head 1.1; branch 1.1.1; access; symbols netbsd-11-0-RC4:1.1.1.8 netbsd-11-0-RC3:1.1.1.8 netbsd-11-0-RC2:1.1.1.8 netbsd-11-0-RC1:1.1.1.8 perseant-exfatfs-base-20250801:1.1.1.8 netbsd-11:1.1.1.8.0.10 netbsd-11-base:1.1.1.8 netbsd-10-1-RELEASE:1.1.1.8 perseant-exfatfs-base-20240630:1.1.1.8 perseant-exfatfs:1.1.1.8.0.8 perseant-exfatfs-base:1.1.1.8 netbsd-8-3-RELEASE:1.1.1.5 netbsd-9-4-RELEASE:1.1.1.7 netbsd-10-0-RELEASE:1.1.1.8 netbsd-10-0-RC6:1.1.1.8 netbsd-10-0-RC5:1.1.1.8 netbsd-10-0-RC4:1.1.1.8 netbsd-10-0-RC3:1.1.1.8 netbsd-10-0-RC2:1.1.1.8 netbsd-10-0-RC1:1.1.1.8 netbsd-10:1.1.1.8.0.6 netbsd-10-base:1.1.1.8 netbsd-9-3-RELEASE:1.1.1.7 cjep_sun2x:1.1.1.8.0.4 cjep_sun2x-base:1.1.1.8 cjep_staticlib_x-base1:1.1.1.8 netbsd-9-2-RELEASE:1.1.1.7 cjep_staticlib_x:1.1.1.8.0.2 cjep_staticlib_x-base:1.1.1.8 netbsd-9-1-RELEASE:1.1.1.7 phil-wifi-20200421:1.1.1.8 phil-wifi-20200411:1.1.1.8 phil-wifi-20200406:1.1.1.8 netbsd-8-2-RELEASE:1.1.1.5 netbsd-9-0-RELEASE:1.1.1.7 netbsd-9-0-RC2:1.1.1.7 netbsd-9-0-RC1:1.1.1.7 netbsd-9:1.1.1.7.0.2 netbsd-9-base:1.1.1.7 phil-wifi-20190609:1.1.1.7 netbsd-8-1-RELEASE:1.1.1.5 netbsd-8-1-RC1:1.1.1.5 pgoyette-compat-merge-20190127:1.1.1.6.2.1 pgoyette-compat-20190127:1.1.1.7 pgoyette-compat-20190118:1.1.1.7 pgoyette-compat-1226:1.1.1.7 pgoyette-compat-1126:1.1.1.7 pgoyette-compat-1020:1.1.1.7 pgoyette-compat-0930:1.1.1.7 pgoyette-compat-0906:1.1.1.7 netbsd-7-2-RELEASE:1.1.1.3 pgoyette-compat-0728:1.1.1.7 clang-337282:1.1.1.7 netbsd-8-0-RELEASE:1.1.1.5 phil-wifi:1.1.1.6.0.4 phil-wifi-base:1.1.1.6 pgoyette-compat-0625:1.1.1.6 netbsd-8-0-RC2:1.1.1.5 pgoyette-compat-0521:1.1.1.6 pgoyette-compat-0502:1.1.1.6 pgoyette-compat-0422:1.1.1.6 netbsd-8-0-RC1:1.1.1.5 pgoyette-compat-0415:1.1.1.6 pgoyette-compat-0407:1.1.1.6 pgoyette-compat-0330:1.1.1.6 pgoyette-compat-0322:1.1.1.6 pgoyette-compat-0315:1.1.1.6 netbsd-7-1-2-RELEASE:1.1.1.3 pgoyette-compat:1.1.1.6.0.2 pgoyette-compat-base:1.1.1.6 netbsd-7-1-1-RELEASE:1.1.1.3 clang-319952:1.1.1.6 matt-nb8-mediatek:1.1.1.5.0.10 matt-nb8-mediatek-base:1.1.1.5 clang-309604:1.1.1.6 perseant-stdc-iso10646:1.1.1.5.0.8 perseant-stdc-iso10646-base:1.1.1.5 netbsd-8:1.1.1.5.0.6 netbsd-8-base:1.1.1.5 prg-localcount2-base3:1.1.1.5 prg-localcount2-base2:1.1.1.5 prg-localcount2-base1:1.1.1.5 prg-localcount2:1.1.1.5.0.4 prg-localcount2-base:1.1.1.5 pgoyette-localcount-20170426:1.1.1.5 bouyer-socketcan-base1:1.1.1.5 pgoyette-localcount-20170320:1.1.1.5 netbsd-7-1:1.1.1.3.0.10 netbsd-7-1-RELEASE:1.1.1.3 netbsd-7-1-RC2:1.1.1.3 clang-294123:1.1.1.5 netbsd-7-nhusb-base-20170116:1.1.1.3 bouyer-socketcan:1.1.1.5.0.2 bouyer-socketcan-base:1.1.1.5 clang-291444:1.1.1.5 pgoyette-localcount-20170107:1.1.1.4 netbsd-7-1-RC1:1.1.1.3 pgoyette-localcount-20161104:1.1.1.4 netbsd-7-0-2-RELEASE:1.1.1.3 localcount-20160914:1.1.1.4 netbsd-7-nhusb:1.1.1.3.0.8 netbsd-7-nhusb-base:1.1.1.3 clang-280599:1.1.1.4 pgoyette-localcount-20160806:1.1.1.4 pgoyette-localcount-20160726:1.1.1.4 pgoyette-localcount:1.1.1.4.0.2 pgoyette-localcount-base:1.1.1.4 netbsd-7-0-1-RELEASE:1.1.1.3 clang-261930:1.1.1.4 netbsd-7-0:1.1.1.3.0.6 netbsd-7-0-RELEASE:1.1.1.3 netbsd-7-0-RC3:1.1.1.3 netbsd-7-0-RC2:1.1.1.3 netbsd-7-0-RC1:1.1.1.3 clang-237755:1.1.1.3 clang-232565:1.1.1.3 clang-227398:1.1.1.3 tls-maxphys-base:1.1.1.3 tls-maxphys:1.1.1.3.0.4 netbsd-7:1.1.1.3.0.2 netbsd-7-base:1.1.1.3 clang-215315:1.1.1.3 clang-209886:1.1.1.3 yamt-pagecache:1.1.1.2.0.4 yamt-pagecache-base9:1.1.1.2 tls-earlyentropy:1.1.1.2.0.2 tls-earlyentropy-base:1.1.1.3 riastradh-xf86-video-intel-2-7-1-pre-2-21-15:1.1.1.2 riastradh-drm2-base3:1.1.1.2 clang-202566:1.1.1.2 clang-201163:1.1.1.2 clang-199312:1.1.1.2 clang-198450:1.1.1.2 clang-196603:1.1.1.1 clang-195771:1.1.1.1 LLVM:1.1.1; locks; strict; comment @// @; 1.1 date 2013.11.28.14.14.53; author joerg; state Exp; branches 1.1.1.1; next ; commitid ow8OybrawrB1f3fx; 1.1.1.1 date 2013.11.28.14.14.53; author joerg; state Exp; branches; next 1.1.1.2; commitid ow8OybrawrB1f3fx; 1.1.1.2 date 2014.01.05.15.39.33; author joerg; state Exp; branches 1.1.1.2.2.1 1.1.1.2.4.1; next 1.1.1.3; commitid wh3aCSIWykURqWjx; 1.1.1.3 date 2014.05.30.18.14.44; author joerg; state Exp; branches 1.1.1.3.4.1; next 1.1.1.4; commitid 8q0kdlBlCn09GACx; 1.1.1.4 date 2016.02.27.22.12.05; author joerg; state Exp; branches 1.1.1.4.2.1; next 1.1.1.5; commitid tIimz3oDlh1NpBWy; 1.1.1.5 date 2017.01.11.10.35.33; author joerg; state Exp; branches; next 1.1.1.6; commitid CNnUNfII1jgNmxBz; 1.1.1.6 date 2017.08.01.19.35.15; author joerg; state Exp; branches 1.1.1.6.2.1 1.1.1.6.4.1; next 1.1.1.7; commitid pMuDy65V0VicSx1A; 1.1.1.7 date 2018.07.17.18.31.06; author joerg; state Exp; branches; next 1.1.1.8; commitid wDzL46ALjrCZgwKA; 1.1.1.8 date 2019.11.13.22.19.29; author joerg; state dead; branches; next ; commitid QD8YATxuNG34YJKB; 1.1.1.2.2.1 date 2014.08.10.07.08.10; author tls; state Exp; branches; next ; commitid t01A1TLTYxkpGMLx; 1.1.1.2.4.1 date 2014.01.05.15.39.33; author yamt; state dead; branches; next 1.1.1.2.4.2; commitid WSrDtL5nYAUyiyBx; 1.1.1.2.4.2 date 2014.05.22.16.18.31; author yamt; state Exp; branches; next ; commitid WSrDtL5nYAUyiyBx; 1.1.1.3.4.1 date 2014.05.30.18.14.44; author tls; state dead; branches; next 1.1.1.3.4.2; commitid jTnpym9Qu0o4R1Nx; 1.1.1.3.4.2 date 2014.08.19.23.47.32; author tls; state Exp; branches; next ; commitid jTnpym9Qu0o4R1Nx; 1.1.1.4.2.1 date 2017.03.20.06.52.42; author pgoyette; state Exp; branches; next ; commitid jjw7cAwgyKq7RfKz; 1.1.1.6.2.1 date 2018.07.28.04.33.24; author pgoyette; state Exp; branches; next ; commitid 1UP1xAIUxv1ZgRLA; 1.1.1.6.4.1 date 2019.06.10.21.45.29; author christos; state Exp; branches; next 1.1.1.6.4.2; commitid jtc8rnCzWiEEHGqB; 1.1.1.6.4.2 date 2020.04.13.07.46.39; author martin; state dead; branches; next ; commitid X01YhRUPVUDaec4C; desc @@ 1.1 log @Initial revision @ text @// SimpleSValBuilder.cpp - A basic SValBuilder -----------------------*- C++ -*- // // The LLVM Compiler Infrastructure // // This file is distributed under the University of Illinois Open Source // License. See LICENSE.TXT for details. // //===----------------------------------------------------------------------===// // // This file defines SimpleSValBuilder, a basic implementation of SValBuilder. // //===----------------------------------------------------------------------===// #include "clang/StaticAnalyzer/Core/PathSensitive/SValBuilder.h" #include "clang/StaticAnalyzer/Core/PathSensitive/APSIntType.h" #include "clang/StaticAnalyzer/Core/PathSensitive/ProgramState.h" using namespace clang; using namespace ento; namespace { class SimpleSValBuilder : public SValBuilder { protected: virtual SVal dispatchCast(SVal val, QualType castTy); virtual SVal evalCastFromNonLoc(NonLoc val, QualType castTy); virtual SVal evalCastFromLoc(Loc val, QualType castTy); public: SimpleSValBuilder(llvm::BumpPtrAllocator &alloc, ASTContext &context, ProgramStateManager &stateMgr) : SValBuilder(alloc, context, stateMgr) {} virtual ~SimpleSValBuilder() {} virtual SVal evalMinus(NonLoc val); virtual SVal evalComplement(NonLoc val); virtual SVal evalBinOpNN(ProgramStateRef state, BinaryOperator::Opcode op, NonLoc lhs, NonLoc rhs, QualType resultTy); virtual SVal evalBinOpLL(ProgramStateRef state, BinaryOperator::Opcode op, Loc lhs, Loc rhs, QualType resultTy); virtual SVal evalBinOpLN(ProgramStateRef state, BinaryOperator::Opcode op, Loc lhs, NonLoc rhs, QualType resultTy); /// getKnownValue - evaluates a given SVal. If the SVal has only one possible /// (integer) value, that value is returned. Otherwise, returns NULL. virtual const llvm::APSInt *getKnownValue(ProgramStateRef state, SVal V); SVal MakeSymIntVal(const SymExpr *LHS, BinaryOperator::Opcode op, const llvm::APSInt &RHS, QualType resultTy); }; } // end anonymous namespace SValBuilder *ento::createSimpleSValBuilder(llvm::BumpPtrAllocator &alloc, ASTContext &context, ProgramStateManager &stateMgr) { return new SimpleSValBuilder(alloc, context, stateMgr); } //===----------------------------------------------------------------------===// // Transfer function for Casts. //===----------------------------------------------------------------------===// SVal SimpleSValBuilder::dispatchCast(SVal Val, QualType CastTy) { assert(Val.getAs() || Val.getAs()); return Val.getAs() ? evalCastFromLoc(Val.castAs(), CastTy) : evalCastFromNonLoc(Val.castAs(), CastTy); } SVal SimpleSValBuilder::evalCastFromNonLoc(NonLoc val, QualType castTy) { bool isLocType = Loc::isLocType(castTy); if (Optional LI = val.getAs()) { if (isLocType) return LI->getLoc(); // FIXME: Correctly support promotions/truncations. unsigned castSize = Context.getTypeSize(castTy); if (castSize == LI->getNumBits()) return val; return makeLocAsInteger(LI->getLoc(), castSize); } if (const SymExpr *se = val.getAsSymbolicExpression()) { QualType T = Context.getCanonicalType(se->getType()); // If types are the same or both are integers, ignore the cast. // FIXME: Remove this hack when we support symbolic truncation/extension. // HACK: If both castTy and T are integers, ignore the cast. This is // not a permanent solution. Eventually we want to precisely handle // extension/truncation of symbolic integers. This prevents us from losing // precision when we assign 'x = y' and 'y' is symbolic and x and y are // different integer types. if (haveSameType(T, castTy)) return val; if (!isLocType) return makeNonLoc(se, T, castTy); return UnknownVal(); } // If value is a non integer constant, produce unknown. if (!val.getAs()) return UnknownVal(); // Handle casts to a boolean type. if (castTy->isBooleanType()) { bool b = val.castAs().getValue().getBoolValue(); return makeTruthVal(b, castTy); } // Only handle casts from integers to integers - if val is an integer constant // being cast to a non integer type, produce unknown. if (!isLocType && !castTy->isIntegralOrEnumerationType()) return UnknownVal(); llvm::APSInt i = val.castAs().getValue(); BasicVals.getAPSIntType(castTy).apply(i); if (isLocType) return makeIntLocVal(i); else return makeIntVal(i); } SVal SimpleSValBuilder::evalCastFromLoc(Loc val, QualType castTy) { // Casts from pointers -> pointers, just return the lval. // // Casts from pointers -> references, just return the lval. These // can be introduced by the frontend for corner cases, e.g // casting from va_list* to __builtin_va_list&. // if (Loc::isLocType(castTy) || castTy->isReferenceType()) return val; // FIXME: Handle transparent unions where a value can be "transparently" // lifted into a union type. if (castTy->isUnionType()) return UnknownVal(); // Casting a Loc to a bool will almost always be true, // unless this is a weak function or a symbolic region. if (castTy->isBooleanType()) { switch (val.getSubKind()) { case loc::MemRegionKind: { const MemRegion *R = val.castAs().getRegion(); if (const FunctionTextRegion *FTR = dyn_cast(R)) if (const FunctionDecl *FD = dyn_cast(FTR->getDecl())) if (FD->isWeak()) // FIXME: Currently we are using an extent symbol here, // because there are no generic region address metadata // symbols to use, only content metadata. return nonloc::SymbolVal(SymMgr.getExtentSymbol(FTR)); if (const SymbolicRegion *SymR = R->getSymbolicBase()) return nonloc::SymbolVal(SymR->getSymbol()); // FALL-THROUGH } case loc::GotoLabelKind: // Labels and non symbolic memory regions are always true. return makeTruthVal(true, castTy); } } if (castTy->isIntegralOrEnumerationType()) { unsigned BitWidth = Context.getTypeSize(castTy); if (!val.getAs()) return makeLocAsInteger(val, BitWidth); llvm::APSInt i = val.castAs().getValue(); BasicVals.getAPSIntType(castTy).apply(i); return makeIntVal(i); } // All other cases: return 'UnknownVal'. This includes casting pointers // to floats, which is probably badness it itself, but this is a good // intermediate solution until we do something better. return UnknownVal(); } //===----------------------------------------------------------------------===// // Transfer function for unary operators. //===----------------------------------------------------------------------===// SVal SimpleSValBuilder::evalMinus(NonLoc val) { switch (val.getSubKind()) { case nonloc::ConcreteIntKind: return val.castAs().evalMinus(*this); default: return UnknownVal(); } } SVal SimpleSValBuilder::evalComplement(NonLoc X) { switch (X.getSubKind()) { case nonloc::ConcreteIntKind: return X.castAs().evalComplement(*this); default: return UnknownVal(); } } //===----------------------------------------------------------------------===// // Transfer function for binary operators. //===----------------------------------------------------------------------===// SVal SimpleSValBuilder::MakeSymIntVal(const SymExpr *LHS, BinaryOperator::Opcode op, const llvm::APSInt &RHS, QualType resultTy) { bool isIdempotent = false; // Check for a few special cases with known reductions first. switch (op) { default: // We can't reduce this case; just treat it normally. break; case BO_Mul: // a*0 and a*1 if (RHS == 0) return makeIntVal(0, resultTy); else if (RHS == 1) isIdempotent = true; break; case BO_Div: // a/0 and a/1 if (RHS == 0) // This is also handled elsewhere. return UndefinedVal(); else if (RHS == 1) isIdempotent = true; break; case BO_Rem: // a%0 and a%1 if (RHS == 0) // This is also handled elsewhere. return UndefinedVal(); else if (RHS == 1) return makeIntVal(0, resultTy); break; case BO_Add: case BO_Sub: case BO_Shl: case BO_Shr: case BO_Xor: // a+0, a-0, a<<0, a>>0, a^0 if (RHS == 0) isIdempotent = true; break; case BO_And: // a&0 and a&(~0) if (RHS == 0) return makeIntVal(0, resultTy); else if (RHS.isAllOnesValue()) isIdempotent = true; break; case BO_Or: // a|0 and a|(~0) if (RHS == 0) isIdempotent = true; else if (RHS.isAllOnesValue()) { const llvm::APSInt &Result = BasicVals.Convert(resultTy, RHS); return nonloc::ConcreteInt(Result); } break; } // Idempotent ops (like a*1) can still change the type of an expression. // Wrap the LHS up in a NonLoc again and let evalCastFromNonLoc do the // dirty work. if (isIdempotent) return evalCastFromNonLoc(nonloc::SymbolVal(LHS), resultTy); // If we reach this point, the expression cannot be simplified. // Make a SymbolVal for the entire expression, after converting the RHS. const llvm::APSInt *ConvertedRHS = &RHS; if (BinaryOperator::isComparisonOp(op)) { // We're looking for a type big enough to compare the symbolic value // with the given constant. // FIXME: This is an approximation of Sema::UsualArithmeticConversions. ASTContext &Ctx = getContext(); QualType SymbolType = LHS->getType(); uint64_t ValWidth = RHS.getBitWidth(); uint64_t TypeWidth = Ctx.getTypeSize(SymbolType); if (ValWidth < TypeWidth) { // If the value is too small, extend it. ConvertedRHS = &BasicVals.Convert(SymbolType, RHS); } else if (ValWidth == TypeWidth) { // If the value is signed but the symbol is unsigned, do the comparison // in unsigned space. [C99 6.3.1.8] // (For the opposite case, the value is already unsigned.) if (RHS.isSigned() && !SymbolType->isSignedIntegerOrEnumerationType()) ConvertedRHS = &BasicVals.Convert(SymbolType, RHS); } } else ConvertedRHS = &BasicVals.Convert(resultTy, RHS); return makeNonLoc(LHS, op, *ConvertedRHS, resultTy); } SVal SimpleSValBuilder::evalBinOpNN(ProgramStateRef state, BinaryOperator::Opcode op, NonLoc lhs, NonLoc rhs, QualType resultTy) { NonLoc InputLHS = lhs; NonLoc InputRHS = rhs; // Handle trivial case where left-side and right-side are the same. if (lhs == rhs) switch (op) { default: break; case BO_EQ: case BO_LE: case BO_GE: return makeTruthVal(true, resultTy); case BO_LT: case BO_GT: case BO_NE: return makeTruthVal(false, resultTy); case BO_Xor: case BO_Sub: if (resultTy->isIntegralOrEnumerationType()) return makeIntVal(0, resultTy); return evalCastFromNonLoc(makeIntVal(0, /*Unsigned=*/false), resultTy); case BO_Or: case BO_And: return evalCastFromNonLoc(lhs, resultTy); } while (1) { switch (lhs.getSubKind()) { default: return makeSymExprValNN(state, op, lhs, rhs, resultTy); case nonloc::LocAsIntegerKind: { Loc lhsL = lhs.castAs().getLoc(); switch (rhs.getSubKind()) { case nonloc::LocAsIntegerKind: return evalBinOpLL(state, op, lhsL, rhs.castAs().getLoc(), resultTy); case nonloc::ConcreteIntKind: { // Transform the integer into a location and compare. llvm::APSInt i = rhs.castAs().getValue(); BasicVals.getAPSIntType(Context.VoidPtrTy).apply(i); return evalBinOpLL(state, op, lhsL, makeLoc(i), resultTy); } default: switch (op) { case BO_EQ: return makeTruthVal(false, resultTy); case BO_NE: return makeTruthVal(true, resultTy); default: // This case also handles pointer arithmetic. return makeSymExprValNN(state, op, InputLHS, InputRHS, resultTy); } } } case nonloc::ConcreteIntKind: { llvm::APSInt LHSValue = lhs.castAs().getValue(); // If we're dealing with two known constants, just perform the operation. if (const llvm::APSInt *KnownRHSValue = getKnownValue(state, rhs)) { llvm::APSInt RHSValue = *KnownRHSValue; if (BinaryOperator::isComparisonOp(op)) { // We're looking for a type big enough to compare the two values. // FIXME: This is not correct. char + short will result in a promotion // to int. Unfortunately we have lost types by this point. APSIntType CompareType = std::max(APSIntType(LHSValue), APSIntType(RHSValue)); CompareType.apply(LHSValue); CompareType.apply(RHSValue); } else if (!BinaryOperator::isShiftOp(op)) { APSIntType IntType = BasicVals.getAPSIntType(resultTy); IntType.apply(LHSValue); IntType.apply(RHSValue); } const llvm::APSInt *Result = BasicVals.evalAPSInt(op, LHSValue, RHSValue); if (!Result) return UndefinedVal(); return nonloc::ConcreteInt(*Result); } // Swap the left and right sides and flip the operator if doing so // allows us to better reason about the expression (this is a form // of expression canonicalization). // While we're at it, catch some special cases for non-commutative ops. switch (op) { case BO_LT: case BO_GT: case BO_LE: case BO_GE: op = BinaryOperator::reverseComparisonOp(op); // FALL-THROUGH case BO_EQ: case BO_NE: case BO_Add: case BO_Mul: case BO_And: case BO_Xor: case BO_Or: std::swap(lhs, rhs); continue; case BO_Shr: // (~0)>>a if (LHSValue.isAllOnesValue() && LHSValue.isSigned()) return evalCastFromNonLoc(lhs, resultTy); // FALL-THROUGH case BO_Shl: // 0<>a if (LHSValue == 0) return evalCastFromNonLoc(lhs, resultTy); return makeSymExprValNN(state, op, InputLHS, InputRHS, resultTy); default: return makeSymExprValNN(state, op, InputLHS, InputRHS, resultTy); } } case nonloc::SymbolValKind: { // We only handle LHS as simple symbols or SymIntExprs. SymbolRef Sym = lhs.castAs().getSymbol(); // LHS is a symbolic expression. if (const SymIntExpr *symIntExpr = dyn_cast(Sym)) { // Is this a logical not? (!x is represented as x == 0.) if (op == BO_EQ && rhs.isZeroConstant()) { // We know how to negate certain expressions. Simplify them here. BinaryOperator::Opcode opc = symIntExpr->getOpcode(); switch (opc) { default: // We don't know how to negate this operation. // Just handle it as if it were a normal comparison to 0. break; case BO_LAnd: case BO_LOr: llvm_unreachable("Logical operators handled by branching logic."); case BO_Assign: case BO_MulAssign: case BO_DivAssign: case BO_RemAssign: case BO_AddAssign: case BO_SubAssign: case BO_ShlAssign: case BO_ShrAssign: case BO_AndAssign: case BO_XorAssign: case BO_OrAssign: case BO_Comma: llvm_unreachable("'=' and ',' operators handled by ExprEngine."); case BO_PtrMemD: case BO_PtrMemI: llvm_unreachable("Pointer arithmetic not handled here."); case BO_LT: case BO_GT: case BO_LE: case BO_GE: case BO_EQ: case BO_NE: assert(resultTy->isBooleanType() || resultTy == getConditionType()); assert(symIntExpr->getType()->isBooleanType() || getContext().hasSameUnqualifiedType(symIntExpr->getType(), getConditionType())); // Negate the comparison and make a value. opc = BinaryOperator::negateComparisonOp(opc); return makeNonLoc(symIntExpr->getLHS(), opc, symIntExpr->getRHS(), resultTy); } } // For now, only handle expressions whose RHS is a constant. if (const llvm::APSInt *RHSValue = getKnownValue(state, rhs)) { // If both the LHS and the current expression are additive, // fold their constants and try again. if (BinaryOperator::isAdditiveOp(op)) { BinaryOperator::Opcode lop = symIntExpr->getOpcode(); if (BinaryOperator::isAdditiveOp(lop)) { // Convert the two constants to a common type, then combine them. // resultTy may not be the best type to convert to, but it's // probably the best choice in expressions with mixed type // (such as x+1U+2LL). The rules for implicit conversions should // choose a reasonable type to preserve the expression, and will // at least match how the value is going to be used. APSIntType IntType = BasicVals.getAPSIntType(resultTy); const llvm::APSInt &first = IntType.convert(symIntExpr->getRHS()); const llvm::APSInt &second = IntType.convert(*RHSValue); const llvm::APSInt *newRHS; if (lop == op) newRHS = BasicVals.evalAPSInt(BO_Add, first, second); else newRHS = BasicVals.evalAPSInt(BO_Sub, first, second); assert(newRHS && "Invalid operation despite common type!"); rhs = nonloc::ConcreteInt(*newRHS); lhs = nonloc::SymbolVal(symIntExpr->getLHS()); op = lop; continue; } } // Otherwise, make a SymIntExpr out of the expression. return MakeSymIntVal(symIntExpr, op, *RHSValue, resultTy); } } // Does the symbolic expression simplify to a constant? // If so, "fold" the constant by setting 'lhs' to a ConcreteInt // and try again. ConstraintManager &CMgr = state->getConstraintManager(); if (const llvm::APSInt *Constant = CMgr.getSymVal(state, Sym)) { lhs = nonloc::ConcreteInt(*Constant); continue; } // Is the RHS a constant? if (const llvm::APSInt *RHSValue = getKnownValue(state, rhs)) return MakeSymIntVal(Sym, op, *RHSValue, resultTy); // Give up -- this is not a symbolic expression we can handle. return makeSymExprValNN(state, op, InputLHS, InputRHS, resultTy); } } } } static SVal evalBinOpFieldRegionFieldRegion(const FieldRegion *LeftFR, const FieldRegion *RightFR, BinaryOperator::Opcode op, QualType resultTy, SimpleSValBuilder &SVB) { // Only comparisons are meaningful here! if (!BinaryOperator::isComparisonOp(op)) return UnknownVal(); // Next, see if the two FRs have the same super-region. // FIXME: This doesn't handle casts yet, and simply stripping the casts // doesn't help. if (LeftFR->getSuperRegion() != RightFR->getSuperRegion()) return UnknownVal(); const FieldDecl *LeftFD = LeftFR->getDecl(); const FieldDecl *RightFD = RightFR->getDecl(); const RecordDecl *RD = LeftFD->getParent(); // Make sure the two FRs are from the same kind of record. Just in case! // FIXME: This is probably where inheritance would be a problem. if (RD != RightFD->getParent()) return UnknownVal(); // We know for sure that the two fields are not the same, since that // would have given us the same SVal. if (op == BO_EQ) return SVB.makeTruthVal(false, resultTy); if (op == BO_NE) return SVB.makeTruthVal(true, resultTy); // Iterate through the fields and see which one comes first. // [C99 6.7.2.1.13] "Within a structure object, the non-bit-field // members and the units in which bit-fields reside have addresses that // increase in the order in which they are declared." bool leftFirst = (op == BO_LT || op == BO_LE); for (RecordDecl::field_iterator I = RD->field_begin(), E = RD->field_end(); I!=E; ++I) { if (*I == LeftFD) return SVB.makeTruthVal(leftFirst, resultTy); if (*I == RightFD) return SVB.makeTruthVal(!leftFirst, resultTy); } llvm_unreachable("Fields not found in parent record's definition"); } // FIXME: all this logic will change if/when we have MemRegion::getLocation(). SVal SimpleSValBuilder::evalBinOpLL(ProgramStateRef state, BinaryOperator::Opcode op, Loc lhs, Loc rhs, QualType resultTy) { // Only comparisons and subtractions are valid operations on two pointers. // See [C99 6.5.5 through 6.5.14] or [C++0x 5.6 through 5.15]. // However, if a pointer is casted to an integer, evalBinOpNN may end up // calling this function with another operation (PR7527). We don't attempt to // model this for now, but it could be useful, particularly when the // "location" is actually an integer value that's been passed through a void*. if (!(BinaryOperator::isComparisonOp(op) || op == BO_Sub)) return UnknownVal(); // Special cases for when both sides are identical. if (lhs == rhs) { switch (op) { default: llvm_unreachable("Unimplemented operation for two identical values"); case BO_Sub: return makeZeroVal(resultTy); case BO_EQ: case BO_LE: case BO_GE: return makeTruthVal(true, resultTy); case BO_NE: case BO_LT: case BO_GT: return makeTruthVal(false, resultTy); } } switch (lhs.getSubKind()) { default: llvm_unreachable("Ordering not implemented for this Loc."); case loc::GotoLabelKind: // The only thing we know about labels is that they're non-null. if (rhs.isZeroConstant()) { switch (op) { default: break; case BO_Sub: return evalCastFromLoc(lhs, resultTy); case BO_EQ: case BO_LE: case BO_LT: return makeTruthVal(false, resultTy); case BO_NE: case BO_GT: case BO_GE: return makeTruthVal(true, resultTy); } } // There may be two labels for the same location, and a function region may // have the same address as a label at the start of the function (depending // on the ABI). // FIXME: we can probably do a comparison against other MemRegions, though. // FIXME: is there a way to tell if two labels refer to the same location? return UnknownVal(); case loc::ConcreteIntKind: { // If one of the operands is a symbol and the other is a constant, // build an expression for use by the constraint manager. if (SymbolRef rSym = rhs.getAsLocSymbol()) { // We can only build expressions with symbols on the left, // so we need a reversible operator. if (!BinaryOperator::isComparisonOp(op)) return UnknownVal(); const llvm::APSInt &lVal = lhs.castAs().getValue(); op = BinaryOperator::reverseComparisonOp(op); return makeNonLoc(rSym, op, lVal, resultTy); } // If both operands are constants, just perform the operation. if (Optional rInt = rhs.getAs()) { SVal ResultVal = lhs.castAs().evalBinOp(BasicVals, op, *rInt); if (Optional Result = ResultVal.getAs()) return evalCastFromNonLoc(*Result, resultTy); assert(!ResultVal.getAs() && "Loc-Loc ops should not produce Locs"); return UnknownVal(); } // Special case comparisons against NULL. // This must come after the test if the RHS is a symbol, which is used to // build constraints. The address of any non-symbolic region is guaranteed // to be non-NULL, as is any label. assert(rhs.getAs() || rhs.getAs()); if (lhs.isZeroConstant()) { switch (op) { default: break; case BO_EQ: case BO_GT: case BO_GE: return makeTruthVal(false, resultTy); case BO_NE: case BO_LT: case BO_LE: return makeTruthVal(true, resultTy); } } // Comparing an arbitrary integer to a region or label address is // completely unknowable. return UnknownVal(); } case loc::MemRegionKind: { if (Optional rInt = rhs.getAs()) { // If one of the operands is a symbol and the other is a constant, // build an expression for use by the constraint manager. if (SymbolRef lSym = lhs.getAsLocSymbol(true)) return MakeSymIntVal(lSym, op, rInt->getValue(), resultTy); // Special case comparisons to NULL. // This must come after the test if the LHS is a symbol, which is used to // build constraints. The address of any non-symbolic region is guaranteed // to be non-NULL. if (rInt->isZeroConstant()) { if (op == BO_Sub) return evalCastFromLoc(lhs, resultTy); if (BinaryOperator::isComparisonOp(op)) { QualType boolType = getContext().BoolTy; NonLoc l = evalCastFromLoc(lhs, boolType).castAs(); NonLoc r = makeTruthVal(false, boolType).castAs(); return evalBinOpNN(state, op, l, r, resultTy); } } // Comparing a region to an arbitrary integer is completely unknowable. return UnknownVal(); } // Get both values as regions, if possible. const MemRegion *LeftMR = lhs.getAsRegion(); assert(LeftMR && "MemRegionKind SVal doesn't have a region!"); const MemRegion *RightMR = rhs.getAsRegion(); if (!RightMR) // The RHS is probably a label, which in theory could address a region. // FIXME: we can probably make a more useful statement about non-code // regions, though. return UnknownVal(); const MemRegion *LeftBase = LeftMR->getBaseRegion(); const MemRegion *RightBase = RightMR->getBaseRegion(); const MemSpaceRegion *LeftMS = LeftBase->getMemorySpace(); const MemSpaceRegion *RightMS = RightBase->getMemorySpace(); const MemSpaceRegion *UnknownMS = MemMgr.getUnknownRegion(); // If the two regions are from different known memory spaces they cannot be // equal. Also, assume that no symbolic region (whose memory space is // unknown) is on the stack. if (LeftMS != RightMS && ((LeftMS != UnknownMS && RightMS != UnknownMS) || (isa(LeftMS) || isa(RightMS)))) { switch (op) { default: return UnknownVal(); case BO_EQ: return makeTruthVal(false, resultTy); case BO_NE: return makeTruthVal(true, resultTy); } } // If both values wrap regions, see if they're from different base regions. // Note, heap base symbolic regions are assumed to not alias with // each other; for example, we assume that malloc returns different address // on each invocation. if (LeftBase != RightBase && ((!isa(LeftBase) && !isa(RightBase)) || (isa(LeftMS) || isa(RightMS))) ){ switch (op) { default: return UnknownVal(); case BO_EQ: return makeTruthVal(false, resultTy); case BO_NE: return makeTruthVal(true, resultTy); } } // Handle special cases for when both regions are element regions. const ElementRegion *RightER = dyn_cast(RightMR); const ElementRegion *LeftER = dyn_cast(LeftMR); if (RightER && LeftER) { // Next, see if the two ERs have the same super-region and matching types. // FIXME: This should do something useful even if the types don't match, // though if both indexes are constant the RegionRawOffset path will // give the correct answer. if (LeftER->getSuperRegion() == RightER->getSuperRegion() && LeftER->getElementType() == RightER->getElementType()) { // Get the left index and cast it to the correct type. // If the index is unknown or undefined, bail out here. SVal LeftIndexVal = LeftER->getIndex(); Optional LeftIndex = LeftIndexVal.getAs(); if (!LeftIndex) return UnknownVal(); LeftIndexVal = evalCastFromNonLoc(*LeftIndex, ArrayIndexTy); LeftIndex = LeftIndexVal.getAs(); if (!LeftIndex) return UnknownVal(); // Do the same for the right index. SVal RightIndexVal = RightER->getIndex(); Optional RightIndex = RightIndexVal.getAs(); if (!RightIndex) return UnknownVal(); RightIndexVal = evalCastFromNonLoc(*RightIndex, ArrayIndexTy); RightIndex = RightIndexVal.getAs(); if (!RightIndex) return UnknownVal(); // Actually perform the operation. // evalBinOpNN expects the two indexes to already be the right type. return evalBinOpNN(state, op, *LeftIndex, *RightIndex, resultTy); } } // Special handling of the FieldRegions, even with symbolic offsets. const FieldRegion *RightFR = dyn_cast(RightMR); const FieldRegion *LeftFR = dyn_cast(LeftMR); if (RightFR && LeftFR) { SVal R = evalBinOpFieldRegionFieldRegion(LeftFR, RightFR, op, resultTy, *this); if (!R.isUnknown()) return R; } // Compare the regions using the raw offsets. RegionOffset LeftOffset = LeftMR->getAsOffset(); RegionOffset RightOffset = RightMR->getAsOffset(); if (LeftOffset.getRegion() != NULL && LeftOffset.getRegion() == RightOffset.getRegion() && !LeftOffset.hasSymbolicOffset() && !RightOffset.hasSymbolicOffset()) { int64_t left = LeftOffset.getOffset(); int64_t right = RightOffset.getOffset(); switch (op) { default: return UnknownVal(); case BO_LT: return makeTruthVal(left < right, resultTy); case BO_GT: return makeTruthVal(left > right, resultTy); case BO_LE: return makeTruthVal(left <= right, resultTy); case BO_GE: return makeTruthVal(left >= right, resultTy); case BO_EQ: return makeTruthVal(left == right, resultTy); case BO_NE: return makeTruthVal(left != right, resultTy); } } // At this point we're not going to get a good answer, but we can try // conjuring an expression instead. SymbolRef LHSSym = lhs.getAsLocSymbol(); SymbolRef RHSSym = rhs.getAsLocSymbol(); if (LHSSym && RHSSym) return makeNonLoc(LHSSym, op, RHSSym, resultTy); // If we get here, we have no way of comparing the regions. return UnknownVal(); } } } SVal SimpleSValBuilder::evalBinOpLN(ProgramStateRef state, BinaryOperator::Opcode op, Loc lhs, NonLoc rhs, QualType resultTy) { assert(!BinaryOperator::isComparisonOp(op) && "arguments to comparison ops must be of the same type"); // Special case: rhs is a zero constant. if (rhs.isZeroConstant()) return lhs; // We are dealing with pointer arithmetic. // Handle pointer arithmetic on constant values. if (Optional rhsInt = rhs.getAs()) { if (Optional lhsInt = lhs.getAs()) { const llvm::APSInt &leftI = lhsInt->getValue(); assert(leftI.isUnsigned()); llvm::APSInt rightI(rhsInt->getValue(), /* isUnsigned */ true); // Convert the bitwidth of rightI. This should deal with overflow // since we are dealing with concrete values. rightI = rightI.extOrTrunc(leftI.getBitWidth()); // Offset the increment by the pointer size. llvm::APSInt Multiplicand(rightI.getBitWidth(), /* isUnsigned */ true); rightI *= Multiplicand; // Compute the adjusted pointer. switch (op) { case BO_Add: rightI = leftI + rightI; break; case BO_Sub: rightI = leftI - rightI; break; default: llvm_unreachable("Invalid pointer arithmetic operation"); } return loc::ConcreteInt(getBasicValueFactory().getValue(rightI)); } } // Handle cases where 'lhs' is a region. if (const MemRegion *region = lhs.getAsRegion()) { rhs = convertToArrayIndex(rhs).castAs(); SVal index = UnknownVal(); const MemRegion *superR = 0; QualType elementType; if (const ElementRegion *elemReg = dyn_cast(region)) { assert(op == BO_Add || op == BO_Sub); index = evalBinOpNN(state, op, elemReg->getIndex(), rhs, getArrayIndexType()); superR = elemReg->getSuperRegion(); elementType = elemReg->getElementType(); } else if (isa(region)) { superR = region; index = rhs; if (resultTy->isAnyPointerType()) elementType = resultTy->getPointeeType(); } if (Optional indexV = index.getAs()) { return loc::MemRegionVal(MemMgr.getElementRegion(elementType, *indexV, superR, getContext())); } } return UnknownVal(); } const llvm::APSInt *SimpleSValBuilder::getKnownValue(ProgramStateRef state, SVal V) { if (V.isUnknownOrUndef()) return NULL; if (Optional X = V.getAs()) return &X->getValue(); if (Optional X = V.getAs()) return &X->getValue(); if (SymbolRef Sym = V.getAsSymbol()) return state->getConstraintManager().getSymVal(state, Sym); // FIXME: Add support for SymExprs. return NULL; } @ 1.1.1.1 log @Import Clang 3.4rc1 r195771. @ text @@ 1.1.1.2 log @Import clang 3.5svn r198450. @ text @d100 1 a100 1 // If value is a non-integer constant, produce unknown. d111 1 a111 1 // being cast to a non-integer type, produce unknown. d161 1 a161 1 // Labels and non-symbolic memory regions are always true. @ 1.1.1.2.2.1 log @Rebase. @ text @d24 3 a26 3 SVal dispatchCast(SVal val, QualType castTy) override; SVal evalCastFromNonLoc(NonLoc val, QualType castTy) override; SVal evalCastFromLoc(Loc val, QualType castTy) override; d34 8 a41 8 SVal evalMinus(NonLoc val) override; SVal evalComplement(NonLoc val) override; SVal evalBinOpNN(ProgramStateRef state, BinaryOperator::Opcode op, NonLoc lhs, NonLoc rhs, QualType resultTy) override; SVal evalBinOpLL(ProgramStateRef state, BinaryOperator::Opcode op, Loc lhs, Loc rhs, QualType resultTy) override; SVal evalBinOpLN(ProgramStateRef state, BinaryOperator::Opcode op, Loc lhs, NonLoc rhs, QualType resultTy) override; d45 2 a46 2 const llvm::APSInt *getKnownValue(ProgramStateRef state, SVal V) override; d572 3 a574 2 for (const auto *I : RD->fields()) { if (I == LeftFD) d576 1 a576 1 if (I == RightFD) d821 1 a821 1 if (LeftOffset.getRegion() != nullptr && d904 1 a904 1 const MemRegion *superR = nullptr; d932 1 a932 1 return nullptr; d944 1 a944 1 return nullptr; @ 1.1.1.3 log @Import Clang 3.5svn r209886. @ text @d24 3 a26 3 SVal dispatchCast(SVal val, QualType castTy) override; SVal evalCastFromNonLoc(NonLoc val, QualType castTy) override; SVal evalCastFromLoc(Loc val, QualType castTy) override; d34 8 a41 8 SVal evalMinus(NonLoc val) override; SVal evalComplement(NonLoc val) override; SVal evalBinOpNN(ProgramStateRef state, BinaryOperator::Opcode op, NonLoc lhs, NonLoc rhs, QualType resultTy) override; SVal evalBinOpLL(ProgramStateRef state, BinaryOperator::Opcode op, Loc lhs, Loc rhs, QualType resultTy) override; SVal evalBinOpLN(ProgramStateRef state, BinaryOperator::Opcode op, Loc lhs, NonLoc rhs, QualType resultTy) override; d45 2 a46 2 const llvm::APSInt *getKnownValue(ProgramStateRef state, SVal V) override; d572 3 a574 2 for (const auto *I : RD->fields()) { if (I == LeftFD) d576 1 a576 1 if (I == RightFD) d821 1 a821 1 if (LeftOffset.getRegion() != nullptr && d904 1 a904 1 const MemRegion *superR = nullptr; d932 1 a932 1 return nullptr; d944 1 a944 1 return nullptr; @ 1.1.1.4 log @Import Clang 3.8.0rc3 r261930. @ text @d32 1 a32 1 ~SimpleSValBuilder() override {} d144 1 a144 1 case loc::MemRegionValKind: { d146 1 a146 1 if (const FunctionCodeRegion *FTR = dyn_cast(R)) d641 1 a641 1 return UnknownVal(); d692 1 a692 1 case loc::MemRegionValKind: { d721 1 a721 1 assert(LeftMR && "MemRegionValKind SVal doesn't have a region!"); d866 1 a866 1 d883 1 a883 1 a913 2 assert(op == BO_Add || op == BO_Sub); index = (op == BO_Add) ? rhs : evalMinus(rhs); d915 1 d925 1 a925 1 return UnknownVal(); @ 1.1.1.4.2.1 log @Sync with HEAD @ text @a71 3 if (val.getAs()) return val; a337 15 case nonloc::PointerToMemberKind: { assert(rhs.getSubKind() == nonloc::PointerToMemberKind && "Both SVals should have pointer-to-member-type"); auto LPTM = lhs.castAs(), RPTM = rhs.castAs(); auto LPTMD = LPTM.getPTMData(), RPTMD = RPTM.getPTMData(); switch (op) { case BO_EQ: return makeTruthVal(LPTMD == RPTMD, resultTy); case BO_NE: return makeTruthVal(LPTMD != RPTMD, resultTy); default: return UnknownVal(); } } a755 6 // FIXME: ObjC object pointers always reside on the heap, but currently // we treat their memory space as unknown, because symbolic pointers // to ObjC objects may alias. There should be a way to construct // possibly-aliasing heap-based regions. For instance, MacOSXApiChecker // guesses memory space for ObjC object pointers manually instead of // relying on us. a859 17 if (op >= BO_PtrMemD && op <= BO_PtrMemI) { if (auto PTMSV = rhs.getAs()) { if (PTMSV->isNullMemberPointer()) return UndefinedVal(); if (const FieldDecl *FD = PTMSV->getDeclAs()) { SVal Result = lhs; for (const auto &I : *PTMSV) Result = StateMgr.getStoreManager().evalDerivedToBase( Result, I->getType(),I->isVirtual()); return state->getLValue(FD, Result); } } return rhs; } @ 1.1.1.5 log @Import Clang pre-4.0.0 r291444. @ text @a71 3 if (val.getAs()) return val; a337 15 case nonloc::PointerToMemberKind: { assert(rhs.getSubKind() == nonloc::PointerToMemberKind && "Both SVals should have pointer-to-member-type"); auto LPTM = lhs.castAs(), RPTM = rhs.castAs(); auto LPTMD = LPTM.getPTMData(), RPTMD = RPTM.getPTMData(); switch (op) { case BO_EQ: return makeTruthVal(LPTMD == RPTMD, resultTy); case BO_NE: return makeTruthVal(LPTMD != RPTMD, resultTy); default: return UnknownVal(); } } a755 6 // FIXME: ObjC object pointers always reside on the heap, but currently // we treat their memory space as unknown, because symbolic pointers // to ObjC objects may alias. There should be a way to construct // possibly-aliasing heap-based regions. For instance, MacOSXApiChecker // guesses memory space for ObjC object pointers manually instead of // relying on us. a859 17 if (op >= BO_PtrMemD && op <= BO_PtrMemI) { if (auto PTMSV = rhs.getAs()) { if (PTMSV->isNullMemberPointer()) return UndefinedVal(); if (const FieldDecl *FD = PTMSV->getDeclAs()) { SVal Result = lhs; for (const auto &I : *PTMSV) Result = StateMgr.getStoreManager().evalDerivedToBase( Result, I->getType(),I->isVirtual()); return state->getLValue(FD, Result); } } return rhs; } @ 1.1.1.6 log @Import clang r309604 from branches/release_50 @ text @a16 1 #include "clang/StaticAnalyzer/Core/PathSensitive/SValVisitor.h" a46 4 /// Recursively descends into symbolic expressions and replaces symbols /// with their known values (in the sense of the getKnownValue() method). SVal simplifySVal(ProgramStateRef State, SVal V) override; d69 1 d71 1 d78 1 d80 1 a80 1 unsigned castSize = Context.getIntWidth(castTy); a160 1 LLVM_FALLTHROUGH; d170 1 a170 1 unsigned BitWidth = Context.getIntWidth(castTy); a364 3 // FIXME: This only makes sense for comparisons. If we want to, say, // add 1 to a LocAsInteger, we'd better unpack the Loc and add to it, // then pack it back into a LocAsInteger. d537 5 a541 6 SVal simplifiedLhs = simplifySVal(state, lhs); if (simplifiedLhs != lhs) if (auto simplifiedLhsAsNonLoc = simplifiedLhs.getAs()) { lhs = *simplifiedLhsAsNonLoc; continue; } d944 1 a944 3 const SubRegion *superR = nullptr; // We need to know the type of the pointer in order to add an integer to it. // Depending on the type, different amount of bytes is added. d951 1 a951 1 superR = cast(elemReg->getSuperRegion()); d957 1 a957 5 superR = cast(region); // TODO: Is this actually reliable? Maybe improving our MemRegion // hierarchy to provide typed regions for all non-void pointers would be // better. For instance, we cannot extend this towards LocAsInteger // operations, where result type of the expression is integer. a986 71 SVal SimpleSValBuilder::simplifySVal(ProgramStateRef State, SVal V) { // For now, this function tries to constant-fold symbols inside a // nonloc::SymbolVal, and does nothing else. More simplifications should // be possible, such as constant-folding an index in an ElementRegion. class Simplifier : public FullSValVisitor { ProgramStateRef State; SValBuilder &SVB; public: Simplifier(ProgramStateRef State) : State(State), SVB(State->getStateManager().getSValBuilder()) {} SVal VisitSymbolData(const SymbolData *S) { if (const llvm::APSInt *I = SVB.getKnownValue(State, nonloc::SymbolVal(S))) return Loc::isLocType(S->getType()) ? (SVal)SVB.makeIntLocVal(*I) : (SVal)SVB.makeIntVal(*I); return nonloc::SymbolVal(S); } // TODO: Support SymbolCast. Support IntSymExpr when/if we actually // start producing them. SVal VisitSymIntExpr(const SymIntExpr *S) { SVal LHS = Visit(S->getLHS()); SVal RHS; // By looking at the APSInt in the right-hand side of S, we cannot // figure out if it should be treated as a Loc or as a NonLoc. // So make our guess by recalling that we cannot multiply pointers // or compare a pointer to an integer. if (Loc::isLocType(S->getLHS()->getType()) && BinaryOperator::isComparisonOp(S->getOpcode())) { // The usual conversion of $sym to &SymRegion{$sym}, as they have // the same meaning for Loc-type symbols, but the latter form // is preferred in SVal computations for being Loc itself. if (SymbolRef Sym = LHS.getAsSymbol()) { assert(Loc::isLocType(Sym->getType())); LHS = SVB.makeLoc(Sym); } RHS = SVB.makeIntLocVal(S->getRHS()); } else { RHS = SVB.makeIntVal(S->getRHS()); } return SVB.evalBinOp(State, S->getOpcode(), LHS, RHS, S->getType()); } SVal VisitSymSymExpr(const SymSymExpr *S) { SVal LHS = Visit(S->getLHS()); SVal RHS = Visit(S->getRHS()); return SVB.evalBinOp(State, S->getOpcode(), LHS, RHS, S->getType()); } SVal VisitSymExpr(SymbolRef S) { return nonloc::SymbolVal(S); } SVal VisitMemRegion(const MemRegion *R) { return loc::MemRegionVal(R); } SVal VisitNonLocSymbolVal(nonloc::SymbolVal V) { // Simplification is much more costly than computing complexity. // For high complexity, it may be not worth it. if (V.getSymbol()->computeComplexity() > 100) return V; return Visit(V.getSymbol()); } SVal VisitSVal(SVal V) { return V; } }; return Simplifier(State).Visit(V); } @ 1.1.1.6.4.1 log @Sync with HEAD @ text @a14 1 #include "clang/StaticAnalyzer/Core/PathSensitive/AnalysisManager.h" a16 1 #include "clang/StaticAnalyzer/Core/PathSensitive/SubEngine.h" d160 1 a160 2 return makeNonLoc(SymR->getSymbol(), BO_NE, BasicVals.getZeroWithPtrWidth(), castTy); a309 188 // See if Sym is known to be a relation Rel with Bound. static bool isInRelation(BinaryOperator::Opcode Rel, SymbolRef Sym, llvm::APSInt Bound, ProgramStateRef State) { SValBuilder &SVB = State->getStateManager().getSValBuilder(); SVal Result = SVB.evalBinOpNN(State, Rel, nonloc::SymbolVal(Sym), nonloc::ConcreteInt(Bound), SVB.getConditionType()); if (auto DV = Result.getAs()) { return !State->assume(*DV, false); } return false; } // See if Sym is known to be within [min/4, max/4], where min and max // are the bounds of the symbol's integral type. With such symbols, // some manipulations can be performed without the risk of overflow. // assume() doesn't cause infinite recursion because we should be dealing // with simpler symbols on every recursive call. static bool isWithinConstantOverflowBounds(SymbolRef Sym, ProgramStateRef State) { SValBuilder &SVB = State->getStateManager().getSValBuilder(); BasicValueFactory &BV = SVB.getBasicValueFactory(); QualType T = Sym->getType(); assert(T->isSignedIntegerOrEnumerationType() && "This only works with signed integers!"); APSIntType AT = BV.getAPSIntType(T); llvm::APSInt Max = AT.getMaxValue() / AT.getValue(4), Min = -Max; return isInRelation(BO_LE, Sym, Max, State) && isInRelation(BO_GE, Sym, Min, State); } // Same for the concrete integers: see if I is within [min/4, max/4]. static bool isWithinConstantOverflowBounds(llvm::APSInt I) { APSIntType AT(I); assert(!AT.isUnsigned() && "This only works with signed integers!"); llvm::APSInt Max = AT.getMaxValue() / AT.getValue(4), Min = -Max; return (I <= Max) && (I >= -Max); } static std::pair decomposeSymbol(SymbolRef Sym, BasicValueFactory &BV) { if (const auto *SymInt = dyn_cast(Sym)) if (BinaryOperator::isAdditiveOp(SymInt->getOpcode())) return std::make_pair(SymInt->getLHS(), (SymInt->getOpcode() == BO_Add) ? (SymInt->getRHS()) : (-SymInt->getRHS())); // Fail to decompose: "reduce" the problem to the "$x + 0" case. return std::make_pair(Sym, BV.getValue(0, Sym->getType())); } // Simplify "(LSym + LInt) Op (RSym + RInt)" assuming all values are of the // same signed integral type and no overflows occur (which should be checked // by the caller). static NonLoc doRearrangeUnchecked(ProgramStateRef State, BinaryOperator::Opcode Op, SymbolRef LSym, llvm::APSInt LInt, SymbolRef RSym, llvm::APSInt RInt) { SValBuilder &SVB = State->getStateManager().getSValBuilder(); BasicValueFactory &BV = SVB.getBasicValueFactory(); SymbolManager &SymMgr = SVB.getSymbolManager(); QualType SymTy = LSym->getType(); assert(SymTy == RSym->getType() && "Symbols are not of the same type!"); assert(APSIntType(LInt) == BV.getAPSIntType(SymTy) && "Integers are not of the same type as symbols!"); assert(APSIntType(RInt) == BV.getAPSIntType(SymTy) && "Integers are not of the same type as symbols!"); QualType ResultTy; if (BinaryOperator::isComparisonOp(Op)) ResultTy = SVB.getConditionType(); else if (BinaryOperator::isAdditiveOp(Op)) ResultTy = SymTy; else llvm_unreachable("Operation not suitable for unchecked rearrangement!"); // FIXME: Can we use assume() without getting into an infinite recursion? if (LSym == RSym) return SVB.evalBinOpNN(State, Op, nonloc::ConcreteInt(LInt), nonloc::ConcreteInt(RInt), ResultTy) .castAs(); SymbolRef ResultSym = nullptr; BinaryOperator::Opcode ResultOp; llvm::APSInt ResultInt; if (BinaryOperator::isComparisonOp(Op)) { // Prefer comparing to a non-negative number. // FIXME: Maybe it'd be better to have consistency in // "$x - $y" vs. "$y - $x" because those are solver's keys. if (LInt > RInt) { ResultSym = SymMgr.getSymSymExpr(RSym, BO_Sub, LSym, SymTy); ResultOp = BinaryOperator::reverseComparisonOp(Op); ResultInt = LInt - RInt; // Opposite order! } else { ResultSym = SymMgr.getSymSymExpr(LSym, BO_Sub, RSym, SymTy); ResultOp = Op; ResultInt = RInt - LInt; // Opposite order! } } else { ResultSym = SymMgr.getSymSymExpr(LSym, Op, RSym, SymTy); ResultInt = (Op == BO_Add) ? (LInt + RInt) : (LInt - RInt); ResultOp = BO_Add; // Bring back the cosmetic difference. if (ResultInt < 0) { ResultInt = -ResultInt; ResultOp = BO_Sub; } else if (ResultInt == 0) { // Shortcut: Simplify "$x + 0" to "$x". return nonloc::SymbolVal(ResultSym); } } const llvm::APSInt &PersistentResultInt = BV.getValue(ResultInt); return nonloc::SymbolVal( SymMgr.getSymIntExpr(ResultSym, ResultOp, PersistentResultInt, ResultTy)); } // Rearrange if symbol type matches the result type and if the operator is a // comparison operator, both symbol and constant must be within constant // overflow bounds. static bool shouldRearrange(ProgramStateRef State, BinaryOperator::Opcode Op, SymbolRef Sym, llvm::APSInt Int, QualType Ty) { return Sym->getType() == Ty && (!BinaryOperator::isComparisonOp(Op) || (isWithinConstantOverflowBounds(Sym, State) && isWithinConstantOverflowBounds(Int))); } static Optional tryRearrange(ProgramStateRef State, BinaryOperator::Opcode Op, NonLoc Lhs, NonLoc Rhs, QualType ResultTy) { ProgramStateManager &StateMgr = State->getStateManager(); SValBuilder &SVB = StateMgr.getSValBuilder(); // We expect everything to be of the same type - this type. QualType SingleTy; auto &Opts = StateMgr.getOwningEngine()->getAnalysisManager().getAnalyzerOptions(); SymbolRef LSym = Lhs.getAsSymbol(); if (!LSym) return None; // Always rearrange additive operations but rearrange comparisons only if // option is set. if (BinaryOperator::isComparisonOp(Op) && Opts.shouldAggressivelySimplifyRelationalComparison()) { SingleTy = LSym->getType(); if (ResultTy != SVB.getConditionType()) return None; // Initialize SingleTy later with a symbol's type. } else if (BinaryOperator::isAdditiveOp(Op)) { SingleTy = ResultTy; if (LSym->getType() != SingleTy) return None; // Substracting unsigned integers is a nightmare. if (!SingleTy->isSignedIntegerOrEnumerationType()) return None; } else { // Don't rearrange other operations. return None; } assert(!SingleTy.isNull() && "We should have figured out the type by now!"); SymbolRef RSym = Rhs.getAsSymbol(); if (!RSym || RSym->getType() != SingleTy) return None; BasicValueFactory &BV = State->getBasicVals(); llvm::APSInt LInt, RInt; std::tie(LSym, LInt) = decomposeSymbol(LSym, BV); std::tie(RSym, RInt) = decomposeSymbol(RSym, BV); if (!shouldRearrange(State, Op, LSym, LInt, SingleTy) || !shouldRearrange(State, Op, RSym, RInt, SingleTy)) return None; // We know that no overflows can occur anymore. return doRearrangeUnchecked(State, Op, LSym, LInt, RSym, RInt); } a362 4 // FIXME: at the moment the implementation // of modeling "pointers as integers" is not complete. if (!BinaryOperator::isComparisonOp(op)) return UnknownVal(); a366 4 // FIXME: at the moment the implementation // of modeling "pointers as integers" is not complete. if (!BinaryOperator::isComparisonOp(op)) return UnknownVal(); a553 3 if (Optional V = tryRearrange(state, op, lhs, rhs, resultTy)) return *V; d674 1 a674 1 if (!BinaryOperator::isComparisonOp(op) || op == BO_Cmp) d721 3 a723 5 if (SymbolRef lSym = lhs.getAsLocSymbol(true)) { if (BinaryOperator::isComparisonOp(op)) return MakeSymIntVal(lSym, op, rInt->getValue(), resultTy); return UnknownVal(); } a914 4 // Perserve the null pointer so that it can be found by the DerefChecker. if (lhs.isZeroConstant()) return lhs; a929 2 QualType pointeeType = resultTy->getPointeeType(); Multiplicand = getContext().getTypeSizeInChars(pointeeType).getQuantity(); a974 6 // Represent arithmetic on void pointers as arithmetic on char pointers. // It is fine when a TypedValueRegion of char value type represents // a void pointer. Note that arithmetic on void pointers is a GCC extension. if (elementType->isVoidType()) elementType = getContext().CharTy; a1009 10 // Cache results for the lifetime of the Simplifier. Results change every // time new constraints are added to the program state, which is the whole // point of simplifying, and for that very reason it's pointless to maintain // the same cache for the duration of the whole analysis. llvm::DenseMap Cached; static bool isUnchanged(SymbolRef Sym, SVal Val) { return Sym == Val.getAsSymbol(); } d1016 1 a1016 1 SVB.getKnownValue(State, SVB.makeSymbolVal(S))) d1019 1 a1019 1 return SVB.makeSymbolVal(S); a1025 4 auto I = Cached.find(S); if (I != Cached.end()) return I->second; a1026 5 if (isUnchanged(S->getLHS(), LHS)) { SVal V = SVB.makeSymbolVal(S); Cached[S] = V; return V; } d1045 1 a1045 4 SVal V = SVB.evalBinOp(State, S->getOpcode(), LHS, RHS, S->getType()); Cached[S] = V; return V; a1048 4 auto I = Cached.find(S); if (I != Cached.end()) return I->second; d1051 1 a1051 8 if (isUnchanged(S->getLHS(), LHS) && isUnchanged(S->getRHS(), RHS)) { SVal V = SVB.makeSymbolVal(S); Cached[S] = V; return V; } SVal V = SVB.evalBinOp(State, S->getOpcode(), LHS, RHS, S->getType()); Cached[S] = V; return V; d1061 2 d1069 1 a1069 10 // A crude way of preventing this function from calling itself from evalBinOp. static bool isReentering = false; if (isReentering) return V; isReentering = true; SVal SimplifiedV = Simplifier(State).Visit(V); isReentering = false; return SimplifiedV; @ 1.1.1.6.4.2 log @Mostly merge changes from HEAD upto 20200411 @ text @@ 1.1.1.6.2.1 log @Sync with HEAD @ text @a14 1 #include "clang/StaticAnalyzer/Core/PathSensitive/AnalysisManager.h" a16 1 #include "clang/StaticAnalyzer/Core/PathSensitive/SubEngine.h" d160 1 a160 2 return makeNonLoc(SymR->getSymbol(), BO_NE, BasicVals.getZeroWithPtrWidth(), castTy); a309 188 // See if Sym is known to be a relation Rel with Bound. static bool isInRelation(BinaryOperator::Opcode Rel, SymbolRef Sym, llvm::APSInt Bound, ProgramStateRef State) { SValBuilder &SVB = State->getStateManager().getSValBuilder(); SVal Result = SVB.evalBinOpNN(State, Rel, nonloc::SymbolVal(Sym), nonloc::ConcreteInt(Bound), SVB.getConditionType()); if (auto DV = Result.getAs()) { return !State->assume(*DV, false); } return false; } // See if Sym is known to be within [min/4, max/4], where min and max // are the bounds of the symbol's integral type. With such symbols, // some manipulations can be performed without the risk of overflow. // assume() doesn't cause infinite recursion because we should be dealing // with simpler symbols on every recursive call. static bool isWithinConstantOverflowBounds(SymbolRef Sym, ProgramStateRef State) { SValBuilder &SVB = State->getStateManager().getSValBuilder(); BasicValueFactory &BV = SVB.getBasicValueFactory(); QualType T = Sym->getType(); assert(T->isSignedIntegerOrEnumerationType() && "This only works with signed integers!"); APSIntType AT = BV.getAPSIntType(T); llvm::APSInt Max = AT.getMaxValue() / AT.getValue(4), Min = -Max; return isInRelation(BO_LE, Sym, Max, State) && isInRelation(BO_GE, Sym, Min, State); } // Same for the concrete integers: see if I is within [min/4, max/4]. static bool isWithinConstantOverflowBounds(llvm::APSInt I) { APSIntType AT(I); assert(!AT.isUnsigned() && "This only works with signed integers!"); llvm::APSInt Max = AT.getMaxValue() / AT.getValue(4), Min = -Max; return (I <= Max) && (I >= -Max); } static std::pair decomposeSymbol(SymbolRef Sym, BasicValueFactory &BV) { if (const auto *SymInt = dyn_cast(Sym)) if (BinaryOperator::isAdditiveOp(SymInt->getOpcode())) return std::make_pair(SymInt->getLHS(), (SymInt->getOpcode() == BO_Add) ? (SymInt->getRHS()) : (-SymInt->getRHS())); // Fail to decompose: "reduce" the problem to the "$x + 0" case. return std::make_pair(Sym, BV.getValue(0, Sym->getType())); } // Simplify "(LSym + LInt) Op (RSym + RInt)" assuming all values are of the // same signed integral type and no overflows occur (which should be checked // by the caller). static NonLoc doRearrangeUnchecked(ProgramStateRef State, BinaryOperator::Opcode Op, SymbolRef LSym, llvm::APSInt LInt, SymbolRef RSym, llvm::APSInt RInt) { SValBuilder &SVB = State->getStateManager().getSValBuilder(); BasicValueFactory &BV = SVB.getBasicValueFactory(); SymbolManager &SymMgr = SVB.getSymbolManager(); QualType SymTy = LSym->getType(); assert(SymTy == RSym->getType() && "Symbols are not of the same type!"); assert(APSIntType(LInt) == BV.getAPSIntType(SymTy) && "Integers are not of the same type as symbols!"); assert(APSIntType(RInt) == BV.getAPSIntType(SymTy) && "Integers are not of the same type as symbols!"); QualType ResultTy; if (BinaryOperator::isComparisonOp(Op)) ResultTy = SVB.getConditionType(); else if (BinaryOperator::isAdditiveOp(Op)) ResultTy = SymTy; else llvm_unreachable("Operation not suitable for unchecked rearrangement!"); // FIXME: Can we use assume() without getting into an infinite recursion? if (LSym == RSym) return SVB.evalBinOpNN(State, Op, nonloc::ConcreteInt(LInt), nonloc::ConcreteInt(RInt), ResultTy) .castAs(); SymbolRef ResultSym = nullptr; BinaryOperator::Opcode ResultOp; llvm::APSInt ResultInt; if (BinaryOperator::isComparisonOp(Op)) { // Prefer comparing to a non-negative number. // FIXME: Maybe it'd be better to have consistency in // "$x - $y" vs. "$y - $x" because those are solver's keys. if (LInt > RInt) { ResultSym = SymMgr.getSymSymExpr(RSym, BO_Sub, LSym, SymTy); ResultOp = BinaryOperator::reverseComparisonOp(Op); ResultInt = LInt - RInt; // Opposite order! } else { ResultSym = SymMgr.getSymSymExpr(LSym, BO_Sub, RSym, SymTy); ResultOp = Op; ResultInt = RInt - LInt; // Opposite order! } } else { ResultSym = SymMgr.getSymSymExpr(LSym, Op, RSym, SymTy); ResultInt = (Op == BO_Add) ? (LInt + RInt) : (LInt - RInt); ResultOp = BO_Add; // Bring back the cosmetic difference. if (ResultInt < 0) { ResultInt = -ResultInt; ResultOp = BO_Sub; } else if (ResultInt == 0) { // Shortcut: Simplify "$x + 0" to "$x". return nonloc::SymbolVal(ResultSym); } } const llvm::APSInt &PersistentResultInt = BV.getValue(ResultInt); return nonloc::SymbolVal( SymMgr.getSymIntExpr(ResultSym, ResultOp, PersistentResultInt, ResultTy)); } // Rearrange if symbol type matches the result type and if the operator is a // comparison operator, both symbol and constant must be within constant // overflow bounds. static bool shouldRearrange(ProgramStateRef State, BinaryOperator::Opcode Op, SymbolRef Sym, llvm::APSInt Int, QualType Ty) { return Sym->getType() == Ty && (!BinaryOperator::isComparisonOp(Op) || (isWithinConstantOverflowBounds(Sym, State) && isWithinConstantOverflowBounds(Int))); } static Optional tryRearrange(ProgramStateRef State, BinaryOperator::Opcode Op, NonLoc Lhs, NonLoc Rhs, QualType ResultTy) { ProgramStateManager &StateMgr = State->getStateManager(); SValBuilder &SVB = StateMgr.getSValBuilder(); // We expect everything to be of the same type - this type. QualType SingleTy; auto &Opts = StateMgr.getOwningEngine()->getAnalysisManager().getAnalyzerOptions(); SymbolRef LSym = Lhs.getAsSymbol(); if (!LSym) return None; // Always rearrange additive operations but rearrange comparisons only if // option is set. if (BinaryOperator::isComparisonOp(Op) && Opts.shouldAggressivelySimplifyRelationalComparison()) { SingleTy = LSym->getType(); if (ResultTy != SVB.getConditionType()) return None; // Initialize SingleTy later with a symbol's type. } else if (BinaryOperator::isAdditiveOp(Op)) { SingleTy = ResultTy; if (LSym->getType() != SingleTy) return None; // Substracting unsigned integers is a nightmare. if (!SingleTy->isSignedIntegerOrEnumerationType()) return None; } else { // Don't rearrange other operations. return None; } assert(!SingleTy.isNull() && "We should have figured out the type by now!"); SymbolRef RSym = Rhs.getAsSymbol(); if (!RSym || RSym->getType() != SingleTy) return None; BasicValueFactory &BV = State->getBasicVals(); llvm::APSInt LInt, RInt; std::tie(LSym, LInt) = decomposeSymbol(LSym, BV); std::tie(RSym, RInt) = decomposeSymbol(RSym, BV); if (!shouldRearrange(State, Op, LSym, LInt, SingleTy) || !shouldRearrange(State, Op, RSym, RInt, SingleTy)) return None; // We know that no overflows can occur anymore. return doRearrangeUnchecked(State, Op, LSym, LInt, RSym, RInt); } a362 4 // FIXME: at the moment the implementation // of modeling "pointers as integers" is not complete. if (!BinaryOperator::isComparisonOp(op)) return UnknownVal(); a366 4 // FIXME: at the moment the implementation // of modeling "pointers as integers" is not complete. if (!BinaryOperator::isComparisonOp(op)) return UnknownVal(); a553 3 if (Optional V = tryRearrange(state, op, lhs, rhs, resultTy)) return *V; d674 1 a674 1 if (!BinaryOperator::isComparisonOp(op) || op == BO_Cmp) d721 3 a723 5 if (SymbolRef lSym = lhs.getAsLocSymbol(true)) { if (BinaryOperator::isComparisonOp(op)) return MakeSymIntVal(lSym, op, rInt->getValue(), resultTy); return UnknownVal(); } a914 4 // Perserve the null pointer so that it can be found by the DerefChecker. if (lhs.isZeroConstant()) return lhs; a929 2 QualType pointeeType = resultTy->getPointeeType(); Multiplicand = getContext().getTypeSizeInChars(pointeeType).getQuantity(); a974 6 // Represent arithmetic on void pointers as arithmetic on char pointers. // It is fine when a TypedValueRegion of char value type represents // a void pointer. Note that arithmetic on void pointers is a GCC extension. if (elementType->isVoidType()) elementType = getContext().CharTy; a1009 10 // Cache results for the lifetime of the Simplifier. Results change every // time new constraints are added to the program state, which is the whole // point of simplifying, and for that very reason it's pointless to maintain // the same cache for the duration of the whole analysis. llvm::DenseMap Cached; static bool isUnchanged(SymbolRef Sym, SVal Val) { return Sym == Val.getAsSymbol(); } d1016 1 a1016 1 SVB.getKnownValue(State, SVB.makeSymbolVal(S))) d1019 1 a1019 1 return SVB.makeSymbolVal(S); a1025 4 auto I = Cached.find(S); if (I != Cached.end()) return I->second; a1026 5 if (isUnchanged(S->getLHS(), LHS)) { SVal V = SVB.makeSymbolVal(S); Cached[S] = V; return V; } d1045 1 a1045 4 SVal V = SVB.evalBinOp(State, S->getOpcode(), LHS, RHS, S->getType()); Cached[S] = V; return V; a1048 4 auto I = Cached.find(S); if (I != Cached.end()) return I->second; d1051 1 a1051 8 if (isUnchanged(S->getLHS(), LHS) && isUnchanged(S->getRHS(), RHS)) { SVal V = SVB.makeSymbolVal(S); Cached[S] = V; return V; } SVal V = SVB.evalBinOp(State, S->getOpcode(), LHS, RHS, S->getType()); Cached[S] = V; return V; d1061 2 d1069 1 a1069 10 // A crude way of preventing this function from calling itself from evalBinOp. static bool isReentering = false; if (isReentering) return V; isReentering = true; SVal SimplifiedV = Simplifier(State).Visit(V); isReentering = false; return SimplifiedV; @ 1.1.1.7 log @Import clang r337282 from trunk @ text @a14 1 #include "clang/StaticAnalyzer/Core/PathSensitive/AnalysisManager.h" a16 1 #include "clang/StaticAnalyzer/Core/PathSensitive/SubEngine.h" d160 1 a160 2 return makeNonLoc(SymR->getSymbol(), BO_NE, BasicVals.getZeroWithPtrWidth(), castTy); a309 188 // See if Sym is known to be a relation Rel with Bound. static bool isInRelation(BinaryOperator::Opcode Rel, SymbolRef Sym, llvm::APSInt Bound, ProgramStateRef State) { SValBuilder &SVB = State->getStateManager().getSValBuilder(); SVal Result = SVB.evalBinOpNN(State, Rel, nonloc::SymbolVal(Sym), nonloc::ConcreteInt(Bound), SVB.getConditionType()); if (auto DV = Result.getAs()) { return !State->assume(*DV, false); } return false; } // See if Sym is known to be within [min/4, max/4], where min and max // are the bounds of the symbol's integral type. With such symbols, // some manipulations can be performed without the risk of overflow. // assume() doesn't cause infinite recursion because we should be dealing // with simpler symbols on every recursive call. static bool isWithinConstantOverflowBounds(SymbolRef Sym, ProgramStateRef State) { SValBuilder &SVB = State->getStateManager().getSValBuilder(); BasicValueFactory &BV = SVB.getBasicValueFactory(); QualType T = Sym->getType(); assert(T->isSignedIntegerOrEnumerationType() && "This only works with signed integers!"); APSIntType AT = BV.getAPSIntType(T); llvm::APSInt Max = AT.getMaxValue() / AT.getValue(4), Min = -Max; return isInRelation(BO_LE, Sym, Max, State) && isInRelation(BO_GE, Sym, Min, State); } // Same for the concrete integers: see if I is within [min/4, max/4]. static bool isWithinConstantOverflowBounds(llvm::APSInt I) { APSIntType AT(I); assert(!AT.isUnsigned() && "This only works with signed integers!"); llvm::APSInt Max = AT.getMaxValue() / AT.getValue(4), Min = -Max; return (I <= Max) && (I >= -Max); } static std::pair decomposeSymbol(SymbolRef Sym, BasicValueFactory &BV) { if (const auto *SymInt = dyn_cast(Sym)) if (BinaryOperator::isAdditiveOp(SymInt->getOpcode())) return std::make_pair(SymInt->getLHS(), (SymInt->getOpcode() == BO_Add) ? (SymInt->getRHS()) : (-SymInt->getRHS())); // Fail to decompose: "reduce" the problem to the "$x + 0" case. return std::make_pair(Sym, BV.getValue(0, Sym->getType())); } // Simplify "(LSym + LInt) Op (RSym + RInt)" assuming all values are of the // same signed integral type and no overflows occur (which should be checked // by the caller). static NonLoc doRearrangeUnchecked(ProgramStateRef State, BinaryOperator::Opcode Op, SymbolRef LSym, llvm::APSInt LInt, SymbolRef RSym, llvm::APSInt RInt) { SValBuilder &SVB = State->getStateManager().getSValBuilder(); BasicValueFactory &BV = SVB.getBasicValueFactory(); SymbolManager &SymMgr = SVB.getSymbolManager(); QualType SymTy = LSym->getType(); assert(SymTy == RSym->getType() && "Symbols are not of the same type!"); assert(APSIntType(LInt) == BV.getAPSIntType(SymTy) && "Integers are not of the same type as symbols!"); assert(APSIntType(RInt) == BV.getAPSIntType(SymTy) && "Integers are not of the same type as symbols!"); QualType ResultTy; if (BinaryOperator::isComparisonOp(Op)) ResultTy = SVB.getConditionType(); else if (BinaryOperator::isAdditiveOp(Op)) ResultTy = SymTy; else llvm_unreachable("Operation not suitable for unchecked rearrangement!"); // FIXME: Can we use assume() without getting into an infinite recursion? if (LSym == RSym) return SVB.evalBinOpNN(State, Op, nonloc::ConcreteInt(LInt), nonloc::ConcreteInt(RInt), ResultTy) .castAs(); SymbolRef ResultSym = nullptr; BinaryOperator::Opcode ResultOp; llvm::APSInt ResultInt; if (BinaryOperator::isComparisonOp(Op)) { // Prefer comparing to a non-negative number. // FIXME: Maybe it'd be better to have consistency in // "$x - $y" vs. "$y - $x" because those are solver's keys. if (LInt > RInt) { ResultSym = SymMgr.getSymSymExpr(RSym, BO_Sub, LSym, SymTy); ResultOp = BinaryOperator::reverseComparisonOp(Op); ResultInt = LInt - RInt; // Opposite order! } else { ResultSym = SymMgr.getSymSymExpr(LSym, BO_Sub, RSym, SymTy); ResultOp = Op; ResultInt = RInt - LInt; // Opposite order! } } else { ResultSym = SymMgr.getSymSymExpr(LSym, Op, RSym, SymTy); ResultInt = (Op == BO_Add) ? (LInt + RInt) : (LInt - RInt); ResultOp = BO_Add; // Bring back the cosmetic difference. if (ResultInt < 0) { ResultInt = -ResultInt; ResultOp = BO_Sub; } else if (ResultInt == 0) { // Shortcut: Simplify "$x + 0" to "$x". return nonloc::SymbolVal(ResultSym); } } const llvm::APSInt &PersistentResultInt = BV.getValue(ResultInt); return nonloc::SymbolVal( SymMgr.getSymIntExpr(ResultSym, ResultOp, PersistentResultInt, ResultTy)); } // Rearrange if symbol type matches the result type and if the operator is a // comparison operator, both symbol and constant must be within constant // overflow bounds. static bool shouldRearrange(ProgramStateRef State, BinaryOperator::Opcode Op, SymbolRef Sym, llvm::APSInt Int, QualType Ty) { return Sym->getType() == Ty && (!BinaryOperator::isComparisonOp(Op) || (isWithinConstantOverflowBounds(Sym, State) && isWithinConstantOverflowBounds(Int))); } static Optional tryRearrange(ProgramStateRef State, BinaryOperator::Opcode Op, NonLoc Lhs, NonLoc Rhs, QualType ResultTy) { ProgramStateManager &StateMgr = State->getStateManager(); SValBuilder &SVB = StateMgr.getSValBuilder(); // We expect everything to be of the same type - this type. QualType SingleTy; auto &Opts = StateMgr.getOwningEngine()->getAnalysisManager().getAnalyzerOptions(); SymbolRef LSym = Lhs.getAsSymbol(); if (!LSym) return None; // Always rearrange additive operations but rearrange comparisons only if // option is set. if (BinaryOperator::isComparisonOp(Op) && Opts.shouldAggressivelySimplifyRelationalComparison()) { SingleTy = LSym->getType(); if (ResultTy != SVB.getConditionType()) return None; // Initialize SingleTy later with a symbol's type. } else if (BinaryOperator::isAdditiveOp(Op)) { SingleTy = ResultTy; if (LSym->getType() != SingleTy) return None; // Substracting unsigned integers is a nightmare. if (!SingleTy->isSignedIntegerOrEnumerationType()) return None; } else { // Don't rearrange other operations. return None; } assert(!SingleTy.isNull() && "We should have figured out the type by now!"); SymbolRef RSym = Rhs.getAsSymbol(); if (!RSym || RSym->getType() != SingleTy) return None; BasicValueFactory &BV = State->getBasicVals(); llvm::APSInt LInt, RInt; std::tie(LSym, LInt) = decomposeSymbol(LSym, BV); std::tie(RSym, RInt) = decomposeSymbol(RSym, BV); if (!shouldRearrange(State, Op, LSym, LInt, SingleTy) || !shouldRearrange(State, Op, RSym, RInt, SingleTy)) return None; // We know that no overflows can occur anymore. return doRearrangeUnchecked(State, Op, LSym, LInt, RSym, RInt); } a362 4 // FIXME: at the moment the implementation // of modeling "pointers as integers" is not complete. if (!BinaryOperator::isComparisonOp(op)) return UnknownVal(); a366 4 // FIXME: at the moment the implementation // of modeling "pointers as integers" is not complete. if (!BinaryOperator::isComparisonOp(op)) return UnknownVal(); a553 3 if (Optional V = tryRearrange(state, op, lhs, rhs, resultTy)) return *V; d674 1 a674 1 if (!BinaryOperator::isComparisonOp(op) || op == BO_Cmp) d721 3 a723 5 if (SymbolRef lSym = lhs.getAsLocSymbol(true)) { if (BinaryOperator::isComparisonOp(op)) return MakeSymIntVal(lSym, op, rInt->getValue(), resultTy); return UnknownVal(); } a914 4 // Perserve the null pointer so that it can be found by the DerefChecker. if (lhs.isZeroConstant()) return lhs; a929 2 QualType pointeeType = resultTy->getPointeeType(); Multiplicand = getContext().getTypeSizeInChars(pointeeType).getQuantity(); a974 6 // Represent arithmetic on void pointers as arithmetic on char pointers. // It is fine when a TypedValueRegion of char value type represents // a void pointer. Note that arithmetic on void pointers is a GCC extension. if (elementType->isVoidType()) elementType = getContext().CharTy; a1009 10 // Cache results for the lifetime of the Simplifier. Results change every // time new constraints are added to the program state, which is the whole // point of simplifying, and for that very reason it's pointless to maintain // the same cache for the duration of the whole analysis. llvm::DenseMap Cached; static bool isUnchanged(SymbolRef Sym, SVal Val) { return Sym == Val.getAsSymbol(); } d1016 1 a1016 1 SVB.getKnownValue(State, SVB.makeSymbolVal(S))) d1019 1 a1019 1 return SVB.makeSymbolVal(S); a1025 4 auto I = Cached.find(S); if (I != Cached.end()) return I->second; a1026 5 if (isUnchanged(S->getLHS(), LHS)) { SVal V = SVB.makeSymbolVal(S); Cached[S] = V; return V; } d1045 1 a1045 4 SVal V = SVB.evalBinOp(State, S->getOpcode(), LHS, RHS, S->getType()); Cached[S] = V; return V; a1048 4 auto I = Cached.find(S); if (I != Cached.end()) return I->second; d1051 1 a1051 8 if (isUnchanged(S->getLHS(), LHS) && isUnchanged(S->getRHS(), RHS)) { SVal V = SVB.makeSymbolVal(S); Cached[S] = V; return V; } SVal V = SVB.evalBinOp(State, S->getOpcode(), LHS, RHS, S->getType()); Cached[S] = V; return V; d1061 2 d1069 1 a1069 10 // A crude way of preventing this function from calling itself from evalBinOp. static bool isReentering = false; if (isReentering) return V; isReentering = true; SVal SimplifiedV = Simplifier(State).Visit(V); isReentering = false; return SimplifiedV; @ 1.1.1.8 log @Mark old LLVM instance as dead. @ text @@ 1.1.1.3.4.1 log @file SimpleSValBuilder.cpp was added on branch tls-maxphys on 2014-08-19 23:47:32 +0000 @ text @d1 944 @ 1.1.1.3.4.2 log @Rebase to HEAD as of a few days ago. @ text @a0 944 // SimpleSValBuilder.cpp - A basic SValBuilder -----------------------*- C++ -*- // // The LLVM Compiler Infrastructure // // This file is distributed under the University of Illinois Open Source // License. See LICENSE.TXT for details. // //===----------------------------------------------------------------------===// // // This file defines SimpleSValBuilder, a basic implementation of SValBuilder. // //===----------------------------------------------------------------------===// #include "clang/StaticAnalyzer/Core/PathSensitive/SValBuilder.h" #include "clang/StaticAnalyzer/Core/PathSensitive/APSIntType.h" #include "clang/StaticAnalyzer/Core/PathSensitive/ProgramState.h" using namespace clang; using namespace ento; namespace { class SimpleSValBuilder : public SValBuilder { protected: SVal dispatchCast(SVal val, QualType castTy) override; SVal evalCastFromNonLoc(NonLoc val, QualType castTy) override; SVal evalCastFromLoc(Loc val, QualType castTy) override; public: SimpleSValBuilder(llvm::BumpPtrAllocator &alloc, ASTContext &context, ProgramStateManager &stateMgr) : SValBuilder(alloc, context, stateMgr) {} virtual ~SimpleSValBuilder() {} SVal evalMinus(NonLoc val) override; SVal evalComplement(NonLoc val) override; SVal evalBinOpNN(ProgramStateRef state, BinaryOperator::Opcode op, NonLoc lhs, NonLoc rhs, QualType resultTy) override; SVal evalBinOpLL(ProgramStateRef state, BinaryOperator::Opcode op, Loc lhs, Loc rhs, QualType resultTy) override; SVal evalBinOpLN(ProgramStateRef state, BinaryOperator::Opcode op, Loc lhs, NonLoc rhs, QualType resultTy) override; /// getKnownValue - evaluates a given SVal. If the SVal has only one possible /// (integer) value, that value is returned. Otherwise, returns NULL. const llvm::APSInt *getKnownValue(ProgramStateRef state, SVal V) override; SVal MakeSymIntVal(const SymExpr *LHS, BinaryOperator::Opcode op, const llvm::APSInt &RHS, QualType resultTy); }; } // end anonymous namespace SValBuilder *ento::createSimpleSValBuilder(llvm::BumpPtrAllocator &alloc, ASTContext &context, ProgramStateManager &stateMgr) { return new SimpleSValBuilder(alloc, context, stateMgr); } //===----------------------------------------------------------------------===// // Transfer function for Casts. //===----------------------------------------------------------------------===// SVal SimpleSValBuilder::dispatchCast(SVal Val, QualType CastTy) { assert(Val.getAs() || Val.getAs()); return Val.getAs() ? evalCastFromLoc(Val.castAs(), CastTy) : evalCastFromNonLoc(Val.castAs(), CastTy); } SVal SimpleSValBuilder::evalCastFromNonLoc(NonLoc val, QualType castTy) { bool isLocType = Loc::isLocType(castTy); if (Optional LI = val.getAs()) { if (isLocType) return LI->getLoc(); // FIXME: Correctly support promotions/truncations. unsigned castSize = Context.getTypeSize(castTy); if (castSize == LI->getNumBits()) return val; return makeLocAsInteger(LI->getLoc(), castSize); } if (const SymExpr *se = val.getAsSymbolicExpression()) { QualType T = Context.getCanonicalType(se->getType()); // If types are the same or both are integers, ignore the cast. // FIXME: Remove this hack when we support symbolic truncation/extension. // HACK: If both castTy and T are integers, ignore the cast. This is // not a permanent solution. Eventually we want to precisely handle // extension/truncation of symbolic integers. This prevents us from losing // precision when we assign 'x = y' and 'y' is symbolic and x and y are // different integer types. if (haveSameType(T, castTy)) return val; if (!isLocType) return makeNonLoc(se, T, castTy); return UnknownVal(); } // If value is a non-integer constant, produce unknown. if (!val.getAs()) return UnknownVal(); // Handle casts to a boolean type. if (castTy->isBooleanType()) { bool b = val.castAs().getValue().getBoolValue(); return makeTruthVal(b, castTy); } // Only handle casts from integers to integers - if val is an integer constant // being cast to a non-integer type, produce unknown. if (!isLocType && !castTy->isIntegralOrEnumerationType()) return UnknownVal(); llvm::APSInt i = val.castAs().getValue(); BasicVals.getAPSIntType(castTy).apply(i); if (isLocType) return makeIntLocVal(i); else return makeIntVal(i); } SVal SimpleSValBuilder::evalCastFromLoc(Loc val, QualType castTy) { // Casts from pointers -> pointers, just return the lval. // // Casts from pointers -> references, just return the lval. These // can be introduced by the frontend for corner cases, e.g // casting from va_list* to __builtin_va_list&. // if (Loc::isLocType(castTy) || castTy->isReferenceType()) return val; // FIXME: Handle transparent unions where a value can be "transparently" // lifted into a union type. if (castTy->isUnionType()) return UnknownVal(); // Casting a Loc to a bool will almost always be true, // unless this is a weak function or a symbolic region. if (castTy->isBooleanType()) { switch (val.getSubKind()) { case loc::MemRegionKind: { const MemRegion *R = val.castAs().getRegion(); if (const FunctionTextRegion *FTR = dyn_cast(R)) if (const FunctionDecl *FD = dyn_cast(FTR->getDecl())) if (FD->isWeak()) // FIXME: Currently we are using an extent symbol here, // because there are no generic region address metadata // symbols to use, only content metadata. return nonloc::SymbolVal(SymMgr.getExtentSymbol(FTR)); if (const SymbolicRegion *SymR = R->getSymbolicBase()) return nonloc::SymbolVal(SymR->getSymbol()); // FALL-THROUGH } case loc::GotoLabelKind: // Labels and non-symbolic memory regions are always true. return makeTruthVal(true, castTy); } } if (castTy->isIntegralOrEnumerationType()) { unsigned BitWidth = Context.getTypeSize(castTy); if (!val.getAs()) return makeLocAsInteger(val, BitWidth); llvm::APSInt i = val.castAs().getValue(); BasicVals.getAPSIntType(castTy).apply(i); return makeIntVal(i); } // All other cases: return 'UnknownVal'. This includes casting pointers // to floats, which is probably badness it itself, but this is a good // intermediate solution until we do something better. return UnknownVal(); } //===----------------------------------------------------------------------===// // Transfer function for unary operators. //===----------------------------------------------------------------------===// SVal SimpleSValBuilder::evalMinus(NonLoc val) { switch (val.getSubKind()) { case nonloc::ConcreteIntKind: return val.castAs().evalMinus(*this); default: return UnknownVal(); } } SVal SimpleSValBuilder::evalComplement(NonLoc X) { switch (X.getSubKind()) { case nonloc::ConcreteIntKind: return X.castAs().evalComplement(*this); default: return UnknownVal(); } } //===----------------------------------------------------------------------===// // Transfer function for binary operators. //===----------------------------------------------------------------------===// SVal SimpleSValBuilder::MakeSymIntVal(const SymExpr *LHS, BinaryOperator::Opcode op, const llvm::APSInt &RHS, QualType resultTy) { bool isIdempotent = false; // Check for a few special cases with known reductions first. switch (op) { default: // We can't reduce this case; just treat it normally. break; case BO_Mul: // a*0 and a*1 if (RHS == 0) return makeIntVal(0, resultTy); else if (RHS == 1) isIdempotent = true; break; case BO_Div: // a/0 and a/1 if (RHS == 0) // This is also handled elsewhere. return UndefinedVal(); else if (RHS == 1) isIdempotent = true; break; case BO_Rem: // a%0 and a%1 if (RHS == 0) // This is also handled elsewhere. return UndefinedVal(); else if (RHS == 1) return makeIntVal(0, resultTy); break; case BO_Add: case BO_Sub: case BO_Shl: case BO_Shr: case BO_Xor: // a+0, a-0, a<<0, a>>0, a^0 if (RHS == 0) isIdempotent = true; break; case BO_And: // a&0 and a&(~0) if (RHS == 0) return makeIntVal(0, resultTy); else if (RHS.isAllOnesValue()) isIdempotent = true; break; case BO_Or: // a|0 and a|(~0) if (RHS == 0) isIdempotent = true; else if (RHS.isAllOnesValue()) { const llvm::APSInt &Result = BasicVals.Convert(resultTy, RHS); return nonloc::ConcreteInt(Result); } break; } // Idempotent ops (like a*1) can still change the type of an expression. // Wrap the LHS up in a NonLoc again and let evalCastFromNonLoc do the // dirty work. if (isIdempotent) return evalCastFromNonLoc(nonloc::SymbolVal(LHS), resultTy); // If we reach this point, the expression cannot be simplified. // Make a SymbolVal for the entire expression, after converting the RHS. const llvm::APSInt *ConvertedRHS = &RHS; if (BinaryOperator::isComparisonOp(op)) { // We're looking for a type big enough to compare the symbolic value // with the given constant. // FIXME: This is an approximation of Sema::UsualArithmeticConversions. ASTContext &Ctx = getContext(); QualType SymbolType = LHS->getType(); uint64_t ValWidth = RHS.getBitWidth(); uint64_t TypeWidth = Ctx.getTypeSize(SymbolType); if (ValWidth < TypeWidth) { // If the value is too small, extend it. ConvertedRHS = &BasicVals.Convert(SymbolType, RHS); } else if (ValWidth == TypeWidth) { // If the value is signed but the symbol is unsigned, do the comparison // in unsigned space. [C99 6.3.1.8] // (For the opposite case, the value is already unsigned.) if (RHS.isSigned() && !SymbolType->isSignedIntegerOrEnumerationType()) ConvertedRHS = &BasicVals.Convert(SymbolType, RHS); } } else ConvertedRHS = &BasicVals.Convert(resultTy, RHS); return makeNonLoc(LHS, op, *ConvertedRHS, resultTy); } SVal SimpleSValBuilder::evalBinOpNN(ProgramStateRef state, BinaryOperator::Opcode op, NonLoc lhs, NonLoc rhs, QualType resultTy) { NonLoc InputLHS = lhs; NonLoc InputRHS = rhs; // Handle trivial case where left-side and right-side are the same. if (lhs == rhs) switch (op) { default: break; case BO_EQ: case BO_LE: case BO_GE: return makeTruthVal(true, resultTy); case BO_LT: case BO_GT: case BO_NE: return makeTruthVal(false, resultTy); case BO_Xor: case BO_Sub: if (resultTy->isIntegralOrEnumerationType()) return makeIntVal(0, resultTy); return evalCastFromNonLoc(makeIntVal(0, /*Unsigned=*/false), resultTy); case BO_Or: case BO_And: return evalCastFromNonLoc(lhs, resultTy); } while (1) { switch (lhs.getSubKind()) { default: return makeSymExprValNN(state, op, lhs, rhs, resultTy); case nonloc::LocAsIntegerKind: { Loc lhsL = lhs.castAs().getLoc(); switch (rhs.getSubKind()) { case nonloc::LocAsIntegerKind: return evalBinOpLL(state, op, lhsL, rhs.castAs().getLoc(), resultTy); case nonloc::ConcreteIntKind: { // Transform the integer into a location and compare. llvm::APSInt i = rhs.castAs().getValue(); BasicVals.getAPSIntType(Context.VoidPtrTy).apply(i); return evalBinOpLL(state, op, lhsL, makeLoc(i), resultTy); } default: switch (op) { case BO_EQ: return makeTruthVal(false, resultTy); case BO_NE: return makeTruthVal(true, resultTy); default: // This case also handles pointer arithmetic. return makeSymExprValNN(state, op, InputLHS, InputRHS, resultTy); } } } case nonloc::ConcreteIntKind: { llvm::APSInt LHSValue = lhs.castAs().getValue(); // If we're dealing with two known constants, just perform the operation. if (const llvm::APSInt *KnownRHSValue = getKnownValue(state, rhs)) { llvm::APSInt RHSValue = *KnownRHSValue; if (BinaryOperator::isComparisonOp(op)) { // We're looking for a type big enough to compare the two values. // FIXME: This is not correct. char + short will result in a promotion // to int. Unfortunately we have lost types by this point. APSIntType CompareType = std::max(APSIntType(LHSValue), APSIntType(RHSValue)); CompareType.apply(LHSValue); CompareType.apply(RHSValue); } else if (!BinaryOperator::isShiftOp(op)) { APSIntType IntType = BasicVals.getAPSIntType(resultTy); IntType.apply(LHSValue); IntType.apply(RHSValue); } const llvm::APSInt *Result = BasicVals.evalAPSInt(op, LHSValue, RHSValue); if (!Result) return UndefinedVal(); return nonloc::ConcreteInt(*Result); } // Swap the left and right sides and flip the operator if doing so // allows us to better reason about the expression (this is a form // of expression canonicalization). // While we're at it, catch some special cases for non-commutative ops. switch (op) { case BO_LT: case BO_GT: case BO_LE: case BO_GE: op = BinaryOperator::reverseComparisonOp(op); // FALL-THROUGH case BO_EQ: case BO_NE: case BO_Add: case BO_Mul: case BO_And: case BO_Xor: case BO_Or: std::swap(lhs, rhs); continue; case BO_Shr: // (~0)>>a if (LHSValue.isAllOnesValue() && LHSValue.isSigned()) return evalCastFromNonLoc(lhs, resultTy); // FALL-THROUGH case BO_Shl: // 0<>a if (LHSValue == 0) return evalCastFromNonLoc(lhs, resultTy); return makeSymExprValNN(state, op, InputLHS, InputRHS, resultTy); default: return makeSymExprValNN(state, op, InputLHS, InputRHS, resultTy); } } case nonloc::SymbolValKind: { // We only handle LHS as simple symbols or SymIntExprs. SymbolRef Sym = lhs.castAs().getSymbol(); // LHS is a symbolic expression. if (const SymIntExpr *symIntExpr = dyn_cast(Sym)) { // Is this a logical not? (!x is represented as x == 0.) if (op == BO_EQ && rhs.isZeroConstant()) { // We know how to negate certain expressions. Simplify them here. BinaryOperator::Opcode opc = symIntExpr->getOpcode(); switch (opc) { default: // We don't know how to negate this operation. // Just handle it as if it were a normal comparison to 0. break; case BO_LAnd: case BO_LOr: llvm_unreachable("Logical operators handled by branching logic."); case BO_Assign: case BO_MulAssign: case BO_DivAssign: case BO_RemAssign: case BO_AddAssign: case BO_SubAssign: case BO_ShlAssign: case BO_ShrAssign: case BO_AndAssign: case BO_XorAssign: case BO_OrAssign: case BO_Comma: llvm_unreachable("'=' and ',' operators handled by ExprEngine."); case BO_PtrMemD: case BO_PtrMemI: llvm_unreachable("Pointer arithmetic not handled here."); case BO_LT: case BO_GT: case BO_LE: case BO_GE: case BO_EQ: case BO_NE: assert(resultTy->isBooleanType() || resultTy == getConditionType()); assert(symIntExpr->getType()->isBooleanType() || getContext().hasSameUnqualifiedType(symIntExpr->getType(), getConditionType())); // Negate the comparison and make a value. opc = BinaryOperator::negateComparisonOp(opc); return makeNonLoc(symIntExpr->getLHS(), opc, symIntExpr->getRHS(), resultTy); } } // For now, only handle expressions whose RHS is a constant. if (const llvm::APSInt *RHSValue = getKnownValue(state, rhs)) { // If both the LHS and the current expression are additive, // fold their constants and try again. if (BinaryOperator::isAdditiveOp(op)) { BinaryOperator::Opcode lop = symIntExpr->getOpcode(); if (BinaryOperator::isAdditiveOp(lop)) { // Convert the two constants to a common type, then combine them. // resultTy may not be the best type to convert to, but it's // probably the best choice in expressions with mixed type // (such as x+1U+2LL). The rules for implicit conversions should // choose a reasonable type to preserve the expression, and will // at least match how the value is going to be used. APSIntType IntType = BasicVals.getAPSIntType(resultTy); const llvm::APSInt &first = IntType.convert(symIntExpr->getRHS()); const llvm::APSInt &second = IntType.convert(*RHSValue); const llvm::APSInt *newRHS; if (lop == op) newRHS = BasicVals.evalAPSInt(BO_Add, first, second); else newRHS = BasicVals.evalAPSInt(BO_Sub, first, second); assert(newRHS && "Invalid operation despite common type!"); rhs = nonloc::ConcreteInt(*newRHS); lhs = nonloc::SymbolVal(symIntExpr->getLHS()); op = lop; continue; } } // Otherwise, make a SymIntExpr out of the expression. return MakeSymIntVal(symIntExpr, op, *RHSValue, resultTy); } } // Does the symbolic expression simplify to a constant? // If so, "fold" the constant by setting 'lhs' to a ConcreteInt // and try again. ConstraintManager &CMgr = state->getConstraintManager(); if (const llvm::APSInt *Constant = CMgr.getSymVal(state, Sym)) { lhs = nonloc::ConcreteInt(*Constant); continue; } // Is the RHS a constant? if (const llvm::APSInt *RHSValue = getKnownValue(state, rhs)) return MakeSymIntVal(Sym, op, *RHSValue, resultTy); // Give up -- this is not a symbolic expression we can handle. return makeSymExprValNN(state, op, InputLHS, InputRHS, resultTy); } } } } static SVal evalBinOpFieldRegionFieldRegion(const FieldRegion *LeftFR, const FieldRegion *RightFR, BinaryOperator::Opcode op, QualType resultTy, SimpleSValBuilder &SVB) { // Only comparisons are meaningful here! if (!BinaryOperator::isComparisonOp(op)) return UnknownVal(); // Next, see if the two FRs have the same super-region. // FIXME: This doesn't handle casts yet, and simply stripping the casts // doesn't help. if (LeftFR->getSuperRegion() != RightFR->getSuperRegion()) return UnknownVal(); const FieldDecl *LeftFD = LeftFR->getDecl(); const FieldDecl *RightFD = RightFR->getDecl(); const RecordDecl *RD = LeftFD->getParent(); // Make sure the two FRs are from the same kind of record. Just in case! // FIXME: This is probably where inheritance would be a problem. if (RD != RightFD->getParent()) return UnknownVal(); // We know for sure that the two fields are not the same, since that // would have given us the same SVal. if (op == BO_EQ) return SVB.makeTruthVal(false, resultTy); if (op == BO_NE) return SVB.makeTruthVal(true, resultTy); // Iterate through the fields and see which one comes first. // [C99 6.7.2.1.13] "Within a structure object, the non-bit-field // members and the units in which bit-fields reside have addresses that // increase in the order in which they are declared." bool leftFirst = (op == BO_LT || op == BO_LE); for (const auto *I : RD->fields()) { if (I == LeftFD) return SVB.makeTruthVal(leftFirst, resultTy); if (I == RightFD) return SVB.makeTruthVal(!leftFirst, resultTy); } llvm_unreachable("Fields not found in parent record's definition"); } // FIXME: all this logic will change if/when we have MemRegion::getLocation(). SVal SimpleSValBuilder::evalBinOpLL(ProgramStateRef state, BinaryOperator::Opcode op, Loc lhs, Loc rhs, QualType resultTy) { // Only comparisons and subtractions are valid operations on two pointers. // See [C99 6.5.5 through 6.5.14] or [C++0x 5.6 through 5.15]. // However, if a pointer is casted to an integer, evalBinOpNN may end up // calling this function with another operation (PR7527). We don't attempt to // model this for now, but it could be useful, particularly when the // "location" is actually an integer value that's been passed through a void*. if (!(BinaryOperator::isComparisonOp(op) || op == BO_Sub)) return UnknownVal(); // Special cases for when both sides are identical. if (lhs == rhs) { switch (op) { default: llvm_unreachable("Unimplemented operation for two identical values"); case BO_Sub: return makeZeroVal(resultTy); case BO_EQ: case BO_LE: case BO_GE: return makeTruthVal(true, resultTy); case BO_NE: case BO_LT: case BO_GT: return makeTruthVal(false, resultTy); } } switch (lhs.getSubKind()) { default: llvm_unreachable("Ordering not implemented for this Loc."); case loc::GotoLabelKind: // The only thing we know about labels is that they're non-null. if (rhs.isZeroConstant()) { switch (op) { default: break; case BO_Sub: return evalCastFromLoc(lhs, resultTy); case BO_EQ: case BO_LE: case BO_LT: return makeTruthVal(false, resultTy); case BO_NE: case BO_GT: case BO_GE: return makeTruthVal(true, resultTy); } } // There may be two labels for the same location, and a function region may // have the same address as a label at the start of the function (depending // on the ABI). // FIXME: we can probably do a comparison against other MemRegions, though. // FIXME: is there a way to tell if two labels refer to the same location? return UnknownVal(); case loc::ConcreteIntKind: { // If one of the operands is a symbol and the other is a constant, // build an expression for use by the constraint manager. if (SymbolRef rSym = rhs.getAsLocSymbol()) { // We can only build expressions with symbols on the left, // so we need a reversible operator. if (!BinaryOperator::isComparisonOp(op)) return UnknownVal(); const llvm::APSInt &lVal = lhs.castAs().getValue(); op = BinaryOperator::reverseComparisonOp(op); return makeNonLoc(rSym, op, lVal, resultTy); } // If both operands are constants, just perform the operation. if (Optional rInt = rhs.getAs()) { SVal ResultVal = lhs.castAs().evalBinOp(BasicVals, op, *rInt); if (Optional Result = ResultVal.getAs()) return evalCastFromNonLoc(*Result, resultTy); assert(!ResultVal.getAs() && "Loc-Loc ops should not produce Locs"); return UnknownVal(); } // Special case comparisons against NULL. // This must come after the test if the RHS is a symbol, which is used to // build constraints. The address of any non-symbolic region is guaranteed // to be non-NULL, as is any label. assert(rhs.getAs() || rhs.getAs()); if (lhs.isZeroConstant()) { switch (op) { default: break; case BO_EQ: case BO_GT: case BO_GE: return makeTruthVal(false, resultTy); case BO_NE: case BO_LT: case BO_LE: return makeTruthVal(true, resultTy); } } // Comparing an arbitrary integer to a region or label address is // completely unknowable. return UnknownVal(); } case loc::MemRegionKind: { if (Optional rInt = rhs.getAs()) { // If one of the operands is a symbol and the other is a constant, // build an expression for use by the constraint manager. if (SymbolRef lSym = lhs.getAsLocSymbol(true)) return MakeSymIntVal(lSym, op, rInt->getValue(), resultTy); // Special case comparisons to NULL. // This must come after the test if the LHS is a symbol, which is used to // build constraints. The address of any non-symbolic region is guaranteed // to be non-NULL. if (rInt->isZeroConstant()) { if (op == BO_Sub) return evalCastFromLoc(lhs, resultTy); if (BinaryOperator::isComparisonOp(op)) { QualType boolType = getContext().BoolTy; NonLoc l = evalCastFromLoc(lhs, boolType).castAs(); NonLoc r = makeTruthVal(false, boolType).castAs(); return evalBinOpNN(state, op, l, r, resultTy); } } // Comparing a region to an arbitrary integer is completely unknowable. return UnknownVal(); } // Get both values as regions, if possible. const MemRegion *LeftMR = lhs.getAsRegion(); assert(LeftMR && "MemRegionKind SVal doesn't have a region!"); const MemRegion *RightMR = rhs.getAsRegion(); if (!RightMR) // The RHS is probably a label, which in theory could address a region. // FIXME: we can probably make a more useful statement about non-code // regions, though. return UnknownVal(); const MemRegion *LeftBase = LeftMR->getBaseRegion(); const MemRegion *RightBase = RightMR->getBaseRegion(); const MemSpaceRegion *LeftMS = LeftBase->getMemorySpace(); const MemSpaceRegion *RightMS = RightBase->getMemorySpace(); const MemSpaceRegion *UnknownMS = MemMgr.getUnknownRegion(); // If the two regions are from different known memory spaces they cannot be // equal. Also, assume that no symbolic region (whose memory space is // unknown) is on the stack. if (LeftMS != RightMS && ((LeftMS != UnknownMS && RightMS != UnknownMS) || (isa(LeftMS) || isa(RightMS)))) { switch (op) { default: return UnknownVal(); case BO_EQ: return makeTruthVal(false, resultTy); case BO_NE: return makeTruthVal(true, resultTy); } } // If both values wrap regions, see if they're from different base regions. // Note, heap base symbolic regions are assumed to not alias with // each other; for example, we assume that malloc returns different address // on each invocation. if (LeftBase != RightBase && ((!isa(LeftBase) && !isa(RightBase)) || (isa(LeftMS) || isa(RightMS))) ){ switch (op) { default: return UnknownVal(); case BO_EQ: return makeTruthVal(false, resultTy); case BO_NE: return makeTruthVal(true, resultTy); } } // Handle special cases for when both regions are element regions. const ElementRegion *RightER = dyn_cast(RightMR); const ElementRegion *LeftER = dyn_cast(LeftMR); if (RightER && LeftER) { // Next, see if the two ERs have the same super-region and matching types. // FIXME: This should do something useful even if the types don't match, // though if both indexes are constant the RegionRawOffset path will // give the correct answer. if (LeftER->getSuperRegion() == RightER->getSuperRegion() && LeftER->getElementType() == RightER->getElementType()) { // Get the left index and cast it to the correct type. // If the index is unknown or undefined, bail out here. SVal LeftIndexVal = LeftER->getIndex(); Optional LeftIndex = LeftIndexVal.getAs(); if (!LeftIndex) return UnknownVal(); LeftIndexVal = evalCastFromNonLoc(*LeftIndex, ArrayIndexTy); LeftIndex = LeftIndexVal.getAs(); if (!LeftIndex) return UnknownVal(); // Do the same for the right index. SVal RightIndexVal = RightER->getIndex(); Optional RightIndex = RightIndexVal.getAs(); if (!RightIndex) return UnknownVal(); RightIndexVal = evalCastFromNonLoc(*RightIndex, ArrayIndexTy); RightIndex = RightIndexVal.getAs(); if (!RightIndex) return UnknownVal(); // Actually perform the operation. // evalBinOpNN expects the two indexes to already be the right type. return evalBinOpNN(state, op, *LeftIndex, *RightIndex, resultTy); } } // Special handling of the FieldRegions, even with symbolic offsets. const FieldRegion *RightFR = dyn_cast(RightMR); const FieldRegion *LeftFR = dyn_cast(LeftMR); if (RightFR && LeftFR) { SVal R = evalBinOpFieldRegionFieldRegion(LeftFR, RightFR, op, resultTy, *this); if (!R.isUnknown()) return R; } // Compare the regions using the raw offsets. RegionOffset LeftOffset = LeftMR->getAsOffset(); RegionOffset RightOffset = RightMR->getAsOffset(); if (LeftOffset.getRegion() != nullptr && LeftOffset.getRegion() == RightOffset.getRegion() && !LeftOffset.hasSymbolicOffset() && !RightOffset.hasSymbolicOffset()) { int64_t left = LeftOffset.getOffset(); int64_t right = RightOffset.getOffset(); switch (op) { default: return UnknownVal(); case BO_LT: return makeTruthVal(left < right, resultTy); case BO_GT: return makeTruthVal(left > right, resultTy); case BO_LE: return makeTruthVal(left <= right, resultTy); case BO_GE: return makeTruthVal(left >= right, resultTy); case BO_EQ: return makeTruthVal(left == right, resultTy); case BO_NE: return makeTruthVal(left != right, resultTy); } } // At this point we're not going to get a good answer, but we can try // conjuring an expression instead. SymbolRef LHSSym = lhs.getAsLocSymbol(); SymbolRef RHSSym = rhs.getAsLocSymbol(); if (LHSSym && RHSSym) return makeNonLoc(LHSSym, op, RHSSym, resultTy); // If we get here, we have no way of comparing the regions. return UnknownVal(); } } } SVal SimpleSValBuilder::evalBinOpLN(ProgramStateRef state, BinaryOperator::Opcode op, Loc lhs, NonLoc rhs, QualType resultTy) { assert(!BinaryOperator::isComparisonOp(op) && "arguments to comparison ops must be of the same type"); // Special case: rhs is a zero constant. if (rhs.isZeroConstant()) return lhs; // We are dealing with pointer arithmetic. // Handle pointer arithmetic on constant values. if (Optional rhsInt = rhs.getAs()) { if (Optional lhsInt = lhs.getAs()) { const llvm::APSInt &leftI = lhsInt->getValue(); assert(leftI.isUnsigned()); llvm::APSInt rightI(rhsInt->getValue(), /* isUnsigned */ true); // Convert the bitwidth of rightI. This should deal with overflow // since we are dealing with concrete values. rightI = rightI.extOrTrunc(leftI.getBitWidth()); // Offset the increment by the pointer size. llvm::APSInt Multiplicand(rightI.getBitWidth(), /* isUnsigned */ true); rightI *= Multiplicand; // Compute the adjusted pointer. switch (op) { case BO_Add: rightI = leftI + rightI; break; case BO_Sub: rightI = leftI - rightI; break; default: llvm_unreachable("Invalid pointer arithmetic operation"); } return loc::ConcreteInt(getBasicValueFactory().getValue(rightI)); } } // Handle cases where 'lhs' is a region. if (const MemRegion *region = lhs.getAsRegion()) { rhs = convertToArrayIndex(rhs).castAs(); SVal index = UnknownVal(); const MemRegion *superR = nullptr; QualType elementType; if (const ElementRegion *elemReg = dyn_cast(region)) { assert(op == BO_Add || op == BO_Sub); index = evalBinOpNN(state, op, elemReg->getIndex(), rhs, getArrayIndexType()); superR = elemReg->getSuperRegion(); elementType = elemReg->getElementType(); } else if (isa(region)) { superR = region; index = rhs; if (resultTy->isAnyPointerType()) elementType = resultTy->getPointeeType(); } if (Optional indexV = index.getAs()) { return loc::MemRegionVal(MemMgr.getElementRegion(elementType, *indexV, superR, getContext())); } } return UnknownVal(); } const llvm::APSInt *SimpleSValBuilder::getKnownValue(ProgramStateRef state, SVal V) { if (V.isUnknownOrUndef()) return nullptr; if (Optional X = V.getAs()) return &X->getValue(); if (Optional X = V.getAs()) return &X->getValue(); if (SymbolRef Sym = V.getAsSymbol()) return state->getConstraintManager().getSymVal(state, Sym); // FIXME: Add support for SymExprs. return nullptr; } @ 1.1.1.2.4.1 log @file SimpleSValBuilder.cpp was added on branch yamt-pagecache on 2014-05-22 16:18:31 +0000 @ text @d1 945 @ 1.1.1.2.4.2 log @sync with head. for a reference, the tree before this commit was tagged as yamt-pagecache-tag8. this commit was splitted into small chunks to avoid a limitation of cvs. ("Protocol error: too many arguments") @ text @a0 945 // SimpleSValBuilder.cpp - A basic SValBuilder -----------------------*- C++ -*- // // The LLVM Compiler Infrastructure // // This file is distributed under the University of Illinois Open Source // License. See LICENSE.TXT for details. // //===----------------------------------------------------------------------===// // // This file defines SimpleSValBuilder, a basic implementation of SValBuilder. // //===----------------------------------------------------------------------===// #include "clang/StaticAnalyzer/Core/PathSensitive/SValBuilder.h" #include "clang/StaticAnalyzer/Core/PathSensitive/APSIntType.h" #include "clang/StaticAnalyzer/Core/PathSensitive/ProgramState.h" using namespace clang; using namespace ento; namespace { class SimpleSValBuilder : public SValBuilder { protected: virtual SVal dispatchCast(SVal val, QualType castTy); virtual SVal evalCastFromNonLoc(NonLoc val, QualType castTy); virtual SVal evalCastFromLoc(Loc val, QualType castTy); public: SimpleSValBuilder(llvm::BumpPtrAllocator &alloc, ASTContext &context, ProgramStateManager &stateMgr) : SValBuilder(alloc, context, stateMgr) {} virtual ~SimpleSValBuilder() {} virtual SVal evalMinus(NonLoc val); virtual SVal evalComplement(NonLoc val); virtual SVal evalBinOpNN(ProgramStateRef state, BinaryOperator::Opcode op, NonLoc lhs, NonLoc rhs, QualType resultTy); virtual SVal evalBinOpLL(ProgramStateRef state, BinaryOperator::Opcode op, Loc lhs, Loc rhs, QualType resultTy); virtual SVal evalBinOpLN(ProgramStateRef state, BinaryOperator::Opcode op, Loc lhs, NonLoc rhs, QualType resultTy); /// getKnownValue - evaluates a given SVal. If the SVal has only one possible /// (integer) value, that value is returned. Otherwise, returns NULL. virtual const llvm::APSInt *getKnownValue(ProgramStateRef state, SVal V); SVal MakeSymIntVal(const SymExpr *LHS, BinaryOperator::Opcode op, const llvm::APSInt &RHS, QualType resultTy); }; } // end anonymous namespace SValBuilder *ento::createSimpleSValBuilder(llvm::BumpPtrAllocator &alloc, ASTContext &context, ProgramStateManager &stateMgr) { return new SimpleSValBuilder(alloc, context, stateMgr); } //===----------------------------------------------------------------------===// // Transfer function for Casts. //===----------------------------------------------------------------------===// SVal SimpleSValBuilder::dispatchCast(SVal Val, QualType CastTy) { assert(Val.getAs() || Val.getAs()); return Val.getAs() ? evalCastFromLoc(Val.castAs(), CastTy) : evalCastFromNonLoc(Val.castAs(), CastTy); } SVal SimpleSValBuilder::evalCastFromNonLoc(NonLoc val, QualType castTy) { bool isLocType = Loc::isLocType(castTy); if (Optional LI = val.getAs()) { if (isLocType) return LI->getLoc(); // FIXME: Correctly support promotions/truncations. unsigned castSize = Context.getTypeSize(castTy); if (castSize == LI->getNumBits()) return val; return makeLocAsInteger(LI->getLoc(), castSize); } if (const SymExpr *se = val.getAsSymbolicExpression()) { QualType T = Context.getCanonicalType(se->getType()); // If types are the same or both are integers, ignore the cast. // FIXME: Remove this hack when we support symbolic truncation/extension. // HACK: If both castTy and T are integers, ignore the cast. This is // not a permanent solution. Eventually we want to precisely handle // extension/truncation of symbolic integers. This prevents us from losing // precision when we assign 'x = y' and 'y' is symbolic and x and y are // different integer types. if (haveSameType(T, castTy)) return val; if (!isLocType) return makeNonLoc(se, T, castTy); return UnknownVal(); } // If value is a non-integer constant, produce unknown. if (!val.getAs()) return UnknownVal(); // Handle casts to a boolean type. if (castTy->isBooleanType()) { bool b = val.castAs().getValue().getBoolValue(); return makeTruthVal(b, castTy); } // Only handle casts from integers to integers - if val is an integer constant // being cast to a non-integer type, produce unknown. if (!isLocType && !castTy->isIntegralOrEnumerationType()) return UnknownVal(); llvm::APSInt i = val.castAs().getValue(); BasicVals.getAPSIntType(castTy).apply(i); if (isLocType) return makeIntLocVal(i); else return makeIntVal(i); } SVal SimpleSValBuilder::evalCastFromLoc(Loc val, QualType castTy) { // Casts from pointers -> pointers, just return the lval. // // Casts from pointers -> references, just return the lval. These // can be introduced by the frontend for corner cases, e.g // casting from va_list* to __builtin_va_list&. // if (Loc::isLocType(castTy) || castTy->isReferenceType()) return val; // FIXME: Handle transparent unions where a value can be "transparently" // lifted into a union type. if (castTy->isUnionType()) return UnknownVal(); // Casting a Loc to a bool will almost always be true, // unless this is a weak function or a symbolic region. if (castTy->isBooleanType()) { switch (val.getSubKind()) { case loc::MemRegionKind: { const MemRegion *R = val.castAs().getRegion(); if (const FunctionTextRegion *FTR = dyn_cast(R)) if (const FunctionDecl *FD = dyn_cast(FTR->getDecl())) if (FD->isWeak()) // FIXME: Currently we are using an extent symbol here, // because there are no generic region address metadata // symbols to use, only content metadata. return nonloc::SymbolVal(SymMgr.getExtentSymbol(FTR)); if (const SymbolicRegion *SymR = R->getSymbolicBase()) return nonloc::SymbolVal(SymR->getSymbol()); // FALL-THROUGH } case loc::GotoLabelKind: // Labels and non-symbolic memory regions are always true. return makeTruthVal(true, castTy); } } if (castTy->isIntegralOrEnumerationType()) { unsigned BitWidth = Context.getTypeSize(castTy); if (!val.getAs()) return makeLocAsInteger(val, BitWidth); llvm::APSInt i = val.castAs().getValue(); BasicVals.getAPSIntType(castTy).apply(i); return makeIntVal(i); } // All other cases: return 'UnknownVal'. This includes casting pointers // to floats, which is probably badness it itself, but this is a good // intermediate solution until we do something better. return UnknownVal(); } //===----------------------------------------------------------------------===// // Transfer function for unary operators. //===----------------------------------------------------------------------===// SVal SimpleSValBuilder::evalMinus(NonLoc val) { switch (val.getSubKind()) { case nonloc::ConcreteIntKind: return val.castAs().evalMinus(*this); default: return UnknownVal(); } } SVal SimpleSValBuilder::evalComplement(NonLoc X) { switch (X.getSubKind()) { case nonloc::ConcreteIntKind: return X.castAs().evalComplement(*this); default: return UnknownVal(); } } //===----------------------------------------------------------------------===// // Transfer function for binary operators. //===----------------------------------------------------------------------===// SVal SimpleSValBuilder::MakeSymIntVal(const SymExpr *LHS, BinaryOperator::Opcode op, const llvm::APSInt &RHS, QualType resultTy) { bool isIdempotent = false; // Check for a few special cases with known reductions first. switch (op) { default: // We can't reduce this case; just treat it normally. break; case BO_Mul: // a*0 and a*1 if (RHS == 0) return makeIntVal(0, resultTy); else if (RHS == 1) isIdempotent = true; break; case BO_Div: // a/0 and a/1 if (RHS == 0) // This is also handled elsewhere. return UndefinedVal(); else if (RHS == 1) isIdempotent = true; break; case BO_Rem: // a%0 and a%1 if (RHS == 0) // This is also handled elsewhere. return UndefinedVal(); else if (RHS == 1) return makeIntVal(0, resultTy); break; case BO_Add: case BO_Sub: case BO_Shl: case BO_Shr: case BO_Xor: // a+0, a-0, a<<0, a>>0, a^0 if (RHS == 0) isIdempotent = true; break; case BO_And: // a&0 and a&(~0) if (RHS == 0) return makeIntVal(0, resultTy); else if (RHS.isAllOnesValue()) isIdempotent = true; break; case BO_Or: // a|0 and a|(~0) if (RHS == 0) isIdempotent = true; else if (RHS.isAllOnesValue()) { const llvm::APSInt &Result = BasicVals.Convert(resultTy, RHS); return nonloc::ConcreteInt(Result); } break; } // Idempotent ops (like a*1) can still change the type of an expression. // Wrap the LHS up in a NonLoc again and let evalCastFromNonLoc do the // dirty work. if (isIdempotent) return evalCastFromNonLoc(nonloc::SymbolVal(LHS), resultTy); // If we reach this point, the expression cannot be simplified. // Make a SymbolVal for the entire expression, after converting the RHS. const llvm::APSInt *ConvertedRHS = &RHS; if (BinaryOperator::isComparisonOp(op)) { // We're looking for a type big enough to compare the symbolic value // with the given constant. // FIXME: This is an approximation of Sema::UsualArithmeticConversions. ASTContext &Ctx = getContext(); QualType SymbolType = LHS->getType(); uint64_t ValWidth = RHS.getBitWidth(); uint64_t TypeWidth = Ctx.getTypeSize(SymbolType); if (ValWidth < TypeWidth) { // If the value is too small, extend it. ConvertedRHS = &BasicVals.Convert(SymbolType, RHS); } else if (ValWidth == TypeWidth) { // If the value is signed but the symbol is unsigned, do the comparison // in unsigned space. [C99 6.3.1.8] // (For the opposite case, the value is already unsigned.) if (RHS.isSigned() && !SymbolType->isSignedIntegerOrEnumerationType()) ConvertedRHS = &BasicVals.Convert(SymbolType, RHS); } } else ConvertedRHS = &BasicVals.Convert(resultTy, RHS); return makeNonLoc(LHS, op, *ConvertedRHS, resultTy); } SVal SimpleSValBuilder::evalBinOpNN(ProgramStateRef state, BinaryOperator::Opcode op, NonLoc lhs, NonLoc rhs, QualType resultTy) { NonLoc InputLHS = lhs; NonLoc InputRHS = rhs; // Handle trivial case where left-side and right-side are the same. if (lhs == rhs) switch (op) { default: break; case BO_EQ: case BO_LE: case BO_GE: return makeTruthVal(true, resultTy); case BO_LT: case BO_GT: case BO_NE: return makeTruthVal(false, resultTy); case BO_Xor: case BO_Sub: if (resultTy->isIntegralOrEnumerationType()) return makeIntVal(0, resultTy); return evalCastFromNonLoc(makeIntVal(0, /*Unsigned=*/false), resultTy); case BO_Or: case BO_And: return evalCastFromNonLoc(lhs, resultTy); } while (1) { switch (lhs.getSubKind()) { default: return makeSymExprValNN(state, op, lhs, rhs, resultTy); case nonloc::LocAsIntegerKind: { Loc lhsL = lhs.castAs().getLoc(); switch (rhs.getSubKind()) { case nonloc::LocAsIntegerKind: return evalBinOpLL(state, op, lhsL, rhs.castAs().getLoc(), resultTy); case nonloc::ConcreteIntKind: { // Transform the integer into a location and compare. llvm::APSInt i = rhs.castAs().getValue(); BasicVals.getAPSIntType(Context.VoidPtrTy).apply(i); return evalBinOpLL(state, op, lhsL, makeLoc(i), resultTy); } default: switch (op) { case BO_EQ: return makeTruthVal(false, resultTy); case BO_NE: return makeTruthVal(true, resultTy); default: // This case also handles pointer arithmetic. return makeSymExprValNN(state, op, InputLHS, InputRHS, resultTy); } } } case nonloc::ConcreteIntKind: { llvm::APSInt LHSValue = lhs.castAs().getValue(); // If we're dealing with two known constants, just perform the operation. if (const llvm::APSInt *KnownRHSValue = getKnownValue(state, rhs)) { llvm::APSInt RHSValue = *KnownRHSValue; if (BinaryOperator::isComparisonOp(op)) { // We're looking for a type big enough to compare the two values. // FIXME: This is not correct. char + short will result in a promotion // to int. Unfortunately we have lost types by this point. APSIntType CompareType = std::max(APSIntType(LHSValue), APSIntType(RHSValue)); CompareType.apply(LHSValue); CompareType.apply(RHSValue); } else if (!BinaryOperator::isShiftOp(op)) { APSIntType IntType = BasicVals.getAPSIntType(resultTy); IntType.apply(LHSValue); IntType.apply(RHSValue); } const llvm::APSInt *Result = BasicVals.evalAPSInt(op, LHSValue, RHSValue); if (!Result) return UndefinedVal(); return nonloc::ConcreteInt(*Result); } // Swap the left and right sides and flip the operator if doing so // allows us to better reason about the expression (this is a form // of expression canonicalization). // While we're at it, catch some special cases for non-commutative ops. switch (op) { case BO_LT: case BO_GT: case BO_LE: case BO_GE: op = BinaryOperator::reverseComparisonOp(op); // FALL-THROUGH case BO_EQ: case BO_NE: case BO_Add: case BO_Mul: case BO_And: case BO_Xor: case BO_Or: std::swap(lhs, rhs); continue; case BO_Shr: // (~0)>>a if (LHSValue.isAllOnesValue() && LHSValue.isSigned()) return evalCastFromNonLoc(lhs, resultTy); // FALL-THROUGH case BO_Shl: // 0<>a if (LHSValue == 0) return evalCastFromNonLoc(lhs, resultTy); return makeSymExprValNN(state, op, InputLHS, InputRHS, resultTy); default: return makeSymExprValNN(state, op, InputLHS, InputRHS, resultTy); } } case nonloc::SymbolValKind: { // We only handle LHS as simple symbols or SymIntExprs. SymbolRef Sym = lhs.castAs().getSymbol(); // LHS is a symbolic expression. if (const SymIntExpr *symIntExpr = dyn_cast(Sym)) { // Is this a logical not? (!x is represented as x == 0.) if (op == BO_EQ && rhs.isZeroConstant()) { // We know how to negate certain expressions. Simplify them here. BinaryOperator::Opcode opc = symIntExpr->getOpcode(); switch (opc) { default: // We don't know how to negate this operation. // Just handle it as if it were a normal comparison to 0. break; case BO_LAnd: case BO_LOr: llvm_unreachable("Logical operators handled by branching logic."); case BO_Assign: case BO_MulAssign: case BO_DivAssign: case BO_RemAssign: case BO_AddAssign: case BO_SubAssign: case BO_ShlAssign: case BO_ShrAssign: case BO_AndAssign: case BO_XorAssign: case BO_OrAssign: case BO_Comma: llvm_unreachable("'=' and ',' operators handled by ExprEngine."); case BO_PtrMemD: case BO_PtrMemI: llvm_unreachable("Pointer arithmetic not handled here."); case BO_LT: case BO_GT: case BO_LE: case BO_GE: case BO_EQ: case BO_NE: assert(resultTy->isBooleanType() || resultTy == getConditionType()); assert(symIntExpr->getType()->isBooleanType() || getContext().hasSameUnqualifiedType(symIntExpr->getType(), getConditionType())); // Negate the comparison and make a value. opc = BinaryOperator::negateComparisonOp(opc); return makeNonLoc(symIntExpr->getLHS(), opc, symIntExpr->getRHS(), resultTy); } } // For now, only handle expressions whose RHS is a constant. if (const llvm::APSInt *RHSValue = getKnownValue(state, rhs)) { // If both the LHS and the current expression are additive, // fold their constants and try again. if (BinaryOperator::isAdditiveOp(op)) { BinaryOperator::Opcode lop = symIntExpr->getOpcode(); if (BinaryOperator::isAdditiveOp(lop)) { // Convert the two constants to a common type, then combine them. // resultTy may not be the best type to convert to, but it's // probably the best choice in expressions with mixed type // (such as x+1U+2LL). The rules for implicit conversions should // choose a reasonable type to preserve the expression, and will // at least match how the value is going to be used. APSIntType IntType = BasicVals.getAPSIntType(resultTy); const llvm::APSInt &first = IntType.convert(symIntExpr->getRHS()); const llvm::APSInt &second = IntType.convert(*RHSValue); const llvm::APSInt *newRHS; if (lop == op) newRHS = BasicVals.evalAPSInt(BO_Add, first, second); else newRHS = BasicVals.evalAPSInt(BO_Sub, first, second); assert(newRHS && "Invalid operation despite common type!"); rhs = nonloc::ConcreteInt(*newRHS); lhs = nonloc::SymbolVal(symIntExpr->getLHS()); op = lop; continue; } } // Otherwise, make a SymIntExpr out of the expression. return MakeSymIntVal(symIntExpr, op, *RHSValue, resultTy); } } // Does the symbolic expression simplify to a constant? // If so, "fold" the constant by setting 'lhs' to a ConcreteInt // and try again. ConstraintManager &CMgr = state->getConstraintManager(); if (const llvm::APSInt *Constant = CMgr.getSymVal(state, Sym)) { lhs = nonloc::ConcreteInt(*Constant); continue; } // Is the RHS a constant? if (const llvm::APSInt *RHSValue = getKnownValue(state, rhs)) return MakeSymIntVal(Sym, op, *RHSValue, resultTy); // Give up -- this is not a symbolic expression we can handle. return makeSymExprValNN(state, op, InputLHS, InputRHS, resultTy); } } } } static SVal evalBinOpFieldRegionFieldRegion(const FieldRegion *LeftFR, const FieldRegion *RightFR, BinaryOperator::Opcode op, QualType resultTy, SimpleSValBuilder &SVB) { // Only comparisons are meaningful here! if (!BinaryOperator::isComparisonOp(op)) return UnknownVal(); // Next, see if the two FRs have the same super-region. // FIXME: This doesn't handle casts yet, and simply stripping the casts // doesn't help. if (LeftFR->getSuperRegion() != RightFR->getSuperRegion()) return UnknownVal(); const FieldDecl *LeftFD = LeftFR->getDecl(); const FieldDecl *RightFD = RightFR->getDecl(); const RecordDecl *RD = LeftFD->getParent(); // Make sure the two FRs are from the same kind of record. Just in case! // FIXME: This is probably where inheritance would be a problem. if (RD != RightFD->getParent()) return UnknownVal(); // We know for sure that the two fields are not the same, since that // would have given us the same SVal. if (op == BO_EQ) return SVB.makeTruthVal(false, resultTy); if (op == BO_NE) return SVB.makeTruthVal(true, resultTy); // Iterate through the fields and see which one comes first. // [C99 6.7.2.1.13] "Within a structure object, the non-bit-field // members and the units in which bit-fields reside have addresses that // increase in the order in which they are declared." bool leftFirst = (op == BO_LT || op == BO_LE); for (RecordDecl::field_iterator I = RD->field_begin(), E = RD->field_end(); I!=E; ++I) { if (*I == LeftFD) return SVB.makeTruthVal(leftFirst, resultTy); if (*I == RightFD) return SVB.makeTruthVal(!leftFirst, resultTy); } llvm_unreachable("Fields not found in parent record's definition"); } // FIXME: all this logic will change if/when we have MemRegion::getLocation(). SVal SimpleSValBuilder::evalBinOpLL(ProgramStateRef state, BinaryOperator::Opcode op, Loc lhs, Loc rhs, QualType resultTy) { // Only comparisons and subtractions are valid operations on two pointers. // See [C99 6.5.5 through 6.5.14] or [C++0x 5.6 through 5.15]. // However, if a pointer is casted to an integer, evalBinOpNN may end up // calling this function with another operation (PR7527). We don't attempt to // model this for now, but it could be useful, particularly when the // "location" is actually an integer value that's been passed through a void*. if (!(BinaryOperator::isComparisonOp(op) || op == BO_Sub)) return UnknownVal(); // Special cases for when both sides are identical. if (lhs == rhs) { switch (op) { default: llvm_unreachable("Unimplemented operation for two identical values"); case BO_Sub: return makeZeroVal(resultTy); case BO_EQ: case BO_LE: case BO_GE: return makeTruthVal(true, resultTy); case BO_NE: case BO_LT: case BO_GT: return makeTruthVal(false, resultTy); } } switch (lhs.getSubKind()) { default: llvm_unreachable("Ordering not implemented for this Loc."); case loc::GotoLabelKind: // The only thing we know about labels is that they're non-null. if (rhs.isZeroConstant()) { switch (op) { default: break; case BO_Sub: return evalCastFromLoc(lhs, resultTy); case BO_EQ: case BO_LE: case BO_LT: return makeTruthVal(false, resultTy); case BO_NE: case BO_GT: case BO_GE: return makeTruthVal(true, resultTy); } } // There may be two labels for the same location, and a function region may // have the same address as a label at the start of the function (depending // on the ABI). // FIXME: we can probably do a comparison against other MemRegions, though. // FIXME: is there a way to tell if two labels refer to the same location? return UnknownVal(); case loc::ConcreteIntKind: { // If one of the operands is a symbol and the other is a constant, // build an expression for use by the constraint manager. if (SymbolRef rSym = rhs.getAsLocSymbol()) { // We can only build expressions with symbols on the left, // so we need a reversible operator. if (!BinaryOperator::isComparisonOp(op)) return UnknownVal(); const llvm::APSInt &lVal = lhs.castAs().getValue(); op = BinaryOperator::reverseComparisonOp(op); return makeNonLoc(rSym, op, lVal, resultTy); } // If both operands are constants, just perform the operation. if (Optional rInt = rhs.getAs()) { SVal ResultVal = lhs.castAs().evalBinOp(BasicVals, op, *rInt); if (Optional Result = ResultVal.getAs()) return evalCastFromNonLoc(*Result, resultTy); assert(!ResultVal.getAs() && "Loc-Loc ops should not produce Locs"); return UnknownVal(); } // Special case comparisons against NULL. // This must come after the test if the RHS is a symbol, which is used to // build constraints. The address of any non-symbolic region is guaranteed // to be non-NULL, as is any label. assert(rhs.getAs() || rhs.getAs()); if (lhs.isZeroConstant()) { switch (op) { default: break; case BO_EQ: case BO_GT: case BO_GE: return makeTruthVal(false, resultTy); case BO_NE: case BO_LT: case BO_LE: return makeTruthVal(true, resultTy); } } // Comparing an arbitrary integer to a region or label address is // completely unknowable. return UnknownVal(); } case loc::MemRegionKind: { if (Optional rInt = rhs.getAs()) { // If one of the operands is a symbol and the other is a constant, // build an expression for use by the constraint manager. if (SymbolRef lSym = lhs.getAsLocSymbol(true)) return MakeSymIntVal(lSym, op, rInt->getValue(), resultTy); // Special case comparisons to NULL. // This must come after the test if the LHS is a symbol, which is used to // build constraints. The address of any non-symbolic region is guaranteed // to be non-NULL. if (rInt->isZeroConstant()) { if (op == BO_Sub) return evalCastFromLoc(lhs, resultTy); if (BinaryOperator::isComparisonOp(op)) { QualType boolType = getContext().BoolTy; NonLoc l = evalCastFromLoc(lhs, boolType).castAs(); NonLoc r = makeTruthVal(false, boolType).castAs(); return evalBinOpNN(state, op, l, r, resultTy); } } // Comparing a region to an arbitrary integer is completely unknowable. return UnknownVal(); } // Get both values as regions, if possible. const MemRegion *LeftMR = lhs.getAsRegion(); assert(LeftMR && "MemRegionKind SVal doesn't have a region!"); const MemRegion *RightMR = rhs.getAsRegion(); if (!RightMR) // The RHS is probably a label, which in theory could address a region. // FIXME: we can probably make a more useful statement about non-code // regions, though. return UnknownVal(); const MemRegion *LeftBase = LeftMR->getBaseRegion(); const MemRegion *RightBase = RightMR->getBaseRegion(); const MemSpaceRegion *LeftMS = LeftBase->getMemorySpace(); const MemSpaceRegion *RightMS = RightBase->getMemorySpace(); const MemSpaceRegion *UnknownMS = MemMgr.getUnknownRegion(); // If the two regions are from different known memory spaces they cannot be // equal. Also, assume that no symbolic region (whose memory space is // unknown) is on the stack. if (LeftMS != RightMS && ((LeftMS != UnknownMS && RightMS != UnknownMS) || (isa(LeftMS) || isa(RightMS)))) { switch (op) { default: return UnknownVal(); case BO_EQ: return makeTruthVal(false, resultTy); case BO_NE: return makeTruthVal(true, resultTy); } } // If both values wrap regions, see if they're from different base regions. // Note, heap base symbolic regions are assumed to not alias with // each other; for example, we assume that malloc returns different address // on each invocation. if (LeftBase != RightBase && ((!isa(LeftBase) && !isa(RightBase)) || (isa(LeftMS) || isa(RightMS))) ){ switch (op) { default: return UnknownVal(); case BO_EQ: return makeTruthVal(false, resultTy); case BO_NE: return makeTruthVal(true, resultTy); } } // Handle special cases for when both regions are element regions. const ElementRegion *RightER = dyn_cast(RightMR); const ElementRegion *LeftER = dyn_cast(LeftMR); if (RightER && LeftER) { // Next, see if the two ERs have the same super-region and matching types. // FIXME: This should do something useful even if the types don't match, // though if both indexes are constant the RegionRawOffset path will // give the correct answer. if (LeftER->getSuperRegion() == RightER->getSuperRegion() && LeftER->getElementType() == RightER->getElementType()) { // Get the left index and cast it to the correct type. // If the index is unknown or undefined, bail out here. SVal LeftIndexVal = LeftER->getIndex(); Optional LeftIndex = LeftIndexVal.getAs(); if (!LeftIndex) return UnknownVal(); LeftIndexVal = evalCastFromNonLoc(*LeftIndex, ArrayIndexTy); LeftIndex = LeftIndexVal.getAs(); if (!LeftIndex) return UnknownVal(); // Do the same for the right index. SVal RightIndexVal = RightER->getIndex(); Optional RightIndex = RightIndexVal.getAs(); if (!RightIndex) return UnknownVal(); RightIndexVal = evalCastFromNonLoc(*RightIndex, ArrayIndexTy); RightIndex = RightIndexVal.getAs(); if (!RightIndex) return UnknownVal(); // Actually perform the operation. // evalBinOpNN expects the two indexes to already be the right type. return evalBinOpNN(state, op, *LeftIndex, *RightIndex, resultTy); } } // Special handling of the FieldRegions, even with symbolic offsets. const FieldRegion *RightFR = dyn_cast(RightMR); const FieldRegion *LeftFR = dyn_cast(LeftMR); if (RightFR && LeftFR) { SVal R = evalBinOpFieldRegionFieldRegion(LeftFR, RightFR, op, resultTy, *this); if (!R.isUnknown()) return R; } // Compare the regions using the raw offsets. RegionOffset LeftOffset = LeftMR->getAsOffset(); RegionOffset RightOffset = RightMR->getAsOffset(); if (LeftOffset.getRegion() != NULL && LeftOffset.getRegion() == RightOffset.getRegion() && !LeftOffset.hasSymbolicOffset() && !RightOffset.hasSymbolicOffset()) { int64_t left = LeftOffset.getOffset(); int64_t right = RightOffset.getOffset(); switch (op) { default: return UnknownVal(); case BO_LT: return makeTruthVal(left < right, resultTy); case BO_GT: return makeTruthVal(left > right, resultTy); case BO_LE: return makeTruthVal(left <= right, resultTy); case BO_GE: return makeTruthVal(left >= right, resultTy); case BO_EQ: return makeTruthVal(left == right, resultTy); case BO_NE: return makeTruthVal(left != right, resultTy); } } // At this point we're not going to get a good answer, but we can try // conjuring an expression instead. SymbolRef LHSSym = lhs.getAsLocSymbol(); SymbolRef RHSSym = rhs.getAsLocSymbol(); if (LHSSym && RHSSym) return makeNonLoc(LHSSym, op, RHSSym, resultTy); // If we get here, we have no way of comparing the regions. return UnknownVal(); } } } SVal SimpleSValBuilder::evalBinOpLN(ProgramStateRef state, BinaryOperator::Opcode op, Loc lhs, NonLoc rhs, QualType resultTy) { assert(!BinaryOperator::isComparisonOp(op) && "arguments to comparison ops must be of the same type"); // Special case: rhs is a zero constant. if (rhs.isZeroConstant()) return lhs; // We are dealing with pointer arithmetic. // Handle pointer arithmetic on constant values. if (Optional rhsInt = rhs.getAs()) { if (Optional lhsInt = lhs.getAs()) { const llvm::APSInt &leftI = lhsInt->getValue(); assert(leftI.isUnsigned()); llvm::APSInt rightI(rhsInt->getValue(), /* isUnsigned */ true); // Convert the bitwidth of rightI. This should deal with overflow // since we are dealing with concrete values. rightI = rightI.extOrTrunc(leftI.getBitWidth()); // Offset the increment by the pointer size. llvm::APSInt Multiplicand(rightI.getBitWidth(), /* isUnsigned */ true); rightI *= Multiplicand; // Compute the adjusted pointer. switch (op) { case BO_Add: rightI = leftI + rightI; break; case BO_Sub: rightI = leftI - rightI; break; default: llvm_unreachable("Invalid pointer arithmetic operation"); } return loc::ConcreteInt(getBasicValueFactory().getValue(rightI)); } } // Handle cases where 'lhs' is a region. if (const MemRegion *region = lhs.getAsRegion()) { rhs = convertToArrayIndex(rhs).castAs(); SVal index = UnknownVal(); const MemRegion *superR = 0; QualType elementType; if (const ElementRegion *elemReg = dyn_cast(region)) { assert(op == BO_Add || op == BO_Sub); index = evalBinOpNN(state, op, elemReg->getIndex(), rhs, getArrayIndexType()); superR = elemReg->getSuperRegion(); elementType = elemReg->getElementType(); } else if (isa(region)) { superR = region; index = rhs; if (resultTy->isAnyPointerType()) elementType = resultTy->getPointeeType(); } if (Optional indexV = index.getAs()) { return loc::MemRegionVal(MemMgr.getElementRegion(elementType, *indexV, superR, getContext())); } } return UnknownVal(); } const llvm::APSInt *SimpleSValBuilder::getKnownValue(ProgramStateRef state, SVal V) { if (V.isUnknownOrUndef()) return NULL; if (Optional X = V.getAs()) return &X->getValue(); if (Optional X = V.getAs()) return &X->getValue(); if (SymbolRef Sym = V.getAsSymbol()) return state->getConstraintManager().getSymVal(state, Sym); // FIXME: Add support for SymExprs. return NULL; } @