1 //== RegionStore.cpp - Field-sensitive store model --------------*- C++ -*--==//
2 //
3 // The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This file defines a basic region store model. In this model, we do have field
11 // sensitivity. But we assume nothing about the heap shape. So recursive data
12 // structures are largely ignored. Basically we do 1-limiting analysis.
13 // Parameter pointers are assumed with no aliasing. Pointee objects of
14 // parameters are created lazily.
15 //
16 //===----------------------------------------------------------------------===//
17 #include "clang/AST/Attr.h"
18 #include "clang/AST/CharUnits.h"
19 #include "clang/Analysis/Analyses/LiveVariables.h"
20 #include "clang/Analysis/AnalysisContext.h"
21 #include "clang/Basic/TargetInfo.h"
22 #include "clang/StaticAnalyzer/Core/PathSensitive/CallEvent.h"
23 #include "clang/StaticAnalyzer/Core/PathSensitive/MemRegion.h"
24 #include "clang/StaticAnalyzer/Core/PathSensitive/ProgramState.h"
25 #include "clang/StaticAnalyzer/Core/PathSensitive/ProgramStateTrait.h"
26 #include "llvm/ADT/ImmutableList.h"
27 #include "llvm/ADT/ImmutableMap.h"
28 #include "llvm/ADT/Optional.h"
29 #include "llvm/Support/raw_ostream.h"
30
31 using namespace clang;
32 using namespace ento;
33
34 //===----------------------------------------------------------------------===//
35 // Representation of binding keys.
36 //===----------------------------------------------------------------------===//
37
38 namespace {
39 class BindingKey {
40 public:
41 enum Kind { Default = 0x0, Direct = 0x1 };
42 private:
43 enum { Symbolic = 0x2 };
44
45 llvm::PointerIntPair<const MemRegion *, 2> P;
46 uint64_t Data;
47
48 /// Create a key for a binding to region \p r, which has a symbolic offset
49 /// from region \p Base.
BindingKey(const SubRegion * r,const SubRegion * Base,Kind k)50 explicit BindingKey(const SubRegion *r, const SubRegion *Base, Kind k)
51 : P(r, k | Symbolic), Data(reinterpret_cast<uintptr_t>(Base)) {
52 assert(r && Base && "Must have known regions.");
53 assert(getConcreteOffsetRegion() == Base && "Failed to store base region");
54 }
55
56 /// Create a key for a binding at \p offset from base region \p r.
BindingKey(const MemRegion * r,uint64_t offset,Kind k)57 explicit BindingKey(const MemRegion *r, uint64_t offset, Kind k)
58 : P(r, k), Data(offset) {
59 assert(r && "Must have known regions.");
60 assert(getOffset() == offset && "Failed to store offset");
61 assert((r == r->getBaseRegion() || isa<ObjCIvarRegion>(r)) && "Not a base");
62 }
63 public:
64
isDirect() const65 bool isDirect() const { return P.getInt() & Direct; }
hasSymbolicOffset() const66 bool hasSymbolicOffset() const { return P.getInt() & Symbolic; }
67
getRegion() const68 const MemRegion *getRegion() const { return P.getPointer(); }
getOffset() const69 uint64_t getOffset() const {
70 assert(!hasSymbolicOffset());
71 return Data;
72 }
73
getConcreteOffsetRegion() const74 const SubRegion *getConcreteOffsetRegion() const {
75 assert(hasSymbolicOffset());
76 return reinterpret_cast<const SubRegion *>(static_cast<uintptr_t>(Data));
77 }
78
getBaseRegion() const79 const MemRegion *getBaseRegion() const {
80 if (hasSymbolicOffset())
81 return getConcreteOffsetRegion()->getBaseRegion();
82 return getRegion()->getBaseRegion();
83 }
84
Profile(llvm::FoldingSetNodeID & ID) const85 void Profile(llvm::FoldingSetNodeID& ID) const {
86 ID.AddPointer(P.getOpaqueValue());
87 ID.AddInteger(Data);
88 }
89
90 static BindingKey Make(const MemRegion *R, Kind k);
91
operator <(const BindingKey & X) const92 bool operator<(const BindingKey &X) const {
93 if (P.getOpaqueValue() < X.P.getOpaqueValue())
94 return true;
95 if (P.getOpaqueValue() > X.P.getOpaqueValue())
96 return false;
97 return Data < X.Data;
98 }
99
operator ==(const BindingKey & X) const100 bool operator==(const BindingKey &X) const {
101 return P.getOpaqueValue() == X.P.getOpaqueValue() &&
102 Data == X.Data;
103 }
104
105 LLVM_ATTRIBUTE_USED void dump() const;
106 };
107 } // end anonymous namespace
108
Make(const MemRegion * R,Kind k)109 BindingKey BindingKey::Make(const MemRegion *R, Kind k) {
110 const RegionOffset &RO = R->getAsOffset();
111 if (RO.hasSymbolicOffset())
112 return BindingKey(cast<SubRegion>(R), cast<SubRegion>(RO.getRegion()), k);
113
114 return BindingKey(RO.getRegion(), RO.getOffset(), k);
115 }
116
117 namespace llvm {
118 static inline
operator <<(raw_ostream & os,BindingKey K)119 raw_ostream &operator<<(raw_ostream &os, BindingKey K) {
120 os << '(' << K.getRegion();
121 if (!K.hasSymbolicOffset())
122 os << ',' << K.getOffset();
123 os << ',' << (K.isDirect() ? "direct" : "default")
124 << ')';
125 return os;
126 }
127
128 template <typename T> struct isPodLike;
129 template <> struct isPodLike<BindingKey> {
130 static const bool value = true;
131 };
132 } // end llvm namespace
133
dump() const134 void BindingKey::dump() const {
135 llvm::errs() << *this;
136 }
137
138 //===----------------------------------------------------------------------===//
139 // Actual Store type.
140 //===----------------------------------------------------------------------===//
141
142 typedef llvm::ImmutableMap<BindingKey, SVal> ClusterBindings;
143 typedef llvm::ImmutableMapRef<BindingKey, SVal> ClusterBindingsRef;
144 typedef std::pair<BindingKey, SVal> BindingPair;
145
146 typedef llvm::ImmutableMap<const MemRegion *, ClusterBindings>
147 RegionBindings;
148
149 namespace {
150 class RegionBindingsRef : public llvm::ImmutableMapRef<const MemRegion *,
151 ClusterBindings> {
152 ClusterBindings::Factory &CBFactory;
153 public:
154 typedef llvm::ImmutableMapRef<const MemRegion *, ClusterBindings>
155 ParentTy;
156
RegionBindingsRef(ClusterBindings::Factory & CBFactory,const RegionBindings::TreeTy * T,RegionBindings::TreeTy::Factory * F)157 RegionBindingsRef(ClusterBindings::Factory &CBFactory,
158 const RegionBindings::TreeTy *T,
159 RegionBindings::TreeTy::Factory *F)
160 : llvm::ImmutableMapRef<const MemRegion *, ClusterBindings>(T, F),
161 CBFactory(CBFactory) {}
162
RegionBindingsRef(const ParentTy & P,ClusterBindings::Factory & CBFactory)163 RegionBindingsRef(const ParentTy &P, ClusterBindings::Factory &CBFactory)
164 : llvm::ImmutableMapRef<const MemRegion *, ClusterBindings>(P),
165 CBFactory(CBFactory) {}
166
add(key_type_ref K,data_type_ref D) const167 RegionBindingsRef add(key_type_ref K, data_type_ref D) const {
168 return RegionBindingsRef(static_cast<const ParentTy*>(this)->add(K, D),
169 CBFactory);
170 }
171
remove(key_type_ref K) const172 RegionBindingsRef remove(key_type_ref K) const {
173 return RegionBindingsRef(static_cast<const ParentTy*>(this)->remove(K),
174 CBFactory);
175 }
176
177 RegionBindingsRef addBinding(BindingKey K, SVal V) const;
178
179 RegionBindingsRef addBinding(const MemRegion *R,
180 BindingKey::Kind k, SVal V) const;
181
operator =(const RegionBindingsRef & X)182 RegionBindingsRef &operator=(const RegionBindingsRef &X) {
183 *static_cast<ParentTy*>(this) = X;
184 return *this;
185 }
186
187 const SVal *lookup(BindingKey K) const;
188 const SVal *lookup(const MemRegion *R, BindingKey::Kind k) const;
lookup(const MemRegion * R) const189 const ClusterBindings *lookup(const MemRegion *R) const {
190 return static_cast<const ParentTy*>(this)->lookup(R);
191 }
192
193 RegionBindingsRef removeBinding(BindingKey K);
194
195 RegionBindingsRef removeBinding(const MemRegion *R,
196 BindingKey::Kind k);
197
removeBinding(const MemRegion * R)198 RegionBindingsRef removeBinding(const MemRegion *R) {
199 return removeBinding(R, BindingKey::Direct).
200 removeBinding(R, BindingKey::Default);
201 }
202
203 Optional<SVal> getDirectBinding(const MemRegion *R) const;
204
205 /// getDefaultBinding - Returns an SVal* representing an optional default
206 /// binding associated with a region and its subregions.
207 Optional<SVal> getDefaultBinding(const MemRegion *R) const;
208
209 /// Return the internal tree as a Store.
asStore() const210 Store asStore() const {
211 return asImmutableMap().getRootWithoutRetain();
212 }
213
dump(raw_ostream & OS,const char * nl) const214 void dump(raw_ostream &OS, const char *nl) const {
215 for (iterator I = begin(), E = end(); I != E; ++I) {
216 const ClusterBindings &Cluster = I.getData();
217 for (ClusterBindings::iterator CI = Cluster.begin(), CE = Cluster.end();
218 CI != CE; ++CI) {
219 OS << ' ' << CI.getKey() << " : " << CI.getData() << nl;
220 }
221 OS << nl;
222 }
223 }
224
dump() const225 LLVM_ATTRIBUTE_USED void dump() const {
226 dump(llvm::errs(), "\n");
227 }
228 };
229 } // end anonymous namespace
230
231 typedef const RegionBindingsRef& RegionBindingsConstRef;
232
getDirectBinding(const MemRegion * R) const233 Optional<SVal> RegionBindingsRef::getDirectBinding(const MemRegion *R) const {
234 return Optional<SVal>::create(lookup(R, BindingKey::Direct));
235 }
236
getDefaultBinding(const MemRegion * R) const237 Optional<SVal> RegionBindingsRef::getDefaultBinding(const MemRegion *R) const {
238 if (R->isBoundable())
239 if (const TypedValueRegion *TR = dyn_cast<TypedValueRegion>(R))
240 if (TR->getValueType()->isUnionType())
241 return UnknownVal();
242
243 return Optional<SVal>::create(lookup(R, BindingKey::Default));
244 }
245
addBinding(BindingKey K,SVal V) const246 RegionBindingsRef RegionBindingsRef::addBinding(BindingKey K, SVal V) const {
247 const MemRegion *Base = K.getBaseRegion();
248
249 const ClusterBindings *ExistingCluster = lookup(Base);
250 ClusterBindings Cluster = (ExistingCluster ? *ExistingCluster
251 : CBFactory.getEmptyMap());
252
253 ClusterBindings NewCluster = CBFactory.add(Cluster, K, V);
254 return add(Base, NewCluster);
255 }
256
257
addBinding(const MemRegion * R,BindingKey::Kind k,SVal V) const258 RegionBindingsRef RegionBindingsRef::addBinding(const MemRegion *R,
259 BindingKey::Kind k,
260 SVal V) const {
261 return addBinding(BindingKey::Make(R, k), V);
262 }
263
lookup(BindingKey K) const264 const SVal *RegionBindingsRef::lookup(BindingKey K) const {
265 const ClusterBindings *Cluster = lookup(K.getBaseRegion());
266 if (!Cluster)
267 return 0;
268 return Cluster->lookup(K);
269 }
270
lookup(const MemRegion * R,BindingKey::Kind k) const271 const SVal *RegionBindingsRef::lookup(const MemRegion *R,
272 BindingKey::Kind k) const {
273 return lookup(BindingKey::Make(R, k));
274 }
275
removeBinding(BindingKey K)276 RegionBindingsRef RegionBindingsRef::removeBinding(BindingKey K) {
277 const MemRegion *Base = K.getBaseRegion();
278 const ClusterBindings *Cluster = lookup(Base);
279 if (!Cluster)
280 return *this;
281
282 ClusterBindings NewCluster = CBFactory.remove(*Cluster, K);
283 if (NewCluster.isEmpty())
284 return remove(Base);
285 return add(Base, NewCluster);
286 }
287
removeBinding(const MemRegion * R,BindingKey::Kind k)288 RegionBindingsRef RegionBindingsRef::removeBinding(const MemRegion *R,
289 BindingKey::Kind k){
290 return removeBinding(BindingKey::Make(R, k));
291 }
292
293 //===----------------------------------------------------------------------===//
294 // Fine-grained control of RegionStoreManager.
295 //===----------------------------------------------------------------------===//
296
297 namespace {
298 struct minimal_features_tag {};
299 struct maximal_features_tag {};
300
301 class RegionStoreFeatures {
302 bool SupportsFields;
303 public:
RegionStoreFeatures(minimal_features_tag)304 RegionStoreFeatures(minimal_features_tag) :
305 SupportsFields(false) {}
306
RegionStoreFeatures(maximal_features_tag)307 RegionStoreFeatures(maximal_features_tag) :
308 SupportsFields(true) {}
309
enableFields(bool t)310 void enableFields(bool t) { SupportsFields = t; }
311
supportsFields() const312 bool supportsFields() const { return SupportsFields; }
313 };
314 }
315
316 //===----------------------------------------------------------------------===//
317 // Main RegionStore logic.
318 //===----------------------------------------------------------------------===//
319
320 namespace {
321
322 class RegionStoreManager : public StoreManager {
323 public:
324 const RegionStoreFeatures Features;
325 RegionBindings::Factory RBFactory;
326 mutable ClusterBindings::Factory CBFactory;
327
328 typedef std::vector<SVal> SValListTy;
329 private:
330 typedef llvm::DenseMap<const LazyCompoundValData *,
331 SValListTy> LazyBindingsMapTy;
332 LazyBindingsMapTy LazyBindingsMap;
333
334 public:
RegionStoreManager(ProgramStateManager & mgr,const RegionStoreFeatures & f)335 RegionStoreManager(ProgramStateManager& mgr, const RegionStoreFeatures &f)
336 : StoreManager(mgr), Features(f),
337 RBFactory(mgr.getAllocator()), CBFactory(mgr.getAllocator()) {}
338
339
340 /// setImplicitDefaultValue - Set the default binding for the provided
341 /// MemRegion to the value implicitly defined for compound literals when
342 /// the value is not specified.
343 RegionBindingsRef setImplicitDefaultValue(RegionBindingsConstRef B,
344 const MemRegion *R, QualType T);
345
346 /// ArrayToPointer - Emulates the "decay" of an array to a pointer
347 /// type. 'Array' represents the lvalue of the array being decayed
348 /// to a pointer, and the returned SVal represents the decayed
349 /// version of that lvalue (i.e., a pointer to the first element of
350 /// the array). This is called by ExprEngine when evaluating
351 /// casts from arrays to pointers.
352 SVal ArrayToPointer(Loc Array);
353
getInitialStore(const LocationContext * InitLoc)354 StoreRef getInitialStore(const LocationContext *InitLoc) {
355 return StoreRef(RBFactory.getEmptyMap().getRootWithoutRetain(), *this);
356 }
357
358 //===-------------------------------------------------------------------===//
359 // Binding values to regions.
360 //===-------------------------------------------------------------------===//
361 RegionBindingsRef invalidateGlobalRegion(MemRegion::Kind K,
362 const Expr *Ex,
363 unsigned Count,
364 const LocationContext *LCtx,
365 RegionBindingsRef B,
366 InvalidatedRegions *Invalidated);
367
368 StoreRef invalidateRegions(Store store, ArrayRef<const MemRegion *> Regions,
369 const Expr *E, unsigned Count,
370 const LocationContext *LCtx,
371 InvalidatedSymbols &IS,
372 const CallEvent *Call,
373 InvalidatedRegions *Invalidated);
374
375 bool scanReachableSymbols(Store S, const MemRegion *R,
376 ScanReachableSymbols &Callbacks);
377
378 RegionBindingsRef removeSubRegionBindings(RegionBindingsConstRef B,
379 const SubRegion *R);
380
381 public: // Part of public interface to class.
382
Bind(Store store,Loc LV,SVal V)383 virtual StoreRef Bind(Store store, Loc LV, SVal V) {
384 return StoreRef(bind(getRegionBindings(store), LV, V).asStore(), *this);
385 }
386
387 RegionBindingsRef bind(RegionBindingsConstRef B, Loc LV, SVal V);
388
389 // BindDefault is only used to initialize a region with a default value.
BindDefault(Store store,const MemRegion * R,SVal V)390 StoreRef BindDefault(Store store, const MemRegion *R, SVal V) {
391 RegionBindingsRef B = getRegionBindings(store);
392 assert(!B.lookup(R, BindingKey::Default));
393 assert(!B.lookup(R, BindingKey::Direct));
394 return StoreRef(B.addBinding(R, BindingKey::Default, V)
395 .asImmutableMap()
396 .getRootWithoutRetain(), *this);
397 }
398
399 /// \brief Create a new store that binds a value to a compound literal.
400 ///
401 /// \param ST The original store whose bindings are the basis for the new
402 /// store.
403 ///
404 /// \param CL The compound literal to bind (the binding key).
405 ///
406 /// \param LC The LocationContext for the binding.
407 ///
408 /// \param V The value to bind to the compound literal.
409 StoreRef bindCompoundLiteral(Store ST,
410 const CompoundLiteralExpr *CL,
411 const LocationContext *LC, SVal V);
412
413 /// BindStruct - Bind a compound value to a structure.
414 RegionBindingsRef bindStruct(RegionBindingsConstRef B,
415 const TypedValueRegion* R, SVal V);
416
417 /// BindVector - Bind a compound value to a vector.
418 RegionBindingsRef bindVector(RegionBindingsConstRef B,
419 const TypedValueRegion* R, SVal V);
420
421 RegionBindingsRef bindArray(RegionBindingsConstRef B,
422 const TypedValueRegion* R,
423 SVal V);
424
425 /// Clears out all bindings in the given region and assigns a new value
426 /// as a Default binding.
427 RegionBindingsRef bindAggregate(RegionBindingsConstRef B,
428 const TypedRegion *R,
429 SVal DefaultVal);
430
431 /// \brief Create a new store with the specified binding removed.
432 /// \param ST the original store, that is the basis for the new store.
433 /// \param L the location whose binding should be removed.
434 virtual StoreRef killBinding(Store ST, Loc L);
435
incrementReferenceCount(Store store)436 void incrementReferenceCount(Store store) {
437 getRegionBindings(store).manualRetain();
438 }
439
440 /// If the StoreManager supports it, decrement the reference count of
441 /// the specified Store object. If the reference count hits 0, the memory
442 /// associated with the object is recycled.
decrementReferenceCount(Store store)443 void decrementReferenceCount(Store store) {
444 getRegionBindings(store).manualRelease();
445 }
446
447 bool includedInBindings(Store store, const MemRegion *region) const;
448
449 /// \brief Return the value bound to specified location in a given state.
450 ///
451 /// The high level logic for this method is this:
452 /// getBinding (L)
453 /// if L has binding
454 /// return L's binding
455 /// else if L is in killset
456 /// return unknown
457 /// else
458 /// if L is on stack or heap
459 /// return undefined
460 /// else
461 /// return symbolic
getBinding(Store S,Loc L,QualType T)462 virtual SVal getBinding(Store S, Loc L, QualType T) {
463 return getBinding(getRegionBindings(S), L, T);
464 }
465
466 SVal getBinding(RegionBindingsConstRef B, Loc L, QualType T = QualType());
467
468 SVal getBindingForElement(RegionBindingsConstRef B, const ElementRegion *R);
469
470 SVal getBindingForField(RegionBindingsConstRef B, const FieldRegion *R);
471
472 SVal getBindingForObjCIvar(RegionBindingsConstRef B, const ObjCIvarRegion *R);
473
474 SVal getBindingForVar(RegionBindingsConstRef B, const VarRegion *R);
475
476 SVal getBindingForLazySymbol(const TypedValueRegion *R);
477
478 SVal getBindingForFieldOrElementCommon(RegionBindingsConstRef B,
479 const TypedValueRegion *R,
480 QualType Ty,
481 const MemRegion *superR);
482
483 SVal getLazyBinding(const SubRegion *LazyBindingRegion,
484 RegionBindingsRef LazyBinding);
485
486 /// Get bindings for the values in a struct and return a CompoundVal, used
487 /// when doing struct copy:
488 /// struct s x, y;
489 /// x = y;
490 /// y's value is retrieved by this method.
491 SVal getBindingForStruct(RegionBindingsConstRef B, const TypedValueRegion *R);
492 SVal getBindingForArray(RegionBindingsConstRef B, const TypedValueRegion *R);
493 NonLoc createLazyBinding(RegionBindingsConstRef B, const TypedValueRegion *R);
494
495 /// Used to lazily generate derived symbols for bindings that are defined
496 /// implicitly by default bindings in a super region.
497 ///
498 /// Note that callers may need to specially handle LazyCompoundVals, which
499 /// are returned as is in case the caller needs to treat them differently.
500 Optional<SVal> getBindingForDerivedDefaultValue(RegionBindingsConstRef B,
501 const MemRegion *superR,
502 const TypedValueRegion *R,
503 QualType Ty);
504
505 /// Get the state and region whose binding this region \p R corresponds to.
506 ///
507 /// If there is no lazy binding for \p R, the returned value will have a null
508 /// \c second. Note that a null pointer can represents a valid Store.
509 std::pair<Store, const SubRegion *>
510 findLazyBinding(RegionBindingsConstRef B, const SubRegion *R,
511 const SubRegion *originalRegion);
512
513 /// Returns the cached set of interesting SVals contained within a lazy
514 /// binding.
515 ///
516 /// The precise value of "interesting" is determined for the purposes of
517 /// RegionStore's internal analysis. It must always contain all regions and
518 /// symbols, but may omit constants and other kinds of SVal.
519 const SValListTy &getInterestingValues(nonloc::LazyCompoundVal LCV);
520
521 //===------------------------------------------------------------------===//
522 // State pruning.
523 //===------------------------------------------------------------------===//
524
525 /// removeDeadBindings - Scans the RegionStore of 'state' for dead values.
526 /// It returns a new Store with these values removed.
527 StoreRef removeDeadBindings(Store store, const StackFrameContext *LCtx,
528 SymbolReaper& SymReaper);
529
530 //===------------------------------------------------------------------===//
531 // Region "extents".
532 //===------------------------------------------------------------------===//
533
534 // FIXME: This method will soon be eliminated; see the note in Store.h.
535 DefinedOrUnknownSVal getSizeInElements(ProgramStateRef state,
536 const MemRegion* R, QualType EleTy);
537
538 //===------------------------------------------------------------------===//
539 // Utility methods.
540 //===------------------------------------------------------------------===//
541
getRegionBindings(Store store) const542 RegionBindingsRef getRegionBindings(Store store) const {
543 return RegionBindingsRef(CBFactory,
544 static_cast<const RegionBindings::TreeTy*>(store),
545 RBFactory.getTreeFactory());
546 }
547
548 void print(Store store, raw_ostream &Out, const char* nl,
549 const char *sep);
550
iterBindings(Store store,BindingsHandler & f)551 void iterBindings(Store store, BindingsHandler& f) {
552 RegionBindingsRef B = getRegionBindings(store);
553 for (RegionBindingsRef::iterator I = B.begin(), E = B.end(); I != E; ++I) {
554 const ClusterBindings &Cluster = I.getData();
555 for (ClusterBindings::iterator CI = Cluster.begin(), CE = Cluster.end();
556 CI != CE; ++CI) {
557 const BindingKey &K = CI.getKey();
558 if (!K.isDirect())
559 continue;
560 if (const SubRegion *R = dyn_cast<SubRegion>(K.getRegion())) {
561 // FIXME: Possibly incorporate the offset?
562 if (!f.HandleBinding(*this, store, R, CI.getData()))
563 return;
564 }
565 }
566 }
567 }
568 };
569
570 } // end anonymous namespace
571
572 //===----------------------------------------------------------------------===//
573 // RegionStore creation.
574 //===----------------------------------------------------------------------===//
575
CreateRegionStoreManager(ProgramStateManager & StMgr)576 StoreManager *ento::CreateRegionStoreManager(ProgramStateManager& StMgr) {
577 RegionStoreFeatures F = maximal_features_tag();
578 return new RegionStoreManager(StMgr, F);
579 }
580
581 StoreManager *
CreateFieldsOnlyRegionStoreManager(ProgramStateManager & StMgr)582 ento::CreateFieldsOnlyRegionStoreManager(ProgramStateManager &StMgr) {
583 RegionStoreFeatures F = minimal_features_tag();
584 F.enableFields(true);
585 return new RegionStoreManager(StMgr, F);
586 }
587
588
589 //===----------------------------------------------------------------------===//
590 // Region Cluster analysis.
591 //===----------------------------------------------------------------------===//
592
593 namespace {
594 template <typename DERIVED>
595 class ClusterAnalysis {
596 protected:
597 typedef llvm::DenseMap<const MemRegion *, const ClusterBindings *> ClusterMap;
598 typedef SmallVector<const MemRegion *, 10> WorkList;
599
600 llvm::SmallPtrSet<const ClusterBindings *, 16> Visited;
601
602 WorkList WL;
603
604 RegionStoreManager &RM;
605 ASTContext &Ctx;
606 SValBuilder &svalBuilder;
607
608 RegionBindingsRef B;
609
610 const bool includeGlobals;
611
getCluster(const MemRegion * R)612 const ClusterBindings *getCluster(const MemRegion *R) {
613 return B.lookup(R);
614 }
615
616 public:
ClusterAnalysis(RegionStoreManager & rm,ProgramStateManager & StateMgr,RegionBindingsRef b,const bool includeGlobals)617 ClusterAnalysis(RegionStoreManager &rm, ProgramStateManager &StateMgr,
618 RegionBindingsRef b, const bool includeGlobals)
619 : RM(rm), Ctx(StateMgr.getContext()),
620 svalBuilder(StateMgr.getSValBuilder()),
621 B(b), includeGlobals(includeGlobals) {}
622
getRegionBindings() const623 RegionBindingsRef getRegionBindings() const { return B; }
624
isVisited(const MemRegion * R)625 bool isVisited(const MemRegion *R) {
626 return Visited.count(getCluster(R));
627 }
628
GenerateClusters()629 void GenerateClusters() {
630 // Scan the entire set of bindings and record the region clusters.
631 for (RegionBindingsRef::iterator RI = B.begin(), RE = B.end();
632 RI != RE; ++RI){
633 const MemRegion *Base = RI.getKey();
634
635 const ClusterBindings &Cluster = RI.getData();
636 assert(!Cluster.isEmpty() && "Empty clusters should be removed");
637 static_cast<DERIVED*>(this)->VisitAddedToCluster(Base, Cluster);
638
639 if (includeGlobals)
640 if (isa<NonStaticGlobalSpaceRegion>(Base->getMemorySpace()))
641 AddToWorkList(Base, &Cluster);
642 }
643 }
644
AddToWorkList(const MemRegion * R,const ClusterBindings * C)645 bool AddToWorkList(const MemRegion *R, const ClusterBindings *C) {
646 if (C && !Visited.insert(C))
647 return false;
648 WL.push_back(R);
649 return true;
650 }
651
AddToWorkList(const MemRegion * R)652 bool AddToWorkList(const MemRegion *R) {
653 const MemRegion *baseR = R->getBaseRegion();
654 return AddToWorkList(baseR, getCluster(baseR));
655 }
656
RunWorkList()657 void RunWorkList() {
658 while (!WL.empty()) {
659 const MemRegion *baseR = WL.pop_back_val();
660
661 // First visit the cluster.
662 if (const ClusterBindings *Cluster = getCluster(baseR))
663 static_cast<DERIVED*>(this)->VisitCluster(baseR, *Cluster);
664
665 // Next, visit the base region.
666 static_cast<DERIVED*>(this)->VisitBaseRegion(baseR);
667 }
668 }
669
670 public:
VisitAddedToCluster(const MemRegion * baseR,const ClusterBindings & C)671 void VisitAddedToCluster(const MemRegion *baseR, const ClusterBindings &C) {}
VisitCluster(const MemRegion * baseR,const ClusterBindings & C)672 void VisitCluster(const MemRegion *baseR, const ClusterBindings &C) {}
VisitBaseRegion(const MemRegion * baseR)673 void VisitBaseRegion(const MemRegion *baseR) {}
674 };
675 }
676
677 //===----------------------------------------------------------------------===//
678 // Binding invalidation.
679 //===----------------------------------------------------------------------===//
680
scanReachableSymbols(Store S,const MemRegion * R,ScanReachableSymbols & Callbacks)681 bool RegionStoreManager::scanReachableSymbols(Store S, const MemRegion *R,
682 ScanReachableSymbols &Callbacks) {
683 assert(R == R->getBaseRegion() && "Should only be called for base regions");
684 RegionBindingsRef B = getRegionBindings(S);
685 const ClusterBindings *Cluster = B.lookup(R);
686
687 if (!Cluster)
688 return true;
689
690 for (ClusterBindings::iterator RI = Cluster->begin(), RE = Cluster->end();
691 RI != RE; ++RI) {
692 if (!Callbacks.scan(RI.getData()))
693 return false;
694 }
695
696 return true;
697 }
698
isUnionField(const FieldRegion * FR)699 static inline bool isUnionField(const FieldRegion *FR) {
700 return FR->getDecl()->getParent()->isUnion();
701 }
702
703 typedef SmallVector<const FieldDecl *, 8> FieldVector;
704
getSymbolicOffsetFields(BindingKey K,FieldVector & Fields)705 void getSymbolicOffsetFields(BindingKey K, FieldVector &Fields) {
706 assert(K.hasSymbolicOffset() && "Not implemented for concrete offset keys");
707
708 const MemRegion *Base = K.getConcreteOffsetRegion();
709 const MemRegion *R = K.getRegion();
710
711 while (R != Base) {
712 if (const FieldRegion *FR = dyn_cast<FieldRegion>(R))
713 if (!isUnionField(FR))
714 Fields.push_back(FR->getDecl());
715
716 R = cast<SubRegion>(R)->getSuperRegion();
717 }
718 }
719
isCompatibleWithFields(BindingKey K,const FieldVector & Fields)720 static bool isCompatibleWithFields(BindingKey K, const FieldVector &Fields) {
721 assert(K.hasSymbolicOffset() && "Not implemented for concrete offset keys");
722
723 if (Fields.empty())
724 return true;
725
726 FieldVector FieldsInBindingKey;
727 getSymbolicOffsetFields(K, FieldsInBindingKey);
728
729 ptrdiff_t Delta = FieldsInBindingKey.size() - Fields.size();
730 if (Delta >= 0)
731 return std::equal(FieldsInBindingKey.begin() + Delta,
732 FieldsInBindingKey.end(),
733 Fields.begin());
734 else
735 return std::equal(FieldsInBindingKey.begin(), FieldsInBindingKey.end(),
736 Fields.begin() - Delta);
737 }
738
739 /// Collects all bindings in \p Cluster that may refer to bindings within
740 /// \p Top.
741 ///
742 /// Each binding is a pair whose \c first is the key (a BindingKey) and whose
743 /// \c second is the value (an SVal).
744 ///
745 /// The \p IncludeAllDefaultBindings parameter specifies whether to include
746 /// default bindings that may extend beyond \p Top itself, e.g. if \p Top is
747 /// an aggregate within a larger aggregate with a default binding.
748 static void
collectSubRegionBindings(SmallVectorImpl<BindingPair> & Bindings,SValBuilder & SVB,const ClusterBindings & Cluster,const SubRegion * Top,BindingKey TopKey,bool IncludeAllDefaultBindings)749 collectSubRegionBindings(SmallVectorImpl<BindingPair> &Bindings,
750 SValBuilder &SVB, const ClusterBindings &Cluster,
751 const SubRegion *Top, BindingKey TopKey,
752 bool IncludeAllDefaultBindings) {
753 FieldVector FieldsInSymbolicSubregions;
754 if (TopKey.hasSymbolicOffset()) {
755 getSymbolicOffsetFields(TopKey, FieldsInSymbolicSubregions);
756 Top = cast<SubRegion>(TopKey.getConcreteOffsetRegion());
757 TopKey = BindingKey::Make(Top, BindingKey::Default);
758 }
759
760 // Find the length (in bits) of the region being invalidated.
761 uint64_t Length = UINT64_MAX;
762 SVal Extent = Top->getExtent(SVB);
763 if (Optional<nonloc::ConcreteInt> ExtentCI =
764 Extent.getAs<nonloc::ConcreteInt>()) {
765 const llvm::APSInt &ExtentInt = ExtentCI->getValue();
766 assert(ExtentInt.isNonNegative() || ExtentInt.isUnsigned());
767 // Extents are in bytes but region offsets are in bits. Be careful!
768 Length = ExtentInt.getLimitedValue() * SVB.getContext().getCharWidth();
769 } else if (const FieldRegion *FR = dyn_cast<FieldRegion>(Top)) {
770 if (FR->getDecl()->isBitField())
771 Length = FR->getDecl()->getBitWidthValue(SVB.getContext());
772 }
773
774 for (ClusterBindings::iterator I = Cluster.begin(), E = Cluster.end();
775 I != E; ++I) {
776 BindingKey NextKey = I.getKey();
777 if (NextKey.getRegion() == TopKey.getRegion()) {
778 // FIXME: This doesn't catch the case where we're really invalidating a
779 // region with a symbolic offset. Example:
780 // R: points[i].y
781 // Next: points[0].x
782
783 if (NextKey.getOffset() > TopKey.getOffset() &&
784 NextKey.getOffset() - TopKey.getOffset() < Length) {
785 // Case 1: The next binding is inside the region we're invalidating.
786 // Include it.
787 Bindings.push_back(*I);
788
789 } else if (NextKey.getOffset() == TopKey.getOffset()) {
790 // Case 2: The next binding is at the same offset as the region we're
791 // invalidating. In this case, we need to leave default bindings alone,
792 // since they may be providing a default value for a regions beyond what
793 // we're invalidating.
794 // FIXME: This is probably incorrect; consider invalidating an outer
795 // struct whose first field is bound to a LazyCompoundVal.
796 if (IncludeAllDefaultBindings || NextKey.isDirect())
797 Bindings.push_back(*I);
798 }
799
800 } else if (NextKey.hasSymbolicOffset()) {
801 const MemRegion *Base = NextKey.getConcreteOffsetRegion();
802 if (Top->isSubRegionOf(Base)) {
803 // Case 3: The next key is symbolic and we just changed something within
804 // its concrete region. We don't know if the binding is still valid, so
805 // we'll be conservative and include it.
806 if (IncludeAllDefaultBindings || NextKey.isDirect())
807 if (isCompatibleWithFields(NextKey, FieldsInSymbolicSubregions))
808 Bindings.push_back(*I);
809 } else if (const SubRegion *BaseSR = dyn_cast<SubRegion>(Base)) {
810 // Case 4: The next key is symbolic, but we changed a known
811 // super-region. In this case the binding is certainly included.
812 if (Top == Base || BaseSR->isSubRegionOf(Top))
813 if (isCompatibleWithFields(NextKey, FieldsInSymbolicSubregions))
814 Bindings.push_back(*I);
815 }
816 }
817 }
818 }
819
820 static void
collectSubRegionBindings(SmallVectorImpl<BindingPair> & Bindings,SValBuilder & SVB,const ClusterBindings & Cluster,const SubRegion * Top,bool IncludeAllDefaultBindings)821 collectSubRegionBindings(SmallVectorImpl<BindingPair> &Bindings,
822 SValBuilder &SVB, const ClusterBindings &Cluster,
823 const SubRegion *Top, bool IncludeAllDefaultBindings) {
824 collectSubRegionBindings(Bindings, SVB, Cluster, Top,
825 BindingKey::Make(Top, BindingKey::Default),
826 IncludeAllDefaultBindings);
827 }
828
829 RegionBindingsRef
removeSubRegionBindings(RegionBindingsConstRef B,const SubRegion * Top)830 RegionStoreManager::removeSubRegionBindings(RegionBindingsConstRef B,
831 const SubRegion *Top) {
832 BindingKey TopKey = BindingKey::Make(Top, BindingKey::Default);
833 const MemRegion *ClusterHead = TopKey.getBaseRegion();
834 if (Top == ClusterHead) {
835 // We can remove an entire cluster's bindings all in one go.
836 return B.remove(Top);
837 }
838
839 const ClusterBindings *Cluster = B.lookup(ClusterHead);
840 if (!Cluster)
841 return B;
842
843 SmallVector<BindingPair, 32> Bindings;
844 collectSubRegionBindings(Bindings, svalBuilder, *Cluster, Top, TopKey,
845 /*IncludeAllDefaultBindings=*/false);
846
847 ClusterBindingsRef Result(*Cluster, CBFactory);
848 for (SmallVectorImpl<BindingPair>::const_iterator I = Bindings.begin(),
849 E = Bindings.end();
850 I != E; ++I)
851 Result = Result.remove(I->first);
852
853 // If we're invalidating a region with a symbolic offset, we need to make sure
854 // we don't treat the base region as uninitialized anymore.
855 // FIXME: This isn't very precise; see the example in the loop.
856 if (TopKey.hasSymbolicOffset()) {
857 const SubRegion *Concrete = TopKey.getConcreteOffsetRegion();
858 Result = Result.add(BindingKey::Make(Concrete, BindingKey::Default),
859 UnknownVal());
860 }
861
862 if (Result.isEmpty())
863 return B.remove(ClusterHead);
864 return B.add(ClusterHead, Result.asImmutableMap());
865 }
866
867 namespace {
868 class invalidateRegionsWorker : public ClusterAnalysis<invalidateRegionsWorker>
869 {
870 const Expr *Ex;
871 unsigned Count;
872 const LocationContext *LCtx;
873 InvalidatedSymbols &IS;
874 StoreManager::InvalidatedRegions *Regions;
875 public:
invalidateRegionsWorker(RegionStoreManager & rm,ProgramStateManager & stateMgr,RegionBindingsRef b,const Expr * ex,unsigned count,const LocationContext * lctx,InvalidatedSymbols & is,StoreManager::InvalidatedRegions * r,bool includeGlobals)876 invalidateRegionsWorker(RegionStoreManager &rm,
877 ProgramStateManager &stateMgr,
878 RegionBindingsRef b,
879 const Expr *ex, unsigned count,
880 const LocationContext *lctx,
881 InvalidatedSymbols &is,
882 StoreManager::InvalidatedRegions *r,
883 bool includeGlobals)
884 : ClusterAnalysis<invalidateRegionsWorker>(rm, stateMgr, b, includeGlobals),
885 Ex(ex), Count(count), LCtx(lctx), IS(is), Regions(r) {}
886
887 void VisitCluster(const MemRegion *baseR, const ClusterBindings &C);
888 void VisitBaseRegion(const MemRegion *baseR);
889
890 private:
891 void VisitBinding(SVal V);
892 };
893 }
894
VisitBinding(SVal V)895 void invalidateRegionsWorker::VisitBinding(SVal V) {
896 // A symbol? Mark it touched by the invalidation.
897 if (SymbolRef Sym = V.getAsSymbol())
898 IS.insert(Sym);
899
900 if (const MemRegion *R = V.getAsRegion()) {
901 AddToWorkList(R);
902 return;
903 }
904
905 // Is it a LazyCompoundVal? All references get invalidated as well.
906 if (Optional<nonloc::LazyCompoundVal> LCS =
907 V.getAs<nonloc::LazyCompoundVal>()) {
908
909 const RegionStoreManager::SValListTy &Vals = RM.getInterestingValues(*LCS);
910
911 for (RegionStoreManager::SValListTy::const_iterator I = Vals.begin(),
912 E = Vals.end();
913 I != E; ++I)
914 VisitBinding(*I);
915
916 return;
917 }
918 }
919
VisitCluster(const MemRegion * BaseR,const ClusterBindings & C)920 void invalidateRegionsWorker::VisitCluster(const MemRegion *BaseR,
921 const ClusterBindings &C) {
922 for (ClusterBindings::iterator I = C.begin(), E = C.end(); I != E; ++I)
923 VisitBinding(I.getData());
924
925 B = B.remove(BaseR);
926 }
927
VisitBaseRegion(const MemRegion * baseR)928 void invalidateRegionsWorker::VisitBaseRegion(const MemRegion *baseR) {
929 // Symbolic region? Mark that symbol touched by the invalidation.
930 if (const SymbolicRegion *SR = dyn_cast<SymbolicRegion>(baseR))
931 IS.insert(SR->getSymbol());
932
933 // BlockDataRegion? If so, invalidate captured variables that are passed
934 // by reference.
935 if (const BlockDataRegion *BR = dyn_cast<BlockDataRegion>(baseR)) {
936 for (BlockDataRegion::referenced_vars_iterator
937 BI = BR->referenced_vars_begin(), BE = BR->referenced_vars_end() ;
938 BI != BE; ++BI) {
939 const VarRegion *VR = BI.getCapturedRegion();
940 const VarDecl *VD = VR->getDecl();
941 if (VD->getAttr<BlocksAttr>() || !VD->hasLocalStorage()) {
942 AddToWorkList(VR);
943 }
944 else if (Loc::isLocType(VR->getValueType())) {
945 // Map the current bindings to a Store to retrieve the value
946 // of the binding. If that binding itself is a region, we should
947 // invalidate that region. This is because a block may capture
948 // a pointer value, but the thing pointed by that pointer may
949 // get invalidated.
950 SVal V = RM.getBinding(B, loc::MemRegionVal(VR));
951 if (Optional<Loc> L = V.getAs<Loc>()) {
952 if (const MemRegion *LR = L->getAsRegion())
953 AddToWorkList(LR);
954 }
955 }
956 }
957 return;
958 }
959
960 // Otherwise, we have a normal data region. Record that we touched the region.
961 if (Regions)
962 Regions->push_back(baseR);
963
964 if (isa<AllocaRegion>(baseR) || isa<SymbolicRegion>(baseR)) {
965 // Invalidate the region by setting its default value to
966 // conjured symbol. The type of the symbol is irrelavant.
967 DefinedOrUnknownSVal V =
968 svalBuilder.conjureSymbolVal(baseR, Ex, LCtx, Ctx.IntTy, Count);
969 B = B.addBinding(baseR, BindingKey::Default, V);
970 return;
971 }
972
973 if (!baseR->isBoundable())
974 return;
975
976 const TypedValueRegion *TR = cast<TypedValueRegion>(baseR);
977 QualType T = TR->getValueType();
978
979 // Invalidate the binding.
980 if (T->isStructureOrClassType()) {
981 // Invalidate the region by setting its default value to
982 // conjured symbol. The type of the symbol is irrelavant.
983 DefinedOrUnknownSVal V = svalBuilder.conjureSymbolVal(baseR, Ex, LCtx,
984 Ctx.IntTy, Count);
985 B = B.addBinding(baseR, BindingKey::Default, V);
986 return;
987 }
988
989 if (const ArrayType *AT = Ctx.getAsArrayType(T)) {
990 // Set the default value of the array to conjured symbol.
991 DefinedOrUnknownSVal V =
992 svalBuilder.conjureSymbolVal(baseR, Ex, LCtx,
993 AT->getElementType(), Count);
994 B = B.addBinding(baseR, BindingKey::Default, V);
995 return;
996 }
997
998 if (includeGlobals &&
999 isa<NonStaticGlobalSpaceRegion>(baseR->getMemorySpace())) {
1000 // If the region is a global and we are invalidating all globals,
1001 // just erase the entry. This causes all globals to be lazily
1002 // symbolicated from the same base symbol.
1003 B = B.removeBinding(baseR);
1004 return;
1005 }
1006
1007
1008 DefinedOrUnknownSVal V = svalBuilder.conjureSymbolVal(baseR, Ex, LCtx,
1009 T,Count);
1010 assert(SymbolManager::canSymbolicate(T) || V.isUnknown());
1011 B = B.addBinding(baseR, BindingKey::Direct, V);
1012 }
1013
1014 RegionBindingsRef
invalidateGlobalRegion(MemRegion::Kind K,const Expr * Ex,unsigned Count,const LocationContext * LCtx,RegionBindingsRef B,InvalidatedRegions * Invalidated)1015 RegionStoreManager::invalidateGlobalRegion(MemRegion::Kind K,
1016 const Expr *Ex,
1017 unsigned Count,
1018 const LocationContext *LCtx,
1019 RegionBindingsRef B,
1020 InvalidatedRegions *Invalidated) {
1021 // Bind the globals memory space to a new symbol that we will use to derive
1022 // the bindings for all globals.
1023 const GlobalsSpaceRegion *GS = MRMgr.getGlobalsRegion(K);
1024 SVal V = svalBuilder.conjureSymbolVal(/* SymbolTag = */ (const void*) GS, Ex, LCtx,
1025 /* type does not matter */ Ctx.IntTy,
1026 Count);
1027
1028 B = B.removeBinding(GS)
1029 .addBinding(BindingKey::Make(GS, BindingKey::Default), V);
1030
1031 // Even if there are no bindings in the global scope, we still need to
1032 // record that we touched it.
1033 if (Invalidated)
1034 Invalidated->push_back(GS);
1035
1036 return B;
1037 }
1038
1039 StoreRef
invalidateRegions(Store store,ArrayRef<const MemRegion * > Regions,const Expr * Ex,unsigned Count,const LocationContext * LCtx,InvalidatedSymbols & IS,const CallEvent * Call,InvalidatedRegions * Invalidated)1040 RegionStoreManager::invalidateRegions(Store store,
1041 ArrayRef<const MemRegion *> Regions,
1042 const Expr *Ex, unsigned Count,
1043 const LocationContext *LCtx,
1044 InvalidatedSymbols &IS,
1045 const CallEvent *Call,
1046 InvalidatedRegions *Invalidated) {
1047 invalidateRegionsWorker W(*this, StateMgr,
1048 RegionStoreManager::getRegionBindings(store),
1049 Ex, Count, LCtx, IS, Invalidated, false);
1050
1051 // Scan the bindings and generate the clusters.
1052 W.GenerateClusters();
1053
1054 // Add the regions to the worklist.
1055 for (ArrayRef<const MemRegion *>::iterator
1056 I = Regions.begin(), E = Regions.end(); I != E; ++I)
1057 W.AddToWorkList(*I);
1058
1059 W.RunWorkList();
1060
1061 // Return the new bindings.
1062 RegionBindingsRef B = W.getRegionBindings();
1063
1064 // For all globals which are not static nor immutable: determine which global
1065 // regions should be invalidated and invalidate them.
1066 // TODO: This could possibly be more precise with modules.
1067 //
1068 // System calls invalidate only system globals.
1069 if (Call && Call->isInSystemHeader()) {
1070 B = invalidateGlobalRegion(MemRegion::GlobalSystemSpaceRegionKind,
1071 Ex, Count, LCtx, B, Invalidated);
1072 // Internal calls might invalidate both system and internal globals.
1073 } else {
1074 B = invalidateGlobalRegion(MemRegion::GlobalSystemSpaceRegionKind,
1075 Ex, Count, LCtx, B, Invalidated);
1076 B = invalidateGlobalRegion(MemRegion::GlobalInternalSpaceRegionKind,
1077 Ex, Count, LCtx, B, Invalidated);
1078 }
1079
1080 return StoreRef(B.asStore(), *this);
1081 }
1082
1083 //===----------------------------------------------------------------------===//
1084 // Extents for regions.
1085 //===----------------------------------------------------------------------===//
1086
1087 DefinedOrUnknownSVal
getSizeInElements(ProgramStateRef state,const MemRegion * R,QualType EleTy)1088 RegionStoreManager::getSizeInElements(ProgramStateRef state,
1089 const MemRegion *R,
1090 QualType EleTy) {
1091 SVal Size = cast<SubRegion>(R)->getExtent(svalBuilder);
1092 const llvm::APSInt *SizeInt = svalBuilder.getKnownValue(state, Size);
1093 if (!SizeInt)
1094 return UnknownVal();
1095
1096 CharUnits RegionSize = CharUnits::fromQuantity(SizeInt->getSExtValue());
1097
1098 if (Ctx.getAsVariableArrayType(EleTy)) {
1099 // FIXME: We need to track extra state to properly record the size
1100 // of VLAs. Returning UnknownVal here, however, is a stop-gap so that
1101 // we don't have a divide-by-zero below.
1102 return UnknownVal();
1103 }
1104
1105 CharUnits EleSize = Ctx.getTypeSizeInChars(EleTy);
1106
1107 // If a variable is reinterpreted as a type that doesn't fit into a larger
1108 // type evenly, round it down.
1109 // This is a signed value, since it's used in arithmetic with signed indices.
1110 return svalBuilder.makeIntVal(RegionSize / EleSize, false);
1111 }
1112
1113 //===----------------------------------------------------------------------===//
1114 // Location and region casting.
1115 //===----------------------------------------------------------------------===//
1116
1117 /// ArrayToPointer - Emulates the "decay" of an array to a pointer
1118 /// type. 'Array' represents the lvalue of the array being decayed
1119 /// to a pointer, and the returned SVal represents the decayed
1120 /// version of that lvalue (i.e., a pointer to the first element of
1121 /// the array). This is called by ExprEngine when evaluating casts
1122 /// from arrays to pointers.
ArrayToPointer(Loc Array)1123 SVal RegionStoreManager::ArrayToPointer(Loc Array) {
1124 if (!Array.getAs<loc::MemRegionVal>())
1125 return UnknownVal();
1126
1127 const MemRegion* R = Array.castAs<loc::MemRegionVal>().getRegion();
1128 const TypedValueRegion* ArrayR = dyn_cast<TypedValueRegion>(R);
1129
1130 if (!ArrayR)
1131 return UnknownVal();
1132
1133 // Strip off typedefs from the ArrayRegion's ValueType.
1134 QualType T = ArrayR->getValueType().getDesugaredType(Ctx);
1135 const ArrayType *AT = cast<ArrayType>(T);
1136 T = AT->getElementType();
1137
1138 NonLoc ZeroIdx = svalBuilder.makeZeroArrayIndex();
1139 return loc::MemRegionVal(MRMgr.getElementRegion(T, ZeroIdx, ArrayR, Ctx));
1140 }
1141
1142 //===----------------------------------------------------------------------===//
1143 // Loading values from regions.
1144 //===----------------------------------------------------------------------===//
1145
getBinding(RegionBindingsConstRef B,Loc L,QualType T)1146 SVal RegionStoreManager::getBinding(RegionBindingsConstRef B, Loc L, QualType T) {
1147 assert(!L.getAs<UnknownVal>() && "location unknown");
1148 assert(!L.getAs<UndefinedVal>() && "location undefined");
1149
1150 // For access to concrete addresses, return UnknownVal. Checks
1151 // for null dereferences (and similar errors) are done by checkers, not
1152 // the Store.
1153 // FIXME: We can consider lazily symbolicating such memory, but we really
1154 // should defer this when we can reason easily about symbolicating arrays
1155 // of bytes.
1156 if (L.getAs<loc::ConcreteInt>()) {
1157 return UnknownVal();
1158 }
1159 if (!L.getAs<loc::MemRegionVal>()) {
1160 return UnknownVal();
1161 }
1162
1163 const MemRegion *MR = L.castAs<loc::MemRegionVal>().getRegion();
1164
1165 if (isa<AllocaRegion>(MR) ||
1166 isa<SymbolicRegion>(MR) ||
1167 isa<CodeTextRegion>(MR)) {
1168 if (T.isNull()) {
1169 if (const TypedRegion *TR = dyn_cast<TypedRegion>(MR))
1170 T = TR->getLocationType();
1171 else {
1172 const SymbolicRegion *SR = cast<SymbolicRegion>(MR);
1173 T = SR->getSymbol()->getType();
1174 }
1175 }
1176 MR = GetElementZeroRegion(MR, T);
1177 }
1178
1179 // FIXME: Perhaps this method should just take a 'const MemRegion*' argument
1180 // instead of 'Loc', and have the other Loc cases handled at a higher level.
1181 const TypedValueRegion *R = cast<TypedValueRegion>(MR);
1182 QualType RTy = R->getValueType();
1183
1184 // FIXME: we do not yet model the parts of a complex type, so treat the
1185 // whole thing as "unknown".
1186 if (RTy->isAnyComplexType())
1187 return UnknownVal();
1188
1189 // FIXME: We should eventually handle funny addressing. e.g.:
1190 //
1191 // int x = ...;
1192 // int *p = &x;
1193 // char *q = (char*) p;
1194 // char c = *q; // returns the first byte of 'x'.
1195 //
1196 // Such funny addressing will occur due to layering of regions.
1197 if (RTy->isStructureOrClassType())
1198 return getBindingForStruct(B, R);
1199
1200 // FIXME: Handle unions.
1201 if (RTy->isUnionType())
1202 return UnknownVal();
1203
1204 if (RTy->isArrayType()) {
1205 if (RTy->isConstantArrayType())
1206 return getBindingForArray(B, R);
1207 else
1208 return UnknownVal();
1209 }
1210
1211 // FIXME: handle Vector types.
1212 if (RTy->isVectorType())
1213 return UnknownVal();
1214
1215 if (const FieldRegion* FR = dyn_cast<FieldRegion>(R))
1216 return CastRetrievedVal(getBindingForField(B, FR), FR, T, false);
1217
1218 if (const ElementRegion* ER = dyn_cast<ElementRegion>(R)) {
1219 // FIXME: Here we actually perform an implicit conversion from the loaded
1220 // value to the element type. Eventually we want to compose these values
1221 // more intelligently. For example, an 'element' can encompass multiple
1222 // bound regions (e.g., several bound bytes), or could be a subset of
1223 // a larger value.
1224 return CastRetrievedVal(getBindingForElement(B, ER), ER, T, false);
1225 }
1226
1227 if (const ObjCIvarRegion *IVR = dyn_cast<ObjCIvarRegion>(R)) {
1228 // FIXME: Here we actually perform an implicit conversion from the loaded
1229 // value to the ivar type. What we should model is stores to ivars
1230 // that blow past the extent of the ivar. If the address of the ivar is
1231 // reinterpretted, it is possible we stored a different value that could
1232 // fit within the ivar. Either we need to cast these when storing them
1233 // or reinterpret them lazily (as we do here).
1234 return CastRetrievedVal(getBindingForObjCIvar(B, IVR), IVR, T, false);
1235 }
1236
1237 if (const VarRegion *VR = dyn_cast<VarRegion>(R)) {
1238 // FIXME: Here we actually perform an implicit conversion from the loaded
1239 // value to the variable type. What we should model is stores to variables
1240 // that blow past the extent of the variable. If the address of the
1241 // variable is reinterpretted, it is possible we stored a different value
1242 // that could fit within the variable. Either we need to cast these when
1243 // storing them or reinterpret them lazily (as we do here).
1244 return CastRetrievedVal(getBindingForVar(B, VR), VR, T, false);
1245 }
1246
1247 const SVal *V = B.lookup(R, BindingKey::Direct);
1248
1249 // Check if the region has a binding.
1250 if (V)
1251 return *V;
1252
1253 // The location does not have a bound value. This means that it has
1254 // the value it had upon its creation and/or entry to the analyzed
1255 // function/method. These are either symbolic values or 'undefined'.
1256 if (R->hasStackNonParametersStorage()) {
1257 // All stack variables are considered to have undefined values
1258 // upon creation. All heap allocated blocks are considered to
1259 // have undefined values as well unless they are explicitly bound
1260 // to specific values.
1261 return UndefinedVal();
1262 }
1263
1264 // All other values are symbolic.
1265 return svalBuilder.getRegionValueSymbolVal(R);
1266 }
1267
1268 /// Checks to see if store \p B has a lazy binding for region \p R.
1269 ///
1270 /// If \p AllowSubregionBindings is \c false, a lazy binding will be rejected
1271 /// if there are additional bindings within \p R.
1272 ///
1273 /// Note that unlike RegionStoreManager::findLazyBinding, this will not search
1274 /// for lazy bindings for super-regions of \p R.
1275 static Optional<nonloc::LazyCompoundVal>
getExistingLazyBinding(SValBuilder & SVB,RegionBindingsConstRef B,const SubRegion * R,bool AllowSubregionBindings)1276 getExistingLazyBinding(SValBuilder &SVB, RegionBindingsConstRef B,
1277 const SubRegion *R, bool AllowSubregionBindings) {
1278 Optional<SVal> V = B.getDefaultBinding(R);
1279 if (!V)
1280 return None;
1281
1282 Optional<nonloc::LazyCompoundVal> LCV = V->getAs<nonloc::LazyCompoundVal>();
1283 if (!LCV)
1284 return None;
1285
1286 // If the LCV is for a subregion, the types won't match, and we shouldn't
1287 // reuse the binding. Unfortuately we can only check this if the destination
1288 // region is a TypedValueRegion.
1289 if (const TypedValueRegion *TVR = dyn_cast<TypedValueRegion>(R)) {
1290 QualType RegionTy = TVR->getValueType();
1291 QualType SourceRegionTy = LCV->getRegion()->getValueType();
1292 if (!SVB.getContext().hasSameUnqualifiedType(RegionTy, SourceRegionTy))
1293 return None;
1294 }
1295
1296 if (!AllowSubregionBindings) {
1297 // If there are any other bindings within this region, we shouldn't reuse
1298 // the top-level binding.
1299 SmallVector<BindingPair, 16> Bindings;
1300 collectSubRegionBindings(Bindings, SVB, *B.lookup(R->getBaseRegion()), R,
1301 /*IncludeAllDefaultBindings=*/true);
1302 if (Bindings.size() > 1)
1303 return None;
1304 }
1305
1306 return *LCV;
1307 }
1308
1309
1310 std::pair<Store, const SubRegion *>
findLazyBinding(RegionBindingsConstRef B,const SubRegion * R,const SubRegion * originalRegion)1311 RegionStoreManager::findLazyBinding(RegionBindingsConstRef B,
1312 const SubRegion *R,
1313 const SubRegion *originalRegion) {
1314 if (originalRegion != R) {
1315 if (Optional<nonloc::LazyCompoundVal> V =
1316 getExistingLazyBinding(svalBuilder, B, R, true))
1317 return std::make_pair(V->getStore(), V->getRegion());
1318 }
1319
1320 typedef std::pair<Store, const SubRegion *> StoreRegionPair;
1321 StoreRegionPair Result = StoreRegionPair();
1322
1323 if (const ElementRegion *ER = dyn_cast<ElementRegion>(R)) {
1324 Result = findLazyBinding(B, cast<SubRegion>(ER->getSuperRegion()),
1325 originalRegion);
1326
1327 if (Result.second)
1328 Result.second = MRMgr.getElementRegionWithSuper(ER, Result.second);
1329
1330 } else if (const FieldRegion *FR = dyn_cast<FieldRegion>(R)) {
1331 Result = findLazyBinding(B, cast<SubRegion>(FR->getSuperRegion()),
1332 originalRegion);
1333
1334 if (Result.second)
1335 Result.second = MRMgr.getFieldRegionWithSuper(FR, Result.second);
1336
1337 } else if (const CXXBaseObjectRegion *BaseReg =
1338 dyn_cast<CXXBaseObjectRegion>(R)) {
1339 // C++ base object region is another kind of region that we should blast
1340 // through to look for lazy compound value. It is like a field region.
1341 Result = findLazyBinding(B, cast<SubRegion>(BaseReg->getSuperRegion()),
1342 originalRegion);
1343
1344 if (Result.second)
1345 Result.second = MRMgr.getCXXBaseObjectRegionWithSuper(BaseReg,
1346 Result.second);
1347 }
1348
1349 return Result;
1350 }
1351
getBindingForElement(RegionBindingsConstRef B,const ElementRegion * R)1352 SVal RegionStoreManager::getBindingForElement(RegionBindingsConstRef B,
1353 const ElementRegion* R) {
1354 // We do not currently model bindings of the CompoundLiteralregion.
1355 if (isa<CompoundLiteralRegion>(R->getBaseRegion()))
1356 return UnknownVal();
1357
1358 // Check if the region has a binding.
1359 if (const Optional<SVal> &V = B.getDirectBinding(R))
1360 return *V;
1361
1362 const MemRegion* superR = R->getSuperRegion();
1363
1364 // Check if the region is an element region of a string literal.
1365 if (const StringRegion *StrR=dyn_cast<StringRegion>(superR)) {
1366 // FIXME: Handle loads from strings where the literal is treated as
1367 // an integer, e.g., *((unsigned int*)"hello")
1368 QualType T = Ctx.getAsArrayType(StrR->getValueType())->getElementType();
1369 if (T != Ctx.getCanonicalType(R->getElementType()))
1370 return UnknownVal();
1371
1372 const StringLiteral *Str = StrR->getStringLiteral();
1373 SVal Idx = R->getIndex();
1374 if (Optional<nonloc::ConcreteInt> CI = Idx.getAs<nonloc::ConcreteInt>()) {
1375 int64_t i = CI->getValue().getSExtValue();
1376 // Abort on string underrun. This can be possible by arbitrary
1377 // clients of getBindingForElement().
1378 if (i < 0)
1379 return UndefinedVal();
1380 int64_t length = Str->getLength();
1381 // Technically, only i == length is guaranteed to be null.
1382 // However, such overflows should be caught before reaching this point;
1383 // the only time such an access would be made is if a string literal was
1384 // used to initialize a larger array.
1385 char c = (i >= length) ? '\0' : Str->getCodeUnit(i);
1386 return svalBuilder.makeIntVal(c, T);
1387 }
1388 }
1389
1390 // Check for loads from a code text region. For such loads, just give up.
1391 if (isa<CodeTextRegion>(superR))
1392 return UnknownVal();
1393
1394 // Handle the case where we are indexing into a larger scalar object.
1395 // For example, this handles:
1396 // int x = ...
1397 // char *y = &x;
1398 // return *y;
1399 // FIXME: This is a hack, and doesn't do anything really intelligent yet.
1400 const RegionRawOffset &O = R->getAsArrayOffset();
1401
1402 // If we cannot reason about the offset, return an unknown value.
1403 if (!O.getRegion())
1404 return UnknownVal();
1405
1406 if (const TypedValueRegion *baseR =
1407 dyn_cast_or_null<TypedValueRegion>(O.getRegion())) {
1408 QualType baseT = baseR->getValueType();
1409 if (baseT->isScalarType()) {
1410 QualType elemT = R->getElementType();
1411 if (elemT->isScalarType()) {
1412 if (Ctx.getTypeSizeInChars(baseT) >= Ctx.getTypeSizeInChars(elemT)) {
1413 if (const Optional<SVal> &V = B.getDirectBinding(superR)) {
1414 if (SymbolRef parentSym = V->getAsSymbol())
1415 return svalBuilder.getDerivedRegionValueSymbolVal(parentSym, R);
1416
1417 if (V->isUnknownOrUndef())
1418 return *V;
1419 // Other cases: give up. We are indexing into a larger object
1420 // that has some value, but we don't know how to handle that yet.
1421 return UnknownVal();
1422 }
1423 }
1424 }
1425 }
1426 }
1427 return getBindingForFieldOrElementCommon(B, R, R->getElementType(),
1428 superR);
1429 }
1430
getBindingForField(RegionBindingsConstRef B,const FieldRegion * R)1431 SVal RegionStoreManager::getBindingForField(RegionBindingsConstRef B,
1432 const FieldRegion* R) {
1433
1434 // Check if the region has a binding.
1435 if (const Optional<SVal> &V = B.getDirectBinding(R))
1436 return *V;
1437
1438 QualType Ty = R->getValueType();
1439 return getBindingForFieldOrElementCommon(B, R, Ty, R->getSuperRegion());
1440 }
1441
1442 Optional<SVal>
getBindingForDerivedDefaultValue(RegionBindingsConstRef B,const MemRegion * superR,const TypedValueRegion * R,QualType Ty)1443 RegionStoreManager::getBindingForDerivedDefaultValue(RegionBindingsConstRef B,
1444 const MemRegion *superR,
1445 const TypedValueRegion *R,
1446 QualType Ty) {
1447
1448 if (const Optional<SVal> &D = B.getDefaultBinding(superR)) {
1449 const SVal &val = D.getValue();
1450 if (SymbolRef parentSym = val.getAsSymbol())
1451 return svalBuilder.getDerivedRegionValueSymbolVal(parentSym, R);
1452
1453 if (val.isZeroConstant())
1454 return svalBuilder.makeZeroVal(Ty);
1455
1456 if (val.isUnknownOrUndef())
1457 return val;
1458
1459 // Lazy bindings are usually handled through getExistingLazyBinding().
1460 // We should unify these two code paths at some point.
1461 if (val.getAs<nonloc::LazyCompoundVal>())
1462 return val;
1463
1464 llvm_unreachable("Unknown default value");
1465 }
1466
1467 return None;
1468 }
1469
getLazyBinding(const SubRegion * LazyBindingRegion,RegionBindingsRef LazyBinding)1470 SVal RegionStoreManager::getLazyBinding(const SubRegion *LazyBindingRegion,
1471 RegionBindingsRef LazyBinding) {
1472 SVal Result;
1473 if (const ElementRegion *ER = dyn_cast<ElementRegion>(LazyBindingRegion))
1474 Result = getBindingForElement(LazyBinding, ER);
1475 else
1476 Result = getBindingForField(LazyBinding,
1477 cast<FieldRegion>(LazyBindingRegion));
1478
1479 // FIXME: This is a hack to deal with RegionStore's inability to distinguish a
1480 // default value for /part/ of an aggregate from a default value for the
1481 // /entire/ aggregate. The most common case of this is when struct Outer
1482 // has as its first member a struct Inner, which is copied in from a stack
1483 // variable. In this case, even if the Outer's default value is symbolic, 0,
1484 // or unknown, it gets overridden by the Inner's default value of undefined.
1485 //
1486 // This is a general problem -- if the Inner is zero-initialized, the Outer
1487 // will now look zero-initialized. The proper way to solve this is with a
1488 // new version of RegionStore that tracks the extent of a binding as well
1489 // as the offset.
1490 //
1491 // This hack only takes care of the undefined case because that can very
1492 // quickly result in a warning.
1493 if (Result.isUndef())
1494 Result = UnknownVal();
1495
1496 return Result;
1497 }
1498
1499 SVal
getBindingForFieldOrElementCommon(RegionBindingsConstRef B,const TypedValueRegion * R,QualType Ty,const MemRegion * superR)1500 RegionStoreManager::getBindingForFieldOrElementCommon(RegionBindingsConstRef B,
1501 const TypedValueRegion *R,
1502 QualType Ty,
1503 const MemRegion *superR) {
1504
1505 // At this point we have already checked in either getBindingForElement or
1506 // getBindingForField if 'R' has a direct binding.
1507
1508 // Lazy binding?
1509 Store lazyBindingStore = NULL;
1510 const SubRegion *lazyBindingRegion = NULL;
1511 llvm::tie(lazyBindingStore, lazyBindingRegion) = findLazyBinding(B, R, R);
1512 if (lazyBindingRegion)
1513 return getLazyBinding(lazyBindingRegion,
1514 getRegionBindings(lazyBindingStore));
1515
1516 // Record whether or not we see a symbolic index. That can completely
1517 // be out of scope of our lookup.
1518 bool hasSymbolicIndex = false;
1519
1520 // FIXME: This is a hack to deal with RegionStore's inability to distinguish a
1521 // default value for /part/ of an aggregate from a default value for the
1522 // /entire/ aggregate. The most common case of this is when struct Outer
1523 // has as its first member a struct Inner, which is copied in from a stack
1524 // variable. In this case, even if the Outer's default value is symbolic, 0,
1525 // or unknown, it gets overridden by the Inner's default value of undefined.
1526 //
1527 // This is a general problem -- if the Inner is zero-initialized, the Outer
1528 // will now look zero-initialized. The proper way to solve this is with a
1529 // new version of RegionStore that tracks the extent of a binding as well
1530 // as the offset.
1531 //
1532 // This hack only takes care of the undefined case because that can very
1533 // quickly result in a warning.
1534 bool hasPartialLazyBinding = false;
1535
1536 const SubRegion *Base = dyn_cast<SubRegion>(superR);
1537 while (Base) {
1538 if (Optional<SVal> D = getBindingForDerivedDefaultValue(B, Base, R, Ty)) {
1539 if (D->getAs<nonloc::LazyCompoundVal>()) {
1540 hasPartialLazyBinding = true;
1541 break;
1542 }
1543
1544 return *D;
1545 }
1546
1547 if (const ElementRegion *ER = dyn_cast<ElementRegion>(Base)) {
1548 NonLoc index = ER->getIndex();
1549 if (!index.isConstant())
1550 hasSymbolicIndex = true;
1551 }
1552
1553 // If our super region is a field or element itself, walk up the region
1554 // hierarchy to see if there is a default value installed in an ancestor.
1555 Base = dyn_cast<SubRegion>(Base->getSuperRegion());
1556 }
1557
1558 if (R->hasStackNonParametersStorage()) {
1559 if (isa<ElementRegion>(R)) {
1560 // Currently we don't reason specially about Clang-style vectors. Check
1561 // if superR is a vector and if so return Unknown.
1562 if (const TypedValueRegion *typedSuperR =
1563 dyn_cast<TypedValueRegion>(superR)) {
1564 if (typedSuperR->getValueType()->isVectorType())
1565 return UnknownVal();
1566 }
1567 }
1568
1569 // FIXME: We also need to take ElementRegions with symbolic indexes into
1570 // account. This case handles both directly accessing an ElementRegion
1571 // with a symbolic offset, but also fields within an element with
1572 // a symbolic offset.
1573 if (hasSymbolicIndex)
1574 return UnknownVal();
1575
1576 if (!hasPartialLazyBinding)
1577 return UndefinedVal();
1578 }
1579
1580 // All other values are symbolic.
1581 return svalBuilder.getRegionValueSymbolVal(R);
1582 }
1583
getBindingForObjCIvar(RegionBindingsConstRef B,const ObjCIvarRegion * R)1584 SVal RegionStoreManager::getBindingForObjCIvar(RegionBindingsConstRef B,
1585 const ObjCIvarRegion* R) {
1586 // Check if the region has a binding.
1587 if (const Optional<SVal> &V = B.getDirectBinding(R))
1588 return *V;
1589
1590 const MemRegion *superR = R->getSuperRegion();
1591
1592 // Check if the super region has a default binding.
1593 if (const Optional<SVal> &V = B.getDefaultBinding(superR)) {
1594 if (SymbolRef parentSym = V->getAsSymbol())
1595 return svalBuilder.getDerivedRegionValueSymbolVal(parentSym, R);
1596
1597 // Other cases: give up.
1598 return UnknownVal();
1599 }
1600
1601 return getBindingForLazySymbol(R);
1602 }
1603
getConstValue(SValBuilder & SVB,const VarDecl * VD)1604 static Optional<SVal> getConstValue(SValBuilder &SVB, const VarDecl *VD) {
1605 ASTContext &Ctx = SVB.getContext();
1606 if (!VD->getType().isConstQualified())
1607 return None;
1608
1609 const Expr *Init = VD->getInit();
1610 if (!Init)
1611 return None;
1612
1613 llvm::APSInt Result;
1614 if (!Init->isGLValue() && Init->EvaluateAsInt(Result, Ctx))
1615 return SVB.makeIntVal(Result);
1616
1617 if (Init->isNullPointerConstant(Ctx, Expr::NPC_ValueDependentIsNotNull))
1618 return SVB.makeNull();
1619
1620 // FIXME: Handle other possible constant expressions.
1621 return None;
1622 }
1623
getBindingForVar(RegionBindingsConstRef B,const VarRegion * R)1624 SVal RegionStoreManager::getBindingForVar(RegionBindingsConstRef B,
1625 const VarRegion *R) {
1626
1627 // Check if the region has a binding.
1628 if (const Optional<SVal> &V = B.getDirectBinding(R))
1629 return *V;
1630
1631 // Lazily derive a value for the VarRegion.
1632 const VarDecl *VD = R->getDecl();
1633 const MemSpaceRegion *MS = R->getMemorySpace();
1634
1635 // Arguments are always symbolic.
1636 if (isa<StackArgumentsSpaceRegion>(MS))
1637 return svalBuilder.getRegionValueSymbolVal(R);
1638
1639 // Is 'VD' declared constant? If so, retrieve the constant value.
1640 if (Optional<SVal> V = getConstValue(svalBuilder, VD))
1641 return *V;
1642
1643 // This must come after the check for constants because closure-captured
1644 // constant variables may appear in UnknownSpaceRegion.
1645 if (isa<UnknownSpaceRegion>(MS))
1646 return svalBuilder.getRegionValueSymbolVal(R);
1647
1648 if (isa<GlobalsSpaceRegion>(MS)) {
1649 QualType T = VD->getType();
1650
1651 // Function-scoped static variables are default-initialized to 0; if they
1652 // have an initializer, it would have been processed by now.
1653 if (isa<StaticGlobalSpaceRegion>(MS))
1654 return svalBuilder.makeZeroVal(T);
1655
1656 if (Optional<SVal> V = getBindingForDerivedDefaultValue(B, MS, R, T)) {
1657 assert(!V->getAs<nonloc::LazyCompoundVal>());
1658 return V.getValue();
1659 }
1660
1661 return svalBuilder.getRegionValueSymbolVal(R);
1662 }
1663
1664 return UndefinedVal();
1665 }
1666
getBindingForLazySymbol(const TypedValueRegion * R)1667 SVal RegionStoreManager::getBindingForLazySymbol(const TypedValueRegion *R) {
1668 // All other values are symbolic.
1669 return svalBuilder.getRegionValueSymbolVal(R);
1670 }
1671
1672 const RegionStoreManager::SValListTy &
getInterestingValues(nonloc::LazyCompoundVal LCV)1673 RegionStoreManager::getInterestingValues(nonloc::LazyCompoundVal LCV) {
1674 // First, check the cache.
1675 LazyBindingsMapTy::iterator I = LazyBindingsMap.find(LCV.getCVData());
1676 if (I != LazyBindingsMap.end())
1677 return I->second;
1678
1679 // If we don't have a list of values cached, start constructing it.
1680 SValListTy List;
1681
1682 const SubRegion *LazyR = LCV.getRegion();
1683 RegionBindingsRef B = getRegionBindings(LCV.getStore());
1684
1685 // If this region had /no/ bindings at the time, there are no interesting
1686 // values to return.
1687 const ClusterBindings *Cluster = B.lookup(LazyR->getBaseRegion());
1688 if (!Cluster)
1689 return (LazyBindingsMap[LCV.getCVData()] = llvm_move(List));
1690
1691 SmallVector<BindingPair, 32> Bindings;
1692 collectSubRegionBindings(Bindings, svalBuilder, *Cluster, LazyR,
1693 /*IncludeAllDefaultBindings=*/true);
1694 for (SmallVectorImpl<BindingPair>::const_iterator I = Bindings.begin(),
1695 E = Bindings.end();
1696 I != E; ++I) {
1697 SVal V = I->second;
1698 if (V.isUnknownOrUndef() || V.isConstant())
1699 continue;
1700
1701 if (Optional<nonloc::LazyCompoundVal> InnerLCV =
1702 V.getAs<nonloc::LazyCompoundVal>()) {
1703 const SValListTy &InnerList = getInterestingValues(*InnerLCV);
1704 List.insert(List.end(), InnerList.begin(), InnerList.end());
1705 continue;
1706 }
1707
1708 List.push_back(V);
1709 }
1710
1711 return (LazyBindingsMap[LCV.getCVData()] = llvm_move(List));
1712 }
1713
createLazyBinding(RegionBindingsConstRef B,const TypedValueRegion * R)1714 NonLoc RegionStoreManager::createLazyBinding(RegionBindingsConstRef B,
1715 const TypedValueRegion *R) {
1716 if (Optional<nonloc::LazyCompoundVal> V =
1717 getExistingLazyBinding(svalBuilder, B, R, false))
1718 return *V;
1719
1720 return svalBuilder.makeLazyCompoundVal(StoreRef(B.asStore(), *this), R);
1721 }
1722
getBindingForStruct(RegionBindingsConstRef B,const TypedValueRegion * R)1723 SVal RegionStoreManager::getBindingForStruct(RegionBindingsConstRef B,
1724 const TypedValueRegion *R) {
1725 const RecordDecl *RD = R->getValueType()->castAs<RecordType>()->getDecl();
1726 if (RD->field_empty())
1727 return UnknownVal();
1728
1729 return createLazyBinding(B, R);
1730 }
1731
getBindingForArray(RegionBindingsConstRef B,const TypedValueRegion * R)1732 SVal RegionStoreManager::getBindingForArray(RegionBindingsConstRef B,
1733 const TypedValueRegion *R) {
1734 assert(Ctx.getAsConstantArrayType(R->getValueType()) &&
1735 "Only constant array types can have compound bindings.");
1736
1737 return createLazyBinding(B, R);
1738 }
1739
includedInBindings(Store store,const MemRegion * region) const1740 bool RegionStoreManager::includedInBindings(Store store,
1741 const MemRegion *region) const {
1742 RegionBindingsRef B = getRegionBindings(store);
1743 region = region->getBaseRegion();
1744
1745 // Quick path: if the base is the head of a cluster, the region is live.
1746 if (B.lookup(region))
1747 return true;
1748
1749 // Slow path: if the region is the VALUE of any binding, it is live.
1750 for (RegionBindingsRef::iterator RI = B.begin(), RE = B.end(); RI != RE; ++RI) {
1751 const ClusterBindings &Cluster = RI.getData();
1752 for (ClusterBindings::iterator CI = Cluster.begin(), CE = Cluster.end();
1753 CI != CE; ++CI) {
1754 const SVal &D = CI.getData();
1755 if (const MemRegion *R = D.getAsRegion())
1756 if (R->getBaseRegion() == region)
1757 return true;
1758 }
1759 }
1760
1761 return false;
1762 }
1763
1764 //===----------------------------------------------------------------------===//
1765 // Binding values to regions.
1766 //===----------------------------------------------------------------------===//
1767
killBinding(Store ST,Loc L)1768 StoreRef RegionStoreManager::killBinding(Store ST, Loc L) {
1769 if (Optional<loc::MemRegionVal> LV = L.getAs<loc::MemRegionVal>())
1770 if (const MemRegion* R = LV->getRegion())
1771 return StoreRef(getRegionBindings(ST).removeBinding(R)
1772 .asImmutableMap()
1773 .getRootWithoutRetain(),
1774 *this);
1775
1776 return StoreRef(ST, *this);
1777 }
1778
1779 RegionBindingsRef
bind(RegionBindingsConstRef B,Loc L,SVal V)1780 RegionStoreManager::bind(RegionBindingsConstRef B, Loc L, SVal V) {
1781 if (L.getAs<loc::ConcreteInt>())
1782 return B;
1783
1784 // If we get here, the location should be a region.
1785 const MemRegion *R = L.castAs<loc::MemRegionVal>().getRegion();
1786
1787 // Check if the region is a struct region.
1788 if (const TypedValueRegion* TR = dyn_cast<TypedValueRegion>(R)) {
1789 QualType Ty = TR->getValueType();
1790 if (Ty->isArrayType())
1791 return bindArray(B, TR, V);
1792 if (Ty->isStructureOrClassType())
1793 return bindStruct(B, TR, V);
1794 if (Ty->isVectorType())
1795 return bindVector(B, TR, V);
1796 }
1797
1798 if (const SymbolicRegion *SR = dyn_cast<SymbolicRegion>(R)) {
1799 // Binding directly to a symbolic region should be treated as binding
1800 // to element 0.
1801 QualType T = SR->getSymbol()->getType();
1802 if (T->isAnyPointerType() || T->isReferenceType())
1803 T = T->getPointeeType();
1804
1805 R = GetElementZeroRegion(SR, T);
1806 }
1807
1808 // Clear out bindings that may overlap with this binding.
1809 RegionBindingsRef NewB = removeSubRegionBindings(B, cast<SubRegion>(R));
1810 return NewB.addBinding(BindingKey::Make(R, BindingKey::Direct), V);
1811 }
1812
1813 // FIXME: this method should be merged into Bind().
bindCompoundLiteral(Store ST,const CompoundLiteralExpr * CL,const LocationContext * LC,SVal V)1814 StoreRef RegionStoreManager::bindCompoundLiteral(Store ST,
1815 const CompoundLiteralExpr *CL,
1816 const LocationContext *LC,
1817 SVal V) {
1818 return Bind(ST, loc::MemRegionVal(MRMgr.getCompoundLiteralRegion(CL, LC)), V);
1819 }
1820
1821 RegionBindingsRef
setImplicitDefaultValue(RegionBindingsConstRef B,const MemRegion * R,QualType T)1822 RegionStoreManager::setImplicitDefaultValue(RegionBindingsConstRef B,
1823 const MemRegion *R,
1824 QualType T) {
1825 SVal V;
1826
1827 if (Loc::isLocType(T))
1828 V = svalBuilder.makeNull();
1829 else if (T->isIntegerType())
1830 V = svalBuilder.makeZeroVal(T);
1831 else if (T->isStructureOrClassType() || T->isArrayType()) {
1832 // Set the default value to a zero constant when it is a structure
1833 // or array. The type doesn't really matter.
1834 V = svalBuilder.makeZeroVal(Ctx.IntTy);
1835 }
1836 else {
1837 // We can't represent values of this type, but we still need to set a value
1838 // to record that the region has been initialized.
1839 // If this assertion ever fires, a new case should be added above -- we
1840 // should know how to default-initialize any value we can symbolicate.
1841 assert(!SymbolManager::canSymbolicate(T) && "This type is representable");
1842 V = UnknownVal();
1843 }
1844
1845 return B.addBinding(R, BindingKey::Default, V);
1846 }
1847
1848 RegionBindingsRef
bindArray(RegionBindingsConstRef B,const TypedValueRegion * R,SVal Init)1849 RegionStoreManager::bindArray(RegionBindingsConstRef B,
1850 const TypedValueRegion* R,
1851 SVal Init) {
1852
1853 const ArrayType *AT =cast<ArrayType>(Ctx.getCanonicalType(R->getValueType()));
1854 QualType ElementTy = AT->getElementType();
1855 Optional<uint64_t> Size;
1856
1857 if (const ConstantArrayType* CAT = dyn_cast<ConstantArrayType>(AT))
1858 Size = CAT->getSize().getZExtValue();
1859
1860 // Check if the init expr is a string literal.
1861 if (Optional<loc::MemRegionVal> MRV = Init.getAs<loc::MemRegionVal>()) {
1862 const StringRegion *S = cast<StringRegion>(MRV->getRegion());
1863
1864 // Treat the string as a lazy compound value.
1865 StoreRef store(B.asStore(), *this);
1866 nonloc::LazyCompoundVal LCV = svalBuilder.makeLazyCompoundVal(store, S)
1867 .castAs<nonloc::LazyCompoundVal>();
1868 return bindAggregate(B, R, LCV);
1869 }
1870
1871 // Handle lazy compound values.
1872 if (Init.getAs<nonloc::LazyCompoundVal>())
1873 return bindAggregate(B, R, Init);
1874
1875 // Remaining case: explicit compound values.
1876
1877 if (Init.isUnknown())
1878 return setImplicitDefaultValue(B, R, ElementTy);
1879
1880 const nonloc::CompoundVal& CV = Init.castAs<nonloc::CompoundVal>();
1881 nonloc::CompoundVal::iterator VI = CV.begin(), VE = CV.end();
1882 uint64_t i = 0;
1883
1884 RegionBindingsRef NewB(B);
1885
1886 for (; Size.hasValue() ? i < Size.getValue() : true ; ++i, ++VI) {
1887 // The init list might be shorter than the array length.
1888 if (VI == VE)
1889 break;
1890
1891 const NonLoc &Idx = svalBuilder.makeArrayIndex(i);
1892 const ElementRegion *ER = MRMgr.getElementRegion(ElementTy, Idx, R, Ctx);
1893
1894 if (ElementTy->isStructureOrClassType())
1895 NewB = bindStruct(NewB, ER, *VI);
1896 else if (ElementTy->isArrayType())
1897 NewB = bindArray(NewB, ER, *VI);
1898 else
1899 NewB = bind(NewB, svalBuilder.makeLoc(ER), *VI);
1900 }
1901
1902 // If the init list is shorter than the array length, set the
1903 // array default value.
1904 if (Size.hasValue() && i < Size.getValue())
1905 NewB = setImplicitDefaultValue(NewB, R, ElementTy);
1906
1907 return NewB;
1908 }
1909
bindVector(RegionBindingsConstRef B,const TypedValueRegion * R,SVal V)1910 RegionBindingsRef RegionStoreManager::bindVector(RegionBindingsConstRef B,
1911 const TypedValueRegion* R,
1912 SVal V) {
1913 QualType T = R->getValueType();
1914 assert(T->isVectorType());
1915 const VectorType *VT = T->getAs<VectorType>(); // Use getAs for typedefs.
1916
1917 // Handle lazy compound values and symbolic values.
1918 if (V.getAs<nonloc::LazyCompoundVal>() || V.getAs<nonloc::SymbolVal>())
1919 return bindAggregate(B, R, V);
1920
1921 // We may get non-CompoundVal accidentally due to imprecise cast logic or
1922 // that we are binding symbolic struct value. Kill the field values, and if
1923 // the value is symbolic go and bind it as a "default" binding.
1924 if (!V.getAs<nonloc::CompoundVal>()) {
1925 return bindAggregate(B, R, UnknownVal());
1926 }
1927
1928 QualType ElemType = VT->getElementType();
1929 nonloc::CompoundVal CV = V.castAs<nonloc::CompoundVal>();
1930 nonloc::CompoundVal::iterator VI = CV.begin(), VE = CV.end();
1931 unsigned index = 0, numElements = VT->getNumElements();
1932 RegionBindingsRef NewB(B);
1933
1934 for ( ; index != numElements ; ++index) {
1935 if (VI == VE)
1936 break;
1937
1938 NonLoc Idx = svalBuilder.makeArrayIndex(index);
1939 const ElementRegion *ER = MRMgr.getElementRegion(ElemType, Idx, R, Ctx);
1940
1941 if (ElemType->isArrayType())
1942 NewB = bindArray(NewB, ER, *VI);
1943 else if (ElemType->isStructureOrClassType())
1944 NewB = bindStruct(NewB, ER, *VI);
1945 else
1946 NewB = bind(NewB, svalBuilder.makeLoc(ER), *VI);
1947 }
1948 return NewB;
1949 }
1950
bindStruct(RegionBindingsConstRef B,const TypedValueRegion * R,SVal V)1951 RegionBindingsRef RegionStoreManager::bindStruct(RegionBindingsConstRef B,
1952 const TypedValueRegion* R,
1953 SVal V) {
1954 if (!Features.supportsFields())
1955 return B;
1956
1957 QualType T = R->getValueType();
1958 assert(T->isStructureOrClassType());
1959
1960 const RecordType* RT = T->getAs<RecordType>();
1961 RecordDecl *RD = RT->getDecl();
1962
1963 if (!RD->isCompleteDefinition())
1964 return B;
1965
1966 // Handle lazy compound values and symbolic values.
1967 if (V.getAs<nonloc::LazyCompoundVal>() || V.getAs<nonloc::SymbolVal>())
1968 return bindAggregate(B, R, V);
1969
1970 // We may get non-CompoundVal accidentally due to imprecise cast logic or
1971 // that we are binding symbolic struct value. Kill the field values, and if
1972 // the value is symbolic go and bind it as a "default" binding.
1973 if (V.isUnknown() || !V.getAs<nonloc::CompoundVal>())
1974 return bindAggregate(B, R, UnknownVal());
1975
1976 const nonloc::CompoundVal& CV = V.castAs<nonloc::CompoundVal>();
1977 nonloc::CompoundVal::iterator VI = CV.begin(), VE = CV.end();
1978
1979 RecordDecl::field_iterator FI, FE;
1980 RegionBindingsRef NewB(B);
1981
1982 for (FI = RD->field_begin(), FE = RD->field_end(); FI != FE; ++FI) {
1983
1984 if (VI == VE)
1985 break;
1986
1987 // Skip any unnamed bitfields to stay in sync with the initializers.
1988 if (FI->isUnnamedBitfield())
1989 continue;
1990
1991 QualType FTy = FI->getType();
1992 const FieldRegion* FR = MRMgr.getFieldRegion(*FI, R);
1993
1994 if (FTy->isArrayType())
1995 NewB = bindArray(NewB, FR, *VI);
1996 else if (FTy->isStructureOrClassType())
1997 NewB = bindStruct(NewB, FR, *VI);
1998 else
1999 NewB = bind(NewB, svalBuilder.makeLoc(FR), *VI);
2000 ++VI;
2001 }
2002
2003 // There may be fewer values in the initialize list than the fields of struct.
2004 if (FI != FE) {
2005 NewB = NewB.addBinding(R, BindingKey::Default,
2006 svalBuilder.makeIntVal(0, false));
2007 }
2008
2009 return NewB;
2010 }
2011
2012 RegionBindingsRef
bindAggregate(RegionBindingsConstRef B,const TypedRegion * R,SVal Val)2013 RegionStoreManager::bindAggregate(RegionBindingsConstRef B,
2014 const TypedRegion *R,
2015 SVal Val) {
2016 // Remove the old bindings, using 'R' as the root of all regions
2017 // we will invalidate. Then add the new binding.
2018 return removeSubRegionBindings(B, R).addBinding(R, BindingKey::Default, Val);
2019 }
2020
2021 //===----------------------------------------------------------------------===//
2022 // State pruning.
2023 //===----------------------------------------------------------------------===//
2024
2025 namespace {
2026 class removeDeadBindingsWorker :
2027 public ClusterAnalysis<removeDeadBindingsWorker> {
2028 SmallVector<const SymbolicRegion*, 12> Postponed;
2029 SymbolReaper &SymReaper;
2030 const StackFrameContext *CurrentLCtx;
2031
2032 public:
removeDeadBindingsWorker(RegionStoreManager & rm,ProgramStateManager & stateMgr,RegionBindingsRef b,SymbolReaper & symReaper,const StackFrameContext * LCtx)2033 removeDeadBindingsWorker(RegionStoreManager &rm,
2034 ProgramStateManager &stateMgr,
2035 RegionBindingsRef b, SymbolReaper &symReaper,
2036 const StackFrameContext *LCtx)
2037 : ClusterAnalysis<removeDeadBindingsWorker>(rm, stateMgr, b,
2038 /* includeGlobals = */ false),
2039 SymReaper(symReaper), CurrentLCtx(LCtx) {}
2040
2041 // Called by ClusterAnalysis.
2042 void VisitAddedToCluster(const MemRegion *baseR, const ClusterBindings &C);
2043 void VisitCluster(const MemRegion *baseR, const ClusterBindings &C);
2044
2045 bool UpdatePostponed();
2046 void VisitBinding(SVal V);
2047 };
2048 }
2049
VisitAddedToCluster(const MemRegion * baseR,const ClusterBindings & C)2050 void removeDeadBindingsWorker::VisitAddedToCluster(const MemRegion *baseR,
2051 const ClusterBindings &C) {
2052
2053 if (const VarRegion *VR = dyn_cast<VarRegion>(baseR)) {
2054 if (SymReaper.isLive(VR))
2055 AddToWorkList(baseR, &C);
2056
2057 return;
2058 }
2059
2060 if (const SymbolicRegion *SR = dyn_cast<SymbolicRegion>(baseR)) {
2061 if (SymReaper.isLive(SR->getSymbol()))
2062 AddToWorkList(SR, &C);
2063 else
2064 Postponed.push_back(SR);
2065
2066 return;
2067 }
2068
2069 if (isa<NonStaticGlobalSpaceRegion>(baseR)) {
2070 AddToWorkList(baseR, &C);
2071 return;
2072 }
2073
2074 // CXXThisRegion in the current or parent location context is live.
2075 if (const CXXThisRegion *TR = dyn_cast<CXXThisRegion>(baseR)) {
2076 const StackArgumentsSpaceRegion *StackReg =
2077 cast<StackArgumentsSpaceRegion>(TR->getSuperRegion());
2078 const StackFrameContext *RegCtx = StackReg->getStackFrame();
2079 if (CurrentLCtx &&
2080 (RegCtx == CurrentLCtx || RegCtx->isParentOf(CurrentLCtx)))
2081 AddToWorkList(TR, &C);
2082 }
2083 }
2084
VisitCluster(const MemRegion * baseR,const ClusterBindings & C)2085 void removeDeadBindingsWorker::VisitCluster(const MemRegion *baseR,
2086 const ClusterBindings &C) {
2087 // Mark the symbol for any SymbolicRegion with live bindings as live itself.
2088 // This means we should continue to track that symbol.
2089 if (const SymbolicRegion *SymR = dyn_cast<SymbolicRegion>(baseR))
2090 SymReaper.markLive(SymR->getSymbol());
2091
2092 for (ClusterBindings::iterator I = C.begin(), E = C.end(); I != E; ++I)
2093 VisitBinding(I.getData());
2094 }
2095
VisitBinding(SVal V)2096 void removeDeadBindingsWorker::VisitBinding(SVal V) {
2097 // Is it a LazyCompoundVal? All referenced regions are live as well.
2098 if (Optional<nonloc::LazyCompoundVal> LCS =
2099 V.getAs<nonloc::LazyCompoundVal>()) {
2100
2101 const RegionStoreManager::SValListTy &Vals = RM.getInterestingValues(*LCS);
2102
2103 for (RegionStoreManager::SValListTy::const_iterator I = Vals.begin(),
2104 E = Vals.end();
2105 I != E; ++I)
2106 VisitBinding(*I);
2107
2108 return;
2109 }
2110
2111 // If V is a region, then add it to the worklist.
2112 if (const MemRegion *R = V.getAsRegion()) {
2113 AddToWorkList(R);
2114
2115 // All regions captured by a block are also live.
2116 if (const BlockDataRegion *BR = dyn_cast<BlockDataRegion>(R)) {
2117 BlockDataRegion::referenced_vars_iterator I = BR->referenced_vars_begin(),
2118 E = BR->referenced_vars_end();
2119 for ( ; I != E; ++I)
2120 AddToWorkList(I.getCapturedRegion());
2121 }
2122 }
2123
2124
2125 // Update the set of live symbols.
2126 for (SymExpr::symbol_iterator SI = V.symbol_begin(), SE = V.symbol_end();
2127 SI!=SE; ++SI)
2128 SymReaper.markLive(*SI);
2129 }
2130
UpdatePostponed()2131 bool removeDeadBindingsWorker::UpdatePostponed() {
2132 // See if any postponed SymbolicRegions are actually live now, after
2133 // having done a scan.
2134 bool changed = false;
2135
2136 for (SmallVectorImpl<const SymbolicRegion*>::iterator
2137 I = Postponed.begin(), E = Postponed.end() ; I != E ; ++I) {
2138 if (const SymbolicRegion *SR = *I) {
2139 if (SymReaper.isLive(SR->getSymbol())) {
2140 changed |= AddToWorkList(SR);
2141 *I = NULL;
2142 }
2143 }
2144 }
2145
2146 return changed;
2147 }
2148
removeDeadBindings(Store store,const StackFrameContext * LCtx,SymbolReaper & SymReaper)2149 StoreRef RegionStoreManager::removeDeadBindings(Store store,
2150 const StackFrameContext *LCtx,
2151 SymbolReaper& SymReaper) {
2152 RegionBindingsRef B = getRegionBindings(store);
2153 removeDeadBindingsWorker W(*this, StateMgr, B, SymReaper, LCtx);
2154 W.GenerateClusters();
2155
2156 // Enqueue the region roots onto the worklist.
2157 for (SymbolReaper::region_iterator I = SymReaper.region_begin(),
2158 E = SymReaper.region_end(); I != E; ++I) {
2159 W.AddToWorkList(*I);
2160 }
2161
2162 do W.RunWorkList(); while (W.UpdatePostponed());
2163
2164 // We have now scanned the store, marking reachable regions and symbols
2165 // as live. We now remove all the regions that are dead from the store
2166 // as well as update DSymbols with the set symbols that are now dead.
2167 for (RegionBindingsRef::iterator I = B.begin(), E = B.end(); I != E; ++I) {
2168 const MemRegion *Base = I.getKey();
2169
2170 // If the cluster has been visited, we know the region has been marked.
2171 if (W.isVisited(Base))
2172 continue;
2173
2174 // Remove the dead entry.
2175 B = B.remove(Base);
2176
2177 if (const SymbolicRegion *SymR = dyn_cast<SymbolicRegion>(Base))
2178 SymReaper.maybeDead(SymR->getSymbol());
2179
2180 // Mark all non-live symbols that this binding references as dead.
2181 const ClusterBindings &Cluster = I.getData();
2182 for (ClusterBindings::iterator CI = Cluster.begin(), CE = Cluster.end();
2183 CI != CE; ++CI) {
2184 SVal X = CI.getData();
2185 SymExpr::symbol_iterator SI = X.symbol_begin(), SE = X.symbol_end();
2186 for (; SI != SE; ++SI)
2187 SymReaper.maybeDead(*SI);
2188 }
2189 }
2190
2191 return StoreRef(B.asStore(), *this);
2192 }
2193
2194 //===----------------------------------------------------------------------===//
2195 // Utility methods.
2196 //===----------------------------------------------------------------------===//
2197
print(Store store,raw_ostream & OS,const char * nl,const char * sep)2198 void RegionStoreManager::print(Store store, raw_ostream &OS,
2199 const char* nl, const char *sep) {
2200 RegionBindingsRef B = getRegionBindings(store);
2201 OS << "Store (direct and default bindings), "
2202 << B.asStore()
2203 << " :" << nl;
2204 B.dump(OS, nl);
2205 }
2206