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 <string>
63 #include <utility>
64
65 using namespace clang;
66 using namespace ento;
67 using namespace bugreporter;
68
69 //===----------------------------------------------------------------------===//
70 // Utility functions.
71 //===----------------------------------------------------------------------===//
72
peelOffPointerArithmetic(const BinaryOperator * B)73 static const Expr *peelOffPointerArithmetic(const BinaryOperator *B) {
74 if (B->isAdditiveOp() && B->getType()->isPointerType()) {
75 if (B->getLHS()->getType()->isPointerType()) {
76 return B->getLHS();
77 } else if (B->getRHS()->getType()->isPointerType()) {
78 return B->getRHS();
79 }
80 }
81 return nullptr;
82 }
83
84 /// \return A subexpression of @c Ex which represents the
85 /// expression-of-interest.
86 static const Expr *peelOffOuterExpr(const Expr *Ex, const ExplodedNode *N);
87
88 /// Given that expression S represents a pointer that would be dereferenced,
89 /// try to find a sub-expression from which the pointer came from.
90 /// This is used for tracking down origins of a null or undefined value:
91 /// "this is null because that is null because that is null" etc.
92 /// We wipe away field and element offsets because they merely add offsets.
93 /// We also wipe away all casts except lvalue-to-rvalue casts, because the
94 /// latter represent an actual pointer dereference; however, we remove
95 /// the final lvalue-to-rvalue cast before returning from this function
96 /// because it demonstrates more clearly from where the pointer rvalue was
97 /// loaded. Examples:
98 /// x->y.z ==> x (lvalue)
99 /// foo()->y.z ==> foo() (rvalue)
getDerefExpr(const Stmt * S)100 const Expr *bugreporter::getDerefExpr(const Stmt *S) {
101 const auto *E = dyn_cast<Expr>(S);
102 if (!E)
103 return nullptr;
104
105 while (true) {
106 if (const auto *CE = dyn_cast<CastExpr>(E)) {
107 if (CE->getCastKind() == CK_LValueToRValue) {
108 // This cast represents the load we're looking for.
109 break;
110 }
111 E = CE->getSubExpr();
112 } else if (const auto *B = dyn_cast<BinaryOperator>(E)) {
113 // Pointer arithmetic: '*(x + 2)' -> 'x') etc.
114 if (const Expr *Inner = peelOffPointerArithmetic(B)) {
115 E = Inner;
116 } else {
117 // Probably more arithmetic can be pattern-matched here,
118 // but for now give up.
119 break;
120 }
121 } else if (const auto *U = dyn_cast<UnaryOperator>(E)) {
122 if (U->getOpcode() == UO_Deref || U->getOpcode() == UO_AddrOf ||
123 (U->isIncrementDecrementOp() && U->getType()->isPointerType())) {
124 // Operators '*' and '&' don't actually mean anything.
125 // We look at casts instead.
126 E = U->getSubExpr();
127 } else {
128 // Probably more arithmetic can be pattern-matched here,
129 // but for now give up.
130 break;
131 }
132 }
133 // Pattern match for a few useful cases: a[0], p->f, *p etc.
134 else if (const auto *ME = dyn_cast<MemberExpr>(E)) {
135 E = ME->getBase();
136 } else if (const auto *IvarRef = dyn_cast<ObjCIvarRefExpr>(E)) {
137 E = IvarRef->getBase();
138 } else if (const auto *AE = dyn_cast<ArraySubscriptExpr>(E)) {
139 E = AE->getBase();
140 } else if (const auto *PE = dyn_cast<ParenExpr>(E)) {
141 E = PE->getSubExpr();
142 } else if (const auto *FE = dyn_cast<FullExpr>(E)) {
143 E = FE->getSubExpr();
144 } else {
145 // Other arbitrary stuff.
146 break;
147 }
148 }
149
150 // Special case: remove the final lvalue-to-rvalue cast, but do not recurse
151 // deeper into the sub-expression. This way we return the lvalue from which
152 // our pointer rvalue was loaded.
153 if (const auto *CE = dyn_cast<ImplicitCastExpr>(E))
154 if (CE->getCastKind() == CK_LValueToRValue)
155 E = CE->getSubExpr();
156
157 return E;
158 }
159
160 static const MemRegion *
getLocationRegionIfReference(const Expr * E,const ExplodedNode * N,bool LookingForReference=true)161 getLocationRegionIfReference(const Expr *E, const ExplodedNode *N,
162 bool LookingForReference = true) {
163 if (const auto *DR = dyn_cast<DeclRefExpr>(E)) {
164 if (const auto *VD = dyn_cast<VarDecl>(DR->getDecl())) {
165 if (LookingForReference && !VD->getType()->isReferenceType())
166 return nullptr;
167 return N->getState()
168 ->getLValue(VD, N->getLocationContext())
169 .getAsRegion();
170 }
171 }
172
173 // FIXME: This does not handle other kinds of null references,
174 // for example, references from FieldRegions:
175 // struct Wrapper { int &ref; };
176 // Wrapper w = { *(int *)0 };
177 // w.ref = 1;
178
179 return nullptr;
180 }
181
182 /// Comparing internal representations of symbolic values (via
183 /// SVal::operator==()) is a valid way to check if the value was updated,
184 /// unless it's a LazyCompoundVal that may have a different internal
185 /// representation every time it is loaded from the state. In this function we
186 /// do an approximate comparison for lazy compound values, checking that they
187 /// are the immediate snapshots of the tracked region's bindings within the
188 /// node's respective states but not really checking that these snapshots
189 /// actually contain the same set of bindings.
hasVisibleUpdate(const ExplodedNode * LeftNode,SVal LeftVal,const ExplodedNode * RightNode,SVal RightVal)190 static bool hasVisibleUpdate(const ExplodedNode *LeftNode, SVal LeftVal,
191 const ExplodedNode *RightNode, SVal RightVal) {
192 if (LeftVal == RightVal)
193 return true;
194
195 const auto LLCV = LeftVal.getAs<nonloc::LazyCompoundVal>();
196 if (!LLCV)
197 return false;
198
199 const auto RLCV = RightVal.getAs<nonloc::LazyCompoundVal>();
200 if (!RLCV)
201 return false;
202
203 return LLCV->getRegion() == RLCV->getRegion() &&
204 LLCV->getStore() == LeftNode->getState()->getStore() &&
205 RLCV->getStore() == RightNode->getState()->getStore();
206 }
207
getSValForVar(const Expr * CondVarExpr,const ExplodedNode * N)208 static std::optional<SVal> getSValForVar(const Expr *CondVarExpr,
209 const ExplodedNode *N) {
210 ProgramStateRef State = N->getState();
211 const LocationContext *LCtx = N->getLocationContext();
212
213 assert(CondVarExpr);
214 CondVarExpr = CondVarExpr->IgnoreImpCasts();
215
216 // The declaration of the value may rely on a pointer so take its l-value.
217 // FIXME: As seen in VisitCommonDeclRefExpr, sometimes DeclRefExpr may
218 // evaluate to a FieldRegion when it refers to a declaration of a lambda
219 // capture variable. We most likely need to duplicate that logic here.
220 if (const auto *DRE = dyn_cast<DeclRefExpr>(CondVarExpr))
221 if (const auto *VD = dyn_cast<VarDecl>(DRE->getDecl()))
222 return State->getSVal(State->getLValue(VD, LCtx));
223
224 if (const auto *ME = dyn_cast<MemberExpr>(CondVarExpr))
225 if (const auto *FD = dyn_cast<FieldDecl>(ME->getMemberDecl()))
226 if (auto FieldL = State->getSVal(ME, LCtx).getAs<Loc>())
227 return State->getRawSVal(*FieldL, FD->getType());
228
229 return std::nullopt;
230 }
231
232 static std::optional<const llvm::APSInt *>
getConcreteIntegerValue(const Expr * CondVarExpr,const ExplodedNode * N)233 getConcreteIntegerValue(const Expr *CondVarExpr, const ExplodedNode *N) {
234
235 if (std::optional<SVal> V = getSValForVar(CondVarExpr, N))
236 if (auto CI = V->getAs<nonloc::ConcreteInt>())
237 return &CI->getValue();
238 return std::nullopt;
239 }
240
isVarAnInterestingCondition(const Expr * CondVarExpr,const ExplodedNode * N,const PathSensitiveBugReport * B)241 static bool isVarAnInterestingCondition(const Expr *CondVarExpr,
242 const ExplodedNode *N,
243 const PathSensitiveBugReport *B) {
244 // Even if this condition is marked as interesting, it isn't *that*
245 // interesting if it didn't happen in a nested stackframe, the user could just
246 // follow the arrows.
247 if (!B->getErrorNode()->getStackFrame()->isParentOf(N->getStackFrame()))
248 return false;
249
250 if (std::optional<SVal> V = getSValForVar(CondVarExpr, N))
251 if (std::optional<bugreporter::TrackingKind> K =
252 B->getInterestingnessKind(*V))
253 return *K == bugreporter::TrackingKind::Condition;
254
255 return false;
256 }
257
isInterestingExpr(const Expr * E,const ExplodedNode * N,const PathSensitiveBugReport * B)258 static bool isInterestingExpr(const Expr *E, const ExplodedNode *N,
259 const PathSensitiveBugReport *B) {
260 if (std::optional<SVal> V = getSValForVar(E, N))
261 return B->getInterestingnessKind(*V).has_value();
262 return false;
263 }
264
265 /// \return name of the macro inside the location \p Loc.
getMacroName(SourceLocation Loc,BugReporterContext & BRC)266 static StringRef getMacroName(SourceLocation Loc,
267 BugReporterContext &BRC) {
268 return Lexer::getImmediateMacroName(
269 Loc,
270 BRC.getSourceManager(),
271 BRC.getASTContext().getLangOpts());
272 }
273
274 /// \return Whether given spelling location corresponds to an expansion
275 /// of a function-like macro.
isFunctionMacroExpansion(SourceLocation Loc,const SourceManager & SM)276 static bool isFunctionMacroExpansion(SourceLocation Loc,
277 const SourceManager &SM) {
278 if (!Loc.isMacroID())
279 return false;
280 while (SM.isMacroArgExpansion(Loc))
281 Loc = SM.getImmediateExpansionRange(Loc).getBegin();
282 std::pair<FileID, unsigned> TLInfo = SM.getDecomposedLoc(Loc);
283 SrcMgr::SLocEntry SE = SM.getSLocEntry(TLInfo.first);
284 const SrcMgr::ExpansionInfo &EInfo = SE.getExpansion();
285 return EInfo.isFunctionMacroExpansion();
286 }
287
288 /// \return Whether \c RegionOfInterest was modified at \p N,
289 /// where \p ValueAfter is \c RegionOfInterest's value at the end of the
290 /// stack frame.
wasRegionOfInterestModifiedAt(const SubRegion * RegionOfInterest,const ExplodedNode * N,SVal ValueAfter)291 static bool wasRegionOfInterestModifiedAt(const SubRegion *RegionOfInterest,
292 const ExplodedNode *N,
293 SVal ValueAfter) {
294 ProgramStateRef State = N->getState();
295 ProgramStateManager &Mgr = N->getState()->getStateManager();
296
297 if (!N->getLocationAs<PostStore>() && !N->getLocationAs<PostInitializer>() &&
298 !N->getLocationAs<PostStmt>())
299 return false;
300
301 // Writing into region of interest.
302 if (auto PS = N->getLocationAs<PostStmt>())
303 if (auto *BO = PS->getStmtAs<BinaryOperator>())
304 if (BO->isAssignmentOp() && RegionOfInterest->isSubRegionOf(
305 N->getSVal(BO->getLHS()).getAsRegion()))
306 return true;
307
308 // SVal after the state is possibly different.
309 SVal ValueAtN = N->getState()->getSVal(RegionOfInterest);
310 if (!Mgr.getSValBuilder()
311 .areEqual(State, ValueAtN, ValueAfter)
312 .isConstrainedTrue() &&
313 (!ValueAtN.isUndef() || !ValueAfter.isUndef()))
314 return true;
315
316 return false;
317 }
318
319 //===----------------------------------------------------------------------===//
320 // Implementation of BugReporterVisitor.
321 //===----------------------------------------------------------------------===//
322
getEndPath(BugReporterContext &,const ExplodedNode *,PathSensitiveBugReport &)323 PathDiagnosticPieceRef BugReporterVisitor::getEndPath(BugReporterContext &,
324 const ExplodedNode *,
325 PathSensitiveBugReport &) {
326 return nullptr;
327 }
328
finalizeVisitor(BugReporterContext &,const ExplodedNode *,PathSensitiveBugReport &)329 void BugReporterVisitor::finalizeVisitor(BugReporterContext &,
330 const ExplodedNode *,
331 PathSensitiveBugReport &) {}
332
333 PathDiagnosticPieceRef
getDefaultEndPath(const BugReporterContext & BRC,const ExplodedNode * EndPathNode,const PathSensitiveBugReport & BR)334 BugReporterVisitor::getDefaultEndPath(const BugReporterContext &BRC,
335 const ExplodedNode *EndPathNode,
336 const PathSensitiveBugReport &BR) {
337 PathDiagnosticLocation L = BR.getLocation();
338 const auto &Ranges = BR.getRanges();
339
340 // Only add the statement itself as a range if we didn't specify any
341 // special ranges for this report.
342 auto P = std::make_shared<PathDiagnosticEventPiece>(
343 L, BR.getDescription(), Ranges.begin() == Ranges.end());
344 for (SourceRange Range : Ranges)
345 P->addRange(Range);
346
347 return P;
348 }
349
350 //===----------------------------------------------------------------------===//
351 // Implementation of NoStateChangeFuncVisitor.
352 //===----------------------------------------------------------------------===//
353
isModifiedInFrame(const ExplodedNode * N)354 bool NoStateChangeFuncVisitor::isModifiedInFrame(const ExplodedNode *N) {
355 const LocationContext *Ctx = N->getLocationContext();
356 const StackFrameContext *SCtx = Ctx->getStackFrame();
357 if (!FramesModifyingCalculated.count(SCtx))
358 findModifyingFrames(N);
359 return FramesModifying.count(SCtx);
360 }
361
markFrameAsModifying(const StackFrameContext * SCtx)362 void NoStateChangeFuncVisitor::markFrameAsModifying(
363 const StackFrameContext *SCtx) {
364 while (!SCtx->inTopFrame()) {
365 auto p = FramesModifying.insert(SCtx);
366 if (!p.second)
367 break; // Frame and all its parents already inserted.
368
369 SCtx = SCtx->getParent()->getStackFrame();
370 }
371 }
372
getMatchingCallExitEnd(const ExplodedNode * N)373 static const ExplodedNode *getMatchingCallExitEnd(const ExplodedNode *N) {
374 assert(N->getLocationAs<CallEnter>());
375 // The stackframe of the callee is only found in the nodes succeeding
376 // the CallEnter node. CallEnter's stack frame refers to the caller.
377 const StackFrameContext *OrigSCtx = N->getFirstSucc()->getStackFrame();
378
379 // Similarly, the nodes preceding CallExitEnd refer to the callee's stack
380 // frame.
381 auto IsMatchingCallExitEnd = [OrigSCtx](const ExplodedNode *N) {
382 return N->getLocationAs<CallExitEnd>() &&
383 OrigSCtx == N->getFirstPred()->getStackFrame();
384 };
385 while (N && !IsMatchingCallExitEnd(N)) {
386 assert(N->succ_size() <= 1 &&
387 "This function is to be used on the trimmed ExplodedGraph!");
388 N = N->getFirstSucc();
389 }
390 return N;
391 }
392
findModifyingFrames(const ExplodedNode * const CallExitBeginN)393 void NoStateChangeFuncVisitor::findModifyingFrames(
394 const ExplodedNode *const CallExitBeginN) {
395
396 assert(CallExitBeginN->getLocationAs<CallExitBegin>());
397
398 const StackFrameContext *const OriginalSCtx =
399 CallExitBeginN->getLocationContext()->getStackFrame();
400
401 const ExplodedNode *CurrCallExitBeginN = CallExitBeginN;
402 const StackFrameContext *CurrentSCtx = OriginalSCtx;
403
404 for (const ExplodedNode *CurrN = CallExitBeginN; CurrN;
405 CurrN = CurrN->getFirstPred()) {
406 // Found a new inlined call.
407 if (CurrN->getLocationAs<CallExitBegin>()) {
408 CurrCallExitBeginN = CurrN;
409 CurrentSCtx = CurrN->getStackFrame();
410 FramesModifyingCalculated.insert(CurrentSCtx);
411 // We won't see a change in between two identical exploded nodes: skip.
412 continue;
413 }
414
415 if (auto CE = CurrN->getLocationAs<CallEnter>()) {
416 if (const ExplodedNode *CallExitEndN = getMatchingCallExitEnd(CurrN))
417 if (wasModifiedInFunction(CurrN, CallExitEndN))
418 markFrameAsModifying(CurrentSCtx);
419
420 // We exited this inlined call, lets actualize the stack frame.
421 CurrentSCtx = CurrN->getStackFrame();
422
423 // Stop calculating at the current function, but always regard it as
424 // modifying, so we can avoid notes like this:
425 // void f(Foo &F) {
426 // F.field = 0; // note: 0 assigned to 'F.field'
427 // // note: returning without writing to 'F.field'
428 // }
429 if (CE->getCalleeContext() == OriginalSCtx) {
430 markFrameAsModifying(CurrentSCtx);
431 break;
432 }
433 }
434
435 if (wasModifiedBeforeCallExit(CurrN, CurrCallExitBeginN))
436 markFrameAsModifying(CurrentSCtx);
437 }
438 }
439
VisitNode(const ExplodedNode * N,BugReporterContext & BR,PathSensitiveBugReport & R)440 PathDiagnosticPieceRef NoStateChangeFuncVisitor::VisitNode(
441 const ExplodedNode *N, BugReporterContext &BR, PathSensitiveBugReport &R) {
442
443 const LocationContext *Ctx = N->getLocationContext();
444 const StackFrameContext *SCtx = Ctx->getStackFrame();
445 ProgramStateRef State = N->getState();
446 auto CallExitLoc = N->getLocationAs<CallExitBegin>();
447
448 // No diagnostic if region was modified inside the frame.
449 if (!CallExitLoc || isModifiedInFrame(N))
450 return nullptr;
451
452 CallEventRef<> Call =
453 BR.getStateManager().getCallEventManager().getCaller(SCtx, State);
454
455 // Optimistically suppress uninitialized value bugs that result
456 // from system headers having a chance to initialize the value
457 // but failing to do so. It's too unlikely a system header's fault.
458 // It's much more likely a situation in which the function has a failure
459 // mode that the user decided not to check. If we want to hunt such
460 // omitted checks, we should provide an explicit function-specific note
461 // describing the precondition under which the function isn't supposed to
462 // initialize its out-parameter, and additionally check that such
463 // precondition can actually be fulfilled on the current path.
464 if (Call->isInSystemHeader()) {
465 // We make an exception for system header functions that have no branches.
466 // Such functions unconditionally fail to initialize the variable.
467 // If they call other functions that have more paths within them,
468 // this suppression would still apply when we visit these inner functions.
469 // One common example of a standard function that doesn't ever initialize
470 // its out parameter is operator placement new; it's up to the follow-up
471 // constructor (if any) to initialize the memory.
472 if (!N->getStackFrame()->getCFG()->isLinear()) {
473 static int i = 0;
474 R.markInvalid(&i, nullptr);
475 }
476 return nullptr;
477 }
478
479 if (const auto *MC = dyn_cast<ObjCMethodCall>(Call)) {
480 // If we failed to construct a piece for self, we still want to check
481 // whether the entity of interest is in a parameter.
482 if (PathDiagnosticPieceRef Piece = maybeEmitNoteForObjCSelf(R, *MC, N))
483 return Piece;
484 }
485
486 if (const auto *CCall = dyn_cast<CXXConstructorCall>(Call)) {
487 // Do not generate diagnostics for not modified parameters in
488 // constructors.
489 return maybeEmitNoteForCXXThis(R, *CCall, N);
490 }
491
492 return maybeEmitNoteForParameters(R, *Call, N);
493 }
494
495 //===----------------------------------------------------------------------===//
496 // Implementation of NoStoreFuncVisitor.
497 //===----------------------------------------------------------------------===//
498
499 namespace {
500 /// Put a diagnostic on return statement of all inlined functions
501 /// for which the region of interest \p RegionOfInterest was passed into,
502 /// but not written inside, and it has caused an undefined read or a null
503 /// pointer dereference outside.
504 class NoStoreFuncVisitor final : public NoStateChangeFuncVisitor {
505 const SubRegion *RegionOfInterest;
506 MemRegionManager &MmrMgr;
507 const SourceManager &SM;
508 const PrintingPolicy &PP;
509
510 /// Recursion limit for dereferencing fields when looking for the
511 /// region of interest.
512 /// The limit of two indicates that we will dereference fields only once.
513 static const unsigned DEREFERENCE_LIMIT = 2;
514
515 using RegionVector = SmallVector<const MemRegion *, 5>;
516
517 public:
NoStoreFuncVisitor(const SubRegion * R,bugreporter::TrackingKind TKind)518 NoStoreFuncVisitor(const SubRegion *R, bugreporter::TrackingKind TKind)
519 : NoStateChangeFuncVisitor(TKind), RegionOfInterest(R),
520 MmrMgr(R->getMemRegionManager()),
521 SM(MmrMgr.getContext().getSourceManager()),
522 PP(MmrMgr.getContext().getPrintingPolicy()) {}
523
Profile(llvm::FoldingSetNodeID & ID) const524 void Profile(llvm::FoldingSetNodeID &ID) const override {
525 static int Tag = 0;
526 ID.AddPointer(&Tag);
527 ID.AddPointer(RegionOfInterest);
528 }
529
530 private:
531 /// \return Whether \c RegionOfInterest was modified at \p CurrN compared to
532 /// the value it holds in \p CallExitBeginN.
533 bool wasModifiedBeforeCallExit(const ExplodedNode *CurrN,
534 const ExplodedNode *CallExitBeginN) override;
535
536 /// Attempts to find the region of interest in a given record decl,
537 /// by either following the base classes or fields.
538 /// Dereferences fields up to a given recursion limit.
539 /// Note that \p Vec is passed by value, leading to quadratic copying cost,
540 /// but it's OK in practice since its length is limited to DEREFERENCE_LIMIT.
541 /// \return A chain fields leading to the region of interest or std::nullopt.
542 const std::optional<RegionVector>
543 findRegionOfInterestInRecord(const RecordDecl *RD, ProgramStateRef State,
544 const MemRegion *R, const RegionVector &Vec = {},
545 int depth = 0);
546
547 // Region of interest corresponds to an IVar, exiting a method
548 // which could have written into that IVar, but did not.
549 PathDiagnosticPieceRef maybeEmitNoteForObjCSelf(PathSensitiveBugReport &R,
550 const ObjCMethodCall &Call,
551 const ExplodedNode *N) final;
552
553 PathDiagnosticPieceRef maybeEmitNoteForCXXThis(PathSensitiveBugReport &R,
554 const CXXConstructorCall &Call,
555 const ExplodedNode *N) final;
556
557 PathDiagnosticPieceRef
558 maybeEmitNoteForParameters(PathSensitiveBugReport &R, const CallEvent &Call,
559 const ExplodedNode *N) final;
560
561 /// Consume the information on the no-store stack frame in order to
562 /// either emit a note or suppress the report enirely.
563 /// \return Diagnostics piece for region not modified in the current function,
564 /// if it decides to emit one.
565 PathDiagnosticPieceRef
566 maybeEmitNote(PathSensitiveBugReport &R, const CallEvent &Call,
567 const ExplodedNode *N, const RegionVector &FieldChain,
568 const MemRegion *MatchedRegion, StringRef FirstElement,
569 bool FirstIsReferenceType, unsigned IndirectionLevel);
570
571 bool prettyPrintRegionName(const RegionVector &FieldChain,
572 const MemRegion *MatchedRegion,
573 StringRef FirstElement, bool FirstIsReferenceType,
574 unsigned IndirectionLevel,
575 llvm::raw_svector_ostream &os);
576
577 StringRef prettyPrintFirstElement(StringRef FirstElement,
578 bool MoreItemsExpected,
579 int IndirectionLevel,
580 llvm::raw_svector_ostream &os);
581 };
582 } // namespace
583
584 /// \return Whether the method declaration \p Parent
585 /// syntactically has a binary operation writing into the ivar \p Ivar.
potentiallyWritesIntoIvar(const Decl * Parent,const ObjCIvarDecl * Ivar)586 static bool potentiallyWritesIntoIvar(const Decl *Parent,
587 const ObjCIvarDecl *Ivar) {
588 using namespace ast_matchers;
589 const char *IvarBind = "Ivar";
590 if (!Parent || !Parent->hasBody())
591 return false;
592 StatementMatcher WriteIntoIvarM = binaryOperator(
593 hasOperatorName("="),
594 hasLHS(ignoringParenImpCasts(
595 objcIvarRefExpr(hasDeclaration(equalsNode(Ivar))).bind(IvarBind))));
596 StatementMatcher ParentM = stmt(hasDescendant(WriteIntoIvarM));
597 auto Matches = match(ParentM, *Parent->getBody(), Parent->getASTContext());
598 for (BoundNodes &Match : Matches) {
599 auto IvarRef = Match.getNodeAs<ObjCIvarRefExpr>(IvarBind);
600 if (IvarRef->isFreeIvar())
601 return true;
602
603 const Expr *Base = IvarRef->getBase();
604 if (const auto *ICE = dyn_cast<ImplicitCastExpr>(Base))
605 Base = ICE->getSubExpr();
606
607 if (const auto *DRE = dyn_cast<DeclRefExpr>(Base))
608 if (const auto *ID = dyn_cast<ImplicitParamDecl>(DRE->getDecl()))
609 if (ID->getParameterKind() == ImplicitParamDecl::ObjCSelf)
610 return true;
611
612 return false;
613 }
614 return false;
615 }
616
617 /// Attempts to find the region of interest in a given CXX decl,
618 /// by either following the base classes or fields.
619 /// Dereferences fields up to a given recursion limit.
620 /// Note that \p Vec is passed by value, leading to quadratic copying cost,
621 /// but it's OK in practice since its length is limited to DEREFERENCE_LIMIT.
622 /// \return A chain fields leading to the region of interest or std::nullopt.
623 const std::optional<NoStoreFuncVisitor::RegionVector>
findRegionOfInterestInRecord(const RecordDecl * RD,ProgramStateRef State,const MemRegion * R,const NoStoreFuncVisitor::RegionVector & Vec,int depth)624 NoStoreFuncVisitor::findRegionOfInterestInRecord(
625 const RecordDecl *RD, ProgramStateRef State, const MemRegion *R,
626 const NoStoreFuncVisitor::RegionVector &Vec /* = {} */,
627 int depth /* = 0 */) {
628
629 if (depth == DEREFERENCE_LIMIT) // Limit the recursion depth.
630 return std::nullopt;
631
632 if (const auto *RDX = dyn_cast<CXXRecordDecl>(RD))
633 if (!RDX->hasDefinition())
634 return std::nullopt;
635
636 // Recursively examine the base classes.
637 // Note that following base classes does not increase the recursion depth.
638 if (const auto *RDX = dyn_cast<CXXRecordDecl>(RD))
639 for (const auto &II : RDX->bases())
640 if (const RecordDecl *RRD = II.getType()->getAsRecordDecl())
641 if (std::optional<RegionVector> Out =
642 findRegionOfInterestInRecord(RRD, State, R, Vec, depth))
643 return Out;
644
645 for (const FieldDecl *I : RD->fields()) {
646 QualType FT = I->getType();
647 const FieldRegion *FR = MmrMgr.getFieldRegion(I, cast<SubRegion>(R));
648 const SVal V = State->getSVal(FR);
649 const MemRegion *VR = V.getAsRegion();
650
651 RegionVector VecF = Vec;
652 VecF.push_back(FR);
653
654 if (RegionOfInterest == VR)
655 return VecF;
656
657 if (const RecordDecl *RRD = FT->getAsRecordDecl())
658 if (auto Out =
659 findRegionOfInterestInRecord(RRD, State, FR, VecF, depth + 1))
660 return Out;
661
662 QualType PT = FT->getPointeeType();
663 if (PT.isNull() || PT->isVoidType() || !VR)
664 continue;
665
666 if (const RecordDecl *RRD = PT->getAsRecordDecl())
667 if (std::optional<RegionVector> Out =
668 findRegionOfInterestInRecord(RRD, State, VR, VecF, depth + 1))
669 return Out;
670 }
671
672 return std::nullopt;
673 }
674
675 PathDiagnosticPieceRef
maybeEmitNoteForObjCSelf(PathSensitiveBugReport & R,const ObjCMethodCall & Call,const ExplodedNode * N)676 NoStoreFuncVisitor::maybeEmitNoteForObjCSelf(PathSensitiveBugReport &R,
677 const ObjCMethodCall &Call,
678 const ExplodedNode *N) {
679 if (const auto *IvarR = dyn_cast<ObjCIvarRegion>(RegionOfInterest)) {
680 const MemRegion *SelfRegion = Call.getReceiverSVal().getAsRegion();
681 if (RegionOfInterest->isSubRegionOf(SelfRegion) &&
682 potentiallyWritesIntoIvar(Call.getRuntimeDefinition().getDecl(),
683 IvarR->getDecl()))
684 return maybeEmitNote(R, Call, N, {}, SelfRegion, "self",
685 /*FirstIsReferenceType=*/false, 1);
686 }
687 return nullptr;
688 }
689
690 PathDiagnosticPieceRef
maybeEmitNoteForCXXThis(PathSensitiveBugReport & R,const CXXConstructorCall & Call,const ExplodedNode * N)691 NoStoreFuncVisitor::maybeEmitNoteForCXXThis(PathSensitiveBugReport &R,
692 const CXXConstructorCall &Call,
693 const ExplodedNode *N) {
694 const MemRegion *ThisR = Call.getCXXThisVal().getAsRegion();
695 if (RegionOfInterest->isSubRegionOf(ThisR) && !Call.getDecl()->isImplicit())
696 return maybeEmitNote(R, Call, N, {}, ThisR, "this",
697 /*FirstIsReferenceType=*/false, 1);
698
699 // Do not generate diagnostics for not modified parameters in
700 // constructors.
701 return nullptr;
702 }
703
704 /// \return whether \p Ty points to a const type, or is a const reference.
isPointerToConst(QualType Ty)705 static bool isPointerToConst(QualType Ty) {
706 return !Ty->getPointeeType().isNull() &&
707 Ty->getPointeeType().getCanonicalType().isConstQualified();
708 }
709
maybeEmitNoteForParameters(PathSensitiveBugReport & R,const CallEvent & Call,const ExplodedNode * N)710 PathDiagnosticPieceRef NoStoreFuncVisitor::maybeEmitNoteForParameters(
711 PathSensitiveBugReport &R, const CallEvent &Call, const ExplodedNode *N) {
712 ArrayRef<ParmVarDecl *> Parameters = Call.parameters();
713 for (unsigned I = 0; I < Call.getNumArgs() && I < Parameters.size(); ++I) {
714 const ParmVarDecl *PVD = Parameters[I];
715 SVal V = Call.getArgSVal(I);
716 bool ParamIsReferenceType = PVD->getType()->isReferenceType();
717 std::string ParamName = PVD->getNameAsString();
718
719 unsigned IndirectionLevel = 1;
720 QualType T = PVD->getType();
721 while (const MemRegion *MR = V.getAsRegion()) {
722 if (RegionOfInterest->isSubRegionOf(MR) && !isPointerToConst(T))
723 return maybeEmitNote(R, Call, N, {}, MR, ParamName,
724 ParamIsReferenceType, IndirectionLevel);
725
726 QualType PT = T->getPointeeType();
727 if (PT.isNull() || PT->isVoidType())
728 break;
729
730 ProgramStateRef State = N->getState();
731
732 if (const RecordDecl *RD = PT->getAsRecordDecl())
733 if (std::optional<RegionVector> P =
734 findRegionOfInterestInRecord(RD, State, MR))
735 return maybeEmitNote(R, Call, N, *P, RegionOfInterest, ParamName,
736 ParamIsReferenceType, IndirectionLevel);
737
738 V = State->getSVal(MR, PT);
739 T = PT;
740 IndirectionLevel++;
741 }
742 }
743
744 return nullptr;
745 }
746
wasModifiedBeforeCallExit(const ExplodedNode * CurrN,const ExplodedNode * CallExitBeginN)747 bool NoStoreFuncVisitor::wasModifiedBeforeCallExit(
748 const ExplodedNode *CurrN, const ExplodedNode *CallExitBeginN) {
749 return ::wasRegionOfInterestModifiedAt(
750 RegionOfInterest, CurrN,
751 CallExitBeginN->getState()->getSVal(RegionOfInterest));
752 }
753
754 static llvm::StringLiteral WillBeUsedForACondition =
755 ", which participates in a condition later";
756
maybeEmitNote(PathSensitiveBugReport & R,const CallEvent & Call,const ExplodedNode * N,const RegionVector & FieldChain,const MemRegion * MatchedRegion,StringRef FirstElement,bool FirstIsReferenceType,unsigned IndirectionLevel)757 PathDiagnosticPieceRef NoStoreFuncVisitor::maybeEmitNote(
758 PathSensitiveBugReport &R, const CallEvent &Call, const ExplodedNode *N,
759 const RegionVector &FieldChain, const MemRegion *MatchedRegion,
760 StringRef FirstElement, bool FirstIsReferenceType,
761 unsigned IndirectionLevel) {
762
763 PathDiagnosticLocation L =
764 PathDiagnosticLocation::create(N->getLocation(), SM);
765
766 // For now this shouldn't trigger, but once it does (as we add more
767 // functions to the body farm), we'll need to decide if these reports
768 // are worth suppressing as well.
769 if (!L.hasValidLocation())
770 return nullptr;
771
772 SmallString<256> sbuf;
773 llvm::raw_svector_ostream os(sbuf);
774 os << "Returning without writing to '";
775
776 // Do not generate the note if failed to pretty-print.
777 if (!prettyPrintRegionName(FieldChain, MatchedRegion, FirstElement,
778 FirstIsReferenceType, IndirectionLevel, os))
779 return nullptr;
780
781 os << "'";
782 if (TKind == bugreporter::TrackingKind::Condition)
783 os << WillBeUsedForACondition;
784 return std::make_shared<PathDiagnosticEventPiece>(L, os.str());
785 }
786
prettyPrintRegionName(const RegionVector & FieldChain,const MemRegion * MatchedRegion,StringRef FirstElement,bool FirstIsReferenceType,unsigned IndirectionLevel,llvm::raw_svector_ostream & os)787 bool NoStoreFuncVisitor::prettyPrintRegionName(const RegionVector &FieldChain,
788 const MemRegion *MatchedRegion,
789 StringRef FirstElement,
790 bool FirstIsReferenceType,
791 unsigned IndirectionLevel,
792 llvm::raw_svector_ostream &os) {
793
794 if (FirstIsReferenceType)
795 IndirectionLevel--;
796
797 RegionVector RegionSequence;
798
799 // Add the regions in the reverse order, then reverse the resulting array.
800 assert(RegionOfInterest->isSubRegionOf(MatchedRegion));
801 const MemRegion *R = RegionOfInterest;
802 while (R != MatchedRegion) {
803 RegionSequence.push_back(R);
804 R = cast<SubRegion>(R)->getSuperRegion();
805 }
806 std::reverse(RegionSequence.begin(), RegionSequence.end());
807 RegionSequence.append(FieldChain.begin(), FieldChain.end());
808
809 StringRef Sep;
810 for (const MemRegion *R : RegionSequence) {
811
812 // Just keep going up to the base region.
813 // Element regions may appear due to casts.
814 if (isa<CXXBaseObjectRegion, CXXTempObjectRegion>(R))
815 continue;
816
817 if (Sep.empty())
818 Sep = prettyPrintFirstElement(FirstElement,
819 /*MoreItemsExpected=*/true,
820 IndirectionLevel, os);
821
822 os << Sep;
823
824 // Can only reasonably pretty-print DeclRegions.
825 if (!isa<DeclRegion>(R))
826 return false;
827
828 const auto *DR = cast<DeclRegion>(R);
829 Sep = DR->getValueType()->isAnyPointerType() ? "->" : ".";
830 DR->getDecl()->getDeclName().print(os, PP);
831 }
832
833 if (Sep.empty())
834 prettyPrintFirstElement(FirstElement,
835 /*MoreItemsExpected=*/false, IndirectionLevel, os);
836 return true;
837 }
838
prettyPrintFirstElement(StringRef FirstElement,bool MoreItemsExpected,int IndirectionLevel,llvm::raw_svector_ostream & os)839 StringRef NoStoreFuncVisitor::prettyPrintFirstElement(
840 StringRef FirstElement, bool MoreItemsExpected, int IndirectionLevel,
841 llvm::raw_svector_ostream &os) {
842 StringRef Out = ".";
843
844 if (IndirectionLevel > 0 && MoreItemsExpected) {
845 IndirectionLevel--;
846 Out = "->";
847 }
848
849 if (IndirectionLevel > 0 && MoreItemsExpected)
850 os << "(";
851
852 for (int i = 0; i < IndirectionLevel; i++)
853 os << "*";
854 os << FirstElement;
855
856 if (IndirectionLevel > 0 && MoreItemsExpected)
857 os << ")";
858
859 return Out;
860 }
861
862 //===----------------------------------------------------------------------===//
863 // Implementation of MacroNullReturnSuppressionVisitor.
864 //===----------------------------------------------------------------------===//
865
866 namespace {
867
868 /// Suppress null-pointer-dereference bugs where dereferenced null was returned
869 /// the macro.
870 class MacroNullReturnSuppressionVisitor final : public BugReporterVisitor {
871 const SubRegion *RegionOfInterest;
872 const SVal ValueAtDereference;
873
874 // Do not invalidate the reports where the value was modified
875 // after it got assigned to from the macro.
876 bool WasModified = false;
877
878 public:
MacroNullReturnSuppressionVisitor(const SubRegion * R,const SVal V)879 MacroNullReturnSuppressionVisitor(const SubRegion *R, const SVal V)
880 : RegionOfInterest(R), ValueAtDereference(V) {}
881
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)882 PathDiagnosticPieceRef VisitNode(const ExplodedNode *N,
883 BugReporterContext &BRC,
884 PathSensitiveBugReport &BR) override {
885 if (WasModified)
886 return nullptr;
887
888 auto BugPoint = BR.getErrorNode()->getLocation().getAs<StmtPoint>();
889 if (!BugPoint)
890 return nullptr;
891
892 const SourceManager &SMgr = BRC.getSourceManager();
893 if (auto Loc = matchAssignment(N)) {
894 if (isFunctionMacroExpansion(*Loc, SMgr)) {
895 std::string MacroName = std::string(getMacroName(*Loc, BRC));
896 SourceLocation BugLoc = BugPoint->getStmt()->getBeginLoc();
897 if (!BugLoc.isMacroID() || getMacroName(BugLoc, BRC) != MacroName)
898 BR.markInvalid(getTag(), MacroName.c_str());
899 }
900 }
901
902 if (wasRegionOfInterestModifiedAt(RegionOfInterest, N, ValueAtDereference))
903 WasModified = true;
904
905 return nullptr;
906 }
907
addMacroVisitorIfNecessary(const ExplodedNode * N,const MemRegion * R,bool EnableNullFPSuppression,PathSensitiveBugReport & BR,const SVal V)908 static void addMacroVisitorIfNecessary(
909 const ExplodedNode *N, const MemRegion *R,
910 bool EnableNullFPSuppression, PathSensitiveBugReport &BR,
911 const SVal V) {
912 AnalyzerOptions &Options = N->getState()->getAnalysisManager().options;
913 if (EnableNullFPSuppression && Options.ShouldSuppressNullReturnPaths &&
914 isa<Loc>(V))
915 BR.addVisitor<MacroNullReturnSuppressionVisitor>(R->getAs<SubRegion>(),
916 V);
917 }
918
getTag() const919 void* getTag() const {
920 static int Tag = 0;
921 return static_cast<void *>(&Tag);
922 }
923
Profile(llvm::FoldingSetNodeID & ID) const924 void Profile(llvm::FoldingSetNodeID &ID) const override {
925 ID.AddPointer(getTag());
926 }
927
928 private:
929 /// \return Source location of right hand side of an assignment
930 /// into \c RegionOfInterest, empty optional if none found.
matchAssignment(const ExplodedNode * N)931 std::optional<SourceLocation> matchAssignment(const ExplodedNode *N) {
932 const Stmt *S = N->getStmtForDiagnostics();
933 ProgramStateRef State = N->getState();
934 auto *LCtx = N->getLocationContext();
935 if (!S)
936 return std::nullopt;
937
938 if (const auto *DS = dyn_cast<DeclStmt>(S)) {
939 if (const auto *VD = dyn_cast<VarDecl>(DS->getSingleDecl()))
940 if (const Expr *RHS = VD->getInit())
941 if (RegionOfInterest->isSubRegionOf(
942 State->getLValue(VD, LCtx).getAsRegion()))
943 return RHS->getBeginLoc();
944 } else if (const auto *BO = dyn_cast<BinaryOperator>(S)) {
945 const MemRegion *R = N->getSVal(BO->getLHS()).getAsRegion();
946 const Expr *RHS = BO->getRHS();
947 if (BO->isAssignmentOp() && RegionOfInterest->isSubRegionOf(R)) {
948 return RHS->getBeginLoc();
949 }
950 }
951 return std::nullopt;
952 }
953 };
954
955 } // end of anonymous namespace
956
957 namespace {
958
959 /// Emits an extra note at the return statement of an interesting stack frame.
960 ///
961 /// The returned value is marked as an interesting value, and if it's null,
962 /// adds a visitor to track where it became null.
963 ///
964 /// This visitor is intended to be used when another visitor discovers that an
965 /// interesting value comes from an inlined function call.
966 class ReturnVisitor : public TrackingBugReporterVisitor {
967 const StackFrameContext *CalleeSFC;
968 enum {
969 Initial,
970 MaybeUnsuppress,
971 Satisfied
972 } Mode = Initial;
973
974 bool EnableNullFPSuppression;
975 bool ShouldInvalidate = true;
976 AnalyzerOptions& Options;
977 bugreporter::TrackingKind TKind;
978
979 public:
ReturnVisitor(TrackerRef ParentTracker,const StackFrameContext * Frame,bool Suppressed,AnalyzerOptions & Options,bugreporter::TrackingKind TKind)980 ReturnVisitor(TrackerRef ParentTracker, const StackFrameContext *Frame,
981 bool Suppressed, AnalyzerOptions &Options,
982 bugreporter::TrackingKind TKind)
983 : TrackingBugReporterVisitor(ParentTracker), CalleeSFC(Frame),
984 EnableNullFPSuppression(Suppressed), Options(Options), TKind(TKind) {}
985
getTag()986 static void *getTag() {
987 static int Tag = 0;
988 return static_cast<void *>(&Tag);
989 }
990
Profile(llvm::FoldingSetNodeID & ID) const991 void Profile(llvm::FoldingSetNodeID &ID) const override {
992 ID.AddPointer(ReturnVisitor::getTag());
993 ID.AddPointer(CalleeSFC);
994 ID.AddBoolean(EnableNullFPSuppression);
995 }
996
visitNodeInitial(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)997 PathDiagnosticPieceRef visitNodeInitial(const ExplodedNode *N,
998 BugReporterContext &BRC,
999 PathSensitiveBugReport &BR) {
1000 // Only print a message at the interesting return statement.
1001 if (N->getLocationContext() != CalleeSFC)
1002 return nullptr;
1003
1004 std::optional<StmtPoint> SP = N->getLocationAs<StmtPoint>();
1005 if (!SP)
1006 return nullptr;
1007
1008 const auto *Ret = dyn_cast<ReturnStmt>(SP->getStmt());
1009 if (!Ret)
1010 return nullptr;
1011
1012 // Okay, we're at the right return statement, but do we have the return
1013 // value available?
1014 ProgramStateRef State = N->getState();
1015 SVal V = State->getSVal(Ret, CalleeSFC);
1016 if (V.isUnknownOrUndef())
1017 return nullptr;
1018
1019 // Don't print any more notes after this one.
1020 Mode = Satisfied;
1021
1022 const Expr *RetE = Ret->getRetValue();
1023 assert(RetE && "Tracking a return value for a void function");
1024
1025 // Handle cases where a reference is returned and then immediately used.
1026 std::optional<Loc> LValue;
1027 if (RetE->isGLValue()) {
1028 if ((LValue = V.getAs<Loc>())) {
1029 SVal RValue = State->getRawSVal(*LValue, RetE->getType());
1030 if (isa<DefinedSVal>(RValue))
1031 V = RValue;
1032 }
1033 }
1034
1035 // Ignore aggregate rvalues.
1036 if (isa<nonloc::LazyCompoundVal, nonloc::CompoundVal>(V))
1037 return nullptr;
1038
1039 RetE = RetE->IgnoreParenCasts();
1040
1041 // Let's track the return value.
1042 getParentTracker().track(RetE, N, {TKind, EnableNullFPSuppression});
1043
1044 // Build an appropriate message based on the return value.
1045 SmallString<64> Msg;
1046 llvm::raw_svector_ostream Out(Msg);
1047
1048 bool WouldEventBeMeaningless = false;
1049
1050 if (State->isNull(V).isConstrainedTrue()) {
1051 if (isa<Loc>(V)) {
1052
1053 // If we have counter-suppression enabled, make sure we keep visiting
1054 // future nodes. We want to emit a path note as well, in case
1055 // the report is resurrected as valid later on.
1056 if (EnableNullFPSuppression &&
1057 Options.ShouldAvoidSuppressingNullArgumentPaths)
1058 Mode = MaybeUnsuppress;
1059
1060 if (RetE->getType()->isObjCObjectPointerType()) {
1061 Out << "Returning nil";
1062 } else {
1063 Out << "Returning null pointer";
1064 }
1065 } else {
1066 Out << "Returning zero";
1067 }
1068
1069 } else {
1070 if (auto CI = V.getAs<nonloc::ConcreteInt>()) {
1071 Out << "Returning the value " << CI->getValue();
1072 } else {
1073 // There is nothing interesting about returning a value, when it is
1074 // plain value without any constraints, and the function is guaranteed
1075 // to return that every time. We could use CFG::isLinear() here, but
1076 // constexpr branches are obvious to the compiler, not necesserily to
1077 // the programmer.
1078 if (N->getCFG().size() == 3)
1079 WouldEventBeMeaningless = true;
1080
1081 Out << (isa<Loc>(V) ? "Returning pointer" : "Returning value");
1082 }
1083 }
1084
1085 if (LValue) {
1086 if (const MemRegion *MR = LValue->getAsRegion()) {
1087 if (MR->canPrintPretty()) {
1088 Out << " (reference to ";
1089 MR->printPretty(Out);
1090 Out << ")";
1091 }
1092 }
1093 } else {
1094 // FIXME: We should have a more generalized location printing mechanism.
1095 if (const auto *DR = dyn_cast<DeclRefExpr>(RetE))
1096 if (const auto *DD = dyn_cast<DeclaratorDecl>(DR->getDecl()))
1097 Out << " (loaded from '" << *DD << "')";
1098 }
1099
1100 PathDiagnosticLocation L(Ret, BRC.getSourceManager(), CalleeSFC);
1101 if (!L.isValid() || !L.asLocation().isValid())
1102 return nullptr;
1103
1104 if (TKind == bugreporter::TrackingKind::Condition)
1105 Out << WillBeUsedForACondition;
1106
1107 auto EventPiece = std::make_shared<PathDiagnosticEventPiece>(L, Out.str());
1108
1109 // If we determined that the note is meaningless, make it prunable, and
1110 // don't mark the stackframe interesting.
1111 if (WouldEventBeMeaningless)
1112 EventPiece->setPrunable(true);
1113 else
1114 BR.markInteresting(CalleeSFC);
1115
1116 return EventPiece;
1117 }
1118
visitNodeMaybeUnsuppress(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)1119 PathDiagnosticPieceRef visitNodeMaybeUnsuppress(const ExplodedNode *N,
1120 BugReporterContext &BRC,
1121 PathSensitiveBugReport &BR) {
1122 assert(Options.ShouldAvoidSuppressingNullArgumentPaths);
1123
1124 // Are we at the entry node for this call?
1125 std::optional<CallEnter> CE = N->getLocationAs<CallEnter>();
1126 if (!CE)
1127 return nullptr;
1128
1129 if (CE->getCalleeContext() != CalleeSFC)
1130 return nullptr;
1131
1132 Mode = Satisfied;
1133
1134 // Don't automatically suppress a report if one of the arguments is
1135 // known to be a null pointer. Instead, start tracking /that/ null
1136 // value back to its origin.
1137 ProgramStateManager &StateMgr = BRC.getStateManager();
1138 CallEventManager &CallMgr = StateMgr.getCallEventManager();
1139
1140 ProgramStateRef State = N->getState();
1141 CallEventRef<> Call = CallMgr.getCaller(CalleeSFC, State);
1142 for (unsigned I = 0, E = Call->getNumArgs(); I != E; ++I) {
1143 std::optional<Loc> ArgV = Call->getArgSVal(I).getAs<Loc>();
1144 if (!ArgV)
1145 continue;
1146
1147 const Expr *ArgE = Call->getArgExpr(I);
1148 if (!ArgE)
1149 continue;
1150
1151 // Is it possible for this argument to be non-null?
1152 if (!State->isNull(*ArgV).isConstrainedTrue())
1153 continue;
1154
1155 if (getParentTracker()
1156 .track(ArgE, N, {TKind, EnableNullFPSuppression})
1157 .FoundSomethingToTrack)
1158 ShouldInvalidate = false;
1159
1160 // If we /can't/ track the null pointer, we should err on the side of
1161 // false negatives, and continue towards marking this report invalid.
1162 // (We will still look at the other arguments, though.)
1163 }
1164
1165 return nullptr;
1166 }
1167
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)1168 PathDiagnosticPieceRef VisitNode(const ExplodedNode *N,
1169 BugReporterContext &BRC,
1170 PathSensitiveBugReport &BR) override {
1171 switch (Mode) {
1172 case Initial:
1173 return visitNodeInitial(N, BRC, BR);
1174 case MaybeUnsuppress:
1175 return visitNodeMaybeUnsuppress(N, BRC, BR);
1176 case Satisfied:
1177 return nullptr;
1178 }
1179
1180 llvm_unreachable("Invalid visit mode!");
1181 }
1182
finalizeVisitor(BugReporterContext &,const ExplodedNode *,PathSensitiveBugReport & BR)1183 void finalizeVisitor(BugReporterContext &, const ExplodedNode *,
1184 PathSensitiveBugReport &BR) override {
1185 if (EnableNullFPSuppression && ShouldInvalidate)
1186 BR.markInvalid(ReturnVisitor::getTag(), CalleeSFC);
1187 }
1188 };
1189
1190 //===----------------------------------------------------------------------===//
1191 // StoreSiteFinder
1192 //===----------------------------------------------------------------------===//
1193
1194 /// Finds last store into the given region,
1195 /// which is different from a given symbolic value.
1196 class StoreSiteFinder final : public TrackingBugReporterVisitor {
1197 const MemRegion *R;
1198 SVal V;
1199 bool Satisfied = false;
1200
1201 TrackingOptions Options;
1202 const StackFrameContext *OriginSFC;
1203
1204 public:
1205 /// \param V We're searching for the store where \c R received this value.
1206 /// \param R The region we're tracking.
1207 /// \param Options Tracking behavior options.
1208 /// \param OriginSFC Only adds notes when the last store happened in a
1209 /// different stackframe to this one. Disregarded if the tracking kind
1210 /// is thorough.
1211 /// This is useful, because for non-tracked regions, notes about
1212 /// changes to its value in a nested stackframe could be pruned, and
1213 /// this visitor can prevent that without polluting the bugpath too
1214 /// much.
StoreSiteFinder(bugreporter::TrackerRef ParentTracker,KnownSVal V,const MemRegion * R,TrackingOptions Options,const StackFrameContext * OriginSFC=nullptr)1215 StoreSiteFinder(bugreporter::TrackerRef ParentTracker, KnownSVal V,
1216 const MemRegion *R, TrackingOptions Options,
1217 const StackFrameContext *OriginSFC = nullptr)
1218 : TrackingBugReporterVisitor(ParentTracker), R(R), V(V), Options(Options),
1219 OriginSFC(OriginSFC) {
1220 assert(R);
1221 }
1222
1223 void Profile(llvm::FoldingSetNodeID &ID) const override;
1224
1225 PathDiagnosticPieceRef VisitNode(const ExplodedNode *N,
1226 BugReporterContext &BRC,
1227 PathSensitiveBugReport &BR) override;
1228 };
1229 } // namespace
1230
Profile(llvm::FoldingSetNodeID & ID) const1231 void StoreSiteFinder::Profile(llvm::FoldingSetNodeID &ID) const {
1232 static int tag = 0;
1233 ID.AddPointer(&tag);
1234 ID.AddPointer(R);
1235 ID.Add(V);
1236 ID.AddInteger(static_cast<int>(Options.Kind));
1237 ID.AddBoolean(Options.EnableNullFPSuppression);
1238 }
1239
1240 /// Returns true if \p N represents the DeclStmt declaring and initializing
1241 /// \p VR.
isInitializationOfVar(const ExplodedNode * N,const VarRegion * VR)1242 static bool isInitializationOfVar(const ExplodedNode *N, const VarRegion *VR) {
1243 std::optional<PostStmt> P = N->getLocationAs<PostStmt>();
1244 if (!P)
1245 return false;
1246
1247 const DeclStmt *DS = P->getStmtAs<DeclStmt>();
1248 if (!DS)
1249 return false;
1250
1251 if (DS->getSingleDecl() != VR->getDecl())
1252 return false;
1253
1254 const MemSpaceRegion *VarSpace = VR->getMemorySpace();
1255 const auto *FrameSpace = dyn_cast<StackSpaceRegion>(VarSpace);
1256 if (!FrameSpace) {
1257 // If we ever directly evaluate global DeclStmts, this assertion will be
1258 // invalid, but this still seems preferable to silently accepting an
1259 // initialization that may be for a path-sensitive variable.
1260 assert(VR->getDecl()->isStaticLocal() && "non-static stackless VarRegion");
1261 return true;
1262 }
1263
1264 assert(VR->getDecl()->hasLocalStorage());
1265 const LocationContext *LCtx = N->getLocationContext();
1266 return FrameSpace->getStackFrame() == LCtx->getStackFrame();
1267 }
1268
isObjCPointer(const MemRegion * R)1269 static bool isObjCPointer(const MemRegion *R) {
1270 if (R->isBoundable())
1271 if (const auto *TR = dyn_cast<TypedValueRegion>(R))
1272 return TR->getValueType()->isObjCObjectPointerType();
1273
1274 return false;
1275 }
1276
isObjCPointer(const ValueDecl * D)1277 static bool isObjCPointer(const ValueDecl *D) {
1278 return D->getType()->isObjCObjectPointerType();
1279 }
1280
1281 /// Show diagnostics for initializing or declaring a region \p R with a bad value.
showBRDiagnostics(llvm::raw_svector_ostream & OS,StoreInfo SI)1282 static void showBRDiagnostics(llvm::raw_svector_ostream &OS, StoreInfo SI) {
1283 const bool HasPrefix = SI.Dest->canPrintPretty();
1284
1285 if (HasPrefix) {
1286 SI.Dest->printPretty(OS);
1287 OS << " ";
1288 }
1289
1290 const char *Action = nullptr;
1291
1292 switch (SI.StoreKind) {
1293 case StoreInfo::Initialization:
1294 Action = HasPrefix ? "initialized to " : "Initializing to ";
1295 break;
1296 case StoreInfo::BlockCapture:
1297 Action = HasPrefix ? "captured by block as " : "Captured by block as ";
1298 break;
1299 default:
1300 llvm_unreachable("Unexpected store kind");
1301 }
1302
1303 if (isa<loc::ConcreteInt>(SI.Value)) {
1304 OS << Action << (isObjCPointer(SI.Dest) ? "nil" : "a null pointer value");
1305
1306 } else if (auto CVal = SI.Value.getAs<nonloc::ConcreteInt>()) {
1307 OS << Action << CVal->getValue();
1308
1309 } else if (SI.Origin && SI.Origin->canPrintPretty()) {
1310 OS << Action << "the value of ";
1311 SI.Origin->printPretty(OS);
1312
1313 } else if (SI.StoreKind == StoreInfo::Initialization) {
1314 // We don't need to check here, all these conditions were
1315 // checked by StoreSiteFinder, when it figured out that it is
1316 // initialization.
1317 const auto *DS =
1318 cast<DeclStmt>(SI.StoreSite->getLocationAs<PostStmt>()->getStmt());
1319
1320 if (SI.Value.isUndef()) {
1321 if (isa<VarRegion>(SI.Dest)) {
1322 const auto *VD = cast<VarDecl>(DS->getSingleDecl());
1323
1324 if (VD->getInit()) {
1325 OS << (HasPrefix ? "initialized" : "Initializing")
1326 << " to a garbage value";
1327 } else {
1328 OS << (HasPrefix ? "declared" : "Declaring")
1329 << " without an initial value";
1330 }
1331 }
1332 } else {
1333 OS << (HasPrefix ? "initialized" : "Initialized") << " here";
1334 }
1335 }
1336 }
1337
1338 /// Display diagnostics for passing bad region as a parameter.
showBRParamDiagnostics(llvm::raw_svector_ostream & OS,StoreInfo SI)1339 static void showBRParamDiagnostics(llvm::raw_svector_ostream &OS,
1340 StoreInfo SI) {
1341 const auto *VR = cast<VarRegion>(SI.Dest);
1342 const auto *D = VR->getDecl();
1343
1344 OS << "Passing ";
1345
1346 if (isa<loc::ConcreteInt>(SI.Value)) {
1347 OS << (isObjCPointer(D) ? "nil object reference" : "null pointer value");
1348
1349 } else if (SI.Value.isUndef()) {
1350 OS << "uninitialized value";
1351
1352 } else if (auto CI = SI.Value.getAs<nonloc::ConcreteInt>()) {
1353 OS << "the value " << CI->getValue();
1354
1355 } else if (SI.Origin && SI.Origin->canPrintPretty()) {
1356 SI.Origin->printPretty(OS);
1357
1358 } else {
1359 OS << "value";
1360 }
1361
1362 if (const auto *Param = dyn_cast<ParmVarDecl>(VR->getDecl())) {
1363 // Printed parameter indexes are 1-based, not 0-based.
1364 unsigned Idx = Param->getFunctionScopeIndex() + 1;
1365 OS << " via " << Idx << llvm::getOrdinalSuffix(Idx) << " parameter";
1366 if (VR->canPrintPretty()) {
1367 OS << " ";
1368 VR->printPretty(OS);
1369 }
1370 } else if (const auto *ImplParam = dyn_cast<ImplicitParamDecl>(D)) {
1371 if (ImplParam->getParameterKind() ==
1372 ImplicitParamDecl::ImplicitParamKind::ObjCSelf) {
1373 OS << " via implicit parameter 'self'";
1374 }
1375 }
1376 }
1377
1378 /// Show default diagnostics for storing bad region.
showBRDefaultDiagnostics(llvm::raw_svector_ostream & OS,StoreInfo SI)1379 static void showBRDefaultDiagnostics(llvm::raw_svector_ostream &OS,
1380 StoreInfo SI) {
1381 const bool HasSuffix = SI.Dest->canPrintPretty();
1382
1383 if (isa<loc::ConcreteInt>(SI.Value)) {
1384 OS << (isObjCPointer(SI.Dest) ? "nil object reference stored"
1385 : (HasSuffix ? "Null pointer value stored"
1386 : "Storing null pointer value"));
1387
1388 } else if (SI.Value.isUndef()) {
1389 OS << (HasSuffix ? "Uninitialized value stored"
1390 : "Storing uninitialized value");
1391
1392 } else if (auto CV = SI.Value.getAs<nonloc::ConcreteInt>()) {
1393 if (HasSuffix)
1394 OS << "The value " << CV->getValue() << " is assigned";
1395 else
1396 OS << "Assigning " << CV->getValue();
1397
1398 } else if (SI.Origin && SI.Origin->canPrintPretty()) {
1399 if (HasSuffix) {
1400 OS << "The value of ";
1401 SI.Origin->printPretty(OS);
1402 OS << " is assigned";
1403 } else {
1404 OS << "Assigning the value of ";
1405 SI.Origin->printPretty(OS);
1406 }
1407
1408 } else {
1409 OS << (HasSuffix ? "Value assigned" : "Assigning value");
1410 }
1411
1412 if (HasSuffix) {
1413 OS << " to ";
1414 SI.Dest->printPretty(OS);
1415 }
1416 }
1417
isTrivialCopyOrMoveCtor(const CXXConstructExpr * CE)1418 static bool isTrivialCopyOrMoveCtor(const CXXConstructExpr *CE) {
1419 if (!CE)
1420 return false;
1421
1422 const auto *CtorDecl = CE->getConstructor();
1423
1424 return CtorDecl->isCopyOrMoveConstructor() && CtorDecl->isTrivial();
1425 }
1426
tryExtractInitializerFromList(const InitListExpr * ILE,const MemRegion * R)1427 static const Expr *tryExtractInitializerFromList(const InitListExpr *ILE,
1428 const MemRegion *R) {
1429
1430 const auto *TVR = dyn_cast_or_null<TypedValueRegion>(R);
1431
1432 if (!TVR)
1433 return nullptr;
1434
1435 const auto ITy = ILE->getType().getCanonicalType();
1436
1437 // Push each sub-region onto the stack.
1438 std::stack<const TypedValueRegion *> TVRStack;
1439 while (isa<FieldRegion>(TVR) || isa<ElementRegion>(TVR)) {
1440 // We found a region that matches the type of the init list,
1441 // so we assume this is the outer-most region. This can happen
1442 // if the initializer list is inside a class. If our assumption
1443 // is wrong, we return a nullptr in the end.
1444 if (ITy == TVR->getValueType().getCanonicalType())
1445 break;
1446
1447 TVRStack.push(TVR);
1448 TVR = cast<TypedValueRegion>(TVR->getSuperRegion());
1449 }
1450
1451 // If the type of the outer most region doesn't match the type
1452 // of the ILE, we can't match the ILE and the region.
1453 if (ITy != TVR->getValueType().getCanonicalType())
1454 return nullptr;
1455
1456 const Expr *Init = ILE;
1457 while (!TVRStack.empty()) {
1458 TVR = TVRStack.top();
1459 TVRStack.pop();
1460
1461 // We hit something that's not an init list before
1462 // running out of regions, so we most likely failed.
1463 if (!isa<InitListExpr>(Init))
1464 return nullptr;
1465
1466 ILE = cast<InitListExpr>(Init);
1467 auto NumInits = ILE->getNumInits();
1468
1469 if (const auto *FR = dyn_cast<FieldRegion>(TVR)) {
1470 const auto *FD = FR->getDecl();
1471
1472 if (FD->getFieldIndex() >= NumInits)
1473 return nullptr;
1474
1475 Init = ILE->getInit(FD->getFieldIndex());
1476 } else if (const auto *ER = dyn_cast<ElementRegion>(TVR)) {
1477 const auto Ind = ER->getIndex();
1478
1479 // If index is symbolic, we can't figure out which expression
1480 // belongs to the region.
1481 if (!Ind.isConstant())
1482 return nullptr;
1483
1484 const auto IndVal = Ind.getAsInteger()->getLimitedValue();
1485 if (IndVal >= NumInits)
1486 return nullptr;
1487
1488 Init = ILE->getInit(IndVal);
1489 }
1490 }
1491
1492 return Init;
1493 }
1494
VisitNode(const ExplodedNode * Succ,BugReporterContext & BRC,PathSensitiveBugReport & BR)1495 PathDiagnosticPieceRef StoreSiteFinder::VisitNode(const ExplodedNode *Succ,
1496 BugReporterContext &BRC,
1497 PathSensitiveBugReport &BR) {
1498 if (Satisfied)
1499 return nullptr;
1500
1501 const ExplodedNode *StoreSite = nullptr;
1502 const ExplodedNode *Pred = Succ->getFirstPred();
1503 const Expr *InitE = nullptr;
1504 bool IsParam = false;
1505
1506 // First see if we reached the declaration of the region.
1507 if (const auto *VR = dyn_cast<VarRegion>(R)) {
1508 if (isInitializationOfVar(Pred, VR)) {
1509 StoreSite = Pred;
1510 InitE = VR->getDecl()->getInit();
1511 }
1512 }
1513
1514 // If this is a post initializer expression, initializing the region, we
1515 // should track the initializer expression.
1516 if (std::optional<PostInitializer> PIP =
1517 Pred->getLocationAs<PostInitializer>()) {
1518 const MemRegion *FieldReg = (const MemRegion *)PIP->getLocationValue();
1519 if (FieldReg == R) {
1520 StoreSite = Pred;
1521 InitE = PIP->getInitializer()->getInit();
1522 }
1523 }
1524
1525 // Otherwise, see if this is the store site:
1526 // (1) Succ has this binding and Pred does not, i.e. this is
1527 // where the binding first occurred.
1528 // (2) Succ has this binding and is a PostStore node for this region, i.e.
1529 // the same binding was re-assigned here.
1530 if (!StoreSite) {
1531 if (Succ->getState()->getSVal(R) != V)
1532 return nullptr;
1533
1534 if (hasVisibleUpdate(Pred, Pred->getState()->getSVal(R), Succ, V)) {
1535 std::optional<PostStore> PS = Succ->getLocationAs<PostStore>();
1536 if (!PS || PS->getLocationValue() != R)
1537 return nullptr;
1538 }
1539
1540 StoreSite = Succ;
1541
1542 if (std::optional<PostStmt> P = Succ->getLocationAs<PostStmt>()) {
1543 // If this is an assignment expression, we can track the value
1544 // being assigned.
1545 if (const BinaryOperator *BO = P->getStmtAs<BinaryOperator>()) {
1546 if (BO->isAssignmentOp())
1547 InitE = BO->getRHS();
1548 }
1549 // If we have a declaration like 'S s{1,2}' that needs special
1550 // handling, we handle it here.
1551 else if (const auto *DS = P->getStmtAs<DeclStmt>()) {
1552 const auto *Decl = DS->getSingleDecl();
1553 if (isa<VarDecl>(Decl)) {
1554 const auto *VD = cast<VarDecl>(Decl);
1555
1556 // FIXME: Here we only track the inner most region, so we lose
1557 // information, but it's still better than a crash or no information
1558 // at all.
1559 //
1560 // E.g.: The region we have is 's.s2.s3.s4.y' and we only track 'y',
1561 // and throw away the rest.
1562 if (const auto *ILE = dyn_cast<InitListExpr>(VD->getInit()))
1563 InitE = tryExtractInitializerFromList(ILE, R);
1564 }
1565 } else if (const auto *CE = P->getStmtAs<CXXConstructExpr>()) {
1566
1567 const auto State = Succ->getState();
1568
1569 if (isTrivialCopyOrMoveCtor(CE) && isa<SubRegion>(R)) {
1570 // Migrate the field regions from the current object to
1571 // the parent object. If we track 'a.y.e' and encounter
1572 // 'S a = b' then we need to track 'b.y.e'.
1573
1574 // Push the regions to a stack, from last to first, so
1575 // considering the example above the stack will look like
1576 // (bottom) 'e' -> 'y' (top).
1577
1578 std::stack<const SubRegion *> SRStack;
1579 const SubRegion *SR = cast<SubRegion>(R);
1580 while (isa<FieldRegion>(SR) || isa<ElementRegion>(SR)) {
1581 SRStack.push(SR);
1582 SR = cast<SubRegion>(SR->getSuperRegion());
1583 }
1584
1585 // Get the region for the object we copied/moved from.
1586 const auto *OriginEx = CE->getArg(0);
1587 const auto OriginVal =
1588 State->getSVal(OriginEx, Succ->getLocationContext());
1589
1590 // Pop the stored field regions and apply them to the origin
1591 // object in the same order we had them on the copy.
1592 // OriginField will evolve like 'b' -> 'b.y' -> 'b.y.e'.
1593 SVal OriginField = OriginVal;
1594 while (!SRStack.empty()) {
1595 const auto *TopR = SRStack.top();
1596 SRStack.pop();
1597
1598 if (const auto *FR = dyn_cast<FieldRegion>(TopR)) {
1599 OriginField = State->getLValue(FR->getDecl(), OriginField);
1600 } else if (const auto *ER = dyn_cast<ElementRegion>(TopR)) {
1601 OriginField = State->getLValue(ER->getElementType(),
1602 ER->getIndex(), OriginField);
1603 } else {
1604 // FIXME: handle other region type
1605 }
1606 }
1607
1608 // Track 'b.y.e'.
1609 getParentTracker().track(V, OriginField.getAsRegion(), Options);
1610 InitE = OriginEx;
1611 }
1612 }
1613 // This branch can occur in cases like `Ctor() : field{ x, y } {}'.
1614 else if (const auto *ILE = P->getStmtAs<InitListExpr>()) {
1615 // FIXME: Here we only track the top level region, so we lose
1616 // information, but it's still better than a crash or no information
1617 // at all.
1618 //
1619 // E.g.: The region we have is 's.s2.s3.s4.y' and we only track 'y', and
1620 // throw away the rest.
1621 InitE = tryExtractInitializerFromList(ILE, R);
1622 }
1623 }
1624
1625 // If this is a call entry, the variable should be a parameter.
1626 // FIXME: Handle CXXThisRegion as well. (This is not a priority because
1627 // 'this' should never be NULL, but this visitor isn't just for NULL and
1628 // UndefinedVal.)
1629 if (std::optional<CallEnter> CE = Succ->getLocationAs<CallEnter>()) {
1630 if (const auto *VR = dyn_cast<VarRegion>(R)) {
1631
1632 if (const auto *Param = dyn_cast<ParmVarDecl>(VR->getDecl())) {
1633 ProgramStateManager &StateMgr = BRC.getStateManager();
1634 CallEventManager &CallMgr = StateMgr.getCallEventManager();
1635
1636 CallEventRef<> Call = CallMgr.getCaller(CE->getCalleeContext(),
1637 Succ->getState());
1638 InitE = Call->getArgExpr(Param->getFunctionScopeIndex());
1639 } else {
1640 // Handle Objective-C 'self'.
1641 assert(isa<ImplicitParamDecl>(VR->getDecl()));
1642 InitE = cast<ObjCMessageExpr>(CE->getCalleeContext()->getCallSite())
1643 ->getInstanceReceiver()->IgnoreParenCasts();
1644 }
1645 IsParam = true;
1646 }
1647 }
1648
1649 // If this is a CXXTempObjectRegion, the Expr responsible for its creation
1650 // is wrapped inside of it.
1651 if (const auto *TmpR = dyn_cast<CXXTempObjectRegion>(R))
1652 InitE = TmpR->getExpr();
1653 }
1654
1655 if (!StoreSite)
1656 return nullptr;
1657
1658 Satisfied = true;
1659
1660 // If we have an expression that provided the value, try to track where it
1661 // came from.
1662 if (InitE) {
1663 if (!IsParam)
1664 InitE = InitE->IgnoreParenCasts();
1665
1666 getParentTracker().track(InitE, StoreSite, Options);
1667 }
1668
1669 // Let's try to find the region where the value came from.
1670 const MemRegion *OldRegion = nullptr;
1671
1672 // If we have init expression, it might be simply a reference
1673 // to a variable, so we can use it.
1674 if (InitE) {
1675 // That region might still be not exactly what we are looking for.
1676 // In situations like `int &ref = val;`, we can't say that
1677 // `ref` is initialized with `val`, rather refers to `val`.
1678 //
1679 // In order, to mitigate situations like this, we check if the last
1680 // stored value in that region is the value that we track.
1681 //
1682 // TODO: support other situations better.
1683 if (const MemRegion *Candidate =
1684 getLocationRegionIfReference(InitE, Succ, false)) {
1685 const StoreManager &SM = BRC.getStateManager().getStoreManager();
1686
1687 // Here we traverse the graph up to find the last node where the
1688 // candidate region is still in the store.
1689 for (const ExplodedNode *N = StoreSite; N; N = N->getFirstPred()) {
1690 if (SM.includedInBindings(N->getState()->getStore(), Candidate)) {
1691 // And if it was bound to the target value, we can use it.
1692 if (N->getState()->getSVal(Candidate) == V) {
1693 OldRegion = Candidate;
1694 }
1695 break;
1696 }
1697 }
1698 }
1699 }
1700
1701 // Otherwise, if the current region does indeed contain the value
1702 // we are looking for, we can look for a region where this value
1703 // was before.
1704 //
1705 // It can be useful for situations like:
1706 // new = identity(old)
1707 // where the analyzer knows that 'identity' returns the value of its
1708 // first argument.
1709 //
1710 // NOTE: If the region R is not a simple var region, it can contain
1711 // V in one of its subregions.
1712 if (!OldRegion && StoreSite->getState()->getSVal(R) == V) {
1713 // Let's go up the graph to find the node where the region is
1714 // bound to V.
1715 const ExplodedNode *NodeWithoutBinding = StoreSite->getFirstPred();
1716 for (;
1717 NodeWithoutBinding && NodeWithoutBinding->getState()->getSVal(R) == V;
1718 NodeWithoutBinding = NodeWithoutBinding->getFirstPred()) {
1719 }
1720
1721 if (NodeWithoutBinding) {
1722 // Let's try to find a unique binding for the value in that node.
1723 // We want to use this to find unique bindings because of the following
1724 // situations:
1725 // b = a;
1726 // c = identity(b);
1727 //
1728 // Telling the user that the value of 'a' is assigned to 'c', while
1729 // correct, can be confusing.
1730 StoreManager::FindUniqueBinding FB(V.getAsLocSymbol());
1731 BRC.getStateManager().iterBindings(NodeWithoutBinding->getState(), FB);
1732 if (FB)
1733 OldRegion = FB.getRegion();
1734 }
1735 }
1736
1737 if (Options.Kind == TrackingKind::Condition && OriginSFC &&
1738 !OriginSFC->isParentOf(StoreSite->getStackFrame()))
1739 return nullptr;
1740
1741 // Okay, we've found the binding. Emit an appropriate message.
1742 SmallString<256> sbuf;
1743 llvm::raw_svector_ostream os(sbuf);
1744
1745 StoreInfo SI = {StoreInfo::Assignment, // default kind
1746 StoreSite,
1747 InitE,
1748 V,
1749 R,
1750 OldRegion};
1751
1752 if (std::optional<PostStmt> PS = StoreSite->getLocationAs<PostStmt>()) {
1753 const Stmt *S = PS->getStmt();
1754 const auto *DS = dyn_cast<DeclStmt>(S);
1755 const auto *VR = dyn_cast<VarRegion>(R);
1756
1757 if (DS) {
1758 SI.StoreKind = StoreInfo::Initialization;
1759 } else if (isa<BlockExpr>(S)) {
1760 SI.StoreKind = StoreInfo::BlockCapture;
1761 if (VR) {
1762 // See if we can get the BlockVarRegion.
1763 ProgramStateRef State = StoreSite->getState();
1764 SVal V = StoreSite->getSVal(S);
1765 if (const auto *BDR =
1766 dyn_cast_or_null<BlockDataRegion>(V.getAsRegion())) {
1767 if (const VarRegion *OriginalR = BDR->getOriginalRegion(VR)) {
1768 getParentTracker().track(State->getSVal(OriginalR), OriginalR,
1769 Options, OriginSFC);
1770 }
1771 }
1772 }
1773 }
1774 } else if (SI.StoreSite->getLocation().getAs<CallEnter>() &&
1775 isa<VarRegion>(SI.Dest)) {
1776 SI.StoreKind = StoreInfo::CallArgument;
1777 }
1778
1779 return getParentTracker().handle(SI, BRC, Options);
1780 }
1781
1782 //===----------------------------------------------------------------------===//
1783 // Implementation of TrackConstraintBRVisitor.
1784 //===----------------------------------------------------------------------===//
1785
Profile(llvm::FoldingSetNodeID & ID) const1786 void TrackConstraintBRVisitor::Profile(llvm::FoldingSetNodeID &ID) const {
1787 static int tag = 0;
1788 ID.AddPointer(&tag);
1789 ID.AddBoolean(Assumption);
1790 ID.Add(Constraint);
1791 }
1792
1793 /// Return the tag associated with this visitor. This tag will be used
1794 /// to make all PathDiagnosticPieces created by this visitor.
getTag()1795 const char *TrackConstraintBRVisitor::getTag() {
1796 return "TrackConstraintBRVisitor";
1797 }
1798
isUnderconstrained(const ExplodedNode * N) const1799 bool TrackConstraintBRVisitor::isUnderconstrained(const ExplodedNode *N) const {
1800 if (IsZeroCheck)
1801 return N->getState()->isNull(Constraint).isUnderconstrained();
1802 return (bool)N->getState()->assume(Constraint, !Assumption);
1803 }
1804
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport &)1805 PathDiagnosticPieceRef TrackConstraintBRVisitor::VisitNode(
1806 const ExplodedNode *N, BugReporterContext &BRC, PathSensitiveBugReport &) {
1807 const ExplodedNode *PrevN = N->getFirstPred();
1808 if (IsSatisfied)
1809 return nullptr;
1810
1811 // Start tracking after we see the first state in which the value is
1812 // constrained.
1813 if (!IsTrackingTurnedOn)
1814 if (!isUnderconstrained(N))
1815 IsTrackingTurnedOn = true;
1816 if (!IsTrackingTurnedOn)
1817 return nullptr;
1818
1819 // Check if in the previous state it was feasible for this constraint
1820 // to *not* be true.
1821 if (isUnderconstrained(PrevN)) {
1822 IsSatisfied = true;
1823
1824 // At this point, the negation of the constraint should be infeasible. If it
1825 // is feasible, make sure that the negation of the constrainti was
1826 // infeasible in the current state. If it is feasible, we somehow missed
1827 // the transition point.
1828 assert(!isUnderconstrained(N));
1829
1830 // We found the transition point for the constraint. We now need to
1831 // pretty-print the constraint. (work-in-progress)
1832 SmallString<64> sbuf;
1833 llvm::raw_svector_ostream os(sbuf);
1834
1835 if (isa<Loc>(Constraint)) {
1836 os << "Assuming pointer value is ";
1837 os << (Assumption ? "non-null" : "null");
1838 }
1839
1840 if (os.str().empty())
1841 return nullptr;
1842
1843 // Construct a new PathDiagnosticPiece.
1844 ProgramPoint P = N->getLocation();
1845
1846 // If this node already have a specialized note, it's probably better
1847 // than our generic note.
1848 // FIXME: This only looks for note tags, not for other ways to add a note.
1849 if (isa_and_nonnull<NoteTag>(P.getTag()))
1850 return nullptr;
1851
1852 PathDiagnosticLocation L =
1853 PathDiagnosticLocation::create(P, BRC.getSourceManager());
1854 if (!L.isValid())
1855 return nullptr;
1856
1857 auto X = std::make_shared<PathDiagnosticEventPiece>(L, os.str());
1858 X->setTag(getTag());
1859 return std::move(X);
1860 }
1861
1862 return nullptr;
1863 }
1864
1865 //===----------------------------------------------------------------------===//
1866 // Implementation of SuppressInlineDefensiveChecksVisitor.
1867 //===----------------------------------------------------------------------===//
1868
1869 SuppressInlineDefensiveChecksVisitor::
SuppressInlineDefensiveChecksVisitor(DefinedSVal Value,const ExplodedNode * N)1870 SuppressInlineDefensiveChecksVisitor(DefinedSVal Value, const ExplodedNode *N)
1871 : V(Value) {
1872 // Check if the visitor is disabled.
1873 AnalyzerOptions &Options = N->getState()->getAnalysisManager().options;
1874 if (!Options.ShouldSuppressInlinedDefensiveChecks)
1875 IsSatisfied = true;
1876 }
1877
Profile(llvm::FoldingSetNodeID & ID) const1878 void SuppressInlineDefensiveChecksVisitor::Profile(
1879 llvm::FoldingSetNodeID &ID) const {
1880 static int id = 0;
1881 ID.AddPointer(&id);
1882 ID.Add(V);
1883 }
1884
getTag()1885 const char *SuppressInlineDefensiveChecksVisitor::getTag() {
1886 return "IDCVisitor";
1887 }
1888
1889 PathDiagnosticPieceRef
VisitNode(const ExplodedNode * Succ,BugReporterContext & BRC,PathSensitiveBugReport & BR)1890 SuppressInlineDefensiveChecksVisitor::VisitNode(const ExplodedNode *Succ,
1891 BugReporterContext &BRC,
1892 PathSensitiveBugReport &BR) {
1893 const ExplodedNode *Pred = Succ->getFirstPred();
1894 if (IsSatisfied)
1895 return nullptr;
1896
1897 // Start tracking after we see the first state in which the value is null.
1898 if (!IsTrackingTurnedOn)
1899 if (Succ->getState()->isNull(V).isConstrainedTrue())
1900 IsTrackingTurnedOn = true;
1901 if (!IsTrackingTurnedOn)
1902 return nullptr;
1903
1904 // Check if in the previous state it was feasible for this value
1905 // to *not* be null.
1906 if (!Pred->getState()->isNull(V).isConstrainedTrue() &&
1907 Succ->getState()->isNull(V).isConstrainedTrue()) {
1908 IsSatisfied = true;
1909
1910 // Check if this is inlined defensive checks.
1911 const LocationContext *CurLC = Succ->getLocationContext();
1912 const LocationContext *ReportLC = BR.getErrorNode()->getLocationContext();
1913 if (CurLC != ReportLC && !CurLC->isParentOf(ReportLC)) {
1914 BR.markInvalid("Suppress IDC", CurLC);
1915 return nullptr;
1916 }
1917
1918 // Treat defensive checks in function-like macros as if they were an inlined
1919 // defensive check. If the bug location is not in a macro and the
1920 // terminator for the current location is in a macro then suppress the
1921 // warning.
1922 auto BugPoint = BR.getErrorNode()->getLocation().getAs<StmtPoint>();
1923
1924 if (!BugPoint)
1925 return nullptr;
1926
1927 ProgramPoint CurPoint = Succ->getLocation();
1928 const Stmt *CurTerminatorStmt = nullptr;
1929 if (auto BE = CurPoint.getAs<BlockEdge>()) {
1930 CurTerminatorStmt = BE->getSrc()->getTerminator().getStmt();
1931 } else if (auto SP = CurPoint.getAs<StmtPoint>()) {
1932 const Stmt *CurStmt = SP->getStmt();
1933 if (!CurStmt->getBeginLoc().isMacroID())
1934 return nullptr;
1935
1936 CFGStmtMap *Map = CurLC->getAnalysisDeclContext()->getCFGStmtMap();
1937 CurTerminatorStmt = Map->getBlock(CurStmt)->getTerminatorStmt();
1938 } else {
1939 return nullptr;
1940 }
1941
1942 if (!CurTerminatorStmt)
1943 return nullptr;
1944
1945 SourceLocation TerminatorLoc = CurTerminatorStmt->getBeginLoc();
1946 if (TerminatorLoc.isMacroID()) {
1947 SourceLocation BugLoc = BugPoint->getStmt()->getBeginLoc();
1948
1949 // Suppress reports unless we are in that same macro.
1950 if (!BugLoc.isMacroID() ||
1951 getMacroName(BugLoc, BRC) != getMacroName(TerminatorLoc, BRC)) {
1952 BR.markInvalid("Suppress Macro IDC", CurLC);
1953 }
1954 return nullptr;
1955 }
1956 }
1957 return nullptr;
1958 }
1959
1960 //===----------------------------------------------------------------------===//
1961 // TrackControlDependencyCondBRVisitor.
1962 //===----------------------------------------------------------------------===//
1963
1964 namespace {
1965 /// Tracks the expressions that are a control dependency of the node that was
1966 /// supplied to the constructor.
1967 /// For example:
1968 ///
1969 /// cond = 1;
1970 /// if (cond)
1971 /// 10 / 0;
1972 ///
1973 /// An error is emitted at line 3. This visitor realizes that the branch
1974 /// on line 2 is a control dependency of line 3, and tracks it's condition via
1975 /// trackExpressionValue().
1976 class TrackControlDependencyCondBRVisitor final
1977 : public TrackingBugReporterVisitor {
1978 const ExplodedNode *Origin;
1979 ControlDependencyCalculator ControlDeps;
1980 llvm::SmallSet<const CFGBlock *, 32> VisitedBlocks;
1981
1982 public:
TrackControlDependencyCondBRVisitor(TrackerRef ParentTracker,const ExplodedNode * O)1983 TrackControlDependencyCondBRVisitor(TrackerRef ParentTracker,
1984 const ExplodedNode *O)
1985 : TrackingBugReporterVisitor(ParentTracker), Origin(O),
1986 ControlDeps(&O->getCFG()) {}
1987
Profile(llvm::FoldingSetNodeID & ID) const1988 void Profile(llvm::FoldingSetNodeID &ID) const override {
1989 static int x = 0;
1990 ID.AddPointer(&x);
1991 }
1992
1993 PathDiagnosticPieceRef VisitNode(const ExplodedNode *N,
1994 BugReporterContext &BRC,
1995 PathSensitiveBugReport &BR) override;
1996 };
1997 } // end of anonymous namespace
1998
1999 static std::shared_ptr<PathDiagnosticEventPiece>
constructDebugPieceForTrackedCondition(const Expr * Cond,const ExplodedNode * N,BugReporterContext & BRC)2000 constructDebugPieceForTrackedCondition(const Expr *Cond,
2001 const ExplodedNode *N,
2002 BugReporterContext &BRC) {
2003
2004 if (BRC.getAnalyzerOptions().AnalysisDiagOpt == PD_NONE ||
2005 !BRC.getAnalyzerOptions().ShouldTrackConditionsDebug)
2006 return nullptr;
2007
2008 std::string ConditionText = std::string(Lexer::getSourceText(
2009 CharSourceRange::getTokenRange(Cond->getSourceRange()),
2010 BRC.getSourceManager(), BRC.getASTContext().getLangOpts()));
2011
2012 return std::make_shared<PathDiagnosticEventPiece>(
2013 PathDiagnosticLocation::createBegin(
2014 Cond, BRC.getSourceManager(), N->getLocationContext()),
2015 (Twine() + "Tracking condition '" + ConditionText + "'").str());
2016 }
2017
isAssertlikeBlock(const CFGBlock * B,ASTContext & Context)2018 static bool isAssertlikeBlock(const CFGBlock *B, ASTContext &Context) {
2019 if (B->succ_size() != 2)
2020 return false;
2021
2022 const CFGBlock *Then = B->succ_begin()->getReachableBlock();
2023 const CFGBlock *Else = (B->succ_begin() + 1)->getReachableBlock();
2024
2025 if (!Then || !Else)
2026 return false;
2027
2028 if (Then->isInevitablySinking() != Else->isInevitablySinking())
2029 return true;
2030
2031 // For the following condition the following CFG would be built:
2032 //
2033 // ------------->
2034 // / \
2035 // [B1] -> [B2] -> [B3] -> [sink]
2036 // assert(A && B || C); \ \
2037 // -----------> [go on with the execution]
2038 //
2039 // It so happens that CFGBlock::getTerminatorCondition returns 'A' for block
2040 // B1, 'A && B' for B2, and 'A && B || C' for B3. Let's check whether we
2041 // reached the end of the condition!
2042 if (const Stmt *ElseCond = Else->getTerminatorCondition())
2043 if (const auto *BinOp = dyn_cast<BinaryOperator>(ElseCond))
2044 if (BinOp->isLogicalOp())
2045 return isAssertlikeBlock(Else, Context);
2046
2047 return false;
2048 }
2049
2050 PathDiagnosticPieceRef
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)2051 TrackControlDependencyCondBRVisitor::VisitNode(const ExplodedNode *N,
2052 BugReporterContext &BRC,
2053 PathSensitiveBugReport &BR) {
2054 // We can only reason about control dependencies within the same stack frame.
2055 if (Origin->getStackFrame() != N->getStackFrame())
2056 return nullptr;
2057
2058 CFGBlock *NB = const_cast<CFGBlock *>(N->getCFGBlock());
2059
2060 // Skip if we already inspected this block.
2061 if (!VisitedBlocks.insert(NB).second)
2062 return nullptr;
2063
2064 CFGBlock *OriginB = const_cast<CFGBlock *>(Origin->getCFGBlock());
2065
2066 // TODO: Cache CFGBlocks for each ExplodedNode.
2067 if (!OriginB || !NB)
2068 return nullptr;
2069
2070 if (isAssertlikeBlock(NB, BRC.getASTContext()))
2071 return nullptr;
2072
2073 if (ControlDeps.isControlDependent(OriginB, NB)) {
2074 // We don't really want to explain for range loops. Evidence suggests that
2075 // the only thing that leads to is the addition of calls to operator!=.
2076 if (llvm::isa_and_nonnull<CXXForRangeStmt>(NB->getTerminatorStmt()))
2077 return nullptr;
2078
2079 if (const Expr *Condition = NB->getLastCondition()) {
2080
2081 // If we can't retrieve a sensible condition, just bail out.
2082 const Expr *InnerExpr = peelOffOuterExpr(Condition, N);
2083 if (!InnerExpr)
2084 return nullptr;
2085
2086 // If the condition was a function call, we likely won't gain much from
2087 // tracking it either. Evidence suggests that it will mostly trigger in
2088 // scenarios like this:
2089 //
2090 // void f(int *x) {
2091 // x = nullptr;
2092 // if (alwaysTrue()) // We don't need a whole lot of explanation
2093 // // here, the function name is good enough.
2094 // *x = 5;
2095 // }
2096 //
2097 // Its easy to create a counterexample where this heuristic would make us
2098 // lose valuable information, but we've never really seen one in practice.
2099 if (isa<CallExpr>(InnerExpr))
2100 return nullptr;
2101
2102 // Keeping track of the already tracked conditions on a visitor level
2103 // isn't sufficient, because a new visitor is created for each tracked
2104 // expression, hence the BugReport level set.
2105 if (BR.addTrackedCondition(N)) {
2106 getParentTracker().track(InnerExpr, N,
2107 {bugreporter::TrackingKind::Condition,
2108 /*EnableNullFPSuppression=*/false});
2109 return constructDebugPieceForTrackedCondition(Condition, N, BRC);
2110 }
2111 }
2112 }
2113
2114 return nullptr;
2115 }
2116
2117 //===----------------------------------------------------------------------===//
2118 // Implementation of trackExpressionValue.
2119 //===----------------------------------------------------------------------===//
2120
peelOffOuterExpr(const Expr * Ex,const ExplodedNode * N)2121 static const Expr *peelOffOuterExpr(const Expr *Ex, const ExplodedNode *N) {
2122
2123 Ex = Ex->IgnoreParenCasts();
2124 if (const auto *FE = dyn_cast<FullExpr>(Ex))
2125 return peelOffOuterExpr(FE->getSubExpr(), N);
2126 if (const auto *OVE = dyn_cast<OpaqueValueExpr>(Ex))
2127 return peelOffOuterExpr(OVE->getSourceExpr(), N);
2128 if (const auto *POE = dyn_cast<PseudoObjectExpr>(Ex)) {
2129 const auto *PropRef = dyn_cast<ObjCPropertyRefExpr>(POE->getSyntacticForm());
2130 if (PropRef && PropRef->isMessagingGetter()) {
2131 const Expr *GetterMessageSend =
2132 POE->getSemanticExpr(POE->getNumSemanticExprs() - 1);
2133 assert(isa<ObjCMessageExpr>(GetterMessageSend->IgnoreParenCasts()));
2134 return peelOffOuterExpr(GetterMessageSend, N);
2135 }
2136 }
2137
2138 // Peel off the ternary operator.
2139 if (const auto *CO = dyn_cast<ConditionalOperator>(Ex)) {
2140 // Find a node where the branching occurred and find out which branch
2141 // we took (true/false) by looking at the ExplodedGraph.
2142 const ExplodedNode *NI = N;
2143 do {
2144 ProgramPoint ProgPoint = NI->getLocation();
2145 if (std::optional<BlockEdge> BE = ProgPoint.getAs<BlockEdge>()) {
2146 const CFGBlock *srcBlk = BE->getSrc();
2147 if (const Stmt *term = srcBlk->getTerminatorStmt()) {
2148 if (term == CO) {
2149 bool TookTrueBranch = (*(srcBlk->succ_begin()) == BE->getDst());
2150 if (TookTrueBranch)
2151 return peelOffOuterExpr(CO->getTrueExpr(), N);
2152 else
2153 return peelOffOuterExpr(CO->getFalseExpr(), N);
2154 }
2155 }
2156 }
2157 NI = NI->getFirstPred();
2158 } while (NI);
2159 }
2160
2161 if (auto *BO = dyn_cast<BinaryOperator>(Ex))
2162 if (const Expr *SubEx = peelOffPointerArithmetic(BO))
2163 return peelOffOuterExpr(SubEx, N);
2164
2165 if (auto *UO = dyn_cast<UnaryOperator>(Ex)) {
2166 if (UO->getOpcode() == UO_LNot)
2167 return peelOffOuterExpr(UO->getSubExpr(), N);
2168
2169 // FIXME: There's a hack in our Store implementation that always computes
2170 // field offsets around null pointers as if they are always equal to 0.
2171 // The idea here is to report accesses to fields as null dereferences
2172 // even though the pointer value that's being dereferenced is actually
2173 // the offset of the field rather than exactly 0.
2174 // See the FIXME in StoreManager's getLValueFieldOrIvar() method.
2175 // This code interacts heavily with this hack; otherwise the value
2176 // would not be null at all for most fields, so we'd be unable to track it.
2177 if (UO->getOpcode() == UO_AddrOf && UO->getSubExpr()->isLValue())
2178 if (const Expr *DerefEx = bugreporter::getDerefExpr(UO->getSubExpr()))
2179 return peelOffOuterExpr(DerefEx, N);
2180 }
2181
2182 return Ex;
2183 }
2184
2185 /// Find the ExplodedNode where the lvalue (the value of 'Ex')
2186 /// was computed.
findNodeForExpression(const ExplodedNode * N,const Expr * Inner)2187 static const ExplodedNode* findNodeForExpression(const ExplodedNode *N,
2188 const Expr *Inner) {
2189 while (N) {
2190 if (N->getStmtForDiagnostics() == Inner)
2191 return N;
2192 N = N->getFirstPred();
2193 }
2194 return N;
2195 }
2196
2197 //===----------------------------------------------------------------------===//
2198 // Tracker implementation
2199 //===----------------------------------------------------------------------===//
2200
constructNote(StoreInfo SI,BugReporterContext & BRC,StringRef NodeText)2201 PathDiagnosticPieceRef StoreHandler::constructNote(StoreInfo SI,
2202 BugReporterContext &BRC,
2203 StringRef NodeText) {
2204 // Construct a new PathDiagnosticPiece.
2205 ProgramPoint P = SI.StoreSite->getLocation();
2206 PathDiagnosticLocation L;
2207 if (P.getAs<CallEnter>() && SI.SourceOfTheValue)
2208 L = PathDiagnosticLocation(SI.SourceOfTheValue, BRC.getSourceManager(),
2209 P.getLocationContext());
2210
2211 if (!L.isValid() || !L.asLocation().isValid())
2212 L = PathDiagnosticLocation::create(P, BRC.getSourceManager());
2213
2214 if (!L.isValid() || !L.asLocation().isValid())
2215 return nullptr;
2216
2217 return std::make_shared<PathDiagnosticEventPiece>(L, NodeText);
2218 }
2219
2220 namespace {
2221 class DefaultStoreHandler final : public StoreHandler {
2222 public:
2223 using StoreHandler::StoreHandler;
2224
handle(StoreInfo SI,BugReporterContext & BRC,TrackingOptions Opts)2225 PathDiagnosticPieceRef handle(StoreInfo SI, BugReporterContext &BRC,
2226 TrackingOptions Opts) override {
2227 // Okay, we've found the binding. Emit an appropriate message.
2228 SmallString<256> Buffer;
2229 llvm::raw_svector_ostream OS(Buffer);
2230
2231 switch (SI.StoreKind) {
2232 case StoreInfo::Initialization:
2233 case StoreInfo::BlockCapture:
2234 showBRDiagnostics(OS, SI);
2235 break;
2236 case StoreInfo::CallArgument:
2237 showBRParamDiagnostics(OS, SI);
2238 break;
2239 case StoreInfo::Assignment:
2240 showBRDefaultDiagnostics(OS, SI);
2241 break;
2242 }
2243
2244 if (Opts.Kind == bugreporter::TrackingKind::Condition)
2245 OS << WillBeUsedForACondition;
2246
2247 return constructNote(SI, BRC, OS.str());
2248 }
2249 };
2250
2251 class ControlDependencyHandler final : public ExpressionHandler {
2252 public:
2253 using ExpressionHandler::ExpressionHandler;
2254
handle(const Expr * Inner,const ExplodedNode * InputNode,const ExplodedNode * LVNode,TrackingOptions Opts)2255 Tracker::Result handle(const Expr *Inner, const ExplodedNode *InputNode,
2256 const ExplodedNode *LVNode,
2257 TrackingOptions Opts) override {
2258 PathSensitiveBugReport &Report = getParentTracker().getReport();
2259
2260 // We only track expressions if we believe that they are important. Chances
2261 // are good that control dependencies to the tracking point are also
2262 // important because of this, let's explain why we believe control reached
2263 // this point.
2264 // TODO: Shouldn't we track control dependencies of every bug location,
2265 // rather than only tracked expressions?
2266 if (LVNode->getState()
2267 ->getAnalysisManager()
2268 .getAnalyzerOptions()
2269 .ShouldTrackConditions) {
2270 Report.addVisitor<TrackControlDependencyCondBRVisitor>(
2271 &getParentTracker(), InputNode);
2272 return {/*FoundSomethingToTrack=*/true};
2273 }
2274
2275 return {};
2276 }
2277 };
2278
2279 class NilReceiverHandler final : public ExpressionHandler {
2280 public:
2281 using ExpressionHandler::ExpressionHandler;
2282
handle(const Expr * Inner,const ExplodedNode * InputNode,const ExplodedNode * LVNode,TrackingOptions Opts)2283 Tracker::Result handle(const Expr *Inner, const ExplodedNode *InputNode,
2284 const ExplodedNode *LVNode,
2285 TrackingOptions Opts) override {
2286 // The message send could be nil due to the receiver being nil.
2287 // At this point in the path, the receiver should be live since we are at
2288 // the message send expr. If it is nil, start tracking it.
2289 if (const Expr *Receiver =
2290 NilReceiverBRVisitor::getNilReceiver(Inner, LVNode))
2291 return getParentTracker().track(Receiver, LVNode, Opts);
2292
2293 return {};
2294 }
2295 };
2296
2297 class ArrayIndexHandler final : public ExpressionHandler {
2298 public:
2299 using ExpressionHandler::ExpressionHandler;
2300
handle(const Expr * Inner,const ExplodedNode * InputNode,const ExplodedNode * LVNode,TrackingOptions Opts)2301 Tracker::Result handle(const Expr *Inner, const ExplodedNode *InputNode,
2302 const ExplodedNode *LVNode,
2303 TrackingOptions Opts) override {
2304 // Track the index if this is an array subscript.
2305 if (const auto *Arr = dyn_cast<ArraySubscriptExpr>(Inner))
2306 return getParentTracker().track(
2307 Arr->getIdx(), LVNode,
2308 {Opts.Kind, /*EnableNullFPSuppression*/ false});
2309
2310 return {};
2311 }
2312 };
2313
2314 // TODO: extract it into more handlers
2315 class InterestingLValueHandler final : public ExpressionHandler {
2316 public:
2317 using ExpressionHandler::ExpressionHandler;
2318
handle(const Expr * Inner,const ExplodedNode * InputNode,const ExplodedNode * LVNode,TrackingOptions Opts)2319 Tracker::Result handle(const Expr *Inner, const ExplodedNode *InputNode,
2320 const ExplodedNode *LVNode,
2321 TrackingOptions Opts) override {
2322 ProgramStateRef LVState = LVNode->getState();
2323 const StackFrameContext *SFC = LVNode->getStackFrame();
2324 PathSensitiveBugReport &Report = getParentTracker().getReport();
2325 Tracker::Result Result;
2326
2327 // See if the expression we're interested refers to a variable.
2328 // If so, we can track both its contents and constraints on its value.
2329 if (ExplodedGraph::isInterestingLValueExpr(Inner)) {
2330 SVal LVal = LVNode->getSVal(Inner);
2331
2332 const MemRegion *RR = getLocationRegionIfReference(Inner, LVNode);
2333 bool LVIsNull = LVState->isNull(LVal).isConstrainedTrue();
2334
2335 // If this is a C++ reference to a null pointer, we are tracking the
2336 // pointer. In addition, we should find the store at which the reference
2337 // got initialized.
2338 if (RR && !LVIsNull)
2339 Result.combineWith(getParentTracker().track(LVal, RR, Opts, SFC));
2340
2341 // In case of C++ references, we want to differentiate between a null
2342 // reference and reference to null pointer.
2343 // If the LVal is null, check if we are dealing with null reference.
2344 // For those, we want to track the location of the reference.
2345 const MemRegion *R =
2346 (RR && LVIsNull) ? RR : LVNode->getSVal(Inner).getAsRegion();
2347
2348 if (R) {
2349
2350 // Mark both the variable region and its contents as interesting.
2351 SVal V = LVState->getRawSVal(loc::MemRegionVal(R));
2352 Report.addVisitor<NoStoreFuncVisitor>(cast<SubRegion>(R), Opts.Kind);
2353
2354 // When we got here, we do have something to track, and we will
2355 // interrupt.
2356 Result.FoundSomethingToTrack = true;
2357 Result.WasInterrupted = true;
2358
2359 MacroNullReturnSuppressionVisitor::addMacroVisitorIfNecessary(
2360 LVNode, R, Opts.EnableNullFPSuppression, Report, V);
2361
2362 Report.markInteresting(V, Opts.Kind);
2363 Report.addVisitor<UndefOrNullArgVisitor>(R);
2364
2365 // If the contents are symbolic and null, find out when they became
2366 // null.
2367 if (V.getAsLocSymbol(/*IncludeBaseRegions=*/true))
2368 if (LVState->isNull(V).isConstrainedTrue())
2369 Report.addVisitor<TrackConstraintBRVisitor>(V.castAs<DefinedSVal>(),
2370 false);
2371
2372 // Add visitor, which will suppress inline defensive checks.
2373 if (auto DV = V.getAs<DefinedSVal>())
2374 if (!DV->isZeroConstant() && Opts.EnableNullFPSuppression)
2375 // Note that LVNode may be too late (i.e., too far from the
2376 // InputNode) because the lvalue may have been computed before the
2377 // inlined call was evaluated. InputNode may as well be too early
2378 // here, because the symbol is already dead; this, however, is fine
2379 // because we can still find the node in which it collapsed to null
2380 // previously.
2381 Report.addVisitor<SuppressInlineDefensiveChecksVisitor>(*DV,
2382 InputNode);
2383 getParentTracker().track(V, R, Opts, SFC);
2384 }
2385 }
2386
2387 return Result;
2388 }
2389 };
2390
2391 /// Adds a ReturnVisitor if the given statement represents a call that was
2392 /// inlined.
2393 ///
2394 /// This will search back through the ExplodedGraph, starting from the given
2395 /// node, looking for when the given statement was processed. If it turns out
2396 /// the statement is a call that was inlined, we add the visitor to the
2397 /// bug report, so it can print a note later.
2398 class InlinedFunctionCallHandler final : public ExpressionHandler {
2399 using ExpressionHandler::ExpressionHandler;
2400
handle(const Expr * E,const ExplodedNode * InputNode,const ExplodedNode * ExprNode,TrackingOptions Opts)2401 Tracker::Result handle(const Expr *E, const ExplodedNode *InputNode,
2402 const ExplodedNode *ExprNode,
2403 TrackingOptions Opts) override {
2404 if (!CallEvent::isCallStmt(E))
2405 return {};
2406
2407 // First, find when we processed the statement.
2408 // If we work with a 'CXXNewExpr' that is going to be purged away before
2409 // its call take place. We would catch that purge in the last condition
2410 // as a 'StmtPoint' so we have to bypass it.
2411 const bool BypassCXXNewExprEval = isa<CXXNewExpr>(E);
2412
2413 // This is moving forward when we enter into another context.
2414 const StackFrameContext *CurrentSFC = ExprNode->getStackFrame();
2415
2416 do {
2417 // If that is satisfied we found our statement as an inlined call.
2418 if (std::optional<CallExitEnd> CEE =
2419 ExprNode->getLocationAs<CallExitEnd>())
2420 if (CEE->getCalleeContext()->getCallSite() == E)
2421 break;
2422
2423 // Try to move forward to the end of the call-chain.
2424 ExprNode = ExprNode->getFirstPred();
2425 if (!ExprNode)
2426 break;
2427
2428 const StackFrameContext *PredSFC = ExprNode->getStackFrame();
2429
2430 // If that is satisfied we found our statement.
2431 // FIXME: This code currently bypasses the call site for the
2432 // conservatively evaluated allocator.
2433 if (!BypassCXXNewExprEval)
2434 if (std::optional<StmtPoint> SP = ExprNode->getLocationAs<StmtPoint>())
2435 // See if we do not enter into another context.
2436 if (SP->getStmt() == E && CurrentSFC == PredSFC)
2437 break;
2438
2439 CurrentSFC = PredSFC;
2440 } while (ExprNode->getStackFrame() == CurrentSFC);
2441
2442 // Next, step over any post-statement checks.
2443 while (ExprNode && ExprNode->getLocation().getAs<PostStmt>())
2444 ExprNode = ExprNode->getFirstPred();
2445 if (!ExprNode)
2446 return {};
2447
2448 // Finally, see if we inlined the call.
2449 std::optional<CallExitEnd> CEE = ExprNode->getLocationAs<CallExitEnd>();
2450 if (!CEE)
2451 return {};
2452
2453 const StackFrameContext *CalleeContext = CEE->getCalleeContext();
2454 if (CalleeContext->getCallSite() != E)
2455 return {};
2456
2457 // Check the return value.
2458 ProgramStateRef State = ExprNode->getState();
2459 SVal RetVal = ExprNode->getSVal(E);
2460
2461 // Handle cases where a reference is returned and then immediately used.
2462 if (cast<Expr>(E)->isGLValue())
2463 if (std::optional<Loc> LValue = RetVal.getAs<Loc>())
2464 RetVal = State->getSVal(*LValue);
2465
2466 // See if the return value is NULL. If so, suppress the report.
2467 AnalyzerOptions &Options = State->getAnalysisManager().options;
2468
2469 bool EnableNullFPSuppression = false;
2470 if (Opts.EnableNullFPSuppression && Options.ShouldSuppressNullReturnPaths)
2471 if (std::optional<Loc> RetLoc = RetVal.getAs<Loc>())
2472 EnableNullFPSuppression = State->isNull(*RetLoc).isConstrainedTrue();
2473
2474 PathSensitiveBugReport &Report = getParentTracker().getReport();
2475 Report.addVisitor<ReturnVisitor>(&getParentTracker(), CalleeContext,
2476 EnableNullFPSuppression, Options,
2477 Opts.Kind);
2478 return {true};
2479 }
2480 };
2481
2482 class DefaultExpressionHandler final : public ExpressionHandler {
2483 public:
2484 using ExpressionHandler::ExpressionHandler;
2485
handle(const Expr * Inner,const ExplodedNode * InputNode,const ExplodedNode * LVNode,TrackingOptions Opts)2486 Tracker::Result handle(const Expr *Inner, const ExplodedNode *InputNode,
2487 const ExplodedNode *LVNode,
2488 TrackingOptions Opts) override {
2489 ProgramStateRef LVState = LVNode->getState();
2490 const StackFrameContext *SFC = LVNode->getStackFrame();
2491 PathSensitiveBugReport &Report = getParentTracker().getReport();
2492 Tracker::Result Result;
2493
2494 // If the expression is not an "lvalue expression", we can still
2495 // track the constraints on its contents.
2496 SVal V = LVState->getSValAsScalarOrLoc(Inner, LVNode->getLocationContext());
2497
2498 // Is it a symbolic value?
2499 if (auto L = V.getAs<loc::MemRegionVal>()) {
2500 // FIXME: this is a hack for fixing a later crash when attempting to
2501 // dereference a void* pointer.
2502 // We should not try to dereference pointers at all when we don't care
2503 // what is written inside the pointer.
2504 bool CanDereference = true;
2505 if (const auto *SR = L->getRegionAs<SymbolicRegion>()) {
2506 if (SR->getPointeeStaticType()->isVoidType())
2507 CanDereference = false;
2508 } else if (L->getRegionAs<AllocaRegion>())
2509 CanDereference = false;
2510
2511 // At this point we are dealing with the region's LValue.
2512 // However, if the rvalue is a symbolic region, we should track it as
2513 // well. Try to use the correct type when looking up the value.
2514 SVal RVal;
2515 if (ExplodedGraph::isInterestingLValueExpr(Inner))
2516 RVal = LVState->getRawSVal(*L, Inner->getType());
2517 else if (CanDereference)
2518 RVal = LVState->getSVal(L->getRegion());
2519
2520 if (CanDereference) {
2521 Report.addVisitor<UndefOrNullArgVisitor>(L->getRegion());
2522 Result.FoundSomethingToTrack = true;
2523
2524 if (auto KV = RVal.getAs<KnownSVal>())
2525 Result.combineWith(
2526 getParentTracker().track(*KV, L->getRegion(), Opts, SFC));
2527 }
2528
2529 const MemRegion *RegionRVal = RVal.getAsRegion();
2530 if (isa_and_nonnull<SymbolicRegion>(RegionRVal)) {
2531 Report.markInteresting(RegionRVal, Opts.Kind);
2532 Report.addVisitor<TrackConstraintBRVisitor>(
2533 loc::MemRegionVal(RegionRVal),
2534 /*assumption=*/false);
2535 Result.FoundSomethingToTrack = true;
2536 }
2537 }
2538
2539 return Result;
2540 }
2541 };
2542
2543 /// Attempts to add visitors to track an RValue expression back to its point of
2544 /// origin.
2545 class PRValueHandler final : public ExpressionHandler {
2546 public:
2547 using ExpressionHandler::ExpressionHandler;
2548
handle(const Expr * E,const ExplodedNode * InputNode,const ExplodedNode * ExprNode,TrackingOptions Opts)2549 Tracker::Result handle(const Expr *E, const ExplodedNode *InputNode,
2550 const ExplodedNode *ExprNode,
2551 TrackingOptions Opts) override {
2552 if (!E->isPRValue())
2553 return {};
2554
2555 const ExplodedNode *RVNode = findNodeForExpression(ExprNode, E);
2556 if (!RVNode)
2557 return {};
2558
2559 Tracker::Result CombinedResult;
2560 Tracker &Parent = getParentTracker();
2561
2562 const auto track = [&CombinedResult, &Parent, ExprNode,
2563 Opts](const Expr *Inner) {
2564 CombinedResult.combineWith(Parent.track(Inner, ExprNode, Opts));
2565 };
2566
2567 // FIXME: Initializer lists can appear in many different contexts
2568 // and most of them needs a special handling. For now let's handle
2569 // what we can. If the initializer list only has 1 element, we track
2570 // that.
2571 // This snippet even handles nesting, e.g.: int *x{{{{{y}}}}};
2572 if (const auto *ILE = dyn_cast<InitListExpr>(E)) {
2573 if (ILE->getNumInits() == 1) {
2574 track(ILE->getInit(0));
2575
2576 return CombinedResult;
2577 }
2578
2579 return {};
2580 }
2581
2582 ProgramStateRef RVState = RVNode->getState();
2583 SVal V = RVState->getSValAsScalarOrLoc(E, RVNode->getLocationContext());
2584 const auto *BO = dyn_cast<BinaryOperator>(E);
2585
2586 if (!BO || !BO->isMultiplicativeOp() || !V.isZeroConstant())
2587 return {};
2588
2589 SVal RHSV = RVState->getSVal(BO->getRHS(), RVNode->getLocationContext());
2590 SVal LHSV = RVState->getSVal(BO->getLHS(), RVNode->getLocationContext());
2591
2592 // Track both LHS and RHS of a multiplication.
2593 if (BO->getOpcode() == BO_Mul) {
2594 if (LHSV.isZeroConstant())
2595 track(BO->getLHS());
2596 if (RHSV.isZeroConstant())
2597 track(BO->getRHS());
2598 } else { // Track only the LHS of a division or a modulo.
2599 if (LHSV.isZeroConstant())
2600 track(BO->getLHS());
2601 }
2602
2603 return CombinedResult;
2604 }
2605 };
2606 } // namespace
2607
Tracker(PathSensitiveBugReport & Report)2608 Tracker::Tracker(PathSensitiveBugReport &Report) : Report(Report) {
2609 // Default expression handlers.
2610 addLowPriorityHandler<ControlDependencyHandler>();
2611 addLowPriorityHandler<NilReceiverHandler>();
2612 addLowPriorityHandler<ArrayIndexHandler>();
2613 addLowPriorityHandler<InterestingLValueHandler>();
2614 addLowPriorityHandler<InlinedFunctionCallHandler>();
2615 addLowPriorityHandler<DefaultExpressionHandler>();
2616 addLowPriorityHandler<PRValueHandler>();
2617 // Default store handlers.
2618 addHighPriorityHandler<DefaultStoreHandler>();
2619 }
2620
track(const Expr * E,const ExplodedNode * N,TrackingOptions Opts)2621 Tracker::Result Tracker::track(const Expr *E, const ExplodedNode *N,
2622 TrackingOptions Opts) {
2623 if (!E || !N)
2624 return {};
2625
2626 const Expr *Inner = peelOffOuterExpr(E, N);
2627 const ExplodedNode *LVNode = findNodeForExpression(N, Inner);
2628 if (!LVNode)
2629 return {};
2630
2631 Result CombinedResult;
2632 // Iterate through the handlers in the order according to their priorities.
2633 for (ExpressionHandlerPtr &Handler : ExpressionHandlers) {
2634 CombinedResult.combineWith(Handler->handle(Inner, N, LVNode, Opts));
2635 if (CombinedResult.WasInterrupted) {
2636 // There is no need to confuse our users here.
2637 // We got interrupted, but our users don't need to know about it.
2638 CombinedResult.WasInterrupted = false;
2639 break;
2640 }
2641 }
2642
2643 return CombinedResult;
2644 }
2645
track(SVal V,const MemRegion * R,TrackingOptions Opts,const StackFrameContext * Origin)2646 Tracker::Result Tracker::track(SVal V, const MemRegion *R, TrackingOptions Opts,
2647 const StackFrameContext *Origin) {
2648 if (auto KV = V.getAs<KnownSVal>()) {
2649 Report.addVisitor<StoreSiteFinder>(this, *KV, R, Opts, Origin);
2650 return {true};
2651 }
2652 return {};
2653 }
2654
handle(StoreInfo SI,BugReporterContext & BRC,TrackingOptions Opts)2655 PathDiagnosticPieceRef Tracker::handle(StoreInfo SI, BugReporterContext &BRC,
2656 TrackingOptions Opts) {
2657 // Iterate through the handlers in the order according to their priorities.
2658 for (StoreHandlerPtr &Handler : StoreHandlers) {
2659 if (PathDiagnosticPieceRef Result = Handler->handle(SI, BRC, Opts))
2660 // If the handler produced a non-null piece, return it.
2661 // There is no need in asking other handlers.
2662 return Result;
2663 }
2664 return {};
2665 }
2666
trackExpressionValue(const ExplodedNode * InputNode,const Expr * E,PathSensitiveBugReport & Report,TrackingOptions Opts)2667 bool bugreporter::trackExpressionValue(const ExplodedNode *InputNode,
2668 const Expr *E,
2669
2670 PathSensitiveBugReport &Report,
2671 TrackingOptions Opts) {
2672 return Tracker::create(Report)
2673 ->track(E, InputNode, Opts)
2674 .FoundSomethingToTrack;
2675 }
2676
trackStoredValue(KnownSVal V,const MemRegion * R,PathSensitiveBugReport & Report,TrackingOptions Opts,const StackFrameContext * Origin)2677 void bugreporter::trackStoredValue(KnownSVal V, const MemRegion *R,
2678 PathSensitiveBugReport &Report,
2679 TrackingOptions Opts,
2680 const StackFrameContext *Origin) {
2681 Tracker::create(Report)->track(V, R, Opts, Origin);
2682 }
2683
2684 //===----------------------------------------------------------------------===//
2685 // Implementation of NulReceiverBRVisitor.
2686 //===----------------------------------------------------------------------===//
2687
getNilReceiver(const Stmt * S,const ExplodedNode * N)2688 const Expr *NilReceiverBRVisitor::getNilReceiver(const Stmt *S,
2689 const ExplodedNode *N) {
2690 const auto *ME = dyn_cast<ObjCMessageExpr>(S);
2691 if (!ME)
2692 return nullptr;
2693 if (const Expr *Receiver = ME->getInstanceReceiver()) {
2694 ProgramStateRef state = N->getState();
2695 SVal V = N->getSVal(Receiver);
2696 if (state->isNull(V).isConstrainedTrue())
2697 return Receiver;
2698 }
2699 return nullptr;
2700 }
2701
2702 PathDiagnosticPieceRef
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)2703 NilReceiverBRVisitor::VisitNode(const ExplodedNode *N, BugReporterContext &BRC,
2704 PathSensitiveBugReport &BR) {
2705 std::optional<PreStmt> P = N->getLocationAs<PreStmt>();
2706 if (!P)
2707 return nullptr;
2708
2709 const Stmt *S = P->getStmt();
2710 const Expr *Receiver = getNilReceiver(S, N);
2711 if (!Receiver)
2712 return nullptr;
2713
2714 llvm::SmallString<256> Buf;
2715 llvm::raw_svector_ostream OS(Buf);
2716
2717 if (const auto *ME = dyn_cast<ObjCMessageExpr>(S)) {
2718 OS << "'";
2719 ME->getSelector().print(OS);
2720 OS << "' not called";
2721 }
2722 else {
2723 OS << "No method is called";
2724 }
2725 OS << " because the receiver is nil";
2726
2727 // The receiver was nil, and hence the method was skipped.
2728 // Register a BugReporterVisitor to issue a message telling us how
2729 // the receiver was null.
2730 bugreporter::trackExpressionValue(N, Receiver, BR,
2731 {bugreporter::TrackingKind::Thorough,
2732 /*EnableNullFPSuppression*/ false});
2733 // Issue a message saying that the method was skipped.
2734 PathDiagnosticLocation L(Receiver, BRC.getSourceManager(),
2735 N->getLocationContext());
2736 return std::make_shared<PathDiagnosticEventPiece>(L, OS.str());
2737 }
2738
2739 //===----------------------------------------------------------------------===//
2740 // Visitor that tries to report interesting diagnostics from conditions.
2741 //===----------------------------------------------------------------------===//
2742
2743 /// Return the tag associated with this visitor. This tag will be used
2744 /// to make all PathDiagnosticPieces created by this visitor.
getTag()2745 const char *ConditionBRVisitor::getTag() { return "ConditionBRVisitor"; }
2746
2747 PathDiagnosticPieceRef
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)2748 ConditionBRVisitor::VisitNode(const ExplodedNode *N, BugReporterContext &BRC,
2749 PathSensitiveBugReport &BR) {
2750 auto piece = VisitNodeImpl(N, BRC, BR);
2751 if (piece) {
2752 piece->setTag(getTag());
2753 if (auto *ev = dyn_cast<PathDiagnosticEventPiece>(piece.get()))
2754 ev->setPrunable(true, /* override */ false);
2755 }
2756 return piece;
2757 }
2758
2759 PathDiagnosticPieceRef
VisitNodeImpl(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)2760 ConditionBRVisitor::VisitNodeImpl(const ExplodedNode *N,
2761 BugReporterContext &BRC,
2762 PathSensitiveBugReport &BR) {
2763 ProgramPoint ProgPoint = N->getLocation();
2764 const std::pair<const ProgramPointTag *, const ProgramPointTag *> &Tags =
2765 ExprEngine::geteagerlyAssumeBinOpBifurcationTags();
2766
2767 // If an assumption was made on a branch, it should be caught
2768 // here by looking at the state transition.
2769 if (std::optional<BlockEdge> BE = ProgPoint.getAs<BlockEdge>()) {
2770 const CFGBlock *SrcBlock = BE->getSrc();
2771 if (const Stmt *Term = SrcBlock->getTerminatorStmt()) {
2772 // If the tag of the previous node is 'Eagerly Assume...' the current
2773 // 'BlockEdge' has the same constraint information. We do not want to
2774 // report the value as it is just an assumption on the predecessor node
2775 // which will be caught in the next VisitNode() iteration as a 'PostStmt'.
2776 const ProgramPointTag *PreviousNodeTag =
2777 N->getFirstPred()->getLocation().getTag();
2778 if (PreviousNodeTag == Tags.first || PreviousNodeTag == Tags.second)
2779 return nullptr;
2780
2781 return VisitTerminator(Term, N, SrcBlock, BE->getDst(), BR, BRC);
2782 }
2783 return nullptr;
2784 }
2785
2786 if (std::optional<PostStmt> PS = ProgPoint.getAs<PostStmt>()) {
2787 const ProgramPointTag *CurrentNodeTag = PS->getTag();
2788 if (CurrentNodeTag != Tags.first && CurrentNodeTag != Tags.second)
2789 return nullptr;
2790
2791 bool TookTrue = CurrentNodeTag == Tags.first;
2792 return VisitTrueTest(cast<Expr>(PS->getStmt()), BRC, BR, N, TookTrue);
2793 }
2794
2795 return nullptr;
2796 }
2797
VisitTerminator(const Stmt * Term,const ExplodedNode * N,const CFGBlock * srcBlk,const CFGBlock * dstBlk,PathSensitiveBugReport & R,BugReporterContext & BRC)2798 PathDiagnosticPieceRef ConditionBRVisitor::VisitTerminator(
2799 const Stmt *Term, const ExplodedNode *N, const CFGBlock *srcBlk,
2800 const CFGBlock *dstBlk, PathSensitiveBugReport &R,
2801 BugReporterContext &BRC) {
2802 const Expr *Cond = nullptr;
2803
2804 // In the code below, Term is a CFG terminator and Cond is a branch condition
2805 // expression upon which the decision is made on this terminator.
2806 //
2807 // For example, in "if (x == 0)", the "if (x == 0)" statement is a terminator,
2808 // and "x == 0" is the respective condition.
2809 //
2810 // Another example: in "if (x && y)", we've got two terminators and two
2811 // conditions due to short-circuit nature of operator "&&":
2812 // 1. The "if (x && y)" statement is a terminator,
2813 // and "y" is the respective condition.
2814 // 2. Also "x && ..." is another terminator,
2815 // and "x" is its condition.
2816
2817 switch (Term->getStmtClass()) {
2818 // FIXME: Stmt::SwitchStmtClass is worth handling, however it is a bit
2819 // more tricky because there are more than two branches to account for.
2820 default:
2821 return nullptr;
2822 case Stmt::IfStmtClass:
2823 Cond = cast<IfStmt>(Term)->getCond();
2824 break;
2825 case Stmt::ConditionalOperatorClass:
2826 Cond = cast<ConditionalOperator>(Term)->getCond();
2827 break;
2828 case Stmt::BinaryOperatorClass:
2829 // When we encounter a logical operator (&& or ||) as a CFG terminator,
2830 // then the condition is actually its LHS; otherwise, we'd encounter
2831 // the parent, such as if-statement, as a terminator.
2832 const auto *BO = cast<BinaryOperator>(Term);
2833 assert(BO->isLogicalOp() &&
2834 "CFG terminator is not a short-circuit operator!");
2835 Cond = BO->getLHS();
2836 break;
2837 }
2838
2839 Cond = Cond->IgnoreParens();
2840
2841 // However, when we encounter a logical operator as a branch condition,
2842 // then the condition is actually its RHS, because LHS would be
2843 // the condition for the logical operator terminator.
2844 while (const auto *InnerBO = dyn_cast<BinaryOperator>(Cond)) {
2845 if (!InnerBO->isLogicalOp())
2846 break;
2847 Cond = InnerBO->getRHS()->IgnoreParens();
2848 }
2849
2850 assert(Cond);
2851 assert(srcBlk->succ_size() == 2);
2852 const bool TookTrue = *(srcBlk->succ_begin()) == dstBlk;
2853 return VisitTrueTest(Cond, BRC, R, N, TookTrue);
2854 }
2855
2856 PathDiagnosticPieceRef
VisitTrueTest(const Expr * Cond,BugReporterContext & BRC,PathSensitiveBugReport & R,const ExplodedNode * N,bool TookTrue)2857 ConditionBRVisitor::VisitTrueTest(const Expr *Cond, BugReporterContext &BRC,
2858 PathSensitiveBugReport &R,
2859 const ExplodedNode *N, bool TookTrue) {
2860 ProgramStateRef CurrentState = N->getState();
2861 ProgramStateRef PrevState = N->getFirstPred()->getState();
2862 const LocationContext *LCtx = N->getLocationContext();
2863
2864 // If the constraint information is changed between the current and the
2865 // previous program state we assuming the newly seen constraint information.
2866 // If we cannot evaluate the condition (and the constraints are the same)
2867 // the analyzer has no information about the value and just assuming it.
2868 bool IsAssuming =
2869 !BRC.getStateManager().haveEqualConstraints(CurrentState, PrevState) ||
2870 CurrentState->getSVal(Cond, LCtx).isUnknownOrUndef();
2871
2872 // These will be modified in code below, but we need to preserve the original
2873 // values in case we want to throw the generic message.
2874 const Expr *CondTmp = Cond;
2875 bool TookTrueTmp = TookTrue;
2876
2877 while (true) {
2878 CondTmp = CondTmp->IgnoreParenCasts();
2879 switch (CondTmp->getStmtClass()) {
2880 default:
2881 break;
2882 case Stmt::BinaryOperatorClass:
2883 if (auto P = VisitTrueTest(Cond, cast<BinaryOperator>(CondTmp),
2884 BRC, R, N, TookTrueTmp, IsAssuming))
2885 return P;
2886 break;
2887 case Stmt::DeclRefExprClass:
2888 if (auto P = VisitTrueTest(Cond, cast<DeclRefExpr>(CondTmp),
2889 BRC, R, N, TookTrueTmp, IsAssuming))
2890 return P;
2891 break;
2892 case Stmt::MemberExprClass:
2893 if (auto P = VisitTrueTest(Cond, cast<MemberExpr>(CondTmp),
2894 BRC, R, N, TookTrueTmp, IsAssuming))
2895 return P;
2896 break;
2897 case Stmt::UnaryOperatorClass: {
2898 const auto *UO = cast<UnaryOperator>(CondTmp);
2899 if (UO->getOpcode() == UO_LNot) {
2900 TookTrueTmp = !TookTrueTmp;
2901 CondTmp = UO->getSubExpr();
2902 continue;
2903 }
2904 break;
2905 }
2906 }
2907 break;
2908 }
2909
2910 // Condition too complex to explain? Just say something so that the user
2911 // knew we've made some path decision at this point.
2912 // If it is too complex and we know the evaluation of the condition do not
2913 // repeat the note from 'BugReporter.cpp'
2914 if (!IsAssuming)
2915 return nullptr;
2916
2917 PathDiagnosticLocation Loc(Cond, BRC.getSourceManager(), LCtx);
2918 if (!Loc.isValid() || !Loc.asLocation().isValid())
2919 return nullptr;
2920
2921 return std::make_shared<PathDiagnosticEventPiece>(
2922 Loc, TookTrue ? GenericTrueMessage : GenericFalseMessage);
2923 }
2924
patternMatch(const Expr * Ex,const Expr * ParentEx,raw_ostream & Out,BugReporterContext & BRC,PathSensitiveBugReport & report,const ExplodedNode * N,std::optional<bool> & prunable,bool IsSameFieldName)2925 bool ConditionBRVisitor::patternMatch(const Expr *Ex, const Expr *ParentEx,
2926 raw_ostream &Out, BugReporterContext &BRC,
2927 PathSensitiveBugReport &report,
2928 const ExplodedNode *N,
2929 std::optional<bool> &prunable,
2930 bool IsSameFieldName) {
2931 const Expr *OriginalExpr = Ex;
2932 Ex = Ex->IgnoreParenCasts();
2933
2934 if (isa<GNUNullExpr, ObjCBoolLiteralExpr, CXXBoolLiteralExpr, IntegerLiteral,
2935 FloatingLiteral>(Ex)) {
2936 // Use heuristics to determine if the expression is a macro
2937 // expanding to a literal and if so, use the macro's name.
2938 SourceLocation BeginLoc = OriginalExpr->getBeginLoc();
2939 SourceLocation EndLoc = OriginalExpr->getEndLoc();
2940 if (BeginLoc.isMacroID() && EndLoc.isMacroID()) {
2941 const SourceManager &SM = BRC.getSourceManager();
2942 const LangOptions &LO = BRC.getASTContext().getLangOpts();
2943 if (Lexer::isAtStartOfMacroExpansion(BeginLoc, SM, LO) &&
2944 Lexer::isAtEndOfMacroExpansion(EndLoc, SM, LO)) {
2945 CharSourceRange R = Lexer::getAsCharRange({BeginLoc, EndLoc}, SM, LO);
2946 Out << Lexer::getSourceText(R, SM, LO);
2947 return false;
2948 }
2949 }
2950 }
2951
2952 if (const auto *DR = dyn_cast<DeclRefExpr>(Ex)) {
2953 const bool quotes = isa<VarDecl>(DR->getDecl());
2954 if (quotes) {
2955 Out << '\'';
2956 const LocationContext *LCtx = N->getLocationContext();
2957 const ProgramState *state = N->getState().get();
2958 if (const MemRegion *R = state->getLValue(cast<VarDecl>(DR->getDecl()),
2959 LCtx).getAsRegion()) {
2960 if (report.isInteresting(R))
2961 prunable = false;
2962 else {
2963 const ProgramState *state = N->getState().get();
2964 SVal V = state->getSVal(R);
2965 if (report.isInteresting(V))
2966 prunable = false;
2967 }
2968 }
2969 }
2970 Out << DR->getDecl()->getDeclName().getAsString();
2971 if (quotes)
2972 Out << '\'';
2973 return quotes;
2974 }
2975
2976 if (const auto *IL = dyn_cast<IntegerLiteral>(Ex)) {
2977 QualType OriginalTy = OriginalExpr->getType();
2978 if (OriginalTy->isPointerType()) {
2979 if (IL->getValue() == 0) {
2980 Out << "null";
2981 return false;
2982 }
2983 }
2984 else if (OriginalTy->isObjCObjectPointerType()) {
2985 if (IL->getValue() == 0) {
2986 Out << "nil";
2987 return false;
2988 }
2989 }
2990
2991 Out << IL->getValue();
2992 return false;
2993 }
2994
2995 if (const auto *ME = dyn_cast<MemberExpr>(Ex)) {
2996 if (!IsSameFieldName)
2997 Out << "field '" << ME->getMemberDecl()->getName() << '\'';
2998 else
2999 Out << '\''
3000 << Lexer::getSourceText(
3001 CharSourceRange::getTokenRange(Ex->getSourceRange()),
3002 BRC.getSourceManager(), BRC.getASTContext().getLangOpts(),
3003 nullptr)
3004 << '\'';
3005 }
3006
3007 return false;
3008 }
3009
VisitTrueTest(const Expr * Cond,const BinaryOperator * BExpr,BugReporterContext & BRC,PathSensitiveBugReport & R,const ExplodedNode * N,bool TookTrue,bool IsAssuming)3010 PathDiagnosticPieceRef ConditionBRVisitor::VisitTrueTest(
3011 const Expr *Cond, const BinaryOperator *BExpr, BugReporterContext &BRC,
3012 PathSensitiveBugReport &R, const ExplodedNode *N, bool TookTrue,
3013 bool IsAssuming) {
3014 bool shouldInvert = false;
3015 std::optional<bool> shouldPrune;
3016
3017 // Check if the field name of the MemberExprs is ambiguous. Example:
3018 // " 'a.d' is equal to 'h.d' " in 'test/Analysis/null-deref-path-notes.cpp'.
3019 bool IsSameFieldName = false;
3020 const auto *LhsME = dyn_cast<MemberExpr>(BExpr->getLHS()->IgnoreParenCasts());
3021 const auto *RhsME = dyn_cast<MemberExpr>(BExpr->getRHS()->IgnoreParenCasts());
3022
3023 if (LhsME && RhsME)
3024 IsSameFieldName =
3025 LhsME->getMemberDecl()->getName() == RhsME->getMemberDecl()->getName();
3026
3027 SmallString<128> LhsString, RhsString;
3028 {
3029 llvm::raw_svector_ostream OutLHS(LhsString), OutRHS(RhsString);
3030 const bool isVarLHS = patternMatch(BExpr->getLHS(), BExpr, OutLHS, BRC, R,
3031 N, shouldPrune, IsSameFieldName);
3032 const bool isVarRHS = patternMatch(BExpr->getRHS(), BExpr, OutRHS, BRC, R,
3033 N, shouldPrune, IsSameFieldName);
3034
3035 shouldInvert = !isVarLHS && isVarRHS;
3036 }
3037
3038 BinaryOperator::Opcode Op = BExpr->getOpcode();
3039
3040 if (BinaryOperator::isAssignmentOp(Op)) {
3041 // For assignment operators, all that we care about is that the LHS
3042 // evaluates to "true" or "false".
3043 return VisitConditionVariable(LhsString, BExpr->getLHS(), BRC, R, N,
3044 TookTrue);
3045 }
3046
3047 // For non-assignment operations, we require that we can understand
3048 // both the LHS and RHS.
3049 if (LhsString.empty() || RhsString.empty() ||
3050 !BinaryOperator::isComparisonOp(Op) || Op == BO_Cmp)
3051 return nullptr;
3052
3053 // Should we invert the strings if the LHS is not a variable name?
3054 SmallString<256> buf;
3055 llvm::raw_svector_ostream Out(buf);
3056 Out << (IsAssuming ? "Assuming " : "")
3057 << (shouldInvert ? RhsString : LhsString) << " is ";
3058
3059 // Do we need to invert the opcode?
3060 if (shouldInvert)
3061 switch (Op) {
3062 default: break;
3063 case BO_LT: Op = BO_GT; break;
3064 case BO_GT: Op = BO_LT; break;
3065 case BO_LE: Op = BO_GE; break;
3066 case BO_GE: Op = BO_LE; break;
3067 }
3068
3069 if (!TookTrue)
3070 switch (Op) {
3071 case BO_EQ: Op = BO_NE; break;
3072 case BO_NE: Op = BO_EQ; break;
3073 case BO_LT: Op = BO_GE; break;
3074 case BO_GT: Op = BO_LE; break;
3075 case BO_LE: Op = BO_GT; break;
3076 case BO_GE: Op = BO_LT; break;
3077 default:
3078 return nullptr;
3079 }
3080
3081 switch (Op) {
3082 case BO_EQ:
3083 Out << "equal to ";
3084 break;
3085 case BO_NE:
3086 Out << "not equal to ";
3087 break;
3088 default:
3089 Out << BinaryOperator::getOpcodeStr(Op) << ' ';
3090 break;
3091 }
3092
3093 Out << (shouldInvert ? LhsString : RhsString);
3094 const LocationContext *LCtx = N->getLocationContext();
3095 const SourceManager &SM = BRC.getSourceManager();
3096
3097 if (isVarAnInterestingCondition(BExpr->getLHS(), N, &R) ||
3098 isVarAnInterestingCondition(BExpr->getRHS(), N, &R))
3099 Out << WillBeUsedForACondition;
3100
3101 // Convert 'field ...' to 'Field ...' if it is a MemberExpr.
3102 std::string Message = std::string(Out.str());
3103 Message[0] = toupper(Message[0]);
3104
3105 // If we know the value create a pop-up note to the value part of 'BExpr'.
3106 if (!IsAssuming) {
3107 PathDiagnosticLocation Loc;
3108 if (!shouldInvert) {
3109 if (LhsME && LhsME->getMemberLoc().isValid())
3110 Loc = PathDiagnosticLocation(LhsME->getMemberLoc(), SM);
3111 else
3112 Loc = PathDiagnosticLocation(BExpr->getLHS(), SM, LCtx);
3113 } else {
3114 if (RhsME && RhsME->getMemberLoc().isValid())
3115 Loc = PathDiagnosticLocation(RhsME->getMemberLoc(), SM);
3116 else
3117 Loc = PathDiagnosticLocation(BExpr->getRHS(), SM, LCtx);
3118 }
3119
3120 return std::make_shared<PathDiagnosticPopUpPiece>(Loc, Message);
3121 }
3122
3123 PathDiagnosticLocation Loc(Cond, SM, LCtx);
3124 auto event = std::make_shared<PathDiagnosticEventPiece>(Loc, Message);
3125 if (shouldPrune)
3126 event->setPrunable(*shouldPrune);
3127 return event;
3128 }
3129
VisitConditionVariable(StringRef LhsString,const Expr * CondVarExpr,BugReporterContext & BRC,PathSensitiveBugReport & report,const ExplodedNode * N,bool TookTrue)3130 PathDiagnosticPieceRef ConditionBRVisitor::VisitConditionVariable(
3131 StringRef LhsString, const Expr *CondVarExpr, BugReporterContext &BRC,
3132 PathSensitiveBugReport &report, const ExplodedNode *N, bool TookTrue) {
3133 // FIXME: If there's already a constraint tracker for this variable,
3134 // we shouldn't emit anything here (c.f. the double note in
3135 // test/Analysis/inlining/path-notes.c)
3136 SmallString<256> buf;
3137 llvm::raw_svector_ostream Out(buf);
3138 Out << "Assuming " << LhsString << " is ";
3139
3140 if (!printValue(CondVarExpr, Out, N, TookTrue, /*IsAssuming=*/true))
3141 return nullptr;
3142
3143 const LocationContext *LCtx = N->getLocationContext();
3144 PathDiagnosticLocation Loc(CondVarExpr, BRC.getSourceManager(), LCtx);
3145
3146 if (isVarAnInterestingCondition(CondVarExpr, N, &report))
3147 Out << WillBeUsedForACondition;
3148
3149 auto event = std::make_shared<PathDiagnosticEventPiece>(Loc, Out.str());
3150
3151 if (isInterestingExpr(CondVarExpr, N, &report))
3152 event->setPrunable(false);
3153
3154 return event;
3155 }
3156
VisitTrueTest(const Expr * Cond,const DeclRefExpr * DRE,BugReporterContext & BRC,PathSensitiveBugReport & report,const ExplodedNode * N,bool TookTrue,bool IsAssuming)3157 PathDiagnosticPieceRef ConditionBRVisitor::VisitTrueTest(
3158 const Expr *Cond, const DeclRefExpr *DRE, BugReporterContext &BRC,
3159 PathSensitiveBugReport &report, const ExplodedNode *N, bool TookTrue,
3160 bool IsAssuming) {
3161 const auto *VD = dyn_cast<VarDecl>(DRE->getDecl());
3162 if (!VD)
3163 return nullptr;
3164
3165 SmallString<256> Buf;
3166 llvm::raw_svector_ostream Out(Buf);
3167
3168 Out << (IsAssuming ? "Assuming '" : "'") << VD->getDeclName() << "' is ";
3169
3170 if (!printValue(DRE, Out, N, TookTrue, IsAssuming))
3171 return nullptr;
3172
3173 const LocationContext *LCtx = N->getLocationContext();
3174
3175 if (isVarAnInterestingCondition(DRE, N, &report))
3176 Out << WillBeUsedForACondition;
3177
3178 // If we know the value create a pop-up note to the 'DRE'.
3179 if (!IsAssuming) {
3180 PathDiagnosticLocation Loc(DRE, BRC.getSourceManager(), LCtx);
3181 return std::make_shared<PathDiagnosticPopUpPiece>(Loc, Out.str());
3182 }
3183
3184 PathDiagnosticLocation Loc(Cond, BRC.getSourceManager(), LCtx);
3185 auto event = std::make_shared<PathDiagnosticEventPiece>(Loc, Out.str());
3186
3187 if (isInterestingExpr(DRE, N, &report))
3188 event->setPrunable(false);
3189
3190 return std::move(event);
3191 }
3192
VisitTrueTest(const Expr * Cond,const MemberExpr * ME,BugReporterContext & BRC,PathSensitiveBugReport & report,const ExplodedNode * N,bool TookTrue,bool IsAssuming)3193 PathDiagnosticPieceRef ConditionBRVisitor::VisitTrueTest(
3194 const Expr *Cond, const MemberExpr *ME, BugReporterContext &BRC,
3195 PathSensitiveBugReport &report, const ExplodedNode *N, bool TookTrue,
3196 bool IsAssuming) {
3197 SmallString<256> Buf;
3198 llvm::raw_svector_ostream Out(Buf);
3199
3200 Out << (IsAssuming ? "Assuming field '" : "Field '")
3201 << ME->getMemberDecl()->getName() << "' is ";
3202
3203 if (!printValue(ME, Out, N, TookTrue, IsAssuming))
3204 return nullptr;
3205
3206 const LocationContext *LCtx = N->getLocationContext();
3207 PathDiagnosticLocation Loc;
3208
3209 // If we know the value create a pop-up note to the member of the MemberExpr.
3210 if (!IsAssuming && ME->getMemberLoc().isValid())
3211 Loc = PathDiagnosticLocation(ME->getMemberLoc(), BRC.getSourceManager());
3212 else
3213 Loc = PathDiagnosticLocation(Cond, BRC.getSourceManager(), LCtx);
3214
3215 if (!Loc.isValid() || !Loc.asLocation().isValid())
3216 return nullptr;
3217
3218 if (isVarAnInterestingCondition(ME, N, &report))
3219 Out << WillBeUsedForACondition;
3220
3221 // If we know the value create a pop-up note.
3222 if (!IsAssuming)
3223 return std::make_shared<PathDiagnosticPopUpPiece>(Loc, Out.str());
3224
3225 auto event = std::make_shared<PathDiagnosticEventPiece>(Loc, Out.str());
3226 if (isInterestingExpr(ME, N, &report))
3227 event->setPrunable(false);
3228 return event;
3229 }
3230
printValue(const Expr * CondVarExpr,raw_ostream & Out,const ExplodedNode * N,bool TookTrue,bool IsAssuming)3231 bool ConditionBRVisitor::printValue(const Expr *CondVarExpr, raw_ostream &Out,
3232 const ExplodedNode *N, bool TookTrue,
3233 bool IsAssuming) {
3234 QualType Ty = CondVarExpr->getType();
3235
3236 if (Ty->isPointerType()) {
3237 Out << (TookTrue ? "non-null" : "null");
3238 return true;
3239 }
3240
3241 if (Ty->isObjCObjectPointerType()) {
3242 Out << (TookTrue ? "non-nil" : "nil");
3243 return true;
3244 }
3245
3246 if (!Ty->isIntegralOrEnumerationType())
3247 return false;
3248
3249 std::optional<const llvm::APSInt *> IntValue;
3250 if (!IsAssuming)
3251 IntValue = getConcreteIntegerValue(CondVarExpr, N);
3252
3253 if (IsAssuming || !IntValue) {
3254 if (Ty->isBooleanType())
3255 Out << (TookTrue ? "true" : "false");
3256 else
3257 Out << (TookTrue ? "not equal to 0" : "0");
3258 } else {
3259 if (Ty->isBooleanType())
3260 Out << ((*IntValue)->getBoolValue() ? "true" : "false");
3261 else
3262 Out << **IntValue;
3263 }
3264
3265 return true;
3266 }
3267
3268 constexpr llvm::StringLiteral ConditionBRVisitor::GenericTrueMessage;
3269 constexpr llvm::StringLiteral ConditionBRVisitor::GenericFalseMessage;
3270
isPieceMessageGeneric(const PathDiagnosticPiece * Piece)3271 bool ConditionBRVisitor::isPieceMessageGeneric(
3272 const PathDiagnosticPiece *Piece) {
3273 return Piece->getString() == GenericTrueMessage ||
3274 Piece->getString() == GenericFalseMessage;
3275 }
3276
3277 //===----------------------------------------------------------------------===//
3278 // Implementation of LikelyFalsePositiveSuppressionBRVisitor.
3279 //===----------------------------------------------------------------------===//
3280
finalizeVisitor(BugReporterContext & BRC,const ExplodedNode * N,PathSensitiveBugReport & BR)3281 void LikelyFalsePositiveSuppressionBRVisitor::finalizeVisitor(
3282 BugReporterContext &BRC, const ExplodedNode *N,
3283 PathSensitiveBugReport &BR) {
3284 // Here we suppress false positives coming from system headers. This list is
3285 // based on known issues.
3286 const AnalyzerOptions &Options = BRC.getAnalyzerOptions();
3287 const Decl *D = N->getLocationContext()->getDecl();
3288
3289 if (AnalysisDeclContext::isInStdNamespace(D)) {
3290 // Skip reports within the 'std' namespace. Although these can sometimes be
3291 // the user's fault, we currently don't report them very well, and
3292 // Note that this will not help for any other data structure libraries, like
3293 // TR1, Boost, or llvm/ADT.
3294 if (Options.ShouldSuppressFromCXXStandardLibrary) {
3295 BR.markInvalid(getTag(), nullptr);
3296 return;
3297 } else {
3298 // If the complete 'std' suppression is not enabled, suppress reports
3299 // from the 'std' namespace that are known to produce false positives.
3300
3301 // The analyzer issues a false use-after-free when std::list::pop_front
3302 // or std::list::pop_back are called multiple times because we cannot
3303 // reason about the internal invariants of the data structure.
3304 if (const auto *MD = dyn_cast<CXXMethodDecl>(D)) {
3305 const CXXRecordDecl *CD = MD->getParent();
3306 if (CD->getName() == "list") {
3307 BR.markInvalid(getTag(), nullptr);
3308 return;
3309 }
3310 }
3311
3312 // The analyzer issues a false positive when the constructor of
3313 // std::__independent_bits_engine from algorithms is used.
3314 if (const auto *MD = dyn_cast<CXXConstructorDecl>(D)) {
3315 const CXXRecordDecl *CD = MD->getParent();
3316 if (CD->getName() == "__independent_bits_engine") {
3317 BR.markInvalid(getTag(), nullptr);
3318 return;
3319 }
3320 }
3321
3322 for (const LocationContext *LCtx = N->getLocationContext(); LCtx;
3323 LCtx = LCtx->getParent()) {
3324 const auto *MD = dyn_cast<CXXMethodDecl>(LCtx->getDecl());
3325 if (!MD)
3326 continue;
3327
3328 const CXXRecordDecl *CD = MD->getParent();
3329 // The analyzer issues a false positive on
3330 // std::basic_string<uint8_t> v; v.push_back(1);
3331 // and
3332 // std::u16string s; s += u'a';
3333 // because we cannot reason about the internal invariants of the
3334 // data structure.
3335 if (CD->getName() == "basic_string") {
3336 BR.markInvalid(getTag(), nullptr);
3337 return;
3338 }
3339
3340 // The analyzer issues a false positive on
3341 // std::shared_ptr<int> p(new int(1)); p = nullptr;
3342 // because it does not reason properly about temporary destructors.
3343 if (CD->getName() == "shared_ptr") {
3344 BR.markInvalid(getTag(), nullptr);
3345 return;
3346 }
3347 }
3348 }
3349 }
3350
3351 // Skip reports within the sys/queue.h macros as we do not have the ability to
3352 // reason about data structure shapes.
3353 const SourceManager &SM = BRC.getSourceManager();
3354 FullSourceLoc Loc = BR.getLocation().asLocation();
3355 while (Loc.isMacroID()) {
3356 Loc = Loc.getSpellingLoc();
3357 if (SM.getFilename(Loc).endswith("sys/queue.h")) {
3358 BR.markInvalid(getTag(), nullptr);
3359 return;
3360 }
3361 }
3362 }
3363
3364 //===----------------------------------------------------------------------===//
3365 // Implementation of UndefOrNullArgVisitor.
3366 //===----------------------------------------------------------------------===//
3367
3368 PathDiagnosticPieceRef
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & BR)3369 UndefOrNullArgVisitor::VisitNode(const ExplodedNode *N, BugReporterContext &BRC,
3370 PathSensitiveBugReport &BR) {
3371 ProgramStateRef State = N->getState();
3372 ProgramPoint ProgLoc = N->getLocation();
3373
3374 // We are only interested in visiting CallEnter nodes.
3375 std::optional<CallEnter> CEnter = ProgLoc.getAs<CallEnter>();
3376 if (!CEnter)
3377 return nullptr;
3378
3379 // Check if one of the arguments is the region the visitor is tracking.
3380 CallEventManager &CEMgr = BRC.getStateManager().getCallEventManager();
3381 CallEventRef<> Call = CEMgr.getCaller(CEnter->getCalleeContext(), State);
3382 unsigned Idx = 0;
3383 ArrayRef<ParmVarDecl *> parms = Call->parameters();
3384
3385 for (const auto ParamDecl : parms) {
3386 const MemRegion *ArgReg = Call->getArgSVal(Idx).getAsRegion();
3387 ++Idx;
3388
3389 // Are we tracking the argument or its subregion?
3390 if ( !ArgReg || !R->isSubRegionOf(ArgReg->StripCasts()))
3391 continue;
3392
3393 // Check the function parameter type.
3394 assert(ParamDecl && "Formal parameter has no decl?");
3395 QualType T = ParamDecl->getType();
3396
3397 if (!(T->isAnyPointerType() || T->isReferenceType())) {
3398 // Function can only change the value passed in by address.
3399 continue;
3400 }
3401
3402 // If it is a const pointer value, the function does not intend to
3403 // change the value.
3404 if (T->getPointeeType().isConstQualified())
3405 continue;
3406
3407 // Mark the call site (LocationContext) as interesting if the value of the
3408 // argument is undefined or '0'/'NULL'.
3409 SVal BoundVal = State->getSVal(R);
3410 if (BoundVal.isUndef() || BoundVal.isZeroConstant()) {
3411 BR.markInteresting(CEnter->getCalleeContext());
3412 return nullptr;
3413 }
3414 }
3415 return nullptr;
3416 }
3417
3418 //===----------------------------------------------------------------------===//
3419 // Implementation of FalsePositiveRefutationBRVisitor.
3420 //===----------------------------------------------------------------------===//
3421
FalsePositiveRefutationBRVisitor()3422 FalsePositiveRefutationBRVisitor::FalsePositiveRefutationBRVisitor()
3423 : Constraints(ConstraintMap::Factory().getEmptyMap()) {}
3424
finalizeVisitor(BugReporterContext & BRC,const ExplodedNode * EndPathNode,PathSensitiveBugReport & BR)3425 void FalsePositiveRefutationBRVisitor::finalizeVisitor(
3426 BugReporterContext &BRC, const ExplodedNode *EndPathNode,
3427 PathSensitiveBugReport &BR) {
3428 // Collect new constraints
3429 addConstraints(EndPathNode, /*OverwriteConstraintsOnExistingSyms=*/true);
3430
3431 // Create a refutation manager
3432 llvm::SMTSolverRef RefutationSolver = llvm::CreateZ3Solver();
3433 ASTContext &Ctx = BRC.getASTContext();
3434
3435 // Add constraints to the solver
3436 for (const auto &I : Constraints) {
3437 const SymbolRef Sym = I.first;
3438 auto RangeIt = I.second.begin();
3439
3440 llvm::SMTExprRef SMTConstraints = SMTConv::getRangeExpr(
3441 RefutationSolver, Ctx, Sym, RangeIt->From(), RangeIt->To(),
3442 /*InRange=*/true);
3443 while ((++RangeIt) != I.second.end()) {
3444 SMTConstraints = RefutationSolver->mkOr(
3445 SMTConstraints, SMTConv::getRangeExpr(RefutationSolver, Ctx, Sym,
3446 RangeIt->From(), RangeIt->To(),
3447 /*InRange=*/true));
3448 }
3449
3450 RefutationSolver->addConstraint(SMTConstraints);
3451 }
3452
3453 // And check for satisfiability
3454 std::optional<bool> IsSAT = RefutationSolver->check();
3455 if (!IsSAT)
3456 return;
3457
3458 if (!*IsSAT)
3459 BR.markInvalid("Infeasible constraints", EndPathNode->getLocationContext());
3460 }
3461
addConstraints(const ExplodedNode * N,bool OverwriteConstraintsOnExistingSyms)3462 void FalsePositiveRefutationBRVisitor::addConstraints(
3463 const ExplodedNode *N, bool OverwriteConstraintsOnExistingSyms) {
3464 // Collect new constraints
3465 ConstraintMap NewCs = getConstraintMap(N->getState());
3466 ConstraintMap::Factory &CF = N->getState()->get_context<ConstraintMap>();
3467
3468 // Add constraints if we don't have them yet
3469 for (auto const &C : NewCs) {
3470 const SymbolRef &Sym = C.first;
3471 if (!Constraints.contains(Sym)) {
3472 // This symbol is new, just add the constraint.
3473 Constraints = CF.add(Constraints, Sym, C.second);
3474 } else if (OverwriteConstraintsOnExistingSyms) {
3475 // Overwrite the associated constraint of the Symbol.
3476 Constraints = CF.remove(Constraints, Sym);
3477 Constraints = CF.add(Constraints, Sym, C.second);
3478 }
3479 }
3480 }
3481
VisitNode(const ExplodedNode * N,BugReporterContext &,PathSensitiveBugReport &)3482 PathDiagnosticPieceRef FalsePositiveRefutationBRVisitor::VisitNode(
3483 const ExplodedNode *N, BugReporterContext &, PathSensitiveBugReport &) {
3484 addConstraints(N, /*OverwriteConstraintsOnExistingSyms=*/false);
3485 return nullptr;
3486 }
3487
Profile(llvm::FoldingSetNodeID & ID) const3488 void FalsePositiveRefutationBRVisitor::Profile(
3489 llvm::FoldingSetNodeID &ID) const {
3490 static int Tag = 0;
3491 ID.AddPointer(&Tag);
3492 }
3493
3494 //===----------------------------------------------------------------------===//
3495 // Implementation of TagVisitor.
3496 //===----------------------------------------------------------------------===//
3497
3498 int NoteTag::Kind = 0;
3499
Profile(llvm::FoldingSetNodeID & ID) const3500 void TagVisitor::Profile(llvm::FoldingSetNodeID &ID) const {
3501 static int Tag = 0;
3502 ID.AddPointer(&Tag);
3503 }
3504
VisitNode(const ExplodedNode * N,BugReporterContext & BRC,PathSensitiveBugReport & R)3505 PathDiagnosticPieceRef TagVisitor::VisitNode(const ExplodedNode *N,
3506 BugReporterContext &BRC,
3507 PathSensitiveBugReport &R) {
3508 ProgramPoint PP = N->getLocation();
3509 const NoteTag *T = dyn_cast_or_null<NoteTag>(PP.getTag());
3510 if (!T)
3511 return nullptr;
3512
3513 if (std::optional<std::string> Msg = T->generateMessage(BRC, R)) {
3514 PathDiagnosticLocation Loc =
3515 PathDiagnosticLocation::create(PP, BRC.getSourceManager());
3516 auto Piece = std::make_shared<PathDiagnosticEventPiece>(Loc, *Msg);
3517 Piece->setPrunable(T->isPrunable());
3518 return Piece;
3519 }
3520
3521 return nullptr;
3522 }
3523