xref: /netbsd-src/external/apache2/llvm/dist/clang/lib/StaticAnalyzer/Core/BugReporterVisitors.cpp (revision e038c9c4676b0f19b1b7dd08a940c6ed64a6d5ae)
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 static const MemRegion *
getLocationRegionIfReference(const Expr * E,const ExplodedNode * N,bool LookingForReference=true)157 getLocationRegionIfReference(const Expr *E, const ExplodedNode *N,
158                              bool LookingForReference = true) {
159   if (const auto *DR = dyn_cast<DeclRefExpr>(E)) {
160     if (const auto *VD = dyn_cast<VarDecl>(DR->getDecl())) {
161       if (LookingForReference && !VD->getType()->isReferenceType())
162         return nullptr;
163       return N->getState()
164           ->getLValue(VD, N->getLocationContext())
165           .getAsRegion();
166     }
167   }
168 
169   // FIXME: This does not handle other kinds of null references,
170   // for example, references from FieldRegions:
171   //   struct Wrapper { int &ref; };
172   //   Wrapper w = { *(int *)0 };
173   //   w.ref = 1;
174 
175   return nullptr;
176 }
177 
178 /// Comparing internal representations of symbolic values (via
179 /// SVal::operator==()) is a valid way to check if the value was updated,
180 /// unless it's a LazyCompoundVal that may have a different internal
181 /// representation every time it is loaded from the state. In this function we
182 /// do an approximate comparison for lazy compound values, checking that they
183 /// are the immediate snapshots of the tracked region's bindings within the
184 /// node's respective states but not really checking that these snapshots
185 /// actually contain the same set of bindings.
hasVisibleUpdate(const ExplodedNode * LeftNode,SVal LeftVal,const ExplodedNode * RightNode,SVal RightVal)186 static bool hasVisibleUpdate(const ExplodedNode *LeftNode, SVal LeftVal,
187                              const ExplodedNode *RightNode, SVal RightVal) {
188   if (LeftVal == RightVal)
189     return true;
190 
191   const auto LLCV = LeftVal.getAs<nonloc::LazyCompoundVal>();
192   if (!LLCV)
193     return false;
194 
195   const auto RLCV = RightVal.getAs<nonloc::LazyCompoundVal>();
196   if (!RLCV)
197     return false;
198 
199   return LLCV->getRegion() == RLCV->getRegion() &&
200     LLCV->getStore() == LeftNode->getState()->getStore() &&
201     RLCV->getStore() == RightNode->getState()->getStore();
202 }
203 
getSValForVar(const Expr * CondVarExpr,const ExplodedNode * N)204 static Optional<SVal> getSValForVar(const Expr *CondVarExpr,
205                                     const ExplodedNode *N) {
206   ProgramStateRef State = N->getState();
207   const LocationContext *LCtx = N->getLocationContext();
208 
209   assert(CondVarExpr);
210   CondVarExpr = CondVarExpr->IgnoreImpCasts();
211 
212   // The declaration of the value may rely on a pointer so take its l-value.
213   // FIXME: As seen in VisitCommonDeclRefExpr, sometimes DeclRefExpr may
214   // evaluate to a FieldRegion when it refers to a declaration of a lambda
215   // capture variable. We most likely need to duplicate that logic here.
216   if (const auto *DRE = dyn_cast<DeclRefExpr>(CondVarExpr))
217     if (const auto *VD = dyn_cast<VarDecl>(DRE->getDecl()))
218       return State->getSVal(State->getLValue(VD, LCtx));
219 
220   if (const auto *ME = dyn_cast<MemberExpr>(CondVarExpr))
221     if (const auto *FD = dyn_cast<FieldDecl>(ME->getMemberDecl()))
222       if (auto FieldL = State->getSVal(ME, LCtx).getAs<Loc>())
223         return State->getRawSVal(*FieldL, FD->getType());
224 
225   return None;
226 }
227 
228 static Optional<const llvm::APSInt *>
getConcreteIntegerValue(const Expr * CondVarExpr,const ExplodedNode * N)229 getConcreteIntegerValue(const Expr *CondVarExpr, const ExplodedNode *N) {
230 
231   if (Optional<SVal> V = getSValForVar(CondVarExpr, N))
232     if (auto CI = V->getAs<nonloc::ConcreteInt>())
233       return &CI->getValue();
234   return None;
235 }
236 
isVarAnInterestingCondition(const Expr * CondVarExpr,const ExplodedNode * N,const PathSensitiveBugReport * B)237 static bool isVarAnInterestingCondition(const Expr *CondVarExpr,
238                                         const ExplodedNode *N,
239                                         const PathSensitiveBugReport *B) {
240   // Even if this condition is marked as interesting, it isn't *that*
241   // interesting if it didn't happen in a nested stackframe, the user could just
242   // follow the arrows.
243   if (!B->getErrorNode()->getStackFrame()->isParentOf(N->getStackFrame()))
244     return false;
245 
246   if (Optional<SVal> V = getSValForVar(CondVarExpr, N))
247     if (Optional<bugreporter::TrackingKind> K = B->getInterestingnessKind(*V))
248       return *K == bugreporter::TrackingKind::Condition;
249 
250   return false;
251 }
252 
isInterestingExpr(const Expr * E,const ExplodedNode * N,const PathSensitiveBugReport * B)253 static bool isInterestingExpr(const Expr *E, const ExplodedNode *N,
254                               const PathSensitiveBugReport *B) {
255   if (Optional<SVal> V = getSValForVar(E, N))
256     return B->getInterestingnessKind(*V).hasValue();
257   return false;
258 }
259 
260 /// \return name of the macro inside the location \p Loc.
getMacroName(SourceLocation Loc,BugReporterContext & BRC)261 static StringRef getMacroName(SourceLocation Loc,
262     BugReporterContext &BRC) {
263   return Lexer::getImmediateMacroName(
264       Loc,
265       BRC.getSourceManager(),
266       BRC.getASTContext().getLangOpts());
267 }
268 
269 /// \return Whether given spelling location corresponds to an expansion
270 /// of a function-like macro.
isFunctionMacroExpansion(SourceLocation Loc,const SourceManager & SM)271 static bool isFunctionMacroExpansion(SourceLocation Loc,
272                                 const SourceManager &SM) {
273   if (!Loc.isMacroID())
274     return false;
275   while (SM.isMacroArgExpansion(Loc))
276     Loc = SM.getImmediateExpansionRange(Loc).getBegin();
277   std::pair<FileID, unsigned> TLInfo = SM.getDecomposedLoc(Loc);
278   SrcMgr::SLocEntry SE = SM.getSLocEntry(TLInfo.first);
279   const SrcMgr::ExpansionInfo &EInfo = SE.getExpansion();
280   return EInfo.isFunctionMacroExpansion();
281 }
282 
283 /// \return Whether \c RegionOfInterest was modified at \p N,
284 /// where \p ValueAfter is \c RegionOfInterest's value at the end of the
285 /// stack frame.
wasRegionOfInterestModifiedAt(const SubRegion * RegionOfInterest,const ExplodedNode * N,SVal ValueAfter)286 static bool wasRegionOfInterestModifiedAt(const SubRegion *RegionOfInterest,
287                                           const ExplodedNode *N,
288                                           SVal ValueAfter) {
289   ProgramStateRef State = N->getState();
290   ProgramStateManager &Mgr = N->getState()->getStateManager();
291 
292   if (!N->getLocationAs<PostStore>() && !N->getLocationAs<PostInitializer>() &&
293       !N->getLocationAs<PostStmt>())
294     return false;
295 
296   // Writing into region of interest.
297   if (auto PS = N->getLocationAs<PostStmt>())
298     if (auto *BO = PS->getStmtAs<BinaryOperator>())
299       if (BO->isAssignmentOp() && RegionOfInterest->isSubRegionOf(
300                                       N->getSVal(BO->getLHS()).getAsRegion()))
301         return true;
302 
303   // SVal after the state is possibly different.
304   SVal ValueAtN = N->getState()->getSVal(RegionOfInterest);
305   if (!Mgr.getSValBuilder()
306            .areEqual(State, ValueAtN, ValueAfter)
307            .isConstrainedTrue() &&
308       (!ValueAtN.isUndef() || !ValueAfter.isUndef()))
309     return true;
310 
311   return false;
312 }
313 
314 //===----------------------------------------------------------------------===//
315 // Implementation of BugReporterVisitor.
316 //===----------------------------------------------------------------------===//
317 
getEndPath(BugReporterContext &,const ExplodedNode *,PathSensitiveBugReport &)318 PathDiagnosticPieceRef BugReporterVisitor::getEndPath(BugReporterContext &,
319                                                       const ExplodedNode *,
320                                                       PathSensitiveBugReport &) {
321   return nullptr;
322 }
323 
finalizeVisitor(BugReporterContext &,const ExplodedNode *,PathSensitiveBugReport &)324 void BugReporterVisitor::finalizeVisitor(BugReporterContext &,
325                                          const ExplodedNode *,
326                                          PathSensitiveBugReport &) {}
327 
328 PathDiagnosticPieceRef
getDefaultEndPath(const BugReporterContext & BRC,const ExplodedNode * EndPathNode,const PathSensitiveBugReport & BR)329 BugReporterVisitor::getDefaultEndPath(const BugReporterContext &BRC,
330                                       const ExplodedNode *EndPathNode,
331                                       const PathSensitiveBugReport &BR) {
332   PathDiagnosticLocation L = BR.getLocation();
333   const auto &Ranges = BR.getRanges();
334 
335   // Only add the statement itself as a range if we didn't specify any
336   // special ranges for this report.
337   auto P = std::make_shared<PathDiagnosticEventPiece>(
338       L, BR.getDescription(), Ranges.begin() == Ranges.end());
339   for (SourceRange Range : Ranges)
340     P->addRange(Range);
341 
342   return P;
343 }
344 
345 //===----------------------------------------------------------------------===//
346 // Implementation of NoStoreFuncVisitor.
347 //===----------------------------------------------------------------------===//
348 
349 namespace {
350 
351 /// Put a diagnostic on return statement of all inlined functions
352 /// for which  the region of interest \p RegionOfInterest was passed into,
353 /// but not written inside, and it has caused an undefined read or a null
354 /// pointer dereference outside.
355 class NoStoreFuncVisitor final : public BugReporterVisitor {
356   const SubRegion *RegionOfInterest;
357   MemRegionManager &MmrMgr;
358   const SourceManager &SM;
359   const PrintingPolicy &PP;
360   bugreporter::TrackingKind TKind;
361 
362   /// Recursion limit for dereferencing fields when looking for the
363   /// region of interest.
364   /// The limit of two indicates that we will dereference fields only once.
365   static const unsigned DEREFERENCE_LIMIT = 2;
366 
367   /// Frames writing into \c RegionOfInterest.
368   /// This visitor generates a note only if a function does not write into
369   /// a region of interest. This information is not immediately available
370   /// by looking at the node associated with the exit from the function
371   /// (usually the return statement). To avoid recomputing the same information
372   /// many times (going up the path for each node and checking whether the
373   /// region was written into) we instead lazily compute the
374   /// stack frames along the path which write into the region of interest.
375   llvm::SmallPtrSet<const StackFrameContext *, 32> FramesModifyingRegion;
376   llvm::SmallPtrSet<const StackFrameContext *, 32> FramesModifyingCalculated;
377 
378   using RegionVector = SmallVector<const MemRegion *, 5>;
379 
380 public:
NoStoreFuncVisitor(const SubRegion * R,bugreporter::TrackingKind TKind)381   NoStoreFuncVisitor(const SubRegion *R, bugreporter::TrackingKind TKind)
382       : RegionOfInterest(R), MmrMgr(R->getMemRegionManager()),
383         SM(MmrMgr.getContext().getSourceManager()),
384         PP(MmrMgr.getContext().getPrintingPolicy()), TKind(TKind) {}
385 
Profile(llvm::FoldingSetNodeID & ID) const386   void Profile(llvm::FoldingSetNodeID &ID) const override {
387     static int Tag = 0;
388     ID.AddPointer(&Tag);
389     ID.AddPointer(RegionOfInterest);
390   }
391 
getTag() const392   void *getTag() const {
393     static int Tag = 0;
394     return static_cast<void *>(&Tag);
395   }
396 
397   PathDiagnosticPieceRef VisitNode(const ExplodedNode *N,
398                                    BugReporterContext &BR,
399                                    PathSensitiveBugReport &R) override;
400 
401 private:
402   /// Attempts to find the region of interest in a given record decl,
403   /// by either following the base classes or fields.
404   /// Dereferences fields up to a given recursion limit.
405   /// Note that \p Vec is passed by value, leading to quadratic copying cost,
406   /// but it's OK in practice since its length is limited to DEREFERENCE_LIMIT.
407   /// \return A chain fields leading to the region of interest or None.
408   const Optional<RegionVector>
409   findRegionOfInterestInRecord(const RecordDecl *RD, ProgramStateRef State,
410                                const MemRegion *R, const RegionVector &Vec = {},
411                                int depth = 0);
412 
413   /// Check and lazily calculate whether the region of interest is
414   /// modified in the stack frame to which \p N belongs.
415   /// The calculation is cached in FramesModifyingRegion.
isRegionOfInterestModifiedInFrame(const ExplodedNode * N)416   bool isRegionOfInterestModifiedInFrame(const ExplodedNode *N) {
417     const LocationContext *Ctx = N->getLocationContext();
418     const StackFrameContext *SCtx = Ctx->getStackFrame();
419     if (!FramesModifyingCalculated.count(SCtx))
420       findModifyingFrames(N);
421     return FramesModifyingRegion.count(SCtx);
422   }
423 
424   /// Write to \c FramesModifyingRegion all stack frames along
425   /// the path in the current stack frame which modify \c RegionOfInterest.
426   void findModifyingFrames(const ExplodedNode *N);
427 
428   /// Consume the information on the no-store stack frame in order to
429   /// either emit a note or suppress the report enirely.
430   /// \return Diagnostics piece for region not modified in the current function,
431   /// if it decides to emit one.
432   PathDiagnosticPieceRef
433   maybeEmitNote(PathSensitiveBugReport &R, const CallEvent &Call,
434                 const ExplodedNode *N, const RegionVector &FieldChain,
435                 const MemRegion *MatchedRegion, StringRef FirstElement,
436                 bool FirstIsReferenceType, unsigned IndirectionLevel);
437 
438   /// Pretty-print region \p MatchedRegion to \p os.
439   /// \return Whether printing succeeded.
440   bool prettyPrintRegionName(StringRef FirstElement, bool FirstIsReferenceType,
441                              const MemRegion *MatchedRegion,
442                              const RegionVector &FieldChain,
443                              int IndirectionLevel,
444                              llvm::raw_svector_ostream &os);
445 
446   /// Print first item in the chain, return new separator.
447   static StringRef prettyPrintFirstElement(StringRef FirstElement,
448                                            bool MoreItemsExpected,
449                                            int IndirectionLevel,
450                                            llvm::raw_svector_ostream &os);
451 };
452 
453 } // end of anonymous namespace
454 
455 /// \return Whether the method declaration \p Parent
456 /// syntactically has a binary operation writing into the ivar \p Ivar.
potentiallyWritesIntoIvar(const Decl * Parent,const ObjCIvarDecl * Ivar)457 static bool potentiallyWritesIntoIvar(const Decl *Parent,
458                                       const ObjCIvarDecl *Ivar) {
459   using namespace ast_matchers;
460   const char *IvarBind = "Ivar";
461   if (!Parent || !Parent->hasBody())
462     return false;
463   StatementMatcher WriteIntoIvarM = binaryOperator(
464       hasOperatorName("="),
465       hasLHS(ignoringParenImpCasts(
466           objcIvarRefExpr(hasDeclaration(equalsNode(Ivar))).bind(IvarBind))));
467   StatementMatcher ParentM = stmt(hasDescendant(WriteIntoIvarM));
468   auto Matches = match(ParentM, *Parent->getBody(), Parent->getASTContext());
469   for (BoundNodes &Match : Matches) {
470     auto IvarRef = Match.getNodeAs<ObjCIvarRefExpr>(IvarBind);
471     if (IvarRef->isFreeIvar())
472       return true;
473 
474     const Expr *Base = IvarRef->getBase();
475     if (const auto *ICE = dyn_cast<ImplicitCastExpr>(Base))
476       Base = ICE->getSubExpr();
477 
478     if (const auto *DRE = dyn_cast<DeclRefExpr>(Base))
479       if (const auto *ID = dyn_cast<ImplicitParamDecl>(DRE->getDecl()))
480         if (ID->getParameterKind() == ImplicitParamDecl::ObjCSelf)
481           return true;
482 
483     return false;
484   }
485   return false;
486 }
487 
488 /// Get parameters associated with runtime definition in order
489 /// to get the correct parameter name.
getCallParameters(CallEventRef<> Call)490 static ArrayRef<ParmVarDecl *> getCallParameters(CallEventRef<> Call) {
491   // Use runtime definition, if available.
492   RuntimeDefinition RD = Call->getRuntimeDefinition();
493   if (const auto *FD = dyn_cast_or_null<FunctionDecl>(RD.getDecl()))
494     return FD->parameters();
495   if (const auto *MD = dyn_cast_or_null<ObjCMethodDecl>(RD.getDecl()))
496     return MD->parameters();
497 
498   return Call->parameters();
499 }
500 
501 /// \return whether \p Ty points to a const type, or is a const reference.
isPointerToConst(QualType Ty)502 static bool isPointerToConst(QualType Ty) {
503   return !Ty->getPointeeType().isNull() &&
504          Ty->getPointeeType().getCanonicalType().isConstQualified();
505 }
506 
507 /// Attempts to find the region of interest in a given CXX decl,
508 /// by either following the base classes or fields.
509 /// Dereferences fields up to a given recursion limit.
510 /// Note that \p Vec is passed by value, leading to quadratic copying cost,
511 /// but it's OK in practice since its length is limited to DEREFERENCE_LIMIT.
512 /// \return A chain fields leading to the region of interest or None.
513 const Optional<NoStoreFuncVisitor::RegionVector>
findRegionOfInterestInRecord(const RecordDecl * RD,ProgramStateRef State,const MemRegion * R,const NoStoreFuncVisitor::RegionVector & Vec,int depth)514 NoStoreFuncVisitor::findRegionOfInterestInRecord(
515     const RecordDecl *RD, ProgramStateRef State, const MemRegion *R,
516     const NoStoreFuncVisitor::RegionVector &Vec /* = {} */,
517     int depth /* = 0 */) {
518 
519   if (depth == DEREFERENCE_LIMIT) // Limit the recursion depth.
520     return None;
521 
522   if (const auto *RDX = dyn_cast<CXXRecordDecl>(RD))
523     if (!RDX->hasDefinition())
524       return None;
525 
526   // Recursively examine the base classes.
527   // Note that following base classes does not increase the recursion depth.
528   if (const auto *RDX = dyn_cast<CXXRecordDecl>(RD))
529     for (const auto &II : RDX->bases())
530       if (const RecordDecl *RRD = II.getType()->getAsRecordDecl())
531         if (Optional<RegionVector> Out =
532                 findRegionOfInterestInRecord(RRD, State, R, Vec, depth))
533           return Out;
534 
535   for (const FieldDecl *I : RD->fields()) {
536     QualType FT = I->getType();
537     const FieldRegion *FR = MmrMgr.getFieldRegion(I, cast<SubRegion>(R));
538     const SVal V = State->getSVal(FR);
539     const MemRegion *VR = V.getAsRegion();
540 
541     RegionVector VecF = Vec;
542     VecF.push_back(FR);
543 
544     if (RegionOfInterest == VR)
545       return VecF;
546 
547     if (const RecordDecl *RRD = FT->getAsRecordDecl())
548       if (auto Out =
549               findRegionOfInterestInRecord(RRD, State, FR, VecF, depth + 1))
550         return Out;
551 
552     QualType PT = FT->getPointeeType();
553     if (PT.isNull() || PT->isVoidType() || !VR)
554       continue;
555 
556     if (const RecordDecl *RRD = PT->getAsRecordDecl())
557       if (Optional<RegionVector> Out =
558               findRegionOfInterestInRecord(RRD, State, VR, VecF, depth + 1))
559         return Out;
560   }
561 
562   return None;
563 }
564 
565 PathDiagnosticPieceRef
VisitNode(const ExplodedNode * N,BugReporterContext & BR,PathSensitiveBugReport & R)566 NoStoreFuncVisitor::VisitNode(const ExplodedNode *N, BugReporterContext &BR,
567                               PathSensitiveBugReport &R) {
568 
569   const LocationContext *Ctx = N->getLocationContext();
570   const StackFrameContext *SCtx = Ctx->getStackFrame();
571   ProgramStateRef State = N->getState();
572   auto CallExitLoc = N->getLocationAs<CallExitBegin>();
573 
574   // No diagnostic if region was modified inside the frame.
575   if (!CallExitLoc || isRegionOfInterestModifiedInFrame(N))
576     return nullptr;
577 
578   CallEventRef<> Call =
579       BR.getStateManager().getCallEventManager().getCaller(SCtx, State);
580 
581   // Region of interest corresponds to an IVar, exiting a method
582   // which could have written into that IVar, but did not.
583   if (const auto *MC = dyn_cast<ObjCMethodCall>(Call)) {
584     if (const auto *IvarR = dyn_cast<ObjCIvarRegion>(RegionOfInterest)) {
585       const MemRegion *SelfRegion = MC->getReceiverSVal().getAsRegion();
586       if (RegionOfInterest->isSubRegionOf(SelfRegion) &&
587           potentiallyWritesIntoIvar(Call->getRuntimeDefinition().getDecl(),
588                                     IvarR->getDecl()))
589         return maybeEmitNote(R, *Call, N, {}, SelfRegion, "self",
590                              /*FirstIsReferenceType=*/false, 1);
591     }
592   }
593 
594   if (const auto *CCall = dyn_cast<CXXConstructorCall>(Call)) {
595     const MemRegion *ThisR = CCall->getCXXThisVal().getAsRegion();
596     if (RegionOfInterest->isSubRegionOf(ThisR) &&
597         !CCall->getDecl()->isImplicit())
598       return maybeEmitNote(R, *Call, N, {}, ThisR, "this",
599                            /*FirstIsReferenceType=*/false, 1);
600 
601     // Do not generate diagnostics for not modified parameters in
602     // constructors.
603     return nullptr;
604   }
605 
606   ArrayRef<ParmVarDecl *> parameters = getCallParameters(Call);
607   for (unsigned I = 0; I < Call->getNumArgs() && I < parameters.size(); ++I) {
608     const ParmVarDecl *PVD = parameters[I];
609     SVal V = Call->getArgSVal(I);
610     bool ParamIsReferenceType = PVD->getType()->isReferenceType();
611     std::string ParamName = PVD->getNameAsString();
612 
613     int IndirectionLevel = 1;
614     QualType T = PVD->getType();
615     while (const MemRegion *MR = V.getAsRegion()) {
616       if (RegionOfInterest->isSubRegionOf(MR) && !isPointerToConst(T))
617         return maybeEmitNote(R, *Call, N, {}, MR, ParamName,
618                              ParamIsReferenceType, IndirectionLevel);
619 
620       QualType PT = T->getPointeeType();
621       if (PT.isNull() || PT->isVoidType())
622         break;
623 
624       if (const RecordDecl *RD = PT->getAsRecordDecl())
625         if (Optional<RegionVector> P =
626                 findRegionOfInterestInRecord(RD, State, MR))
627           return maybeEmitNote(R, *Call, N, *P, RegionOfInterest, ParamName,
628                                ParamIsReferenceType, IndirectionLevel);
629 
630       V = State->getSVal(MR, PT);
631       T = PT;
632       IndirectionLevel++;
633     }
634   }
635 
636   return nullptr;
637 }
638 
findModifyingFrames(const ExplodedNode * N)639 void NoStoreFuncVisitor::findModifyingFrames(const ExplodedNode *N) {
640   assert(N->getLocationAs<CallExitBegin>());
641   ProgramStateRef LastReturnState = N->getState();
642   SVal ValueAtReturn = LastReturnState->getSVal(RegionOfInterest);
643   const LocationContext *Ctx = N->getLocationContext();
644   const StackFrameContext *OriginalSCtx = Ctx->getStackFrame();
645 
646   do {
647     ProgramStateRef State = N->getState();
648     auto CallExitLoc = N->getLocationAs<CallExitBegin>();
649     if (CallExitLoc) {
650       LastReturnState = State;
651       ValueAtReturn = LastReturnState->getSVal(RegionOfInterest);
652     }
653 
654     FramesModifyingCalculated.insert(N->getLocationContext()->getStackFrame());
655 
656     if (wasRegionOfInterestModifiedAt(RegionOfInterest, N, ValueAtReturn)) {
657       const StackFrameContext *SCtx = N->getStackFrame();
658       while (!SCtx->inTopFrame()) {
659         auto p = FramesModifyingRegion.insert(SCtx);
660         if (!p.second)
661           break; // Frame and all its parents already inserted.
662         SCtx = SCtx->getParent()->getStackFrame();
663       }
664     }
665 
666     // Stop calculation at the call to the current function.
667     if (auto CE = N->getLocationAs<CallEnter>())
668       if (CE->getCalleeContext() == OriginalSCtx)
669         break;
670 
671     N = N->getFirstPred();
672   } while (N);
673 }
674 
675 static llvm::StringLiteral WillBeUsedForACondition =
676     ", which participates in a condition later";
677 
maybeEmitNote(PathSensitiveBugReport & R,const CallEvent & Call,const ExplodedNode * N,const RegionVector & FieldChain,const MemRegion * MatchedRegion,StringRef FirstElement,bool FirstIsReferenceType,unsigned IndirectionLevel)678 PathDiagnosticPieceRef NoStoreFuncVisitor::maybeEmitNote(
679     PathSensitiveBugReport &R, const CallEvent &Call, const ExplodedNode *N,
680     const RegionVector &FieldChain, const MemRegion *MatchedRegion,
681     StringRef FirstElement, bool FirstIsReferenceType,
682     unsigned IndirectionLevel) {
683   // Optimistically suppress uninitialized value bugs that result
684   // from system headers having a chance to initialize the value
685   // but failing to do so. It's too unlikely a system header's fault.
686   // It's much more likely a situation in which the function has a failure
687   // mode that the user decided not to check. If we want to hunt such
688   // omitted checks, we should provide an explicit function-specific note
689   // describing the precondition under which the function isn't supposed to
690   // initialize its out-parameter, and additionally check that such
691   // precondition can actually be fulfilled on the current path.
692   if (Call.isInSystemHeader()) {
693     // We make an exception for system header functions that have no branches.
694     // Such functions unconditionally fail to initialize the variable.
695     // If they call other functions that have more paths within them,
696     // this suppression would still apply when we visit these inner functions.
697     // One common example of a standard function that doesn't ever initialize
698     // its out parameter is operator placement new; it's up to the follow-up
699     // constructor (if any) to initialize the memory.
700     if (!N->getStackFrame()->getCFG()->isLinear())
701       R.markInvalid(getTag(), nullptr);
702     return nullptr;
703   }
704 
705   PathDiagnosticLocation L =
706       PathDiagnosticLocation::create(N->getLocation(), SM);
707 
708   // For now this shouldn't trigger, but once it does (as we add more
709   // functions to the body farm), we'll need to decide if these reports
710   // are worth suppressing as well.
711   if (!L.hasValidLocation())
712     return nullptr;
713 
714   SmallString<256> sbuf;
715   llvm::raw_svector_ostream os(sbuf);
716   os << "Returning without writing to '";
717 
718   // Do not generate the note if failed to pretty-print.
719   if (!prettyPrintRegionName(FirstElement, FirstIsReferenceType, MatchedRegion,
720                              FieldChain, IndirectionLevel, os))
721     return nullptr;
722 
723   os << "'";
724   if (TKind == bugreporter::TrackingKind::Condition)
725     os << WillBeUsedForACondition;
726   return std::make_shared<PathDiagnosticEventPiece>(L, os.str());
727 }
728 
prettyPrintRegionName(StringRef FirstElement,bool FirstIsReferenceType,const MemRegion * MatchedRegion,const RegionVector & FieldChain,int IndirectionLevel,llvm::raw_svector_ostream & os)729 bool NoStoreFuncVisitor::prettyPrintRegionName(StringRef FirstElement,
730                                                bool FirstIsReferenceType,
731                                                const MemRegion *MatchedRegion,
732                                                const RegionVector &FieldChain,
733                                                int IndirectionLevel,
734                                                llvm::raw_svector_ostream &os) {
735 
736   if (FirstIsReferenceType)
737     IndirectionLevel--;
738 
739   RegionVector RegionSequence;
740 
741   // Add the regions in the reverse order, then reverse the resulting array.
742   assert(RegionOfInterest->isSubRegionOf(MatchedRegion));
743   const MemRegion *R = RegionOfInterest;
744   while (R != MatchedRegion) {
745     RegionSequence.push_back(R);
746     R = cast<SubRegion>(R)->getSuperRegion();
747   }
748   std::reverse(RegionSequence.begin(), RegionSequence.end());
749   RegionSequence.append(FieldChain.begin(), FieldChain.end());
750 
751   StringRef Sep;
752   for (const MemRegion *R : RegionSequence) {
753 
754     // Just keep going up to the base region.
755     // Element regions may appear due to casts.
756     if (isa<CXXBaseObjectRegion>(R) || isa<CXXTempObjectRegion>(R))
757       continue;
758 
759     if (Sep.empty())
760       Sep = prettyPrintFirstElement(FirstElement,
761                                     /*MoreItemsExpected=*/true,
762                                     IndirectionLevel, os);
763 
764     os << Sep;
765 
766     // Can only reasonably pretty-print DeclRegions.
767     if (!isa<DeclRegion>(R))
768       return false;
769 
770     const auto *DR = cast<DeclRegion>(R);
771     Sep = DR->getValueType()->isAnyPointerType() ? "->" : ".";
772     DR->getDecl()->getDeclName().print(os, PP);
773   }
774 
775   if (Sep.empty())
776     prettyPrintFirstElement(FirstElement,
777                             /*MoreItemsExpected=*/false, IndirectionLevel, os);
778   return true;
779 }
780 
prettyPrintFirstElement(StringRef FirstElement,bool MoreItemsExpected,int IndirectionLevel,llvm::raw_svector_ostream & os)781 StringRef NoStoreFuncVisitor::prettyPrintFirstElement(
782     StringRef FirstElement, bool MoreItemsExpected, int IndirectionLevel,
783     llvm::raw_svector_ostream &os) {
784   StringRef Out = ".";
785 
786   if (IndirectionLevel > 0 && MoreItemsExpected) {
787     IndirectionLevel--;
788     Out = "->";
789   }
790 
791   if (IndirectionLevel > 0 && MoreItemsExpected)
792     os << "(";
793 
794   for (int i = 0; i < IndirectionLevel; i++)
795     os << "*";
796   os << FirstElement;
797 
798   if (IndirectionLevel > 0 && MoreItemsExpected)
799     os << ")";
800 
801   return Out;
802 }
803 
804 //===----------------------------------------------------------------------===//
805 // Implementation of MacroNullReturnSuppressionVisitor.
806 //===----------------------------------------------------------------------===//
807 
808 namespace {
809 
810 /// Suppress null-pointer-dereference bugs where dereferenced null was returned
811 /// the macro.
812 class MacroNullReturnSuppressionVisitor final : public BugReporterVisitor {
813   const SubRegion *RegionOfInterest;
814   const SVal ValueAtDereference;
815 
816   // Do not invalidate the reports where the value was modified
817   // after it got assigned to from the macro.
818   bool WasModified = false;
819 
820 public:
MacroNullReturnSuppressionVisitor(const SubRegion * R,const SVal V)821   MacroNullReturnSuppressionVisitor(const SubRegion *R, const SVal V)
822       : RegionOfInterest(R), ValueAtDereference(V) {}
823 
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)824   PathDiagnosticPieceRef VisitNode(const ExplodedNode *N,
825                                    BugReporterContext &BRC,
826                                    PathSensitiveBugReport &BR) override {
827     if (WasModified)
828       return nullptr;
829 
830     auto BugPoint = BR.getErrorNode()->getLocation().getAs<StmtPoint>();
831     if (!BugPoint)
832       return nullptr;
833 
834     const SourceManager &SMgr = BRC.getSourceManager();
835     if (auto Loc = matchAssignment(N)) {
836       if (isFunctionMacroExpansion(*Loc, SMgr)) {
837         std::string MacroName = std::string(getMacroName(*Loc, BRC));
838         SourceLocation BugLoc = BugPoint->getStmt()->getBeginLoc();
839         if (!BugLoc.isMacroID() || getMacroName(BugLoc, BRC) != MacroName)
840           BR.markInvalid(getTag(), MacroName.c_str());
841       }
842     }
843 
844     if (wasRegionOfInterestModifiedAt(RegionOfInterest, N, ValueAtDereference))
845       WasModified = true;
846 
847     return nullptr;
848   }
849 
addMacroVisitorIfNecessary(const ExplodedNode * N,const MemRegion * R,bool EnableNullFPSuppression,PathSensitiveBugReport & BR,const SVal V)850   static void addMacroVisitorIfNecessary(
851         const ExplodedNode *N, const MemRegion *R,
852         bool EnableNullFPSuppression, PathSensitiveBugReport &BR,
853         const SVal V) {
854     AnalyzerOptions &Options = N->getState()->getAnalysisManager().options;
855     if (EnableNullFPSuppression &&
856         Options.ShouldSuppressNullReturnPaths && V.getAs<Loc>())
857       BR.addVisitor(std::make_unique<MacroNullReturnSuppressionVisitor>(
858               R->getAs<SubRegion>(), V));
859   }
860 
getTag() const861   void* getTag() const {
862     static int Tag = 0;
863     return static_cast<void *>(&Tag);
864   }
865 
Profile(llvm::FoldingSetNodeID & ID) const866   void Profile(llvm::FoldingSetNodeID &ID) const override {
867     ID.AddPointer(getTag());
868   }
869 
870 private:
871   /// \return Source location of right hand side of an assignment
872   /// into \c RegionOfInterest, empty optional if none found.
matchAssignment(const ExplodedNode * N)873   Optional<SourceLocation> matchAssignment(const ExplodedNode *N) {
874     const Stmt *S = N->getStmtForDiagnostics();
875     ProgramStateRef State = N->getState();
876     auto *LCtx = N->getLocationContext();
877     if (!S)
878       return None;
879 
880     if (const auto *DS = dyn_cast<DeclStmt>(S)) {
881       if (const auto *VD = dyn_cast<VarDecl>(DS->getSingleDecl()))
882         if (const Expr *RHS = VD->getInit())
883           if (RegionOfInterest->isSubRegionOf(
884                   State->getLValue(VD, LCtx).getAsRegion()))
885             return RHS->getBeginLoc();
886     } else if (const auto *BO = dyn_cast<BinaryOperator>(S)) {
887       const MemRegion *R = N->getSVal(BO->getLHS()).getAsRegion();
888       const Expr *RHS = BO->getRHS();
889       if (BO->isAssignmentOp() && RegionOfInterest->isSubRegionOf(R)) {
890         return RHS->getBeginLoc();
891       }
892     }
893     return None;
894   }
895 };
896 
897 } // end of anonymous namespace
898 
899 namespace {
900 
901 /// Emits an extra note at the return statement of an interesting stack frame.
902 ///
903 /// The returned value is marked as an interesting value, and if it's null,
904 /// adds a visitor to track where it became null.
905 ///
906 /// This visitor is intended to be used when another visitor discovers that an
907 /// interesting value comes from an inlined function call.
908 class ReturnVisitor : public BugReporterVisitor {
909   const StackFrameContext *CalleeSFC;
910   enum {
911     Initial,
912     MaybeUnsuppress,
913     Satisfied
914   } Mode = Initial;
915 
916   bool EnableNullFPSuppression;
917   bool ShouldInvalidate = true;
918   AnalyzerOptions& Options;
919   bugreporter::TrackingKind TKind;
920 
921 public:
ReturnVisitor(const StackFrameContext * Frame,bool Suppressed,AnalyzerOptions & Options,bugreporter::TrackingKind TKind)922   ReturnVisitor(const StackFrameContext *Frame, bool Suppressed,
923                 AnalyzerOptions &Options, bugreporter::TrackingKind TKind)
924       : CalleeSFC(Frame), EnableNullFPSuppression(Suppressed),
925         Options(Options), TKind(TKind) {}
926 
getTag()927   static void *getTag() {
928     static int Tag = 0;
929     return static_cast<void *>(&Tag);
930   }
931 
Profile(llvm::FoldingSetNodeID & ID) const932   void Profile(llvm::FoldingSetNodeID &ID) const override {
933     ID.AddPointer(ReturnVisitor::getTag());
934     ID.AddPointer(CalleeSFC);
935     ID.AddBoolean(EnableNullFPSuppression);
936   }
937 
938   /// Adds a ReturnVisitor if the given statement represents a call that was
939   /// inlined.
940   ///
941   /// This will search back through the ExplodedGraph, starting from the given
942   /// node, looking for when the given statement was processed. If it turns out
943   /// the statement is a call that was inlined, we add the visitor to the
944   /// bug report, so it can print a note later.
addVisitorIfNecessary(const ExplodedNode * Node,const Stmt * S,PathSensitiveBugReport & BR,bool InEnableNullFPSuppression,bugreporter::TrackingKind TKind)945   static void addVisitorIfNecessary(const ExplodedNode *Node, const Stmt *S,
946                                     PathSensitiveBugReport &BR,
947                                     bool InEnableNullFPSuppression,
948                                     bugreporter::TrackingKind TKind) {
949     if (!CallEvent::isCallStmt(S))
950       return;
951 
952     // First, find when we processed the statement.
953     // If we work with a 'CXXNewExpr' that is going to be purged away before
954     // its call take place. We would catch that purge in the last condition
955     // as a 'StmtPoint' so we have to bypass it.
956     const bool BypassCXXNewExprEval = isa<CXXNewExpr>(S);
957 
958     // This is moving forward when we enter into another context.
959     const StackFrameContext *CurrentSFC = Node->getStackFrame();
960 
961     do {
962       // If that is satisfied we found our statement as an inlined call.
963       if (Optional<CallExitEnd> CEE = Node->getLocationAs<CallExitEnd>())
964         if (CEE->getCalleeContext()->getCallSite() == S)
965           break;
966 
967       // Try to move forward to the end of the call-chain.
968       Node = Node->getFirstPred();
969       if (!Node)
970         break;
971 
972       const StackFrameContext *PredSFC = Node->getStackFrame();
973 
974       // If that is satisfied we found our statement.
975       // FIXME: This code currently bypasses the call site for the
976       //        conservatively evaluated allocator.
977       if (!BypassCXXNewExprEval)
978         if (Optional<StmtPoint> SP = Node->getLocationAs<StmtPoint>())
979           // See if we do not enter into another context.
980           if (SP->getStmt() == S && CurrentSFC == PredSFC)
981             break;
982 
983       CurrentSFC = PredSFC;
984     } while (Node->getStackFrame() == CurrentSFC);
985 
986     // Next, step over any post-statement checks.
987     while (Node && Node->getLocation().getAs<PostStmt>())
988       Node = Node->getFirstPred();
989     if (!Node)
990       return;
991 
992     // Finally, see if we inlined the call.
993     Optional<CallExitEnd> CEE = Node->getLocationAs<CallExitEnd>();
994     if (!CEE)
995       return;
996 
997     const StackFrameContext *CalleeContext = CEE->getCalleeContext();
998     if (CalleeContext->getCallSite() != S)
999       return;
1000 
1001     // Check the return value.
1002     ProgramStateRef State = Node->getState();
1003     SVal RetVal = Node->getSVal(S);
1004 
1005     // Handle cases where a reference is returned and then immediately used.
1006     if (cast<Expr>(S)->isGLValue())
1007       if (Optional<Loc> LValue = RetVal.getAs<Loc>())
1008         RetVal = State->getSVal(*LValue);
1009 
1010     // See if the return value is NULL. If so, suppress the report.
1011     AnalyzerOptions &Options = State->getAnalysisManager().options;
1012 
1013     bool EnableNullFPSuppression = false;
1014     if (InEnableNullFPSuppression &&
1015         Options.ShouldSuppressNullReturnPaths)
1016       if (Optional<Loc> RetLoc = RetVal.getAs<Loc>())
1017         EnableNullFPSuppression = State->isNull(*RetLoc).isConstrainedTrue();
1018 
1019     BR.addVisitor(std::make_unique<ReturnVisitor>(CalleeContext,
1020                                                    EnableNullFPSuppression,
1021                                                    Options, TKind));
1022   }
1023 
visitNodeInitial(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)1024   PathDiagnosticPieceRef visitNodeInitial(const ExplodedNode *N,
1025                                           BugReporterContext &BRC,
1026                                           PathSensitiveBugReport &BR) {
1027     // Only print a message at the interesting return statement.
1028     if (N->getLocationContext() != CalleeSFC)
1029       return nullptr;
1030 
1031     Optional<StmtPoint> SP = N->getLocationAs<StmtPoint>();
1032     if (!SP)
1033       return nullptr;
1034 
1035     const auto *Ret = dyn_cast<ReturnStmt>(SP->getStmt());
1036     if (!Ret)
1037       return nullptr;
1038 
1039     // Okay, we're at the right return statement, but do we have the return
1040     // value available?
1041     ProgramStateRef State = N->getState();
1042     SVal V = State->getSVal(Ret, CalleeSFC);
1043     if (V.isUnknownOrUndef())
1044       return nullptr;
1045 
1046     // Don't print any more notes after this one.
1047     Mode = Satisfied;
1048 
1049     const Expr *RetE = Ret->getRetValue();
1050     assert(RetE && "Tracking a return value for a void function");
1051 
1052     // Handle cases where a reference is returned and then immediately used.
1053     Optional<Loc> LValue;
1054     if (RetE->isGLValue()) {
1055       if ((LValue = V.getAs<Loc>())) {
1056         SVal RValue = State->getRawSVal(*LValue, RetE->getType());
1057         if (RValue.getAs<DefinedSVal>())
1058           V = RValue;
1059       }
1060     }
1061 
1062     // Ignore aggregate rvalues.
1063     if (V.getAs<nonloc::LazyCompoundVal>() ||
1064         V.getAs<nonloc::CompoundVal>())
1065       return nullptr;
1066 
1067     RetE = RetE->IgnoreParenCasts();
1068 
1069     // Let's track the return value.
1070     bugreporter::trackExpressionValue(
1071         N, RetE, BR, TKind, EnableNullFPSuppression);
1072 
1073     // Build an appropriate message based on the return value.
1074     SmallString<64> Msg;
1075     llvm::raw_svector_ostream Out(Msg);
1076 
1077     bool WouldEventBeMeaningless = false;
1078 
1079     if (State->isNull(V).isConstrainedTrue()) {
1080       if (V.getAs<Loc>()) {
1081 
1082         // If we have counter-suppression enabled, make sure we keep visiting
1083         // future nodes. We want to emit a path note as well, in case
1084         // the report is resurrected as valid later on.
1085         if (EnableNullFPSuppression &&
1086             Options.ShouldAvoidSuppressingNullArgumentPaths)
1087           Mode = MaybeUnsuppress;
1088 
1089         if (RetE->getType()->isObjCObjectPointerType()) {
1090           Out << "Returning nil";
1091         } else {
1092           Out << "Returning null pointer";
1093         }
1094       } else {
1095         Out << "Returning zero";
1096       }
1097 
1098     } else {
1099       if (auto CI = V.getAs<nonloc::ConcreteInt>()) {
1100         Out << "Returning the value " << CI->getValue();
1101       } else {
1102         // There is nothing interesting about returning a value, when it is
1103         // plain value without any constraints, and the function is guaranteed
1104         // to return that every time. We could use CFG::isLinear() here, but
1105         // constexpr branches are obvious to the compiler, not necesserily to
1106         // the programmer.
1107         if (N->getCFG().size() == 3)
1108           WouldEventBeMeaningless = true;
1109 
1110         if (V.getAs<Loc>())
1111           Out << "Returning pointer";
1112         else
1113           Out << "Returning value";
1114       }
1115     }
1116 
1117     if (LValue) {
1118       if (const MemRegion *MR = LValue->getAsRegion()) {
1119         if (MR->canPrintPretty()) {
1120           Out << " (reference to ";
1121           MR->printPretty(Out);
1122           Out << ")";
1123         }
1124       }
1125     } else {
1126       // FIXME: We should have a more generalized location printing mechanism.
1127       if (const auto *DR = dyn_cast<DeclRefExpr>(RetE))
1128         if (const auto *DD = dyn_cast<DeclaratorDecl>(DR->getDecl()))
1129           Out << " (loaded from '" << *DD << "')";
1130     }
1131 
1132     PathDiagnosticLocation L(Ret, BRC.getSourceManager(), CalleeSFC);
1133     if (!L.isValid() || !L.asLocation().isValid())
1134       return nullptr;
1135 
1136     if (TKind == bugreporter::TrackingKind::Condition)
1137       Out << WillBeUsedForACondition;
1138 
1139     auto EventPiece = std::make_shared<PathDiagnosticEventPiece>(L, Out.str());
1140 
1141     // If we determined that the note is meaningless, make it prunable, and
1142     // don't mark the stackframe interesting.
1143     if (WouldEventBeMeaningless)
1144       EventPiece->setPrunable(true);
1145     else
1146       BR.markInteresting(CalleeSFC);
1147 
1148     return EventPiece;
1149   }
1150 
visitNodeMaybeUnsuppress(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)1151   PathDiagnosticPieceRef visitNodeMaybeUnsuppress(const ExplodedNode *N,
1152                                                   BugReporterContext &BRC,
1153                                                   PathSensitiveBugReport &BR) {
1154     assert(Options.ShouldAvoidSuppressingNullArgumentPaths);
1155 
1156     // Are we at the entry node for this call?
1157     Optional<CallEnter> CE = N->getLocationAs<CallEnter>();
1158     if (!CE)
1159       return nullptr;
1160 
1161     if (CE->getCalleeContext() != CalleeSFC)
1162       return nullptr;
1163 
1164     Mode = Satisfied;
1165 
1166     // Don't automatically suppress a report if one of the arguments is
1167     // known to be a null pointer. Instead, start tracking /that/ null
1168     // value back to its origin.
1169     ProgramStateManager &StateMgr = BRC.getStateManager();
1170     CallEventManager &CallMgr = StateMgr.getCallEventManager();
1171 
1172     ProgramStateRef State = N->getState();
1173     CallEventRef<> Call = CallMgr.getCaller(CalleeSFC, State);
1174     for (unsigned I = 0, E = Call->getNumArgs(); I != E; ++I) {
1175       Optional<Loc> ArgV = Call->getArgSVal(I).getAs<Loc>();
1176       if (!ArgV)
1177         continue;
1178 
1179       const Expr *ArgE = Call->getArgExpr(I);
1180       if (!ArgE)
1181         continue;
1182 
1183       // Is it possible for this argument to be non-null?
1184       if (!State->isNull(*ArgV).isConstrainedTrue())
1185         continue;
1186 
1187       if (trackExpressionValue(N, ArgE, BR, TKind, EnableNullFPSuppression))
1188         ShouldInvalidate = false;
1189 
1190       // If we /can't/ track the null pointer, we should err on the side of
1191       // false negatives, and continue towards marking this report invalid.
1192       // (We will still look at the other arguments, though.)
1193     }
1194 
1195     return nullptr;
1196   }
1197 
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)1198   PathDiagnosticPieceRef VisitNode(const ExplodedNode *N,
1199                                    BugReporterContext &BRC,
1200                                    PathSensitiveBugReport &BR) override {
1201     switch (Mode) {
1202     case Initial:
1203       return visitNodeInitial(N, BRC, BR);
1204     case MaybeUnsuppress:
1205       return visitNodeMaybeUnsuppress(N, BRC, BR);
1206     case Satisfied:
1207       return nullptr;
1208     }
1209 
1210     llvm_unreachable("Invalid visit mode!");
1211   }
1212 
finalizeVisitor(BugReporterContext &,const ExplodedNode *,PathSensitiveBugReport & BR)1213   void finalizeVisitor(BugReporterContext &, const ExplodedNode *,
1214                        PathSensitiveBugReport &BR) override {
1215     if (EnableNullFPSuppression && ShouldInvalidate)
1216       BR.markInvalid(ReturnVisitor::getTag(), CalleeSFC);
1217   }
1218 };
1219 
1220 } // end of anonymous namespace
1221 
1222 //===----------------------------------------------------------------------===//
1223 // Implementation of FindLastStoreBRVisitor.
1224 //===----------------------------------------------------------------------===//
1225 
Profile(llvm::FoldingSetNodeID & ID) const1226 void FindLastStoreBRVisitor::Profile(llvm::FoldingSetNodeID &ID) const {
1227   static int tag = 0;
1228   ID.AddPointer(&tag);
1229   ID.AddPointer(R);
1230   ID.Add(V);
1231   ID.AddInteger(static_cast<int>(TKind));
1232   ID.AddBoolean(EnableNullFPSuppression);
1233 }
1234 
1235 /// Returns true if \p N represents the DeclStmt declaring and initializing
1236 /// \p VR.
isInitializationOfVar(const ExplodedNode * N,const VarRegion * VR)1237 static bool isInitializationOfVar(const ExplodedNode *N, const VarRegion *VR) {
1238   Optional<PostStmt> P = N->getLocationAs<PostStmt>();
1239   if (!P)
1240     return false;
1241 
1242   const DeclStmt *DS = P->getStmtAs<DeclStmt>();
1243   if (!DS)
1244     return false;
1245 
1246   if (DS->getSingleDecl() != VR->getDecl())
1247     return false;
1248 
1249   const MemSpaceRegion *VarSpace = VR->getMemorySpace();
1250   const auto *FrameSpace = dyn_cast<StackSpaceRegion>(VarSpace);
1251   if (!FrameSpace) {
1252     // If we ever directly evaluate global DeclStmts, this assertion will be
1253     // invalid, but this still seems preferable to silently accepting an
1254     // initialization that may be for a path-sensitive variable.
1255     assert(VR->getDecl()->isStaticLocal() && "non-static stackless VarRegion");
1256     return true;
1257   }
1258 
1259   assert(VR->getDecl()->hasLocalStorage());
1260   const LocationContext *LCtx = N->getLocationContext();
1261   return FrameSpace->getStackFrame() == LCtx->getStackFrame();
1262 }
1263 
1264 /// 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 * NewR,SVal V,const MemRegion * OldR,const DeclStmt * DS)1265 static void showBRDiagnostics(const char *action, llvm::raw_svector_ostream &os,
1266                               const MemRegion *NewR, SVal V,
1267                               const MemRegion *OldR, const DeclStmt *DS) {
1268   if (NewR->canPrintPretty()) {
1269     NewR->printPretty(os);
1270     os << " ";
1271   }
1272 
1273   if (V.getAs<loc::ConcreteInt>()) {
1274     bool b = false;
1275     if (NewR->isBoundable()) {
1276       if (const auto *TR = dyn_cast<TypedValueRegion>(NewR)) {
1277         if (TR->getValueType()->isObjCObjectPointerType()) {
1278           os << action << "nil";
1279           b = true;
1280         }
1281       }
1282     }
1283     if (!b)
1284       os << action << "a null pointer value";
1285 
1286   } else if (auto CVal = V.getAs<nonloc::ConcreteInt>()) {
1287     os << action << CVal->getValue();
1288   } else if (OldR && OldR->canPrintPretty()) {
1289     os << action << "the value of ";
1290     OldR->printPretty(os);
1291   } else if (DS) {
1292     if (V.isUndef()) {
1293       if (isa<VarRegion>(NewR)) {
1294         const auto *VD = cast<VarDecl>(DS->getSingleDecl());
1295         if (VD->getInit()) {
1296           os << (NewR->canPrintPretty() ? "initialized" : "Initializing")
1297              << " to a garbage value";
1298         } else {
1299           os << (NewR->canPrintPretty() ? "declared" : "Declaring")
1300              << " without an initial value";
1301         }
1302       }
1303     } else {
1304       os << (NewR->canPrintPretty() ? "initialized" : "Initialized") << " here";
1305     }
1306   }
1307 }
1308 
1309 /// Display diagnostics for passing bad region as a parameter.
showBRParamDiagnostics(llvm::raw_svector_ostream & os,const VarRegion * VR,SVal V,const MemRegion * ValueR)1310 static void showBRParamDiagnostics(llvm::raw_svector_ostream &os,
1311                                    const VarRegion *VR, SVal V,
1312                                    const MemRegion *ValueR) {
1313   const auto *Param = cast<ParmVarDecl>(VR->getDecl());
1314 
1315   os << "Passing ";
1316 
1317   if (V.getAs<loc::ConcreteInt>()) {
1318     if (Param->getType()->isObjCObjectPointerType())
1319       os << "nil object reference";
1320     else
1321       os << "null pointer value";
1322   } else if (V.isUndef()) {
1323     os << "uninitialized value";
1324   } else if (auto CI = V.getAs<nonloc::ConcreteInt>()) {
1325     os << "the value " << CI->getValue();
1326   } else if (ValueR && ValueR->canPrintPretty()) {
1327     ValueR->printPretty(os);
1328   } else {
1329     os << "value";
1330   }
1331 
1332   // Printed parameter indexes are 1-based, not 0-based.
1333   unsigned Idx = Param->getFunctionScopeIndex() + 1;
1334   os << " via " << Idx << llvm::getOrdinalSuffix(Idx) << " parameter";
1335   if (VR->canPrintPretty()) {
1336     os << " ";
1337     VR->printPretty(os);
1338   }
1339 }
1340 
1341 /// Show default diagnostics for storing bad region.
showBRDefaultDiagnostics(llvm::raw_svector_ostream & os,const MemRegion * NewR,SVal V,const MemRegion * OldR)1342 static void showBRDefaultDiagnostics(llvm::raw_svector_ostream &os,
1343                                      const MemRegion *NewR, SVal V,
1344                                      const MemRegion *OldR) {
1345   if (V.getAs<loc::ConcreteInt>()) {
1346     bool b = false;
1347     if (NewR->isBoundable()) {
1348       if (const auto *TR = dyn_cast<TypedValueRegion>(NewR)) {
1349         if (TR->getValueType()->isObjCObjectPointerType()) {
1350           os << "nil object reference stored";
1351           b = true;
1352         }
1353       }
1354     }
1355     if (!b) {
1356       if (NewR->canPrintPretty())
1357         os << "Null pointer value stored";
1358       else
1359         os << "Storing null pointer value";
1360     }
1361 
1362   } else if (V.isUndef()) {
1363     if (NewR->canPrintPretty())
1364       os << "Uninitialized value stored";
1365     else
1366       os << "Storing uninitialized value";
1367 
1368   } else if (auto CV = V.getAs<nonloc::ConcreteInt>()) {
1369     if (NewR->canPrintPretty())
1370       os << "The value " << CV->getValue() << " is assigned";
1371     else
1372       os << "Assigning " << CV->getValue();
1373 
1374   } else if (OldR && OldR->canPrintPretty()) {
1375     if (NewR->canPrintPretty()) {
1376       os << "The value of ";
1377       OldR->printPretty(os);
1378       os << " is assigned";
1379     } else {
1380       os << "Assigning the value of ";
1381       OldR->printPretty(os);
1382     }
1383 
1384   } else {
1385     if (NewR->canPrintPretty())
1386       os << "Value assigned";
1387     else
1388       os << "Assigning value";
1389   }
1390 
1391   if (NewR->canPrintPretty()) {
1392     os << " to ";
1393     NewR->printPretty(os);
1394   }
1395 }
1396 
1397 PathDiagnosticPieceRef
VisitNode(const ExplodedNode * Succ,BugReporterContext & BRC,PathSensitiveBugReport & BR)1398 FindLastStoreBRVisitor::VisitNode(const ExplodedNode *Succ,
1399                                   BugReporterContext &BRC,
1400                                   PathSensitiveBugReport &BR) {
1401   if (Satisfied)
1402     return nullptr;
1403 
1404   const ExplodedNode *StoreSite = nullptr;
1405   const ExplodedNode *Pred = Succ->getFirstPred();
1406   const Expr *InitE = nullptr;
1407   bool IsParam = false;
1408 
1409   // First see if we reached the declaration of the region.
1410   if (const auto *VR = dyn_cast<VarRegion>(R)) {
1411     if (isInitializationOfVar(Pred, VR)) {
1412       StoreSite = Pred;
1413       InitE = VR->getDecl()->getInit();
1414     }
1415   }
1416 
1417   // If this is a post initializer expression, initializing the region, we
1418   // should track the initializer expression.
1419   if (Optional<PostInitializer> PIP = Pred->getLocationAs<PostInitializer>()) {
1420     const MemRegion *FieldReg = (const MemRegion *)PIP->getLocationValue();
1421     if (FieldReg == R) {
1422       StoreSite = Pred;
1423       InitE = PIP->getInitializer()->getInit();
1424     }
1425   }
1426 
1427   // Otherwise, see if this is the store site:
1428   // (1) Succ has this binding and Pred does not, i.e. this is
1429   //     where the binding first occurred.
1430   // (2) Succ has this binding and is a PostStore node for this region, i.e.
1431   //     the same binding was re-assigned here.
1432   if (!StoreSite) {
1433     if (Succ->getState()->getSVal(R) != V)
1434       return nullptr;
1435 
1436     if (hasVisibleUpdate(Pred, Pred->getState()->getSVal(R), Succ, V)) {
1437       Optional<PostStore> PS = Succ->getLocationAs<PostStore>();
1438       if (!PS || PS->getLocationValue() != R)
1439         return nullptr;
1440     }
1441 
1442     StoreSite = Succ;
1443 
1444     // If this is an assignment expression, we can track the value
1445     // being assigned.
1446     if (Optional<PostStmt> P = Succ->getLocationAs<PostStmt>())
1447       if (const BinaryOperator *BO = P->getStmtAs<BinaryOperator>())
1448         if (BO->isAssignmentOp())
1449           InitE = BO->getRHS();
1450 
1451     // If this is a call entry, the variable should be a parameter.
1452     // FIXME: Handle CXXThisRegion as well. (This is not a priority because
1453     // 'this' should never be NULL, but this visitor isn't just for NULL and
1454     // UndefinedVal.)
1455     if (Optional<CallEnter> CE = Succ->getLocationAs<CallEnter>()) {
1456       if (const auto *VR = dyn_cast<VarRegion>(R)) {
1457 
1458         if (const auto *Param = dyn_cast<ParmVarDecl>(VR->getDecl())) {
1459           ProgramStateManager &StateMgr = BRC.getStateManager();
1460           CallEventManager &CallMgr = StateMgr.getCallEventManager();
1461 
1462           CallEventRef<> Call = CallMgr.getCaller(CE->getCalleeContext(),
1463                                                   Succ->getState());
1464           InitE = Call->getArgExpr(Param->getFunctionScopeIndex());
1465         } else {
1466           // Handle Objective-C 'self'.
1467           assert(isa<ImplicitParamDecl>(VR->getDecl()));
1468           InitE = cast<ObjCMessageExpr>(CE->getCalleeContext()->getCallSite())
1469                       ->getInstanceReceiver()->IgnoreParenCasts();
1470         }
1471         IsParam = true;
1472       }
1473     }
1474 
1475     // If this is a CXXTempObjectRegion, the Expr responsible for its creation
1476     // is wrapped inside of it.
1477     if (const auto *TmpR = dyn_cast<CXXTempObjectRegion>(R))
1478       InitE = TmpR->getExpr();
1479   }
1480 
1481   if (!StoreSite)
1482     return nullptr;
1483 
1484   Satisfied = true;
1485 
1486   // If we have an expression that provided the value, try to track where it
1487   // came from.
1488   if (InitE) {
1489     if (!IsParam)
1490       InitE = InitE->IgnoreParenCasts();
1491 
1492     bugreporter::trackExpressionValue(StoreSite, InitE, BR, TKind,
1493                                       EnableNullFPSuppression);
1494   }
1495 
1496   // Let's try to find the region where the value came from.
1497   const MemRegion *OldRegion = nullptr;
1498 
1499   // If we have init expression, it might be simply a reference
1500   // to a variable, so we can use it.
1501   if (InitE) {
1502     // That region might still be not exactly what we are looking for.
1503     // In situations like `int &ref = val;`, we can't say that
1504     // `ref` is initialized with `val`, rather refers to `val`.
1505     //
1506     // In order, to mitigate situations like this, we check if the last
1507     // stored value in that region is the value that we track.
1508     //
1509     // TODO: support other situations better.
1510     if (const MemRegion *Candidate =
1511             getLocationRegionIfReference(InitE, Succ, false)) {
1512       const StoreManager &SM = BRC.getStateManager().getStoreManager();
1513 
1514       // Here we traverse the graph up to find the last node where the
1515       // candidate region is still in the store.
1516       for (const ExplodedNode *N = StoreSite; N; N = N->getFirstPred()) {
1517         if (SM.includedInBindings(N->getState()->getStore(), Candidate)) {
1518           // And if it was bound to the target value, we can use it.
1519           if (N->getState()->getSVal(Candidate) == V) {
1520             OldRegion = Candidate;
1521           }
1522           break;
1523         }
1524       }
1525     }
1526   }
1527 
1528   // Otherwise, if the current region does indeed contain the value
1529   // we are looking for, we can look for a region where this value
1530   // was before.
1531   //
1532   // It can be useful for situations like:
1533   //     new = identity(old)
1534   // where the analyzer knows that 'identity' returns the value of its
1535   // first argument.
1536   //
1537   // NOTE: If the region R is not a simple var region, it can contain
1538   //       V in one of its subregions.
1539   if (!OldRegion && StoreSite->getState()->getSVal(R) == V) {
1540     // Let's go up the graph to find the node where the region is
1541     // bound to V.
1542     const ExplodedNode *NodeWithoutBinding = StoreSite->getFirstPred();
1543     for (;
1544          NodeWithoutBinding && NodeWithoutBinding->getState()->getSVal(R) == V;
1545          NodeWithoutBinding = NodeWithoutBinding->getFirstPred()) {
1546     }
1547 
1548     if (NodeWithoutBinding) {
1549       // Let's try to find a unique binding for the value in that node.
1550       // We want to use this to find unique bindings because of the following
1551       // situations:
1552       //     b = a;
1553       //     c = identity(b);
1554       //
1555       // Telling the user that the value of 'a' is assigned to 'c', while
1556       // correct, can be confusing.
1557       StoreManager::FindUniqueBinding FB(V.getAsLocSymbol());
1558       BRC.getStateManager().iterBindings(NodeWithoutBinding->getState(), FB);
1559       if (FB)
1560         OldRegion = FB.getRegion();
1561     }
1562   }
1563 
1564   if (TKind == TrackingKind::Condition &&
1565       !OriginSFC->isParentOf(StoreSite->getStackFrame()))
1566     return nullptr;
1567 
1568   // Okay, we've found the binding. Emit an appropriate message.
1569   SmallString<256> sbuf;
1570   llvm::raw_svector_ostream os(sbuf);
1571 
1572   if (Optional<PostStmt> PS = StoreSite->getLocationAs<PostStmt>()) {
1573     const Stmt *S = PS->getStmt();
1574     const char *action = nullptr;
1575     const auto *DS = dyn_cast<DeclStmt>(S);
1576     const auto *VR = dyn_cast<VarRegion>(R);
1577 
1578     if (DS) {
1579       action = R->canPrintPretty() ? "initialized to " :
1580                                      "Initializing to ";
1581     } else if (isa<BlockExpr>(S)) {
1582       action = R->canPrintPretty() ? "captured by block as " :
1583                                      "Captured by block as ";
1584       if (VR) {
1585         // See if we can get the BlockVarRegion.
1586         ProgramStateRef State = StoreSite->getState();
1587         SVal V = StoreSite->getSVal(S);
1588         if (const auto *BDR =
1589               dyn_cast_or_null<BlockDataRegion>(V.getAsRegion())) {
1590           if (const VarRegion *OriginalR = BDR->getOriginalRegion(VR)) {
1591             if (auto KV = State->getSVal(OriginalR).getAs<KnownSVal>())
1592               BR.addVisitor(std::make_unique<FindLastStoreBRVisitor>(
1593                   *KV, OriginalR, EnableNullFPSuppression, TKind, OriginSFC));
1594           }
1595         }
1596       }
1597     }
1598     if (action)
1599       showBRDiagnostics(action, os, R, V, OldRegion, DS);
1600 
1601   } else if (StoreSite->getLocation().getAs<CallEnter>()) {
1602     if (const auto *VR = dyn_cast<VarRegion>(R))
1603       showBRParamDiagnostics(os, VR, V, OldRegion);
1604   }
1605 
1606   if (os.str().empty())
1607     showBRDefaultDiagnostics(os, R, V, OldRegion);
1608 
1609   if (TKind == bugreporter::TrackingKind::Condition)
1610     os << WillBeUsedForACondition;
1611 
1612   // Construct a new PathDiagnosticPiece.
1613   ProgramPoint P = StoreSite->getLocation();
1614   PathDiagnosticLocation L;
1615   if (P.getAs<CallEnter>() && InitE)
1616     L = PathDiagnosticLocation(InitE, BRC.getSourceManager(),
1617                                P.getLocationContext());
1618 
1619   if (!L.isValid() || !L.asLocation().isValid())
1620     L = PathDiagnosticLocation::create(P, BRC.getSourceManager());
1621 
1622   if (!L.isValid() || !L.asLocation().isValid())
1623     return nullptr;
1624 
1625   return std::make_shared<PathDiagnosticEventPiece>(L, os.str());
1626 }
1627 
1628 //===----------------------------------------------------------------------===//
1629 // Implementation of TrackConstraintBRVisitor.
1630 //===----------------------------------------------------------------------===//
1631 
Profile(llvm::FoldingSetNodeID & ID) const1632 void TrackConstraintBRVisitor::Profile(llvm::FoldingSetNodeID &ID) const {
1633   static int tag = 0;
1634   ID.AddPointer(&tag);
1635   ID.AddBoolean(Assumption);
1636   ID.Add(Constraint);
1637 }
1638 
1639 /// Return the tag associated with this visitor.  This tag will be used
1640 /// to make all PathDiagnosticPieces created by this visitor.
getTag()1641 const char *TrackConstraintBRVisitor::getTag() {
1642   return "TrackConstraintBRVisitor";
1643 }
1644 
isUnderconstrained(const ExplodedNode * N) const1645 bool TrackConstraintBRVisitor::isUnderconstrained(const ExplodedNode *N) const {
1646   if (IsZeroCheck)
1647     return N->getState()->isNull(Constraint).isUnderconstrained();
1648   return (bool)N->getState()->assume(Constraint, !Assumption);
1649 }
1650 
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport &)1651 PathDiagnosticPieceRef TrackConstraintBRVisitor::VisitNode(
1652     const ExplodedNode *N, BugReporterContext &BRC, PathSensitiveBugReport &) {
1653   const ExplodedNode *PrevN = N->getFirstPred();
1654   if (IsSatisfied)
1655     return nullptr;
1656 
1657   // Start tracking after we see the first state in which the value is
1658   // constrained.
1659   if (!IsTrackingTurnedOn)
1660     if (!isUnderconstrained(N))
1661       IsTrackingTurnedOn = true;
1662   if (!IsTrackingTurnedOn)
1663     return nullptr;
1664 
1665   // Check if in the previous state it was feasible for this constraint
1666   // to *not* be true.
1667   if (isUnderconstrained(PrevN)) {
1668     IsSatisfied = true;
1669 
1670     // As a sanity check, make sure that the negation of the constraint
1671     // was infeasible in the current state.  If it is feasible, we somehow
1672     // missed the transition point.
1673     assert(!isUnderconstrained(N));
1674 
1675     // We found the transition point for the constraint.  We now need to
1676     // pretty-print the constraint. (work-in-progress)
1677     SmallString<64> sbuf;
1678     llvm::raw_svector_ostream os(sbuf);
1679 
1680     if (Constraint.getAs<Loc>()) {
1681       os << "Assuming pointer value is ";
1682       os << (Assumption ? "non-null" : "null");
1683     }
1684 
1685     if (os.str().empty())
1686       return nullptr;
1687 
1688     // Construct a new PathDiagnosticPiece.
1689     ProgramPoint P = N->getLocation();
1690     PathDiagnosticLocation L =
1691       PathDiagnosticLocation::create(P, BRC.getSourceManager());
1692     if (!L.isValid())
1693       return nullptr;
1694 
1695     auto X = std::make_shared<PathDiagnosticEventPiece>(L, os.str());
1696     X->setTag(getTag());
1697     return std::move(X);
1698   }
1699 
1700   return nullptr;
1701 }
1702 
1703 //===----------------------------------------------------------------------===//
1704 // Implementation of SuppressInlineDefensiveChecksVisitor.
1705 //===----------------------------------------------------------------------===//
1706 
1707 SuppressInlineDefensiveChecksVisitor::
SuppressInlineDefensiveChecksVisitor(DefinedSVal Value,const ExplodedNode * N)1708 SuppressInlineDefensiveChecksVisitor(DefinedSVal Value, const ExplodedNode *N)
1709     : V(Value) {
1710   // Check if the visitor is disabled.
1711   AnalyzerOptions &Options = N->getState()->getAnalysisManager().options;
1712   if (!Options.ShouldSuppressInlinedDefensiveChecks)
1713     IsSatisfied = true;
1714 }
1715 
Profile(llvm::FoldingSetNodeID & ID) const1716 void SuppressInlineDefensiveChecksVisitor::Profile(
1717     llvm::FoldingSetNodeID &ID) const {
1718   static int id = 0;
1719   ID.AddPointer(&id);
1720   ID.Add(V);
1721 }
1722 
getTag()1723 const char *SuppressInlineDefensiveChecksVisitor::getTag() {
1724   return "IDCVisitor";
1725 }
1726 
1727 PathDiagnosticPieceRef
VisitNode(const ExplodedNode * Succ,BugReporterContext & BRC,PathSensitiveBugReport & BR)1728 SuppressInlineDefensiveChecksVisitor::VisitNode(const ExplodedNode *Succ,
1729                                                 BugReporterContext &BRC,
1730                                                 PathSensitiveBugReport &BR) {
1731   const ExplodedNode *Pred = Succ->getFirstPred();
1732   if (IsSatisfied)
1733     return nullptr;
1734 
1735   // Start tracking after we see the first state in which the value is null.
1736   if (!IsTrackingTurnedOn)
1737     if (Succ->getState()->isNull(V).isConstrainedTrue())
1738       IsTrackingTurnedOn = true;
1739   if (!IsTrackingTurnedOn)
1740     return nullptr;
1741 
1742   // Check if in the previous state it was feasible for this value
1743   // to *not* be null.
1744   if (!Pred->getState()->isNull(V).isConstrainedTrue() &&
1745       Succ->getState()->isNull(V).isConstrainedTrue()) {
1746     IsSatisfied = true;
1747 
1748     // Check if this is inlined defensive checks.
1749     const LocationContext *CurLC = Succ->getLocationContext();
1750     const LocationContext *ReportLC = BR.getErrorNode()->getLocationContext();
1751     if (CurLC != ReportLC && !CurLC->isParentOf(ReportLC)) {
1752       BR.markInvalid("Suppress IDC", CurLC);
1753       return nullptr;
1754     }
1755 
1756     // Treat defensive checks in function-like macros as if they were an inlined
1757     // defensive check. If the bug location is not in a macro and the
1758     // terminator for the current location is in a macro then suppress the
1759     // warning.
1760     auto BugPoint = BR.getErrorNode()->getLocation().getAs<StmtPoint>();
1761 
1762     if (!BugPoint)
1763       return nullptr;
1764 
1765     ProgramPoint CurPoint = Succ->getLocation();
1766     const Stmt *CurTerminatorStmt = nullptr;
1767     if (auto BE = CurPoint.getAs<BlockEdge>()) {
1768       CurTerminatorStmt = BE->getSrc()->getTerminator().getStmt();
1769     } else if (auto SP = CurPoint.getAs<StmtPoint>()) {
1770       const Stmt *CurStmt = SP->getStmt();
1771       if (!CurStmt->getBeginLoc().isMacroID())
1772         return nullptr;
1773 
1774       CFGStmtMap *Map = CurLC->getAnalysisDeclContext()->getCFGStmtMap();
1775       CurTerminatorStmt = Map->getBlock(CurStmt)->getTerminatorStmt();
1776     } else {
1777       return nullptr;
1778     }
1779 
1780     if (!CurTerminatorStmt)
1781       return nullptr;
1782 
1783     SourceLocation TerminatorLoc = CurTerminatorStmt->getBeginLoc();
1784     if (TerminatorLoc.isMacroID()) {
1785       SourceLocation BugLoc = BugPoint->getStmt()->getBeginLoc();
1786 
1787       // Suppress reports unless we are in that same macro.
1788       if (!BugLoc.isMacroID() ||
1789           getMacroName(BugLoc, BRC) != getMacroName(TerminatorLoc, BRC)) {
1790         BR.markInvalid("Suppress Macro IDC", CurLC);
1791       }
1792       return nullptr;
1793     }
1794   }
1795   return nullptr;
1796 }
1797 
1798 //===----------------------------------------------------------------------===//
1799 // TrackControlDependencyCondBRVisitor.
1800 //===----------------------------------------------------------------------===//
1801 
1802 namespace {
1803 /// Tracks the expressions that are a control dependency of the node that was
1804 /// supplied to the constructor.
1805 /// For example:
1806 ///
1807 ///   cond = 1;
1808 ///   if (cond)
1809 ///     10 / 0;
1810 ///
1811 /// An error is emitted at line 3. This visitor realizes that the branch
1812 /// on line 2 is a control dependency of line 3, and tracks it's condition via
1813 /// trackExpressionValue().
1814 class TrackControlDependencyCondBRVisitor final : public BugReporterVisitor {
1815   const ExplodedNode *Origin;
1816   ControlDependencyCalculator ControlDeps;
1817   llvm::SmallSet<const CFGBlock *, 32> VisitedBlocks;
1818 
1819 public:
TrackControlDependencyCondBRVisitor(const ExplodedNode * O)1820   TrackControlDependencyCondBRVisitor(const ExplodedNode *O)
1821   : Origin(O), ControlDeps(&O->getCFG()) {}
1822 
Profile(llvm::FoldingSetNodeID & ID) const1823   void Profile(llvm::FoldingSetNodeID &ID) const override {
1824     static int x = 0;
1825     ID.AddPointer(&x);
1826   }
1827 
1828   PathDiagnosticPieceRef VisitNode(const ExplodedNode *N,
1829                                    BugReporterContext &BRC,
1830                                    PathSensitiveBugReport &BR) override;
1831 };
1832 } // end of anonymous namespace
1833 
1834 static std::shared_ptr<PathDiagnosticEventPiece>
constructDebugPieceForTrackedCondition(const Expr * Cond,const ExplodedNode * N,BugReporterContext & BRC)1835 constructDebugPieceForTrackedCondition(const Expr *Cond,
1836                                        const ExplodedNode *N,
1837                                        BugReporterContext &BRC) {
1838 
1839   if (BRC.getAnalyzerOptions().AnalysisDiagOpt == PD_NONE ||
1840       !BRC.getAnalyzerOptions().ShouldTrackConditionsDebug)
1841     return nullptr;
1842 
1843   std::string ConditionText = std::string(Lexer::getSourceText(
1844       CharSourceRange::getTokenRange(Cond->getSourceRange()),
1845       BRC.getSourceManager(), BRC.getASTContext().getLangOpts()));
1846 
1847   return std::make_shared<PathDiagnosticEventPiece>(
1848       PathDiagnosticLocation::createBegin(
1849           Cond, BRC.getSourceManager(), N->getLocationContext()),
1850           (Twine() + "Tracking condition '" + ConditionText + "'").str());
1851 }
1852 
isAssertlikeBlock(const CFGBlock * B,ASTContext & Context)1853 static bool isAssertlikeBlock(const CFGBlock *B, ASTContext &Context) {
1854   if (B->succ_size() != 2)
1855     return false;
1856 
1857   const CFGBlock *Then = B->succ_begin()->getReachableBlock();
1858   const CFGBlock *Else = (B->succ_begin() + 1)->getReachableBlock();
1859 
1860   if (!Then || !Else)
1861     return false;
1862 
1863   if (Then->isInevitablySinking() != Else->isInevitablySinking())
1864     return true;
1865 
1866   // For the following condition the following CFG would be built:
1867   //
1868   //                          ------------->
1869   //                         /              \
1870   //                       [B1] -> [B2] -> [B3] -> [sink]
1871   // assert(A && B || C);            \       \
1872   //                                  -----------> [go on with the execution]
1873   //
1874   // It so happens that CFGBlock::getTerminatorCondition returns 'A' for block
1875   // B1, 'A && B' for B2, and 'A && B || C' for B3. Let's check whether we
1876   // reached the end of the condition!
1877   if (const Stmt *ElseCond = Else->getTerminatorCondition())
1878     if (const auto *BinOp = dyn_cast<BinaryOperator>(ElseCond))
1879       if (BinOp->isLogicalOp())
1880         return isAssertlikeBlock(Else, Context);
1881 
1882   return false;
1883 }
1884 
1885 PathDiagnosticPieceRef
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)1886 TrackControlDependencyCondBRVisitor::VisitNode(const ExplodedNode *N,
1887                                                BugReporterContext &BRC,
1888                                                PathSensitiveBugReport &BR) {
1889   // We can only reason about control dependencies within the same stack frame.
1890   if (Origin->getStackFrame() != N->getStackFrame())
1891     return nullptr;
1892 
1893   CFGBlock *NB = const_cast<CFGBlock *>(N->getCFGBlock());
1894 
1895   // Skip if we already inspected this block.
1896   if (!VisitedBlocks.insert(NB).second)
1897     return nullptr;
1898 
1899   CFGBlock *OriginB = const_cast<CFGBlock *>(Origin->getCFGBlock());
1900 
1901   // TODO: Cache CFGBlocks for each ExplodedNode.
1902   if (!OriginB || !NB)
1903     return nullptr;
1904 
1905   if (isAssertlikeBlock(NB, BRC.getASTContext()))
1906     return nullptr;
1907 
1908   if (ControlDeps.isControlDependent(OriginB, NB)) {
1909     // We don't really want to explain for range loops. Evidence suggests that
1910     // the only thing that leads to is the addition of calls to operator!=.
1911     if (llvm::isa_and_nonnull<CXXForRangeStmt>(NB->getTerminatorStmt()))
1912       return nullptr;
1913 
1914     if (const Expr *Condition = NB->getLastCondition()) {
1915       // Keeping track of the already tracked conditions on a visitor level
1916       // isn't sufficient, because a new visitor is created for each tracked
1917       // expression, hence the BugReport level set.
1918       if (BR.addTrackedCondition(N)) {
1919         bugreporter::trackExpressionValue(
1920             N, Condition, BR, bugreporter::TrackingKind::Condition,
1921             /*EnableNullFPSuppression=*/false);
1922         return constructDebugPieceForTrackedCondition(Condition, N, BRC);
1923       }
1924     }
1925   }
1926 
1927   return nullptr;
1928 }
1929 
1930 //===----------------------------------------------------------------------===//
1931 // Implementation of trackExpressionValue.
1932 //===----------------------------------------------------------------------===//
1933 
1934 /// \return A subexpression of @c Ex which represents the
1935 /// expression-of-interest.
peelOffOuterExpr(const Expr * Ex,const ExplodedNode * N)1936 static const Expr *peelOffOuterExpr(const Expr *Ex,
1937                                     const ExplodedNode *N) {
1938   Ex = Ex->IgnoreParenCasts();
1939   if (const auto *FE = dyn_cast<FullExpr>(Ex))
1940     return peelOffOuterExpr(FE->getSubExpr(), N);
1941   if (const auto *OVE = dyn_cast<OpaqueValueExpr>(Ex))
1942     return peelOffOuterExpr(OVE->getSourceExpr(), N);
1943   if (const auto *POE = dyn_cast<PseudoObjectExpr>(Ex)) {
1944     const auto *PropRef = dyn_cast<ObjCPropertyRefExpr>(POE->getSyntacticForm());
1945     if (PropRef && PropRef->isMessagingGetter()) {
1946       const Expr *GetterMessageSend =
1947           POE->getSemanticExpr(POE->getNumSemanticExprs() - 1);
1948       assert(isa<ObjCMessageExpr>(GetterMessageSend->IgnoreParenCasts()));
1949       return peelOffOuterExpr(GetterMessageSend, N);
1950     }
1951   }
1952 
1953   // Peel off the ternary operator.
1954   if (const auto *CO = dyn_cast<ConditionalOperator>(Ex)) {
1955     // Find a node where the branching occurred and find out which branch
1956     // we took (true/false) by looking at the ExplodedGraph.
1957     const ExplodedNode *NI = N;
1958     do {
1959       ProgramPoint ProgPoint = NI->getLocation();
1960       if (Optional<BlockEdge> BE = ProgPoint.getAs<BlockEdge>()) {
1961         const CFGBlock *srcBlk = BE->getSrc();
1962         if (const Stmt *term = srcBlk->getTerminatorStmt()) {
1963           if (term == CO) {
1964             bool TookTrueBranch = (*(srcBlk->succ_begin()) == BE->getDst());
1965             if (TookTrueBranch)
1966               return peelOffOuterExpr(CO->getTrueExpr(), N);
1967             else
1968               return peelOffOuterExpr(CO->getFalseExpr(), N);
1969           }
1970         }
1971       }
1972       NI = NI->getFirstPred();
1973     } while (NI);
1974   }
1975 
1976   if (auto *BO = dyn_cast<BinaryOperator>(Ex))
1977     if (const Expr *SubEx = peelOffPointerArithmetic(BO))
1978       return peelOffOuterExpr(SubEx, N);
1979 
1980   if (auto *UO = dyn_cast<UnaryOperator>(Ex)) {
1981     if (UO->getOpcode() == UO_LNot)
1982       return peelOffOuterExpr(UO->getSubExpr(), N);
1983 
1984     // FIXME: There's a hack in our Store implementation that always computes
1985     // field offsets around null pointers as if they are always equal to 0.
1986     // The idea here is to report accesses to fields as null dereferences
1987     // even though the pointer value that's being dereferenced is actually
1988     // the offset of the field rather than exactly 0.
1989     // See the FIXME in StoreManager's getLValueFieldOrIvar() method.
1990     // This code interacts heavily with this hack; otherwise the value
1991     // would not be null at all for most fields, so we'd be unable to track it.
1992     if (UO->getOpcode() == UO_AddrOf && UO->getSubExpr()->isLValue())
1993       if (const Expr *DerefEx = bugreporter::getDerefExpr(UO->getSubExpr()))
1994         return peelOffOuterExpr(DerefEx, N);
1995   }
1996 
1997   return Ex;
1998 }
1999 
2000 /// Find the ExplodedNode where the lvalue (the value of 'Ex')
2001 /// was computed.
findNodeForExpression(const ExplodedNode * N,const Expr * Inner)2002 static const ExplodedNode* findNodeForExpression(const ExplodedNode *N,
2003                                                  const Expr *Inner) {
2004   while (N) {
2005     if (N->getStmtForDiagnostics() == Inner)
2006       return N;
2007     N = N->getFirstPred();
2008   }
2009   return N;
2010 }
2011 
2012 /// Attempts to add visitors to track an RValue expression back to its point of
2013 /// origin. Works similarly to trackExpressionValue, but accepts only RValues.
trackRValueExpression(const ExplodedNode * InputNode,const Expr * E,PathSensitiveBugReport & report,bugreporter::TrackingKind TKind,bool EnableNullFPSuppression)2014 static void trackRValueExpression(const ExplodedNode *InputNode, const Expr *E,
2015                                   PathSensitiveBugReport &report,
2016                                   bugreporter::TrackingKind TKind,
2017                                   bool EnableNullFPSuppression) {
2018   assert(E->isRValue() && "The expression is not an rvalue!");
2019   const ExplodedNode *RVNode = findNodeForExpression(InputNode, E);
2020   if (!RVNode)
2021     return;
2022   ProgramStateRef RVState = RVNode->getState();
2023   SVal V = RVState->getSValAsScalarOrLoc(E, RVNode->getLocationContext());
2024   const auto *BO = dyn_cast<BinaryOperator>(E);
2025   if (!BO)
2026     return;
2027   if (!V.isZeroConstant())
2028     return;
2029   if (!BO->isMultiplicativeOp())
2030     return;
2031 
2032   SVal RHSV = RVState->getSVal(BO->getRHS(), RVNode->getLocationContext());
2033   SVal LHSV = RVState->getSVal(BO->getLHS(), RVNode->getLocationContext());
2034 
2035   // Track both LHS and RHS of a multiplication.
2036   if (BO->getOpcode() == BO_Mul) {
2037     if (LHSV.isZeroConstant())
2038       trackExpressionValue(InputNode, BO->getLHS(), report, TKind,
2039                            EnableNullFPSuppression);
2040     if (RHSV.isZeroConstant())
2041       trackExpressionValue(InputNode, BO->getRHS(), report, TKind,
2042                            EnableNullFPSuppression);
2043   } else { // Track only the LHS of a division or a modulo.
2044     if (LHSV.isZeroConstant())
2045       trackExpressionValue(InputNode, BO->getLHS(), report, TKind,
2046                            EnableNullFPSuppression);
2047   }
2048 }
2049 
trackExpressionValue(const ExplodedNode * InputNode,const Expr * E,PathSensitiveBugReport & report,bugreporter::TrackingKind TKind,bool EnableNullFPSuppression)2050 bool bugreporter::trackExpressionValue(const ExplodedNode *InputNode,
2051                                        const Expr *E,
2052                                        PathSensitiveBugReport &report,
2053                                        bugreporter::TrackingKind TKind,
2054                                        bool EnableNullFPSuppression) {
2055 
2056   if (!E || !InputNode)
2057     return false;
2058 
2059   const Expr *Inner = peelOffOuterExpr(E, InputNode);
2060   const ExplodedNode *LVNode = findNodeForExpression(InputNode, Inner);
2061   if (!LVNode)
2062     return false;
2063 
2064   ProgramStateRef LVState = LVNode->getState();
2065   const StackFrameContext *SFC = LVNode->getStackFrame();
2066 
2067   // We only track expressions if we believe that they are important. Chances
2068   // are good that control dependencies to the tracking point are also important
2069   // because of this, let's explain why we believe control reached this point.
2070   // TODO: Shouldn't we track control dependencies of every bug location, rather
2071   // than only tracked expressions?
2072   if (LVState->getAnalysisManager().getAnalyzerOptions().ShouldTrackConditions)
2073     report.addVisitor(std::make_unique<TrackControlDependencyCondBRVisitor>(
2074           InputNode));
2075 
2076   // The message send could be nil due to the receiver being nil.
2077   // At this point in the path, the receiver should be live since we are at the
2078   // message send expr. If it is nil, start tracking it.
2079   if (const Expr *Receiver = NilReceiverBRVisitor::getNilReceiver(Inner, LVNode))
2080     trackExpressionValue(
2081         LVNode, Receiver, report, TKind, EnableNullFPSuppression);
2082 
2083   // Track the index if this is an array subscript.
2084   if (const auto *Arr = dyn_cast<ArraySubscriptExpr>(Inner))
2085     trackExpressionValue(
2086         LVNode, Arr->getIdx(), report, TKind, /*EnableNullFPSuppression*/false);
2087 
2088   // See if the expression we're interested refers to a variable.
2089   // If so, we can track both its contents and constraints on its value.
2090   if (ExplodedGraph::isInterestingLValueExpr(Inner)) {
2091     SVal LVal = LVNode->getSVal(Inner);
2092 
2093     const MemRegion *RR = getLocationRegionIfReference(Inner, LVNode);
2094     bool LVIsNull = LVState->isNull(LVal).isConstrainedTrue();
2095 
2096     // If this is a C++ reference to a null pointer, we are tracking the
2097     // pointer. In addition, we should find the store at which the reference
2098     // got initialized.
2099     if (RR && !LVIsNull)
2100       if (auto KV = LVal.getAs<KnownSVal>())
2101         report.addVisitor(std::make_unique<FindLastStoreBRVisitor>(
2102             *KV, RR, EnableNullFPSuppression, TKind, SFC));
2103 
2104     // In case of C++ references, we want to differentiate between a null
2105     // reference and reference to null pointer.
2106     // If the LVal is null, check if we are dealing with null reference.
2107     // For those, we want to track the location of the reference.
2108     const MemRegion *R = (RR && LVIsNull) ? RR :
2109         LVNode->getSVal(Inner).getAsRegion();
2110 
2111     if (R) {
2112 
2113       // Mark both the variable region and its contents as interesting.
2114       SVal V = LVState->getRawSVal(loc::MemRegionVal(R));
2115       report.addVisitor(
2116           std::make_unique<NoStoreFuncVisitor>(cast<SubRegion>(R), TKind));
2117 
2118       MacroNullReturnSuppressionVisitor::addMacroVisitorIfNecessary(
2119           LVNode, R, EnableNullFPSuppression, report, V);
2120 
2121       report.markInteresting(V, TKind);
2122       report.addVisitor(std::make_unique<UndefOrNullArgVisitor>(R));
2123 
2124       // If the contents are symbolic and null, find out when they became null.
2125       if (V.getAsLocSymbol(/*IncludeBaseRegions=*/true))
2126         if (LVState->isNull(V).isConstrainedTrue())
2127           report.addVisitor(std::make_unique<TrackConstraintBRVisitor>(
2128               V.castAs<DefinedSVal>(), false));
2129 
2130       // Add visitor, which will suppress inline defensive checks.
2131       if (auto DV = V.getAs<DefinedSVal>())
2132         if (!DV->isZeroConstant() && EnableNullFPSuppression) {
2133           // Note that LVNode may be too late (i.e., too far from the InputNode)
2134           // because the lvalue may have been computed before the inlined call
2135           // was evaluated. InputNode may as well be too early here, because
2136           // the symbol is already dead; this, however, is fine because we can
2137           // still find the node in which it collapsed to null previously.
2138           report.addVisitor(
2139               std::make_unique<SuppressInlineDefensiveChecksVisitor>(
2140                   *DV, InputNode));
2141         }
2142 
2143       if (auto KV = V.getAs<KnownSVal>())
2144         report.addVisitor(std::make_unique<FindLastStoreBRVisitor>(
2145             *KV, R, EnableNullFPSuppression, TKind, SFC));
2146       return true;
2147     }
2148   }
2149 
2150   // If the expression is not an "lvalue expression", we can still
2151   // track the constraints on its contents.
2152   SVal V = LVState->getSValAsScalarOrLoc(Inner, LVNode->getLocationContext());
2153 
2154   ReturnVisitor::addVisitorIfNecessary(
2155     LVNode, Inner, report, EnableNullFPSuppression, TKind);
2156 
2157   // Is it a symbolic value?
2158   if (auto L = V.getAs<loc::MemRegionVal>()) {
2159     // FIXME: this is a hack for fixing a later crash when attempting to
2160     // dereference a void* pointer.
2161     // We should not try to dereference pointers at all when we don't care
2162     // what is written inside the pointer.
2163     bool CanDereference = true;
2164     if (const auto *SR = L->getRegionAs<SymbolicRegion>()) {
2165       if (SR->getSymbol()->getType()->getPointeeType()->isVoidType())
2166         CanDereference = false;
2167     } else if (L->getRegionAs<AllocaRegion>())
2168       CanDereference = false;
2169 
2170     // At this point we are dealing with the region's LValue.
2171     // However, if the rvalue is a symbolic region, we should track it as well.
2172     // Try to use the correct type when looking up the value.
2173     SVal RVal;
2174     if (ExplodedGraph::isInterestingLValueExpr(Inner))
2175       RVal = LVState->getRawSVal(L.getValue(), Inner->getType());
2176     else if (CanDereference)
2177       RVal = LVState->getSVal(L->getRegion());
2178 
2179     if (CanDereference) {
2180       report.addVisitor(
2181           std::make_unique<UndefOrNullArgVisitor>(L->getRegion()));
2182 
2183       if (auto KV = RVal.getAs<KnownSVal>())
2184         report.addVisitor(std::make_unique<FindLastStoreBRVisitor>(
2185             *KV, L->getRegion(), EnableNullFPSuppression, TKind, SFC));
2186     }
2187 
2188     const MemRegion *RegionRVal = RVal.getAsRegion();
2189     if (RegionRVal && isa<SymbolicRegion>(RegionRVal)) {
2190       report.markInteresting(RegionRVal, TKind);
2191       report.addVisitor(std::make_unique<TrackConstraintBRVisitor>(
2192             loc::MemRegionVal(RegionRVal), /*assumption=*/false));
2193     }
2194   }
2195 
2196   if (Inner->isRValue())
2197     trackRValueExpression(LVNode, Inner, report, TKind,
2198                           EnableNullFPSuppression);
2199 
2200   return true;
2201 }
2202 
2203 //===----------------------------------------------------------------------===//
2204 // Implementation of NulReceiverBRVisitor.
2205 //===----------------------------------------------------------------------===//
2206 
getNilReceiver(const Stmt * S,const ExplodedNode * N)2207 const Expr *NilReceiverBRVisitor::getNilReceiver(const Stmt *S,
2208                                                  const ExplodedNode *N) {
2209   const auto *ME = dyn_cast<ObjCMessageExpr>(S);
2210   if (!ME)
2211     return nullptr;
2212   if (const Expr *Receiver = ME->getInstanceReceiver()) {
2213     ProgramStateRef state = N->getState();
2214     SVal V = N->getSVal(Receiver);
2215     if (state->isNull(V).isConstrainedTrue())
2216       return Receiver;
2217   }
2218   return nullptr;
2219 }
2220 
2221 PathDiagnosticPieceRef
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)2222 NilReceiverBRVisitor::VisitNode(const ExplodedNode *N, BugReporterContext &BRC,
2223                                 PathSensitiveBugReport &BR) {
2224   Optional<PreStmt> P = N->getLocationAs<PreStmt>();
2225   if (!P)
2226     return nullptr;
2227 
2228   const Stmt *S = P->getStmt();
2229   const Expr *Receiver = getNilReceiver(S, N);
2230   if (!Receiver)
2231     return nullptr;
2232 
2233   llvm::SmallString<256> Buf;
2234   llvm::raw_svector_ostream OS(Buf);
2235 
2236   if (const auto *ME = dyn_cast<ObjCMessageExpr>(S)) {
2237     OS << "'";
2238     ME->getSelector().print(OS);
2239     OS << "' not called";
2240   }
2241   else {
2242     OS << "No method is called";
2243   }
2244   OS << " because the receiver is nil";
2245 
2246   // The receiver was nil, and hence the method was skipped.
2247   // Register a BugReporterVisitor to issue a message telling us how
2248   // the receiver was null.
2249   bugreporter::trackExpressionValue(
2250       N, Receiver, BR, bugreporter::TrackingKind::Thorough,
2251       /*EnableNullFPSuppression*/ false);
2252   // Issue a message saying that the method was skipped.
2253   PathDiagnosticLocation L(Receiver, BRC.getSourceManager(),
2254                                      N->getLocationContext());
2255   return std::make_shared<PathDiagnosticEventPiece>(L, OS.str());
2256 }
2257 
2258 //===----------------------------------------------------------------------===//
2259 // Visitor that tries to report interesting diagnostics from conditions.
2260 //===----------------------------------------------------------------------===//
2261 
2262 /// Return the tag associated with this visitor.  This tag will be used
2263 /// to make all PathDiagnosticPieces created by this visitor.
getTag()2264 const char *ConditionBRVisitor::getTag() { return "ConditionBRVisitor"; }
2265 
2266 PathDiagnosticPieceRef
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)2267 ConditionBRVisitor::VisitNode(const ExplodedNode *N, BugReporterContext &BRC,
2268                               PathSensitiveBugReport &BR) {
2269   auto piece = VisitNodeImpl(N, BRC, BR);
2270   if (piece) {
2271     piece->setTag(getTag());
2272     if (auto *ev = dyn_cast<PathDiagnosticEventPiece>(piece.get()))
2273       ev->setPrunable(true, /* override */ false);
2274   }
2275   return piece;
2276 }
2277 
2278 PathDiagnosticPieceRef
VisitNodeImpl(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)2279 ConditionBRVisitor::VisitNodeImpl(const ExplodedNode *N,
2280                                   BugReporterContext &BRC,
2281                                   PathSensitiveBugReport &BR) {
2282   ProgramPoint ProgPoint = N->getLocation();
2283   const std::pair<const ProgramPointTag *, const ProgramPointTag *> &Tags =
2284       ExprEngine::geteagerlyAssumeBinOpBifurcationTags();
2285 
2286   // If an assumption was made on a branch, it should be caught
2287   // here by looking at the state transition.
2288   if (Optional<BlockEdge> BE = ProgPoint.getAs<BlockEdge>()) {
2289     const CFGBlock *SrcBlock = BE->getSrc();
2290     if (const Stmt *Term = SrcBlock->getTerminatorStmt()) {
2291       // If the tag of the previous node is 'Eagerly Assume...' the current
2292       // 'BlockEdge' has the same constraint information. We do not want to
2293       // report the value as it is just an assumption on the predecessor node
2294       // which will be caught in the next VisitNode() iteration as a 'PostStmt'.
2295       const ProgramPointTag *PreviousNodeTag =
2296           N->getFirstPred()->getLocation().getTag();
2297       if (PreviousNodeTag == Tags.first || PreviousNodeTag == Tags.second)
2298         return nullptr;
2299 
2300       return VisitTerminator(Term, N, SrcBlock, BE->getDst(), BR, BRC);
2301     }
2302     return nullptr;
2303   }
2304 
2305   if (Optional<PostStmt> PS = ProgPoint.getAs<PostStmt>()) {
2306     const ProgramPointTag *CurrentNodeTag = PS->getTag();
2307     if (CurrentNodeTag != Tags.first && CurrentNodeTag != Tags.second)
2308       return nullptr;
2309 
2310     bool TookTrue = CurrentNodeTag == Tags.first;
2311     return VisitTrueTest(cast<Expr>(PS->getStmt()), BRC, BR, N, TookTrue);
2312   }
2313 
2314   return nullptr;
2315 }
2316 
VisitTerminator(const Stmt * Term,const ExplodedNode * N,const CFGBlock * srcBlk,const CFGBlock * dstBlk,PathSensitiveBugReport & R,BugReporterContext & BRC)2317 PathDiagnosticPieceRef ConditionBRVisitor::VisitTerminator(
2318     const Stmt *Term, const ExplodedNode *N, const CFGBlock *srcBlk,
2319     const CFGBlock *dstBlk, PathSensitiveBugReport &R,
2320     BugReporterContext &BRC) {
2321   const Expr *Cond = nullptr;
2322 
2323   // In the code below, Term is a CFG terminator and Cond is a branch condition
2324   // expression upon which the decision is made on this terminator.
2325   //
2326   // For example, in "if (x == 0)", the "if (x == 0)" statement is a terminator,
2327   // and "x == 0" is the respective condition.
2328   //
2329   // Another example: in "if (x && y)", we've got two terminators and two
2330   // conditions due to short-circuit nature of operator "&&":
2331   // 1. The "if (x && y)" statement is a terminator,
2332   //    and "y" is the respective condition.
2333   // 2. Also "x && ..." is another terminator,
2334   //    and "x" is its condition.
2335 
2336   switch (Term->getStmtClass()) {
2337   // FIXME: Stmt::SwitchStmtClass is worth handling, however it is a bit
2338   // more tricky because there are more than two branches to account for.
2339   default:
2340     return nullptr;
2341   case Stmt::IfStmtClass:
2342     Cond = cast<IfStmt>(Term)->getCond();
2343     break;
2344   case Stmt::ConditionalOperatorClass:
2345     Cond = cast<ConditionalOperator>(Term)->getCond();
2346     break;
2347   case Stmt::BinaryOperatorClass:
2348     // When we encounter a logical operator (&& or ||) as a CFG terminator,
2349     // then the condition is actually its LHS; otherwise, we'd encounter
2350     // the parent, such as if-statement, as a terminator.
2351     const auto *BO = cast<BinaryOperator>(Term);
2352     assert(BO->isLogicalOp() &&
2353            "CFG terminator is not a short-circuit operator!");
2354     Cond = BO->getLHS();
2355     break;
2356   }
2357 
2358   Cond = Cond->IgnoreParens();
2359 
2360   // However, when we encounter a logical operator as a branch condition,
2361   // then the condition is actually its RHS, because LHS would be
2362   // the condition for the logical operator terminator.
2363   while (const auto *InnerBO = dyn_cast<BinaryOperator>(Cond)) {
2364     if (!InnerBO->isLogicalOp())
2365       break;
2366     Cond = InnerBO->getRHS()->IgnoreParens();
2367   }
2368 
2369   assert(Cond);
2370   assert(srcBlk->succ_size() == 2);
2371   const bool TookTrue = *(srcBlk->succ_begin()) == dstBlk;
2372   return VisitTrueTest(Cond, BRC, R, N, TookTrue);
2373 }
2374 
2375 PathDiagnosticPieceRef
VisitTrueTest(const Expr * Cond,BugReporterContext & BRC,PathSensitiveBugReport & R,const ExplodedNode * N,bool TookTrue)2376 ConditionBRVisitor::VisitTrueTest(const Expr *Cond, BugReporterContext &BRC,
2377                                   PathSensitiveBugReport &R,
2378                                   const ExplodedNode *N, bool TookTrue) {
2379   ProgramStateRef CurrentState = N->getState();
2380   ProgramStateRef PrevState = N->getFirstPred()->getState();
2381   const LocationContext *LCtx = N->getLocationContext();
2382 
2383   // If the constraint information is changed between the current and the
2384   // previous program state we assuming the newly seen constraint information.
2385   // If we cannot evaluate the condition (and the constraints are the same)
2386   // the analyzer has no information about the value and just assuming it.
2387   bool IsAssuming =
2388       !BRC.getStateManager().haveEqualConstraints(CurrentState, PrevState) ||
2389       CurrentState->getSVal(Cond, LCtx).isUnknownOrUndef();
2390 
2391   // These will be modified in code below, but we need to preserve the original
2392   //  values in case we want to throw the generic message.
2393   const Expr *CondTmp = Cond;
2394   bool TookTrueTmp = TookTrue;
2395 
2396   while (true) {
2397     CondTmp = CondTmp->IgnoreParenCasts();
2398     switch (CondTmp->getStmtClass()) {
2399       default:
2400         break;
2401       case Stmt::BinaryOperatorClass:
2402         if (auto P = VisitTrueTest(Cond, cast<BinaryOperator>(CondTmp),
2403                                    BRC, R, N, TookTrueTmp, IsAssuming))
2404           return P;
2405         break;
2406       case Stmt::DeclRefExprClass:
2407         if (auto P = VisitTrueTest(Cond, cast<DeclRefExpr>(CondTmp),
2408                                    BRC, R, N, TookTrueTmp, IsAssuming))
2409           return P;
2410         break;
2411       case Stmt::MemberExprClass:
2412         if (auto P = VisitTrueTest(Cond, cast<MemberExpr>(CondTmp),
2413                                    BRC, R, N, TookTrueTmp, IsAssuming))
2414           return P;
2415         break;
2416       case Stmt::UnaryOperatorClass: {
2417         const auto *UO = cast<UnaryOperator>(CondTmp);
2418         if (UO->getOpcode() == UO_LNot) {
2419           TookTrueTmp = !TookTrueTmp;
2420           CondTmp = UO->getSubExpr();
2421           continue;
2422         }
2423         break;
2424       }
2425     }
2426     break;
2427   }
2428 
2429   // Condition too complex to explain? Just say something so that the user
2430   // knew we've made some path decision at this point.
2431   // If it is too complex and we know the evaluation of the condition do not
2432   // repeat the note from 'BugReporter.cpp'
2433   if (!IsAssuming)
2434     return nullptr;
2435 
2436   PathDiagnosticLocation Loc(Cond, BRC.getSourceManager(), LCtx);
2437   if (!Loc.isValid() || !Loc.asLocation().isValid())
2438     return nullptr;
2439 
2440   return std::make_shared<PathDiagnosticEventPiece>(
2441       Loc, TookTrue ? GenericTrueMessage : GenericFalseMessage);
2442 }
2443 
patternMatch(const Expr * Ex,const Expr * ParentEx,raw_ostream & Out,BugReporterContext & BRC,PathSensitiveBugReport & report,const ExplodedNode * N,Optional<bool> & prunable,bool IsSameFieldName)2444 bool ConditionBRVisitor::patternMatch(const Expr *Ex,
2445                                       const Expr *ParentEx,
2446                                       raw_ostream &Out,
2447                                       BugReporterContext &BRC,
2448                                       PathSensitiveBugReport &report,
2449                                       const ExplodedNode *N,
2450                                       Optional<bool> &prunable,
2451                                       bool IsSameFieldName) {
2452   const Expr *OriginalExpr = Ex;
2453   Ex = Ex->IgnoreParenCasts();
2454 
2455   if (isa<GNUNullExpr>(Ex) || isa<ObjCBoolLiteralExpr>(Ex) ||
2456       isa<CXXBoolLiteralExpr>(Ex) || isa<IntegerLiteral>(Ex) ||
2457       isa<FloatingLiteral>(Ex)) {
2458     // Use heuristics to determine if the expression is a macro
2459     // expanding to a literal and if so, use the macro's name.
2460     SourceLocation BeginLoc = OriginalExpr->getBeginLoc();
2461     SourceLocation EndLoc = OriginalExpr->getEndLoc();
2462     if (BeginLoc.isMacroID() && EndLoc.isMacroID()) {
2463       const SourceManager &SM = BRC.getSourceManager();
2464       const LangOptions &LO = BRC.getASTContext().getLangOpts();
2465       if (Lexer::isAtStartOfMacroExpansion(BeginLoc, SM, LO) &&
2466           Lexer::isAtEndOfMacroExpansion(EndLoc, SM, LO)) {
2467         CharSourceRange R = Lexer::getAsCharRange({BeginLoc, EndLoc}, SM, LO);
2468         Out << Lexer::getSourceText(R, SM, LO);
2469         return false;
2470       }
2471     }
2472   }
2473 
2474   if (const auto *DR = dyn_cast<DeclRefExpr>(Ex)) {
2475     const bool quotes = isa<VarDecl>(DR->getDecl());
2476     if (quotes) {
2477       Out << '\'';
2478       const LocationContext *LCtx = N->getLocationContext();
2479       const ProgramState *state = N->getState().get();
2480       if (const MemRegion *R = state->getLValue(cast<VarDecl>(DR->getDecl()),
2481                                                 LCtx).getAsRegion()) {
2482         if (report.isInteresting(R))
2483           prunable = false;
2484         else {
2485           const ProgramState *state = N->getState().get();
2486           SVal V = state->getSVal(R);
2487           if (report.isInteresting(V))
2488             prunable = false;
2489         }
2490       }
2491     }
2492     Out << DR->getDecl()->getDeclName().getAsString();
2493     if (quotes)
2494       Out << '\'';
2495     return quotes;
2496   }
2497 
2498   if (const auto *IL = dyn_cast<IntegerLiteral>(Ex)) {
2499     QualType OriginalTy = OriginalExpr->getType();
2500     if (OriginalTy->isPointerType()) {
2501       if (IL->getValue() == 0) {
2502         Out << "null";
2503         return false;
2504       }
2505     }
2506     else if (OriginalTy->isObjCObjectPointerType()) {
2507       if (IL->getValue() == 0) {
2508         Out << "nil";
2509         return false;
2510       }
2511     }
2512 
2513     Out << IL->getValue();
2514     return false;
2515   }
2516 
2517   if (const auto *ME = dyn_cast<MemberExpr>(Ex)) {
2518     if (!IsSameFieldName)
2519       Out << "field '" << ME->getMemberDecl()->getName() << '\'';
2520     else
2521       Out << '\''
2522           << Lexer::getSourceText(
2523                  CharSourceRange::getTokenRange(Ex->getSourceRange()),
2524                  BRC.getSourceManager(), BRC.getASTContext().getLangOpts(), 0)
2525           << '\'';
2526   }
2527 
2528   return false;
2529 }
2530 
VisitTrueTest(const Expr * Cond,const BinaryOperator * BExpr,BugReporterContext & BRC,PathSensitiveBugReport & R,const ExplodedNode * N,bool TookTrue,bool IsAssuming)2531 PathDiagnosticPieceRef ConditionBRVisitor::VisitTrueTest(
2532     const Expr *Cond, const BinaryOperator *BExpr, BugReporterContext &BRC,
2533     PathSensitiveBugReport &R, const ExplodedNode *N, bool TookTrue,
2534     bool IsAssuming) {
2535   bool shouldInvert = false;
2536   Optional<bool> shouldPrune;
2537 
2538   // Check if the field name of the MemberExprs is ambiguous. Example:
2539   // " 'a.d' is equal to 'h.d' " in 'test/Analysis/null-deref-path-notes.cpp'.
2540   bool IsSameFieldName = false;
2541   const auto *LhsME = dyn_cast<MemberExpr>(BExpr->getLHS()->IgnoreParenCasts());
2542   const auto *RhsME = dyn_cast<MemberExpr>(BExpr->getRHS()->IgnoreParenCasts());
2543 
2544   if (LhsME && RhsME)
2545     IsSameFieldName =
2546         LhsME->getMemberDecl()->getName() == RhsME->getMemberDecl()->getName();
2547 
2548   SmallString<128> LhsString, RhsString;
2549   {
2550     llvm::raw_svector_ostream OutLHS(LhsString), OutRHS(RhsString);
2551     const bool isVarLHS = patternMatch(BExpr->getLHS(), BExpr, OutLHS, BRC, R,
2552                                        N, shouldPrune, IsSameFieldName);
2553     const bool isVarRHS = patternMatch(BExpr->getRHS(), BExpr, OutRHS, BRC, R,
2554                                        N, shouldPrune, IsSameFieldName);
2555 
2556     shouldInvert = !isVarLHS && isVarRHS;
2557   }
2558 
2559   BinaryOperator::Opcode Op = BExpr->getOpcode();
2560 
2561   if (BinaryOperator::isAssignmentOp(Op)) {
2562     // For assignment operators, all that we care about is that the LHS
2563     // evaluates to "true" or "false".
2564     return VisitConditionVariable(LhsString, BExpr->getLHS(), BRC, R, N,
2565                                   TookTrue);
2566   }
2567 
2568   // For non-assignment operations, we require that we can understand
2569   // both the LHS and RHS.
2570   if (LhsString.empty() || RhsString.empty() ||
2571       !BinaryOperator::isComparisonOp(Op) || Op == BO_Cmp)
2572     return nullptr;
2573 
2574   // Should we invert the strings if the LHS is not a variable name?
2575   SmallString<256> buf;
2576   llvm::raw_svector_ostream Out(buf);
2577   Out << (IsAssuming ? "Assuming " : "")
2578       << (shouldInvert ? RhsString : LhsString) << " is ";
2579 
2580   // Do we need to invert the opcode?
2581   if (shouldInvert)
2582     switch (Op) {
2583       default: break;
2584       case BO_LT: Op = BO_GT; break;
2585       case BO_GT: Op = BO_LT; break;
2586       case BO_LE: Op = BO_GE; break;
2587       case BO_GE: Op = BO_LE; break;
2588     }
2589 
2590   if (!TookTrue)
2591     switch (Op) {
2592       case BO_EQ: Op = BO_NE; break;
2593       case BO_NE: Op = BO_EQ; break;
2594       case BO_LT: Op = BO_GE; break;
2595       case BO_GT: Op = BO_LE; break;
2596       case BO_LE: Op = BO_GT; break;
2597       case BO_GE: Op = BO_LT; break;
2598       default:
2599         return nullptr;
2600     }
2601 
2602   switch (Op) {
2603     case BO_EQ:
2604       Out << "equal to ";
2605       break;
2606     case BO_NE:
2607       Out << "not equal to ";
2608       break;
2609     default:
2610       Out << BinaryOperator::getOpcodeStr(Op) << ' ';
2611       break;
2612   }
2613 
2614   Out << (shouldInvert ? LhsString : RhsString);
2615   const LocationContext *LCtx = N->getLocationContext();
2616   const SourceManager &SM = BRC.getSourceManager();
2617 
2618   if (isVarAnInterestingCondition(BExpr->getLHS(), N, &R) ||
2619       isVarAnInterestingCondition(BExpr->getRHS(), N, &R))
2620     Out << WillBeUsedForACondition;
2621 
2622   // Convert 'field ...' to 'Field ...' if it is a MemberExpr.
2623   std::string Message = std::string(Out.str());
2624   Message[0] = toupper(Message[0]);
2625 
2626   // If we know the value create a pop-up note to the value part of 'BExpr'.
2627   if (!IsAssuming) {
2628     PathDiagnosticLocation Loc;
2629     if (!shouldInvert) {
2630       if (LhsME && LhsME->getMemberLoc().isValid())
2631         Loc = PathDiagnosticLocation(LhsME->getMemberLoc(), SM);
2632       else
2633         Loc = PathDiagnosticLocation(BExpr->getLHS(), SM, LCtx);
2634     } else {
2635       if (RhsME && RhsME->getMemberLoc().isValid())
2636         Loc = PathDiagnosticLocation(RhsME->getMemberLoc(), SM);
2637       else
2638         Loc = PathDiagnosticLocation(BExpr->getRHS(), SM, LCtx);
2639     }
2640 
2641     return std::make_shared<PathDiagnosticPopUpPiece>(Loc, Message);
2642   }
2643 
2644   PathDiagnosticLocation Loc(Cond, SM, LCtx);
2645   auto event = std::make_shared<PathDiagnosticEventPiece>(Loc, Message);
2646   if (shouldPrune.hasValue())
2647     event->setPrunable(shouldPrune.getValue());
2648   return event;
2649 }
2650 
VisitConditionVariable(StringRef LhsString,const Expr * CondVarExpr,BugReporterContext & BRC,PathSensitiveBugReport & report,const ExplodedNode * N,bool TookTrue)2651 PathDiagnosticPieceRef ConditionBRVisitor::VisitConditionVariable(
2652     StringRef LhsString, const Expr *CondVarExpr, BugReporterContext &BRC,
2653     PathSensitiveBugReport &report, const ExplodedNode *N, bool TookTrue) {
2654   // FIXME: If there's already a constraint tracker for this variable,
2655   // we shouldn't emit anything here (c.f. the double note in
2656   // test/Analysis/inlining/path-notes.c)
2657   SmallString<256> buf;
2658   llvm::raw_svector_ostream Out(buf);
2659   Out << "Assuming " << LhsString << " is ";
2660 
2661   if (!printValue(CondVarExpr, Out, N, TookTrue, /*IsAssuming=*/true))
2662     return nullptr;
2663 
2664   const LocationContext *LCtx = N->getLocationContext();
2665   PathDiagnosticLocation Loc(CondVarExpr, BRC.getSourceManager(), LCtx);
2666 
2667   if (isVarAnInterestingCondition(CondVarExpr, N, &report))
2668     Out << WillBeUsedForACondition;
2669 
2670   auto event = std::make_shared<PathDiagnosticEventPiece>(Loc, Out.str());
2671 
2672   if (isInterestingExpr(CondVarExpr, N, &report))
2673     event->setPrunable(false);
2674 
2675   return event;
2676 }
2677 
VisitTrueTest(const Expr * Cond,const DeclRefExpr * DRE,BugReporterContext & BRC,PathSensitiveBugReport & report,const ExplodedNode * N,bool TookTrue,bool IsAssuming)2678 PathDiagnosticPieceRef ConditionBRVisitor::VisitTrueTest(
2679     const Expr *Cond, const DeclRefExpr *DRE, BugReporterContext &BRC,
2680     PathSensitiveBugReport &report, const ExplodedNode *N, bool TookTrue,
2681     bool IsAssuming) {
2682   const auto *VD = dyn_cast<VarDecl>(DRE->getDecl());
2683   if (!VD)
2684     return nullptr;
2685 
2686   SmallString<256> Buf;
2687   llvm::raw_svector_ostream Out(Buf);
2688 
2689   Out << (IsAssuming ? "Assuming '" : "'") << VD->getDeclName() << "' is ";
2690 
2691   if (!printValue(DRE, Out, N, TookTrue, IsAssuming))
2692     return nullptr;
2693 
2694   const LocationContext *LCtx = N->getLocationContext();
2695 
2696   if (isVarAnInterestingCondition(DRE, N, &report))
2697     Out << WillBeUsedForACondition;
2698 
2699   // If we know the value create a pop-up note to the 'DRE'.
2700   if (!IsAssuming) {
2701     PathDiagnosticLocation Loc(DRE, BRC.getSourceManager(), LCtx);
2702     return std::make_shared<PathDiagnosticPopUpPiece>(Loc, Out.str());
2703   }
2704 
2705   PathDiagnosticLocation Loc(Cond, BRC.getSourceManager(), LCtx);
2706   auto event = std::make_shared<PathDiagnosticEventPiece>(Loc, Out.str());
2707 
2708   if (isInterestingExpr(DRE, N, &report))
2709     event->setPrunable(false);
2710 
2711   return std::move(event);
2712 }
2713 
VisitTrueTest(const Expr * Cond,const MemberExpr * ME,BugReporterContext & BRC,PathSensitiveBugReport & report,const ExplodedNode * N,bool TookTrue,bool IsAssuming)2714 PathDiagnosticPieceRef ConditionBRVisitor::VisitTrueTest(
2715     const Expr *Cond, const MemberExpr *ME, BugReporterContext &BRC,
2716     PathSensitiveBugReport &report, const ExplodedNode *N, bool TookTrue,
2717     bool IsAssuming) {
2718   SmallString<256> Buf;
2719   llvm::raw_svector_ostream Out(Buf);
2720 
2721   Out << (IsAssuming ? "Assuming field '" : "Field '")
2722       << ME->getMemberDecl()->getName() << "' is ";
2723 
2724   if (!printValue(ME, Out, N, TookTrue, IsAssuming))
2725     return nullptr;
2726 
2727   const LocationContext *LCtx = N->getLocationContext();
2728   PathDiagnosticLocation Loc;
2729 
2730   // If we know the value create a pop-up note to the member of the MemberExpr.
2731   if (!IsAssuming && ME->getMemberLoc().isValid())
2732     Loc = PathDiagnosticLocation(ME->getMemberLoc(), BRC.getSourceManager());
2733   else
2734     Loc = PathDiagnosticLocation(Cond, BRC.getSourceManager(), LCtx);
2735 
2736   if (!Loc.isValid() || !Loc.asLocation().isValid())
2737     return nullptr;
2738 
2739   if (isVarAnInterestingCondition(ME, N, &report))
2740     Out << WillBeUsedForACondition;
2741 
2742   // If we know the value create a pop-up note.
2743   if (!IsAssuming)
2744     return std::make_shared<PathDiagnosticPopUpPiece>(Loc, Out.str());
2745 
2746   auto event = std::make_shared<PathDiagnosticEventPiece>(Loc, Out.str());
2747   if (isInterestingExpr(ME, N, &report))
2748     event->setPrunable(false);
2749   return event;
2750 }
2751 
printValue(const Expr * CondVarExpr,raw_ostream & Out,const ExplodedNode * N,bool TookTrue,bool IsAssuming)2752 bool ConditionBRVisitor::printValue(const Expr *CondVarExpr, raw_ostream &Out,
2753                                     const ExplodedNode *N, bool TookTrue,
2754                                     bool IsAssuming) {
2755   QualType Ty = CondVarExpr->getType();
2756 
2757   if (Ty->isPointerType()) {
2758     Out << (TookTrue ? "non-null" : "null");
2759     return true;
2760   }
2761 
2762   if (Ty->isObjCObjectPointerType()) {
2763     Out << (TookTrue ? "non-nil" : "nil");
2764     return true;
2765   }
2766 
2767   if (!Ty->isIntegralOrEnumerationType())
2768     return false;
2769 
2770   Optional<const llvm::APSInt *> IntValue;
2771   if (!IsAssuming)
2772     IntValue = getConcreteIntegerValue(CondVarExpr, N);
2773 
2774   if (IsAssuming || !IntValue.hasValue()) {
2775     if (Ty->isBooleanType())
2776       Out << (TookTrue ? "true" : "false");
2777     else
2778       Out << (TookTrue ? "not equal to 0" : "0");
2779   } else {
2780     if (Ty->isBooleanType())
2781       Out << (IntValue.getValue()->getBoolValue() ? "true" : "false");
2782     else
2783       Out << *IntValue.getValue();
2784   }
2785 
2786   return true;
2787 }
2788 
2789 constexpr llvm::StringLiteral ConditionBRVisitor::GenericTrueMessage;
2790 constexpr llvm::StringLiteral ConditionBRVisitor::GenericFalseMessage;
2791 
isPieceMessageGeneric(const PathDiagnosticPiece * Piece)2792 bool ConditionBRVisitor::isPieceMessageGeneric(
2793     const PathDiagnosticPiece *Piece) {
2794   return Piece->getString() == GenericTrueMessage ||
2795          Piece->getString() == GenericFalseMessage;
2796 }
2797 
2798 //===----------------------------------------------------------------------===//
2799 // Implementation of LikelyFalsePositiveSuppressionBRVisitor.
2800 //===----------------------------------------------------------------------===//
2801 
finalizeVisitor(BugReporterContext & BRC,const ExplodedNode * N,PathSensitiveBugReport & BR)2802 void LikelyFalsePositiveSuppressionBRVisitor::finalizeVisitor(
2803     BugReporterContext &BRC, const ExplodedNode *N,
2804     PathSensitiveBugReport &BR) {
2805   // Here we suppress false positives coming from system headers. This list is
2806   // based on known issues.
2807   const AnalyzerOptions &Options = BRC.getAnalyzerOptions();
2808   const Decl *D = N->getLocationContext()->getDecl();
2809 
2810   if (AnalysisDeclContext::isInStdNamespace(D)) {
2811     // Skip reports within the 'std' namespace. Although these can sometimes be
2812     // the user's fault, we currently don't report them very well, and
2813     // Note that this will not help for any other data structure libraries, like
2814     // TR1, Boost, or llvm/ADT.
2815     if (Options.ShouldSuppressFromCXXStandardLibrary) {
2816       BR.markInvalid(getTag(), nullptr);
2817       return;
2818     } else {
2819       // If the complete 'std' suppression is not enabled, suppress reports
2820       // from the 'std' namespace that are known to produce false positives.
2821 
2822       // The analyzer issues a false use-after-free when std::list::pop_front
2823       // or std::list::pop_back are called multiple times because we cannot
2824       // reason about the internal invariants of the data structure.
2825       if (const auto *MD = dyn_cast<CXXMethodDecl>(D)) {
2826         const CXXRecordDecl *CD = MD->getParent();
2827         if (CD->getName() == "list") {
2828           BR.markInvalid(getTag(), nullptr);
2829           return;
2830         }
2831       }
2832 
2833       // The analyzer issues a false positive when the constructor of
2834       // std::__independent_bits_engine from algorithms is used.
2835       if (const auto *MD = dyn_cast<CXXConstructorDecl>(D)) {
2836         const CXXRecordDecl *CD = MD->getParent();
2837         if (CD->getName() == "__independent_bits_engine") {
2838           BR.markInvalid(getTag(), nullptr);
2839           return;
2840         }
2841       }
2842 
2843       for (const LocationContext *LCtx = N->getLocationContext(); LCtx;
2844            LCtx = LCtx->getParent()) {
2845         const auto *MD = dyn_cast<CXXMethodDecl>(LCtx->getDecl());
2846         if (!MD)
2847           continue;
2848 
2849         const CXXRecordDecl *CD = MD->getParent();
2850         // The analyzer issues a false positive on
2851         //   std::basic_string<uint8_t> v; v.push_back(1);
2852         // and
2853         //   std::u16string s; s += u'a';
2854         // because we cannot reason about the internal invariants of the
2855         // data structure.
2856         if (CD->getName() == "basic_string") {
2857           BR.markInvalid(getTag(), nullptr);
2858           return;
2859         }
2860 
2861         // The analyzer issues a false positive on
2862         //    std::shared_ptr<int> p(new int(1)); p = nullptr;
2863         // because it does not reason properly about temporary destructors.
2864         if (CD->getName() == "shared_ptr") {
2865           BR.markInvalid(getTag(), nullptr);
2866           return;
2867         }
2868       }
2869     }
2870   }
2871 
2872   // Skip reports within the sys/queue.h macros as we do not have the ability to
2873   // reason about data structure shapes.
2874   const SourceManager &SM = BRC.getSourceManager();
2875   FullSourceLoc Loc = BR.getLocation().asLocation();
2876   while (Loc.isMacroID()) {
2877     Loc = Loc.getSpellingLoc();
2878     if (SM.getFilename(Loc).endswith("sys/queue.h")) {
2879       BR.markInvalid(getTag(), nullptr);
2880       return;
2881     }
2882   }
2883 }
2884 
2885 //===----------------------------------------------------------------------===//
2886 // Implementation of UndefOrNullArgVisitor.
2887 //===----------------------------------------------------------------------===//
2888 
2889 PathDiagnosticPieceRef
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)2890 UndefOrNullArgVisitor::VisitNode(const ExplodedNode *N, BugReporterContext &BRC,
2891                                  PathSensitiveBugReport &BR) {
2892   ProgramStateRef State = N->getState();
2893   ProgramPoint ProgLoc = N->getLocation();
2894 
2895   // We are only interested in visiting CallEnter nodes.
2896   Optional<CallEnter> CEnter = ProgLoc.getAs<CallEnter>();
2897   if (!CEnter)
2898     return nullptr;
2899 
2900   // Check if one of the arguments is the region the visitor is tracking.
2901   CallEventManager &CEMgr = BRC.getStateManager().getCallEventManager();
2902   CallEventRef<> Call = CEMgr.getCaller(CEnter->getCalleeContext(), State);
2903   unsigned Idx = 0;
2904   ArrayRef<ParmVarDecl *> parms = Call->parameters();
2905 
2906   for (const auto ParamDecl : parms) {
2907     const MemRegion *ArgReg = Call->getArgSVal(Idx).getAsRegion();
2908     ++Idx;
2909 
2910     // Are we tracking the argument or its subregion?
2911     if ( !ArgReg || !R->isSubRegionOf(ArgReg->StripCasts()))
2912       continue;
2913 
2914     // Check the function parameter type.
2915     assert(ParamDecl && "Formal parameter has no decl?");
2916     QualType T = ParamDecl->getType();
2917 
2918     if (!(T->isAnyPointerType() || T->isReferenceType())) {
2919       // Function can only change the value passed in by address.
2920       continue;
2921     }
2922 
2923     // If it is a const pointer value, the function does not intend to
2924     // change the value.
2925     if (T->getPointeeType().isConstQualified())
2926       continue;
2927 
2928     // Mark the call site (LocationContext) as interesting if the value of the
2929     // argument is undefined or '0'/'NULL'.
2930     SVal BoundVal = State->getSVal(R);
2931     if (BoundVal.isUndef() || BoundVal.isZeroConstant()) {
2932       BR.markInteresting(CEnter->getCalleeContext());
2933       return nullptr;
2934     }
2935   }
2936   return nullptr;
2937 }
2938 
2939 //===----------------------------------------------------------------------===//
2940 // Implementation of FalsePositiveRefutationBRVisitor.
2941 //===----------------------------------------------------------------------===//
2942 
FalsePositiveRefutationBRVisitor()2943 FalsePositiveRefutationBRVisitor::FalsePositiveRefutationBRVisitor()
2944     : Constraints(ConstraintMap::Factory().getEmptyMap()) {}
2945 
finalizeVisitor(BugReporterContext & BRC,const ExplodedNode * EndPathNode,PathSensitiveBugReport & BR)2946 void FalsePositiveRefutationBRVisitor::finalizeVisitor(
2947     BugReporterContext &BRC, const ExplodedNode *EndPathNode,
2948     PathSensitiveBugReport &BR) {
2949   // Collect new constraints
2950   addConstraints(EndPathNode, /*OverwriteConstraintsOnExistingSyms=*/true);
2951 
2952   // Create a refutation manager
2953   llvm::SMTSolverRef RefutationSolver = llvm::CreateZ3Solver();
2954   ASTContext &Ctx = BRC.getASTContext();
2955 
2956   // Add constraints to the solver
2957   for (const auto &I : Constraints) {
2958     const SymbolRef Sym = I.first;
2959     auto RangeIt = I.second.begin();
2960 
2961     llvm::SMTExprRef SMTConstraints = SMTConv::getRangeExpr(
2962         RefutationSolver, Ctx, Sym, RangeIt->From(), RangeIt->To(),
2963         /*InRange=*/true);
2964     while ((++RangeIt) != I.second.end()) {
2965       SMTConstraints = RefutationSolver->mkOr(
2966           SMTConstraints, SMTConv::getRangeExpr(RefutationSolver, Ctx, Sym,
2967                                                 RangeIt->From(), RangeIt->To(),
2968                                                 /*InRange=*/true));
2969     }
2970 
2971     RefutationSolver->addConstraint(SMTConstraints);
2972   }
2973 
2974   // And check for satisfiability
2975   Optional<bool> IsSAT = RefutationSolver->check();
2976   if (!IsSAT.hasValue())
2977     return;
2978 
2979   if (!IsSAT.getValue())
2980     BR.markInvalid("Infeasible constraints", EndPathNode->getLocationContext());
2981 }
2982 
addConstraints(const ExplodedNode * N,bool OverwriteConstraintsOnExistingSyms)2983 void FalsePositiveRefutationBRVisitor::addConstraints(
2984     const ExplodedNode *N, bool OverwriteConstraintsOnExistingSyms) {
2985   // Collect new constraints
2986   ConstraintMap NewCs = getConstraintMap(N->getState());
2987   ConstraintMap::Factory &CF = N->getState()->get_context<ConstraintMap>();
2988 
2989   // Add constraints if we don't have them yet
2990   for (auto const &C : NewCs) {
2991     const SymbolRef &Sym = C.first;
2992     if (!Constraints.contains(Sym)) {
2993       // This symbol is new, just add the constraint.
2994       Constraints = CF.add(Constraints, Sym, C.second);
2995     } else if (OverwriteConstraintsOnExistingSyms) {
2996       // Overwrite the associated constraint of the Symbol.
2997       Constraints = CF.remove(Constraints, Sym);
2998       Constraints = CF.add(Constraints, Sym, C.second);
2999     }
3000   }
3001 }
3002 
VisitNode(const ExplodedNode * N,BugReporterContext &,PathSensitiveBugReport &)3003 PathDiagnosticPieceRef FalsePositiveRefutationBRVisitor::VisitNode(
3004     const ExplodedNode *N, BugReporterContext &, PathSensitiveBugReport &) {
3005   addConstraints(N, /*OverwriteConstraintsOnExistingSyms=*/false);
3006   return nullptr;
3007 }
3008 
Profile(llvm::FoldingSetNodeID & ID) const3009 void FalsePositiveRefutationBRVisitor::Profile(
3010     llvm::FoldingSetNodeID &ID) const {
3011   static int Tag = 0;
3012   ID.AddPointer(&Tag);
3013 }
3014 
3015 //===----------------------------------------------------------------------===//
3016 // Implementation of TagVisitor.
3017 //===----------------------------------------------------------------------===//
3018 
3019 int NoteTag::Kind = 0;
3020 
Profile(llvm::FoldingSetNodeID & ID) const3021 void TagVisitor::Profile(llvm::FoldingSetNodeID &ID) const {
3022   static int Tag = 0;
3023   ID.AddPointer(&Tag);
3024 }
3025 
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & R)3026 PathDiagnosticPieceRef TagVisitor::VisitNode(const ExplodedNode *N,
3027                                              BugReporterContext &BRC,
3028                                              PathSensitiveBugReport &R) {
3029   ProgramPoint PP = N->getLocation();
3030   const NoteTag *T = dyn_cast_or_null<NoteTag>(PP.getTag());
3031   if (!T)
3032     return nullptr;
3033 
3034   if (Optional<std::string> Msg = T->generateMessage(BRC, R)) {
3035     PathDiagnosticLocation Loc =
3036         PathDiagnosticLocation::create(PP, BRC.getSourceManager());
3037     auto Piece = std::make_shared<PathDiagnosticEventPiece>(Loc, *Msg);
3038     Piece->setPrunable(T->isPrunable());
3039     return Piece;
3040   }
3041 
3042   return nullptr;
3043 }
3044