1 //===- Store.cpp - Interface for maps from Locations to Values ------------===//
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
7 //===----------------------------------------------------------------------===//
9 // This file defined the types Store and StoreManager.
11 //===----------------------------------------------------------------------===//
13 #include "clang/StaticAnalyzer/Core/PathSensitive/Store.h"
14 #include "clang/AST/ASTContext.h"
15 #include "clang/AST/CXXInheritance.h"
16 #include "clang/AST/CharUnits.h"
17 #include "clang/AST/Decl.h"
18 #include "clang/AST/DeclCXX.h"
19 #include "clang/AST/DeclObjC.h"
20 #include "clang/AST/Expr.h"
21 #include "clang/AST/Type.h"
22 #include "clang/Basic/LLVM.h"
23 #include "clang/StaticAnalyzer/Core/PathSensitive/BasicValueFactory.h"
24 #include "clang/StaticAnalyzer/Core/PathSensitive/CallEvent.h"
25 #include "clang/StaticAnalyzer/Core/PathSensitive/MemRegion.h"
26 #include "clang/StaticAnalyzer/Core/PathSensitive/ProgramState.h"
27 #include "clang/StaticAnalyzer/Core/PathSensitive/SValBuilder.h"
28 #include "clang/StaticAnalyzer/Core/PathSensitive/SVals.h"
29 #include "clang/StaticAnalyzer/Core/PathSensitive/StoreRef.h"
30 #include "clang/StaticAnalyzer/Core/PathSensitive/SymExpr.h"
31 #include "llvm/ADT/APSInt.h"
32 #include "llvm/ADT/Optional.h"
33 #include "llvm/ADT/SmallVector.h"
34 #include "llvm/Support/Casting.h"
35 #include "llvm/Support/ErrorHandling.h"
39 using namespace clang;
42 StoreManager::StoreManager(ProgramStateManager &stateMgr)
43 : svalBuilder(stateMgr.getSValBuilder()), StateMgr(stateMgr),
44 MRMgr(svalBuilder.getRegionManager()), Ctx(stateMgr.getContext()) {}
46 StoreRef StoreManager::enterStackFrame(Store OldStore,
47 const CallEvent &Call,
48 const StackFrameContext *LCtx) {
49 StoreRef Store = StoreRef(OldStore, *this);
51 SmallVector<CallEvent::FrameBindingTy, 16> InitialBindings;
52 Call.getInitialStackFrameContents(LCtx, InitialBindings);
54 for (const auto &I : InitialBindings)
55 Store = Bind(Store.getStore(), I.first.castAs<Loc>(), I.second);
60 const ElementRegion *StoreManager::MakeElementRegion(const SubRegion *Base,
63 NonLoc idx = svalBuilder.makeArrayIndex(index);
64 return MRMgr.getElementRegion(EleTy, idx, Base, svalBuilder.getContext());
67 const ElementRegion *StoreManager::GetElementZeroRegion(const SubRegion *R,
69 NonLoc idx = svalBuilder.makeZeroArrayIndex();
71 return MRMgr.getElementRegion(T, idx, R, Ctx);
74 const MemRegion *StoreManager::castRegion(const MemRegion *R, QualType CastToTy) {
75 ASTContext &Ctx = StateMgr.getContext();
77 // Handle casts to Objective-C objects.
78 if (CastToTy->isObjCObjectPointerType())
79 return R->StripCasts();
81 if (CastToTy->isBlockPointerType()) {
82 // FIXME: We may need different solutions, depending on the symbol
83 // involved. Blocks can be casted to/from 'id', as they can be treated
84 // as Objective-C objects. This could possibly be handled by enhancing
85 // our reasoning of downcasts of symbolic objects.
86 if (isa<CodeTextRegion>(R) || isa<SymbolicRegion>(R))
89 // We don't know what to make of it. Return a NULL region, which
90 // will be interpreted as UnknownVal.
94 // Now assume we are casting from pointer to pointer. Other cases should
95 // already be handled.
96 QualType PointeeTy = CastToTy->getPointeeType();
97 QualType CanonPointeeTy = Ctx.getCanonicalType(PointeeTy);
99 // Handle casts to void*. We just pass the region through.
100 if (CanonPointeeTy.getLocalUnqualifiedType() == Ctx.VoidTy)
103 // Handle casts from compatible types.
104 if (R->isBoundable())
105 if (const auto *TR = dyn_cast<TypedValueRegion>(R)) {
106 QualType ObjTy = Ctx.getCanonicalType(TR->getValueType());
107 if (CanonPointeeTy == ObjTy)
111 // Process region cast according to the kind of the region being cast.
112 switch (R->getKind()) {
113 case MemRegion::CXXThisRegionKind:
114 case MemRegion::CodeSpaceRegionKind:
115 case MemRegion::StackLocalsSpaceRegionKind:
116 case MemRegion::StackArgumentsSpaceRegionKind:
117 case MemRegion::HeapSpaceRegionKind:
118 case MemRegion::UnknownSpaceRegionKind:
119 case MemRegion::StaticGlobalSpaceRegionKind:
120 case MemRegion::GlobalInternalSpaceRegionKind:
121 case MemRegion::GlobalSystemSpaceRegionKind:
122 case MemRegion::GlobalImmutableSpaceRegionKind: {
123 llvm_unreachable("Invalid region cast");
126 case MemRegion::FunctionCodeRegionKind:
127 case MemRegion::BlockCodeRegionKind:
128 case MemRegion::BlockDataRegionKind:
129 case MemRegion::StringRegionKind:
130 // FIXME: Need to handle arbitrary downcasts.
131 case MemRegion::SymbolicRegionKind:
132 case MemRegion::AllocaRegionKind:
133 case MemRegion::CompoundLiteralRegionKind:
134 case MemRegion::FieldRegionKind:
135 case MemRegion::ObjCIvarRegionKind:
136 case MemRegion::ObjCStringRegionKind:
137 case MemRegion::NonParamVarRegionKind:
138 case MemRegion::ParamVarRegionKind:
139 case MemRegion::CXXTempObjectRegionKind:
140 case MemRegion::CXXBaseObjectRegionKind:
141 case MemRegion::CXXDerivedObjectRegionKind:
142 return MakeElementRegion(cast<SubRegion>(R), PointeeTy);
144 case MemRegion::ElementRegionKind: {
145 // If we are casting from an ElementRegion to another type, the
146 // algorithm is as follows:
148 // (1) Compute the "raw offset" of the ElementRegion from the
149 // base region. This is done by calling 'getAsRawOffset()'.
151 // (2a) If we get a 'RegionRawOffset' after calling
152 // 'getAsRawOffset()', determine if the absolute offset
153 // can be exactly divided into chunks of the size of the
154 // casted-pointee type. If so, create a new ElementRegion with
155 // the pointee-cast type as the new ElementType and the index
156 // being the offset divded by the chunk size. If not, create
157 // a new ElementRegion at offset 0 off the raw offset region.
159 // (2b) If we don't a get a 'RegionRawOffset' after calling
160 // 'getAsRawOffset()', it means that we are at offset 0.
162 // FIXME: Handle symbolic raw offsets.
164 const ElementRegion *elementR = cast<ElementRegion>(R);
165 const RegionRawOffset &rawOff = elementR->getAsArrayOffset();
166 const MemRegion *baseR = rawOff.getRegion();
168 // If we cannot compute a raw offset, throw up our hands and return
169 // a NULL MemRegion*.
173 CharUnits off = rawOff.getOffset();
176 // Edge case: we are at 0 bytes off the beginning of baseR. We
177 // check to see if type we are casting to is the same as the base
178 // region. If so, just return the base region.
179 if (const auto *TR = dyn_cast<TypedValueRegion>(baseR)) {
180 QualType ObjTy = Ctx.getCanonicalType(TR->getValueType());
181 QualType CanonPointeeTy = Ctx.getCanonicalType(PointeeTy);
182 if (CanonPointeeTy == ObjTy)
186 // Otherwise, create a new ElementRegion at offset 0.
187 return MakeElementRegion(cast<SubRegion>(baseR), PointeeTy);
190 // We have a non-zero offset from the base region. We want to determine
191 // if the offset can be evenly divided by sizeof(PointeeTy). If so,
192 // we create an ElementRegion whose index is that value. Otherwise, we
193 // create two ElementRegions, one that reflects a raw offset and the other
194 // that reflects the cast.
196 // Compute the index for the new ElementRegion.
197 int64_t newIndex = 0;
198 const MemRegion *newSuperR = nullptr;
200 // We can only compute sizeof(PointeeTy) if it is a complete type.
201 if (!PointeeTy->isIncompleteType()) {
202 // Compute the size in **bytes**.
203 CharUnits pointeeTySize = Ctx.getTypeSizeInChars(PointeeTy);
204 if (!pointeeTySize.isZero()) {
205 // Is the offset a multiple of the size? If so, we can layer the
206 // ElementRegion (with elementType == PointeeTy) directly on top of
208 if (off % pointeeTySize == 0) {
209 newIndex = off / pointeeTySize;
216 // Create an intermediate ElementRegion to represent the raw byte.
217 // This will be the super region of the final ElementRegion.
218 newSuperR = MakeElementRegion(cast<SubRegion>(baseR), Ctx.CharTy,
222 return MakeElementRegion(cast<SubRegion>(newSuperR), PointeeTy, newIndex);
226 llvm_unreachable("unreachable");
229 static bool regionMatchesCXXRecordType(SVal V, QualType Ty) {
230 const MemRegion *MR = V.getAsRegion();
234 const auto *TVR = dyn_cast<TypedValueRegion>(MR);
238 const CXXRecordDecl *RD = TVR->getValueType()->getAsCXXRecordDecl();
242 const CXXRecordDecl *Expected = Ty->getPointeeCXXRecordDecl();
244 Expected = Ty->getAsCXXRecordDecl();
246 return Expected->getCanonicalDecl() == RD->getCanonicalDecl();
249 SVal StoreManager::evalDerivedToBase(SVal Derived, const CastExpr *Cast) {
250 // Sanity check to avoid doing the wrong thing in the face of
252 if (!regionMatchesCXXRecordType(Derived, Cast->getSubExpr()->getType()))
255 // Walk through the cast path to create nested CXXBaseRegions.
256 SVal Result = Derived;
257 for (CastExpr::path_const_iterator I = Cast->path_begin(),
258 E = Cast->path_end();
260 Result = evalDerivedToBase(Result, (*I)->getType(), (*I)->isVirtual());
265 SVal StoreManager::evalDerivedToBase(SVal Derived, const CXXBasePath &Path) {
266 // Walk through the path to create nested CXXBaseRegions.
267 SVal Result = Derived;
268 for (const auto &I : Path)
269 Result = evalDerivedToBase(Result, I.Base->getType(),
270 I.Base->isVirtual());
274 SVal StoreManager::evalDerivedToBase(SVal Derived, QualType BaseType,
276 const MemRegion *DerivedReg = Derived.getAsRegion();
280 const CXXRecordDecl *BaseDecl = BaseType->getPointeeCXXRecordDecl();
282 BaseDecl = BaseType->getAsCXXRecordDecl();
283 assert(BaseDecl && "not a C++ object?");
285 if (const auto *AlreadyDerivedReg =
286 dyn_cast<CXXDerivedObjectRegion>(DerivedReg)) {
288 dyn_cast<SymbolicRegion>(AlreadyDerivedReg->getSuperRegion()))
289 if (SR->getSymbol()->getType()->getPointeeCXXRecordDecl() == BaseDecl)
290 return loc::MemRegionVal(SR);
292 DerivedReg = AlreadyDerivedReg->getSuperRegion();
295 const MemRegion *BaseReg = MRMgr.getCXXBaseObjectRegion(
296 BaseDecl, cast<SubRegion>(DerivedReg), IsVirtual);
298 return loc::MemRegionVal(BaseReg);
301 /// Returns the static type of the given region, if it represents a C++ class
304 /// This handles both fully-typed regions, where the dynamic type is known, and
305 /// symbolic regions, where the dynamic type is merely bounded (and even then,
306 /// only ostensibly!), but does not take advantage of any dynamic type info.
307 static const CXXRecordDecl *getCXXRecordType(const MemRegion *MR) {
308 if (const auto *TVR = dyn_cast<TypedValueRegion>(MR))
309 return TVR->getValueType()->getAsCXXRecordDecl();
310 if (const auto *SR = dyn_cast<SymbolicRegion>(MR))
311 return SR->getSymbol()->getType()->getPointeeCXXRecordDecl();
315 SVal StoreManager::attemptDownCast(SVal Base, QualType TargetType,
319 const MemRegion *MR = Base.getAsRegion();
323 // Assume the derived class is a pointer or a reference to a CXX record.
324 TargetType = TargetType->getPointeeType();
325 assert(!TargetType.isNull());
326 const CXXRecordDecl *TargetClass = TargetType->getAsCXXRecordDecl();
327 if (!TargetClass && !TargetType->isVoidType())
330 // Drill down the CXXBaseObject chains, which represent upcasts (casts from
332 while (const CXXRecordDecl *MRClass = getCXXRecordType(MR)) {
333 // If found the derived class, the cast succeeds.
334 if (MRClass == TargetClass)
335 return loc::MemRegionVal(MR);
337 // We skip over incomplete types. They must be the result of an earlier
338 // reinterpret_cast, as one can only dynamic_cast between types in the same
340 if (!TargetType->isVoidType() && MRClass->hasDefinition()) {
341 // Static upcasts are marked as DerivedToBase casts by Sema, so this will
342 // only happen when multiple or virtual inheritance is involved.
343 CXXBasePaths Paths(/*FindAmbiguities=*/false, /*RecordPaths=*/true,
344 /*DetectVirtual=*/false);
345 if (MRClass->isDerivedFrom(TargetClass, Paths))
346 return evalDerivedToBase(loc::MemRegionVal(MR), Paths.front());
349 if (const auto *BaseR = dyn_cast<CXXBaseObjectRegion>(MR)) {
350 // Drill down the chain to get the derived classes.
351 MR = BaseR->getSuperRegion();
355 // If this is a cast to void*, return the region.
356 if (TargetType->isVoidType())
357 return loc::MemRegionVal(MR);
359 // Strange use of reinterpret_cast can give us paths we don't reason
360 // about well, by putting in ElementRegions where we'd expect
361 // CXXBaseObjectRegions. If it's a valid reinterpret_cast (i.e. if the
362 // derived class has a zero offset from the base class), then it's safe
363 // to strip the cast; if it's invalid, -Wreinterpret-base-class should
364 // catch it. In the interest of performance, the analyzer will silently
365 // do the wrong thing in the invalid case (because offsets for subregions
367 const MemRegion *Uncasted = MR->StripCasts(/*IncludeBaseCasts=*/false);
368 if (Uncasted == MR) {
369 // We reached the bottom of the hierarchy and did not find the derived
370 // class. We must be casting the base to derived, so the cast should
378 // If we're casting a symbolic base pointer to a derived class, use
379 // CXXDerivedObjectRegion to represent the cast. If it's a pointer to an
380 // unrelated type, it must be a weird reinterpret_cast and we have to
381 // be fine with ElementRegion. TODO: Should we instead make
382 // Derived{TargetClass, Element{SourceClass, SR}}?
383 if (const auto *SR = dyn_cast<SymbolicRegion>(MR)) {
384 QualType T = SR->getSymbol()->getType();
385 const CXXRecordDecl *SourceClass = T->getPointeeCXXRecordDecl();
386 if (TargetClass && SourceClass && TargetClass->isDerivedFrom(SourceClass))
387 return loc::MemRegionVal(
388 MRMgr.getCXXDerivedObjectRegion(TargetClass, SR));
389 return loc::MemRegionVal(GetElementZeroRegion(SR, TargetType));
392 // We failed if the region we ended up with has perfect type info.
393 Failed = isa<TypedValueRegion>(MR);
397 static bool hasSameUnqualifiedPointeeType(QualType ty1, QualType ty2) {
398 return ty1->getPointeeType().getCanonicalType().getTypePtr() ==
399 ty2->getPointeeType().getCanonicalType().getTypePtr();
402 /// CastRetrievedVal - Used by subclasses of StoreManager to implement
403 /// implicit casts that arise from loads from regions that are reinterpreted
404 /// as another region.
405 SVal StoreManager::CastRetrievedVal(SVal V, const TypedValueRegion *R,
407 if (castTy.isNull() || V.isUnknownOrUndef())
410 // The dispatchCast() call below would convert the int into a float.
411 // What we want, however, is a bit-by-bit reinterpretation of the int
412 // as a float, which usually yields nothing garbage. For now skip casts
413 // from ints to floats.
414 // TODO: What other combinations of types are affected?
415 if (castTy->isFloatingType()) {
416 SymbolRef Sym = V.getAsSymbol();
417 if (Sym && !Sym->getType()->isFloatingType())
421 // When retrieving symbolic pointer and expecting a non-void pointer,
422 // wrap them into element regions of the expected type if necessary.
423 // SValBuilder::dispatchCast() doesn't do that, but it is necessary to
424 // make sure that the retrieved value makes sense, because there's no other
425 // cast in the AST that would tell us to cast it to the correct pointer type.
426 // We might need to do that for non-void pointers as well.
427 // FIXME: We really need a single good function to perform casts for us
428 // correctly every time we need it.
429 if (castTy->isPointerType() && !castTy->isVoidPointerType())
430 if (const auto *SR = dyn_cast_or_null<SymbolicRegion>(V.getAsRegion())) {
431 QualType sr = SR->getSymbol()->getType();
432 if (!hasSameUnqualifiedPointeeType(sr, castTy))
433 return loc::MemRegionVal(castRegion(SR, castTy));
436 return svalBuilder.dispatchCast(V, castTy);
439 SVal StoreManager::getLValueFieldOrIvar(const Decl *D, SVal Base) {
440 if (Base.isUnknownOrUndef())
443 Loc BaseL = Base.castAs<Loc>();
444 const SubRegion* BaseR = nullptr;
446 switch (BaseL.getSubKind()) {
447 case loc::MemRegionValKind:
448 BaseR = cast<SubRegion>(BaseL.castAs<loc::MemRegionVal>().getRegion());
451 case loc::GotoLabelKind:
452 // These are anormal cases. Flag an undefined value.
453 return UndefinedVal();
455 case loc::ConcreteIntKind:
456 // While these seem funny, this can happen through casts.
457 // FIXME: What we should return is the field offset, not base. For example,
458 // add the field offset to the integer value. That way things
459 // like this work properly: &(((struct foo *) 0xa)->f)
460 // However, that's not easy to fix without reducing our abilities
461 // to catch null pointer dereference. Eg., ((struct foo *)0x0)->f = 7
462 // is a null dereference even though we're dereferencing offset of f
463 // rather than null. Coming up with an approach that computes offsets
464 // over null pointers properly while still being able to catch null
465 // dereferences might be worth it.
469 llvm_unreachable("Unhandled Base.");
472 // NOTE: We must have this check first because ObjCIvarDecl is a subclass
474 if (const auto *ID = dyn_cast<ObjCIvarDecl>(D))
475 return loc::MemRegionVal(MRMgr.getObjCIvarRegion(ID, BaseR));
477 return loc::MemRegionVal(MRMgr.getFieldRegion(cast<FieldDecl>(D), BaseR));
480 SVal StoreManager::getLValueIvar(const ObjCIvarDecl *decl, SVal base) {
481 return getLValueFieldOrIvar(decl, base);
484 SVal StoreManager::getLValueElement(QualType elementType, NonLoc Offset,
486 // If the base is an unknown or undefined value, just return it back.
487 // FIXME: For absolute pointer addresses, we just return that value back as
488 // well, although in reality we should return the offset added to that
489 // value. See also the similar FIXME in getLValueFieldOrIvar().
490 if (Base.isUnknownOrUndef() || Base.getAs<loc::ConcreteInt>())
493 if (Base.getAs<loc::GotoLabel>())
496 const SubRegion *BaseRegion =
497 Base.castAs<loc::MemRegionVal>().getRegionAs<SubRegion>();
499 // Pointer of any type can be cast and used as array base.
500 const auto *ElemR = dyn_cast<ElementRegion>(BaseRegion);
502 // Convert the offset to the appropriate size and signedness.
503 Offset = svalBuilder.convertToArrayIndex(Offset).castAs<NonLoc>();
506 // If the base region is not an ElementRegion, create one.
507 // This can happen in the following example:
509 // char *p = __builtin_alloc(10);
512 // Observe that 'p' binds to an AllocaRegion.
513 return loc::MemRegionVal(MRMgr.getElementRegion(elementType, Offset,
517 SVal BaseIdx = ElemR->getIndex();
519 if (!BaseIdx.getAs<nonloc::ConcreteInt>())
522 const llvm::APSInt &BaseIdxI =
523 BaseIdx.castAs<nonloc::ConcreteInt>().getValue();
525 // Only allow non-integer offsets if the base region has no offset itself.
526 // FIXME: This is a somewhat arbitrary restriction. We should be using
527 // SValBuilder here to add the two offsets without checking their types.
528 if (!Offset.getAs<nonloc::ConcreteInt>()) {
529 if (isa<ElementRegion>(BaseRegion->StripCasts()))
532 return loc::MemRegionVal(MRMgr.getElementRegion(
533 elementType, Offset, cast<SubRegion>(ElemR->getSuperRegion()), Ctx));
536 const llvm::APSInt& OffI = Offset.castAs<nonloc::ConcreteInt>().getValue();
537 assert(BaseIdxI.isSigned());
539 // Compute the new index.
540 nonloc::ConcreteInt NewIdx(svalBuilder.getBasicValueFactory().getValue(BaseIdxI +
543 // Construct the new ElementRegion.
544 const SubRegion *ArrayR = cast<SubRegion>(ElemR->getSuperRegion());
545 return loc::MemRegionVal(MRMgr.getElementRegion(elementType, NewIdx, ArrayR,
549 StoreManager::BindingsHandler::~BindingsHandler() = default;
551 bool StoreManager::FindUniqueBinding::HandleBinding(StoreManager& SMgr,
555 SymbolRef SymV = val.getAsLocSymbol();
556 if (!SymV || SymV != Sym)