SemaChecking.cpp revision e2c60667d1bc29b7e148f9c9828dcbdbfc5b82b0
1//===--- SemaChecking.cpp - Extra Semantic Checking -----------------------===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// This file implements extra semantic analysis beyond what is enforced 11// by the C type system. 12// 13//===----------------------------------------------------------------------===// 14 15#include "clang/Sema/Initialization.h" 16#include "clang/Sema/Sema.h" 17#include "clang/Sema/SemaInternal.h" 18#include "clang/Sema/Initialization.h" 19#include "clang/Sema/ScopeInfo.h" 20#include "clang/Analysis/Analyses/FormatString.h" 21#include "clang/AST/ASTContext.h" 22#include "clang/AST/CharUnits.h" 23#include "clang/AST/DeclCXX.h" 24#include "clang/AST/DeclObjC.h" 25#include "clang/AST/ExprCXX.h" 26#include "clang/AST/ExprObjC.h" 27#include "clang/AST/EvaluatedExprVisitor.h" 28#include "clang/AST/DeclObjC.h" 29#include "clang/AST/StmtCXX.h" 30#include "clang/AST/StmtObjC.h" 31#include "clang/Lex/Preprocessor.h" 32#include "llvm/ADT/BitVector.h" 33#include "llvm/ADT/SmallString.h" 34#include "llvm/ADT/STLExtras.h" 35#include "llvm/Support/raw_ostream.h" 36#include "clang/Basic/TargetBuiltins.h" 37#include "clang/Basic/TargetInfo.h" 38#include "clang/Basic/ConvertUTF.h" 39#include <limits> 40using namespace clang; 41using namespace sema; 42 43SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 44 unsigned ByteNo) const { 45 return SL->getLocationOfByte(ByteNo, PP.getSourceManager(), 46 PP.getLangOpts(), PP.getTargetInfo()); 47} 48 49/// Checks that a call expression's argument count is the desired number. 50/// This is useful when doing custom type-checking. Returns true on error. 51static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) { 52 unsigned argCount = call->getNumArgs(); 53 if (argCount == desiredArgCount) return false; 54 55 if (argCount < desiredArgCount) 56 return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args) 57 << 0 /*function call*/ << desiredArgCount << argCount 58 << call->getSourceRange(); 59 60 // Highlight all the excess arguments. 61 SourceRange range(call->getArg(desiredArgCount)->getLocStart(), 62 call->getArg(argCount - 1)->getLocEnd()); 63 64 return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args) 65 << 0 /*function call*/ << desiredArgCount << argCount 66 << call->getArg(1)->getSourceRange(); 67} 68 69/// Check that the first argument to __builtin_annotation is an integer 70/// and the second argument is a non-wide string literal. 71static bool SemaBuiltinAnnotation(Sema &S, CallExpr *TheCall) { 72 if (checkArgCount(S, TheCall, 2)) 73 return true; 74 75 // First argument should be an integer. 76 Expr *ValArg = TheCall->getArg(0); 77 QualType Ty = ValArg->getType(); 78 if (!Ty->isIntegerType()) { 79 S.Diag(ValArg->getLocStart(), diag::err_builtin_annotation_first_arg) 80 << ValArg->getSourceRange(); 81 return true; 82 } 83 84 // Second argument should be a constant string. 85 Expr *StrArg = TheCall->getArg(1)->IgnoreParenCasts(); 86 StringLiteral *Literal = dyn_cast<StringLiteral>(StrArg); 87 if (!Literal || !Literal->isAscii()) { 88 S.Diag(StrArg->getLocStart(), diag::err_builtin_annotation_second_arg) 89 << StrArg->getSourceRange(); 90 return true; 91 } 92 93 TheCall->setType(Ty); 94 return false; 95} 96 97ExprResult 98Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 99 ExprResult TheCallResult(Owned(TheCall)); 100 101 // Find out if any arguments are required to be integer constant expressions. 102 unsigned ICEArguments = 0; 103 ASTContext::GetBuiltinTypeError Error; 104 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments); 105 if (Error != ASTContext::GE_None) 106 ICEArguments = 0; // Don't diagnose previously diagnosed errors. 107 108 // If any arguments are required to be ICE's, check and diagnose. 109 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) { 110 // Skip arguments not required to be ICE's. 111 if ((ICEArguments & (1 << ArgNo)) == 0) continue; 112 113 llvm::APSInt Result; 114 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 115 return true; 116 ICEArguments &= ~(1 << ArgNo); 117 } 118 119 switch (BuiltinID) { 120 case Builtin::BI__builtin___CFStringMakeConstantString: 121 assert(TheCall->getNumArgs() == 1 && 122 "Wrong # arguments to builtin CFStringMakeConstantString"); 123 if (CheckObjCString(TheCall->getArg(0))) 124 return ExprError(); 125 break; 126 case Builtin::BI__builtin_stdarg_start: 127 case Builtin::BI__builtin_va_start: 128 if (SemaBuiltinVAStart(TheCall)) 129 return ExprError(); 130 break; 131 case Builtin::BI__builtin_isgreater: 132 case Builtin::BI__builtin_isgreaterequal: 133 case Builtin::BI__builtin_isless: 134 case Builtin::BI__builtin_islessequal: 135 case Builtin::BI__builtin_islessgreater: 136 case Builtin::BI__builtin_isunordered: 137 if (SemaBuiltinUnorderedCompare(TheCall)) 138 return ExprError(); 139 break; 140 case Builtin::BI__builtin_fpclassify: 141 if (SemaBuiltinFPClassification(TheCall, 6)) 142 return ExprError(); 143 break; 144 case Builtin::BI__builtin_isfinite: 145 case Builtin::BI__builtin_isinf: 146 case Builtin::BI__builtin_isinf_sign: 147 case Builtin::BI__builtin_isnan: 148 case Builtin::BI__builtin_isnormal: 149 if (SemaBuiltinFPClassification(TheCall, 1)) 150 return ExprError(); 151 break; 152 case Builtin::BI__builtin_shufflevector: 153 return SemaBuiltinShuffleVector(TheCall); 154 // TheCall will be freed by the smart pointer here, but that's fine, since 155 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 156 case Builtin::BI__builtin_prefetch: 157 if (SemaBuiltinPrefetch(TheCall)) 158 return ExprError(); 159 break; 160 case Builtin::BI__builtin_object_size: 161 if (SemaBuiltinObjectSize(TheCall)) 162 return ExprError(); 163 break; 164 case Builtin::BI__builtin_longjmp: 165 if (SemaBuiltinLongjmp(TheCall)) 166 return ExprError(); 167 break; 168 169 case Builtin::BI__builtin_classify_type: 170 if (checkArgCount(*this, TheCall, 1)) return true; 171 TheCall->setType(Context.IntTy); 172 break; 173 case Builtin::BI__builtin_constant_p: 174 if (checkArgCount(*this, TheCall, 1)) return true; 175 TheCall->setType(Context.IntTy); 176 break; 177 case Builtin::BI__sync_fetch_and_add: 178 case Builtin::BI__sync_fetch_and_add_1: 179 case Builtin::BI__sync_fetch_and_add_2: 180 case Builtin::BI__sync_fetch_and_add_4: 181 case Builtin::BI__sync_fetch_and_add_8: 182 case Builtin::BI__sync_fetch_and_add_16: 183 case Builtin::BI__sync_fetch_and_sub: 184 case Builtin::BI__sync_fetch_and_sub_1: 185 case Builtin::BI__sync_fetch_and_sub_2: 186 case Builtin::BI__sync_fetch_and_sub_4: 187 case Builtin::BI__sync_fetch_and_sub_8: 188 case Builtin::BI__sync_fetch_and_sub_16: 189 case Builtin::BI__sync_fetch_and_or: 190 case Builtin::BI__sync_fetch_and_or_1: 191 case Builtin::BI__sync_fetch_and_or_2: 192 case Builtin::BI__sync_fetch_and_or_4: 193 case Builtin::BI__sync_fetch_and_or_8: 194 case Builtin::BI__sync_fetch_and_or_16: 195 case Builtin::BI__sync_fetch_and_and: 196 case Builtin::BI__sync_fetch_and_and_1: 197 case Builtin::BI__sync_fetch_and_and_2: 198 case Builtin::BI__sync_fetch_and_and_4: 199 case Builtin::BI__sync_fetch_and_and_8: 200 case Builtin::BI__sync_fetch_and_and_16: 201 case Builtin::BI__sync_fetch_and_xor: 202 case Builtin::BI__sync_fetch_and_xor_1: 203 case Builtin::BI__sync_fetch_and_xor_2: 204 case Builtin::BI__sync_fetch_and_xor_4: 205 case Builtin::BI__sync_fetch_and_xor_8: 206 case Builtin::BI__sync_fetch_and_xor_16: 207 case Builtin::BI__sync_add_and_fetch: 208 case Builtin::BI__sync_add_and_fetch_1: 209 case Builtin::BI__sync_add_and_fetch_2: 210 case Builtin::BI__sync_add_and_fetch_4: 211 case Builtin::BI__sync_add_and_fetch_8: 212 case Builtin::BI__sync_add_and_fetch_16: 213 case Builtin::BI__sync_sub_and_fetch: 214 case Builtin::BI__sync_sub_and_fetch_1: 215 case Builtin::BI__sync_sub_and_fetch_2: 216 case Builtin::BI__sync_sub_and_fetch_4: 217 case Builtin::BI__sync_sub_and_fetch_8: 218 case Builtin::BI__sync_sub_and_fetch_16: 219 case Builtin::BI__sync_and_and_fetch: 220 case Builtin::BI__sync_and_and_fetch_1: 221 case Builtin::BI__sync_and_and_fetch_2: 222 case Builtin::BI__sync_and_and_fetch_4: 223 case Builtin::BI__sync_and_and_fetch_8: 224 case Builtin::BI__sync_and_and_fetch_16: 225 case Builtin::BI__sync_or_and_fetch: 226 case Builtin::BI__sync_or_and_fetch_1: 227 case Builtin::BI__sync_or_and_fetch_2: 228 case Builtin::BI__sync_or_and_fetch_4: 229 case Builtin::BI__sync_or_and_fetch_8: 230 case Builtin::BI__sync_or_and_fetch_16: 231 case Builtin::BI__sync_xor_and_fetch: 232 case Builtin::BI__sync_xor_and_fetch_1: 233 case Builtin::BI__sync_xor_and_fetch_2: 234 case Builtin::BI__sync_xor_and_fetch_4: 235 case Builtin::BI__sync_xor_and_fetch_8: 236 case Builtin::BI__sync_xor_and_fetch_16: 237 case Builtin::BI__sync_val_compare_and_swap: 238 case Builtin::BI__sync_val_compare_and_swap_1: 239 case Builtin::BI__sync_val_compare_and_swap_2: 240 case Builtin::BI__sync_val_compare_and_swap_4: 241 case Builtin::BI__sync_val_compare_and_swap_8: 242 case Builtin::BI__sync_val_compare_and_swap_16: 243 case Builtin::BI__sync_bool_compare_and_swap: 244 case Builtin::BI__sync_bool_compare_and_swap_1: 245 case Builtin::BI__sync_bool_compare_and_swap_2: 246 case Builtin::BI__sync_bool_compare_and_swap_4: 247 case Builtin::BI__sync_bool_compare_and_swap_8: 248 case Builtin::BI__sync_bool_compare_and_swap_16: 249 case Builtin::BI__sync_lock_test_and_set: 250 case Builtin::BI__sync_lock_test_and_set_1: 251 case Builtin::BI__sync_lock_test_and_set_2: 252 case Builtin::BI__sync_lock_test_and_set_4: 253 case Builtin::BI__sync_lock_test_and_set_8: 254 case Builtin::BI__sync_lock_test_and_set_16: 255 case Builtin::BI__sync_lock_release: 256 case Builtin::BI__sync_lock_release_1: 257 case Builtin::BI__sync_lock_release_2: 258 case Builtin::BI__sync_lock_release_4: 259 case Builtin::BI__sync_lock_release_8: 260 case Builtin::BI__sync_lock_release_16: 261 case Builtin::BI__sync_swap: 262 case Builtin::BI__sync_swap_1: 263 case Builtin::BI__sync_swap_2: 264 case Builtin::BI__sync_swap_4: 265 case Builtin::BI__sync_swap_8: 266 case Builtin::BI__sync_swap_16: 267 return SemaBuiltinAtomicOverloaded(move(TheCallResult)); 268#define BUILTIN(ID, TYPE, ATTRS) 269#define ATOMIC_BUILTIN(ID, TYPE, ATTRS) \ 270 case Builtin::BI##ID: \ 271 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::AO##ID); 272#include "clang/Basic/Builtins.def" 273 case Builtin::BI__builtin_annotation: 274 if (SemaBuiltinAnnotation(*this, TheCall)) 275 return ExprError(); 276 break; 277 } 278 279 // Since the target specific builtins for each arch overlap, only check those 280 // of the arch we are compiling for. 281 if (BuiltinID >= Builtin::FirstTSBuiltin) { 282 switch (Context.getTargetInfo().getTriple().getArch()) { 283 case llvm::Triple::arm: 284 case llvm::Triple::thumb: 285 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 286 return ExprError(); 287 break; 288 default: 289 break; 290 } 291 } 292 293 return move(TheCallResult); 294} 295 296// Get the valid immediate range for the specified NEON type code. 297static unsigned RFT(unsigned t, bool shift = false) { 298 NeonTypeFlags Type(t); 299 int IsQuad = Type.isQuad(); 300 switch (Type.getEltType()) { 301 case NeonTypeFlags::Int8: 302 case NeonTypeFlags::Poly8: 303 return shift ? 7 : (8 << IsQuad) - 1; 304 case NeonTypeFlags::Int16: 305 case NeonTypeFlags::Poly16: 306 return shift ? 15 : (4 << IsQuad) - 1; 307 case NeonTypeFlags::Int32: 308 return shift ? 31 : (2 << IsQuad) - 1; 309 case NeonTypeFlags::Int64: 310 return shift ? 63 : (1 << IsQuad) - 1; 311 case NeonTypeFlags::Float16: 312 assert(!shift && "cannot shift float types!"); 313 return (4 << IsQuad) - 1; 314 case NeonTypeFlags::Float32: 315 assert(!shift && "cannot shift float types!"); 316 return (2 << IsQuad) - 1; 317 } 318 llvm_unreachable("Invalid NeonTypeFlag!"); 319} 320 321/// getNeonEltType - Return the QualType corresponding to the elements of 322/// the vector type specified by the NeonTypeFlags. This is used to check 323/// the pointer arguments for Neon load/store intrinsics. 324static QualType getNeonEltType(NeonTypeFlags Flags, ASTContext &Context) { 325 switch (Flags.getEltType()) { 326 case NeonTypeFlags::Int8: 327 return Flags.isUnsigned() ? Context.UnsignedCharTy : Context.SignedCharTy; 328 case NeonTypeFlags::Int16: 329 return Flags.isUnsigned() ? Context.UnsignedShortTy : Context.ShortTy; 330 case NeonTypeFlags::Int32: 331 return Flags.isUnsigned() ? Context.UnsignedIntTy : Context.IntTy; 332 case NeonTypeFlags::Int64: 333 return Flags.isUnsigned() ? Context.UnsignedLongLongTy : Context.LongLongTy; 334 case NeonTypeFlags::Poly8: 335 return Context.SignedCharTy; 336 case NeonTypeFlags::Poly16: 337 return Context.ShortTy; 338 case NeonTypeFlags::Float16: 339 return Context.UnsignedShortTy; 340 case NeonTypeFlags::Float32: 341 return Context.FloatTy; 342 } 343 llvm_unreachable("Invalid NeonTypeFlag!"); 344} 345 346bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 347 llvm::APSInt Result; 348 349 unsigned mask = 0; 350 unsigned TV = 0; 351 int PtrArgNum = -1; 352 bool HasConstPtr = false; 353 switch (BuiltinID) { 354#define GET_NEON_OVERLOAD_CHECK 355#include "clang/Basic/arm_neon.inc" 356#undef GET_NEON_OVERLOAD_CHECK 357 } 358 359 // For NEON intrinsics which are overloaded on vector element type, validate 360 // the immediate which specifies which variant to emit. 361 unsigned ImmArg = TheCall->getNumArgs()-1; 362 if (mask) { 363 if (SemaBuiltinConstantArg(TheCall, ImmArg, Result)) 364 return true; 365 366 TV = Result.getLimitedValue(64); 367 if ((TV > 63) || (mask & (1 << TV)) == 0) 368 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 369 << TheCall->getArg(ImmArg)->getSourceRange(); 370 } 371 372 if (PtrArgNum >= 0) { 373 // Check that pointer arguments have the specified type. 374 Expr *Arg = TheCall->getArg(PtrArgNum); 375 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) 376 Arg = ICE->getSubExpr(); 377 ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg); 378 QualType RHSTy = RHS.get()->getType(); 379 QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context); 380 if (HasConstPtr) 381 EltTy = EltTy.withConst(); 382 QualType LHSTy = Context.getPointerType(EltTy); 383 AssignConvertType ConvTy; 384 ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS); 385 if (RHS.isInvalid()) 386 return true; 387 if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy, 388 RHS.get(), AA_Assigning)) 389 return true; 390 } 391 392 // For NEON intrinsics which take an immediate value as part of the 393 // instruction, range check them here. 394 unsigned i = 0, l = 0, u = 0; 395 switch (BuiltinID) { 396 default: return false; 397 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 398 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 399 case ARM::BI__builtin_arm_vcvtr_f: 400 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 401#define GET_NEON_IMMEDIATE_CHECK 402#include "clang/Basic/arm_neon.inc" 403#undef GET_NEON_IMMEDIATE_CHECK 404 }; 405 406 // Check that the immediate argument is actually a constant. 407 if (SemaBuiltinConstantArg(TheCall, i, Result)) 408 return true; 409 410 // Range check against the upper/lower values for this isntruction. 411 unsigned Val = Result.getZExtValue(); 412 if (Val < l || Val > (u + l)) 413 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 414 << l << u+l << TheCall->getArg(i)->getSourceRange(); 415 416 // FIXME: VFP Intrinsics should error if VFP not present. 417 return false; 418} 419 420/// CheckFunctionCall - Check a direct function call for various correctness 421/// and safety properties not strictly enforced by the C type system. 422bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 423 // Get the IdentifierInfo* for the called function. 424 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 425 426 // None of the checks below are needed for functions that don't have 427 // simple names (e.g., C++ conversion functions). 428 if (!FnInfo) 429 return false; 430 431 // FIXME: This mechanism should be abstracted to be less fragile and 432 // more efficient. For example, just map function ids to custom 433 // handlers. 434 435 // Printf and scanf checking. 436 for (specific_attr_iterator<FormatAttr> 437 i = FDecl->specific_attr_begin<FormatAttr>(), 438 e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { 439 CheckFormatArguments(*i, TheCall); 440 } 441 442 for (specific_attr_iterator<NonNullAttr> 443 i = FDecl->specific_attr_begin<NonNullAttr>(), 444 e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) { 445 CheckNonNullArguments(*i, TheCall->getArgs(), 446 TheCall->getCallee()->getLocStart()); 447 } 448 449 unsigned CMId = FDecl->getMemoryFunctionKind(); 450 if (CMId == 0) 451 return false; 452 453 // Handle memory setting and copying functions. 454 if (CMId == Builtin::BIstrlcpy || CMId == Builtin::BIstrlcat) 455 CheckStrlcpycatArguments(TheCall, FnInfo); 456 else if (CMId == Builtin::BIstrncat) 457 CheckStrncatArguments(TheCall, FnInfo); 458 else 459 CheckMemaccessArguments(TheCall, CMId, FnInfo); 460 461 return false; 462} 463 464bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac, 465 Expr **Args, unsigned NumArgs) { 466 for (specific_attr_iterator<FormatAttr> 467 i = Method->specific_attr_begin<FormatAttr>(), 468 e = Method->specific_attr_end<FormatAttr>(); i != e ; ++i) { 469 470 CheckFormatArguments(*i, Args, NumArgs, false, lbrac, 471 Method->getSourceRange()); 472 } 473 474 // diagnose nonnull arguments. 475 for (specific_attr_iterator<NonNullAttr> 476 i = Method->specific_attr_begin<NonNullAttr>(), 477 e = Method->specific_attr_end<NonNullAttr>(); i != e; ++i) { 478 CheckNonNullArguments(*i, Args, lbrac); 479 } 480 481 return false; 482} 483 484bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 485 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 486 if (!V) 487 return false; 488 489 QualType Ty = V->getType(); 490 if (!Ty->isBlockPointerType()) 491 return false; 492 493 // format string checking. 494 for (specific_attr_iterator<FormatAttr> 495 i = NDecl->specific_attr_begin<FormatAttr>(), 496 e = NDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { 497 CheckFormatArguments(*i, TheCall); 498 } 499 500 return false; 501} 502 503ExprResult Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, 504 AtomicExpr::AtomicOp Op) { 505 CallExpr *TheCall = cast<CallExpr>(TheCallResult.get()); 506 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 507 508 // All these operations take one of the following forms: 509 enum { 510 // C __c11_atomic_init(A *, C) 511 Init, 512 // C __c11_atomic_load(A *, int) 513 Load, 514 // void __atomic_load(A *, CP, int) 515 Copy, 516 // C __c11_atomic_add(A *, M, int) 517 Arithmetic, 518 // C __atomic_exchange_n(A *, CP, int) 519 Xchg, 520 // void __atomic_exchange(A *, C *, CP, int) 521 GNUXchg, 522 // bool __c11_atomic_compare_exchange_strong(A *, C *, CP, int, int) 523 C11CmpXchg, 524 // bool __atomic_compare_exchange(A *, C *, CP, bool, int, int) 525 GNUCmpXchg 526 } Form = Init; 527 const unsigned NumArgs[] = { 2, 2, 3, 3, 3, 4, 5, 6 }; 528 const unsigned NumVals[] = { 1, 0, 1, 1, 1, 2, 2, 3 }; 529 // where: 530 // C is an appropriate type, 531 // A is volatile _Atomic(C) for __c11 builtins and is C for GNU builtins, 532 // CP is C for __c11 builtins and GNU _n builtins and is C * otherwise, 533 // M is C if C is an integer, and ptrdiff_t if C is a pointer, and 534 // the int parameters are for orderings. 535 536 assert(AtomicExpr::AO__c11_atomic_init == 0 && 537 AtomicExpr::AO__c11_atomic_fetch_xor + 1 == AtomicExpr::AO__atomic_load 538 && "need to update code for modified C11 atomics"); 539 bool IsC11 = Op >= AtomicExpr::AO__c11_atomic_init && 540 Op <= AtomicExpr::AO__c11_atomic_fetch_xor; 541 bool IsN = Op == AtomicExpr::AO__atomic_load_n || 542 Op == AtomicExpr::AO__atomic_store_n || 543 Op == AtomicExpr::AO__atomic_exchange_n || 544 Op == AtomicExpr::AO__atomic_compare_exchange_n; 545 bool IsAddSub = false; 546 547 switch (Op) { 548 case AtomicExpr::AO__c11_atomic_init: 549 Form = Init; 550 break; 551 552 case AtomicExpr::AO__c11_atomic_load: 553 case AtomicExpr::AO__atomic_load_n: 554 Form = Load; 555 break; 556 557 case AtomicExpr::AO__c11_atomic_store: 558 case AtomicExpr::AO__atomic_load: 559 case AtomicExpr::AO__atomic_store: 560 case AtomicExpr::AO__atomic_store_n: 561 Form = Copy; 562 break; 563 564 case AtomicExpr::AO__c11_atomic_fetch_add: 565 case AtomicExpr::AO__c11_atomic_fetch_sub: 566 case AtomicExpr::AO__atomic_fetch_add: 567 case AtomicExpr::AO__atomic_fetch_sub: 568 case AtomicExpr::AO__atomic_add_fetch: 569 case AtomicExpr::AO__atomic_sub_fetch: 570 IsAddSub = true; 571 // Fall through. 572 case AtomicExpr::AO__c11_atomic_fetch_and: 573 case AtomicExpr::AO__c11_atomic_fetch_or: 574 case AtomicExpr::AO__c11_atomic_fetch_xor: 575 case AtomicExpr::AO__atomic_fetch_and: 576 case AtomicExpr::AO__atomic_fetch_or: 577 case AtomicExpr::AO__atomic_fetch_xor: 578 case AtomicExpr::AO__atomic_fetch_nand: 579 case AtomicExpr::AO__atomic_and_fetch: 580 case AtomicExpr::AO__atomic_or_fetch: 581 case AtomicExpr::AO__atomic_xor_fetch: 582 case AtomicExpr::AO__atomic_nand_fetch: 583 Form = Arithmetic; 584 break; 585 586 case AtomicExpr::AO__c11_atomic_exchange: 587 case AtomicExpr::AO__atomic_exchange_n: 588 Form = Xchg; 589 break; 590 591 case AtomicExpr::AO__atomic_exchange: 592 Form = GNUXchg; 593 break; 594 595 case AtomicExpr::AO__c11_atomic_compare_exchange_strong: 596 case AtomicExpr::AO__c11_atomic_compare_exchange_weak: 597 Form = C11CmpXchg; 598 break; 599 600 case AtomicExpr::AO__atomic_compare_exchange: 601 case AtomicExpr::AO__atomic_compare_exchange_n: 602 Form = GNUCmpXchg; 603 break; 604 } 605 606 // Check we have the right number of arguments. 607 if (TheCall->getNumArgs() < NumArgs[Form]) { 608 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 609 << 0 << NumArgs[Form] << TheCall->getNumArgs() 610 << TheCall->getCallee()->getSourceRange(); 611 return ExprError(); 612 } else if (TheCall->getNumArgs() > NumArgs[Form]) { 613 Diag(TheCall->getArg(NumArgs[Form])->getLocStart(), 614 diag::err_typecheck_call_too_many_args) 615 << 0 << NumArgs[Form] << TheCall->getNumArgs() 616 << TheCall->getCallee()->getSourceRange(); 617 return ExprError(); 618 } 619 620 // Inspect the first argument of the atomic operation. 621 Expr *Ptr = TheCall->getArg(0); 622 Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get(); 623 const PointerType *pointerType = Ptr->getType()->getAs<PointerType>(); 624 if (!pointerType) { 625 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 626 << Ptr->getType() << Ptr->getSourceRange(); 627 return ExprError(); 628 } 629 630 // For a __c11 builtin, this should be a pointer to an _Atomic type. 631 QualType AtomTy = pointerType->getPointeeType(); // 'A' 632 QualType ValType = AtomTy; // 'C' 633 if (IsC11) { 634 if (!AtomTy->isAtomicType()) { 635 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 636 << Ptr->getType() << Ptr->getSourceRange(); 637 return ExprError(); 638 } 639 ValType = AtomTy->getAs<AtomicType>()->getValueType(); 640 } 641 642 // For an arithmetic operation, the implied arithmetic must be well-formed. 643 if (Form == Arithmetic) { 644 // gcc does not enforce these rules for GNU atomics, but we do so for sanity. 645 if (IsAddSub && !ValType->isIntegerType() && !ValType->isPointerType()) { 646 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 647 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 648 return ExprError(); 649 } 650 if (!IsAddSub && !ValType->isIntegerType()) { 651 Diag(DRE->getLocStart(), diag::err_atomic_op_bitwise_needs_atomic_int) 652 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 653 return ExprError(); 654 } 655 } else if (IsN && !ValType->isIntegerType() && !ValType->isPointerType()) { 656 // For __atomic_*_n operations, the value type must be a scalar integral or 657 // pointer type which is 1, 2, 4, 8 or 16 bytes in length. 658 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 659 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 660 return ExprError(); 661 } 662 663 if (!IsC11 && !AtomTy.isTriviallyCopyableType(Context)) { 664 // For GNU atomics, require a trivially-copyable type. This is not part of 665 // the GNU atomics specification, but we enforce it for sanity. 666 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_trivial_copy) 667 << Ptr->getType() << Ptr->getSourceRange(); 668 return ExprError(); 669 } 670 671 // FIXME: For any builtin other than a load, the ValType must not be 672 // const-qualified. 673 674 switch (ValType.getObjCLifetime()) { 675 case Qualifiers::OCL_None: 676 case Qualifiers::OCL_ExplicitNone: 677 // okay 678 break; 679 680 case Qualifiers::OCL_Weak: 681 case Qualifiers::OCL_Strong: 682 case Qualifiers::OCL_Autoreleasing: 683 // FIXME: Can this happen? By this point, ValType should be known 684 // to be trivially copyable. 685 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 686 << ValType << Ptr->getSourceRange(); 687 return ExprError(); 688 } 689 690 QualType ResultType = ValType; 691 if (Form == Copy || Form == GNUXchg || Form == Init) 692 ResultType = Context.VoidTy; 693 else if (Form == C11CmpXchg || Form == GNUCmpXchg) 694 ResultType = Context.BoolTy; 695 696 // The type of a parameter passed 'by value'. In the GNU atomics, such 697 // arguments are actually passed as pointers. 698 QualType ByValType = ValType; // 'CP' 699 if (!IsC11 && !IsN) 700 ByValType = Ptr->getType(); 701 702 // The first argument --- the pointer --- has a fixed type; we 703 // deduce the types of the rest of the arguments accordingly. Walk 704 // the remaining arguments, converting them to the deduced value type. 705 for (unsigned i = 1; i != NumArgs[Form]; ++i) { 706 QualType Ty; 707 if (i < NumVals[Form] + 1) { 708 switch (i) { 709 case 1: 710 // The second argument is the non-atomic operand. For arithmetic, this 711 // is always passed by value, and for a compare_exchange it is always 712 // passed by address. For the rest, GNU uses by-address and C11 uses 713 // by-value. 714 assert(Form != Load); 715 if (Form == Init || (Form == Arithmetic && ValType->isIntegerType())) 716 Ty = ValType; 717 else if (Form == Copy || Form == Xchg) 718 Ty = ByValType; 719 else if (Form == Arithmetic) 720 Ty = Context.getPointerDiffType(); 721 else 722 Ty = Context.getPointerType(ValType.getUnqualifiedType()); 723 break; 724 case 2: 725 // The third argument to compare_exchange / GNU exchange is a 726 // (pointer to a) desired value. 727 Ty = ByValType; 728 break; 729 case 3: 730 // The fourth argument to GNU compare_exchange is a 'weak' flag. 731 Ty = Context.BoolTy; 732 break; 733 } 734 } else { 735 // The order(s) are always converted to int. 736 Ty = Context.IntTy; 737 } 738 739 InitializedEntity Entity = 740 InitializedEntity::InitializeParameter(Context, Ty, false); 741 ExprResult Arg = TheCall->getArg(i); 742 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 743 if (Arg.isInvalid()) 744 return true; 745 TheCall->setArg(i, Arg.get()); 746 } 747 748 // Permute the arguments into a 'consistent' order. 749 SmallVector<Expr*, 5> SubExprs; 750 SubExprs.push_back(Ptr); 751 switch (Form) { 752 case Init: 753 // Note, AtomicExpr::getVal1() has a special case for this atomic. 754 SubExprs.push_back(TheCall->getArg(1)); // Val1 755 break; 756 case Load: 757 SubExprs.push_back(TheCall->getArg(1)); // Order 758 break; 759 case Copy: 760 case Arithmetic: 761 case Xchg: 762 SubExprs.push_back(TheCall->getArg(2)); // Order 763 SubExprs.push_back(TheCall->getArg(1)); // Val1 764 break; 765 case GNUXchg: 766 // Note, AtomicExpr::getVal2() has a special case for this atomic. 767 SubExprs.push_back(TheCall->getArg(3)); // Order 768 SubExprs.push_back(TheCall->getArg(1)); // Val1 769 SubExprs.push_back(TheCall->getArg(2)); // Val2 770 break; 771 case C11CmpXchg: 772 SubExprs.push_back(TheCall->getArg(3)); // Order 773 SubExprs.push_back(TheCall->getArg(1)); // Val1 774 SubExprs.push_back(TheCall->getArg(4)); // OrderFail 775 SubExprs.push_back(TheCall->getArg(2)); // Val2 776 break; 777 case GNUCmpXchg: 778 SubExprs.push_back(TheCall->getArg(4)); // Order 779 SubExprs.push_back(TheCall->getArg(1)); // Val1 780 SubExprs.push_back(TheCall->getArg(5)); // OrderFail 781 SubExprs.push_back(TheCall->getArg(2)); // Val2 782 SubExprs.push_back(TheCall->getArg(3)); // Weak 783 break; 784 } 785 786 return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 787 SubExprs.data(), SubExprs.size(), 788 ResultType, Op, 789 TheCall->getRParenLoc())); 790} 791 792 793/// checkBuiltinArgument - Given a call to a builtin function, perform 794/// normal type-checking on the given argument, updating the call in 795/// place. This is useful when a builtin function requires custom 796/// type-checking for some of its arguments but not necessarily all of 797/// them. 798/// 799/// Returns true on error. 800static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 801 FunctionDecl *Fn = E->getDirectCallee(); 802 assert(Fn && "builtin call without direct callee!"); 803 804 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 805 InitializedEntity Entity = 806 InitializedEntity::InitializeParameter(S.Context, Param); 807 808 ExprResult Arg = E->getArg(0); 809 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 810 if (Arg.isInvalid()) 811 return true; 812 813 E->setArg(ArgIndex, Arg.take()); 814 return false; 815} 816 817/// SemaBuiltinAtomicOverloaded - We have a call to a function like 818/// __sync_fetch_and_add, which is an overloaded function based on the pointer 819/// type of its first argument. The main ActOnCallExpr routines have already 820/// promoted the types of arguments because all of these calls are prototyped as 821/// void(...). 822/// 823/// This function goes through and does final semantic checking for these 824/// builtins, 825ExprResult 826Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 827 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 828 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 829 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 830 831 // Ensure that we have at least one argument to do type inference from. 832 if (TheCall->getNumArgs() < 1) { 833 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 834 << 0 << 1 << TheCall->getNumArgs() 835 << TheCall->getCallee()->getSourceRange(); 836 return ExprError(); 837 } 838 839 // Inspect the first argument of the atomic builtin. This should always be 840 // a pointer type, whose element is an integral scalar or pointer type. 841 // Because it is a pointer type, we don't have to worry about any implicit 842 // casts here. 843 // FIXME: We don't allow floating point scalars as input. 844 Expr *FirstArg = TheCall->getArg(0); 845 ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg); 846 if (FirstArgResult.isInvalid()) 847 return ExprError(); 848 FirstArg = FirstArgResult.take(); 849 TheCall->setArg(0, FirstArg); 850 851 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 852 if (!pointerType) { 853 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 854 << FirstArg->getType() << FirstArg->getSourceRange(); 855 return ExprError(); 856 } 857 858 QualType ValType = pointerType->getPointeeType(); 859 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 860 !ValType->isBlockPointerType()) { 861 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 862 << FirstArg->getType() << FirstArg->getSourceRange(); 863 return ExprError(); 864 } 865 866 switch (ValType.getObjCLifetime()) { 867 case Qualifiers::OCL_None: 868 case Qualifiers::OCL_ExplicitNone: 869 // okay 870 break; 871 872 case Qualifiers::OCL_Weak: 873 case Qualifiers::OCL_Strong: 874 case Qualifiers::OCL_Autoreleasing: 875 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 876 << ValType << FirstArg->getSourceRange(); 877 return ExprError(); 878 } 879 880 // Strip any qualifiers off ValType. 881 ValType = ValType.getUnqualifiedType(); 882 883 // The majority of builtins return a value, but a few have special return 884 // types, so allow them to override appropriately below. 885 QualType ResultType = ValType; 886 887 // We need to figure out which concrete builtin this maps onto. For example, 888 // __sync_fetch_and_add with a 2 byte object turns into 889 // __sync_fetch_and_add_2. 890#define BUILTIN_ROW(x) \ 891 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 892 Builtin::BI##x##_8, Builtin::BI##x##_16 } 893 894 static const unsigned BuiltinIndices[][5] = { 895 BUILTIN_ROW(__sync_fetch_and_add), 896 BUILTIN_ROW(__sync_fetch_and_sub), 897 BUILTIN_ROW(__sync_fetch_and_or), 898 BUILTIN_ROW(__sync_fetch_and_and), 899 BUILTIN_ROW(__sync_fetch_and_xor), 900 901 BUILTIN_ROW(__sync_add_and_fetch), 902 BUILTIN_ROW(__sync_sub_and_fetch), 903 BUILTIN_ROW(__sync_and_and_fetch), 904 BUILTIN_ROW(__sync_or_and_fetch), 905 BUILTIN_ROW(__sync_xor_and_fetch), 906 907 BUILTIN_ROW(__sync_val_compare_and_swap), 908 BUILTIN_ROW(__sync_bool_compare_and_swap), 909 BUILTIN_ROW(__sync_lock_test_and_set), 910 BUILTIN_ROW(__sync_lock_release), 911 BUILTIN_ROW(__sync_swap) 912 }; 913#undef BUILTIN_ROW 914 915 // Determine the index of the size. 916 unsigned SizeIndex; 917 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 918 case 1: SizeIndex = 0; break; 919 case 2: SizeIndex = 1; break; 920 case 4: SizeIndex = 2; break; 921 case 8: SizeIndex = 3; break; 922 case 16: SizeIndex = 4; break; 923 default: 924 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 925 << FirstArg->getType() << FirstArg->getSourceRange(); 926 return ExprError(); 927 } 928 929 // Each of these builtins has one pointer argument, followed by some number of 930 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 931 // that we ignore. Find out which row of BuiltinIndices to read from as well 932 // as the number of fixed args. 933 unsigned BuiltinID = FDecl->getBuiltinID(); 934 unsigned BuiltinIndex, NumFixed = 1; 935 switch (BuiltinID) { 936 default: llvm_unreachable("Unknown overloaded atomic builtin!"); 937 case Builtin::BI__sync_fetch_and_add: 938 case Builtin::BI__sync_fetch_and_add_1: 939 case Builtin::BI__sync_fetch_and_add_2: 940 case Builtin::BI__sync_fetch_and_add_4: 941 case Builtin::BI__sync_fetch_and_add_8: 942 case Builtin::BI__sync_fetch_and_add_16: 943 BuiltinIndex = 0; 944 break; 945 946 case Builtin::BI__sync_fetch_and_sub: 947 case Builtin::BI__sync_fetch_and_sub_1: 948 case Builtin::BI__sync_fetch_and_sub_2: 949 case Builtin::BI__sync_fetch_and_sub_4: 950 case Builtin::BI__sync_fetch_and_sub_8: 951 case Builtin::BI__sync_fetch_and_sub_16: 952 BuiltinIndex = 1; 953 break; 954 955 case Builtin::BI__sync_fetch_and_or: 956 case Builtin::BI__sync_fetch_and_or_1: 957 case Builtin::BI__sync_fetch_and_or_2: 958 case Builtin::BI__sync_fetch_and_or_4: 959 case Builtin::BI__sync_fetch_and_or_8: 960 case Builtin::BI__sync_fetch_and_or_16: 961 BuiltinIndex = 2; 962 break; 963 964 case Builtin::BI__sync_fetch_and_and: 965 case Builtin::BI__sync_fetch_and_and_1: 966 case Builtin::BI__sync_fetch_and_and_2: 967 case Builtin::BI__sync_fetch_and_and_4: 968 case Builtin::BI__sync_fetch_and_and_8: 969 case Builtin::BI__sync_fetch_and_and_16: 970 BuiltinIndex = 3; 971 break; 972 973 case Builtin::BI__sync_fetch_and_xor: 974 case Builtin::BI__sync_fetch_and_xor_1: 975 case Builtin::BI__sync_fetch_and_xor_2: 976 case Builtin::BI__sync_fetch_and_xor_4: 977 case Builtin::BI__sync_fetch_and_xor_8: 978 case Builtin::BI__sync_fetch_and_xor_16: 979 BuiltinIndex = 4; 980 break; 981 982 case Builtin::BI__sync_add_and_fetch: 983 case Builtin::BI__sync_add_and_fetch_1: 984 case Builtin::BI__sync_add_and_fetch_2: 985 case Builtin::BI__sync_add_and_fetch_4: 986 case Builtin::BI__sync_add_and_fetch_8: 987 case Builtin::BI__sync_add_and_fetch_16: 988 BuiltinIndex = 5; 989 break; 990 991 case Builtin::BI__sync_sub_and_fetch: 992 case Builtin::BI__sync_sub_and_fetch_1: 993 case Builtin::BI__sync_sub_and_fetch_2: 994 case Builtin::BI__sync_sub_and_fetch_4: 995 case Builtin::BI__sync_sub_and_fetch_8: 996 case Builtin::BI__sync_sub_and_fetch_16: 997 BuiltinIndex = 6; 998 break; 999 1000 case Builtin::BI__sync_and_and_fetch: 1001 case Builtin::BI__sync_and_and_fetch_1: 1002 case Builtin::BI__sync_and_and_fetch_2: 1003 case Builtin::BI__sync_and_and_fetch_4: 1004 case Builtin::BI__sync_and_and_fetch_8: 1005 case Builtin::BI__sync_and_and_fetch_16: 1006 BuiltinIndex = 7; 1007 break; 1008 1009 case Builtin::BI__sync_or_and_fetch: 1010 case Builtin::BI__sync_or_and_fetch_1: 1011 case Builtin::BI__sync_or_and_fetch_2: 1012 case Builtin::BI__sync_or_and_fetch_4: 1013 case Builtin::BI__sync_or_and_fetch_8: 1014 case Builtin::BI__sync_or_and_fetch_16: 1015 BuiltinIndex = 8; 1016 break; 1017 1018 case Builtin::BI__sync_xor_and_fetch: 1019 case Builtin::BI__sync_xor_and_fetch_1: 1020 case Builtin::BI__sync_xor_and_fetch_2: 1021 case Builtin::BI__sync_xor_and_fetch_4: 1022 case Builtin::BI__sync_xor_and_fetch_8: 1023 case Builtin::BI__sync_xor_and_fetch_16: 1024 BuiltinIndex = 9; 1025 break; 1026 1027 case Builtin::BI__sync_val_compare_and_swap: 1028 case Builtin::BI__sync_val_compare_and_swap_1: 1029 case Builtin::BI__sync_val_compare_and_swap_2: 1030 case Builtin::BI__sync_val_compare_and_swap_4: 1031 case Builtin::BI__sync_val_compare_and_swap_8: 1032 case Builtin::BI__sync_val_compare_and_swap_16: 1033 BuiltinIndex = 10; 1034 NumFixed = 2; 1035 break; 1036 1037 case Builtin::BI__sync_bool_compare_and_swap: 1038 case Builtin::BI__sync_bool_compare_and_swap_1: 1039 case Builtin::BI__sync_bool_compare_and_swap_2: 1040 case Builtin::BI__sync_bool_compare_and_swap_4: 1041 case Builtin::BI__sync_bool_compare_and_swap_8: 1042 case Builtin::BI__sync_bool_compare_and_swap_16: 1043 BuiltinIndex = 11; 1044 NumFixed = 2; 1045 ResultType = Context.BoolTy; 1046 break; 1047 1048 case Builtin::BI__sync_lock_test_and_set: 1049 case Builtin::BI__sync_lock_test_and_set_1: 1050 case Builtin::BI__sync_lock_test_and_set_2: 1051 case Builtin::BI__sync_lock_test_and_set_4: 1052 case Builtin::BI__sync_lock_test_and_set_8: 1053 case Builtin::BI__sync_lock_test_and_set_16: 1054 BuiltinIndex = 12; 1055 break; 1056 1057 case Builtin::BI__sync_lock_release: 1058 case Builtin::BI__sync_lock_release_1: 1059 case Builtin::BI__sync_lock_release_2: 1060 case Builtin::BI__sync_lock_release_4: 1061 case Builtin::BI__sync_lock_release_8: 1062 case Builtin::BI__sync_lock_release_16: 1063 BuiltinIndex = 13; 1064 NumFixed = 0; 1065 ResultType = Context.VoidTy; 1066 break; 1067 1068 case Builtin::BI__sync_swap: 1069 case Builtin::BI__sync_swap_1: 1070 case Builtin::BI__sync_swap_2: 1071 case Builtin::BI__sync_swap_4: 1072 case Builtin::BI__sync_swap_8: 1073 case Builtin::BI__sync_swap_16: 1074 BuiltinIndex = 14; 1075 break; 1076 } 1077 1078 // Now that we know how many fixed arguments we expect, first check that we 1079 // have at least that many. 1080 if (TheCall->getNumArgs() < 1+NumFixed) { 1081 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 1082 << 0 << 1+NumFixed << TheCall->getNumArgs() 1083 << TheCall->getCallee()->getSourceRange(); 1084 return ExprError(); 1085 } 1086 1087 // Get the decl for the concrete builtin from this, we can tell what the 1088 // concrete integer type we should convert to is. 1089 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 1090 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 1091 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 1092 FunctionDecl *NewBuiltinDecl = 1093 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 1094 TUScope, false, DRE->getLocStart())); 1095 1096 // The first argument --- the pointer --- has a fixed type; we 1097 // deduce the types of the rest of the arguments accordingly. Walk 1098 // the remaining arguments, converting them to the deduced value type. 1099 for (unsigned i = 0; i != NumFixed; ++i) { 1100 ExprResult Arg = TheCall->getArg(i+1); 1101 1102 // GCC does an implicit conversion to the pointer or integer ValType. This 1103 // can fail in some cases (1i -> int**), check for this error case now. 1104 // Initialize the argument. 1105 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context, 1106 ValType, /*consume*/ false); 1107 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 1108 if (Arg.isInvalid()) 1109 return ExprError(); 1110 1111 // Okay, we have something that *can* be converted to the right type. Check 1112 // to see if there is a potentially weird extension going on here. This can 1113 // happen when you do an atomic operation on something like an char* and 1114 // pass in 42. The 42 gets converted to char. This is even more strange 1115 // for things like 45.123 -> char, etc. 1116 // FIXME: Do this check. 1117 TheCall->setArg(i+1, Arg.take()); 1118 } 1119 1120 ASTContext& Context = this->getASTContext(); 1121 1122 // Create a new DeclRefExpr to refer to the new decl. 1123 DeclRefExpr* NewDRE = DeclRefExpr::Create( 1124 Context, 1125 DRE->getQualifierLoc(), 1126 SourceLocation(), 1127 NewBuiltinDecl, 1128 /*enclosing*/ false, 1129 DRE->getLocation(), 1130 NewBuiltinDecl->getType(), 1131 DRE->getValueKind()); 1132 1133 // Set the callee in the CallExpr. 1134 // FIXME: This leaks the original parens and implicit casts. 1135 ExprResult PromotedCall = UsualUnaryConversions(NewDRE); 1136 if (PromotedCall.isInvalid()) 1137 return ExprError(); 1138 TheCall->setCallee(PromotedCall.take()); 1139 1140 // Change the result type of the call to match the original value type. This 1141 // is arbitrary, but the codegen for these builtins ins design to handle it 1142 // gracefully. 1143 TheCall->setType(ResultType); 1144 1145 return move(TheCallResult); 1146} 1147 1148/// CheckObjCString - Checks that the argument to the builtin 1149/// CFString constructor is correct 1150/// Note: It might also make sense to do the UTF-16 conversion here (would 1151/// simplify the backend). 1152bool Sema::CheckObjCString(Expr *Arg) { 1153 Arg = Arg->IgnoreParenCasts(); 1154 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 1155 1156 if (!Literal || !Literal->isAscii()) { 1157 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 1158 << Arg->getSourceRange(); 1159 return true; 1160 } 1161 1162 if (Literal->containsNonAsciiOrNull()) { 1163 StringRef String = Literal->getString(); 1164 unsigned NumBytes = String.size(); 1165 SmallVector<UTF16, 128> ToBuf(NumBytes); 1166 const UTF8 *FromPtr = (UTF8 *)String.data(); 1167 UTF16 *ToPtr = &ToBuf[0]; 1168 1169 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 1170 &ToPtr, ToPtr + NumBytes, 1171 strictConversion); 1172 // Check for conversion failure. 1173 if (Result != conversionOK) 1174 Diag(Arg->getLocStart(), 1175 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 1176 } 1177 return false; 1178} 1179 1180/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 1181/// Emit an error and return true on failure, return false on success. 1182bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 1183 Expr *Fn = TheCall->getCallee(); 1184 if (TheCall->getNumArgs() > 2) { 1185 Diag(TheCall->getArg(2)->getLocStart(), 1186 diag::err_typecheck_call_too_many_args) 1187 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1188 << Fn->getSourceRange() 1189 << SourceRange(TheCall->getArg(2)->getLocStart(), 1190 (*(TheCall->arg_end()-1))->getLocEnd()); 1191 return true; 1192 } 1193 1194 if (TheCall->getNumArgs() < 2) { 1195 return Diag(TheCall->getLocEnd(), 1196 diag::err_typecheck_call_too_few_args_at_least) 1197 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 1198 } 1199 1200 // Type-check the first argument normally. 1201 if (checkBuiltinArgument(*this, TheCall, 0)) 1202 return true; 1203 1204 // Determine whether the current function is variadic or not. 1205 BlockScopeInfo *CurBlock = getCurBlock(); 1206 bool isVariadic; 1207 if (CurBlock) 1208 isVariadic = CurBlock->TheDecl->isVariadic(); 1209 else if (FunctionDecl *FD = getCurFunctionDecl()) 1210 isVariadic = FD->isVariadic(); 1211 else 1212 isVariadic = getCurMethodDecl()->isVariadic(); 1213 1214 if (!isVariadic) { 1215 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 1216 return true; 1217 } 1218 1219 // Verify that the second argument to the builtin is the last argument of the 1220 // current function or method. 1221 bool SecondArgIsLastNamedArgument = false; 1222 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 1223 1224 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 1225 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 1226 // FIXME: This isn't correct for methods (results in bogus warning). 1227 // Get the last formal in the current function. 1228 const ParmVarDecl *LastArg; 1229 if (CurBlock) 1230 LastArg = *(CurBlock->TheDecl->param_end()-1); 1231 else if (FunctionDecl *FD = getCurFunctionDecl()) 1232 LastArg = *(FD->param_end()-1); 1233 else 1234 LastArg = *(getCurMethodDecl()->param_end()-1); 1235 SecondArgIsLastNamedArgument = PV == LastArg; 1236 } 1237 } 1238 1239 if (!SecondArgIsLastNamedArgument) 1240 Diag(TheCall->getArg(1)->getLocStart(), 1241 diag::warn_second_parameter_of_va_start_not_last_named_argument); 1242 return false; 1243} 1244 1245/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 1246/// friends. This is declared to take (...), so we have to check everything. 1247bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 1248 if (TheCall->getNumArgs() < 2) 1249 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1250 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 1251 if (TheCall->getNumArgs() > 2) 1252 return Diag(TheCall->getArg(2)->getLocStart(), 1253 diag::err_typecheck_call_too_many_args) 1254 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1255 << SourceRange(TheCall->getArg(2)->getLocStart(), 1256 (*(TheCall->arg_end()-1))->getLocEnd()); 1257 1258 ExprResult OrigArg0 = TheCall->getArg(0); 1259 ExprResult OrigArg1 = TheCall->getArg(1); 1260 1261 // Do standard promotions between the two arguments, returning their common 1262 // type. 1263 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 1264 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 1265 return true; 1266 1267 // Make sure any conversions are pushed back into the call; this is 1268 // type safe since unordered compare builtins are declared as "_Bool 1269 // foo(...)". 1270 TheCall->setArg(0, OrigArg0.get()); 1271 TheCall->setArg(1, OrigArg1.get()); 1272 1273 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 1274 return false; 1275 1276 // If the common type isn't a real floating type, then the arguments were 1277 // invalid for this operation. 1278 if (!Res->isRealFloatingType()) 1279 return Diag(OrigArg0.get()->getLocStart(), 1280 diag::err_typecheck_call_invalid_ordered_compare) 1281 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 1282 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 1283 1284 return false; 1285} 1286 1287/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 1288/// __builtin_isnan and friends. This is declared to take (...), so we have 1289/// to check everything. We expect the last argument to be a floating point 1290/// value. 1291bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 1292 if (TheCall->getNumArgs() < NumArgs) 1293 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1294 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 1295 if (TheCall->getNumArgs() > NumArgs) 1296 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 1297 diag::err_typecheck_call_too_many_args) 1298 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 1299 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 1300 (*(TheCall->arg_end()-1))->getLocEnd()); 1301 1302 Expr *OrigArg = TheCall->getArg(NumArgs-1); 1303 1304 if (OrigArg->isTypeDependent()) 1305 return false; 1306 1307 // This operation requires a non-_Complex floating-point number. 1308 if (!OrigArg->getType()->isRealFloatingType()) 1309 return Diag(OrigArg->getLocStart(), 1310 diag::err_typecheck_call_invalid_unary_fp) 1311 << OrigArg->getType() << OrigArg->getSourceRange(); 1312 1313 // If this is an implicit conversion from float -> double, remove it. 1314 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 1315 Expr *CastArg = Cast->getSubExpr(); 1316 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 1317 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 1318 "promotion from float to double is the only expected cast here"); 1319 Cast->setSubExpr(0); 1320 TheCall->setArg(NumArgs-1, CastArg); 1321 } 1322 } 1323 1324 return false; 1325} 1326 1327/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1328// This is declared to take (...), so we have to check everything. 1329ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1330 if (TheCall->getNumArgs() < 2) 1331 return ExprError(Diag(TheCall->getLocEnd(), 1332 diag::err_typecheck_call_too_few_args_at_least) 1333 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1334 << TheCall->getSourceRange()); 1335 1336 // Determine which of the following types of shufflevector we're checking: 1337 // 1) unary, vector mask: (lhs, mask) 1338 // 2) binary, vector mask: (lhs, rhs, mask) 1339 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1340 QualType resType = TheCall->getArg(0)->getType(); 1341 unsigned numElements = 0; 1342 1343 if (!TheCall->getArg(0)->isTypeDependent() && 1344 !TheCall->getArg(1)->isTypeDependent()) { 1345 QualType LHSType = TheCall->getArg(0)->getType(); 1346 QualType RHSType = TheCall->getArg(1)->getType(); 1347 1348 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 1349 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 1350 << SourceRange(TheCall->getArg(0)->getLocStart(), 1351 TheCall->getArg(1)->getLocEnd()); 1352 return ExprError(); 1353 } 1354 1355 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1356 unsigned numResElements = TheCall->getNumArgs() - 2; 1357 1358 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1359 // with mask. If so, verify that RHS is an integer vector type with the 1360 // same number of elts as lhs. 1361 if (TheCall->getNumArgs() == 2) { 1362 if (!RHSType->hasIntegerRepresentation() || 1363 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1364 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1365 << SourceRange(TheCall->getArg(1)->getLocStart(), 1366 TheCall->getArg(1)->getLocEnd()); 1367 numResElements = numElements; 1368 } 1369 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1370 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1371 << SourceRange(TheCall->getArg(0)->getLocStart(), 1372 TheCall->getArg(1)->getLocEnd()); 1373 return ExprError(); 1374 } else if (numElements != numResElements) { 1375 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1376 resType = Context.getVectorType(eltType, numResElements, 1377 VectorType::GenericVector); 1378 } 1379 } 1380 1381 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1382 if (TheCall->getArg(i)->isTypeDependent() || 1383 TheCall->getArg(i)->isValueDependent()) 1384 continue; 1385 1386 llvm::APSInt Result(32); 1387 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1388 return ExprError(Diag(TheCall->getLocStart(), 1389 diag::err_shufflevector_nonconstant_argument) 1390 << TheCall->getArg(i)->getSourceRange()); 1391 1392 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1393 return ExprError(Diag(TheCall->getLocStart(), 1394 diag::err_shufflevector_argument_too_large) 1395 << TheCall->getArg(i)->getSourceRange()); 1396 } 1397 1398 SmallVector<Expr*, 32> exprs; 1399 1400 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1401 exprs.push_back(TheCall->getArg(i)); 1402 TheCall->setArg(i, 0); 1403 } 1404 1405 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 1406 exprs.size(), resType, 1407 TheCall->getCallee()->getLocStart(), 1408 TheCall->getRParenLoc())); 1409} 1410 1411/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1412// This is declared to take (const void*, ...) and can take two 1413// optional constant int args. 1414bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1415 unsigned NumArgs = TheCall->getNumArgs(); 1416 1417 if (NumArgs > 3) 1418 return Diag(TheCall->getLocEnd(), 1419 diag::err_typecheck_call_too_many_args_at_most) 1420 << 0 /*function call*/ << 3 << NumArgs 1421 << TheCall->getSourceRange(); 1422 1423 // Argument 0 is checked for us and the remaining arguments must be 1424 // constant integers. 1425 for (unsigned i = 1; i != NumArgs; ++i) { 1426 Expr *Arg = TheCall->getArg(i); 1427 1428 llvm::APSInt Result; 1429 if (SemaBuiltinConstantArg(TheCall, i, Result)) 1430 return true; 1431 1432 // FIXME: gcc issues a warning and rewrites these to 0. These 1433 // seems especially odd for the third argument since the default 1434 // is 3. 1435 if (i == 1) { 1436 if (Result.getLimitedValue() > 1) 1437 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1438 << "0" << "1" << Arg->getSourceRange(); 1439 } else { 1440 if (Result.getLimitedValue() > 3) 1441 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1442 << "0" << "3" << Arg->getSourceRange(); 1443 } 1444 } 1445 1446 return false; 1447} 1448 1449/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1450/// TheCall is a constant expression. 1451bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1452 llvm::APSInt &Result) { 1453 Expr *Arg = TheCall->getArg(ArgNum); 1454 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1455 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1456 1457 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 1458 1459 if (!Arg->isIntegerConstantExpr(Result, Context)) 1460 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 1461 << FDecl->getDeclName() << Arg->getSourceRange(); 1462 1463 return false; 1464} 1465 1466/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 1467/// int type). This simply type checks that type is one of the defined 1468/// constants (0-3). 1469// For compatibility check 0-3, llvm only handles 0 and 2. 1470bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 1471 llvm::APSInt Result; 1472 1473 // Check constant-ness first. 1474 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1475 return true; 1476 1477 Expr *Arg = TheCall->getArg(1); 1478 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 1479 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1480 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1481 } 1482 1483 return false; 1484} 1485 1486/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1487/// This checks that val is a constant 1. 1488bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1489 Expr *Arg = TheCall->getArg(1); 1490 llvm::APSInt Result; 1491 1492 // TODO: This is less than ideal. Overload this to take a value. 1493 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1494 return true; 1495 1496 if (Result != 1) 1497 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1498 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1499 1500 return false; 1501} 1502 1503// Handle i > 1 ? "x" : "y", recursively. 1504bool Sema::SemaCheckStringLiteral(const Expr *E, Expr **Args, 1505 unsigned NumArgs, bool HasVAListArg, 1506 unsigned format_idx, unsigned firstDataArg, 1507 FormatStringType Type, bool inFunctionCall) { 1508 tryAgain: 1509 if (E->isTypeDependent() || E->isValueDependent()) 1510 return false; 1511 1512 E = E->IgnoreParenCasts(); 1513 1514 if (E->isNullPointerConstant(Context, Expr::NPC_ValueDependentIsNotNull)) 1515 // Technically -Wformat-nonliteral does not warn about this case. 1516 // The behavior of printf and friends in this case is implementation 1517 // dependent. Ideally if the format string cannot be null then 1518 // it should have a 'nonnull' attribute in the function prototype. 1519 return true; 1520 1521 switch (E->getStmtClass()) { 1522 case Stmt::BinaryConditionalOperatorClass: 1523 case Stmt::ConditionalOperatorClass: { 1524 const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E); 1525 return SemaCheckStringLiteral(C->getTrueExpr(), Args, NumArgs, HasVAListArg, 1526 format_idx, firstDataArg, Type, 1527 inFunctionCall) 1528 && SemaCheckStringLiteral(C->getFalseExpr(), Args, NumArgs, HasVAListArg, 1529 format_idx, firstDataArg, Type, 1530 inFunctionCall); 1531 } 1532 1533 case Stmt::ImplicitCastExprClass: { 1534 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 1535 goto tryAgain; 1536 } 1537 1538 case Stmt::OpaqueValueExprClass: 1539 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 1540 E = src; 1541 goto tryAgain; 1542 } 1543 return false; 1544 1545 case Stmt::PredefinedExprClass: 1546 // While __func__, etc., are technically not string literals, they 1547 // cannot contain format specifiers and thus are not a security 1548 // liability. 1549 return true; 1550 1551 case Stmt::DeclRefExprClass: { 1552 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1553 1554 // As an exception, do not flag errors for variables binding to 1555 // const string literals. 1556 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1557 bool isConstant = false; 1558 QualType T = DR->getType(); 1559 1560 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1561 isConstant = AT->getElementType().isConstant(Context); 1562 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1563 isConstant = T.isConstant(Context) && 1564 PT->getPointeeType().isConstant(Context); 1565 } else if (T->isObjCObjectPointerType()) { 1566 // In ObjC, there is usually no "const ObjectPointer" type, 1567 // so don't check if the pointee type is constant. 1568 isConstant = T.isConstant(Context); 1569 } 1570 1571 if (isConstant) { 1572 if (const Expr *Init = VD->getAnyInitializer()) { 1573 // Look through initializers like const char c[] = { "foo" } 1574 if (const InitListExpr *InitList = dyn_cast<InitListExpr>(Init)) { 1575 if (InitList->isStringLiteralInit()) 1576 Init = InitList->getInit(0)->IgnoreParenImpCasts(); 1577 } 1578 return SemaCheckStringLiteral(Init, Args, NumArgs, 1579 HasVAListArg, format_idx, firstDataArg, 1580 Type, /*inFunctionCall*/false); 1581 } 1582 } 1583 1584 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1585 // special check to see if the format string is a function parameter 1586 // of the function calling the printf function. If the function 1587 // has an attribute indicating it is a printf-like function, then we 1588 // should suppress warnings concerning non-literals being used in a call 1589 // to a vprintf function. For example: 1590 // 1591 // void 1592 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1593 // va_list ap; 1594 // va_start(ap, fmt); 1595 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1596 // ... 1597 // 1598 if (HasVAListArg) { 1599 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(VD)) { 1600 if (const NamedDecl *ND = dyn_cast<NamedDecl>(PV->getDeclContext())) { 1601 int PVIndex = PV->getFunctionScopeIndex() + 1; 1602 for (specific_attr_iterator<FormatAttr> 1603 i = ND->specific_attr_begin<FormatAttr>(), 1604 e = ND->specific_attr_end<FormatAttr>(); i != e ; ++i) { 1605 FormatAttr *PVFormat = *i; 1606 // adjust for implicit parameter 1607 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1608 if (MD->isInstance()) 1609 ++PVIndex; 1610 // We also check if the formats are compatible. 1611 // We can't pass a 'scanf' string to a 'printf' function. 1612 if (PVIndex == PVFormat->getFormatIdx() && 1613 Type == GetFormatStringType(PVFormat)) 1614 return true; 1615 } 1616 } 1617 } 1618 } 1619 } 1620 1621 return false; 1622 } 1623 1624 case Stmt::CallExprClass: 1625 case Stmt::CXXMemberCallExprClass: { 1626 const CallExpr *CE = cast<CallExpr>(E); 1627 if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(CE->getCalleeDecl())) { 1628 if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) { 1629 unsigned ArgIndex = FA->getFormatIdx(); 1630 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1631 if (MD->isInstance()) 1632 --ArgIndex; 1633 const Expr *Arg = CE->getArg(ArgIndex - 1); 1634 1635 return SemaCheckStringLiteral(Arg, Args, NumArgs, HasVAListArg, 1636 format_idx, firstDataArg, Type, 1637 inFunctionCall); 1638 } 1639 } 1640 1641 return false; 1642 } 1643 case Stmt::ObjCStringLiteralClass: 1644 case Stmt::StringLiteralClass: { 1645 const StringLiteral *StrE = NULL; 1646 1647 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1648 StrE = ObjCFExpr->getString(); 1649 else 1650 StrE = cast<StringLiteral>(E); 1651 1652 if (StrE) { 1653 CheckFormatString(StrE, E, Args, NumArgs, HasVAListArg, format_idx, 1654 firstDataArg, Type, inFunctionCall); 1655 return true; 1656 } 1657 1658 return false; 1659 } 1660 1661 default: 1662 return false; 1663 } 1664} 1665 1666void 1667Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1668 const Expr * const *ExprArgs, 1669 SourceLocation CallSiteLoc) { 1670 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1671 e = NonNull->args_end(); 1672 i != e; ++i) { 1673 const Expr *ArgExpr = ExprArgs[*i]; 1674 if (ArgExpr->isNullPointerConstant(Context, 1675 Expr::NPC_ValueDependentIsNotNull)) 1676 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 1677 } 1678} 1679 1680Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) { 1681 return llvm::StringSwitch<FormatStringType>(Format->getType()) 1682 .Case("scanf", FST_Scanf) 1683 .Cases("printf", "printf0", FST_Printf) 1684 .Cases("NSString", "CFString", FST_NSString) 1685 .Case("strftime", FST_Strftime) 1686 .Case("strfmon", FST_Strfmon) 1687 .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf) 1688 .Default(FST_Unknown); 1689} 1690 1691/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar 1692/// functions) for correct use of format strings. 1693void Sema::CheckFormatArguments(const FormatAttr *Format, CallExpr *TheCall) { 1694 bool IsCXXMember = false; 1695 // The way the format attribute works in GCC, the implicit this argument 1696 // of member functions is counted. However, it doesn't appear in our own 1697 // lists, so decrement format_idx in that case. 1698 IsCXXMember = isa<CXXMemberCallExpr>(TheCall); 1699 CheckFormatArguments(Format, TheCall->getArgs(), TheCall->getNumArgs(), 1700 IsCXXMember, TheCall->getRParenLoc(), 1701 TheCall->getCallee()->getSourceRange()); 1702} 1703 1704void Sema::CheckFormatArguments(const FormatAttr *Format, Expr **Args, 1705 unsigned NumArgs, bool IsCXXMember, 1706 SourceLocation Loc, SourceRange Range) { 1707 bool HasVAListArg = Format->getFirstArg() == 0; 1708 unsigned format_idx = Format->getFormatIdx() - 1; 1709 unsigned firstDataArg = HasVAListArg ? 0 : Format->getFirstArg() - 1; 1710 if (IsCXXMember) { 1711 if (format_idx == 0) 1712 return; 1713 --format_idx; 1714 if(firstDataArg != 0) 1715 --firstDataArg; 1716 } 1717 CheckFormatArguments(Args, NumArgs, HasVAListArg, format_idx, 1718 firstDataArg, GetFormatStringType(Format), Loc, Range); 1719} 1720 1721void Sema::CheckFormatArguments(Expr **Args, unsigned NumArgs, 1722 bool HasVAListArg, unsigned format_idx, 1723 unsigned firstDataArg, FormatStringType Type, 1724 SourceLocation Loc, SourceRange Range) { 1725 // CHECK: printf/scanf-like function is called with no format string. 1726 if (format_idx >= NumArgs) { 1727 Diag(Loc, diag::warn_missing_format_string) << Range; 1728 return; 1729 } 1730 1731 const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts(); 1732 1733 // CHECK: format string is not a string literal. 1734 // 1735 // Dynamically generated format strings are difficult to 1736 // automatically vet at compile time. Requiring that format strings 1737 // are string literals: (1) permits the checking of format strings by 1738 // the compiler and thereby (2) can practically remove the source of 1739 // many format string exploits. 1740 1741 // Format string can be either ObjC string (e.g. @"%d") or 1742 // C string (e.g. "%d") 1743 // ObjC string uses the same format specifiers as C string, so we can use 1744 // the same format string checking logic for both ObjC and C strings. 1745 if (SemaCheckStringLiteral(OrigFormatExpr, Args, NumArgs, HasVAListArg, 1746 format_idx, firstDataArg, Type)) 1747 return; // Literal format string found, check done! 1748 1749 // Strftime is particular as it always uses a single 'time' argument, 1750 // so it is safe to pass a non-literal string. 1751 if (Type == FST_Strftime) 1752 return; 1753 1754 // Do not emit diag when the string param is a macro expansion and the 1755 // format is either NSString or CFString. This is a hack to prevent 1756 // diag when using the NSLocalizedString and CFCopyLocalizedString macros 1757 // which are usually used in place of NS and CF string literals. 1758 if (Type == FST_NSString && 1759 SourceMgr.isInSystemMacro(Args[format_idx]->getLocStart())) 1760 return; 1761 1762 // If there are no arguments specified, warn with -Wformat-security, otherwise 1763 // warn only with -Wformat-nonliteral. 1764 if (NumArgs == format_idx+1) 1765 Diag(Args[format_idx]->getLocStart(), 1766 diag::warn_format_nonliteral_noargs) 1767 << OrigFormatExpr->getSourceRange(); 1768 else 1769 Diag(Args[format_idx]->getLocStart(), 1770 diag::warn_format_nonliteral) 1771 << OrigFormatExpr->getSourceRange(); 1772} 1773 1774namespace { 1775class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1776protected: 1777 Sema &S; 1778 const StringLiteral *FExpr; 1779 const Expr *OrigFormatExpr; 1780 const unsigned FirstDataArg; 1781 const unsigned NumDataArgs; 1782 const bool IsObjCLiteral; 1783 const char *Beg; // Start of format string. 1784 const bool HasVAListArg; 1785 const Expr * const *Args; 1786 const unsigned NumArgs; 1787 unsigned FormatIdx; 1788 llvm::BitVector CoveredArgs; 1789 bool usesPositionalArgs; 1790 bool atFirstArg; 1791 bool inFunctionCall; 1792public: 1793 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1794 const Expr *origFormatExpr, unsigned firstDataArg, 1795 unsigned numDataArgs, bool isObjCLiteral, 1796 const char *beg, bool hasVAListArg, 1797 Expr **args, unsigned numArgs, 1798 unsigned formatIdx, bool inFunctionCall) 1799 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1800 FirstDataArg(firstDataArg), 1801 NumDataArgs(numDataArgs), 1802 IsObjCLiteral(isObjCLiteral), Beg(beg), 1803 HasVAListArg(hasVAListArg), 1804 Args(args), NumArgs(numArgs), FormatIdx(formatIdx), 1805 usesPositionalArgs(false), atFirstArg(true), 1806 inFunctionCall(inFunctionCall) { 1807 CoveredArgs.resize(numDataArgs); 1808 CoveredArgs.reset(); 1809 } 1810 1811 void DoneProcessing(); 1812 1813 void HandleIncompleteSpecifier(const char *startSpecifier, 1814 unsigned specifierLen); 1815 1816 void HandleNonStandardLengthModifier( 1817 const analyze_format_string::LengthModifier &LM, 1818 const char *startSpecifier, unsigned specifierLen); 1819 1820 void HandleNonStandardConversionSpecifier( 1821 const analyze_format_string::ConversionSpecifier &CS, 1822 const char *startSpecifier, unsigned specifierLen); 1823 1824 void HandleNonStandardConversionSpecification( 1825 const analyze_format_string::LengthModifier &LM, 1826 const analyze_format_string::ConversionSpecifier &CS, 1827 const char *startSpecifier, unsigned specifierLen); 1828 1829 virtual void HandlePosition(const char *startPos, unsigned posLen); 1830 1831 virtual void HandleInvalidPosition(const char *startSpecifier, 1832 unsigned specifierLen, 1833 analyze_format_string::PositionContext p); 1834 1835 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1836 1837 void HandleNullChar(const char *nullCharacter); 1838 1839 template <typename Range> 1840 static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall, 1841 const Expr *ArgumentExpr, 1842 PartialDiagnostic PDiag, 1843 SourceLocation StringLoc, 1844 bool IsStringLocation, Range StringRange, 1845 FixItHint Fixit = FixItHint()); 1846 1847protected: 1848 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1849 const char *startSpec, 1850 unsigned specifierLen, 1851 const char *csStart, unsigned csLen); 1852 1853 void HandlePositionalNonpositionalArgs(SourceLocation Loc, 1854 const char *startSpec, 1855 unsigned specifierLen); 1856 1857 SourceRange getFormatStringRange(); 1858 CharSourceRange getSpecifierRange(const char *startSpecifier, 1859 unsigned specifierLen); 1860 SourceLocation getLocationOfByte(const char *x); 1861 1862 const Expr *getDataArg(unsigned i) const; 1863 1864 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1865 const analyze_format_string::ConversionSpecifier &CS, 1866 const char *startSpecifier, unsigned specifierLen, 1867 unsigned argIndex); 1868 1869 template <typename Range> 1870 void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc, 1871 bool IsStringLocation, Range StringRange, 1872 FixItHint Fixit = FixItHint()); 1873 1874 void CheckPositionalAndNonpositionalArgs( 1875 const analyze_format_string::FormatSpecifier *FS); 1876}; 1877} 1878 1879SourceRange CheckFormatHandler::getFormatStringRange() { 1880 return OrigFormatExpr->getSourceRange(); 1881} 1882 1883CharSourceRange CheckFormatHandler:: 1884getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1885 SourceLocation Start = getLocationOfByte(startSpecifier); 1886 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1887 1888 // Advance the end SourceLocation by one due to half-open ranges. 1889 End = End.getLocWithOffset(1); 1890 1891 return CharSourceRange::getCharRange(Start, End); 1892} 1893 1894SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 1895 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1896} 1897 1898void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 1899 unsigned specifierLen){ 1900 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier), 1901 getLocationOfByte(startSpecifier), 1902 /*IsStringLocation*/true, 1903 getSpecifierRange(startSpecifier, specifierLen)); 1904} 1905 1906void CheckFormatHandler::HandleNonStandardLengthModifier( 1907 const analyze_format_string::LengthModifier &LM, 1908 const char *startSpecifier, unsigned specifierLen) { 1909 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) << LM.toString() 1910 << 0, 1911 getLocationOfByte(LM.getStart()), 1912 /*IsStringLocation*/true, 1913 getSpecifierRange(startSpecifier, specifierLen)); 1914} 1915 1916void CheckFormatHandler::HandleNonStandardConversionSpecifier( 1917 const analyze_format_string::ConversionSpecifier &CS, 1918 const char *startSpecifier, unsigned specifierLen) { 1919 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) << CS.toString() 1920 << 1, 1921 getLocationOfByte(CS.getStart()), 1922 /*IsStringLocation*/true, 1923 getSpecifierRange(startSpecifier, specifierLen)); 1924} 1925 1926void CheckFormatHandler::HandleNonStandardConversionSpecification( 1927 const analyze_format_string::LengthModifier &LM, 1928 const analyze_format_string::ConversionSpecifier &CS, 1929 const char *startSpecifier, unsigned specifierLen) { 1930 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_conversion_spec) 1931 << LM.toString() << CS.toString(), 1932 getLocationOfByte(LM.getStart()), 1933 /*IsStringLocation*/true, 1934 getSpecifierRange(startSpecifier, specifierLen)); 1935} 1936 1937void CheckFormatHandler::HandlePosition(const char *startPos, 1938 unsigned posLen) { 1939 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_positional_arg), 1940 getLocationOfByte(startPos), 1941 /*IsStringLocation*/true, 1942 getSpecifierRange(startPos, posLen)); 1943} 1944 1945void 1946CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1947 analyze_format_string::PositionContext p) { 1948 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier) 1949 << (unsigned) p, 1950 getLocationOfByte(startPos), /*IsStringLocation*/true, 1951 getSpecifierRange(startPos, posLen)); 1952} 1953 1954void CheckFormatHandler::HandleZeroPosition(const char *startPos, 1955 unsigned posLen) { 1956 EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier), 1957 getLocationOfByte(startPos), 1958 /*IsStringLocation*/true, 1959 getSpecifierRange(startPos, posLen)); 1960} 1961 1962void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 1963 if (!IsObjCLiteral) { 1964 // The presence of a null character is likely an error. 1965 EmitFormatDiagnostic( 1966 S.PDiag(diag::warn_printf_format_string_contains_null_char), 1967 getLocationOfByte(nullCharacter), /*IsStringLocation*/true, 1968 getFormatStringRange()); 1969 } 1970} 1971 1972const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 1973 return Args[FirstDataArg + i]; 1974} 1975 1976void CheckFormatHandler::DoneProcessing() { 1977 // Does the number of data arguments exceed the number of 1978 // format conversions in the format string? 1979 if (!HasVAListArg) { 1980 // Find any arguments that weren't covered. 1981 CoveredArgs.flip(); 1982 signed notCoveredArg = CoveredArgs.find_first(); 1983 if (notCoveredArg >= 0) { 1984 assert((unsigned)notCoveredArg < NumDataArgs); 1985 SourceLocation Loc = getDataArg((unsigned) notCoveredArg)->getLocStart(); 1986 if (!S.getSourceManager().isInSystemMacro(Loc)) { 1987 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used), 1988 Loc, 1989 /*IsStringLocation*/false, getFormatStringRange()); 1990 } 1991 } 1992 } 1993} 1994 1995bool 1996CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 1997 SourceLocation Loc, 1998 const char *startSpec, 1999 unsigned specifierLen, 2000 const char *csStart, 2001 unsigned csLen) { 2002 2003 bool keepGoing = true; 2004 if (argIndex < NumDataArgs) { 2005 // Consider the argument coverered, even though the specifier doesn't 2006 // make sense. 2007 CoveredArgs.set(argIndex); 2008 } 2009 else { 2010 // If argIndex exceeds the number of data arguments we 2011 // don't issue a warning because that is just a cascade of warnings (and 2012 // they may have intended '%%' anyway). We don't want to continue processing 2013 // the format string after this point, however, as we will like just get 2014 // gibberish when trying to match arguments. 2015 keepGoing = false; 2016 } 2017 2018 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion) 2019 << StringRef(csStart, csLen), 2020 Loc, /*IsStringLocation*/true, 2021 getSpecifierRange(startSpec, specifierLen)); 2022 2023 return keepGoing; 2024} 2025 2026void 2027CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc, 2028 const char *startSpec, 2029 unsigned specifierLen) { 2030 EmitFormatDiagnostic( 2031 S.PDiag(diag::warn_format_mix_positional_nonpositional_args), 2032 Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen)); 2033} 2034 2035bool 2036CheckFormatHandler::CheckNumArgs( 2037 const analyze_format_string::FormatSpecifier &FS, 2038 const analyze_format_string::ConversionSpecifier &CS, 2039 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 2040 2041 if (argIndex >= NumDataArgs) { 2042 PartialDiagnostic PDiag = FS.usesPositionalArg() 2043 ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args) 2044 << (argIndex+1) << NumDataArgs) 2045 : S.PDiag(diag::warn_printf_insufficient_data_args); 2046 EmitFormatDiagnostic( 2047 PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true, 2048 getSpecifierRange(startSpecifier, specifierLen)); 2049 return false; 2050 } 2051 return true; 2052} 2053 2054template<typename Range> 2055void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag, 2056 SourceLocation Loc, 2057 bool IsStringLocation, 2058 Range StringRange, 2059 FixItHint FixIt) { 2060 EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag, 2061 Loc, IsStringLocation, StringRange, FixIt); 2062} 2063 2064/// \brief If the format string is not within the funcion call, emit a note 2065/// so that the function call and string are in diagnostic messages. 2066/// 2067/// \param inFunctionCall if true, the format string is within the function 2068/// call and only one diagnostic message will be produced. Otherwise, an 2069/// extra note will be emitted pointing to location of the format string. 2070/// 2071/// \param ArgumentExpr the expression that is passed as the format string 2072/// argument in the function call. Used for getting locations when two 2073/// diagnostics are emitted. 2074/// 2075/// \param PDiag the callee should already have provided any strings for the 2076/// diagnostic message. This function only adds locations and fixits 2077/// to diagnostics. 2078/// 2079/// \param Loc primary location for diagnostic. If two diagnostics are 2080/// required, one will be at Loc and a new SourceLocation will be created for 2081/// the other one. 2082/// 2083/// \param IsStringLocation if true, Loc points to the format string should be 2084/// used for the note. Otherwise, Loc points to the argument list and will 2085/// be used with PDiag. 2086/// 2087/// \param StringRange some or all of the string to highlight. This is 2088/// templated so it can accept either a CharSourceRange or a SourceRange. 2089/// 2090/// \param Fixit optional fix it hint for the format string. 2091template<typename Range> 2092void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall, 2093 const Expr *ArgumentExpr, 2094 PartialDiagnostic PDiag, 2095 SourceLocation Loc, 2096 bool IsStringLocation, 2097 Range StringRange, 2098 FixItHint FixIt) { 2099 if (InFunctionCall) 2100 S.Diag(Loc, PDiag) << StringRange << FixIt; 2101 else { 2102 S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag) 2103 << ArgumentExpr->getSourceRange(); 2104 S.Diag(IsStringLocation ? Loc : StringRange.getBegin(), 2105 diag::note_format_string_defined) 2106 << StringRange << FixIt; 2107 } 2108} 2109 2110//===--- CHECK: Printf format string checking ------------------------------===// 2111 2112namespace { 2113class CheckPrintfHandler : public CheckFormatHandler { 2114public: 2115 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 2116 const Expr *origFormatExpr, unsigned firstDataArg, 2117 unsigned numDataArgs, bool isObjCLiteral, 2118 const char *beg, bool hasVAListArg, 2119 Expr **Args, unsigned NumArgs, 2120 unsigned formatIdx, bool inFunctionCall) 2121 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2122 numDataArgs, isObjCLiteral, beg, hasVAListArg, 2123 Args, NumArgs, formatIdx, inFunctionCall) {} 2124 2125 2126 bool HandleInvalidPrintfConversionSpecifier( 2127 const analyze_printf::PrintfSpecifier &FS, 2128 const char *startSpecifier, 2129 unsigned specifierLen); 2130 2131 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 2132 const char *startSpecifier, 2133 unsigned specifierLen); 2134 2135 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 2136 const char *startSpecifier, unsigned specifierLen); 2137 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 2138 const analyze_printf::OptionalAmount &Amt, 2139 unsigned type, 2140 const char *startSpecifier, unsigned specifierLen); 2141 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2142 const analyze_printf::OptionalFlag &flag, 2143 const char *startSpecifier, unsigned specifierLen); 2144 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 2145 const analyze_printf::OptionalFlag &ignoredFlag, 2146 const analyze_printf::OptionalFlag &flag, 2147 const char *startSpecifier, unsigned specifierLen); 2148}; 2149} 2150 2151bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 2152 const analyze_printf::PrintfSpecifier &FS, 2153 const char *startSpecifier, 2154 unsigned specifierLen) { 2155 const analyze_printf::PrintfConversionSpecifier &CS = 2156 FS.getConversionSpecifier(); 2157 2158 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2159 getLocationOfByte(CS.getStart()), 2160 startSpecifier, specifierLen, 2161 CS.getStart(), CS.getLength()); 2162} 2163 2164bool CheckPrintfHandler::HandleAmount( 2165 const analyze_format_string::OptionalAmount &Amt, 2166 unsigned k, const char *startSpecifier, 2167 unsigned specifierLen) { 2168 2169 if (Amt.hasDataArgument()) { 2170 if (!HasVAListArg) { 2171 unsigned argIndex = Amt.getArgIndex(); 2172 if (argIndex >= NumDataArgs) { 2173 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg) 2174 << k, 2175 getLocationOfByte(Amt.getStart()), 2176 /*IsStringLocation*/true, 2177 getSpecifierRange(startSpecifier, specifierLen)); 2178 // Don't do any more checking. We will just emit 2179 // spurious errors. 2180 return false; 2181 } 2182 2183 // Type check the data argument. It should be an 'int'. 2184 // Although not in conformance with C99, we also allow the argument to be 2185 // an 'unsigned int' as that is a reasonably safe case. GCC also 2186 // doesn't emit a warning for that case. 2187 CoveredArgs.set(argIndex); 2188 const Expr *Arg = getDataArg(argIndex); 2189 QualType T = Arg->getType(); 2190 2191 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 2192 assert(ATR.isValid()); 2193 2194 if (!ATR.matchesType(S.Context, T)) { 2195 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type) 2196 << k << ATR.getRepresentativeTypeName(S.Context) 2197 << T << Arg->getSourceRange(), 2198 getLocationOfByte(Amt.getStart()), 2199 /*IsStringLocation*/true, 2200 getSpecifierRange(startSpecifier, specifierLen)); 2201 // Don't do any more checking. We will just emit 2202 // spurious errors. 2203 return false; 2204 } 2205 } 2206 } 2207 return true; 2208} 2209 2210void CheckPrintfHandler::HandleInvalidAmount( 2211 const analyze_printf::PrintfSpecifier &FS, 2212 const analyze_printf::OptionalAmount &Amt, 2213 unsigned type, 2214 const char *startSpecifier, 2215 unsigned specifierLen) { 2216 const analyze_printf::PrintfConversionSpecifier &CS = 2217 FS.getConversionSpecifier(); 2218 2219 FixItHint fixit = 2220 Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant 2221 ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 2222 Amt.getConstantLength())) 2223 : FixItHint(); 2224 2225 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount) 2226 << type << CS.toString(), 2227 getLocationOfByte(Amt.getStart()), 2228 /*IsStringLocation*/true, 2229 getSpecifierRange(startSpecifier, specifierLen), 2230 fixit); 2231} 2232 2233void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2234 const analyze_printf::OptionalFlag &flag, 2235 const char *startSpecifier, 2236 unsigned specifierLen) { 2237 // Warn about pointless flag with a fixit removal. 2238 const analyze_printf::PrintfConversionSpecifier &CS = 2239 FS.getConversionSpecifier(); 2240 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag) 2241 << flag.toString() << CS.toString(), 2242 getLocationOfByte(flag.getPosition()), 2243 /*IsStringLocation*/true, 2244 getSpecifierRange(startSpecifier, specifierLen), 2245 FixItHint::CreateRemoval( 2246 getSpecifierRange(flag.getPosition(), 1))); 2247} 2248 2249void CheckPrintfHandler::HandleIgnoredFlag( 2250 const analyze_printf::PrintfSpecifier &FS, 2251 const analyze_printf::OptionalFlag &ignoredFlag, 2252 const analyze_printf::OptionalFlag &flag, 2253 const char *startSpecifier, 2254 unsigned specifierLen) { 2255 // Warn about ignored flag with a fixit removal. 2256 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag) 2257 << ignoredFlag.toString() << flag.toString(), 2258 getLocationOfByte(ignoredFlag.getPosition()), 2259 /*IsStringLocation*/true, 2260 getSpecifierRange(startSpecifier, specifierLen), 2261 FixItHint::CreateRemoval( 2262 getSpecifierRange(ignoredFlag.getPosition(), 1))); 2263} 2264 2265bool 2266CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 2267 &FS, 2268 const char *startSpecifier, 2269 unsigned specifierLen) { 2270 2271 using namespace analyze_format_string; 2272 using namespace analyze_printf; 2273 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 2274 2275 if (FS.consumesDataArgument()) { 2276 if (atFirstArg) { 2277 atFirstArg = false; 2278 usesPositionalArgs = FS.usesPositionalArg(); 2279 } 2280 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2281 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2282 startSpecifier, specifierLen); 2283 return false; 2284 } 2285 } 2286 2287 // First check if the field width, precision, and conversion specifier 2288 // have matching data arguments. 2289 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 2290 startSpecifier, specifierLen)) { 2291 return false; 2292 } 2293 2294 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 2295 startSpecifier, specifierLen)) { 2296 return false; 2297 } 2298 2299 if (!CS.consumesDataArgument()) { 2300 // FIXME: Technically specifying a precision or field width here 2301 // makes no sense. Worth issuing a warning at some point. 2302 return true; 2303 } 2304 2305 // Consume the argument. 2306 unsigned argIndex = FS.getArgIndex(); 2307 if (argIndex < NumDataArgs) { 2308 // The check to see if the argIndex is valid will come later. 2309 // We set the bit here because we may exit early from this 2310 // function if we encounter some other error. 2311 CoveredArgs.set(argIndex); 2312 } 2313 2314 // Check for using an Objective-C specific conversion specifier 2315 // in a non-ObjC literal. 2316 if (!IsObjCLiteral && CS.isObjCArg()) { 2317 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 2318 specifierLen); 2319 } 2320 2321 // Check for invalid use of field width 2322 if (!FS.hasValidFieldWidth()) { 2323 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 2324 startSpecifier, specifierLen); 2325 } 2326 2327 // Check for invalid use of precision 2328 if (!FS.hasValidPrecision()) { 2329 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 2330 startSpecifier, specifierLen); 2331 } 2332 2333 // Check each flag does not conflict with any other component. 2334 if (!FS.hasValidThousandsGroupingPrefix()) 2335 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 2336 if (!FS.hasValidLeadingZeros()) 2337 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 2338 if (!FS.hasValidPlusPrefix()) 2339 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 2340 if (!FS.hasValidSpacePrefix()) 2341 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 2342 if (!FS.hasValidAlternativeForm()) 2343 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 2344 if (!FS.hasValidLeftJustified()) 2345 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 2346 2347 // Check that flags are not ignored by another flag 2348 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 2349 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 2350 startSpecifier, specifierLen); 2351 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 2352 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 2353 startSpecifier, specifierLen); 2354 2355 // Check the length modifier is valid with the given conversion specifier. 2356 const LengthModifier &LM = FS.getLengthModifier(); 2357 if (!FS.hasValidLengthModifier()) 2358 EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length) 2359 << LM.toString() << CS.toString(), 2360 getLocationOfByte(LM.getStart()), 2361 /*IsStringLocation*/true, 2362 getSpecifierRange(startSpecifier, specifierLen), 2363 FixItHint::CreateRemoval( 2364 getSpecifierRange(LM.getStart(), 2365 LM.getLength()))); 2366 if (!FS.hasStandardLengthModifier()) 2367 HandleNonStandardLengthModifier(LM, startSpecifier, specifierLen); 2368 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 2369 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 2370 if (!FS.hasStandardLengthConversionCombination()) 2371 HandleNonStandardConversionSpecification(LM, CS, startSpecifier, 2372 specifierLen); 2373 2374 // Are we using '%n'? 2375 if (CS.getKind() == ConversionSpecifier::nArg) { 2376 // Issue a warning about this being a possible security issue. 2377 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_write_back), 2378 getLocationOfByte(CS.getStart()), 2379 /*IsStringLocation*/true, 2380 getSpecifierRange(startSpecifier, specifierLen)); 2381 // Continue checking the other format specifiers. 2382 return true; 2383 } 2384 2385 // The remaining checks depend on the data arguments. 2386 if (HasVAListArg) 2387 return true; 2388 2389 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2390 return false; 2391 2392 // Now type check the data expression that matches the 2393 // format specifier. 2394 const Expr *Ex = getDataArg(argIndex); 2395 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context, 2396 IsObjCLiteral); 2397 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 2398 // Check if we didn't match because of an implicit cast from a 'char' 2399 // or 'short' to an 'int'. This is done because printf is a varargs 2400 // function. 2401 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 2402 if (ICE->getType() == S.Context.IntTy || 2403 ICE->getType() == S.Context.UnsignedIntTy) { 2404 // All further checking is done on the subexpression. 2405 Ex = ICE->getSubExpr(); 2406 if (ATR.matchesType(S.Context, Ex->getType())) 2407 return true; 2408 } 2409 2410 // We may be able to offer a FixItHint if it is a supported type. 2411 PrintfSpecifier fixedFS = FS; 2412 bool success = fixedFS.fixType(Ex->getType(), S.getLangOpts(), 2413 S.Context, IsObjCLiteral); 2414 2415 if (success) { 2416 // Get the fix string from the fixed format specifier 2417 SmallString<128> buf; 2418 llvm::raw_svector_ostream os(buf); 2419 fixedFS.toString(os); 2420 2421 EmitFormatDiagnostic( 2422 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2423 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2424 << Ex->getSourceRange(), 2425 getLocationOfByte(CS.getStart()), 2426 /*IsStringLocation*/true, 2427 getSpecifierRange(startSpecifier, specifierLen), 2428 FixItHint::CreateReplacement( 2429 getSpecifierRange(startSpecifier, specifierLen), 2430 os.str())); 2431 } 2432 else { 2433 EmitFormatDiagnostic( 2434 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2435 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2436 << getSpecifierRange(startSpecifier, specifierLen) 2437 << Ex->getSourceRange(), 2438 getLocationOfByte(CS.getStart()), 2439 true, 2440 getSpecifierRange(startSpecifier, specifierLen)); 2441 } 2442 } 2443 2444 return true; 2445} 2446 2447//===--- CHECK: Scanf format string checking ------------------------------===// 2448 2449namespace { 2450class CheckScanfHandler : public CheckFormatHandler { 2451public: 2452 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 2453 const Expr *origFormatExpr, unsigned firstDataArg, 2454 unsigned numDataArgs, bool isObjCLiteral, 2455 const char *beg, bool hasVAListArg, 2456 Expr **Args, unsigned NumArgs, 2457 unsigned formatIdx, bool inFunctionCall) 2458 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2459 numDataArgs, isObjCLiteral, beg, hasVAListArg, 2460 Args, NumArgs, formatIdx, inFunctionCall) {} 2461 2462 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 2463 const char *startSpecifier, 2464 unsigned specifierLen); 2465 2466 bool HandleInvalidScanfConversionSpecifier( 2467 const analyze_scanf::ScanfSpecifier &FS, 2468 const char *startSpecifier, 2469 unsigned specifierLen); 2470 2471 void HandleIncompleteScanList(const char *start, const char *end); 2472}; 2473} 2474 2475void CheckScanfHandler::HandleIncompleteScanList(const char *start, 2476 const char *end) { 2477 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 2478 getLocationOfByte(end), /*IsStringLocation*/true, 2479 getSpecifierRange(start, end - start)); 2480} 2481 2482bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 2483 const analyze_scanf::ScanfSpecifier &FS, 2484 const char *startSpecifier, 2485 unsigned specifierLen) { 2486 2487 const analyze_scanf::ScanfConversionSpecifier &CS = 2488 FS.getConversionSpecifier(); 2489 2490 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2491 getLocationOfByte(CS.getStart()), 2492 startSpecifier, specifierLen, 2493 CS.getStart(), CS.getLength()); 2494} 2495 2496bool CheckScanfHandler::HandleScanfSpecifier( 2497 const analyze_scanf::ScanfSpecifier &FS, 2498 const char *startSpecifier, 2499 unsigned specifierLen) { 2500 2501 using namespace analyze_scanf; 2502 using namespace analyze_format_string; 2503 2504 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 2505 2506 // Handle case where '%' and '*' don't consume an argument. These shouldn't 2507 // be used to decide if we are using positional arguments consistently. 2508 if (FS.consumesDataArgument()) { 2509 if (atFirstArg) { 2510 atFirstArg = false; 2511 usesPositionalArgs = FS.usesPositionalArg(); 2512 } 2513 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2514 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2515 startSpecifier, specifierLen); 2516 return false; 2517 } 2518 } 2519 2520 // Check if the field with is non-zero. 2521 const OptionalAmount &Amt = FS.getFieldWidth(); 2522 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 2523 if (Amt.getConstantAmount() == 0) { 2524 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 2525 Amt.getConstantLength()); 2526 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 2527 getLocationOfByte(Amt.getStart()), 2528 /*IsStringLocation*/true, R, 2529 FixItHint::CreateRemoval(R)); 2530 } 2531 } 2532 2533 if (!FS.consumesDataArgument()) { 2534 // FIXME: Technically specifying a precision or field width here 2535 // makes no sense. Worth issuing a warning at some point. 2536 return true; 2537 } 2538 2539 // Consume the argument. 2540 unsigned argIndex = FS.getArgIndex(); 2541 if (argIndex < NumDataArgs) { 2542 // The check to see if the argIndex is valid will come later. 2543 // We set the bit here because we may exit early from this 2544 // function if we encounter some other error. 2545 CoveredArgs.set(argIndex); 2546 } 2547 2548 // Check the length modifier is valid with the given conversion specifier. 2549 const LengthModifier &LM = FS.getLengthModifier(); 2550 if (!FS.hasValidLengthModifier()) { 2551 const CharSourceRange &R = getSpecifierRange(LM.getStart(), LM.getLength()); 2552 EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length) 2553 << LM.toString() << CS.toString() 2554 << getSpecifierRange(startSpecifier, specifierLen), 2555 getLocationOfByte(LM.getStart()), 2556 /*IsStringLocation*/true, R, 2557 FixItHint::CreateRemoval(R)); 2558 } 2559 2560 if (!FS.hasStandardLengthModifier()) 2561 HandleNonStandardLengthModifier(LM, startSpecifier, specifierLen); 2562 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 2563 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 2564 if (!FS.hasStandardLengthConversionCombination()) 2565 HandleNonStandardConversionSpecification(LM, CS, startSpecifier, 2566 specifierLen); 2567 2568 // The remaining checks depend on the data arguments. 2569 if (HasVAListArg) 2570 return true; 2571 2572 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2573 return false; 2574 2575 // Check that the argument type matches the format specifier. 2576 const Expr *Ex = getDataArg(argIndex); 2577 const analyze_scanf::ScanfArgTypeResult &ATR = FS.getArgType(S.Context); 2578 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 2579 ScanfSpecifier fixedFS = FS; 2580 bool success = fixedFS.fixType(Ex->getType(), S.getLangOpts(), 2581 S.Context); 2582 2583 if (success) { 2584 // Get the fix string from the fixed format specifier. 2585 SmallString<128> buf; 2586 llvm::raw_svector_ostream os(buf); 2587 fixedFS.toString(os); 2588 2589 EmitFormatDiagnostic( 2590 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2591 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2592 << Ex->getSourceRange(), 2593 getLocationOfByte(CS.getStart()), 2594 /*IsStringLocation*/true, 2595 getSpecifierRange(startSpecifier, specifierLen), 2596 FixItHint::CreateReplacement( 2597 getSpecifierRange(startSpecifier, specifierLen), 2598 os.str())); 2599 } else { 2600 EmitFormatDiagnostic( 2601 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2602 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2603 << Ex->getSourceRange(), 2604 getLocationOfByte(CS.getStart()), 2605 /*IsStringLocation*/true, 2606 getSpecifierRange(startSpecifier, specifierLen)); 2607 } 2608 } 2609 2610 return true; 2611} 2612 2613void Sema::CheckFormatString(const StringLiteral *FExpr, 2614 const Expr *OrigFormatExpr, 2615 Expr **Args, unsigned NumArgs, 2616 bool HasVAListArg, unsigned format_idx, 2617 unsigned firstDataArg, FormatStringType Type, 2618 bool inFunctionCall) { 2619 2620 // CHECK: is the format string a wide literal? 2621 if (!FExpr->isAscii()) { 2622 CheckFormatHandler::EmitFormatDiagnostic( 2623 *this, inFunctionCall, Args[format_idx], 2624 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 2625 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2626 return; 2627 } 2628 2629 // Str - The format string. NOTE: this is NOT null-terminated! 2630 StringRef StrRef = FExpr->getString(); 2631 const char *Str = StrRef.data(); 2632 unsigned StrLen = StrRef.size(); 2633 const unsigned numDataArgs = NumArgs - firstDataArg; 2634 2635 // CHECK: empty format string? 2636 if (StrLen == 0 && numDataArgs > 0) { 2637 CheckFormatHandler::EmitFormatDiagnostic( 2638 *this, inFunctionCall, Args[format_idx], 2639 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 2640 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2641 return; 2642 } 2643 2644 if (Type == FST_Printf || Type == FST_NSString) { 2645 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2646 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2647 Str, HasVAListArg, Args, NumArgs, format_idx, 2648 inFunctionCall); 2649 2650 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen, 2651 getLangOpts())) 2652 H.DoneProcessing(); 2653 } else if (Type == FST_Scanf) { 2654 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2655 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2656 Str, HasVAListArg, Args, NumArgs, format_idx, 2657 inFunctionCall); 2658 2659 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen, 2660 getLangOpts())) 2661 H.DoneProcessing(); 2662 } // TODO: handle other formats 2663} 2664 2665//===--- CHECK: Standard memory functions ---------------------------------===// 2666 2667/// \brief Determine whether the given type is a dynamic class type (e.g., 2668/// whether it has a vtable). 2669static bool isDynamicClassType(QualType T) { 2670 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 2671 if (CXXRecordDecl *Definition = Record->getDefinition()) 2672 if (Definition->isDynamicClass()) 2673 return true; 2674 2675 return false; 2676} 2677 2678/// \brief If E is a sizeof expression, returns its argument expression, 2679/// otherwise returns NULL. 2680static const Expr *getSizeOfExprArg(const Expr* E) { 2681 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2682 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2683 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 2684 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 2685 2686 return 0; 2687} 2688 2689/// \brief If E is a sizeof expression, returns its argument type. 2690static QualType getSizeOfArgType(const Expr* E) { 2691 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2692 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2693 if (SizeOf->getKind() == clang::UETT_SizeOf) 2694 return SizeOf->getTypeOfArgument(); 2695 2696 return QualType(); 2697} 2698 2699/// \brief Check for dangerous or invalid arguments to memset(). 2700/// 2701/// This issues warnings on known problematic, dangerous or unspecified 2702/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 2703/// function calls. 2704/// 2705/// \param Call The call expression to diagnose. 2706void Sema::CheckMemaccessArguments(const CallExpr *Call, 2707 unsigned BId, 2708 IdentifierInfo *FnName) { 2709 assert(BId != 0); 2710 2711 // It is possible to have a non-standard definition of memset. Validate 2712 // we have enough arguments, and if not, abort further checking. 2713 unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3); 2714 if (Call->getNumArgs() < ExpectedNumArgs) 2715 return; 2716 2717 unsigned LastArg = (BId == Builtin::BImemset || 2718 BId == Builtin::BIstrndup ? 1 : 2); 2719 unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2); 2720 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 2721 2722 // We have special checking when the length is a sizeof expression. 2723 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 2724 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 2725 llvm::FoldingSetNodeID SizeOfArgID; 2726 2727 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 2728 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 2729 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 2730 2731 QualType DestTy = Dest->getType(); 2732 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 2733 QualType PointeeTy = DestPtrTy->getPointeeType(); 2734 2735 // Never warn about void type pointers. This can be used to suppress 2736 // false positives. 2737 if (PointeeTy->isVoidType()) 2738 continue; 2739 2740 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 2741 // actually comparing the expressions for equality. Because computing the 2742 // expression IDs can be expensive, we only do this if the diagnostic is 2743 // enabled. 2744 if (SizeOfArg && 2745 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 2746 SizeOfArg->getExprLoc())) { 2747 // We only compute IDs for expressions if the warning is enabled, and 2748 // cache the sizeof arg's ID. 2749 if (SizeOfArgID == llvm::FoldingSetNodeID()) 2750 SizeOfArg->Profile(SizeOfArgID, Context, true); 2751 llvm::FoldingSetNodeID DestID; 2752 Dest->Profile(DestID, Context, true); 2753 if (DestID == SizeOfArgID) { 2754 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 2755 // over sizeof(src) as well. 2756 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 2757 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 2758 if (UnaryOp->getOpcode() == UO_AddrOf) 2759 ActionIdx = 1; // If its an address-of operator, just remove it. 2760 if (Context.getTypeSize(PointeeTy) == Context.getCharWidth()) 2761 ActionIdx = 2; // If the pointee's size is sizeof(char), 2762 // suggest an explicit length. 2763 unsigned DestSrcSelect = 2764 (BId == Builtin::BIstrndup ? 1 : ArgIdx); 2765 DiagRuntimeBehavior(SizeOfArg->getExprLoc(), Dest, 2766 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 2767 << FnName << DestSrcSelect << ActionIdx 2768 << Dest->getSourceRange() 2769 << SizeOfArg->getSourceRange()); 2770 break; 2771 } 2772 } 2773 2774 // Also check for cases where the sizeof argument is the exact same 2775 // type as the memory argument, and where it points to a user-defined 2776 // record type. 2777 if (SizeOfArgTy != QualType()) { 2778 if (PointeeTy->isRecordType() && 2779 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 2780 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 2781 PDiag(diag::warn_sizeof_pointer_type_memaccess) 2782 << FnName << SizeOfArgTy << ArgIdx 2783 << PointeeTy << Dest->getSourceRange() 2784 << LenExpr->getSourceRange()); 2785 break; 2786 } 2787 } 2788 2789 // Always complain about dynamic classes. 2790 if (isDynamicClassType(PointeeTy)) { 2791 2792 unsigned OperationType = 0; 2793 // "overwritten" if we're warning about the destination for any call 2794 // but memcmp; otherwise a verb appropriate to the call. 2795 if (ArgIdx != 0 || BId == Builtin::BImemcmp) { 2796 if (BId == Builtin::BImemcpy) 2797 OperationType = 1; 2798 else if(BId == Builtin::BImemmove) 2799 OperationType = 2; 2800 else if (BId == Builtin::BImemcmp) 2801 OperationType = 3; 2802 } 2803 2804 DiagRuntimeBehavior( 2805 Dest->getExprLoc(), Dest, 2806 PDiag(diag::warn_dyn_class_memaccess) 2807 << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx) 2808 << FnName << PointeeTy 2809 << OperationType 2810 << Call->getCallee()->getSourceRange()); 2811 } else if (PointeeTy.hasNonTrivialObjCLifetime() && 2812 BId != Builtin::BImemset) 2813 DiagRuntimeBehavior( 2814 Dest->getExprLoc(), Dest, 2815 PDiag(diag::warn_arc_object_memaccess) 2816 << ArgIdx << FnName << PointeeTy 2817 << Call->getCallee()->getSourceRange()); 2818 else 2819 continue; 2820 2821 DiagRuntimeBehavior( 2822 Dest->getExprLoc(), Dest, 2823 PDiag(diag::note_bad_memaccess_silence) 2824 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 2825 break; 2826 } 2827 } 2828} 2829 2830// A little helper routine: ignore addition and subtraction of integer literals. 2831// This intentionally does not ignore all integer constant expressions because 2832// we don't want to remove sizeof(). 2833static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 2834 Ex = Ex->IgnoreParenCasts(); 2835 2836 for (;;) { 2837 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 2838 if (!BO || !BO->isAdditiveOp()) 2839 break; 2840 2841 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 2842 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 2843 2844 if (isa<IntegerLiteral>(RHS)) 2845 Ex = LHS; 2846 else if (isa<IntegerLiteral>(LHS)) 2847 Ex = RHS; 2848 else 2849 break; 2850 } 2851 2852 return Ex; 2853} 2854 2855// Warn if the user has made the 'size' argument to strlcpy or strlcat 2856// be the size of the source, instead of the destination. 2857void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 2858 IdentifierInfo *FnName) { 2859 2860 // Don't crash if the user has the wrong number of arguments 2861 if (Call->getNumArgs() != 3) 2862 return; 2863 2864 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 2865 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 2866 const Expr *CompareWithSrc = NULL; 2867 2868 // Look for 'strlcpy(dst, x, sizeof(x))' 2869 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 2870 CompareWithSrc = Ex; 2871 else { 2872 // Look for 'strlcpy(dst, x, strlen(x))' 2873 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 2874 if (SizeCall->isBuiltinCall() == Builtin::BIstrlen 2875 && SizeCall->getNumArgs() == 1) 2876 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 2877 } 2878 } 2879 2880 if (!CompareWithSrc) 2881 return; 2882 2883 // Determine if the argument to sizeof/strlen is equal to the source 2884 // argument. In principle there's all kinds of things you could do 2885 // here, for instance creating an == expression and evaluating it with 2886 // EvaluateAsBooleanCondition, but this uses a more direct technique: 2887 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 2888 if (!SrcArgDRE) 2889 return; 2890 2891 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 2892 if (!CompareWithSrcDRE || 2893 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 2894 return; 2895 2896 const Expr *OriginalSizeArg = Call->getArg(2); 2897 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 2898 << OriginalSizeArg->getSourceRange() << FnName; 2899 2900 // Output a FIXIT hint if the destination is an array (rather than a 2901 // pointer to an array). This could be enhanced to handle some 2902 // pointers if we know the actual size, like if DstArg is 'array+2' 2903 // we could say 'sizeof(array)-2'. 2904 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 2905 QualType DstArgTy = DstArg->getType(); 2906 2907 // Only handle constant-sized or VLAs, but not flexible members. 2908 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 2909 // Only issue the FIXIT for arrays of size > 1. 2910 if (CAT->getSize().getSExtValue() <= 1) 2911 return; 2912 } else if (!DstArgTy->isVariableArrayType()) { 2913 return; 2914 } 2915 2916 SmallString<128> sizeString; 2917 llvm::raw_svector_ostream OS(sizeString); 2918 OS << "sizeof("; 2919 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 2920 OS << ")"; 2921 2922 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 2923 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 2924 OS.str()); 2925} 2926 2927/// Check if two expressions refer to the same declaration. 2928static bool referToTheSameDecl(const Expr *E1, const Expr *E2) { 2929 if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1)) 2930 if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2)) 2931 return D1->getDecl() == D2->getDecl(); 2932 return false; 2933} 2934 2935static const Expr *getStrlenExprArg(const Expr *E) { 2936 if (const CallExpr *CE = dyn_cast<CallExpr>(E)) { 2937 const FunctionDecl *FD = CE->getDirectCallee(); 2938 if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen) 2939 return 0; 2940 return CE->getArg(0)->IgnoreParenCasts(); 2941 } 2942 return 0; 2943} 2944 2945// Warn on anti-patterns as the 'size' argument to strncat. 2946// The correct size argument should look like following: 2947// strncat(dst, src, sizeof(dst) - strlen(dest) - 1); 2948void Sema::CheckStrncatArguments(const CallExpr *CE, 2949 IdentifierInfo *FnName) { 2950 // Don't crash if the user has the wrong number of arguments. 2951 if (CE->getNumArgs() < 3) 2952 return; 2953 const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts(); 2954 const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts(); 2955 const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts(); 2956 2957 // Identify common expressions, which are wrongly used as the size argument 2958 // to strncat and may lead to buffer overflows. 2959 unsigned PatternType = 0; 2960 if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) { 2961 // - sizeof(dst) 2962 if (referToTheSameDecl(SizeOfArg, DstArg)) 2963 PatternType = 1; 2964 // - sizeof(src) 2965 else if (referToTheSameDecl(SizeOfArg, SrcArg)) 2966 PatternType = 2; 2967 } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) { 2968 if (BE->getOpcode() == BO_Sub) { 2969 const Expr *L = BE->getLHS()->IgnoreParenCasts(); 2970 const Expr *R = BE->getRHS()->IgnoreParenCasts(); 2971 // - sizeof(dst) - strlen(dst) 2972 if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) && 2973 referToTheSameDecl(DstArg, getStrlenExprArg(R))) 2974 PatternType = 1; 2975 // - sizeof(src) - (anything) 2976 else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L))) 2977 PatternType = 2; 2978 } 2979 } 2980 2981 if (PatternType == 0) 2982 return; 2983 2984 // Generate the diagnostic. 2985 SourceLocation SL = LenArg->getLocStart(); 2986 SourceRange SR = LenArg->getSourceRange(); 2987 SourceManager &SM = PP.getSourceManager(); 2988 2989 // If the function is defined as a builtin macro, do not show macro expansion. 2990 if (SM.isMacroArgExpansion(SL)) { 2991 SL = SM.getSpellingLoc(SL); 2992 SR = SourceRange(SM.getSpellingLoc(SR.getBegin()), 2993 SM.getSpellingLoc(SR.getEnd())); 2994 } 2995 2996 if (PatternType == 1) 2997 Diag(SL, diag::warn_strncat_large_size) << SR; 2998 else 2999 Diag(SL, diag::warn_strncat_src_size) << SR; 3000 3001 // Output a FIXIT hint if the destination is an array (rather than a 3002 // pointer to an array). This could be enhanced to handle some 3003 // pointers if we know the actual size, like if DstArg is 'array+2' 3004 // we could say 'sizeof(array)-2'. 3005 QualType DstArgTy = DstArg->getType(); 3006 3007 // Only handle constant-sized or VLAs, but not flexible members. 3008 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 3009 // Only issue the FIXIT for arrays of size > 1. 3010 if (CAT->getSize().getSExtValue() <= 1) 3011 return; 3012 } else if (!DstArgTy->isVariableArrayType()) { 3013 return; 3014 } 3015 3016 SmallString<128> sizeString; 3017 llvm::raw_svector_ostream OS(sizeString); 3018 OS << "sizeof("; 3019 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 3020 OS << ") - "; 3021 OS << "strlen("; 3022 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 3023 OS << ") - 1"; 3024 3025 Diag(SL, diag::note_strncat_wrong_size) 3026 << FixItHint::CreateReplacement(SR, OS.str()); 3027} 3028 3029//===--- CHECK: Return Address of Stack Variable --------------------------===// 3030 3031static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3032 Decl *ParentDecl); 3033static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars, 3034 Decl *ParentDecl); 3035 3036/// CheckReturnStackAddr - Check if a return statement returns the address 3037/// of a stack variable. 3038void 3039Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 3040 SourceLocation ReturnLoc) { 3041 3042 Expr *stackE = 0; 3043 SmallVector<DeclRefExpr *, 8> refVars; 3044 3045 // Perform checking for returned stack addresses, local blocks, 3046 // label addresses or references to temporaries. 3047 if (lhsType->isPointerType() || 3048 (!getLangOpts().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 3049 stackE = EvalAddr(RetValExp, refVars, /*ParentDecl=*/0); 3050 } else if (lhsType->isReferenceType()) { 3051 stackE = EvalVal(RetValExp, refVars, /*ParentDecl=*/0); 3052 } 3053 3054 if (stackE == 0) 3055 return; // Nothing suspicious was found. 3056 3057 SourceLocation diagLoc; 3058 SourceRange diagRange; 3059 if (refVars.empty()) { 3060 diagLoc = stackE->getLocStart(); 3061 diagRange = stackE->getSourceRange(); 3062 } else { 3063 // We followed through a reference variable. 'stackE' contains the 3064 // problematic expression but we will warn at the return statement pointing 3065 // at the reference variable. We will later display the "trail" of 3066 // reference variables using notes. 3067 diagLoc = refVars[0]->getLocStart(); 3068 diagRange = refVars[0]->getSourceRange(); 3069 } 3070 3071 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 3072 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 3073 : diag::warn_ret_stack_addr) 3074 << DR->getDecl()->getDeclName() << diagRange; 3075 } else if (isa<BlockExpr>(stackE)) { // local block. 3076 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 3077 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 3078 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 3079 } else { // local temporary. 3080 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 3081 : diag::warn_ret_local_temp_addr) 3082 << diagRange; 3083 } 3084 3085 // Display the "trail" of reference variables that we followed until we 3086 // found the problematic expression using notes. 3087 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 3088 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 3089 // If this var binds to another reference var, show the range of the next 3090 // var, otherwise the var binds to the problematic expression, in which case 3091 // show the range of the expression. 3092 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 3093 : stackE->getSourceRange(); 3094 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 3095 << VD->getDeclName() << range; 3096 } 3097} 3098 3099/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 3100/// check if the expression in a return statement evaluates to an address 3101/// to a location on the stack, a local block, an address of a label, or a 3102/// reference to local temporary. The recursion is used to traverse the 3103/// AST of the return expression, with recursion backtracking when we 3104/// encounter a subexpression that (1) clearly does not lead to one of the 3105/// above problematic expressions (2) is something we cannot determine leads to 3106/// a problematic expression based on such local checking. 3107/// 3108/// Both EvalAddr and EvalVal follow through reference variables to evaluate 3109/// the expression that they point to. Such variables are added to the 3110/// 'refVars' vector so that we know what the reference variable "trail" was. 3111/// 3112/// EvalAddr processes expressions that are pointers that are used as 3113/// references (and not L-values). EvalVal handles all other values. 3114/// At the base case of the recursion is a check for the above problematic 3115/// expressions. 3116/// 3117/// This implementation handles: 3118/// 3119/// * pointer-to-pointer casts 3120/// * implicit conversions from array references to pointers 3121/// * taking the address of fields 3122/// * arbitrary interplay between "&" and "*" operators 3123/// * pointer arithmetic from an address of a stack variable 3124/// * taking the address of an array element where the array is on the stack 3125static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3126 Decl *ParentDecl) { 3127 if (E->isTypeDependent()) 3128 return NULL; 3129 3130 // We should only be called for evaluating pointer expressions. 3131 assert((E->getType()->isAnyPointerType() || 3132 E->getType()->isBlockPointerType() || 3133 E->getType()->isObjCQualifiedIdType()) && 3134 "EvalAddr only works on pointers"); 3135 3136 E = E->IgnoreParens(); 3137 3138 // Our "symbolic interpreter" is just a dispatch off the currently 3139 // viewed AST node. We then recursively traverse the AST by calling 3140 // EvalAddr and EvalVal appropriately. 3141 switch (E->getStmtClass()) { 3142 case Stmt::DeclRefExprClass: { 3143 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3144 3145 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 3146 // If this is a reference variable, follow through to the expression that 3147 // it points to. 3148 if (V->hasLocalStorage() && 3149 V->getType()->isReferenceType() && V->hasInit()) { 3150 // Add the reference variable to the "trail". 3151 refVars.push_back(DR); 3152 return EvalAddr(V->getInit(), refVars, ParentDecl); 3153 } 3154 3155 return NULL; 3156 } 3157 3158 case Stmt::UnaryOperatorClass: { 3159 // The only unary operator that make sense to handle here 3160 // is AddrOf. All others don't make sense as pointers. 3161 UnaryOperator *U = cast<UnaryOperator>(E); 3162 3163 if (U->getOpcode() == UO_AddrOf) 3164 return EvalVal(U->getSubExpr(), refVars, ParentDecl); 3165 else 3166 return NULL; 3167 } 3168 3169 case Stmt::BinaryOperatorClass: { 3170 // Handle pointer arithmetic. All other binary operators are not valid 3171 // in this context. 3172 BinaryOperator *B = cast<BinaryOperator>(E); 3173 BinaryOperatorKind op = B->getOpcode(); 3174 3175 if (op != BO_Add && op != BO_Sub) 3176 return NULL; 3177 3178 Expr *Base = B->getLHS(); 3179 3180 // Determine which argument is the real pointer base. It could be 3181 // the RHS argument instead of the LHS. 3182 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 3183 3184 assert (Base->getType()->isPointerType()); 3185 return EvalAddr(Base, refVars, ParentDecl); 3186 } 3187 3188 // For conditional operators we need to see if either the LHS or RHS are 3189 // valid DeclRefExpr*s. If one of them is valid, we return it. 3190 case Stmt::ConditionalOperatorClass: { 3191 ConditionalOperator *C = cast<ConditionalOperator>(E); 3192 3193 // Handle the GNU extension for missing LHS. 3194 if (Expr *lhsExpr = C->getLHS()) { 3195 // In C++, we can have a throw-expression, which has 'void' type. 3196 if (!lhsExpr->getType()->isVoidType()) 3197 if (Expr* LHS = EvalAddr(lhsExpr, refVars, ParentDecl)) 3198 return LHS; 3199 } 3200 3201 // In C++, we can have a throw-expression, which has 'void' type. 3202 if (C->getRHS()->getType()->isVoidType()) 3203 return NULL; 3204 3205 return EvalAddr(C->getRHS(), refVars, ParentDecl); 3206 } 3207 3208 case Stmt::BlockExprClass: 3209 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 3210 return E; // local block. 3211 return NULL; 3212 3213 case Stmt::AddrLabelExprClass: 3214 return E; // address of label. 3215 3216 case Stmt::ExprWithCleanupsClass: 3217 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars, 3218 ParentDecl); 3219 3220 // For casts, we need to handle conversions from arrays to 3221 // pointer values, and pointer-to-pointer conversions. 3222 case Stmt::ImplicitCastExprClass: 3223 case Stmt::CStyleCastExprClass: 3224 case Stmt::CXXFunctionalCastExprClass: 3225 case Stmt::ObjCBridgedCastExprClass: 3226 case Stmt::CXXStaticCastExprClass: 3227 case Stmt::CXXDynamicCastExprClass: 3228 case Stmt::CXXConstCastExprClass: 3229 case Stmt::CXXReinterpretCastExprClass: { 3230 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 3231 switch (cast<CastExpr>(E)->getCastKind()) { 3232 case CK_BitCast: 3233 case CK_LValueToRValue: 3234 case CK_NoOp: 3235 case CK_BaseToDerived: 3236 case CK_DerivedToBase: 3237 case CK_UncheckedDerivedToBase: 3238 case CK_Dynamic: 3239 case CK_CPointerToObjCPointerCast: 3240 case CK_BlockPointerToObjCPointerCast: 3241 case CK_AnyPointerToBlockPointerCast: 3242 return EvalAddr(SubExpr, refVars, ParentDecl); 3243 3244 case CK_ArrayToPointerDecay: 3245 return EvalVal(SubExpr, refVars, ParentDecl); 3246 3247 default: 3248 return 0; 3249 } 3250 } 3251 3252 case Stmt::MaterializeTemporaryExprClass: 3253 if (Expr *Result = EvalAddr( 3254 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3255 refVars, ParentDecl)) 3256 return Result; 3257 3258 return E; 3259 3260 // Everything else: we simply don't reason about them. 3261 default: 3262 return NULL; 3263 } 3264} 3265 3266 3267/// EvalVal - This function is complements EvalAddr in the mutual recursion. 3268/// See the comments for EvalAddr for more details. 3269static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3270 Decl *ParentDecl) { 3271do { 3272 // We should only be called for evaluating non-pointer expressions, or 3273 // expressions with a pointer type that are not used as references but instead 3274 // are l-values (e.g., DeclRefExpr with a pointer type). 3275 3276 // Our "symbolic interpreter" is just a dispatch off the currently 3277 // viewed AST node. We then recursively traverse the AST by calling 3278 // EvalAddr and EvalVal appropriately. 3279 3280 E = E->IgnoreParens(); 3281 switch (E->getStmtClass()) { 3282 case Stmt::ImplicitCastExprClass: { 3283 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 3284 if (IE->getValueKind() == VK_LValue) { 3285 E = IE->getSubExpr(); 3286 continue; 3287 } 3288 return NULL; 3289 } 3290 3291 case Stmt::ExprWithCleanupsClass: 3292 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,ParentDecl); 3293 3294 case Stmt::DeclRefExprClass: { 3295 // When we hit a DeclRefExpr we are looking at code that refers to a 3296 // variable's name. If it's not a reference variable we check if it has 3297 // local storage within the function, and if so, return the expression. 3298 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3299 3300 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) { 3301 // Check if it refers to itself, e.g. "int& i = i;". 3302 if (V == ParentDecl) 3303 return DR; 3304 3305 if (V->hasLocalStorage()) { 3306 if (!V->getType()->isReferenceType()) 3307 return DR; 3308 3309 // Reference variable, follow through to the expression that 3310 // it points to. 3311 if (V->hasInit()) { 3312 // Add the reference variable to the "trail". 3313 refVars.push_back(DR); 3314 return EvalVal(V->getInit(), refVars, V); 3315 } 3316 } 3317 } 3318 3319 return NULL; 3320 } 3321 3322 case Stmt::UnaryOperatorClass: { 3323 // The only unary operator that make sense to handle here 3324 // is Deref. All others don't resolve to a "name." This includes 3325 // handling all sorts of rvalues passed to a unary operator. 3326 UnaryOperator *U = cast<UnaryOperator>(E); 3327 3328 if (U->getOpcode() == UO_Deref) 3329 return EvalAddr(U->getSubExpr(), refVars, ParentDecl); 3330 3331 return NULL; 3332 } 3333 3334 case Stmt::ArraySubscriptExprClass: { 3335 // Array subscripts are potential references to data on the stack. We 3336 // retrieve the DeclRefExpr* for the array variable if it indeed 3337 // has local storage. 3338 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars,ParentDecl); 3339 } 3340 3341 case Stmt::ConditionalOperatorClass: { 3342 // For conditional operators we need to see if either the LHS or RHS are 3343 // non-NULL Expr's. If one is non-NULL, we return it. 3344 ConditionalOperator *C = cast<ConditionalOperator>(E); 3345 3346 // Handle the GNU extension for missing LHS. 3347 if (Expr *lhsExpr = C->getLHS()) 3348 if (Expr *LHS = EvalVal(lhsExpr, refVars, ParentDecl)) 3349 return LHS; 3350 3351 return EvalVal(C->getRHS(), refVars, ParentDecl); 3352 } 3353 3354 // Accesses to members are potential references to data on the stack. 3355 case Stmt::MemberExprClass: { 3356 MemberExpr *M = cast<MemberExpr>(E); 3357 3358 // Check for indirect access. We only want direct field accesses. 3359 if (M->isArrow()) 3360 return NULL; 3361 3362 // Check whether the member type is itself a reference, in which case 3363 // we're not going to refer to the member, but to what the member refers to. 3364 if (M->getMemberDecl()->getType()->isReferenceType()) 3365 return NULL; 3366 3367 return EvalVal(M->getBase(), refVars, ParentDecl); 3368 } 3369 3370 case Stmt::MaterializeTemporaryExprClass: 3371 if (Expr *Result = EvalVal( 3372 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3373 refVars, ParentDecl)) 3374 return Result; 3375 3376 return E; 3377 3378 default: 3379 // Check that we don't return or take the address of a reference to a 3380 // temporary. This is only useful in C++. 3381 if (!E->isTypeDependent() && E->isRValue()) 3382 return E; 3383 3384 // Everything else: we simply don't reason about them. 3385 return NULL; 3386 } 3387} while (true); 3388} 3389 3390//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 3391 3392/// Check for comparisons of floating point operands using != and ==. 3393/// Issue a warning if these are no self-comparisons, as they are not likely 3394/// to do what the programmer intended. 3395void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 3396 bool EmitWarning = true; 3397 3398 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 3399 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 3400 3401 // Special case: check for x == x (which is OK). 3402 // Do not emit warnings for such cases. 3403 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 3404 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 3405 if (DRL->getDecl() == DRR->getDecl()) 3406 EmitWarning = false; 3407 3408 3409 // Special case: check for comparisons against literals that can be exactly 3410 // represented by APFloat. In such cases, do not emit a warning. This 3411 // is a heuristic: often comparison against such literals are used to 3412 // detect if a value in a variable has not changed. This clearly can 3413 // lead to false negatives. 3414 if (EmitWarning) { 3415 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 3416 if (FLL->isExact()) 3417 EmitWarning = false; 3418 } else 3419 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 3420 if (FLR->isExact()) 3421 EmitWarning = false; 3422 } 3423 } 3424 3425 // Check for comparisons with builtin types. 3426 if (EmitWarning) 3427 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 3428 if (CL->isBuiltinCall()) 3429 EmitWarning = false; 3430 3431 if (EmitWarning) 3432 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 3433 if (CR->isBuiltinCall()) 3434 EmitWarning = false; 3435 3436 // Emit the diagnostic. 3437 if (EmitWarning) 3438 Diag(Loc, diag::warn_floatingpoint_eq) 3439 << LHS->getSourceRange() << RHS->getSourceRange(); 3440} 3441 3442//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 3443//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 3444 3445namespace { 3446 3447/// Structure recording the 'active' range of an integer-valued 3448/// expression. 3449struct IntRange { 3450 /// The number of bits active in the int. 3451 unsigned Width; 3452 3453 /// True if the int is known not to have negative values. 3454 bool NonNegative; 3455 3456 IntRange(unsigned Width, bool NonNegative) 3457 : Width(Width), NonNegative(NonNegative) 3458 {} 3459 3460 /// Returns the range of the bool type. 3461 static IntRange forBoolType() { 3462 return IntRange(1, true); 3463 } 3464 3465 /// Returns the range of an opaque value of the given integral type. 3466 static IntRange forValueOfType(ASTContext &C, QualType T) { 3467 return forValueOfCanonicalType(C, 3468 T->getCanonicalTypeInternal().getTypePtr()); 3469 } 3470 3471 /// Returns the range of an opaque value of a canonical integral type. 3472 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 3473 assert(T->isCanonicalUnqualified()); 3474 3475 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3476 T = VT->getElementType().getTypePtr(); 3477 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3478 T = CT->getElementType().getTypePtr(); 3479 3480 // For enum types, use the known bit width of the enumerators. 3481 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 3482 EnumDecl *Enum = ET->getDecl(); 3483 if (!Enum->isCompleteDefinition()) 3484 return IntRange(C.getIntWidth(QualType(T, 0)), false); 3485 3486 unsigned NumPositive = Enum->getNumPositiveBits(); 3487 unsigned NumNegative = Enum->getNumNegativeBits(); 3488 3489 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 3490 } 3491 3492 const BuiltinType *BT = cast<BuiltinType>(T); 3493 assert(BT->isInteger()); 3494 3495 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3496 } 3497 3498 /// Returns the "target" range of a canonical integral type, i.e. 3499 /// the range of values expressible in the type. 3500 /// 3501 /// This matches forValueOfCanonicalType except that enums have the 3502 /// full range of their type, not the range of their enumerators. 3503 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 3504 assert(T->isCanonicalUnqualified()); 3505 3506 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3507 T = VT->getElementType().getTypePtr(); 3508 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3509 T = CT->getElementType().getTypePtr(); 3510 if (const EnumType *ET = dyn_cast<EnumType>(T)) 3511 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 3512 3513 const BuiltinType *BT = cast<BuiltinType>(T); 3514 assert(BT->isInteger()); 3515 3516 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3517 } 3518 3519 /// Returns the supremum of two ranges: i.e. their conservative merge. 3520 static IntRange join(IntRange L, IntRange R) { 3521 return IntRange(std::max(L.Width, R.Width), 3522 L.NonNegative && R.NonNegative); 3523 } 3524 3525 /// Returns the infinum of two ranges: i.e. their aggressive merge. 3526 static IntRange meet(IntRange L, IntRange R) { 3527 return IntRange(std::min(L.Width, R.Width), 3528 L.NonNegative || R.NonNegative); 3529 } 3530}; 3531 3532static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, 3533 unsigned MaxWidth) { 3534 if (value.isSigned() && value.isNegative()) 3535 return IntRange(value.getMinSignedBits(), false); 3536 3537 if (value.getBitWidth() > MaxWidth) 3538 value = value.trunc(MaxWidth); 3539 3540 // isNonNegative() just checks the sign bit without considering 3541 // signedness. 3542 return IntRange(value.getActiveBits(), true); 3543} 3544 3545static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 3546 unsigned MaxWidth) { 3547 if (result.isInt()) 3548 return GetValueRange(C, result.getInt(), MaxWidth); 3549 3550 if (result.isVector()) { 3551 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 3552 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 3553 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 3554 R = IntRange::join(R, El); 3555 } 3556 return R; 3557 } 3558 3559 if (result.isComplexInt()) { 3560 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 3561 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 3562 return IntRange::join(R, I); 3563 } 3564 3565 // This can happen with lossless casts to intptr_t of "based" lvalues. 3566 // Assume it might use arbitrary bits. 3567 // FIXME: The only reason we need to pass the type in here is to get 3568 // the sign right on this one case. It would be nice if APValue 3569 // preserved this. 3570 assert(result.isLValue() || result.isAddrLabelDiff()); 3571 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 3572} 3573 3574/// Pseudo-evaluate the given integer expression, estimating the 3575/// range of values it might take. 3576/// 3577/// \param MaxWidth - the width to which the value will be truncated 3578static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 3579 E = E->IgnoreParens(); 3580 3581 // Try a full evaluation first. 3582 Expr::EvalResult result; 3583 if (E->EvaluateAsRValue(result, C)) 3584 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 3585 3586 // I think we only want to look through implicit casts here; if the 3587 // user has an explicit widening cast, we should treat the value as 3588 // being of the new, wider type. 3589 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 3590 if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue) 3591 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 3592 3593 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 3594 3595 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 3596 3597 // Assume that non-integer casts can span the full range of the type. 3598 if (!isIntegerCast) 3599 return OutputTypeRange; 3600 3601 IntRange SubRange 3602 = GetExprRange(C, CE->getSubExpr(), 3603 std::min(MaxWidth, OutputTypeRange.Width)); 3604 3605 // Bail out if the subexpr's range is as wide as the cast type. 3606 if (SubRange.Width >= OutputTypeRange.Width) 3607 return OutputTypeRange; 3608 3609 // Otherwise, we take the smaller width, and we're non-negative if 3610 // either the output type or the subexpr is. 3611 return IntRange(SubRange.Width, 3612 SubRange.NonNegative || OutputTypeRange.NonNegative); 3613 } 3614 3615 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 3616 // If we can fold the condition, just take that operand. 3617 bool CondResult; 3618 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 3619 return GetExprRange(C, CondResult ? CO->getTrueExpr() 3620 : CO->getFalseExpr(), 3621 MaxWidth); 3622 3623 // Otherwise, conservatively merge. 3624 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 3625 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 3626 return IntRange::join(L, R); 3627 } 3628 3629 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 3630 switch (BO->getOpcode()) { 3631 3632 // Boolean-valued operations are single-bit and positive. 3633 case BO_LAnd: 3634 case BO_LOr: 3635 case BO_LT: 3636 case BO_GT: 3637 case BO_LE: 3638 case BO_GE: 3639 case BO_EQ: 3640 case BO_NE: 3641 return IntRange::forBoolType(); 3642 3643 // The type of the assignments is the type of the LHS, so the RHS 3644 // is not necessarily the same type. 3645 case BO_MulAssign: 3646 case BO_DivAssign: 3647 case BO_RemAssign: 3648 case BO_AddAssign: 3649 case BO_SubAssign: 3650 case BO_XorAssign: 3651 case BO_OrAssign: 3652 // TODO: bitfields? 3653 return IntRange::forValueOfType(C, E->getType()); 3654 3655 // Simple assignments just pass through the RHS, which will have 3656 // been coerced to the LHS type. 3657 case BO_Assign: 3658 // TODO: bitfields? 3659 return GetExprRange(C, BO->getRHS(), MaxWidth); 3660 3661 // Operations with opaque sources are black-listed. 3662 case BO_PtrMemD: 3663 case BO_PtrMemI: 3664 return IntRange::forValueOfType(C, E->getType()); 3665 3666 // Bitwise-and uses the *infinum* of the two source ranges. 3667 case BO_And: 3668 case BO_AndAssign: 3669 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 3670 GetExprRange(C, BO->getRHS(), MaxWidth)); 3671 3672 // Left shift gets black-listed based on a judgement call. 3673 case BO_Shl: 3674 // ...except that we want to treat '1 << (blah)' as logically 3675 // positive. It's an important idiom. 3676 if (IntegerLiteral *I 3677 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 3678 if (I->getValue() == 1) { 3679 IntRange R = IntRange::forValueOfType(C, E->getType()); 3680 return IntRange(R.Width, /*NonNegative*/ true); 3681 } 3682 } 3683 // fallthrough 3684 3685 case BO_ShlAssign: 3686 return IntRange::forValueOfType(C, E->getType()); 3687 3688 // Right shift by a constant can narrow its left argument. 3689 case BO_Shr: 3690 case BO_ShrAssign: { 3691 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3692 3693 // If the shift amount is a positive constant, drop the width by 3694 // that much. 3695 llvm::APSInt shift; 3696 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 3697 shift.isNonNegative()) { 3698 unsigned zext = shift.getZExtValue(); 3699 if (zext >= L.Width) 3700 L.Width = (L.NonNegative ? 0 : 1); 3701 else 3702 L.Width -= zext; 3703 } 3704 3705 return L; 3706 } 3707 3708 // Comma acts as its right operand. 3709 case BO_Comma: 3710 return GetExprRange(C, BO->getRHS(), MaxWidth); 3711 3712 // Black-list pointer subtractions. 3713 case BO_Sub: 3714 if (BO->getLHS()->getType()->isPointerType()) 3715 return IntRange::forValueOfType(C, E->getType()); 3716 break; 3717 3718 // The width of a division result is mostly determined by the size 3719 // of the LHS. 3720 case BO_Div: { 3721 // Don't 'pre-truncate' the operands. 3722 unsigned opWidth = C.getIntWidth(E->getType()); 3723 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3724 3725 // If the divisor is constant, use that. 3726 llvm::APSInt divisor; 3727 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 3728 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 3729 if (log2 >= L.Width) 3730 L.Width = (L.NonNegative ? 0 : 1); 3731 else 3732 L.Width = std::min(L.Width - log2, MaxWidth); 3733 return L; 3734 } 3735 3736 // Otherwise, just use the LHS's width. 3737 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3738 return IntRange(L.Width, L.NonNegative && R.NonNegative); 3739 } 3740 3741 // The result of a remainder can't be larger than the result of 3742 // either side. 3743 case BO_Rem: { 3744 // Don't 'pre-truncate' the operands. 3745 unsigned opWidth = C.getIntWidth(E->getType()); 3746 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3747 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3748 3749 IntRange meet = IntRange::meet(L, R); 3750 meet.Width = std::min(meet.Width, MaxWidth); 3751 return meet; 3752 } 3753 3754 // The default behavior is okay for these. 3755 case BO_Mul: 3756 case BO_Add: 3757 case BO_Xor: 3758 case BO_Or: 3759 break; 3760 } 3761 3762 // The default case is to treat the operation as if it were closed 3763 // on the narrowest type that encompasses both operands. 3764 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3765 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 3766 return IntRange::join(L, R); 3767 } 3768 3769 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 3770 switch (UO->getOpcode()) { 3771 // Boolean-valued operations are white-listed. 3772 case UO_LNot: 3773 return IntRange::forBoolType(); 3774 3775 // Operations with opaque sources are black-listed. 3776 case UO_Deref: 3777 case UO_AddrOf: // should be impossible 3778 return IntRange::forValueOfType(C, E->getType()); 3779 3780 default: 3781 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 3782 } 3783 } 3784 3785 if (dyn_cast<OffsetOfExpr>(E)) { 3786 IntRange::forValueOfType(C, E->getType()); 3787 } 3788 3789 if (FieldDecl *BitField = E->getBitField()) 3790 return IntRange(BitField->getBitWidthValue(C), 3791 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 3792 3793 return IntRange::forValueOfType(C, E->getType()); 3794} 3795 3796static IntRange GetExprRange(ASTContext &C, Expr *E) { 3797 return GetExprRange(C, E, C.getIntWidth(E->getType())); 3798} 3799 3800/// Checks whether the given value, which currently has the given 3801/// source semantics, has the same value when coerced through the 3802/// target semantics. 3803static bool IsSameFloatAfterCast(const llvm::APFloat &value, 3804 const llvm::fltSemantics &Src, 3805 const llvm::fltSemantics &Tgt) { 3806 llvm::APFloat truncated = value; 3807 3808 bool ignored; 3809 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 3810 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 3811 3812 return truncated.bitwiseIsEqual(value); 3813} 3814 3815/// Checks whether the given value, which currently has the given 3816/// source semantics, has the same value when coerced through the 3817/// target semantics. 3818/// 3819/// The value might be a vector of floats (or a complex number). 3820static bool IsSameFloatAfterCast(const APValue &value, 3821 const llvm::fltSemantics &Src, 3822 const llvm::fltSemantics &Tgt) { 3823 if (value.isFloat()) 3824 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 3825 3826 if (value.isVector()) { 3827 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 3828 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 3829 return false; 3830 return true; 3831 } 3832 3833 assert(value.isComplexFloat()); 3834 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 3835 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 3836} 3837 3838static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 3839 3840static bool IsZero(Sema &S, Expr *E) { 3841 // Suppress cases where we are comparing against an enum constant. 3842 if (const DeclRefExpr *DR = 3843 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 3844 if (isa<EnumConstantDecl>(DR->getDecl())) 3845 return false; 3846 3847 // Suppress cases where the '0' value is expanded from a macro. 3848 if (E->getLocStart().isMacroID()) 3849 return false; 3850 3851 llvm::APSInt Value; 3852 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 3853} 3854 3855static bool HasEnumType(Expr *E) { 3856 // Strip off implicit integral promotions. 3857 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 3858 if (ICE->getCastKind() != CK_IntegralCast && 3859 ICE->getCastKind() != CK_NoOp) 3860 break; 3861 E = ICE->getSubExpr(); 3862 } 3863 3864 return E->getType()->isEnumeralType(); 3865} 3866 3867static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 3868 BinaryOperatorKind op = E->getOpcode(); 3869 if (E->isValueDependent()) 3870 return; 3871 3872 if (op == BO_LT && IsZero(S, E->getRHS())) { 3873 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3874 << "< 0" << "false" << HasEnumType(E->getLHS()) 3875 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3876 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 3877 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3878 << ">= 0" << "true" << HasEnumType(E->getLHS()) 3879 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3880 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 3881 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3882 << "0 >" << "false" << HasEnumType(E->getRHS()) 3883 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3884 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 3885 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3886 << "0 <=" << "true" << HasEnumType(E->getRHS()) 3887 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3888 } 3889} 3890 3891/// Analyze the operands of the given comparison. Implements the 3892/// fallback case from AnalyzeComparison. 3893static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 3894 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3895 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3896} 3897 3898/// \brief Implements -Wsign-compare. 3899/// 3900/// \param E the binary operator to check for warnings 3901static void AnalyzeComparison(Sema &S, BinaryOperator *E) { 3902 // The type the comparison is being performed in. 3903 QualType T = E->getLHS()->getType(); 3904 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 3905 && "comparison with mismatched types"); 3906 3907 // We don't do anything special if this isn't an unsigned integral 3908 // comparison: we're only interested in integral comparisons, and 3909 // signed comparisons only happen in cases we don't care to warn about. 3910 // 3911 // We also don't care about value-dependent expressions or expressions 3912 // whose result is a constant. 3913 if (!T->hasUnsignedIntegerRepresentation() 3914 || E->isValueDependent() || E->isIntegerConstantExpr(S.Context)) 3915 return AnalyzeImpConvsInComparison(S, E); 3916 3917 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 3918 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 3919 3920 // Check to see if one of the (unmodified) operands is of different 3921 // signedness. 3922 Expr *signedOperand, *unsignedOperand; 3923 if (LHS->getType()->hasSignedIntegerRepresentation()) { 3924 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 3925 "unsigned comparison between two signed integer expressions?"); 3926 signedOperand = LHS; 3927 unsignedOperand = RHS; 3928 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 3929 signedOperand = RHS; 3930 unsignedOperand = LHS; 3931 } else { 3932 CheckTrivialUnsignedComparison(S, E); 3933 return AnalyzeImpConvsInComparison(S, E); 3934 } 3935 3936 // Otherwise, calculate the effective range of the signed operand. 3937 IntRange signedRange = GetExprRange(S.Context, signedOperand); 3938 3939 // Go ahead and analyze implicit conversions in the operands. Note 3940 // that we skip the implicit conversions on both sides. 3941 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 3942 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 3943 3944 // If the signed range is non-negative, -Wsign-compare won't fire, 3945 // but we should still check for comparisons which are always true 3946 // or false. 3947 if (signedRange.NonNegative) 3948 return CheckTrivialUnsignedComparison(S, E); 3949 3950 // For (in)equality comparisons, if the unsigned operand is a 3951 // constant which cannot collide with a overflowed signed operand, 3952 // then reinterpreting the signed operand as unsigned will not 3953 // change the result of the comparison. 3954 if (E->isEqualityOp()) { 3955 unsigned comparisonWidth = S.Context.getIntWidth(T); 3956 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 3957 3958 // We should never be unable to prove that the unsigned operand is 3959 // non-negative. 3960 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 3961 3962 if (unsignedRange.Width < comparisonWidth) 3963 return; 3964 } 3965 3966 S.DiagRuntimeBehavior(E->getOperatorLoc(), E, 3967 S.PDiag(diag::warn_mixed_sign_comparison) 3968 << LHS->getType() << RHS->getType() 3969 << LHS->getSourceRange() << RHS->getSourceRange()); 3970} 3971 3972/// Analyzes an attempt to assign the given value to a bitfield. 3973/// 3974/// Returns true if there was something fishy about the attempt. 3975static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 3976 SourceLocation InitLoc) { 3977 assert(Bitfield->isBitField()); 3978 if (Bitfield->isInvalidDecl()) 3979 return false; 3980 3981 // White-list bool bitfields. 3982 if (Bitfield->getType()->isBooleanType()) 3983 return false; 3984 3985 // Ignore value- or type-dependent expressions. 3986 if (Bitfield->getBitWidth()->isValueDependent() || 3987 Bitfield->getBitWidth()->isTypeDependent() || 3988 Init->isValueDependent() || 3989 Init->isTypeDependent()) 3990 return false; 3991 3992 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 3993 3994 llvm::APSInt Value; 3995 if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects)) 3996 return false; 3997 3998 unsigned OriginalWidth = Value.getBitWidth(); 3999 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 4000 4001 if (OriginalWidth <= FieldWidth) 4002 return false; 4003 4004 // Compute the value which the bitfield will contain. 4005 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 4006 TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType()); 4007 4008 // Check whether the stored value is equal to the original value. 4009 TruncatedValue = TruncatedValue.extend(OriginalWidth); 4010 if (Value == TruncatedValue) 4011 return false; 4012 4013 // Special-case bitfields of width 1: booleans are naturally 0/1, and 4014 // therefore don't strictly fit into a signed bitfield of width 1. 4015 if (FieldWidth == 1 && Value == 1) 4016 return false; 4017 4018 std::string PrettyValue = Value.toString(10); 4019 std::string PrettyTrunc = TruncatedValue.toString(10); 4020 4021 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 4022 << PrettyValue << PrettyTrunc << OriginalInit->getType() 4023 << Init->getSourceRange(); 4024 4025 return true; 4026} 4027 4028/// Analyze the given simple or compound assignment for warning-worthy 4029/// operations. 4030static void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 4031 // Just recurse on the LHS. 4032 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4033 4034 // We want to recurse on the RHS as normal unless we're assigning to 4035 // a bitfield. 4036 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 4037 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 4038 E->getOperatorLoc())) { 4039 // Recurse, ignoring any implicit conversions on the RHS. 4040 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 4041 E->getOperatorLoc()); 4042 } 4043 } 4044 4045 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4046} 4047 4048/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4049static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 4050 SourceLocation CContext, unsigned diag, 4051 bool pruneControlFlow = false) { 4052 if (pruneControlFlow) { 4053 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4054 S.PDiag(diag) 4055 << SourceType << T << E->getSourceRange() 4056 << SourceRange(CContext)); 4057 return; 4058 } 4059 S.Diag(E->getExprLoc(), diag) 4060 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 4061} 4062 4063/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4064static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, 4065 SourceLocation CContext, unsigned diag, 4066 bool pruneControlFlow = false) { 4067 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow); 4068} 4069 4070/// Diagnose an implicit cast from a literal expression. Does not warn when the 4071/// cast wouldn't lose information. 4072void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 4073 SourceLocation CContext) { 4074 // Try to convert the literal exactly to an integer. If we can, don't warn. 4075 bool isExact = false; 4076 const llvm::APFloat &Value = FL->getValue(); 4077 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 4078 T->hasUnsignedIntegerRepresentation()); 4079 if (Value.convertToInteger(IntegerValue, 4080 llvm::APFloat::rmTowardZero, &isExact) 4081 == llvm::APFloat::opOK && isExact) 4082 return; 4083 4084 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 4085 << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext); 4086} 4087 4088std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 4089 if (!Range.Width) return "0"; 4090 4091 llvm::APSInt ValueInRange = Value; 4092 ValueInRange.setIsSigned(!Range.NonNegative); 4093 ValueInRange = ValueInRange.trunc(Range.Width); 4094 return ValueInRange.toString(10); 4095} 4096 4097void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 4098 SourceLocation CC, bool *ICContext = 0) { 4099 if (E->isTypeDependent() || E->isValueDependent()) return; 4100 4101 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 4102 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 4103 if (Source == Target) return; 4104 if (Target->isDependentType()) return; 4105 4106 // If the conversion context location is invalid don't complain. We also 4107 // don't want to emit a warning if the issue occurs from the expansion of 4108 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 4109 // delay this check as long as possible. Once we detect we are in that 4110 // scenario, we just return. 4111 if (CC.isInvalid()) 4112 return; 4113 4114 // Diagnose implicit casts to bool. 4115 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 4116 if (isa<StringLiteral>(E)) 4117 // Warn on string literal to bool. Checks for string literals in logical 4118 // expressions, for instances, assert(0 && "error here"), is prevented 4119 // by a check in AnalyzeImplicitConversions(). 4120 return DiagnoseImpCast(S, E, T, CC, 4121 diag::warn_impcast_string_literal_to_bool); 4122 if (Source->isFunctionType()) { 4123 // Warn on function to bool. Checks free functions and static member 4124 // functions. Weakly imported functions are excluded from the check, 4125 // since it's common to test their value to check whether the linker 4126 // found a definition for them. 4127 ValueDecl *D = 0; 4128 if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) { 4129 D = R->getDecl(); 4130 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 4131 D = M->getMemberDecl(); 4132 } 4133 4134 if (D && !D->isWeak()) { 4135 if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) { 4136 S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool) 4137 << F << E->getSourceRange() << SourceRange(CC); 4138 S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence) 4139 << FixItHint::CreateInsertion(E->getExprLoc(), "&"); 4140 QualType ReturnType; 4141 UnresolvedSet<4> NonTemplateOverloads; 4142 S.isExprCallable(*E, ReturnType, NonTemplateOverloads); 4143 if (!ReturnType.isNull() 4144 && ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 4145 S.Diag(E->getExprLoc(), diag::note_function_to_bool_call) 4146 << FixItHint::CreateInsertion( 4147 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()"); 4148 return; 4149 } 4150 } 4151 } 4152 return; // Other casts to bool are not checked. 4153 } 4154 4155 // Strip vector types. 4156 if (isa<VectorType>(Source)) { 4157 if (!isa<VectorType>(Target)) { 4158 if (S.SourceMgr.isInSystemMacro(CC)) 4159 return; 4160 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 4161 } 4162 4163 // If the vector cast is cast between two vectors of the same size, it is 4164 // a bitcast, not a conversion. 4165 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 4166 return; 4167 4168 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 4169 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 4170 } 4171 4172 // Strip complex types. 4173 if (isa<ComplexType>(Source)) { 4174 if (!isa<ComplexType>(Target)) { 4175 if (S.SourceMgr.isInSystemMacro(CC)) 4176 return; 4177 4178 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 4179 } 4180 4181 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 4182 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 4183 } 4184 4185 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 4186 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 4187 4188 // If the source is floating point... 4189 if (SourceBT && SourceBT->isFloatingPoint()) { 4190 // ...and the target is floating point... 4191 if (TargetBT && TargetBT->isFloatingPoint()) { 4192 // ...then warn if we're dropping FP rank. 4193 4194 // Builtin FP kinds are ordered by increasing FP rank. 4195 if (SourceBT->getKind() > TargetBT->getKind()) { 4196 // Don't warn about float constants that are precisely 4197 // representable in the target type. 4198 Expr::EvalResult result; 4199 if (E->EvaluateAsRValue(result, S.Context)) { 4200 // Value might be a float, a float vector, or a float complex. 4201 if (IsSameFloatAfterCast(result.Val, 4202 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 4203 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 4204 return; 4205 } 4206 4207 if (S.SourceMgr.isInSystemMacro(CC)) 4208 return; 4209 4210 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 4211 } 4212 return; 4213 } 4214 4215 // If the target is integral, always warn. 4216 if ((TargetBT && TargetBT->isInteger())) { 4217 if (S.SourceMgr.isInSystemMacro(CC)) 4218 return; 4219 4220 Expr *InnerE = E->IgnoreParenImpCasts(); 4221 // We also want to warn on, e.g., "int i = -1.234" 4222 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 4223 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 4224 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 4225 4226 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 4227 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 4228 } else { 4229 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 4230 } 4231 } 4232 4233 return; 4234 } 4235 4236 if (!Source->isIntegerType() || !Target->isIntegerType()) 4237 return; 4238 4239 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 4240 == Expr::NPCK_GNUNull) && Target->isIntegerType()) { 4241 SourceLocation Loc = E->getSourceRange().getBegin(); 4242 if (Loc.isMacroID()) 4243 Loc = S.SourceMgr.getImmediateExpansionRange(Loc).first; 4244 S.Diag(Loc, diag::warn_impcast_null_pointer_to_integer) 4245 << T << clang::SourceRange(CC) 4246 << FixItHint::CreateReplacement(Loc, S.getFixItZeroLiteralForType(T)); 4247 return; 4248 } 4249 4250 IntRange SourceRange = GetExprRange(S.Context, E); 4251 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 4252 4253 if (SourceRange.Width > TargetRange.Width) { 4254 // If the source is a constant, use a default-on diagnostic. 4255 // TODO: this should happen for bitfield stores, too. 4256 llvm::APSInt Value(32); 4257 if (E->isIntegerConstantExpr(Value, S.Context)) { 4258 if (S.SourceMgr.isInSystemMacro(CC)) 4259 return; 4260 4261 std::string PrettySourceValue = Value.toString(10); 4262 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 4263 4264 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4265 S.PDiag(diag::warn_impcast_integer_precision_constant) 4266 << PrettySourceValue << PrettyTargetValue 4267 << E->getType() << T << E->getSourceRange() 4268 << clang::SourceRange(CC)); 4269 return; 4270 } 4271 4272 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 4273 if (S.SourceMgr.isInSystemMacro(CC)) 4274 return; 4275 4276 if (TargetRange.Width == 32 && S.Context.getIntWidth(E->getType()) == 64) 4277 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32, 4278 /* pruneControlFlow */ true); 4279 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 4280 } 4281 4282 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 4283 (!TargetRange.NonNegative && SourceRange.NonNegative && 4284 SourceRange.Width == TargetRange.Width)) { 4285 4286 if (S.SourceMgr.isInSystemMacro(CC)) 4287 return; 4288 4289 unsigned DiagID = diag::warn_impcast_integer_sign; 4290 4291 // Traditionally, gcc has warned about this under -Wsign-compare. 4292 // We also want to warn about it in -Wconversion. 4293 // So if -Wconversion is off, use a completely identical diagnostic 4294 // in the sign-compare group. 4295 // The conditional-checking code will 4296 if (ICContext) { 4297 DiagID = diag::warn_impcast_integer_sign_conditional; 4298 *ICContext = true; 4299 } 4300 4301 return DiagnoseImpCast(S, E, T, CC, DiagID); 4302 } 4303 4304 // Diagnose conversions between different enumeration types. 4305 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 4306 // type, to give us better diagnostics. 4307 QualType SourceType = E->getType(); 4308 if (!S.getLangOpts().CPlusPlus) { 4309 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 4310 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 4311 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 4312 SourceType = S.Context.getTypeDeclType(Enum); 4313 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 4314 } 4315 } 4316 4317 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 4318 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 4319 if ((SourceEnum->getDecl()->getIdentifier() || 4320 SourceEnum->getDecl()->getTypedefNameForAnonDecl()) && 4321 (TargetEnum->getDecl()->getIdentifier() || 4322 TargetEnum->getDecl()->getTypedefNameForAnonDecl()) && 4323 SourceEnum != TargetEnum) { 4324 if (S.SourceMgr.isInSystemMacro(CC)) 4325 return; 4326 4327 return DiagnoseImpCast(S, E, SourceType, T, CC, 4328 diag::warn_impcast_different_enum_types); 4329 } 4330 4331 return; 4332} 4333 4334void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 4335 4336void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 4337 SourceLocation CC, bool &ICContext) { 4338 E = E->IgnoreParenImpCasts(); 4339 4340 if (isa<ConditionalOperator>(E)) 4341 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 4342 4343 AnalyzeImplicitConversions(S, E, CC); 4344 if (E->getType() != T) 4345 return CheckImplicitConversion(S, E, T, CC, &ICContext); 4346 return; 4347} 4348 4349void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 4350 SourceLocation CC = E->getQuestionLoc(); 4351 4352 AnalyzeImplicitConversions(S, E->getCond(), CC); 4353 4354 bool Suspicious = false; 4355 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 4356 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 4357 4358 // If -Wconversion would have warned about either of the candidates 4359 // for a signedness conversion to the context type... 4360 if (!Suspicious) return; 4361 4362 // ...but it's currently ignored... 4363 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 4364 CC)) 4365 return; 4366 4367 // ...then check whether it would have warned about either of the 4368 // candidates for a signedness conversion to the condition type. 4369 if (E->getType() == T) return; 4370 4371 Suspicious = false; 4372 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 4373 E->getType(), CC, &Suspicious); 4374 if (!Suspicious) 4375 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 4376 E->getType(), CC, &Suspicious); 4377} 4378 4379/// AnalyzeImplicitConversions - Find and report any interesting 4380/// implicit conversions in the given expression. There are a couple 4381/// of competing diagnostics here, -Wconversion and -Wsign-compare. 4382void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 4383 QualType T = OrigE->getType(); 4384 Expr *E = OrigE->IgnoreParenImpCasts(); 4385 4386 if (E->isTypeDependent() || E->isValueDependent()) 4387 return; 4388 4389 // For conditional operators, we analyze the arguments as if they 4390 // were being fed directly into the output. 4391 if (isa<ConditionalOperator>(E)) { 4392 ConditionalOperator *CO = cast<ConditionalOperator>(E); 4393 CheckConditionalOperator(S, CO, T); 4394 return; 4395 } 4396 4397 // Go ahead and check any implicit conversions we might have skipped. 4398 // The non-canonical typecheck is just an optimization; 4399 // CheckImplicitConversion will filter out dead implicit conversions. 4400 if (E->getType() != T) 4401 CheckImplicitConversion(S, E, T, CC); 4402 4403 // Now continue drilling into this expression. 4404 4405 // Skip past explicit casts. 4406 if (isa<ExplicitCastExpr>(E)) { 4407 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 4408 return AnalyzeImplicitConversions(S, E, CC); 4409 } 4410 4411 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 4412 // Do a somewhat different check with comparison operators. 4413 if (BO->isComparisonOp()) 4414 return AnalyzeComparison(S, BO); 4415 4416 // And with simple assignments. 4417 if (BO->getOpcode() == BO_Assign) 4418 return AnalyzeAssignment(S, BO); 4419 } 4420 4421 // These break the otherwise-useful invariant below. Fortunately, 4422 // we don't really need to recurse into them, because any internal 4423 // expressions should have been analyzed already when they were 4424 // built into statements. 4425 if (isa<StmtExpr>(E)) return; 4426 4427 // Don't descend into unevaluated contexts. 4428 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 4429 4430 // Now just recurse over the expression's children. 4431 CC = E->getExprLoc(); 4432 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 4433 bool IsLogicalOperator = BO && BO->isLogicalOp(); 4434 for (Stmt::child_range I = E->children(); I; ++I) { 4435 Expr *ChildExpr = dyn_cast_or_null<Expr>(*I); 4436 if (!ChildExpr) 4437 continue; 4438 4439 if (IsLogicalOperator && 4440 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 4441 // Ignore checking string literals that are in logical operators. 4442 continue; 4443 AnalyzeImplicitConversions(S, ChildExpr, CC); 4444 } 4445} 4446 4447} // end anonymous namespace 4448 4449/// Diagnoses "dangerous" implicit conversions within the given 4450/// expression (which is a full expression). Implements -Wconversion 4451/// and -Wsign-compare. 4452/// 4453/// \param CC the "context" location of the implicit conversion, i.e. 4454/// the most location of the syntactic entity requiring the implicit 4455/// conversion 4456void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 4457 // Don't diagnose in unevaluated contexts. 4458 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 4459 return; 4460 4461 // Don't diagnose for value- or type-dependent expressions. 4462 if (E->isTypeDependent() || E->isValueDependent()) 4463 return; 4464 4465 // Check for array bounds violations in cases where the check isn't triggered 4466 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 4467 // ArraySubscriptExpr is on the RHS of a variable initialization. 4468 CheckArrayAccess(E); 4469 4470 // This is not the right CC for (e.g.) a variable initialization. 4471 AnalyzeImplicitConversions(*this, E, CC); 4472} 4473 4474void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 4475 FieldDecl *BitField, 4476 Expr *Init) { 4477 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 4478} 4479 4480/// CheckParmsForFunctionDef - Check that the parameters of the given 4481/// function are appropriate for the definition of a function. This 4482/// takes care of any checks that cannot be performed on the 4483/// declaration itself, e.g., that the types of each of the function 4484/// parameters are complete. 4485bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 4486 bool CheckParameterNames) { 4487 bool HasInvalidParm = false; 4488 for (; P != PEnd; ++P) { 4489 ParmVarDecl *Param = *P; 4490 4491 // C99 6.7.5.3p4: the parameters in a parameter type list in a 4492 // function declarator that is part of a function definition of 4493 // that function shall not have incomplete type. 4494 // 4495 // This is also C++ [dcl.fct]p6. 4496 if (!Param->isInvalidDecl() && 4497 RequireCompleteType(Param->getLocation(), Param->getType(), 4498 diag::err_typecheck_decl_incomplete_type)) { 4499 Param->setInvalidDecl(); 4500 HasInvalidParm = true; 4501 } 4502 4503 // C99 6.9.1p5: If the declarator includes a parameter type list, the 4504 // declaration of each parameter shall include an identifier. 4505 if (CheckParameterNames && 4506 Param->getIdentifier() == 0 && 4507 !Param->isImplicit() && 4508 !getLangOpts().CPlusPlus) 4509 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 4510 4511 // C99 6.7.5.3p12: 4512 // If the function declarator is not part of a definition of that 4513 // function, parameters may have incomplete type and may use the [*] 4514 // notation in their sequences of declarator specifiers to specify 4515 // variable length array types. 4516 QualType PType = Param->getOriginalType(); 4517 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 4518 if (AT->getSizeModifier() == ArrayType::Star) { 4519 // FIXME: This diagnosic should point the the '[*]' if source-location 4520 // information is added for it. 4521 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 4522 } 4523 } 4524 } 4525 4526 return HasInvalidParm; 4527} 4528 4529/// CheckCastAlign - Implements -Wcast-align, which warns when a 4530/// pointer cast increases the alignment requirements. 4531void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 4532 // This is actually a lot of work to potentially be doing on every 4533 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 4534 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 4535 TRange.getBegin()) 4536 == DiagnosticsEngine::Ignored) 4537 return; 4538 4539 // Ignore dependent types. 4540 if (T->isDependentType() || Op->getType()->isDependentType()) 4541 return; 4542 4543 // Require that the destination be a pointer type. 4544 const PointerType *DestPtr = T->getAs<PointerType>(); 4545 if (!DestPtr) return; 4546 4547 // If the destination has alignment 1, we're done. 4548 QualType DestPointee = DestPtr->getPointeeType(); 4549 if (DestPointee->isIncompleteType()) return; 4550 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 4551 if (DestAlign.isOne()) return; 4552 4553 // Require that the source be a pointer type. 4554 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 4555 if (!SrcPtr) return; 4556 QualType SrcPointee = SrcPtr->getPointeeType(); 4557 4558 // Whitelist casts from cv void*. We already implicitly 4559 // whitelisted casts to cv void*, since they have alignment 1. 4560 // Also whitelist casts involving incomplete types, which implicitly 4561 // includes 'void'. 4562 if (SrcPointee->isIncompleteType()) return; 4563 4564 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 4565 if (SrcAlign >= DestAlign) return; 4566 4567 Diag(TRange.getBegin(), diag::warn_cast_align) 4568 << Op->getType() << T 4569 << static_cast<unsigned>(SrcAlign.getQuantity()) 4570 << static_cast<unsigned>(DestAlign.getQuantity()) 4571 << TRange << Op->getSourceRange(); 4572} 4573 4574static const Type* getElementType(const Expr *BaseExpr) { 4575 const Type* EltType = BaseExpr->getType().getTypePtr(); 4576 if (EltType->isAnyPointerType()) 4577 return EltType->getPointeeType().getTypePtr(); 4578 else if (EltType->isArrayType()) 4579 return EltType->getBaseElementTypeUnsafe(); 4580 return EltType; 4581} 4582 4583/// \brief Check whether this array fits the idiom of a size-one tail padded 4584/// array member of a struct. 4585/// 4586/// We avoid emitting out-of-bounds access warnings for such arrays as they are 4587/// commonly used to emulate flexible arrays in C89 code. 4588static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 4589 const NamedDecl *ND) { 4590 if (Size != 1 || !ND) return false; 4591 4592 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 4593 if (!FD) return false; 4594 4595 // Don't consider sizes resulting from macro expansions or template argument 4596 // substitution to form C89 tail-padded arrays. 4597 4598 TypeSourceInfo *TInfo = FD->getTypeSourceInfo(); 4599 while (TInfo) { 4600 TypeLoc TL = TInfo->getTypeLoc(); 4601 // Look through typedefs. 4602 const TypedefTypeLoc *TTL = dyn_cast<TypedefTypeLoc>(&TL); 4603 if (TTL) { 4604 const TypedefNameDecl *TDL = TTL->getTypedefNameDecl(); 4605 TInfo = TDL->getTypeSourceInfo(); 4606 continue; 4607 } 4608 ConstantArrayTypeLoc CTL = cast<ConstantArrayTypeLoc>(TL); 4609 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(CTL.getSizeExpr()); 4610 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 4611 return false; 4612 break; 4613 } 4614 4615 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 4616 if (!RD) return false; 4617 if (RD->isUnion()) return false; 4618 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 4619 if (!CRD->isStandardLayout()) return false; 4620 } 4621 4622 // See if this is the last field decl in the record. 4623 const Decl *D = FD; 4624 while ((D = D->getNextDeclInContext())) 4625 if (isa<FieldDecl>(D)) 4626 return false; 4627 return true; 4628} 4629 4630void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 4631 const ArraySubscriptExpr *ASE, 4632 bool AllowOnePastEnd, bool IndexNegated) { 4633 IndexExpr = IndexExpr->IgnoreParenImpCasts(); 4634 if (IndexExpr->isValueDependent()) 4635 return; 4636 4637 const Type *EffectiveType = getElementType(BaseExpr); 4638 BaseExpr = BaseExpr->IgnoreParenCasts(); 4639 const ConstantArrayType *ArrayTy = 4640 Context.getAsConstantArrayType(BaseExpr->getType()); 4641 if (!ArrayTy) 4642 return; 4643 4644 llvm::APSInt index; 4645 if (!IndexExpr->EvaluateAsInt(index, Context)) 4646 return; 4647 if (IndexNegated) 4648 index = -index; 4649 4650 const NamedDecl *ND = NULL; 4651 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4652 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 4653 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 4654 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 4655 4656 if (index.isUnsigned() || !index.isNegative()) { 4657 llvm::APInt size = ArrayTy->getSize(); 4658 if (!size.isStrictlyPositive()) 4659 return; 4660 4661 const Type* BaseType = getElementType(BaseExpr); 4662 if (BaseType != EffectiveType) { 4663 // Make sure we're comparing apples to apples when comparing index to size 4664 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 4665 uint64_t array_typesize = Context.getTypeSize(BaseType); 4666 // Handle ptrarith_typesize being zero, such as when casting to void* 4667 if (!ptrarith_typesize) ptrarith_typesize = 1; 4668 if (ptrarith_typesize != array_typesize) { 4669 // There's a cast to a different size type involved 4670 uint64_t ratio = array_typesize / ptrarith_typesize; 4671 // TODO: Be smarter about handling cases where array_typesize is not a 4672 // multiple of ptrarith_typesize 4673 if (ptrarith_typesize * ratio == array_typesize) 4674 size *= llvm::APInt(size.getBitWidth(), ratio); 4675 } 4676 } 4677 4678 if (size.getBitWidth() > index.getBitWidth()) 4679 index = index.zext(size.getBitWidth()); 4680 else if (size.getBitWidth() < index.getBitWidth()) 4681 size = size.zext(index.getBitWidth()); 4682 4683 // For array subscripting the index must be less than size, but for pointer 4684 // arithmetic also allow the index (offset) to be equal to size since 4685 // computing the next address after the end of the array is legal and 4686 // commonly done e.g. in C++ iterators and range-based for loops. 4687 if (AllowOnePastEnd ? index.ule(size) : index.ult(size)) 4688 return; 4689 4690 // Also don't warn for arrays of size 1 which are members of some 4691 // structure. These are often used to approximate flexible arrays in C89 4692 // code. 4693 if (IsTailPaddedMemberArray(*this, size, ND)) 4694 return; 4695 4696 // Suppress the warning if the subscript expression (as identified by the 4697 // ']' location) and the index expression are both from macro expansions 4698 // within a system header. 4699 if (ASE) { 4700 SourceLocation RBracketLoc = SourceMgr.getSpellingLoc( 4701 ASE->getRBracketLoc()); 4702 if (SourceMgr.isInSystemHeader(RBracketLoc)) { 4703 SourceLocation IndexLoc = SourceMgr.getSpellingLoc( 4704 IndexExpr->getLocStart()); 4705 if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc)) 4706 return; 4707 } 4708 } 4709 4710 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 4711 if (ASE) 4712 DiagID = diag::warn_array_index_exceeds_bounds; 4713 4714 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4715 PDiag(DiagID) << index.toString(10, true) 4716 << size.toString(10, true) 4717 << (unsigned)size.getLimitedValue(~0U) 4718 << IndexExpr->getSourceRange()); 4719 } else { 4720 unsigned DiagID = diag::warn_array_index_precedes_bounds; 4721 if (!ASE) { 4722 DiagID = diag::warn_ptr_arith_precedes_bounds; 4723 if (index.isNegative()) index = -index; 4724 } 4725 4726 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4727 PDiag(DiagID) << index.toString(10, true) 4728 << IndexExpr->getSourceRange()); 4729 } 4730 4731 if (!ND) { 4732 // Try harder to find a NamedDecl to point at in the note. 4733 while (const ArraySubscriptExpr *ASE = 4734 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 4735 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 4736 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4737 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 4738 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 4739 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 4740 } 4741 4742 if (ND) 4743 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 4744 PDiag(diag::note_array_index_out_of_bounds) 4745 << ND->getDeclName()); 4746} 4747 4748void Sema::CheckArrayAccess(const Expr *expr) { 4749 int AllowOnePastEnd = 0; 4750 while (expr) { 4751 expr = expr->IgnoreParenImpCasts(); 4752 switch (expr->getStmtClass()) { 4753 case Stmt::ArraySubscriptExprClass: { 4754 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 4755 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE, 4756 AllowOnePastEnd > 0); 4757 return; 4758 } 4759 case Stmt::UnaryOperatorClass: { 4760 // Only unwrap the * and & unary operators 4761 const UnaryOperator *UO = cast<UnaryOperator>(expr); 4762 expr = UO->getSubExpr(); 4763 switch (UO->getOpcode()) { 4764 case UO_AddrOf: 4765 AllowOnePastEnd++; 4766 break; 4767 case UO_Deref: 4768 AllowOnePastEnd--; 4769 break; 4770 default: 4771 return; 4772 } 4773 break; 4774 } 4775 case Stmt::ConditionalOperatorClass: { 4776 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 4777 if (const Expr *lhs = cond->getLHS()) 4778 CheckArrayAccess(lhs); 4779 if (const Expr *rhs = cond->getRHS()) 4780 CheckArrayAccess(rhs); 4781 return; 4782 } 4783 default: 4784 return; 4785 } 4786 } 4787} 4788 4789//===--- CHECK: Objective-C retain cycles ----------------------------------// 4790 4791namespace { 4792 struct RetainCycleOwner { 4793 RetainCycleOwner() : Variable(0), Indirect(false) {} 4794 VarDecl *Variable; 4795 SourceRange Range; 4796 SourceLocation Loc; 4797 bool Indirect; 4798 4799 void setLocsFrom(Expr *e) { 4800 Loc = e->getExprLoc(); 4801 Range = e->getSourceRange(); 4802 } 4803 }; 4804} 4805 4806/// Consider whether capturing the given variable can possibly lead to 4807/// a retain cycle. 4808static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 4809 // In ARC, it's captured strongly iff the variable has __strong 4810 // lifetime. In MRR, it's captured strongly if the variable is 4811 // __block and has an appropriate type. 4812 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4813 return false; 4814 4815 owner.Variable = var; 4816 owner.setLocsFrom(ref); 4817 return true; 4818} 4819 4820static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) { 4821 while (true) { 4822 e = e->IgnoreParens(); 4823 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 4824 switch (cast->getCastKind()) { 4825 case CK_BitCast: 4826 case CK_LValueBitCast: 4827 case CK_LValueToRValue: 4828 case CK_ARCReclaimReturnedObject: 4829 e = cast->getSubExpr(); 4830 continue; 4831 4832 default: 4833 return false; 4834 } 4835 } 4836 4837 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 4838 ObjCIvarDecl *ivar = ref->getDecl(); 4839 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4840 return false; 4841 4842 // Try to find a retain cycle in the base. 4843 if (!findRetainCycleOwner(S, ref->getBase(), owner)) 4844 return false; 4845 4846 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 4847 owner.Indirect = true; 4848 return true; 4849 } 4850 4851 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 4852 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 4853 if (!var) return false; 4854 return considerVariable(var, ref, owner); 4855 } 4856 4857 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 4858 if (member->isArrow()) return false; 4859 4860 // Don't count this as an indirect ownership. 4861 e = member->getBase(); 4862 continue; 4863 } 4864 4865 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 4866 // Only pay attention to pseudo-objects on property references. 4867 ObjCPropertyRefExpr *pre 4868 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 4869 ->IgnoreParens()); 4870 if (!pre) return false; 4871 if (pre->isImplicitProperty()) return false; 4872 ObjCPropertyDecl *property = pre->getExplicitProperty(); 4873 if (!property->isRetaining() && 4874 !(property->getPropertyIvarDecl() && 4875 property->getPropertyIvarDecl()->getType() 4876 .getObjCLifetime() == Qualifiers::OCL_Strong)) 4877 return false; 4878 4879 owner.Indirect = true; 4880 if (pre->isSuperReceiver()) { 4881 owner.Variable = S.getCurMethodDecl()->getSelfDecl(); 4882 if (!owner.Variable) 4883 return false; 4884 owner.Loc = pre->getLocation(); 4885 owner.Range = pre->getSourceRange(); 4886 return true; 4887 } 4888 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 4889 ->getSourceExpr()); 4890 continue; 4891 } 4892 4893 // Array ivars? 4894 4895 return false; 4896 } 4897} 4898 4899namespace { 4900 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 4901 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 4902 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 4903 Variable(variable), Capturer(0) {} 4904 4905 VarDecl *Variable; 4906 Expr *Capturer; 4907 4908 void VisitDeclRefExpr(DeclRefExpr *ref) { 4909 if (ref->getDecl() == Variable && !Capturer) 4910 Capturer = ref; 4911 } 4912 4913 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 4914 if (Capturer) return; 4915 Visit(ref->getBase()); 4916 if (Capturer && ref->isFreeIvar()) 4917 Capturer = ref; 4918 } 4919 4920 void VisitBlockExpr(BlockExpr *block) { 4921 // Look inside nested blocks 4922 if (block->getBlockDecl()->capturesVariable(Variable)) 4923 Visit(block->getBlockDecl()->getBody()); 4924 } 4925 }; 4926} 4927 4928/// Check whether the given argument is a block which captures a 4929/// variable. 4930static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 4931 assert(owner.Variable && owner.Loc.isValid()); 4932 4933 e = e->IgnoreParenCasts(); 4934 BlockExpr *block = dyn_cast<BlockExpr>(e); 4935 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 4936 return 0; 4937 4938 FindCaptureVisitor visitor(S.Context, owner.Variable); 4939 visitor.Visit(block->getBlockDecl()->getBody()); 4940 return visitor.Capturer; 4941} 4942 4943static void diagnoseRetainCycle(Sema &S, Expr *capturer, 4944 RetainCycleOwner &owner) { 4945 assert(capturer); 4946 assert(owner.Variable && owner.Loc.isValid()); 4947 4948 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 4949 << owner.Variable << capturer->getSourceRange(); 4950 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 4951 << owner.Indirect << owner.Range; 4952} 4953 4954/// Check for a keyword selector that starts with the word 'add' or 4955/// 'set'. 4956static bool isSetterLikeSelector(Selector sel) { 4957 if (sel.isUnarySelector()) return false; 4958 4959 StringRef str = sel.getNameForSlot(0); 4960 while (!str.empty() && str.front() == '_') str = str.substr(1); 4961 if (str.startswith("set")) 4962 str = str.substr(3); 4963 else if (str.startswith("add")) { 4964 // Specially whitelist 'addOperationWithBlock:'. 4965 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 4966 return false; 4967 str = str.substr(3); 4968 } 4969 else 4970 return false; 4971 4972 if (str.empty()) return true; 4973 return !islower(str.front()); 4974} 4975 4976/// Check a message send to see if it's likely to cause a retain cycle. 4977void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 4978 // Only check instance methods whose selector looks like a setter. 4979 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 4980 return; 4981 4982 // Try to find a variable that the receiver is strongly owned by. 4983 RetainCycleOwner owner; 4984 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 4985 if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner)) 4986 return; 4987 } else { 4988 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 4989 owner.Variable = getCurMethodDecl()->getSelfDecl(); 4990 owner.Loc = msg->getSuperLoc(); 4991 owner.Range = msg->getSuperLoc(); 4992 } 4993 4994 // Check whether the receiver is captured by any of the arguments. 4995 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 4996 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 4997 return diagnoseRetainCycle(*this, capturer, owner); 4998} 4999 5000/// Check a property assign to see if it's likely to cause a retain cycle. 5001void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 5002 RetainCycleOwner owner; 5003 if (!findRetainCycleOwner(*this, receiver, owner)) 5004 return; 5005 5006 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 5007 diagnoseRetainCycle(*this, capturer, owner); 5008} 5009 5010bool Sema::checkUnsafeAssigns(SourceLocation Loc, 5011 QualType LHS, Expr *RHS) { 5012 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 5013 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 5014 return false; 5015 // strip off any implicit cast added to get to the one arc-specific 5016 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 5017 if (cast->getCastKind() == CK_ARCConsumeObject) { 5018 Diag(Loc, diag::warn_arc_retained_assign) 5019 << (LT == Qualifiers::OCL_ExplicitNone) 5020 << RHS->getSourceRange(); 5021 return true; 5022 } 5023 RHS = cast->getSubExpr(); 5024 } 5025 return false; 5026} 5027 5028void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 5029 Expr *LHS, Expr *RHS) { 5030 QualType LHSType; 5031 // PropertyRef on LHS type need be directly obtained from 5032 // its declaration as it has a PsuedoType. 5033 ObjCPropertyRefExpr *PRE 5034 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens()); 5035 if (PRE && !PRE->isImplicitProperty()) { 5036 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 5037 if (PD) 5038 LHSType = PD->getType(); 5039 } 5040 5041 if (LHSType.isNull()) 5042 LHSType = LHS->getType(); 5043 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 5044 return; 5045 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 5046 // FIXME. Check for other life times. 5047 if (LT != Qualifiers::OCL_None) 5048 return; 5049 5050 if (PRE) { 5051 if (PRE->isImplicitProperty()) 5052 return; 5053 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 5054 if (!PD) 5055 return; 5056 5057 unsigned Attributes = PD->getPropertyAttributes(); 5058 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) { 5059 // when 'assign' attribute was not explicitly specified 5060 // by user, ignore it and rely on property type itself 5061 // for lifetime info. 5062 unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten(); 5063 if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) && 5064 LHSType->isObjCRetainableType()) 5065 return; 5066 5067 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 5068 if (cast->getCastKind() == CK_ARCConsumeObject) { 5069 Diag(Loc, diag::warn_arc_retained_property_assign) 5070 << RHS->getSourceRange(); 5071 return; 5072 } 5073 RHS = cast->getSubExpr(); 5074 } 5075 } 5076 } 5077} 5078 5079//===--- CHECK: Empty statement body (-Wempty-body) ---------------------===// 5080 5081namespace { 5082bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr, 5083 SourceLocation StmtLoc, 5084 const NullStmt *Body) { 5085 // Do not warn if the body is a macro that expands to nothing, e.g: 5086 // 5087 // #define CALL(x) 5088 // if (condition) 5089 // CALL(0); 5090 // 5091 if (Body->hasLeadingEmptyMacro()) 5092 return false; 5093 5094 // Get line numbers of statement and body. 5095 bool StmtLineInvalid; 5096 unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc, 5097 &StmtLineInvalid); 5098 if (StmtLineInvalid) 5099 return false; 5100 5101 bool BodyLineInvalid; 5102 unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(), 5103 &BodyLineInvalid); 5104 if (BodyLineInvalid) 5105 return false; 5106 5107 // Warn if null statement and body are on the same line. 5108 if (StmtLine != BodyLine) 5109 return false; 5110 5111 return true; 5112} 5113} // Unnamed namespace 5114 5115void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc, 5116 const Stmt *Body, 5117 unsigned DiagID) { 5118 // Since this is a syntactic check, don't emit diagnostic for template 5119 // instantiations, this just adds noise. 5120 if (CurrentInstantiationScope) 5121 return; 5122 5123 // The body should be a null statement. 5124 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 5125 if (!NBody) 5126 return; 5127 5128 // Do the usual checks. 5129 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 5130 return; 5131 5132 Diag(NBody->getSemiLoc(), DiagID); 5133 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 5134} 5135 5136void Sema::DiagnoseEmptyLoopBody(const Stmt *S, 5137 const Stmt *PossibleBody) { 5138 assert(!CurrentInstantiationScope); // Ensured by caller 5139 5140 SourceLocation StmtLoc; 5141 const Stmt *Body; 5142 unsigned DiagID; 5143 if (const ForStmt *FS = dyn_cast<ForStmt>(S)) { 5144 StmtLoc = FS->getRParenLoc(); 5145 Body = FS->getBody(); 5146 DiagID = diag::warn_empty_for_body; 5147 } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) { 5148 StmtLoc = WS->getCond()->getSourceRange().getEnd(); 5149 Body = WS->getBody(); 5150 DiagID = diag::warn_empty_while_body; 5151 } else 5152 return; // Neither `for' nor `while'. 5153 5154 // The body should be a null statement. 5155 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 5156 if (!NBody) 5157 return; 5158 5159 // Skip expensive checks if diagnostic is disabled. 5160 if (Diags.getDiagnosticLevel(DiagID, NBody->getSemiLoc()) == 5161 DiagnosticsEngine::Ignored) 5162 return; 5163 5164 // Do the usual checks. 5165 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 5166 return; 5167 5168 // `for(...);' and `while(...);' are popular idioms, so in order to keep 5169 // noise level low, emit diagnostics only if for/while is followed by a 5170 // CompoundStmt, e.g.: 5171 // for (int i = 0; i < n; i++); 5172 // { 5173 // a(i); 5174 // } 5175 // or if for/while is followed by a statement with more indentation 5176 // than for/while itself: 5177 // for (int i = 0; i < n; i++); 5178 // a(i); 5179 bool ProbableTypo = isa<CompoundStmt>(PossibleBody); 5180 if (!ProbableTypo) { 5181 bool BodyColInvalid; 5182 unsigned BodyCol = SourceMgr.getPresumedColumnNumber( 5183 PossibleBody->getLocStart(), 5184 &BodyColInvalid); 5185 if (BodyColInvalid) 5186 return; 5187 5188 bool StmtColInvalid; 5189 unsigned StmtCol = SourceMgr.getPresumedColumnNumber( 5190 S->getLocStart(), 5191 &StmtColInvalid); 5192 if (StmtColInvalid) 5193 return; 5194 5195 if (BodyCol > StmtCol) 5196 ProbableTypo = true; 5197 } 5198 5199 if (ProbableTypo) { 5200 Diag(NBody->getSemiLoc(), DiagID); 5201 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 5202 } 5203} 5204