getBeginLoc();
1680 
1681       // Suppress reports unless we are in that same macro.
1682       if (!BugLoc.isMacroID() ||
1683           getMacroName(BugLoc, BRC) != getMacroName(TerminatorLoc, BRC)) {
1684         BR.markInvalid("Suppress Macro IDC", CurLC);
1685       }
1686       return nullptr;
1687     }
1688   }
1689   return nullptr;
1690 }
1691 
1692 //===----------------------------------------------------------------------===//
1693 // TrackControlDependencyCondBRVisitor.
1694 //===----------------------------------------------------------------------===//
1695 
1696 namespace {
1697 /// Tracks the expressions that are a control dependency of the node that was
1698 /// supplied to the constructor.
1699 /// For example:
1700 ///
1701 ///   cond = 1;
1702 ///   if (cond)
1703 ///     10 / 0;
1704 ///
1705 /// An error is emitted at line 3. This visitor realizes that the branch
1706 /// on line 2 is a control dependency of line 3, and tracks it's condition via
1707 /// trackExpressionValue().
1708 class TrackControlDependencyCondBRVisitor final : public BugReporterVisitor {
1709   const ExplodedNode *Origin;
1710   ControlDependencyCalculator ControlDeps;
1711   llvm::SmallSet<const CFGBlock *, 32> VisitedBlocks;
1712 
1713 public:
TrackControlDependencyCondBRVisitor(const ExplodedNode * O)1714   TrackControlDependencyCondBRVisitor(const ExplodedNode *O)
1715   : Origin(O), ControlDeps(&O->getCFG()) {}
1716 
Profile(llvm::FoldingSetNodeID & ID) const1717   void Profile(llvm::FoldingSetNodeID &ID) const override {
1718     static int x = 0;
1719     ID.AddPointer(&x);
1720   }
1721 
1722   PathDiagnosticPieceRef VisitNode(const ExplodedNode *N,
1723                                    BugReporterContext &BRC,
1724                                    PathSensitiveBugReport &BR) override;
1725 };
1726 } // end of anonymous namespace
1727 
1728 static std::shared_ptr<PathDiagnosticEventPiece>
constructDebugPieceForTrackedCondition(const Expr * Cond,const ExplodedNode * N,BugReporterContext & BRC)1729 constructDebugPieceForTrackedCondition(const Expr *Cond,
1730                                        const ExplodedNode *N,
1731                                        BugReporterContext &BRC) {
1732 
1733   if (BRC.getAnalyzerOptions().AnalysisDiagOpt == PD_NONE ||
1734       !BRC.getAnalyzerOptions().ShouldTrackConditionsDebug)
1735     return nullptr;
1736 
1737   std::string ConditionText = std::string(Lexer::getSourceText(
1738       CharSourceRange::getTokenRange(Cond->getSourceRange()),
1739       BRC.getSourceManager(), BRC.getASTContext().getLangOpts()));
1740 
1741   return std::make_shared<PathDiagnosticEventPiece>(
1742       PathDiagnosticLocation::createBegin(
1743           Cond, BRC.getSourceManager(), N->getLocationContext()),
1744           (Twine() + "Tracking condition '" + ConditionText + "'").str());
1745 }
1746 
isAssertlikeBlock(const CFGBlock * B,ASTContext & Context)1747 static bool isAssertlikeBlock(const CFGBlock *B, ASTContext &Context) {
1748   if (B->succ_size() != 2)
1749     return false;
1750 
1751   const CFGBlock *Then = B->succ_begin()->getReachableBlock();
1752   const CFGBlock *Else = (B->succ_begin() + 1)->getReachableBlock();
1753 
1754   if (!Then || !Else)
1755     return false;
1756 
1757   if (Then->isInevitablySinking() != Else->isInevitablySinking())
1758     return true;
1759 
1760   // For the following condition the following CFG would be built:
1761   //
1762   //                          ------------->
1763   //                         /              \
1764   //                       [B1] -> [B2] -> [B3] -> [sink]
1765   // assert(A && B || C);            \       \
1766   //                                  -----------> [go on with the execution]
1767   //
1768   // It so happens that CFGBlock::getTerminatorCondition returns 'A' for block
1769   // B1, 'A && B' for B2, and 'A && B || C' for B3. Let's check whether we
1770   // reached the end of the condition!
1771   if (const Stmt *ElseCond = Else->getTerminatorCondition())
1772     if (const auto *BinOp = dyn_cast<BinaryOperator>(ElseCond))
1773       if (BinOp->isLogicalOp())
1774         return isAssertlikeBlock(Else, Context);
1775 
1776   return false;
1777 }
1778 
1779 PathDiagnosticPieceRef
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)1780 TrackControlDependencyCondBRVisitor::VisitNode(const ExplodedNode *N,
1781                                                BugReporterContext &BRC,
1782                                                PathSensitiveBugReport &BR) {
1783   // We can only reason about control dependencies within the same stack frame.
1784   if (Origin->getStackFrame() != N->getStackFrame())
1785     return nullptr;
1786 
1787   CFGBlock *NB = const_cast<CFGBlock *>(N->getCFGBlock());
1788 
1789   // Skip if we already inspected this block.
1790   if (!VisitedBlocks.insert(NB).second)
1791     return nullptr;
1792 
1793   CFGBlock *OriginB = const_cast<CFGBlock *>(Origin->getCFGBlock());
1794 
1795   // TODO: Cache CFGBlocks for each ExplodedNode.
1796   if (!OriginB || !NB)
1797     return nullptr;
1798 
1799   if (isAssertlikeBlock(NB, BRC.getASTContext()))
1800     return nullptr;
1801 
1802   if (ControlDeps.isControlDependent(OriginB, NB)) {
1803     // We don't really want to explain for range loops. Evidence suggests that
1804     // the only thing that leads to is the addition of calls to operator!=.
1805     if (llvm::isa_and_nonnull<CXXForRangeStmt>(NB->getTerminatorStmt()))
1806       return nullptr;
1807 
1808     if (const Expr *Condition = NB->getLastCondition()) {
1809       // Keeping track of the already tracked conditions on a visitor level
1810       // isn't sufficient, because a new visitor is created for each tracked
1811       // expression, hence the BugReport level set.
1812       if (BR.addTrackedCondition(N)) {
1813         bugreporter::trackExpressionValue(
1814             N, Condition, BR, bugreporter::TrackingKind::Condition,
1815             /*EnableNullFPSuppression=*/false);
1816         return constructDebugPieceForTrackedCondition(Condition, N, BRC);
1817       }
1818     }
1819   }
1820 
1821   return nullptr;
1822 }
1823 
1824 //===----------------------------------------------------------------------===//
1825 // Implementation of trackExpressionValue.
1826 //===----------------------------------------------------------------------===//
1827 
getLocationRegionIfReference(const Expr * E,const ExplodedNode * N)1828 static const MemRegion *getLocationRegionIfReference(const Expr *E,
1829                                                      const ExplodedNode *N) {
1830   if (const auto *DR = dyn_cast<DeclRefExpr>(E)) {
1831     if (const auto *VD = dyn_cast<VarDecl>(DR->getDecl())) {
1832       if (!VD->getType()->isReferenceType())
1833         return nullptr;
1834       ProgramStateManager &StateMgr = N->getState()->getStateManager();
1835       MemRegionManager &MRMgr = StateMgr.getRegionManager();
1836       return MRMgr.getVarRegion(VD, N->getLocationContext());
1837     }
1838   }
1839 
1840   // FIXME: This does not handle other kinds of null references,
1841   // for example, references from FieldRegions:
1842   //   struct Wrapper { int &ref; };
1843   //   Wrapper w = { *(int *)0 };
1844   //   w.ref = 1;
1845 
1846   return nullptr;
1847 }
1848 
1849 /// \return A subexpression of {@code Ex} which represents the
1850 /// expression-of-interest.
peelOffOuterExpr(const Expr * Ex,const ExplodedNode * N)1851 static const Expr *peelOffOuterExpr(const Expr *Ex,
1852                                     const ExplodedNode *N) {
1853   Ex = Ex->IgnoreParenCasts();
1854   if (const auto *FE = dyn_cast<FullExpr>(Ex))
1855     return peelOffOuterExpr(FE->getSubExpr(), N);
1856   if (const auto *OVE = dyn_cast<OpaqueValueExpr>(Ex))
1857     return peelOffOuterExpr(OVE->getSourceExpr(), N);
1858   if (const auto *POE = dyn_cast<PseudoObjectExpr>(Ex)) {
1859     const auto *PropRef = dyn_cast<ObjCPropertyRefExpr>(POE->getSyntacticForm());
1860     if (PropRef && PropRef->isMessagingGetter()) {
1861       const Expr *GetterMessageSend =
1862           POE->getSemanticExpr(POE->getNumSemanticExprs() - 1);
1863       assert(isa<ObjCMessageExpr>(GetterMessageSend->IgnoreParenCasts()));
1864       return peelOffOuterExpr(GetterMessageSend, N);
1865     }
1866   }
1867 
1868   // Peel off the ternary operator.
1869   if (const auto *CO = dyn_cast<ConditionalOperator>(Ex)) {
1870     // Find a node where the branching occurred and find out which branch
1871     // we took (true/false) by looking at the ExplodedGraph.
1872     const ExplodedNode *NI = N;
1873     do {
1874       ProgramPoint ProgPoint = NI->getLocation();
1875       if (Optional<BlockEdge> BE = ProgPoint.getAs<BlockEdge>()) {
1876         const CFGBlock *srcBlk = BE->getSrc();
1877         if (const Stmt *term = srcBlk->getTerminatorStmt()) {
1878           if (term == CO) {
1879             bool TookTrueBranch = (*(srcBlk->succ_begin()) == BE->getDst());
1880             if (TookTrueBranch)
1881               return peelOffOuterExpr(CO->getTrueExpr(), N);
1882             else
1883               return peelOffOuterExpr(CO->getFalseExpr(), N);
1884           }
1885         }
1886       }
1887       NI = NI->getFirstPred();
1888     } while (NI);
1889   }
1890 
1891   if (auto *BO = dyn_cast<BinaryOperator>(Ex))
1892     if (const Expr *SubEx = peelOffPointerArithmetic(BO))
1893       return peelOffOuterExpr(SubEx, N);
1894 
1895   if (auto *UO = dyn_cast<UnaryOperator>(Ex)) {
1896     if (UO->getOpcode() == UO_LNot)
1897       return peelOffOuterExpr(UO->getSubExpr(), N);
1898 
1899     // FIXME: There's a hack in our Store implementation that always computes
1900     // field offsets around null pointers as if they are always equal to 0.
1901     // The idea here is to report accesses to fields as null dereferences
1902     // even though the pointer value that's being dereferenced is actually
1903     // the offset of the field rather than exactly 0.
1904     // See the FIXME in StoreManager's getLValueFieldOrIvar() method.
1905     // This code interacts heavily with this hack; otherwise the value
1906     // would not be null at all for most fields, so we'd be unable to track it.
1907     if (UO->getOpcode() == UO_AddrOf && UO->getSubExpr()->isLValue())
1908       if (const Expr *DerefEx = bugreporter::getDerefExpr(UO->getSubExpr()))
1909         return peelOffOuterExpr(DerefEx, N);
1910   }
1911 
1912   return Ex;
1913 }
1914 
1915 /// Find the ExplodedNode where the lvalue (the value of 'Ex')
1916 /// was computed.
findNodeForExpression(const ExplodedNode * N,const Expr * Inner)1917 static const ExplodedNode* findNodeForExpression(const ExplodedNode *N,
1918                                                  const Expr *Inner) {
1919   while (N) {
1920     if (N->getStmtForDiagnostics() == Inner)
1921       return N;
1922     N = N->getFirstPred();
1923   }
1924   return N;
1925 }
1926 
trackExpressionValue(const ExplodedNode * InputNode,const Expr * E,PathSensitiveBugReport & report,bugreporter::TrackingKind TKind,bool EnableNullFPSuppression)1927 bool bugreporter::trackExpressionValue(const ExplodedNode *InputNode,
1928                                        const Expr *E,
1929                                        PathSensitiveBugReport &report,
1930                                        bugreporter::TrackingKind TKind,
1931                                        bool EnableNullFPSuppression) {
1932 
1933   if (!E || !InputNode)
1934     return false;
1935 
1936   const Expr *Inner = peelOffOuterExpr(E, InputNode);
1937   const ExplodedNode *LVNode = findNodeForExpression(InputNode, Inner);
1938   if (!LVNode)
1939     return false;
1940 
1941   ProgramStateRef LVState = LVNode->getState();
1942   const StackFrameContext *SFC = LVNode->getStackFrame();
1943 
1944   // We only track expressions if we believe that they are important. Chances
1945   // are good that control dependencies to the tracking point are also improtant
1946   // because of this, let's explain why we believe control reached this point.
1947   // TODO: Shouldn't we track control dependencies of every bug location, rather
1948   // than only tracked expressions?
1949   if (LVState->getAnalysisManager().getAnalyzerOptions().ShouldTrackConditions)
1950     report.addVisitor(std::make_unique<TrackControlDependencyCondBRVisitor>(
1951           InputNode));
1952 
1953   // The message send could be nil due to the receiver being nil.
1954   // At this point in the path, the receiver should be live since we are at the
1955   // message send expr. If it is nil, start tracking it.
1956   if (const Expr *Receiver = NilReceiverBRVisitor::getNilReceiver(Inner, LVNode))
1957     trackExpressionValue(
1958         LVNode, Receiver, report, TKind, EnableNullFPSuppression);
1959 
1960   // Track the index if this is an array subscript.
1961   if (const auto *Arr = dyn_cast<ArraySubscriptExpr>(Inner))
1962     trackExpressionValue(
1963         LVNode, Arr->getIdx(), report, TKind, /*EnableNullFPSuppression*/false);
1964 
1965   // See if the expression we're interested refers to a variable.
1966   // If so, we can track both its contents and constraints on its value.
1967   if (ExplodedGraph::isInterestingLValueExpr(Inner)) {
1968     SVal LVal = LVNode->getSVal(Inner);
1969 
1970     const MemRegion *RR = getLocationRegionIfReference(Inner, LVNode);
1971     bool LVIsNull = LVState->isNull(LVal).isConstrainedTrue();
1972 
1973     // If this is a C++ reference to a null pointer, we are tracking the
1974     // pointer. In addition, we should find the store at which the reference
1975     // got initialized.
1976     if (RR && !LVIsNull)
1977       if (auto KV = LVal.getAs<KnownSVal>())
1978         report.addVisitor(std::make_unique<FindLastStoreBRVisitor>(
1979             *KV, RR, EnableNullFPSuppression, TKind, SFC));
1980 
1981     // In case of C++ references, we want to differentiate between a null
1982     // reference and reference to null pointer.
1983     // If the LVal is null, check if we are dealing with null reference.
1984     // For those, we want to track the location of the reference.
1985     const MemRegion *R = (RR && LVIsNull) ? RR :
1986         LVNode->getSVal(Inner).getAsRegion();
1987 
1988     if (R) {
1989 
1990       // Mark both the variable region and its contents as interesting.
1991       SVal V = LVState->getRawSVal(loc::MemRegionVal(R));
1992       report.addVisitor(
1993           std::make_unique<NoStoreFuncVisitor>(cast<SubRegion>(R), TKind));
1994 
1995       MacroNullReturnSuppressionVisitor::addMacroVisitorIfNecessary(
1996           LVNode, R, EnableNullFPSuppression, report, V);
1997 
1998       report.markInteresting(V, TKind);
1999       report.addVisitor(std::make_unique<UndefOrNullArgVisitor>(R));
2000 
2001       // If the contents are symbolic and null, find out when they became null.
2002       if (V.getAsLocSymbol(/*IncludeBaseRegions=*/true))
2003         if (LVState->isNull(V).isConstrainedTrue())
2004           report.addVisitor(std::make_unique<TrackConstraintBRVisitor>(
2005               V.castAs<DefinedSVal>(), false));
2006 
2007       // Add visitor, which will suppress inline defensive checks.
2008       if (auto DV = V.getAs<DefinedSVal>())
2009         if (!DV->isZeroConstant() && EnableNullFPSuppression) {
2010           // Note that LVNode may be too late (i.e., too far from the InputNode)
2011           // because the lvalue may have been computed before the inlined call
2012           // was evaluated. InputNode may as well be too early here, because
2013           // the symbol is already dead; this, however, is fine because we can
2014           // still find the node in which it collapsed to null previously.
2015           report.addVisitor(
2016               std::make_unique<SuppressInlineDefensiveChecksVisitor>(
2017                   *DV, InputNode));
2018         }
2019 
2020       if (auto KV = V.getAs<KnownSVal>())
2021         report.addVisitor(std::make_unique<FindLastStoreBRVisitor>(
2022             *KV, R, EnableNullFPSuppression, TKind, SFC));
2023       return true;
2024     }
2025   }
2026 
2027   // If the expression is not an "lvalue expression", we can still
2028   // track the constraints on its contents.
2029   SVal V = LVState->getSValAsScalarOrLoc(Inner, LVNode->getLocationContext());
2030 
2031   ReturnVisitor::addVisitorIfNecessary(
2032     LVNode, Inner, report, EnableNullFPSuppression, TKind);
2033 
2034   // Is it a symbolic value?
2035   if (auto L = V.getAs<loc::MemRegionVal>()) {
2036     // FIXME: this is a hack for fixing a later crash when attempting to
2037     // dereference a void* pointer.
2038     // We should not try to dereference pointers at all when we don't care
2039     // what is written inside the pointer.
2040     bool CanDereference = true;
2041     if (const auto *SR = L->getRegionAs<SymbolicRegion>()) {
2042       if (SR->getSymbol()->getType()->getPointeeType()->isVoidType())
2043         CanDereference = false;
2044     } else if (L->getRegionAs<AllocaRegion>())
2045       CanDereference = false;
2046 
2047     // At this point we are dealing with the region's LValue.
2048     // However, if the rvalue is a symbolic region, we should track it as well.
2049     // Try to use the correct type when looking up the value.
2050     SVal RVal;
2051     if (ExplodedGraph::isInterestingLValueExpr(Inner))
2052       RVal = LVState->getRawSVal(L.getValue(), Inner->getType());
2053     else if (CanDereference)
2054       RVal = LVState->getSVal(L->getRegion());
2055 
2056     if (CanDereference) {
2057       report.addVisitor(
2058           std::make_unique<UndefOrNullArgVisitor>(L->getRegion()));
2059 
2060       if (auto KV = RVal.getAs<KnownSVal>())
2061         report.addVisitor(std::make_unique<FindLastStoreBRVisitor>(
2062             *KV, L->getRegion(), EnableNullFPSuppression, TKind, SFC));
2063     }
2064 
2065     const MemRegion *RegionRVal = RVal.getAsRegion();
2066     if (RegionRVal && isa<SymbolicRegion>(RegionRVal)) {
2067       report.markInteresting(RegionRVal, TKind);
2068       report.addVisitor(std::make_unique<TrackConstraintBRVisitor>(
2069             loc::MemRegionVal(RegionRVal), /*assumption=*/false));
2070     }
2071   }
2072   return true;
2073 }
2074 
2075 //===----------------------------------------------------------------------===//
2076 // Implementation of NulReceiverBRVisitor.
2077 //===----------------------------------------------------------------------===//
2078 
getNilReceiver(const Stmt * S,const ExplodedNode * N)2079 const Expr *NilReceiverBRVisitor::getNilReceiver(const Stmt *S,
2080                                                  const ExplodedNode *N) {
2081   const auto *ME = dyn_cast<ObjCMessageExpr>(S);
2082   if (!ME)
2083     return nullptr;
2084   if (const Expr *Receiver = ME->getInstanceReceiver()) {
2085     ProgramStateRef state = N->getState();
2086     SVal V = N->getSVal(Receiver);
2087     if (state->isNull(V).isConstrainedTrue())
2088       return Receiver;
2089   }
2090   return nullptr;
2091 }
2092 
2093 PathDiagnosticPieceRef
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)2094 NilReceiverBRVisitor::VisitNode(const ExplodedNode *N, BugReporterContext &BRC,
2095                                 PathSensitiveBugReport &BR) {
2096   Optional<PreStmt> P = N->getLocationAs<PreStmt>();
2097   if (!P)
2098     return nullptr;
2099 
2100   const Stmt *S = P->getStmt();
2101   const Expr *Receiver = getNilReceiver(S, N);
2102   if (!Receiver)
2103     return nullptr;
2104 
2105   llvm::SmallString<256> Buf;
2106   llvm::raw_svector_ostream OS(Buf);
2107 
2108   if (const auto *ME = dyn_cast<ObjCMessageExpr>(S)) {
2109     OS << "'";
2110     ME->getSelector().print(OS);
2111     OS << "' not called";
2112   }
2113   else {
2114     OS << "No method is called";
2115   }
2116   OS << " because the receiver is nil";
2117 
2118   // The receiver was nil, and hence the method was skipped.
2119   // Register a BugReporterVisitor to issue a message telling us how
2120   // the receiver was null.
2121   bugreporter::trackExpressionValue(
2122       N, Receiver, BR, bugreporter::TrackingKind::Thorough,
2123       /*EnableNullFPSuppression*/ false);
2124   // Issue a message saying that the method was skipped.
2125   PathDiagnosticLocation L(Receiver, BRC.getSourceManager(),
2126                                      N->getLocationContext());
2127   return std::make_shared<PathDiagnosticEventPiece>(L, OS.str());
2128 }
2129 
2130 //===----------------------------------------------------------------------===//
2131 // Visitor that tries to report interesting diagnostics from conditions.
2132 //===----------------------------------------------------------------------===//
2133 
2134 /// Return the tag associated with this visitor.  This tag will be used
2135 /// to make all PathDiagnosticPieces created by this visitor.
getTag()2136 const char *ConditionBRVisitor::getTag() { return "ConditionBRVisitor"; }
2137 
2138 PathDiagnosticPieceRef
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)2139 ConditionBRVisitor::VisitNode(const ExplodedNode *N, BugReporterContext &BRC,
2140                               PathSensitiveBugReport &BR) {
2141   auto piece = VisitNodeImpl(N, BRC, BR);
2142   if (piece) {
2143     piece->setTag(getTag());
2144     if (auto *ev = dyn_cast<PathDiagnosticEventPiece>(piece.get()))
2145       ev->setPrunable(true, /* override */ false);
2146   }
2147   return piece;
2148 }
2149 
2150 PathDiagnosticPieceRef
VisitNodeImpl(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)2151 ConditionBRVisitor::VisitNodeImpl(const ExplodedNode *N,
2152                                   BugReporterContext &BRC,
2153                                   PathSensitiveBugReport &BR) {
2154   ProgramPoint ProgPoint = N->getLocation();
2155   const std::pair<const ProgramPointTag *, const ProgramPointTag *> &Tags =
2156       ExprEngine::geteagerlyAssumeBinOpBifurcationTags();
2157 
2158   // If an assumption was made on a branch, it should be caught
2159   // here by looking at the state transition.
2160   if (Optional<BlockEdge> BE = ProgPoint.getAs<BlockEdge>()) {
2161     const CFGBlock *SrcBlock = BE->getSrc();
2162     if (const Stmt *Term = SrcBlock->getTerminatorStmt()) {
2163       // If the tag of the previous node is 'Eagerly Assume...' the current
2164       // 'BlockEdge' has the same constraint information. We do not want to
2165       // report the value as it is just an assumption on the predecessor node
2166       // which will be caught in the next VisitNode() iteration as a 'PostStmt'.
2167       const ProgramPointTag *PreviousNodeTag =
2168           N->getFirstPred()->getLocation().getTag();
2169       if (PreviousNodeTag == Tags.first || PreviousNodeTag == Tags.second)
2170         return nullptr;
2171 
2172       return VisitTerminator(Term, N, SrcBlock, BE->getDst(), BR, BRC);
2173     }
2174     return nullptr;
2175   }
2176 
2177   if (Optional<PostStmt> PS = ProgPoint.getAs<PostStmt>()) {
2178     const ProgramPointTag *CurrentNodeTag = PS->getTag();
2179     if (CurrentNodeTag != Tags.first && CurrentNodeTag != Tags.second)
2180       return nullptr;
2181 
2182     bool TookTrue = CurrentNodeTag == Tags.first;
2183     return VisitTrueTest(cast<Expr>(PS->getStmt()), BRC, BR, N, TookTrue);
2184   }
2185 
2186   return nullptr;
2187 }
2188 
VisitTerminator(const Stmt * Term,const ExplodedNode * N,const CFGBlock * srcBlk,const CFGBlock * dstBlk,PathSensitiveBugReport & R,BugReporterContext & BRC)2189 PathDiagnosticPieceRef ConditionBRVisitor::VisitTerminator(
2190     const Stmt *Term, const ExplodedNode *N, const CFGBlock *srcBlk,
2191     const CFGBlock *dstBlk, PathSensitiveBugReport &R,
2192     BugReporterContext &BRC) {
2193   const Expr *Cond = nullptr;
2194 
2195   // In the code below, Term is a CFG terminator and Cond is a branch condition
2196   // expression upon which the decision is made on this terminator.
2197   //
2198   // For example, in "if (x == 0)", the "if (x == 0)" statement is a terminator,
2199   // and "x == 0" is the respective condition.
2200   //
2201   // Another example: in "if (x && y)", we've got two terminators and two
2202   // conditions due to short-circuit nature of operator "&&":
2203   // 1. The "if (x && y)" statement is a terminator,
2204   //    and "y" is the respective condition.
2205   // 2. Also "x && ..." is another terminator,
2206   //    and "x" is its condition.
2207 
2208   switch (Term->getStmtClass()) {
2209   // FIXME: Stmt::SwitchStmtClass is worth handling, however it is a bit
2210   // more tricky because there are more than two branches to account for.
2211   default:
2212     return nullptr;
2213   case Stmt::IfStmtClass:
2214     Cond = cast<IfStmt>(Term)->getCond();
2215     break;
2216   case Stmt::ConditionalOperatorClass:
2217     Cond = cast<ConditionalOperator>(Term)->getCond();
2218     break;
2219   case Stmt::BinaryOperatorClass:
2220     // When we encounter a logical operator (&& or ||) as a CFG terminator,
2221     // then the condition is actually its LHS; otherwise, we'd encounter
2222     // the parent, such as if-statement, as a terminator.
2223     const auto *BO = cast<BinaryOperator>(Term);
2224     assert(BO->isLogicalOp() &&
2225            "CFG terminator is not a short-circuit operator!");
2226     Cond = BO->getLHS();
2227     break;
2228   }
2229 
2230   Cond = Cond->IgnoreParens();
2231 
2232   // However, when we encounter a logical operator as a branch condition,
2233   // then the condition is actually its RHS, because LHS would be
2234   // the condition for the logical operator terminator.
2235   while (const auto *InnerBO = dyn_cast<BinaryOperator>(Cond)) {
2236     if (!InnerBO->isLogicalOp())
2237       break;
2238     Cond = InnerBO->getRHS()->IgnoreParens();
2239   }
2240 
2241   assert(Cond);
2242   assert(srcBlk->succ_size() == 2);
2243   const bool TookTrue = *(srcBlk->succ_begin()) == dstBlk;
2244   return VisitTrueTest(Cond, BRC, R, N, TookTrue);
2245 }
2246 
2247 PathDiagnosticPieceRef
VisitTrueTest(const Expr * Cond,BugReporterContext & BRC,PathSensitiveBugReport & R,const ExplodedNode * N,bool TookTrue)2248 ConditionBRVisitor::VisitTrueTest(const Expr *Cond, BugReporterContext &BRC,
2249                                   PathSensitiveBugReport &R,
2250                                   const ExplodedNode *N, bool TookTrue) {
2251   ProgramStateRef CurrentState = N->getState();
2252   ProgramStateRef PrevState = N->getFirstPred()->getState();
2253   const LocationContext *LCtx = N->getLocationContext();
2254 
2255   // If the constraint information is changed between the current and the
2256   // previous program state we assuming the newly seen constraint information.
2257   // If we cannot evaluate the condition (and the constraints are the same)
2258   // the analyzer has no information about the value and just assuming it.
2259   bool IsAssuming =
2260       !BRC.getStateManager().haveEqualConstraints(CurrentState, PrevState) ||
2261       CurrentState->getSVal(Cond, LCtx).isUnknownOrUndef();
2262 
2263   // These will be modified in code below, but we need to preserve the original
2264   //  values in case we want to throw the generic message.
2265   const Expr *CondTmp = Cond;
2266   bool TookTrueTmp = TookTrue;
2267 
2268   while (true) {
2269     CondTmp = CondTmp->IgnoreParenCasts();
2270     switch (CondTmp->getStmtClass()) {
2271       default:
2272         break;
2273       case Stmt::BinaryOperatorClass:
2274         if (auto P = VisitTrueTest(Cond, cast<BinaryOperator>(CondTmp),
2275                                    BRC, R, N, TookTrueTmp, IsAssuming))
2276           return P;
2277         break;
2278       case Stmt::DeclRefExprClass:
2279         if (auto P = VisitTrueTest(Cond, cast<DeclRefExpr>(CondTmp),
2280                                    BRC, R, N, TookTrueTmp, IsAssuming))
2281           return P;
2282         break;
2283       case Stmt::MemberExprClass:
2284         if (auto P = VisitTrueTest(Cond, cast<MemberExpr>(CondTmp),
2285                                    BRC, R, N, TookTrueTmp, IsAssuming))
2286           return P;
2287         break;
2288       case Stmt::UnaryOperatorClass: {
2289         const auto *UO = cast<UnaryOperator>(CondTmp);
2290         if (UO->getOpcode() == UO_LNot) {
2291           TookTrueTmp = !TookTrueTmp;
2292           CondTmp = UO->getSubExpr();
2293           continue;
2294         }
2295         break;
2296       }
2297     }
2298     break;
2299   }
2300 
2301   // Condition too complex to explain? Just say something so that the user
2302   // knew we've made some path decision at this point.
2303   // If it is too complex and we know the evaluation of the condition do not
2304   // repeat the note from 'BugReporter.cpp'
2305   if (!IsAssuming)
2306     return nullptr;
2307 
2308   PathDiagnosticLocation Loc(Cond, BRC.getSourceManager(), LCtx);
2309   if (!Loc.isValid() || !Loc.asLocation().isValid())
2310     return nullptr;
2311 
2312   return std::make_shared<PathDiagnosticEventPiece>(
2313       Loc, TookTrue ? GenericTrueMessage : GenericFalseMessage);
2314 }
2315 
patternMatch(const Expr * Ex,const Expr * ParentEx,raw_ostream & Out,BugReporterContext & BRC,PathSensitiveBugReport & report,const ExplodedNode * N,Optional<bool> & prunable,bool IsSameFieldName)2316 bool ConditionBRVisitor::patternMatch(const Expr *Ex,
2317                                       const Expr *ParentEx,
2318                                       raw_ostream &Out,
2319                                       BugReporterContext &BRC,
2320                                       PathSensitiveBugReport &report,
2321                                       const ExplodedNode *N,
2322                                       Optional<bool> &prunable,
2323                                       bool IsSameFieldName) {
2324   const Expr *OriginalExpr = Ex;
2325   Ex = Ex->IgnoreParenCasts();
2326 
2327   if (isa<GNUNullExpr>(Ex) || isa<ObjCBoolLiteralExpr>(Ex) ||
2328       isa<CXXBoolLiteralExpr>(Ex) || isa<IntegerLiteral>(Ex) ||
2329       isa<FloatingLiteral>(Ex)) {
2330     // Use heuristics to determine if the expression is a macro
2331     // expanding to a literal and if so, use the macro's name.
2332     SourceLocation BeginLoc = OriginalExpr->getBeginLoc();
2333     SourceLocation EndLoc = OriginalExpr->getEndLoc();
2334     if (BeginLoc.isMacroID() && EndLoc.isMacroID()) {
2335       const SourceManager &SM = BRC.getSourceManager();
2336       const LangOptions &LO = BRC.getASTContext().getLangOpts();
2337       if (Lexer::isAtStartOfMacroExpansion(BeginLoc, SM, LO) &&
2338           Lexer::isAtEndOfMacroExpansion(EndLoc, SM, LO)) {
2339         CharSourceRange R = Lexer::getAsCharRange({BeginLoc, EndLoc}, SM, LO);
2340         Out << Lexer::getSourceText(R, SM, LO);
2341         return false;
2342       }
2343     }
2344   }
2345 
2346   if (const auto *DR = dyn_cast<DeclRefExpr>(Ex)) {
2347     const bool quotes = isa<VarDecl>(DR->getDecl());
2348     if (quotes) {
2349       Out << '\'';
2350       const LocationContext *LCtx = N->getLocationContext();
2351       const ProgramState *state = N->getState().get();
2352       if (const MemRegion *R = state->getLValue(cast<VarDecl>(DR->getDecl()),
2353                                                 LCtx).getAsRegion()) {
2354         if (report.isInteresting(R))
2355           prunable = false;
2356         else {
2357           const ProgramState *state = N->getState().get();
2358           SVal V = state->getSVal(R);
2359           if (report.isInteresting(V))
2360             prunable = false;
2361         }
2362       }
2363     }
2364     Out << DR->getDecl()->getDeclName().getAsString();
2365     if (quotes)
2366       Out << '\'';
2367     return quotes;
2368   }
2369 
2370   if (const auto *IL = dyn_cast<IntegerLiteral>(Ex)) {
2371     QualType OriginalTy = OriginalExpr->getType();
2372     if (OriginalTy->isPointerType()) {
2373       if (IL->getValue() == 0) {
2374         Out << "null";
2375         return false;
2376       }
2377     }
2378     else if (OriginalTy->isObjCObjectPointerType()) {
2379       if (IL->getValue() == 0) {
2380         Out << "nil";
2381         return false;
2382       }
2383     }
2384 
2385     Out << IL->getValue();
2386     return false;
2387   }
2388 
2389   if (const auto *ME = dyn_cast<MemberExpr>(Ex)) {
2390     if (!IsSameFieldName)
2391       Out << "field '" << ME->getMemberDecl()->getName() << '\'';
2392     else
2393       Out << '\''
2394           << Lexer::getSourceText(
2395                  CharSourceRange::getTokenRange(Ex->getSourceRange()),
2396                  BRC.getSourceManager(), BRC.getASTContext().getLangOpts(), 0)
2397           << '\'';
2398   }
2399 
2400   return false;
2401 }
2402 
VisitTrueTest(const Expr * Cond,const BinaryOperator * BExpr,BugReporterContext & BRC,PathSensitiveBugReport & R,const ExplodedNode * N,bool TookTrue,bool IsAssuming)2403 PathDiagnosticPieceRef ConditionBRVisitor::VisitTrueTest(
2404     const Expr *Cond, const BinaryOperator *BExpr, BugReporterContext &BRC,
2405     PathSensitiveBugReport &R, const ExplodedNode *N, bool TookTrue,
2406     bool IsAssuming) {
2407   bool shouldInvert = false;
2408   Optional<bool> shouldPrune;
2409 
2410   // Check if the field name of the MemberExprs is ambiguous. Example:
2411   // " 'a.d' is equal to 'h.d' " in 'test/Analysis/null-deref-path-notes.cpp'.
2412   bool IsSameFieldName = false;
2413   const auto *LhsME = dyn_cast<MemberExpr>(BExpr->getLHS()->IgnoreParenCasts());
2414   const auto *RhsME = dyn_cast<MemberExpr>(BExpr->getRHS()->IgnoreParenCasts());
2415 
2416   if (LhsME && RhsME)
2417     IsSameFieldName =
2418         LhsME->getMemberDecl()->getName() == RhsME->getMemberDecl()->getName();
2419 
2420   SmallString<128> LhsString, RhsString;
2421   {
2422     llvm::raw_svector_ostream OutLHS(LhsString), OutRHS(RhsString);
2423     const bool isVarLHS = patternMatch(BExpr->getLHS(), BExpr, OutLHS, BRC, R,
2424                                        N, shouldPrune, IsSameFieldName);
2425     const bool isVarRHS = patternMatch(BExpr->getRHS(), BExpr, OutRHS, BRC, R,
2426                                        N, shouldPrune, IsSameFieldName);
2427 
2428     shouldInvert = !isVarLHS && isVarRHS;
2429   }
2430 
2431   BinaryOperator::Opcode Op = BExpr->getOpcode();
2432 
2433   if (BinaryOperator::isAssignmentOp(Op)) {
2434     // For assignment operators, all that we care about is that the LHS
2435     // evaluates to "true" or "false".
2436     return VisitConditionVariable(LhsString, BExpr->getLHS(), BRC, R, N,
2437                                   TookTrue);
2438   }
2439 
2440   // For non-assignment operations, we require that we can understand
2441   // both the LHS and RHS.
2442   if (LhsString.empty() || RhsString.empty() ||
2443       !BinaryOperator::isComparisonOp(Op) || Op == BO_Cmp)
2444     return nullptr;
2445 
2446   // Should we invert the strings if the LHS is not a variable name?
2447   SmallString<256> buf;
2448   llvm::raw_svector_ostream Out(buf);
2449   Out << (IsAssuming ? "Assuming " : "")
2450       << (shouldInvert ? RhsString : LhsString) << " is ";
2451 
2452   // Do we need to invert the opcode?
2453   if (shouldInvert)
2454     switch (Op) {
2455       default: break;
2456       case BO_LT: Op = BO_GT; break;
2457       case BO_GT: Op = BO_LT; break;
2458       case BO_LE: Op = BO_GE; break;
2459       case BO_GE: Op = BO_LE; break;
2460     }
2461 
2462   if (!TookTrue)
2463     switch (Op) {
2464       case BO_EQ: Op = BO_NE; break;
2465       case BO_NE: Op = BO_EQ; break;
2466       case BO_LT: Op = BO_GE; break;
2467       case BO_GT: Op = BO_LE; break;
2468       case BO_LE: Op = BO_GT; break;
2469       case BO_GE: Op = BO_LT; break;
2470       default:
2471         return nullptr;
2472     }
2473 
2474   switch (Op) {
2475     case BO_EQ:
2476       Out << "equal to ";
2477       break;
2478     case BO_NE:
2479       Out << "not equal to ";
2480       break;
2481     default:
2482       Out << BinaryOperator::getOpcodeStr(Op) << ' ';
2483       break;
2484   }
2485 
2486   Out << (shouldInvert ? LhsString : RhsString);
2487   const LocationContext *LCtx = N->getLocationContext();
2488   const SourceManager &SM = BRC.getSourceManager();
2489 
2490   if (isVarAnInterestingCondition(BExpr->getLHS(), N, &R) ||
2491       isVarAnInterestingCondition(BExpr->getRHS(), N, &R))
2492     Out << WillBeUsedForACondition;
2493 
2494   // Convert 'field ...' to 'Field ...' if it is a MemberExpr.
2495   std::string Message = std::string(Out.str());
2496   Message[0] = toupper(Message[0]);
2497 
2498   // If we know the value create a pop-up note to the value part of 'BExpr'.
2499   if (!IsAssuming) {
2500     PathDiagnosticLocation Loc;
2501     if (!shouldInvert) {
2502       if (LhsME && LhsME->getMemberLoc().isValid())
2503         Loc = PathDiagnosticLocation(LhsME->getMemberLoc(), SM);
2504       else
2505         Loc = PathDiagnosticLocation(BExpr->getLHS(), SM, LCtx);
2506     } else {
2507       if (RhsME && RhsME->getMemberLoc().isValid())
2508         Loc = PathDiagnosticLocation(RhsME->getMemberLoc(), SM);
2509       else
2510         Loc = PathDiagnosticLocation(BExpr->getRHS(), SM, LCtx);
2511     }
2512 
2513     return std::make_shared<PathDiagnosticPopUpPiece>(Loc, Message);
2514   }
2515 
2516   PathDiagnosticLocation Loc(Cond, SM, LCtx);
2517   auto event = std::make_shared<PathDiagnosticEventPiece>(Loc, Message);
2518   if (shouldPrune.hasValue())
2519     event->setPrunable(shouldPrune.getValue());
2520   return event;
2521 }
2522 
VisitConditionVariable(StringRef LhsString,const Expr * CondVarExpr,BugReporterContext & BRC,PathSensitiveBugReport & report,const ExplodedNode * N,bool TookTrue)2523 PathDiagnosticPieceRef ConditionBRVisitor::VisitConditionVariable(
2524     StringRef LhsString, const Expr *CondVarExpr, BugReporterContext &BRC,
2525     PathSensitiveBugReport &report, const ExplodedNode *N, bool TookTrue) {
2526   // FIXME: If there's already a constraint tracker for this variable,
2527   // we shouldn't emit anything here (c.f. the double note in
2528   // test/Analysis/inlining/path-notes.c)
2529   SmallString<256> buf;
2530   llvm::raw_svector_ostream Out(buf);
2531   Out << "Assuming " << LhsString << " is ";
2532 
2533   if (!printValue(CondVarExpr, Out, N, TookTrue, /*IsAssuming=*/true))
2534     return nullptr;
2535 
2536   const LocationContext *LCtx = N->getLocationContext();
2537   PathDiagnosticLocation Loc(CondVarExpr, BRC.getSourceManager(), LCtx);
2538 
2539   if (isVarAnInterestingCondition(CondVarExpr, N, &report))
2540     Out << WillBeUsedForACondition;
2541 
2542   auto event = std::make_shared<PathDiagnosticEventPiece>(Loc, Out.str());
2543 
2544   if (isInterestingExpr(CondVarExpr, N, &report))
2545     event->setPrunable(false);
2546 
2547   return event;
2548 }
2549 
VisitTrueTest(const Expr * Cond,const DeclRefExpr * DRE,BugReporterContext & BRC,PathSensitiveBugReport & report,const ExplodedNode * N,bool TookTrue,bool IsAssuming)2550 PathDiagnosticPieceRef ConditionBRVisitor::VisitTrueTest(
2551     const Expr *Cond, const DeclRefExpr *DRE, BugReporterContext &BRC,
2552     PathSensitiveBugReport &report, const ExplodedNode *N, bool TookTrue,
2553     bool IsAssuming) {
2554   const auto *VD = dyn_cast<VarDecl>(DRE->getDecl());
2555   if (!VD)
2556     return nullptr;
2557 
2558   SmallString<256> Buf;
2559   llvm::raw_svector_ostream Out(Buf);
2560 
2561   Out << (IsAssuming ? "Assuming '" : "'") << VD->getDeclName() << "' is ";
2562 
2563   if (!printValue(DRE, Out, N, TookTrue, IsAssuming))
2564     return nullptr;
2565 
2566   const LocationContext *LCtx = N->getLocationContext();
2567 
2568   if (isVarAnInterestingCondition(DRE, N, &report))
2569     Out << WillBeUsedForACondition;
2570 
2571   // If we know the value create a pop-up note to the 'DRE'.
2572   if (!IsAssuming) {
2573     PathDiagnosticLocation Loc(DRE, BRC.getSourceManager(), LCtx);
2574     return std::make_shared<PathDiagnosticPopUpPiece>(Loc, Out.str());
2575   }
2576 
2577   PathDiagnosticLocation Loc(Cond, BRC.getSourceManager(), LCtx);
2578   auto event = std::make_shared<PathDiagnosticEventPiece>(Loc, Out.str());
2579 
2580   if (isInterestingExpr(DRE, N, &report))
2581     event->setPrunable(false);
2582 
2583   return std::move(event);
2584 }
2585 
VisitTrueTest(const Expr * Cond,const MemberExpr * ME,BugReporterContext & BRC,PathSensitiveBugReport & report,const ExplodedNode * N,bool TookTrue,bool IsAssuming)2586 PathDiagnosticPieceRef ConditionBRVisitor::VisitTrueTest(
2587     const Expr *Cond, const MemberExpr *ME, BugReporterContext &BRC,
2588     PathSensitiveBugReport &report, const ExplodedNode *N, bool TookTrue,
2589     bool IsAssuming) {
2590   SmallString<256> Buf;
2591   llvm::raw_svector_ostream Out(Buf);
2592 
2593   Out << (IsAssuming ? "Assuming field '" : "Field '")
2594       << ME->getMemberDecl()->getName() << "' is ";
2595 
2596   if (!printValue(ME, Out, N, TookTrue, IsAssuming))
2597     return nullptr;
2598 
2599   const LocationContext *LCtx = N->getLocationContext();
2600   PathDiagnosticLocation Loc;
2601 
2602   // If we know the value create a pop-up note to the member of the MemberExpr.
2603   if (!IsAssuming && ME->getMemberLoc().isValid())
2604     Loc = PathDiagnosticLocation(ME->getMemberLoc(), BRC.getSourceManager());
2605   else
2606     Loc = PathDiagnosticLocation(Cond, BRC.getSourceManager(), LCtx);
2607 
2608   if (!Loc.isValid() || !Loc.asLocation().isValid())
2609     return nullptr;
2610 
2611   if (isVarAnInterestingCondition(ME, N, &report))
2612     Out << WillBeUsedForACondition;
2613 
2614   // If we know the value create a pop-up note.
2615   if (!IsAssuming)
2616     return std::make_shared<PathDiagnosticPopUpPiece>(Loc, Out.str());
2617 
2618   auto event = std::make_shared<PathDiagnosticEventPiece>(Loc, Out.str());
2619   if (isInterestingExpr(ME, N, &report))
2620     event->setPrunable(false);
2621   return event;
2622 }
2623 
printValue(const Expr * CondVarExpr,raw_ostream & Out,const ExplodedNode * N,bool TookTrue,bool IsAssuming)2624 bool ConditionBRVisitor::printValue(const Expr *CondVarExpr, raw_ostream &Out,
2625                                     const ExplodedNode *N, bool TookTrue,
2626                                     bool IsAssuming) {
2627   QualType Ty = CondVarExpr->getType();
2628 
2629   if (Ty->isPointerType()) {
2630     Out << (TookTrue ? "non-null" : "null");
2631     return true;
2632   }
2633 
2634   if (Ty->isObjCObjectPointerType()) {
2635     Out << (TookTrue ? "non-nil" : "nil");
2636     return true;
2637   }
2638 
2639   if (!Ty->isIntegralOrEnumerationType())
2640     return false;
2641 
2642   Optional<const llvm::APSInt *> IntValue;
2643   if (!IsAssuming)
2644     IntValue = getConcreteIntegerValue(CondVarExpr, N);
2645 
2646   if (IsAssuming || !IntValue.hasValue()) {
2647     if (Ty->isBooleanType())
2648       Out << (TookTrue ? "true" : "false");
2649     else
2650       Out << (TookTrue ? "not equal to 0" : "0");
2651   } else {
2652     if (Ty->isBooleanType())
2653       Out << (IntValue.getValue()->getBoolValue() ? "true" : "false");
2654     else
2655       Out << *IntValue.getValue();
2656   }
2657 
2658   return true;
2659 }
2660 
2661 constexpr llvm::StringLiteral ConditionBRVisitor::GenericTrueMessage;
2662 constexpr llvm::StringLiteral ConditionBRVisitor::GenericFalseMessage;
2663 
isPieceMessageGeneric(const PathDiagnosticPiece * Piece)2664 bool ConditionBRVisitor::isPieceMessageGeneric(
2665     const PathDiagnosticPiece *Piece) {
2666   return Piece->getString() == GenericTrueMessage ||
2667          Piece->getString() == GenericFalseMessage;
2668 }
2669 
2670 //===----------------------------------------------------------------------===//
2671 // Implementation of LikelyFalsePositiveSuppressionBRVisitor.
2672 //===----------------------------------------------------------------------===//
2673 
finalizeVisitor(BugReporterContext & BRC,const ExplodedNode * N,PathSensitiveBugReport & BR)2674 void LikelyFalsePositiveSuppressionBRVisitor::finalizeVisitor(
2675     BugReporterContext &BRC, const ExplodedNode *N,
2676     PathSensitiveBugReport &BR) {
2677   // Here we suppress false positives coming from system headers. This list is
2678   // based on known issues.
2679   const AnalyzerOptions &Options = BRC.getAnalyzerOptions();
2680   const Decl *D = N->getLocationContext()->getDecl();
2681 
2682   if (AnalysisDeclContext::isInStdNamespace(D)) {
2683     // Skip reports within the 'std' namespace. Although these can sometimes be
2684     // the user's fault, we currently don't report them very well, and
2685     // Note that this will not help for any other data structure libraries, like
2686     // TR1, Boost, or llvm/ADT.
2687     if (Options.ShouldSuppressFromCXXStandardLibrary) {
2688       BR.markInvalid(getTag(), nullptr);
2689       return;
2690     } else {
2691       // If the complete 'std' suppression is not enabled, suppress reports
2692       // from the 'std' namespace that are known to produce false positives.
2693 
2694       // The analyzer issues a false use-after-free when std::list::pop_front
2695       // or std::list::pop_back are called multiple times because we cannot
2696       // reason about the internal invariants of the data structure.
2697       if (const auto *MD = dyn_cast<CXXMethodDecl>(D)) {
2698         const CXXRecordDecl *CD = MD->getParent();
2699         if (CD->getName() == "list") {
2700           BR.markInvalid(getTag(), nullptr);
2701           return;
2702         }
2703       }
2704 
2705       // The analyzer issues a false positive when the constructor of
2706       // std::__independent_bits_engine from algorithms is used.
2707       if (const auto *MD = dyn_cast<CXXConstructorDecl>(D)) {
2708         const CXXRecordDecl *CD = MD->getParent();
2709         if (CD->getName() == "__independent_bits_engine") {
2710           BR.markInvalid(getTag(), nullptr);
2711           return;
2712         }
2713       }
2714 
2715       for (const LocationContext *LCtx = N->getLocationContext(); LCtx;
2716            LCtx = LCtx->getParent()) {
2717         const auto *MD = dyn_cast<CXXMethodDecl>(LCtx->getDecl());
2718         if (!MD)
2719           continue;
2720 
2721         const CXXRecordDecl *CD = MD->getParent();
2722         // The analyzer issues a false positive on
2723         //   std::basic_string<uint8_t> v; v.push_back(1);
2724         // and
2725         //   std::u16string s; s += u'a';
2726         // because we cannot reason about the internal invariants of the
2727         // data structure.
2728         if (CD->getName() == "basic_string") {
2729           BR.markInvalid(getTag(), nullptr);
2730           return;
2731         }
2732 
2733         // The analyzer issues a false positive on
2734         //    std::shared_ptr<int> p(new int(1)); p = nullptr;
2735         // because it does not reason properly about temporary destructors.
2736         if (CD->getName() == "shared_ptr") {
2737           BR.markInvalid(getTag(), nullptr);
2738           return;
2739         }
2740       }
2741     }
2742   }
2743 
2744   // Skip reports within the sys/queue.h macros as we do not have the ability to
2745   // reason about data structure shapes.
2746   const SourceManager &SM = BRC.getSourceManager();
2747   FullSourceLoc Loc = BR.getLocation().asLocation();
2748   while (Loc.isMacroID()) {
2749     Loc = Loc.getSpellingLoc();
2750     if (SM.getFilename(Loc).endswith("sys/queue.h")) {
2751       BR.markInvalid(getTag(), nullptr);
2752       return;
2753     }
2754   }
2755 }
2756 
2757 //===----------------------------------------------------------------------===//
2758 // Implementation of UndefOrNullArgVisitor.
2759 //===----------------------------------------------------------------------===//
2760 
2761 PathDiagnosticPieceRef
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)2762 UndefOrNullArgVisitor::VisitNode(const ExplodedNode *N, BugReporterContext &BRC,
2763                                  PathSensitiveBugReport &BR) {
2764   ProgramStateRef State = N->getState();
2765   ProgramPoint ProgLoc = N->getLocation();
2766 
2767   // We are only interested in visiting CallEnter nodes.
2768   Optional<CallEnter> CEnter = ProgLoc.getAs<CallEnter>();
2769   if (!CEnter)
2770     return nullptr;
2771 
2772   // Check if one of the arguments is the region the visitor is tracking.
2773   CallEventManager &CEMgr = BRC.getStateManager().getCallEventManager();
2774   CallEventRef<> Call = CEMgr.getCaller(CEnter->getCalleeContext(), State);
2775   unsigned Idx = 0;
2776   ArrayRef<ParmVarDecl *> parms = Call->parameters();
2777 
2778   for (const auto ParamDecl : parms) {
2779     const MemRegion *ArgReg = Call->getArgSVal(Idx).getAsRegion();
2780     ++Idx;
2781 
2782     // Are we tracking the argument or its subregion?
2783     if ( !ArgReg || !R->isSubRegionOf(ArgReg->StripCasts()))
2784       continue;
2785 
2786     // Check the function parameter type.
2787     assert(ParamDecl && "Formal parameter has no decl?");
2788     QualType T = ParamDecl->getType();
2789 
2790     if (!(T->isAnyPointerType() || T->isReferenceType())) {
2791       // Function can only change the value passed in by address.
2792       continue;
2793     }
2794 
2795     // If it is a const pointer value, the function does not intend to
2796     // change the value.
2797     if (T->getPointeeType().isConstQualified())
2798       continue;
2799 
2800     // Mark the call site (LocationContext) as interesting if the value of the
2801     // argument is undefined or '0'/'NULL'.
2802     SVal BoundVal = State->getSVal(R);
2803     if (BoundVal.isUndef() || BoundVal.isZeroConstant()) {
2804       BR.markInteresting(CEnter->getCalleeContext());
2805       return nullptr;
2806     }
2807   }
2808   return nullptr;
2809 }
2810 
2811 //===----------------------------------------------------------------------===//
2812 // Implementation of FalsePositiveRefutationBRVisitor.
2813 //===----------------------------------------------------------------------===//
2814 
FalsePositiveRefutationBRVisitor()2815 FalsePositiveRefutationBRVisitor::FalsePositiveRefutationBRVisitor()
2816     : Constraints(ConstraintRangeTy::Factory().getEmptyMap()) {}
2817 
finalizeVisitor(BugReporterContext & BRC,const ExplodedNode * EndPathNode,PathSensitiveBugReport & BR)2818 void FalsePositiveRefutationBRVisitor::finalizeVisitor(
2819     BugReporterContext &BRC, const ExplodedNode *EndPathNode,
2820     PathSensitiveBugReport &BR) {
2821   // Collect new constraints
2822   addConstraints(EndPathNode, /*OverwriteConstraintsOnExistingSyms=*/true);
2823 
2824   // Create a refutation manager
2825   llvm::SMTSolverRef RefutationSolver = llvm::CreateZ3Solver();
2826   ASTContext &Ctx = BRC.getASTContext();
2827 
2828   // Add constraints to the solver
2829   for (const auto &I : Constraints) {
2830     const SymbolRef Sym = I.first;
2831     auto RangeIt = I.second.begin();
2832 
2833     llvm::SMTExprRef SMTConstraints = SMTConv::getRangeExpr(
2834         RefutationSolver, Ctx, Sym, RangeIt->From(), RangeIt->To(),
2835         /*InRange=*/true);
2836     while ((++RangeIt) != I.second.end()) {
2837       SMTConstraints = RefutationSolver->mkOr(
2838           SMTConstraints, SMTConv::getRangeExpr(RefutationSolver, Ctx, Sym,
2839                                                 RangeIt->From(), RangeIt->To(),
2840                                                 /*InRange=*/true));
2841     }
2842 
2843     RefutationSolver->addConstraint(SMTConstraints);
2844   }
2845 
2846   // And check for satisfiability
2847   Optional<bool> IsSAT = RefutationSolver->check();
2848   if (!IsSAT.hasValue())
2849     return;
2850 
2851   if (!IsSAT.getValue())
2852     BR.markInvalid("Infeasible constraints", EndPathNode->getLocationContext());
2853 }
2854 
addConstraints(const ExplodedNode * N,bool OverwriteConstraintsOnExistingSyms)2855 void FalsePositiveRefutationBRVisitor::addConstraints(
2856     const ExplodedNode *N, bool OverwriteConstraintsOnExistingSyms) {
2857   // Collect new constraints
2858   const ConstraintRangeTy &NewCs = N->getState()->get<ConstraintRange>();
2859   ConstraintRangeTy::Factory &CF =
2860       N->getState()->get_context<ConstraintRange>();
2861 
2862   // Add constraints if we don't have them yet
2863   for (auto const &C : NewCs) {
2864     const SymbolRef &Sym = C.first;
2865     if (!Constraints.contains(Sym)) {
2866       // This symbol is new, just add the constraint.
2867       Constraints = CF.add(Constraints, Sym, C.second);
2868     } else if (OverwriteConstraintsOnExistingSyms) {
2869       // Overwrite the associated constraint of the Symbol.
2870       Constraints = CF.remove(Constraints, Sym);
2871       Constraints = CF.add(Constraints, Sym, C.second);
2872     }
2873   }
2874 }
2875 
VisitNode(const ExplodedNode * N,BugReporterContext &,PathSensitiveBugReport &)2876 PathDiagnosticPieceRef FalsePositiveRefutationBRVisitor::VisitNode(
2877     const ExplodedNode *N, BugReporterContext &, PathSensitiveBugReport &) {
2878   addConstraints(N, /*OverwriteConstraintsOnExistingSyms=*/false);
2879   return nullptr;
2880 }
2881 
Profile(llvm::FoldingSetNodeID & ID) const2882 void FalsePositiveRefutationBRVisitor::Profile(
2883     llvm::FoldingSetNodeID &ID) const {
2884   static int Tag = 0;
2885   ID.AddPointer(&Tag);
2886 }
2887 
2888 //===----------------------------------------------------------------------===//
2889 // Implementation of TagVisitor.
2890 //===----------------------------------------------------------------------===//
2891 
2892 int NoteTag::Kind = 0;
2893 
Profile(llvm::FoldingSetNodeID & ID) const2894 void TagVisitor::Profile(llvm::FoldingSetNodeID &ID) const {
2895   static int Tag = 0;
2896   ID.AddPointer(&Tag);
2897 }
2898 
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & R)2899 PathDiagnosticPieceRef TagVisitor::VisitNode(const ExplodedNode *N,
2900                                              BugReporterContext &BRC,
2901                                              PathSensitiveBugReport &R) {
2902   ProgramPoint PP = N->getLocation();
2903   const NoteTag *T = dyn_cast_or_null<NoteTag>(PP.getTag());
2904   if (!T)
2905     return nullptr;
2906 
2907   if (Optional<std::string> Msg = T->generateMessage(BRC, R)) {
2908     PathDiagnosticLocation Loc =
2909         PathDiagnosticLocation::create(PP, BRC.getSourceManager());
2910     auto Piece = std::make_shared<PathDiagnosticEventPiece>(Loc, *Msg);
2911     Piece->setPrunable(T->isPrunable());
2912     return Piece;
2913   }
2914 
2915   return nullptr;
2916 }
2917