xref: /freebsd/contrib/llvm-project/clang/lib/StaticAnalyzer/Core/BugReporterVisitors.cpp (revision 0fca6ea1d4eea4c934cfff25ac9ee8ad6fe95583)
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/STLExtras.h"
50 #include "llvm/ADT/SmallPtrSet.h"
51 #include "llvm/ADT/SmallString.h"
52 #include "llvm/ADT/SmallVector.h"
53 #include "llvm/ADT/StringExtras.h"
54 #include "llvm/ADT/StringRef.h"
55 #include "llvm/Support/Casting.h"
56 #include "llvm/Support/ErrorHandling.h"
57 #include "llvm/Support/raw_ostream.h"
58 #include <cassert>
59 #include <deque>
60 #include <memory>
61 #include <optional>
62 #include <stack>
63 #include <string>
64 #include <utility>
65 
66 using namespace clang;
67 using namespace ento;
68 using namespace bugreporter;
69 
70 //===----------------------------------------------------------------------===//
71 // Utility functions.
72 //===----------------------------------------------------------------------===//
73 
peelOffPointerArithmetic(const BinaryOperator * B)74 static const Expr *peelOffPointerArithmetic(const BinaryOperator *B) {
75   if (B->isAdditiveOp() && B->getType()->isPointerType()) {
76     if (B->getLHS()->getType()->isPointerType()) {
77       return B->getLHS();
78     } else if (B->getRHS()->getType()->isPointerType()) {
79       return B->getRHS();
80     }
81   }
82   return nullptr;
83 }
84 
85 /// \return A subexpression of @c Ex which represents the
86 /// expression-of-interest.
87 static const Expr *peelOffOuterExpr(const Expr *Ex, const ExplodedNode *N);
88 
89 /// Given that expression S represents a pointer that would be dereferenced,
90 /// try to find a sub-expression from which the pointer came from.
91 /// This is used for tracking down origins of a null or undefined value:
92 /// "this is null because that is null because that is null" etc.
93 /// We wipe away field and element offsets because they merely add offsets.
94 /// We also wipe away all casts except lvalue-to-rvalue casts, because the
95 /// latter represent an actual pointer dereference; however, we remove
96 /// the final lvalue-to-rvalue cast before returning from this function
97 /// because it demonstrates more clearly from where the pointer rvalue was
98 /// loaded. Examples:
99 ///   x->y.z      ==>  x (lvalue)
100 ///   foo()->y.z  ==>  foo() (rvalue)
getDerefExpr(const Stmt * S)101 const Expr *bugreporter::getDerefExpr(const Stmt *S) {
102   const auto *E = dyn_cast<Expr>(S);
103   if (!E)
104     return nullptr;
105 
106   while (true) {
107     if (const auto *CE = dyn_cast<CastExpr>(E)) {
108       if (CE->getCastKind() == CK_LValueToRValue) {
109         // This cast represents the load we're looking for.
110         break;
111       }
112       E = CE->getSubExpr();
113     } else if (const auto *B = dyn_cast<BinaryOperator>(E)) {
114       // Pointer arithmetic: '*(x + 2)' -> 'x') etc.
115       if (const Expr *Inner = peelOffPointerArithmetic(B)) {
116         E = Inner;
117       } else if (B->isAssignmentOp()) {
118         // Follow LHS of assignments: '*p = 404' -> 'p'.
119         E = B->getLHS();
120       } else {
121         // Probably more arithmetic can be pattern-matched here,
122         // but for now give up.
123         break;
124       }
125     } else if (const auto *U = dyn_cast<UnaryOperator>(E)) {
126       if (U->getOpcode() == UO_Deref || U->getOpcode() == UO_AddrOf ||
127           (U->isIncrementDecrementOp() && U->getType()->isPointerType())) {
128         // Operators '*' and '&' don't actually mean anything.
129         // We look at casts instead.
130         E = U->getSubExpr();
131       } else {
132         // Probably more arithmetic can be pattern-matched here,
133         // but for now give up.
134         break;
135       }
136     }
137     // Pattern match for a few useful cases: a[0], p->f, *p etc.
138     else if (const auto *ME = dyn_cast<MemberExpr>(E)) {
139       // This handles the case when the dereferencing of a member reference
140       // happens. This is needed, because the AST for dereferencing a
141       // member reference looks like the following:
142       // |-MemberExpr
143       //  `-DeclRefExpr
144       // Without this special case the notes would refer to the whole object
145       // (struct, class or union variable) instead of just the relevant member.
146 
147       if (ME->getMemberDecl()->getType()->isReferenceType())
148         break;
149       E = ME->getBase();
150     } else if (const auto *IvarRef = dyn_cast<ObjCIvarRefExpr>(E)) {
151       E = IvarRef->getBase();
152     } else if (const auto *AE = dyn_cast<ArraySubscriptExpr>(E)) {
153       E = AE->getBase();
154     } else if (const auto *PE = dyn_cast<ParenExpr>(E)) {
155       E = PE->getSubExpr();
156     } else if (const auto *FE = dyn_cast<FullExpr>(E)) {
157       E = FE->getSubExpr();
158     } else {
159       // Other arbitrary stuff.
160       break;
161     }
162   }
163 
164   // Special case: remove the final lvalue-to-rvalue cast, but do not recurse
165   // deeper into the sub-expression. This way we return the lvalue from which
166   // our pointer rvalue was loaded.
167   if (const auto *CE = dyn_cast<ImplicitCastExpr>(E))
168     if (CE->getCastKind() == CK_LValueToRValue)
169       E = CE->getSubExpr();
170 
171   return E;
172 }
173 
getVarDeclForExpression(const Expr * E)174 static const VarDecl *getVarDeclForExpression(const Expr *E) {
175   if (const auto *DR = dyn_cast<DeclRefExpr>(E))
176     return dyn_cast<VarDecl>(DR->getDecl());
177   return nullptr;
178 }
179 
180 static const MemRegion *
getLocationRegionIfReference(const Expr * E,const ExplodedNode * N,bool LookingForReference=true)181 getLocationRegionIfReference(const Expr *E, const ExplodedNode *N,
182                              bool LookingForReference = true) {
183   if (const auto *ME = dyn_cast<MemberExpr>(E)) {
184     // This handles null references from FieldRegions, for example:
185     //   struct Wrapper { int &ref; };
186     //   Wrapper w = { *(int *)0 };
187     //   w.ref = 1;
188     const Expr *Base = ME->getBase();
189     const VarDecl *VD = getVarDeclForExpression(Base);
190     if (!VD)
191       return nullptr;
192 
193     const auto *FD = dyn_cast<FieldDecl>(ME->getMemberDecl());
194     if (!FD)
195       return nullptr;
196 
197     if (FD->getType()->isReferenceType()) {
198       SVal StructSVal = N->getState()->getLValue(VD, N->getLocationContext());
199       return N->getState()->getLValue(FD, StructSVal).getAsRegion();
200     }
201     return nullptr;
202   }
203 
204   const VarDecl *VD = getVarDeclForExpression(E);
205   if (!VD)
206     return nullptr;
207   if (LookingForReference && !VD->getType()->isReferenceType())
208     return nullptr;
209   return N->getState()->getLValue(VD, N->getLocationContext()).getAsRegion();
210 }
211 
212 /// Comparing internal representations of symbolic values (via
213 /// SVal::operator==()) is a valid way to check if the value was updated,
214 /// unless it's a LazyCompoundVal that may have a different internal
215 /// representation every time it is loaded from the state. In this function we
216 /// do an approximate comparison for lazy compound values, checking that they
217 /// are the immediate snapshots of the tracked region's bindings within the
218 /// node's respective states but not really checking that these snapshots
219 /// actually contain the same set of bindings.
hasVisibleUpdate(const ExplodedNode * LeftNode,SVal LeftVal,const ExplodedNode * RightNode,SVal RightVal)220 static bool hasVisibleUpdate(const ExplodedNode *LeftNode, SVal LeftVal,
221                              const ExplodedNode *RightNode, SVal RightVal) {
222   if (LeftVal == RightVal)
223     return true;
224 
225   const auto LLCV = LeftVal.getAs<nonloc::LazyCompoundVal>();
226   if (!LLCV)
227     return false;
228 
229   const auto RLCV = RightVal.getAs<nonloc::LazyCompoundVal>();
230   if (!RLCV)
231     return false;
232 
233   return LLCV->getRegion() == RLCV->getRegion() &&
234     LLCV->getStore() == LeftNode->getState()->getStore() &&
235     RLCV->getStore() == RightNode->getState()->getStore();
236 }
237 
getSValForVar(const Expr * CondVarExpr,const ExplodedNode * N)238 static std::optional<SVal> getSValForVar(const Expr *CondVarExpr,
239                                          const ExplodedNode *N) {
240   ProgramStateRef State = N->getState();
241   const LocationContext *LCtx = N->getLocationContext();
242 
243   assert(CondVarExpr);
244   CondVarExpr = CondVarExpr->IgnoreImpCasts();
245 
246   // The declaration of the value may rely on a pointer so take its l-value.
247   // FIXME: As seen in VisitCommonDeclRefExpr, sometimes DeclRefExpr may
248   // evaluate to a FieldRegion when it refers to a declaration of a lambda
249   // capture variable. We most likely need to duplicate that logic here.
250   if (const auto *DRE = dyn_cast<DeclRefExpr>(CondVarExpr))
251     if (const auto *VD = dyn_cast<VarDecl>(DRE->getDecl()))
252       return State->getSVal(State->getLValue(VD, LCtx));
253 
254   if (const auto *ME = dyn_cast<MemberExpr>(CondVarExpr))
255     if (const auto *FD = dyn_cast<FieldDecl>(ME->getMemberDecl()))
256       if (auto FieldL = State->getSVal(ME, LCtx).getAs<Loc>())
257         return State->getRawSVal(*FieldL, FD->getType());
258 
259   return std::nullopt;
260 }
261 
262 static std::optional<const llvm::APSInt *>
getConcreteIntegerValue(const Expr * CondVarExpr,const ExplodedNode * N)263 getConcreteIntegerValue(const Expr *CondVarExpr, const ExplodedNode *N) {
264 
265   if (std::optional<SVal> V = getSValForVar(CondVarExpr, N))
266     if (auto CI = V->getAs<nonloc::ConcreteInt>())
267       return &CI->getValue();
268   return std::nullopt;
269 }
270 
isVarAnInterestingCondition(const Expr * CondVarExpr,const ExplodedNode * N,const PathSensitiveBugReport * B)271 static bool isVarAnInterestingCondition(const Expr *CondVarExpr,
272                                         const ExplodedNode *N,
273                                         const PathSensitiveBugReport *B) {
274   // Even if this condition is marked as interesting, it isn't *that*
275   // interesting if it didn't happen in a nested stackframe, the user could just
276   // follow the arrows.
277   if (!B->getErrorNode()->getStackFrame()->isParentOf(N->getStackFrame()))
278     return false;
279 
280   if (std::optional<SVal> V = getSValForVar(CondVarExpr, N))
281     if (std::optional<bugreporter::TrackingKind> K =
282             B->getInterestingnessKind(*V))
283       return *K == bugreporter::TrackingKind::Condition;
284 
285   return false;
286 }
287 
isInterestingExpr(const Expr * E,const ExplodedNode * N,const PathSensitiveBugReport * B)288 static bool isInterestingExpr(const Expr *E, const ExplodedNode *N,
289                               const PathSensitiveBugReport *B) {
290   if (std::optional<SVal> V = getSValForVar(E, N))
291     return B->getInterestingnessKind(*V).has_value();
292   return false;
293 }
294 
295 /// \return name of the macro inside the location \p Loc.
getMacroName(SourceLocation Loc,BugReporterContext & BRC)296 static StringRef getMacroName(SourceLocation Loc,
297     BugReporterContext &BRC) {
298   return Lexer::getImmediateMacroName(
299       Loc,
300       BRC.getSourceManager(),
301       BRC.getASTContext().getLangOpts());
302 }
303 
304 /// \return Whether given spelling location corresponds to an expansion
305 /// of a function-like macro.
isFunctionMacroExpansion(SourceLocation Loc,const SourceManager & SM)306 static bool isFunctionMacroExpansion(SourceLocation Loc,
307                                 const SourceManager &SM) {
308   if (!Loc.isMacroID())
309     return false;
310   while (SM.isMacroArgExpansion(Loc))
311     Loc = SM.getImmediateExpansionRange(Loc).getBegin();
312   std::pair<FileID, unsigned> TLInfo = SM.getDecomposedLoc(Loc);
313   SrcMgr::SLocEntry SE = SM.getSLocEntry(TLInfo.first);
314   const SrcMgr::ExpansionInfo &EInfo = SE.getExpansion();
315   return EInfo.isFunctionMacroExpansion();
316 }
317 
318 /// \return Whether \c RegionOfInterest was modified at \p N,
319 /// where \p ValueAfter is \c RegionOfInterest's value at the end of the
320 /// stack frame.
wasRegionOfInterestModifiedAt(const SubRegion * RegionOfInterest,const ExplodedNode * N,SVal ValueAfter)321 static bool wasRegionOfInterestModifiedAt(const SubRegion *RegionOfInterest,
322                                           const ExplodedNode *N,
323                                           SVal ValueAfter) {
324   ProgramStateRef State = N->getState();
325   ProgramStateManager &Mgr = N->getState()->getStateManager();
326 
327   if (!N->getLocationAs<PostStore>() && !N->getLocationAs<PostInitializer>() &&
328       !N->getLocationAs<PostStmt>())
329     return false;
330 
331   // Writing into region of interest.
332   if (auto PS = N->getLocationAs<PostStmt>())
333     if (auto *BO = PS->getStmtAs<BinaryOperator>())
334       if (BO->isAssignmentOp() && RegionOfInterest->isSubRegionOf(
335                                       N->getSVal(BO->getLHS()).getAsRegion()))
336         return true;
337 
338   // SVal after the state is possibly different.
339   SVal ValueAtN = N->getState()->getSVal(RegionOfInterest);
340   if (!Mgr.getSValBuilder()
341            .areEqual(State, ValueAtN, ValueAfter)
342            .isConstrainedTrue() &&
343       (!ValueAtN.isUndef() || !ValueAfter.isUndef()))
344     return true;
345 
346   return false;
347 }
348 
349 //===----------------------------------------------------------------------===//
350 // Implementation of BugReporterVisitor.
351 //===----------------------------------------------------------------------===//
352 
getEndPath(BugReporterContext &,const ExplodedNode *,PathSensitiveBugReport &)353 PathDiagnosticPieceRef BugReporterVisitor::getEndPath(BugReporterContext &,
354                                                       const ExplodedNode *,
355                                                       PathSensitiveBugReport &) {
356   return nullptr;
357 }
358 
finalizeVisitor(BugReporterContext &,const ExplodedNode *,PathSensitiveBugReport &)359 void BugReporterVisitor::finalizeVisitor(BugReporterContext &,
360                                          const ExplodedNode *,
361                                          PathSensitiveBugReport &) {}
362 
363 PathDiagnosticPieceRef
getDefaultEndPath(const BugReporterContext & BRC,const ExplodedNode * EndPathNode,const PathSensitiveBugReport & BR)364 BugReporterVisitor::getDefaultEndPath(const BugReporterContext &BRC,
365                                       const ExplodedNode *EndPathNode,
366                                       const PathSensitiveBugReport &BR) {
367   PathDiagnosticLocation L = BR.getLocation();
368   const auto &Ranges = BR.getRanges();
369 
370   // Only add the statement itself as a range if we didn't specify any
371   // special ranges for this report.
372   auto P = std::make_shared<PathDiagnosticEventPiece>(
373       L, BR.getDescription(), Ranges.begin() == Ranges.end());
374   for (SourceRange Range : Ranges)
375     P->addRange(Range);
376 
377   return P;
378 }
379 
380 //===----------------------------------------------------------------------===//
381 // Implementation of NoStateChangeFuncVisitor.
382 //===----------------------------------------------------------------------===//
383 
isModifiedInFrame(const ExplodedNode * N)384 bool NoStateChangeFuncVisitor::isModifiedInFrame(const ExplodedNode *N) {
385   const LocationContext *Ctx = N->getLocationContext();
386   const StackFrameContext *SCtx = Ctx->getStackFrame();
387   if (!FramesModifyingCalculated.count(SCtx))
388     findModifyingFrames(N);
389   return FramesModifying.count(SCtx);
390 }
391 
markFrameAsModifying(const StackFrameContext * SCtx)392 void NoStateChangeFuncVisitor::markFrameAsModifying(
393     const StackFrameContext *SCtx) {
394   while (!SCtx->inTopFrame()) {
395     auto p = FramesModifying.insert(SCtx);
396     if (!p.second)
397       break; // Frame and all its parents already inserted.
398 
399     SCtx = SCtx->getParent()->getStackFrame();
400   }
401 }
402 
getMatchingCallExitEnd(const ExplodedNode * N)403 static const ExplodedNode *getMatchingCallExitEnd(const ExplodedNode *N) {
404   assert(N->getLocationAs<CallEnter>());
405   // The stackframe of the callee is only found in the nodes succeeding
406   // the CallEnter node. CallEnter's stack frame refers to the caller.
407   const StackFrameContext *OrigSCtx = N->getFirstSucc()->getStackFrame();
408 
409   // Similarly, the nodes preceding CallExitEnd refer to the callee's stack
410   // frame.
411   auto IsMatchingCallExitEnd = [OrigSCtx](const ExplodedNode *N) {
412     return N->getLocationAs<CallExitEnd>() &&
413            OrigSCtx == N->getFirstPred()->getStackFrame();
414   };
415   while (N && !IsMatchingCallExitEnd(N)) {
416     assert(N->succ_size() <= 1 &&
417            "This function is to be used on the trimmed ExplodedGraph!");
418     N = N->getFirstSucc();
419   }
420   return N;
421 }
422 
findModifyingFrames(const ExplodedNode * const CallExitBeginN)423 void NoStateChangeFuncVisitor::findModifyingFrames(
424     const ExplodedNode *const CallExitBeginN) {
425 
426   assert(CallExitBeginN->getLocationAs<CallExitBegin>());
427 
428   const StackFrameContext *const OriginalSCtx =
429       CallExitBeginN->getLocationContext()->getStackFrame();
430 
431   const ExplodedNode *CurrCallExitBeginN = CallExitBeginN;
432   const StackFrameContext *CurrentSCtx = OriginalSCtx;
433 
434   for (const ExplodedNode *CurrN = CallExitBeginN; CurrN;
435        CurrN = CurrN->getFirstPred()) {
436     // Found a new inlined call.
437     if (CurrN->getLocationAs<CallExitBegin>()) {
438       CurrCallExitBeginN = CurrN;
439       CurrentSCtx = CurrN->getStackFrame();
440       FramesModifyingCalculated.insert(CurrentSCtx);
441       // We won't see a change in between two identical exploded nodes: skip.
442       continue;
443     }
444 
445     if (auto CE = CurrN->getLocationAs<CallEnter>()) {
446       if (const ExplodedNode *CallExitEndN = getMatchingCallExitEnd(CurrN))
447         if (wasModifiedInFunction(CurrN, CallExitEndN))
448           markFrameAsModifying(CurrentSCtx);
449 
450       // We exited this inlined call, lets actualize the stack frame.
451       CurrentSCtx = CurrN->getStackFrame();
452 
453       // Stop calculating at the current function, but always regard it as
454       // modifying, so we can avoid notes like this:
455       //   void f(Foo &F) {
456       //     F.field = 0; // note: 0 assigned to 'F.field'
457       //                  // note: returning without writing to 'F.field'
458       //   }
459       if (CE->getCalleeContext() == OriginalSCtx) {
460         markFrameAsModifying(CurrentSCtx);
461         break;
462       }
463     }
464 
465     if (wasModifiedBeforeCallExit(CurrN, CurrCallExitBeginN))
466       markFrameAsModifying(CurrentSCtx);
467   }
468 }
469 
VisitNode(const ExplodedNode * N,BugReporterContext & BR,PathSensitiveBugReport & R)470 PathDiagnosticPieceRef NoStateChangeFuncVisitor::VisitNode(
471     const ExplodedNode *N, BugReporterContext &BR, PathSensitiveBugReport &R) {
472 
473   const LocationContext *Ctx = N->getLocationContext();
474   const StackFrameContext *SCtx = Ctx->getStackFrame();
475   ProgramStateRef State = N->getState();
476   auto CallExitLoc = N->getLocationAs<CallExitBegin>();
477 
478   // No diagnostic if region was modified inside the frame.
479   if (!CallExitLoc || isModifiedInFrame(N))
480     return nullptr;
481 
482   CallEventRef<> Call =
483       BR.getStateManager().getCallEventManager().getCaller(SCtx, State);
484 
485   // Optimistically suppress uninitialized value bugs that result
486   // from system headers having a chance to initialize the value
487   // but failing to do so. It's too unlikely a system header's fault.
488   // It's much more likely a situation in which the function has a failure
489   // mode that the user decided not to check. If we want to hunt such
490   // omitted checks, we should provide an explicit function-specific note
491   // describing the precondition under which the function isn't supposed to
492   // initialize its out-parameter, and additionally check that such
493   // precondition can actually be fulfilled on the current path.
494   if (Call->isInSystemHeader()) {
495     // We make an exception for system header functions that have no branches.
496     // Such functions unconditionally fail to initialize the variable.
497     // If they call other functions that have more paths within them,
498     // this suppression would still apply when we visit these inner functions.
499     // One common example of a standard function that doesn't ever initialize
500     // its out parameter is operator placement new; it's up to the follow-up
501     // constructor (if any) to initialize the memory.
502     if (!N->getStackFrame()->getCFG()->isLinear()) {
503       static int i = 0;
504       R.markInvalid(&i, nullptr);
505     }
506     return nullptr;
507   }
508 
509   if (const auto *MC = dyn_cast<ObjCMethodCall>(Call)) {
510     // If we failed to construct a piece for self, we still want to check
511     // whether the entity of interest is in a parameter.
512     if (PathDiagnosticPieceRef Piece = maybeEmitNoteForObjCSelf(R, *MC, N))
513       return Piece;
514   }
515 
516   if (const auto *CCall = dyn_cast<CXXConstructorCall>(Call)) {
517     // Do not generate diagnostics for not modified parameters in
518     // constructors.
519     return maybeEmitNoteForCXXThis(R, *CCall, N);
520   }
521 
522   return maybeEmitNoteForParameters(R, *Call, N);
523 }
524 
525 //===----------------------------------------------------------------------===//
526 // Implementation of NoStoreFuncVisitor.
527 //===----------------------------------------------------------------------===//
528 
529 namespace {
530 /// Put a diagnostic on return statement of all inlined functions
531 /// for which  the region of interest \p RegionOfInterest was passed into,
532 /// but not written inside, and it has caused an undefined read or a null
533 /// pointer dereference outside.
534 class NoStoreFuncVisitor final : public NoStateChangeFuncVisitor {
535   const SubRegion *RegionOfInterest;
536   MemRegionManager &MmrMgr;
537   const SourceManager &SM;
538   const PrintingPolicy &PP;
539 
540   /// Recursion limit for dereferencing fields when looking for the
541   /// region of interest.
542   /// The limit of two indicates that we will dereference fields only once.
543   static const unsigned DEREFERENCE_LIMIT = 2;
544 
545   using RegionVector = SmallVector<const MemRegion *, 5>;
546 
547 public:
NoStoreFuncVisitor(const SubRegion * R,bugreporter::TrackingKind TKind)548   NoStoreFuncVisitor(const SubRegion *R, bugreporter::TrackingKind TKind)
549       : NoStateChangeFuncVisitor(TKind), RegionOfInterest(R),
550         MmrMgr(R->getMemRegionManager()),
551         SM(MmrMgr.getContext().getSourceManager()),
552         PP(MmrMgr.getContext().getPrintingPolicy()) {}
553 
Profile(llvm::FoldingSetNodeID & ID) const554   void Profile(llvm::FoldingSetNodeID &ID) const override {
555     static int Tag = 0;
556     ID.AddPointer(&Tag);
557     ID.AddPointer(RegionOfInterest);
558   }
559 
560 private:
561   /// \return Whether \c RegionOfInterest was modified at \p CurrN compared to
562   /// the value it holds in \p CallExitBeginN.
563   bool wasModifiedBeforeCallExit(const ExplodedNode *CurrN,
564                                  const ExplodedNode *CallExitBeginN) override;
565 
566   /// Attempts to find the region of interest in a given record decl,
567   /// by either following the base classes or fields.
568   /// Dereferences fields up to a given recursion limit.
569   /// Note that \p Vec is passed by value, leading to quadratic copying cost,
570   /// but it's OK in practice since its length is limited to DEREFERENCE_LIMIT.
571   /// \return A chain fields leading to the region of interest or std::nullopt.
572   const std::optional<RegionVector>
573   findRegionOfInterestInRecord(const RecordDecl *RD, ProgramStateRef State,
574                                const MemRegion *R, const RegionVector &Vec = {},
575                                int depth = 0);
576 
577   // Region of interest corresponds to an IVar, exiting a method
578   // which could have written into that IVar, but did not.
579   PathDiagnosticPieceRef maybeEmitNoteForObjCSelf(PathSensitiveBugReport &R,
580                                                   const ObjCMethodCall &Call,
581                                                   const ExplodedNode *N) final;
582 
583   PathDiagnosticPieceRef maybeEmitNoteForCXXThis(PathSensitiveBugReport &R,
584                                                  const CXXConstructorCall &Call,
585                                                  const ExplodedNode *N) final;
586 
587   PathDiagnosticPieceRef
588   maybeEmitNoteForParameters(PathSensitiveBugReport &R, const CallEvent &Call,
589                              const ExplodedNode *N) final;
590 
591   /// Consume the information on the no-store stack frame in order to
592   /// either emit a note or suppress the report enirely.
593   /// \return Diagnostics piece for region not modified in the current function,
594   /// if it decides to emit one.
595   PathDiagnosticPieceRef
596   maybeEmitNote(PathSensitiveBugReport &R, const CallEvent &Call,
597                 const ExplodedNode *N, const RegionVector &FieldChain,
598                 const MemRegion *MatchedRegion, StringRef FirstElement,
599                 bool FirstIsReferenceType, unsigned IndirectionLevel);
600 
601   bool prettyPrintRegionName(const RegionVector &FieldChain,
602                              const MemRegion *MatchedRegion,
603                              StringRef FirstElement, bool FirstIsReferenceType,
604                              unsigned IndirectionLevel,
605                              llvm::raw_svector_ostream &os);
606 
607   StringRef prettyPrintFirstElement(StringRef FirstElement,
608                                     bool MoreItemsExpected,
609                                     int IndirectionLevel,
610                                     llvm::raw_svector_ostream &os);
611 };
612 } // namespace
613 
614 /// \return Whether the method declaration \p Parent
615 /// syntactically has a binary operation writing into the ivar \p Ivar.
potentiallyWritesIntoIvar(const Decl * Parent,const ObjCIvarDecl * Ivar)616 static bool potentiallyWritesIntoIvar(const Decl *Parent,
617                                       const ObjCIvarDecl *Ivar) {
618   using namespace ast_matchers;
619   const char *IvarBind = "Ivar";
620   if (!Parent || !Parent->hasBody())
621     return false;
622   StatementMatcher WriteIntoIvarM = binaryOperator(
623       hasOperatorName("="),
624       hasLHS(ignoringParenImpCasts(
625           objcIvarRefExpr(hasDeclaration(equalsNode(Ivar))).bind(IvarBind))));
626   StatementMatcher ParentM = stmt(hasDescendant(WriteIntoIvarM));
627   auto Matches = match(ParentM, *Parent->getBody(), Parent->getASTContext());
628   for (BoundNodes &Match : Matches) {
629     auto IvarRef = Match.getNodeAs<ObjCIvarRefExpr>(IvarBind);
630     if (IvarRef->isFreeIvar())
631       return true;
632 
633     const Expr *Base = IvarRef->getBase();
634     if (const auto *ICE = dyn_cast<ImplicitCastExpr>(Base))
635       Base = ICE->getSubExpr();
636 
637     if (const auto *DRE = dyn_cast<DeclRefExpr>(Base))
638       if (const auto *ID = dyn_cast<ImplicitParamDecl>(DRE->getDecl()))
639         if (ID->getParameterKind() == ImplicitParamKind::ObjCSelf)
640           return true;
641 
642     return false;
643   }
644   return false;
645 }
646 
647 /// Attempts to find the region of interest in a given CXX decl,
648 /// by either following the base classes or fields.
649 /// Dereferences fields up to a given recursion limit.
650 /// Note that \p Vec is passed by value, leading to quadratic copying cost,
651 /// but it's OK in practice since its length is limited to DEREFERENCE_LIMIT.
652 /// \return A chain fields leading to the region of interest or std::nullopt.
653 const std::optional<NoStoreFuncVisitor::RegionVector>
findRegionOfInterestInRecord(const RecordDecl * RD,ProgramStateRef State,const MemRegion * R,const NoStoreFuncVisitor::RegionVector & Vec,int depth)654 NoStoreFuncVisitor::findRegionOfInterestInRecord(
655     const RecordDecl *RD, ProgramStateRef State, const MemRegion *R,
656     const NoStoreFuncVisitor::RegionVector &Vec /* = {} */,
657     int depth /* = 0 */) {
658 
659   if (depth == DEREFERENCE_LIMIT) // Limit the recursion depth.
660     return std::nullopt;
661 
662   if (const auto *RDX = dyn_cast<CXXRecordDecl>(RD))
663     if (!RDX->hasDefinition())
664       return std::nullopt;
665 
666   // Recursively examine the base classes.
667   // Note that following base classes does not increase the recursion depth.
668   if (const auto *RDX = dyn_cast<CXXRecordDecl>(RD))
669     for (const auto &II : RDX->bases())
670       if (const RecordDecl *RRD = II.getType()->getAsRecordDecl())
671         if (std::optional<RegionVector> Out =
672                 findRegionOfInterestInRecord(RRD, State, R, Vec, depth))
673           return Out;
674 
675   for (const FieldDecl *I : RD->fields()) {
676     QualType FT = I->getType();
677     const FieldRegion *FR = MmrMgr.getFieldRegion(I, cast<SubRegion>(R));
678     const SVal V = State->getSVal(FR);
679     const MemRegion *VR = V.getAsRegion();
680 
681     RegionVector VecF = Vec;
682     VecF.push_back(FR);
683 
684     if (RegionOfInterest == VR)
685       return VecF;
686 
687     if (const RecordDecl *RRD = FT->getAsRecordDecl())
688       if (auto Out =
689               findRegionOfInterestInRecord(RRD, State, FR, VecF, depth + 1))
690         return Out;
691 
692     QualType PT = FT->getPointeeType();
693     if (PT.isNull() || PT->isVoidType() || !VR)
694       continue;
695 
696     if (const RecordDecl *RRD = PT->getAsRecordDecl())
697       if (std::optional<RegionVector> Out =
698               findRegionOfInterestInRecord(RRD, State, VR, VecF, depth + 1))
699         return Out;
700   }
701 
702   return std::nullopt;
703 }
704 
705 PathDiagnosticPieceRef
maybeEmitNoteForObjCSelf(PathSensitiveBugReport & R,const ObjCMethodCall & Call,const ExplodedNode * N)706 NoStoreFuncVisitor::maybeEmitNoteForObjCSelf(PathSensitiveBugReport &R,
707                                              const ObjCMethodCall &Call,
708                                              const ExplodedNode *N) {
709   if (const auto *IvarR = dyn_cast<ObjCIvarRegion>(RegionOfInterest)) {
710     const MemRegion *SelfRegion = Call.getReceiverSVal().getAsRegion();
711     if (RegionOfInterest->isSubRegionOf(SelfRegion) &&
712         potentiallyWritesIntoIvar(Call.getRuntimeDefinition().getDecl(),
713                                   IvarR->getDecl()))
714       return maybeEmitNote(R, Call, N, {}, SelfRegion, "self",
715                            /*FirstIsReferenceType=*/false, 1);
716   }
717   return nullptr;
718 }
719 
720 PathDiagnosticPieceRef
maybeEmitNoteForCXXThis(PathSensitiveBugReport & R,const CXXConstructorCall & Call,const ExplodedNode * N)721 NoStoreFuncVisitor::maybeEmitNoteForCXXThis(PathSensitiveBugReport &R,
722                                             const CXXConstructorCall &Call,
723                                             const ExplodedNode *N) {
724   const MemRegion *ThisR = Call.getCXXThisVal().getAsRegion();
725   if (RegionOfInterest->isSubRegionOf(ThisR) && !Call.getDecl()->isImplicit())
726     return maybeEmitNote(R, Call, N, {}, ThisR, "this",
727                          /*FirstIsReferenceType=*/false, 1);
728 
729   // Do not generate diagnostics for not modified parameters in
730   // constructors.
731   return nullptr;
732 }
733 
734 /// \return whether \p Ty points to a const type, or is a const reference.
isPointerToConst(QualType Ty)735 static bool isPointerToConst(QualType Ty) {
736   return !Ty->getPointeeType().isNull() &&
737          Ty->getPointeeType().getCanonicalType().isConstQualified();
738 }
739 
maybeEmitNoteForParameters(PathSensitiveBugReport & R,const CallEvent & Call,const ExplodedNode * N)740 PathDiagnosticPieceRef NoStoreFuncVisitor::maybeEmitNoteForParameters(
741     PathSensitiveBugReport &R, const CallEvent &Call, const ExplodedNode *N) {
742   ArrayRef<ParmVarDecl *> Parameters = Call.parameters();
743   for (unsigned I = 0; I < Call.getNumArgs() && I < Parameters.size(); ++I) {
744     const ParmVarDecl *PVD = Parameters[I];
745     SVal V = Call.getArgSVal(I);
746     bool ParamIsReferenceType = PVD->getType()->isReferenceType();
747     std::string ParamName = PVD->getNameAsString();
748 
749     unsigned IndirectionLevel = 1;
750     QualType T = PVD->getType();
751     while (const MemRegion *MR = V.getAsRegion()) {
752       if (RegionOfInterest->isSubRegionOf(MR) && !isPointerToConst(T))
753         return maybeEmitNote(R, Call, N, {}, MR, ParamName,
754                              ParamIsReferenceType, IndirectionLevel);
755 
756       QualType PT = T->getPointeeType();
757       if (PT.isNull() || PT->isVoidType())
758         break;
759 
760       ProgramStateRef State = N->getState();
761 
762       if (const RecordDecl *RD = PT->getAsRecordDecl())
763         if (std::optional<RegionVector> P =
764                 findRegionOfInterestInRecord(RD, State, MR))
765           return maybeEmitNote(R, Call, N, *P, RegionOfInterest, ParamName,
766                                ParamIsReferenceType, IndirectionLevel);
767 
768       V = State->getSVal(MR, PT);
769       T = PT;
770       IndirectionLevel++;
771     }
772   }
773 
774   return nullptr;
775 }
776 
wasModifiedBeforeCallExit(const ExplodedNode * CurrN,const ExplodedNode * CallExitBeginN)777 bool NoStoreFuncVisitor::wasModifiedBeforeCallExit(
778     const ExplodedNode *CurrN, const ExplodedNode *CallExitBeginN) {
779   return ::wasRegionOfInterestModifiedAt(
780       RegionOfInterest, CurrN,
781       CallExitBeginN->getState()->getSVal(RegionOfInterest));
782 }
783 
784 static llvm::StringLiteral WillBeUsedForACondition =
785     ", which participates in a condition later";
786 
maybeEmitNote(PathSensitiveBugReport & R,const CallEvent & Call,const ExplodedNode * N,const RegionVector & FieldChain,const MemRegion * MatchedRegion,StringRef FirstElement,bool FirstIsReferenceType,unsigned IndirectionLevel)787 PathDiagnosticPieceRef NoStoreFuncVisitor::maybeEmitNote(
788     PathSensitiveBugReport &R, const CallEvent &Call, const ExplodedNode *N,
789     const RegionVector &FieldChain, const MemRegion *MatchedRegion,
790     StringRef FirstElement, bool FirstIsReferenceType,
791     unsigned IndirectionLevel) {
792 
793   PathDiagnosticLocation L =
794       PathDiagnosticLocation::create(N->getLocation(), SM);
795 
796   // For now this shouldn't trigger, but once it does (as we add more
797   // functions to the body farm), we'll need to decide if these reports
798   // are worth suppressing as well.
799   if (!L.hasValidLocation())
800     return nullptr;
801 
802   SmallString<256> sbuf;
803   llvm::raw_svector_ostream os(sbuf);
804   os << "Returning without writing to '";
805 
806   // Do not generate the note if failed to pretty-print.
807   if (!prettyPrintRegionName(FieldChain, MatchedRegion, FirstElement,
808                              FirstIsReferenceType, IndirectionLevel, os))
809     return nullptr;
810 
811   os << "'";
812   if (TKind == bugreporter::TrackingKind::Condition)
813     os << WillBeUsedForACondition;
814   return std::make_shared<PathDiagnosticEventPiece>(L, os.str());
815 }
816 
prettyPrintRegionName(const RegionVector & FieldChain,const MemRegion * MatchedRegion,StringRef FirstElement,bool FirstIsReferenceType,unsigned IndirectionLevel,llvm::raw_svector_ostream & os)817 bool NoStoreFuncVisitor::prettyPrintRegionName(const RegionVector &FieldChain,
818                                                const MemRegion *MatchedRegion,
819                                                StringRef FirstElement,
820                                                bool FirstIsReferenceType,
821                                                unsigned IndirectionLevel,
822                                                llvm::raw_svector_ostream &os) {
823 
824   if (FirstIsReferenceType)
825     IndirectionLevel--;
826 
827   RegionVector RegionSequence;
828 
829   // Add the regions in the reverse order, then reverse the resulting array.
830   assert(RegionOfInterest->isSubRegionOf(MatchedRegion));
831   const MemRegion *R = RegionOfInterest;
832   while (R != MatchedRegion) {
833     RegionSequence.push_back(R);
834     R = cast<SubRegion>(R)->getSuperRegion();
835   }
836   std::reverse(RegionSequence.begin(), RegionSequence.end());
837   RegionSequence.append(FieldChain.begin(), FieldChain.end());
838 
839   StringRef Sep;
840   for (const MemRegion *R : RegionSequence) {
841 
842     // Just keep going up to the base region.
843     // Element regions may appear due to casts.
844     if (isa<CXXBaseObjectRegion, CXXTempObjectRegion>(R))
845       continue;
846 
847     if (Sep.empty())
848       Sep = prettyPrintFirstElement(FirstElement,
849                                     /*MoreItemsExpected=*/true,
850                                     IndirectionLevel, os);
851 
852     os << Sep;
853 
854     // Can only reasonably pretty-print DeclRegions.
855     if (!isa<DeclRegion>(R))
856       return false;
857 
858     const auto *DR = cast<DeclRegion>(R);
859     Sep = DR->getValueType()->isAnyPointerType() ? "->" : ".";
860     DR->getDecl()->getDeclName().print(os, PP);
861   }
862 
863   if (Sep.empty())
864     prettyPrintFirstElement(FirstElement,
865                             /*MoreItemsExpected=*/false, IndirectionLevel, os);
866   return true;
867 }
868 
prettyPrintFirstElement(StringRef FirstElement,bool MoreItemsExpected,int IndirectionLevel,llvm::raw_svector_ostream & os)869 StringRef NoStoreFuncVisitor::prettyPrintFirstElement(
870     StringRef FirstElement, bool MoreItemsExpected, int IndirectionLevel,
871     llvm::raw_svector_ostream &os) {
872   StringRef Out = ".";
873 
874   if (IndirectionLevel > 0 && MoreItemsExpected) {
875     IndirectionLevel--;
876     Out = "->";
877   }
878 
879   if (IndirectionLevel > 0 && MoreItemsExpected)
880     os << "(";
881 
882   for (int i = 0; i < IndirectionLevel; i++)
883     os << "*";
884   os << FirstElement;
885 
886   if (IndirectionLevel > 0 && MoreItemsExpected)
887     os << ")";
888 
889   return Out;
890 }
891 
892 //===----------------------------------------------------------------------===//
893 // Implementation of MacroNullReturnSuppressionVisitor.
894 //===----------------------------------------------------------------------===//
895 
896 namespace {
897 
898 /// Suppress null-pointer-dereference bugs where dereferenced null was returned
899 /// the macro.
900 class MacroNullReturnSuppressionVisitor final : public BugReporterVisitor {
901   const SubRegion *RegionOfInterest;
902   const SVal ValueAtDereference;
903 
904   // Do not invalidate the reports where the value was modified
905   // after it got assigned to from the macro.
906   bool WasModified = false;
907 
908 public:
MacroNullReturnSuppressionVisitor(const SubRegion * R,const SVal V)909   MacroNullReturnSuppressionVisitor(const SubRegion *R, const SVal V)
910       : RegionOfInterest(R), ValueAtDereference(V) {}
911 
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)912   PathDiagnosticPieceRef VisitNode(const ExplodedNode *N,
913                                    BugReporterContext &BRC,
914                                    PathSensitiveBugReport &BR) override {
915     if (WasModified)
916       return nullptr;
917 
918     auto BugPoint = BR.getErrorNode()->getLocation().getAs<StmtPoint>();
919     if (!BugPoint)
920       return nullptr;
921 
922     const SourceManager &SMgr = BRC.getSourceManager();
923     if (auto Loc = matchAssignment(N)) {
924       if (isFunctionMacroExpansion(*Loc, SMgr)) {
925         std::string MacroName = std::string(getMacroName(*Loc, BRC));
926         SourceLocation BugLoc = BugPoint->getStmt()->getBeginLoc();
927         if (!BugLoc.isMacroID() || getMacroName(BugLoc, BRC) != MacroName)
928           BR.markInvalid(getTag(), MacroName.c_str());
929       }
930     }
931 
932     if (wasRegionOfInterestModifiedAt(RegionOfInterest, N, ValueAtDereference))
933       WasModified = true;
934 
935     return nullptr;
936   }
937 
addMacroVisitorIfNecessary(const ExplodedNode * N,const MemRegion * R,bool EnableNullFPSuppression,PathSensitiveBugReport & BR,const SVal V)938   static void addMacroVisitorIfNecessary(
939         const ExplodedNode *N, const MemRegion *R,
940         bool EnableNullFPSuppression, PathSensitiveBugReport &BR,
941         const SVal V) {
942     AnalyzerOptions &Options = N->getState()->getAnalysisManager().options;
943     if (EnableNullFPSuppression && Options.ShouldSuppressNullReturnPaths &&
944         isa<Loc>(V))
945       BR.addVisitor<MacroNullReturnSuppressionVisitor>(R->getAs<SubRegion>(),
946                                                        V);
947   }
948 
getTag() const949   void* getTag() const {
950     static int Tag = 0;
951     return static_cast<void *>(&Tag);
952   }
953 
Profile(llvm::FoldingSetNodeID & ID) const954   void Profile(llvm::FoldingSetNodeID &ID) const override {
955     ID.AddPointer(getTag());
956   }
957 
958 private:
959   /// \return Source location of right hand side of an assignment
960   /// into \c RegionOfInterest, empty optional if none found.
matchAssignment(const ExplodedNode * N)961   std::optional<SourceLocation> matchAssignment(const ExplodedNode *N) {
962     const Stmt *S = N->getStmtForDiagnostics();
963     ProgramStateRef State = N->getState();
964     auto *LCtx = N->getLocationContext();
965     if (!S)
966       return std::nullopt;
967 
968     if (const auto *DS = dyn_cast<DeclStmt>(S)) {
969       if (const auto *VD = dyn_cast<VarDecl>(DS->getSingleDecl()))
970         if (const Expr *RHS = VD->getInit())
971           if (RegionOfInterest->isSubRegionOf(
972                   State->getLValue(VD, LCtx).getAsRegion()))
973             return RHS->getBeginLoc();
974     } else if (const auto *BO = dyn_cast<BinaryOperator>(S)) {
975       const MemRegion *R = N->getSVal(BO->getLHS()).getAsRegion();
976       const Expr *RHS = BO->getRHS();
977       if (BO->isAssignmentOp() && RegionOfInterest->isSubRegionOf(R)) {
978         return RHS->getBeginLoc();
979       }
980     }
981     return std::nullopt;
982   }
983 };
984 
985 } // end of anonymous namespace
986 
987 namespace {
988 
989 /// Emits an extra note at the return statement of an interesting stack frame.
990 ///
991 /// The returned value is marked as an interesting value, and if it's null,
992 /// adds a visitor to track where it became null.
993 ///
994 /// This visitor is intended to be used when another visitor discovers that an
995 /// interesting value comes from an inlined function call.
996 class ReturnVisitor : public TrackingBugReporterVisitor {
997   const StackFrameContext *CalleeSFC;
998   enum {
999     Initial,
1000     MaybeUnsuppress,
1001     Satisfied
1002   } Mode = Initial;
1003 
1004   bool EnableNullFPSuppression;
1005   bool ShouldInvalidate = true;
1006   AnalyzerOptions& Options;
1007   bugreporter::TrackingKind TKind;
1008 
1009 public:
ReturnVisitor(TrackerRef ParentTracker,const StackFrameContext * Frame,bool Suppressed,AnalyzerOptions & Options,bugreporter::TrackingKind TKind)1010   ReturnVisitor(TrackerRef ParentTracker, const StackFrameContext *Frame,
1011                 bool Suppressed, AnalyzerOptions &Options,
1012                 bugreporter::TrackingKind TKind)
1013       : TrackingBugReporterVisitor(ParentTracker), CalleeSFC(Frame),
1014         EnableNullFPSuppression(Suppressed), Options(Options), TKind(TKind) {}
1015 
getTag()1016   static void *getTag() {
1017     static int Tag = 0;
1018     return static_cast<void *>(&Tag);
1019   }
1020 
Profile(llvm::FoldingSetNodeID & ID) const1021   void Profile(llvm::FoldingSetNodeID &ID) const override {
1022     ID.AddPointer(ReturnVisitor::getTag());
1023     ID.AddPointer(CalleeSFC);
1024     ID.AddBoolean(EnableNullFPSuppression);
1025   }
1026 
visitNodeInitial(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)1027   PathDiagnosticPieceRef visitNodeInitial(const ExplodedNode *N,
1028                                           BugReporterContext &BRC,
1029                                           PathSensitiveBugReport &BR) {
1030     // Only print a message at the interesting return statement.
1031     if (N->getLocationContext() != CalleeSFC)
1032       return nullptr;
1033 
1034     std::optional<StmtPoint> SP = N->getLocationAs<StmtPoint>();
1035     if (!SP)
1036       return nullptr;
1037 
1038     const auto *Ret = dyn_cast<ReturnStmt>(SP->getStmt());
1039     if (!Ret)
1040       return nullptr;
1041 
1042     // Okay, we're at the right return statement, but do we have the return
1043     // value available?
1044     ProgramStateRef State = N->getState();
1045     SVal V = State->getSVal(Ret, CalleeSFC);
1046     if (V.isUnknownOrUndef())
1047       return nullptr;
1048 
1049     // Don't print any more notes after this one.
1050     Mode = Satisfied;
1051 
1052     const Expr *RetE = Ret->getRetValue();
1053     assert(RetE && "Tracking a return value for a void function");
1054 
1055     // Handle cases where a reference is returned and then immediately used.
1056     std::optional<Loc> LValue;
1057     if (RetE->isGLValue()) {
1058       if ((LValue = V.getAs<Loc>())) {
1059         SVal RValue = State->getRawSVal(*LValue, RetE->getType());
1060         if (isa<DefinedSVal>(RValue))
1061           V = RValue;
1062       }
1063     }
1064 
1065     // Ignore aggregate rvalues.
1066     if (isa<nonloc::LazyCompoundVal, nonloc::CompoundVal>(V))
1067       return nullptr;
1068 
1069     RetE = RetE->IgnoreParenCasts();
1070 
1071     // Let's track the return value.
1072     getParentTracker().track(RetE, N, {TKind, EnableNullFPSuppression});
1073 
1074     // Build an appropriate message based on the return value.
1075     SmallString<64> Msg;
1076     llvm::raw_svector_ostream Out(Msg);
1077 
1078     bool WouldEventBeMeaningless = false;
1079 
1080     if (State->isNull(V).isConstrainedTrue()) {
1081       if (isa<Loc>(V)) {
1082 
1083         // If we have counter-suppression enabled, make sure we keep visiting
1084         // future nodes. We want to emit a path note as well, in case
1085         // the report is resurrected as valid later on.
1086         if (EnableNullFPSuppression &&
1087             Options.ShouldAvoidSuppressingNullArgumentPaths)
1088           Mode = MaybeUnsuppress;
1089 
1090         if (RetE->getType()->isObjCObjectPointerType()) {
1091           Out << "Returning nil";
1092         } else {
1093           Out << "Returning null pointer";
1094         }
1095       } else {
1096         Out << "Returning zero";
1097       }
1098 
1099     } else {
1100       if (auto CI = V.getAs<nonloc::ConcreteInt>()) {
1101         Out << "Returning the value " << CI->getValue();
1102       } else {
1103         // There is nothing interesting about returning a value, when it is
1104         // plain value without any constraints, and the function is guaranteed
1105         // to return that every time. We could use CFG::isLinear() here, but
1106         // constexpr branches are obvious to the compiler, not necesserily to
1107         // the programmer.
1108         if (N->getCFG().size() == 3)
1109           WouldEventBeMeaningless = true;
1110 
1111         Out << (isa<Loc>(V) ? "Returning pointer" : "Returning value");
1112       }
1113     }
1114 
1115     if (LValue) {
1116       if (const MemRegion *MR = LValue->getAsRegion()) {
1117         if (MR->canPrintPretty()) {
1118           Out << " (reference to ";
1119           MR->printPretty(Out);
1120           Out << ")";
1121         }
1122       }
1123     } else {
1124       // FIXME: We should have a more generalized location printing mechanism.
1125       if (const auto *DR = dyn_cast<DeclRefExpr>(RetE))
1126         if (const auto *DD = dyn_cast<DeclaratorDecl>(DR->getDecl()))
1127           Out << " (loaded from '" << *DD << "')";
1128     }
1129 
1130     PathDiagnosticLocation L(Ret, BRC.getSourceManager(), CalleeSFC);
1131     if (!L.isValid() || !L.asLocation().isValid())
1132       return nullptr;
1133 
1134     if (TKind == bugreporter::TrackingKind::Condition)
1135       Out << WillBeUsedForACondition;
1136 
1137     auto EventPiece = std::make_shared<PathDiagnosticEventPiece>(L, Out.str());
1138 
1139     // If we determined that the note is meaningless, make it prunable, and
1140     // don't mark the stackframe interesting.
1141     if (WouldEventBeMeaningless)
1142       EventPiece->setPrunable(true);
1143     else
1144       BR.markInteresting(CalleeSFC);
1145 
1146     return EventPiece;
1147   }
1148 
visitNodeMaybeUnsuppress(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)1149   PathDiagnosticPieceRef visitNodeMaybeUnsuppress(const ExplodedNode *N,
1150                                                   BugReporterContext &BRC,
1151                                                   PathSensitiveBugReport &BR) {
1152     assert(Options.ShouldAvoidSuppressingNullArgumentPaths);
1153 
1154     // Are we at the entry node for this call?
1155     std::optional<CallEnter> CE = N->getLocationAs<CallEnter>();
1156     if (!CE)
1157       return nullptr;
1158 
1159     if (CE->getCalleeContext() != CalleeSFC)
1160       return nullptr;
1161 
1162     Mode = Satisfied;
1163 
1164     // Don't automatically suppress a report if one of the arguments is
1165     // known to be a null pointer. Instead, start tracking /that/ null
1166     // value back to its origin.
1167     ProgramStateManager &StateMgr = BRC.getStateManager();
1168     CallEventManager &CallMgr = StateMgr.getCallEventManager();
1169 
1170     ProgramStateRef State = N->getState();
1171     CallEventRef<> Call = CallMgr.getCaller(CalleeSFC, State);
1172     for (unsigned I = 0, E = Call->getNumArgs(); I != E; ++I) {
1173       std::optional<Loc> ArgV = Call->getArgSVal(I).getAs<Loc>();
1174       if (!ArgV)
1175         continue;
1176 
1177       const Expr *ArgE = Call->getArgExpr(I);
1178       if (!ArgE)
1179         continue;
1180 
1181       // Is it possible for this argument to be non-null?
1182       if (!State->isNull(*ArgV).isConstrainedTrue())
1183         continue;
1184 
1185       if (getParentTracker()
1186               .track(ArgE, N, {TKind, EnableNullFPSuppression})
1187               .FoundSomethingToTrack)
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 //===----------------------------------------------------------------------===//
1221 //                               StoreSiteFinder
1222 //===----------------------------------------------------------------------===//
1223 
1224 /// Finds last store into the given region,
1225 /// which is different from a given symbolic value.
1226 class StoreSiteFinder final : public TrackingBugReporterVisitor {
1227   const MemRegion *R;
1228   SVal V;
1229   bool Satisfied = false;
1230 
1231   TrackingOptions Options;
1232   const StackFrameContext *OriginSFC;
1233 
1234 public:
1235   /// \param V We're searching for the store where \c R received this value.
1236   /// \param R The region we're tracking.
1237   /// \param Options Tracking behavior options.
1238   /// \param OriginSFC Only adds notes when the last store happened in a
1239   ///        different stackframe to this one. Disregarded if the tracking kind
1240   ///        is thorough.
1241   ///        This is useful, because for non-tracked regions, notes about
1242   ///        changes to its value in a nested stackframe could be pruned, and
1243   ///        this visitor can prevent that without polluting the bugpath too
1244   ///        much.
StoreSiteFinder(bugreporter::TrackerRef ParentTracker,SVal V,const MemRegion * R,TrackingOptions Options,const StackFrameContext * OriginSFC=nullptr)1245   StoreSiteFinder(bugreporter::TrackerRef ParentTracker, SVal V,
1246                   const MemRegion *R, TrackingOptions Options,
1247                   const StackFrameContext *OriginSFC = nullptr)
1248       : TrackingBugReporterVisitor(ParentTracker), R(R), V(V), Options(Options),
1249         OriginSFC(OriginSFC) {
1250     assert(R);
1251   }
1252 
1253   void Profile(llvm::FoldingSetNodeID &ID) const override;
1254 
1255   PathDiagnosticPieceRef VisitNode(const ExplodedNode *N,
1256                                    BugReporterContext &BRC,
1257                                    PathSensitiveBugReport &BR) override;
1258 };
1259 } // namespace
1260 
Profile(llvm::FoldingSetNodeID & ID) const1261 void StoreSiteFinder::Profile(llvm::FoldingSetNodeID &ID) const {
1262   static int tag = 0;
1263   ID.AddPointer(&tag);
1264   ID.AddPointer(R);
1265   ID.Add(V);
1266   ID.AddInteger(static_cast<int>(Options.Kind));
1267   ID.AddBoolean(Options.EnableNullFPSuppression);
1268 }
1269 
1270 /// Returns true if \p N represents the DeclStmt declaring and initializing
1271 /// \p VR.
isInitializationOfVar(const ExplodedNode * N,const VarRegion * VR)1272 static bool isInitializationOfVar(const ExplodedNode *N, const VarRegion *VR) {
1273   std::optional<PostStmt> P = N->getLocationAs<PostStmt>();
1274   if (!P)
1275     return false;
1276 
1277   const DeclStmt *DS = P->getStmtAs<DeclStmt>();
1278   if (!DS)
1279     return false;
1280 
1281   if (DS->getSingleDecl() != VR->getDecl())
1282     return false;
1283 
1284   const MemSpaceRegion *VarSpace = VR->getMemorySpace();
1285   const auto *FrameSpace = dyn_cast<StackSpaceRegion>(VarSpace);
1286   if (!FrameSpace) {
1287     // If we ever directly evaluate global DeclStmts, this assertion will be
1288     // invalid, but this still seems preferable to silently accepting an
1289     // initialization that may be for a path-sensitive variable.
1290     assert(VR->getDecl()->isStaticLocal() && "non-static stackless VarRegion");
1291     return true;
1292   }
1293 
1294   assert(VR->getDecl()->hasLocalStorage());
1295   const LocationContext *LCtx = N->getLocationContext();
1296   return FrameSpace->getStackFrame() == LCtx->getStackFrame();
1297 }
1298 
isObjCPointer(const MemRegion * R)1299 static bool isObjCPointer(const MemRegion *R) {
1300   if (R->isBoundable())
1301     if (const auto *TR = dyn_cast<TypedValueRegion>(R))
1302       return TR->getValueType()->isObjCObjectPointerType();
1303 
1304   return false;
1305 }
1306 
isObjCPointer(const ValueDecl * D)1307 static bool isObjCPointer(const ValueDecl *D) {
1308   return D->getType()->isObjCObjectPointerType();
1309 }
1310 
1311 /// Show diagnostics for initializing or declaring a region \p R with a bad value.
showBRDiagnostics(llvm::raw_svector_ostream & OS,StoreInfo SI)1312 static void showBRDiagnostics(llvm::raw_svector_ostream &OS, StoreInfo SI) {
1313   const bool HasPrefix = SI.Dest->canPrintPretty();
1314 
1315   if (HasPrefix) {
1316     SI.Dest->printPretty(OS);
1317     OS << " ";
1318   }
1319 
1320   const char *Action = nullptr;
1321 
1322   switch (SI.StoreKind) {
1323   case StoreInfo::Initialization:
1324     Action = HasPrefix ? "initialized to " : "Initializing to ";
1325     break;
1326   case StoreInfo::BlockCapture:
1327     Action = HasPrefix ? "captured by block as " : "Captured by block as ";
1328     break;
1329   default:
1330     llvm_unreachable("Unexpected store kind");
1331   }
1332 
1333   if (isa<loc::ConcreteInt>(SI.Value)) {
1334     OS << Action << (isObjCPointer(SI.Dest) ? "nil" : "a null pointer value");
1335 
1336   } else if (auto CVal = SI.Value.getAs<nonloc::ConcreteInt>()) {
1337     OS << Action << CVal->getValue();
1338 
1339   } else if (SI.Origin && SI.Origin->canPrintPretty()) {
1340     OS << Action << "the value of ";
1341     SI.Origin->printPretty(OS);
1342 
1343   } else if (SI.StoreKind == StoreInfo::Initialization) {
1344     // We don't need to check here, all these conditions were
1345     // checked by StoreSiteFinder, when it figured out that it is
1346     // initialization.
1347     const auto *DS =
1348         cast<DeclStmt>(SI.StoreSite->getLocationAs<PostStmt>()->getStmt());
1349 
1350     if (SI.Value.isUndef()) {
1351       if (isa<VarRegion>(SI.Dest)) {
1352         const auto *VD = cast<VarDecl>(DS->getSingleDecl());
1353 
1354         if (VD->getInit()) {
1355           OS << (HasPrefix ? "initialized" : "Initializing")
1356              << " to a garbage value";
1357         } else {
1358           OS << (HasPrefix ? "declared" : "Declaring")
1359              << " without an initial value";
1360         }
1361       }
1362     } else {
1363       OS << (HasPrefix ? "initialized" : "Initialized") << " here";
1364     }
1365   }
1366 }
1367 
1368 /// Display diagnostics for passing bad region as a parameter.
showBRParamDiagnostics(llvm::raw_svector_ostream & OS,StoreInfo SI)1369 static void showBRParamDiagnostics(llvm::raw_svector_ostream &OS,
1370                                    StoreInfo SI) {
1371   const auto *VR = cast<VarRegion>(SI.Dest);
1372   const auto *D = VR->getDecl();
1373 
1374   OS << "Passing ";
1375 
1376   if (isa<loc::ConcreteInt>(SI.Value)) {
1377     OS << (isObjCPointer(D) ? "nil object reference" : "null pointer value");
1378 
1379   } else if (SI.Value.isUndef()) {
1380     OS << "uninitialized value";
1381 
1382   } else if (auto CI = SI.Value.getAs<nonloc::ConcreteInt>()) {
1383     OS << "the value " << CI->getValue();
1384 
1385   } else if (SI.Origin && SI.Origin->canPrintPretty()) {
1386     SI.Origin->printPretty(OS);
1387 
1388   } else {
1389     OS << "value";
1390   }
1391 
1392   if (const auto *Param = dyn_cast<ParmVarDecl>(VR->getDecl())) {
1393     // Printed parameter indexes are 1-based, not 0-based.
1394     unsigned Idx = Param->getFunctionScopeIndex() + 1;
1395     OS << " via " << Idx << llvm::getOrdinalSuffix(Idx) << " parameter";
1396     if (VR->canPrintPretty()) {
1397       OS << " ";
1398       VR->printPretty(OS);
1399     }
1400   } else if (const auto *ImplParam = dyn_cast<ImplicitParamDecl>(D)) {
1401     if (ImplParam->getParameterKind() == ImplicitParamKind::ObjCSelf) {
1402       OS << " via implicit parameter 'self'";
1403     }
1404   }
1405 }
1406 
1407 /// Show default diagnostics for storing bad region.
showBRDefaultDiagnostics(llvm::raw_svector_ostream & OS,StoreInfo SI)1408 static void showBRDefaultDiagnostics(llvm::raw_svector_ostream &OS,
1409                                      StoreInfo SI) {
1410   const bool HasSuffix = SI.Dest->canPrintPretty();
1411 
1412   if (isa<loc::ConcreteInt>(SI.Value)) {
1413     OS << (isObjCPointer(SI.Dest) ? "nil object reference stored"
1414                                   : (HasSuffix ? "Null pointer value stored"
1415                                                : "Storing null pointer value"));
1416 
1417   } else if (SI.Value.isUndef()) {
1418     OS << (HasSuffix ? "Uninitialized value stored"
1419                      : "Storing uninitialized value");
1420 
1421   } else if (auto CV = SI.Value.getAs<nonloc::ConcreteInt>()) {
1422     if (HasSuffix)
1423       OS << "The value " << CV->getValue() << " is assigned";
1424     else
1425       OS << "Assigning " << CV->getValue();
1426 
1427   } else if (SI.Origin && SI.Origin->canPrintPretty()) {
1428     if (HasSuffix) {
1429       OS << "The value of ";
1430       SI.Origin->printPretty(OS);
1431       OS << " is assigned";
1432     } else {
1433       OS << "Assigning the value of ";
1434       SI.Origin->printPretty(OS);
1435     }
1436 
1437   } else {
1438     OS << (HasSuffix ? "Value assigned" : "Assigning value");
1439   }
1440 
1441   if (HasSuffix) {
1442     OS << " to ";
1443     SI.Dest->printPretty(OS);
1444   }
1445 }
1446 
isTrivialCopyOrMoveCtor(const CXXConstructExpr * CE)1447 static bool isTrivialCopyOrMoveCtor(const CXXConstructExpr *CE) {
1448   if (!CE)
1449     return false;
1450 
1451   const auto *CtorDecl = CE->getConstructor();
1452 
1453   return CtorDecl->isCopyOrMoveConstructor() && CtorDecl->isTrivial();
1454 }
1455 
tryExtractInitializerFromList(const InitListExpr * ILE,const MemRegion * R)1456 static const Expr *tryExtractInitializerFromList(const InitListExpr *ILE,
1457                                                  const MemRegion *R) {
1458 
1459   const auto *TVR = dyn_cast_or_null<TypedValueRegion>(R);
1460 
1461   if (!TVR)
1462     return nullptr;
1463 
1464   const auto ITy = ILE->getType().getCanonicalType();
1465 
1466   // Push each sub-region onto the stack.
1467   std::stack<const TypedValueRegion *> TVRStack;
1468   while (isa<FieldRegion>(TVR) || isa<ElementRegion>(TVR)) {
1469     // We found a region that matches the type of the init list,
1470     // so we assume this is the outer-most region. This can happen
1471     // if the initializer list is inside a class. If our assumption
1472     // is wrong, we return a nullptr in the end.
1473     if (ITy == TVR->getValueType().getCanonicalType())
1474       break;
1475 
1476     TVRStack.push(TVR);
1477     TVR = cast<TypedValueRegion>(TVR->getSuperRegion());
1478   }
1479 
1480   // If the type of the outer most region doesn't match the type
1481   // of the ILE, we can't match the ILE and the region.
1482   if (ITy != TVR->getValueType().getCanonicalType())
1483     return nullptr;
1484 
1485   const Expr *Init = ILE;
1486   while (!TVRStack.empty()) {
1487     TVR = TVRStack.top();
1488     TVRStack.pop();
1489 
1490     // We hit something that's not an init list before
1491     // running out of regions, so we most likely failed.
1492     if (!isa<InitListExpr>(Init))
1493       return nullptr;
1494 
1495     ILE = cast<InitListExpr>(Init);
1496     auto NumInits = ILE->getNumInits();
1497 
1498     if (const auto *FR = dyn_cast<FieldRegion>(TVR)) {
1499       const auto *FD = FR->getDecl();
1500 
1501       if (FD->getFieldIndex() >= NumInits)
1502         return nullptr;
1503 
1504       Init = ILE->getInit(FD->getFieldIndex());
1505     } else if (const auto *ER = dyn_cast<ElementRegion>(TVR)) {
1506       const auto Ind = ER->getIndex();
1507 
1508       // If index is symbolic, we can't figure out which expression
1509       // belongs to the region.
1510       if (!Ind.isConstant())
1511         return nullptr;
1512 
1513       const auto IndVal = Ind.getAsInteger()->getLimitedValue();
1514       if (IndVal >= NumInits)
1515         return nullptr;
1516 
1517       Init = ILE->getInit(IndVal);
1518     }
1519   }
1520 
1521   return Init;
1522 }
1523 
VisitNode(const ExplodedNode * Succ,BugReporterContext & BRC,PathSensitiveBugReport & BR)1524 PathDiagnosticPieceRef StoreSiteFinder::VisitNode(const ExplodedNode *Succ,
1525                                                   BugReporterContext &BRC,
1526                                                   PathSensitiveBugReport &BR) {
1527   if (Satisfied)
1528     return nullptr;
1529 
1530   const ExplodedNode *StoreSite = nullptr;
1531   const ExplodedNode *Pred = Succ->getFirstPred();
1532   const Expr *InitE = nullptr;
1533   bool IsParam = false;
1534 
1535   // First see if we reached the declaration of the region.
1536   if (const auto *VR = dyn_cast<VarRegion>(R)) {
1537     if (isInitializationOfVar(Pred, VR)) {
1538       StoreSite = Pred;
1539       InitE = VR->getDecl()->getInit();
1540     }
1541   }
1542 
1543   // If this is a post initializer expression, initializing the region, we
1544   // should track the initializer expression.
1545   if (std::optional<PostInitializer> PIP =
1546           Pred->getLocationAs<PostInitializer>()) {
1547     const MemRegion *FieldReg = (const MemRegion *)PIP->getLocationValue();
1548     if (FieldReg == R) {
1549       StoreSite = Pred;
1550       InitE = PIP->getInitializer()->getInit();
1551     }
1552   }
1553 
1554   // Otherwise, see if this is the store site:
1555   // (1) Succ has this binding and Pred does not, i.e. this is
1556   //     where the binding first occurred.
1557   // (2) Succ has this binding and is a PostStore node for this region, i.e.
1558   //     the same binding was re-assigned here.
1559   if (!StoreSite) {
1560     if (Succ->getState()->getSVal(R) != V)
1561       return nullptr;
1562 
1563     if (hasVisibleUpdate(Pred, Pred->getState()->getSVal(R), Succ, V)) {
1564       std::optional<PostStore> PS = Succ->getLocationAs<PostStore>();
1565       if (!PS || PS->getLocationValue() != R)
1566         return nullptr;
1567     }
1568 
1569     StoreSite = Succ;
1570 
1571     if (std::optional<PostStmt> P = Succ->getLocationAs<PostStmt>()) {
1572       // If this is an assignment expression, we can track the value
1573       // being assigned.
1574       if (const BinaryOperator *BO = P->getStmtAs<BinaryOperator>()) {
1575         if (BO->isAssignmentOp())
1576           InitE = BO->getRHS();
1577       }
1578       // If we have a declaration like 'S s{1,2}' that needs special
1579       // handling, we handle it here.
1580       else if (const auto *DS = P->getStmtAs<DeclStmt>()) {
1581         const auto *Decl = DS->getSingleDecl();
1582         if (isa<VarDecl>(Decl)) {
1583           const auto *VD = cast<VarDecl>(Decl);
1584 
1585           // FIXME: Here we only track the inner most region, so we lose
1586           // information, but it's still better than a crash or no information
1587           // at all.
1588           //
1589           // E.g.: The region we have is 's.s2.s3.s4.y' and we only track 'y',
1590           // and throw away the rest.
1591           if (const auto *ILE = dyn_cast<InitListExpr>(VD->getInit()))
1592             InitE = tryExtractInitializerFromList(ILE, R);
1593         }
1594       } else if (const auto *CE = P->getStmtAs<CXXConstructExpr>()) {
1595 
1596         const auto State = Succ->getState();
1597 
1598         if (isTrivialCopyOrMoveCtor(CE) && isa<SubRegion>(R)) {
1599           // Migrate the field regions from the current object to
1600           // the parent object. If we track 'a.y.e' and encounter
1601           // 'S a = b' then we need to track 'b.y.e'.
1602 
1603           // Push the regions to a stack, from last to first, so
1604           // considering the example above the stack will look like
1605           // (bottom) 'e' -> 'y' (top).
1606 
1607           std::stack<const SubRegion *> SRStack;
1608           const SubRegion *SR = cast<SubRegion>(R);
1609           while (isa<FieldRegion>(SR) || isa<ElementRegion>(SR)) {
1610             SRStack.push(SR);
1611             SR = cast<SubRegion>(SR->getSuperRegion());
1612           }
1613 
1614           // Get the region for the object we copied/moved from.
1615           const auto *OriginEx = CE->getArg(0);
1616           const auto OriginVal =
1617               State->getSVal(OriginEx, Succ->getLocationContext());
1618 
1619           // Pop the stored field regions and apply them to the origin
1620           // object in the same order we had them on the copy.
1621           // OriginField will evolve like 'b' -> 'b.y' -> 'b.y.e'.
1622           SVal OriginField = OriginVal;
1623           while (!SRStack.empty()) {
1624             const auto *TopR = SRStack.top();
1625             SRStack.pop();
1626 
1627             if (const auto *FR = dyn_cast<FieldRegion>(TopR)) {
1628               OriginField = State->getLValue(FR->getDecl(), OriginField);
1629             } else if (const auto *ER = dyn_cast<ElementRegion>(TopR)) {
1630               OriginField = State->getLValue(ER->getElementType(),
1631                                              ER->getIndex(), OriginField);
1632             } else {
1633               // FIXME: handle other region type
1634             }
1635           }
1636 
1637           // Track 'b.y.e'.
1638           getParentTracker().track(V, OriginField.getAsRegion(), Options);
1639           InitE = OriginEx;
1640         }
1641       }
1642       // This branch can occur in cases like `Ctor() : field{ x, y } {}'.
1643       else if (const auto *ILE = P->getStmtAs<InitListExpr>()) {
1644         // FIXME: Here we only track the top level region, so we lose
1645         // information, but it's still better than a crash or no information
1646         // at all.
1647         //
1648         // E.g.: The region we have is 's.s2.s3.s4.y' and we only track 'y', and
1649         // throw away the rest.
1650         InitE = tryExtractInitializerFromList(ILE, R);
1651       }
1652     }
1653 
1654     // If this is a call entry, the variable should be a parameter.
1655     // FIXME: Handle CXXThisRegion as well. (This is not a priority because
1656     // 'this' should never be NULL, but this visitor isn't just for NULL and
1657     // UndefinedVal.)
1658     if (std::optional<CallEnter> CE = Succ->getLocationAs<CallEnter>()) {
1659       if (const auto *VR = dyn_cast<VarRegion>(R)) {
1660 
1661         if (const auto *Param = dyn_cast<ParmVarDecl>(VR->getDecl())) {
1662           ProgramStateManager &StateMgr = BRC.getStateManager();
1663           CallEventManager &CallMgr = StateMgr.getCallEventManager();
1664 
1665           CallEventRef<> Call = CallMgr.getCaller(CE->getCalleeContext(),
1666                                                   Succ->getState());
1667           InitE = Call->getArgExpr(Param->getFunctionScopeIndex());
1668         } else {
1669           // Handle Objective-C 'self'.
1670           assert(isa<ImplicitParamDecl>(VR->getDecl()));
1671           InitE = cast<ObjCMessageExpr>(CE->getCalleeContext()->getCallSite())
1672                       ->getInstanceReceiver()->IgnoreParenCasts();
1673         }
1674         IsParam = true;
1675       }
1676     }
1677 
1678     // If this is a CXXTempObjectRegion, the Expr responsible for its creation
1679     // is wrapped inside of it.
1680     if (const auto *TmpR = dyn_cast<CXXTempObjectRegion>(R))
1681       InitE = TmpR->getExpr();
1682   }
1683 
1684   if (!StoreSite)
1685     return nullptr;
1686 
1687   Satisfied = true;
1688 
1689   // If we have an expression that provided the value, try to track where it
1690   // came from.
1691   if (InitE) {
1692     if (!IsParam)
1693       InitE = InitE->IgnoreParenCasts();
1694 
1695     getParentTracker().track(InitE, StoreSite, Options);
1696   }
1697 
1698   // Let's try to find the region where the value came from.
1699   const MemRegion *OldRegion = nullptr;
1700 
1701   // If we have init expression, it might be simply a reference
1702   // to a variable, so we can use it.
1703   if (InitE) {
1704     // That region might still be not exactly what we are looking for.
1705     // In situations like `int &ref = val;`, we can't say that
1706     // `ref` is initialized with `val`, rather refers to `val`.
1707     //
1708     // In order, to mitigate situations like this, we check if the last
1709     // stored value in that region is the value that we track.
1710     //
1711     // TODO: support other situations better.
1712     if (const MemRegion *Candidate =
1713             getLocationRegionIfReference(InitE, Succ, false)) {
1714       const StoreManager &SM = BRC.getStateManager().getStoreManager();
1715 
1716       // Here we traverse the graph up to find the last node where the
1717       // candidate region is still in the store.
1718       for (const ExplodedNode *N = StoreSite; N; N = N->getFirstPred()) {
1719         if (SM.includedInBindings(N->getState()->getStore(), Candidate)) {
1720           // And if it was bound to the target value, we can use it.
1721           if (N->getState()->getSVal(Candidate) == V) {
1722             OldRegion = Candidate;
1723           }
1724           break;
1725         }
1726       }
1727     }
1728   }
1729 
1730   // Otherwise, if the current region does indeed contain the value
1731   // we are looking for, we can look for a region where this value
1732   // was before.
1733   //
1734   // It can be useful for situations like:
1735   //     new = identity(old)
1736   // where the analyzer knows that 'identity' returns the value of its
1737   // first argument.
1738   //
1739   // NOTE: If the region R is not a simple var region, it can contain
1740   //       V in one of its subregions.
1741   if (!OldRegion && StoreSite->getState()->getSVal(R) == V) {
1742     // Let's go up the graph to find the node where the region is
1743     // bound to V.
1744     const ExplodedNode *NodeWithoutBinding = StoreSite->getFirstPred();
1745     for (;
1746          NodeWithoutBinding && NodeWithoutBinding->getState()->getSVal(R) == V;
1747          NodeWithoutBinding = NodeWithoutBinding->getFirstPred()) {
1748     }
1749 
1750     if (NodeWithoutBinding) {
1751       // Let's try to find a unique binding for the value in that node.
1752       // We want to use this to find unique bindings because of the following
1753       // situations:
1754       //     b = a;
1755       //     c = identity(b);
1756       //
1757       // Telling the user that the value of 'a' is assigned to 'c', while
1758       // correct, can be confusing.
1759       StoreManager::FindUniqueBinding FB(V.getAsLocSymbol());
1760       BRC.getStateManager().iterBindings(NodeWithoutBinding->getState(), FB);
1761       if (FB)
1762         OldRegion = FB.getRegion();
1763     }
1764   }
1765 
1766   if (Options.Kind == TrackingKind::Condition && OriginSFC &&
1767       !OriginSFC->isParentOf(StoreSite->getStackFrame()))
1768     return nullptr;
1769 
1770   // Okay, we've found the binding. Emit an appropriate message.
1771   SmallString<256> sbuf;
1772   llvm::raw_svector_ostream os(sbuf);
1773 
1774   StoreInfo SI = {StoreInfo::Assignment, // default kind
1775                   StoreSite,
1776                   InitE,
1777                   V,
1778                   R,
1779                   OldRegion};
1780 
1781   if (std::optional<PostStmt> PS = StoreSite->getLocationAs<PostStmt>()) {
1782     const Stmt *S = PS->getStmt();
1783     const auto *DS = dyn_cast<DeclStmt>(S);
1784     const auto *VR = dyn_cast<VarRegion>(R);
1785 
1786     if (DS) {
1787       SI.StoreKind = StoreInfo::Initialization;
1788     } else if (isa<BlockExpr>(S)) {
1789       SI.StoreKind = StoreInfo::BlockCapture;
1790       if (VR) {
1791         // See if we can get the BlockVarRegion.
1792         ProgramStateRef State = StoreSite->getState();
1793         SVal V = StoreSite->getSVal(S);
1794         if (const auto *BDR =
1795                 dyn_cast_or_null<BlockDataRegion>(V.getAsRegion())) {
1796           if (const VarRegion *OriginalR = BDR->getOriginalRegion(VR)) {
1797             getParentTracker().track(State->getSVal(OriginalR), OriginalR,
1798                                      Options, OriginSFC);
1799           }
1800         }
1801       }
1802     }
1803   } else if (SI.StoreSite->getLocation().getAs<CallEnter>() &&
1804              isa<VarRegion>(SI.Dest)) {
1805     SI.StoreKind = StoreInfo::CallArgument;
1806   }
1807 
1808   return getParentTracker().handle(SI, BRC, Options);
1809 }
1810 
1811 //===----------------------------------------------------------------------===//
1812 // Implementation of TrackConstraintBRVisitor.
1813 //===----------------------------------------------------------------------===//
1814 
Profile(llvm::FoldingSetNodeID & ID) const1815 void TrackConstraintBRVisitor::Profile(llvm::FoldingSetNodeID &ID) const {
1816   static int tag = 0;
1817   ID.AddPointer(&tag);
1818   ID.AddString(Message);
1819   ID.AddBoolean(Assumption);
1820   ID.Add(Constraint);
1821 }
1822 
1823 /// Return the tag associated with this visitor.  This tag will be used
1824 /// to make all PathDiagnosticPieces created by this visitor.
getTag()1825 const char *TrackConstraintBRVisitor::getTag() {
1826   return "TrackConstraintBRVisitor";
1827 }
1828 
isZeroCheck() const1829 bool TrackConstraintBRVisitor::isZeroCheck() const {
1830   return !Assumption && Constraint.getAs<Loc>();
1831 }
1832 
isUnderconstrained(const ExplodedNode * N) const1833 bool TrackConstraintBRVisitor::isUnderconstrained(const ExplodedNode *N) const {
1834   if (isZeroCheck())
1835     return N->getState()->isNull(Constraint).isUnderconstrained();
1836   return (bool)N->getState()->assume(Constraint, !Assumption);
1837 }
1838 
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport &)1839 PathDiagnosticPieceRef TrackConstraintBRVisitor::VisitNode(
1840     const ExplodedNode *N, BugReporterContext &BRC, PathSensitiveBugReport &) {
1841   const ExplodedNode *PrevN = N->getFirstPred();
1842   if (IsSatisfied)
1843     return nullptr;
1844 
1845   // Start tracking after we see the first state in which the value is
1846   // constrained.
1847   if (!IsTrackingTurnedOn)
1848     if (!isUnderconstrained(N))
1849       IsTrackingTurnedOn = true;
1850   if (!IsTrackingTurnedOn)
1851     return nullptr;
1852 
1853   // Check if in the previous state it was feasible for this constraint
1854   // to *not* be true.
1855   if (isUnderconstrained(PrevN)) {
1856     IsSatisfied = true;
1857 
1858     // At this point, the negation of the constraint should be infeasible. If it
1859     // is feasible, make sure that the negation of the constrainti was
1860     // infeasible in the current state.  If it is feasible, we somehow missed
1861     // the transition point.
1862     assert(!isUnderconstrained(N));
1863 
1864     // Construct a new PathDiagnosticPiece.
1865     ProgramPoint P = N->getLocation();
1866 
1867     // If this node already have a specialized note, it's probably better
1868     // than our generic note.
1869     // FIXME: This only looks for note tags, not for other ways to add a note.
1870     if (isa_and_nonnull<NoteTag>(P.getTag()))
1871       return nullptr;
1872 
1873     PathDiagnosticLocation L =
1874       PathDiagnosticLocation::create(P, BRC.getSourceManager());
1875     if (!L.isValid())
1876       return nullptr;
1877 
1878     auto X = std::make_shared<PathDiagnosticEventPiece>(L, Message);
1879     X->setTag(getTag());
1880     return std::move(X);
1881   }
1882 
1883   return nullptr;
1884 }
1885 
1886 //===----------------------------------------------------------------------===//
1887 // Implementation of SuppressInlineDefensiveChecksVisitor.
1888 //===----------------------------------------------------------------------===//
1889 
1890 SuppressInlineDefensiveChecksVisitor::
SuppressInlineDefensiveChecksVisitor(DefinedSVal Value,const ExplodedNode * N)1891 SuppressInlineDefensiveChecksVisitor(DefinedSVal Value, const ExplodedNode *N)
1892     : V(Value) {
1893   // Check if the visitor is disabled.
1894   AnalyzerOptions &Options = N->getState()->getAnalysisManager().options;
1895   if (!Options.ShouldSuppressInlinedDefensiveChecks)
1896     IsSatisfied = true;
1897 }
1898 
Profile(llvm::FoldingSetNodeID & ID) const1899 void SuppressInlineDefensiveChecksVisitor::Profile(
1900     llvm::FoldingSetNodeID &ID) const {
1901   static int id = 0;
1902   ID.AddPointer(&id);
1903   ID.Add(V);
1904 }
1905 
getTag()1906 const char *SuppressInlineDefensiveChecksVisitor::getTag() {
1907   return "IDCVisitor";
1908 }
1909 
1910 PathDiagnosticPieceRef
VisitNode(const ExplodedNode * Succ,BugReporterContext & BRC,PathSensitiveBugReport & BR)1911 SuppressInlineDefensiveChecksVisitor::VisitNode(const ExplodedNode *Succ,
1912                                                 BugReporterContext &BRC,
1913                                                 PathSensitiveBugReport &BR) {
1914   const ExplodedNode *Pred = Succ->getFirstPred();
1915   if (IsSatisfied)
1916     return nullptr;
1917 
1918   // Start tracking after we see the first state in which the value is null.
1919   if (!IsTrackingTurnedOn)
1920     if (Succ->getState()->isNull(V).isConstrainedTrue())
1921       IsTrackingTurnedOn = true;
1922   if (!IsTrackingTurnedOn)
1923     return nullptr;
1924 
1925   // Check if in the previous state it was feasible for this value
1926   // to *not* be null.
1927   if (!Pred->getState()->isNull(V).isConstrainedTrue() &&
1928       Succ->getState()->isNull(V).isConstrainedTrue()) {
1929     IsSatisfied = true;
1930 
1931     // Check if this is inlined defensive checks.
1932     const LocationContext *CurLC = Succ->getLocationContext();
1933     const LocationContext *ReportLC = BR.getErrorNode()->getLocationContext();
1934     if (CurLC != ReportLC && !CurLC->isParentOf(ReportLC)) {
1935       BR.markInvalid("Suppress IDC", CurLC);
1936       return nullptr;
1937     }
1938 
1939     // Treat defensive checks in function-like macros as if they were an inlined
1940     // defensive check. If the bug location is not in a macro and the
1941     // terminator for the current location is in a macro then suppress the
1942     // warning.
1943     auto BugPoint = BR.getErrorNode()->getLocation().getAs<StmtPoint>();
1944 
1945     if (!BugPoint)
1946       return nullptr;
1947 
1948     ProgramPoint CurPoint = Succ->getLocation();
1949     const Stmt *CurTerminatorStmt = nullptr;
1950     if (auto BE = CurPoint.getAs<BlockEdge>()) {
1951       CurTerminatorStmt = BE->getSrc()->getTerminator().getStmt();
1952     } else if (auto SP = CurPoint.getAs<StmtPoint>()) {
1953       const Stmt *CurStmt = SP->getStmt();
1954       if (!CurStmt->getBeginLoc().isMacroID())
1955         return nullptr;
1956 
1957       CFGStmtMap *Map = CurLC->getAnalysisDeclContext()->getCFGStmtMap();
1958       CurTerminatorStmt = Map->getBlock(CurStmt)->getTerminatorStmt();
1959     } else {
1960       return nullptr;
1961     }
1962 
1963     if (!CurTerminatorStmt)
1964       return nullptr;
1965 
1966     SourceLocation TerminatorLoc = CurTerminatorStmt->getBeginLoc();
1967     if (TerminatorLoc.isMacroID()) {
1968       SourceLocation BugLoc = BugPoint->getStmt()->getBeginLoc();
1969 
1970       // Suppress reports unless we are in that same macro.
1971       if (!BugLoc.isMacroID() ||
1972           getMacroName(BugLoc, BRC) != getMacroName(TerminatorLoc, BRC)) {
1973         BR.markInvalid("Suppress Macro IDC", CurLC);
1974       }
1975       return nullptr;
1976     }
1977   }
1978   return nullptr;
1979 }
1980 
1981 //===----------------------------------------------------------------------===//
1982 // TrackControlDependencyCondBRVisitor.
1983 //===----------------------------------------------------------------------===//
1984 
1985 namespace {
1986 /// Tracks the expressions that are a control dependency of the node that was
1987 /// supplied to the constructor.
1988 /// For example:
1989 ///
1990 ///   cond = 1;
1991 ///   if (cond)
1992 ///     10 / 0;
1993 ///
1994 /// An error is emitted at line 3. This visitor realizes that the branch
1995 /// on line 2 is a control dependency of line 3, and tracks it's condition via
1996 /// trackExpressionValue().
1997 class TrackControlDependencyCondBRVisitor final
1998     : public TrackingBugReporterVisitor {
1999   const ExplodedNode *Origin;
2000   ControlDependencyCalculator ControlDeps;
2001   llvm::SmallSet<const CFGBlock *, 32> VisitedBlocks;
2002 
2003 public:
TrackControlDependencyCondBRVisitor(TrackerRef ParentTracker,const ExplodedNode * O)2004   TrackControlDependencyCondBRVisitor(TrackerRef ParentTracker,
2005                                       const ExplodedNode *O)
2006       : TrackingBugReporterVisitor(ParentTracker), Origin(O),
2007         ControlDeps(&O->getCFG()) {}
2008 
Profile(llvm::FoldingSetNodeID & ID) const2009   void Profile(llvm::FoldingSetNodeID &ID) const override {
2010     static int x = 0;
2011     ID.AddPointer(&x);
2012   }
2013 
2014   PathDiagnosticPieceRef VisitNode(const ExplodedNode *N,
2015                                    BugReporterContext &BRC,
2016                                    PathSensitiveBugReport &BR) override;
2017 };
2018 } // end of anonymous namespace
2019 
2020 static std::shared_ptr<PathDiagnosticEventPiece>
constructDebugPieceForTrackedCondition(const Expr * Cond,const ExplodedNode * N,BugReporterContext & BRC)2021 constructDebugPieceForTrackedCondition(const Expr *Cond,
2022                                        const ExplodedNode *N,
2023                                        BugReporterContext &BRC) {
2024 
2025   if (BRC.getAnalyzerOptions().AnalysisDiagOpt == PD_NONE ||
2026       !BRC.getAnalyzerOptions().ShouldTrackConditionsDebug)
2027     return nullptr;
2028 
2029   std::string ConditionText = std::string(Lexer::getSourceText(
2030       CharSourceRange::getTokenRange(Cond->getSourceRange()),
2031       BRC.getSourceManager(), BRC.getASTContext().getLangOpts()));
2032 
2033   return std::make_shared<PathDiagnosticEventPiece>(
2034       PathDiagnosticLocation::createBegin(
2035           Cond, BRC.getSourceManager(), N->getLocationContext()),
2036           (Twine() + "Tracking condition '" + ConditionText + "'").str());
2037 }
2038 
isAssertlikeBlock(const CFGBlock * B,ASTContext & Context)2039 static bool isAssertlikeBlock(const CFGBlock *B, ASTContext &Context) {
2040   if (B->succ_size() != 2)
2041     return false;
2042 
2043   const CFGBlock *Then = B->succ_begin()->getReachableBlock();
2044   const CFGBlock *Else = (B->succ_begin() + 1)->getReachableBlock();
2045 
2046   if (!Then || !Else)
2047     return false;
2048 
2049   if (Then->isInevitablySinking() != Else->isInevitablySinking())
2050     return true;
2051 
2052   // For the following condition the following CFG would be built:
2053   //
2054   //                          ------------->
2055   //                         /              \
2056   //                       [B1] -> [B2] -> [B3] -> [sink]
2057   // assert(A && B || C);            \       \
2058   //                                  -----------> [go on with the execution]
2059   //
2060   // It so happens that CFGBlock::getTerminatorCondition returns 'A' for block
2061   // B1, 'A && B' for B2, and 'A && B || C' for B3. Let's check whether we
2062   // reached the end of the condition!
2063   if (const Stmt *ElseCond = Else->getTerminatorCondition())
2064     if (const auto *BinOp = dyn_cast<BinaryOperator>(ElseCond))
2065       if (BinOp->isLogicalOp())
2066         return isAssertlikeBlock(Else, Context);
2067 
2068   return false;
2069 }
2070 
2071 PathDiagnosticPieceRef
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)2072 TrackControlDependencyCondBRVisitor::VisitNode(const ExplodedNode *N,
2073                                                BugReporterContext &BRC,
2074                                                PathSensitiveBugReport &BR) {
2075   // We can only reason about control dependencies within the same stack frame.
2076   if (Origin->getStackFrame() != N->getStackFrame())
2077     return nullptr;
2078 
2079   CFGBlock *NB = const_cast<CFGBlock *>(N->getCFGBlock());
2080 
2081   // Skip if we already inspected this block.
2082   if (!VisitedBlocks.insert(NB).second)
2083     return nullptr;
2084 
2085   CFGBlock *OriginB = const_cast<CFGBlock *>(Origin->getCFGBlock());
2086 
2087   // TODO: Cache CFGBlocks for each ExplodedNode.
2088   if (!OriginB || !NB)
2089     return nullptr;
2090 
2091   if (isAssertlikeBlock(NB, BRC.getASTContext()))
2092     return nullptr;
2093 
2094   if (ControlDeps.isControlDependent(OriginB, NB)) {
2095     // We don't really want to explain for range loops. Evidence suggests that
2096     // the only thing that leads to is the addition of calls to operator!=.
2097     if (llvm::isa_and_nonnull<CXXForRangeStmt>(NB->getTerminatorStmt()))
2098       return nullptr;
2099 
2100     if (const Expr *Condition = NB->getLastCondition()) {
2101 
2102       // If we can't retrieve a sensible condition, just bail out.
2103       const Expr *InnerExpr = peelOffOuterExpr(Condition, N);
2104       if (!InnerExpr)
2105         return nullptr;
2106 
2107       // If the condition was a function call, we likely won't gain much from
2108       // tracking it either. Evidence suggests that it will mostly trigger in
2109       // scenarios like this:
2110       //
2111       //   void f(int *x) {
2112       //     x = nullptr;
2113       //     if (alwaysTrue()) // We don't need a whole lot of explanation
2114       //                       // here, the function name is good enough.
2115       //       *x = 5;
2116       //   }
2117       //
2118       // Its easy to create a counterexample where this heuristic would make us
2119       // lose valuable information, but we've never really seen one in practice.
2120       if (isa<CallExpr>(InnerExpr))
2121         return nullptr;
2122 
2123       // Keeping track of the already tracked conditions on a visitor level
2124       // isn't sufficient, because a new visitor is created for each tracked
2125       // expression, hence the BugReport level set.
2126       if (BR.addTrackedCondition(N)) {
2127         getParentTracker().track(InnerExpr, N,
2128                                  {bugreporter::TrackingKind::Condition,
2129                                   /*EnableNullFPSuppression=*/false});
2130         return constructDebugPieceForTrackedCondition(Condition, N, BRC);
2131       }
2132     }
2133   }
2134 
2135   return nullptr;
2136 }
2137 
2138 //===----------------------------------------------------------------------===//
2139 // Implementation of trackExpressionValue.
2140 //===----------------------------------------------------------------------===//
2141 
peelOffOuterExpr(const Expr * Ex,const ExplodedNode * N)2142 static const Expr *peelOffOuterExpr(const Expr *Ex, const ExplodedNode *N) {
2143 
2144   Ex = Ex->IgnoreParenCasts();
2145   if (const auto *FE = dyn_cast<FullExpr>(Ex))
2146     return peelOffOuterExpr(FE->getSubExpr(), N);
2147   if (const auto *OVE = dyn_cast<OpaqueValueExpr>(Ex))
2148     return peelOffOuterExpr(OVE->getSourceExpr(), N);
2149   if (const auto *POE = dyn_cast<PseudoObjectExpr>(Ex)) {
2150     const auto *PropRef = dyn_cast<ObjCPropertyRefExpr>(POE->getSyntacticForm());
2151     if (PropRef && PropRef->isMessagingGetter()) {
2152       const Expr *GetterMessageSend =
2153           POE->getSemanticExpr(POE->getNumSemanticExprs() - 1);
2154       assert(isa<ObjCMessageExpr>(GetterMessageSend->IgnoreParenCasts()));
2155       return peelOffOuterExpr(GetterMessageSend, N);
2156     }
2157   }
2158 
2159   // Peel off the ternary operator.
2160   if (const auto *CO = dyn_cast<ConditionalOperator>(Ex)) {
2161     // Find a node where the branching occurred and find out which branch
2162     // we took (true/false) by looking at the ExplodedGraph.
2163     const ExplodedNode *NI = N;
2164     do {
2165       ProgramPoint ProgPoint = NI->getLocation();
2166       if (std::optional<BlockEdge> BE = ProgPoint.getAs<BlockEdge>()) {
2167         const CFGBlock *srcBlk = BE->getSrc();
2168         if (const Stmt *term = srcBlk->getTerminatorStmt()) {
2169           if (term == CO) {
2170             bool TookTrueBranch = (*(srcBlk->succ_begin()) == BE->getDst());
2171             if (TookTrueBranch)
2172               return peelOffOuterExpr(CO->getTrueExpr(), N);
2173             else
2174               return peelOffOuterExpr(CO->getFalseExpr(), N);
2175           }
2176         }
2177       }
2178       NI = NI->getFirstPred();
2179     } while (NI);
2180   }
2181 
2182   if (auto *BO = dyn_cast<BinaryOperator>(Ex))
2183     if (const Expr *SubEx = peelOffPointerArithmetic(BO))
2184       return peelOffOuterExpr(SubEx, N);
2185 
2186   if (auto *UO = dyn_cast<UnaryOperator>(Ex)) {
2187     if (UO->getOpcode() == UO_LNot)
2188       return peelOffOuterExpr(UO->getSubExpr(), N);
2189 
2190     // FIXME: There's a hack in our Store implementation that always computes
2191     // field offsets around null pointers as if they are always equal to 0.
2192     // The idea here is to report accesses to fields as null dereferences
2193     // even though the pointer value that's being dereferenced is actually
2194     // the offset of the field rather than exactly 0.
2195     // See the FIXME in StoreManager's getLValueFieldOrIvar() method.
2196     // This code interacts heavily with this hack; otherwise the value
2197     // would not be null at all for most fields, so we'd be unable to track it.
2198     if (UO->getOpcode() == UO_AddrOf && UO->getSubExpr()->isLValue())
2199       if (const Expr *DerefEx = bugreporter::getDerefExpr(UO->getSubExpr()))
2200         return peelOffOuterExpr(DerefEx, N);
2201   }
2202 
2203   return Ex;
2204 }
2205 
2206 /// Find the ExplodedNode where the lvalue (the value of 'Ex')
2207 /// was computed.
findNodeForExpression(const ExplodedNode * N,const Expr * Inner)2208 static const ExplodedNode* findNodeForExpression(const ExplodedNode *N,
2209                                                  const Expr *Inner) {
2210   while (N) {
2211     if (N->getStmtForDiagnostics() == Inner)
2212       return N;
2213     N = N->getFirstPred();
2214   }
2215   return N;
2216 }
2217 
2218 //===----------------------------------------------------------------------===//
2219 //                            Tracker implementation
2220 //===----------------------------------------------------------------------===//
2221 
constructNote(StoreInfo SI,BugReporterContext & BRC,StringRef NodeText)2222 PathDiagnosticPieceRef StoreHandler::constructNote(StoreInfo SI,
2223                                                    BugReporterContext &BRC,
2224                                                    StringRef NodeText) {
2225   // Construct a new PathDiagnosticPiece.
2226   ProgramPoint P = SI.StoreSite->getLocation();
2227   PathDiagnosticLocation L;
2228   if (P.getAs<CallEnter>() && SI.SourceOfTheValue)
2229     L = PathDiagnosticLocation(SI.SourceOfTheValue, BRC.getSourceManager(),
2230                                P.getLocationContext());
2231 
2232   if (!L.isValid() || !L.asLocation().isValid())
2233     L = PathDiagnosticLocation::create(P, BRC.getSourceManager());
2234 
2235   if (!L.isValid() || !L.asLocation().isValid())
2236     return nullptr;
2237 
2238   return std::make_shared<PathDiagnosticEventPiece>(L, NodeText);
2239 }
2240 
2241 namespace {
2242 class DefaultStoreHandler final : public StoreHandler {
2243 public:
2244   using StoreHandler::StoreHandler;
2245 
handle(StoreInfo SI,BugReporterContext & BRC,TrackingOptions Opts)2246   PathDiagnosticPieceRef handle(StoreInfo SI, BugReporterContext &BRC,
2247                                 TrackingOptions Opts) override {
2248     // Okay, we've found the binding. Emit an appropriate message.
2249     SmallString<256> Buffer;
2250     llvm::raw_svector_ostream OS(Buffer);
2251 
2252     switch (SI.StoreKind) {
2253     case StoreInfo::Initialization:
2254     case StoreInfo::BlockCapture:
2255       showBRDiagnostics(OS, SI);
2256       break;
2257     case StoreInfo::CallArgument:
2258       showBRParamDiagnostics(OS, SI);
2259       break;
2260     case StoreInfo::Assignment:
2261       showBRDefaultDiagnostics(OS, SI);
2262       break;
2263     }
2264 
2265     if (Opts.Kind == bugreporter::TrackingKind::Condition)
2266       OS << WillBeUsedForACondition;
2267 
2268     return constructNote(SI, BRC, OS.str());
2269   }
2270 };
2271 
2272 class ControlDependencyHandler final : public ExpressionHandler {
2273 public:
2274   using ExpressionHandler::ExpressionHandler;
2275 
handle(const Expr * Inner,const ExplodedNode * InputNode,const ExplodedNode * LVNode,TrackingOptions Opts)2276   Tracker::Result handle(const Expr *Inner, const ExplodedNode *InputNode,
2277                          const ExplodedNode *LVNode,
2278                          TrackingOptions Opts) override {
2279     PathSensitiveBugReport &Report = getParentTracker().getReport();
2280 
2281     // We only track expressions if we believe that they are important. Chances
2282     // are good that control dependencies to the tracking point are also
2283     // important because of this, let's explain why we believe control reached
2284     // this point.
2285     // TODO: Shouldn't we track control dependencies of every bug location,
2286     // rather than only tracked expressions?
2287     if (LVNode->getState()
2288             ->getAnalysisManager()
2289             .getAnalyzerOptions()
2290             .ShouldTrackConditions) {
2291       Report.addVisitor<TrackControlDependencyCondBRVisitor>(
2292           &getParentTracker(), InputNode);
2293       return {/*FoundSomethingToTrack=*/true};
2294     }
2295 
2296     return {};
2297   }
2298 };
2299 
2300 class NilReceiverHandler final : public ExpressionHandler {
2301 public:
2302   using ExpressionHandler::ExpressionHandler;
2303 
handle(const Expr * Inner,const ExplodedNode * InputNode,const ExplodedNode * LVNode,TrackingOptions Opts)2304   Tracker::Result handle(const Expr *Inner, const ExplodedNode *InputNode,
2305                          const ExplodedNode *LVNode,
2306                          TrackingOptions Opts) override {
2307     // The message send could be nil due to the receiver being nil.
2308     // At this point in the path, the receiver should be live since we are at
2309     // the message send expr. If it is nil, start tracking it.
2310     if (const Expr *Receiver =
2311             NilReceiverBRVisitor::getNilReceiver(Inner, LVNode))
2312       return getParentTracker().track(Receiver, LVNode, Opts);
2313 
2314     return {};
2315   }
2316 };
2317 
2318 class ArrayIndexHandler final : public ExpressionHandler {
2319 public:
2320   using ExpressionHandler::ExpressionHandler;
2321 
handle(const Expr * Inner,const ExplodedNode * InputNode,const ExplodedNode * LVNode,TrackingOptions Opts)2322   Tracker::Result handle(const Expr *Inner, const ExplodedNode *InputNode,
2323                          const ExplodedNode *LVNode,
2324                          TrackingOptions Opts) override {
2325     // Track the index if this is an array subscript.
2326     if (const auto *Arr = dyn_cast<ArraySubscriptExpr>(Inner))
2327       return getParentTracker().track(
2328           Arr->getIdx(), LVNode,
2329           {Opts.Kind, /*EnableNullFPSuppression*/ false});
2330 
2331     return {};
2332   }
2333 };
2334 
2335 // TODO: extract it into more handlers
2336 class InterestingLValueHandler final : public ExpressionHandler {
2337 public:
2338   using ExpressionHandler::ExpressionHandler;
2339 
handle(const Expr * Inner,const ExplodedNode * InputNode,const ExplodedNode * LVNode,TrackingOptions Opts)2340   Tracker::Result handle(const Expr *Inner, const ExplodedNode *InputNode,
2341                          const ExplodedNode *LVNode,
2342                          TrackingOptions Opts) override {
2343     ProgramStateRef LVState = LVNode->getState();
2344     const StackFrameContext *SFC = LVNode->getStackFrame();
2345     PathSensitiveBugReport &Report = getParentTracker().getReport();
2346     Tracker::Result Result;
2347 
2348     // See if the expression we're interested refers to a variable.
2349     // If so, we can track both its contents and constraints on its value.
2350     if (ExplodedGraph::isInterestingLValueExpr(Inner)) {
2351       SVal LVal = LVNode->getSVal(Inner);
2352 
2353       const MemRegion *RR = getLocationRegionIfReference(Inner, LVNode);
2354       bool LVIsNull = LVState->isNull(LVal).isConstrainedTrue();
2355 
2356       // If this is a C++ reference to a null pointer, we are tracking the
2357       // pointer. In addition, we should find the store at which the reference
2358       // got initialized.
2359       if (RR && !LVIsNull)
2360         Result.combineWith(getParentTracker().track(LVal, RR, Opts, SFC));
2361 
2362       // In case of C++ references, we want to differentiate between a null
2363       // reference and reference to null pointer.
2364       // If the LVal is null, check if we are dealing with null reference.
2365       // For those, we want to track the location of the reference.
2366       const MemRegion *R =
2367           (RR && LVIsNull) ? RR : LVNode->getSVal(Inner).getAsRegion();
2368 
2369       if (R) {
2370 
2371         // Mark both the variable region and its contents as interesting.
2372         SVal V = LVState->getRawSVal(loc::MemRegionVal(R));
2373         Report.addVisitor<NoStoreFuncVisitor>(cast<SubRegion>(R), Opts.Kind);
2374 
2375         // When we got here, we do have something to track, and we will
2376         // interrupt.
2377         Result.FoundSomethingToTrack = true;
2378         Result.WasInterrupted = true;
2379 
2380         MacroNullReturnSuppressionVisitor::addMacroVisitorIfNecessary(
2381             LVNode, R, Opts.EnableNullFPSuppression, Report, V);
2382 
2383         Report.markInteresting(V, Opts.Kind);
2384         Report.addVisitor<UndefOrNullArgVisitor>(R);
2385 
2386         // If the contents are symbolic and null, find out when they became
2387         // null.
2388         if (V.getAsLocSymbol(/*IncludeBaseRegions=*/true))
2389           if (LVState->isNull(V).isConstrainedTrue())
2390             Report.addVisitor<TrackConstraintBRVisitor>(
2391                 V.castAs<DefinedSVal>(),
2392                 /*Assumption=*/false, "Assuming pointer value is null");
2393 
2394         // Add visitor, which will suppress inline defensive checks.
2395         if (auto DV = V.getAs<DefinedSVal>())
2396           if (!DV->isZeroConstant() && Opts.EnableNullFPSuppression)
2397             // Note that LVNode may be too late (i.e., too far from the
2398             // InputNode) because the lvalue may have been computed before the
2399             // inlined call was evaluated. InputNode may as well be too early
2400             // here, because the symbol is already dead; this, however, is fine
2401             // because we can still find the node in which it collapsed to null
2402             // previously.
2403             Report.addVisitor<SuppressInlineDefensiveChecksVisitor>(*DV,
2404                                                                     InputNode);
2405         getParentTracker().track(V, R, Opts, SFC);
2406       }
2407     }
2408 
2409     return Result;
2410   }
2411 };
2412 
2413 /// Adds a ReturnVisitor if the given statement represents a call that was
2414 /// inlined.
2415 ///
2416 /// This will search back through the ExplodedGraph, starting from the given
2417 /// node, looking for when the given statement was processed. If it turns out
2418 /// the statement is a call that was inlined, we add the visitor to the
2419 /// bug report, so it can print a note later.
2420 class InlinedFunctionCallHandler final : public ExpressionHandler {
2421   using ExpressionHandler::ExpressionHandler;
2422 
handle(const Expr * E,const ExplodedNode * InputNode,const ExplodedNode * ExprNode,TrackingOptions Opts)2423   Tracker::Result handle(const Expr *E, const ExplodedNode *InputNode,
2424                          const ExplodedNode *ExprNode,
2425                          TrackingOptions Opts) override {
2426     if (!CallEvent::isCallStmt(E))
2427       return {};
2428 
2429     // First, find when we processed the statement.
2430     // If we work with a 'CXXNewExpr' that is going to be purged away before
2431     // its call take place. We would catch that purge in the last condition
2432     // as a 'StmtPoint' so we have to bypass it.
2433     const bool BypassCXXNewExprEval = isa<CXXNewExpr>(E);
2434 
2435     // This is moving forward when we enter into another context.
2436     const StackFrameContext *CurrentSFC = ExprNode->getStackFrame();
2437 
2438     do {
2439       // If that is satisfied we found our statement as an inlined call.
2440       if (std::optional<CallExitEnd> CEE =
2441               ExprNode->getLocationAs<CallExitEnd>())
2442         if (CEE->getCalleeContext()->getCallSite() == E)
2443           break;
2444 
2445       // Try to move forward to the end of the call-chain.
2446       ExprNode = ExprNode->getFirstPred();
2447       if (!ExprNode)
2448         break;
2449 
2450       const StackFrameContext *PredSFC = ExprNode->getStackFrame();
2451 
2452       // If that is satisfied we found our statement.
2453       // FIXME: This code currently bypasses the call site for the
2454       //        conservatively evaluated allocator.
2455       if (!BypassCXXNewExprEval)
2456         if (std::optional<StmtPoint> SP = ExprNode->getLocationAs<StmtPoint>())
2457           // See if we do not enter into another context.
2458           if (SP->getStmt() == E && CurrentSFC == PredSFC)
2459             break;
2460 
2461       CurrentSFC = PredSFC;
2462     } while (ExprNode->getStackFrame() == CurrentSFC);
2463 
2464     // Next, step over any post-statement checks.
2465     while (ExprNode && ExprNode->getLocation().getAs<PostStmt>())
2466       ExprNode = ExprNode->getFirstPred();
2467     if (!ExprNode)
2468       return {};
2469 
2470     // Finally, see if we inlined the call.
2471     std::optional<CallExitEnd> CEE = ExprNode->getLocationAs<CallExitEnd>();
2472     if (!CEE)
2473       return {};
2474 
2475     const StackFrameContext *CalleeContext = CEE->getCalleeContext();
2476     if (CalleeContext->getCallSite() != E)
2477       return {};
2478 
2479     // Check the return value.
2480     ProgramStateRef State = ExprNode->getState();
2481     SVal RetVal = ExprNode->getSVal(E);
2482 
2483     // Handle cases where a reference is returned and then immediately used.
2484     if (cast<Expr>(E)->isGLValue())
2485       if (std::optional<Loc> LValue = RetVal.getAs<Loc>())
2486         RetVal = State->getSVal(*LValue);
2487 
2488     // See if the return value is NULL. If so, suppress the report.
2489     AnalyzerOptions &Options = State->getAnalysisManager().options;
2490 
2491     bool EnableNullFPSuppression = false;
2492     if (Opts.EnableNullFPSuppression && Options.ShouldSuppressNullReturnPaths)
2493       if (std::optional<Loc> RetLoc = RetVal.getAs<Loc>())
2494         EnableNullFPSuppression = State->isNull(*RetLoc).isConstrainedTrue();
2495 
2496     PathSensitiveBugReport &Report = getParentTracker().getReport();
2497     Report.addVisitor<ReturnVisitor>(&getParentTracker(), CalleeContext,
2498                                      EnableNullFPSuppression, Options,
2499                                      Opts.Kind);
2500     return {true};
2501   }
2502 };
2503 
2504 class DefaultExpressionHandler final : public ExpressionHandler {
2505 public:
2506   using ExpressionHandler::ExpressionHandler;
2507 
handle(const Expr * Inner,const ExplodedNode * InputNode,const ExplodedNode * LVNode,TrackingOptions Opts)2508   Tracker::Result handle(const Expr *Inner, const ExplodedNode *InputNode,
2509                          const ExplodedNode *LVNode,
2510                          TrackingOptions Opts) override {
2511     ProgramStateRef LVState = LVNode->getState();
2512     const StackFrameContext *SFC = LVNode->getStackFrame();
2513     PathSensitiveBugReport &Report = getParentTracker().getReport();
2514     Tracker::Result Result;
2515 
2516     // If the expression is not an "lvalue expression", we can still
2517     // track the constraints on its contents.
2518     SVal V = LVState->getSValAsScalarOrLoc(Inner, LVNode->getLocationContext());
2519 
2520     // Is it a symbolic value?
2521     if (auto L = V.getAs<loc::MemRegionVal>()) {
2522       // FIXME: this is a hack for fixing a later crash when attempting to
2523       // dereference a void* pointer.
2524       // We should not try to dereference pointers at all when we don't care
2525       // what is written inside the pointer.
2526       bool CanDereference = true;
2527       if (const auto *SR = L->getRegionAs<SymbolicRegion>()) {
2528         if (SR->getPointeeStaticType()->isVoidType())
2529           CanDereference = false;
2530       } else if (L->getRegionAs<AllocaRegion>())
2531         CanDereference = false;
2532 
2533       // At this point we are dealing with the region's LValue.
2534       // However, if the rvalue is a symbolic region, we should track it as
2535       // well. Try to use the correct type when looking up the value.
2536       SVal RVal;
2537       if (ExplodedGraph::isInterestingLValueExpr(Inner))
2538         RVal = LVState->getRawSVal(*L, Inner->getType());
2539       else if (CanDereference)
2540         RVal = LVState->getSVal(L->getRegion());
2541 
2542       if (CanDereference) {
2543         Report.addVisitor<UndefOrNullArgVisitor>(L->getRegion());
2544         Result.FoundSomethingToTrack = true;
2545 
2546         if (!RVal.isUnknown())
2547           Result.combineWith(
2548               getParentTracker().track(RVal, L->getRegion(), Opts, SFC));
2549       }
2550 
2551       const MemRegion *RegionRVal = RVal.getAsRegion();
2552       if (isa_and_nonnull<SymbolicRegion>(RegionRVal)) {
2553         Report.markInteresting(RegionRVal, Opts.Kind);
2554         Report.addVisitor<TrackConstraintBRVisitor>(
2555             loc::MemRegionVal(RegionRVal),
2556             /*Assumption=*/false, "Assuming pointer value is null");
2557         Result.FoundSomethingToTrack = true;
2558       }
2559     }
2560 
2561     return Result;
2562   }
2563 };
2564 
2565 /// Attempts to add visitors to track an RValue expression back to its point of
2566 /// origin.
2567 class PRValueHandler final : public ExpressionHandler {
2568 public:
2569   using ExpressionHandler::ExpressionHandler;
2570 
handle(const Expr * E,const ExplodedNode * InputNode,const ExplodedNode * ExprNode,TrackingOptions Opts)2571   Tracker::Result handle(const Expr *E, const ExplodedNode *InputNode,
2572                          const ExplodedNode *ExprNode,
2573                          TrackingOptions Opts) override {
2574     if (!E->isPRValue())
2575       return {};
2576 
2577     const ExplodedNode *RVNode = findNodeForExpression(ExprNode, E);
2578     if (!RVNode)
2579       return {};
2580 
2581     Tracker::Result CombinedResult;
2582     Tracker &Parent = getParentTracker();
2583 
2584     const auto track = [&CombinedResult, &Parent, ExprNode,
2585                         Opts](const Expr *Inner) {
2586       CombinedResult.combineWith(Parent.track(Inner, ExprNode, Opts));
2587     };
2588 
2589     // FIXME: Initializer lists can appear in many different contexts
2590     // and most of them needs a special handling. For now let's handle
2591     // what we can. If the initializer list only has 1 element, we track
2592     // that.
2593     // This snippet even handles nesting, e.g.: int *x{{{{{y}}}}};
2594     if (const auto *ILE = dyn_cast<InitListExpr>(E)) {
2595       if (ILE->getNumInits() == 1) {
2596         track(ILE->getInit(0));
2597 
2598         return CombinedResult;
2599       }
2600 
2601       return {};
2602     }
2603 
2604     ProgramStateRef RVState = RVNode->getState();
2605     SVal V = RVState->getSValAsScalarOrLoc(E, RVNode->getLocationContext());
2606     const auto *BO = dyn_cast<BinaryOperator>(E);
2607 
2608     if (!BO || !BO->isMultiplicativeOp() || !V.isZeroConstant())
2609       return {};
2610 
2611     SVal RHSV = RVState->getSVal(BO->getRHS(), RVNode->getLocationContext());
2612     SVal LHSV = RVState->getSVal(BO->getLHS(), RVNode->getLocationContext());
2613 
2614     // Track both LHS and RHS of a multiplication.
2615     if (BO->getOpcode() == BO_Mul) {
2616       if (LHSV.isZeroConstant())
2617         track(BO->getLHS());
2618       if (RHSV.isZeroConstant())
2619         track(BO->getRHS());
2620     } else { // Track only the LHS of a division or a modulo.
2621       if (LHSV.isZeroConstant())
2622         track(BO->getLHS());
2623     }
2624 
2625     return CombinedResult;
2626   }
2627 };
2628 } // namespace
2629 
Tracker(PathSensitiveBugReport & Report)2630 Tracker::Tracker(PathSensitiveBugReport &Report) : Report(Report) {
2631   // Default expression handlers.
2632   addLowPriorityHandler<ControlDependencyHandler>();
2633   addLowPriorityHandler<NilReceiverHandler>();
2634   addLowPriorityHandler<ArrayIndexHandler>();
2635   addLowPriorityHandler<InterestingLValueHandler>();
2636   addLowPriorityHandler<InlinedFunctionCallHandler>();
2637   addLowPriorityHandler<DefaultExpressionHandler>();
2638   addLowPriorityHandler<PRValueHandler>();
2639   // Default store handlers.
2640   addHighPriorityHandler<DefaultStoreHandler>();
2641 }
2642 
track(const Expr * E,const ExplodedNode * N,TrackingOptions Opts)2643 Tracker::Result Tracker::track(const Expr *E, const ExplodedNode *N,
2644                                TrackingOptions Opts) {
2645   if (!E || !N)
2646     return {};
2647 
2648   const Expr *Inner = peelOffOuterExpr(E, N);
2649   const ExplodedNode *LVNode = findNodeForExpression(N, Inner);
2650   if (!LVNode)
2651     return {};
2652 
2653   Result CombinedResult;
2654   // Iterate through the handlers in the order according to their priorities.
2655   for (ExpressionHandlerPtr &Handler : ExpressionHandlers) {
2656     CombinedResult.combineWith(Handler->handle(Inner, N, LVNode, Opts));
2657     if (CombinedResult.WasInterrupted) {
2658       // There is no need to confuse our users here.
2659       // We got interrupted, but our users don't need to know about it.
2660       CombinedResult.WasInterrupted = false;
2661       break;
2662     }
2663   }
2664 
2665   return CombinedResult;
2666 }
2667 
track(SVal V,const MemRegion * R,TrackingOptions Opts,const StackFrameContext * Origin)2668 Tracker::Result Tracker::track(SVal V, const MemRegion *R, TrackingOptions Opts,
2669                                const StackFrameContext *Origin) {
2670   if (!V.isUnknown()) {
2671     Report.addVisitor<StoreSiteFinder>(this, V, R, Opts, Origin);
2672     return {true};
2673   }
2674   return {};
2675 }
2676 
handle(StoreInfo SI,BugReporterContext & BRC,TrackingOptions Opts)2677 PathDiagnosticPieceRef Tracker::handle(StoreInfo SI, BugReporterContext &BRC,
2678                                        TrackingOptions Opts) {
2679   // Iterate through the handlers in the order according to their priorities.
2680   for (StoreHandlerPtr &Handler : StoreHandlers) {
2681     if (PathDiagnosticPieceRef Result = Handler->handle(SI, BRC, Opts))
2682       // If the handler produced a non-null piece, return it.
2683       // There is no need in asking other handlers.
2684       return Result;
2685   }
2686   return {};
2687 }
2688 
trackExpressionValue(const ExplodedNode * InputNode,const Expr * E,PathSensitiveBugReport & Report,TrackingOptions Opts)2689 bool bugreporter::trackExpressionValue(const ExplodedNode *InputNode,
2690                                        const Expr *E,
2691 
2692                                        PathSensitiveBugReport &Report,
2693                                        TrackingOptions Opts) {
2694   return Tracker::create(Report)
2695       ->track(E, InputNode, Opts)
2696       .FoundSomethingToTrack;
2697 }
2698 
trackStoredValue(SVal V,const MemRegion * R,PathSensitiveBugReport & Report,TrackingOptions Opts,const StackFrameContext * Origin)2699 void bugreporter::trackStoredValue(SVal V, const MemRegion *R,
2700                                    PathSensitiveBugReport &Report,
2701                                    TrackingOptions Opts,
2702                                    const StackFrameContext *Origin) {
2703   Tracker::create(Report)->track(V, R, Opts, Origin);
2704 }
2705 
2706 //===----------------------------------------------------------------------===//
2707 // Implementation of NulReceiverBRVisitor.
2708 //===----------------------------------------------------------------------===//
2709 
getNilReceiver(const Stmt * S,const ExplodedNode * N)2710 const Expr *NilReceiverBRVisitor::getNilReceiver(const Stmt *S,
2711                                                  const ExplodedNode *N) {
2712   const auto *ME = dyn_cast<ObjCMessageExpr>(S);
2713   if (!ME)
2714     return nullptr;
2715   if (const Expr *Receiver = ME->getInstanceReceiver()) {
2716     ProgramStateRef state = N->getState();
2717     SVal V = N->getSVal(Receiver);
2718     if (state->isNull(V).isConstrainedTrue())
2719       return Receiver;
2720   }
2721   return nullptr;
2722 }
2723 
2724 PathDiagnosticPieceRef
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)2725 NilReceiverBRVisitor::VisitNode(const ExplodedNode *N, BugReporterContext &BRC,
2726                                 PathSensitiveBugReport &BR) {
2727   std::optional<PreStmt> P = N->getLocationAs<PreStmt>();
2728   if (!P)
2729     return nullptr;
2730 
2731   const Stmt *S = P->getStmt();
2732   const Expr *Receiver = getNilReceiver(S, N);
2733   if (!Receiver)
2734     return nullptr;
2735 
2736   llvm::SmallString<256> Buf;
2737   llvm::raw_svector_ostream OS(Buf);
2738 
2739   if (const auto *ME = dyn_cast<ObjCMessageExpr>(S)) {
2740     OS << "'";
2741     ME->getSelector().print(OS);
2742     OS << "' not called";
2743   }
2744   else {
2745     OS << "No method is called";
2746   }
2747   OS << " because the receiver is nil";
2748 
2749   // The receiver was nil, and hence the method was skipped.
2750   // Register a BugReporterVisitor to issue a message telling us how
2751   // the receiver was null.
2752   bugreporter::trackExpressionValue(N, Receiver, BR,
2753                                     {bugreporter::TrackingKind::Thorough,
2754                                      /*EnableNullFPSuppression*/ false});
2755   // Issue a message saying that the method was skipped.
2756   PathDiagnosticLocation L(Receiver, BRC.getSourceManager(),
2757                                      N->getLocationContext());
2758   return std::make_shared<PathDiagnosticEventPiece>(L, OS.str());
2759 }
2760 
2761 //===----------------------------------------------------------------------===//
2762 // Visitor that tries to report interesting diagnostics from conditions.
2763 //===----------------------------------------------------------------------===//
2764 
2765 /// Return the tag associated with this visitor.  This tag will be used
2766 /// to make all PathDiagnosticPieces created by this visitor.
getTag()2767 const char *ConditionBRVisitor::getTag() { return "ConditionBRVisitor"; }
2768 
2769 PathDiagnosticPieceRef
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)2770 ConditionBRVisitor::VisitNode(const ExplodedNode *N, BugReporterContext &BRC,
2771                               PathSensitiveBugReport &BR) {
2772   auto piece = VisitNodeImpl(N, BRC, BR);
2773   if (piece) {
2774     piece->setTag(getTag());
2775     if (auto *ev = dyn_cast<PathDiagnosticEventPiece>(piece.get()))
2776       ev->setPrunable(true, /* override */ false);
2777   }
2778   return piece;
2779 }
2780 
2781 PathDiagnosticPieceRef
VisitNodeImpl(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)2782 ConditionBRVisitor::VisitNodeImpl(const ExplodedNode *N,
2783                                   BugReporterContext &BRC,
2784                                   PathSensitiveBugReport &BR) {
2785   ProgramPoint ProgPoint = N->getLocation();
2786   const std::pair<const ProgramPointTag *, const ProgramPointTag *> &Tags =
2787       ExprEngine::geteagerlyAssumeBinOpBifurcationTags();
2788 
2789   // If an assumption was made on a branch, it should be caught
2790   // here by looking at the state transition.
2791   if (std::optional<BlockEdge> BE = ProgPoint.getAs<BlockEdge>()) {
2792     const CFGBlock *SrcBlock = BE->getSrc();
2793     if (const Stmt *Term = SrcBlock->getTerminatorStmt()) {
2794       // If the tag of the previous node is 'Eagerly Assume...' the current
2795       // 'BlockEdge' has the same constraint information. We do not want to
2796       // report the value as it is just an assumption on the predecessor node
2797       // which will be caught in the next VisitNode() iteration as a 'PostStmt'.
2798       const ProgramPointTag *PreviousNodeTag =
2799           N->getFirstPred()->getLocation().getTag();
2800       if (PreviousNodeTag == Tags.first || PreviousNodeTag == Tags.second)
2801         return nullptr;
2802 
2803       return VisitTerminator(Term, N, SrcBlock, BE->getDst(), BR, BRC);
2804     }
2805     return nullptr;
2806   }
2807 
2808   if (std::optional<PostStmt> PS = ProgPoint.getAs<PostStmt>()) {
2809     const ProgramPointTag *CurrentNodeTag = PS->getTag();
2810     if (CurrentNodeTag != Tags.first && CurrentNodeTag != Tags.second)
2811       return nullptr;
2812 
2813     bool TookTrue = CurrentNodeTag == Tags.first;
2814     return VisitTrueTest(cast<Expr>(PS->getStmt()), BRC, BR, N, TookTrue);
2815   }
2816 
2817   return nullptr;
2818 }
2819 
VisitTerminator(const Stmt * Term,const ExplodedNode * N,const CFGBlock * srcBlk,const CFGBlock * dstBlk,PathSensitiveBugReport & R,BugReporterContext & BRC)2820 PathDiagnosticPieceRef ConditionBRVisitor::VisitTerminator(
2821     const Stmt *Term, const ExplodedNode *N, const CFGBlock *srcBlk,
2822     const CFGBlock *dstBlk, PathSensitiveBugReport &R,
2823     BugReporterContext &BRC) {
2824   const Expr *Cond = nullptr;
2825 
2826   // In the code below, Term is a CFG terminator and Cond is a branch condition
2827   // expression upon which the decision is made on this terminator.
2828   //
2829   // For example, in "if (x == 0)", the "if (x == 0)" statement is a terminator,
2830   // and "x == 0" is the respective condition.
2831   //
2832   // Another example: in "if (x && y)", we've got two terminators and two
2833   // conditions due to short-circuit nature of operator "&&":
2834   // 1. The "if (x && y)" statement is a terminator,
2835   //    and "y" is the respective condition.
2836   // 2. Also "x && ..." is another terminator,
2837   //    and "x" is its condition.
2838 
2839   switch (Term->getStmtClass()) {
2840   // FIXME: Stmt::SwitchStmtClass is worth handling, however it is a bit
2841   // more tricky because there are more than two branches to account for.
2842   default:
2843     return nullptr;
2844   case Stmt::IfStmtClass:
2845     Cond = cast<IfStmt>(Term)->getCond();
2846     break;
2847   case Stmt::ConditionalOperatorClass:
2848     Cond = cast<ConditionalOperator>(Term)->getCond();
2849     break;
2850   case Stmt::BinaryOperatorClass:
2851     // When we encounter a logical operator (&& or ||) as a CFG terminator,
2852     // then the condition is actually its LHS; otherwise, we'd encounter
2853     // the parent, such as if-statement, as a terminator.
2854     const auto *BO = cast<BinaryOperator>(Term);
2855     assert(BO->isLogicalOp() &&
2856            "CFG terminator is not a short-circuit operator!");
2857     Cond = BO->getLHS();
2858     break;
2859   }
2860 
2861   Cond = Cond->IgnoreParens();
2862 
2863   // However, when we encounter a logical operator as a branch condition,
2864   // then the condition is actually its RHS, because LHS would be
2865   // the condition for the logical operator terminator.
2866   while (const auto *InnerBO = dyn_cast<BinaryOperator>(Cond)) {
2867     if (!InnerBO->isLogicalOp())
2868       break;
2869     Cond = InnerBO->getRHS()->IgnoreParens();
2870   }
2871 
2872   assert(Cond);
2873   assert(srcBlk->succ_size() == 2);
2874   const bool TookTrue = *(srcBlk->succ_begin()) == dstBlk;
2875   return VisitTrueTest(Cond, BRC, R, N, TookTrue);
2876 }
2877 
2878 PathDiagnosticPieceRef
VisitTrueTest(const Expr * Cond,BugReporterContext & BRC,PathSensitiveBugReport & R,const ExplodedNode * N,bool TookTrue)2879 ConditionBRVisitor::VisitTrueTest(const Expr *Cond, BugReporterContext &BRC,
2880                                   PathSensitiveBugReport &R,
2881                                   const ExplodedNode *N, bool TookTrue) {
2882   ProgramStateRef CurrentState = N->getState();
2883   ProgramStateRef PrevState = N->getFirstPred()->getState();
2884   const LocationContext *LCtx = N->getLocationContext();
2885 
2886   // If the constraint information is changed between the current and the
2887   // previous program state we assuming the newly seen constraint information.
2888   // If we cannot evaluate the condition (and the constraints are the same)
2889   // the analyzer has no information about the value and just assuming it.
2890   // FIXME: This logic is not entirely correct, because e.g. in code like
2891   //   void f(unsigned arg) {
2892   //     if (arg >= 0) {
2893   //       // ...
2894   //     }
2895   //   }
2896   // it will say that the "arg >= 0" check is _assuming_ something new because
2897   // the constraint that "$arg >= 0" is 1 was added to the list of known
2898   // constraints. However, the unsigned value is always >= 0 so semantically
2899   // this is not a "real" assumption.
2900   bool IsAssuming =
2901       !BRC.getStateManager().haveEqualConstraints(CurrentState, PrevState) ||
2902       CurrentState->getSVal(Cond, LCtx).isUnknownOrUndef();
2903 
2904   // These will be modified in code below, but we need to preserve the original
2905   //  values in case we want to throw the generic message.
2906   const Expr *CondTmp = Cond;
2907   bool TookTrueTmp = TookTrue;
2908 
2909   while (true) {
2910     CondTmp = CondTmp->IgnoreParenCasts();
2911     switch (CondTmp->getStmtClass()) {
2912       default:
2913         break;
2914       case Stmt::BinaryOperatorClass:
2915         if (auto P = VisitTrueTest(Cond, cast<BinaryOperator>(CondTmp),
2916                                    BRC, R, N, TookTrueTmp, IsAssuming))
2917           return P;
2918         break;
2919       case Stmt::DeclRefExprClass:
2920         if (auto P = VisitTrueTest(Cond, cast<DeclRefExpr>(CondTmp),
2921                                    BRC, R, N, TookTrueTmp, IsAssuming))
2922           return P;
2923         break;
2924       case Stmt::MemberExprClass:
2925         if (auto P = VisitTrueTest(Cond, cast<MemberExpr>(CondTmp),
2926                                    BRC, R, N, TookTrueTmp, IsAssuming))
2927           return P;
2928         break;
2929       case Stmt::UnaryOperatorClass: {
2930         const auto *UO = cast<UnaryOperator>(CondTmp);
2931         if (UO->getOpcode() == UO_LNot) {
2932           TookTrueTmp = !TookTrueTmp;
2933           CondTmp = UO->getSubExpr();
2934           continue;
2935         }
2936         break;
2937       }
2938     }
2939     break;
2940   }
2941 
2942   // Condition too complex to explain? Just say something so that the user
2943   // knew we've made some path decision at this point.
2944   // If it is too complex and we know the evaluation of the condition do not
2945   // repeat the note from 'BugReporter.cpp'
2946   if (!IsAssuming)
2947     return nullptr;
2948 
2949   PathDiagnosticLocation Loc(Cond, BRC.getSourceManager(), LCtx);
2950   if (!Loc.isValid() || !Loc.asLocation().isValid())
2951     return nullptr;
2952 
2953   return std::make_shared<PathDiagnosticEventPiece>(
2954       Loc, TookTrue ? GenericTrueMessage : GenericFalseMessage);
2955 }
2956 
patternMatch(const Expr * Ex,const Expr * ParentEx,raw_ostream & Out,BugReporterContext & BRC,PathSensitiveBugReport & report,const ExplodedNode * N,std::optional<bool> & prunable,bool IsSameFieldName)2957 bool ConditionBRVisitor::patternMatch(const Expr *Ex, const Expr *ParentEx,
2958                                       raw_ostream &Out, BugReporterContext &BRC,
2959                                       PathSensitiveBugReport &report,
2960                                       const ExplodedNode *N,
2961                                       std::optional<bool> &prunable,
2962                                       bool IsSameFieldName) {
2963   const Expr *OriginalExpr = Ex;
2964   Ex = Ex->IgnoreParenCasts();
2965 
2966   if (isa<GNUNullExpr, ObjCBoolLiteralExpr, CXXBoolLiteralExpr, IntegerLiteral,
2967           FloatingLiteral>(Ex)) {
2968     // Use heuristics to determine if the expression is a macro
2969     // expanding to a literal and if so, use the macro's name.
2970     SourceLocation BeginLoc = OriginalExpr->getBeginLoc();
2971     SourceLocation EndLoc = OriginalExpr->getEndLoc();
2972     if (BeginLoc.isMacroID() && EndLoc.isMacroID()) {
2973       const SourceManager &SM = BRC.getSourceManager();
2974       const LangOptions &LO = BRC.getASTContext().getLangOpts();
2975       if (Lexer::isAtStartOfMacroExpansion(BeginLoc, SM, LO) &&
2976           Lexer::isAtEndOfMacroExpansion(EndLoc, SM, LO)) {
2977         CharSourceRange R = Lexer::getAsCharRange({BeginLoc, EndLoc}, SM, LO);
2978         Out << Lexer::getSourceText(R, SM, LO);
2979         return false;
2980       }
2981     }
2982   }
2983 
2984   if (const auto *DR = dyn_cast<DeclRefExpr>(Ex)) {
2985     const bool quotes = isa<VarDecl>(DR->getDecl());
2986     if (quotes) {
2987       Out << '\'';
2988       const LocationContext *LCtx = N->getLocationContext();
2989       const ProgramState *state = N->getState().get();
2990       if (const MemRegion *R = state->getLValue(cast<VarDecl>(DR->getDecl()),
2991                                                 LCtx).getAsRegion()) {
2992         if (report.isInteresting(R))
2993           prunable = false;
2994         else {
2995           const ProgramState *state = N->getState().get();
2996           SVal V = state->getSVal(R);
2997           if (report.isInteresting(V))
2998             prunable = false;
2999         }
3000       }
3001     }
3002     Out << DR->getDecl()->getDeclName().getAsString();
3003     if (quotes)
3004       Out << '\'';
3005     return quotes;
3006   }
3007 
3008   if (const auto *IL = dyn_cast<IntegerLiteral>(Ex)) {
3009     QualType OriginalTy = OriginalExpr->getType();
3010     if (OriginalTy->isPointerType()) {
3011       if (IL->getValue() == 0) {
3012         Out << "null";
3013         return false;
3014       }
3015     }
3016     else if (OriginalTy->isObjCObjectPointerType()) {
3017       if (IL->getValue() == 0) {
3018         Out << "nil";
3019         return false;
3020       }
3021     }
3022 
3023     Out << IL->getValue();
3024     return false;
3025   }
3026 
3027   if (const auto *ME = dyn_cast<MemberExpr>(Ex)) {
3028     if (!IsSameFieldName)
3029       Out << "field '" << ME->getMemberDecl()->getName() << '\'';
3030     else
3031       Out << '\''
3032           << Lexer::getSourceText(
3033                  CharSourceRange::getTokenRange(Ex->getSourceRange()),
3034                  BRC.getSourceManager(), BRC.getASTContext().getLangOpts(),
3035                  nullptr)
3036           << '\'';
3037   }
3038 
3039   return false;
3040 }
3041 
VisitTrueTest(const Expr * Cond,const BinaryOperator * BExpr,BugReporterContext & BRC,PathSensitiveBugReport & R,const ExplodedNode * N,bool TookTrue,bool IsAssuming)3042 PathDiagnosticPieceRef ConditionBRVisitor::VisitTrueTest(
3043     const Expr *Cond, const BinaryOperator *BExpr, BugReporterContext &BRC,
3044     PathSensitiveBugReport &R, const ExplodedNode *N, bool TookTrue,
3045     bool IsAssuming) {
3046   bool shouldInvert = false;
3047   std::optional<bool> shouldPrune;
3048 
3049   // Check if the field name of the MemberExprs is ambiguous. Example:
3050   // " 'a.d' is equal to 'h.d' " in 'test/Analysis/null-deref-path-notes.cpp'.
3051   bool IsSameFieldName = false;
3052   const auto *LhsME = dyn_cast<MemberExpr>(BExpr->getLHS()->IgnoreParenCasts());
3053   const auto *RhsME = dyn_cast<MemberExpr>(BExpr->getRHS()->IgnoreParenCasts());
3054 
3055   if (LhsME && RhsME)
3056     IsSameFieldName =
3057         LhsME->getMemberDecl()->getName() == RhsME->getMemberDecl()->getName();
3058 
3059   SmallString<128> LhsString, RhsString;
3060   {
3061     llvm::raw_svector_ostream OutLHS(LhsString), OutRHS(RhsString);
3062     const bool isVarLHS = patternMatch(BExpr->getLHS(), BExpr, OutLHS, BRC, R,
3063                                        N, shouldPrune, IsSameFieldName);
3064     const bool isVarRHS = patternMatch(BExpr->getRHS(), BExpr, OutRHS, BRC, R,
3065                                        N, shouldPrune, IsSameFieldName);
3066 
3067     shouldInvert = !isVarLHS && isVarRHS;
3068   }
3069 
3070   BinaryOperator::Opcode Op = BExpr->getOpcode();
3071 
3072   if (BinaryOperator::isAssignmentOp(Op)) {
3073     // For assignment operators, all that we care about is that the LHS
3074     // evaluates to "true" or "false".
3075     return VisitConditionVariable(LhsString, BExpr->getLHS(), BRC, R, N,
3076                                   TookTrue);
3077   }
3078 
3079   // For non-assignment operations, we require that we can understand
3080   // both the LHS and RHS.
3081   if (LhsString.empty() || RhsString.empty() ||
3082       !BinaryOperator::isComparisonOp(Op) || Op == BO_Cmp)
3083     return nullptr;
3084 
3085   // Should we invert the strings if the LHS is not a variable name?
3086   SmallString<256> buf;
3087   llvm::raw_svector_ostream Out(buf);
3088   Out << (IsAssuming ? "Assuming " : "")
3089       << (shouldInvert ? RhsString : LhsString) << " is ";
3090 
3091   // Do we need to invert the opcode?
3092   if (shouldInvert)
3093     switch (Op) {
3094       default: break;
3095       case BO_LT: Op = BO_GT; break;
3096       case BO_GT: Op = BO_LT; break;
3097       case BO_LE: Op = BO_GE; break;
3098       case BO_GE: Op = BO_LE; break;
3099     }
3100 
3101   if (!TookTrue)
3102     switch (Op) {
3103       case BO_EQ: Op = BO_NE; break;
3104       case BO_NE: Op = BO_EQ; break;
3105       case BO_LT: Op = BO_GE; break;
3106       case BO_GT: Op = BO_LE; break;
3107       case BO_LE: Op = BO_GT; break;
3108       case BO_GE: Op = BO_LT; break;
3109       default:
3110         return nullptr;
3111     }
3112 
3113   switch (Op) {
3114     case BO_EQ:
3115       Out << "equal to ";
3116       break;
3117     case BO_NE:
3118       Out << "not equal to ";
3119       break;
3120     default:
3121       Out << BinaryOperator::getOpcodeStr(Op) << ' ';
3122       break;
3123   }
3124 
3125   Out << (shouldInvert ? LhsString : RhsString);
3126   const LocationContext *LCtx = N->getLocationContext();
3127   const SourceManager &SM = BRC.getSourceManager();
3128 
3129   if (isVarAnInterestingCondition(BExpr->getLHS(), N, &R) ||
3130       isVarAnInterestingCondition(BExpr->getRHS(), N, &R))
3131     Out << WillBeUsedForACondition;
3132 
3133   // Convert 'field ...' to 'Field ...' if it is a MemberExpr.
3134   std::string Message = std::string(Out.str());
3135   Message[0] = toupper(Message[0]);
3136 
3137   // If we know the value create a pop-up note to the value part of 'BExpr'.
3138   if (!IsAssuming) {
3139     PathDiagnosticLocation Loc;
3140     if (!shouldInvert) {
3141       if (LhsME && LhsME->getMemberLoc().isValid())
3142         Loc = PathDiagnosticLocation(LhsME->getMemberLoc(), SM);
3143       else
3144         Loc = PathDiagnosticLocation(BExpr->getLHS(), SM, LCtx);
3145     } else {
3146       if (RhsME && RhsME->getMemberLoc().isValid())
3147         Loc = PathDiagnosticLocation(RhsME->getMemberLoc(), SM);
3148       else
3149         Loc = PathDiagnosticLocation(BExpr->getRHS(), SM, LCtx);
3150     }
3151 
3152     return std::make_shared<PathDiagnosticPopUpPiece>(Loc, Message);
3153   }
3154 
3155   PathDiagnosticLocation Loc(Cond, SM, LCtx);
3156   auto event = std::make_shared<PathDiagnosticEventPiece>(Loc, Message);
3157   if (shouldPrune)
3158     event->setPrunable(*shouldPrune);
3159   return event;
3160 }
3161 
VisitConditionVariable(StringRef LhsString,const Expr * CondVarExpr,BugReporterContext & BRC,PathSensitiveBugReport & report,const ExplodedNode * N,bool TookTrue)3162 PathDiagnosticPieceRef ConditionBRVisitor::VisitConditionVariable(
3163     StringRef LhsString, const Expr *CondVarExpr, BugReporterContext &BRC,
3164     PathSensitiveBugReport &report, const ExplodedNode *N, bool TookTrue) {
3165   // FIXME: If there's already a constraint tracker for this variable,
3166   // we shouldn't emit anything here (c.f. the double note in
3167   // test/Analysis/inlining/path-notes.c)
3168   SmallString<256> buf;
3169   llvm::raw_svector_ostream Out(buf);
3170   Out << "Assuming " << LhsString << " is ";
3171 
3172   if (!printValue(CondVarExpr, Out, N, TookTrue, /*IsAssuming=*/true))
3173     return nullptr;
3174 
3175   const LocationContext *LCtx = N->getLocationContext();
3176   PathDiagnosticLocation Loc(CondVarExpr, BRC.getSourceManager(), LCtx);
3177 
3178   if (isVarAnInterestingCondition(CondVarExpr, N, &report))
3179     Out << WillBeUsedForACondition;
3180 
3181   auto event = std::make_shared<PathDiagnosticEventPiece>(Loc, Out.str());
3182 
3183   if (isInterestingExpr(CondVarExpr, N, &report))
3184     event->setPrunable(false);
3185 
3186   return event;
3187 }
3188 
VisitTrueTest(const Expr * Cond,const DeclRefExpr * DRE,BugReporterContext & BRC,PathSensitiveBugReport & report,const ExplodedNode * N,bool TookTrue,bool IsAssuming)3189 PathDiagnosticPieceRef ConditionBRVisitor::VisitTrueTest(
3190     const Expr *Cond, const DeclRefExpr *DRE, BugReporterContext &BRC,
3191     PathSensitiveBugReport &report, const ExplodedNode *N, bool TookTrue,
3192     bool IsAssuming) {
3193   const auto *VD = dyn_cast<VarDecl>(DRE->getDecl());
3194   if (!VD)
3195     return nullptr;
3196 
3197   SmallString<256> Buf;
3198   llvm::raw_svector_ostream Out(Buf);
3199 
3200   Out << (IsAssuming ? "Assuming '" : "'") << VD->getDeclName() << "' is ";
3201 
3202   if (!printValue(DRE, Out, N, TookTrue, IsAssuming))
3203     return nullptr;
3204 
3205   const LocationContext *LCtx = N->getLocationContext();
3206 
3207   if (isVarAnInterestingCondition(DRE, N, &report))
3208     Out << WillBeUsedForACondition;
3209 
3210   // If we know the value create a pop-up note to the 'DRE'.
3211   if (!IsAssuming) {
3212     PathDiagnosticLocation Loc(DRE, BRC.getSourceManager(), LCtx);
3213     return std::make_shared<PathDiagnosticPopUpPiece>(Loc, Out.str());
3214   }
3215 
3216   PathDiagnosticLocation Loc(Cond, BRC.getSourceManager(), LCtx);
3217   auto event = std::make_shared<PathDiagnosticEventPiece>(Loc, Out.str());
3218 
3219   if (isInterestingExpr(DRE, N, &report))
3220     event->setPrunable(false);
3221 
3222   return std::move(event);
3223 }
3224 
VisitTrueTest(const Expr * Cond,const MemberExpr * ME,BugReporterContext & BRC,PathSensitiveBugReport & report,const ExplodedNode * N,bool TookTrue,bool IsAssuming)3225 PathDiagnosticPieceRef ConditionBRVisitor::VisitTrueTest(
3226     const Expr *Cond, const MemberExpr *ME, BugReporterContext &BRC,
3227     PathSensitiveBugReport &report, const ExplodedNode *N, bool TookTrue,
3228     bool IsAssuming) {
3229   SmallString<256> Buf;
3230   llvm::raw_svector_ostream Out(Buf);
3231 
3232   Out << (IsAssuming ? "Assuming field '" : "Field '")
3233       << ME->getMemberDecl()->getName() << "' is ";
3234 
3235   if (!printValue(ME, Out, N, TookTrue, IsAssuming))
3236     return nullptr;
3237 
3238   const LocationContext *LCtx = N->getLocationContext();
3239   PathDiagnosticLocation Loc;
3240 
3241   // If we know the value create a pop-up note to the member of the MemberExpr.
3242   if (!IsAssuming && ME->getMemberLoc().isValid())
3243     Loc = PathDiagnosticLocation(ME->getMemberLoc(), BRC.getSourceManager());
3244   else
3245     Loc = PathDiagnosticLocation(Cond, BRC.getSourceManager(), LCtx);
3246 
3247   if (!Loc.isValid() || !Loc.asLocation().isValid())
3248     return nullptr;
3249 
3250   if (isVarAnInterestingCondition(ME, N, &report))
3251     Out << WillBeUsedForACondition;
3252 
3253   // If we know the value create a pop-up note.
3254   if (!IsAssuming)
3255     return std::make_shared<PathDiagnosticPopUpPiece>(Loc, Out.str());
3256 
3257   auto event = std::make_shared<PathDiagnosticEventPiece>(Loc, Out.str());
3258   if (isInterestingExpr(ME, N, &report))
3259     event->setPrunable(false);
3260   return event;
3261 }
3262 
printValue(const Expr * CondVarExpr,raw_ostream & Out,const ExplodedNode * N,bool TookTrue,bool IsAssuming)3263 bool ConditionBRVisitor::printValue(const Expr *CondVarExpr, raw_ostream &Out,
3264                                     const ExplodedNode *N, bool TookTrue,
3265                                     bool IsAssuming) {
3266   QualType Ty = CondVarExpr->getType();
3267 
3268   if (Ty->isPointerType()) {
3269     Out << (TookTrue ? "non-null" : "null");
3270     return true;
3271   }
3272 
3273   if (Ty->isObjCObjectPointerType()) {
3274     Out << (TookTrue ? "non-nil" : "nil");
3275     return true;
3276   }
3277 
3278   if (!Ty->isIntegralOrEnumerationType())
3279     return false;
3280 
3281   std::optional<const llvm::APSInt *> IntValue;
3282   if (!IsAssuming)
3283     IntValue = getConcreteIntegerValue(CondVarExpr, N);
3284 
3285   if (IsAssuming || !IntValue) {
3286     if (Ty->isBooleanType())
3287       Out << (TookTrue ? "true" : "false");
3288     else
3289       Out << (TookTrue ? "not equal to 0" : "0");
3290   } else {
3291     if (Ty->isBooleanType())
3292       Out << ((*IntValue)->getBoolValue() ? "true" : "false");
3293     else
3294       Out << **IntValue;
3295   }
3296 
3297   return true;
3298 }
3299 
3300 constexpr llvm::StringLiteral ConditionBRVisitor::GenericTrueMessage;
3301 constexpr llvm::StringLiteral ConditionBRVisitor::GenericFalseMessage;
3302 
isPieceMessageGeneric(const PathDiagnosticPiece * Piece)3303 bool ConditionBRVisitor::isPieceMessageGeneric(
3304     const PathDiagnosticPiece *Piece) {
3305   return Piece->getString() == GenericTrueMessage ||
3306          Piece->getString() == GenericFalseMessage;
3307 }
3308 
3309 //===----------------------------------------------------------------------===//
3310 // Implementation of LikelyFalsePositiveSuppressionBRVisitor.
3311 //===----------------------------------------------------------------------===//
3312 
finalizeVisitor(BugReporterContext & BRC,const ExplodedNode * N,PathSensitiveBugReport & BR)3313 void LikelyFalsePositiveSuppressionBRVisitor::finalizeVisitor(
3314     BugReporterContext &BRC, const ExplodedNode *N,
3315     PathSensitiveBugReport &BR) {
3316   // Here we suppress false positives coming from system headers. This list is
3317   // based on known issues.
3318   const AnalyzerOptions &Options = BRC.getAnalyzerOptions();
3319   const Decl *D = N->getLocationContext()->getDecl();
3320 
3321   if (AnalysisDeclContext::isInStdNamespace(D)) {
3322     // Skip reports within the 'std' namespace. Although these can sometimes be
3323     // the user's fault, we currently don't report them very well, and
3324     // Note that this will not help for any other data structure libraries, like
3325     // TR1, Boost, or llvm/ADT.
3326     if (Options.ShouldSuppressFromCXXStandardLibrary) {
3327       BR.markInvalid(getTag(), nullptr);
3328       return;
3329     } else {
3330       // If the complete 'std' suppression is not enabled, suppress reports
3331       // from the 'std' namespace that are known to produce false positives.
3332 
3333       // The analyzer issues a false use-after-free when std::list::pop_front
3334       // or std::list::pop_back are called multiple times because we cannot
3335       // reason about the internal invariants of the data structure.
3336       if (const auto *MD = dyn_cast<CXXMethodDecl>(D)) {
3337         const CXXRecordDecl *CD = MD->getParent();
3338         if (CD->getName() == "list") {
3339           BR.markInvalid(getTag(), nullptr);
3340           return;
3341         }
3342       }
3343 
3344       // The analyzer issues a false positive when the constructor of
3345       // std::__independent_bits_engine from algorithms is used.
3346       if (const auto *MD = dyn_cast<CXXConstructorDecl>(D)) {
3347         const CXXRecordDecl *CD = MD->getParent();
3348         if (CD->getName() == "__independent_bits_engine") {
3349           BR.markInvalid(getTag(), nullptr);
3350           return;
3351         }
3352       }
3353 
3354       for (const LocationContext *LCtx = N->getLocationContext(); LCtx;
3355            LCtx = LCtx->getParent()) {
3356         const auto *MD = dyn_cast<CXXMethodDecl>(LCtx->getDecl());
3357         if (!MD)
3358           continue;
3359 
3360         const CXXRecordDecl *CD = MD->getParent();
3361         // The analyzer issues a false positive on
3362         //   std::basic_string<uint8_t> v; v.push_back(1);
3363         // and
3364         //   std::u16string s; s += u'a';
3365         // because we cannot reason about the internal invariants of the
3366         // data structure.
3367         if (CD->getName() == "basic_string") {
3368           BR.markInvalid(getTag(), nullptr);
3369           return;
3370         }
3371 
3372         // The analyzer issues a false positive on
3373         //    std::shared_ptr<int> p(new int(1)); p = nullptr;
3374         // because it does not reason properly about temporary destructors.
3375         if (CD->getName() == "shared_ptr") {
3376           BR.markInvalid(getTag(), nullptr);
3377           return;
3378         }
3379       }
3380     }
3381   }
3382 
3383   // Skip reports within the sys/queue.h macros as we do not have the ability to
3384   // reason about data structure shapes.
3385   const SourceManager &SM = BRC.getSourceManager();
3386   FullSourceLoc Loc = BR.getLocation().asLocation();
3387   while (Loc.isMacroID()) {
3388     Loc = Loc.getSpellingLoc();
3389     if (SM.getFilename(Loc).ends_with("sys/queue.h")) {
3390       BR.markInvalid(getTag(), nullptr);
3391       return;
3392     }
3393   }
3394 }
3395 
3396 //===----------------------------------------------------------------------===//
3397 // Implementation of UndefOrNullArgVisitor.
3398 //===----------------------------------------------------------------------===//
3399 
3400 PathDiagnosticPieceRef
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)3401 UndefOrNullArgVisitor::VisitNode(const ExplodedNode *N, BugReporterContext &BRC,
3402                                  PathSensitiveBugReport &BR) {
3403   ProgramStateRef State = N->getState();
3404   ProgramPoint ProgLoc = N->getLocation();
3405 
3406   // We are only interested in visiting CallEnter nodes.
3407   std::optional<CallEnter> CEnter = ProgLoc.getAs<CallEnter>();
3408   if (!CEnter)
3409     return nullptr;
3410 
3411   // Check if one of the arguments is the region the visitor is tracking.
3412   CallEventManager &CEMgr = BRC.getStateManager().getCallEventManager();
3413   CallEventRef<> Call = CEMgr.getCaller(CEnter->getCalleeContext(), State);
3414   unsigned Idx = 0;
3415   ArrayRef<ParmVarDecl *> parms = Call->parameters();
3416 
3417   for (const auto ParamDecl : parms) {
3418     const MemRegion *ArgReg = Call->getArgSVal(Idx).getAsRegion();
3419     ++Idx;
3420 
3421     // Are we tracking the argument or its subregion?
3422     if ( !ArgReg || !R->isSubRegionOf(ArgReg->StripCasts()))
3423       continue;
3424 
3425     // Check the function parameter type.
3426     assert(ParamDecl && "Formal parameter has no decl?");
3427     QualType T = ParamDecl->getType();
3428 
3429     if (!(T->isAnyPointerType() || T->isReferenceType())) {
3430       // Function can only change the value passed in by address.
3431       continue;
3432     }
3433 
3434     // If it is a const pointer value, the function does not intend to
3435     // change the value.
3436     if (T->getPointeeType().isConstQualified())
3437       continue;
3438 
3439     // Mark the call site (LocationContext) as interesting if the value of the
3440     // argument is undefined or '0'/'NULL'.
3441     SVal BoundVal = State->getSVal(R);
3442     if (BoundVal.isUndef() || BoundVal.isZeroConstant()) {
3443       BR.markInteresting(CEnter->getCalleeContext());
3444       return nullptr;
3445     }
3446   }
3447   return nullptr;
3448 }
3449 
3450 //===----------------------------------------------------------------------===//
3451 // Implementation of TagVisitor.
3452 //===----------------------------------------------------------------------===//
3453 
3454 int NoteTag::Kind = 0;
3455 
Profile(llvm::FoldingSetNodeID & ID) const3456 void TagVisitor::Profile(llvm::FoldingSetNodeID &ID) const {
3457   static int Tag = 0;
3458   ID.AddPointer(&Tag);
3459 }
3460 
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & R)3461 PathDiagnosticPieceRef TagVisitor::VisitNode(const ExplodedNode *N,
3462                                              BugReporterContext &BRC,
3463                                              PathSensitiveBugReport &R) {
3464   ProgramPoint PP = N->getLocation();
3465   const NoteTag *T = dyn_cast_or_null<NoteTag>(PP.getTag());
3466   if (!T)
3467     return nullptr;
3468 
3469   if (std::optional<std::string> Msg = T->generateMessage(BRC, R)) {
3470     PathDiagnosticLocation Loc =
3471         PathDiagnosticLocation::create(PP, BRC.getSourceManager());
3472     auto Piece = std::make_shared<PathDiagnosticEventPiece>(Loc, *Msg);
3473     Piece->setPrunable(T->isPrunable());
3474     return Piece;
3475   }
3476 
3477   return nullptr;
3478 }
3479