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/SemaInternal.h" 16 #include "clang/AST/ASTContext.h" 17 #include "clang/AST/CharUnits.h" 18 #include "clang/AST/DeclCXX.h" 19 #include "clang/AST/DeclObjC.h" 20 #include "clang/AST/EvaluatedExprVisitor.h" 21 #include "clang/AST/Expr.h" 22 #include "clang/AST/ExprCXX.h" 23 #include "clang/AST/ExprObjC.h" 24 #include "clang/AST/StmtCXX.h" 25 #include "clang/AST/StmtObjC.h" 26 #include "clang/Analysis/Analyses/FormatString.h" 27 #include "clang/Basic/CharInfo.h" 28 #include "clang/Basic/TargetBuiltins.h" 29 #include "clang/Basic/TargetInfo.h" 30 #include "clang/Lex/Lexer.h" // TODO: Extract static functions to fix layering. 31 #include "clang/Sema/Initialization.h" 32 #include "clang/Sema/Lookup.h" 33 #include "clang/Sema/ScopeInfo.h" 34 #include "clang/Sema/Sema.h" 35 #include "llvm/ADT/STLExtras.h" 36 #include "llvm/ADT/SmallBitVector.h" 37 #include "llvm/ADT/SmallString.h" 38 #include "llvm/Support/ConvertUTF.h" 39 #include "llvm/Support/raw_ostream.h" 40 #include <limits> 41 using namespace clang; 42 using namespace sema; 43 44 SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 45 unsigned ByteNo) const { 46 return SL->getLocationOfByte(ByteNo, getSourceManager(), LangOpts, 47 Context.getTargetInfo()); 48 } 49 50 /// Checks that a call expression's argument count is the desired number. 51 /// This is useful when doing custom type-checking. Returns true on error. 52 static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) { 53 unsigned argCount = call->getNumArgs(); 54 if (argCount == desiredArgCount) return false; 55 56 if (argCount < desiredArgCount) 57 return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args) 58 << 0 /*function call*/ << desiredArgCount << argCount 59 << call->getSourceRange(); 60 61 // Highlight all the excess arguments. 62 SourceRange range(call->getArg(desiredArgCount)->getLocStart(), 63 call->getArg(argCount - 1)->getLocEnd()); 64 65 return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args) 66 << 0 /*function call*/ << desiredArgCount << argCount 67 << call->getArg(1)->getSourceRange(); 68 } 69 70 /// Check that the first argument to __builtin_annotation is an integer 71 /// and the second argument is a non-wide string literal. 72 static bool SemaBuiltinAnnotation(Sema &S, CallExpr *TheCall) { 73 if (checkArgCount(S, TheCall, 2)) 74 return true; 75 76 // First argument should be an integer. 77 Expr *ValArg = TheCall->getArg(0); 78 QualType Ty = ValArg->getType(); 79 if (!Ty->isIntegerType()) { 80 S.Diag(ValArg->getLocStart(), diag::err_builtin_annotation_first_arg) 81 << ValArg->getSourceRange(); 82 return true; 83 } 84 85 // Second argument should be a constant string. 86 Expr *StrArg = TheCall->getArg(1)->IgnoreParenCasts(); 87 StringLiteral *Literal = dyn_cast<StringLiteral>(StrArg); 88 if (!Literal || !Literal->isAscii()) { 89 S.Diag(StrArg->getLocStart(), diag::err_builtin_annotation_second_arg) 90 << StrArg->getSourceRange(); 91 return true; 92 } 93 94 TheCall->setType(Ty); 95 return false; 96 } 97 98 /// Check that the argument to __builtin_addressof is a glvalue, and set the 99 /// result type to the corresponding pointer type. 100 static bool SemaBuiltinAddressof(Sema &S, CallExpr *TheCall) { 101 if (checkArgCount(S, TheCall, 1)) 102 return true; 103 104 ExprResult Arg(TheCall->getArg(0)); 105 QualType ResultType = S.CheckAddressOfOperand(Arg, TheCall->getLocStart()); 106 if (ResultType.isNull()) 107 return true; 108 109 TheCall->setArg(0, Arg.get()); 110 TheCall->setType(ResultType); 111 return false; 112 } 113 114 ExprResult 115 Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 116 ExprResult TheCallResult(TheCall); 117 118 // Find out if any arguments are required to be integer constant expressions. 119 unsigned ICEArguments = 0; 120 ASTContext::GetBuiltinTypeError Error; 121 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments); 122 if (Error != ASTContext::GE_None) 123 ICEArguments = 0; // Don't diagnose previously diagnosed errors. 124 125 // If any arguments are required to be ICE's, check and diagnose. 126 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) { 127 // Skip arguments not required to be ICE's. 128 if ((ICEArguments & (1 << ArgNo)) == 0) continue; 129 130 llvm::APSInt Result; 131 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 132 return true; 133 ICEArguments &= ~(1 << ArgNo); 134 } 135 136 switch (BuiltinID) { 137 case Builtin::BI__builtin___CFStringMakeConstantString: 138 assert(TheCall->getNumArgs() == 1 && 139 "Wrong # arguments to builtin CFStringMakeConstantString"); 140 if (CheckObjCString(TheCall->getArg(0))) 141 return ExprError(); 142 break; 143 case Builtin::BI__builtin_stdarg_start: 144 case Builtin::BI__builtin_va_start: 145 case Builtin::BI__va_start: 146 if (SemaBuiltinVAStart(TheCall)) 147 return ExprError(); 148 break; 149 case Builtin::BI__builtin_isgreater: 150 case Builtin::BI__builtin_isgreaterequal: 151 case Builtin::BI__builtin_isless: 152 case Builtin::BI__builtin_islessequal: 153 case Builtin::BI__builtin_islessgreater: 154 case Builtin::BI__builtin_isunordered: 155 if (SemaBuiltinUnorderedCompare(TheCall)) 156 return ExprError(); 157 break; 158 case Builtin::BI__builtin_fpclassify: 159 if (SemaBuiltinFPClassification(TheCall, 6)) 160 return ExprError(); 161 break; 162 case Builtin::BI__builtin_isfinite: 163 case Builtin::BI__builtin_isinf: 164 case Builtin::BI__builtin_isinf_sign: 165 case Builtin::BI__builtin_isnan: 166 case Builtin::BI__builtin_isnormal: 167 if (SemaBuiltinFPClassification(TheCall, 1)) 168 return ExprError(); 169 break; 170 case Builtin::BI__builtin_shufflevector: 171 return SemaBuiltinShuffleVector(TheCall); 172 // TheCall will be freed by the smart pointer here, but that's fine, since 173 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 174 case Builtin::BI__builtin_prefetch: 175 if (SemaBuiltinPrefetch(TheCall)) 176 return ExprError(); 177 break; 178 case Builtin::BI__builtin_object_size: 179 if (SemaBuiltinConstantArgRange(TheCall, 1, 0, 3)) 180 return ExprError(); 181 break; 182 case Builtin::BI__builtin_longjmp: 183 if (SemaBuiltinLongjmp(TheCall)) 184 return ExprError(); 185 break; 186 187 case Builtin::BI__builtin_classify_type: 188 if (checkArgCount(*this, TheCall, 1)) return true; 189 TheCall->setType(Context.IntTy); 190 break; 191 case Builtin::BI__builtin_constant_p: 192 if (checkArgCount(*this, TheCall, 1)) return true; 193 TheCall->setType(Context.IntTy); 194 break; 195 case Builtin::BI__sync_fetch_and_add: 196 case Builtin::BI__sync_fetch_and_add_1: 197 case Builtin::BI__sync_fetch_and_add_2: 198 case Builtin::BI__sync_fetch_and_add_4: 199 case Builtin::BI__sync_fetch_and_add_8: 200 case Builtin::BI__sync_fetch_and_add_16: 201 case Builtin::BI__sync_fetch_and_sub: 202 case Builtin::BI__sync_fetch_and_sub_1: 203 case Builtin::BI__sync_fetch_and_sub_2: 204 case Builtin::BI__sync_fetch_and_sub_4: 205 case Builtin::BI__sync_fetch_and_sub_8: 206 case Builtin::BI__sync_fetch_and_sub_16: 207 case Builtin::BI__sync_fetch_and_or: 208 case Builtin::BI__sync_fetch_and_or_1: 209 case Builtin::BI__sync_fetch_and_or_2: 210 case Builtin::BI__sync_fetch_and_or_4: 211 case Builtin::BI__sync_fetch_and_or_8: 212 case Builtin::BI__sync_fetch_and_or_16: 213 case Builtin::BI__sync_fetch_and_and: 214 case Builtin::BI__sync_fetch_and_and_1: 215 case Builtin::BI__sync_fetch_and_and_2: 216 case Builtin::BI__sync_fetch_and_and_4: 217 case Builtin::BI__sync_fetch_and_and_8: 218 case Builtin::BI__sync_fetch_and_and_16: 219 case Builtin::BI__sync_fetch_and_xor: 220 case Builtin::BI__sync_fetch_and_xor_1: 221 case Builtin::BI__sync_fetch_and_xor_2: 222 case Builtin::BI__sync_fetch_and_xor_4: 223 case Builtin::BI__sync_fetch_and_xor_8: 224 case Builtin::BI__sync_fetch_and_xor_16: 225 case Builtin::BI__sync_add_and_fetch: 226 case Builtin::BI__sync_add_and_fetch_1: 227 case Builtin::BI__sync_add_and_fetch_2: 228 case Builtin::BI__sync_add_and_fetch_4: 229 case Builtin::BI__sync_add_and_fetch_8: 230 case Builtin::BI__sync_add_and_fetch_16: 231 case Builtin::BI__sync_sub_and_fetch: 232 case Builtin::BI__sync_sub_and_fetch_1: 233 case Builtin::BI__sync_sub_and_fetch_2: 234 case Builtin::BI__sync_sub_and_fetch_4: 235 case Builtin::BI__sync_sub_and_fetch_8: 236 case Builtin::BI__sync_sub_and_fetch_16: 237 case Builtin::BI__sync_and_and_fetch: 238 case Builtin::BI__sync_and_and_fetch_1: 239 case Builtin::BI__sync_and_and_fetch_2: 240 case Builtin::BI__sync_and_and_fetch_4: 241 case Builtin::BI__sync_and_and_fetch_8: 242 case Builtin::BI__sync_and_and_fetch_16: 243 case Builtin::BI__sync_or_and_fetch: 244 case Builtin::BI__sync_or_and_fetch_1: 245 case Builtin::BI__sync_or_and_fetch_2: 246 case Builtin::BI__sync_or_and_fetch_4: 247 case Builtin::BI__sync_or_and_fetch_8: 248 case Builtin::BI__sync_or_and_fetch_16: 249 case Builtin::BI__sync_xor_and_fetch: 250 case Builtin::BI__sync_xor_and_fetch_1: 251 case Builtin::BI__sync_xor_and_fetch_2: 252 case Builtin::BI__sync_xor_and_fetch_4: 253 case Builtin::BI__sync_xor_and_fetch_8: 254 case Builtin::BI__sync_xor_and_fetch_16: 255 case Builtin::BI__sync_val_compare_and_swap: 256 case Builtin::BI__sync_val_compare_and_swap_1: 257 case Builtin::BI__sync_val_compare_and_swap_2: 258 case Builtin::BI__sync_val_compare_and_swap_4: 259 case Builtin::BI__sync_val_compare_and_swap_8: 260 case Builtin::BI__sync_val_compare_and_swap_16: 261 case Builtin::BI__sync_bool_compare_and_swap: 262 case Builtin::BI__sync_bool_compare_and_swap_1: 263 case Builtin::BI__sync_bool_compare_and_swap_2: 264 case Builtin::BI__sync_bool_compare_and_swap_4: 265 case Builtin::BI__sync_bool_compare_and_swap_8: 266 case Builtin::BI__sync_bool_compare_and_swap_16: 267 case Builtin::BI__sync_lock_test_and_set: 268 case Builtin::BI__sync_lock_test_and_set_1: 269 case Builtin::BI__sync_lock_test_and_set_2: 270 case Builtin::BI__sync_lock_test_and_set_4: 271 case Builtin::BI__sync_lock_test_and_set_8: 272 case Builtin::BI__sync_lock_test_and_set_16: 273 case Builtin::BI__sync_lock_release: 274 case Builtin::BI__sync_lock_release_1: 275 case Builtin::BI__sync_lock_release_2: 276 case Builtin::BI__sync_lock_release_4: 277 case Builtin::BI__sync_lock_release_8: 278 case Builtin::BI__sync_lock_release_16: 279 case Builtin::BI__sync_swap: 280 case Builtin::BI__sync_swap_1: 281 case Builtin::BI__sync_swap_2: 282 case Builtin::BI__sync_swap_4: 283 case Builtin::BI__sync_swap_8: 284 case Builtin::BI__sync_swap_16: 285 return SemaBuiltinAtomicOverloaded(TheCallResult); 286 #define BUILTIN(ID, TYPE, ATTRS) 287 #define ATOMIC_BUILTIN(ID, TYPE, ATTRS) \ 288 case Builtin::BI##ID: \ 289 return SemaAtomicOpsOverloaded(TheCallResult, AtomicExpr::AO##ID); 290 #include "clang/Basic/Builtins.def" 291 case Builtin::BI__builtin_annotation: 292 if (SemaBuiltinAnnotation(*this, TheCall)) 293 return ExprError(); 294 break; 295 case Builtin::BI__builtin_addressof: 296 if (SemaBuiltinAddressof(*this, TheCall)) 297 return ExprError(); 298 break; 299 case Builtin::BI__builtin_operator_new: 300 case Builtin::BI__builtin_operator_delete: 301 if (!getLangOpts().CPlusPlus) { 302 Diag(TheCall->getExprLoc(), diag::err_builtin_requires_language) 303 << (BuiltinID == Builtin::BI__builtin_operator_new 304 ? "__builtin_operator_new" 305 : "__builtin_operator_delete") 306 << "C++"; 307 return ExprError(); 308 } 309 // CodeGen assumes it can find the global new and delete to call, 310 // so ensure that they are declared. 311 DeclareGlobalNewDelete(); 312 break; 313 } 314 315 // Since the target specific builtins for each arch overlap, only check those 316 // of the arch we are compiling for. 317 if (BuiltinID >= Builtin::FirstTSBuiltin) { 318 switch (Context.getTargetInfo().getTriple().getArch()) { 319 case llvm::Triple::arm: 320 case llvm::Triple::armeb: 321 case llvm::Triple::thumb: 322 case llvm::Triple::thumbeb: 323 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 324 return ExprError(); 325 break; 326 case llvm::Triple::aarch64: 327 case llvm::Triple::aarch64_be: 328 case llvm::Triple::arm64: 329 case llvm::Triple::arm64_be: 330 if (CheckAArch64BuiltinFunctionCall(BuiltinID, TheCall)) 331 return ExprError(); 332 break; 333 case llvm::Triple::mips: 334 case llvm::Triple::mipsel: 335 case llvm::Triple::mips64: 336 case llvm::Triple::mips64el: 337 if (CheckMipsBuiltinFunctionCall(BuiltinID, TheCall)) 338 return ExprError(); 339 break; 340 case llvm::Triple::x86: 341 case llvm::Triple::x86_64: 342 if (CheckX86BuiltinFunctionCall(BuiltinID, TheCall)) 343 return ExprError(); 344 break; 345 default: 346 break; 347 } 348 } 349 350 return TheCallResult; 351 } 352 353 // Get the valid immediate range for the specified NEON type code. 354 static unsigned RFT(unsigned t, bool shift = false, bool ForceQuad = false) { 355 NeonTypeFlags Type(t); 356 int IsQuad = ForceQuad ? true : Type.isQuad(); 357 switch (Type.getEltType()) { 358 case NeonTypeFlags::Int8: 359 case NeonTypeFlags::Poly8: 360 return shift ? 7 : (8 << IsQuad) - 1; 361 case NeonTypeFlags::Int16: 362 case NeonTypeFlags::Poly16: 363 return shift ? 15 : (4 << IsQuad) - 1; 364 case NeonTypeFlags::Int32: 365 return shift ? 31 : (2 << IsQuad) - 1; 366 case NeonTypeFlags::Int64: 367 case NeonTypeFlags::Poly64: 368 return shift ? 63 : (1 << IsQuad) - 1; 369 case NeonTypeFlags::Poly128: 370 return shift ? 127 : (1 << IsQuad) - 1; 371 case NeonTypeFlags::Float16: 372 assert(!shift && "cannot shift float types!"); 373 return (4 << IsQuad) - 1; 374 case NeonTypeFlags::Float32: 375 assert(!shift && "cannot shift float types!"); 376 return (2 << IsQuad) - 1; 377 case NeonTypeFlags::Float64: 378 assert(!shift && "cannot shift float types!"); 379 return (1 << IsQuad) - 1; 380 } 381 llvm_unreachable("Invalid NeonTypeFlag!"); 382 } 383 384 /// getNeonEltType - Return the QualType corresponding to the elements of 385 /// the vector type specified by the NeonTypeFlags. This is used to check 386 /// the pointer arguments for Neon load/store intrinsics. 387 static QualType getNeonEltType(NeonTypeFlags Flags, ASTContext &Context, 388 bool IsPolyUnsigned, bool IsInt64Long) { 389 switch (Flags.getEltType()) { 390 case NeonTypeFlags::Int8: 391 return Flags.isUnsigned() ? Context.UnsignedCharTy : Context.SignedCharTy; 392 case NeonTypeFlags::Int16: 393 return Flags.isUnsigned() ? Context.UnsignedShortTy : Context.ShortTy; 394 case NeonTypeFlags::Int32: 395 return Flags.isUnsigned() ? Context.UnsignedIntTy : Context.IntTy; 396 case NeonTypeFlags::Int64: 397 if (IsInt64Long) 398 return Flags.isUnsigned() ? Context.UnsignedLongTy : Context.LongTy; 399 else 400 return Flags.isUnsigned() ? Context.UnsignedLongLongTy 401 : Context.LongLongTy; 402 case NeonTypeFlags::Poly8: 403 return IsPolyUnsigned ? Context.UnsignedCharTy : Context.SignedCharTy; 404 case NeonTypeFlags::Poly16: 405 return IsPolyUnsigned ? Context.UnsignedShortTy : Context.ShortTy; 406 case NeonTypeFlags::Poly64: 407 return Context.UnsignedLongTy; 408 case NeonTypeFlags::Poly128: 409 break; 410 case NeonTypeFlags::Float16: 411 return Context.HalfTy; 412 case NeonTypeFlags::Float32: 413 return Context.FloatTy; 414 case NeonTypeFlags::Float64: 415 return Context.DoubleTy; 416 } 417 llvm_unreachable("Invalid NeonTypeFlag!"); 418 } 419 420 bool Sema::CheckNeonBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 421 llvm::APSInt Result; 422 uint64_t mask = 0; 423 unsigned TV = 0; 424 int PtrArgNum = -1; 425 bool HasConstPtr = false; 426 switch (BuiltinID) { 427 #define GET_NEON_OVERLOAD_CHECK 428 #include "clang/Basic/arm_neon.inc" 429 #undef GET_NEON_OVERLOAD_CHECK 430 } 431 432 // For NEON intrinsics which are overloaded on vector element type, validate 433 // the immediate which specifies which variant to emit. 434 unsigned ImmArg = TheCall->getNumArgs()-1; 435 if (mask) { 436 if (SemaBuiltinConstantArg(TheCall, ImmArg, Result)) 437 return true; 438 439 TV = Result.getLimitedValue(64); 440 if ((TV > 63) || (mask & (1ULL << TV)) == 0) 441 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 442 << TheCall->getArg(ImmArg)->getSourceRange(); 443 } 444 445 if (PtrArgNum >= 0) { 446 // Check that pointer arguments have the specified type. 447 Expr *Arg = TheCall->getArg(PtrArgNum); 448 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) 449 Arg = ICE->getSubExpr(); 450 ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg); 451 QualType RHSTy = RHS.get()->getType(); 452 453 llvm::Triple::ArchType Arch = Context.getTargetInfo().getTriple().getArch(); 454 bool IsPolyUnsigned = 455 Arch == llvm::Triple::aarch64 || Arch == llvm::Triple::arm64; 456 bool IsInt64Long = 457 Context.getTargetInfo().getInt64Type() == TargetInfo::SignedLong; 458 QualType EltTy = 459 getNeonEltType(NeonTypeFlags(TV), Context, IsPolyUnsigned, IsInt64Long); 460 if (HasConstPtr) 461 EltTy = EltTy.withConst(); 462 QualType LHSTy = Context.getPointerType(EltTy); 463 AssignConvertType ConvTy; 464 ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS); 465 if (RHS.isInvalid()) 466 return true; 467 if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy, 468 RHS.get(), AA_Assigning)) 469 return true; 470 } 471 472 // For NEON intrinsics which take an immediate value as part of the 473 // instruction, range check them here. 474 unsigned i = 0, l = 0, u = 0; 475 switch (BuiltinID) { 476 default: 477 return false; 478 #define GET_NEON_IMMEDIATE_CHECK 479 #include "clang/Basic/arm_neon.inc" 480 #undef GET_NEON_IMMEDIATE_CHECK 481 } 482 483 return SemaBuiltinConstantArgRange(TheCall, i, l, u + l); 484 } 485 486 bool Sema::CheckARMBuiltinExclusiveCall(unsigned BuiltinID, CallExpr *TheCall, 487 unsigned MaxWidth) { 488 assert((BuiltinID == ARM::BI__builtin_arm_ldrex || 489 BuiltinID == ARM::BI__builtin_arm_strex || 490 BuiltinID == AArch64::BI__builtin_arm_ldrex || 491 BuiltinID == AArch64::BI__builtin_arm_strex) && 492 "unexpected ARM builtin"); 493 bool IsLdrex = BuiltinID == ARM::BI__builtin_arm_ldrex || 494 BuiltinID == AArch64::BI__builtin_arm_ldrex; 495 496 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 497 498 // Ensure that we have the proper number of arguments. 499 if (checkArgCount(*this, TheCall, IsLdrex ? 1 : 2)) 500 return true; 501 502 // Inspect the pointer argument of the atomic builtin. This should always be 503 // a pointer type, whose element is an integral scalar or pointer type. 504 // Because it is a pointer type, we don't have to worry about any implicit 505 // casts here. 506 Expr *PointerArg = TheCall->getArg(IsLdrex ? 0 : 1); 507 ExprResult PointerArgRes = DefaultFunctionArrayLvalueConversion(PointerArg); 508 if (PointerArgRes.isInvalid()) 509 return true; 510 PointerArg = PointerArgRes.get(); 511 512 const PointerType *pointerType = PointerArg->getType()->getAs<PointerType>(); 513 if (!pointerType) { 514 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 515 << PointerArg->getType() << PointerArg->getSourceRange(); 516 return true; 517 } 518 519 // ldrex takes a "const volatile T*" and strex takes a "volatile T*". Our next 520 // task is to insert the appropriate casts into the AST. First work out just 521 // what the appropriate type is. 522 QualType ValType = pointerType->getPointeeType(); 523 QualType AddrType = ValType.getUnqualifiedType().withVolatile(); 524 if (IsLdrex) 525 AddrType.addConst(); 526 527 // Issue a warning if the cast is dodgy. 528 CastKind CastNeeded = CK_NoOp; 529 if (!AddrType.isAtLeastAsQualifiedAs(ValType)) { 530 CastNeeded = CK_BitCast; 531 Diag(DRE->getLocStart(), diag::ext_typecheck_convert_discards_qualifiers) 532 << PointerArg->getType() 533 << Context.getPointerType(AddrType) 534 << AA_Passing << PointerArg->getSourceRange(); 535 } 536 537 // Finally, do the cast and replace the argument with the corrected version. 538 AddrType = Context.getPointerType(AddrType); 539 PointerArgRes = ImpCastExprToType(PointerArg, AddrType, CastNeeded); 540 if (PointerArgRes.isInvalid()) 541 return true; 542 PointerArg = PointerArgRes.get(); 543 544 TheCall->setArg(IsLdrex ? 0 : 1, PointerArg); 545 546 // In general, we allow ints, floats and pointers to be loaded and stored. 547 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 548 !ValType->isBlockPointerType() && !ValType->isFloatingType()) { 549 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intfltptr) 550 << PointerArg->getType() << PointerArg->getSourceRange(); 551 return true; 552 } 553 554 // But ARM doesn't have instructions to deal with 128-bit versions. 555 if (Context.getTypeSize(ValType) > MaxWidth) { 556 assert(MaxWidth == 64 && "Diagnostic unexpectedly inaccurate"); 557 Diag(DRE->getLocStart(), diag::err_atomic_exclusive_builtin_pointer_size) 558 << PointerArg->getType() << PointerArg->getSourceRange(); 559 return true; 560 } 561 562 switch (ValType.getObjCLifetime()) { 563 case Qualifiers::OCL_None: 564 case Qualifiers::OCL_ExplicitNone: 565 // okay 566 break; 567 568 case Qualifiers::OCL_Weak: 569 case Qualifiers::OCL_Strong: 570 case Qualifiers::OCL_Autoreleasing: 571 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 572 << ValType << PointerArg->getSourceRange(); 573 return true; 574 } 575 576 577 if (IsLdrex) { 578 TheCall->setType(ValType); 579 return false; 580 } 581 582 // Initialize the argument to be stored. 583 ExprResult ValArg = TheCall->getArg(0); 584 InitializedEntity Entity = InitializedEntity::InitializeParameter( 585 Context, ValType, /*consume*/ false); 586 ValArg = PerformCopyInitialization(Entity, SourceLocation(), ValArg); 587 if (ValArg.isInvalid()) 588 return true; 589 TheCall->setArg(0, ValArg.get()); 590 591 // __builtin_arm_strex always returns an int. It's marked as such in the .def, 592 // but the custom checker bypasses all default analysis. 593 TheCall->setType(Context.IntTy); 594 return false; 595 } 596 597 bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 598 llvm::APSInt Result; 599 600 if (BuiltinID == ARM::BI__builtin_arm_ldrex || 601 BuiltinID == ARM::BI__builtin_arm_strex) { 602 return CheckARMBuiltinExclusiveCall(BuiltinID, TheCall, 64); 603 } 604 605 if (CheckNeonBuiltinFunctionCall(BuiltinID, TheCall)) 606 return true; 607 608 // For NEON intrinsics which take an immediate value as part of the 609 // instruction, range check them here. 610 unsigned i = 0, l = 0, u = 0; 611 switch (BuiltinID) { 612 default: return false; 613 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 614 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 615 case ARM::BI__builtin_arm_vcvtr_f: 616 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 617 case ARM::BI__builtin_arm_dmb: 618 case ARM::BI__builtin_arm_dsb: l = 0; u = 15; break; 619 } 620 621 // FIXME: VFP Intrinsics should error if VFP not present. 622 return SemaBuiltinConstantArgRange(TheCall, i, l, u + l); 623 } 624 625 bool Sema::CheckAArch64BuiltinFunctionCall(unsigned BuiltinID, 626 CallExpr *TheCall) { 627 llvm::APSInt Result; 628 629 if (BuiltinID == AArch64::BI__builtin_arm_ldrex || 630 BuiltinID == AArch64::BI__builtin_arm_strex) { 631 return CheckARMBuiltinExclusiveCall(BuiltinID, TheCall, 128); 632 } 633 634 if (CheckNeonBuiltinFunctionCall(BuiltinID, TheCall)) 635 return true; 636 637 return false; 638 } 639 640 bool Sema::CheckMipsBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 641 unsigned i = 0, l = 0, u = 0; 642 switch (BuiltinID) { 643 default: return false; 644 case Mips::BI__builtin_mips_wrdsp: i = 1; l = 0; u = 63; break; 645 case Mips::BI__builtin_mips_rddsp: i = 0; l = 0; u = 63; break; 646 case Mips::BI__builtin_mips_append: i = 2; l = 0; u = 31; break; 647 case Mips::BI__builtin_mips_balign: i = 2; l = 0; u = 3; break; 648 case Mips::BI__builtin_mips_precr_sra_ph_w: i = 2; l = 0; u = 31; break; 649 case Mips::BI__builtin_mips_precr_sra_r_ph_w: i = 2; l = 0; u = 31; break; 650 case Mips::BI__builtin_mips_prepend: i = 2; l = 0; u = 31; break; 651 } 652 653 return SemaBuiltinConstantArgRange(TheCall, i, l, u); 654 } 655 656 bool Sema::CheckX86BuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 657 switch (BuiltinID) { 658 case X86::BI_mm_prefetch: 659 // This is declared to take (const char*, int) 660 return SemaBuiltinConstantArgRange(TheCall, 1, 0, 3); 661 } 662 return false; 663 } 664 665 /// Given a FunctionDecl's FormatAttr, attempts to populate the FomatStringInfo 666 /// parameter with the FormatAttr's correct format_idx and firstDataArg. 667 /// Returns true when the format fits the function and the FormatStringInfo has 668 /// been populated. 669 bool Sema::getFormatStringInfo(const FormatAttr *Format, bool IsCXXMember, 670 FormatStringInfo *FSI) { 671 FSI->HasVAListArg = Format->getFirstArg() == 0; 672 FSI->FormatIdx = Format->getFormatIdx() - 1; 673 FSI->FirstDataArg = FSI->HasVAListArg ? 0 : Format->getFirstArg() - 1; 674 675 // The way the format attribute works in GCC, the implicit this argument 676 // of member functions is counted. However, it doesn't appear in our own 677 // lists, so decrement format_idx in that case. 678 if (IsCXXMember) { 679 if(FSI->FormatIdx == 0) 680 return false; 681 --FSI->FormatIdx; 682 if (FSI->FirstDataArg != 0) 683 --FSI->FirstDataArg; 684 } 685 return true; 686 } 687 688 /// Checks if a the given expression evaluates to null. 689 /// 690 /// \brief Returns true if the value evaluates to null. 691 static bool CheckNonNullExpr(Sema &S, 692 const Expr *Expr) { 693 // As a special case, transparent unions initialized with zero are 694 // considered null for the purposes of the nonnull attribute. 695 if (const RecordType *UT = Expr->getType()->getAsUnionType()) { 696 if (UT->getDecl()->hasAttr<TransparentUnionAttr>()) 697 if (const CompoundLiteralExpr *CLE = 698 dyn_cast<CompoundLiteralExpr>(Expr)) 699 if (const InitListExpr *ILE = 700 dyn_cast<InitListExpr>(CLE->getInitializer())) 701 Expr = ILE->getInit(0); 702 } 703 704 bool Result; 705 return (!Expr->isValueDependent() && 706 Expr->EvaluateAsBooleanCondition(Result, S.Context) && 707 !Result); 708 } 709 710 static void CheckNonNullArgument(Sema &S, 711 const Expr *ArgExpr, 712 SourceLocation CallSiteLoc) { 713 if (CheckNonNullExpr(S, ArgExpr)) 714 S.Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 715 } 716 717 static void CheckNonNullArguments(Sema &S, 718 const NamedDecl *FDecl, 719 const Expr * const *ExprArgs, 720 SourceLocation CallSiteLoc) { 721 // Check the attributes attached to the method/function itself. 722 for (const auto *NonNull : FDecl->specific_attrs<NonNullAttr>()) { 723 for (const auto &Val : NonNull->args()) 724 CheckNonNullArgument(S, ExprArgs[Val], CallSiteLoc); 725 } 726 727 // Check the attributes on the parameters. 728 ArrayRef<ParmVarDecl*> parms; 729 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(FDecl)) 730 parms = FD->parameters(); 731 else if (const ObjCMethodDecl *MD = dyn_cast<ObjCMethodDecl>(FDecl)) 732 parms = MD->parameters(); 733 734 unsigned argIndex = 0; 735 for (ArrayRef<ParmVarDecl*>::iterator I = parms.begin(), E = parms.end(); 736 I != E; ++I, ++argIndex) { 737 const ParmVarDecl *PVD = *I; 738 if (PVD->hasAttr<NonNullAttr>()) 739 CheckNonNullArgument(S, ExprArgs[argIndex], CallSiteLoc); 740 } 741 } 742 743 /// Handles the checks for format strings, non-POD arguments to vararg 744 /// functions, and NULL arguments passed to non-NULL parameters. 745 void Sema::checkCall(NamedDecl *FDecl, ArrayRef<const Expr *> Args, 746 unsigned NumParams, bool IsMemberFunction, 747 SourceLocation Loc, SourceRange Range, 748 VariadicCallType CallType) { 749 // FIXME: We should check as much as we can in the template definition. 750 if (CurContext->isDependentContext()) 751 return; 752 753 // Printf and scanf checking. 754 llvm::SmallBitVector CheckedVarArgs; 755 if (FDecl) { 756 for (const auto *I : FDecl->specific_attrs<FormatAttr>()) { 757 // Only create vector if there are format attributes. 758 CheckedVarArgs.resize(Args.size()); 759 760 CheckFormatArguments(I, Args, IsMemberFunction, CallType, Loc, Range, 761 CheckedVarArgs); 762 } 763 } 764 765 // Refuse POD arguments that weren't caught by the format string 766 // checks above. 767 if (CallType != VariadicDoesNotApply) { 768 for (unsigned ArgIdx = NumParams; ArgIdx < Args.size(); ++ArgIdx) { 769 // Args[ArgIdx] can be null in malformed code. 770 if (const Expr *Arg = Args[ArgIdx]) { 771 if (CheckedVarArgs.empty() || !CheckedVarArgs[ArgIdx]) 772 checkVariadicArgument(Arg, CallType); 773 } 774 } 775 } 776 777 if (FDecl) { 778 CheckNonNullArguments(*this, FDecl, Args.data(), Loc); 779 780 // Type safety checking. 781 for (const auto *I : FDecl->specific_attrs<ArgumentWithTypeTagAttr>()) 782 CheckArgumentWithTypeTag(I, Args.data()); 783 } 784 } 785 786 /// CheckConstructorCall - Check a constructor call for correctness and safety 787 /// properties not enforced by the C type system. 788 void Sema::CheckConstructorCall(FunctionDecl *FDecl, 789 ArrayRef<const Expr *> Args, 790 const FunctionProtoType *Proto, 791 SourceLocation Loc) { 792 VariadicCallType CallType = 793 Proto->isVariadic() ? VariadicConstructor : VariadicDoesNotApply; 794 checkCall(FDecl, Args, Proto->getNumParams(), 795 /*IsMemberFunction=*/true, Loc, SourceRange(), CallType); 796 } 797 798 /// CheckFunctionCall - Check a direct function call for various correctness 799 /// and safety properties not strictly enforced by the C type system. 800 bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall, 801 const FunctionProtoType *Proto) { 802 bool IsMemberOperatorCall = isa<CXXOperatorCallExpr>(TheCall) && 803 isa<CXXMethodDecl>(FDecl); 804 bool IsMemberFunction = isa<CXXMemberCallExpr>(TheCall) || 805 IsMemberOperatorCall; 806 VariadicCallType CallType = getVariadicCallType(FDecl, Proto, 807 TheCall->getCallee()); 808 unsigned NumParams = Proto ? Proto->getNumParams() : 0; 809 Expr** Args = TheCall->getArgs(); 810 unsigned NumArgs = TheCall->getNumArgs(); 811 if (IsMemberOperatorCall) { 812 // If this is a call to a member operator, hide the first argument 813 // from checkCall. 814 // FIXME: Our choice of AST representation here is less than ideal. 815 ++Args; 816 --NumArgs; 817 } 818 checkCall(FDecl, llvm::makeArrayRef<const Expr *>(Args, NumArgs), NumParams, 819 IsMemberFunction, TheCall->getRParenLoc(), 820 TheCall->getCallee()->getSourceRange(), CallType); 821 822 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 823 // None of the checks below are needed for functions that don't have 824 // simple names (e.g., C++ conversion functions). 825 if (!FnInfo) 826 return false; 827 828 CheckAbsoluteValueFunction(TheCall, FDecl, FnInfo); 829 830 unsigned CMId = FDecl->getMemoryFunctionKind(); 831 if (CMId == 0) 832 return false; 833 834 // Handle memory setting and copying functions. 835 if (CMId == Builtin::BIstrlcpy || CMId == Builtin::BIstrlcat) 836 CheckStrlcpycatArguments(TheCall, FnInfo); 837 else if (CMId == Builtin::BIstrncat) 838 CheckStrncatArguments(TheCall, FnInfo); 839 else 840 CheckMemaccessArguments(TheCall, CMId, FnInfo); 841 842 return false; 843 } 844 845 bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac, 846 ArrayRef<const Expr *> Args) { 847 VariadicCallType CallType = 848 Method->isVariadic() ? VariadicMethod : VariadicDoesNotApply; 849 850 checkCall(Method, Args, Method->param_size(), 851 /*IsMemberFunction=*/false, 852 lbrac, Method->getSourceRange(), CallType); 853 854 return false; 855 } 856 857 bool Sema::CheckPointerCall(NamedDecl *NDecl, CallExpr *TheCall, 858 const FunctionProtoType *Proto) { 859 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 860 if (!V) 861 return false; 862 863 QualType Ty = V->getType(); 864 if (!Ty->isBlockPointerType() && !Ty->isFunctionPointerType()) 865 return false; 866 867 VariadicCallType CallType; 868 if (!Proto || !Proto->isVariadic()) { 869 CallType = VariadicDoesNotApply; 870 } else if (Ty->isBlockPointerType()) { 871 CallType = VariadicBlock; 872 } else { // Ty->isFunctionPointerType() 873 CallType = VariadicFunction; 874 } 875 unsigned NumParams = Proto ? Proto->getNumParams() : 0; 876 877 checkCall(NDecl, llvm::makeArrayRef<const Expr *>(TheCall->getArgs(), 878 TheCall->getNumArgs()), 879 NumParams, /*IsMemberFunction=*/false, TheCall->getRParenLoc(), 880 TheCall->getCallee()->getSourceRange(), CallType); 881 882 return false; 883 } 884 885 /// Checks function calls when a FunctionDecl or a NamedDecl is not available, 886 /// such as function pointers returned from functions. 887 bool Sema::CheckOtherCall(CallExpr *TheCall, const FunctionProtoType *Proto) { 888 VariadicCallType CallType = getVariadicCallType(/*FDecl=*/nullptr, Proto, 889 TheCall->getCallee()); 890 unsigned NumParams = Proto ? Proto->getNumParams() : 0; 891 892 checkCall(/*FDecl=*/nullptr, 893 llvm::makeArrayRef<const Expr *>(TheCall->getArgs(), 894 TheCall->getNumArgs()), 895 NumParams, /*IsMemberFunction=*/false, TheCall->getRParenLoc(), 896 TheCall->getCallee()->getSourceRange(), CallType); 897 898 return false; 899 } 900 901 static bool isValidOrderingForOp(int64_t Ordering, AtomicExpr::AtomicOp Op) { 902 if (Ordering < AtomicExpr::AO_ABI_memory_order_relaxed || 903 Ordering > AtomicExpr::AO_ABI_memory_order_seq_cst) 904 return false; 905 906 switch (Op) { 907 case AtomicExpr::AO__c11_atomic_init: 908 llvm_unreachable("There is no ordering argument for an init"); 909 910 case AtomicExpr::AO__c11_atomic_load: 911 case AtomicExpr::AO__atomic_load_n: 912 case AtomicExpr::AO__atomic_load: 913 return Ordering != AtomicExpr::AO_ABI_memory_order_release && 914 Ordering != AtomicExpr::AO_ABI_memory_order_acq_rel; 915 916 case AtomicExpr::AO__c11_atomic_store: 917 case AtomicExpr::AO__atomic_store: 918 case AtomicExpr::AO__atomic_store_n: 919 return Ordering != AtomicExpr::AO_ABI_memory_order_consume && 920 Ordering != AtomicExpr::AO_ABI_memory_order_acquire && 921 Ordering != AtomicExpr::AO_ABI_memory_order_acq_rel; 922 923 default: 924 return true; 925 } 926 } 927 928 ExprResult Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, 929 AtomicExpr::AtomicOp Op) { 930 CallExpr *TheCall = cast<CallExpr>(TheCallResult.get()); 931 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 932 933 // All these operations take one of the following forms: 934 enum { 935 // C __c11_atomic_init(A *, C) 936 Init, 937 // C __c11_atomic_load(A *, int) 938 Load, 939 // void __atomic_load(A *, CP, int) 940 Copy, 941 // C __c11_atomic_add(A *, M, int) 942 Arithmetic, 943 // C __atomic_exchange_n(A *, CP, int) 944 Xchg, 945 // void __atomic_exchange(A *, C *, CP, int) 946 GNUXchg, 947 // bool __c11_atomic_compare_exchange_strong(A *, C *, CP, int, int) 948 C11CmpXchg, 949 // bool __atomic_compare_exchange(A *, C *, CP, bool, int, int) 950 GNUCmpXchg 951 } Form = Init; 952 const unsigned NumArgs[] = { 2, 2, 3, 3, 3, 4, 5, 6 }; 953 const unsigned NumVals[] = { 1, 0, 1, 1, 1, 2, 2, 3 }; 954 // where: 955 // C is an appropriate type, 956 // A is volatile _Atomic(C) for __c11 builtins and is C for GNU builtins, 957 // CP is C for __c11 builtins and GNU _n builtins and is C * otherwise, 958 // M is C if C is an integer, and ptrdiff_t if C is a pointer, and 959 // the int parameters are for orderings. 960 961 assert(AtomicExpr::AO__c11_atomic_init == 0 && 962 AtomicExpr::AO__c11_atomic_fetch_xor + 1 == AtomicExpr::AO__atomic_load 963 && "need to update code for modified C11 atomics"); 964 bool IsC11 = Op >= AtomicExpr::AO__c11_atomic_init && 965 Op <= AtomicExpr::AO__c11_atomic_fetch_xor; 966 bool IsN = Op == AtomicExpr::AO__atomic_load_n || 967 Op == AtomicExpr::AO__atomic_store_n || 968 Op == AtomicExpr::AO__atomic_exchange_n || 969 Op == AtomicExpr::AO__atomic_compare_exchange_n; 970 bool IsAddSub = false; 971 972 switch (Op) { 973 case AtomicExpr::AO__c11_atomic_init: 974 Form = Init; 975 break; 976 977 case AtomicExpr::AO__c11_atomic_load: 978 case AtomicExpr::AO__atomic_load_n: 979 Form = Load; 980 break; 981 982 case AtomicExpr::AO__c11_atomic_store: 983 case AtomicExpr::AO__atomic_load: 984 case AtomicExpr::AO__atomic_store: 985 case AtomicExpr::AO__atomic_store_n: 986 Form = Copy; 987 break; 988 989 case AtomicExpr::AO__c11_atomic_fetch_add: 990 case AtomicExpr::AO__c11_atomic_fetch_sub: 991 case AtomicExpr::AO__atomic_fetch_add: 992 case AtomicExpr::AO__atomic_fetch_sub: 993 case AtomicExpr::AO__atomic_add_fetch: 994 case AtomicExpr::AO__atomic_sub_fetch: 995 IsAddSub = true; 996 // Fall through. 997 case AtomicExpr::AO__c11_atomic_fetch_and: 998 case AtomicExpr::AO__c11_atomic_fetch_or: 999 case AtomicExpr::AO__c11_atomic_fetch_xor: 1000 case AtomicExpr::AO__atomic_fetch_and: 1001 case AtomicExpr::AO__atomic_fetch_or: 1002 case AtomicExpr::AO__atomic_fetch_xor: 1003 case AtomicExpr::AO__atomic_fetch_nand: 1004 case AtomicExpr::AO__atomic_and_fetch: 1005 case AtomicExpr::AO__atomic_or_fetch: 1006 case AtomicExpr::AO__atomic_xor_fetch: 1007 case AtomicExpr::AO__atomic_nand_fetch: 1008 Form = Arithmetic; 1009 break; 1010 1011 case AtomicExpr::AO__c11_atomic_exchange: 1012 case AtomicExpr::AO__atomic_exchange_n: 1013 Form = Xchg; 1014 break; 1015 1016 case AtomicExpr::AO__atomic_exchange: 1017 Form = GNUXchg; 1018 break; 1019 1020 case AtomicExpr::AO__c11_atomic_compare_exchange_strong: 1021 case AtomicExpr::AO__c11_atomic_compare_exchange_weak: 1022 Form = C11CmpXchg; 1023 break; 1024 1025 case AtomicExpr::AO__atomic_compare_exchange: 1026 case AtomicExpr::AO__atomic_compare_exchange_n: 1027 Form = GNUCmpXchg; 1028 break; 1029 } 1030 1031 // Check we have the right number of arguments. 1032 if (TheCall->getNumArgs() < NumArgs[Form]) { 1033 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1034 << 0 << NumArgs[Form] << TheCall->getNumArgs() 1035 << TheCall->getCallee()->getSourceRange(); 1036 return ExprError(); 1037 } else if (TheCall->getNumArgs() > NumArgs[Form]) { 1038 Diag(TheCall->getArg(NumArgs[Form])->getLocStart(), 1039 diag::err_typecheck_call_too_many_args) 1040 << 0 << NumArgs[Form] << TheCall->getNumArgs() 1041 << TheCall->getCallee()->getSourceRange(); 1042 return ExprError(); 1043 } 1044 1045 // Inspect the first argument of the atomic operation. 1046 Expr *Ptr = TheCall->getArg(0); 1047 Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get(); 1048 const PointerType *pointerType = Ptr->getType()->getAs<PointerType>(); 1049 if (!pointerType) { 1050 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 1051 << Ptr->getType() << Ptr->getSourceRange(); 1052 return ExprError(); 1053 } 1054 1055 // For a __c11 builtin, this should be a pointer to an _Atomic type. 1056 QualType AtomTy = pointerType->getPointeeType(); // 'A' 1057 QualType ValType = AtomTy; // 'C' 1058 if (IsC11) { 1059 if (!AtomTy->isAtomicType()) { 1060 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 1061 << Ptr->getType() << Ptr->getSourceRange(); 1062 return ExprError(); 1063 } 1064 if (AtomTy.isConstQualified()) { 1065 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_non_const_atomic) 1066 << Ptr->getType() << Ptr->getSourceRange(); 1067 return ExprError(); 1068 } 1069 ValType = AtomTy->getAs<AtomicType>()->getValueType(); 1070 } 1071 1072 // For an arithmetic operation, the implied arithmetic must be well-formed. 1073 if (Form == Arithmetic) { 1074 // gcc does not enforce these rules for GNU atomics, but we do so for sanity. 1075 if (IsAddSub && !ValType->isIntegerType() && !ValType->isPointerType()) { 1076 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 1077 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 1078 return ExprError(); 1079 } 1080 if (!IsAddSub && !ValType->isIntegerType()) { 1081 Diag(DRE->getLocStart(), diag::err_atomic_op_bitwise_needs_atomic_int) 1082 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 1083 return ExprError(); 1084 } 1085 } else if (IsN && !ValType->isIntegerType() && !ValType->isPointerType()) { 1086 // For __atomic_*_n operations, the value type must be a scalar integral or 1087 // pointer type which is 1, 2, 4, 8 or 16 bytes in length. 1088 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 1089 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 1090 return ExprError(); 1091 } 1092 1093 if (!IsC11 && !AtomTy.isTriviallyCopyableType(Context) && 1094 !AtomTy->isScalarType()) { 1095 // For GNU atomics, require a trivially-copyable type. This is not part of 1096 // the GNU atomics specification, but we enforce it for sanity. 1097 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_trivial_copy) 1098 << Ptr->getType() << Ptr->getSourceRange(); 1099 return ExprError(); 1100 } 1101 1102 // FIXME: For any builtin other than a load, the ValType must not be 1103 // const-qualified. 1104 1105 switch (ValType.getObjCLifetime()) { 1106 case Qualifiers::OCL_None: 1107 case Qualifiers::OCL_ExplicitNone: 1108 // okay 1109 break; 1110 1111 case Qualifiers::OCL_Weak: 1112 case Qualifiers::OCL_Strong: 1113 case Qualifiers::OCL_Autoreleasing: 1114 // FIXME: Can this happen? By this point, ValType should be known 1115 // to be trivially copyable. 1116 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 1117 << ValType << Ptr->getSourceRange(); 1118 return ExprError(); 1119 } 1120 1121 QualType ResultType = ValType; 1122 if (Form == Copy || Form == GNUXchg || Form == Init) 1123 ResultType = Context.VoidTy; 1124 else if (Form == C11CmpXchg || Form == GNUCmpXchg) 1125 ResultType = Context.BoolTy; 1126 1127 // The type of a parameter passed 'by value'. In the GNU atomics, such 1128 // arguments are actually passed as pointers. 1129 QualType ByValType = ValType; // 'CP' 1130 if (!IsC11 && !IsN) 1131 ByValType = Ptr->getType(); 1132 1133 // The first argument --- the pointer --- has a fixed type; we 1134 // deduce the types of the rest of the arguments accordingly. Walk 1135 // the remaining arguments, converting them to the deduced value type. 1136 for (unsigned i = 1; i != NumArgs[Form]; ++i) { 1137 QualType Ty; 1138 if (i < NumVals[Form] + 1) { 1139 switch (i) { 1140 case 1: 1141 // The second argument is the non-atomic operand. For arithmetic, this 1142 // is always passed by value, and for a compare_exchange it is always 1143 // passed by address. For the rest, GNU uses by-address and C11 uses 1144 // by-value. 1145 assert(Form != Load); 1146 if (Form == Init || (Form == Arithmetic && ValType->isIntegerType())) 1147 Ty = ValType; 1148 else if (Form == Copy || Form == Xchg) 1149 Ty = ByValType; 1150 else if (Form == Arithmetic) 1151 Ty = Context.getPointerDiffType(); 1152 else 1153 Ty = Context.getPointerType(ValType.getUnqualifiedType()); 1154 break; 1155 case 2: 1156 // The third argument to compare_exchange / GNU exchange is a 1157 // (pointer to a) desired value. 1158 Ty = ByValType; 1159 break; 1160 case 3: 1161 // The fourth argument to GNU compare_exchange is a 'weak' flag. 1162 Ty = Context.BoolTy; 1163 break; 1164 } 1165 } else { 1166 // The order(s) are always converted to int. 1167 Ty = Context.IntTy; 1168 } 1169 1170 InitializedEntity Entity = 1171 InitializedEntity::InitializeParameter(Context, Ty, false); 1172 ExprResult Arg = TheCall->getArg(i); 1173 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 1174 if (Arg.isInvalid()) 1175 return true; 1176 TheCall->setArg(i, Arg.get()); 1177 } 1178 1179 // Permute the arguments into a 'consistent' order. 1180 SmallVector<Expr*, 5> SubExprs; 1181 SubExprs.push_back(Ptr); 1182 switch (Form) { 1183 case Init: 1184 // Note, AtomicExpr::getVal1() has a special case for this atomic. 1185 SubExprs.push_back(TheCall->getArg(1)); // Val1 1186 break; 1187 case Load: 1188 SubExprs.push_back(TheCall->getArg(1)); // Order 1189 break; 1190 case Copy: 1191 case Arithmetic: 1192 case Xchg: 1193 SubExprs.push_back(TheCall->getArg(2)); // Order 1194 SubExprs.push_back(TheCall->getArg(1)); // Val1 1195 break; 1196 case GNUXchg: 1197 // Note, AtomicExpr::getVal2() has a special case for this atomic. 1198 SubExprs.push_back(TheCall->getArg(3)); // Order 1199 SubExprs.push_back(TheCall->getArg(1)); // Val1 1200 SubExprs.push_back(TheCall->getArg(2)); // Val2 1201 break; 1202 case C11CmpXchg: 1203 SubExprs.push_back(TheCall->getArg(3)); // Order 1204 SubExprs.push_back(TheCall->getArg(1)); // Val1 1205 SubExprs.push_back(TheCall->getArg(4)); // OrderFail 1206 SubExprs.push_back(TheCall->getArg(2)); // Val2 1207 break; 1208 case GNUCmpXchg: 1209 SubExprs.push_back(TheCall->getArg(4)); // Order 1210 SubExprs.push_back(TheCall->getArg(1)); // Val1 1211 SubExprs.push_back(TheCall->getArg(5)); // OrderFail 1212 SubExprs.push_back(TheCall->getArg(2)); // Val2 1213 SubExprs.push_back(TheCall->getArg(3)); // Weak 1214 break; 1215 } 1216 1217 if (SubExprs.size() >= 2 && Form != Init) { 1218 llvm::APSInt Result(32); 1219 if (SubExprs[1]->isIntegerConstantExpr(Result, Context) && 1220 !isValidOrderingForOp(Result.getSExtValue(), Op)) 1221 Diag(SubExprs[1]->getLocStart(), 1222 diag::warn_atomic_op_has_invalid_memory_order) 1223 << SubExprs[1]->getSourceRange(); 1224 } 1225 1226 AtomicExpr *AE = new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 1227 SubExprs, ResultType, Op, 1228 TheCall->getRParenLoc()); 1229 1230 if ((Op == AtomicExpr::AO__c11_atomic_load || 1231 (Op == AtomicExpr::AO__c11_atomic_store)) && 1232 Context.AtomicUsesUnsupportedLibcall(AE)) 1233 Diag(AE->getLocStart(), diag::err_atomic_load_store_uses_lib) << 1234 ((Op == AtomicExpr::AO__c11_atomic_load) ? 0 : 1); 1235 1236 return AE; 1237 } 1238 1239 1240 /// checkBuiltinArgument - Given a call to a builtin function, perform 1241 /// normal type-checking on the given argument, updating the call in 1242 /// place. This is useful when a builtin function requires custom 1243 /// type-checking for some of its arguments but not necessarily all of 1244 /// them. 1245 /// 1246 /// Returns true on error. 1247 static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 1248 FunctionDecl *Fn = E->getDirectCallee(); 1249 assert(Fn && "builtin call without direct callee!"); 1250 1251 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 1252 InitializedEntity Entity = 1253 InitializedEntity::InitializeParameter(S.Context, Param); 1254 1255 ExprResult Arg = E->getArg(0); 1256 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 1257 if (Arg.isInvalid()) 1258 return true; 1259 1260 E->setArg(ArgIndex, Arg.get()); 1261 return false; 1262 } 1263 1264 /// SemaBuiltinAtomicOverloaded - We have a call to a function like 1265 /// __sync_fetch_and_add, which is an overloaded function based on the pointer 1266 /// type of its first argument. The main ActOnCallExpr routines have already 1267 /// promoted the types of arguments because all of these calls are prototyped as 1268 /// void(...). 1269 /// 1270 /// This function goes through and does final semantic checking for these 1271 /// builtins, 1272 ExprResult 1273 Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 1274 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 1275 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1276 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1277 1278 // Ensure that we have at least one argument to do type inference from. 1279 if (TheCall->getNumArgs() < 1) { 1280 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 1281 << 0 << 1 << TheCall->getNumArgs() 1282 << TheCall->getCallee()->getSourceRange(); 1283 return ExprError(); 1284 } 1285 1286 // Inspect the first argument of the atomic builtin. This should always be 1287 // a pointer type, whose element is an integral scalar or pointer type. 1288 // Because it is a pointer type, we don't have to worry about any implicit 1289 // casts here. 1290 // FIXME: We don't allow floating point scalars as input. 1291 Expr *FirstArg = TheCall->getArg(0); 1292 ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg); 1293 if (FirstArgResult.isInvalid()) 1294 return ExprError(); 1295 FirstArg = FirstArgResult.get(); 1296 TheCall->setArg(0, FirstArg); 1297 1298 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 1299 if (!pointerType) { 1300 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 1301 << FirstArg->getType() << FirstArg->getSourceRange(); 1302 return ExprError(); 1303 } 1304 1305 QualType ValType = pointerType->getPointeeType(); 1306 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 1307 !ValType->isBlockPointerType()) { 1308 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 1309 << FirstArg->getType() << FirstArg->getSourceRange(); 1310 return ExprError(); 1311 } 1312 1313 switch (ValType.getObjCLifetime()) { 1314 case Qualifiers::OCL_None: 1315 case Qualifiers::OCL_ExplicitNone: 1316 // okay 1317 break; 1318 1319 case Qualifiers::OCL_Weak: 1320 case Qualifiers::OCL_Strong: 1321 case Qualifiers::OCL_Autoreleasing: 1322 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 1323 << ValType << FirstArg->getSourceRange(); 1324 return ExprError(); 1325 } 1326 1327 // Strip any qualifiers off ValType. 1328 ValType = ValType.getUnqualifiedType(); 1329 1330 // The majority of builtins return a value, but a few have special return 1331 // types, so allow them to override appropriately below. 1332 QualType ResultType = ValType; 1333 1334 // We need to figure out which concrete builtin this maps onto. For example, 1335 // __sync_fetch_and_add with a 2 byte object turns into 1336 // __sync_fetch_and_add_2. 1337 #define BUILTIN_ROW(x) \ 1338 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 1339 Builtin::BI##x##_8, Builtin::BI##x##_16 } 1340 1341 static const unsigned BuiltinIndices[][5] = { 1342 BUILTIN_ROW(__sync_fetch_and_add), 1343 BUILTIN_ROW(__sync_fetch_and_sub), 1344 BUILTIN_ROW(__sync_fetch_and_or), 1345 BUILTIN_ROW(__sync_fetch_and_and), 1346 BUILTIN_ROW(__sync_fetch_and_xor), 1347 1348 BUILTIN_ROW(__sync_add_and_fetch), 1349 BUILTIN_ROW(__sync_sub_and_fetch), 1350 BUILTIN_ROW(__sync_and_and_fetch), 1351 BUILTIN_ROW(__sync_or_and_fetch), 1352 BUILTIN_ROW(__sync_xor_and_fetch), 1353 1354 BUILTIN_ROW(__sync_val_compare_and_swap), 1355 BUILTIN_ROW(__sync_bool_compare_and_swap), 1356 BUILTIN_ROW(__sync_lock_test_and_set), 1357 BUILTIN_ROW(__sync_lock_release), 1358 BUILTIN_ROW(__sync_swap) 1359 }; 1360 #undef BUILTIN_ROW 1361 1362 // Determine the index of the size. 1363 unsigned SizeIndex; 1364 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 1365 case 1: SizeIndex = 0; break; 1366 case 2: SizeIndex = 1; break; 1367 case 4: SizeIndex = 2; break; 1368 case 8: SizeIndex = 3; break; 1369 case 16: SizeIndex = 4; break; 1370 default: 1371 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 1372 << FirstArg->getType() << FirstArg->getSourceRange(); 1373 return ExprError(); 1374 } 1375 1376 // Each of these builtins has one pointer argument, followed by some number of 1377 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 1378 // that we ignore. Find out which row of BuiltinIndices to read from as well 1379 // as the number of fixed args. 1380 unsigned BuiltinID = FDecl->getBuiltinID(); 1381 unsigned BuiltinIndex, NumFixed = 1; 1382 switch (BuiltinID) { 1383 default: llvm_unreachable("Unknown overloaded atomic builtin!"); 1384 case Builtin::BI__sync_fetch_and_add: 1385 case Builtin::BI__sync_fetch_and_add_1: 1386 case Builtin::BI__sync_fetch_and_add_2: 1387 case Builtin::BI__sync_fetch_and_add_4: 1388 case Builtin::BI__sync_fetch_and_add_8: 1389 case Builtin::BI__sync_fetch_and_add_16: 1390 BuiltinIndex = 0; 1391 break; 1392 1393 case Builtin::BI__sync_fetch_and_sub: 1394 case Builtin::BI__sync_fetch_and_sub_1: 1395 case Builtin::BI__sync_fetch_and_sub_2: 1396 case Builtin::BI__sync_fetch_and_sub_4: 1397 case Builtin::BI__sync_fetch_and_sub_8: 1398 case Builtin::BI__sync_fetch_and_sub_16: 1399 BuiltinIndex = 1; 1400 break; 1401 1402 case Builtin::BI__sync_fetch_and_or: 1403 case Builtin::BI__sync_fetch_and_or_1: 1404 case Builtin::BI__sync_fetch_and_or_2: 1405 case Builtin::BI__sync_fetch_and_or_4: 1406 case Builtin::BI__sync_fetch_and_or_8: 1407 case Builtin::BI__sync_fetch_and_or_16: 1408 BuiltinIndex = 2; 1409 break; 1410 1411 case Builtin::BI__sync_fetch_and_and: 1412 case Builtin::BI__sync_fetch_and_and_1: 1413 case Builtin::BI__sync_fetch_and_and_2: 1414 case Builtin::BI__sync_fetch_and_and_4: 1415 case Builtin::BI__sync_fetch_and_and_8: 1416 case Builtin::BI__sync_fetch_and_and_16: 1417 BuiltinIndex = 3; 1418 break; 1419 1420 case Builtin::BI__sync_fetch_and_xor: 1421 case Builtin::BI__sync_fetch_and_xor_1: 1422 case Builtin::BI__sync_fetch_and_xor_2: 1423 case Builtin::BI__sync_fetch_and_xor_4: 1424 case Builtin::BI__sync_fetch_and_xor_8: 1425 case Builtin::BI__sync_fetch_and_xor_16: 1426 BuiltinIndex = 4; 1427 break; 1428 1429 case Builtin::BI__sync_add_and_fetch: 1430 case Builtin::BI__sync_add_and_fetch_1: 1431 case Builtin::BI__sync_add_and_fetch_2: 1432 case Builtin::BI__sync_add_and_fetch_4: 1433 case Builtin::BI__sync_add_and_fetch_8: 1434 case Builtin::BI__sync_add_and_fetch_16: 1435 BuiltinIndex = 5; 1436 break; 1437 1438 case Builtin::BI__sync_sub_and_fetch: 1439 case Builtin::BI__sync_sub_and_fetch_1: 1440 case Builtin::BI__sync_sub_and_fetch_2: 1441 case Builtin::BI__sync_sub_and_fetch_4: 1442 case Builtin::BI__sync_sub_and_fetch_8: 1443 case Builtin::BI__sync_sub_and_fetch_16: 1444 BuiltinIndex = 6; 1445 break; 1446 1447 case Builtin::BI__sync_and_and_fetch: 1448 case Builtin::BI__sync_and_and_fetch_1: 1449 case Builtin::BI__sync_and_and_fetch_2: 1450 case Builtin::BI__sync_and_and_fetch_4: 1451 case Builtin::BI__sync_and_and_fetch_8: 1452 case Builtin::BI__sync_and_and_fetch_16: 1453 BuiltinIndex = 7; 1454 break; 1455 1456 case Builtin::BI__sync_or_and_fetch: 1457 case Builtin::BI__sync_or_and_fetch_1: 1458 case Builtin::BI__sync_or_and_fetch_2: 1459 case Builtin::BI__sync_or_and_fetch_4: 1460 case Builtin::BI__sync_or_and_fetch_8: 1461 case Builtin::BI__sync_or_and_fetch_16: 1462 BuiltinIndex = 8; 1463 break; 1464 1465 case Builtin::BI__sync_xor_and_fetch: 1466 case Builtin::BI__sync_xor_and_fetch_1: 1467 case Builtin::BI__sync_xor_and_fetch_2: 1468 case Builtin::BI__sync_xor_and_fetch_4: 1469 case Builtin::BI__sync_xor_and_fetch_8: 1470 case Builtin::BI__sync_xor_and_fetch_16: 1471 BuiltinIndex = 9; 1472 break; 1473 1474 case Builtin::BI__sync_val_compare_and_swap: 1475 case Builtin::BI__sync_val_compare_and_swap_1: 1476 case Builtin::BI__sync_val_compare_and_swap_2: 1477 case Builtin::BI__sync_val_compare_and_swap_4: 1478 case Builtin::BI__sync_val_compare_and_swap_8: 1479 case Builtin::BI__sync_val_compare_and_swap_16: 1480 BuiltinIndex = 10; 1481 NumFixed = 2; 1482 break; 1483 1484 case Builtin::BI__sync_bool_compare_and_swap: 1485 case Builtin::BI__sync_bool_compare_and_swap_1: 1486 case Builtin::BI__sync_bool_compare_and_swap_2: 1487 case Builtin::BI__sync_bool_compare_and_swap_4: 1488 case Builtin::BI__sync_bool_compare_and_swap_8: 1489 case Builtin::BI__sync_bool_compare_and_swap_16: 1490 BuiltinIndex = 11; 1491 NumFixed = 2; 1492 ResultType = Context.BoolTy; 1493 break; 1494 1495 case Builtin::BI__sync_lock_test_and_set: 1496 case Builtin::BI__sync_lock_test_and_set_1: 1497 case Builtin::BI__sync_lock_test_and_set_2: 1498 case Builtin::BI__sync_lock_test_and_set_4: 1499 case Builtin::BI__sync_lock_test_and_set_8: 1500 case Builtin::BI__sync_lock_test_and_set_16: 1501 BuiltinIndex = 12; 1502 break; 1503 1504 case Builtin::BI__sync_lock_release: 1505 case Builtin::BI__sync_lock_release_1: 1506 case Builtin::BI__sync_lock_release_2: 1507 case Builtin::BI__sync_lock_release_4: 1508 case Builtin::BI__sync_lock_release_8: 1509 case Builtin::BI__sync_lock_release_16: 1510 BuiltinIndex = 13; 1511 NumFixed = 0; 1512 ResultType = Context.VoidTy; 1513 break; 1514 1515 case Builtin::BI__sync_swap: 1516 case Builtin::BI__sync_swap_1: 1517 case Builtin::BI__sync_swap_2: 1518 case Builtin::BI__sync_swap_4: 1519 case Builtin::BI__sync_swap_8: 1520 case Builtin::BI__sync_swap_16: 1521 BuiltinIndex = 14; 1522 break; 1523 } 1524 1525 // Now that we know how many fixed arguments we expect, first check that we 1526 // have at least that many. 1527 if (TheCall->getNumArgs() < 1+NumFixed) { 1528 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 1529 << 0 << 1+NumFixed << TheCall->getNumArgs() 1530 << TheCall->getCallee()->getSourceRange(); 1531 return ExprError(); 1532 } 1533 1534 // Get the decl for the concrete builtin from this, we can tell what the 1535 // concrete integer type we should convert to is. 1536 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 1537 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 1538 FunctionDecl *NewBuiltinDecl; 1539 if (NewBuiltinID == BuiltinID) 1540 NewBuiltinDecl = FDecl; 1541 else { 1542 // Perform builtin lookup to avoid redeclaring it. 1543 DeclarationName DN(&Context.Idents.get(NewBuiltinName)); 1544 LookupResult Res(*this, DN, DRE->getLocStart(), LookupOrdinaryName); 1545 LookupName(Res, TUScope, /*AllowBuiltinCreation=*/true); 1546 assert(Res.getFoundDecl()); 1547 NewBuiltinDecl = dyn_cast<FunctionDecl>(Res.getFoundDecl()); 1548 if (!NewBuiltinDecl) 1549 return ExprError(); 1550 } 1551 1552 // The first argument --- the pointer --- has a fixed type; we 1553 // deduce the types of the rest of the arguments accordingly. Walk 1554 // the remaining arguments, converting them to the deduced value type. 1555 for (unsigned i = 0; i != NumFixed; ++i) { 1556 ExprResult Arg = TheCall->getArg(i+1); 1557 1558 // GCC does an implicit conversion to the pointer or integer ValType. This 1559 // can fail in some cases (1i -> int**), check for this error case now. 1560 // Initialize the argument. 1561 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context, 1562 ValType, /*consume*/ false); 1563 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 1564 if (Arg.isInvalid()) 1565 return ExprError(); 1566 1567 // Okay, we have something that *can* be converted to the right type. Check 1568 // to see if there is a potentially weird extension going on here. This can 1569 // happen when you do an atomic operation on something like an char* and 1570 // pass in 42. The 42 gets converted to char. This is even more strange 1571 // for things like 45.123 -> char, etc. 1572 // FIXME: Do this check. 1573 TheCall->setArg(i+1, Arg.get()); 1574 } 1575 1576 ASTContext& Context = this->getASTContext(); 1577 1578 // Create a new DeclRefExpr to refer to the new decl. 1579 DeclRefExpr* NewDRE = DeclRefExpr::Create( 1580 Context, 1581 DRE->getQualifierLoc(), 1582 SourceLocation(), 1583 NewBuiltinDecl, 1584 /*enclosing*/ false, 1585 DRE->getLocation(), 1586 Context.BuiltinFnTy, 1587 DRE->getValueKind()); 1588 1589 // Set the callee in the CallExpr. 1590 // FIXME: This loses syntactic information. 1591 QualType CalleePtrTy = Context.getPointerType(NewBuiltinDecl->getType()); 1592 ExprResult PromotedCall = ImpCastExprToType(NewDRE, CalleePtrTy, 1593 CK_BuiltinFnToFnPtr); 1594 TheCall->setCallee(PromotedCall.get()); 1595 1596 // Change the result type of the call to match the original value type. This 1597 // is arbitrary, but the codegen for these builtins ins design to handle it 1598 // gracefully. 1599 TheCall->setType(ResultType); 1600 1601 return TheCallResult; 1602 } 1603 1604 /// CheckObjCString - Checks that the argument to the builtin 1605 /// CFString constructor is correct 1606 /// Note: It might also make sense to do the UTF-16 conversion here (would 1607 /// simplify the backend). 1608 bool Sema::CheckObjCString(Expr *Arg) { 1609 Arg = Arg->IgnoreParenCasts(); 1610 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 1611 1612 if (!Literal || !Literal->isAscii()) { 1613 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 1614 << Arg->getSourceRange(); 1615 return true; 1616 } 1617 1618 if (Literal->containsNonAsciiOrNull()) { 1619 StringRef String = Literal->getString(); 1620 unsigned NumBytes = String.size(); 1621 SmallVector<UTF16, 128> ToBuf(NumBytes); 1622 const UTF8 *FromPtr = (const UTF8 *)String.data(); 1623 UTF16 *ToPtr = &ToBuf[0]; 1624 1625 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 1626 &ToPtr, ToPtr + NumBytes, 1627 strictConversion); 1628 // Check for conversion failure. 1629 if (Result != conversionOK) 1630 Diag(Arg->getLocStart(), 1631 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 1632 } 1633 return false; 1634 } 1635 1636 /// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 1637 /// Emit an error and return true on failure, return false on success. 1638 bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 1639 Expr *Fn = TheCall->getCallee(); 1640 if (TheCall->getNumArgs() > 2) { 1641 Diag(TheCall->getArg(2)->getLocStart(), 1642 diag::err_typecheck_call_too_many_args) 1643 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1644 << Fn->getSourceRange() 1645 << SourceRange(TheCall->getArg(2)->getLocStart(), 1646 (*(TheCall->arg_end()-1))->getLocEnd()); 1647 return true; 1648 } 1649 1650 if (TheCall->getNumArgs() < 2) { 1651 return Diag(TheCall->getLocEnd(), 1652 diag::err_typecheck_call_too_few_args_at_least) 1653 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 1654 } 1655 1656 // Type-check the first argument normally. 1657 if (checkBuiltinArgument(*this, TheCall, 0)) 1658 return true; 1659 1660 // Determine whether the current function is variadic or not. 1661 BlockScopeInfo *CurBlock = getCurBlock(); 1662 bool isVariadic; 1663 if (CurBlock) 1664 isVariadic = CurBlock->TheDecl->isVariadic(); 1665 else if (FunctionDecl *FD = getCurFunctionDecl()) 1666 isVariadic = FD->isVariadic(); 1667 else 1668 isVariadic = getCurMethodDecl()->isVariadic(); 1669 1670 if (!isVariadic) { 1671 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 1672 return true; 1673 } 1674 1675 // Verify that the second argument to the builtin is the last argument of the 1676 // current function or method. 1677 bool SecondArgIsLastNamedArgument = false; 1678 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 1679 1680 // These are valid if SecondArgIsLastNamedArgument is false after the next 1681 // block. 1682 QualType Type; 1683 SourceLocation ParamLoc; 1684 1685 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 1686 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 1687 // FIXME: This isn't correct for methods (results in bogus warning). 1688 // Get the last formal in the current function. 1689 const ParmVarDecl *LastArg; 1690 if (CurBlock) 1691 LastArg = *(CurBlock->TheDecl->param_end()-1); 1692 else if (FunctionDecl *FD = getCurFunctionDecl()) 1693 LastArg = *(FD->param_end()-1); 1694 else 1695 LastArg = *(getCurMethodDecl()->param_end()-1); 1696 SecondArgIsLastNamedArgument = PV == LastArg; 1697 1698 Type = PV->getType(); 1699 ParamLoc = PV->getLocation(); 1700 } 1701 } 1702 1703 if (!SecondArgIsLastNamedArgument) 1704 Diag(TheCall->getArg(1)->getLocStart(), 1705 diag::warn_second_parameter_of_va_start_not_last_named_argument); 1706 else if (Type->isReferenceType()) { 1707 Diag(Arg->getLocStart(), 1708 diag::warn_va_start_of_reference_type_is_undefined); 1709 Diag(ParamLoc, diag::note_parameter_type) << Type; 1710 } 1711 1712 TheCall->setType(Context.VoidTy); 1713 return false; 1714 } 1715 1716 /// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 1717 /// friends. This is declared to take (...), so we have to check everything. 1718 bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 1719 if (TheCall->getNumArgs() < 2) 1720 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1721 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 1722 if (TheCall->getNumArgs() > 2) 1723 return Diag(TheCall->getArg(2)->getLocStart(), 1724 diag::err_typecheck_call_too_many_args) 1725 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1726 << SourceRange(TheCall->getArg(2)->getLocStart(), 1727 (*(TheCall->arg_end()-1))->getLocEnd()); 1728 1729 ExprResult OrigArg0 = TheCall->getArg(0); 1730 ExprResult OrigArg1 = TheCall->getArg(1); 1731 1732 // Do standard promotions between the two arguments, returning their common 1733 // type. 1734 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 1735 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 1736 return true; 1737 1738 // Make sure any conversions are pushed back into the call; this is 1739 // type safe since unordered compare builtins are declared as "_Bool 1740 // foo(...)". 1741 TheCall->setArg(0, OrigArg0.get()); 1742 TheCall->setArg(1, OrigArg1.get()); 1743 1744 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 1745 return false; 1746 1747 // If the common type isn't a real floating type, then the arguments were 1748 // invalid for this operation. 1749 if (Res.isNull() || !Res->isRealFloatingType()) 1750 return Diag(OrigArg0.get()->getLocStart(), 1751 diag::err_typecheck_call_invalid_ordered_compare) 1752 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 1753 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 1754 1755 return false; 1756 } 1757 1758 /// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 1759 /// __builtin_isnan and friends. This is declared to take (...), so we have 1760 /// to check everything. We expect the last argument to be a floating point 1761 /// value. 1762 bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 1763 if (TheCall->getNumArgs() < NumArgs) 1764 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1765 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 1766 if (TheCall->getNumArgs() > NumArgs) 1767 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 1768 diag::err_typecheck_call_too_many_args) 1769 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 1770 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 1771 (*(TheCall->arg_end()-1))->getLocEnd()); 1772 1773 Expr *OrigArg = TheCall->getArg(NumArgs-1); 1774 1775 if (OrigArg->isTypeDependent()) 1776 return false; 1777 1778 // This operation requires a non-_Complex floating-point number. 1779 if (!OrigArg->getType()->isRealFloatingType()) 1780 return Diag(OrigArg->getLocStart(), 1781 diag::err_typecheck_call_invalid_unary_fp) 1782 << OrigArg->getType() << OrigArg->getSourceRange(); 1783 1784 // If this is an implicit conversion from float -> double, remove it. 1785 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 1786 Expr *CastArg = Cast->getSubExpr(); 1787 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 1788 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 1789 "promotion from float to double is the only expected cast here"); 1790 Cast->setSubExpr(nullptr); 1791 TheCall->setArg(NumArgs-1, CastArg); 1792 } 1793 } 1794 1795 return false; 1796 } 1797 1798 /// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1799 // This is declared to take (...), so we have to check everything. 1800 ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1801 if (TheCall->getNumArgs() < 2) 1802 return ExprError(Diag(TheCall->getLocEnd(), 1803 diag::err_typecheck_call_too_few_args_at_least) 1804 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1805 << TheCall->getSourceRange()); 1806 1807 // Determine which of the following types of shufflevector we're checking: 1808 // 1) unary, vector mask: (lhs, mask) 1809 // 2) binary, vector mask: (lhs, rhs, mask) 1810 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1811 QualType resType = TheCall->getArg(0)->getType(); 1812 unsigned numElements = 0; 1813 1814 if (!TheCall->getArg(0)->isTypeDependent() && 1815 !TheCall->getArg(1)->isTypeDependent()) { 1816 QualType LHSType = TheCall->getArg(0)->getType(); 1817 QualType RHSType = TheCall->getArg(1)->getType(); 1818 1819 if (!LHSType->isVectorType() || !RHSType->isVectorType()) 1820 return ExprError(Diag(TheCall->getLocStart(), 1821 diag::err_shufflevector_non_vector) 1822 << SourceRange(TheCall->getArg(0)->getLocStart(), 1823 TheCall->getArg(1)->getLocEnd())); 1824 1825 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1826 unsigned numResElements = TheCall->getNumArgs() - 2; 1827 1828 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1829 // with mask. If so, verify that RHS is an integer vector type with the 1830 // same number of elts as lhs. 1831 if (TheCall->getNumArgs() == 2) { 1832 if (!RHSType->hasIntegerRepresentation() || 1833 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1834 return ExprError(Diag(TheCall->getLocStart(), 1835 diag::err_shufflevector_incompatible_vector) 1836 << SourceRange(TheCall->getArg(1)->getLocStart(), 1837 TheCall->getArg(1)->getLocEnd())); 1838 } else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1839 return ExprError(Diag(TheCall->getLocStart(), 1840 diag::err_shufflevector_incompatible_vector) 1841 << SourceRange(TheCall->getArg(0)->getLocStart(), 1842 TheCall->getArg(1)->getLocEnd())); 1843 } else if (numElements != numResElements) { 1844 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1845 resType = Context.getVectorType(eltType, numResElements, 1846 VectorType::GenericVector); 1847 } 1848 } 1849 1850 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1851 if (TheCall->getArg(i)->isTypeDependent() || 1852 TheCall->getArg(i)->isValueDependent()) 1853 continue; 1854 1855 llvm::APSInt Result(32); 1856 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1857 return ExprError(Diag(TheCall->getLocStart(), 1858 diag::err_shufflevector_nonconstant_argument) 1859 << TheCall->getArg(i)->getSourceRange()); 1860 1861 // Allow -1 which will be translated to undef in the IR. 1862 if (Result.isSigned() && Result.isAllOnesValue()) 1863 continue; 1864 1865 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1866 return ExprError(Diag(TheCall->getLocStart(), 1867 diag::err_shufflevector_argument_too_large) 1868 << TheCall->getArg(i)->getSourceRange()); 1869 } 1870 1871 SmallVector<Expr*, 32> exprs; 1872 1873 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1874 exprs.push_back(TheCall->getArg(i)); 1875 TheCall->setArg(i, nullptr); 1876 } 1877 1878 return new (Context) ShuffleVectorExpr(Context, exprs, resType, 1879 TheCall->getCallee()->getLocStart(), 1880 TheCall->getRParenLoc()); 1881 } 1882 1883 /// SemaConvertVectorExpr - Handle __builtin_convertvector 1884 ExprResult Sema::SemaConvertVectorExpr(Expr *E, TypeSourceInfo *TInfo, 1885 SourceLocation BuiltinLoc, 1886 SourceLocation RParenLoc) { 1887 ExprValueKind VK = VK_RValue; 1888 ExprObjectKind OK = OK_Ordinary; 1889 QualType DstTy = TInfo->getType(); 1890 QualType SrcTy = E->getType(); 1891 1892 if (!SrcTy->isVectorType() && !SrcTy->isDependentType()) 1893 return ExprError(Diag(BuiltinLoc, 1894 diag::err_convertvector_non_vector) 1895 << E->getSourceRange()); 1896 if (!DstTy->isVectorType() && !DstTy->isDependentType()) 1897 return ExprError(Diag(BuiltinLoc, 1898 diag::err_convertvector_non_vector_type)); 1899 1900 if (!SrcTy->isDependentType() && !DstTy->isDependentType()) { 1901 unsigned SrcElts = SrcTy->getAs<VectorType>()->getNumElements(); 1902 unsigned DstElts = DstTy->getAs<VectorType>()->getNumElements(); 1903 if (SrcElts != DstElts) 1904 return ExprError(Diag(BuiltinLoc, 1905 diag::err_convertvector_incompatible_vector) 1906 << E->getSourceRange()); 1907 } 1908 1909 return new (Context) 1910 ConvertVectorExpr(E, TInfo, DstTy, VK, OK, BuiltinLoc, RParenLoc); 1911 } 1912 1913 /// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1914 // This is declared to take (const void*, ...) and can take two 1915 // optional constant int args. 1916 bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1917 unsigned NumArgs = TheCall->getNumArgs(); 1918 1919 if (NumArgs > 3) 1920 return Diag(TheCall->getLocEnd(), 1921 diag::err_typecheck_call_too_many_args_at_most) 1922 << 0 /*function call*/ << 3 << NumArgs 1923 << TheCall->getSourceRange(); 1924 1925 // Argument 0 is checked for us and the remaining arguments must be 1926 // constant integers. 1927 for (unsigned i = 1; i != NumArgs; ++i) 1928 if (SemaBuiltinConstantArgRange(TheCall, i, 0, i == 1 ? 1 : 3)) 1929 return true; 1930 1931 return false; 1932 } 1933 1934 /// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1935 /// TheCall is a constant expression. 1936 bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1937 llvm::APSInt &Result) { 1938 Expr *Arg = TheCall->getArg(ArgNum); 1939 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1940 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1941 1942 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 1943 1944 if (!Arg->isIntegerConstantExpr(Result, Context)) 1945 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 1946 << FDecl->getDeclName() << Arg->getSourceRange(); 1947 1948 return false; 1949 } 1950 1951 /// SemaBuiltinConstantArgRange - Handle a check if argument ArgNum of CallExpr 1952 /// TheCall is a constant expression in the range [Low, High]. 1953 bool Sema::SemaBuiltinConstantArgRange(CallExpr *TheCall, int ArgNum, 1954 int Low, int High) { 1955 llvm::APSInt Result; 1956 1957 // We can't check the value of a dependent argument. 1958 Expr *Arg = TheCall->getArg(ArgNum); 1959 if (Arg->isTypeDependent() || Arg->isValueDependent()) 1960 return false; 1961 1962 // Check constant-ness first. 1963 if (SemaBuiltinConstantArg(TheCall, ArgNum, Result)) 1964 return true; 1965 1966 if (Result.getSExtValue() < Low || Result.getSExtValue() > High) 1967 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1968 << Low << High << Arg->getSourceRange(); 1969 1970 return false; 1971 } 1972 1973 /// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1974 /// This checks that val is a constant 1. 1975 bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1976 Expr *Arg = TheCall->getArg(1); 1977 llvm::APSInt Result; 1978 1979 // TODO: This is less than ideal. Overload this to take a value. 1980 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1981 return true; 1982 1983 if (Result != 1) 1984 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1985 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1986 1987 return false; 1988 } 1989 1990 namespace { 1991 enum StringLiteralCheckType { 1992 SLCT_NotALiteral, 1993 SLCT_UncheckedLiteral, 1994 SLCT_CheckedLiteral 1995 }; 1996 } 1997 1998 // Determine if an expression is a string literal or constant string. 1999 // If this function returns false on the arguments to a function expecting a 2000 // format string, we will usually need to emit a warning. 2001 // True string literals are then checked by CheckFormatString. 2002 static StringLiteralCheckType 2003 checkFormatStringExpr(Sema &S, const Expr *E, ArrayRef<const Expr *> Args, 2004 bool HasVAListArg, unsigned format_idx, 2005 unsigned firstDataArg, Sema::FormatStringType Type, 2006 Sema::VariadicCallType CallType, bool InFunctionCall, 2007 llvm::SmallBitVector &CheckedVarArgs) { 2008 tryAgain: 2009 if (E->isTypeDependent() || E->isValueDependent()) 2010 return SLCT_NotALiteral; 2011 2012 E = E->IgnoreParenCasts(); 2013 2014 if (E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull)) 2015 // Technically -Wformat-nonliteral does not warn about this case. 2016 // The behavior of printf and friends in this case is implementation 2017 // dependent. Ideally if the format string cannot be null then 2018 // it should have a 'nonnull' attribute in the function prototype. 2019 return SLCT_UncheckedLiteral; 2020 2021 switch (E->getStmtClass()) { 2022 case Stmt::BinaryConditionalOperatorClass: 2023 case Stmt::ConditionalOperatorClass: { 2024 // The expression is a literal if both sub-expressions were, and it was 2025 // completely checked only if both sub-expressions were checked. 2026 const AbstractConditionalOperator *C = 2027 cast<AbstractConditionalOperator>(E); 2028 StringLiteralCheckType Left = 2029 checkFormatStringExpr(S, C->getTrueExpr(), Args, 2030 HasVAListArg, format_idx, firstDataArg, 2031 Type, CallType, InFunctionCall, CheckedVarArgs); 2032 if (Left == SLCT_NotALiteral) 2033 return SLCT_NotALiteral; 2034 StringLiteralCheckType Right = 2035 checkFormatStringExpr(S, C->getFalseExpr(), Args, 2036 HasVAListArg, format_idx, firstDataArg, 2037 Type, CallType, InFunctionCall, CheckedVarArgs); 2038 return Left < Right ? Left : Right; 2039 } 2040 2041 case Stmt::ImplicitCastExprClass: { 2042 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 2043 goto tryAgain; 2044 } 2045 2046 case Stmt::OpaqueValueExprClass: 2047 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 2048 E = src; 2049 goto tryAgain; 2050 } 2051 return SLCT_NotALiteral; 2052 2053 case Stmt::PredefinedExprClass: 2054 // While __func__, etc., are technically not string literals, they 2055 // cannot contain format specifiers and thus are not a security 2056 // liability. 2057 return SLCT_UncheckedLiteral; 2058 2059 case Stmt::DeclRefExprClass: { 2060 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 2061 2062 // As an exception, do not flag errors for variables binding to 2063 // const string literals. 2064 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 2065 bool isConstant = false; 2066 QualType T = DR->getType(); 2067 2068 if (const ArrayType *AT = S.Context.getAsArrayType(T)) { 2069 isConstant = AT->getElementType().isConstant(S.Context); 2070 } else if (const PointerType *PT = T->getAs<PointerType>()) { 2071 isConstant = T.isConstant(S.Context) && 2072 PT->getPointeeType().isConstant(S.Context); 2073 } else if (T->isObjCObjectPointerType()) { 2074 // In ObjC, there is usually no "const ObjectPointer" type, 2075 // so don't check if the pointee type is constant. 2076 isConstant = T.isConstant(S.Context); 2077 } 2078 2079 if (isConstant) { 2080 if (const Expr *Init = VD->getAnyInitializer()) { 2081 // Look through initializers like const char c[] = { "foo" } 2082 if (const InitListExpr *InitList = dyn_cast<InitListExpr>(Init)) { 2083 if (InitList->isStringLiteralInit()) 2084 Init = InitList->getInit(0)->IgnoreParenImpCasts(); 2085 } 2086 return checkFormatStringExpr(S, Init, Args, 2087 HasVAListArg, format_idx, 2088 firstDataArg, Type, CallType, 2089 /*InFunctionCall*/false, CheckedVarArgs); 2090 } 2091 } 2092 2093 // For vprintf* functions (i.e., HasVAListArg==true), we add a 2094 // special check to see if the format string is a function parameter 2095 // of the function calling the printf function. If the function 2096 // has an attribute indicating it is a printf-like function, then we 2097 // should suppress warnings concerning non-literals being used in a call 2098 // to a vprintf function. For example: 2099 // 2100 // void 2101 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 2102 // va_list ap; 2103 // va_start(ap, fmt); 2104 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 2105 // ... 2106 // } 2107 if (HasVAListArg) { 2108 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(VD)) { 2109 if (const NamedDecl *ND = dyn_cast<NamedDecl>(PV->getDeclContext())) { 2110 int PVIndex = PV->getFunctionScopeIndex() + 1; 2111 for (const auto *PVFormat : ND->specific_attrs<FormatAttr>()) { 2112 // adjust for implicit parameter 2113 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 2114 if (MD->isInstance()) 2115 ++PVIndex; 2116 // We also check if the formats are compatible. 2117 // We can't pass a 'scanf' string to a 'printf' function. 2118 if (PVIndex == PVFormat->getFormatIdx() && 2119 Type == S.GetFormatStringType(PVFormat)) 2120 return SLCT_UncheckedLiteral; 2121 } 2122 } 2123 } 2124 } 2125 } 2126 2127 return SLCT_NotALiteral; 2128 } 2129 2130 case Stmt::CallExprClass: 2131 case Stmt::CXXMemberCallExprClass: { 2132 const CallExpr *CE = cast<CallExpr>(E); 2133 if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(CE->getCalleeDecl())) { 2134 if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) { 2135 unsigned ArgIndex = FA->getFormatIdx(); 2136 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 2137 if (MD->isInstance()) 2138 --ArgIndex; 2139 const Expr *Arg = CE->getArg(ArgIndex - 1); 2140 2141 return checkFormatStringExpr(S, Arg, Args, 2142 HasVAListArg, format_idx, firstDataArg, 2143 Type, CallType, InFunctionCall, 2144 CheckedVarArgs); 2145 } else if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(ND)) { 2146 unsigned BuiltinID = FD->getBuiltinID(); 2147 if (BuiltinID == Builtin::BI__builtin___CFStringMakeConstantString || 2148 BuiltinID == Builtin::BI__builtin___NSStringMakeConstantString) { 2149 const Expr *Arg = CE->getArg(0); 2150 return checkFormatStringExpr(S, Arg, Args, 2151 HasVAListArg, format_idx, 2152 firstDataArg, Type, CallType, 2153 InFunctionCall, CheckedVarArgs); 2154 } 2155 } 2156 } 2157 2158 return SLCT_NotALiteral; 2159 } 2160 case Stmt::ObjCStringLiteralClass: 2161 case Stmt::StringLiteralClass: { 2162 const StringLiteral *StrE = nullptr; 2163 2164 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 2165 StrE = ObjCFExpr->getString(); 2166 else 2167 StrE = cast<StringLiteral>(E); 2168 2169 if (StrE) { 2170 S.CheckFormatString(StrE, E, Args, HasVAListArg, format_idx, firstDataArg, 2171 Type, InFunctionCall, CallType, CheckedVarArgs); 2172 return SLCT_CheckedLiteral; 2173 } 2174 2175 return SLCT_NotALiteral; 2176 } 2177 2178 default: 2179 return SLCT_NotALiteral; 2180 } 2181 } 2182 2183 Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) { 2184 return llvm::StringSwitch<FormatStringType>(Format->getType()->getName()) 2185 .Case("scanf", FST_Scanf) 2186 .Cases("printf", "printf0", FST_Printf) 2187 .Cases("NSString", "CFString", FST_NSString) 2188 .Case("strftime", FST_Strftime) 2189 .Case("strfmon", FST_Strfmon) 2190 .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf) 2191 .Default(FST_Unknown); 2192 } 2193 2194 /// CheckFormatArguments - Check calls to printf and scanf (and similar 2195 /// functions) for correct use of format strings. 2196 /// Returns true if a format string has been fully checked. 2197 bool Sema::CheckFormatArguments(const FormatAttr *Format, 2198 ArrayRef<const Expr *> Args, 2199 bool IsCXXMember, 2200 VariadicCallType CallType, 2201 SourceLocation Loc, SourceRange Range, 2202 llvm::SmallBitVector &CheckedVarArgs) { 2203 FormatStringInfo FSI; 2204 if (getFormatStringInfo(Format, IsCXXMember, &FSI)) 2205 return CheckFormatArguments(Args, FSI.HasVAListArg, FSI.FormatIdx, 2206 FSI.FirstDataArg, GetFormatStringType(Format), 2207 CallType, Loc, Range, CheckedVarArgs); 2208 return false; 2209 } 2210 2211 bool Sema::CheckFormatArguments(ArrayRef<const Expr *> Args, 2212 bool HasVAListArg, unsigned format_idx, 2213 unsigned firstDataArg, FormatStringType Type, 2214 VariadicCallType CallType, 2215 SourceLocation Loc, SourceRange Range, 2216 llvm::SmallBitVector &CheckedVarArgs) { 2217 // CHECK: printf/scanf-like function is called with no format string. 2218 if (format_idx >= Args.size()) { 2219 Diag(Loc, diag::warn_missing_format_string) << Range; 2220 return false; 2221 } 2222 2223 const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts(); 2224 2225 // CHECK: format string is not a string literal. 2226 // 2227 // Dynamically generated format strings are difficult to 2228 // automatically vet at compile time. Requiring that format strings 2229 // are string literals: (1) permits the checking of format strings by 2230 // the compiler and thereby (2) can practically remove the source of 2231 // many format string exploits. 2232 2233 // Format string can be either ObjC string (e.g. @"%d") or 2234 // C string (e.g. "%d") 2235 // ObjC string uses the same format specifiers as C string, so we can use 2236 // the same format string checking logic for both ObjC and C strings. 2237 StringLiteralCheckType CT = 2238 checkFormatStringExpr(*this, OrigFormatExpr, Args, HasVAListArg, 2239 format_idx, firstDataArg, Type, CallType, 2240 /*IsFunctionCall*/true, CheckedVarArgs); 2241 if (CT != SLCT_NotALiteral) 2242 // Literal format string found, check done! 2243 return CT == SLCT_CheckedLiteral; 2244 2245 // Strftime is particular as it always uses a single 'time' argument, 2246 // so it is safe to pass a non-literal string. 2247 if (Type == FST_Strftime) 2248 return false; 2249 2250 // Do not emit diag when the string param is a macro expansion and the 2251 // format is either NSString or CFString. This is a hack to prevent 2252 // diag when using the NSLocalizedString and CFCopyLocalizedString macros 2253 // which are usually used in place of NS and CF string literals. 2254 if (Type == FST_NSString && 2255 SourceMgr.isInSystemMacro(Args[format_idx]->getLocStart())) 2256 return false; 2257 2258 // If there are no arguments specified, warn with -Wformat-security, otherwise 2259 // warn only with -Wformat-nonliteral. 2260 if (Args.size() == firstDataArg) 2261 Diag(Args[format_idx]->getLocStart(), 2262 diag::warn_format_nonliteral_noargs) 2263 << OrigFormatExpr->getSourceRange(); 2264 else 2265 Diag(Args[format_idx]->getLocStart(), 2266 diag::warn_format_nonliteral) 2267 << OrigFormatExpr->getSourceRange(); 2268 return false; 2269 } 2270 2271 namespace { 2272 class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 2273 protected: 2274 Sema &S; 2275 const StringLiteral *FExpr; 2276 const Expr *OrigFormatExpr; 2277 const unsigned FirstDataArg; 2278 const unsigned NumDataArgs; 2279 const char *Beg; // Start of format string. 2280 const bool HasVAListArg; 2281 ArrayRef<const Expr *> Args; 2282 unsigned FormatIdx; 2283 llvm::SmallBitVector CoveredArgs; 2284 bool usesPositionalArgs; 2285 bool atFirstArg; 2286 bool inFunctionCall; 2287 Sema::VariadicCallType CallType; 2288 llvm::SmallBitVector &CheckedVarArgs; 2289 public: 2290 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 2291 const Expr *origFormatExpr, unsigned firstDataArg, 2292 unsigned numDataArgs, const char *beg, bool hasVAListArg, 2293 ArrayRef<const Expr *> Args, 2294 unsigned formatIdx, bool inFunctionCall, 2295 Sema::VariadicCallType callType, 2296 llvm::SmallBitVector &CheckedVarArgs) 2297 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 2298 FirstDataArg(firstDataArg), NumDataArgs(numDataArgs), 2299 Beg(beg), HasVAListArg(hasVAListArg), 2300 Args(Args), FormatIdx(formatIdx), 2301 usesPositionalArgs(false), atFirstArg(true), 2302 inFunctionCall(inFunctionCall), CallType(callType), 2303 CheckedVarArgs(CheckedVarArgs) { 2304 CoveredArgs.resize(numDataArgs); 2305 CoveredArgs.reset(); 2306 } 2307 2308 void DoneProcessing(); 2309 2310 void HandleIncompleteSpecifier(const char *startSpecifier, 2311 unsigned specifierLen) override; 2312 2313 void HandleInvalidLengthModifier( 2314 const analyze_format_string::FormatSpecifier &FS, 2315 const analyze_format_string::ConversionSpecifier &CS, 2316 const char *startSpecifier, unsigned specifierLen, 2317 unsigned DiagID); 2318 2319 void HandleNonStandardLengthModifier( 2320 const analyze_format_string::FormatSpecifier &FS, 2321 const char *startSpecifier, unsigned specifierLen); 2322 2323 void HandleNonStandardConversionSpecifier( 2324 const analyze_format_string::ConversionSpecifier &CS, 2325 const char *startSpecifier, unsigned specifierLen); 2326 2327 void HandlePosition(const char *startPos, unsigned posLen) override; 2328 2329 void HandleInvalidPosition(const char *startSpecifier, 2330 unsigned specifierLen, 2331 analyze_format_string::PositionContext p) override; 2332 2333 void HandleZeroPosition(const char *startPos, unsigned posLen) override; 2334 2335 void HandleNullChar(const char *nullCharacter) override; 2336 2337 template <typename Range> 2338 static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall, 2339 const Expr *ArgumentExpr, 2340 PartialDiagnostic PDiag, 2341 SourceLocation StringLoc, 2342 bool IsStringLocation, Range StringRange, 2343 ArrayRef<FixItHint> Fixit = None); 2344 2345 protected: 2346 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 2347 const char *startSpec, 2348 unsigned specifierLen, 2349 const char *csStart, unsigned csLen); 2350 2351 void HandlePositionalNonpositionalArgs(SourceLocation Loc, 2352 const char *startSpec, 2353 unsigned specifierLen); 2354 2355 SourceRange getFormatStringRange(); 2356 CharSourceRange getSpecifierRange(const char *startSpecifier, 2357 unsigned specifierLen); 2358 SourceLocation getLocationOfByte(const char *x); 2359 2360 const Expr *getDataArg(unsigned i) const; 2361 2362 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 2363 const analyze_format_string::ConversionSpecifier &CS, 2364 const char *startSpecifier, unsigned specifierLen, 2365 unsigned argIndex); 2366 2367 template <typename Range> 2368 void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc, 2369 bool IsStringLocation, Range StringRange, 2370 ArrayRef<FixItHint> Fixit = None); 2371 }; 2372 } 2373 2374 SourceRange CheckFormatHandler::getFormatStringRange() { 2375 return OrigFormatExpr->getSourceRange(); 2376 } 2377 2378 CharSourceRange CheckFormatHandler:: 2379 getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 2380 SourceLocation Start = getLocationOfByte(startSpecifier); 2381 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 2382 2383 // Advance the end SourceLocation by one due to half-open ranges. 2384 End = End.getLocWithOffset(1); 2385 2386 return CharSourceRange::getCharRange(Start, End); 2387 } 2388 2389 SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 2390 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 2391 } 2392 2393 void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 2394 unsigned specifierLen){ 2395 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier), 2396 getLocationOfByte(startSpecifier), 2397 /*IsStringLocation*/true, 2398 getSpecifierRange(startSpecifier, specifierLen)); 2399 } 2400 2401 void CheckFormatHandler::HandleInvalidLengthModifier( 2402 const analyze_format_string::FormatSpecifier &FS, 2403 const analyze_format_string::ConversionSpecifier &CS, 2404 const char *startSpecifier, unsigned specifierLen, unsigned DiagID) { 2405 using namespace analyze_format_string; 2406 2407 const LengthModifier &LM = FS.getLengthModifier(); 2408 CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength()); 2409 2410 // See if we know how to fix this length modifier. 2411 Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier(); 2412 if (FixedLM) { 2413 EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(), 2414 getLocationOfByte(LM.getStart()), 2415 /*IsStringLocation*/true, 2416 getSpecifierRange(startSpecifier, specifierLen)); 2417 2418 S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier) 2419 << FixedLM->toString() 2420 << FixItHint::CreateReplacement(LMRange, FixedLM->toString()); 2421 2422 } else { 2423 FixItHint Hint; 2424 if (DiagID == diag::warn_format_nonsensical_length) 2425 Hint = FixItHint::CreateRemoval(LMRange); 2426 2427 EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(), 2428 getLocationOfByte(LM.getStart()), 2429 /*IsStringLocation*/true, 2430 getSpecifierRange(startSpecifier, specifierLen), 2431 Hint); 2432 } 2433 } 2434 2435 void CheckFormatHandler::HandleNonStandardLengthModifier( 2436 const analyze_format_string::FormatSpecifier &FS, 2437 const char *startSpecifier, unsigned specifierLen) { 2438 using namespace analyze_format_string; 2439 2440 const LengthModifier &LM = FS.getLengthModifier(); 2441 CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength()); 2442 2443 // See if we know how to fix this length modifier. 2444 Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier(); 2445 if (FixedLM) { 2446 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2447 << LM.toString() << 0, 2448 getLocationOfByte(LM.getStart()), 2449 /*IsStringLocation*/true, 2450 getSpecifierRange(startSpecifier, specifierLen)); 2451 2452 S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier) 2453 << FixedLM->toString() 2454 << FixItHint::CreateReplacement(LMRange, FixedLM->toString()); 2455 2456 } else { 2457 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2458 << LM.toString() << 0, 2459 getLocationOfByte(LM.getStart()), 2460 /*IsStringLocation*/true, 2461 getSpecifierRange(startSpecifier, specifierLen)); 2462 } 2463 } 2464 2465 void CheckFormatHandler::HandleNonStandardConversionSpecifier( 2466 const analyze_format_string::ConversionSpecifier &CS, 2467 const char *startSpecifier, unsigned specifierLen) { 2468 using namespace analyze_format_string; 2469 2470 // See if we know how to fix this conversion specifier. 2471 Optional<ConversionSpecifier> FixedCS = CS.getStandardSpecifier(); 2472 if (FixedCS) { 2473 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2474 << CS.toString() << /*conversion specifier*/1, 2475 getLocationOfByte(CS.getStart()), 2476 /*IsStringLocation*/true, 2477 getSpecifierRange(startSpecifier, specifierLen)); 2478 2479 CharSourceRange CSRange = getSpecifierRange(CS.getStart(), CS.getLength()); 2480 S.Diag(getLocationOfByte(CS.getStart()), diag::note_format_fix_specifier) 2481 << FixedCS->toString() 2482 << FixItHint::CreateReplacement(CSRange, FixedCS->toString()); 2483 } else { 2484 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2485 << CS.toString() << /*conversion specifier*/1, 2486 getLocationOfByte(CS.getStart()), 2487 /*IsStringLocation*/true, 2488 getSpecifierRange(startSpecifier, specifierLen)); 2489 } 2490 } 2491 2492 void CheckFormatHandler::HandlePosition(const char *startPos, 2493 unsigned posLen) { 2494 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_positional_arg), 2495 getLocationOfByte(startPos), 2496 /*IsStringLocation*/true, 2497 getSpecifierRange(startPos, posLen)); 2498 } 2499 2500 void 2501 CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 2502 analyze_format_string::PositionContext p) { 2503 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier) 2504 << (unsigned) p, 2505 getLocationOfByte(startPos), /*IsStringLocation*/true, 2506 getSpecifierRange(startPos, posLen)); 2507 } 2508 2509 void CheckFormatHandler::HandleZeroPosition(const char *startPos, 2510 unsigned posLen) { 2511 EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier), 2512 getLocationOfByte(startPos), 2513 /*IsStringLocation*/true, 2514 getSpecifierRange(startPos, posLen)); 2515 } 2516 2517 void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 2518 if (!isa<ObjCStringLiteral>(OrigFormatExpr)) { 2519 // The presence of a null character is likely an error. 2520 EmitFormatDiagnostic( 2521 S.PDiag(diag::warn_printf_format_string_contains_null_char), 2522 getLocationOfByte(nullCharacter), /*IsStringLocation*/true, 2523 getFormatStringRange()); 2524 } 2525 } 2526 2527 // Note that this may return NULL if there was an error parsing or building 2528 // one of the argument expressions. 2529 const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 2530 return Args[FirstDataArg + i]; 2531 } 2532 2533 void CheckFormatHandler::DoneProcessing() { 2534 // Does the number of data arguments exceed the number of 2535 // format conversions in the format string? 2536 if (!HasVAListArg) { 2537 // Find any arguments that weren't covered. 2538 CoveredArgs.flip(); 2539 signed notCoveredArg = CoveredArgs.find_first(); 2540 if (notCoveredArg >= 0) { 2541 assert((unsigned)notCoveredArg < NumDataArgs); 2542 if (const Expr *E = getDataArg((unsigned) notCoveredArg)) { 2543 SourceLocation Loc = E->getLocStart(); 2544 if (!S.getSourceManager().isInSystemMacro(Loc)) { 2545 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used), 2546 Loc, /*IsStringLocation*/false, 2547 getFormatStringRange()); 2548 } 2549 } 2550 } 2551 } 2552 } 2553 2554 bool 2555 CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 2556 SourceLocation Loc, 2557 const char *startSpec, 2558 unsigned specifierLen, 2559 const char *csStart, 2560 unsigned csLen) { 2561 2562 bool keepGoing = true; 2563 if (argIndex < NumDataArgs) { 2564 // Consider the argument coverered, even though the specifier doesn't 2565 // make sense. 2566 CoveredArgs.set(argIndex); 2567 } 2568 else { 2569 // If argIndex exceeds the number of data arguments we 2570 // don't issue a warning because that is just a cascade of warnings (and 2571 // they may have intended '%%' anyway). We don't want to continue processing 2572 // the format string after this point, however, as we will like just get 2573 // gibberish when trying to match arguments. 2574 keepGoing = false; 2575 } 2576 2577 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion) 2578 << StringRef(csStart, csLen), 2579 Loc, /*IsStringLocation*/true, 2580 getSpecifierRange(startSpec, specifierLen)); 2581 2582 return keepGoing; 2583 } 2584 2585 void 2586 CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc, 2587 const char *startSpec, 2588 unsigned specifierLen) { 2589 EmitFormatDiagnostic( 2590 S.PDiag(diag::warn_format_mix_positional_nonpositional_args), 2591 Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen)); 2592 } 2593 2594 bool 2595 CheckFormatHandler::CheckNumArgs( 2596 const analyze_format_string::FormatSpecifier &FS, 2597 const analyze_format_string::ConversionSpecifier &CS, 2598 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 2599 2600 if (argIndex >= NumDataArgs) { 2601 PartialDiagnostic PDiag = FS.usesPositionalArg() 2602 ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args) 2603 << (argIndex+1) << NumDataArgs) 2604 : S.PDiag(diag::warn_printf_insufficient_data_args); 2605 EmitFormatDiagnostic( 2606 PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true, 2607 getSpecifierRange(startSpecifier, specifierLen)); 2608 return false; 2609 } 2610 return true; 2611 } 2612 2613 template<typename Range> 2614 void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag, 2615 SourceLocation Loc, 2616 bool IsStringLocation, 2617 Range StringRange, 2618 ArrayRef<FixItHint> FixIt) { 2619 EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag, 2620 Loc, IsStringLocation, StringRange, FixIt); 2621 } 2622 2623 /// \brief If the format string is not within the funcion call, emit a note 2624 /// so that the function call and string are in diagnostic messages. 2625 /// 2626 /// \param InFunctionCall if true, the format string is within the function 2627 /// call and only one diagnostic message will be produced. Otherwise, an 2628 /// extra note will be emitted pointing to location of the format string. 2629 /// 2630 /// \param ArgumentExpr the expression that is passed as the format string 2631 /// argument in the function call. Used for getting locations when two 2632 /// diagnostics are emitted. 2633 /// 2634 /// \param PDiag the callee should already have provided any strings for the 2635 /// diagnostic message. This function only adds locations and fixits 2636 /// to diagnostics. 2637 /// 2638 /// \param Loc primary location for diagnostic. If two diagnostics are 2639 /// required, one will be at Loc and a new SourceLocation will be created for 2640 /// the other one. 2641 /// 2642 /// \param IsStringLocation if true, Loc points to the format string should be 2643 /// used for the note. Otherwise, Loc points to the argument list and will 2644 /// be used with PDiag. 2645 /// 2646 /// \param StringRange some or all of the string to highlight. This is 2647 /// templated so it can accept either a CharSourceRange or a SourceRange. 2648 /// 2649 /// \param FixIt optional fix it hint for the format string. 2650 template<typename Range> 2651 void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall, 2652 const Expr *ArgumentExpr, 2653 PartialDiagnostic PDiag, 2654 SourceLocation Loc, 2655 bool IsStringLocation, 2656 Range StringRange, 2657 ArrayRef<FixItHint> FixIt) { 2658 if (InFunctionCall) { 2659 const Sema::SemaDiagnosticBuilder &D = S.Diag(Loc, PDiag); 2660 D << StringRange; 2661 for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end(); 2662 I != E; ++I) { 2663 D << *I; 2664 } 2665 } else { 2666 S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag) 2667 << ArgumentExpr->getSourceRange(); 2668 2669 const Sema::SemaDiagnosticBuilder &Note = 2670 S.Diag(IsStringLocation ? Loc : StringRange.getBegin(), 2671 diag::note_format_string_defined); 2672 2673 Note << StringRange; 2674 for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end(); 2675 I != E; ++I) { 2676 Note << *I; 2677 } 2678 } 2679 } 2680 2681 //===--- CHECK: Printf format string checking ------------------------------===// 2682 2683 namespace { 2684 class CheckPrintfHandler : public CheckFormatHandler { 2685 bool ObjCContext; 2686 public: 2687 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 2688 const Expr *origFormatExpr, unsigned firstDataArg, 2689 unsigned numDataArgs, bool isObjC, 2690 const char *beg, bool hasVAListArg, 2691 ArrayRef<const Expr *> Args, 2692 unsigned formatIdx, bool inFunctionCall, 2693 Sema::VariadicCallType CallType, 2694 llvm::SmallBitVector &CheckedVarArgs) 2695 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2696 numDataArgs, beg, hasVAListArg, Args, 2697 formatIdx, inFunctionCall, CallType, CheckedVarArgs), 2698 ObjCContext(isObjC) 2699 {} 2700 2701 2702 bool HandleInvalidPrintfConversionSpecifier( 2703 const analyze_printf::PrintfSpecifier &FS, 2704 const char *startSpecifier, 2705 unsigned specifierLen) override; 2706 2707 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 2708 const char *startSpecifier, 2709 unsigned specifierLen) override; 2710 bool checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 2711 const char *StartSpecifier, 2712 unsigned SpecifierLen, 2713 const Expr *E); 2714 2715 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 2716 const char *startSpecifier, unsigned specifierLen); 2717 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 2718 const analyze_printf::OptionalAmount &Amt, 2719 unsigned type, 2720 const char *startSpecifier, unsigned specifierLen); 2721 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2722 const analyze_printf::OptionalFlag &flag, 2723 const char *startSpecifier, unsigned specifierLen); 2724 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 2725 const analyze_printf::OptionalFlag &ignoredFlag, 2726 const analyze_printf::OptionalFlag &flag, 2727 const char *startSpecifier, unsigned specifierLen); 2728 bool checkForCStrMembers(const analyze_printf::ArgType &AT, 2729 const Expr *E); 2730 2731 }; 2732 } 2733 2734 bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 2735 const analyze_printf::PrintfSpecifier &FS, 2736 const char *startSpecifier, 2737 unsigned specifierLen) { 2738 const analyze_printf::PrintfConversionSpecifier &CS = 2739 FS.getConversionSpecifier(); 2740 2741 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2742 getLocationOfByte(CS.getStart()), 2743 startSpecifier, specifierLen, 2744 CS.getStart(), CS.getLength()); 2745 } 2746 2747 bool CheckPrintfHandler::HandleAmount( 2748 const analyze_format_string::OptionalAmount &Amt, 2749 unsigned k, const char *startSpecifier, 2750 unsigned specifierLen) { 2751 2752 if (Amt.hasDataArgument()) { 2753 if (!HasVAListArg) { 2754 unsigned argIndex = Amt.getArgIndex(); 2755 if (argIndex >= NumDataArgs) { 2756 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg) 2757 << k, 2758 getLocationOfByte(Amt.getStart()), 2759 /*IsStringLocation*/true, 2760 getSpecifierRange(startSpecifier, specifierLen)); 2761 // Don't do any more checking. We will just emit 2762 // spurious errors. 2763 return false; 2764 } 2765 2766 // Type check the data argument. It should be an 'int'. 2767 // Although not in conformance with C99, we also allow the argument to be 2768 // an 'unsigned int' as that is a reasonably safe case. GCC also 2769 // doesn't emit a warning for that case. 2770 CoveredArgs.set(argIndex); 2771 const Expr *Arg = getDataArg(argIndex); 2772 if (!Arg) 2773 return false; 2774 2775 QualType T = Arg->getType(); 2776 2777 const analyze_printf::ArgType &AT = Amt.getArgType(S.Context); 2778 assert(AT.isValid()); 2779 2780 if (!AT.matchesType(S.Context, T)) { 2781 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type) 2782 << k << AT.getRepresentativeTypeName(S.Context) 2783 << T << Arg->getSourceRange(), 2784 getLocationOfByte(Amt.getStart()), 2785 /*IsStringLocation*/true, 2786 getSpecifierRange(startSpecifier, specifierLen)); 2787 // Don't do any more checking. We will just emit 2788 // spurious errors. 2789 return false; 2790 } 2791 } 2792 } 2793 return true; 2794 } 2795 2796 void CheckPrintfHandler::HandleInvalidAmount( 2797 const analyze_printf::PrintfSpecifier &FS, 2798 const analyze_printf::OptionalAmount &Amt, 2799 unsigned type, 2800 const char *startSpecifier, 2801 unsigned specifierLen) { 2802 const analyze_printf::PrintfConversionSpecifier &CS = 2803 FS.getConversionSpecifier(); 2804 2805 FixItHint fixit = 2806 Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant 2807 ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 2808 Amt.getConstantLength())) 2809 : FixItHint(); 2810 2811 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount) 2812 << type << CS.toString(), 2813 getLocationOfByte(Amt.getStart()), 2814 /*IsStringLocation*/true, 2815 getSpecifierRange(startSpecifier, specifierLen), 2816 fixit); 2817 } 2818 2819 void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2820 const analyze_printf::OptionalFlag &flag, 2821 const char *startSpecifier, 2822 unsigned specifierLen) { 2823 // Warn about pointless flag with a fixit removal. 2824 const analyze_printf::PrintfConversionSpecifier &CS = 2825 FS.getConversionSpecifier(); 2826 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag) 2827 << flag.toString() << CS.toString(), 2828 getLocationOfByte(flag.getPosition()), 2829 /*IsStringLocation*/true, 2830 getSpecifierRange(startSpecifier, specifierLen), 2831 FixItHint::CreateRemoval( 2832 getSpecifierRange(flag.getPosition(), 1))); 2833 } 2834 2835 void CheckPrintfHandler::HandleIgnoredFlag( 2836 const analyze_printf::PrintfSpecifier &FS, 2837 const analyze_printf::OptionalFlag &ignoredFlag, 2838 const analyze_printf::OptionalFlag &flag, 2839 const char *startSpecifier, 2840 unsigned specifierLen) { 2841 // Warn about ignored flag with a fixit removal. 2842 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag) 2843 << ignoredFlag.toString() << flag.toString(), 2844 getLocationOfByte(ignoredFlag.getPosition()), 2845 /*IsStringLocation*/true, 2846 getSpecifierRange(startSpecifier, specifierLen), 2847 FixItHint::CreateRemoval( 2848 getSpecifierRange(ignoredFlag.getPosition(), 1))); 2849 } 2850 2851 // Determines if the specified is a C++ class or struct containing 2852 // a member with the specified name and kind (e.g. a CXXMethodDecl named 2853 // "c_str()"). 2854 template<typename MemberKind> 2855 static llvm::SmallPtrSet<MemberKind*, 1> 2856 CXXRecordMembersNamed(StringRef Name, Sema &S, QualType Ty) { 2857 const RecordType *RT = Ty->getAs<RecordType>(); 2858 llvm::SmallPtrSet<MemberKind*, 1> Results; 2859 2860 if (!RT) 2861 return Results; 2862 const CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(RT->getDecl()); 2863 if (!RD || !RD->getDefinition()) 2864 return Results; 2865 2866 LookupResult R(S, &S.Context.Idents.get(Name), SourceLocation(), 2867 Sema::LookupMemberName); 2868 R.suppressDiagnostics(); 2869 2870 // We just need to include all members of the right kind turned up by the 2871 // filter, at this point. 2872 if (S.LookupQualifiedName(R, RT->getDecl())) 2873 for (LookupResult::iterator I = R.begin(), E = R.end(); I != E; ++I) { 2874 NamedDecl *decl = (*I)->getUnderlyingDecl(); 2875 if (MemberKind *FK = dyn_cast<MemberKind>(decl)) 2876 Results.insert(FK); 2877 } 2878 return Results; 2879 } 2880 2881 /// Check if we could call '.c_str()' on an object. 2882 /// 2883 /// FIXME: This returns the wrong results in some cases (if cv-qualifiers don't 2884 /// allow the call, or if it would be ambiguous). 2885 bool Sema::hasCStrMethod(const Expr *E) { 2886 typedef llvm::SmallPtrSet<CXXMethodDecl*, 1> MethodSet; 2887 MethodSet Results = 2888 CXXRecordMembersNamed<CXXMethodDecl>("c_str", *this, E->getType()); 2889 for (MethodSet::iterator MI = Results.begin(), ME = Results.end(); 2890 MI != ME; ++MI) 2891 if ((*MI)->getMinRequiredArguments() == 0) 2892 return true; 2893 return false; 2894 } 2895 2896 // Check if a (w)string was passed when a (w)char* was needed, and offer a 2897 // better diagnostic if so. AT is assumed to be valid. 2898 // Returns true when a c_str() conversion method is found. 2899 bool CheckPrintfHandler::checkForCStrMembers( 2900 const analyze_printf::ArgType &AT, const Expr *E) { 2901 typedef llvm::SmallPtrSet<CXXMethodDecl*, 1> MethodSet; 2902 2903 MethodSet Results = 2904 CXXRecordMembersNamed<CXXMethodDecl>("c_str", S, E->getType()); 2905 2906 for (MethodSet::iterator MI = Results.begin(), ME = Results.end(); 2907 MI != ME; ++MI) { 2908 const CXXMethodDecl *Method = *MI; 2909 if (Method->getMinRequiredArguments() == 0 && 2910 AT.matchesType(S.Context, Method->getReturnType())) { 2911 // FIXME: Suggest parens if the expression needs them. 2912 SourceLocation EndLoc = S.getLocForEndOfToken(E->getLocEnd()); 2913 S.Diag(E->getLocStart(), diag::note_printf_c_str) 2914 << "c_str()" 2915 << FixItHint::CreateInsertion(EndLoc, ".c_str()"); 2916 return true; 2917 } 2918 } 2919 2920 return false; 2921 } 2922 2923 bool 2924 CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 2925 &FS, 2926 const char *startSpecifier, 2927 unsigned specifierLen) { 2928 2929 using namespace analyze_format_string; 2930 using namespace analyze_printf; 2931 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 2932 2933 if (FS.consumesDataArgument()) { 2934 if (atFirstArg) { 2935 atFirstArg = false; 2936 usesPositionalArgs = FS.usesPositionalArg(); 2937 } 2938 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2939 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2940 startSpecifier, specifierLen); 2941 return false; 2942 } 2943 } 2944 2945 // First check if the field width, precision, and conversion specifier 2946 // have matching data arguments. 2947 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 2948 startSpecifier, specifierLen)) { 2949 return false; 2950 } 2951 2952 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 2953 startSpecifier, specifierLen)) { 2954 return false; 2955 } 2956 2957 if (!CS.consumesDataArgument()) { 2958 // FIXME: Technically specifying a precision or field width here 2959 // makes no sense. Worth issuing a warning at some point. 2960 return true; 2961 } 2962 2963 // Consume the argument. 2964 unsigned argIndex = FS.getArgIndex(); 2965 if (argIndex < NumDataArgs) { 2966 // The check to see if the argIndex is valid will come later. 2967 // We set the bit here because we may exit early from this 2968 // function if we encounter some other error. 2969 CoveredArgs.set(argIndex); 2970 } 2971 2972 // Check for using an Objective-C specific conversion specifier 2973 // in a non-ObjC literal. 2974 if (!ObjCContext && CS.isObjCArg()) { 2975 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 2976 specifierLen); 2977 } 2978 2979 // Check for invalid use of field width 2980 if (!FS.hasValidFieldWidth()) { 2981 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 2982 startSpecifier, specifierLen); 2983 } 2984 2985 // Check for invalid use of precision 2986 if (!FS.hasValidPrecision()) { 2987 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 2988 startSpecifier, specifierLen); 2989 } 2990 2991 // Check each flag does not conflict with any other component. 2992 if (!FS.hasValidThousandsGroupingPrefix()) 2993 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 2994 if (!FS.hasValidLeadingZeros()) 2995 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 2996 if (!FS.hasValidPlusPrefix()) 2997 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 2998 if (!FS.hasValidSpacePrefix()) 2999 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 3000 if (!FS.hasValidAlternativeForm()) 3001 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 3002 if (!FS.hasValidLeftJustified()) 3003 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 3004 3005 // Check that flags are not ignored by another flag 3006 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 3007 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 3008 startSpecifier, specifierLen); 3009 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 3010 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 3011 startSpecifier, specifierLen); 3012 3013 // Check the length modifier is valid with the given conversion specifier. 3014 if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo())) 3015 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3016 diag::warn_format_nonsensical_length); 3017 else if (!FS.hasStandardLengthModifier()) 3018 HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen); 3019 else if (!FS.hasStandardLengthConversionCombination()) 3020 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3021 diag::warn_format_non_standard_conversion_spec); 3022 3023 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 3024 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 3025 3026 // The remaining checks depend on the data arguments. 3027 if (HasVAListArg) 3028 return true; 3029 3030 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 3031 return false; 3032 3033 const Expr *Arg = getDataArg(argIndex); 3034 if (!Arg) 3035 return true; 3036 3037 return checkFormatExpr(FS, startSpecifier, specifierLen, Arg); 3038 } 3039 3040 static bool requiresParensToAddCast(const Expr *E) { 3041 // FIXME: We should have a general way to reason about operator 3042 // precedence and whether parens are actually needed here. 3043 // Take care of a few common cases where they aren't. 3044 const Expr *Inside = E->IgnoreImpCasts(); 3045 if (const PseudoObjectExpr *POE = dyn_cast<PseudoObjectExpr>(Inside)) 3046 Inside = POE->getSyntacticForm()->IgnoreImpCasts(); 3047 3048 switch (Inside->getStmtClass()) { 3049 case Stmt::ArraySubscriptExprClass: 3050 case Stmt::CallExprClass: 3051 case Stmt::CharacterLiteralClass: 3052 case Stmt::CXXBoolLiteralExprClass: 3053 case Stmt::DeclRefExprClass: 3054 case Stmt::FloatingLiteralClass: 3055 case Stmt::IntegerLiteralClass: 3056 case Stmt::MemberExprClass: 3057 case Stmt::ObjCArrayLiteralClass: 3058 case Stmt::ObjCBoolLiteralExprClass: 3059 case Stmt::ObjCBoxedExprClass: 3060 case Stmt::ObjCDictionaryLiteralClass: 3061 case Stmt::ObjCEncodeExprClass: 3062 case Stmt::ObjCIvarRefExprClass: 3063 case Stmt::ObjCMessageExprClass: 3064 case Stmt::ObjCPropertyRefExprClass: 3065 case Stmt::ObjCStringLiteralClass: 3066 case Stmt::ObjCSubscriptRefExprClass: 3067 case Stmt::ParenExprClass: 3068 case Stmt::StringLiteralClass: 3069 case Stmt::UnaryOperatorClass: 3070 return false; 3071 default: 3072 return true; 3073 } 3074 } 3075 3076 bool 3077 CheckPrintfHandler::checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 3078 const char *StartSpecifier, 3079 unsigned SpecifierLen, 3080 const Expr *E) { 3081 using namespace analyze_format_string; 3082 using namespace analyze_printf; 3083 // Now type check the data expression that matches the 3084 // format specifier. 3085 const analyze_printf::ArgType &AT = FS.getArgType(S.Context, 3086 ObjCContext); 3087 if (!AT.isValid()) 3088 return true; 3089 3090 QualType ExprTy = E->getType(); 3091 while (const TypeOfExprType *TET = dyn_cast<TypeOfExprType>(ExprTy)) { 3092 ExprTy = TET->getUnderlyingExpr()->getType(); 3093 } 3094 3095 if (AT.matchesType(S.Context, ExprTy)) 3096 return true; 3097 3098 // Look through argument promotions for our error message's reported type. 3099 // This includes the integral and floating promotions, but excludes array 3100 // and function pointer decay; seeing that an argument intended to be a 3101 // string has type 'char [6]' is probably more confusing than 'char *'. 3102 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 3103 if (ICE->getCastKind() == CK_IntegralCast || 3104 ICE->getCastKind() == CK_FloatingCast) { 3105 E = ICE->getSubExpr(); 3106 ExprTy = E->getType(); 3107 3108 // Check if we didn't match because of an implicit cast from a 'char' 3109 // or 'short' to an 'int'. This is done because printf is a varargs 3110 // function. 3111 if (ICE->getType() == S.Context.IntTy || 3112 ICE->getType() == S.Context.UnsignedIntTy) { 3113 // All further checking is done on the subexpression. 3114 if (AT.matchesType(S.Context, ExprTy)) 3115 return true; 3116 } 3117 } 3118 } else if (const CharacterLiteral *CL = dyn_cast<CharacterLiteral>(E)) { 3119 // Special case for 'a', which has type 'int' in C. 3120 // Note, however, that we do /not/ want to treat multibyte constants like 3121 // 'MooV' as characters! This form is deprecated but still exists. 3122 if (ExprTy == S.Context.IntTy) 3123 if (llvm::isUIntN(S.Context.getCharWidth(), CL->getValue())) 3124 ExprTy = S.Context.CharTy; 3125 } 3126 3127 // Look through enums to their underlying type. 3128 bool IsEnum = false; 3129 if (auto EnumTy = ExprTy->getAs<EnumType>()) { 3130 ExprTy = EnumTy->getDecl()->getIntegerType(); 3131 IsEnum = true; 3132 } 3133 3134 // %C in an Objective-C context prints a unichar, not a wchar_t. 3135 // If the argument is an integer of some kind, believe the %C and suggest 3136 // a cast instead of changing the conversion specifier. 3137 QualType IntendedTy = ExprTy; 3138 if (ObjCContext && 3139 FS.getConversionSpecifier().getKind() == ConversionSpecifier::CArg) { 3140 if (ExprTy->isIntegralOrUnscopedEnumerationType() && 3141 !ExprTy->isCharType()) { 3142 // 'unichar' is defined as a typedef of unsigned short, but we should 3143 // prefer using the typedef if it is visible. 3144 IntendedTy = S.Context.UnsignedShortTy; 3145 3146 // While we are here, check if the value is an IntegerLiteral that happens 3147 // to be within the valid range. 3148 if (const IntegerLiteral *IL = dyn_cast<IntegerLiteral>(E)) { 3149 const llvm::APInt &V = IL->getValue(); 3150 if (V.getActiveBits() <= S.Context.getTypeSize(IntendedTy)) 3151 return true; 3152 } 3153 3154 LookupResult Result(S, &S.Context.Idents.get("unichar"), E->getLocStart(), 3155 Sema::LookupOrdinaryName); 3156 if (S.LookupName(Result, S.getCurScope())) { 3157 NamedDecl *ND = Result.getFoundDecl(); 3158 if (TypedefNameDecl *TD = dyn_cast<TypedefNameDecl>(ND)) 3159 if (TD->getUnderlyingType() == IntendedTy) 3160 IntendedTy = S.Context.getTypedefType(TD); 3161 } 3162 } 3163 } 3164 3165 // Special-case some of Darwin's platform-independence types by suggesting 3166 // casts to primitive types that are known to be large enough. 3167 bool ShouldNotPrintDirectly = false; 3168 if (S.Context.getTargetInfo().getTriple().isOSDarwin()) { 3169 // Use a 'while' to peel off layers of typedefs. 3170 QualType TyTy = IntendedTy; 3171 while (const TypedefType *UserTy = TyTy->getAs<TypedefType>()) { 3172 StringRef Name = UserTy->getDecl()->getName(); 3173 QualType CastTy = llvm::StringSwitch<QualType>(Name) 3174 .Case("NSInteger", S.Context.LongTy) 3175 .Case("NSUInteger", S.Context.UnsignedLongTy) 3176 .Case("SInt32", S.Context.IntTy) 3177 .Case("UInt32", S.Context.UnsignedIntTy) 3178 .Default(QualType()); 3179 3180 if (!CastTy.isNull()) { 3181 ShouldNotPrintDirectly = true; 3182 IntendedTy = CastTy; 3183 break; 3184 } 3185 TyTy = UserTy->desugar(); 3186 } 3187 } 3188 3189 // We may be able to offer a FixItHint if it is a supported type. 3190 PrintfSpecifier fixedFS = FS; 3191 bool success = fixedFS.fixType(IntendedTy, S.getLangOpts(), 3192 S.Context, ObjCContext); 3193 3194 if (success) { 3195 // Get the fix string from the fixed format specifier 3196 SmallString<16> buf; 3197 llvm::raw_svector_ostream os(buf); 3198 fixedFS.toString(os); 3199 3200 CharSourceRange SpecRange = getSpecifierRange(StartSpecifier, SpecifierLen); 3201 3202 if (IntendedTy == ExprTy) { 3203 // In this case, the specifier is wrong and should be changed to match 3204 // the argument. 3205 EmitFormatDiagnostic( 3206 S.PDiag(diag::warn_format_conversion_argument_type_mismatch) 3207 << AT.getRepresentativeTypeName(S.Context) << IntendedTy << IsEnum 3208 << E->getSourceRange(), 3209 E->getLocStart(), 3210 /*IsStringLocation*/false, 3211 SpecRange, 3212 FixItHint::CreateReplacement(SpecRange, os.str())); 3213 3214 } else { 3215 // The canonical type for formatting this value is different from the 3216 // actual type of the expression. (This occurs, for example, with Darwin's 3217 // NSInteger on 32-bit platforms, where it is typedef'd as 'int', but 3218 // should be printed as 'long' for 64-bit compatibility.) 3219 // Rather than emitting a normal format/argument mismatch, we want to 3220 // add a cast to the recommended type (and correct the format string 3221 // if necessary). 3222 SmallString<16> CastBuf; 3223 llvm::raw_svector_ostream CastFix(CastBuf); 3224 CastFix << "("; 3225 IntendedTy.print(CastFix, S.Context.getPrintingPolicy()); 3226 CastFix << ")"; 3227 3228 SmallVector<FixItHint,4> Hints; 3229 if (!AT.matchesType(S.Context, IntendedTy)) 3230 Hints.push_back(FixItHint::CreateReplacement(SpecRange, os.str())); 3231 3232 if (const CStyleCastExpr *CCast = dyn_cast<CStyleCastExpr>(E)) { 3233 // If there's already a cast present, just replace it. 3234 SourceRange CastRange(CCast->getLParenLoc(), CCast->getRParenLoc()); 3235 Hints.push_back(FixItHint::CreateReplacement(CastRange, CastFix.str())); 3236 3237 } else if (!requiresParensToAddCast(E)) { 3238 // If the expression has high enough precedence, 3239 // just write the C-style cast. 3240 Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(), 3241 CastFix.str())); 3242 } else { 3243 // Otherwise, add parens around the expression as well as the cast. 3244 CastFix << "("; 3245 Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(), 3246 CastFix.str())); 3247 3248 SourceLocation After = S.getLocForEndOfToken(E->getLocEnd()); 3249 Hints.push_back(FixItHint::CreateInsertion(After, ")")); 3250 } 3251 3252 if (ShouldNotPrintDirectly) { 3253 // The expression has a type that should not be printed directly. 3254 // We extract the name from the typedef because we don't want to show 3255 // the underlying type in the diagnostic. 3256 StringRef Name = cast<TypedefType>(ExprTy)->getDecl()->getName(); 3257 3258 EmitFormatDiagnostic(S.PDiag(diag::warn_format_argument_needs_cast) 3259 << Name << IntendedTy << IsEnum 3260 << E->getSourceRange(), 3261 E->getLocStart(), /*IsStringLocation=*/false, 3262 SpecRange, Hints); 3263 } else { 3264 // In this case, the expression could be printed using a different 3265 // specifier, but we've decided that the specifier is probably correct 3266 // and we should cast instead. Just use the normal warning message. 3267 EmitFormatDiagnostic( 3268 S.PDiag(diag::warn_format_conversion_argument_type_mismatch) 3269 << AT.getRepresentativeTypeName(S.Context) << ExprTy << IsEnum 3270 << E->getSourceRange(), 3271 E->getLocStart(), /*IsStringLocation*/false, 3272 SpecRange, Hints); 3273 } 3274 } 3275 } else { 3276 const CharSourceRange &CSR = getSpecifierRange(StartSpecifier, 3277 SpecifierLen); 3278 // Since the warning for passing non-POD types to variadic functions 3279 // was deferred until now, we emit a warning for non-POD 3280 // arguments here. 3281 switch (S.isValidVarArgType(ExprTy)) { 3282 case Sema::VAK_Valid: 3283 case Sema::VAK_ValidInCXX11: 3284 EmitFormatDiagnostic( 3285 S.PDiag(diag::warn_format_conversion_argument_type_mismatch) 3286 << AT.getRepresentativeTypeName(S.Context) << ExprTy << IsEnum 3287 << CSR 3288 << E->getSourceRange(), 3289 E->getLocStart(), /*IsStringLocation*/false, CSR); 3290 break; 3291 3292 case Sema::VAK_Undefined: 3293 EmitFormatDiagnostic( 3294 S.PDiag(diag::warn_non_pod_vararg_with_format_string) 3295 << S.getLangOpts().CPlusPlus11 3296 << ExprTy 3297 << CallType 3298 << AT.getRepresentativeTypeName(S.Context) 3299 << CSR 3300 << E->getSourceRange(), 3301 E->getLocStart(), /*IsStringLocation*/false, CSR); 3302 checkForCStrMembers(AT, E); 3303 break; 3304 3305 case Sema::VAK_Invalid: 3306 if (ExprTy->isObjCObjectType()) 3307 EmitFormatDiagnostic( 3308 S.PDiag(diag::err_cannot_pass_objc_interface_to_vararg_format) 3309 << S.getLangOpts().CPlusPlus11 3310 << ExprTy 3311 << CallType 3312 << AT.getRepresentativeTypeName(S.Context) 3313 << CSR 3314 << E->getSourceRange(), 3315 E->getLocStart(), /*IsStringLocation*/false, CSR); 3316 else 3317 // FIXME: If this is an initializer list, suggest removing the braces 3318 // or inserting a cast to the target type. 3319 S.Diag(E->getLocStart(), diag::err_cannot_pass_to_vararg_format) 3320 << isa<InitListExpr>(E) << ExprTy << CallType 3321 << AT.getRepresentativeTypeName(S.Context) 3322 << E->getSourceRange(); 3323 break; 3324 } 3325 3326 assert(FirstDataArg + FS.getArgIndex() < CheckedVarArgs.size() && 3327 "format string specifier index out of range"); 3328 CheckedVarArgs[FirstDataArg + FS.getArgIndex()] = true; 3329 } 3330 3331 return true; 3332 } 3333 3334 //===--- CHECK: Scanf format string checking ------------------------------===// 3335 3336 namespace { 3337 class CheckScanfHandler : public CheckFormatHandler { 3338 public: 3339 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 3340 const Expr *origFormatExpr, unsigned firstDataArg, 3341 unsigned numDataArgs, const char *beg, bool hasVAListArg, 3342 ArrayRef<const Expr *> Args, 3343 unsigned formatIdx, bool inFunctionCall, 3344 Sema::VariadicCallType CallType, 3345 llvm::SmallBitVector &CheckedVarArgs) 3346 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 3347 numDataArgs, beg, hasVAListArg, 3348 Args, formatIdx, inFunctionCall, CallType, 3349 CheckedVarArgs) 3350 {} 3351 3352 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 3353 const char *startSpecifier, 3354 unsigned specifierLen) override; 3355 3356 bool HandleInvalidScanfConversionSpecifier( 3357 const analyze_scanf::ScanfSpecifier &FS, 3358 const char *startSpecifier, 3359 unsigned specifierLen) override; 3360 3361 void HandleIncompleteScanList(const char *start, const char *end) override; 3362 }; 3363 } 3364 3365 void CheckScanfHandler::HandleIncompleteScanList(const char *start, 3366 const char *end) { 3367 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 3368 getLocationOfByte(end), /*IsStringLocation*/true, 3369 getSpecifierRange(start, end - start)); 3370 } 3371 3372 bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 3373 const analyze_scanf::ScanfSpecifier &FS, 3374 const char *startSpecifier, 3375 unsigned specifierLen) { 3376 3377 const analyze_scanf::ScanfConversionSpecifier &CS = 3378 FS.getConversionSpecifier(); 3379 3380 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 3381 getLocationOfByte(CS.getStart()), 3382 startSpecifier, specifierLen, 3383 CS.getStart(), CS.getLength()); 3384 } 3385 3386 bool CheckScanfHandler::HandleScanfSpecifier( 3387 const analyze_scanf::ScanfSpecifier &FS, 3388 const char *startSpecifier, 3389 unsigned specifierLen) { 3390 3391 using namespace analyze_scanf; 3392 using namespace analyze_format_string; 3393 3394 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 3395 3396 // Handle case where '%' and '*' don't consume an argument. These shouldn't 3397 // be used to decide if we are using positional arguments consistently. 3398 if (FS.consumesDataArgument()) { 3399 if (atFirstArg) { 3400 atFirstArg = false; 3401 usesPositionalArgs = FS.usesPositionalArg(); 3402 } 3403 else if (usesPositionalArgs != FS.usesPositionalArg()) { 3404 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 3405 startSpecifier, specifierLen); 3406 return false; 3407 } 3408 } 3409 3410 // Check if the field with is non-zero. 3411 const OptionalAmount &Amt = FS.getFieldWidth(); 3412 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 3413 if (Amt.getConstantAmount() == 0) { 3414 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 3415 Amt.getConstantLength()); 3416 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 3417 getLocationOfByte(Amt.getStart()), 3418 /*IsStringLocation*/true, R, 3419 FixItHint::CreateRemoval(R)); 3420 } 3421 } 3422 3423 if (!FS.consumesDataArgument()) { 3424 // FIXME: Technically specifying a precision or field width here 3425 // makes no sense. Worth issuing a warning at some point. 3426 return true; 3427 } 3428 3429 // Consume the argument. 3430 unsigned argIndex = FS.getArgIndex(); 3431 if (argIndex < NumDataArgs) { 3432 // The check to see if the argIndex is valid will come later. 3433 // We set the bit here because we may exit early from this 3434 // function if we encounter some other error. 3435 CoveredArgs.set(argIndex); 3436 } 3437 3438 // Check the length modifier is valid with the given conversion specifier. 3439 if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo())) 3440 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3441 diag::warn_format_nonsensical_length); 3442 else if (!FS.hasStandardLengthModifier()) 3443 HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen); 3444 else if (!FS.hasStandardLengthConversionCombination()) 3445 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3446 diag::warn_format_non_standard_conversion_spec); 3447 3448 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 3449 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 3450 3451 // The remaining checks depend on the data arguments. 3452 if (HasVAListArg) 3453 return true; 3454 3455 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 3456 return false; 3457 3458 // Check that the argument type matches the format specifier. 3459 const Expr *Ex = getDataArg(argIndex); 3460 if (!Ex) 3461 return true; 3462 3463 const analyze_format_string::ArgType &AT = FS.getArgType(S.Context); 3464 if (AT.isValid() && !AT.matchesType(S.Context, Ex->getType())) { 3465 ScanfSpecifier fixedFS = FS; 3466 bool success = fixedFS.fixType(Ex->getType(), 3467 Ex->IgnoreImpCasts()->getType(), 3468 S.getLangOpts(), S.Context); 3469 3470 if (success) { 3471 // Get the fix string from the fixed format specifier. 3472 SmallString<128> buf; 3473 llvm::raw_svector_ostream os(buf); 3474 fixedFS.toString(os); 3475 3476 EmitFormatDiagnostic( 3477 S.PDiag(diag::warn_format_conversion_argument_type_mismatch) 3478 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() << false 3479 << Ex->getSourceRange(), 3480 Ex->getLocStart(), 3481 /*IsStringLocation*/false, 3482 getSpecifierRange(startSpecifier, specifierLen), 3483 FixItHint::CreateReplacement( 3484 getSpecifierRange(startSpecifier, specifierLen), 3485 os.str())); 3486 } else { 3487 EmitFormatDiagnostic( 3488 S.PDiag(diag::warn_format_conversion_argument_type_mismatch) 3489 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() << false 3490 << Ex->getSourceRange(), 3491 Ex->getLocStart(), 3492 /*IsStringLocation*/false, 3493 getSpecifierRange(startSpecifier, specifierLen)); 3494 } 3495 } 3496 3497 return true; 3498 } 3499 3500 void Sema::CheckFormatString(const StringLiteral *FExpr, 3501 const Expr *OrigFormatExpr, 3502 ArrayRef<const Expr *> Args, 3503 bool HasVAListArg, unsigned format_idx, 3504 unsigned firstDataArg, FormatStringType Type, 3505 bool inFunctionCall, VariadicCallType CallType, 3506 llvm::SmallBitVector &CheckedVarArgs) { 3507 3508 // CHECK: is the format string a wide literal? 3509 if (!FExpr->isAscii() && !FExpr->isUTF8()) { 3510 CheckFormatHandler::EmitFormatDiagnostic( 3511 *this, inFunctionCall, Args[format_idx], 3512 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 3513 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 3514 return; 3515 } 3516 3517 // Str - The format string. NOTE: this is NOT null-terminated! 3518 StringRef StrRef = FExpr->getString(); 3519 const char *Str = StrRef.data(); 3520 // Account for cases where the string literal is truncated in a declaration. 3521 const ConstantArrayType *T = Context.getAsConstantArrayType(FExpr->getType()); 3522 assert(T && "String literal not of constant array type!"); 3523 size_t TypeSize = T->getSize().getZExtValue(); 3524 size_t StrLen = std::min(std::max(TypeSize, size_t(1)) - 1, StrRef.size()); 3525 const unsigned numDataArgs = Args.size() - firstDataArg; 3526 3527 // Emit a warning if the string literal is truncated and does not contain an 3528 // embedded null character. 3529 if (TypeSize <= StrRef.size() && 3530 StrRef.substr(0, TypeSize).find('\0') == StringRef::npos) { 3531 CheckFormatHandler::EmitFormatDiagnostic( 3532 *this, inFunctionCall, Args[format_idx], 3533 PDiag(diag::warn_printf_format_string_not_null_terminated), 3534 FExpr->getLocStart(), 3535 /*IsStringLocation=*/true, OrigFormatExpr->getSourceRange()); 3536 return; 3537 } 3538 3539 // CHECK: empty format string? 3540 if (StrLen == 0 && numDataArgs > 0) { 3541 CheckFormatHandler::EmitFormatDiagnostic( 3542 *this, inFunctionCall, Args[format_idx], 3543 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 3544 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 3545 return; 3546 } 3547 3548 if (Type == FST_Printf || Type == FST_NSString) { 3549 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 3550 numDataArgs, (Type == FST_NSString), 3551 Str, HasVAListArg, Args, format_idx, 3552 inFunctionCall, CallType, CheckedVarArgs); 3553 3554 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen, 3555 getLangOpts(), 3556 Context.getTargetInfo())) 3557 H.DoneProcessing(); 3558 } else if (Type == FST_Scanf) { 3559 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, numDataArgs, 3560 Str, HasVAListArg, Args, format_idx, 3561 inFunctionCall, CallType, CheckedVarArgs); 3562 3563 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen, 3564 getLangOpts(), 3565 Context.getTargetInfo())) 3566 H.DoneProcessing(); 3567 } // TODO: handle other formats 3568 } 3569 3570 //===--- CHECK: Warn on use of wrong absolute value function. -------------===// 3571 3572 // Returns the related absolute value function that is larger, of 0 if one 3573 // does not exist. 3574 static unsigned getLargerAbsoluteValueFunction(unsigned AbsFunction) { 3575 switch (AbsFunction) { 3576 default: 3577 return 0; 3578 3579 case Builtin::BI__builtin_abs: 3580 return Builtin::BI__builtin_labs; 3581 case Builtin::BI__builtin_labs: 3582 return Builtin::BI__builtin_llabs; 3583 case Builtin::BI__builtin_llabs: 3584 return 0; 3585 3586 case Builtin::BI__builtin_fabsf: 3587 return Builtin::BI__builtin_fabs; 3588 case Builtin::BI__builtin_fabs: 3589 return Builtin::BI__builtin_fabsl; 3590 case Builtin::BI__builtin_fabsl: 3591 return 0; 3592 3593 case Builtin::BI__builtin_cabsf: 3594 return Builtin::BI__builtin_cabs; 3595 case Builtin::BI__builtin_cabs: 3596 return Builtin::BI__builtin_cabsl; 3597 case Builtin::BI__builtin_cabsl: 3598 return 0; 3599 3600 case Builtin::BIabs: 3601 return Builtin::BIlabs; 3602 case Builtin::BIlabs: 3603 return Builtin::BIllabs; 3604 case Builtin::BIllabs: 3605 return 0; 3606 3607 case Builtin::BIfabsf: 3608 return Builtin::BIfabs; 3609 case Builtin::BIfabs: 3610 return Builtin::BIfabsl; 3611 case Builtin::BIfabsl: 3612 return 0; 3613 3614 case Builtin::BIcabsf: 3615 return Builtin::BIcabs; 3616 case Builtin::BIcabs: 3617 return Builtin::BIcabsl; 3618 case Builtin::BIcabsl: 3619 return 0; 3620 } 3621 } 3622 3623 // Returns the argument type of the absolute value function. 3624 static QualType getAbsoluteValueArgumentType(ASTContext &Context, 3625 unsigned AbsType) { 3626 if (AbsType == 0) 3627 return QualType(); 3628 3629 ASTContext::GetBuiltinTypeError Error = ASTContext::GE_None; 3630 QualType BuiltinType = Context.GetBuiltinType(AbsType, Error); 3631 if (Error != ASTContext::GE_None) 3632 return QualType(); 3633 3634 const FunctionProtoType *FT = BuiltinType->getAs<FunctionProtoType>(); 3635 if (!FT) 3636 return QualType(); 3637 3638 if (FT->getNumParams() != 1) 3639 return QualType(); 3640 3641 return FT->getParamType(0); 3642 } 3643 3644 // Returns the best absolute value function, or zero, based on type and 3645 // current absolute value function. 3646 static unsigned getBestAbsFunction(ASTContext &Context, QualType ArgType, 3647 unsigned AbsFunctionKind) { 3648 unsigned BestKind = 0; 3649 uint64_t ArgSize = Context.getTypeSize(ArgType); 3650 for (unsigned Kind = AbsFunctionKind; Kind != 0; 3651 Kind = getLargerAbsoluteValueFunction(Kind)) { 3652 QualType ParamType = getAbsoluteValueArgumentType(Context, Kind); 3653 if (Context.getTypeSize(ParamType) >= ArgSize) { 3654 if (BestKind == 0) 3655 BestKind = Kind; 3656 else if (Context.hasSameType(ParamType, ArgType)) { 3657 BestKind = Kind; 3658 break; 3659 } 3660 } 3661 } 3662 return BestKind; 3663 } 3664 3665 enum AbsoluteValueKind { 3666 AVK_Integer, 3667 AVK_Floating, 3668 AVK_Complex 3669 }; 3670 3671 static AbsoluteValueKind getAbsoluteValueKind(QualType T) { 3672 if (T->isIntegralOrEnumerationType()) 3673 return AVK_Integer; 3674 if (T->isRealFloatingType()) 3675 return AVK_Floating; 3676 if (T->isAnyComplexType()) 3677 return AVK_Complex; 3678 3679 llvm_unreachable("Type not integer, floating, or complex"); 3680 } 3681 3682 // Changes the absolute value function to a different type. Preserves whether 3683 // the function is a builtin. 3684 static unsigned changeAbsFunction(unsigned AbsKind, 3685 AbsoluteValueKind ValueKind) { 3686 switch (ValueKind) { 3687 case AVK_Integer: 3688 switch (AbsKind) { 3689 default: 3690 return 0; 3691 case Builtin::BI__builtin_fabsf: 3692 case Builtin::BI__builtin_fabs: 3693 case Builtin::BI__builtin_fabsl: 3694 case Builtin::BI__builtin_cabsf: 3695 case Builtin::BI__builtin_cabs: 3696 case Builtin::BI__builtin_cabsl: 3697 return Builtin::BI__builtin_abs; 3698 case Builtin::BIfabsf: 3699 case Builtin::BIfabs: 3700 case Builtin::BIfabsl: 3701 case Builtin::BIcabsf: 3702 case Builtin::BIcabs: 3703 case Builtin::BIcabsl: 3704 return Builtin::BIabs; 3705 } 3706 case AVK_Floating: 3707 switch (AbsKind) { 3708 default: 3709 return 0; 3710 case Builtin::BI__builtin_abs: 3711 case Builtin::BI__builtin_labs: 3712 case Builtin::BI__builtin_llabs: 3713 case Builtin::BI__builtin_cabsf: 3714 case Builtin::BI__builtin_cabs: 3715 case Builtin::BI__builtin_cabsl: 3716 return Builtin::BI__builtin_fabsf; 3717 case Builtin::BIabs: 3718 case Builtin::BIlabs: 3719 case Builtin::BIllabs: 3720 case Builtin::BIcabsf: 3721 case Builtin::BIcabs: 3722 case Builtin::BIcabsl: 3723 return Builtin::BIfabsf; 3724 } 3725 case AVK_Complex: 3726 switch (AbsKind) { 3727 default: 3728 return 0; 3729 case Builtin::BI__builtin_abs: 3730 case Builtin::BI__builtin_labs: 3731 case Builtin::BI__builtin_llabs: 3732 case Builtin::BI__builtin_fabsf: 3733 case Builtin::BI__builtin_fabs: 3734 case Builtin::BI__builtin_fabsl: 3735 return Builtin::BI__builtin_cabsf; 3736 case Builtin::BIabs: 3737 case Builtin::BIlabs: 3738 case Builtin::BIllabs: 3739 case Builtin::BIfabsf: 3740 case Builtin::BIfabs: 3741 case Builtin::BIfabsl: 3742 return Builtin::BIcabsf; 3743 } 3744 } 3745 llvm_unreachable("Unable to convert function"); 3746 } 3747 3748 static unsigned getAbsoluteValueFunctionKind(const FunctionDecl *FDecl) { 3749 const IdentifierInfo *FnInfo = FDecl->getIdentifier(); 3750 if (!FnInfo) 3751 return 0; 3752 3753 switch (FDecl->getBuiltinID()) { 3754 default: 3755 return 0; 3756 case Builtin::BI__builtin_abs: 3757 case Builtin::BI__builtin_fabs: 3758 case Builtin::BI__builtin_fabsf: 3759 case Builtin::BI__builtin_fabsl: 3760 case Builtin::BI__builtin_labs: 3761 case Builtin::BI__builtin_llabs: 3762 case Builtin::BI__builtin_cabs: 3763 case Builtin::BI__builtin_cabsf: 3764 case Builtin::BI__builtin_cabsl: 3765 case Builtin::BIabs: 3766 case Builtin::BIlabs: 3767 case Builtin::BIllabs: 3768 case Builtin::BIfabs: 3769 case Builtin::BIfabsf: 3770 case Builtin::BIfabsl: 3771 case Builtin::BIcabs: 3772 case Builtin::BIcabsf: 3773 case Builtin::BIcabsl: 3774 return FDecl->getBuiltinID(); 3775 } 3776 llvm_unreachable("Unknown Builtin type"); 3777 } 3778 3779 // If the replacement is valid, emit a note with replacement function. 3780 // Additionally, suggest including the proper header if not already included. 3781 static void emitReplacement(Sema &S, SourceLocation Loc, SourceRange Range, 3782 unsigned AbsKind, QualType ArgType) { 3783 bool EmitHeaderHint = true; 3784 const char *HeaderName = nullptr; 3785 const char *FunctionName = nullptr; 3786 if (S.getLangOpts().CPlusPlus && !ArgType->isAnyComplexType()) { 3787 FunctionName = "std::abs"; 3788 if (ArgType->isIntegralOrEnumerationType()) { 3789 HeaderName = "cstdlib"; 3790 } else if (ArgType->isRealFloatingType()) { 3791 HeaderName = "cmath"; 3792 } else { 3793 llvm_unreachable("Invalid Type"); 3794 } 3795 3796 // Lookup all std::abs 3797 if (NamespaceDecl *Std = S.getStdNamespace()) { 3798 LookupResult R(S, &S.Context.Idents.get("abs"), Loc, Sema::LookupAnyName); 3799 R.suppressDiagnostics(); 3800 S.LookupQualifiedName(R, Std); 3801 3802 for (const auto *I : R) { 3803 const FunctionDecl *FDecl = nullptr; 3804 if (const UsingShadowDecl *UsingD = dyn_cast<UsingShadowDecl>(I)) { 3805 FDecl = dyn_cast<FunctionDecl>(UsingD->getTargetDecl()); 3806 } else { 3807 FDecl = dyn_cast<FunctionDecl>(I); 3808 } 3809 if (!FDecl) 3810 continue; 3811 3812 // Found std::abs(), check that they are the right ones. 3813 if (FDecl->getNumParams() != 1) 3814 continue; 3815 3816 // Check that the parameter type can handle the argument. 3817 QualType ParamType = FDecl->getParamDecl(0)->getType(); 3818 if (getAbsoluteValueKind(ArgType) == getAbsoluteValueKind(ParamType) && 3819 S.Context.getTypeSize(ArgType) <= 3820 S.Context.getTypeSize(ParamType)) { 3821 // Found a function, don't need the header hint. 3822 EmitHeaderHint = false; 3823 break; 3824 } 3825 } 3826 } 3827 } else { 3828 FunctionName = S.Context.BuiltinInfo.GetName(AbsKind); 3829 HeaderName = S.Context.BuiltinInfo.getHeaderName(AbsKind); 3830 3831 if (HeaderName) { 3832 DeclarationName DN(&S.Context.Idents.get(FunctionName)); 3833 LookupResult R(S, DN, Loc, Sema::LookupAnyName); 3834 R.suppressDiagnostics(); 3835 S.LookupName(R, S.getCurScope()); 3836 3837 if (R.isSingleResult()) { 3838 FunctionDecl *FD = dyn_cast<FunctionDecl>(R.getFoundDecl()); 3839 if (FD && FD->getBuiltinID() == AbsKind) { 3840 EmitHeaderHint = false; 3841 } else { 3842 return; 3843 } 3844 } else if (!R.empty()) { 3845 return; 3846 } 3847 } 3848 } 3849 3850 S.Diag(Loc, diag::note_replace_abs_function) 3851 << FunctionName << FixItHint::CreateReplacement(Range, FunctionName); 3852 3853 if (!HeaderName) 3854 return; 3855 3856 if (!EmitHeaderHint) 3857 return; 3858 3859 S.Diag(Loc, diag::note_please_include_header) << HeaderName << FunctionName; 3860 } 3861 3862 static bool IsFunctionStdAbs(const FunctionDecl *FDecl) { 3863 if (!FDecl) 3864 return false; 3865 3866 if (!FDecl->getIdentifier() || !FDecl->getIdentifier()->isStr("abs")) 3867 return false; 3868 3869 const NamespaceDecl *ND = dyn_cast<NamespaceDecl>(FDecl->getDeclContext()); 3870 3871 while (ND && ND->isInlineNamespace()) { 3872 ND = dyn_cast<NamespaceDecl>(ND->getDeclContext()); 3873 } 3874 3875 if (!ND || !ND->getIdentifier() || !ND->getIdentifier()->isStr("std")) 3876 return false; 3877 3878 if (!isa<TranslationUnitDecl>(ND->getDeclContext())) 3879 return false; 3880 3881 return true; 3882 } 3883 3884 // Warn when using the wrong abs() function. 3885 void Sema::CheckAbsoluteValueFunction(const CallExpr *Call, 3886 const FunctionDecl *FDecl, 3887 IdentifierInfo *FnInfo) { 3888 if (Call->getNumArgs() != 1) 3889 return; 3890 3891 unsigned AbsKind = getAbsoluteValueFunctionKind(FDecl); 3892 bool IsStdAbs = IsFunctionStdAbs(FDecl); 3893 if (AbsKind == 0 && !IsStdAbs) 3894 return; 3895 3896 QualType ArgType = Call->getArg(0)->IgnoreParenImpCasts()->getType(); 3897 QualType ParamType = Call->getArg(0)->getType(); 3898 3899 // Unsigned types can not be negative. Suggest to drop the absolute value 3900 // function. 3901 if (ArgType->isUnsignedIntegerType()) { 3902 const char *FunctionName = 3903 IsStdAbs ? "std::abs" : Context.BuiltinInfo.GetName(AbsKind); 3904 Diag(Call->getExprLoc(), diag::warn_unsigned_abs) << ArgType << ParamType; 3905 Diag(Call->getExprLoc(), diag::note_remove_abs) 3906 << FunctionName 3907 << FixItHint::CreateRemoval(Call->getCallee()->getSourceRange()); 3908 return; 3909 } 3910 3911 // std::abs has overloads which prevent most of the absolute value problems 3912 // from occurring. 3913 if (IsStdAbs) 3914 return; 3915 3916 AbsoluteValueKind ArgValueKind = getAbsoluteValueKind(ArgType); 3917 AbsoluteValueKind ParamValueKind = getAbsoluteValueKind(ParamType); 3918 3919 // The argument and parameter are the same kind. Check if they are the right 3920 // size. 3921 if (ArgValueKind == ParamValueKind) { 3922 if (Context.getTypeSize(ArgType) <= Context.getTypeSize(ParamType)) 3923 return; 3924 3925 unsigned NewAbsKind = getBestAbsFunction(Context, ArgType, AbsKind); 3926 Diag(Call->getExprLoc(), diag::warn_abs_too_small) 3927 << FDecl << ArgType << ParamType; 3928 3929 if (NewAbsKind == 0) 3930 return; 3931 3932 emitReplacement(*this, Call->getExprLoc(), 3933 Call->getCallee()->getSourceRange(), NewAbsKind, ArgType); 3934 return; 3935 } 3936 3937 // ArgValueKind != ParamValueKind 3938 // The wrong type of absolute value function was used. Attempt to find the 3939 // proper one. 3940 unsigned NewAbsKind = changeAbsFunction(AbsKind, ArgValueKind); 3941 NewAbsKind = getBestAbsFunction(Context, ArgType, NewAbsKind); 3942 if (NewAbsKind == 0) 3943 return; 3944 3945 Diag(Call->getExprLoc(), diag::warn_wrong_absolute_value_type) 3946 << FDecl << ParamValueKind << ArgValueKind; 3947 3948 emitReplacement(*this, Call->getExprLoc(), 3949 Call->getCallee()->getSourceRange(), NewAbsKind, ArgType); 3950 return; 3951 } 3952 3953 //===--- CHECK: Standard memory functions ---------------------------------===// 3954 3955 /// \brief Takes the expression passed to the size_t parameter of functions 3956 /// such as memcmp, strncat, etc and warns if it's a comparison. 3957 /// 3958 /// This is to catch typos like `if (memcmp(&a, &b, sizeof(a) > 0))`. 3959 static bool CheckMemorySizeofForComparison(Sema &S, const Expr *E, 3960 IdentifierInfo *FnName, 3961 SourceLocation FnLoc, 3962 SourceLocation RParenLoc) { 3963 const BinaryOperator *Size = dyn_cast<BinaryOperator>(E); 3964 if (!Size) 3965 return false; 3966 3967 // if E is binop and op is >, <, >=, <=, ==, &&, ||: 3968 if (!Size->isComparisonOp() && !Size->isEqualityOp() && !Size->isLogicalOp()) 3969 return false; 3970 3971 SourceRange SizeRange = Size->getSourceRange(); 3972 S.Diag(Size->getOperatorLoc(), diag::warn_memsize_comparison) 3973 << SizeRange << FnName; 3974 S.Diag(FnLoc, diag::note_memsize_comparison_paren) 3975 << FnName << FixItHint::CreateInsertion( 3976 S.getLocForEndOfToken(Size->getLHS()->getLocEnd()), ")") 3977 << FixItHint::CreateRemoval(RParenLoc); 3978 S.Diag(SizeRange.getBegin(), diag::note_memsize_comparison_cast_silence) 3979 << FixItHint::CreateInsertion(SizeRange.getBegin(), "(size_t)(") 3980 << FixItHint::CreateInsertion(S.getLocForEndOfToken(SizeRange.getEnd()), 3981 ")"); 3982 3983 return true; 3984 } 3985 3986 /// \brief Determine whether the given type is a dynamic class type (e.g., 3987 /// whether it has a vtable). 3988 static bool isDynamicClassType(QualType T) { 3989 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 3990 if (CXXRecordDecl *Definition = Record->getDefinition()) 3991 if (Definition->isDynamicClass()) 3992 return true; 3993 3994 return false; 3995 } 3996 3997 /// \brief If E is a sizeof expression, returns its argument expression, 3998 /// otherwise returns NULL. 3999 static const Expr *getSizeOfExprArg(const Expr* E) { 4000 if (const UnaryExprOrTypeTraitExpr *SizeOf = 4001 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 4002 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 4003 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 4004 4005 return nullptr; 4006 } 4007 4008 /// \brief If E is a sizeof expression, returns its argument type. 4009 static QualType getSizeOfArgType(const Expr* E) { 4010 if (const UnaryExprOrTypeTraitExpr *SizeOf = 4011 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 4012 if (SizeOf->getKind() == clang::UETT_SizeOf) 4013 return SizeOf->getTypeOfArgument(); 4014 4015 return QualType(); 4016 } 4017 4018 /// \brief Check for dangerous or invalid arguments to memset(). 4019 /// 4020 /// This issues warnings on known problematic, dangerous or unspecified 4021 /// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 4022 /// function calls. 4023 /// 4024 /// \param Call The call expression to diagnose. 4025 void Sema::CheckMemaccessArguments(const CallExpr *Call, 4026 unsigned BId, 4027 IdentifierInfo *FnName) { 4028 assert(BId != 0); 4029 4030 // It is possible to have a non-standard definition of memset. Validate 4031 // we have enough arguments, and if not, abort further checking. 4032 unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3); 4033 if (Call->getNumArgs() < ExpectedNumArgs) 4034 return; 4035 4036 unsigned LastArg = (BId == Builtin::BImemset || 4037 BId == Builtin::BIstrndup ? 1 : 2); 4038 unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2); 4039 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 4040 4041 if (CheckMemorySizeofForComparison(*this, LenExpr, FnName, 4042 Call->getLocStart(), Call->getRParenLoc())) 4043 return; 4044 4045 // We have special checking when the length is a sizeof expression. 4046 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 4047 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 4048 llvm::FoldingSetNodeID SizeOfArgID; 4049 4050 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 4051 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 4052 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 4053 4054 QualType DestTy = Dest->getType(); 4055 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 4056 QualType PointeeTy = DestPtrTy->getPointeeType(); 4057 4058 // Never warn about void type pointers. This can be used to suppress 4059 // false positives. 4060 if (PointeeTy->isVoidType()) 4061 continue; 4062 4063 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 4064 // actually comparing the expressions for equality. Because computing the 4065 // expression IDs can be expensive, we only do this if the diagnostic is 4066 // enabled. 4067 if (SizeOfArg && 4068 !Diags.isIgnored(diag::warn_sizeof_pointer_expr_memaccess, 4069 SizeOfArg->getExprLoc())) { 4070 // We only compute IDs for expressions if the warning is enabled, and 4071 // cache the sizeof arg's ID. 4072 if (SizeOfArgID == llvm::FoldingSetNodeID()) 4073 SizeOfArg->Profile(SizeOfArgID, Context, true); 4074 llvm::FoldingSetNodeID DestID; 4075 Dest->Profile(DestID, Context, true); 4076 if (DestID == SizeOfArgID) { 4077 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 4078 // over sizeof(src) as well. 4079 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 4080 StringRef ReadableName = FnName->getName(); 4081 4082 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 4083 if (UnaryOp->getOpcode() == UO_AddrOf) 4084 ActionIdx = 1; // If its an address-of operator, just remove it. 4085 if (!PointeeTy->isIncompleteType() && 4086 (Context.getTypeSize(PointeeTy) == Context.getCharWidth())) 4087 ActionIdx = 2; // If the pointee's size is sizeof(char), 4088 // suggest an explicit length. 4089 4090 // If the function is defined as a builtin macro, do not show macro 4091 // expansion. 4092 SourceLocation SL = SizeOfArg->getExprLoc(); 4093 SourceRange DSR = Dest->getSourceRange(); 4094 SourceRange SSR = SizeOfArg->getSourceRange(); 4095 SourceManager &SM = getSourceManager(); 4096 4097 if (SM.isMacroArgExpansion(SL)) { 4098 ReadableName = Lexer::getImmediateMacroName(SL, SM, LangOpts); 4099 SL = SM.getSpellingLoc(SL); 4100 DSR = SourceRange(SM.getSpellingLoc(DSR.getBegin()), 4101 SM.getSpellingLoc(DSR.getEnd())); 4102 SSR = SourceRange(SM.getSpellingLoc(SSR.getBegin()), 4103 SM.getSpellingLoc(SSR.getEnd())); 4104 } 4105 4106 DiagRuntimeBehavior(SL, SizeOfArg, 4107 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 4108 << ReadableName 4109 << PointeeTy 4110 << DestTy 4111 << DSR 4112 << SSR); 4113 DiagRuntimeBehavior(SL, SizeOfArg, 4114 PDiag(diag::warn_sizeof_pointer_expr_memaccess_note) 4115 << ActionIdx 4116 << SSR); 4117 4118 break; 4119 } 4120 } 4121 4122 // Also check for cases where the sizeof argument is the exact same 4123 // type as the memory argument, and where it points to a user-defined 4124 // record type. 4125 if (SizeOfArgTy != QualType()) { 4126 if (PointeeTy->isRecordType() && 4127 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 4128 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 4129 PDiag(diag::warn_sizeof_pointer_type_memaccess) 4130 << FnName << SizeOfArgTy << ArgIdx 4131 << PointeeTy << Dest->getSourceRange() 4132 << LenExpr->getSourceRange()); 4133 break; 4134 } 4135 } 4136 4137 // Always complain about dynamic classes. 4138 if (isDynamicClassType(PointeeTy)) { 4139 4140 unsigned OperationType = 0; 4141 // "overwritten" if we're warning about the destination for any call 4142 // but memcmp; otherwise a verb appropriate to the call. 4143 if (ArgIdx != 0 || BId == Builtin::BImemcmp) { 4144 if (BId == Builtin::BImemcpy) 4145 OperationType = 1; 4146 else if(BId == Builtin::BImemmove) 4147 OperationType = 2; 4148 else if (BId == Builtin::BImemcmp) 4149 OperationType = 3; 4150 } 4151 4152 DiagRuntimeBehavior( 4153 Dest->getExprLoc(), Dest, 4154 PDiag(diag::warn_dyn_class_memaccess) 4155 << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx) 4156 << FnName << PointeeTy 4157 << OperationType 4158 << Call->getCallee()->getSourceRange()); 4159 } else if (PointeeTy.hasNonTrivialObjCLifetime() && 4160 BId != Builtin::BImemset) 4161 DiagRuntimeBehavior( 4162 Dest->getExprLoc(), Dest, 4163 PDiag(diag::warn_arc_object_memaccess) 4164 << ArgIdx << FnName << PointeeTy 4165 << Call->getCallee()->getSourceRange()); 4166 else 4167 continue; 4168 4169 DiagRuntimeBehavior( 4170 Dest->getExprLoc(), Dest, 4171 PDiag(diag::note_bad_memaccess_silence) 4172 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 4173 break; 4174 } 4175 } 4176 } 4177 4178 // A little helper routine: ignore addition and subtraction of integer literals. 4179 // This intentionally does not ignore all integer constant expressions because 4180 // we don't want to remove sizeof(). 4181 static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 4182 Ex = Ex->IgnoreParenCasts(); 4183 4184 for (;;) { 4185 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 4186 if (!BO || !BO->isAdditiveOp()) 4187 break; 4188 4189 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 4190 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 4191 4192 if (isa<IntegerLiteral>(RHS)) 4193 Ex = LHS; 4194 else if (isa<IntegerLiteral>(LHS)) 4195 Ex = RHS; 4196 else 4197 break; 4198 } 4199 4200 return Ex; 4201 } 4202 4203 static bool isConstantSizeArrayWithMoreThanOneElement(QualType Ty, 4204 ASTContext &Context) { 4205 // Only handle constant-sized or VLAs, but not flexible members. 4206 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(Ty)) { 4207 // Only issue the FIXIT for arrays of size > 1. 4208 if (CAT->getSize().getSExtValue() <= 1) 4209 return false; 4210 } else if (!Ty->isVariableArrayType()) { 4211 return false; 4212 } 4213 return true; 4214 } 4215 4216 // Warn if the user has made the 'size' argument to strlcpy or strlcat 4217 // be the size of the source, instead of the destination. 4218 void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 4219 IdentifierInfo *FnName) { 4220 4221 // Don't crash if the user has the wrong number of arguments 4222 if (Call->getNumArgs() != 3) 4223 return; 4224 4225 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 4226 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 4227 const Expr *CompareWithSrc = nullptr; 4228 4229 if (CheckMemorySizeofForComparison(*this, SizeArg, FnName, 4230 Call->getLocStart(), Call->getRParenLoc())) 4231 return; 4232 4233 // Look for 'strlcpy(dst, x, sizeof(x))' 4234 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 4235 CompareWithSrc = Ex; 4236 else { 4237 // Look for 'strlcpy(dst, x, strlen(x))' 4238 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 4239 if (SizeCall->getBuiltinCallee() == Builtin::BIstrlen && 4240 SizeCall->getNumArgs() == 1) 4241 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 4242 } 4243 } 4244 4245 if (!CompareWithSrc) 4246 return; 4247 4248 // Determine if the argument to sizeof/strlen is equal to the source 4249 // argument. In principle there's all kinds of things you could do 4250 // here, for instance creating an == expression and evaluating it with 4251 // EvaluateAsBooleanCondition, but this uses a more direct technique: 4252 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 4253 if (!SrcArgDRE) 4254 return; 4255 4256 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 4257 if (!CompareWithSrcDRE || 4258 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 4259 return; 4260 4261 const Expr *OriginalSizeArg = Call->getArg(2); 4262 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 4263 << OriginalSizeArg->getSourceRange() << FnName; 4264 4265 // Output a FIXIT hint if the destination is an array (rather than a 4266 // pointer to an array). This could be enhanced to handle some 4267 // pointers if we know the actual size, like if DstArg is 'array+2' 4268 // we could say 'sizeof(array)-2'. 4269 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 4270 if (!isConstantSizeArrayWithMoreThanOneElement(DstArg->getType(), Context)) 4271 return; 4272 4273 SmallString<128> sizeString; 4274 llvm::raw_svector_ostream OS(sizeString); 4275 OS << "sizeof("; 4276 DstArg->printPretty(OS, nullptr, getPrintingPolicy()); 4277 OS << ")"; 4278 4279 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 4280 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 4281 OS.str()); 4282 } 4283 4284 /// Check if two expressions refer to the same declaration. 4285 static bool referToTheSameDecl(const Expr *E1, const Expr *E2) { 4286 if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1)) 4287 if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2)) 4288 return D1->getDecl() == D2->getDecl(); 4289 return false; 4290 } 4291 4292 static const Expr *getStrlenExprArg(const Expr *E) { 4293 if (const CallExpr *CE = dyn_cast<CallExpr>(E)) { 4294 const FunctionDecl *FD = CE->getDirectCallee(); 4295 if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen) 4296 return nullptr; 4297 return CE->getArg(0)->IgnoreParenCasts(); 4298 } 4299 return nullptr; 4300 } 4301 4302 // Warn on anti-patterns as the 'size' argument to strncat. 4303 // The correct size argument should look like following: 4304 // strncat(dst, src, sizeof(dst) - strlen(dest) - 1); 4305 void Sema::CheckStrncatArguments(const CallExpr *CE, 4306 IdentifierInfo *FnName) { 4307 // Don't crash if the user has the wrong number of arguments. 4308 if (CE->getNumArgs() < 3) 4309 return; 4310 const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts(); 4311 const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts(); 4312 const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts(); 4313 4314 if (CheckMemorySizeofForComparison(*this, LenArg, FnName, CE->getLocStart(), 4315 CE->getRParenLoc())) 4316 return; 4317 4318 // Identify common expressions, which are wrongly used as the size argument 4319 // to strncat and may lead to buffer overflows. 4320 unsigned PatternType = 0; 4321 if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) { 4322 // - sizeof(dst) 4323 if (referToTheSameDecl(SizeOfArg, DstArg)) 4324 PatternType = 1; 4325 // - sizeof(src) 4326 else if (referToTheSameDecl(SizeOfArg, SrcArg)) 4327 PatternType = 2; 4328 } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) { 4329 if (BE->getOpcode() == BO_Sub) { 4330 const Expr *L = BE->getLHS()->IgnoreParenCasts(); 4331 const Expr *R = BE->getRHS()->IgnoreParenCasts(); 4332 // - sizeof(dst) - strlen(dst) 4333 if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) && 4334 referToTheSameDecl(DstArg, getStrlenExprArg(R))) 4335 PatternType = 1; 4336 // - sizeof(src) - (anything) 4337 else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L))) 4338 PatternType = 2; 4339 } 4340 } 4341 4342 if (PatternType == 0) 4343 return; 4344 4345 // Generate the diagnostic. 4346 SourceLocation SL = LenArg->getLocStart(); 4347 SourceRange SR = LenArg->getSourceRange(); 4348 SourceManager &SM = getSourceManager(); 4349 4350 // If the function is defined as a builtin macro, do not show macro expansion. 4351 if (SM.isMacroArgExpansion(SL)) { 4352 SL = SM.getSpellingLoc(SL); 4353 SR = SourceRange(SM.getSpellingLoc(SR.getBegin()), 4354 SM.getSpellingLoc(SR.getEnd())); 4355 } 4356 4357 // Check if the destination is an array (rather than a pointer to an array). 4358 QualType DstTy = DstArg->getType(); 4359 bool isKnownSizeArray = isConstantSizeArrayWithMoreThanOneElement(DstTy, 4360 Context); 4361 if (!isKnownSizeArray) { 4362 if (PatternType == 1) 4363 Diag(SL, diag::warn_strncat_wrong_size) << SR; 4364 else 4365 Diag(SL, diag::warn_strncat_src_size) << SR; 4366 return; 4367 } 4368 4369 if (PatternType == 1) 4370 Diag(SL, diag::warn_strncat_large_size) << SR; 4371 else 4372 Diag(SL, diag::warn_strncat_src_size) << SR; 4373 4374 SmallString<128> sizeString; 4375 llvm::raw_svector_ostream OS(sizeString); 4376 OS << "sizeof("; 4377 DstArg->printPretty(OS, nullptr, getPrintingPolicy()); 4378 OS << ") - "; 4379 OS << "strlen("; 4380 DstArg->printPretty(OS, nullptr, getPrintingPolicy()); 4381 OS << ") - 1"; 4382 4383 Diag(SL, diag::note_strncat_wrong_size) 4384 << FixItHint::CreateReplacement(SR, OS.str()); 4385 } 4386 4387 //===--- CHECK: Return Address of Stack Variable --------------------------===// 4388 4389 static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 4390 Decl *ParentDecl); 4391 static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars, 4392 Decl *ParentDecl); 4393 4394 /// CheckReturnStackAddr - Check if a return statement returns the address 4395 /// of a stack variable. 4396 static void 4397 CheckReturnStackAddr(Sema &S, Expr *RetValExp, QualType lhsType, 4398 SourceLocation ReturnLoc) { 4399 4400 Expr *stackE = nullptr; 4401 SmallVector<DeclRefExpr *, 8> refVars; 4402 4403 // Perform checking for returned stack addresses, local blocks, 4404 // label addresses or references to temporaries. 4405 if (lhsType->isPointerType() || 4406 (!S.getLangOpts().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 4407 stackE = EvalAddr(RetValExp, refVars, /*ParentDecl=*/nullptr); 4408 } else if (lhsType->isReferenceType()) { 4409 stackE = EvalVal(RetValExp, refVars, /*ParentDecl=*/nullptr); 4410 } 4411 4412 if (!stackE) 4413 return; // Nothing suspicious was found. 4414 4415 SourceLocation diagLoc; 4416 SourceRange diagRange; 4417 if (refVars.empty()) { 4418 diagLoc = stackE->getLocStart(); 4419 diagRange = stackE->getSourceRange(); 4420 } else { 4421 // We followed through a reference variable. 'stackE' contains the 4422 // problematic expression but we will warn at the return statement pointing 4423 // at the reference variable. We will later display the "trail" of 4424 // reference variables using notes. 4425 diagLoc = refVars[0]->getLocStart(); 4426 diagRange = refVars[0]->getSourceRange(); 4427 } 4428 4429 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 4430 S.Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 4431 : diag::warn_ret_stack_addr) 4432 << DR->getDecl()->getDeclName() << diagRange; 4433 } else if (isa<BlockExpr>(stackE)) { // local block. 4434 S.Diag(diagLoc, diag::err_ret_local_block) << diagRange; 4435 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 4436 S.Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 4437 } else { // local temporary. 4438 S.Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 4439 : diag::warn_ret_local_temp_addr) 4440 << diagRange; 4441 } 4442 4443 // Display the "trail" of reference variables that we followed until we 4444 // found the problematic expression using notes. 4445 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 4446 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 4447 // If this var binds to another reference var, show the range of the next 4448 // var, otherwise the var binds to the problematic expression, in which case 4449 // show the range of the expression. 4450 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 4451 : stackE->getSourceRange(); 4452 S.Diag(VD->getLocation(), diag::note_ref_var_local_bind) 4453 << VD->getDeclName() << range; 4454 } 4455 } 4456 4457 /// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 4458 /// check if the expression in a return statement evaluates to an address 4459 /// to a location on the stack, a local block, an address of a label, or a 4460 /// reference to local temporary. The recursion is used to traverse the 4461 /// AST of the return expression, with recursion backtracking when we 4462 /// encounter a subexpression that (1) clearly does not lead to one of the 4463 /// above problematic expressions (2) is something we cannot determine leads to 4464 /// a problematic expression based on such local checking. 4465 /// 4466 /// Both EvalAddr and EvalVal follow through reference variables to evaluate 4467 /// the expression that they point to. Such variables are added to the 4468 /// 'refVars' vector so that we know what the reference variable "trail" was. 4469 /// 4470 /// EvalAddr processes expressions that are pointers that are used as 4471 /// references (and not L-values). EvalVal handles all other values. 4472 /// At the base case of the recursion is a check for the above problematic 4473 /// expressions. 4474 /// 4475 /// This implementation handles: 4476 /// 4477 /// * pointer-to-pointer casts 4478 /// * implicit conversions from array references to pointers 4479 /// * taking the address of fields 4480 /// * arbitrary interplay between "&" and "*" operators 4481 /// * pointer arithmetic from an address of a stack variable 4482 /// * taking the address of an array element where the array is on the stack 4483 static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 4484 Decl *ParentDecl) { 4485 if (E->isTypeDependent()) 4486 return nullptr; 4487 4488 // We should only be called for evaluating pointer expressions. 4489 assert((E->getType()->isAnyPointerType() || 4490 E->getType()->isBlockPointerType() || 4491 E->getType()->isObjCQualifiedIdType()) && 4492 "EvalAddr only works on pointers"); 4493 4494 E = E->IgnoreParens(); 4495 4496 // Our "symbolic interpreter" is just a dispatch off the currently 4497 // viewed AST node. We then recursively traverse the AST by calling 4498 // EvalAddr and EvalVal appropriately. 4499 switch (E->getStmtClass()) { 4500 case Stmt::DeclRefExprClass: { 4501 DeclRefExpr *DR = cast<DeclRefExpr>(E); 4502 4503 // If we leave the immediate function, the lifetime isn't about to end. 4504 if (DR->refersToEnclosingLocal()) 4505 return nullptr; 4506 4507 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 4508 // If this is a reference variable, follow through to the expression that 4509 // it points to. 4510 if (V->hasLocalStorage() && 4511 V->getType()->isReferenceType() && V->hasInit()) { 4512 // Add the reference variable to the "trail". 4513 refVars.push_back(DR); 4514 return EvalAddr(V->getInit(), refVars, ParentDecl); 4515 } 4516 4517 return nullptr; 4518 } 4519 4520 case Stmt::UnaryOperatorClass: { 4521 // The only unary operator that make sense to handle here 4522 // is AddrOf. All others don't make sense as pointers. 4523 UnaryOperator *U = cast<UnaryOperator>(E); 4524 4525 if (U->getOpcode() == UO_AddrOf) 4526 return EvalVal(U->getSubExpr(), refVars, ParentDecl); 4527 else 4528 return nullptr; 4529 } 4530 4531 case Stmt::BinaryOperatorClass: { 4532 // Handle pointer arithmetic. All other binary operators are not valid 4533 // in this context. 4534 BinaryOperator *B = cast<BinaryOperator>(E); 4535 BinaryOperatorKind op = B->getOpcode(); 4536 4537 if (op != BO_Add && op != BO_Sub) 4538 return nullptr; 4539 4540 Expr *Base = B->getLHS(); 4541 4542 // Determine which argument is the real pointer base. It could be 4543 // the RHS argument instead of the LHS. 4544 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 4545 4546 assert (Base->getType()->isPointerType()); 4547 return EvalAddr(Base, refVars, ParentDecl); 4548 } 4549 4550 // For conditional operators we need to see if either the LHS or RHS are 4551 // valid DeclRefExpr*s. If one of them is valid, we return it. 4552 case Stmt::ConditionalOperatorClass: { 4553 ConditionalOperator *C = cast<ConditionalOperator>(E); 4554 4555 // Handle the GNU extension for missing LHS. 4556 // FIXME: That isn't a ConditionalOperator, so doesn't get here. 4557 if (Expr *LHSExpr = C->getLHS()) { 4558 // In C++, we can have a throw-expression, which has 'void' type. 4559 if (!LHSExpr->getType()->isVoidType()) 4560 if (Expr *LHS = EvalAddr(LHSExpr, refVars, ParentDecl)) 4561 return LHS; 4562 } 4563 4564 // In C++, we can have a throw-expression, which has 'void' type. 4565 if (C->getRHS()->getType()->isVoidType()) 4566 return nullptr; 4567 4568 return EvalAddr(C->getRHS(), refVars, ParentDecl); 4569 } 4570 4571 case Stmt::BlockExprClass: 4572 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 4573 return E; // local block. 4574 return nullptr; 4575 4576 case Stmt::AddrLabelExprClass: 4577 return E; // address of label. 4578 4579 case Stmt::ExprWithCleanupsClass: 4580 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars, 4581 ParentDecl); 4582 4583 // For casts, we need to handle conversions from arrays to 4584 // pointer values, and pointer-to-pointer conversions. 4585 case Stmt::ImplicitCastExprClass: 4586 case Stmt::CStyleCastExprClass: 4587 case Stmt::CXXFunctionalCastExprClass: 4588 case Stmt::ObjCBridgedCastExprClass: 4589 case Stmt::CXXStaticCastExprClass: 4590 case Stmt::CXXDynamicCastExprClass: 4591 case Stmt::CXXConstCastExprClass: 4592 case Stmt::CXXReinterpretCastExprClass: { 4593 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 4594 switch (cast<CastExpr>(E)->getCastKind()) { 4595 case CK_BitCast: 4596 case CK_LValueToRValue: 4597 case CK_NoOp: 4598 case CK_BaseToDerived: 4599 case CK_DerivedToBase: 4600 case CK_UncheckedDerivedToBase: 4601 case CK_Dynamic: 4602 case CK_CPointerToObjCPointerCast: 4603 case CK_BlockPointerToObjCPointerCast: 4604 case CK_AnyPointerToBlockPointerCast: 4605 return EvalAddr(SubExpr, refVars, ParentDecl); 4606 4607 case CK_ArrayToPointerDecay: 4608 return EvalVal(SubExpr, refVars, ParentDecl); 4609 4610 default: 4611 return nullptr; 4612 } 4613 } 4614 4615 case Stmt::MaterializeTemporaryExprClass: 4616 if (Expr *Result = EvalAddr( 4617 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 4618 refVars, ParentDecl)) 4619 return Result; 4620 4621 return E; 4622 4623 // Everything else: we simply don't reason about them. 4624 default: 4625 return nullptr; 4626 } 4627 } 4628 4629 4630 /// EvalVal - This function is complements EvalAddr in the mutual recursion. 4631 /// See the comments for EvalAddr for more details. 4632 static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 4633 Decl *ParentDecl) { 4634 do { 4635 // We should only be called for evaluating non-pointer expressions, or 4636 // expressions with a pointer type that are not used as references but instead 4637 // are l-values (e.g., DeclRefExpr with a pointer type). 4638 4639 // Our "symbolic interpreter" is just a dispatch off the currently 4640 // viewed AST node. We then recursively traverse the AST by calling 4641 // EvalAddr and EvalVal appropriately. 4642 4643 E = E->IgnoreParens(); 4644 switch (E->getStmtClass()) { 4645 case Stmt::ImplicitCastExprClass: { 4646 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 4647 if (IE->getValueKind() == VK_LValue) { 4648 E = IE->getSubExpr(); 4649 continue; 4650 } 4651 return nullptr; 4652 } 4653 4654 case Stmt::ExprWithCleanupsClass: 4655 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,ParentDecl); 4656 4657 case Stmt::DeclRefExprClass: { 4658 // When we hit a DeclRefExpr we are looking at code that refers to a 4659 // variable's name. If it's not a reference variable we check if it has 4660 // local storage within the function, and if so, return the expression. 4661 DeclRefExpr *DR = cast<DeclRefExpr>(E); 4662 4663 // If we leave the immediate function, the lifetime isn't about to end. 4664 if (DR->refersToEnclosingLocal()) 4665 return nullptr; 4666 4667 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) { 4668 // Check if it refers to itself, e.g. "int& i = i;". 4669 if (V == ParentDecl) 4670 return DR; 4671 4672 if (V->hasLocalStorage()) { 4673 if (!V->getType()->isReferenceType()) 4674 return DR; 4675 4676 // Reference variable, follow through to the expression that 4677 // it points to. 4678 if (V->hasInit()) { 4679 // Add the reference variable to the "trail". 4680 refVars.push_back(DR); 4681 return EvalVal(V->getInit(), refVars, V); 4682 } 4683 } 4684 } 4685 4686 return nullptr; 4687 } 4688 4689 case Stmt::UnaryOperatorClass: { 4690 // The only unary operator that make sense to handle here 4691 // is Deref. All others don't resolve to a "name." This includes 4692 // handling all sorts of rvalues passed to a unary operator. 4693 UnaryOperator *U = cast<UnaryOperator>(E); 4694 4695 if (U->getOpcode() == UO_Deref) 4696 return EvalAddr(U->getSubExpr(), refVars, ParentDecl); 4697 4698 return nullptr; 4699 } 4700 4701 case Stmt::ArraySubscriptExprClass: { 4702 // Array subscripts are potential references to data on the stack. We 4703 // retrieve the DeclRefExpr* for the array variable if it indeed 4704 // has local storage. 4705 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars,ParentDecl); 4706 } 4707 4708 case Stmt::ConditionalOperatorClass: { 4709 // For conditional operators we need to see if either the LHS or RHS are 4710 // non-NULL Expr's. If one is non-NULL, we return it. 4711 ConditionalOperator *C = cast<ConditionalOperator>(E); 4712 4713 // Handle the GNU extension for missing LHS. 4714 if (Expr *LHSExpr = C->getLHS()) { 4715 // In C++, we can have a throw-expression, which has 'void' type. 4716 if (!LHSExpr->getType()->isVoidType()) 4717 if (Expr *LHS = EvalVal(LHSExpr, refVars, ParentDecl)) 4718 return LHS; 4719 } 4720 4721 // In C++, we can have a throw-expression, which has 'void' type. 4722 if (C->getRHS()->getType()->isVoidType()) 4723 return nullptr; 4724 4725 return EvalVal(C->getRHS(), refVars, ParentDecl); 4726 } 4727 4728 // Accesses to members are potential references to data on the stack. 4729 case Stmt::MemberExprClass: { 4730 MemberExpr *M = cast<MemberExpr>(E); 4731 4732 // Check for indirect access. We only want direct field accesses. 4733 if (M->isArrow()) 4734 return nullptr; 4735 4736 // Check whether the member type is itself a reference, in which case 4737 // we're not going to refer to the member, but to what the member refers to. 4738 if (M->getMemberDecl()->getType()->isReferenceType()) 4739 return nullptr; 4740 4741 return EvalVal(M->getBase(), refVars, ParentDecl); 4742 } 4743 4744 case Stmt::MaterializeTemporaryExprClass: 4745 if (Expr *Result = EvalVal( 4746 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 4747 refVars, ParentDecl)) 4748 return Result; 4749 4750 return E; 4751 4752 default: 4753 // Check that we don't return or take the address of a reference to a 4754 // temporary. This is only useful in C++. 4755 if (!E->isTypeDependent() && E->isRValue()) 4756 return E; 4757 4758 // Everything else: we simply don't reason about them. 4759 return nullptr; 4760 } 4761 } while (true); 4762 } 4763 4764 void 4765 Sema::CheckReturnValExpr(Expr *RetValExp, QualType lhsType, 4766 SourceLocation ReturnLoc, 4767 bool isObjCMethod, 4768 const AttrVec *Attrs, 4769 const FunctionDecl *FD) { 4770 CheckReturnStackAddr(*this, RetValExp, lhsType, ReturnLoc); 4771 4772 // Check if the return value is null but should not be. 4773 if (Attrs && hasSpecificAttr<ReturnsNonNullAttr>(*Attrs) && 4774 CheckNonNullExpr(*this, RetValExp)) 4775 Diag(ReturnLoc, diag::warn_null_ret) 4776 << (isObjCMethod ? 1 : 0) << RetValExp->getSourceRange(); 4777 4778 // C++11 [basic.stc.dynamic.allocation]p4: 4779 // If an allocation function declared with a non-throwing 4780 // exception-specification fails to allocate storage, it shall return 4781 // a null pointer. Any other allocation function that fails to allocate 4782 // storage shall indicate failure only by throwing an exception [...] 4783 if (FD) { 4784 OverloadedOperatorKind Op = FD->getOverloadedOperator(); 4785 if (Op == OO_New || Op == OO_Array_New) { 4786 const FunctionProtoType *Proto 4787 = FD->getType()->castAs<FunctionProtoType>(); 4788 if (!Proto->isNothrow(Context, /*ResultIfDependent*/true) && 4789 CheckNonNullExpr(*this, RetValExp)) 4790 Diag(ReturnLoc, diag::warn_operator_new_returns_null) 4791 << FD << getLangOpts().CPlusPlus11; 4792 } 4793 } 4794 } 4795 4796 //===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 4797 4798 /// Check for comparisons of floating point operands using != and ==. 4799 /// Issue a warning if these are no self-comparisons, as they are not likely 4800 /// to do what the programmer intended. 4801 void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 4802 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 4803 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 4804 4805 // Special case: check for x == x (which is OK). 4806 // Do not emit warnings for such cases. 4807 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 4808 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 4809 if (DRL->getDecl() == DRR->getDecl()) 4810 return; 4811 4812 4813 // Special case: check for comparisons against literals that can be exactly 4814 // represented by APFloat. In such cases, do not emit a warning. This 4815 // is a heuristic: often comparison against such literals are used to 4816 // detect if a value in a variable has not changed. This clearly can 4817 // lead to false negatives. 4818 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 4819 if (FLL->isExact()) 4820 return; 4821 } else 4822 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)) 4823 if (FLR->isExact()) 4824 return; 4825 4826 // Check for comparisons with builtin types. 4827 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 4828 if (CL->getBuiltinCallee()) 4829 return; 4830 4831 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 4832 if (CR->getBuiltinCallee()) 4833 return; 4834 4835 // Emit the diagnostic. 4836 Diag(Loc, diag::warn_floatingpoint_eq) 4837 << LHS->getSourceRange() << RHS->getSourceRange(); 4838 } 4839 4840 //===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 4841 //===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 4842 4843 namespace { 4844 4845 /// Structure recording the 'active' range of an integer-valued 4846 /// expression. 4847 struct IntRange { 4848 /// The number of bits active in the int. 4849 unsigned Width; 4850 4851 /// True if the int is known not to have negative values. 4852 bool NonNegative; 4853 4854 IntRange(unsigned Width, bool NonNegative) 4855 : Width(Width), NonNegative(NonNegative) 4856 {} 4857 4858 /// Returns the range of the bool type. 4859 static IntRange forBoolType() { 4860 return IntRange(1, true); 4861 } 4862 4863 /// Returns the range of an opaque value of the given integral type. 4864 static IntRange forValueOfType(ASTContext &C, QualType T) { 4865 return forValueOfCanonicalType(C, 4866 T->getCanonicalTypeInternal().getTypePtr()); 4867 } 4868 4869 /// Returns the range of an opaque value of a canonical integral type. 4870 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 4871 assert(T->isCanonicalUnqualified()); 4872 4873 if (const VectorType *VT = dyn_cast<VectorType>(T)) 4874 T = VT->getElementType().getTypePtr(); 4875 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 4876 T = CT->getElementType().getTypePtr(); 4877 4878 // For enum types, use the known bit width of the enumerators. 4879 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 4880 EnumDecl *Enum = ET->getDecl(); 4881 if (!Enum->isCompleteDefinition()) 4882 return IntRange(C.getIntWidth(QualType(T, 0)), false); 4883 4884 unsigned NumPositive = Enum->getNumPositiveBits(); 4885 unsigned NumNegative = Enum->getNumNegativeBits(); 4886 4887 if (NumNegative == 0) 4888 return IntRange(NumPositive, true/*NonNegative*/); 4889 else 4890 return IntRange(std::max(NumPositive + 1, NumNegative), 4891 false/*NonNegative*/); 4892 } 4893 4894 const BuiltinType *BT = cast<BuiltinType>(T); 4895 assert(BT->isInteger()); 4896 4897 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 4898 } 4899 4900 /// Returns the "target" range of a canonical integral type, i.e. 4901 /// the range of values expressible in the type. 4902 /// 4903 /// This matches forValueOfCanonicalType except that enums have the 4904 /// full range of their type, not the range of their enumerators. 4905 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 4906 assert(T->isCanonicalUnqualified()); 4907 4908 if (const VectorType *VT = dyn_cast<VectorType>(T)) 4909 T = VT->getElementType().getTypePtr(); 4910 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 4911 T = CT->getElementType().getTypePtr(); 4912 if (const EnumType *ET = dyn_cast<EnumType>(T)) 4913 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 4914 4915 const BuiltinType *BT = cast<BuiltinType>(T); 4916 assert(BT->isInteger()); 4917 4918 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 4919 } 4920 4921 /// Returns the supremum of two ranges: i.e. their conservative merge. 4922 static IntRange join(IntRange L, IntRange R) { 4923 return IntRange(std::max(L.Width, R.Width), 4924 L.NonNegative && R.NonNegative); 4925 } 4926 4927 /// Returns the infinum of two ranges: i.e. their aggressive merge. 4928 static IntRange meet(IntRange L, IntRange R) { 4929 return IntRange(std::min(L.Width, R.Width), 4930 L.NonNegative || R.NonNegative); 4931 } 4932 }; 4933 4934 static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, 4935 unsigned MaxWidth) { 4936 if (value.isSigned() && value.isNegative()) 4937 return IntRange(value.getMinSignedBits(), false); 4938 4939 if (value.getBitWidth() > MaxWidth) 4940 value = value.trunc(MaxWidth); 4941 4942 // isNonNegative() just checks the sign bit without considering 4943 // signedness. 4944 return IntRange(value.getActiveBits(), true); 4945 } 4946 4947 static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 4948 unsigned MaxWidth) { 4949 if (result.isInt()) 4950 return GetValueRange(C, result.getInt(), MaxWidth); 4951 4952 if (result.isVector()) { 4953 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 4954 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 4955 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 4956 R = IntRange::join(R, El); 4957 } 4958 return R; 4959 } 4960 4961 if (result.isComplexInt()) { 4962 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 4963 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 4964 return IntRange::join(R, I); 4965 } 4966 4967 // This can happen with lossless casts to intptr_t of "based" lvalues. 4968 // Assume it might use arbitrary bits. 4969 // FIXME: The only reason we need to pass the type in here is to get 4970 // the sign right on this one case. It would be nice if APValue 4971 // preserved this. 4972 assert(result.isLValue() || result.isAddrLabelDiff()); 4973 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 4974 } 4975 4976 static QualType GetExprType(Expr *E) { 4977 QualType Ty = E->getType(); 4978 if (const AtomicType *AtomicRHS = Ty->getAs<AtomicType>()) 4979 Ty = AtomicRHS->getValueType(); 4980 return Ty; 4981 } 4982 4983 /// Pseudo-evaluate the given integer expression, estimating the 4984 /// range of values it might take. 4985 /// 4986 /// \param MaxWidth - the width to which the value will be truncated 4987 static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 4988 E = E->IgnoreParens(); 4989 4990 // Try a full evaluation first. 4991 Expr::EvalResult result; 4992 if (E->EvaluateAsRValue(result, C)) 4993 return GetValueRange(C, result.Val, GetExprType(E), MaxWidth); 4994 4995 // I think we only want to look through implicit casts here; if the 4996 // user has an explicit widening cast, we should treat the value as 4997 // being of the new, wider type. 4998 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 4999 if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue) 5000 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 5001 5002 IntRange OutputTypeRange = IntRange::forValueOfType(C, GetExprType(CE)); 5003 5004 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 5005 5006 // Assume that non-integer casts can span the full range of the type. 5007 if (!isIntegerCast) 5008 return OutputTypeRange; 5009 5010 IntRange SubRange 5011 = GetExprRange(C, CE->getSubExpr(), 5012 std::min(MaxWidth, OutputTypeRange.Width)); 5013 5014 // Bail out if the subexpr's range is as wide as the cast type. 5015 if (SubRange.Width >= OutputTypeRange.Width) 5016 return OutputTypeRange; 5017 5018 // Otherwise, we take the smaller width, and we're non-negative if 5019 // either the output type or the subexpr is. 5020 return IntRange(SubRange.Width, 5021 SubRange.NonNegative || OutputTypeRange.NonNegative); 5022 } 5023 5024 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 5025 // If we can fold the condition, just take that operand. 5026 bool CondResult; 5027 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 5028 return GetExprRange(C, CondResult ? CO->getTrueExpr() 5029 : CO->getFalseExpr(), 5030 MaxWidth); 5031 5032 // Otherwise, conservatively merge. 5033 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 5034 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 5035 return IntRange::join(L, R); 5036 } 5037 5038 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 5039 switch (BO->getOpcode()) { 5040 5041 // Boolean-valued operations are single-bit and positive. 5042 case BO_LAnd: 5043 case BO_LOr: 5044 case BO_LT: 5045 case BO_GT: 5046 case BO_LE: 5047 case BO_GE: 5048 case BO_EQ: 5049 case BO_NE: 5050 return IntRange::forBoolType(); 5051 5052 // The type of the assignments is the type of the LHS, so the RHS 5053 // is not necessarily the same type. 5054 case BO_MulAssign: 5055 case BO_DivAssign: 5056 case BO_RemAssign: 5057 case BO_AddAssign: 5058 case BO_SubAssign: 5059 case BO_XorAssign: 5060 case BO_OrAssign: 5061 // TODO: bitfields? 5062 return IntRange::forValueOfType(C, GetExprType(E)); 5063 5064 // Simple assignments just pass through the RHS, which will have 5065 // been coerced to the LHS type. 5066 case BO_Assign: 5067 // TODO: bitfields? 5068 return GetExprRange(C, BO->getRHS(), MaxWidth); 5069 5070 // Operations with opaque sources are black-listed. 5071 case BO_PtrMemD: 5072 case BO_PtrMemI: 5073 return IntRange::forValueOfType(C, GetExprType(E)); 5074 5075 // Bitwise-and uses the *infinum* of the two source ranges. 5076 case BO_And: 5077 case BO_AndAssign: 5078 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 5079 GetExprRange(C, BO->getRHS(), MaxWidth)); 5080 5081 // Left shift gets black-listed based on a judgement call. 5082 case BO_Shl: 5083 // ...except that we want to treat '1 << (blah)' as logically 5084 // positive. It's an important idiom. 5085 if (IntegerLiteral *I 5086 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 5087 if (I->getValue() == 1) { 5088 IntRange R = IntRange::forValueOfType(C, GetExprType(E)); 5089 return IntRange(R.Width, /*NonNegative*/ true); 5090 } 5091 } 5092 // fallthrough 5093 5094 case BO_ShlAssign: 5095 return IntRange::forValueOfType(C, GetExprType(E)); 5096 5097 // Right shift by a constant can narrow its left argument. 5098 case BO_Shr: 5099 case BO_ShrAssign: { 5100 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 5101 5102 // If the shift amount is a positive constant, drop the width by 5103 // that much. 5104 llvm::APSInt shift; 5105 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 5106 shift.isNonNegative()) { 5107 unsigned zext = shift.getZExtValue(); 5108 if (zext >= L.Width) 5109 L.Width = (L.NonNegative ? 0 : 1); 5110 else 5111 L.Width -= zext; 5112 } 5113 5114 return L; 5115 } 5116 5117 // Comma acts as its right operand. 5118 case BO_Comma: 5119 return GetExprRange(C, BO->getRHS(), MaxWidth); 5120 5121 // Black-list pointer subtractions. 5122 case BO_Sub: 5123 if (BO->getLHS()->getType()->isPointerType()) 5124 return IntRange::forValueOfType(C, GetExprType(E)); 5125 break; 5126 5127 // The width of a division result is mostly determined by the size 5128 // of the LHS. 5129 case BO_Div: { 5130 // Don't 'pre-truncate' the operands. 5131 unsigned opWidth = C.getIntWidth(GetExprType(E)); 5132 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 5133 5134 // If the divisor is constant, use that. 5135 llvm::APSInt divisor; 5136 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 5137 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 5138 if (log2 >= L.Width) 5139 L.Width = (L.NonNegative ? 0 : 1); 5140 else 5141 L.Width = std::min(L.Width - log2, MaxWidth); 5142 return L; 5143 } 5144 5145 // Otherwise, just use the LHS's width. 5146 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 5147 return IntRange(L.Width, L.NonNegative && R.NonNegative); 5148 } 5149 5150 // The result of a remainder can't be larger than the result of 5151 // either side. 5152 case BO_Rem: { 5153 // Don't 'pre-truncate' the operands. 5154 unsigned opWidth = C.getIntWidth(GetExprType(E)); 5155 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 5156 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 5157 5158 IntRange meet = IntRange::meet(L, R); 5159 meet.Width = std::min(meet.Width, MaxWidth); 5160 return meet; 5161 } 5162 5163 // The default behavior is okay for these. 5164 case BO_Mul: 5165 case BO_Add: 5166 case BO_Xor: 5167 case BO_Or: 5168 break; 5169 } 5170 5171 // The default case is to treat the operation as if it were closed 5172 // on the narrowest type that encompasses both operands. 5173 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 5174 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 5175 return IntRange::join(L, R); 5176 } 5177 5178 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 5179 switch (UO->getOpcode()) { 5180 // Boolean-valued operations are white-listed. 5181 case UO_LNot: 5182 return IntRange::forBoolType(); 5183 5184 // Operations with opaque sources are black-listed. 5185 case UO_Deref: 5186 case UO_AddrOf: // should be impossible 5187 return IntRange::forValueOfType(C, GetExprType(E)); 5188 5189 default: 5190 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 5191 } 5192 } 5193 5194 if (OpaqueValueExpr *OVE = dyn_cast<OpaqueValueExpr>(E)) 5195 return GetExprRange(C, OVE->getSourceExpr(), MaxWidth); 5196 5197 if (FieldDecl *BitField = E->getSourceBitField()) 5198 return IntRange(BitField->getBitWidthValue(C), 5199 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 5200 5201 return IntRange::forValueOfType(C, GetExprType(E)); 5202 } 5203 5204 static IntRange GetExprRange(ASTContext &C, Expr *E) { 5205 return GetExprRange(C, E, C.getIntWidth(GetExprType(E))); 5206 } 5207 5208 /// Checks whether the given value, which currently has the given 5209 /// source semantics, has the same value when coerced through the 5210 /// target semantics. 5211 static bool IsSameFloatAfterCast(const llvm::APFloat &value, 5212 const llvm::fltSemantics &Src, 5213 const llvm::fltSemantics &Tgt) { 5214 llvm::APFloat truncated = value; 5215 5216 bool ignored; 5217 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 5218 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 5219 5220 return truncated.bitwiseIsEqual(value); 5221 } 5222 5223 /// Checks whether the given value, which currently has the given 5224 /// source semantics, has the same value when coerced through the 5225 /// target semantics. 5226 /// 5227 /// The value might be a vector of floats (or a complex number). 5228 static bool IsSameFloatAfterCast(const APValue &value, 5229 const llvm::fltSemantics &Src, 5230 const llvm::fltSemantics &Tgt) { 5231 if (value.isFloat()) 5232 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 5233 5234 if (value.isVector()) { 5235 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 5236 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 5237 return false; 5238 return true; 5239 } 5240 5241 assert(value.isComplexFloat()); 5242 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 5243 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 5244 } 5245 5246 static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 5247 5248 static bool IsZero(Sema &S, Expr *E) { 5249 // Suppress cases where we are comparing against an enum constant. 5250 if (const DeclRefExpr *DR = 5251 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 5252 if (isa<EnumConstantDecl>(DR->getDecl())) 5253 return false; 5254 5255 // Suppress cases where the '0' value is expanded from a macro. 5256 if (E->getLocStart().isMacroID()) 5257 return false; 5258 5259 llvm::APSInt Value; 5260 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 5261 } 5262 5263 static bool HasEnumType(Expr *E) { 5264 // Strip off implicit integral promotions. 5265 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 5266 if (ICE->getCastKind() != CK_IntegralCast && 5267 ICE->getCastKind() != CK_NoOp) 5268 break; 5269 E = ICE->getSubExpr(); 5270 } 5271 5272 return E->getType()->isEnumeralType(); 5273 } 5274 5275 static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 5276 // Disable warning in template instantiations. 5277 if (!S.ActiveTemplateInstantiations.empty()) 5278 return; 5279 5280 BinaryOperatorKind op = E->getOpcode(); 5281 if (E->isValueDependent()) 5282 return; 5283 5284 if (op == BO_LT && IsZero(S, E->getRHS())) { 5285 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 5286 << "< 0" << "false" << HasEnumType(E->getLHS()) 5287 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 5288 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 5289 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 5290 << ">= 0" << "true" << HasEnumType(E->getLHS()) 5291 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 5292 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 5293 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 5294 << "0 >" << "false" << HasEnumType(E->getRHS()) 5295 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 5296 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 5297 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 5298 << "0 <=" << "true" << HasEnumType(E->getRHS()) 5299 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 5300 } 5301 } 5302 5303 static void DiagnoseOutOfRangeComparison(Sema &S, BinaryOperator *E, 5304 Expr *Constant, Expr *Other, 5305 llvm::APSInt Value, 5306 bool RhsConstant) { 5307 // Disable warning in template instantiations. 5308 if (!S.ActiveTemplateInstantiations.empty()) 5309 return; 5310 5311 // TODO: Investigate using GetExprRange() to get tighter bounds 5312 // on the bit ranges. 5313 QualType OtherT = Other->getType(); 5314 IntRange OtherRange = IntRange::forValueOfType(S.Context, OtherT); 5315 unsigned OtherWidth = OtherRange.Width; 5316 5317 bool OtherIsBooleanType = Other->isKnownToHaveBooleanValue(); 5318 5319 // 0 values are handled later by CheckTrivialUnsignedComparison(). 5320 if ((Value == 0) && (!OtherIsBooleanType)) 5321 return; 5322 5323 BinaryOperatorKind op = E->getOpcode(); 5324 bool IsTrue = true; 5325 5326 // Used for diagnostic printout. 5327 enum { 5328 LiteralConstant = 0, 5329 CXXBoolLiteralTrue, 5330 CXXBoolLiteralFalse 5331 } LiteralOrBoolConstant = LiteralConstant; 5332 5333 if (!OtherIsBooleanType) { 5334 QualType ConstantT = Constant->getType(); 5335 QualType CommonT = E->getLHS()->getType(); 5336 5337 if (S.Context.hasSameUnqualifiedType(OtherT, ConstantT)) 5338 return; 5339 assert((OtherT->isIntegerType() && ConstantT->isIntegerType()) && 5340 "comparison with non-integer type"); 5341 5342 bool ConstantSigned = ConstantT->isSignedIntegerType(); 5343 bool CommonSigned = CommonT->isSignedIntegerType(); 5344 5345 bool EqualityOnly = false; 5346 5347 if (CommonSigned) { 5348 // The common type is signed, therefore no signed to unsigned conversion. 5349 if (!OtherRange.NonNegative) { 5350 // Check that the constant is representable in type OtherT. 5351 if (ConstantSigned) { 5352 if (OtherWidth >= Value.getMinSignedBits()) 5353 return; 5354 } else { // !ConstantSigned 5355 if (OtherWidth >= Value.getActiveBits() + 1) 5356 return; 5357 } 5358 } else { // !OtherSigned 5359 // Check that the constant is representable in type OtherT. 5360 // Negative values are out of range. 5361 if (ConstantSigned) { 5362 if (Value.isNonNegative() && OtherWidth >= Value.getActiveBits()) 5363 return; 5364 } else { // !ConstantSigned 5365 if (OtherWidth >= Value.getActiveBits()) 5366 return; 5367 } 5368 } 5369 } else { // !CommonSigned 5370 if (OtherRange.NonNegative) { 5371 if (OtherWidth >= Value.getActiveBits()) 5372 return; 5373 } else { // OtherSigned 5374 assert(!ConstantSigned && 5375 "Two signed types converted to unsigned types."); 5376 // Check to see if the constant is representable in OtherT. 5377 if (OtherWidth > Value.getActiveBits()) 5378 return; 5379 // Check to see if the constant is equivalent to a negative value 5380 // cast to CommonT. 5381 if (S.Context.getIntWidth(ConstantT) == 5382 S.Context.getIntWidth(CommonT) && 5383 Value.isNegative() && Value.getMinSignedBits() <= OtherWidth) 5384 return; 5385 // The constant value rests between values that OtherT can represent 5386 // after conversion. Relational comparison still works, but equality 5387 // comparisons will be tautological. 5388 EqualityOnly = true; 5389 } 5390 } 5391 5392 bool PositiveConstant = !ConstantSigned || Value.isNonNegative(); 5393 5394 if (op == BO_EQ || op == BO_NE) { 5395 IsTrue = op == BO_NE; 5396 } else if (EqualityOnly) { 5397 return; 5398 } else if (RhsConstant) { 5399 if (op == BO_GT || op == BO_GE) 5400 IsTrue = !PositiveConstant; 5401 else // op == BO_LT || op == BO_LE 5402 IsTrue = PositiveConstant; 5403 } else { 5404 if (op == BO_LT || op == BO_LE) 5405 IsTrue = !PositiveConstant; 5406 else // op == BO_GT || op == BO_GE 5407 IsTrue = PositiveConstant; 5408 } 5409 } else { 5410 // Other isKnownToHaveBooleanValue 5411 enum CompareBoolWithConstantResult { AFals, ATrue, Unkwn }; 5412 enum ConstantValue { LT_Zero, Zero, One, GT_One, SizeOfConstVal }; 5413 enum ConstantSide { Lhs, Rhs, SizeOfConstSides }; 5414 5415 static const struct LinkedConditions { 5416 CompareBoolWithConstantResult BO_LT_OP[SizeOfConstSides][SizeOfConstVal]; 5417 CompareBoolWithConstantResult BO_GT_OP[SizeOfConstSides][SizeOfConstVal]; 5418 CompareBoolWithConstantResult BO_LE_OP[SizeOfConstSides][SizeOfConstVal]; 5419 CompareBoolWithConstantResult BO_GE_OP[SizeOfConstSides][SizeOfConstVal]; 5420 CompareBoolWithConstantResult BO_EQ_OP[SizeOfConstSides][SizeOfConstVal]; 5421 CompareBoolWithConstantResult BO_NE_OP[SizeOfConstSides][SizeOfConstVal]; 5422 5423 } TruthTable = { 5424 // Constant on LHS. | Constant on RHS. | 5425 // LT_Zero| Zero | One |GT_One| LT_Zero| Zero | One |GT_One| 5426 { { ATrue, Unkwn, AFals, AFals }, { AFals, AFals, Unkwn, ATrue } }, 5427 { { AFals, AFals, Unkwn, ATrue }, { ATrue, Unkwn, AFals, AFals } }, 5428 { { ATrue, ATrue, Unkwn, AFals }, { AFals, Unkwn, ATrue, ATrue } }, 5429 { { AFals, Unkwn, ATrue, ATrue }, { ATrue, ATrue, Unkwn, AFals } }, 5430 { { AFals, Unkwn, Unkwn, AFals }, { AFals, Unkwn, Unkwn, AFals } }, 5431 { { ATrue, Unkwn, Unkwn, ATrue }, { ATrue, Unkwn, Unkwn, ATrue } } 5432 }; 5433 5434 bool ConstantIsBoolLiteral = isa<CXXBoolLiteralExpr>(Constant); 5435 5436 enum ConstantValue ConstVal = Zero; 5437 if (Value.isUnsigned() || Value.isNonNegative()) { 5438 if (Value == 0) { 5439 LiteralOrBoolConstant = 5440 ConstantIsBoolLiteral ? CXXBoolLiteralFalse : LiteralConstant; 5441 ConstVal = Zero; 5442 } else if (Value == 1) { 5443 LiteralOrBoolConstant = 5444 ConstantIsBoolLiteral ? CXXBoolLiteralTrue : LiteralConstant; 5445 ConstVal = One; 5446 } else { 5447 LiteralOrBoolConstant = LiteralConstant; 5448 ConstVal = GT_One; 5449 } 5450 } else { 5451 ConstVal = LT_Zero; 5452 } 5453 5454 CompareBoolWithConstantResult CmpRes; 5455 5456 switch (op) { 5457 case BO_LT: 5458 CmpRes = TruthTable.BO_LT_OP[RhsConstant][ConstVal]; 5459 break; 5460 case BO_GT: 5461 CmpRes = TruthTable.BO_GT_OP[RhsConstant][ConstVal]; 5462 break; 5463 case BO_LE: 5464 CmpRes = TruthTable.BO_LE_OP[RhsConstant][ConstVal]; 5465 break; 5466 case BO_GE: 5467 CmpRes = TruthTable.BO_GE_OP[RhsConstant][ConstVal]; 5468 break; 5469 case BO_EQ: 5470 CmpRes = TruthTable.BO_EQ_OP[RhsConstant][ConstVal]; 5471 break; 5472 case BO_NE: 5473 CmpRes = TruthTable.BO_NE_OP[RhsConstant][ConstVal]; 5474 break; 5475 default: 5476 CmpRes = Unkwn; 5477 break; 5478 } 5479 5480 if (CmpRes == AFals) { 5481 IsTrue = false; 5482 } else if (CmpRes == ATrue) { 5483 IsTrue = true; 5484 } else { 5485 return; 5486 } 5487 } 5488 5489 // If this is a comparison to an enum constant, include that 5490 // constant in the diagnostic. 5491 const EnumConstantDecl *ED = nullptr; 5492 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Constant)) 5493 ED = dyn_cast<EnumConstantDecl>(DR->getDecl()); 5494 5495 SmallString<64> PrettySourceValue; 5496 llvm::raw_svector_ostream OS(PrettySourceValue); 5497 if (ED) 5498 OS << '\'' << *ED << "' (" << Value << ")"; 5499 else 5500 OS << Value; 5501 5502 S.DiagRuntimeBehavior( 5503 E->getOperatorLoc(), E, 5504 S.PDiag(diag::warn_out_of_range_compare) 5505 << OS.str() << LiteralOrBoolConstant 5506 << OtherT << (OtherIsBooleanType && !OtherT->isBooleanType()) << IsTrue 5507 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange()); 5508 } 5509 5510 /// Analyze the operands of the given comparison. Implements the 5511 /// fallback case from AnalyzeComparison. 5512 static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 5513 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 5514 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 5515 } 5516 5517 /// \brief Implements -Wsign-compare. 5518 /// 5519 /// \param E the binary operator to check for warnings 5520 static void AnalyzeComparison(Sema &S, BinaryOperator *E) { 5521 // The type the comparison is being performed in. 5522 QualType T = E->getLHS()->getType(); 5523 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 5524 && "comparison with mismatched types"); 5525 if (E->isValueDependent()) 5526 return AnalyzeImpConvsInComparison(S, E); 5527 5528 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 5529 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 5530 5531 bool IsComparisonConstant = false; 5532 5533 // Check whether an integer constant comparison results in a value 5534 // of 'true' or 'false'. 5535 if (T->isIntegralType(S.Context)) { 5536 llvm::APSInt RHSValue; 5537 bool IsRHSIntegralLiteral = 5538 RHS->isIntegerConstantExpr(RHSValue, S.Context); 5539 llvm::APSInt LHSValue; 5540 bool IsLHSIntegralLiteral = 5541 LHS->isIntegerConstantExpr(LHSValue, S.Context); 5542 if (IsRHSIntegralLiteral && !IsLHSIntegralLiteral) 5543 DiagnoseOutOfRangeComparison(S, E, RHS, LHS, RHSValue, true); 5544 else if (!IsRHSIntegralLiteral && IsLHSIntegralLiteral) 5545 DiagnoseOutOfRangeComparison(S, E, LHS, RHS, LHSValue, false); 5546 else 5547 IsComparisonConstant = 5548 (IsRHSIntegralLiteral && IsLHSIntegralLiteral); 5549 } else if (!T->hasUnsignedIntegerRepresentation()) 5550 IsComparisonConstant = E->isIntegerConstantExpr(S.Context); 5551 5552 // We don't do anything special if this isn't an unsigned integral 5553 // comparison: we're only interested in integral comparisons, and 5554 // signed comparisons only happen in cases we don't care to warn about. 5555 // 5556 // We also don't care about value-dependent expressions or expressions 5557 // whose result is a constant. 5558 if (!T->hasUnsignedIntegerRepresentation() || IsComparisonConstant) 5559 return AnalyzeImpConvsInComparison(S, E); 5560 5561 // Check to see if one of the (unmodified) operands is of different 5562 // signedness. 5563 Expr *signedOperand, *unsignedOperand; 5564 if (LHS->getType()->hasSignedIntegerRepresentation()) { 5565 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 5566 "unsigned comparison between two signed integer expressions?"); 5567 signedOperand = LHS; 5568 unsignedOperand = RHS; 5569 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 5570 signedOperand = RHS; 5571 unsignedOperand = LHS; 5572 } else { 5573 CheckTrivialUnsignedComparison(S, E); 5574 return AnalyzeImpConvsInComparison(S, E); 5575 } 5576 5577 // Otherwise, calculate the effective range of the signed operand. 5578 IntRange signedRange = GetExprRange(S.Context, signedOperand); 5579 5580 // Go ahead and analyze implicit conversions in the operands. Note 5581 // that we skip the implicit conversions on both sides. 5582 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 5583 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 5584 5585 // If the signed range is non-negative, -Wsign-compare won't fire, 5586 // but we should still check for comparisons which are always true 5587 // or false. 5588 if (signedRange.NonNegative) 5589 return CheckTrivialUnsignedComparison(S, E); 5590 5591 // For (in)equality comparisons, if the unsigned operand is a 5592 // constant which cannot collide with a overflowed signed operand, 5593 // then reinterpreting the signed operand as unsigned will not 5594 // change the result of the comparison. 5595 if (E->isEqualityOp()) { 5596 unsigned comparisonWidth = S.Context.getIntWidth(T); 5597 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 5598 5599 // We should never be unable to prove that the unsigned operand is 5600 // non-negative. 5601 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 5602 5603 if (unsignedRange.Width < comparisonWidth) 5604 return; 5605 } 5606 5607 S.DiagRuntimeBehavior(E->getOperatorLoc(), E, 5608 S.PDiag(diag::warn_mixed_sign_comparison) 5609 << LHS->getType() << RHS->getType() 5610 << LHS->getSourceRange() << RHS->getSourceRange()); 5611 } 5612 5613 /// Analyzes an attempt to assign the given value to a bitfield. 5614 /// 5615 /// Returns true if there was something fishy about the attempt. 5616 static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 5617 SourceLocation InitLoc) { 5618 assert(Bitfield->isBitField()); 5619 if (Bitfield->isInvalidDecl()) 5620 return false; 5621 5622 // White-list bool bitfields. 5623 if (Bitfield->getType()->isBooleanType()) 5624 return false; 5625 5626 // Ignore value- or type-dependent expressions. 5627 if (Bitfield->getBitWidth()->isValueDependent() || 5628 Bitfield->getBitWidth()->isTypeDependent() || 5629 Init->isValueDependent() || 5630 Init->isTypeDependent()) 5631 return false; 5632 5633 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 5634 5635 llvm::APSInt Value; 5636 if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects)) 5637 return false; 5638 5639 unsigned OriginalWidth = Value.getBitWidth(); 5640 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 5641 5642 if (OriginalWidth <= FieldWidth) 5643 return false; 5644 5645 // Compute the value which the bitfield will contain. 5646 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 5647 TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType()); 5648 5649 // Check whether the stored value is equal to the original value. 5650 TruncatedValue = TruncatedValue.extend(OriginalWidth); 5651 if (llvm::APSInt::isSameValue(Value, TruncatedValue)) 5652 return false; 5653 5654 // Special-case bitfields of width 1: booleans are naturally 0/1, and 5655 // therefore don't strictly fit into a signed bitfield of width 1. 5656 if (FieldWidth == 1 && Value == 1) 5657 return false; 5658 5659 std::string PrettyValue = Value.toString(10); 5660 std::string PrettyTrunc = TruncatedValue.toString(10); 5661 5662 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 5663 << PrettyValue << PrettyTrunc << OriginalInit->getType() 5664 << Init->getSourceRange(); 5665 5666 return true; 5667 } 5668 5669 /// Analyze the given simple or compound assignment for warning-worthy 5670 /// operations. 5671 static void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 5672 // Just recurse on the LHS. 5673 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 5674 5675 // We want to recurse on the RHS as normal unless we're assigning to 5676 // a bitfield. 5677 if (FieldDecl *Bitfield = E->getLHS()->getSourceBitField()) { 5678 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 5679 E->getOperatorLoc())) { 5680 // Recurse, ignoring any implicit conversions on the RHS. 5681 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 5682 E->getOperatorLoc()); 5683 } 5684 } 5685 5686 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 5687 } 5688 5689 /// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 5690 static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 5691 SourceLocation CContext, unsigned diag, 5692 bool pruneControlFlow = false) { 5693 if (pruneControlFlow) { 5694 S.DiagRuntimeBehavior(E->getExprLoc(), E, 5695 S.PDiag(diag) 5696 << SourceType << T << E->getSourceRange() 5697 << SourceRange(CContext)); 5698 return; 5699 } 5700 S.Diag(E->getExprLoc(), diag) 5701 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 5702 } 5703 5704 /// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 5705 static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, 5706 SourceLocation CContext, unsigned diag, 5707 bool pruneControlFlow = false) { 5708 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow); 5709 } 5710 5711 /// Diagnose an implicit cast from a literal expression. Does not warn when the 5712 /// cast wouldn't lose information. 5713 void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 5714 SourceLocation CContext) { 5715 // Try to convert the literal exactly to an integer. If we can, don't warn. 5716 bool isExact = false; 5717 const llvm::APFloat &Value = FL->getValue(); 5718 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 5719 T->hasUnsignedIntegerRepresentation()); 5720 if (Value.convertToInteger(IntegerValue, 5721 llvm::APFloat::rmTowardZero, &isExact) 5722 == llvm::APFloat::opOK && isExact) 5723 return; 5724 5725 // FIXME: Force the precision of the source value down so we don't print 5726 // digits which are usually useless (we don't really care here if we 5727 // truncate a digit by accident in edge cases). Ideally, APFloat::toString 5728 // would automatically print the shortest representation, but it's a bit 5729 // tricky to implement. 5730 SmallString<16> PrettySourceValue; 5731 unsigned precision = llvm::APFloat::semanticsPrecision(Value.getSemantics()); 5732 precision = (precision * 59 + 195) / 196; 5733 Value.toString(PrettySourceValue, precision); 5734 5735 SmallString<16> PrettyTargetValue; 5736 if (T->isSpecificBuiltinType(BuiltinType::Bool)) 5737 PrettyTargetValue = IntegerValue == 0 ? "false" : "true"; 5738 else 5739 IntegerValue.toString(PrettyTargetValue); 5740 5741 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 5742 << FL->getType() << T.getUnqualifiedType() << PrettySourceValue 5743 << PrettyTargetValue << FL->getSourceRange() << SourceRange(CContext); 5744 } 5745 5746 std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 5747 if (!Range.Width) return "0"; 5748 5749 llvm::APSInt ValueInRange = Value; 5750 ValueInRange.setIsSigned(!Range.NonNegative); 5751 ValueInRange = ValueInRange.trunc(Range.Width); 5752 return ValueInRange.toString(10); 5753 } 5754 5755 static bool IsImplicitBoolFloatConversion(Sema &S, Expr *Ex, bool ToBool) { 5756 if (!isa<ImplicitCastExpr>(Ex)) 5757 return false; 5758 5759 Expr *InnerE = Ex->IgnoreParenImpCasts(); 5760 const Type *Target = S.Context.getCanonicalType(Ex->getType()).getTypePtr(); 5761 const Type *Source = 5762 S.Context.getCanonicalType(InnerE->getType()).getTypePtr(); 5763 if (Target->isDependentType()) 5764 return false; 5765 5766 const BuiltinType *FloatCandidateBT = 5767 dyn_cast<BuiltinType>(ToBool ? Source : Target); 5768 const Type *BoolCandidateType = ToBool ? Target : Source; 5769 5770 return (BoolCandidateType->isSpecificBuiltinType(BuiltinType::Bool) && 5771 FloatCandidateBT && (FloatCandidateBT->isFloatingPoint())); 5772 } 5773 5774 void CheckImplicitArgumentConversions(Sema &S, CallExpr *TheCall, 5775 SourceLocation CC) { 5776 unsigned NumArgs = TheCall->getNumArgs(); 5777 for (unsigned i = 0; i < NumArgs; ++i) { 5778 Expr *CurrA = TheCall->getArg(i); 5779 if (!IsImplicitBoolFloatConversion(S, CurrA, true)) 5780 continue; 5781 5782 bool IsSwapped = ((i > 0) && 5783 IsImplicitBoolFloatConversion(S, TheCall->getArg(i - 1), false)); 5784 IsSwapped |= ((i < (NumArgs - 1)) && 5785 IsImplicitBoolFloatConversion(S, TheCall->getArg(i + 1), false)); 5786 if (IsSwapped) { 5787 // Warn on this floating-point to bool conversion. 5788 DiagnoseImpCast(S, CurrA->IgnoreParenImpCasts(), 5789 CurrA->getType(), CC, 5790 diag::warn_impcast_floating_point_to_bool); 5791 } 5792 } 5793 } 5794 5795 void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 5796 SourceLocation CC, bool *ICContext = nullptr) { 5797 if (E->isTypeDependent() || E->isValueDependent()) return; 5798 5799 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 5800 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 5801 if (Source == Target) return; 5802 if (Target->isDependentType()) return; 5803 5804 // If the conversion context location is invalid don't complain. We also 5805 // don't want to emit a warning if the issue occurs from the expansion of 5806 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 5807 // delay this check as long as possible. Once we detect we are in that 5808 // scenario, we just return. 5809 if (CC.isInvalid()) 5810 return; 5811 5812 // Diagnose implicit casts to bool. 5813 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 5814 if (isa<StringLiteral>(E)) 5815 // Warn on string literal to bool. Checks for string literals in logical 5816 // and expressions, for instance, assert(0 && "error here"), are 5817 // prevented by a check in AnalyzeImplicitConversions(). 5818 return DiagnoseImpCast(S, E, T, CC, 5819 diag::warn_impcast_string_literal_to_bool); 5820 if (isa<ObjCStringLiteral>(E) || isa<ObjCArrayLiteral>(E) || 5821 isa<ObjCDictionaryLiteral>(E) || isa<ObjCBoxedExpr>(E)) { 5822 // This covers the literal expressions that evaluate to Objective-C 5823 // objects. 5824 return DiagnoseImpCast(S, E, T, CC, 5825 diag::warn_impcast_objective_c_literal_to_bool); 5826 } 5827 if (Source->isPointerType() || Source->canDecayToPointerType()) { 5828 // Warn on pointer to bool conversion that is always true. 5829 S.DiagnoseAlwaysNonNullPointer(E, Expr::NPCK_NotNull, /*IsEqual*/ false, 5830 SourceRange(CC)); 5831 } 5832 } 5833 5834 // Strip vector types. 5835 if (isa<VectorType>(Source)) { 5836 if (!isa<VectorType>(Target)) { 5837 if (S.SourceMgr.isInSystemMacro(CC)) 5838 return; 5839 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 5840 } 5841 5842 // If the vector cast is cast between two vectors of the same size, it is 5843 // a bitcast, not a conversion. 5844 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 5845 return; 5846 5847 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 5848 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 5849 } 5850 if (auto VecTy = dyn_cast<VectorType>(Target)) 5851 Target = VecTy->getElementType().getTypePtr(); 5852 5853 // Strip complex types. 5854 if (isa<ComplexType>(Source)) { 5855 if (!isa<ComplexType>(Target)) { 5856 if (S.SourceMgr.isInSystemMacro(CC)) 5857 return; 5858 5859 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 5860 } 5861 5862 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 5863 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 5864 } 5865 5866 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 5867 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 5868 5869 // If the source is floating point... 5870 if (SourceBT && SourceBT->isFloatingPoint()) { 5871 // ...and the target is floating point... 5872 if (TargetBT && TargetBT->isFloatingPoint()) { 5873 // ...then warn if we're dropping FP rank. 5874 5875 // Builtin FP kinds are ordered by increasing FP rank. 5876 if (SourceBT->getKind() > TargetBT->getKind()) { 5877 // Don't warn about float constants that are precisely 5878 // representable in the target type. 5879 Expr::EvalResult result; 5880 if (E->EvaluateAsRValue(result, S.Context)) { 5881 // Value might be a float, a float vector, or a float complex. 5882 if (IsSameFloatAfterCast(result.Val, 5883 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 5884 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 5885 return; 5886 } 5887 5888 if (S.SourceMgr.isInSystemMacro(CC)) 5889 return; 5890 5891 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 5892 } 5893 return; 5894 } 5895 5896 // If the target is integral, always warn. 5897 if (TargetBT && TargetBT->isInteger()) { 5898 if (S.SourceMgr.isInSystemMacro(CC)) 5899 return; 5900 5901 Expr *InnerE = E->IgnoreParenImpCasts(); 5902 // We also want to warn on, e.g., "int i = -1.234" 5903 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 5904 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 5905 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 5906 5907 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 5908 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 5909 } else { 5910 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 5911 } 5912 } 5913 5914 // If the target is bool, warn if expr is a function or method call. 5915 if (Target->isSpecificBuiltinType(BuiltinType::Bool) && 5916 isa<CallExpr>(E)) { 5917 // Check last argument of function call to see if it is an 5918 // implicit cast from a type matching the type the result 5919 // is being cast to. 5920 CallExpr *CEx = cast<CallExpr>(E); 5921 unsigned NumArgs = CEx->getNumArgs(); 5922 if (NumArgs > 0) { 5923 Expr *LastA = CEx->getArg(NumArgs - 1); 5924 Expr *InnerE = LastA->IgnoreParenImpCasts(); 5925 const Type *InnerType = 5926 S.Context.getCanonicalType(InnerE->getType()).getTypePtr(); 5927 if (isa<ImplicitCastExpr>(LastA) && (InnerType == Target)) { 5928 // Warn on this floating-point to bool conversion 5929 DiagnoseImpCast(S, E, T, CC, 5930 diag::warn_impcast_floating_point_to_bool); 5931 } 5932 } 5933 } 5934 return; 5935 } 5936 5937 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 5938 == Expr::NPCK_GNUNull) && !Target->isAnyPointerType() 5939 && !Target->isBlockPointerType() && !Target->isMemberPointerType() 5940 && Target->isScalarType() && !Target->isNullPtrType()) { 5941 SourceLocation Loc = E->getSourceRange().getBegin(); 5942 if (Loc.isMacroID()) 5943 Loc = S.SourceMgr.getImmediateExpansionRange(Loc).first; 5944 if (!Loc.isMacroID() || CC.isMacroID()) 5945 S.Diag(Loc, diag::warn_impcast_null_pointer_to_integer) 5946 << T << clang::SourceRange(CC) 5947 << FixItHint::CreateReplacement(Loc, 5948 S.getFixItZeroLiteralForType(T, Loc)); 5949 } 5950 5951 if (!Source->isIntegerType() || !Target->isIntegerType()) 5952 return; 5953 5954 // TODO: remove this early return once the false positives for constant->bool 5955 // in templates, macros, etc, are reduced or removed. 5956 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 5957 return; 5958 5959 IntRange SourceRange = GetExprRange(S.Context, E); 5960 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 5961 5962 if (SourceRange.Width > TargetRange.Width) { 5963 // If the source is a constant, use a default-on diagnostic. 5964 // TODO: this should happen for bitfield stores, too. 5965 llvm::APSInt Value(32); 5966 if (E->isIntegerConstantExpr(Value, S.Context)) { 5967 if (S.SourceMgr.isInSystemMacro(CC)) 5968 return; 5969 5970 std::string PrettySourceValue = Value.toString(10); 5971 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 5972 5973 S.DiagRuntimeBehavior(E->getExprLoc(), E, 5974 S.PDiag(diag::warn_impcast_integer_precision_constant) 5975 << PrettySourceValue << PrettyTargetValue 5976 << E->getType() << T << E->getSourceRange() 5977 << clang::SourceRange(CC)); 5978 return; 5979 } 5980 5981 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 5982 if (S.SourceMgr.isInSystemMacro(CC)) 5983 return; 5984 5985 if (TargetRange.Width == 32 && S.Context.getIntWidth(E->getType()) == 64) 5986 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32, 5987 /* pruneControlFlow */ true); 5988 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 5989 } 5990 5991 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 5992 (!TargetRange.NonNegative && SourceRange.NonNegative && 5993 SourceRange.Width == TargetRange.Width)) { 5994 5995 if (S.SourceMgr.isInSystemMacro(CC)) 5996 return; 5997 5998 unsigned DiagID = diag::warn_impcast_integer_sign; 5999 6000 // Traditionally, gcc has warned about this under -Wsign-compare. 6001 // We also want to warn about it in -Wconversion. 6002 // So if -Wconversion is off, use a completely identical diagnostic 6003 // in the sign-compare group. 6004 // The conditional-checking code will 6005 if (ICContext) { 6006 DiagID = diag::warn_impcast_integer_sign_conditional; 6007 *ICContext = true; 6008 } 6009 6010 return DiagnoseImpCast(S, E, T, CC, DiagID); 6011 } 6012 6013 // Diagnose conversions between different enumeration types. 6014 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 6015 // type, to give us better diagnostics. 6016 QualType SourceType = E->getType(); 6017 if (!S.getLangOpts().CPlusPlus) { 6018 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 6019 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 6020 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 6021 SourceType = S.Context.getTypeDeclType(Enum); 6022 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 6023 } 6024 } 6025 6026 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 6027 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 6028 if (SourceEnum->getDecl()->hasNameForLinkage() && 6029 TargetEnum->getDecl()->hasNameForLinkage() && 6030 SourceEnum != TargetEnum) { 6031 if (S.SourceMgr.isInSystemMacro(CC)) 6032 return; 6033 6034 return DiagnoseImpCast(S, E, SourceType, T, CC, 6035 diag::warn_impcast_different_enum_types); 6036 } 6037 6038 return; 6039 } 6040 6041 void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 6042 SourceLocation CC, QualType T); 6043 6044 void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 6045 SourceLocation CC, bool &ICContext) { 6046 E = E->IgnoreParenImpCasts(); 6047 6048 if (isa<ConditionalOperator>(E)) 6049 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), CC, T); 6050 6051 AnalyzeImplicitConversions(S, E, CC); 6052 if (E->getType() != T) 6053 return CheckImplicitConversion(S, E, T, CC, &ICContext); 6054 return; 6055 } 6056 6057 void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 6058 SourceLocation CC, QualType T) { 6059 AnalyzeImplicitConversions(S, E->getCond(), CC); 6060 6061 bool Suspicious = false; 6062 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 6063 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 6064 6065 // If -Wconversion would have warned about either of the candidates 6066 // for a signedness conversion to the context type... 6067 if (!Suspicious) return; 6068 6069 // ...but it's currently ignored... 6070 if (!S.Diags.isIgnored(diag::warn_impcast_integer_sign_conditional, CC)) 6071 return; 6072 6073 // ...then check whether it would have warned about either of the 6074 // candidates for a signedness conversion to the condition type. 6075 if (E->getType() == T) return; 6076 6077 Suspicious = false; 6078 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 6079 E->getType(), CC, &Suspicious); 6080 if (!Suspicious) 6081 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 6082 E->getType(), CC, &Suspicious); 6083 } 6084 6085 /// AnalyzeImplicitConversions - Find and report any interesting 6086 /// implicit conversions in the given expression. There are a couple 6087 /// of competing diagnostics here, -Wconversion and -Wsign-compare. 6088 void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 6089 QualType T = OrigE->getType(); 6090 Expr *E = OrigE->IgnoreParenImpCasts(); 6091 6092 if (E->isTypeDependent() || E->isValueDependent()) 6093 return; 6094 6095 // For conditional operators, we analyze the arguments as if they 6096 // were being fed directly into the output. 6097 if (isa<ConditionalOperator>(E)) { 6098 ConditionalOperator *CO = cast<ConditionalOperator>(E); 6099 CheckConditionalOperator(S, CO, CC, T); 6100 return; 6101 } 6102 6103 // Check implicit argument conversions for function calls. 6104 if (CallExpr *Call = dyn_cast<CallExpr>(E)) 6105 CheckImplicitArgumentConversions(S, Call, CC); 6106 6107 // Go ahead and check any implicit conversions we might have skipped. 6108 // The non-canonical typecheck is just an optimization; 6109 // CheckImplicitConversion will filter out dead implicit conversions. 6110 if (E->getType() != T) 6111 CheckImplicitConversion(S, E, T, CC); 6112 6113 // Now continue drilling into this expression. 6114 6115 if (PseudoObjectExpr * POE = dyn_cast<PseudoObjectExpr>(E)) { 6116 if (POE->getResultExpr()) 6117 E = POE->getResultExpr(); 6118 } 6119 6120 if (const OpaqueValueExpr *OVE = dyn_cast<OpaqueValueExpr>(E)) 6121 return AnalyzeImplicitConversions(S, OVE->getSourceExpr(), CC); 6122 6123 // Skip past explicit casts. 6124 if (isa<ExplicitCastExpr>(E)) { 6125 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 6126 return AnalyzeImplicitConversions(S, E, CC); 6127 } 6128 6129 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 6130 // Do a somewhat different check with comparison operators. 6131 if (BO->isComparisonOp()) 6132 return AnalyzeComparison(S, BO); 6133 6134 // And with simple assignments. 6135 if (BO->getOpcode() == BO_Assign) 6136 return AnalyzeAssignment(S, BO); 6137 } 6138 6139 // These break the otherwise-useful invariant below. Fortunately, 6140 // we don't really need to recurse into them, because any internal 6141 // expressions should have been analyzed already when they were 6142 // built into statements. 6143 if (isa<StmtExpr>(E)) return; 6144 6145 // Don't descend into unevaluated contexts. 6146 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 6147 6148 // Now just recurse over the expression's children. 6149 CC = E->getExprLoc(); 6150 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 6151 bool IsLogicalAndOperator = BO && BO->getOpcode() == BO_LAnd; 6152 for (Stmt::child_range I = E->children(); I; ++I) { 6153 Expr *ChildExpr = dyn_cast_or_null<Expr>(*I); 6154 if (!ChildExpr) 6155 continue; 6156 6157 if (IsLogicalAndOperator && 6158 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 6159 // Ignore checking string literals that are in logical and operators. 6160 // This is a common pattern for asserts. 6161 continue; 6162 AnalyzeImplicitConversions(S, ChildExpr, CC); 6163 } 6164 } 6165 6166 } // end anonymous namespace 6167 6168 enum { 6169 AddressOf, 6170 FunctionPointer, 6171 ArrayPointer 6172 }; 6173 6174 /// \brief Diagnose pointers that are always non-null. 6175 /// \param E the expression containing the pointer 6176 /// \param NullKind NPCK_NotNull if E is a cast to bool, otherwise, E is 6177 /// compared to a null pointer 6178 /// \param IsEqual True when the comparison is equal to a null pointer 6179 /// \param Range Extra SourceRange to highlight in the diagnostic 6180 void Sema::DiagnoseAlwaysNonNullPointer(Expr *E, 6181 Expr::NullPointerConstantKind NullKind, 6182 bool IsEqual, SourceRange Range) { 6183 if (!E) 6184 return; 6185 6186 // Don't warn inside macros. 6187 if (E->getExprLoc().isMacroID()) 6188 return; 6189 E = E->IgnoreImpCasts(); 6190 6191 const bool IsCompare = NullKind != Expr::NPCK_NotNull; 6192 6193 if (isa<CXXThisExpr>(E)) { 6194 unsigned DiagID = IsCompare ? diag::warn_this_null_compare 6195 : diag::warn_this_bool_conversion; 6196 Diag(E->getExprLoc(), DiagID) << E->getSourceRange() << Range << IsEqual; 6197 return; 6198 } 6199 6200 bool IsAddressOf = false; 6201 6202 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 6203 if (UO->getOpcode() != UO_AddrOf) 6204 return; 6205 IsAddressOf = true; 6206 E = UO->getSubExpr(); 6207 } 6208 6209 // Expect to find a single Decl. Skip anything more complicated. 6210 ValueDecl *D = nullptr; 6211 if (DeclRefExpr *R = dyn_cast<DeclRefExpr>(E)) { 6212 D = R->getDecl(); 6213 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 6214 D = M->getMemberDecl(); 6215 } 6216 6217 // Weak Decls can be null. 6218 if (!D || D->isWeak()) 6219 return; 6220 6221 QualType T = D->getType(); 6222 const bool IsArray = T->isArrayType(); 6223 const bool IsFunction = T->isFunctionType(); 6224 6225 if (IsAddressOf) { 6226 // Address of function is used to silence the function warning. 6227 if (IsFunction) 6228 return; 6229 6230 if (T->isReferenceType()) { 6231 unsigned DiagID = IsCompare 6232 ? diag::warn_address_of_reference_null_compare 6233 : diag::warn_address_of_reference_bool_conversion; 6234 Diag(E->getExprLoc(), DiagID) << E->getSourceRange() << Range << IsEqual; 6235 return; 6236 } 6237 } 6238 6239 // Found nothing. 6240 if (!IsAddressOf && !IsFunction && !IsArray) 6241 return; 6242 6243 // Pretty print the expression for the diagnostic. 6244 std::string Str; 6245 llvm::raw_string_ostream S(Str); 6246 E->printPretty(S, nullptr, getPrintingPolicy()); 6247 6248 unsigned DiagID = IsCompare ? diag::warn_null_pointer_compare 6249 : diag::warn_impcast_pointer_to_bool; 6250 unsigned DiagType; 6251 if (IsAddressOf) 6252 DiagType = AddressOf; 6253 else if (IsFunction) 6254 DiagType = FunctionPointer; 6255 else if (IsArray) 6256 DiagType = ArrayPointer; 6257 else 6258 llvm_unreachable("Could not determine diagnostic."); 6259 Diag(E->getExprLoc(), DiagID) << DiagType << S.str() << E->getSourceRange() 6260 << Range << IsEqual; 6261 6262 if (!IsFunction) 6263 return; 6264 6265 // Suggest '&' to silence the function warning. 6266 Diag(E->getExprLoc(), diag::note_function_warning_silence) 6267 << FixItHint::CreateInsertion(E->getLocStart(), "&"); 6268 6269 // Check to see if '()' fixit should be emitted. 6270 QualType ReturnType; 6271 UnresolvedSet<4> NonTemplateOverloads; 6272 tryExprAsCall(*E, ReturnType, NonTemplateOverloads); 6273 if (ReturnType.isNull()) 6274 return; 6275 6276 if (IsCompare) { 6277 // There are two cases here. If there is null constant, the only suggest 6278 // for a pointer return type. If the null is 0, then suggest if the return 6279 // type is a pointer or an integer type. 6280 if (!ReturnType->isPointerType()) { 6281 if (NullKind == Expr::NPCK_ZeroExpression || 6282 NullKind == Expr::NPCK_ZeroLiteral) { 6283 if (!ReturnType->isIntegerType()) 6284 return; 6285 } else { 6286 return; 6287 } 6288 } 6289 } else { // !IsCompare 6290 // For function to bool, only suggest if the function pointer has bool 6291 // return type. 6292 if (!ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 6293 return; 6294 } 6295 Diag(E->getExprLoc(), diag::note_function_to_function_call) 6296 << FixItHint::CreateInsertion(getLocForEndOfToken(E->getLocEnd()), "()"); 6297 } 6298 6299 6300 /// Diagnoses "dangerous" implicit conversions within the given 6301 /// expression (which is a full expression). Implements -Wconversion 6302 /// and -Wsign-compare. 6303 /// 6304 /// \param CC the "context" location of the implicit conversion, i.e. 6305 /// the most location of the syntactic entity requiring the implicit 6306 /// conversion 6307 void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 6308 // Don't diagnose in unevaluated contexts. 6309 if (isUnevaluatedContext()) 6310 return; 6311 6312 // Don't diagnose for value- or type-dependent expressions. 6313 if (E->isTypeDependent() || E->isValueDependent()) 6314 return; 6315 6316 // Check for array bounds violations in cases where the check isn't triggered 6317 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 6318 // ArraySubscriptExpr is on the RHS of a variable initialization. 6319 CheckArrayAccess(E); 6320 6321 // This is not the right CC for (e.g.) a variable initialization. 6322 AnalyzeImplicitConversions(*this, E, CC); 6323 } 6324 6325 /// Diagnose when expression is an integer constant expression and its evaluation 6326 /// results in integer overflow 6327 void Sema::CheckForIntOverflow (Expr *E) { 6328 if (isa<BinaryOperator>(E->IgnoreParens())) 6329 E->EvaluateForOverflow(Context); 6330 } 6331 6332 namespace { 6333 /// \brief Visitor for expressions which looks for unsequenced operations on the 6334 /// same object. 6335 class SequenceChecker : public EvaluatedExprVisitor<SequenceChecker> { 6336 typedef EvaluatedExprVisitor<SequenceChecker> Base; 6337 6338 /// \brief A tree of sequenced regions within an expression. Two regions are 6339 /// unsequenced if one is an ancestor or a descendent of the other. When we 6340 /// finish processing an expression with sequencing, such as a comma 6341 /// expression, we fold its tree nodes into its parent, since they are 6342 /// unsequenced with respect to nodes we will visit later. 6343 class SequenceTree { 6344 struct Value { 6345 explicit Value(unsigned Parent) : Parent(Parent), Merged(false) {} 6346 unsigned Parent : 31; 6347 bool Merged : 1; 6348 }; 6349 SmallVector<Value, 8> Values; 6350 6351 public: 6352 /// \brief A region within an expression which may be sequenced with respect 6353 /// to some other region. 6354 class Seq { 6355 explicit Seq(unsigned N) : Index(N) {} 6356 unsigned Index; 6357 friend class SequenceTree; 6358 public: 6359 Seq() : Index(0) {} 6360 }; 6361 6362 SequenceTree() { Values.push_back(Value(0)); } 6363 Seq root() const { return Seq(0); } 6364 6365 /// \brief Create a new sequence of operations, which is an unsequenced 6366 /// subset of \p Parent. This sequence of operations is sequenced with 6367 /// respect to other children of \p Parent. 6368 Seq allocate(Seq Parent) { 6369 Values.push_back(Value(Parent.Index)); 6370 return Seq(Values.size() - 1); 6371 } 6372 6373 /// \brief Merge a sequence of operations into its parent. 6374 void merge(Seq S) { 6375 Values[S.Index].Merged = true; 6376 } 6377 6378 /// \brief Determine whether two operations are unsequenced. This operation 6379 /// is asymmetric: \p Cur should be the more recent sequence, and \p Old 6380 /// should have been merged into its parent as appropriate. 6381 bool isUnsequenced(Seq Cur, Seq Old) { 6382 unsigned C = representative(Cur.Index); 6383 unsigned Target = representative(Old.Index); 6384 while (C >= Target) { 6385 if (C == Target) 6386 return true; 6387 C = Values[C].Parent; 6388 } 6389 return false; 6390 } 6391 6392 private: 6393 /// \brief Pick a representative for a sequence. 6394 unsigned representative(unsigned K) { 6395 if (Values[K].Merged) 6396 // Perform path compression as we go. 6397 return Values[K].Parent = representative(Values[K].Parent); 6398 return K; 6399 } 6400 }; 6401 6402 /// An object for which we can track unsequenced uses. 6403 typedef NamedDecl *Object; 6404 6405 /// Different flavors of object usage which we track. We only track the 6406 /// least-sequenced usage of each kind. 6407 enum UsageKind { 6408 /// A read of an object. Multiple unsequenced reads are OK. 6409 UK_Use, 6410 /// A modification of an object which is sequenced before the value 6411 /// computation of the expression, such as ++n in C++. 6412 UK_ModAsValue, 6413 /// A modification of an object which is not sequenced before the value 6414 /// computation of the expression, such as n++. 6415 UK_ModAsSideEffect, 6416 6417 UK_Count = UK_ModAsSideEffect + 1 6418 }; 6419 6420 struct Usage { 6421 Usage() : Use(nullptr), Seq() {} 6422 Expr *Use; 6423 SequenceTree::Seq Seq; 6424 }; 6425 6426 struct UsageInfo { 6427 UsageInfo() : Diagnosed(false) {} 6428 Usage Uses[UK_Count]; 6429 /// Have we issued a diagnostic for this variable already? 6430 bool Diagnosed; 6431 }; 6432 typedef llvm::SmallDenseMap<Object, UsageInfo, 16> UsageInfoMap; 6433 6434 Sema &SemaRef; 6435 /// Sequenced regions within the expression. 6436 SequenceTree Tree; 6437 /// Declaration modifications and references which we have seen. 6438 UsageInfoMap UsageMap; 6439 /// The region we are currently within. 6440 SequenceTree::Seq Region; 6441 /// Filled in with declarations which were modified as a side-effect 6442 /// (that is, post-increment operations). 6443 SmallVectorImpl<std::pair<Object, Usage> > *ModAsSideEffect; 6444 /// Expressions to check later. We defer checking these to reduce 6445 /// stack usage. 6446 SmallVectorImpl<Expr *> &WorkList; 6447 6448 /// RAII object wrapping the visitation of a sequenced subexpression of an 6449 /// expression. At the end of this process, the side-effects of the evaluation 6450 /// become sequenced with respect to the value computation of the result, so 6451 /// we downgrade any UK_ModAsSideEffect within the evaluation to 6452 /// UK_ModAsValue. 6453 struct SequencedSubexpression { 6454 SequencedSubexpression(SequenceChecker &Self) 6455 : Self(Self), OldModAsSideEffect(Self.ModAsSideEffect) { 6456 Self.ModAsSideEffect = &ModAsSideEffect; 6457 } 6458 ~SequencedSubexpression() { 6459 for (unsigned I = 0, E = ModAsSideEffect.size(); I != E; ++I) { 6460 UsageInfo &U = Self.UsageMap[ModAsSideEffect[I].first]; 6461 U.Uses[UK_ModAsSideEffect] = ModAsSideEffect[I].second; 6462 Self.addUsage(U, ModAsSideEffect[I].first, 6463 ModAsSideEffect[I].second.Use, UK_ModAsValue); 6464 } 6465 Self.ModAsSideEffect = OldModAsSideEffect; 6466 } 6467 6468 SequenceChecker &Self; 6469 SmallVector<std::pair<Object, Usage>, 4> ModAsSideEffect; 6470 SmallVectorImpl<std::pair<Object, Usage> > *OldModAsSideEffect; 6471 }; 6472 6473 /// RAII object wrapping the visitation of a subexpression which we might 6474 /// choose to evaluate as a constant. If any subexpression is evaluated and 6475 /// found to be non-constant, this allows us to suppress the evaluation of 6476 /// the outer expression. 6477 class EvaluationTracker { 6478 public: 6479 EvaluationTracker(SequenceChecker &Self) 6480 : Self(Self), Prev(Self.EvalTracker), EvalOK(true) { 6481 Self.EvalTracker = this; 6482 } 6483 ~EvaluationTracker() { 6484 Self.EvalTracker = Prev; 6485 if (Prev) 6486 Prev->EvalOK &= EvalOK; 6487 } 6488 6489 bool evaluate(const Expr *E, bool &Result) { 6490 if (!EvalOK || E->isValueDependent()) 6491 return false; 6492 EvalOK = E->EvaluateAsBooleanCondition(Result, Self.SemaRef.Context); 6493 return EvalOK; 6494 } 6495 6496 private: 6497 SequenceChecker &Self; 6498 EvaluationTracker *Prev; 6499 bool EvalOK; 6500 } *EvalTracker; 6501 6502 /// \brief Find the object which is produced by the specified expression, 6503 /// if any. 6504 Object getObject(Expr *E, bool Mod) const { 6505 E = E->IgnoreParenCasts(); 6506 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 6507 if (Mod && (UO->getOpcode() == UO_PreInc || UO->getOpcode() == UO_PreDec)) 6508 return getObject(UO->getSubExpr(), Mod); 6509 } else if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 6510 if (BO->getOpcode() == BO_Comma) 6511 return getObject(BO->getRHS(), Mod); 6512 if (Mod && BO->isAssignmentOp()) 6513 return getObject(BO->getLHS(), Mod); 6514 } else if (MemberExpr *ME = dyn_cast<MemberExpr>(E)) { 6515 // FIXME: Check for more interesting cases, like "x.n = ++x.n". 6516 if (isa<CXXThisExpr>(ME->getBase()->IgnoreParenCasts())) 6517 return ME->getMemberDecl(); 6518 } else if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 6519 // FIXME: If this is a reference, map through to its value. 6520 return DRE->getDecl(); 6521 return nullptr; 6522 } 6523 6524 /// \brief Note that an object was modified or used by an expression. 6525 void addUsage(UsageInfo &UI, Object O, Expr *Ref, UsageKind UK) { 6526 Usage &U = UI.Uses[UK]; 6527 if (!U.Use || !Tree.isUnsequenced(Region, U.Seq)) { 6528 if (UK == UK_ModAsSideEffect && ModAsSideEffect) 6529 ModAsSideEffect->push_back(std::make_pair(O, U)); 6530 U.Use = Ref; 6531 U.Seq = Region; 6532 } 6533 } 6534 /// \brief Check whether a modification or use conflicts with a prior usage. 6535 void checkUsage(Object O, UsageInfo &UI, Expr *Ref, UsageKind OtherKind, 6536 bool IsModMod) { 6537 if (UI.Diagnosed) 6538 return; 6539 6540 const Usage &U = UI.Uses[OtherKind]; 6541 if (!U.Use || !Tree.isUnsequenced(Region, U.Seq)) 6542 return; 6543 6544 Expr *Mod = U.Use; 6545 Expr *ModOrUse = Ref; 6546 if (OtherKind == UK_Use) 6547 std::swap(Mod, ModOrUse); 6548 6549 SemaRef.Diag(Mod->getExprLoc(), 6550 IsModMod ? diag::warn_unsequenced_mod_mod 6551 : diag::warn_unsequenced_mod_use) 6552 << O << SourceRange(ModOrUse->getExprLoc()); 6553 UI.Diagnosed = true; 6554 } 6555 6556 void notePreUse(Object O, Expr *Use) { 6557 UsageInfo &U = UsageMap[O]; 6558 // Uses conflict with other modifications. 6559 checkUsage(O, U, Use, UK_ModAsValue, false); 6560 } 6561 void notePostUse(Object O, Expr *Use) { 6562 UsageInfo &U = UsageMap[O]; 6563 checkUsage(O, U, Use, UK_ModAsSideEffect, false); 6564 addUsage(U, O, Use, UK_Use); 6565 } 6566 6567 void notePreMod(Object O, Expr *Mod) { 6568 UsageInfo &U = UsageMap[O]; 6569 // Modifications conflict with other modifications and with uses. 6570 checkUsage(O, U, Mod, UK_ModAsValue, true); 6571 checkUsage(O, U, Mod, UK_Use, false); 6572 } 6573 void notePostMod(Object O, Expr *Use, UsageKind UK) { 6574 UsageInfo &U = UsageMap[O]; 6575 checkUsage(O, U, Use, UK_ModAsSideEffect, true); 6576 addUsage(U, O, Use, UK); 6577 } 6578 6579 public: 6580 SequenceChecker(Sema &S, Expr *E, SmallVectorImpl<Expr *> &WorkList) 6581 : Base(S.Context), SemaRef(S), Region(Tree.root()), 6582 ModAsSideEffect(nullptr), WorkList(WorkList), EvalTracker(nullptr) { 6583 Visit(E); 6584 } 6585 6586 void VisitStmt(Stmt *S) { 6587 // Skip all statements which aren't expressions for now. 6588 } 6589 6590 void VisitExpr(Expr *E) { 6591 // By default, just recurse to evaluated subexpressions. 6592 Base::VisitStmt(E); 6593 } 6594 6595 void VisitCastExpr(CastExpr *E) { 6596 Object O = Object(); 6597 if (E->getCastKind() == CK_LValueToRValue) 6598 O = getObject(E->getSubExpr(), false); 6599 6600 if (O) 6601 notePreUse(O, E); 6602 VisitExpr(E); 6603 if (O) 6604 notePostUse(O, E); 6605 } 6606 6607 void VisitBinComma(BinaryOperator *BO) { 6608 // C++11 [expr.comma]p1: 6609 // Every value computation and side effect associated with the left 6610 // expression is sequenced before every value computation and side 6611 // effect associated with the right expression. 6612 SequenceTree::Seq LHS = Tree.allocate(Region); 6613 SequenceTree::Seq RHS = Tree.allocate(Region); 6614 SequenceTree::Seq OldRegion = Region; 6615 6616 { 6617 SequencedSubexpression SeqLHS(*this); 6618 Region = LHS; 6619 Visit(BO->getLHS()); 6620 } 6621 6622 Region = RHS; 6623 Visit(BO->getRHS()); 6624 6625 Region = OldRegion; 6626 6627 // Forget that LHS and RHS are sequenced. They are both unsequenced 6628 // with respect to other stuff. 6629 Tree.merge(LHS); 6630 Tree.merge(RHS); 6631 } 6632 6633 void VisitBinAssign(BinaryOperator *BO) { 6634 // The modification is sequenced after the value computation of the LHS 6635 // and RHS, so check it before inspecting the operands and update the 6636 // map afterwards. 6637 Object O = getObject(BO->getLHS(), true); 6638 if (!O) 6639 return VisitExpr(BO); 6640 6641 notePreMod(O, BO); 6642 6643 // C++11 [expr.ass]p7: 6644 // E1 op= E2 is equivalent to E1 = E1 op E2, except that E1 is evaluated 6645 // only once. 6646 // 6647 // Therefore, for a compound assignment operator, O is considered used 6648 // everywhere except within the evaluation of E1 itself. 6649 if (isa<CompoundAssignOperator>(BO)) 6650 notePreUse(O, BO); 6651 6652 Visit(BO->getLHS()); 6653 6654 if (isa<CompoundAssignOperator>(BO)) 6655 notePostUse(O, BO); 6656 6657 Visit(BO->getRHS()); 6658 6659 // C++11 [expr.ass]p1: 6660 // the assignment is sequenced [...] before the value computation of the 6661 // assignment expression. 6662 // C11 6.5.16/3 has no such rule. 6663 notePostMod(O, BO, SemaRef.getLangOpts().CPlusPlus ? UK_ModAsValue 6664 : UK_ModAsSideEffect); 6665 } 6666 void VisitCompoundAssignOperator(CompoundAssignOperator *CAO) { 6667 VisitBinAssign(CAO); 6668 } 6669 6670 void VisitUnaryPreInc(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); } 6671 void VisitUnaryPreDec(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); } 6672 void VisitUnaryPreIncDec(UnaryOperator *UO) { 6673 Object O = getObject(UO->getSubExpr(), true); 6674 if (!O) 6675 return VisitExpr(UO); 6676 6677 notePreMod(O, UO); 6678 Visit(UO->getSubExpr()); 6679 // C++11 [expr.pre.incr]p1: 6680 // the expression ++x is equivalent to x+=1 6681 notePostMod(O, UO, SemaRef.getLangOpts().CPlusPlus ? UK_ModAsValue 6682 : UK_ModAsSideEffect); 6683 } 6684 6685 void VisitUnaryPostInc(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); } 6686 void VisitUnaryPostDec(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); } 6687 void VisitUnaryPostIncDec(UnaryOperator *UO) { 6688 Object O = getObject(UO->getSubExpr(), true); 6689 if (!O) 6690 return VisitExpr(UO); 6691 6692 notePreMod(O, UO); 6693 Visit(UO->getSubExpr()); 6694 notePostMod(O, UO, UK_ModAsSideEffect); 6695 } 6696 6697 /// Don't visit the RHS of '&&' or '||' if it might not be evaluated. 6698 void VisitBinLOr(BinaryOperator *BO) { 6699 // The side-effects of the LHS of an '&&' are sequenced before the 6700 // value computation of the RHS, and hence before the value computation 6701 // of the '&&' itself, unless the LHS evaluates to zero. We treat them 6702 // as if they were unconditionally sequenced. 6703 EvaluationTracker Eval(*this); 6704 { 6705 SequencedSubexpression Sequenced(*this); 6706 Visit(BO->getLHS()); 6707 } 6708 6709 bool Result; 6710 if (Eval.evaluate(BO->getLHS(), Result)) { 6711 if (!Result) 6712 Visit(BO->getRHS()); 6713 } else { 6714 // Check for unsequenced operations in the RHS, treating it as an 6715 // entirely separate evaluation. 6716 // 6717 // FIXME: If there are operations in the RHS which are unsequenced 6718 // with respect to operations outside the RHS, and those operations 6719 // are unconditionally evaluated, diagnose them. 6720 WorkList.push_back(BO->getRHS()); 6721 } 6722 } 6723 void VisitBinLAnd(BinaryOperator *BO) { 6724 EvaluationTracker Eval(*this); 6725 { 6726 SequencedSubexpression Sequenced(*this); 6727 Visit(BO->getLHS()); 6728 } 6729 6730 bool Result; 6731 if (Eval.evaluate(BO->getLHS(), Result)) { 6732 if (Result) 6733 Visit(BO->getRHS()); 6734 } else { 6735 WorkList.push_back(BO->getRHS()); 6736 } 6737 } 6738 6739 // Only visit the condition, unless we can be sure which subexpression will 6740 // be chosen. 6741 void VisitAbstractConditionalOperator(AbstractConditionalOperator *CO) { 6742 EvaluationTracker Eval(*this); 6743 { 6744 SequencedSubexpression Sequenced(*this); 6745 Visit(CO->getCond()); 6746 } 6747 6748 bool Result; 6749 if (Eval.evaluate(CO->getCond(), Result)) 6750 Visit(Result ? CO->getTrueExpr() : CO->getFalseExpr()); 6751 else { 6752 WorkList.push_back(CO->getTrueExpr()); 6753 WorkList.push_back(CO->getFalseExpr()); 6754 } 6755 } 6756 6757 void VisitCallExpr(CallExpr *CE) { 6758 // C++11 [intro.execution]p15: 6759 // When calling a function [...], every value computation and side effect 6760 // associated with any argument expression, or with the postfix expression 6761 // designating the called function, is sequenced before execution of every 6762 // expression or statement in the body of the function [and thus before 6763 // the value computation of its result]. 6764 SequencedSubexpression Sequenced(*this); 6765 Base::VisitCallExpr(CE); 6766 6767 // FIXME: CXXNewExpr and CXXDeleteExpr implicitly call functions. 6768 } 6769 6770 void VisitCXXConstructExpr(CXXConstructExpr *CCE) { 6771 // This is a call, so all subexpressions are sequenced before the result. 6772 SequencedSubexpression Sequenced(*this); 6773 6774 if (!CCE->isListInitialization()) 6775 return VisitExpr(CCE); 6776 6777 // In C++11, list initializations are sequenced. 6778 SmallVector<SequenceTree::Seq, 32> Elts; 6779 SequenceTree::Seq Parent = Region; 6780 for (CXXConstructExpr::arg_iterator I = CCE->arg_begin(), 6781 E = CCE->arg_end(); 6782 I != E; ++I) { 6783 Region = Tree.allocate(Parent); 6784 Elts.push_back(Region); 6785 Visit(*I); 6786 } 6787 6788 // Forget that the initializers are sequenced. 6789 Region = Parent; 6790 for (unsigned I = 0; I < Elts.size(); ++I) 6791 Tree.merge(Elts[I]); 6792 } 6793 6794 void VisitInitListExpr(InitListExpr *ILE) { 6795 if (!SemaRef.getLangOpts().CPlusPlus11) 6796 return VisitExpr(ILE); 6797 6798 // In C++11, list initializations are sequenced. 6799 SmallVector<SequenceTree::Seq, 32> Elts; 6800 SequenceTree::Seq Parent = Region; 6801 for (unsigned I = 0; I < ILE->getNumInits(); ++I) { 6802 Expr *E = ILE->getInit(I); 6803 if (!E) continue; 6804 Region = Tree.allocate(Parent); 6805 Elts.push_back(Region); 6806 Visit(E); 6807 } 6808 6809 // Forget that the initializers are sequenced. 6810 Region = Parent; 6811 for (unsigned I = 0; I < Elts.size(); ++I) 6812 Tree.merge(Elts[I]); 6813 } 6814 }; 6815 } 6816 6817 void Sema::CheckUnsequencedOperations(Expr *E) { 6818 SmallVector<Expr *, 8> WorkList; 6819 WorkList.push_back(E); 6820 while (!WorkList.empty()) { 6821 Expr *Item = WorkList.pop_back_val(); 6822 SequenceChecker(*this, Item, WorkList); 6823 } 6824 } 6825 6826 void Sema::CheckCompletedExpr(Expr *E, SourceLocation CheckLoc, 6827 bool IsConstexpr) { 6828 CheckImplicitConversions(E, CheckLoc); 6829 CheckUnsequencedOperations(E); 6830 if (!IsConstexpr && !E->isValueDependent()) 6831 CheckForIntOverflow(E); 6832 } 6833 6834 void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 6835 FieldDecl *BitField, 6836 Expr *Init) { 6837 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 6838 } 6839 6840 /// CheckParmsForFunctionDef - Check that the parameters of the given 6841 /// function are appropriate for the definition of a function. This 6842 /// takes care of any checks that cannot be performed on the 6843 /// declaration itself, e.g., that the types of each of the function 6844 /// parameters are complete. 6845 bool Sema::CheckParmsForFunctionDef(ParmVarDecl *const *P, 6846 ParmVarDecl *const *PEnd, 6847 bool CheckParameterNames) { 6848 bool HasInvalidParm = false; 6849 for (; P != PEnd; ++P) { 6850 ParmVarDecl *Param = *P; 6851 6852 // C99 6.7.5.3p4: the parameters in a parameter type list in a 6853 // function declarator that is part of a function definition of 6854 // that function shall not have incomplete type. 6855 // 6856 // This is also C++ [dcl.fct]p6. 6857 if (!Param->isInvalidDecl() && 6858 RequireCompleteType(Param->getLocation(), Param->getType(), 6859 diag::err_typecheck_decl_incomplete_type)) { 6860 Param->setInvalidDecl(); 6861 HasInvalidParm = true; 6862 } 6863 6864 // C99 6.9.1p5: If the declarator includes a parameter type list, the 6865 // declaration of each parameter shall include an identifier. 6866 if (CheckParameterNames && 6867 Param->getIdentifier() == nullptr && 6868 !Param->isImplicit() && 6869 !getLangOpts().CPlusPlus) 6870 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 6871 6872 // C99 6.7.5.3p12: 6873 // If the function declarator is not part of a definition of that 6874 // function, parameters may have incomplete type and may use the [*] 6875 // notation in their sequences of declarator specifiers to specify 6876 // variable length array types. 6877 QualType PType = Param->getOriginalType(); 6878 while (const ArrayType *AT = Context.getAsArrayType(PType)) { 6879 if (AT->getSizeModifier() == ArrayType::Star) { 6880 // FIXME: This diagnostic should point the '[*]' if source-location 6881 // information is added for it. 6882 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 6883 break; 6884 } 6885 PType= AT->getElementType(); 6886 } 6887 6888 // MSVC destroys objects passed by value in the callee. Therefore a 6889 // function definition which takes such a parameter must be able to call the 6890 // object's destructor. However, we don't perform any direct access check 6891 // on the dtor. 6892 if (getLangOpts().CPlusPlus && Context.getTargetInfo() 6893 .getCXXABI() 6894 .areArgsDestroyedLeftToRightInCallee()) { 6895 if (!Param->isInvalidDecl()) { 6896 if (const RecordType *RT = Param->getType()->getAs<RecordType>()) { 6897 CXXRecordDecl *ClassDecl = cast<CXXRecordDecl>(RT->getDecl()); 6898 if (!ClassDecl->isInvalidDecl() && 6899 !ClassDecl->hasIrrelevantDestructor() && 6900 !ClassDecl->isDependentContext()) { 6901 CXXDestructorDecl *Destructor = LookupDestructor(ClassDecl); 6902 MarkFunctionReferenced(Param->getLocation(), Destructor); 6903 DiagnoseUseOfDecl(Destructor, Param->getLocation()); 6904 } 6905 } 6906 } 6907 } 6908 } 6909 6910 return HasInvalidParm; 6911 } 6912 6913 /// CheckCastAlign - Implements -Wcast-align, which warns when a 6914 /// pointer cast increases the alignment requirements. 6915 void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 6916 // This is actually a lot of work to potentially be doing on every 6917 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 6918 if (getDiagnostics().isIgnored(diag::warn_cast_align, TRange.getBegin())) 6919 return; 6920 6921 // Ignore dependent types. 6922 if (T->isDependentType() || Op->getType()->isDependentType()) 6923 return; 6924 6925 // Require that the destination be a pointer type. 6926 const PointerType *DestPtr = T->getAs<PointerType>(); 6927 if (!DestPtr) return; 6928 6929 // If the destination has alignment 1, we're done. 6930 QualType DestPointee = DestPtr->getPointeeType(); 6931 if (DestPointee->isIncompleteType()) return; 6932 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 6933 if (DestAlign.isOne()) return; 6934 6935 // Require that the source be a pointer type. 6936 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 6937 if (!SrcPtr) return; 6938 QualType SrcPointee = SrcPtr->getPointeeType(); 6939 6940 // Whitelist casts from cv void*. We already implicitly 6941 // whitelisted casts to cv void*, since they have alignment 1. 6942 // Also whitelist casts involving incomplete types, which implicitly 6943 // includes 'void'. 6944 if (SrcPointee->isIncompleteType()) return; 6945 6946 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 6947 if (SrcAlign >= DestAlign) return; 6948 6949 Diag(TRange.getBegin(), diag::warn_cast_align) 6950 << Op->getType() << T 6951 << static_cast<unsigned>(SrcAlign.getQuantity()) 6952 << static_cast<unsigned>(DestAlign.getQuantity()) 6953 << TRange << Op->getSourceRange(); 6954 } 6955 6956 static const Type* getElementType(const Expr *BaseExpr) { 6957 const Type* EltType = BaseExpr->getType().getTypePtr(); 6958 if (EltType->isAnyPointerType()) 6959 return EltType->getPointeeType().getTypePtr(); 6960 else if (EltType->isArrayType()) 6961 return EltType->getBaseElementTypeUnsafe(); 6962 return EltType; 6963 } 6964 6965 /// \brief Check whether this array fits the idiom of a size-one tail padded 6966 /// array member of a struct. 6967 /// 6968 /// We avoid emitting out-of-bounds access warnings for such arrays as they are 6969 /// commonly used to emulate flexible arrays in C89 code. 6970 static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 6971 const NamedDecl *ND) { 6972 if (Size != 1 || !ND) return false; 6973 6974 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 6975 if (!FD) return false; 6976 6977 // Don't consider sizes resulting from macro expansions or template argument 6978 // substitution to form C89 tail-padded arrays. 6979 6980 TypeSourceInfo *TInfo = FD->getTypeSourceInfo(); 6981 while (TInfo) { 6982 TypeLoc TL = TInfo->getTypeLoc(); 6983 // Look through typedefs. 6984 if (TypedefTypeLoc TTL = TL.getAs<TypedefTypeLoc>()) { 6985 const TypedefNameDecl *TDL = TTL.getTypedefNameDecl(); 6986 TInfo = TDL->getTypeSourceInfo(); 6987 continue; 6988 } 6989 if (ConstantArrayTypeLoc CTL = TL.getAs<ConstantArrayTypeLoc>()) { 6990 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(CTL.getSizeExpr()); 6991 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 6992 return false; 6993 } 6994 break; 6995 } 6996 6997 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 6998 if (!RD) return false; 6999 if (RD->isUnion()) return false; 7000 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 7001 if (!CRD->isStandardLayout()) return false; 7002 } 7003 7004 // See if this is the last field decl in the record. 7005 const Decl *D = FD; 7006 while ((D = D->getNextDeclInContext())) 7007 if (isa<FieldDecl>(D)) 7008 return false; 7009 return true; 7010 } 7011 7012 void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 7013 const ArraySubscriptExpr *ASE, 7014 bool AllowOnePastEnd, bool IndexNegated) { 7015 IndexExpr = IndexExpr->IgnoreParenImpCasts(); 7016 if (IndexExpr->isValueDependent()) 7017 return; 7018 7019 const Type *EffectiveType = getElementType(BaseExpr); 7020 BaseExpr = BaseExpr->IgnoreParenCasts(); 7021 const ConstantArrayType *ArrayTy = 7022 Context.getAsConstantArrayType(BaseExpr->getType()); 7023 if (!ArrayTy) 7024 return; 7025 7026 llvm::APSInt index; 7027 if (!IndexExpr->EvaluateAsInt(index, Context)) 7028 return; 7029 if (IndexNegated) 7030 index = -index; 7031 7032 const NamedDecl *ND = nullptr; 7033 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 7034 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 7035 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 7036 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 7037 7038 if (index.isUnsigned() || !index.isNegative()) { 7039 llvm::APInt size = ArrayTy->getSize(); 7040 if (!size.isStrictlyPositive()) 7041 return; 7042 7043 const Type* BaseType = getElementType(BaseExpr); 7044 if (BaseType != EffectiveType) { 7045 // Make sure we're comparing apples to apples when comparing index to size 7046 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 7047 uint64_t array_typesize = Context.getTypeSize(BaseType); 7048 // Handle ptrarith_typesize being zero, such as when casting to void* 7049 if (!ptrarith_typesize) ptrarith_typesize = 1; 7050 if (ptrarith_typesize != array_typesize) { 7051 // There's a cast to a different size type involved 7052 uint64_t ratio = array_typesize / ptrarith_typesize; 7053 // TODO: Be smarter about handling cases where array_typesize is not a 7054 // multiple of ptrarith_typesize 7055 if (ptrarith_typesize * ratio == array_typesize) 7056 size *= llvm::APInt(size.getBitWidth(), ratio); 7057 } 7058 } 7059 7060 if (size.getBitWidth() > index.getBitWidth()) 7061 index = index.zext(size.getBitWidth()); 7062 else if (size.getBitWidth() < index.getBitWidth()) 7063 size = size.zext(index.getBitWidth()); 7064 7065 // For array subscripting the index must be less than size, but for pointer 7066 // arithmetic also allow the index (offset) to be equal to size since 7067 // computing the next address after the end of the array is legal and 7068 // commonly done e.g. in C++ iterators and range-based for loops. 7069 if (AllowOnePastEnd ? index.ule(size) : index.ult(size)) 7070 return; 7071 7072 // Also don't warn for arrays of size 1 which are members of some 7073 // structure. These are often used to approximate flexible arrays in C89 7074 // code. 7075 if (IsTailPaddedMemberArray(*this, size, ND)) 7076 return; 7077 7078 // Suppress the warning if the subscript expression (as identified by the 7079 // ']' location) and the index expression are both from macro expansions 7080 // within a system header. 7081 if (ASE) { 7082 SourceLocation RBracketLoc = SourceMgr.getSpellingLoc( 7083 ASE->getRBracketLoc()); 7084 if (SourceMgr.isInSystemHeader(RBracketLoc)) { 7085 SourceLocation IndexLoc = SourceMgr.getSpellingLoc( 7086 IndexExpr->getLocStart()); 7087 if (SourceMgr.isWrittenInSameFile(RBracketLoc, IndexLoc)) 7088 return; 7089 } 7090 } 7091 7092 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 7093 if (ASE) 7094 DiagID = diag::warn_array_index_exceeds_bounds; 7095 7096 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 7097 PDiag(DiagID) << index.toString(10, true) 7098 << size.toString(10, true) 7099 << (unsigned)size.getLimitedValue(~0U) 7100 << IndexExpr->getSourceRange()); 7101 } else { 7102 unsigned DiagID = diag::warn_array_index_precedes_bounds; 7103 if (!ASE) { 7104 DiagID = diag::warn_ptr_arith_precedes_bounds; 7105 if (index.isNegative()) index = -index; 7106 } 7107 7108 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 7109 PDiag(DiagID) << index.toString(10, true) 7110 << IndexExpr->getSourceRange()); 7111 } 7112 7113 if (!ND) { 7114 // Try harder to find a NamedDecl to point at in the note. 7115 while (const ArraySubscriptExpr *ASE = 7116 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 7117 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 7118 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 7119 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 7120 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 7121 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 7122 } 7123 7124 if (ND) 7125 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 7126 PDiag(diag::note_array_index_out_of_bounds) 7127 << ND->getDeclName()); 7128 } 7129 7130 void Sema::CheckArrayAccess(const Expr *expr) { 7131 int AllowOnePastEnd = 0; 7132 while (expr) { 7133 expr = expr->IgnoreParenImpCasts(); 7134 switch (expr->getStmtClass()) { 7135 case Stmt::ArraySubscriptExprClass: { 7136 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 7137 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE, 7138 AllowOnePastEnd > 0); 7139 return; 7140 } 7141 case Stmt::UnaryOperatorClass: { 7142 // Only unwrap the * and & unary operators 7143 const UnaryOperator *UO = cast<UnaryOperator>(expr); 7144 expr = UO->getSubExpr(); 7145 switch (UO->getOpcode()) { 7146 case UO_AddrOf: 7147 AllowOnePastEnd++; 7148 break; 7149 case UO_Deref: 7150 AllowOnePastEnd--; 7151 break; 7152 default: 7153 return; 7154 } 7155 break; 7156 } 7157 case Stmt::ConditionalOperatorClass: { 7158 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 7159 if (const Expr *lhs = cond->getLHS()) 7160 CheckArrayAccess(lhs); 7161 if (const Expr *rhs = cond->getRHS()) 7162 CheckArrayAccess(rhs); 7163 return; 7164 } 7165 default: 7166 return; 7167 } 7168 } 7169 } 7170 7171 //===--- CHECK: Objective-C retain cycles ----------------------------------// 7172 7173 namespace { 7174 struct RetainCycleOwner { 7175 RetainCycleOwner() : Variable(nullptr), Indirect(false) {} 7176 VarDecl *Variable; 7177 SourceRange Range; 7178 SourceLocation Loc; 7179 bool Indirect; 7180 7181 void setLocsFrom(Expr *e) { 7182 Loc = e->getExprLoc(); 7183 Range = e->getSourceRange(); 7184 } 7185 }; 7186 } 7187 7188 /// Consider whether capturing the given variable can possibly lead to 7189 /// a retain cycle. 7190 static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 7191 // In ARC, it's captured strongly iff the variable has __strong 7192 // lifetime. In MRR, it's captured strongly if the variable is 7193 // __block and has an appropriate type. 7194 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 7195 return false; 7196 7197 owner.Variable = var; 7198 if (ref) 7199 owner.setLocsFrom(ref); 7200 return true; 7201 } 7202 7203 static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) { 7204 while (true) { 7205 e = e->IgnoreParens(); 7206 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 7207 switch (cast->getCastKind()) { 7208 case CK_BitCast: 7209 case CK_LValueBitCast: 7210 case CK_LValueToRValue: 7211 case CK_ARCReclaimReturnedObject: 7212 e = cast->getSubExpr(); 7213 continue; 7214 7215 default: 7216 return false; 7217 } 7218 } 7219 7220 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 7221 ObjCIvarDecl *ivar = ref->getDecl(); 7222 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 7223 return false; 7224 7225 // Try to find a retain cycle in the base. 7226 if (!findRetainCycleOwner(S, ref->getBase(), owner)) 7227 return false; 7228 7229 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 7230 owner.Indirect = true; 7231 return true; 7232 } 7233 7234 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 7235 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 7236 if (!var) return false; 7237 return considerVariable(var, ref, owner); 7238 } 7239 7240 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 7241 if (member->isArrow()) return false; 7242 7243 // Don't count this as an indirect ownership. 7244 e = member->getBase(); 7245 continue; 7246 } 7247 7248 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 7249 // Only pay attention to pseudo-objects on property references. 7250 ObjCPropertyRefExpr *pre 7251 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 7252 ->IgnoreParens()); 7253 if (!pre) return false; 7254 if (pre->isImplicitProperty()) return false; 7255 ObjCPropertyDecl *property = pre->getExplicitProperty(); 7256 if (!property->isRetaining() && 7257 !(property->getPropertyIvarDecl() && 7258 property->getPropertyIvarDecl()->getType() 7259 .getObjCLifetime() == Qualifiers::OCL_Strong)) 7260 return false; 7261 7262 owner.Indirect = true; 7263 if (pre->isSuperReceiver()) { 7264 owner.Variable = S.getCurMethodDecl()->getSelfDecl(); 7265 if (!owner.Variable) 7266 return false; 7267 owner.Loc = pre->getLocation(); 7268 owner.Range = pre->getSourceRange(); 7269 return true; 7270 } 7271 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 7272 ->getSourceExpr()); 7273 continue; 7274 } 7275 7276 // Array ivars? 7277 7278 return false; 7279 } 7280 } 7281 7282 namespace { 7283 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 7284 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 7285 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 7286 Context(Context), Variable(variable), Capturer(nullptr), 7287 VarWillBeReased(false) {} 7288 ASTContext &Context; 7289 VarDecl *Variable; 7290 Expr *Capturer; 7291 bool VarWillBeReased; 7292 7293 void VisitDeclRefExpr(DeclRefExpr *ref) { 7294 if (ref->getDecl() == Variable && !Capturer) 7295 Capturer = ref; 7296 } 7297 7298 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 7299 if (Capturer) return; 7300 Visit(ref->getBase()); 7301 if (Capturer && ref->isFreeIvar()) 7302 Capturer = ref; 7303 } 7304 7305 void VisitBlockExpr(BlockExpr *block) { 7306 // Look inside nested blocks 7307 if (block->getBlockDecl()->capturesVariable(Variable)) 7308 Visit(block->getBlockDecl()->getBody()); 7309 } 7310 7311 void VisitOpaqueValueExpr(OpaqueValueExpr *OVE) { 7312 if (Capturer) return; 7313 if (OVE->getSourceExpr()) 7314 Visit(OVE->getSourceExpr()); 7315 } 7316 void VisitBinaryOperator(BinaryOperator *BinOp) { 7317 if (!Variable || VarWillBeReased || BinOp->getOpcode() != BO_Assign) 7318 return; 7319 Expr *LHS = BinOp->getLHS(); 7320 if (const DeclRefExpr *DRE = dyn_cast_or_null<DeclRefExpr>(LHS)) { 7321 if (DRE->getDecl() != Variable) 7322 return; 7323 if (Expr *RHS = BinOp->getRHS()) { 7324 RHS = RHS->IgnoreParenCasts(); 7325 llvm::APSInt Value; 7326 VarWillBeReased = 7327 (RHS && RHS->isIntegerConstantExpr(Value, Context) && Value == 0); 7328 } 7329 } 7330 } 7331 }; 7332 } 7333 7334 /// Check whether the given argument is a block which captures a 7335 /// variable. 7336 static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 7337 assert(owner.Variable && owner.Loc.isValid()); 7338 7339 e = e->IgnoreParenCasts(); 7340 7341 // Look through [^{...} copy] and Block_copy(^{...}). 7342 if (ObjCMessageExpr *ME = dyn_cast<ObjCMessageExpr>(e)) { 7343 Selector Cmd = ME->getSelector(); 7344 if (Cmd.isUnarySelector() && Cmd.getNameForSlot(0) == "copy") { 7345 e = ME->getInstanceReceiver(); 7346 if (!e) 7347 return nullptr; 7348 e = e->IgnoreParenCasts(); 7349 } 7350 } else if (CallExpr *CE = dyn_cast<CallExpr>(e)) { 7351 if (CE->getNumArgs() == 1) { 7352 FunctionDecl *Fn = dyn_cast_or_null<FunctionDecl>(CE->getCalleeDecl()); 7353 if (Fn) { 7354 const IdentifierInfo *FnI = Fn->getIdentifier(); 7355 if (FnI && FnI->isStr("_Block_copy")) { 7356 e = CE->getArg(0)->IgnoreParenCasts(); 7357 } 7358 } 7359 } 7360 } 7361 7362 BlockExpr *block = dyn_cast<BlockExpr>(e); 7363 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 7364 return nullptr; 7365 7366 FindCaptureVisitor visitor(S.Context, owner.Variable); 7367 visitor.Visit(block->getBlockDecl()->getBody()); 7368 return visitor.VarWillBeReased ? nullptr : visitor.Capturer; 7369 } 7370 7371 static void diagnoseRetainCycle(Sema &S, Expr *capturer, 7372 RetainCycleOwner &owner) { 7373 assert(capturer); 7374 assert(owner.Variable && owner.Loc.isValid()); 7375 7376 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 7377 << owner.Variable << capturer->getSourceRange(); 7378 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 7379 << owner.Indirect << owner.Range; 7380 } 7381 7382 /// Check for a keyword selector that starts with the word 'add' or 7383 /// 'set'. 7384 static bool isSetterLikeSelector(Selector sel) { 7385 if (sel.isUnarySelector()) return false; 7386 7387 StringRef str = sel.getNameForSlot(0); 7388 while (!str.empty() && str.front() == '_') str = str.substr(1); 7389 if (str.startswith("set")) 7390 str = str.substr(3); 7391 else if (str.startswith("add")) { 7392 // Specially whitelist 'addOperationWithBlock:'. 7393 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 7394 return false; 7395 str = str.substr(3); 7396 } 7397 else 7398 return false; 7399 7400 if (str.empty()) return true; 7401 return !isLowercase(str.front()); 7402 } 7403 7404 /// Check a message send to see if it's likely to cause a retain cycle. 7405 void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 7406 // Only check instance methods whose selector looks like a setter. 7407 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 7408 return; 7409 7410 // Try to find a variable that the receiver is strongly owned by. 7411 RetainCycleOwner owner; 7412 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 7413 if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner)) 7414 return; 7415 } else { 7416 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 7417 owner.Variable = getCurMethodDecl()->getSelfDecl(); 7418 owner.Loc = msg->getSuperLoc(); 7419 owner.Range = msg->getSuperLoc(); 7420 } 7421 7422 // Check whether the receiver is captured by any of the arguments. 7423 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 7424 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 7425 return diagnoseRetainCycle(*this, capturer, owner); 7426 } 7427 7428 /// Check a property assign to see if it's likely to cause a retain cycle. 7429 void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 7430 RetainCycleOwner owner; 7431 if (!findRetainCycleOwner(*this, receiver, owner)) 7432 return; 7433 7434 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 7435 diagnoseRetainCycle(*this, capturer, owner); 7436 } 7437 7438 void Sema::checkRetainCycles(VarDecl *Var, Expr *Init) { 7439 RetainCycleOwner Owner; 7440 if (!considerVariable(Var, /*DeclRefExpr=*/nullptr, Owner)) 7441 return; 7442 7443 // Because we don't have an expression for the variable, we have to set the 7444 // location explicitly here. 7445 Owner.Loc = Var->getLocation(); 7446 Owner.Range = Var->getSourceRange(); 7447 7448 if (Expr *Capturer = findCapturingExpr(*this, Init, Owner)) 7449 diagnoseRetainCycle(*this, Capturer, Owner); 7450 } 7451 7452 static bool checkUnsafeAssignLiteral(Sema &S, SourceLocation Loc, 7453 Expr *RHS, bool isProperty) { 7454 // Check if RHS is an Objective-C object literal, which also can get 7455 // immediately zapped in a weak reference. Note that we explicitly 7456 // allow ObjCStringLiterals, since those are designed to never really die. 7457 RHS = RHS->IgnoreParenImpCasts(); 7458 7459 // This enum needs to match with the 'select' in 7460 // warn_objc_arc_literal_assign (off-by-1). 7461 Sema::ObjCLiteralKind Kind = S.CheckLiteralKind(RHS); 7462 if (Kind == Sema::LK_String || Kind == Sema::LK_None) 7463 return false; 7464 7465 S.Diag(Loc, diag::warn_arc_literal_assign) 7466 << (unsigned) Kind 7467 << (isProperty ? 0 : 1) 7468 << RHS->getSourceRange(); 7469 7470 return true; 7471 } 7472 7473 static bool checkUnsafeAssignObject(Sema &S, SourceLocation Loc, 7474 Qualifiers::ObjCLifetime LT, 7475 Expr *RHS, bool isProperty) { 7476 // Strip off any implicit cast added to get to the one ARC-specific. 7477 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 7478 if (cast->getCastKind() == CK_ARCConsumeObject) { 7479 S.Diag(Loc, diag::warn_arc_retained_assign) 7480 << (LT == Qualifiers::OCL_ExplicitNone) 7481 << (isProperty ? 0 : 1) 7482 << RHS->getSourceRange(); 7483 return true; 7484 } 7485 RHS = cast->getSubExpr(); 7486 } 7487 7488 if (LT == Qualifiers::OCL_Weak && 7489 checkUnsafeAssignLiteral(S, Loc, RHS, isProperty)) 7490 return true; 7491 7492 return false; 7493 } 7494 7495 bool Sema::checkUnsafeAssigns(SourceLocation Loc, 7496 QualType LHS, Expr *RHS) { 7497 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 7498 7499 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 7500 return false; 7501 7502 if (checkUnsafeAssignObject(*this, Loc, LT, RHS, false)) 7503 return true; 7504 7505 return false; 7506 } 7507 7508 void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 7509 Expr *LHS, Expr *RHS) { 7510 QualType LHSType; 7511 // PropertyRef on LHS type need be directly obtained from 7512 // its declaration as it has a PseudoType. 7513 ObjCPropertyRefExpr *PRE 7514 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens()); 7515 if (PRE && !PRE->isImplicitProperty()) { 7516 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 7517 if (PD) 7518 LHSType = PD->getType(); 7519 } 7520 7521 if (LHSType.isNull()) 7522 LHSType = LHS->getType(); 7523 7524 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 7525 7526 if (LT == Qualifiers::OCL_Weak) { 7527 if (!Diags.isIgnored(diag::warn_arc_repeated_use_of_weak, Loc)) 7528 getCurFunction()->markSafeWeakUse(LHS); 7529 } 7530 7531 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 7532 return; 7533 7534 // FIXME. Check for other life times. 7535 if (LT != Qualifiers::OCL_None) 7536 return; 7537 7538 if (PRE) { 7539 if (PRE->isImplicitProperty()) 7540 return; 7541 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 7542 if (!PD) 7543 return; 7544 7545 unsigned Attributes = PD->getPropertyAttributes(); 7546 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) { 7547 // when 'assign' attribute was not explicitly specified 7548 // by user, ignore it and rely on property type itself 7549 // for lifetime info. 7550 unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten(); 7551 if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) && 7552 LHSType->isObjCRetainableType()) 7553 return; 7554 7555 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 7556 if (cast->getCastKind() == CK_ARCConsumeObject) { 7557 Diag(Loc, diag::warn_arc_retained_property_assign) 7558 << RHS->getSourceRange(); 7559 return; 7560 } 7561 RHS = cast->getSubExpr(); 7562 } 7563 } 7564 else if (Attributes & ObjCPropertyDecl::OBJC_PR_weak) { 7565 if (checkUnsafeAssignObject(*this, Loc, Qualifiers::OCL_Weak, RHS, true)) 7566 return; 7567 } 7568 } 7569 } 7570 7571 //===--- CHECK: Empty statement body (-Wempty-body) ---------------------===// 7572 7573 namespace { 7574 bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr, 7575 SourceLocation StmtLoc, 7576 const NullStmt *Body) { 7577 // Do not warn if the body is a macro that expands to nothing, e.g: 7578 // 7579 // #define CALL(x) 7580 // if (condition) 7581 // CALL(0); 7582 // 7583 if (Body->hasLeadingEmptyMacro()) 7584 return false; 7585 7586 // Get line numbers of statement and body. 7587 bool StmtLineInvalid; 7588 unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc, 7589 &StmtLineInvalid); 7590 if (StmtLineInvalid) 7591 return false; 7592 7593 bool BodyLineInvalid; 7594 unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(), 7595 &BodyLineInvalid); 7596 if (BodyLineInvalid) 7597 return false; 7598 7599 // Warn if null statement and body are on the same line. 7600 if (StmtLine != BodyLine) 7601 return false; 7602 7603 return true; 7604 } 7605 } // Unnamed namespace 7606 7607 void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc, 7608 const Stmt *Body, 7609 unsigned DiagID) { 7610 // Since this is a syntactic check, don't emit diagnostic for template 7611 // instantiations, this just adds noise. 7612 if (CurrentInstantiationScope) 7613 return; 7614 7615 // The body should be a null statement. 7616 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 7617 if (!NBody) 7618 return; 7619 7620 // Do the usual checks. 7621 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 7622 return; 7623 7624 Diag(NBody->getSemiLoc(), DiagID); 7625 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 7626 } 7627 7628 void Sema::DiagnoseEmptyLoopBody(const Stmt *S, 7629 const Stmt *PossibleBody) { 7630 assert(!CurrentInstantiationScope); // Ensured by caller 7631 7632 SourceLocation StmtLoc; 7633 const Stmt *Body; 7634 unsigned DiagID; 7635 if (const ForStmt *FS = dyn_cast<ForStmt>(S)) { 7636 StmtLoc = FS->getRParenLoc(); 7637 Body = FS->getBody(); 7638 DiagID = diag::warn_empty_for_body; 7639 } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) { 7640 StmtLoc = WS->getCond()->getSourceRange().getEnd(); 7641 Body = WS->getBody(); 7642 DiagID = diag::warn_empty_while_body; 7643 } else 7644 return; // Neither `for' nor `while'. 7645 7646 // The body should be a null statement. 7647 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 7648 if (!NBody) 7649 return; 7650 7651 // Skip expensive checks if diagnostic is disabled. 7652 if (Diags.isIgnored(DiagID, NBody->getSemiLoc())) 7653 return; 7654 7655 // Do the usual checks. 7656 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 7657 return; 7658 7659 // `for(...);' and `while(...);' are popular idioms, so in order to keep 7660 // noise level low, emit diagnostics only if for/while is followed by a 7661 // CompoundStmt, e.g.: 7662 // for (int i = 0; i < n; i++); 7663 // { 7664 // a(i); 7665 // } 7666 // or if for/while is followed by a statement with more indentation 7667 // than for/while itself: 7668 // for (int i = 0; i < n; i++); 7669 // a(i); 7670 bool ProbableTypo = isa<CompoundStmt>(PossibleBody); 7671 if (!ProbableTypo) { 7672 bool BodyColInvalid; 7673 unsigned BodyCol = SourceMgr.getPresumedColumnNumber( 7674 PossibleBody->getLocStart(), 7675 &BodyColInvalid); 7676 if (BodyColInvalid) 7677 return; 7678 7679 bool StmtColInvalid; 7680 unsigned StmtCol = SourceMgr.getPresumedColumnNumber( 7681 S->getLocStart(), 7682 &StmtColInvalid); 7683 if (StmtColInvalid) 7684 return; 7685 7686 if (BodyCol > StmtCol) 7687 ProbableTypo = true; 7688 } 7689 7690 if (ProbableTypo) { 7691 Diag(NBody->getSemiLoc(), DiagID); 7692 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 7693 } 7694 } 7695 7696 //===--- Layout compatibility ----------------------------------------------// 7697 7698 namespace { 7699 7700 bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2); 7701 7702 /// \brief Check if two enumeration types are layout-compatible. 7703 bool isLayoutCompatible(ASTContext &C, EnumDecl *ED1, EnumDecl *ED2) { 7704 // C++11 [dcl.enum] p8: 7705 // Two enumeration types are layout-compatible if they have the same 7706 // underlying type. 7707 return ED1->isComplete() && ED2->isComplete() && 7708 C.hasSameType(ED1->getIntegerType(), ED2->getIntegerType()); 7709 } 7710 7711 /// \brief Check if two fields are layout-compatible. 7712 bool isLayoutCompatible(ASTContext &C, FieldDecl *Field1, FieldDecl *Field2) { 7713 if (!isLayoutCompatible(C, Field1->getType(), Field2->getType())) 7714 return false; 7715 7716 if (Field1->isBitField() != Field2->isBitField()) 7717 return false; 7718 7719 if (Field1->isBitField()) { 7720 // Make sure that the bit-fields are the same length. 7721 unsigned Bits1 = Field1->getBitWidthValue(C); 7722 unsigned Bits2 = Field2->getBitWidthValue(C); 7723 7724 if (Bits1 != Bits2) 7725 return false; 7726 } 7727 7728 return true; 7729 } 7730 7731 /// \brief Check if two standard-layout structs are layout-compatible. 7732 /// (C++11 [class.mem] p17) 7733 bool isLayoutCompatibleStruct(ASTContext &C, 7734 RecordDecl *RD1, 7735 RecordDecl *RD2) { 7736 // If both records are C++ classes, check that base classes match. 7737 if (const CXXRecordDecl *D1CXX = dyn_cast<CXXRecordDecl>(RD1)) { 7738 // If one of records is a CXXRecordDecl we are in C++ mode, 7739 // thus the other one is a CXXRecordDecl, too. 7740 const CXXRecordDecl *D2CXX = cast<CXXRecordDecl>(RD2); 7741 // Check number of base classes. 7742 if (D1CXX->getNumBases() != D2CXX->getNumBases()) 7743 return false; 7744 7745 // Check the base classes. 7746 for (CXXRecordDecl::base_class_const_iterator 7747 Base1 = D1CXX->bases_begin(), 7748 BaseEnd1 = D1CXX->bases_end(), 7749 Base2 = D2CXX->bases_begin(); 7750 Base1 != BaseEnd1; 7751 ++Base1, ++Base2) { 7752 if (!isLayoutCompatible(C, Base1->getType(), Base2->getType())) 7753 return false; 7754 } 7755 } else if (const CXXRecordDecl *D2CXX = dyn_cast<CXXRecordDecl>(RD2)) { 7756 // If only RD2 is a C++ class, it should have zero base classes. 7757 if (D2CXX->getNumBases() > 0) 7758 return false; 7759 } 7760 7761 // Check the fields. 7762 RecordDecl::field_iterator Field2 = RD2->field_begin(), 7763 Field2End = RD2->field_end(), 7764 Field1 = RD1->field_begin(), 7765 Field1End = RD1->field_end(); 7766 for ( ; Field1 != Field1End && Field2 != Field2End; ++Field1, ++Field2) { 7767 if (!isLayoutCompatible(C, *Field1, *Field2)) 7768 return false; 7769 } 7770 if (Field1 != Field1End || Field2 != Field2End) 7771 return false; 7772 7773 return true; 7774 } 7775 7776 /// \brief Check if two standard-layout unions are layout-compatible. 7777 /// (C++11 [class.mem] p18) 7778 bool isLayoutCompatibleUnion(ASTContext &C, 7779 RecordDecl *RD1, 7780 RecordDecl *RD2) { 7781 llvm::SmallPtrSet<FieldDecl *, 8> UnmatchedFields; 7782 for (auto *Field2 : RD2->fields()) 7783 UnmatchedFields.insert(Field2); 7784 7785 for (auto *Field1 : RD1->fields()) { 7786 llvm::SmallPtrSet<FieldDecl *, 8>::iterator 7787 I = UnmatchedFields.begin(), 7788 E = UnmatchedFields.end(); 7789 7790 for ( ; I != E; ++I) { 7791 if (isLayoutCompatible(C, Field1, *I)) { 7792 bool Result = UnmatchedFields.erase(*I); 7793 (void) Result; 7794 assert(Result); 7795 break; 7796 } 7797 } 7798 if (I == E) 7799 return false; 7800 } 7801 7802 return UnmatchedFields.empty(); 7803 } 7804 7805 bool isLayoutCompatible(ASTContext &C, RecordDecl *RD1, RecordDecl *RD2) { 7806 if (RD1->isUnion() != RD2->isUnion()) 7807 return false; 7808 7809 if (RD1->isUnion()) 7810 return isLayoutCompatibleUnion(C, RD1, RD2); 7811 else 7812 return isLayoutCompatibleStruct(C, RD1, RD2); 7813 } 7814 7815 /// \brief Check if two types are layout-compatible in C++11 sense. 7816 bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2) { 7817 if (T1.isNull() || T2.isNull()) 7818 return false; 7819 7820 // C++11 [basic.types] p11: 7821 // If two types T1 and T2 are the same type, then T1 and T2 are 7822 // layout-compatible types. 7823 if (C.hasSameType(T1, T2)) 7824 return true; 7825 7826 T1 = T1.getCanonicalType().getUnqualifiedType(); 7827 T2 = T2.getCanonicalType().getUnqualifiedType(); 7828 7829 const Type::TypeClass TC1 = T1->getTypeClass(); 7830 const Type::TypeClass TC2 = T2->getTypeClass(); 7831 7832 if (TC1 != TC2) 7833 return false; 7834 7835 if (TC1 == Type::Enum) { 7836 return isLayoutCompatible(C, 7837 cast<EnumType>(T1)->getDecl(), 7838 cast<EnumType>(T2)->getDecl()); 7839 } else if (TC1 == Type::Record) { 7840 if (!T1->isStandardLayoutType() || !T2->isStandardLayoutType()) 7841 return false; 7842 7843 return isLayoutCompatible(C, 7844 cast<RecordType>(T1)->getDecl(), 7845 cast<RecordType>(T2)->getDecl()); 7846 } 7847 7848 return false; 7849 } 7850 } 7851 7852 //===--- CHECK: pointer_with_type_tag attribute: datatypes should match ----// 7853 7854 namespace { 7855 /// \brief Given a type tag expression find the type tag itself. 7856 /// 7857 /// \param TypeExpr Type tag expression, as it appears in user's code. 7858 /// 7859 /// \param VD Declaration of an identifier that appears in a type tag. 7860 /// 7861 /// \param MagicValue Type tag magic value. 7862 bool FindTypeTagExpr(const Expr *TypeExpr, const ASTContext &Ctx, 7863 const ValueDecl **VD, uint64_t *MagicValue) { 7864 while(true) { 7865 if (!TypeExpr) 7866 return false; 7867 7868 TypeExpr = TypeExpr->IgnoreParenImpCasts()->IgnoreParenCasts(); 7869 7870 switch (TypeExpr->getStmtClass()) { 7871 case Stmt::UnaryOperatorClass: { 7872 const UnaryOperator *UO = cast<UnaryOperator>(TypeExpr); 7873 if (UO->getOpcode() == UO_AddrOf || UO->getOpcode() == UO_Deref) { 7874 TypeExpr = UO->getSubExpr(); 7875 continue; 7876 } 7877 return false; 7878 } 7879 7880 case Stmt::DeclRefExprClass: { 7881 const DeclRefExpr *DRE = cast<DeclRefExpr>(TypeExpr); 7882 *VD = DRE->getDecl(); 7883 return true; 7884 } 7885 7886 case Stmt::IntegerLiteralClass: { 7887 const IntegerLiteral *IL = cast<IntegerLiteral>(TypeExpr); 7888 llvm::APInt MagicValueAPInt = IL->getValue(); 7889 if (MagicValueAPInt.getActiveBits() <= 64) { 7890 *MagicValue = MagicValueAPInt.getZExtValue(); 7891 return true; 7892 } else 7893 return false; 7894 } 7895 7896 case Stmt::BinaryConditionalOperatorClass: 7897 case Stmt::ConditionalOperatorClass: { 7898 const AbstractConditionalOperator *ACO = 7899 cast<AbstractConditionalOperator>(TypeExpr); 7900 bool Result; 7901 if (ACO->getCond()->EvaluateAsBooleanCondition(Result, Ctx)) { 7902 if (Result) 7903 TypeExpr = ACO->getTrueExpr(); 7904 else 7905 TypeExpr = ACO->getFalseExpr(); 7906 continue; 7907 } 7908 return false; 7909 } 7910 7911 case Stmt::BinaryOperatorClass: { 7912 const BinaryOperator *BO = cast<BinaryOperator>(TypeExpr); 7913 if (BO->getOpcode() == BO_Comma) { 7914 TypeExpr = BO->getRHS(); 7915 continue; 7916 } 7917 return false; 7918 } 7919 7920 default: 7921 return false; 7922 } 7923 } 7924 } 7925 7926 /// \brief Retrieve the C type corresponding to type tag TypeExpr. 7927 /// 7928 /// \param TypeExpr Expression that specifies a type tag. 7929 /// 7930 /// \param MagicValues Registered magic values. 7931 /// 7932 /// \param FoundWrongKind Set to true if a type tag was found, but of a wrong 7933 /// kind. 7934 /// 7935 /// \param TypeInfo Information about the corresponding C type. 7936 /// 7937 /// \returns true if the corresponding C type was found. 7938 bool GetMatchingCType( 7939 const IdentifierInfo *ArgumentKind, 7940 const Expr *TypeExpr, const ASTContext &Ctx, 7941 const llvm::DenseMap<Sema::TypeTagMagicValue, 7942 Sema::TypeTagData> *MagicValues, 7943 bool &FoundWrongKind, 7944 Sema::TypeTagData &TypeInfo) { 7945 FoundWrongKind = false; 7946 7947 // Variable declaration that has type_tag_for_datatype attribute. 7948 const ValueDecl *VD = nullptr; 7949 7950 uint64_t MagicValue; 7951 7952 if (!FindTypeTagExpr(TypeExpr, Ctx, &VD, &MagicValue)) 7953 return false; 7954 7955 if (VD) { 7956 if (TypeTagForDatatypeAttr *I = VD->getAttr<TypeTagForDatatypeAttr>()) { 7957 if (I->getArgumentKind() != ArgumentKind) { 7958 FoundWrongKind = true; 7959 return false; 7960 } 7961 TypeInfo.Type = I->getMatchingCType(); 7962 TypeInfo.LayoutCompatible = I->getLayoutCompatible(); 7963 TypeInfo.MustBeNull = I->getMustBeNull(); 7964 return true; 7965 } 7966 return false; 7967 } 7968 7969 if (!MagicValues) 7970 return false; 7971 7972 llvm::DenseMap<Sema::TypeTagMagicValue, 7973 Sema::TypeTagData>::const_iterator I = 7974 MagicValues->find(std::make_pair(ArgumentKind, MagicValue)); 7975 if (I == MagicValues->end()) 7976 return false; 7977 7978 TypeInfo = I->second; 7979 return true; 7980 } 7981 } // unnamed namespace 7982 7983 void Sema::RegisterTypeTagForDatatype(const IdentifierInfo *ArgumentKind, 7984 uint64_t MagicValue, QualType Type, 7985 bool LayoutCompatible, 7986 bool MustBeNull) { 7987 if (!TypeTagForDatatypeMagicValues) 7988 TypeTagForDatatypeMagicValues.reset( 7989 new llvm::DenseMap<TypeTagMagicValue, TypeTagData>); 7990 7991 TypeTagMagicValue Magic(ArgumentKind, MagicValue); 7992 (*TypeTagForDatatypeMagicValues)[Magic] = 7993 TypeTagData(Type, LayoutCompatible, MustBeNull); 7994 } 7995 7996 namespace { 7997 bool IsSameCharType(QualType T1, QualType T2) { 7998 const BuiltinType *BT1 = T1->getAs<BuiltinType>(); 7999 if (!BT1) 8000 return false; 8001 8002 const BuiltinType *BT2 = T2->getAs<BuiltinType>(); 8003 if (!BT2) 8004 return false; 8005 8006 BuiltinType::Kind T1Kind = BT1->getKind(); 8007 BuiltinType::Kind T2Kind = BT2->getKind(); 8008 8009 return (T1Kind == BuiltinType::SChar && T2Kind == BuiltinType::Char_S) || 8010 (T1Kind == BuiltinType::UChar && T2Kind == BuiltinType::Char_U) || 8011 (T1Kind == BuiltinType::Char_U && T2Kind == BuiltinType::UChar) || 8012 (T1Kind == BuiltinType::Char_S && T2Kind == BuiltinType::SChar); 8013 } 8014 } // unnamed namespace 8015 8016 void Sema::CheckArgumentWithTypeTag(const ArgumentWithTypeTagAttr *Attr, 8017 const Expr * const *ExprArgs) { 8018 const IdentifierInfo *ArgumentKind = Attr->getArgumentKind(); 8019 bool IsPointerAttr = Attr->getIsPointer(); 8020 8021 const Expr *TypeTagExpr = ExprArgs[Attr->getTypeTagIdx()]; 8022 bool FoundWrongKind; 8023 TypeTagData TypeInfo; 8024 if (!GetMatchingCType(ArgumentKind, TypeTagExpr, Context, 8025 TypeTagForDatatypeMagicValues.get(), 8026 FoundWrongKind, TypeInfo)) { 8027 if (FoundWrongKind) 8028 Diag(TypeTagExpr->getExprLoc(), 8029 diag::warn_type_tag_for_datatype_wrong_kind) 8030 << TypeTagExpr->getSourceRange(); 8031 return; 8032 } 8033 8034 const Expr *ArgumentExpr = ExprArgs[Attr->getArgumentIdx()]; 8035 if (IsPointerAttr) { 8036 // Skip implicit cast of pointer to `void *' (as a function argument). 8037 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(ArgumentExpr)) 8038 if (ICE->getType()->isVoidPointerType() && 8039 ICE->getCastKind() == CK_BitCast) 8040 ArgumentExpr = ICE->getSubExpr(); 8041 } 8042 QualType ArgumentType = ArgumentExpr->getType(); 8043 8044 // Passing a `void*' pointer shouldn't trigger a warning. 8045 if (IsPointerAttr && ArgumentType->isVoidPointerType()) 8046 return; 8047 8048 if (TypeInfo.MustBeNull) { 8049 // Type tag with matching void type requires a null pointer. 8050 if (!ArgumentExpr->isNullPointerConstant(Context, 8051 Expr::NPC_ValueDependentIsNotNull)) { 8052 Diag(ArgumentExpr->getExprLoc(), 8053 diag::warn_type_safety_null_pointer_required) 8054 << ArgumentKind->getName() 8055 << ArgumentExpr->getSourceRange() 8056 << TypeTagExpr->getSourceRange(); 8057 } 8058 return; 8059 } 8060 8061 QualType RequiredType = TypeInfo.Type; 8062 if (IsPointerAttr) 8063 RequiredType = Context.getPointerType(RequiredType); 8064 8065 bool mismatch = false; 8066 if (!TypeInfo.LayoutCompatible) { 8067 mismatch = !Context.hasSameType(ArgumentType, RequiredType); 8068 8069 // C++11 [basic.fundamental] p1: 8070 // Plain char, signed char, and unsigned char are three distinct types. 8071 // 8072 // But we treat plain `char' as equivalent to `signed char' or `unsigned 8073 // char' depending on the current char signedness mode. 8074 if (mismatch) 8075 if ((IsPointerAttr && IsSameCharType(ArgumentType->getPointeeType(), 8076 RequiredType->getPointeeType())) || 8077 (!IsPointerAttr && IsSameCharType(ArgumentType, RequiredType))) 8078 mismatch = false; 8079 } else 8080 if (IsPointerAttr) 8081 mismatch = !isLayoutCompatible(Context, 8082 ArgumentType->getPointeeType(), 8083 RequiredType->getPointeeType()); 8084 else 8085 mismatch = !isLayoutCompatible(Context, ArgumentType, RequiredType); 8086 8087 if (mismatch) 8088 Diag(ArgumentExpr->getExprLoc(), diag::warn_type_safety_type_mismatch) 8089 << ArgumentType << ArgumentKind 8090 << TypeInfo.LayoutCompatible << RequiredType 8091 << ArgumentExpr->getSourceRange() 8092 << TypeTagExpr->getSourceRange(); 8093 } 8094 8095