SValBuilder.cpp revision 7f1fd2f182717d5ce6cde60398128910c90f98be
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
23using namespace clang;
24using namespace ento;
25
26//===----------------------------------------------------------------------===//
27// Basic SVal creation.
28//===----------------------------------------------------------------------===//
29
30void SValBuilder::anchor() { }
31
32DefinedOrUnknownSVal 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
44NonLoc 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
54NonLoc 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
62NonLoc 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
69NonLoc 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
76SVal 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
90nonloc::ConcreteInt SValBuilder::makeBoolVal(const CXXBoolLiteralExpr *boolean){
91  return makeTruthVal(boolean->getValue());
92}
93
94DefinedOrUnknownSVal
95SValBuilder::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
109DefinedOrUnknownSVal 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
124DefinedOrUnknownSVal 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
141DefinedOrUnknownSVal 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
156DefinedOrUnknownSVal
157SValBuilder::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
168DefinedSVal 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
183DefinedOrUnknownSVal
184SValBuilder::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
199DefinedSVal SValBuilder::getFunctionPointer(const FunctionDecl *func) {
200  return loc::MemRegionVal(MemMgr.getFunctionTextRegion(func));
201}
202
203DefinedSVal SValBuilder::getBlockPointer(const BlockDecl *block,
204                                         CanQualType locTy,
205                                         const LocationContext *locContext) {
206  const BlockTextRegion *BC =
207    MemMgr.getBlockTextRegion(block, locTy, locContext->getAnalysisDeclContext());
208  const BlockDataRegion *BD = MemMgr.getBlockDataRegion(BC, locContext);
209  return loc::MemRegionVal(BD);
210}
211
212/// Return a memory region for the 'this' object reference.
213loc::MemRegionVal SValBuilder::getCXXThis(const CXXMethodDecl *D,
214                                          const StackFrameContext *SFC) {
215  return loc::MemRegionVal(getRegionManager().
216                           getCXXThisRegion(D->getThisType(getContext()), SFC));
217}
218
219/// Return a memory region for the 'this' object reference.
220loc::MemRegionVal SValBuilder::getCXXThis(const CXXRecordDecl *D,
221                                          const StackFrameContext *SFC) {
222  const Type *T = D->getTypeForDecl();
223  QualType PT = getContext().getPointerType(QualType(T, 0));
224  return loc::MemRegionVal(getRegionManager().getCXXThisRegion(PT, SFC));
225}
226
227Optional<SVal> SValBuilder::getConstantVal(const Expr *E) {
228  E = E->IgnoreParens();
229
230  switch (E->getStmtClass()) {
231  // Handle expressions that we treat differently from the AST's constant
232  // evaluator.
233  case Stmt::AddrLabelExprClass:
234    return makeLoc(cast<AddrLabelExpr>(E));
235
236  case Stmt::CXXScalarValueInitExprClass:
237  case Stmt::ImplicitValueInitExprClass:
238    return makeZeroVal(E->getType());
239
240  case Stmt::ObjCStringLiteralClass: {
241    const ObjCStringLiteral *SL = cast<ObjCStringLiteral>(E);
242    return makeLoc(getRegionManager().getObjCStringRegion(SL));
243  }
244
245  case Stmt::StringLiteralClass: {
246    const StringLiteral *SL = cast<StringLiteral>(E);
247    return makeLoc(getRegionManager().getStringRegion(SL));
248  }
249
250  // Fast-path some expressions to avoid the overhead of going through the AST's
251  // constant evaluator
252  case Stmt::CharacterLiteralClass: {
253    const CharacterLiteral *C = cast<CharacterLiteral>(E);
254    return makeIntVal(C->getValue(), C->getType());
255  }
256
257  case Stmt::CXXBoolLiteralExprClass:
258    return makeBoolVal(cast<CXXBoolLiteralExpr>(E));
259
260  case Stmt::IntegerLiteralClass:
261    return makeIntVal(cast<IntegerLiteral>(E));
262
263  case Stmt::ObjCBoolLiteralExprClass:
264    return makeBoolVal(cast<ObjCBoolLiteralExpr>(E));
265
266  case Stmt::CXXNullPtrLiteralExprClass:
267    return makeNull();
268
269  // If we don't have a special case, fall back to the AST's constant evaluator.
270  default: {
271    // Don't try to come up with a value for materialized temporaries.
272    if (E->isGLValue())
273      return None;
274
275    ASTContext &Ctx = getContext();
276    llvm::APSInt Result;
277    if (E->EvaluateAsInt(Result, Ctx))
278      return makeIntVal(Result);
279
280    if (Loc::isLocType(E->getType()))
281      if (E->isNullPointerConstant(Ctx, Expr::NPC_ValueDependentIsNotNull))
282        return makeNull();
283
284    return None;
285  }
286  }
287}
288
289//===----------------------------------------------------------------------===//
290
291SVal SValBuilder::makeSymExprValNN(ProgramStateRef State,
292                                   BinaryOperator::Opcode Op,
293                                   NonLoc LHS, NonLoc RHS,
294                                   QualType ResultTy) {
295  if (!State->isTainted(RHS) && !State->isTainted(LHS))
296    return UnknownVal();
297
298  const SymExpr *symLHS = LHS.getAsSymExpr();
299  const SymExpr *symRHS = RHS.getAsSymExpr();
300  // TODO: When the Max Complexity is reached, we should conjure a symbol
301  // instead of generating an Unknown value and propagate the taint info to it.
302  const unsigned MaxComp = 10000; // 100000 28X
303
304  if (symLHS && symRHS &&
305      (symLHS->computeComplexity() + symRHS->computeComplexity()) <  MaxComp)
306    return makeNonLoc(symLHS, Op, symRHS, ResultTy);
307
308  if (symLHS && symLHS->computeComplexity() < MaxComp)
309    if (Optional<nonloc::ConcreteInt> rInt = RHS.getAs<nonloc::ConcreteInt>())
310      return makeNonLoc(symLHS, Op, rInt->getValue(), ResultTy);
311
312  if (symRHS && symRHS->computeComplexity() < MaxComp)
313    if (Optional<nonloc::ConcreteInt> lInt = LHS.getAs<nonloc::ConcreteInt>())
314      return makeNonLoc(lInt->getValue(), Op, symRHS, ResultTy);
315
316  return UnknownVal();
317}
318
319
320SVal SValBuilder::evalBinOp(ProgramStateRef state, BinaryOperator::Opcode op,
321                            SVal lhs, SVal rhs, QualType type) {
322
323  if (lhs.isUndef() || rhs.isUndef())
324    return UndefinedVal();
325
326  if (lhs.isUnknown() || rhs.isUnknown())
327    return UnknownVal();
328
329  if (Optional<Loc> LV = lhs.getAs<Loc>()) {
330    if (Optional<Loc> RV = rhs.getAs<Loc>())
331      return evalBinOpLL(state, op, *LV, *RV, type);
332
333    return evalBinOpLN(state, op, *LV, rhs.castAs<NonLoc>(), type);
334  }
335
336  if (Optional<Loc> RV = rhs.getAs<Loc>()) {
337    // Support pointer arithmetic where the addend is on the left
338    // and the pointer on the right.
339    assert(op == BO_Add);
340
341    // Commute the operands.
342    return evalBinOpLN(state, op, *RV, lhs.castAs<NonLoc>(), type);
343  }
344
345  return evalBinOpNN(state, op, lhs.castAs<NonLoc>(), rhs.castAs<NonLoc>(),
346                     type);
347}
348
349DefinedOrUnknownSVal SValBuilder::evalEQ(ProgramStateRef state,
350                                         DefinedOrUnknownSVal lhs,
351                                         DefinedOrUnknownSVal rhs) {
352  return evalBinOp(state, BO_EQ, lhs, rhs, Context.IntTy)
353      .castAs<DefinedOrUnknownSVal>();
354}
355
356/// Recursively check if the pointer types are equal modulo const, volatile,
357/// and restrict qualifiers. Also, assume that all types are similar to 'void'.
358/// Assumes the input types are canonical.
359static bool shouldBeModeledWithNoOp(ASTContext &Context, QualType ToTy,
360                                                         QualType FromTy) {
361  while (Context.UnwrapSimilarPointerTypes(ToTy, FromTy)) {
362    Qualifiers Quals1, Quals2;
363    ToTy = Context.getUnqualifiedArrayType(ToTy, Quals1);
364    FromTy = Context.getUnqualifiedArrayType(FromTy, Quals2);
365
366    // Make sure that non cvr-qualifiers the other qualifiers (e.g., address
367    // spaces) are identical.
368    Quals1.removeCVRQualifiers();
369    Quals2.removeCVRQualifiers();
370    if (Quals1 != Quals2)
371      return false;
372  }
373
374  // If we are casting to void, the 'From' value can be used to represent the
375  // 'To' value.
376  if (ToTy->isVoidType())
377    return true;
378
379  if (ToTy != FromTy)
380    return false;
381
382  return true;
383}
384
385// FIXME: should rewrite according to the cast kind.
386SVal SValBuilder::evalCast(SVal val, QualType castTy, QualType originalTy) {
387  castTy = Context.getCanonicalType(castTy);
388  originalTy = Context.getCanonicalType(originalTy);
389  if (val.isUnknownOrUndef() || castTy == originalTy)
390    return val;
391
392  if (castTy->isBooleanType()) {
393    if (val.isUnknownOrUndef())
394      return val;
395    if (val.isConstant())
396      return makeTruthVal(!val.isZeroConstant(), castTy);
397    if (SymbolRef Sym = val.getAsSymbol()) {
398      BasicValueFactory &BVF = getBasicValueFactory();
399      // FIXME: If we had a state here, we could see if the symbol is known to
400      // be zero, but we don't.
401      return makeNonLoc(Sym, BO_NE, BVF.getValue(0, Sym->getType()), castTy);
402    }
403
404    assert(val.getAs<Loc>());
405    return makeTruthVal(true, castTy);
406  }
407
408  // For const casts, casts to void, just propagate the value.
409  if (!castTy->isVariableArrayType() && !originalTy->isVariableArrayType())
410    if (shouldBeModeledWithNoOp(Context, Context.getPointerType(castTy),
411                                         Context.getPointerType(originalTy)))
412      return val;
413
414  // Check for casts from pointers to integers.
415  if (castTy->isIntegralOrEnumerationType() && Loc::isLocType(originalTy))
416    return evalCastFromLoc(val.castAs<Loc>(), castTy);
417
418  // Check for casts from integers to pointers.
419  if (Loc::isLocType(castTy) && originalTy->isIntegralOrEnumerationType()) {
420    if (Optional<nonloc::LocAsInteger> LV = val.getAs<nonloc::LocAsInteger>()) {
421      if (const MemRegion *R = LV->getLoc().getAsRegion()) {
422        StoreManager &storeMgr = StateMgr.getStoreManager();
423        R = storeMgr.castRegion(R, castTy);
424        return R ? SVal(loc::MemRegionVal(R)) : UnknownVal();
425      }
426      return LV->getLoc();
427    }
428    return dispatchCast(val, castTy);
429  }
430
431  // Just pass through function and block pointers.
432  if (originalTy->isBlockPointerType() || originalTy->isFunctionPointerType()) {
433    assert(Loc::isLocType(castTy));
434    return val;
435  }
436
437  // Check for casts from array type to another type.
438  if (const ArrayType *arrayT =
439                      dyn_cast<ArrayType>(originalTy.getCanonicalType())) {
440    // We will always decay to a pointer.
441    QualType elemTy = arrayT->getElementType();
442    val = StateMgr.ArrayToPointer(val.castAs<Loc>(), elemTy);
443
444    // Are we casting from an array to a pointer?  If so just pass on
445    // the decayed value.
446    if (castTy->isPointerType() || castTy->isReferenceType())
447      return val;
448
449    // Are we casting from an array to an integer?  If so, cast the decayed
450    // pointer value to an integer.
451    assert(castTy->isIntegralOrEnumerationType());
452
453    // FIXME: Keep these here for now in case we decide soon that we
454    // need the original decayed type.
455    //    QualType elemTy = cast<ArrayType>(originalTy)->getElementType();
456    //    QualType pointerTy = C.getPointerType(elemTy);
457    return evalCastFromLoc(val.castAs<Loc>(), castTy);
458  }
459
460  // Check for casts from a region to a specific type.
461  if (const MemRegion *R = val.getAsRegion()) {
462    // Handle other casts of locations to integers.
463    if (castTy->isIntegralOrEnumerationType())
464      return evalCastFromLoc(loc::MemRegionVal(R), castTy);
465
466    // FIXME: We should handle the case where we strip off view layers to get
467    //  to a desugared type.
468    if (!Loc::isLocType(castTy)) {
469      // FIXME: There can be gross cases where one casts the result of a function
470      // (that returns a pointer) to some other value that happens to fit
471      // within that pointer value.  We currently have no good way to
472      // model such operations.  When this happens, the underlying operation
473      // is that the caller is reasoning about bits.  Conceptually we are
474      // layering a "view" of a location on top of those bits.  Perhaps
475      // we need to be more lazy about mutual possible views, even on an
476      // SVal?  This may be necessary for bit-level reasoning as well.
477      return UnknownVal();
478    }
479
480    // We get a symbolic function pointer for a dereference of a function
481    // pointer, but it is of function type. Example:
482
483    //  struct FPRec {
484    //    void (*my_func)(int * x);
485    //  };
486    //
487    //  int bar(int x);
488    //
489    //  int f1_a(struct FPRec* foo) {
490    //    int x;
491    //    (*foo->my_func)(&x);
492    //    return bar(x)+1; // no-warning
493    //  }
494
495    assert(Loc::isLocType(originalTy) || originalTy->isFunctionType() ||
496           originalTy->isBlockPointerType() || castTy->isReferenceType());
497
498    StoreManager &storeMgr = StateMgr.getStoreManager();
499
500    // Delegate to store manager to get the result of casting a region to a
501    // different type.  If the MemRegion* returned is NULL, this expression
502    // Evaluates to UnknownVal.
503    R = storeMgr.castRegion(R, castTy);
504    return R ? SVal(loc::MemRegionVal(R)) : UnknownVal();
505  }
506
507  return dispatchCast(val, castTy);
508}
509