1 // SValBuilder.cpp - Basic class for all SValBuilder implementations -*- 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 SValBuilder, the base class for all (complete) SValBuilder
11 //  implementations.
12 //
13 //===----------------------------------------------------------------------===//
14 
15 #include "clang/StaticAnalyzer/Core/PathSensitive/SValBuilder.h"
16 #include "clang/AST/DeclCXX.h"
17 #include "clang/AST/ExprCXX.h"
18 #include "clang/StaticAnalyzer/Core/PathSensitive/BasicValueFactory.h"
19 #include "clang/StaticAnalyzer/Core/PathSensitive/MemRegion.h"
20 #include "clang/StaticAnalyzer/Core/PathSensitive/ProgramState.h"
21 #include "clang/StaticAnalyzer/Core/PathSensitive/SVals.h"
22 
23 using namespace clang;
24 using namespace ento;
25 
26 //===----------------------------------------------------------------------===//
27 // Basic SVal creation.
28 //===----------------------------------------------------------------------===//
29 
anchor()30 void SValBuilder::anchor() { }
31 
makeZeroVal(QualType type)32 DefinedOrUnknownSVal SValBuilder::makeZeroVal(QualType type) {
33   if (Loc::isLocType(type))
34     return makeNull();
35 
36   if (type->isIntegralOrEnumerationType())
37     return makeIntVal(0, type);
38 
39   // FIXME: Handle floats.
40   // FIXME: Handle structs.
41   return UnknownVal();
42 }
43 
makeNonLoc(const SymExpr * lhs,BinaryOperator::Opcode op,const llvm::APSInt & rhs,QualType type)44 NonLoc SValBuilder::makeNonLoc(const SymExpr *lhs, BinaryOperator::Opcode op,
45                                 const llvm::APSInt& rhs, QualType type) {
46   // The Environment ensures we always get a persistent APSInt in
47   // BasicValueFactory, so we don't need to get the APSInt from
48   // BasicValueFactory again.
49   assert(lhs);
50   assert(!Loc::isLocType(type));
51   return nonloc::SymbolVal(SymMgr.getSymIntExpr(lhs, op, rhs, type));
52 }
53 
makeNonLoc(const llvm::APSInt & lhs,BinaryOperator::Opcode op,const SymExpr * rhs,QualType type)54 NonLoc SValBuilder::makeNonLoc(const llvm::APSInt& lhs,
55                                BinaryOperator::Opcode op, const SymExpr *rhs,
56                                QualType type) {
57   assert(rhs);
58   assert(!Loc::isLocType(type));
59   return nonloc::SymbolVal(SymMgr.getIntSymExpr(lhs, op, rhs, type));
60 }
61 
makeNonLoc(const SymExpr * lhs,BinaryOperator::Opcode op,const SymExpr * rhs,QualType type)62 NonLoc SValBuilder::makeNonLoc(const SymExpr *lhs, BinaryOperator::Opcode op,
63                                const SymExpr *rhs, QualType type) {
64   assert(lhs && rhs);
65   assert(!Loc::isLocType(type));
66   return nonloc::SymbolVal(SymMgr.getSymSymExpr(lhs, op, rhs, type));
67 }
68 
makeNonLoc(const SymExpr * operand,QualType fromTy,QualType toTy)69 NonLoc SValBuilder::makeNonLoc(const SymExpr *operand,
70                                QualType fromTy, QualType toTy) {
71   assert(operand);
72   assert(!Loc::isLocType(toTy));
73   return nonloc::SymbolVal(SymMgr.getCastSymbol(operand, fromTy, toTy));
74 }
75 
convertToArrayIndex(SVal val)76 SVal SValBuilder::convertToArrayIndex(SVal val) {
77   if (val.isUnknownOrUndef())
78     return val;
79 
80   // Common case: we have an appropriately sized integer.
81   if (Optional<nonloc::ConcreteInt> CI = val.getAs<nonloc::ConcreteInt>()) {
82     const llvm::APSInt& I = CI->getValue();
83     if (I.getBitWidth() == ArrayIndexWidth && I.isSigned())
84       return val;
85   }
86 
87   return evalCastFromNonLoc(val.castAs<NonLoc>(), ArrayIndexTy);
88 }
89 
makeBoolVal(const CXXBoolLiteralExpr * boolean)90 nonloc::ConcreteInt SValBuilder::makeBoolVal(const CXXBoolLiteralExpr *boolean){
91   return makeTruthVal(boolean->getValue());
92 }
93 
94 DefinedOrUnknownSVal
getRegionValueSymbolVal(const TypedValueRegion * region)95 SValBuilder::getRegionValueSymbolVal(const TypedValueRegion* region) {
96   QualType T = region->getValueType();
97 
98   if (!SymbolManager::canSymbolicate(T))
99     return UnknownVal();
100 
101   SymbolRef sym = SymMgr.getRegionValueSymbol(region);
102 
103   if (Loc::isLocType(T))
104     return loc::MemRegionVal(MemMgr.getSymbolicRegion(sym));
105 
106   return nonloc::SymbolVal(sym);
107 }
108 
conjureSymbolVal(const void * SymbolTag,const Expr * Ex,const LocationContext * LCtx,unsigned Count)109 DefinedOrUnknownSVal SValBuilder::conjureSymbolVal(const void *SymbolTag,
110                                                    const Expr *Ex,
111                                                    const LocationContext *LCtx,
112                                                    unsigned Count) {
113   QualType T = Ex->getType();
114 
115   // Compute the type of the result. If the expression is not an R-value, the
116   // result should be a location.
117   QualType ExType = Ex->getType();
118   if (Ex->isGLValue())
119     T = LCtx->getAnalysisDeclContext()->getASTContext().getPointerType(ExType);
120 
121   return conjureSymbolVal(SymbolTag, Ex, LCtx, T, Count);
122 }
123 
conjureSymbolVal(const void * symbolTag,const Expr * expr,const LocationContext * LCtx,QualType type,unsigned count)124 DefinedOrUnknownSVal SValBuilder::conjureSymbolVal(const void *symbolTag,
125                                                    const Expr *expr,
126                                                    const LocationContext *LCtx,
127                                                    QualType type,
128                                                    unsigned count) {
129   if (!SymbolManager::canSymbolicate(type))
130     return UnknownVal();
131 
132   SymbolRef sym = SymMgr.conjureSymbol(expr, LCtx, type, count, symbolTag);
133 
134   if (Loc::isLocType(type))
135     return loc::MemRegionVal(MemMgr.getSymbolicRegion(sym));
136 
137   return nonloc::SymbolVal(sym);
138 }
139 
140 
conjureSymbolVal(const Stmt * stmt,const LocationContext * LCtx,QualType type,unsigned visitCount)141 DefinedOrUnknownSVal SValBuilder::conjureSymbolVal(const Stmt *stmt,
142                                                    const LocationContext *LCtx,
143                                                    QualType type,
144                                                    unsigned visitCount) {
145   if (!SymbolManager::canSymbolicate(type))
146     return UnknownVal();
147 
148   SymbolRef sym = SymMgr.conjureSymbol(stmt, LCtx, type, visitCount);
149 
150   if (Loc::isLocType(type))
151     return loc::MemRegionVal(MemMgr.getSymbolicRegion(sym));
152 
153   return nonloc::SymbolVal(sym);
154 }
155 
156 DefinedOrUnknownSVal
getConjuredHeapSymbolVal(const Expr * E,const LocationContext * LCtx,unsigned VisitCount)157 SValBuilder::getConjuredHeapSymbolVal(const Expr *E,
158                                       const LocationContext *LCtx,
159                                       unsigned VisitCount) {
160   QualType T = E->getType();
161   assert(Loc::isLocType(T));
162   assert(SymbolManager::canSymbolicate(T));
163 
164   SymbolRef sym = SymMgr.conjureSymbol(E, LCtx, T, VisitCount);
165   return loc::MemRegionVal(MemMgr.getSymbolicHeapRegion(sym));
166 }
167 
getMetadataSymbolVal(const void * symbolTag,const MemRegion * region,const Expr * expr,QualType type,unsigned count)168 DefinedSVal SValBuilder::getMetadataSymbolVal(const void *symbolTag,
169                                               const MemRegion *region,
170                                               const Expr *expr, QualType type,
171                                               unsigned count) {
172   assert(SymbolManager::canSymbolicate(type) && "Invalid metadata symbol type");
173 
174   SymbolRef sym =
175       SymMgr.getMetadataSymbol(region, expr, type, count, symbolTag);
176 
177   if (Loc::isLocType(type))
178     return loc::MemRegionVal(MemMgr.getSymbolicRegion(sym));
179 
180   return nonloc::SymbolVal(sym);
181 }
182 
183 DefinedOrUnknownSVal
getDerivedRegionValueSymbolVal(SymbolRef parentSymbol,const TypedValueRegion * region)184 SValBuilder::getDerivedRegionValueSymbolVal(SymbolRef parentSymbol,
185                                              const TypedValueRegion *region) {
186   QualType T = region->getValueType();
187 
188   if (!SymbolManager::canSymbolicate(T))
189     return UnknownVal();
190 
191   SymbolRef sym = SymMgr.getDerivedSymbol(parentSymbol, region);
192 
193   if (Loc::isLocType(T))
194     return loc::MemRegionVal(MemMgr.getSymbolicRegion(sym));
195 
196   return nonloc::SymbolVal(sym);
197 }
198 
getFunctionPointer(const FunctionDecl * func)199 DefinedSVal SValBuilder::getFunctionPointer(const FunctionDecl *func) {
200   return loc::MemRegionVal(MemMgr.getFunctionTextRegion(func));
201 }
202 
getBlockPointer(const BlockDecl * block,CanQualType locTy,const LocationContext * locContext,unsigned blockCount)203 DefinedSVal SValBuilder::getBlockPointer(const BlockDecl *block,
204                                          CanQualType locTy,
205                                          const LocationContext *locContext,
206                                          unsigned blockCount) {
207   const BlockTextRegion *BC =
208     MemMgr.getBlockTextRegion(block, locTy, locContext->getAnalysisDeclContext());
209   const BlockDataRegion *BD = MemMgr.getBlockDataRegion(BC, locContext,
210                                                         blockCount);
211   return loc::MemRegionVal(BD);
212 }
213 
214 /// Return a memory region for the 'this' object reference.
getCXXThis(const CXXMethodDecl * D,const StackFrameContext * SFC)215 loc::MemRegionVal SValBuilder::getCXXThis(const CXXMethodDecl *D,
216                                           const StackFrameContext *SFC) {
217   return loc::MemRegionVal(getRegionManager().
218                            getCXXThisRegion(D->getThisType(getContext()), SFC));
219 }
220 
221 /// Return a memory region for the 'this' object reference.
getCXXThis(const CXXRecordDecl * D,const StackFrameContext * SFC)222 loc::MemRegionVal SValBuilder::getCXXThis(const CXXRecordDecl *D,
223                                           const StackFrameContext *SFC) {
224   const Type *T = D->getTypeForDecl();
225   QualType PT = getContext().getPointerType(QualType(T, 0));
226   return loc::MemRegionVal(getRegionManager().getCXXThisRegion(PT, SFC));
227 }
228 
getConstantVal(const Expr * E)229 Optional<SVal> SValBuilder::getConstantVal(const Expr *E) {
230   E = E->IgnoreParens();
231 
232   switch (E->getStmtClass()) {
233   // Handle expressions that we treat differently from the AST's constant
234   // evaluator.
235   case Stmt::AddrLabelExprClass:
236     return makeLoc(cast<AddrLabelExpr>(E));
237 
238   case Stmt::CXXScalarValueInitExprClass:
239   case Stmt::ImplicitValueInitExprClass:
240     return makeZeroVal(E->getType());
241 
242   case Stmt::ObjCStringLiteralClass: {
243     const ObjCStringLiteral *SL = cast<ObjCStringLiteral>(E);
244     return makeLoc(getRegionManager().getObjCStringRegion(SL));
245   }
246 
247   case Stmt::StringLiteralClass: {
248     const StringLiteral *SL = cast<StringLiteral>(E);
249     return makeLoc(getRegionManager().getStringRegion(SL));
250   }
251 
252   // Fast-path some expressions to avoid the overhead of going through the AST's
253   // constant evaluator
254   case Stmt::CharacterLiteralClass: {
255     const CharacterLiteral *C = cast<CharacterLiteral>(E);
256     return makeIntVal(C->getValue(), C->getType());
257   }
258 
259   case Stmt::CXXBoolLiteralExprClass:
260     return makeBoolVal(cast<CXXBoolLiteralExpr>(E));
261 
262   case Stmt::IntegerLiteralClass:
263     return makeIntVal(cast<IntegerLiteral>(E));
264 
265   case Stmt::ObjCBoolLiteralExprClass:
266     return makeBoolVal(cast<ObjCBoolLiteralExpr>(E));
267 
268   case Stmt::CXXNullPtrLiteralExprClass:
269     return makeNull();
270 
271   case Stmt::ImplicitCastExprClass: {
272     const CastExpr *CE = cast<CastExpr>(E);
273     if (CE->getCastKind() == CK_ArrayToPointerDecay) {
274       Optional<SVal> ArrayVal = getConstantVal(CE->getSubExpr());
275       if (!ArrayVal)
276         return None;
277       return evalCast(*ArrayVal, CE->getType(), CE->getSubExpr()->getType());
278     }
279     // FALLTHROUGH
280   }
281 
282   // If we don't have a special case, fall back to the AST's constant evaluator.
283   default: {
284     // Don't try to come up with a value for materialized temporaries.
285     if (E->isGLValue())
286       return None;
287 
288     ASTContext &Ctx = getContext();
289     llvm::APSInt Result;
290     if (E->EvaluateAsInt(Result, Ctx))
291       return makeIntVal(Result);
292 
293     if (Loc::isLocType(E->getType()))
294       if (E->isNullPointerConstant(Ctx, Expr::NPC_ValueDependentIsNotNull))
295         return makeNull();
296 
297     return None;
298   }
299   }
300 }
301 
302 //===----------------------------------------------------------------------===//
303 
makeSymExprValNN(ProgramStateRef State,BinaryOperator::Opcode Op,NonLoc LHS,NonLoc RHS,QualType ResultTy)304 SVal SValBuilder::makeSymExprValNN(ProgramStateRef State,
305                                    BinaryOperator::Opcode Op,
306                                    NonLoc LHS, NonLoc RHS,
307                                    QualType ResultTy) {
308   if (!State->isTainted(RHS) && !State->isTainted(LHS))
309     return UnknownVal();
310 
311   const SymExpr *symLHS = LHS.getAsSymExpr();
312   const SymExpr *symRHS = RHS.getAsSymExpr();
313   // TODO: When the Max Complexity is reached, we should conjure a symbol
314   // instead of generating an Unknown value and propagate the taint info to it.
315   const unsigned MaxComp = 10000; // 100000 28X
316 
317   if (symLHS && symRHS &&
318       (symLHS->computeComplexity() + symRHS->computeComplexity()) <  MaxComp)
319     return makeNonLoc(symLHS, Op, symRHS, ResultTy);
320 
321   if (symLHS && symLHS->computeComplexity() < MaxComp)
322     if (Optional<nonloc::ConcreteInt> rInt = RHS.getAs<nonloc::ConcreteInt>())
323       return makeNonLoc(symLHS, Op, rInt->getValue(), ResultTy);
324 
325   if (symRHS && symRHS->computeComplexity() < MaxComp)
326     if (Optional<nonloc::ConcreteInt> lInt = LHS.getAs<nonloc::ConcreteInt>())
327       return makeNonLoc(lInt->getValue(), Op, symRHS, ResultTy);
328 
329   return UnknownVal();
330 }
331 
332 
evalBinOp(ProgramStateRef state,BinaryOperator::Opcode op,SVal lhs,SVal rhs,QualType type)333 SVal SValBuilder::evalBinOp(ProgramStateRef state, BinaryOperator::Opcode op,
334                             SVal lhs, SVal rhs, QualType type) {
335 
336   if (lhs.isUndef() || rhs.isUndef())
337     return UndefinedVal();
338 
339   if (lhs.isUnknown() || rhs.isUnknown())
340     return UnknownVal();
341 
342   if (Optional<Loc> LV = lhs.getAs<Loc>()) {
343     if (Optional<Loc> RV = rhs.getAs<Loc>())
344       return evalBinOpLL(state, op, *LV, *RV, type);
345 
346     return evalBinOpLN(state, op, *LV, rhs.castAs<NonLoc>(), type);
347   }
348 
349   if (Optional<Loc> RV = rhs.getAs<Loc>()) {
350     // Support pointer arithmetic where the addend is on the left
351     // and the pointer on the right.
352     assert(op == BO_Add);
353 
354     // Commute the operands.
355     return evalBinOpLN(state, op, *RV, lhs.castAs<NonLoc>(), type);
356   }
357 
358   return evalBinOpNN(state, op, lhs.castAs<NonLoc>(), rhs.castAs<NonLoc>(),
359                      type);
360 }
361 
evalEQ(ProgramStateRef state,DefinedOrUnknownSVal lhs,DefinedOrUnknownSVal rhs)362 DefinedOrUnknownSVal SValBuilder::evalEQ(ProgramStateRef state,
363                                          DefinedOrUnknownSVal lhs,
364                                          DefinedOrUnknownSVal rhs) {
365   return evalBinOp(state, BO_EQ, lhs, rhs, getConditionType())
366       .castAs<DefinedOrUnknownSVal>();
367 }
368 
369 /// Recursively check if the pointer types are equal modulo const, volatile,
370 /// and restrict qualifiers. Also, assume that all types are similar to 'void'.
371 /// Assumes the input types are canonical.
shouldBeModeledWithNoOp(ASTContext & Context,QualType ToTy,QualType FromTy)372 static bool shouldBeModeledWithNoOp(ASTContext &Context, QualType ToTy,
373                                                          QualType FromTy) {
374   while (Context.UnwrapSimilarPointerTypes(ToTy, FromTy)) {
375     Qualifiers Quals1, Quals2;
376     ToTy = Context.getUnqualifiedArrayType(ToTy, Quals1);
377     FromTy = Context.getUnqualifiedArrayType(FromTy, Quals2);
378 
379     // Make sure that non-cvr-qualifiers the other qualifiers (e.g., address
380     // spaces) are identical.
381     Quals1.removeCVRQualifiers();
382     Quals2.removeCVRQualifiers();
383     if (Quals1 != Quals2)
384       return false;
385   }
386 
387   // If we are casting to void, the 'From' value can be used to represent the
388   // 'To' value.
389   if (ToTy->isVoidType())
390     return true;
391 
392   if (ToTy != FromTy)
393     return false;
394 
395   return true;
396 }
397 
398 // FIXME: should rewrite according to the cast kind.
evalCast(SVal val,QualType castTy,QualType originalTy)399 SVal SValBuilder::evalCast(SVal val, QualType castTy, QualType originalTy) {
400   castTy = Context.getCanonicalType(castTy);
401   originalTy = Context.getCanonicalType(originalTy);
402   if (val.isUnknownOrUndef() || castTy == originalTy)
403     return val;
404 
405   if (castTy->isBooleanType()) {
406     if (val.isUnknownOrUndef())
407       return val;
408     if (val.isConstant())
409       return makeTruthVal(!val.isZeroConstant(), castTy);
410     if (!Loc::isLocType(originalTy) &&
411         !originalTy->isIntegralOrEnumerationType() &&
412         !originalTy->isMemberPointerType())
413       return UnknownVal();
414     if (SymbolRef Sym = val.getAsSymbol(true)) {
415       BasicValueFactory &BVF = getBasicValueFactory();
416       // FIXME: If we had a state here, we could see if the symbol is known to
417       // be zero, but we don't.
418       return makeNonLoc(Sym, BO_NE, BVF.getValue(0, Sym->getType()), castTy);
419     }
420     // Loc values are not always true, they could be weakly linked functions.
421     if (Optional<Loc> L = val.getAs<Loc>())
422       return evalCastFromLoc(*L, castTy);
423 
424     Loc L = val.castAs<nonloc::LocAsInteger>().getLoc();
425     return evalCastFromLoc(L, castTy);
426   }
427 
428   // For const casts, casts to void, just propagate the value.
429   if (!castTy->isVariableArrayType() && !originalTy->isVariableArrayType())
430     if (shouldBeModeledWithNoOp(Context, Context.getPointerType(castTy),
431                                          Context.getPointerType(originalTy)))
432       return val;
433 
434   // Check for casts from pointers to integers.
435   if (castTy->isIntegralOrEnumerationType() && Loc::isLocType(originalTy))
436     return evalCastFromLoc(val.castAs<Loc>(), castTy);
437 
438   // Check for casts from integers to pointers.
439   if (Loc::isLocType(castTy) && originalTy->isIntegralOrEnumerationType()) {
440     if (Optional<nonloc::LocAsInteger> LV = val.getAs<nonloc::LocAsInteger>()) {
441       if (const MemRegion *R = LV->getLoc().getAsRegion()) {
442         StoreManager &storeMgr = StateMgr.getStoreManager();
443         R = storeMgr.castRegion(R, castTy);
444         return R ? SVal(loc::MemRegionVal(R)) : UnknownVal();
445       }
446       return LV->getLoc();
447     }
448     return dispatchCast(val, castTy);
449   }
450 
451   // Just pass through function and block pointers.
452   if (originalTy->isBlockPointerType() || originalTy->isFunctionPointerType()) {
453     assert(Loc::isLocType(castTy));
454     return val;
455   }
456 
457   // Check for casts from array type to another type.
458   if (const ArrayType *arrayT =
459                       dyn_cast<ArrayType>(originalTy.getCanonicalType())) {
460     // We will always decay to a pointer.
461     QualType elemTy = arrayT->getElementType();
462     val = StateMgr.ArrayToPointer(val.castAs<Loc>(), elemTy);
463 
464     // Are we casting from an array to a pointer?  If so just pass on
465     // the decayed value.
466     if (castTy->isPointerType() || castTy->isReferenceType())
467       return val;
468 
469     // Are we casting from an array to an integer?  If so, cast the decayed
470     // pointer value to an integer.
471     assert(castTy->isIntegralOrEnumerationType());
472 
473     // FIXME: Keep these here for now in case we decide soon that we
474     // need the original decayed type.
475     //    QualType elemTy = cast<ArrayType>(originalTy)->getElementType();
476     //    QualType pointerTy = C.getPointerType(elemTy);
477     return evalCastFromLoc(val.castAs<Loc>(), castTy);
478   }
479 
480   // Check for casts from a region to a specific type.
481   if (const MemRegion *R = val.getAsRegion()) {
482     // Handle other casts of locations to integers.
483     if (castTy->isIntegralOrEnumerationType())
484       return evalCastFromLoc(loc::MemRegionVal(R), castTy);
485 
486     // FIXME: We should handle the case where we strip off view layers to get
487     //  to a desugared type.
488     if (!Loc::isLocType(castTy)) {
489       // FIXME: There can be gross cases where one casts the result of a function
490       // (that returns a pointer) to some other value that happens to fit
491       // within that pointer value.  We currently have no good way to
492       // model such operations.  When this happens, the underlying operation
493       // is that the caller is reasoning about bits.  Conceptually we are
494       // layering a "view" of a location on top of those bits.  Perhaps
495       // we need to be more lazy about mutual possible views, even on an
496       // SVal?  This may be necessary for bit-level reasoning as well.
497       return UnknownVal();
498     }
499 
500     // We get a symbolic function pointer for a dereference of a function
501     // pointer, but it is of function type. Example:
502 
503     //  struct FPRec {
504     //    void (*my_func)(int * x);
505     //  };
506     //
507     //  int bar(int x);
508     //
509     //  int f1_a(struct FPRec* foo) {
510     //    int x;
511     //    (*foo->my_func)(&x);
512     //    return bar(x)+1; // no-warning
513     //  }
514 
515     assert(Loc::isLocType(originalTy) || originalTy->isFunctionType() ||
516            originalTy->isBlockPointerType() || castTy->isReferenceType());
517 
518     StoreManager &storeMgr = StateMgr.getStoreManager();
519 
520     // Delegate to store manager to get the result of casting a region to a
521     // different type.  If the MemRegion* returned is NULL, this expression
522     // Evaluates to UnknownVal.
523     R = storeMgr.castRegion(R, castTy);
524     return R ? SVal(loc::MemRegionVal(R)) : UnknownVal();
525   }
526 
527   return dispatchCast(val, castTy);
528 }
529