1 // RetainCountDiagnostics.cpp - Checks for leaks and other issues -*- C++ -*--//
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 diagnostics for RetainCountChecker, which implements
10 //  a reference count checker for Core Foundation and Cocoa on (Mac OS X).
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "RetainCountDiagnostics.h"
15 #include "RetainCountChecker.h"
16 #include "llvm/ADT/STLExtras.h"
17 #include "llvm/ADT/SmallVector.h"
18 #include <optional>
19 
20 using namespace clang;
21 using namespace ento;
22 using namespace retaincountchecker;
23 
24 StringRef RefCountBug::bugTypeToName(RefCountBug::RefCountBugKind BT) {
25   switch (BT) {
26   case UseAfterRelease:
27     return "Use-after-release";
28   case ReleaseNotOwned:
29     return "Bad release";
30   case DeallocNotOwned:
31     return "-dealloc sent to non-exclusively owned object";
32   case FreeNotOwned:
33     return "freeing non-exclusively owned object";
34   case OverAutorelease:
35     return "Object autoreleased too many times";
36   case ReturnNotOwnedForOwned:
37     return "Method should return an owned object";
38   case LeakWithinFunction:
39     return "Leak";
40   case LeakAtReturn:
41     return "Leak of returned object";
42   }
43   llvm_unreachable("Unknown RefCountBugKind");
44 }
45 
46 StringRef RefCountBug::getDescription() const {
47   switch (BT) {
48   case UseAfterRelease:
49     return "Reference-counted object is used after it is released";
50   case ReleaseNotOwned:
51     return "Incorrect decrement of the reference count of an object that is "
52            "not owned at this point by the caller";
53   case DeallocNotOwned:
54     return "-dealloc sent to object that may be referenced elsewhere";
55   case FreeNotOwned:
56     return  "'free' called on an object that may be referenced elsewhere";
57   case OverAutorelease:
58     return "Object autoreleased too many times";
59   case ReturnNotOwnedForOwned:
60     return "Object with a +0 retain count returned to caller where a +1 "
61            "(owning) retain count is expected";
62   case LeakWithinFunction:
63   case LeakAtReturn:
64     return "";
65   }
66   llvm_unreachable("Unknown RefCountBugKind");
67 }
68 
69 RefCountBug::RefCountBug(CheckerNameRef Checker, RefCountBugKind BT)
70     : BugType(Checker, bugTypeToName(BT), categories::MemoryRefCount,
71               /*SuppressOnSink=*/BT == LeakWithinFunction ||
72                   BT == LeakAtReturn),
73       BT(BT) {}
74 
75 static bool isNumericLiteralExpression(const Expr *E) {
76   // FIXME: This set of cases was copied from SemaExprObjC.
77   return isa<IntegerLiteral, CharacterLiteral, FloatingLiteral,
78              ObjCBoolLiteralExpr, CXXBoolLiteralExpr>(E);
79 }
80 
81 /// If type represents a pointer to CXXRecordDecl,
82 /// and is not a typedef, return the decl name.
83 /// Otherwise, return the serialization of type.
84 static std::string getPrettyTypeName(QualType QT) {
85   QualType PT = QT->getPointeeType();
86   if (!PT.isNull() && !QT->getAs<TypedefType>())
87     if (const auto *RD = PT->getAsCXXRecordDecl())
88       return std::string(RD->getName());
89   return QT.getAsString();
90 }
91 
92 /// Write information about the type state change to @c os,
93 /// return whether the note should be generated.
94 static bool shouldGenerateNote(llvm::raw_string_ostream &os,
95                                const RefVal *PrevT,
96                                const RefVal &CurrV,
97                                bool DeallocSent) {
98   // Get the previous type state.
99   RefVal PrevV = *PrevT;
100 
101   // Specially handle -dealloc.
102   if (DeallocSent) {
103     // Determine if the object's reference count was pushed to zero.
104     assert(!PrevV.hasSameState(CurrV) && "The state should have changed.");
105     // We may not have transitioned to 'release' if we hit an error.
106     // This case is handled elsewhere.
107     if (CurrV.getKind() == RefVal::Released) {
108       assert(CurrV.getCombinedCounts() == 0);
109       os << "Object released by directly sending the '-dealloc' message";
110       return true;
111     }
112   }
113 
114   // Determine if the typestate has changed.
115   if (!PrevV.hasSameState(CurrV))
116     switch (CurrV.getKind()) {
117     case RefVal::Owned:
118     case RefVal::NotOwned:
119       if (PrevV.getCount() == CurrV.getCount()) {
120         // Did an autorelease message get sent?
121         if (PrevV.getAutoreleaseCount() == CurrV.getAutoreleaseCount())
122           return false;
123 
124         assert(PrevV.getAutoreleaseCount() < CurrV.getAutoreleaseCount());
125         os << "Object autoreleased";
126         return true;
127       }
128 
129       if (PrevV.getCount() > CurrV.getCount())
130         os << "Reference count decremented.";
131       else
132         os << "Reference count incremented.";
133 
134       if (unsigned Count = CurrV.getCount())
135         os << " The object now has a +" << Count << " retain count.";
136 
137       return true;
138 
139     case RefVal::Released:
140       if (CurrV.getIvarAccessHistory() ==
141               RefVal::IvarAccessHistory::ReleasedAfterDirectAccess &&
142           CurrV.getIvarAccessHistory() != PrevV.getIvarAccessHistory()) {
143         os << "Strong instance variable relinquished. ";
144       }
145       os << "Object released.";
146       return true;
147 
148     case RefVal::ReturnedOwned:
149       // Autoreleases can be applied after marking a node ReturnedOwned.
150       if (CurrV.getAutoreleaseCount())
151         return false;
152 
153       os << "Object returned to caller as an owning reference (single "
154             "retain count transferred to caller)";
155       return true;
156 
157     case RefVal::ReturnedNotOwned:
158       os << "Object returned to caller with a +0 retain count";
159       return true;
160 
161     default:
162       return false;
163     }
164   return true;
165 }
166 
167 /// Finds argument index of the out paramter in the call @c S
168 /// corresponding to the symbol @c Sym.
169 /// If none found, returns std::nullopt.
170 static std::optional<unsigned>
171 findArgIdxOfSymbol(ProgramStateRef CurrSt, const LocationContext *LCtx,
172                    SymbolRef &Sym, std::optional<CallEventRef<>> CE) {
173   if (!CE)
174     return std::nullopt;
175 
176   for (unsigned Idx = 0; Idx < (*CE)->getNumArgs(); Idx++)
177     if (const MemRegion *MR = (*CE)->getArgSVal(Idx).getAsRegion())
178       if (const auto *TR = dyn_cast<TypedValueRegion>(MR))
179         if (CurrSt->getSVal(MR, TR->getValueType()).getAsSymbol() == Sym)
180           return Idx;
181 
182   return std::nullopt;
183 }
184 
185 static std::optional<std::string> findMetaClassAlloc(const Expr *Callee) {
186   if (const auto *ME = dyn_cast<MemberExpr>(Callee)) {
187     if (ME->getMemberDecl()->getNameAsString() != "alloc")
188       return std::nullopt;
189     const Expr *This = ME->getBase()->IgnoreParenImpCasts();
190     if (const auto *DRE = dyn_cast<DeclRefExpr>(This)) {
191       const ValueDecl *VD = DRE->getDecl();
192       if (VD->getNameAsString() != "metaClass")
193         return std::nullopt;
194 
195       if (const auto *RD = dyn_cast<CXXRecordDecl>(VD->getDeclContext()))
196         return RD->getNameAsString();
197 
198     }
199   }
200   return std::nullopt;
201 }
202 
203 static std::string findAllocatedObjectName(const Stmt *S, QualType QT) {
204   if (const auto *CE = dyn_cast<CallExpr>(S))
205     if (auto Out = findMetaClassAlloc(CE->getCallee()))
206       return *Out;
207   return getPrettyTypeName(QT);
208 }
209 
210 static void generateDiagnosticsForCallLike(ProgramStateRef CurrSt,
211                                            const LocationContext *LCtx,
212                                            const RefVal &CurrV, SymbolRef &Sym,
213                                            const Stmt *S,
214                                            llvm::raw_string_ostream &os) {
215   CallEventManager &Mgr = CurrSt->getStateManager().getCallEventManager();
216   if (const CallExpr *CE = dyn_cast<CallExpr>(S)) {
217     // Get the name of the callee (if it is available)
218     // from the tracked SVal.
219     SVal X = CurrSt->getSValAsScalarOrLoc(CE->getCallee(), LCtx);
220     const FunctionDecl *FD = X.getAsFunctionDecl();
221 
222     // If failed, try to get it from AST.
223     if (!FD)
224       FD = dyn_cast<FunctionDecl>(CE->getCalleeDecl());
225 
226     if (const auto *MD = dyn_cast<CXXMethodDecl>(CE->getCalleeDecl())) {
227       os << "Call to method '" << MD->getQualifiedNameAsString() << '\'';
228     } else if (FD) {
229       os << "Call to function '" << FD->getQualifiedNameAsString() << '\'';
230     } else {
231       os << "function call";
232     }
233   } else if (isa<CXXNewExpr>(S)) {
234     os << "Operator 'new'";
235   } else {
236     assert(isa<ObjCMessageExpr>(S));
237     CallEventRef<ObjCMethodCall> Call = Mgr.getObjCMethodCall(
238         cast<ObjCMessageExpr>(S), CurrSt, LCtx, {nullptr, 0});
239 
240     switch (Call->getMessageKind()) {
241     case OCM_Message:
242       os << "Method";
243       break;
244     case OCM_PropertyAccess:
245       os << "Property";
246       break;
247     case OCM_Subscript:
248       os << "Subscript";
249       break;
250     }
251   }
252 
253   std::optional<CallEventRef<>> CE = Mgr.getCall(S, CurrSt, LCtx, {nullptr, 0});
254   auto Idx = findArgIdxOfSymbol(CurrSt, LCtx, Sym, CE);
255 
256   // If index is not found, we assume that the symbol was returned.
257   if (!Idx) {
258     os << " returns ";
259   } else {
260     os << " writes ";
261   }
262 
263   if (CurrV.getObjKind() == ObjKind::CF) {
264     os << "a Core Foundation object of type '" << Sym->getType() << "' with a ";
265   } else if (CurrV.getObjKind() == ObjKind::OS) {
266     os << "an OSObject of type '" << findAllocatedObjectName(S, Sym->getType())
267        << "' with a ";
268   } else if (CurrV.getObjKind() == ObjKind::Generalized) {
269     os << "an object of type '" << Sym->getType() << "' with a ";
270   } else {
271     assert(CurrV.getObjKind() == ObjKind::ObjC);
272     QualType T = Sym->getType();
273     if (!isa<ObjCObjectPointerType>(T)) {
274       os << "an Objective-C object with a ";
275     } else {
276       const ObjCObjectPointerType *PT = cast<ObjCObjectPointerType>(T);
277       os << "an instance of " << PT->getPointeeType() << " with a ";
278     }
279   }
280 
281   if (CurrV.isOwned()) {
282     os << "+1 retain count";
283   } else {
284     assert(CurrV.isNotOwned());
285     os << "+0 retain count";
286   }
287 
288   if (Idx) {
289     os << " into an out parameter '";
290     const ParmVarDecl *PVD = (*CE)->parameters()[*Idx];
291     PVD->getNameForDiagnostic(os, PVD->getASTContext().getPrintingPolicy(),
292                               /*Qualified=*/false);
293     os << "'";
294 
295     QualType RT = (*CE)->getResultType();
296     if (!RT.isNull() && !RT->isVoidType()) {
297       SVal RV = (*CE)->getReturnValue();
298       if (CurrSt->isNull(RV).isConstrainedTrue()) {
299         os << " (assuming the call returns zero)";
300       } else if (CurrSt->isNonNull(RV).isConstrainedTrue()) {
301         os << " (assuming the call returns non-zero)";
302       }
303 
304     }
305   }
306 }
307 
308 namespace clang {
309 namespace ento {
310 namespace retaincountchecker {
311 
312 class RefCountReportVisitor : public BugReporterVisitor {
313 protected:
314   SymbolRef Sym;
315 
316 public:
317   RefCountReportVisitor(SymbolRef sym) : Sym(sym) {}
318 
319   void Profile(llvm::FoldingSetNodeID &ID) const override {
320     static int x = 0;
321     ID.AddPointer(&x);
322     ID.AddPointer(Sym);
323   }
324 
325   PathDiagnosticPieceRef VisitNode(const ExplodedNode *N,
326                                    BugReporterContext &BRC,
327                                    PathSensitiveBugReport &BR) override;
328 
329   PathDiagnosticPieceRef getEndPath(BugReporterContext &BRC,
330                                     const ExplodedNode *N,
331                                     PathSensitiveBugReport &BR) override;
332 };
333 
334 class RefLeakReportVisitor : public RefCountReportVisitor {
335 public:
336   RefLeakReportVisitor(SymbolRef Sym, const MemRegion *LastBinding)
337       : RefCountReportVisitor(Sym), LastBinding(LastBinding) {}
338 
339   PathDiagnosticPieceRef getEndPath(BugReporterContext &BRC,
340                                     const ExplodedNode *N,
341                                     PathSensitiveBugReport &BR) override;
342 
343 private:
344   const MemRegion *LastBinding;
345 };
346 
347 } // end namespace retaincountchecker
348 } // end namespace ento
349 } // end namespace clang
350 
351 
352 /// Find the first node with the parent stack frame.
353 static const ExplodedNode *getCalleeNode(const ExplodedNode *Pred) {
354   const StackFrameContext *SC = Pred->getStackFrame();
355   if (SC->inTopFrame())
356     return nullptr;
357   const StackFrameContext *PC = SC->getParent()->getStackFrame();
358   if (!PC)
359     return nullptr;
360 
361   const ExplodedNode *N = Pred;
362   while (N && N->getStackFrame() != PC) {
363     N = N->getFirstPred();
364   }
365   return N;
366 }
367 
368 
369 /// Insert a diagnostic piece at function exit
370 /// if a function parameter is annotated as "os_consumed",
371 /// but it does not actually consume the reference.
372 static std::shared_ptr<PathDiagnosticEventPiece>
373 annotateConsumedSummaryMismatch(const ExplodedNode *N,
374                                 CallExitBegin &CallExitLoc,
375                                 const SourceManager &SM,
376                                 CallEventManager &CEMgr) {
377 
378   const ExplodedNode *CN = getCalleeNode(N);
379   if (!CN)
380     return nullptr;
381 
382   CallEventRef<> Call = CEMgr.getCaller(N->getStackFrame(), N->getState());
383 
384   std::string sbuf;
385   llvm::raw_string_ostream os(sbuf);
386   ArrayRef<const ParmVarDecl *> Parameters = Call->parameters();
387   for (unsigned I=0; I < Call->getNumArgs() && I < Parameters.size(); ++I) {
388     const ParmVarDecl *PVD = Parameters[I];
389 
390     if (!PVD->hasAttr<OSConsumedAttr>())
391       continue;
392 
393     if (SymbolRef SR = Call->getArgSVal(I).getAsLocSymbol()) {
394       const RefVal *CountBeforeCall = getRefBinding(CN->getState(), SR);
395       const RefVal *CountAtExit = getRefBinding(N->getState(), SR);
396 
397       if (!CountBeforeCall || !CountAtExit)
398         continue;
399 
400       unsigned CountBefore = CountBeforeCall->getCount();
401       unsigned CountAfter = CountAtExit->getCount();
402 
403       bool AsExpected = CountBefore > 0 && CountAfter == CountBefore - 1;
404       if (!AsExpected) {
405         os << "Parameter '";
406         PVD->getNameForDiagnostic(os, PVD->getASTContext().getPrintingPolicy(),
407                                   /*Qualified=*/false);
408         os << "' is marked as consuming, but the function did not consume "
409            << "the reference\n";
410       }
411     }
412   }
413 
414   if (os.str().empty())
415     return nullptr;
416 
417   PathDiagnosticLocation L = PathDiagnosticLocation::create(CallExitLoc, SM);
418   return std::make_shared<PathDiagnosticEventPiece>(L, os.str());
419 }
420 
421 /// Annotate the parameter at the analysis entry point.
422 static std::shared_ptr<PathDiagnosticEventPiece>
423 annotateStartParameter(const ExplodedNode *N, SymbolRef Sym,
424                        const SourceManager &SM) {
425   auto PP = N->getLocationAs<BlockEdge>();
426   if (!PP)
427     return nullptr;
428 
429   const CFGBlock *Src = PP->getSrc();
430   const RefVal *CurrT = getRefBinding(N->getState(), Sym);
431 
432   if (&Src->getParent()->getEntry() != Src || !CurrT ||
433       getRefBinding(N->getFirstPred()->getState(), Sym))
434     return nullptr;
435 
436   const auto *VR = cast<VarRegion>(cast<SymbolRegionValue>(Sym)->getRegion());
437   const auto *PVD = cast<ParmVarDecl>(VR->getDecl());
438   PathDiagnosticLocation L = PathDiagnosticLocation(PVD, SM);
439 
440   std::string s;
441   llvm::raw_string_ostream os(s);
442   os << "Parameter '" << PVD->getDeclName() << "' starts at +";
443   if (CurrT->getCount() == 1) {
444     os << "1, as it is marked as consuming";
445   } else {
446     assert(CurrT->getCount() == 0);
447     os << "0";
448   }
449   return std::make_shared<PathDiagnosticEventPiece>(L, os.str());
450 }
451 
452 PathDiagnosticPieceRef
453 RefCountReportVisitor::VisitNode(const ExplodedNode *N, BugReporterContext &BRC,
454                                  PathSensitiveBugReport &BR) {
455 
456   const auto &BT = static_cast<const RefCountBug&>(BR.getBugType());
457 
458   bool IsFreeUnowned = BT.getBugType() == RefCountBug::FreeNotOwned ||
459                        BT.getBugType() == RefCountBug::DeallocNotOwned;
460 
461   const SourceManager &SM = BRC.getSourceManager();
462   CallEventManager &CEMgr = BRC.getStateManager().getCallEventManager();
463   if (auto CE = N->getLocationAs<CallExitBegin>())
464     if (auto PD = annotateConsumedSummaryMismatch(N, *CE, SM, CEMgr))
465       return PD;
466 
467   if (auto PD = annotateStartParameter(N, Sym, SM))
468     return PD;
469 
470   // FIXME: We will eventually need to handle non-statement-based events
471   // (__attribute__((cleanup))).
472   if (!N->getLocation().getAs<StmtPoint>())
473     return nullptr;
474 
475   // Check if the type state has changed.
476   const ExplodedNode *PrevNode = N->getFirstPred();
477   ProgramStateRef PrevSt = PrevNode->getState();
478   ProgramStateRef CurrSt = N->getState();
479   const LocationContext *LCtx = N->getLocationContext();
480 
481   const RefVal* CurrT = getRefBinding(CurrSt, Sym);
482   if (!CurrT)
483     return nullptr;
484 
485   const RefVal &CurrV = *CurrT;
486   const RefVal *PrevT = getRefBinding(PrevSt, Sym);
487 
488   // Create a string buffer to constain all the useful things we want
489   // to tell the user.
490   std::string sbuf;
491   llvm::raw_string_ostream os(sbuf);
492 
493   if (PrevT && IsFreeUnowned && CurrV.isNotOwned() && PrevT->isOwned()) {
494     os << "Object is now not exclusively owned";
495     auto Pos = PathDiagnosticLocation::create(N->getLocation(), SM);
496     return std::make_shared<PathDiagnosticEventPiece>(Pos, os.str());
497   }
498 
499   // This is the allocation site since the previous node had no bindings
500   // for this symbol.
501   if (!PrevT) {
502     const Stmt *S = N->getLocation().castAs<StmtPoint>().getStmt();
503 
504     if (isa<ObjCIvarRefExpr>(S) &&
505         isSynthesizedAccessor(LCtx->getStackFrame())) {
506       S = LCtx->getStackFrame()->getCallSite();
507     }
508 
509     if (isa<ObjCArrayLiteral>(S)) {
510       os << "NSArray literal is an object with a +0 retain count";
511     } else if (isa<ObjCDictionaryLiteral>(S)) {
512       os << "NSDictionary literal is an object with a +0 retain count";
513     } else if (const ObjCBoxedExpr *BL = dyn_cast<ObjCBoxedExpr>(S)) {
514       if (isNumericLiteralExpression(BL->getSubExpr()))
515         os << "NSNumber literal is an object with a +0 retain count";
516       else {
517         const ObjCInterfaceDecl *BoxClass = nullptr;
518         if (const ObjCMethodDecl *Method = BL->getBoxingMethod())
519           BoxClass = Method->getClassInterface();
520 
521         // We should always be able to find the boxing class interface,
522         // but consider this future-proofing.
523         if (BoxClass) {
524           os << *BoxClass << " b";
525         } else {
526           os << "B";
527         }
528 
529         os << "oxed expression produces an object with a +0 retain count";
530       }
531     } else if (isa<ObjCIvarRefExpr>(S)) {
532       os << "Object loaded from instance variable";
533     } else {
534       generateDiagnosticsForCallLike(CurrSt, LCtx, CurrV, Sym, S, os);
535     }
536 
537     PathDiagnosticLocation Pos(S, SM, N->getLocationContext());
538     return std::make_shared<PathDiagnosticEventPiece>(Pos, os.str());
539   }
540 
541   // Gather up the effects that were performed on the object at this
542   // program point
543   bool DeallocSent = false;
544 
545   const ProgramPointTag *Tag = N->getLocation().getTag();
546 
547   if (Tag == &RetainCountChecker::getCastFailTag()) {
548     os << "Assuming dynamic cast returns null due to type mismatch";
549   }
550 
551   if (Tag == &RetainCountChecker::getDeallocSentTag()) {
552     // We only have summaries attached to nodes after evaluating CallExpr and
553     // ObjCMessageExprs.
554     const Stmt *S = N->getLocation().castAs<StmtPoint>().getStmt();
555 
556     if (const CallExpr *CE = dyn_cast<CallExpr>(S)) {
557       // Iterate through the parameter expressions and see if the symbol
558       // was ever passed as an argument.
559       unsigned i = 0;
560 
561       for (auto AI=CE->arg_begin(), AE=CE->arg_end(); AI!=AE; ++AI, ++i) {
562 
563         // Retrieve the value of the argument.  Is it the symbol
564         // we are interested in?
565         if (CurrSt->getSValAsScalarOrLoc(*AI, LCtx).getAsLocSymbol() != Sym)
566           continue;
567 
568         // We have an argument.  Get the effect!
569         DeallocSent = true;
570       }
571     } else if (const ObjCMessageExpr *ME = dyn_cast<ObjCMessageExpr>(S)) {
572       if (const Expr *receiver = ME->getInstanceReceiver()) {
573         if (CurrSt->getSValAsScalarOrLoc(receiver, LCtx)
574               .getAsLocSymbol() == Sym) {
575           // The symbol we are tracking is the receiver.
576           DeallocSent = true;
577         }
578       }
579     }
580   }
581 
582   if (!shouldGenerateNote(os, PrevT, CurrV, DeallocSent))
583     return nullptr;
584 
585   if (os.str().empty())
586     return nullptr; // We have nothing to say!
587 
588   const Stmt *S = N->getLocation().castAs<StmtPoint>().getStmt();
589   PathDiagnosticLocation Pos(S, BRC.getSourceManager(),
590                                 N->getLocationContext());
591   auto P = std::make_shared<PathDiagnosticEventPiece>(Pos, os.str());
592 
593   // Add the range by scanning the children of the statement for any bindings
594   // to Sym.
595   for (const Stmt *Child : S->children())
596     if (const Expr *Exp = dyn_cast_or_null<Expr>(Child))
597       if (CurrSt->getSValAsScalarOrLoc(Exp, LCtx).getAsLocSymbol() == Sym) {
598         P->addRange(Exp->getSourceRange());
599         break;
600       }
601 
602   return std::move(P);
603 }
604 
605 static std::optional<std::string> describeRegion(const MemRegion *MR) {
606   if (const auto *VR = dyn_cast_or_null<VarRegion>(MR))
607     return std::string(VR->getDecl()->getName());
608   // Once we support more storage locations for bindings,
609   // this would need to be improved.
610   return std::nullopt;
611 }
612 
613 using Bindings = llvm::SmallVector<std::pair<const MemRegion *, SVal>, 4>;
614 
615 namespace {
616 class VarBindingsCollector : public StoreManager::BindingsHandler {
617   SymbolRef Sym;
618   Bindings &Result;
619 
620 public:
621   VarBindingsCollector(SymbolRef Sym, Bindings &ToFill)
622       : Sym(Sym), Result(ToFill) {}
623 
624   bool HandleBinding(StoreManager &SMgr, Store Store, const MemRegion *R,
625                      SVal Val) override {
626     SymbolRef SymV = Val.getAsLocSymbol();
627     if (!SymV || SymV != Sym)
628       return true;
629 
630     if (isa<NonParamVarRegion>(R))
631       Result.emplace_back(R, Val);
632 
633     return true;
634   }
635 };
636 } // namespace
637 
638 Bindings getAllVarBindingsForSymbol(ProgramStateManager &Manager,
639                                     const ExplodedNode *Node, SymbolRef Sym) {
640   Bindings Result;
641   VarBindingsCollector Collector{Sym, Result};
642   while (Result.empty() && Node) {
643     Manager.iterBindings(Node->getState(), Collector);
644     Node = Node->getFirstPred();
645   }
646 
647   return Result;
648 }
649 
650 namespace {
651 // Find the first node in the current function context that referred to the
652 // tracked symbol and the memory location that value was stored to. Note, the
653 // value is only reported if the allocation occurred in the same function as
654 // the leak. The function can also return a location context, which should be
655 // treated as interesting.
656 struct AllocationInfo {
657   const ExplodedNode* N;
658   const MemRegion *R;
659   const LocationContext *InterestingMethodContext;
660   AllocationInfo(const ExplodedNode *InN,
661                  const MemRegion *InR,
662                  const LocationContext *InInterestingMethodContext) :
663     N(InN), R(InR), InterestingMethodContext(InInterestingMethodContext) {}
664 };
665 } // end anonymous namespace
666 
667 static AllocationInfo GetAllocationSite(ProgramStateManager &StateMgr,
668                                         const ExplodedNode *N, SymbolRef Sym) {
669   const ExplodedNode *AllocationNode = N;
670   const ExplodedNode *AllocationNodeInCurrentOrParentContext = N;
671   const MemRegion *FirstBinding = nullptr;
672   const LocationContext *LeakContext = N->getLocationContext();
673 
674   // The location context of the init method called on the leaked object, if
675   // available.
676   const LocationContext *InitMethodContext = nullptr;
677 
678   while (N) {
679     ProgramStateRef St = N->getState();
680     const LocationContext *NContext = N->getLocationContext();
681 
682     if (!getRefBinding(St, Sym))
683       break;
684 
685     StoreManager::FindUniqueBinding FB(Sym);
686     StateMgr.iterBindings(St, FB);
687 
688     if (FB) {
689       const MemRegion *R = FB.getRegion();
690       // Do not show local variables belonging to a function other than
691       // where the error is reported.
692       if (auto MR = dyn_cast<StackSpaceRegion>(R->getMemorySpace()))
693         if (MR->getStackFrame() == LeakContext->getStackFrame())
694           FirstBinding = R;
695     }
696 
697     // AllocationNode is the last node in which the symbol was tracked.
698     AllocationNode = N;
699 
700     // AllocationNodeInCurrentContext, is the last node in the current or
701     // parent context in which the symbol was tracked.
702     //
703     // Note that the allocation site might be in the parent context. For example,
704     // the case where an allocation happens in a block that captures a reference
705     // to it and that reference is overwritten/dropped by another call to
706     // the block.
707     if (NContext == LeakContext || NContext->isParentOf(LeakContext))
708       AllocationNodeInCurrentOrParentContext = N;
709 
710     // Find the last init that was called on the given symbol and store the
711     // init method's location context.
712     if (!InitMethodContext)
713       if (auto CEP = N->getLocation().getAs<CallEnter>()) {
714         const Stmt *CE = CEP->getCallExpr();
715         if (const auto *ME = dyn_cast_or_null<ObjCMessageExpr>(CE)) {
716           const Stmt *RecExpr = ME->getInstanceReceiver();
717           if (RecExpr) {
718             SVal RecV = St->getSVal(RecExpr, NContext);
719             if (ME->getMethodFamily() == OMF_init && RecV.getAsSymbol() == Sym)
720               InitMethodContext = CEP->getCalleeContext();
721           }
722         }
723       }
724 
725     N = N->getFirstPred();
726   }
727 
728   // If we are reporting a leak of the object that was allocated with alloc,
729   // mark its init method as interesting.
730   const LocationContext *InterestingMethodContext = nullptr;
731   if (InitMethodContext) {
732     const ProgramPoint AllocPP = AllocationNode->getLocation();
733     if (std::optional<StmtPoint> SP = AllocPP.getAs<StmtPoint>())
734       if (const ObjCMessageExpr *ME = SP->getStmtAs<ObjCMessageExpr>())
735         if (ME->getMethodFamily() == OMF_alloc)
736           InterestingMethodContext = InitMethodContext;
737   }
738 
739   // If allocation happened in a function different from the leak node context,
740   // do not report the binding.
741   assert(N && "Could not find allocation node");
742 
743   if (AllocationNodeInCurrentOrParentContext &&
744       AllocationNodeInCurrentOrParentContext->getLocationContext() !=
745       LeakContext)
746     FirstBinding = nullptr;
747 
748   return AllocationInfo(AllocationNodeInCurrentOrParentContext, FirstBinding,
749                         InterestingMethodContext);
750 }
751 
752 PathDiagnosticPieceRef
753 RefCountReportVisitor::getEndPath(BugReporterContext &BRC,
754                                   const ExplodedNode *EndN,
755                                   PathSensitiveBugReport &BR) {
756   BR.markInteresting(Sym);
757   return BugReporterVisitor::getDefaultEndPath(BRC, EndN, BR);
758 }
759 
760 PathDiagnosticPieceRef
761 RefLeakReportVisitor::getEndPath(BugReporterContext &BRC,
762                                  const ExplodedNode *EndN,
763                                  PathSensitiveBugReport &BR) {
764 
765   // Tell the BugReporterContext to report cases when the tracked symbol is
766   // assigned to different variables, etc.
767   BR.markInteresting(Sym);
768 
769   PathDiagnosticLocation L = cast<RefLeakReport>(BR).getEndOfPath();
770 
771   std::string sbuf;
772   llvm::raw_string_ostream os(sbuf);
773 
774   os << "Object leaked: ";
775 
776   std::optional<std::string> RegionDescription = describeRegion(LastBinding);
777   if (RegionDescription) {
778     os << "object allocated and stored into '" << *RegionDescription << '\'';
779   } else {
780     os << "allocated object of type '" << getPrettyTypeName(Sym->getType())
781        << "'";
782   }
783 
784   // Get the retain count.
785   const RefVal *RV = getRefBinding(EndN->getState(), Sym);
786   assert(RV);
787 
788   if (RV->getKind() == RefVal::ErrorLeakReturned) {
789     // FIXME: Per comments in rdar://6320065, "create" only applies to CF
790     // objects.  Only "copy", "alloc", "retain" and "new" transfer ownership
791     // to the caller for NS objects.
792     const Decl *D = &EndN->getCodeDecl();
793 
794     os << (isa<ObjCMethodDecl>(D) ? " is returned from a method "
795                                   : " is returned from a function ");
796 
797     if (D->hasAttr<CFReturnsNotRetainedAttr>()) {
798       os << "that is annotated as CF_RETURNS_NOT_RETAINED";
799     } else if (D->hasAttr<NSReturnsNotRetainedAttr>()) {
800       os << "that is annotated as NS_RETURNS_NOT_RETAINED";
801     } else if (D->hasAttr<OSReturnsNotRetainedAttr>()) {
802       os << "that is annotated as OS_RETURNS_NOT_RETAINED";
803     } else {
804       if (const ObjCMethodDecl *MD = dyn_cast<ObjCMethodDecl>(D)) {
805         if (BRC.getASTContext().getLangOpts().ObjCAutoRefCount) {
806           os << "managed by Automatic Reference Counting";
807         } else {
808           os << "whose name ('" << MD->getSelector().getAsString()
809              << "') does not start with "
810                 "'copy', 'mutableCopy', 'alloc' or 'new'."
811                 "  This violates the naming convention rules"
812                 " given in the Memory Management Guide for Cocoa";
813         }
814       } else {
815         const FunctionDecl *FD = cast<FunctionDecl>(D);
816         ObjKind K = RV->getObjKind();
817         if (K == ObjKind::ObjC || K == ObjKind::CF) {
818           os << "whose name ('" << *FD
819              << "') does not contain 'Copy' or 'Create'.  This violates the "
820                 "naming"
821                 " convention rules given in the Memory Management Guide for "
822                 "Core"
823                 " Foundation";
824         } else if (RV->getObjKind() == ObjKind::OS) {
825           std::string FuncName = FD->getNameAsString();
826           os << "whose name ('" << FuncName << "') starts with '"
827              << StringRef(FuncName).substr(0, 3) << "'";
828         }
829       }
830     }
831   } else {
832     os << " is not referenced later in this execution path and has a retain "
833           "count of +"
834        << RV->getCount();
835   }
836 
837   return std::make_shared<PathDiagnosticEventPiece>(L, os.str());
838 }
839 
840 RefCountReport::RefCountReport(const RefCountBug &D, const LangOptions &LOpts,
841                                ExplodedNode *n, SymbolRef sym, bool isLeak)
842     : PathSensitiveBugReport(D, D.getDescription(), n), Sym(sym),
843       isLeak(isLeak) {
844   if (!isLeak)
845     addVisitor<RefCountReportVisitor>(sym);
846 }
847 
848 RefCountReport::RefCountReport(const RefCountBug &D, const LangOptions &LOpts,
849                                ExplodedNode *n, SymbolRef sym,
850                                StringRef endText)
851     : PathSensitiveBugReport(D, D.getDescription(), endText, n) {
852 
853   addVisitor<RefCountReportVisitor>(sym);
854 }
855 
856 void RefLeakReport::deriveParamLocation(CheckerContext &Ctx) {
857   const SourceManager &SMgr = Ctx.getSourceManager();
858 
859   if (!Sym->getOriginRegion())
860     return;
861 
862   auto *Region = dyn_cast<DeclRegion>(Sym->getOriginRegion());
863   if (Region) {
864     const Decl *PDecl = Region->getDecl();
865     if (isa_and_nonnull<ParmVarDecl>(PDecl)) {
866       PathDiagnosticLocation ParamLocation =
867           PathDiagnosticLocation::create(PDecl, SMgr);
868       Location = ParamLocation;
869       UniqueingLocation = ParamLocation;
870       UniqueingDecl = Ctx.getLocationContext()->getDecl();
871     }
872   }
873 }
874 
875 void RefLeakReport::deriveAllocLocation(CheckerContext &Ctx) {
876   // Most bug reports are cached at the location where they occurred.
877   // With leaks, we want to unique them by the location where they were
878   // allocated, and only report a single path.  To do this, we need to find
879   // the allocation site of a piece of tracked memory, which we do via a
880   // call to GetAllocationSite.  This will walk the ExplodedGraph backwards.
881   // Note that this is *not* the trimmed graph; we are guaranteed, however,
882   // that all ancestor nodes that represent the allocation site have the
883   // same SourceLocation.
884   const ExplodedNode *AllocNode = nullptr;
885 
886   const SourceManager &SMgr = Ctx.getSourceManager();
887 
888   AllocationInfo AllocI =
889       GetAllocationSite(Ctx.getStateManager(), getErrorNode(), Sym);
890 
891   AllocNode = AllocI.N;
892   AllocFirstBinding = AllocI.R;
893   markInteresting(AllocI.InterestingMethodContext);
894 
895   // Get the SourceLocation for the allocation site.
896   // FIXME: This will crash the analyzer if an allocation comes from an
897   // implicit call (ex: a destructor call).
898   // (Currently there are no such allocations in Cocoa, though.)
899   AllocStmt = AllocNode->getStmtForDiagnostics();
900 
901   if (!AllocStmt) {
902     AllocFirstBinding = nullptr;
903     return;
904   }
905 
906   PathDiagnosticLocation AllocLocation = PathDiagnosticLocation::createBegin(
907       AllocStmt, SMgr, AllocNode->getLocationContext());
908   Location = AllocLocation;
909 
910   // Set uniqieing info, which will be used for unique the bug reports. The
911   // leaks should be uniqued on the allocation site.
912   UniqueingLocation = AllocLocation;
913   UniqueingDecl = AllocNode->getLocationContext()->getDecl();
914 }
915 
916 void RefLeakReport::createDescription(CheckerContext &Ctx) {
917   assert(Location.isValid() && UniqueingDecl && UniqueingLocation.isValid());
918   Description.clear();
919   llvm::raw_string_ostream os(Description);
920   os << "Potential leak of an object";
921 
922   std::optional<std::string> RegionDescription =
923       describeRegion(AllocBindingToReport);
924   if (RegionDescription) {
925     os << " stored into '" << *RegionDescription << '\'';
926   } else {
927 
928     // If we can't figure out the name, just supply the type information.
929     os << " of type '" << getPrettyTypeName(Sym->getType()) << "'";
930   }
931 }
932 
933 void RefLeakReport::findBindingToReport(CheckerContext &Ctx,
934                                         ExplodedNode *Node) {
935   if (!AllocFirstBinding)
936     // If we don't have any bindings, we won't be able to find any
937     // better binding to report.
938     return;
939 
940   // If the original region still contains the leaking symbol...
941   if (Node->getState()->getSVal(AllocFirstBinding).getAsSymbol() == Sym) {
942     // ...it is the best binding to report.
943     AllocBindingToReport = AllocFirstBinding;
944     return;
945   }
946 
947   // At this point, we know that the original region doesn't contain the leaking
948   // when the actual leak happens.  It means that it can be confusing for the
949   // user to see such description in the message.
950   //
951   // Let's consider the following example:
952   //   Object *Original = allocate(...);
953   //   Object *New = Original;
954   //   Original = allocate(...);
955   //   Original->release();
956   //
957   // Complaining about a leaking object "stored into Original" might cause a
958   // rightful confusion because 'Original' is actually released.
959   // We should complain about 'New' instead.
960   Bindings AllVarBindings =
961       getAllVarBindingsForSymbol(Ctx.getStateManager(), Node, Sym);
962 
963   // While looking for the last var bindings, we can still find
964   // `AllocFirstBinding` to be one of them.  In situations like this,
965   // it would still be the easiest case to explain to our users.
966   if (!AllVarBindings.empty() &&
967       llvm::count_if(AllVarBindings,
968                      [this](const std::pair<const MemRegion *, SVal> Binding) {
969                        return Binding.first == AllocFirstBinding;
970                      }) == 0) {
971     // Let's pick one of them at random (if there is something to pick from).
972     AllocBindingToReport = AllVarBindings[0].first;
973 
974     // Because 'AllocBindingToReport' is not the same as
975     // 'AllocFirstBinding', we need to explain how the leaking object
976     // got from one to another.
977     //
978     // NOTE: We use the actual SVal stored in AllocBindingToReport here because
979     //       trackStoredValue compares SVal's and it can get trickier for
980     //       something like derived regions if we want to construct SVal from
981     //       Sym. Instead, we take the value that is definitely stored in that
982     //       region, thus guaranteeing that trackStoredValue will work.
983     bugreporter::trackStoredValue(AllVarBindings[0].second.castAs<KnownSVal>(),
984                                   AllocBindingToReport, *this);
985   } else {
986     AllocBindingToReport = AllocFirstBinding;
987   }
988 }
989 
990 RefLeakReport::RefLeakReport(const RefCountBug &D, const LangOptions &LOpts,
991                              ExplodedNode *N, SymbolRef Sym,
992                              CheckerContext &Ctx)
993     : RefCountReport(D, LOpts, N, Sym, /*isLeak=*/true) {
994 
995   deriveAllocLocation(Ctx);
996   findBindingToReport(Ctx, N);
997 
998   if (!AllocFirstBinding)
999     deriveParamLocation(Ctx);
1000 
1001   createDescription(Ctx);
1002 
1003   addVisitor<RefLeakReportVisitor>(Sym, AllocBindingToReport);
1004 }
1005