1 //== RegionStore.cpp - Field-sensitive store model --------------*- C++ -*--==//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This file defines a basic region store model. In this model, we do have field
10 // sensitivity. But we assume nothing about the heap shape. So recursive data
11 // structures are largely ignored. Basically we do 1-limiting analysis.
12 // Parameter pointers are assumed with no aliasing. Pointee objects of
13 // parameters are created lazily.
14 //
15 //===----------------------------------------------------------------------===//
16
17 #include "clang/AST/Attr.h"
18 #include "clang/AST/CharUnits.h"
19 #include "clang/ASTMatchers/ASTMatchFinder.h"
20 #include "clang/Analysis/Analyses/LiveVariables.h"
21 #include "clang/Analysis/AnalysisDeclContext.h"
22 #include "clang/Basic/JsonSupport.h"
23 #include "clang/Basic/TargetInfo.h"
24 #include "clang/StaticAnalyzer/Core/PathSensitive/AnalysisManager.h"
25 #include "clang/StaticAnalyzer/Core/PathSensitive/CallEvent.h"
26 #include "clang/StaticAnalyzer/Core/PathSensitive/MemRegion.h"
27 #include "clang/StaticAnalyzer/Core/PathSensitive/ProgramState.h"
28 #include "clang/StaticAnalyzer/Core/PathSensitive/ProgramStateTrait.h"
29 #include "clang/StaticAnalyzer/Core/PathSensitive/SubEngine.h"
30 #include "llvm/ADT/ImmutableMap.h"
31 #include "llvm/ADT/Optional.h"
32 #include "llvm/Support/raw_ostream.h"
33 #include <utility>
34
35 using namespace clang;
36 using namespace ento;
37
38 //===----------------------------------------------------------------------===//
39 // Representation of binding keys.
40 //===----------------------------------------------------------------------===//
41
42 namespace {
43 class BindingKey {
44 public:
45 enum Kind { Default = 0x0, Direct = 0x1 };
46 private:
47 enum { Symbolic = 0x2 };
48
49 llvm::PointerIntPair<const MemRegion *, 2> P;
50 uint64_t Data;
51
52 /// Create a key for a binding to region \p r, which has a symbolic offset
53 /// from region \p Base.
BindingKey(const SubRegion * r,const SubRegion * Base,Kind k)54 explicit BindingKey(const SubRegion *r, const SubRegion *Base, Kind k)
55 : P(r, k | Symbolic), Data(reinterpret_cast<uintptr_t>(Base)) {
56 assert(r && Base && "Must have known regions.");
57 assert(getConcreteOffsetRegion() == Base && "Failed to store base region");
58 }
59
60 /// Create a key for a binding at \p offset from base region \p r.
BindingKey(const MemRegion * r,uint64_t offset,Kind k)61 explicit BindingKey(const MemRegion *r, uint64_t offset, Kind k)
62 : P(r, k), Data(offset) {
63 assert(r && "Must have known regions.");
64 assert(getOffset() == offset && "Failed to store offset");
65 assert((r == r->getBaseRegion() || isa<ObjCIvarRegion>(r) ||
66 isa <CXXDerivedObjectRegion>(r)) &&
67 "Not a base");
68 }
69 public:
70
isDirect() const71 bool isDirect() const { return P.getInt() & Direct; }
hasSymbolicOffset() const72 bool hasSymbolicOffset() const { return P.getInt() & Symbolic; }
73
getRegion() const74 const MemRegion *getRegion() const { return P.getPointer(); }
getOffset() const75 uint64_t getOffset() const {
76 assert(!hasSymbolicOffset());
77 return Data;
78 }
79
getConcreteOffsetRegion() const80 const SubRegion *getConcreteOffsetRegion() const {
81 assert(hasSymbolicOffset());
82 return reinterpret_cast<const SubRegion *>(static_cast<uintptr_t>(Data));
83 }
84
getBaseRegion() const85 const MemRegion *getBaseRegion() const {
86 if (hasSymbolicOffset())
87 return getConcreteOffsetRegion()->getBaseRegion();
88 return getRegion()->getBaseRegion();
89 }
90
Profile(llvm::FoldingSetNodeID & ID) const91 void Profile(llvm::FoldingSetNodeID& ID) const {
92 ID.AddPointer(P.getOpaqueValue());
93 ID.AddInteger(Data);
94 }
95
96 static BindingKey Make(const MemRegion *R, Kind k);
97
operator <(const BindingKey & X) const98 bool operator<(const BindingKey &X) const {
99 if (P.getOpaqueValue() < X.P.getOpaqueValue())
100 return true;
101 if (P.getOpaqueValue() > X.P.getOpaqueValue())
102 return false;
103 return Data < X.Data;
104 }
105
operator ==(const BindingKey & X) const106 bool operator==(const BindingKey &X) const {
107 return P.getOpaqueValue() == X.P.getOpaqueValue() &&
108 Data == X.Data;
109 }
110
111 LLVM_DUMP_METHOD void dump() const;
112 };
113 } // end anonymous namespace
114
Make(const MemRegion * R,Kind k)115 BindingKey BindingKey::Make(const MemRegion *R, Kind k) {
116 const RegionOffset &RO = R->getAsOffset();
117 if (RO.hasSymbolicOffset())
118 return BindingKey(cast<SubRegion>(R), cast<SubRegion>(RO.getRegion()), k);
119
120 return BindingKey(RO.getRegion(), RO.getOffset(), k);
121 }
122
123 namespace llvm {
operator <<(raw_ostream & Out,BindingKey K)124 static inline raw_ostream &operator<<(raw_ostream &Out, BindingKey K) {
125 Out << "\"kind\": \"" << (K.isDirect() ? "Direct" : "Default")
126 << "\", \"offset\": ";
127
128 if (!K.hasSymbolicOffset())
129 Out << K.getOffset();
130 else
131 Out << "null";
132
133 return Out;
134 }
135
136 } // namespace llvm
137
138 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
dump() const139 void BindingKey::dump() const { llvm::errs() << *this; }
140 #endif
141
142 //===----------------------------------------------------------------------===//
143 // Actual Store type.
144 //===----------------------------------------------------------------------===//
145
146 typedef llvm::ImmutableMap<BindingKey, SVal> ClusterBindings;
147 typedef llvm::ImmutableMapRef<BindingKey, SVal> ClusterBindingsRef;
148 typedef std::pair<BindingKey, SVal> BindingPair;
149
150 typedef llvm::ImmutableMap<const MemRegion *, ClusterBindings>
151 RegionBindings;
152
153 namespace {
154 class RegionBindingsRef : public llvm::ImmutableMapRef<const MemRegion *,
155 ClusterBindings> {
156 ClusterBindings::Factory *CBFactory;
157
158 // This flag indicates whether the current bindings are within the analysis
159 // that has started from main(). It affects how we perform loads from
160 // global variables that have initializers: if we have observed the
161 // program execution from the start and we know that these variables
162 // have not been overwritten yet, we can be sure that their initializers
163 // are still relevant. This flag never gets changed when the bindings are
164 // updated, so it could potentially be moved into RegionStoreManager
165 // (as if it's the same bindings but a different loading procedure)
166 // however that would have made the manager needlessly stateful.
167 bool IsMainAnalysis;
168
169 public:
170 typedef llvm::ImmutableMapRef<const MemRegion *, ClusterBindings>
171 ParentTy;
172
RegionBindingsRef(ClusterBindings::Factory & CBFactory,const RegionBindings::TreeTy * T,RegionBindings::TreeTy::Factory * F,bool IsMainAnalysis)173 RegionBindingsRef(ClusterBindings::Factory &CBFactory,
174 const RegionBindings::TreeTy *T,
175 RegionBindings::TreeTy::Factory *F,
176 bool IsMainAnalysis)
177 : llvm::ImmutableMapRef<const MemRegion *, ClusterBindings>(T, F),
178 CBFactory(&CBFactory), IsMainAnalysis(IsMainAnalysis) {}
179
RegionBindingsRef(const ParentTy & P,ClusterBindings::Factory & CBFactory,bool IsMainAnalysis)180 RegionBindingsRef(const ParentTy &P,
181 ClusterBindings::Factory &CBFactory,
182 bool IsMainAnalysis)
183 : llvm::ImmutableMapRef<const MemRegion *, ClusterBindings>(P),
184 CBFactory(&CBFactory), IsMainAnalysis(IsMainAnalysis) {}
185
add(key_type_ref K,data_type_ref D) const186 RegionBindingsRef add(key_type_ref K, data_type_ref D) const {
187 return RegionBindingsRef(static_cast<const ParentTy *>(this)->add(K, D),
188 *CBFactory, IsMainAnalysis);
189 }
190
remove(key_type_ref K) const191 RegionBindingsRef remove(key_type_ref K) const {
192 return RegionBindingsRef(static_cast<const ParentTy *>(this)->remove(K),
193 *CBFactory, IsMainAnalysis);
194 }
195
196 RegionBindingsRef addBinding(BindingKey K, SVal V) const;
197
198 RegionBindingsRef addBinding(const MemRegion *R,
199 BindingKey::Kind k, SVal V) const;
200
201 const SVal *lookup(BindingKey K) const;
202 const SVal *lookup(const MemRegion *R, BindingKey::Kind k) const;
203 using llvm::ImmutableMapRef<const MemRegion *, ClusterBindings>::lookup;
204
205 RegionBindingsRef removeBinding(BindingKey K);
206
207 RegionBindingsRef removeBinding(const MemRegion *R,
208 BindingKey::Kind k);
209
removeBinding(const MemRegion * R)210 RegionBindingsRef removeBinding(const MemRegion *R) {
211 return removeBinding(R, BindingKey::Direct).
212 removeBinding(R, BindingKey::Default);
213 }
214
215 Optional<SVal> getDirectBinding(const MemRegion *R) const;
216
217 /// getDefaultBinding - Returns an SVal* representing an optional default
218 /// binding associated with a region and its subregions.
219 Optional<SVal> getDefaultBinding(const MemRegion *R) const;
220
221 /// Return the internal tree as a Store.
asStore() const222 Store asStore() const {
223 llvm::PointerIntPair<Store, 1, bool> Ptr = {
224 asImmutableMap().getRootWithoutRetain(), IsMainAnalysis};
225 return reinterpret_cast<Store>(Ptr.getOpaqueValue());
226 }
227
isMainAnalysis() const228 bool isMainAnalysis() const {
229 return IsMainAnalysis;
230 }
231
printJson(raw_ostream & Out,const char * NL="\\n",unsigned int Space=0,bool IsDot=false) const232 void printJson(raw_ostream &Out, const char *NL = "\n",
233 unsigned int Space = 0, bool IsDot = false) const {
234 for (iterator I = begin(); I != end(); ++I) {
235 // TODO: We might need a .printJson for I.getKey() as well.
236 Indent(Out, Space, IsDot)
237 << "{ \"cluster\": \"" << I.getKey() << "\", \"pointer\": \""
238 << (const void *)I.getKey() << "\", \"items\": [" << NL;
239
240 ++Space;
241 const ClusterBindings &CB = I.getData();
242 for (ClusterBindings::iterator CI = CB.begin(); CI != CB.end(); ++CI) {
243 Indent(Out, Space, IsDot) << "{ " << CI.getKey() << ", \"value\": ";
244 CI.getData().printJson(Out, /*AddQuotes=*/true);
245 Out << " }";
246 if (std::next(CI) != CB.end())
247 Out << ',';
248 Out << NL;
249 }
250
251 --Space;
252 Indent(Out, Space, IsDot) << "]}";
253 if (std::next(I) != end())
254 Out << ',';
255 Out << NL;
256 }
257 }
258
dump() const259 LLVM_DUMP_METHOD void dump() const { printJson(llvm::errs()); }
260 };
261 } // end anonymous namespace
262
263 typedef const RegionBindingsRef& RegionBindingsConstRef;
264
getDirectBinding(const MemRegion * R) const265 Optional<SVal> RegionBindingsRef::getDirectBinding(const MemRegion *R) const {
266 return Optional<SVal>::create(lookup(R, BindingKey::Direct));
267 }
268
getDefaultBinding(const MemRegion * R) const269 Optional<SVal> RegionBindingsRef::getDefaultBinding(const MemRegion *R) const {
270 return Optional<SVal>::create(lookup(R, BindingKey::Default));
271 }
272
addBinding(BindingKey K,SVal V) const273 RegionBindingsRef RegionBindingsRef::addBinding(BindingKey K, SVal V) const {
274 const MemRegion *Base = K.getBaseRegion();
275
276 const ClusterBindings *ExistingCluster = lookup(Base);
277 ClusterBindings Cluster =
278 (ExistingCluster ? *ExistingCluster : CBFactory->getEmptyMap());
279
280 ClusterBindings NewCluster = CBFactory->add(Cluster, K, V);
281 return add(Base, NewCluster);
282 }
283
284
addBinding(const MemRegion * R,BindingKey::Kind k,SVal V) const285 RegionBindingsRef RegionBindingsRef::addBinding(const MemRegion *R,
286 BindingKey::Kind k,
287 SVal V) const {
288 return addBinding(BindingKey::Make(R, k), V);
289 }
290
lookup(BindingKey K) const291 const SVal *RegionBindingsRef::lookup(BindingKey K) const {
292 const ClusterBindings *Cluster = lookup(K.getBaseRegion());
293 if (!Cluster)
294 return nullptr;
295 return Cluster->lookup(K);
296 }
297
lookup(const MemRegion * R,BindingKey::Kind k) const298 const SVal *RegionBindingsRef::lookup(const MemRegion *R,
299 BindingKey::Kind k) const {
300 return lookup(BindingKey::Make(R, k));
301 }
302
removeBinding(BindingKey K)303 RegionBindingsRef RegionBindingsRef::removeBinding(BindingKey K) {
304 const MemRegion *Base = K.getBaseRegion();
305 const ClusterBindings *Cluster = lookup(Base);
306 if (!Cluster)
307 return *this;
308
309 ClusterBindings NewCluster = CBFactory->remove(*Cluster, K);
310 if (NewCluster.isEmpty())
311 return remove(Base);
312 return add(Base, NewCluster);
313 }
314
removeBinding(const MemRegion * R,BindingKey::Kind k)315 RegionBindingsRef RegionBindingsRef::removeBinding(const MemRegion *R,
316 BindingKey::Kind k){
317 return removeBinding(BindingKey::Make(R, k));
318 }
319
320 //===----------------------------------------------------------------------===//
321 // Fine-grained control of RegionStoreManager.
322 //===----------------------------------------------------------------------===//
323
324 namespace {
325 struct minimal_features_tag {};
326 struct maximal_features_tag {};
327
328 class RegionStoreFeatures {
329 bool SupportsFields;
330 public:
RegionStoreFeatures(minimal_features_tag)331 RegionStoreFeatures(minimal_features_tag) :
332 SupportsFields(false) {}
333
RegionStoreFeatures(maximal_features_tag)334 RegionStoreFeatures(maximal_features_tag) :
335 SupportsFields(true) {}
336
enableFields(bool t)337 void enableFields(bool t) { SupportsFields = t; }
338
supportsFields() const339 bool supportsFields() const { return SupportsFields; }
340 };
341 }
342
343 //===----------------------------------------------------------------------===//
344 // Main RegionStore logic.
345 //===----------------------------------------------------------------------===//
346
347 namespace {
348 class InvalidateRegionsWorker;
349
350 class RegionStoreManager : public StoreManager {
351 public:
352 const RegionStoreFeatures Features;
353
354 RegionBindings::Factory RBFactory;
355 mutable ClusterBindings::Factory CBFactory;
356
357 typedef std::vector<SVal> SValListTy;
358 private:
359 typedef llvm::DenseMap<const LazyCompoundValData *,
360 SValListTy> LazyBindingsMapTy;
361 LazyBindingsMapTy LazyBindingsMap;
362
363 /// The largest number of fields a struct can have and still be
364 /// considered "small".
365 ///
366 /// This is currently used to decide whether or not it is worth "forcing" a
367 /// LazyCompoundVal on bind.
368 ///
369 /// This is controlled by 'region-store-small-struct-limit' option.
370 /// To disable all small-struct-dependent behavior, set the option to "0".
371 unsigned SmallStructLimit;
372
373 /// A helper used to populate the work list with the given set of
374 /// regions.
375 void populateWorkList(InvalidateRegionsWorker &W,
376 ArrayRef<SVal> Values,
377 InvalidatedRegions *TopLevelRegions);
378
379 public:
RegionStoreManager(ProgramStateManager & mgr,const RegionStoreFeatures & f)380 RegionStoreManager(ProgramStateManager& mgr, const RegionStoreFeatures &f)
381 : StoreManager(mgr), Features(f),
382 RBFactory(mgr.getAllocator()), CBFactory(mgr.getAllocator()),
383 SmallStructLimit(0) {
384 SubEngine &Eng = StateMgr.getOwningEngine();
385 AnalyzerOptions &Options = Eng.getAnalysisManager().options;
386 SmallStructLimit = Options.RegionStoreSmallStructLimit;
387 }
388
389
390 /// setImplicitDefaultValue - Set the default binding for the provided
391 /// MemRegion to the value implicitly defined for compound literals when
392 /// the value is not specified.
393 RegionBindingsRef setImplicitDefaultValue(RegionBindingsConstRef B,
394 const MemRegion *R, QualType T);
395
396 /// ArrayToPointer - Emulates the "decay" of an array to a pointer
397 /// type. 'Array' represents the lvalue of the array being decayed
398 /// to a pointer, and the returned SVal represents the decayed
399 /// version of that lvalue (i.e., a pointer to the first element of
400 /// the array). This is called by ExprEngine when evaluating
401 /// casts from arrays to pointers.
402 SVal ArrayToPointer(Loc Array, QualType ElementTy) override;
403
404 /// Creates the Store that correctly represents memory contents before
405 /// the beginning of the analysis of the given top-level stack frame.
getInitialStore(const LocationContext * InitLoc)406 StoreRef getInitialStore(const LocationContext *InitLoc) override {
407 bool IsMainAnalysis = false;
408 if (const auto *FD = dyn_cast<FunctionDecl>(InitLoc->getDecl()))
409 IsMainAnalysis = FD->isMain() && !Ctx.getLangOpts().CPlusPlus;
410 return StoreRef(RegionBindingsRef(
411 RegionBindingsRef::ParentTy(RBFactory.getEmptyMap(), RBFactory),
412 CBFactory, IsMainAnalysis).asStore(), *this);
413 }
414
415 //===-------------------------------------------------------------------===//
416 // Binding values to regions.
417 //===-------------------------------------------------------------------===//
418 RegionBindingsRef invalidateGlobalRegion(MemRegion::Kind K,
419 const Expr *Ex,
420 unsigned Count,
421 const LocationContext *LCtx,
422 RegionBindingsRef B,
423 InvalidatedRegions *Invalidated);
424
425 StoreRef invalidateRegions(Store store,
426 ArrayRef<SVal> Values,
427 const Expr *E, unsigned Count,
428 const LocationContext *LCtx,
429 const CallEvent *Call,
430 InvalidatedSymbols &IS,
431 RegionAndSymbolInvalidationTraits &ITraits,
432 InvalidatedRegions *Invalidated,
433 InvalidatedRegions *InvalidatedTopLevel) override;
434
435 bool scanReachableSymbols(Store S, const MemRegion *R,
436 ScanReachableSymbols &Callbacks) override;
437
438 RegionBindingsRef removeSubRegionBindings(RegionBindingsConstRef B,
439 const SubRegion *R);
440
441 public: // Part of public interface to class.
442
Bind(Store store,Loc LV,SVal V)443 StoreRef Bind(Store store, Loc LV, SVal V) override {
444 return StoreRef(bind(getRegionBindings(store), LV, V).asStore(), *this);
445 }
446
447 RegionBindingsRef bind(RegionBindingsConstRef B, Loc LV, SVal V);
448
449 // BindDefaultInitial is only used to initialize a region with
450 // a default value.
BindDefaultInitial(Store store,const MemRegion * R,SVal V)451 StoreRef BindDefaultInitial(Store store, const MemRegion *R,
452 SVal V) override {
453 RegionBindingsRef B = getRegionBindings(store);
454 // Use other APIs when you have to wipe the region that was initialized
455 // earlier.
456 assert(!(B.getDefaultBinding(R) || B.getDirectBinding(R)) &&
457 "Double initialization!");
458 B = B.addBinding(BindingKey::Make(R, BindingKey::Default), V);
459 return StoreRef(B.asImmutableMap().getRootWithoutRetain(), *this);
460 }
461
462 // BindDefaultZero is used for zeroing constructors that may accidentally
463 // overwrite existing bindings.
BindDefaultZero(Store store,const MemRegion * R)464 StoreRef BindDefaultZero(Store store, const MemRegion *R) override {
465 // FIXME: The offsets of empty bases can be tricky because of
466 // of the so called "empty base class optimization".
467 // If a base class has been optimized out
468 // we should not try to create a binding, otherwise we should.
469 // Unfortunately, at the moment ASTRecordLayout doesn't expose
470 // the actual sizes of the empty bases
471 // and trying to infer them from offsets/alignments
472 // seems to be error-prone and non-trivial because of the trailing padding.
473 // As a temporary mitigation we don't create bindings for empty bases.
474 if (const auto *BR = dyn_cast<CXXBaseObjectRegion>(R))
475 if (BR->getDecl()->isEmpty())
476 return StoreRef(store, *this);
477
478 RegionBindingsRef B = getRegionBindings(store);
479 SVal V = svalBuilder.makeZeroVal(Ctx.CharTy);
480 B = removeSubRegionBindings(B, cast<SubRegion>(R));
481 B = B.addBinding(BindingKey::Make(R, BindingKey::Default), V);
482 return StoreRef(B.asImmutableMap().getRootWithoutRetain(), *this);
483 }
484
485 /// Attempt to extract the fields of \p LCV and bind them to the struct region
486 /// \p R.
487 ///
488 /// This path is used when it seems advantageous to "force" loading the values
489 /// within a LazyCompoundVal to bind memberwise to the struct region, rather
490 /// than using a Default binding at the base of the entire region. This is a
491 /// heuristic attempting to avoid building long chains of LazyCompoundVals.
492 ///
493 /// \returns The updated store bindings, or \c None if binding non-lazily
494 /// would be too expensive.
495 Optional<RegionBindingsRef> tryBindSmallStruct(RegionBindingsConstRef B,
496 const TypedValueRegion *R,
497 const RecordDecl *RD,
498 nonloc::LazyCompoundVal LCV);
499
500 /// BindStruct - Bind a compound value to a structure.
501 RegionBindingsRef bindStruct(RegionBindingsConstRef B,
502 const TypedValueRegion* R, SVal V);
503
504 /// BindVector - Bind a compound value to a vector.
505 RegionBindingsRef bindVector(RegionBindingsConstRef B,
506 const TypedValueRegion* R, SVal V);
507
508 RegionBindingsRef bindArray(RegionBindingsConstRef B,
509 const TypedValueRegion* R,
510 SVal V);
511
512 /// Clears out all bindings in the given region and assigns a new value
513 /// as a Default binding.
514 RegionBindingsRef bindAggregate(RegionBindingsConstRef B,
515 const TypedRegion *R,
516 SVal DefaultVal);
517
518 /// Create a new store with the specified binding removed.
519 /// \param ST the original store, that is the basis for the new store.
520 /// \param L the location whose binding should be removed.
521 StoreRef killBinding(Store ST, Loc L) override;
522
incrementReferenceCount(Store store)523 void incrementReferenceCount(Store store) override {
524 getRegionBindings(store).manualRetain();
525 }
526
527 /// If the StoreManager supports it, decrement the reference count of
528 /// the specified Store object. If the reference count hits 0, the memory
529 /// associated with the object is recycled.
decrementReferenceCount(Store store)530 void decrementReferenceCount(Store store) override {
531 getRegionBindings(store).manualRelease();
532 }
533
534 bool includedInBindings(Store store, const MemRegion *region) const override;
535
536 /// Return the value bound to specified location in a given state.
537 ///
538 /// The high level logic for this method is this:
539 /// getBinding (L)
540 /// if L has binding
541 /// return L's binding
542 /// else if L is in killset
543 /// return unknown
544 /// else
545 /// if L is on stack or heap
546 /// return undefined
547 /// else
548 /// return symbolic
getBinding(Store S,Loc L,QualType T)549 SVal getBinding(Store S, Loc L, QualType T) override {
550 return getBinding(getRegionBindings(S), L, T);
551 }
552
getDefaultBinding(Store S,const MemRegion * R)553 Optional<SVal> getDefaultBinding(Store S, const MemRegion *R) override {
554 RegionBindingsRef B = getRegionBindings(S);
555 // Default bindings are always applied over a base region so look up the
556 // base region's default binding, otherwise the lookup will fail when R
557 // is at an offset from R->getBaseRegion().
558 return B.getDefaultBinding(R->getBaseRegion());
559 }
560
561 SVal getBinding(RegionBindingsConstRef B, Loc L, QualType T = QualType());
562
563 SVal getBindingForElement(RegionBindingsConstRef B, const ElementRegion *R);
564
565 SVal getBindingForField(RegionBindingsConstRef B, const FieldRegion *R);
566
567 SVal getBindingForObjCIvar(RegionBindingsConstRef B, const ObjCIvarRegion *R);
568
569 SVal getBindingForVar(RegionBindingsConstRef B, const VarRegion *R);
570
571 SVal getBindingForLazySymbol(const TypedValueRegion *R);
572
573 SVal getBindingForFieldOrElementCommon(RegionBindingsConstRef B,
574 const TypedValueRegion *R,
575 QualType Ty);
576
577 SVal getLazyBinding(const SubRegion *LazyBindingRegion,
578 RegionBindingsRef LazyBinding);
579
580 /// Get bindings for the values in a struct and return a CompoundVal, used
581 /// when doing struct copy:
582 /// struct s x, y;
583 /// x = y;
584 /// y's value is retrieved by this method.
585 SVal getBindingForStruct(RegionBindingsConstRef B, const TypedValueRegion *R);
586 SVal getBindingForArray(RegionBindingsConstRef B, const TypedValueRegion *R);
587 NonLoc createLazyBinding(RegionBindingsConstRef B, const TypedValueRegion *R);
588
589 /// Used to lazily generate derived symbols for bindings that are defined
590 /// implicitly by default bindings in a super region.
591 ///
592 /// Note that callers may need to specially handle LazyCompoundVals, which
593 /// are returned as is in case the caller needs to treat them differently.
594 Optional<SVal> getBindingForDerivedDefaultValue(RegionBindingsConstRef B,
595 const MemRegion *superR,
596 const TypedValueRegion *R,
597 QualType Ty);
598
599 /// Get the state and region whose binding this region \p R corresponds to.
600 ///
601 /// If there is no lazy binding for \p R, the returned value will have a null
602 /// \c second. Note that a null pointer can represents a valid Store.
603 std::pair<Store, const SubRegion *>
604 findLazyBinding(RegionBindingsConstRef B, const SubRegion *R,
605 const SubRegion *originalRegion);
606
607 /// Returns the cached set of interesting SVals contained within a lazy
608 /// binding.
609 ///
610 /// The precise value of "interesting" is determined for the purposes of
611 /// RegionStore's internal analysis. It must always contain all regions and
612 /// symbols, but may omit constants and other kinds of SVal.
613 const SValListTy &getInterestingValues(nonloc::LazyCompoundVal LCV);
614
615 //===------------------------------------------------------------------===//
616 // State pruning.
617 //===------------------------------------------------------------------===//
618
619 /// removeDeadBindings - Scans the RegionStore of 'state' for dead values.
620 /// It returns a new Store with these values removed.
621 StoreRef removeDeadBindings(Store store, const StackFrameContext *LCtx,
622 SymbolReaper& SymReaper) override;
623
624 //===------------------------------------------------------------------===//
625 // Region "extents".
626 //===------------------------------------------------------------------===//
627
628 // FIXME: This method will soon be eliminated; see the note in Store.h.
629 DefinedOrUnknownSVal getSizeInElements(ProgramStateRef state,
630 const MemRegion* R,
631 QualType EleTy) override;
632
633 //===------------------------------------------------------------------===//
634 // Utility methods.
635 //===------------------------------------------------------------------===//
636
getRegionBindings(Store store) const637 RegionBindingsRef getRegionBindings(Store store) const {
638 llvm::PointerIntPair<Store, 1, bool> Ptr;
639 Ptr.setFromOpaqueValue(const_cast<void *>(store));
640 return RegionBindingsRef(
641 CBFactory,
642 static_cast<const RegionBindings::TreeTy *>(Ptr.getPointer()),
643 RBFactory.getTreeFactory(),
644 Ptr.getInt());
645 }
646
647 void printJson(raw_ostream &Out, Store S, const char *NL = "\n",
648 unsigned int Space = 0, bool IsDot = false) const override;
649
iterBindings(Store store,BindingsHandler & f)650 void iterBindings(Store store, BindingsHandler& f) override {
651 RegionBindingsRef B = getRegionBindings(store);
652 for (RegionBindingsRef::iterator I = B.begin(), E = B.end(); I != E; ++I) {
653 const ClusterBindings &Cluster = I.getData();
654 for (ClusterBindings::iterator CI = Cluster.begin(), CE = Cluster.end();
655 CI != CE; ++CI) {
656 const BindingKey &K = CI.getKey();
657 if (!K.isDirect())
658 continue;
659 if (const SubRegion *R = dyn_cast<SubRegion>(K.getRegion())) {
660 // FIXME: Possibly incorporate the offset?
661 if (!f.HandleBinding(*this, store, R, CI.getData()))
662 return;
663 }
664 }
665 }
666 }
667 };
668
669 } // end anonymous namespace
670
671 //===----------------------------------------------------------------------===//
672 // RegionStore creation.
673 //===----------------------------------------------------------------------===//
674
675 std::unique_ptr<StoreManager>
CreateRegionStoreManager(ProgramStateManager & StMgr)676 ento::CreateRegionStoreManager(ProgramStateManager &StMgr) {
677 RegionStoreFeatures F = maximal_features_tag();
678 return std::make_unique<RegionStoreManager>(StMgr, F);
679 }
680
681 std::unique_ptr<StoreManager>
CreateFieldsOnlyRegionStoreManager(ProgramStateManager & StMgr)682 ento::CreateFieldsOnlyRegionStoreManager(ProgramStateManager &StMgr) {
683 RegionStoreFeatures F = minimal_features_tag();
684 F.enableFields(true);
685 return std::make_unique<RegionStoreManager>(StMgr, F);
686 }
687
688
689 //===----------------------------------------------------------------------===//
690 // Region Cluster analysis.
691 //===----------------------------------------------------------------------===//
692
693 namespace {
694 /// Used to determine which global regions are automatically included in the
695 /// initial worklist of a ClusterAnalysis.
696 enum GlobalsFilterKind {
697 /// Don't include any global regions.
698 GFK_None,
699 /// Only include system globals.
700 GFK_SystemOnly,
701 /// Include all global regions.
702 GFK_All
703 };
704
705 template <typename DERIVED>
706 class ClusterAnalysis {
707 protected:
708 typedef llvm::DenseMap<const MemRegion *, const ClusterBindings *> ClusterMap;
709 typedef const MemRegion * WorkListElement;
710 typedef SmallVector<WorkListElement, 10> WorkList;
711
712 llvm::SmallPtrSet<const ClusterBindings *, 16> Visited;
713
714 WorkList WL;
715
716 RegionStoreManager &RM;
717 ASTContext &Ctx;
718 SValBuilder &svalBuilder;
719
720 RegionBindingsRef B;
721
722
723 protected:
getCluster(const MemRegion * R)724 const ClusterBindings *getCluster(const MemRegion *R) {
725 return B.lookup(R);
726 }
727
728 /// Returns true if all clusters in the given memspace should be initially
729 /// included in the cluster analysis. Subclasses may provide their
730 /// own implementation.
includeEntireMemorySpace(const MemRegion * Base)731 bool includeEntireMemorySpace(const MemRegion *Base) {
732 return false;
733 }
734
735 public:
ClusterAnalysis(RegionStoreManager & rm,ProgramStateManager & StateMgr,RegionBindingsRef b)736 ClusterAnalysis(RegionStoreManager &rm, ProgramStateManager &StateMgr,
737 RegionBindingsRef b)
738 : RM(rm), Ctx(StateMgr.getContext()),
739 svalBuilder(StateMgr.getSValBuilder()), B(std::move(b)) {}
740
getRegionBindings() const741 RegionBindingsRef getRegionBindings() const { return B; }
742
isVisited(const MemRegion * R)743 bool isVisited(const MemRegion *R) {
744 return Visited.count(getCluster(R));
745 }
746
GenerateClusters()747 void GenerateClusters() {
748 // Scan the entire set of bindings and record the region clusters.
749 for (RegionBindingsRef::iterator RI = B.begin(), RE = B.end();
750 RI != RE; ++RI){
751 const MemRegion *Base = RI.getKey();
752
753 const ClusterBindings &Cluster = RI.getData();
754 assert(!Cluster.isEmpty() && "Empty clusters should be removed");
755 static_cast<DERIVED*>(this)->VisitAddedToCluster(Base, Cluster);
756
757 // If the base's memspace should be entirely invalidated, add the cluster
758 // to the workspace up front.
759 if (static_cast<DERIVED*>(this)->includeEntireMemorySpace(Base))
760 AddToWorkList(WorkListElement(Base), &Cluster);
761 }
762 }
763
AddToWorkList(WorkListElement E,const ClusterBindings * C)764 bool AddToWorkList(WorkListElement E, const ClusterBindings *C) {
765 if (C && !Visited.insert(C).second)
766 return false;
767 WL.push_back(E);
768 return true;
769 }
770
AddToWorkList(const MemRegion * R)771 bool AddToWorkList(const MemRegion *R) {
772 return static_cast<DERIVED*>(this)->AddToWorkList(R);
773 }
774
RunWorkList()775 void RunWorkList() {
776 while (!WL.empty()) {
777 WorkListElement E = WL.pop_back_val();
778 const MemRegion *BaseR = E;
779
780 static_cast<DERIVED*>(this)->VisitCluster(BaseR, getCluster(BaseR));
781 }
782 }
783
VisitAddedToCluster(const MemRegion * baseR,const ClusterBindings & C)784 void VisitAddedToCluster(const MemRegion *baseR, const ClusterBindings &C) {}
VisitCluster(const MemRegion * baseR,const ClusterBindings * C)785 void VisitCluster(const MemRegion *baseR, const ClusterBindings *C) {}
786
VisitCluster(const MemRegion * BaseR,const ClusterBindings * C,bool Flag)787 void VisitCluster(const MemRegion *BaseR, const ClusterBindings *C,
788 bool Flag) {
789 static_cast<DERIVED*>(this)->VisitCluster(BaseR, C);
790 }
791 };
792 }
793
794 //===----------------------------------------------------------------------===//
795 // Binding invalidation.
796 //===----------------------------------------------------------------------===//
797
scanReachableSymbols(Store S,const MemRegion * R,ScanReachableSymbols & Callbacks)798 bool RegionStoreManager::scanReachableSymbols(Store S, const MemRegion *R,
799 ScanReachableSymbols &Callbacks) {
800 assert(R == R->getBaseRegion() && "Should only be called for base regions");
801 RegionBindingsRef B = getRegionBindings(S);
802 const ClusterBindings *Cluster = B.lookup(R);
803
804 if (!Cluster)
805 return true;
806
807 for (ClusterBindings::iterator RI = Cluster->begin(), RE = Cluster->end();
808 RI != RE; ++RI) {
809 if (!Callbacks.scan(RI.getData()))
810 return false;
811 }
812
813 return true;
814 }
815
isUnionField(const FieldRegion * FR)816 static inline bool isUnionField(const FieldRegion *FR) {
817 return FR->getDecl()->getParent()->isUnion();
818 }
819
820 typedef SmallVector<const FieldDecl *, 8> FieldVector;
821
getSymbolicOffsetFields(BindingKey K,FieldVector & Fields)822 static void getSymbolicOffsetFields(BindingKey K, FieldVector &Fields) {
823 assert(K.hasSymbolicOffset() && "Not implemented for concrete offset keys");
824
825 const MemRegion *Base = K.getConcreteOffsetRegion();
826 const MemRegion *R = K.getRegion();
827
828 while (R != Base) {
829 if (const FieldRegion *FR = dyn_cast<FieldRegion>(R))
830 if (!isUnionField(FR))
831 Fields.push_back(FR->getDecl());
832
833 R = cast<SubRegion>(R)->getSuperRegion();
834 }
835 }
836
isCompatibleWithFields(BindingKey K,const FieldVector & Fields)837 static bool isCompatibleWithFields(BindingKey K, const FieldVector &Fields) {
838 assert(K.hasSymbolicOffset() && "Not implemented for concrete offset keys");
839
840 if (Fields.empty())
841 return true;
842
843 FieldVector FieldsInBindingKey;
844 getSymbolicOffsetFields(K, FieldsInBindingKey);
845
846 ptrdiff_t Delta = FieldsInBindingKey.size() - Fields.size();
847 if (Delta >= 0)
848 return std::equal(FieldsInBindingKey.begin() + Delta,
849 FieldsInBindingKey.end(),
850 Fields.begin());
851 else
852 return std::equal(FieldsInBindingKey.begin(), FieldsInBindingKey.end(),
853 Fields.begin() - Delta);
854 }
855
856 /// Collects all bindings in \p Cluster that may refer to bindings within
857 /// \p Top.
858 ///
859 /// Each binding is a pair whose \c first is the key (a BindingKey) and whose
860 /// \c second is the value (an SVal).
861 ///
862 /// The \p IncludeAllDefaultBindings parameter specifies whether to include
863 /// default bindings that may extend beyond \p Top itself, e.g. if \p Top is
864 /// an aggregate within a larger aggregate with a default binding.
865 static void
collectSubRegionBindings(SmallVectorImpl<BindingPair> & Bindings,SValBuilder & SVB,const ClusterBindings & Cluster,const SubRegion * Top,BindingKey TopKey,bool IncludeAllDefaultBindings)866 collectSubRegionBindings(SmallVectorImpl<BindingPair> &Bindings,
867 SValBuilder &SVB, const ClusterBindings &Cluster,
868 const SubRegion *Top, BindingKey TopKey,
869 bool IncludeAllDefaultBindings) {
870 FieldVector FieldsInSymbolicSubregions;
871 if (TopKey.hasSymbolicOffset()) {
872 getSymbolicOffsetFields(TopKey, FieldsInSymbolicSubregions);
873 Top = TopKey.getConcreteOffsetRegion();
874 TopKey = BindingKey::Make(Top, BindingKey::Default);
875 }
876
877 // Find the length (in bits) of the region being invalidated.
878 uint64_t Length = UINT64_MAX;
879 SVal Extent = Top->getExtent(SVB);
880 if (Optional<nonloc::ConcreteInt> ExtentCI =
881 Extent.getAs<nonloc::ConcreteInt>()) {
882 const llvm::APSInt &ExtentInt = ExtentCI->getValue();
883 assert(ExtentInt.isNonNegative() || ExtentInt.isUnsigned());
884 // Extents are in bytes but region offsets are in bits. Be careful!
885 Length = ExtentInt.getLimitedValue() * SVB.getContext().getCharWidth();
886 } else if (const FieldRegion *FR = dyn_cast<FieldRegion>(Top)) {
887 if (FR->getDecl()->isBitField())
888 Length = FR->getDecl()->getBitWidthValue(SVB.getContext());
889 }
890
891 for (ClusterBindings::iterator I = Cluster.begin(), E = Cluster.end();
892 I != E; ++I) {
893 BindingKey NextKey = I.getKey();
894 if (NextKey.getRegion() == TopKey.getRegion()) {
895 // FIXME: This doesn't catch the case where we're really invalidating a
896 // region with a symbolic offset. Example:
897 // R: points[i].y
898 // Next: points[0].x
899
900 if (NextKey.getOffset() > TopKey.getOffset() &&
901 NextKey.getOffset() - TopKey.getOffset() < Length) {
902 // Case 1: The next binding is inside the region we're invalidating.
903 // Include it.
904 Bindings.push_back(*I);
905
906 } else if (NextKey.getOffset() == TopKey.getOffset()) {
907 // Case 2: The next binding is at the same offset as the region we're
908 // invalidating. In this case, we need to leave default bindings alone,
909 // since they may be providing a default value for a regions beyond what
910 // we're invalidating.
911 // FIXME: This is probably incorrect; consider invalidating an outer
912 // struct whose first field is bound to a LazyCompoundVal.
913 if (IncludeAllDefaultBindings || NextKey.isDirect())
914 Bindings.push_back(*I);
915 }
916
917 } else if (NextKey.hasSymbolicOffset()) {
918 const MemRegion *Base = NextKey.getConcreteOffsetRegion();
919 if (Top->isSubRegionOf(Base) && Top != Base) {
920 // Case 3: The next key is symbolic and we just changed something within
921 // its concrete region. We don't know if the binding is still valid, so
922 // we'll be conservative and include it.
923 if (IncludeAllDefaultBindings || NextKey.isDirect())
924 if (isCompatibleWithFields(NextKey, FieldsInSymbolicSubregions))
925 Bindings.push_back(*I);
926 } else if (const SubRegion *BaseSR = dyn_cast<SubRegion>(Base)) {
927 // Case 4: The next key is symbolic, but we changed a known
928 // super-region. In this case the binding is certainly included.
929 if (BaseSR->isSubRegionOf(Top))
930 if (isCompatibleWithFields(NextKey, FieldsInSymbolicSubregions))
931 Bindings.push_back(*I);
932 }
933 }
934 }
935 }
936
937 static void
collectSubRegionBindings(SmallVectorImpl<BindingPair> & Bindings,SValBuilder & SVB,const ClusterBindings & Cluster,const SubRegion * Top,bool IncludeAllDefaultBindings)938 collectSubRegionBindings(SmallVectorImpl<BindingPair> &Bindings,
939 SValBuilder &SVB, const ClusterBindings &Cluster,
940 const SubRegion *Top, bool IncludeAllDefaultBindings) {
941 collectSubRegionBindings(Bindings, SVB, Cluster, Top,
942 BindingKey::Make(Top, BindingKey::Default),
943 IncludeAllDefaultBindings);
944 }
945
946 RegionBindingsRef
removeSubRegionBindings(RegionBindingsConstRef B,const SubRegion * Top)947 RegionStoreManager::removeSubRegionBindings(RegionBindingsConstRef B,
948 const SubRegion *Top) {
949 BindingKey TopKey = BindingKey::Make(Top, BindingKey::Default);
950 const MemRegion *ClusterHead = TopKey.getBaseRegion();
951
952 if (Top == ClusterHead) {
953 // We can remove an entire cluster's bindings all in one go.
954 return B.remove(Top);
955 }
956
957 const ClusterBindings *Cluster = B.lookup(ClusterHead);
958 if (!Cluster) {
959 // If we're invalidating a region with a symbolic offset, we need to make
960 // sure we don't treat the base region as uninitialized anymore.
961 if (TopKey.hasSymbolicOffset()) {
962 const SubRegion *Concrete = TopKey.getConcreteOffsetRegion();
963 return B.addBinding(Concrete, BindingKey::Default, UnknownVal());
964 }
965 return B;
966 }
967
968 SmallVector<BindingPair, 32> Bindings;
969 collectSubRegionBindings(Bindings, svalBuilder, *Cluster, Top, TopKey,
970 /*IncludeAllDefaultBindings=*/false);
971
972 ClusterBindingsRef Result(*Cluster, CBFactory);
973 for (SmallVectorImpl<BindingPair>::const_iterator I = Bindings.begin(),
974 E = Bindings.end();
975 I != E; ++I)
976 Result = Result.remove(I->first);
977
978 // If we're invalidating a region with a symbolic offset, we need to make sure
979 // we don't treat the base region as uninitialized anymore.
980 // FIXME: This isn't very precise; see the example in
981 // collectSubRegionBindings.
982 if (TopKey.hasSymbolicOffset()) {
983 const SubRegion *Concrete = TopKey.getConcreteOffsetRegion();
984 Result = Result.add(BindingKey::Make(Concrete, BindingKey::Default),
985 UnknownVal());
986 }
987
988 if (Result.isEmpty())
989 return B.remove(ClusterHead);
990 return B.add(ClusterHead, Result.asImmutableMap());
991 }
992
993 namespace {
994 class InvalidateRegionsWorker : public ClusterAnalysis<InvalidateRegionsWorker>
995 {
996 const Expr *Ex;
997 unsigned Count;
998 const LocationContext *LCtx;
999 InvalidatedSymbols &IS;
1000 RegionAndSymbolInvalidationTraits &ITraits;
1001 StoreManager::InvalidatedRegions *Regions;
1002 GlobalsFilterKind GlobalsFilter;
1003 public:
InvalidateRegionsWorker(RegionStoreManager & rm,ProgramStateManager & stateMgr,RegionBindingsRef b,const Expr * ex,unsigned count,const LocationContext * lctx,InvalidatedSymbols & is,RegionAndSymbolInvalidationTraits & ITraitsIn,StoreManager::InvalidatedRegions * r,GlobalsFilterKind GFK)1004 InvalidateRegionsWorker(RegionStoreManager &rm,
1005 ProgramStateManager &stateMgr,
1006 RegionBindingsRef b,
1007 const Expr *ex, unsigned count,
1008 const LocationContext *lctx,
1009 InvalidatedSymbols &is,
1010 RegionAndSymbolInvalidationTraits &ITraitsIn,
1011 StoreManager::InvalidatedRegions *r,
1012 GlobalsFilterKind GFK)
1013 : ClusterAnalysis<InvalidateRegionsWorker>(rm, stateMgr, b),
1014 Ex(ex), Count(count), LCtx(lctx), IS(is), ITraits(ITraitsIn), Regions(r),
1015 GlobalsFilter(GFK) {}
1016
1017 void VisitCluster(const MemRegion *baseR, const ClusterBindings *C);
1018 void VisitBinding(SVal V);
1019
1020 using ClusterAnalysis::AddToWorkList;
1021
1022 bool AddToWorkList(const MemRegion *R);
1023
1024 /// Returns true if all clusters in the memory space for \p Base should be
1025 /// be invalidated.
1026 bool includeEntireMemorySpace(const MemRegion *Base);
1027
1028 /// Returns true if the memory space of the given region is one of the global
1029 /// regions specially included at the start of invalidation.
1030 bool isInitiallyIncludedGlobalRegion(const MemRegion *R);
1031 };
1032 }
1033
AddToWorkList(const MemRegion * R)1034 bool InvalidateRegionsWorker::AddToWorkList(const MemRegion *R) {
1035 bool doNotInvalidateSuperRegion = ITraits.hasTrait(
1036 R, RegionAndSymbolInvalidationTraits::TK_DoNotInvalidateSuperRegion);
1037 const MemRegion *BaseR = doNotInvalidateSuperRegion ? R : R->getBaseRegion();
1038 return AddToWorkList(WorkListElement(BaseR), getCluster(BaseR));
1039 }
1040
VisitBinding(SVal V)1041 void InvalidateRegionsWorker::VisitBinding(SVal V) {
1042 // A symbol? Mark it touched by the invalidation.
1043 if (SymbolRef Sym = V.getAsSymbol())
1044 IS.insert(Sym);
1045
1046 if (const MemRegion *R = V.getAsRegion()) {
1047 AddToWorkList(R);
1048 return;
1049 }
1050
1051 // Is it a LazyCompoundVal? All references get invalidated as well.
1052 if (Optional<nonloc::LazyCompoundVal> LCS =
1053 V.getAs<nonloc::LazyCompoundVal>()) {
1054
1055 const RegionStoreManager::SValListTy &Vals = RM.getInterestingValues(*LCS);
1056
1057 for (RegionStoreManager::SValListTy::const_iterator I = Vals.begin(),
1058 E = Vals.end();
1059 I != E; ++I)
1060 VisitBinding(*I);
1061
1062 return;
1063 }
1064 }
1065
VisitCluster(const MemRegion * baseR,const ClusterBindings * C)1066 void InvalidateRegionsWorker::VisitCluster(const MemRegion *baseR,
1067 const ClusterBindings *C) {
1068
1069 bool PreserveRegionsContents =
1070 ITraits.hasTrait(baseR,
1071 RegionAndSymbolInvalidationTraits::TK_PreserveContents);
1072
1073 if (C) {
1074 for (ClusterBindings::iterator I = C->begin(), E = C->end(); I != E; ++I)
1075 VisitBinding(I.getData());
1076
1077 // Invalidate regions contents.
1078 if (!PreserveRegionsContents)
1079 B = B.remove(baseR);
1080 }
1081
1082 if (const auto *TO = dyn_cast<TypedValueRegion>(baseR)) {
1083 if (const auto *RD = TO->getValueType()->getAsCXXRecordDecl()) {
1084
1085 // Lambdas can affect all static local variables without explicitly
1086 // capturing those.
1087 // We invalidate all static locals referenced inside the lambda body.
1088 if (RD->isLambda() && RD->getLambdaCallOperator()->getBody()) {
1089 using namespace ast_matchers;
1090
1091 const char *DeclBind = "DeclBind";
1092 StatementMatcher RefToStatic = stmt(hasDescendant(declRefExpr(
1093 to(varDecl(hasStaticStorageDuration()).bind(DeclBind)))));
1094 auto Matches =
1095 match(RefToStatic, *RD->getLambdaCallOperator()->getBody(),
1096 RD->getASTContext());
1097
1098 for (BoundNodes &Match : Matches) {
1099 auto *VD = Match.getNodeAs<VarDecl>(DeclBind);
1100 const VarRegion *ToInvalidate =
1101 RM.getRegionManager().getVarRegion(VD, LCtx);
1102 AddToWorkList(ToInvalidate);
1103 }
1104 }
1105 }
1106 }
1107
1108 // BlockDataRegion? If so, invalidate captured variables that are passed
1109 // by reference.
1110 if (const BlockDataRegion *BR = dyn_cast<BlockDataRegion>(baseR)) {
1111 for (BlockDataRegion::referenced_vars_iterator
1112 BI = BR->referenced_vars_begin(), BE = BR->referenced_vars_end() ;
1113 BI != BE; ++BI) {
1114 const VarRegion *VR = BI.getCapturedRegion();
1115 const VarDecl *VD = VR->getDecl();
1116 if (VD->hasAttr<BlocksAttr>() || !VD->hasLocalStorage()) {
1117 AddToWorkList(VR);
1118 }
1119 else if (Loc::isLocType(VR->getValueType())) {
1120 // Map the current bindings to a Store to retrieve the value
1121 // of the binding. If that binding itself is a region, we should
1122 // invalidate that region. This is because a block may capture
1123 // a pointer value, but the thing pointed by that pointer may
1124 // get invalidated.
1125 SVal V = RM.getBinding(B, loc::MemRegionVal(VR));
1126 if (Optional<Loc> L = V.getAs<Loc>()) {
1127 if (const MemRegion *LR = L->getAsRegion())
1128 AddToWorkList(LR);
1129 }
1130 }
1131 }
1132 return;
1133 }
1134
1135 // Symbolic region?
1136 if (const SymbolicRegion *SR = dyn_cast<SymbolicRegion>(baseR))
1137 IS.insert(SR->getSymbol());
1138
1139 // Nothing else should be done in the case when we preserve regions context.
1140 if (PreserveRegionsContents)
1141 return;
1142
1143 // Otherwise, we have a normal data region. Record that we touched the region.
1144 if (Regions)
1145 Regions->push_back(baseR);
1146
1147 if (isa<AllocaRegion>(baseR) || isa<SymbolicRegion>(baseR)) {
1148 // Invalidate the region by setting its default value to
1149 // conjured symbol. The type of the symbol is irrelevant.
1150 DefinedOrUnknownSVal V =
1151 svalBuilder.conjureSymbolVal(baseR, Ex, LCtx, Ctx.IntTy, Count);
1152 B = B.addBinding(baseR, BindingKey::Default, V);
1153 return;
1154 }
1155
1156 if (!baseR->isBoundable())
1157 return;
1158
1159 const TypedValueRegion *TR = cast<TypedValueRegion>(baseR);
1160 QualType T = TR->getValueType();
1161
1162 if (isInitiallyIncludedGlobalRegion(baseR)) {
1163 // If the region is a global and we are invalidating all globals,
1164 // erasing the entry is good enough. This causes all globals to be lazily
1165 // symbolicated from the same base symbol.
1166 return;
1167 }
1168
1169 if (T->isRecordType()) {
1170 // Invalidate the region by setting its default value to
1171 // conjured symbol. The type of the symbol is irrelevant.
1172 DefinedOrUnknownSVal V = svalBuilder.conjureSymbolVal(baseR, Ex, LCtx,
1173 Ctx.IntTy, Count);
1174 B = B.addBinding(baseR, BindingKey::Default, V);
1175 return;
1176 }
1177
1178 if (const ArrayType *AT = Ctx.getAsArrayType(T)) {
1179 bool doNotInvalidateSuperRegion = ITraits.hasTrait(
1180 baseR,
1181 RegionAndSymbolInvalidationTraits::TK_DoNotInvalidateSuperRegion);
1182
1183 if (doNotInvalidateSuperRegion) {
1184 // We are not doing blank invalidation of the whole array region so we
1185 // have to manually invalidate each elements.
1186 Optional<uint64_t> NumElements;
1187
1188 // Compute lower and upper offsets for region within array.
1189 if (const ConstantArrayType *CAT = dyn_cast<ConstantArrayType>(AT))
1190 NumElements = CAT->getSize().getZExtValue();
1191 if (!NumElements) // We are not dealing with a constant size array
1192 goto conjure_default;
1193 QualType ElementTy = AT->getElementType();
1194 uint64_t ElemSize = Ctx.getTypeSize(ElementTy);
1195 const RegionOffset &RO = baseR->getAsOffset();
1196 const MemRegion *SuperR = baseR->getBaseRegion();
1197 if (RO.hasSymbolicOffset()) {
1198 // If base region has a symbolic offset,
1199 // we revert to invalidating the super region.
1200 if (SuperR)
1201 AddToWorkList(SuperR);
1202 goto conjure_default;
1203 }
1204
1205 uint64_t LowerOffset = RO.getOffset();
1206 uint64_t UpperOffset = LowerOffset + *NumElements * ElemSize;
1207 bool UpperOverflow = UpperOffset < LowerOffset;
1208
1209 // Invalidate regions which are within array boundaries,
1210 // or have a symbolic offset.
1211 if (!SuperR)
1212 goto conjure_default;
1213
1214 const ClusterBindings *C = B.lookup(SuperR);
1215 if (!C)
1216 goto conjure_default;
1217
1218 for (ClusterBindings::iterator I = C->begin(), E = C->end(); I != E;
1219 ++I) {
1220 const BindingKey &BK = I.getKey();
1221 Optional<uint64_t> ROffset =
1222 BK.hasSymbolicOffset() ? Optional<uint64_t>() : BK.getOffset();
1223
1224 // Check offset is not symbolic and within array's boundaries.
1225 // Handles arrays of 0 elements and of 0-sized elements as well.
1226 if (!ROffset ||
1227 ((*ROffset >= LowerOffset && *ROffset < UpperOffset) ||
1228 (UpperOverflow &&
1229 (*ROffset >= LowerOffset || *ROffset < UpperOffset)) ||
1230 (LowerOffset == UpperOffset && *ROffset == LowerOffset))) {
1231 B = B.removeBinding(I.getKey());
1232 // Bound symbolic regions need to be invalidated for dead symbol
1233 // detection.
1234 SVal V = I.getData();
1235 const MemRegion *R = V.getAsRegion();
1236 if (R && isa<SymbolicRegion>(R))
1237 VisitBinding(V);
1238 }
1239 }
1240 }
1241 conjure_default:
1242 // Set the default value of the array to conjured symbol.
1243 DefinedOrUnknownSVal V =
1244 svalBuilder.conjureSymbolVal(baseR, Ex, LCtx,
1245 AT->getElementType(), Count);
1246 B = B.addBinding(baseR, BindingKey::Default, V);
1247 return;
1248 }
1249
1250 DefinedOrUnknownSVal V = svalBuilder.conjureSymbolVal(baseR, Ex, LCtx,
1251 T,Count);
1252 assert(SymbolManager::canSymbolicate(T) || V.isUnknown());
1253 B = B.addBinding(baseR, BindingKey::Direct, V);
1254 }
1255
isInitiallyIncludedGlobalRegion(const MemRegion * R)1256 bool InvalidateRegionsWorker::isInitiallyIncludedGlobalRegion(
1257 const MemRegion *R) {
1258 switch (GlobalsFilter) {
1259 case GFK_None:
1260 return false;
1261 case GFK_SystemOnly:
1262 return isa<GlobalSystemSpaceRegion>(R->getMemorySpace());
1263 case GFK_All:
1264 return isa<NonStaticGlobalSpaceRegion>(R->getMemorySpace());
1265 }
1266
1267 llvm_unreachable("unknown globals filter");
1268 }
1269
includeEntireMemorySpace(const MemRegion * Base)1270 bool InvalidateRegionsWorker::includeEntireMemorySpace(const MemRegion *Base) {
1271 if (isInitiallyIncludedGlobalRegion(Base))
1272 return true;
1273
1274 const MemSpaceRegion *MemSpace = Base->getMemorySpace();
1275 return ITraits.hasTrait(MemSpace,
1276 RegionAndSymbolInvalidationTraits::TK_EntireMemSpace);
1277 }
1278
1279 RegionBindingsRef
invalidateGlobalRegion(MemRegion::Kind K,const Expr * Ex,unsigned Count,const LocationContext * LCtx,RegionBindingsRef B,InvalidatedRegions * Invalidated)1280 RegionStoreManager::invalidateGlobalRegion(MemRegion::Kind K,
1281 const Expr *Ex,
1282 unsigned Count,
1283 const LocationContext *LCtx,
1284 RegionBindingsRef B,
1285 InvalidatedRegions *Invalidated) {
1286 // Bind the globals memory space to a new symbol that we will use to derive
1287 // the bindings for all globals.
1288 const GlobalsSpaceRegion *GS = MRMgr.getGlobalsRegion(K);
1289 SVal V = svalBuilder.conjureSymbolVal(/* symbolTag = */ (const void*) GS, Ex, LCtx,
1290 /* type does not matter */ Ctx.IntTy,
1291 Count);
1292
1293 B = B.removeBinding(GS)
1294 .addBinding(BindingKey::Make(GS, BindingKey::Default), V);
1295
1296 // Even if there are no bindings in the global scope, we still need to
1297 // record that we touched it.
1298 if (Invalidated)
1299 Invalidated->push_back(GS);
1300
1301 return B;
1302 }
1303
populateWorkList(InvalidateRegionsWorker & W,ArrayRef<SVal> Values,InvalidatedRegions * TopLevelRegions)1304 void RegionStoreManager::populateWorkList(InvalidateRegionsWorker &W,
1305 ArrayRef<SVal> Values,
1306 InvalidatedRegions *TopLevelRegions) {
1307 for (ArrayRef<SVal>::iterator I = Values.begin(),
1308 E = Values.end(); I != E; ++I) {
1309 SVal V = *I;
1310 if (Optional<nonloc::LazyCompoundVal> LCS =
1311 V.getAs<nonloc::LazyCompoundVal>()) {
1312
1313 const SValListTy &Vals = getInterestingValues(*LCS);
1314
1315 for (SValListTy::const_iterator I = Vals.begin(),
1316 E = Vals.end(); I != E; ++I) {
1317 // Note: the last argument is false here because these are
1318 // non-top-level regions.
1319 if (const MemRegion *R = (*I).getAsRegion())
1320 W.AddToWorkList(R);
1321 }
1322 continue;
1323 }
1324
1325 if (const MemRegion *R = V.getAsRegion()) {
1326 if (TopLevelRegions)
1327 TopLevelRegions->push_back(R);
1328 W.AddToWorkList(R);
1329 continue;
1330 }
1331 }
1332 }
1333
1334 StoreRef
invalidateRegions(Store store,ArrayRef<SVal> Values,const Expr * Ex,unsigned Count,const LocationContext * LCtx,const CallEvent * Call,InvalidatedSymbols & IS,RegionAndSymbolInvalidationTraits & ITraits,InvalidatedRegions * TopLevelRegions,InvalidatedRegions * Invalidated)1335 RegionStoreManager::invalidateRegions(Store store,
1336 ArrayRef<SVal> Values,
1337 const Expr *Ex, unsigned Count,
1338 const LocationContext *LCtx,
1339 const CallEvent *Call,
1340 InvalidatedSymbols &IS,
1341 RegionAndSymbolInvalidationTraits &ITraits,
1342 InvalidatedRegions *TopLevelRegions,
1343 InvalidatedRegions *Invalidated) {
1344 GlobalsFilterKind GlobalsFilter;
1345 if (Call) {
1346 if (Call->isInSystemHeader())
1347 GlobalsFilter = GFK_SystemOnly;
1348 else
1349 GlobalsFilter = GFK_All;
1350 } else {
1351 GlobalsFilter = GFK_None;
1352 }
1353
1354 RegionBindingsRef B = getRegionBindings(store);
1355 InvalidateRegionsWorker W(*this, StateMgr, B, Ex, Count, LCtx, IS, ITraits,
1356 Invalidated, GlobalsFilter);
1357
1358 // Scan the bindings and generate the clusters.
1359 W.GenerateClusters();
1360
1361 // Add the regions to the worklist.
1362 populateWorkList(W, Values, TopLevelRegions);
1363
1364 W.RunWorkList();
1365
1366 // Return the new bindings.
1367 B = W.getRegionBindings();
1368
1369 // For calls, determine which global regions should be invalidated and
1370 // invalidate them. (Note that function-static and immutable globals are never
1371 // invalidated by this.)
1372 // TODO: This could possibly be more precise with modules.
1373 switch (GlobalsFilter) {
1374 case GFK_All:
1375 B = invalidateGlobalRegion(MemRegion::GlobalInternalSpaceRegionKind,
1376 Ex, Count, LCtx, B, Invalidated);
1377 LLVM_FALLTHROUGH;
1378 case GFK_SystemOnly:
1379 B = invalidateGlobalRegion(MemRegion::GlobalSystemSpaceRegionKind,
1380 Ex, Count, LCtx, B, Invalidated);
1381 LLVM_FALLTHROUGH;
1382 case GFK_None:
1383 break;
1384 }
1385
1386 return StoreRef(B.asStore(), *this);
1387 }
1388
1389 //===----------------------------------------------------------------------===//
1390 // Extents for regions.
1391 //===----------------------------------------------------------------------===//
1392
1393 DefinedOrUnknownSVal
getSizeInElements(ProgramStateRef state,const MemRegion * R,QualType EleTy)1394 RegionStoreManager::getSizeInElements(ProgramStateRef state,
1395 const MemRegion *R,
1396 QualType EleTy) {
1397 SVal Size = cast<SubRegion>(R)->getExtent(svalBuilder);
1398 const llvm::APSInt *SizeInt = svalBuilder.getKnownValue(state, Size);
1399 if (!SizeInt)
1400 return UnknownVal();
1401
1402 CharUnits RegionSize = CharUnits::fromQuantity(SizeInt->getSExtValue());
1403
1404 if (Ctx.getAsVariableArrayType(EleTy)) {
1405 // FIXME: We need to track extra state to properly record the size
1406 // of VLAs. Returning UnknownVal here, however, is a stop-gap so that
1407 // we don't have a divide-by-zero below.
1408 return UnknownVal();
1409 }
1410
1411 CharUnits EleSize = Ctx.getTypeSizeInChars(EleTy);
1412
1413 // If a variable is reinterpreted as a type that doesn't fit into a larger
1414 // type evenly, round it down.
1415 // This is a signed value, since it's used in arithmetic with signed indices.
1416 return svalBuilder.makeIntVal(RegionSize / EleSize,
1417 svalBuilder.getArrayIndexType());
1418 }
1419
1420 //===----------------------------------------------------------------------===//
1421 // Location and region casting.
1422 //===----------------------------------------------------------------------===//
1423
1424 /// ArrayToPointer - Emulates the "decay" of an array to a pointer
1425 /// type. 'Array' represents the lvalue of the array being decayed
1426 /// to a pointer, and the returned SVal represents the decayed
1427 /// version of that lvalue (i.e., a pointer to the first element of
1428 /// the array). This is called by ExprEngine when evaluating casts
1429 /// from arrays to pointers.
ArrayToPointer(Loc Array,QualType T)1430 SVal RegionStoreManager::ArrayToPointer(Loc Array, QualType T) {
1431 if (Array.getAs<loc::ConcreteInt>())
1432 return Array;
1433
1434 if (!Array.getAs<loc::MemRegionVal>())
1435 return UnknownVal();
1436
1437 const SubRegion *R =
1438 cast<SubRegion>(Array.castAs<loc::MemRegionVal>().getRegion());
1439 NonLoc ZeroIdx = svalBuilder.makeZeroArrayIndex();
1440 return loc::MemRegionVal(MRMgr.getElementRegion(T, ZeroIdx, R, Ctx));
1441 }
1442
1443 //===----------------------------------------------------------------------===//
1444 // Loading values from regions.
1445 //===----------------------------------------------------------------------===//
1446
getBinding(RegionBindingsConstRef B,Loc L,QualType T)1447 SVal RegionStoreManager::getBinding(RegionBindingsConstRef B, Loc L, QualType T) {
1448 assert(!L.getAs<UnknownVal>() && "location unknown");
1449 assert(!L.getAs<UndefinedVal>() && "location undefined");
1450
1451 // For access to concrete addresses, return UnknownVal. Checks
1452 // for null dereferences (and similar errors) are done by checkers, not
1453 // the Store.
1454 // FIXME: We can consider lazily symbolicating such memory, but we really
1455 // should defer this when we can reason easily about symbolicating arrays
1456 // of bytes.
1457 if (L.getAs<loc::ConcreteInt>()) {
1458 return UnknownVal();
1459 }
1460 if (!L.getAs<loc::MemRegionVal>()) {
1461 return UnknownVal();
1462 }
1463
1464 const MemRegion *MR = L.castAs<loc::MemRegionVal>().getRegion();
1465
1466 if (isa<BlockDataRegion>(MR)) {
1467 return UnknownVal();
1468 }
1469
1470 if (!isa<TypedValueRegion>(MR)) {
1471 if (T.isNull()) {
1472 if (const TypedRegion *TR = dyn_cast<TypedRegion>(MR))
1473 T = TR->getLocationType()->getPointeeType();
1474 else if (const SymbolicRegion *SR = dyn_cast<SymbolicRegion>(MR))
1475 T = SR->getSymbol()->getType()->getPointeeType();
1476 }
1477 assert(!T.isNull() && "Unable to auto-detect binding type!");
1478 assert(!T->isVoidType() && "Attempting to dereference a void pointer!");
1479 MR = GetElementZeroRegion(cast<SubRegion>(MR), T);
1480 } else {
1481 T = cast<TypedValueRegion>(MR)->getValueType();
1482 }
1483
1484 // FIXME: Perhaps this method should just take a 'const MemRegion*' argument
1485 // instead of 'Loc', and have the other Loc cases handled at a higher level.
1486 const TypedValueRegion *R = cast<TypedValueRegion>(MR);
1487 QualType RTy = R->getValueType();
1488
1489 // FIXME: we do not yet model the parts of a complex type, so treat the
1490 // whole thing as "unknown".
1491 if (RTy->isAnyComplexType())
1492 return UnknownVal();
1493
1494 // FIXME: We should eventually handle funny addressing. e.g.:
1495 //
1496 // int x = ...;
1497 // int *p = &x;
1498 // char *q = (char*) p;
1499 // char c = *q; // returns the first byte of 'x'.
1500 //
1501 // Such funny addressing will occur due to layering of regions.
1502 if (RTy->isStructureOrClassType())
1503 return getBindingForStruct(B, R);
1504
1505 // FIXME: Handle unions.
1506 if (RTy->isUnionType())
1507 return createLazyBinding(B, R);
1508
1509 if (RTy->isArrayType()) {
1510 if (RTy->isConstantArrayType())
1511 return getBindingForArray(B, R);
1512 else
1513 return UnknownVal();
1514 }
1515
1516 // FIXME: handle Vector types.
1517 if (RTy->isVectorType())
1518 return UnknownVal();
1519
1520 if (const FieldRegion* FR = dyn_cast<FieldRegion>(R))
1521 return CastRetrievedVal(getBindingForField(B, FR), FR, T);
1522
1523 if (const ElementRegion* ER = dyn_cast<ElementRegion>(R)) {
1524 // FIXME: Here we actually perform an implicit conversion from the loaded
1525 // value to the element type. Eventually we want to compose these values
1526 // more intelligently. For example, an 'element' can encompass multiple
1527 // bound regions (e.g., several bound bytes), or could be a subset of
1528 // a larger value.
1529 return CastRetrievedVal(getBindingForElement(B, ER), ER, T);
1530 }
1531
1532 if (const ObjCIvarRegion *IVR = dyn_cast<ObjCIvarRegion>(R)) {
1533 // FIXME: Here we actually perform an implicit conversion from the loaded
1534 // value to the ivar type. What we should model is stores to ivars
1535 // that blow past the extent of the ivar. If the address of the ivar is
1536 // reinterpretted, it is possible we stored a different value that could
1537 // fit within the ivar. Either we need to cast these when storing them
1538 // or reinterpret them lazily (as we do here).
1539 return CastRetrievedVal(getBindingForObjCIvar(B, IVR), IVR, T);
1540 }
1541
1542 if (const VarRegion *VR = dyn_cast<VarRegion>(R)) {
1543 // FIXME: Here we actually perform an implicit conversion from the loaded
1544 // value to the variable type. What we should model is stores to variables
1545 // that blow past the extent of the variable. If the address of the
1546 // variable is reinterpretted, it is possible we stored a different value
1547 // that could fit within the variable. Either we need to cast these when
1548 // storing them or reinterpret them lazily (as we do here).
1549 return CastRetrievedVal(getBindingForVar(B, VR), VR, T);
1550 }
1551
1552 const SVal *V = B.lookup(R, BindingKey::Direct);
1553
1554 // Check if the region has a binding.
1555 if (V)
1556 return *V;
1557
1558 // The location does not have a bound value. This means that it has
1559 // the value it had upon its creation and/or entry to the analyzed
1560 // function/method. These are either symbolic values or 'undefined'.
1561 if (R->hasStackNonParametersStorage()) {
1562 // All stack variables are considered to have undefined values
1563 // upon creation. All heap allocated blocks are considered to
1564 // have undefined values as well unless they are explicitly bound
1565 // to specific values.
1566 return UndefinedVal();
1567 }
1568
1569 // All other values are symbolic.
1570 return svalBuilder.getRegionValueSymbolVal(R);
1571 }
1572
getUnderlyingType(const SubRegion * R)1573 static QualType getUnderlyingType(const SubRegion *R) {
1574 QualType RegionTy;
1575 if (const TypedValueRegion *TVR = dyn_cast<TypedValueRegion>(R))
1576 RegionTy = TVR->getValueType();
1577
1578 if (const SymbolicRegion *SR = dyn_cast<SymbolicRegion>(R))
1579 RegionTy = SR->getSymbol()->getType();
1580
1581 return RegionTy;
1582 }
1583
1584 /// Checks to see if store \p B has a lazy binding for region \p R.
1585 ///
1586 /// If \p AllowSubregionBindings is \c false, a lazy binding will be rejected
1587 /// if there are additional bindings within \p R.
1588 ///
1589 /// Note that unlike RegionStoreManager::findLazyBinding, this will not search
1590 /// for lazy bindings for super-regions of \p R.
1591 static Optional<nonloc::LazyCompoundVal>
getExistingLazyBinding(SValBuilder & SVB,RegionBindingsConstRef B,const SubRegion * R,bool AllowSubregionBindings)1592 getExistingLazyBinding(SValBuilder &SVB, RegionBindingsConstRef B,
1593 const SubRegion *R, bool AllowSubregionBindings) {
1594 Optional<SVal> V = B.getDefaultBinding(R);
1595 if (!V)
1596 return None;
1597
1598 Optional<nonloc::LazyCompoundVal> LCV = V->getAs<nonloc::LazyCompoundVal>();
1599 if (!LCV)
1600 return None;
1601
1602 // If the LCV is for a subregion, the types might not match, and we shouldn't
1603 // reuse the binding.
1604 QualType RegionTy = getUnderlyingType(R);
1605 if (!RegionTy.isNull() &&
1606 !RegionTy->isVoidPointerType()) {
1607 QualType SourceRegionTy = LCV->getRegion()->getValueType();
1608 if (!SVB.getContext().hasSameUnqualifiedType(RegionTy, SourceRegionTy))
1609 return None;
1610 }
1611
1612 if (!AllowSubregionBindings) {
1613 // If there are any other bindings within this region, we shouldn't reuse
1614 // the top-level binding.
1615 SmallVector<BindingPair, 16> Bindings;
1616 collectSubRegionBindings(Bindings, SVB, *B.lookup(R->getBaseRegion()), R,
1617 /*IncludeAllDefaultBindings=*/true);
1618 if (Bindings.size() > 1)
1619 return None;
1620 }
1621
1622 return *LCV;
1623 }
1624
1625
1626 std::pair<Store, const SubRegion *>
findLazyBinding(RegionBindingsConstRef B,const SubRegion * R,const SubRegion * originalRegion)1627 RegionStoreManager::findLazyBinding(RegionBindingsConstRef B,
1628 const SubRegion *R,
1629 const SubRegion *originalRegion) {
1630 if (originalRegion != R) {
1631 if (Optional<nonloc::LazyCompoundVal> V =
1632 getExistingLazyBinding(svalBuilder, B, R, true))
1633 return std::make_pair(V->getStore(), V->getRegion());
1634 }
1635
1636 typedef std::pair<Store, const SubRegion *> StoreRegionPair;
1637 StoreRegionPair Result = StoreRegionPair();
1638
1639 if (const ElementRegion *ER = dyn_cast<ElementRegion>(R)) {
1640 Result = findLazyBinding(B, cast<SubRegion>(ER->getSuperRegion()),
1641 originalRegion);
1642
1643 if (Result.second)
1644 Result.second = MRMgr.getElementRegionWithSuper(ER, Result.second);
1645
1646 } else if (const FieldRegion *FR = dyn_cast<FieldRegion>(R)) {
1647 Result = findLazyBinding(B, cast<SubRegion>(FR->getSuperRegion()),
1648 originalRegion);
1649
1650 if (Result.second)
1651 Result.second = MRMgr.getFieldRegionWithSuper(FR, Result.second);
1652
1653 } else if (const CXXBaseObjectRegion *BaseReg =
1654 dyn_cast<CXXBaseObjectRegion>(R)) {
1655 // C++ base object region is another kind of region that we should blast
1656 // through to look for lazy compound value. It is like a field region.
1657 Result = findLazyBinding(B, cast<SubRegion>(BaseReg->getSuperRegion()),
1658 originalRegion);
1659
1660 if (Result.second)
1661 Result.second = MRMgr.getCXXBaseObjectRegionWithSuper(BaseReg,
1662 Result.second);
1663 }
1664
1665 return Result;
1666 }
1667
getBindingForElement(RegionBindingsConstRef B,const ElementRegion * R)1668 SVal RegionStoreManager::getBindingForElement(RegionBindingsConstRef B,
1669 const ElementRegion* R) {
1670 // We do not currently model bindings of the CompoundLiteralregion.
1671 if (isa<CompoundLiteralRegion>(R->getBaseRegion()))
1672 return UnknownVal();
1673
1674 // Check if the region has a binding.
1675 if (const Optional<SVal> &V = B.getDirectBinding(R))
1676 return *V;
1677
1678 const MemRegion* superR = R->getSuperRegion();
1679
1680 // Check if the region is an element region of a string literal.
1681 if (const StringRegion *StrR = dyn_cast<StringRegion>(superR)) {
1682 // FIXME: Handle loads from strings where the literal is treated as
1683 // an integer, e.g., *((unsigned int*)"hello")
1684 QualType T = Ctx.getAsArrayType(StrR->getValueType())->getElementType();
1685 if (!Ctx.hasSameUnqualifiedType(T, R->getElementType()))
1686 return UnknownVal();
1687
1688 const StringLiteral *Str = StrR->getStringLiteral();
1689 SVal Idx = R->getIndex();
1690 if (Optional<nonloc::ConcreteInt> CI = Idx.getAs<nonloc::ConcreteInt>()) {
1691 int64_t i = CI->getValue().getSExtValue();
1692 // Abort on string underrun. This can be possible by arbitrary
1693 // clients of getBindingForElement().
1694 if (i < 0)
1695 return UndefinedVal();
1696 int64_t length = Str->getLength();
1697 // Technically, only i == length is guaranteed to be null.
1698 // However, such overflows should be caught before reaching this point;
1699 // the only time such an access would be made is if a string literal was
1700 // used to initialize a larger array.
1701 char c = (i >= length) ? '\0' : Str->getCodeUnit(i);
1702 return svalBuilder.makeIntVal(c, T);
1703 }
1704 } else if (const VarRegion *VR = dyn_cast<VarRegion>(superR)) {
1705 // Check if the containing array has an initialized value that we can trust.
1706 // We can trust a const value or a value of a global initializer in main().
1707 const VarDecl *VD = VR->getDecl();
1708 if (VD->getType().isConstQualified() ||
1709 R->getElementType().isConstQualified() ||
1710 (B.isMainAnalysis() && VD->hasGlobalStorage())) {
1711 if (const Expr *Init = VD->getAnyInitializer()) {
1712 if (const auto *InitList = dyn_cast<InitListExpr>(Init)) {
1713 // The array index has to be known.
1714 if (auto CI = R->getIndex().getAs<nonloc::ConcreteInt>()) {
1715 int64_t i = CI->getValue().getSExtValue();
1716 // If it is known that the index is out of bounds, we can return
1717 // an undefined value.
1718 if (i < 0)
1719 return UndefinedVal();
1720
1721 if (auto CAT = Ctx.getAsConstantArrayType(VD->getType()))
1722 if (CAT->getSize().sle(i))
1723 return UndefinedVal();
1724
1725 // If there is a list, but no init, it must be zero.
1726 if (i >= InitList->getNumInits())
1727 return svalBuilder.makeZeroVal(R->getElementType());
1728
1729 if (const Expr *ElemInit = InitList->getInit(i))
1730 if (Optional<SVal> V = svalBuilder.getConstantVal(ElemInit))
1731 return *V;
1732 }
1733 }
1734 }
1735 }
1736 }
1737
1738 // Check for loads from a code text region. For such loads, just give up.
1739 if (isa<CodeTextRegion>(superR))
1740 return UnknownVal();
1741
1742 // Handle the case where we are indexing into a larger scalar object.
1743 // For example, this handles:
1744 // int x = ...
1745 // char *y = &x;
1746 // return *y;
1747 // FIXME: This is a hack, and doesn't do anything really intelligent yet.
1748 const RegionRawOffset &O = R->getAsArrayOffset();
1749
1750 // If we cannot reason about the offset, return an unknown value.
1751 if (!O.getRegion())
1752 return UnknownVal();
1753
1754 if (const TypedValueRegion *baseR =
1755 dyn_cast_or_null<TypedValueRegion>(O.getRegion())) {
1756 QualType baseT = baseR->getValueType();
1757 if (baseT->isScalarType()) {
1758 QualType elemT = R->getElementType();
1759 if (elemT->isScalarType()) {
1760 if (Ctx.getTypeSizeInChars(baseT) >= Ctx.getTypeSizeInChars(elemT)) {
1761 if (const Optional<SVal> &V = B.getDirectBinding(superR)) {
1762 if (SymbolRef parentSym = V->getAsSymbol())
1763 return svalBuilder.getDerivedRegionValueSymbolVal(parentSym, R);
1764
1765 if (V->isUnknownOrUndef())
1766 return *V;
1767 // Other cases: give up. We are indexing into a larger object
1768 // that has some value, but we don't know how to handle that yet.
1769 return UnknownVal();
1770 }
1771 }
1772 }
1773 }
1774 }
1775 return getBindingForFieldOrElementCommon(B, R, R->getElementType());
1776 }
1777
getBindingForField(RegionBindingsConstRef B,const FieldRegion * R)1778 SVal RegionStoreManager::getBindingForField(RegionBindingsConstRef B,
1779 const FieldRegion* R) {
1780
1781 // Check if the region has a binding.
1782 if (const Optional<SVal> &V = B.getDirectBinding(R))
1783 return *V;
1784
1785 // Is the field declared constant and has an in-class initializer?
1786 const FieldDecl *FD = R->getDecl();
1787 QualType Ty = FD->getType();
1788 if (Ty.isConstQualified())
1789 if (const Expr *Init = FD->getInClassInitializer())
1790 if (Optional<SVal> V = svalBuilder.getConstantVal(Init))
1791 return *V;
1792
1793 // If the containing record was initialized, try to get its constant value.
1794 const MemRegion* superR = R->getSuperRegion();
1795 if (const auto *VR = dyn_cast<VarRegion>(superR)) {
1796 const VarDecl *VD = VR->getDecl();
1797 QualType RecordVarTy = VD->getType();
1798 unsigned Index = FD->getFieldIndex();
1799 // Either the record variable or the field has an initializer that we can
1800 // trust. We trust initializers of constants and, additionally, respect
1801 // initializers of globals when analyzing main().
1802 if (RecordVarTy.isConstQualified() || Ty.isConstQualified() ||
1803 (B.isMainAnalysis() && VD->hasGlobalStorage()))
1804 if (const Expr *Init = VD->getAnyInitializer())
1805 if (const auto *InitList = dyn_cast<InitListExpr>(Init)) {
1806 if (Index < InitList->getNumInits()) {
1807 if (const Expr *FieldInit = InitList->getInit(Index))
1808 if (Optional<SVal> V = svalBuilder.getConstantVal(FieldInit))
1809 return *V;
1810 } else {
1811 return svalBuilder.makeZeroVal(Ty);
1812 }
1813 }
1814 }
1815
1816 return getBindingForFieldOrElementCommon(B, R, Ty);
1817 }
1818
1819 Optional<SVal>
getBindingForDerivedDefaultValue(RegionBindingsConstRef B,const MemRegion * superR,const TypedValueRegion * R,QualType Ty)1820 RegionStoreManager::getBindingForDerivedDefaultValue(RegionBindingsConstRef B,
1821 const MemRegion *superR,
1822 const TypedValueRegion *R,
1823 QualType Ty) {
1824
1825 if (const Optional<SVal> &D = B.getDefaultBinding(superR)) {
1826 const SVal &val = D.getValue();
1827 if (SymbolRef parentSym = val.getAsSymbol())
1828 return svalBuilder.getDerivedRegionValueSymbolVal(parentSym, R);
1829
1830 if (val.isZeroConstant())
1831 return svalBuilder.makeZeroVal(Ty);
1832
1833 if (val.isUnknownOrUndef())
1834 return val;
1835
1836 // Lazy bindings are usually handled through getExistingLazyBinding().
1837 // We should unify these two code paths at some point.
1838 if (val.getAs<nonloc::LazyCompoundVal>() ||
1839 val.getAs<nonloc::CompoundVal>())
1840 return val;
1841
1842 llvm_unreachable("Unknown default value");
1843 }
1844
1845 return None;
1846 }
1847
getLazyBinding(const SubRegion * LazyBindingRegion,RegionBindingsRef LazyBinding)1848 SVal RegionStoreManager::getLazyBinding(const SubRegion *LazyBindingRegion,
1849 RegionBindingsRef LazyBinding) {
1850 SVal Result;
1851 if (const ElementRegion *ER = dyn_cast<ElementRegion>(LazyBindingRegion))
1852 Result = getBindingForElement(LazyBinding, ER);
1853 else
1854 Result = getBindingForField(LazyBinding,
1855 cast<FieldRegion>(LazyBindingRegion));
1856
1857 // FIXME: This is a hack to deal with RegionStore's inability to distinguish a
1858 // default value for /part/ of an aggregate from a default value for the
1859 // /entire/ aggregate. The most common case of this is when struct Outer
1860 // has as its first member a struct Inner, which is copied in from a stack
1861 // variable. In this case, even if the Outer's default value is symbolic, 0,
1862 // or unknown, it gets overridden by the Inner's default value of undefined.
1863 //
1864 // This is a general problem -- if the Inner is zero-initialized, the Outer
1865 // will now look zero-initialized. The proper way to solve this is with a
1866 // new version of RegionStore that tracks the extent of a binding as well
1867 // as the offset.
1868 //
1869 // This hack only takes care of the undefined case because that can very
1870 // quickly result in a warning.
1871 if (Result.isUndef())
1872 Result = UnknownVal();
1873
1874 return Result;
1875 }
1876
1877 SVal
getBindingForFieldOrElementCommon(RegionBindingsConstRef B,const TypedValueRegion * R,QualType Ty)1878 RegionStoreManager::getBindingForFieldOrElementCommon(RegionBindingsConstRef B,
1879 const TypedValueRegion *R,
1880 QualType Ty) {
1881
1882 // At this point we have already checked in either getBindingForElement or
1883 // getBindingForField if 'R' has a direct binding.
1884
1885 // Lazy binding?
1886 Store lazyBindingStore = nullptr;
1887 const SubRegion *lazyBindingRegion = nullptr;
1888 std::tie(lazyBindingStore, lazyBindingRegion) = findLazyBinding(B, R, R);
1889 if (lazyBindingRegion)
1890 return getLazyBinding(lazyBindingRegion,
1891 getRegionBindings(lazyBindingStore));
1892
1893 // Record whether or not we see a symbolic index. That can completely
1894 // be out of scope of our lookup.
1895 bool hasSymbolicIndex = false;
1896
1897 // FIXME: This is a hack to deal with RegionStore's inability to distinguish a
1898 // default value for /part/ of an aggregate from a default value for the
1899 // /entire/ aggregate. The most common case of this is when struct Outer
1900 // has as its first member a struct Inner, which is copied in from a stack
1901 // variable. In this case, even if the Outer's default value is symbolic, 0,
1902 // or unknown, it gets overridden by the Inner's default value of undefined.
1903 //
1904 // This is a general problem -- if the Inner is zero-initialized, the Outer
1905 // will now look zero-initialized. The proper way to solve this is with a
1906 // new version of RegionStore that tracks the extent of a binding as well
1907 // as the offset.
1908 //
1909 // This hack only takes care of the undefined case because that can very
1910 // quickly result in a warning.
1911 bool hasPartialLazyBinding = false;
1912
1913 const SubRegion *SR = R;
1914 while (SR) {
1915 const MemRegion *Base = SR->getSuperRegion();
1916 if (Optional<SVal> D = getBindingForDerivedDefaultValue(B, Base, R, Ty)) {
1917 if (D->getAs<nonloc::LazyCompoundVal>()) {
1918 hasPartialLazyBinding = true;
1919 break;
1920 }
1921
1922 return *D;
1923 }
1924
1925 if (const ElementRegion *ER = dyn_cast<ElementRegion>(Base)) {
1926 NonLoc index = ER->getIndex();
1927 if (!index.isConstant())
1928 hasSymbolicIndex = true;
1929 }
1930
1931 // If our super region is a field or element itself, walk up the region
1932 // hierarchy to see if there is a default value installed in an ancestor.
1933 SR = dyn_cast<SubRegion>(Base);
1934 }
1935
1936 if (R->hasStackNonParametersStorage()) {
1937 if (isa<ElementRegion>(R)) {
1938 // Currently we don't reason specially about Clang-style vectors. Check
1939 // if superR is a vector and if so return Unknown.
1940 if (const TypedValueRegion *typedSuperR =
1941 dyn_cast<TypedValueRegion>(R->getSuperRegion())) {
1942 if (typedSuperR->getValueType()->isVectorType())
1943 return UnknownVal();
1944 }
1945 }
1946
1947 // FIXME: We also need to take ElementRegions with symbolic indexes into
1948 // account. This case handles both directly accessing an ElementRegion
1949 // with a symbolic offset, but also fields within an element with
1950 // a symbolic offset.
1951 if (hasSymbolicIndex)
1952 return UnknownVal();
1953
1954 // Additionally allow introspection of a block's internal layout.
1955 if (!hasPartialLazyBinding && !isa<BlockDataRegion>(R->getBaseRegion()))
1956 return UndefinedVal();
1957 }
1958
1959 // All other values are symbolic.
1960 return svalBuilder.getRegionValueSymbolVal(R);
1961 }
1962
getBindingForObjCIvar(RegionBindingsConstRef B,const ObjCIvarRegion * R)1963 SVal RegionStoreManager::getBindingForObjCIvar(RegionBindingsConstRef B,
1964 const ObjCIvarRegion* R) {
1965 // Check if the region has a binding.
1966 if (const Optional<SVal> &V = B.getDirectBinding(R))
1967 return *V;
1968
1969 const MemRegion *superR = R->getSuperRegion();
1970
1971 // Check if the super region has a default binding.
1972 if (const Optional<SVal> &V = B.getDefaultBinding(superR)) {
1973 if (SymbolRef parentSym = V->getAsSymbol())
1974 return svalBuilder.getDerivedRegionValueSymbolVal(parentSym, R);
1975
1976 // Other cases: give up.
1977 return UnknownVal();
1978 }
1979
1980 return getBindingForLazySymbol(R);
1981 }
1982
getBindingForVar(RegionBindingsConstRef B,const VarRegion * R)1983 SVal RegionStoreManager::getBindingForVar(RegionBindingsConstRef B,
1984 const VarRegion *R) {
1985
1986 // Check if the region has a binding.
1987 if (Optional<SVal> V = B.getDirectBinding(R))
1988 return *V;
1989
1990 if (Optional<SVal> V = B.getDefaultBinding(R))
1991 return *V;
1992
1993 // Lazily derive a value for the VarRegion.
1994 const VarDecl *VD = R->getDecl();
1995 const MemSpaceRegion *MS = R->getMemorySpace();
1996
1997 // Arguments are always symbolic.
1998 if (isa<StackArgumentsSpaceRegion>(MS))
1999 return svalBuilder.getRegionValueSymbolVal(R);
2000
2001 // Is 'VD' declared constant? If so, retrieve the constant value.
2002 if (VD->getType().isConstQualified()) {
2003 if (const Expr *Init = VD->getAnyInitializer()) {
2004 if (Optional<SVal> V = svalBuilder.getConstantVal(Init))
2005 return *V;
2006
2007 // If the variable is const qualified and has an initializer but
2008 // we couldn't evaluate initializer to a value, treat the value as
2009 // unknown.
2010 return UnknownVal();
2011 }
2012 }
2013
2014 // This must come after the check for constants because closure-captured
2015 // constant variables may appear in UnknownSpaceRegion.
2016 if (isa<UnknownSpaceRegion>(MS))
2017 return svalBuilder.getRegionValueSymbolVal(R);
2018
2019 if (isa<GlobalsSpaceRegion>(MS)) {
2020 QualType T = VD->getType();
2021
2022 // If we're in main(), then global initializers have not become stale yet.
2023 if (B.isMainAnalysis())
2024 if (const Expr *Init = VD->getAnyInitializer())
2025 if (Optional<SVal> V = svalBuilder.getConstantVal(Init))
2026 return *V;
2027
2028 // Function-scoped static variables are default-initialized to 0; if they
2029 // have an initializer, it would have been processed by now.
2030 // FIXME: This is only true when we're starting analysis from main().
2031 // We're losing a lot of coverage here.
2032 if (isa<StaticGlobalSpaceRegion>(MS))
2033 return svalBuilder.makeZeroVal(T);
2034
2035 if (Optional<SVal> V = getBindingForDerivedDefaultValue(B, MS, R, T)) {
2036 assert(!V->getAs<nonloc::LazyCompoundVal>());
2037 return V.getValue();
2038 }
2039
2040 return svalBuilder.getRegionValueSymbolVal(R);
2041 }
2042
2043 return UndefinedVal();
2044 }
2045
getBindingForLazySymbol(const TypedValueRegion * R)2046 SVal RegionStoreManager::getBindingForLazySymbol(const TypedValueRegion *R) {
2047 // All other values are symbolic.
2048 return svalBuilder.getRegionValueSymbolVal(R);
2049 }
2050
2051 const RegionStoreManager::SValListTy &
getInterestingValues(nonloc::LazyCompoundVal LCV)2052 RegionStoreManager::getInterestingValues(nonloc::LazyCompoundVal LCV) {
2053 // First, check the cache.
2054 LazyBindingsMapTy::iterator I = LazyBindingsMap.find(LCV.getCVData());
2055 if (I != LazyBindingsMap.end())
2056 return I->second;
2057
2058 // If we don't have a list of values cached, start constructing it.
2059 SValListTy List;
2060
2061 const SubRegion *LazyR = LCV.getRegion();
2062 RegionBindingsRef B = getRegionBindings(LCV.getStore());
2063
2064 // If this region had /no/ bindings at the time, there are no interesting
2065 // values to return.
2066 const ClusterBindings *Cluster = B.lookup(LazyR->getBaseRegion());
2067 if (!Cluster)
2068 return (LazyBindingsMap[LCV.getCVData()] = std::move(List));
2069
2070 SmallVector<BindingPair, 32> Bindings;
2071 collectSubRegionBindings(Bindings, svalBuilder, *Cluster, LazyR,
2072 /*IncludeAllDefaultBindings=*/true);
2073 for (SmallVectorImpl<BindingPair>::const_iterator I = Bindings.begin(),
2074 E = Bindings.end();
2075 I != E; ++I) {
2076 SVal V = I->second;
2077 if (V.isUnknownOrUndef() || V.isConstant())
2078 continue;
2079
2080 if (Optional<nonloc::LazyCompoundVal> InnerLCV =
2081 V.getAs<nonloc::LazyCompoundVal>()) {
2082 const SValListTy &InnerList = getInterestingValues(*InnerLCV);
2083 List.insert(List.end(), InnerList.begin(), InnerList.end());
2084 continue;
2085 }
2086
2087 List.push_back(V);
2088 }
2089
2090 return (LazyBindingsMap[LCV.getCVData()] = std::move(List));
2091 }
2092
createLazyBinding(RegionBindingsConstRef B,const TypedValueRegion * R)2093 NonLoc RegionStoreManager::createLazyBinding(RegionBindingsConstRef B,
2094 const TypedValueRegion *R) {
2095 if (Optional<nonloc::LazyCompoundVal> V =
2096 getExistingLazyBinding(svalBuilder, B, R, false))
2097 return *V;
2098
2099 return svalBuilder.makeLazyCompoundVal(StoreRef(B.asStore(), *this), R);
2100 }
2101
isRecordEmpty(const RecordDecl * RD)2102 static bool isRecordEmpty(const RecordDecl *RD) {
2103 if (!RD->field_empty())
2104 return false;
2105 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD))
2106 return CRD->getNumBases() == 0;
2107 return true;
2108 }
2109
getBindingForStruct(RegionBindingsConstRef B,const TypedValueRegion * R)2110 SVal RegionStoreManager::getBindingForStruct(RegionBindingsConstRef B,
2111 const TypedValueRegion *R) {
2112 const RecordDecl *RD = R->getValueType()->castAs<RecordType>()->getDecl();
2113 if (!RD->getDefinition() || isRecordEmpty(RD))
2114 return UnknownVal();
2115
2116 return createLazyBinding(B, R);
2117 }
2118
getBindingForArray(RegionBindingsConstRef B,const TypedValueRegion * R)2119 SVal RegionStoreManager::getBindingForArray(RegionBindingsConstRef B,
2120 const TypedValueRegion *R) {
2121 assert(Ctx.getAsConstantArrayType(R->getValueType()) &&
2122 "Only constant array types can have compound bindings.");
2123
2124 return createLazyBinding(B, R);
2125 }
2126
includedInBindings(Store store,const MemRegion * region) const2127 bool RegionStoreManager::includedInBindings(Store store,
2128 const MemRegion *region) const {
2129 RegionBindingsRef B = getRegionBindings(store);
2130 region = region->getBaseRegion();
2131
2132 // Quick path: if the base is the head of a cluster, the region is live.
2133 if (B.lookup(region))
2134 return true;
2135
2136 // Slow path: if the region is the VALUE of any binding, it is live.
2137 for (RegionBindingsRef::iterator RI = B.begin(), RE = B.end(); RI != RE; ++RI) {
2138 const ClusterBindings &Cluster = RI.getData();
2139 for (ClusterBindings::iterator CI = Cluster.begin(), CE = Cluster.end();
2140 CI != CE; ++CI) {
2141 const SVal &D = CI.getData();
2142 if (const MemRegion *R = D.getAsRegion())
2143 if (R->getBaseRegion() == region)
2144 return true;
2145 }
2146 }
2147
2148 return false;
2149 }
2150
2151 //===----------------------------------------------------------------------===//
2152 // Binding values to regions.
2153 //===----------------------------------------------------------------------===//
2154
killBinding(Store ST,Loc L)2155 StoreRef RegionStoreManager::killBinding(Store ST, Loc L) {
2156 if (Optional<loc::MemRegionVal> LV = L.getAs<loc::MemRegionVal>())
2157 if (const MemRegion* R = LV->getRegion())
2158 return StoreRef(getRegionBindings(ST).removeBinding(R)
2159 .asImmutableMap()
2160 .getRootWithoutRetain(),
2161 *this);
2162
2163 return StoreRef(ST, *this);
2164 }
2165
2166 RegionBindingsRef
bind(RegionBindingsConstRef B,Loc L,SVal V)2167 RegionStoreManager::bind(RegionBindingsConstRef B, Loc L, SVal V) {
2168 if (L.getAs<loc::ConcreteInt>())
2169 return B;
2170
2171 // If we get here, the location should be a region.
2172 const MemRegion *R = L.castAs<loc::MemRegionVal>().getRegion();
2173
2174 // Check if the region is a struct region.
2175 if (const TypedValueRegion* TR = dyn_cast<TypedValueRegion>(R)) {
2176 QualType Ty = TR->getValueType();
2177 if (Ty->isArrayType())
2178 return bindArray(B, TR, V);
2179 if (Ty->isStructureOrClassType())
2180 return bindStruct(B, TR, V);
2181 if (Ty->isVectorType())
2182 return bindVector(B, TR, V);
2183 if (Ty->isUnionType())
2184 return bindAggregate(B, TR, V);
2185 }
2186
2187 if (const SymbolicRegion *SR = dyn_cast<SymbolicRegion>(R)) {
2188 // Binding directly to a symbolic region should be treated as binding
2189 // to element 0.
2190 QualType T = SR->getSymbol()->getType();
2191 if (T->isAnyPointerType() || T->isReferenceType())
2192 T = T->getPointeeType();
2193
2194 R = GetElementZeroRegion(SR, T);
2195 }
2196
2197 assert((!isa<CXXThisRegion>(R) || !B.lookup(R)) &&
2198 "'this' pointer is not an l-value and is not assignable");
2199
2200 // Clear out bindings that may overlap with this binding.
2201 RegionBindingsRef NewB = removeSubRegionBindings(B, cast<SubRegion>(R));
2202 return NewB.addBinding(BindingKey::Make(R, BindingKey::Direct), V);
2203 }
2204
2205 RegionBindingsRef
setImplicitDefaultValue(RegionBindingsConstRef B,const MemRegion * R,QualType T)2206 RegionStoreManager::setImplicitDefaultValue(RegionBindingsConstRef B,
2207 const MemRegion *R,
2208 QualType T) {
2209 SVal V;
2210
2211 if (Loc::isLocType(T))
2212 V = svalBuilder.makeNull();
2213 else if (T->isIntegralOrEnumerationType())
2214 V = svalBuilder.makeZeroVal(T);
2215 else if (T->isStructureOrClassType() || T->isArrayType()) {
2216 // Set the default value to a zero constant when it is a structure
2217 // or array. The type doesn't really matter.
2218 V = svalBuilder.makeZeroVal(Ctx.IntTy);
2219 }
2220 else {
2221 // We can't represent values of this type, but we still need to set a value
2222 // to record that the region has been initialized.
2223 // If this assertion ever fires, a new case should be added above -- we
2224 // should know how to default-initialize any value we can symbolicate.
2225 assert(!SymbolManager::canSymbolicate(T) && "This type is representable");
2226 V = UnknownVal();
2227 }
2228
2229 return B.addBinding(R, BindingKey::Default, V);
2230 }
2231
2232 RegionBindingsRef
bindArray(RegionBindingsConstRef B,const TypedValueRegion * R,SVal Init)2233 RegionStoreManager::bindArray(RegionBindingsConstRef B,
2234 const TypedValueRegion* R,
2235 SVal Init) {
2236
2237 const ArrayType *AT =cast<ArrayType>(Ctx.getCanonicalType(R->getValueType()));
2238 QualType ElementTy = AT->getElementType();
2239 Optional<uint64_t> Size;
2240
2241 if (const ConstantArrayType* CAT = dyn_cast<ConstantArrayType>(AT))
2242 Size = CAT->getSize().getZExtValue();
2243
2244 // Check if the init expr is a literal. If so, bind the rvalue instead.
2245 // FIXME: It's not responsibility of the Store to transform this lvalue
2246 // to rvalue. ExprEngine or maybe even CFG should do this before binding.
2247 if (Optional<loc::MemRegionVal> MRV = Init.getAs<loc::MemRegionVal>()) {
2248 SVal V = getBinding(B.asStore(), *MRV, R->getValueType());
2249 return bindAggregate(B, R, V);
2250 }
2251
2252 // Handle lazy compound values.
2253 if (Init.getAs<nonloc::LazyCompoundVal>())
2254 return bindAggregate(B, R, Init);
2255
2256 if (Init.isUnknown())
2257 return bindAggregate(B, R, UnknownVal());
2258
2259 // Remaining case: explicit compound values.
2260 const nonloc::CompoundVal& CV = Init.castAs<nonloc::CompoundVal>();
2261 nonloc::CompoundVal::iterator VI = CV.begin(), VE = CV.end();
2262 uint64_t i = 0;
2263
2264 RegionBindingsRef NewB(B);
2265
2266 for (; Size.hasValue() ? i < Size.getValue() : true ; ++i, ++VI) {
2267 // The init list might be shorter than the array length.
2268 if (VI == VE)
2269 break;
2270
2271 const NonLoc &Idx = svalBuilder.makeArrayIndex(i);
2272 const ElementRegion *ER = MRMgr.getElementRegion(ElementTy, Idx, R, Ctx);
2273
2274 if (ElementTy->isStructureOrClassType())
2275 NewB = bindStruct(NewB, ER, *VI);
2276 else if (ElementTy->isArrayType())
2277 NewB = bindArray(NewB, ER, *VI);
2278 else
2279 NewB = bind(NewB, loc::MemRegionVal(ER), *VI);
2280 }
2281
2282 // If the init list is shorter than the array length (or the array has
2283 // variable length), set the array default value. Values that are already set
2284 // are not overwritten.
2285 if (!Size.hasValue() || i < Size.getValue())
2286 NewB = setImplicitDefaultValue(NewB, R, ElementTy);
2287
2288 return NewB;
2289 }
2290
bindVector(RegionBindingsConstRef B,const TypedValueRegion * R,SVal V)2291 RegionBindingsRef RegionStoreManager::bindVector(RegionBindingsConstRef B,
2292 const TypedValueRegion* R,
2293 SVal V) {
2294 QualType T = R->getValueType();
2295 const VectorType *VT = T->castAs<VectorType>(); // Use castAs for typedefs.
2296
2297 // Handle lazy compound values and symbolic values.
2298 if (V.getAs<nonloc::LazyCompoundVal>() || V.getAs<nonloc::SymbolVal>())
2299 return bindAggregate(B, R, V);
2300
2301 // We may get non-CompoundVal accidentally due to imprecise cast logic or
2302 // that we are binding symbolic struct value. Kill the field values, and if
2303 // the value is symbolic go and bind it as a "default" binding.
2304 if (!V.getAs<nonloc::CompoundVal>()) {
2305 return bindAggregate(B, R, UnknownVal());
2306 }
2307
2308 QualType ElemType = VT->getElementType();
2309 nonloc::CompoundVal CV = V.castAs<nonloc::CompoundVal>();
2310 nonloc::CompoundVal::iterator VI = CV.begin(), VE = CV.end();
2311 unsigned index = 0, numElements = VT->getNumElements();
2312 RegionBindingsRef NewB(B);
2313
2314 for ( ; index != numElements ; ++index) {
2315 if (VI == VE)
2316 break;
2317
2318 NonLoc Idx = svalBuilder.makeArrayIndex(index);
2319 const ElementRegion *ER = MRMgr.getElementRegion(ElemType, Idx, R, Ctx);
2320
2321 if (ElemType->isArrayType())
2322 NewB = bindArray(NewB, ER, *VI);
2323 else if (ElemType->isStructureOrClassType())
2324 NewB = bindStruct(NewB, ER, *VI);
2325 else
2326 NewB = bind(NewB, loc::MemRegionVal(ER), *VI);
2327 }
2328 return NewB;
2329 }
2330
2331 Optional<RegionBindingsRef>
tryBindSmallStruct(RegionBindingsConstRef B,const TypedValueRegion * R,const RecordDecl * RD,nonloc::LazyCompoundVal LCV)2332 RegionStoreManager::tryBindSmallStruct(RegionBindingsConstRef B,
2333 const TypedValueRegion *R,
2334 const RecordDecl *RD,
2335 nonloc::LazyCompoundVal LCV) {
2336 FieldVector Fields;
2337
2338 if (const CXXRecordDecl *Class = dyn_cast<CXXRecordDecl>(RD))
2339 if (Class->getNumBases() != 0 || Class->getNumVBases() != 0)
2340 return None;
2341
2342 for (const auto *FD : RD->fields()) {
2343 if (FD->isUnnamedBitfield())
2344 continue;
2345
2346 // If there are too many fields, or if any of the fields are aggregates,
2347 // just use the LCV as a default binding.
2348 if (Fields.size() == SmallStructLimit)
2349 return None;
2350
2351 QualType Ty = FD->getType();
2352 if (!(Ty->isScalarType() || Ty->isReferenceType()))
2353 return None;
2354
2355 Fields.push_back(FD);
2356 }
2357
2358 RegionBindingsRef NewB = B;
2359
2360 for (FieldVector::iterator I = Fields.begin(), E = Fields.end(); I != E; ++I){
2361 const FieldRegion *SourceFR = MRMgr.getFieldRegion(*I, LCV.getRegion());
2362 SVal V = getBindingForField(getRegionBindings(LCV.getStore()), SourceFR);
2363
2364 const FieldRegion *DestFR = MRMgr.getFieldRegion(*I, R);
2365 NewB = bind(NewB, loc::MemRegionVal(DestFR), V);
2366 }
2367
2368 return NewB;
2369 }
2370
bindStruct(RegionBindingsConstRef B,const TypedValueRegion * R,SVal V)2371 RegionBindingsRef RegionStoreManager::bindStruct(RegionBindingsConstRef B,
2372 const TypedValueRegion* R,
2373 SVal V) {
2374 if (!Features.supportsFields())
2375 return B;
2376
2377 QualType T = R->getValueType();
2378 assert(T->isStructureOrClassType());
2379
2380 const RecordType* RT = T->castAs<RecordType>();
2381 const RecordDecl *RD = RT->getDecl();
2382
2383 if (!RD->isCompleteDefinition())
2384 return B;
2385
2386 // Handle lazy compound values and symbolic values.
2387 if (Optional<nonloc::LazyCompoundVal> LCV =
2388 V.getAs<nonloc::LazyCompoundVal>()) {
2389 if (Optional<RegionBindingsRef> NewB = tryBindSmallStruct(B, R, RD, *LCV))
2390 return *NewB;
2391 return bindAggregate(B, R, V);
2392 }
2393 if (V.getAs<nonloc::SymbolVal>())
2394 return bindAggregate(B, R, V);
2395
2396 // We may get non-CompoundVal accidentally due to imprecise cast logic or
2397 // that we are binding symbolic struct value. Kill the field values, and if
2398 // the value is symbolic go and bind it as a "default" binding.
2399 if (V.isUnknown() || !V.getAs<nonloc::CompoundVal>())
2400 return bindAggregate(B, R, UnknownVal());
2401
2402 // The raw CompoundVal is essentially a symbolic InitListExpr: an (immutable)
2403 // list of other values. It appears pretty much only when there's an actual
2404 // initializer list expression in the program, and the analyzer tries to
2405 // unwrap it as soon as possible.
2406 // This code is where such unwrap happens: when the compound value is put into
2407 // the object that it was supposed to initialize (it's an *initializer* list,
2408 // after all), instead of binding the whole value to the whole object, we bind
2409 // sub-values to sub-objects. Sub-values may themselves be compound values,
2410 // and in this case the procedure becomes recursive.
2411 // FIXME: The annoying part about compound values is that they don't carry
2412 // any sort of information about which value corresponds to which sub-object.
2413 // It's simply a list of values in the middle of nowhere; we expect to match
2414 // them to sub-objects, essentially, "by index": first value binds to
2415 // the first field, second value binds to the second field, etc.
2416 // It would have been much safer to organize non-lazy compound values as
2417 // a mapping from fields/bases to values.
2418 const nonloc::CompoundVal& CV = V.castAs<nonloc::CompoundVal>();
2419 nonloc::CompoundVal::iterator VI = CV.begin(), VE = CV.end();
2420
2421 RegionBindingsRef NewB(B);
2422
2423 // In C++17 aggregates may have base classes, handle those as well.
2424 // They appear before fields in the initializer list / compound value.
2425 if (const auto *CRD = dyn_cast<CXXRecordDecl>(RD)) {
2426 // If the object was constructed with a constructor, its value is a
2427 // LazyCompoundVal. If it's a raw CompoundVal, it means that we're
2428 // performing aggregate initialization. The only exception from this
2429 // rule is sending an Objective-C++ message that returns a C++ object
2430 // to a nil receiver; in this case the semantics is to return a
2431 // zero-initialized object even if it's a C++ object that doesn't have
2432 // this sort of constructor; the CompoundVal is empty in this case.
2433 assert((CRD->isAggregate() || (Ctx.getLangOpts().ObjC && VI == VE)) &&
2434 "Non-aggregates are constructed with a constructor!");
2435
2436 for (const auto &B : CRD->bases()) {
2437 // (Multiple inheritance is fine though.)
2438 assert(!B.isVirtual() && "Aggregates cannot have virtual base classes!");
2439
2440 if (VI == VE)
2441 break;
2442
2443 QualType BTy = B.getType();
2444 assert(BTy->isStructureOrClassType() && "Base classes must be classes!");
2445
2446 const CXXRecordDecl *BRD = BTy->getAsCXXRecordDecl();
2447 assert(BRD && "Base classes must be C++ classes!");
2448
2449 const CXXBaseObjectRegion *BR =
2450 MRMgr.getCXXBaseObjectRegion(BRD, R, /*IsVirtual=*/false);
2451
2452 NewB = bindStruct(NewB, BR, *VI);
2453
2454 ++VI;
2455 }
2456 }
2457
2458 RecordDecl::field_iterator FI, FE;
2459
2460 for (FI = RD->field_begin(), FE = RD->field_end(); FI != FE; ++FI) {
2461
2462 if (VI == VE)
2463 break;
2464
2465 // Skip any unnamed bitfields to stay in sync with the initializers.
2466 if (FI->isUnnamedBitfield())
2467 continue;
2468
2469 QualType FTy = FI->getType();
2470 const FieldRegion* FR = MRMgr.getFieldRegion(*FI, R);
2471
2472 if (FTy->isArrayType())
2473 NewB = bindArray(NewB, FR, *VI);
2474 else if (FTy->isStructureOrClassType())
2475 NewB = bindStruct(NewB, FR, *VI);
2476 else
2477 NewB = bind(NewB, loc::MemRegionVal(FR), *VI);
2478 ++VI;
2479 }
2480
2481 // There may be fewer values in the initialize list than the fields of struct.
2482 if (FI != FE) {
2483 NewB = NewB.addBinding(R, BindingKey::Default,
2484 svalBuilder.makeIntVal(0, false));
2485 }
2486
2487 return NewB;
2488 }
2489
2490 RegionBindingsRef
bindAggregate(RegionBindingsConstRef B,const TypedRegion * R,SVal Val)2491 RegionStoreManager::bindAggregate(RegionBindingsConstRef B,
2492 const TypedRegion *R,
2493 SVal Val) {
2494 // Remove the old bindings, using 'R' as the root of all regions
2495 // we will invalidate. Then add the new binding.
2496 return removeSubRegionBindings(B, R).addBinding(R, BindingKey::Default, Val);
2497 }
2498
2499 //===----------------------------------------------------------------------===//
2500 // State pruning.
2501 //===----------------------------------------------------------------------===//
2502
2503 namespace {
2504 class RemoveDeadBindingsWorker
2505 : public ClusterAnalysis<RemoveDeadBindingsWorker> {
2506 SmallVector<const SymbolicRegion *, 12> Postponed;
2507 SymbolReaper &SymReaper;
2508 const StackFrameContext *CurrentLCtx;
2509
2510 public:
RemoveDeadBindingsWorker(RegionStoreManager & rm,ProgramStateManager & stateMgr,RegionBindingsRef b,SymbolReaper & symReaper,const StackFrameContext * LCtx)2511 RemoveDeadBindingsWorker(RegionStoreManager &rm,
2512 ProgramStateManager &stateMgr,
2513 RegionBindingsRef b, SymbolReaper &symReaper,
2514 const StackFrameContext *LCtx)
2515 : ClusterAnalysis<RemoveDeadBindingsWorker>(rm, stateMgr, b),
2516 SymReaper(symReaper), CurrentLCtx(LCtx) {}
2517
2518 // Called by ClusterAnalysis.
2519 void VisitAddedToCluster(const MemRegion *baseR, const ClusterBindings &C);
2520 void VisitCluster(const MemRegion *baseR, const ClusterBindings *C);
2521 using ClusterAnalysis<RemoveDeadBindingsWorker>::VisitCluster;
2522
2523 using ClusterAnalysis::AddToWorkList;
2524
2525 bool AddToWorkList(const MemRegion *R);
2526
2527 bool UpdatePostponed();
2528 void VisitBinding(SVal V);
2529 };
2530 }
2531
AddToWorkList(const MemRegion * R)2532 bool RemoveDeadBindingsWorker::AddToWorkList(const MemRegion *R) {
2533 const MemRegion *BaseR = R->getBaseRegion();
2534 return AddToWorkList(WorkListElement(BaseR), getCluster(BaseR));
2535 }
2536
VisitAddedToCluster(const MemRegion * baseR,const ClusterBindings & C)2537 void RemoveDeadBindingsWorker::VisitAddedToCluster(const MemRegion *baseR,
2538 const ClusterBindings &C) {
2539
2540 if (const VarRegion *VR = dyn_cast<VarRegion>(baseR)) {
2541 if (SymReaper.isLive(VR))
2542 AddToWorkList(baseR, &C);
2543
2544 return;
2545 }
2546
2547 if (const SymbolicRegion *SR = dyn_cast<SymbolicRegion>(baseR)) {
2548 if (SymReaper.isLive(SR->getSymbol()))
2549 AddToWorkList(SR, &C);
2550 else
2551 Postponed.push_back(SR);
2552
2553 return;
2554 }
2555
2556 if (isa<NonStaticGlobalSpaceRegion>(baseR)) {
2557 AddToWorkList(baseR, &C);
2558 return;
2559 }
2560
2561 // CXXThisRegion in the current or parent location context is live.
2562 if (const CXXThisRegion *TR = dyn_cast<CXXThisRegion>(baseR)) {
2563 const auto *StackReg =
2564 cast<StackArgumentsSpaceRegion>(TR->getSuperRegion());
2565 const StackFrameContext *RegCtx = StackReg->getStackFrame();
2566 if (CurrentLCtx &&
2567 (RegCtx == CurrentLCtx || RegCtx->isParentOf(CurrentLCtx)))
2568 AddToWorkList(TR, &C);
2569 }
2570 }
2571
VisitCluster(const MemRegion * baseR,const ClusterBindings * C)2572 void RemoveDeadBindingsWorker::VisitCluster(const MemRegion *baseR,
2573 const ClusterBindings *C) {
2574 if (!C)
2575 return;
2576
2577 // Mark the symbol for any SymbolicRegion with live bindings as live itself.
2578 // This means we should continue to track that symbol.
2579 if (const SymbolicRegion *SymR = dyn_cast<SymbolicRegion>(baseR))
2580 SymReaper.markLive(SymR->getSymbol());
2581
2582 for (ClusterBindings::iterator I = C->begin(), E = C->end(); I != E; ++I) {
2583 // Element index of a binding key is live.
2584 SymReaper.markElementIndicesLive(I.getKey().getRegion());
2585
2586 VisitBinding(I.getData());
2587 }
2588 }
2589
VisitBinding(SVal V)2590 void RemoveDeadBindingsWorker::VisitBinding(SVal V) {
2591 // Is it a LazyCompoundVal? All referenced regions are live as well.
2592 if (Optional<nonloc::LazyCompoundVal> LCS =
2593 V.getAs<nonloc::LazyCompoundVal>()) {
2594
2595 const RegionStoreManager::SValListTy &Vals = RM.getInterestingValues(*LCS);
2596
2597 for (RegionStoreManager::SValListTy::const_iterator I = Vals.begin(),
2598 E = Vals.end();
2599 I != E; ++I)
2600 VisitBinding(*I);
2601
2602 return;
2603 }
2604
2605 // If V is a region, then add it to the worklist.
2606 if (const MemRegion *R = V.getAsRegion()) {
2607 AddToWorkList(R);
2608 SymReaper.markLive(R);
2609
2610 // All regions captured by a block are also live.
2611 if (const BlockDataRegion *BR = dyn_cast<BlockDataRegion>(R)) {
2612 BlockDataRegion::referenced_vars_iterator I = BR->referenced_vars_begin(),
2613 E = BR->referenced_vars_end();
2614 for ( ; I != E; ++I)
2615 AddToWorkList(I.getCapturedRegion());
2616 }
2617 }
2618
2619
2620 // Update the set of live symbols.
2621 for (auto SI = V.symbol_begin(), SE = V.symbol_end(); SI!=SE; ++SI)
2622 SymReaper.markLive(*SI);
2623 }
2624
UpdatePostponed()2625 bool RemoveDeadBindingsWorker::UpdatePostponed() {
2626 // See if any postponed SymbolicRegions are actually live now, after
2627 // having done a scan.
2628 bool Changed = false;
2629
2630 for (auto I = Postponed.begin(), E = Postponed.end(); I != E; ++I) {
2631 if (const SymbolicRegion *SR = *I) {
2632 if (SymReaper.isLive(SR->getSymbol())) {
2633 Changed |= AddToWorkList(SR);
2634 *I = nullptr;
2635 }
2636 }
2637 }
2638
2639 return Changed;
2640 }
2641
removeDeadBindings(Store store,const StackFrameContext * LCtx,SymbolReaper & SymReaper)2642 StoreRef RegionStoreManager::removeDeadBindings(Store store,
2643 const StackFrameContext *LCtx,
2644 SymbolReaper& SymReaper) {
2645 RegionBindingsRef B = getRegionBindings(store);
2646 RemoveDeadBindingsWorker W(*this, StateMgr, B, SymReaper, LCtx);
2647 W.GenerateClusters();
2648
2649 // Enqueue the region roots onto the worklist.
2650 for (SymbolReaper::region_iterator I = SymReaper.region_begin(),
2651 E = SymReaper.region_end(); I != E; ++I) {
2652 W.AddToWorkList(*I);
2653 }
2654
2655 do W.RunWorkList(); while (W.UpdatePostponed());
2656
2657 // We have now scanned the store, marking reachable regions and symbols
2658 // as live. We now remove all the regions that are dead from the store
2659 // as well as update DSymbols with the set symbols that are now dead.
2660 for (RegionBindingsRef::iterator I = B.begin(), E = B.end(); I != E; ++I) {
2661 const MemRegion *Base = I.getKey();
2662
2663 // If the cluster has been visited, we know the region has been marked.
2664 // Otherwise, remove the dead entry.
2665 if (!W.isVisited(Base))
2666 B = B.remove(Base);
2667 }
2668
2669 return StoreRef(B.asStore(), *this);
2670 }
2671
2672 //===----------------------------------------------------------------------===//
2673 // Utility methods.
2674 //===----------------------------------------------------------------------===//
2675
printJson(raw_ostream & Out,Store S,const char * NL,unsigned int Space,bool IsDot) const2676 void RegionStoreManager::printJson(raw_ostream &Out, Store S, const char *NL,
2677 unsigned int Space, bool IsDot) const {
2678 RegionBindingsRef Bindings = getRegionBindings(S);
2679
2680 Indent(Out, Space, IsDot) << "\"store\": ";
2681
2682 if (Bindings.isEmpty()) {
2683 Out << "null," << NL;
2684 return;
2685 }
2686
2687 Out << "{ \"pointer\": \"" << Bindings.asStore() << "\", \"items\": [" << NL;
2688 Bindings.printJson(Out, NL, Space + 1, IsDot);
2689 Indent(Out, Space, IsDot) << "]}," << NL;
2690 }
2691