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 30 void SValBuilder::anchor() { } 31 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 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 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 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 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 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 90 nonloc::ConcreteInt SValBuilder::makeBoolVal(const CXXBoolLiteralExpr *boolean){ 91 return makeTruthVal(boolean->getValue()); 92 } 93 94 DefinedOrUnknownSVal 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 } 108 109 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 } 123 124 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 } 139 140 141 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 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 } 167 168 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 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 } 198 199 DefinedSVal SValBuilder::getFunctionPointer(const FunctionDecl *func) { 200 return loc::MemRegionVal(MemMgr.getFunctionTextRegion(func)); 201 } 202 203 DefinedSVal 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. 213 loc::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. 220 loc::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 } 226 227 Optional<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 case Stmt::ImplicitCastExprClass: { 270 const CastExpr *CE = cast<CastExpr>(E); 271 if (CE->getCastKind() == CK_ArrayToPointerDecay) { 272 Optional<SVal> ArrayVal = getConstantVal(CE->getSubExpr()); 273 if (!ArrayVal) 274 return None; 275 return evalCast(*ArrayVal, CE->getType(), CE->getSubExpr()->getType()); 276 } 277 // FALLTHROUGH 278 } 279 280 // If we don't have a special case, fall back to the AST's constant evaluator. 281 default: { 282 // Don't try to come up with a value for materialized temporaries. 283 if (E->isGLValue()) 284 return None; 285 286 ASTContext &Ctx = getContext(); 287 llvm::APSInt Result; 288 if (E->EvaluateAsInt(Result, Ctx)) 289 return makeIntVal(Result); 290 291 if (Loc::isLocType(E->getType())) 292 if (E->isNullPointerConstant(Ctx, Expr::NPC_ValueDependentIsNotNull)) 293 return makeNull(); 294 295 return None; 296 } 297 } 298 } 299 300 //===----------------------------------------------------------------------===// 301 302 SVal SValBuilder::makeSymExprValNN(ProgramStateRef State, 303 BinaryOperator::Opcode Op, 304 NonLoc LHS, NonLoc RHS, 305 QualType ResultTy) { 306 if (!State->isTainted(RHS) && !State->isTainted(LHS)) 307 return UnknownVal(); 308 309 const SymExpr *symLHS = LHS.getAsSymExpr(); 310 const SymExpr *symRHS = RHS.getAsSymExpr(); 311 // TODO: When the Max Complexity is reached, we should conjure a symbol 312 // instead of generating an Unknown value and propagate the taint info to it. 313 const unsigned MaxComp = 10000; // 100000 28X 314 315 if (symLHS && symRHS && 316 (symLHS->computeComplexity() + symRHS->computeComplexity()) < MaxComp) 317 return makeNonLoc(symLHS, Op, symRHS, ResultTy); 318 319 if (symLHS && symLHS->computeComplexity() < MaxComp) 320 if (Optional<nonloc::ConcreteInt> rInt = RHS.getAs<nonloc::ConcreteInt>()) 321 return makeNonLoc(symLHS, Op, rInt->getValue(), ResultTy); 322 323 if (symRHS && symRHS->computeComplexity() < MaxComp) 324 if (Optional<nonloc::ConcreteInt> lInt = LHS.getAs<nonloc::ConcreteInt>()) 325 return makeNonLoc(lInt->getValue(), Op, symRHS, ResultTy); 326 327 return UnknownVal(); 328 } 329 330 331 SVal SValBuilder::evalBinOp(ProgramStateRef state, BinaryOperator::Opcode op, 332 SVal lhs, SVal rhs, QualType type) { 333 334 if (lhs.isUndef() || rhs.isUndef()) 335 return UndefinedVal(); 336 337 if (lhs.isUnknown() || rhs.isUnknown()) 338 return UnknownVal(); 339 340 if (Optional<Loc> LV = lhs.getAs<Loc>()) { 341 if (Optional<Loc> RV = rhs.getAs<Loc>()) 342 return evalBinOpLL(state, op, *LV, *RV, type); 343 344 return evalBinOpLN(state, op, *LV, rhs.castAs<NonLoc>(), type); 345 } 346 347 if (Optional<Loc> RV = rhs.getAs<Loc>()) { 348 // Support pointer arithmetic where the addend is on the left 349 // and the pointer on the right. 350 assert(op == BO_Add); 351 352 // Commute the operands. 353 return evalBinOpLN(state, op, *RV, lhs.castAs<NonLoc>(), type); 354 } 355 356 return evalBinOpNN(state, op, lhs.castAs<NonLoc>(), rhs.castAs<NonLoc>(), 357 type); 358 } 359 360 DefinedOrUnknownSVal SValBuilder::evalEQ(ProgramStateRef state, 361 DefinedOrUnknownSVal lhs, 362 DefinedOrUnknownSVal rhs) { 363 return evalBinOp(state, BO_EQ, lhs, rhs, Context.IntTy) 364 .castAs<DefinedOrUnknownSVal>(); 365 } 366 367 /// Recursively check if the pointer types are equal modulo const, volatile, 368 /// and restrict qualifiers. Also, assume that all types are similar to 'void'. 369 /// Assumes the input types are canonical. 370 static bool shouldBeModeledWithNoOp(ASTContext &Context, QualType ToTy, 371 QualType FromTy) { 372 while (Context.UnwrapSimilarPointerTypes(ToTy, FromTy)) { 373 Qualifiers Quals1, Quals2; 374 ToTy = Context.getUnqualifiedArrayType(ToTy, Quals1); 375 FromTy = Context.getUnqualifiedArrayType(FromTy, Quals2); 376 377 // Make sure that non cvr-qualifiers the other qualifiers (e.g., address 378 // spaces) are identical. 379 Quals1.removeCVRQualifiers(); 380 Quals2.removeCVRQualifiers(); 381 if (Quals1 != Quals2) 382 return false; 383 } 384 385 // If we are casting to void, the 'From' value can be used to represent the 386 // 'To' value. 387 if (ToTy->isVoidType()) 388 return true; 389 390 if (ToTy != FromTy) 391 return false; 392 393 return true; 394 } 395 396 // FIXME: should rewrite according to the cast kind. 397 SVal SValBuilder::evalCast(SVal val, QualType castTy, QualType originalTy) { 398 castTy = Context.getCanonicalType(castTy); 399 originalTy = Context.getCanonicalType(originalTy); 400 if (val.isUnknownOrUndef() || castTy == originalTy) 401 return val; 402 403 if (castTy->isBooleanType()) { 404 if (val.isUnknownOrUndef()) 405 return val; 406 if (val.isConstant()) 407 return makeTruthVal(!val.isZeroConstant(), castTy); 408 if (SymbolRef Sym = val.getAsSymbol()) { 409 BasicValueFactory &BVF = getBasicValueFactory(); 410 // FIXME: If we had a state here, we could see if the symbol is known to 411 // be zero, but we don't. 412 return makeNonLoc(Sym, BO_NE, BVF.getValue(0, Sym->getType()), castTy); 413 } 414 415 assert(val.getAs<Loc>() || val.getAs<nonloc::LocAsInteger>()); 416 return makeTruthVal(true, castTy); 417 } 418 419 // For const casts, casts to void, just propagate the value. 420 if (!castTy->isVariableArrayType() && !originalTy->isVariableArrayType()) 421 if (shouldBeModeledWithNoOp(Context, Context.getPointerType(castTy), 422 Context.getPointerType(originalTy))) 423 return val; 424 425 // Check for casts from pointers to integers. 426 if (castTy->isIntegralOrEnumerationType() && Loc::isLocType(originalTy)) 427 return evalCastFromLoc(val.castAs<Loc>(), castTy); 428 429 // Check for casts from integers to pointers. 430 if (Loc::isLocType(castTy) && originalTy->isIntegralOrEnumerationType()) { 431 if (Optional<nonloc::LocAsInteger> LV = val.getAs<nonloc::LocAsInteger>()) { 432 if (const MemRegion *R = LV->getLoc().getAsRegion()) { 433 StoreManager &storeMgr = StateMgr.getStoreManager(); 434 R = storeMgr.castRegion(R, castTy); 435 return R ? SVal(loc::MemRegionVal(R)) : UnknownVal(); 436 } 437 return LV->getLoc(); 438 } 439 return dispatchCast(val, castTy); 440 } 441 442 // Just pass through function and block pointers. 443 if (originalTy->isBlockPointerType() || originalTy->isFunctionPointerType()) { 444 assert(Loc::isLocType(castTy)); 445 return val; 446 } 447 448 // Check for casts from array type to another type. 449 if (const ArrayType *arrayT = 450 dyn_cast<ArrayType>(originalTy.getCanonicalType())) { 451 // We will always decay to a pointer. 452 QualType elemTy = arrayT->getElementType(); 453 val = StateMgr.ArrayToPointer(val.castAs<Loc>(), elemTy); 454 455 // Are we casting from an array to a pointer? If so just pass on 456 // the decayed value. 457 if (castTy->isPointerType() || castTy->isReferenceType()) 458 return val; 459 460 // Are we casting from an array to an integer? If so, cast the decayed 461 // pointer value to an integer. 462 assert(castTy->isIntegralOrEnumerationType()); 463 464 // FIXME: Keep these here for now in case we decide soon that we 465 // need the original decayed type. 466 // QualType elemTy = cast<ArrayType>(originalTy)->getElementType(); 467 // QualType pointerTy = C.getPointerType(elemTy); 468 return evalCastFromLoc(val.castAs<Loc>(), castTy); 469 } 470 471 // Check for casts from a region to a specific type. 472 if (const MemRegion *R = val.getAsRegion()) { 473 // Handle other casts of locations to integers. 474 if (castTy->isIntegralOrEnumerationType()) 475 return evalCastFromLoc(loc::MemRegionVal(R), castTy); 476 477 // FIXME: We should handle the case where we strip off view layers to get 478 // to a desugared type. 479 if (!Loc::isLocType(castTy)) { 480 // FIXME: There can be gross cases where one casts the result of a function 481 // (that returns a pointer) to some other value that happens to fit 482 // within that pointer value. We currently have no good way to 483 // model such operations. When this happens, the underlying operation 484 // is that the caller is reasoning about bits. Conceptually we are 485 // layering a "view" of a location on top of those bits. Perhaps 486 // we need to be more lazy about mutual possible views, even on an 487 // SVal? This may be necessary for bit-level reasoning as well. 488 return UnknownVal(); 489 } 490 491 // We get a symbolic function pointer for a dereference of a function 492 // pointer, but it is of function type. Example: 493 494 // struct FPRec { 495 // void (*my_func)(int * x); 496 // }; 497 // 498 // int bar(int x); 499 // 500 // int f1_a(struct FPRec* foo) { 501 // int x; 502 // (*foo->my_func)(&x); 503 // return bar(x)+1; // no-warning 504 // } 505 506 assert(Loc::isLocType(originalTy) || originalTy->isFunctionType() || 507 originalTy->isBlockPointerType() || castTy->isReferenceType()); 508 509 StoreManager &storeMgr = StateMgr.getStoreManager(); 510 511 // Delegate to store manager to get the result of casting a region to a 512 // different type. If the MemRegion* returned is NULL, this expression 513 // Evaluates to UnknownVal. 514 R = storeMgr.castRegion(R, castTy); 515 return R ? SVal(loc::MemRegionVal(R)) : UnknownVal(); 516 } 517 518 return dispatchCast(val, castTy); 519 } 520