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/Sema.h" 16 #include "clang/Sema/SemaInternal.h" 17 #include "clang/Sema/ScopeInfo.h" 18 #include "clang/Analysis/Analyses/FormatString.h" 19 #include "clang/AST/ASTContext.h" 20 #include "clang/AST/CharUnits.h" 21 #include "clang/AST/DeclCXX.h" 22 #include "clang/AST/DeclObjC.h" 23 #include "clang/AST/ExprCXX.h" 24 #include "clang/AST/ExprObjC.h" 25 #include "clang/AST/EvaluatedExprVisitor.h" 26 #include "clang/AST/DeclObjC.h" 27 #include "clang/AST/StmtCXX.h" 28 #include "clang/AST/StmtObjC.h" 29 #include "clang/Lex/Preprocessor.h" 30 #include "llvm/ADT/BitVector.h" 31 #include "llvm/ADT/STLExtras.h" 32 #include "llvm/Support/raw_ostream.h" 33 #include "clang/Basic/TargetBuiltins.h" 34 #include "clang/Basic/TargetInfo.h" 35 #include "clang/Basic/ConvertUTF.h" 36 #include <limits> 37 using namespace clang; 38 using namespace sema; 39 40 SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 41 unsigned ByteNo) const { 42 return SL->getLocationOfByte(ByteNo, PP.getSourceManager(), 43 PP.getLangOptions(), PP.getTargetInfo()); 44 } 45 46 47 /// CheckablePrintfAttr - does a function call have a "printf" attribute 48 /// and arguments that merit checking? 49 bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) { 50 if (Format->getType() == "printf") return true; 51 if (Format->getType() == "printf0") { 52 // printf0 allows null "format" string; if so don't check format/args 53 unsigned format_idx = Format->getFormatIdx() - 1; 54 // Does the index refer to the implicit object argument? 55 if (isa<CXXMemberCallExpr>(TheCall)) { 56 if (format_idx == 0) 57 return false; 58 --format_idx; 59 } 60 if (format_idx < TheCall->getNumArgs()) { 61 Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts(); 62 if (!Format->isNullPointerConstant(Context, 63 Expr::NPC_ValueDependentIsNull)) 64 return true; 65 } 66 } 67 return false; 68 } 69 70 /// Checks that a call expression's argument count is the desired number. 71 /// This is useful when doing custom type-checking. Returns true on error. 72 static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) { 73 unsigned argCount = call->getNumArgs(); 74 if (argCount == desiredArgCount) return false; 75 76 if (argCount < desiredArgCount) 77 return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args) 78 << 0 /*function call*/ << desiredArgCount << argCount 79 << call->getSourceRange(); 80 81 // Highlight all the excess arguments. 82 SourceRange range(call->getArg(desiredArgCount)->getLocStart(), 83 call->getArg(argCount - 1)->getLocEnd()); 84 85 return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args) 86 << 0 /*function call*/ << desiredArgCount << argCount 87 << call->getArg(1)->getSourceRange(); 88 } 89 90 ExprResult 91 Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 92 ExprResult TheCallResult(Owned(TheCall)); 93 94 // Find out if any arguments are required to be integer constant expressions. 95 unsigned ICEArguments = 0; 96 ASTContext::GetBuiltinTypeError Error; 97 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments); 98 if (Error != ASTContext::GE_None) 99 ICEArguments = 0; // Don't diagnose previously diagnosed errors. 100 101 // If any arguments are required to be ICE's, check and diagnose. 102 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) { 103 // Skip arguments not required to be ICE's. 104 if ((ICEArguments & (1 << ArgNo)) == 0) continue; 105 106 llvm::APSInt Result; 107 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 108 return true; 109 ICEArguments &= ~(1 << ArgNo); 110 } 111 112 switch (BuiltinID) { 113 case Builtin::BI__builtin___CFStringMakeConstantString: 114 assert(TheCall->getNumArgs() == 1 && 115 "Wrong # arguments to builtin CFStringMakeConstantString"); 116 if (CheckObjCString(TheCall->getArg(0))) 117 return ExprError(); 118 break; 119 case Builtin::BI__builtin_stdarg_start: 120 case Builtin::BI__builtin_va_start: 121 if (SemaBuiltinVAStart(TheCall)) 122 return ExprError(); 123 break; 124 case Builtin::BI__builtin_isgreater: 125 case Builtin::BI__builtin_isgreaterequal: 126 case Builtin::BI__builtin_isless: 127 case Builtin::BI__builtin_islessequal: 128 case Builtin::BI__builtin_islessgreater: 129 case Builtin::BI__builtin_isunordered: 130 if (SemaBuiltinUnorderedCompare(TheCall)) 131 return ExprError(); 132 break; 133 case Builtin::BI__builtin_fpclassify: 134 if (SemaBuiltinFPClassification(TheCall, 6)) 135 return ExprError(); 136 break; 137 case Builtin::BI__builtin_isfinite: 138 case Builtin::BI__builtin_isinf: 139 case Builtin::BI__builtin_isinf_sign: 140 case Builtin::BI__builtin_isnan: 141 case Builtin::BI__builtin_isnormal: 142 if (SemaBuiltinFPClassification(TheCall, 1)) 143 return ExprError(); 144 break; 145 case Builtin::BI__builtin_shufflevector: 146 return SemaBuiltinShuffleVector(TheCall); 147 // TheCall will be freed by the smart pointer here, but that's fine, since 148 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 149 case Builtin::BI__builtin_prefetch: 150 if (SemaBuiltinPrefetch(TheCall)) 151 return ExprError(); 152 break; 153 case Builtin::BI__builtin_object_size: 154 if (SemaBuiltinObjectSize(TheCall)) 155 return ExprError(); 156 break; 157 case Builtin::BI__builtin_longjmp: 158 if (SemaBuiltinLongjmp(TheCall)) 159 return ExprError(); 160 break; 161 162 case Builtin::BI__builtin_classify_type: 163 if (checkArgCount(*this, TheCall, 1)) return true; 164 TheCall->setType(Context.IntTy); 165 break; 166 case Builtin::BI__builtin_constant_p: 167 if (checkArgCount(*this, TheCall, 1)) return true; 168 TheCall->setType(Context.IntTy); 169 break; 170 case Builtin::BI__sync_fetch_and_add: 171 case Builtin::BI__sync_fetch_and_sub: 172 case Builtin::BI__sync_fetch_and_or: 173 case Builtin::BI__sync_fetch_and_and: 174 case Builtin::BI__sync_fetch_and_xor: 175 case Builtin::BI__sync_add_and_fetch: 176 case Builtin::BI__sync_sub_and_fetch: 177 case Builtin::BI__sync_and_and_fetch: 178 case Builtin::BI__sync_or_and_fetch: 179 case Builtin::BI__sync_xor_and_fetch: 180 case Builtin::BI__sync_val_compare_and_swap: 181 case Builtin::BI__sync_bool_compare_and_swap: 182 case Builtin::BI__sync_lock_test_and_set: 183 case Builtin::BI__sync_lock_release: 184 case Builtin::BI__sync_swap: 185 return SemaBuiltinAtomicOverloaded(move(TheCallResult)); 186 } 187 188 // Since the target specific builtins for each arch overlap, only check those 189 // of the arch we are compiling for. 190 if (BuiltinID >= Builtin::FirstTSBuiltin) { 191 switch (Context.Target.getTriple().getArch()) { 192 case llvm::Triple::arm: 193 case llvm::Triple::thumb: 194 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 195 return ExprError(); 196 break; 197 default: 198 break; 199 } 200 } 201 202 return move(TheCallResult); 203 } 204 205 // Get the valid immediate range for the specified NEON type code. 206 static unsigned RFT(unsigned t, bool shift = false) { 207 bool quad = t & 0x10; 208 209 switch (t & 0x7) { 210 case 0: // i8 211 return shift ? 7 : (8 << (int)quad) - 1; 212 case 1: // i16 213 return shift ? 15 : (4 << (int)quad) - 1; 214 case 2: // i32 215 return shift ? 31 : (2 << (int)quad) - 1; 216 case 3: // i64 217 return shift ? 63 : (1 << (int)quad) - 1; 218 case 4: // f32 219 assert(!shift && "cannot shift float types!"); 220 return (2 << (int)quad) - 1; 221 case 5: // poly8 222 return shift ? 7 : (8 << (int)quad) - 1; 223 case 6: // poly16 224 return shift ? 15 : (4 << (int)quad) - 1; 225 case 7: // float16 226 assert(!shift && "cannot shift float types!"); 227 return (4 << (int)quad) - 1; 228 } 229 return 0; 230 } 231 232 bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 233 llvm::APSInt Result; 234 235 unsigned mask = 0; 236 unsigned TV = 0; 237 switch (BuiltinID) { 238 #define GET_NEON_OVERLOAD_CHECK 239 #include "clang/Basic/arm_neon.inc" 240 #undef GET_NEON_OVERLOAD_CHECK 241 } 242 243 // For NEON intrinsics which are overloaded on vector element type, validate 244 // the immediate which specifies which variant to emit. 245 if (mask) { 246 unsigned ArgNo = TheCall->getNumArgs()-1; 247 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 248 return true; 249 250 TV = Result.getLimitedValue(32); 251 if ((TV > 31) || (mask & (1 << TV)) == 0) 252 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 253 << TheCall->getArg(ArgNo)->getSourceRange(); 254 } 255 256 // For NEON intrinsics which take an immediate value as part of the 257 // instruction, range check them here. 258 unsigned i = 0, l = 0, u = 0; 259 switch (BuiltinID) { 260 default: return false; 261 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 262 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 263 case ARM::BI__builtin_arm_vcvtr_f: 264 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 265 #define GET_NEON_IMMEDIATE_CHECK 266 #include "clang/Basic/arm_neon.inc" 267 #undef GET_NEON_IMMEDIATE_CHECK 268 }; 269 270 // Check that the immediate argument is actually a constant. 271 if (SemaBuiltinConstantArg(TheCall, i, Result)) 272 return true; 273 274 // Range check against the upper/lower values for this isntruction. 275 unsigned Val = Result.getZExtValue(); 276 if (Val < l || Val > (u + l)) 277 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 278 << l << u+l << TheCall->getArg(i)->getSourceRange(); 279 280 // FIXME: VFP Intrinsics should error if VFP not present. 281 return false; 282 } 283 284 /// CheckFunctionCall - Check a direct function call for various correctness 285 /// and safety properties not strictly enforced by the C type system. 286 bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 287 // Get the IdentifierInfo* for the called function. 288 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 289 290 // None of the checks below are needed for functions that don't have 291 // simple names (e.g., C++ conversion functions). 292 if (!FnInfo) 293 return false; 294 295 // FIXME: This mechanism should be abstracted to be less fragile and 296 // more efficient. For example, just map function ids to custom 297 // handlers. 298 299 // Printf and scanf checking. 300 for (specific_attr_iterator<FormatAttr> 301 i = FDecl->specific_attr_begin<FormatAttr>(), 302 e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { 303 304 const FormatAttr *Format = *i; 305 const bool b = Format->getType() == "scanf"; 306 if (b || CheckablePrintfAttr(Format, TheCall)) { 307 bool HasVAListArg = Format->getFirstArg() == 0; 308 CheckPrintfScanfArguments(TheCall, HasVAListArg, 309 Format->getFormatIdx() - 1, 310 HasVAListArg ? 0 : Format->getFirstArg() - 1, 311 !b); 312 } 313 } 314 315 for (specific_attr_iterator<NonNullAttr> 316 i = FDecl->specific_attr_begin<NonNullAttr>(), 317 e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) { 318 CheckNonNullArguments(*i, TheCall->getArgs(), 319 TheCall->getCallee()->getLocStart()); 320 } 321 322 // Memset/memcpy/memmove handling 323 int CMF = -1; 324 switch (FDecl->getBuiltinID()) { 325 case Builtin::BI__builtin_memset: 326 case Builtin::BI__builtin___memset_chk: 327 case Builtin::BImemset: 328 CMF = CMF_Memset; 329 break; 330 331 case Builtin::BI__builtin_memcpy: 332 case Builtin::BI__builtin___memcpy_chk: 333 case Builtin::BImemcpy: 334 CMF = CMF_Memcpy; 335 break; 336 337 case Builtin::BI__builtin_memmove: 338 case Builtin::BI__builtin___memmove_chk: 339 case Builtin::BImemmove: 340 CMF = CMF_Memmove; 341 break; 342 343 default: 344 if (FDecl->getLinkage() == ExternalLinkage && 345 (!getLangOptions().CPlusPlus || FDecl->isExternC())) { 346 if (FnInfo->isStr("memset")) 347 CMF = CMF_Memset; 348 else if (FnInfo->isStr("memcpy")) 349 CMF = CMF_Memcpy; 350 else if (FnInfo->isStr("memmove")) 351 CMF = CMF_Memmove; 352 } 353 break; 354 } 355 356 if (CMF != -1) 357 CheckMemsetcpymoveArguments(TheCall, CheckedMemoryFunction(CMF), FnInfo); 358 359 return false; 360 } 361 362 bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 363 // Printf checking. 364 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 365 if (!Format) 366 return false; 367 368 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 369 if (!V) 370 return false; 371 372 QualType Ty = V->getType(); 373 if (!Ty->isBlockPointerType()) 374 return false; 375 376 const bool b = Format->getType() == "scanf"; 377 if (!b && !CheckablePrintfAttr(Format, TheCall)) 378 return false; 379 380 bool HasVAListArg = Format->getFirstArg() == 0; 381 CheckPrintfScanfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 382 HasVAListArg ? 0 : Format->getFirstArg() - 1, !b); 383 384 return false; 385 } 386 387 /// SemaBuiltinAtomicOverloaded - We have a call to a function like 388 /// __sync_fetch_and_add, which is an overloaded function based on the pointer 389 /// type of its first argument. The main ActOnCallExpr routines have already 390 /// promoted the types of arguments because all of these calls are prototyped as 391 /// void(...). 392 /// 393 /// This function goes through and does final semantic checking for these 394 /// builtins, 395 ExprResult 396 Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 397 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 398 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 399 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 400 401 // Ensure that we have at least one argument to do type inference from. 402 if (TheCall->getNumArgs() < 1) { 403 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 404 << 0 << 1 << TheCall->getNumArgs() 405 << TheCall->getCallee()->getSourceRange(); 406 return ExprError(); 407 } 408 409 // Inspect the first argument of the atomic builtin. This should always be 410 // a pointer type, whose element is an integral scalar or pointer type. 411 // Because it is a pointer type, we don't have to worry about any implicit 412 // casts here. 413 // FIXME: We don't allow floating point scalars as input. 414 Expr *FirstArg = TheCall->getArg(0); 415 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 416 if (!pointerType) { 417 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 418 << FirstArg->getType() << FirstArg->getSourceRange(); 419 return ExprError(); 420 } 421 422 QualType ValType = pointerType->getPointeeType(); 423 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 424 !ValType->isBlockPointerType()) { 425 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 426 << FirstArg->getType() << FirstArg->getSourceRange(); 427 return ExprError(); 428 } 429 430 switch (ValType.getObjCLifetime()) { 431 case Qualifiers::OCL_None: 432 case Qualifiers::OCL_ExplicitNone: 433 // okay 434 break; 435 436 case Qualifiers::OCL_Weak: 437 case Qualifiers::OCL_Strong: 438 case Qualifiers::OCL_Autoreleasing: 439 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 440 << ValType << FirstArg->getSourceRange(); 441 return ExprError(); 442 } 443 444 // The majority of builtins return a value, but a few have special return 445 // types, so allow them to override appropriately below. 446 QualType ResultType = ValType; 447 448 // We need to figure out which concrete builtin this maps onto. For example, 449 // __sync_fetch_and_add with a 2 byte object turns into 450 // __sync_fetch_and_add_2. 451 #define BUILTIN_ROW(x) \ 452 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 453 Builtin::BI##x##_8, Builtin::BI##x##_16 } 454 455 static const unsigned BuiltinIndices[][5] = { 456 BUILTIN_ROW(__sync_fetch_and_add), 457 BUILTIN_ROW(__sync_fetch_and_sub), 458 BUILTIN_ROW(__sync_fetch_and_or), 459 BUILTIN_ROW(__sync_fetch_and_and), 460 BUILTIN_ROW(__sync_fetch_and_xor), 461 462 BUILTIN_ROW(__sync_add_and_fetch), 463 BUILTIN_ROW(__sync_sub_and_fetch), 464 BUILTIN_ROW(__sync_and_and_fetch), 465 BUILTIN_ROW(__sync_or_and_fetch), 466 BUILTIN_ROW(__sync_xor_and_fetch), 467 468 BUILTIN_ROW(__sync_val_compare_and_swap), 469 BUILTIN_ROW(__sync_bool_compare_and_swap), 470 BUILTIN_ROW(__sync_lock_test_and_set), 471 BUILTIN_ROW(__sync_lock_release), 472 BUILTIN_ROW(__sync_swap) 473 }; 474 #undef BUILTIN_ROW 475 476 // Determine the index of the size. 477 unsigned SizeIndex; 478 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 479 case 1: SizeIndex = 0; break; 480 case 2: SizeIndex = 1; break; 481 case 4: SizeIndex = 2; break; 482 case 8: SizeIndex = 3; break; 483 case 16: SizeIndex = 4; break; 484 default: 485 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 486 << FirstArg->getType() << FirstArg->getSourceRange(); 487 return ExprError(); 488 } 489 490 // Each of these builtins has one pointer argument, followed by some number of 491 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 492 // that we ignore. Find out which row of BuiltinIndices to read from as well 493 // as the number of fixed args. 494 unsigned BuiltinID = FDecl->getBuiltinID(); 495 unsigned BuiltinIndex, NumFixed = 1; 496 switch (BuiltinID) { 497 default: assert(0 && "Unknown overloaded atomic builtin!"); 498 case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break; 499 case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break; 500 case Builtin::BI__sync_fetch_and_or: BuiltinIndex = 2; break; 501 case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break; 502 case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break; 503 504 case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break; 505 case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break; 506 case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break; 507 case Builtin::BI__sync_or_and_fetch: BuiltinIndex = 8; break; 508 case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break; 509 510 case Builtin::BI__sync_val_compare_and_swap: 511 BuiltinIndex = 10; 512 NumFixed = 2; 513 break; 514 case Builtin::BI__sync_bool_compare_and_swap: 515 BuiltinIndex = 11; 516 NumFixed = 2; 517 ResultType = Context.BoolTy; 518 break; 519 case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break; 520 case Builtin::BI__sync_lock_release: 521 BuiltinIndex = 13; 522 NumFixed = 0; 523 ResultType = Context.VoidTy; 524 break; 525 case Builtin::BI__sync_swap: BuiltinIndex = 14; break; 526 } 527 528 // Now that we know how many fixed arguments we expect, first check that we 529 // have at least that many. 530 if (TheCall->getNumArgs() < 1+NumFixed) { 531 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 532 << 0 << 1+NumFixed << TheCall->getNumArgs() 533 << TheCall->getCallee()->getSourceRange(); 534 return ExprError(); 535 } 536 537 // Get the decl for the concrete builtin from this, we can tell what the 538 // concrete integer type we should convert to is. 539 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 540 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 541 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 542 FunctionDecl *NewBuiltinDecl = 543 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 544 TUScope, false, DRE->getLocStart())); 545 546 // The first argument --- the pointer --- has a fixed type; we 547 // deduce the types of the rest of the arguments accordingly. Walk 548 // the remaining arguments, converting them to the deduced value type. 549 for (unsigned i = 0; i != NumFixed; ++i) { 550 ExprResult Arg = TheCall->getArg(i+1); 551 552 // If the argument is an implicit cast, then there was a promotion due to 553 // "...", just remove it now. 554 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg.get())) { 555 Arg = ICE->getSubExpr(); 556 ICE->setSubExpr(0); 557 TheCall->setArg(i+1, Arg.get()); 558 } 559 560 // GCC does an implicit conversion to the pointer or integer ValType. This 561 // can fail in some cases (1i -> int**), check for this error case now. 562 CastKind Kind = CK_Invalid; 563 ExprValueKind VK = VK_RValue; 564 CXXCastPath BasePath; 565 Arg = CheckCastTypes(Arg.get()->getLocStart(), Arg.get()->getSourceRange(), 566 ValType, Arg.take(), Kind, VK, BasePath); 567 if (Arg.isInvalid()) 568 return ExprError(); 569 570 // Okay, we have something that *can* be converted to the right type. Check 571 // to see if there is a potentially weird extension going on here. This can 572 // happen when you do an atomic operation on something like an char* and 573 // pass in 42. The 42 gets converted to char. This is even more strange 574 // for things like 45.123 -> char, etc. 575 // FIXME: Do this check. 576 Arg = ImpCastExprToType(Arg.take(), ValType, Kind, VK, &BasePath); 577 TheCall->setArg(i+1, Arg.get()); 578 } 579 580 // Switch the DeclRefExpr to refer to the new decl. 581 DRE->setDecl(NewBuiltinDecl); 582 DRE->setType(NewBuiltinDecl->getType()); 583 584 // Set the callee in the CallExpr. 585 // FIXME: This leaks the original parens and implicit casts. 586 ExprResult PromotedCall = UsualUnaryConversions(DRE); 587 if (PromotedCall.isInvalid()) 588 return ExprError(); 589 TheCall->setCallee(PromotedCall.take()); 590 591 // Change the result type of the call to match the original value type. This 592 // is arbitrary, but the codegen for these builtins ins design to handle it 593 // gracefully. 594 TheCall->setType(ResultType); 595 596 return move(TheCallResult); 597 } 598 599 600 /// CheckObjCString - Checks that the argument to the builtin 601 /// CFString constructor is correct 602 /// Note: It might also make sense to do the UTF-16 conversion here (would 603 /// simplify the backend). 604 bool Sema::CheckObjCString(Expr *Arg) { 605 Arg = Arg->IgnoreParenCasts(); 606 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 607 608 if (!Literal || Literal->isWide()) { 609 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 610 << Arg->getSourceRange(); 611 return true; 612 } 613 614 if (Literal->containsNonAsciiOrNull()) { 615 llvm::StringRef String = Literal->getString(); 616 unsigned NumBytes = String.size(); 617 llvm::SmallVector<UTF16, 128> ToBuf(NumBytes); 618 const UTF8 *FromPtr = (UTF8 *)String.data(); 619 UTF16 *ToPtr = &ToBuf[0]; 620 621 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 622 &ToPtr, ToPtr + NumBytes, 623 strictConversion); 624 // Check for conversion failure. 625 if (Result != conversionOK) 626 Diag(Arg->getLocStart(), 627 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 628 } 629 return false; 630 } 631 632 /// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 633 /// Emit an error and return true on failure, return false on success. 634 bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 635 Expr *Fn = TheCall->getCallee(); 636 if (TheCall->getNumArgs() > 2) { 637 Diag(TheCall->getArg(2)->getLocStart(), 638 diag::err_typecheck_call_too_many_args) 639 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 640 << Fn->getSourceRange() 641 << SourceRange(TheCall->getArg(2)->getLocStart(), 642 (*(TheCall->arg_end()-1))->getLocEnd()); 643 return true; 644 } 645 646 if (TheCall->getNumArgs() < 2) { 647 return Diag(TheCall->getLocEnd(), 648 diag::err_typecheck_call_too_few_args_at_least) 649 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 650 } 651 652 // Determine whether the current function is variadic or not. 653 BlockScopeInfo *CurBlock = getCurBlock(); 654 bool isVariadic; 655 if (CurBlock) 656 isVariadic = CurBlock->TheDecl->isVariadic(); 657 else if (FunctionDecl *FD = getCurFunctionDecl()) 658 isVariadic = FD->isVariadic(); 659 else 660 isVariadic = getCurMethodDecl()->isVariadic(); 661 662 if (!isVariadic) { 663 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 664 return true; 665 } 666 667 // Verify that the second argument to the builtin is the last argument of the 668 // current function or method. 669 bool SecondArgIsLastNamedArgument = false; 670 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 671 672 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 673 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 674 // FIXME: This isn't correct for methods (results in bogus warning). 675 // Get the last formal in the current function. 676 const ParmVarDecl *LastArg; 677 if (CurBlock) 678 LastArg = *(CurBlock->TheDecl->param_end()-1); 679 else if (FunctionDecl *FD = getCurFunctionDecl()) 680 LastArg = *(FD->param_end()-1); 681 else 682 LastArg = *(getCurMethodDecl()->param_end()-1); 683 SecondArgIsLastNamedArgument = PV == LastArg; 684 } 685 } 686 687 if (!SecondArgIsLastNamedArgument) 688 Diag(TheCall->getArg(1)->getLocStart(), 689 diag::warn_second_parameter_of_va_start_not_last_named_argument); 690 return false; 691 } 692 693 /// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 694 /// friends. This is declared to take (...), so we have to check everything. 695 bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 696 if (TheCall->getNumArgs() < 2) 697 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 698 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 699 if (TheCall->getNumArgs() > 2) 700 return Diag(TheCall->getArg(2)->getLocStart(), 701 diag::err_typecheck_call_too_many_args) 702 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 703 << SourceRange(TheCall->getArg(2)->getLocStart(), 704 (*(TheCall->arg_end()-1))->getLocEnd()); 705 706 ExprResult OrigArg0 = TheCall->getArg(0); 707 ExprResult OrigArg1 = TheCall->getArg(1); 708 709 // Do standard promotions between the two arguments, returning their common 710 // type. 711 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 712 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 713 return true; 714 715 // Make sure any conversions are pushed back into the call; this is 716 // type safe since unordered compare builtins are declared as "_Bool 717 // foo(...)". 718 TheCall->setArg(0, OrigArg0.get()); 719 TheCall->setArg(1, OrigArg1.get()); 720 721 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 722 return false; 723 724 // If the common type isn't a real floating type, then the arguments were 725 // invalid for this operation. 726 if (!Res->isRealFloatingType()) 727 return Diag(OrigArg0.get()->getLocStart(), 728 diag::err_typecheck_call_invalid_ordered_compare) 729 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 730 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 731 732 return false; 733 } 734 735 /// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 736 /// __builtin_isnan and friends. This is declared to take (...), so we have 737 /// to check everything. We expect the last argument to be a floating point 738 /// value. 739 bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 740 if (TheCall->getNumArgs() < NumArgs) 741 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 742 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 743 if (TheCall->getNumArgs() > NumArgs) 744 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 745 diag::err_typecheck_call_too_many_args) 746 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 747 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 748 (*(TheCall->arg_end()-1))->getLocEnd()); 749 750 Expr *OrigArg = TheCall->getArg(NumArgs-1); 751 752 if (OrigArg->isTypeDependent()) 753 return false; 754 755 // This operation requires a non-_Complex floating-point number. 756 if (!OrigArg->getType()->isRealFloatingType()) 757 return Diag(OrigArg->getLocStart(), 758 diag::err_typecheck_call_invalid_unary_fp) 759 << OrigArg->getType() << OrigArg->getSourceRange(); 760 761 // If this is an implicit conversion from float -> double, remove it. 762 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 763 Expr *CastArg = Cast->getSubExpr(); 764 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 765 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 766 "promotion from float to double is the only expected cast here"); 767 Cast->setSubExpr(0); 768 TheCall->setArg(NumArgs-1, CastArg); 769 OrigArg = CastArg; 770 } 771 } 772 773 return false; 774 } 775 776 /// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 777 // This is declared to take (...), so we have to check everything. 778 ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 779 if (TheCall->getNumArgs() < 2) 780 return ExprError(Diag(TheCall->getLocEnd(), 781 diag::err_typecheck_call_too_few_args_at_least) 782 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 783 << TheCall->getSourceRange()); 784 785 // Determine which of the following types of shufflevector we're checking: 786 // 1) unary, vector mask: (lhs, mask) 787 // 2) binary, vector mask: (lhs, rhs, mask) 788 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 789 QualType resType = TheCall->getArg(0)->getType(); 790 unsigned numElements = 0; 791 792 if (!TheCall->getArg(0)->isTypeDependent() && 793 !TheCall->getArg(1)->isTypeDependent()) { 794 QualType LHSType = TheCall->getArg(0)->getType(); 795 QualType RHSType = TheCall->getArg(1)->getType(); 796 797 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 798 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 799 << SourceRange(TheCall->getArg(0)->getLocStart(), 800 TheCall->getArg(1)->getLocEnd()); 801 return ExprError(); 802 } 803 804 numElements = LHSType->getAs<VectorType>()->getNumElements(); 805 unsigned numResElements = TheCall->getNumArgs() - 2; 806 807 // Check to see if we have a call with 2 vector arguments, the unary shuffle 808 // with mask. If so, verify that RHS is an integer vector type with the 809 // same number of elts as lhs. 810 if (TheCall->getNumArgs() == 2) { 811 if (!RHSType->hasIntegerRepresentation() || 812 RHSType->getAs<VectorType>()->getNumElements() != numElements) 813 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 814 << SourceRange(TheCall->getArg(1)->getLocStart(), 815 TheCall->getArg(1)->getLocEnd()); 816 numResElements = numElements; 817 } 818 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 819 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 820 << SourceRange(TheCall->getArg(0)->getLocStart(), 821 TheCall->getArg(1)->getLocEnd()); 822 return ExprError(); 823 } else if (numElements != numResElements) { 824 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 825 resType = Context.getVectorType(eltType, numResElements, 826 VectorType::GenericVector); 827 } 828 } 829 830 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 831 if (TheCall->getArg(i)->isTypeDependent() || 832 TheCall->getArg(i)->isValueDependent()) 833 continue; 834 835 llvm::APSInt Result(32); 836 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 837 return ExprError(Diag(TheCall->getLocStart(), 838 diag::err_shufflevector_nonconstant_argument) 839 << TheCall->getArg(i)->getSourceRange()); 840 841 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 842 return ExprError(Diag(TheCall->getLocStart(), 843 diag::err_shufflevector_argument_too_large) 844 << TheCall->getArg(i)->getSourceRange()); 845 } 846 847 llvm::SmallVector<Expr*, 32> exprs; 848 849 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 850 exprs.push_back(TheCall->getArg(i)); 851 TheCall->setArg(i, 0); 852 } 853 854 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 855 exprs.size(), resType, 856 TheCall->getCallee()->getLocStart(), 857 TheCall->getRParenLoc())); 858 } 859 860 /// SemaBuiltinPrefetch - Handle __builtin_prefetch. 861 // This is declared to take (const void*, ...) and can take two 862 // optional constant int args. 863 bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 864 unsigned NumArgs = TheCall->getNumArgs(); 865 866 if (NumArgs > 3) 867 return Diag(TheCall->getLocEnd(), 868 diag::err_typecheck_call_too_many_args_at_most) 869 << 0 /*function call*/ << 3 << NumArgs 870 << TheCall->getSourceRange(); 871 872 // Argument 0 is checked for us and the remaining arguments must be 873 // constant integers. 874 for (unsigned i = 1; i != NumArgs; ++i) { 875 Expr *Arg = TheCall->getArg(i); 876 877 llvm::APSInt Result; 878 if (SemaBuiltinConstantArg(TheCall, i, Result)) 879 return true; 880 881 // FIXME: gcc issues a warning and rewrites these to 0. These 882 // seems especially odd for the third argument since the default 883 // is 3. 884 if (i == 1) { 885 if (Result.getLimitedValue() > 1) 886 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 887 << "0" << "1" << Arg->getSourceRange(); 888 } else { 889 if (Result.getLimitedValue() > 3) 890 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 891 << "0" << "3" << Arg->getSourceRange(); 892 } 893 } 894 895 return false; 896 } 897 898 /// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 899 /// TheCall is a constant expression. 900 bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 901 llvm::APSInt &Result) { 902 Expr *Arg = TheCall->getArg(ArgNum); 903 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 904 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 905 906 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 907 908 if (!Arg->isIntegerConstantExpr(Result, Context)) 909 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 910 << FDecl->getDeclName() << Arg->getSourceRange(); 911 912 return false; 913 } 914 915 /// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 916 /// int type). This simply type checks that type is one of the defined 917 /// constants (0-3). 918 // For compatibility check 0-3, llvm only handles 0 and 2. 919 bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 920 llvm::APSInt Result; 921 922 // Check constant-ness first. 923 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 924 return true; 925 926 Expr *Arg = TheCall->getArg(1); 927 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 928 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 929 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 930 } 931 932 return false; 933 } 934 935 /// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 936 /// This checks that val is a constant 1. 937 bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 938 Expr *Arg = TheCall->getArg(1); 939 llvm::APSInt Result; 940 941 // TODO: This is less than ideal. Overload this to take a value. 942 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 943 return true; 944 945 if (Result != 1) 946 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 947 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 948 949 return false; 950 } 951 952 // Handle i > 1 ? "x" : "y", recursively. 953 bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 954 bool HasVAListArg, 955 unsigned format_idx, unsigned firstDataArg, 956 bool isPrintf) { 957 tryAgain: 958 if (E->isTypeDependent() || E->isValueDependent()) 959 return false; 960 961 E = E->IgnoreParens(); 962 963 switch (E->getStmtClass()) { 964 case Stmt::BinaryConditionalOperatorClass: 965 case Stmt::ConditionalOperatorClass: { 966 const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E); 967 return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, HasVAListArg, 968 format_idx, firstDataArg, isPrintf) 969 && SemaCheckStringLiteral(C->getFalseExpr(), TheCall, HasVAListArg, 970 format_idx, firstDataArg, isPrintf); 971 } 972 973 case Stmt::IntegerLiteralClass: 974 // Technically -Wformat-nonliteral does not warn about this case. 975 // The behavior of printf and friends in this case is implementation 976 // dependent. Ideally if the format string cannot be null then 977 // it should have a 'nonnull' attribute in the function prototype. 978 return true; 979 980 case Stmt::ImplicitCastExprClass: { 981 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 982 goto tryAgain; 983 } 984 985 case Stmt::OpaqueValueExprClass: 986 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 987 E = src; 988 goto tryAgain; 989 } 990 return false; 991 992 case Stmt::PredefinedExprClass: 993 // While __func__, etc., are technically not string literals, they 994 // cannot contain format specifiers and thus are not a security 995 // liability. 996 return true; 997 998 case Stmt::DeclRefExprClass: { 999 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1000 1001 // As an exception, do not flag errors for variables binding to 1002 // const string literals. 1003 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1004 bool isConstant = false; 1005 QualType T = DR->getType(); 1006 1007 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1008 isConstant = AT->getElementType().isConstant(Context); 1009 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1010 isConstant = T.isConstant(Context) && 1011 PT->getPointeeType().isConstant(Context); 1012 } 1013 1014 if (isConstant) { 1015 if (const Expr *Init = VD->getAnyInitializer()) 1016 return SemaCheckStringLiteral(Init, TheCall, 1017 HasVAListArg, format_idx, firstDataArg, 1018 isPrintf); 1019 } 1020 1021 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1022 // special check to see if the format string is a function parameter 1023 // of the function calling the printf function. If the function 1024 // has an attribute indicating it is a printf-like function, then we 1025 // should suppress warnings concerning non-literals being used in a call 1026 // to a vprintf function. For example: 1027 // 1028 // void 1029 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1030 // va_list ap; 1031 // va_start(ap, fmt); 1032 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1033 // ... 1034 // 1035 // 1036 // FIXME: We don't have full attribute support yet, so just check to see 1037 // if the argument is a DeclRefExpr that references a parameter. We'll 1038 // add proper support for checking the attribute later. 1039 if (HasVAListArg) 1040 if (isa<ParmVarDecl>(VD)) 1041 return true; 1042 } 1043 1044 return false; 1045 } 1046 1047 case Stmt::CallExprClass: { 1048 const CallExpr *CE = cast<CallExpr>(E); 1049 if (const ImplicitCastExpr *ICE 1050 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 1051 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 1052 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 1053 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 1054 unsigned ArgIndex = FA->getFormatIdx(); 1055 const Expr *Arg = CE->getArg(ArgIndex - 1); 1056 1057 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 1058 format_idx, firstDataArg, isPrintf); 1059 } 1060 } 1061 } 1062 } 1063 1064 return false; 1065 } 1066 case Stmt::ObjCStringLiteralClass: 1067 case Stmt::StringLiteralClass: { 1068 const StringLiteral *StrE = NULL; 1069 1070 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1071 StrE = ObjCFExpr->getString(); 1072 else 1073 StrE = cast<StringLiteral>(E); 1074 1075 if (StrE) { 1076 CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx, 1077 firstDataArg, isPrintf); 1078 return true; 1079 } 1080 1081 return false; 1082 } 1083 1084 default: 1085 return false; 1086 } 1087 } 1088 1089 void 1090 Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1091 const Expr * const *ExprArgs, 1092 SourceLocation CallSiteLoc) { 1093 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1094 e = NonNull->args_end(); 1095 i != e; ++i) { 1096 const Expr *ArgExpr = ExprArgs[*i]; 1097 if (ArgExpr->isNullPointerConstant(Context, 1098 Expr::NPC_ValueDependentIsNotNull)) 1099 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 1100 } 1101 } 1102 1103 /// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar 1104 /// functions) for correct use of format strings. 1105 void 1106 Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg, 1107 unsigned format_idx, unsigned firstDataArg, 1108 bool isPrintf) { 1109 1110 const Expr *Fn = TheCall->getCallee(); 1111 1112 // The way the format attribute works in GCC, the implicit this argument 1113 // of member functions is counted. However, it doesn't appear in our own 1114 // lists, so decrement format_idx in that case. 1115 if (isa<CXXMemberCallExpr>(TheCall)) { 1116 const CXXMethodDecl *method_decl = 1117 dyn_cast<CXXMethodDecl>(TheCall->getCalleeDecl()); 1118 if (method_decl && method_decl->isInstance()) { 1119 // Catch a format attribute mistakenly referring to the object argument. 1120 if (format_idx == 0) 1121 return; 1122 --format_idx; 1123 if(firstDataArg != 0) 1124 --firstDataArg; 1125 } 1126 } 1127 1128 // CHECK: printf/scanf-like function is called with no format string. 1129 if (format_idx >= TheCall->getNumArgs()) { 1130 Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string) 1131 << Fn->getSourceRange(); 1132 return; 1133 } 1134 1135 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 1136 1137 // CHECK: format string is not a string literal. 1138 // 1139 // Dynamically generated format strings are difficult to 1140 // automatically vet at compile time. Requiring that format strings 1141 // are string literals: (1) permits the checking of format strings by 1142 // the compiler and thereby (2) can practically remove the source of 1143 // many format string exploits. 1144 1145 // Format string can be either ObjC string (e.g. @"%d") or 1146 // C string (e.g. "%d") 1147 // ObjC string uses the same format specifiers as C string, so we can use 1148 // the same format string checking logic for both ObjC and C strings. 1149 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 1150 firstDataArg, isPrintf)) 1151 return; // Literal format string found, check done! 1152 1153 // If there are no arguments specified, warn with -Wformat-security, otherwise 1154 // warn only with -Wformat-nonliteral. 1155 if (TheCall->getNumArgs() == format_idx+1) 1156 Diag(TheCall->getArg(format_idx)->getLocStart(), 1157 diag::warn_format_nonliteral_noargs) 1158 << OrigFormatExpr->getSourceRange(); 1159 else 1160 Diag(TheCall->getArg(format_idx)->getLocStart(), 1161 diag::warn_format_nonliteral) 1162 << OrigFormatExpr->getSourceRange(); 1163 } 1164 1165 namespace { 1166 class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1167 protected: 1168 Sema &S; 1169 const StringLiteral *FExpr; 1170 const Expr *OrigFormatExpr; 1171 const unsigned FirstDataArg; 1172 const unsigned NumDataArgs; 1173 const bool IsObjCLiteral; 1174 const char *Beg; // Start of format string. 1175 const bool HasVAListArg; 1176 const CallExpr *TheCall; 1177 unsigned FormatIdx; 1178 llvm::BitVector CoveredArgs; 1179 bool usesPositionalArgs; 1180 bool atFirstArg; 1181 public: 1182 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1183 const Expr *origFormatExpr, unsigned firstDataArg, 1184 unsigned numDataArgs, bool isObjCLiteral, 1185 const char *beg, bool hasVAListArg, 1186 const CallExpr *theCall, unsigned formatIdx) 1187 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1188 FirstDataArg(firstDataArg), 1189 NumDataArgs(numDataArgs), 1190 IsObjCLiteral(isObjCLiteral), Beg(beg), 1191 HasVAListArg(hasVAListArg), 1192 TheCall(theCall), FormatIdx(formatIdx), 1193 usesPositionalArgs(false), atFirstArg(true) { 1194 CoveredArgs.resize(numDataArgs); 1195 CoveredArgs.reset(); 1196 } 1197 1198 void DoneProcessing(); 1199 1200 void HandleIncompleteSpecifier(const char *startSpecifier, 1201 unsigned specifierLen); 1202 1203 virtual void HandleInvalidPosition(const char *startSpecifier, 1204 unsigned specifierLen, 1205 analyze_format_string::PositionContext p); 1206 1207 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1208 1209 void HandleNullChar(const char *nullCharacter); 1210 1211 protected: 1212 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1213 const char *startSpec, 1214 unsigned specifierLen, 1215 const char *csStart, unsigned csLen); 1216 1217 SourceRange getFormatStringRange(); 1218 CharSourceRange getSpecifierRange(const char *startSpecifier, 1219 unsigned specifierLen); 1220 SourceLocation getLocationOfByte(const char *x); 1221 1222 const Expr *getDataArg(unsigned i) const; 1223 1224 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1225 const analyze_format_string::ConversionSpecifier &CS, 1226 const char *startSpecifier, unsigned specifierLen, 1227 unsigned argIndex); 1228 }; 1229 } 1230 1231 SourceRange CheckFormatHandler::getFormatStringRange() { 1232 return OrigFormatExpr->getSourceRange(); 1233 } 1234 1235 CharSourceRange CheckFormatHandler:: 1236 getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1237 SourceLocation Start = getLocationOfByte(startSpecifier); 1238 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1239 1240 // Advance the end SourceLocation by one due to half-open ranges. 1241 End = End.getFileLocWithOffset(1); 1242 1243 return CharSourceRange::getCharRange(Start, End); 1244 } 1245 1246 SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 1247 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1248 } 1249 1250 void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 1251 unsigned specifierLen){ 1252 SourceLocation Loc = getLocationOfByte(startSpecifier); 1253 S.Diag(Loc, diag::warn_printf_incomplete_specifier) 1254 << getSpecifierRange(startSpecifier, specifierLen); 1255 } 1256 1257 void 1258 CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1259 analyze_format_string::PositionContext p) { 1260 SourceLocation Loc = getLocationOfByte(startPos); 1261 S.Diag(Loc, diag::warn_format_invalid_positional_specifier) 1262 << (unsigned) p << getSpecifierRange(startPos, posLen); 1263 } 1264 1265 void CheckFormatHandler::HandleZeroPosition(const char *startPos, 1266 unsigned posLen) { 1267 SourceLocation Loc = getLocationOfByte(startPos); 1268 S.Diag(Loc, diag::warn_format_zero_positional_specifier) 1269 << getSpecifierRange(startPos, posLen); 1270 } 1271 1272 void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 1273 if (!IsObjCLiteral) { 1274 // The presence of a null character is likely an error. 1275 S.Diag(getLocationOfByte(nullCharacter), 1276 diag::warn_printf_format_string_contains_null_char) 1277 << getFormatStringRange(); 1278 } 1279 } 1280 1281 const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 1282 return TheCall->getArg(FirstDataArg + i); 1283 } 1284 1285 void CheckFormatHandler::DoneProcessing() { 1286 // Does the number of data arguments exceed the number of 1287 // format conversions in the format string? 1288 if (!HasVAListArg) { 1289 // Find any arguments that weren't covered. 1290 CoveredArgs.flip(); 1291 signed notCoveredArg = CoveredArgs.find_first(); 1292 if (notCoveredArg >= 0) { 1293 assert((unsigned)notCoveredArg < NumDataArgs); 1294 S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(), 1295 diag::warn_printf_data_arg_not_used) 1296 << getFormatStringRange(); 1297 } 1298 } 1299 } 1300 1301 bool 1302 CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 1303 SourceLocation Loc, 1304 const char *startSpec, 1305 unsigned specifierLen, 1306 const char *csStart, 1307 unsigned csLen) { 1308 1309 bool keepGoing = true; 1310 if (argIndex < NumDataArgs) { 1311 // Consider the argument coverered, even though the specifier doesn't 1312 // make sense. 1313 CoveredArgs.set(argIndex); 1314 } 1315 else { 1316 // If argIndex exceeds the number of data arguments we 1317 // don't issue a warning because that is just a cascade of warnings (and 1318 // they may have intended '%%' anyway). We don't want to continue processing 1319 // the format string after this point, however, as we will like just get 1320 // gibberish when trying to match arguments. 1321 keepGoing = false; 1322 } 1323 1324 S.Diag(Loc, diag::warn_format_invalid_conversion) 1325 << llvm::StringRef(csStart, csLen) 1326 << getSpecifierRange(startSpec, specifierLen); 1327 1328 return keepGoing; 1329 } 1330 1331 bool 1332 CheckFormatHandler::CheckNumArgs( 1333 const analyze_format_string::FormatSpecifier &FS, 1334 const analyze_format_string::ConversionSpecifier &CS, 1335 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 1336 1337 if (argIndex >= NumDataArgs) { 1338 if (FS.usesPositionalArg()) { 1339 S.Diag(getLocationOfByte(CS.getStart()), 1340 diag::warn_printf_positional_arg_exceeds_data_args) 1341 << (argIndex+1) << NumDataArgs 1342 << getSpecifierRange(startSpecifier, specifierLen); 1343 } 1344 else { 1345 S.Diag(getLocationOfByte(CS.getStart()), 1346 diag::warn_printf_insufficient_data_args) 1347 << getSpecifierRange(startSpecifier, specifierLen); 1348 } 1349 1350 return false; 1351 } 1352 return true; 1353 } 1354 1355 //===--- CHECK: Printf format string checking ------------------------------===// 1356 1357 namespace { 1358 class CheckPrintfHandler : public CheckFormatHandler { 1359 public: 1360 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1361 const Expr *origFormatExpr, unsigned firstDataArg, 1362 unsigned numDataArgs, bool isObjCLiteral, 1363 const char *beg, bool hasVAListArg, 1364 const CallExpr *theCall, unsigned formatIdx) 1365 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1366 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1367 theCall, formatIdx) {} 1368 1369 1370 bool HandleInvalidPrintfConversionSpecifier( 1371 const analyze_printf::PrintfSpecifier &FS, 1372 const char *startSpecifier, 1373 unsigned specifierLen); 1374 1375 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 1376 const char *startSpecifier, 1377 unsigned specifierLen); 1378 1379 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 1380 const char *startSpecifier, unsigned specifierLen); 1381 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 1382 const analyze_printf::OptionalAmount &Amt, 1383 unsigned type, 1384 const char *startSpecifier, unsigned specifierLen); 1385 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1386 const analyze_printf::OptionalFlag &flag, 1387 const char *startSpecifier, unsigned specifierLen); 1388 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 1389 const analyze_printf::OptionalFlag &ignoredFlag, 1390 const analyze_printf::OptionalFlag &flag, 1391 const char *startSpecifier, unsigned specifierLen); 1392 }; 1393 } 1394 1395 bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 1396 const analyze_printf::PrintfSpecifier &FS, 1397 const char *startSpecifier, 1398 unsigned specifierLen) { 1399 const analyze_printf::PrintfConversionSpecifier &CS = 1400 FS.getConversionSpecifier(); 1401 1402 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1403 getLocationOfByte(CS.getStart()), 1404 startSpecifier, specifierLen, 1405 CS.getStart(), CS.getLength()); 1406 } 1407 1408 bool CheckPrintfHandler::HandleAmount( 1409 const analyze_format_string::OptionalAmount &Amt, 1410 unsigned k, const char *startSpecifier, 1411 unsigned specifierLen) { 1412 1413 if (Amt.hasDataArgument()) { 1414 if (!HasVAListArg) { 1415 unsigned argIndex = Amt.getArgIndex(); 1416 if (argIndex >= NumDataArgs) { 1417 S.Diag(getLocationOfByte(Amt.getStart()), 1418 diag::warn_printf_asterisk_missing_arg) 1419 << k << getSpecifierRange(startSpecifier, specifierLen); 1420 // Don't do any more checking. We will just emit 1421 // spurious errors. 1422 return false; 1423 } 1424 1425 // Type check the data argument. It should be an 'int'. 1426 // Although not in conformance with C99, we also allow the argument to be 1427 // an 'unsigned int' as that is a reasonably safe case. GCC also 1428 // doesn't emit a warning for that case. 1429 CoveredArgs.set(argIndex); 1430 const Expr *Arg = getDataArg(argIndex); 1431 QualType T = Arg->getType(); 1432 1433 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1434 assert(ATR.isValid()); 1435 1436 if (!ATR.matchesType(S.Context, T)) { 1437 S.Diag(getLocationOfByte(Amt.getStart()), 1438 diag::warn_printf_asterisk_wrong_type) 1439 << k 1440 << ATR.getRepresentativeType(S.Context) << T 1441 << getSpecifierRange(startSpecifier, specifierLen) 1442 << Arg->getSourceRange(); 1443 // Don't do any more checking. We will just emit 1444 // spurious errors. 1445 return false; 1446 } 1447 } 1448 } 1449 return true; 1450 } 1451 1452 void CheckPrintfHandler::HandleInvalidAmount( 1453 const analyze_printf::PrintfSpecifier &FS, 1454 const analyze_printf::OptionalAmount &Amt, 1455 unsigned type, 1456 const char *startSpecifier, 1457 unsigned specifierLen) { 1458 const analyze_printf::PrintfConversionSpecifier &CS = 1459 FS.getConversionSpecifier(); 1460 switch (Amt.getHowSpecified()) { 1461 case analyze_printf::OptionalAmount::Constant: 1462 S.Diag(getLocationOfByte(Amt.getStart()), 1463 diag::warn_printf_nonsensical_optional_amount) 1464 << type 1465 << CS.toString() 1466 << getSpecifierRange(startSpecifier, specifierLen) 1467 << FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 1468 Amt.getConstantLength())); 1469 break; 1470 1471 default: 1472 S.Diag(getLocationOfByte(Amt.getStart()), 1473 diag::warn_printf_nonsensical_optional_amount) 1474 << type 1475 << CS.toString() 1476 << getSpecifierRange(startSpecifier, specifierLen); 1477 break; 1478 } 1479 } 1480 1481 void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1482 const analyze_printf::OptionalFlag &flag, 1483 const char *startSpecifier, 1484 unsigned specifierLen) { 1485 // Warn about pointless flag with a fixit removal. 1486 const analyze_printf::PrintfConversionSpecifier &CS = 1487 FS.getConversionSpecifier(); 1488 S.Diag(getLocationOfByte(flag.getPosition()), 1489 diag::warn_printf_nonsensical_flag) 1490 << flag.toString() << CS.toString() 1491 << getSpecifierRange(startSpecifier, specifierLen) 1492 << FixItHint::CreateRemoval(getSpecifierRange(flag.getPosition(), 1)); 1493 } 1494 1495 void CheckPrintfHandler::HandleIgnoredFlag( 1496 const analyze_printf::PrintfSpecifier &FS, 1497 const analyze_printf::OptionalFlag &ignoredFlag, 1498 const analyze_printf::OptionalFlag &flag, 1499 const char *startSpecifier, 1500 unsigned specifierLen) { 1501 // Warn about ignored flag with a fixit removal. 1502 S.Diag(getLocationOfByte(ignoredFlag.getPosition()), 1503 diag::warn_printf_ignored_flag) 1504 << ignoredFlag.toString() << flag.toString() 1505 << getSpecifierRange(startSpecifier, specifierLen) 1506 << FixItHint::CreateRemoval(getSpecifierRange( 1507 ignoredFlag.getPosition(), 1)); 1508 } 1509 1510 bool 1511 CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 1512 &FS, 1513 const char *startSpecifier, 1514 unsigned specifierLen) { 1515 1516 using namespace analyze_format_string; 1517 using namespace analyze_printf; 1518 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 1519 1520 if (FS.consumesDataArgument()) { 1521 if (atFirstArg) { 1522 atFirstArg = false; 1523 usesPositionalArgs = FS.usesPositionalArg(); 1524 } 1525 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1526 // Cannot mix-and-match positional and non-positional arguments. 1527 S.Diag(getLocationOfByte(CS.getStart()), 1528 diag::warn_format_mix_positional_nonpositional_args) 1529 << getSpecifierRange(startSpecifier, specifierLen); 1530 return false; 1531 } 1532 } 1533 1534 // First check if the field width, precision, and conversion specifier 1535 // have matching data arguments. 1536 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 1537 startSpecifier, specifierLen)) { 1538 return false; 1539 } 1540 1541 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 1542 startSpecifier, specifierLen)) { 1543 return false; 1544 } 1545 1546 if (!CS.consumesDataArgument()) { 1547 // FIXME: Technically specifying a precision or field width here 1548 // makes no sense. Worth issuing a warning at some point. 1549 return true; 1550 } 1551 1552 // Consume the argument. 1553 unsigned argIndex = FS.getArgIndex(); 1554 if (argIndex < NumDataArgs) { 1555 // The check to see if the argIndex is valid will come later. 1556 // We set the bit here because we may exit early from this 1557 // function if we encounter some other error. 1558 CoveredArgs.set(argIndex); 1559 } 1560 1561 // Check for using an Objective-C specific conversion specifier 1562 // in a non-ObjC literal. 1563 if (!IsObjCLiteral && CS.isObjCArg()) { 1564 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 1565 specifierLen); 1566 } 1567 1568 // Check for invalid use of field width 1569 if (!FS.hasValidFieldWidth()) { 1570 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 1571 startSpecifier, specifierLen); 1572 } 1573 1574 // Check for invalid use of precision 1575 if (!FS.hasValidPrecision()) { 1576 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 1577 startSpecifier, specifierLen); 1578 } 1579 1580 // Check each flag does not conflict with any other component. 1581 if (!FS.hasValidThousandsGroupingPrefix()) 1582 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 1583 if (!FS.hasValidLeadingZeros()) 1584 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 1585 if (!FS.hasValidPlusPrefix()) 1586 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 1587 if (!FS.hasValidSpacePrefix()) 1588 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 1589 if (!FS.hasValidAlternativeForm()) 1590 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 1591 if (!FS.hasValidLeftJustified()) 1592 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 1593 1594 // Check that flags are not ignored by another flag 1595 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 1596 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 1597 startSpecifier, specifierLen); 1598 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 1599 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 1600 startSpecifier, specifierLen); 1601 1602 // Check the length modifier is valid with the given conversion specifier. 1603 const LengthModifier &LM = FS.getLengthModifier(); 1604 if (!FS.hasValidLengthModifier()) 1605 S.Diag(getLocationOfByte(LM.getStart()), 1606 diag::warn_format_nonsensical_length) 1607 << LM.toString() << CS.toString() 1608 << getSpecifierRange(startSpecifier, specifierLen) 1609 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 1610 LM.getLength())); 1611 1612 // Are we using '%n'? 1613 if (CS.getKind() == ConversionSpecifier::nArg) { 1614 // Issue a warning about this being a possible security issue. 1615 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back) 1616 << getSpecifierRange(startSpecifier, specifierLen); 1617 // Continue checking the other format specifiers. 1618 return true; 1619 } 1620 1621 // The remaining checks depend on the data arguments. 1622 if (HasVAListArg) 1623 return true; 1624 1625 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 1626 return false; 1627 1628 // Now type check the data expression that matches the 1629 // format specifier. 1630 const Expr *Ex = getDataArg(argIndex); 1631 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context); 1632 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 1633 // Check if we didn't match because of an implicit cast from a 'char' 1634 // or 'short' to an 'int'. This is done because printf is a varargs 1635 // function. 1636 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 1637 if (ICE->getType() == S.Context.IntTy) { 1638 // All further checking is done on the subexpression. 1639 Ex = ICE->getSubExpr(); 1640 if (ATR.matchesType(S.Context, Ex->getType())) 1641 return true; 1642 } 1643 1644 // We may be able to offer a FixItHint if it is a supported type. 1645 PrintfSpecifier fixedFS = FS; 1646 bool success = fixedFS.fixType(Ex->getType()); 1647 1648 if (success) { 1649 // Get the fix string from the fixed format specifier 1650 llvm::SmallString<128> buf; 1651 llvm::raw_svector_ostream os(buf); 1652 fixedFS.toString(os); 1653 1654 // FIXME: getRepresentativeType() perhaps should return a string 1655 // instead of a QualType to better handle when the representative 1656 // type is 'wint_t' (which is defined in the system headers). 1657 S.Diag(getLocationOfByte(CS.getStart()), 1658 diag::warn_printf_conversion_argument_type_mismatch) 1659 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1660 << getSpecifierRange(startSpecifier, specifierLen) 1661 << Ex->getSourceRange() 1662 << FixItHint::CreateReplacement( 1663 getSpecifierRange(startSpecifier, specifierLen), 1664 os.str()); 1665 } 1666 else { 1667 S.Diag(getLocationOfByte(CS.getStart()), 1668 diag::warn_printf_conversion_argument_type_mismatch) 1669 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1670 << getSpecifierRange(startSpecifier, specifierLen) 1671 << Ex->getSourceRange(); 1672 } 1673 } 1674 1675 return true; 1676 } 1677 1678 //===--- CHECK: Scanf format string checking ------------------------------===// 1679 1680 namespace { 1681 class CheckScanfHandler : public CheckFormatHandler { 1682 public: 1683 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 1684 const Expr *origFormatExpr, unsigned firstDataArg, 1685 unsigned numDataArgs, bool isObjCLiteral, 1686 const char *beg, bool hasVAListArg, 1687 const CallExpr *theCall, unsigned formatIdx) 1688 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1689 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1690 theCall, formatIdx) {} 1691 1692 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 1693 const char *startSpecifier, 1694 unsigned specifierLen); 1695 1696 bool HandleInvalidScanfConversionSpecifier( 1697 const analyze_scanf::ScanfSpecifier &FS, 1698 const char *startSpecifier, 1699 unsigned specifierLen); 1700 1701 void HandleIncompleteScanList(const char *start, const char *end); 1702 }; 1703 } 1704 1705 void CheckScanfHandler::HandleIncompleteScanList(const char *start, 1706 const char *end) { 1707 S.Diag(getLocationOfByte(end), diag::warn_scanf_scanlist_incomplete) 1708 << getSpecifierRange(start, end - start); 1709 } 1710 1711 bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 1712 const analyze_scanf::ScanfSpecifier &FS, 1713 const char *startSpecifier, 1714 unsigned specifierLen) { 1715 1716 const analyze_scanf::ScanfConversionSpecifier &CS = 1717 FS.getConversionSpecifier(); 1718 1719 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1720 getLocationOfByte(CS.getStart()), 1721 startSpecifier, specifierLen, 1722 CS.getStart(), CS.getLength()); 1723 } 1724 1725 bool CheckScanfHandler::HandleScanfSpecifier( 1726 const analyze_scanf::ScanfSpecifier &FS, 1727 const char *startSpecifier, 1728 unsigned specifierLen) { 1729 1730 using namespace analyze_scanf; 1731 using namespace analyze_format_string; 1732 1733 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 1734 1735 // Handle case where '%' and '*' don't consume an argument. These shouldn't 1736 // be used to decide if we are using positional arguments consistently. 1737 if (FS.consumesDataArgument()) { 1738 if (atFirstArg) { 1739 atFirstArg = false; 1740 usesPositionalArgs = FS.usesPositionalArg(); 1741 } 1742 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1743 // Cannot mix-and-match positional and non-positional arguments. 1744 S.Diag(getLocationOfByte(CS.getStart()), 1745 diag::warn_format_mix_positional_nonpositional_args) 1746 << getSpecifierRange(startSpecifier, specifierLen); 1747 return false; 1748 } 1749 } 1750 1751 // Check if the field with is non-zero. 1752 const OptionalAmount &Amt = FS.getFieldWidth(); 1753 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 1754 if (Amt.getConstantAmount() == 0) { 1755 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 1756 Amt.getConstantLength()); 1757 S.Diag(getLocationOfByte(Amt.getStart()), 1758 diag::warn_scanf_nonzero_width) 1759 << R << FixItHint::CreateRemoval(R); 1760 } 1761 } 1762 1763 if (!FS.consumesDataArgument()) { 1764 // FIXME: Technically specifying a precision or field width here 1765 // makes no sense. Worth issuing a warning at some point. 1766 return true; 1767 } 1768 1769 // Consume the argument. 1770 unsigned argIndex = FS.getArgIndex(); 1771 if (argIndex < NumDataArgs) { 1772 // The check to see if the argIndex is valid will come later. 1773 // We set the bit here because we may exit early from this 1774 // function if we encounter some other error. 1775 CoveredArgs.set(argIndex); 1776 } 1777 1778 // Check the length modifier is valid with the given conversion specifier. 1779 const LengthModifier &LM = FS.getLengthModifier(); 1780 if (!FS.hasValidLengthModifier()) { 1781 S.Diag(getLocationOfByte(LM.getStart()), 1782 diag::warn_format_nonsensical_length) 1783 << LM.toString() << CS.toString() 1784 << getSpecifierRange(startSpecifier, specifierLen) 1785 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 1786 LM.getLength())); 1787 } 1788 1789 // The remaining checks depend on the data arguments. 1790 if (HasVAListArg) 1791 return true; 1792 1793 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 1794 return false; 1795 1796 // FIXME: Check that the argument type matches the format specifier. 1797 1798 return true; 1799 } 1800 1801 void Sema::CheckFormatString(const StringLiteral *FExpr, 1802 const Expr *OrigFormatExpr, 1803 const CallExpr *TheCall, bool HasVAListArg, 1804 unsigned format_idx, unsigned firstDataArg, 1805 bool isPrintf) { 1806 1807 // CHECK: is the format string a wide literal? 1808 if (FExpr->isWide()) { 1809 Diag(FExpr->getLocStart(), 1810 diag::warn_format_string_is_wide_literal) 1811 << OrigFormatExpr->getSourceRange(); 1812 return; 1813 } 1814 1815 // Str - The format string. NOTE: this is NOT null-terminated! 1816 llvm::StringRef StrRef = FExpr->getString(); 1817 const char *Str = StrRef.data(); 1818 unsigned StrLen = StrRef.size(); 1819 1820 // CHECK: empty format string? 1821 if (StrLen == 0) { 1822 Diag(FExpr->getLocStart(), diag::warn_empty_format_string) 1823 << OrigFormatExpr->getSourceRange(); 1824 return; 1825 } 1826 1827 if (isPrintf) { 1828 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1829 TheCall->getNumArgs() - firstDataArg, 1830 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1831 HasVAListArg, TheCall, format_idx); 1832 1833 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen)) 1834 H.DoneProcessing(); 1835 } 1836 else { 1837 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1838 TheCall->getNumArgs() - firstDataArg, 1839 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1840 HasVAListArg, TheCall, format_idx); 1841 1842 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen)) 1843 H.DoneProcessing(); 1844 } 1845 } 1846 1847 //===--- CHECK: Standard memory functions ---------------------------------===// 1848 1849 /// \brief Determine whether the given type is a dynamic class type (e.g., 1850 /// whether it has a vtable). 1851 static bool isDynamicClassType(QualType T) { 1852 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 1853 if (CXXRecordDecl *Definition = Record->getDefinition()) 1854 if (Definition->isDynamicClass()) 1855 return true; 1856 1857 return false; 1858 } 1859 1860 /// \brief If E is a sizeof expression, returns its argument expression, 1861 /// otherwise returns NULL. 1862 static const Expr *getSizeOfExprArg(const Expr* E) { 1863 if (const UnaryExprOrTypeTraitExpr *SizeOf = 1864 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 1865 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 1866 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 1867 1868 return 0; 1869 } 1870 1871 /// \brief If E is a sizeof expression, returns its argument type. 1872 static QualType getSizeOfArgType(const Expr* E) { 1873 if (const UnaryExprOrTypeTraitExpr *SizeOf = 1874 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 1875 if (SizeOf->getKind() == clang::UETT_SizeOf) 1876 return SizeOf->getTypeOfArgument(); 1877 1878 return QualType(); 1879 } 1880 1881 /// \brief Check for dangerous or invalid arguments to memset(). 1882 /// 1883 /// This issues warnings on known problematic, dangerous or unspecified 1884 /// arguments to the standard 'memset', 'memcpy', and 'memmove' function calls. 1885 /// 1886 /// \param Call The call expression to diagnose. 1887 void Sema::CheckMemsetcpymoveArguments(const CallExpr *Call, 1888 CheckedMemoryFunction CMF, 1889 IdentifierInfo *FnName) { 1890 // It is possible to have a non-standard definition of memset. Validate 1891 // we have enough arguments, and if not, abort further checking. 1892 if (Call->getNumArgs() < 3) 1893 return; 1894 1895 unsigned LastArg = (CMF == CMF_Memset? 1 : 2); 1896 const Expr *LenExpr = Call->getArg(2)->IgnoreParenImpCasts(); 1897 1898 // We have special checking when the length is a sizeof expression. 1899 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 1900 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 1901 llvm::FoldingSetNodeID SizeOfArgID; 1902 1903 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 1904 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 1905 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 1906 1907 QualType DestTy = Dest->getType(); 1908 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 1909 QualType PointeeTy = DestPtrTy->getPointeeType(); 1910 1911 // Never warn about void type pointers. This can be used to suppress 1912 // false positives. 1913 if (PointeeTy->isVoidType()) 1914 continue; 1915 1916 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 1917 // actually comparing the expressions for equality. Because computing the 1918 // expression IDs can be expensive, we only do this if the diagnostic is 1919 // enabled. 1920 if (SizeOfArg && 1921 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 1922 SizeOfArg->getExprLoc())) { 1923 // We only compute IDs for expressions if the warning is enabled, and 1924 // cache the sizeof arg's ID. 1925 if (SizeOfArgID == llvm::FoldingSetNodeID()) 1926 SizeOfArg->Profile(SizeOfArgID, Context, true); 1927 llvm::FoldingSetNodeID DestID; 1928 Dest->Profile(DestID, Context, true); 1929 if (DestID == SizeOfArgID) { 1930 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 1931 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 1932 if (UnaryOp->getOpcode() == UO_AddrOf) 1933 ActionIdx = 1; // If its an address-of operator, just remove it. 1934 if (Context.getTypeSize(PointeeTy) == Context.getCharWidth()) 1935 ActionIdx = 2; // If the pointee's size is sizeof(char), 1936 // suggest an explicit length. 1937 DiagRuntimeBehavior(SizeOfArg->getExprLoc(), Dest, 1938 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 1939 << FnName << ArgIdx << ActionIdx 1940 << Dest->getSourceRange() 1941 << SizeOfArg->getSourceRange()); 1942 break; 1943 } 1944 } 1945 1946 // Also check for cases where the sizeof argument is the exact same 1947 // type as the memory argument, and where it points to a user-defined 1948 // record type. 1949 if (SizeOfArgTy != QualType()) { 1950 if (PointeeTy->isRecordType() && 1951 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 1952 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 1953 PDiag(diag::warn_sizeof_pointer_type_memaccess) 1954 << FnName << SizeOfArgTy << ArgIdx 1955 << PointeeTy << Dest->getSourceRange() 1956 << LenExpr->getSourceRange()); 1957 break; 1958 } 1959 } 1960 1961 unsigned DiagID; 1962 1963 // Always complain about dynamic classes. 1964 if (isDynamicClassType(PointeeTy)) 1965 DiagID = diag::warn_dyn_class_memaccess; 1966 else if (PointeeTy.hasNonTrivialObjCLifetime() && CMF != CMF_Memset) 1967 DiagID = diag::warn_arc_object_memaccess; 1968 else 1969 continue; 1970 1971 DiagRuntimeBehavior( 1972 Dest->getExprLoc(), Dest, 1973 PDiag(DiagID) 1974 << ArgIdx << FnName << PointeeTy 1975 << Call->getCallee()->getSourceRange()); 1976 1977 DiagRuntimeBehavior( 1978 Dest->getExprLoc(), Dest, 1979 PDiag(diag::note_bad_memaccess_silence) 1980 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 1981 break; 1982 } 1983 } 1984 } 1985 1986 //===--- CHECK: Return Address of Stack Variable --------------------------===// 1987 1988 static Expr *EvalVal(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars); 1989 static Expr *EvalAddr(Expr* E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars); 1990 1991 /// CheckReturnStackAddr - Check if a return statement returns the address 1992 /// of a stack variable. 1993 void 1994 Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 1995 SourceLocation ReturnLoc) { 1996 1997 Expr *stackE = 0; 1998 llvm::SmallVector<DeclRefExpr *, 8> refVars; 1999 2000 // Perform checking for returned stack addresses, local blocks, 2001 // label addresses or references to temporaries. 2002 if (lhsType->isPointerType() || 2003 (!getLangOptions().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 2004 stackE = EvalAddr(RetValExp, refVars); 2005 } else if (lhsType->isReferenceType()) { 2006 stackE = EvalVal(RetValExp, refVars); 2007 } 2008 2009 if (stackE == 0) 2010 return; // Nothing suspicious was found. 2011 2012 SourceLocation diagLoc; 2013 SourceRange diagRange; 2014 if (refVars.empty()) { 2015 diagLoc = stackE->getLocStart(); 2016 diagRange = stackE->getSourceRange(); 2017 } else { 2018 // We followed through a reference variable. 'stackE' contains the 2019 // problematic expression but we will warn at the return statement pointing 2020 // at the reference variable. We will later display the "trail" of 2021 // reference variables using notes. 2022 diagLoc = refVars[0]->getLocStart(); 2023 diagRange = refVars[0]->getSourceRange(); 2024 } 2025 2026 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 2027 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 2028 : diag::warn_ret_stack_addr) 2029 << DR->getDecl()->getDeclName() << diagRange; 2030 } else if (isa<BlockExpr>(stackE)) { // local block. 2031 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 2032 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 2033 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 2034 } else { // local temporary. 2035 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 2036 : diag::warn_ret_local_temp_addr) 2037 << diagRange; 2038 } 2039 2040 // Display the "trail" of reference variables that we followed until we 2041 // found the problematic expression using notes. 2042 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 2043 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 2044 // If this var binds to another reference var, show the range of the next 2045 // var, otherwise the var binds to the problematic expression, in which case 2046 // show the range of the expression. 2047 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 2048 : stackE->getSourceRange(); 2049 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 2050 << VD->getDeclName() << range; 2051 } 2052 } 2053 2054 /// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 2055 /// check if the expression in a return statement evaluates to an address 2056 /// to a location on the stack, a local block, an address of a label, or a 2057 /// reference to local temporary. The recursion is used to traverse the 2058 /// AST of the return expression, with recursion backtracking when we 2059 /// encounter a subexpression that (1) clearly does not lead to one of the 2060 /// above problematic expressions (2) is something we cannot determine leads to 2061 /// a problematic expression based on such local checking. 2062 /// 2063 /// Both EvalAddr and EvalVal follow through reference variables to evaluate 2064 /// the expression that they point to. Such variables are added to the 2065 /// 'refVars' vector so that we know what the reference variable "trail" was. 2066 /// 2067 /// EvalAddr processes expressions that are pointers that are used as 2068 /// references (and not L-values). EvalVal handles all other values. 2069 /// At the base case of the recursion is a check for the above problematic 2070 /// expressions. 2071 /// 2072 /// This implementation handles: 2073 /// 2074 /// * pointer-to-pointer casts 2075 /// * implicit conversions from array references to pointers 2076 /// * taking the address of fields 2077 /// * arbitrary interplay between "&" and "*" operators 2078 /// * pointer arithmetic from an address of a stack variable 2079 /// * taking the address of an array element where the array is on the stack 2080 static Expr *EvalAddr(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars) { 2081 if (E->isTypeDependent()) 2082 return NULL; 2083 2084 // We should only be called for evaluating pointer expressions. 2085 assert((E->getType()->isAnyPointerType() || 2086 E->getType()->isBlockPointerType() || 2087 E->getType()->isObjCQualifiedIdType()) && 2088 "EvalAddr only works on pointers"); 2089 2090 E = E->IgnoreParens(); 2091 2092 // Our "symbolic interpreter" is just a dispatch off the currently 2093 // viewed AST node. We then recursively traverse the AST by calling 2094 // EvalAddr and EvalVal appropriately. 2095 switch (E->getStmtClass()) { 2096 case Stmt::DeclRefExprClass: { 2097 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2098 2099 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2100 // If this is a reference variable, follow through to the expression that 2101 // it points to. 2102 if (V->hasLocalStorage() && 2103 V->getType()->isReferenceType() && V->hasInit()) { 2104 // Add the reference variable to the "trail". 2105 refVars.push_back(DR); 2106 return EvalAddr(V->getInit(), refVars); 2107 } 2108 2109 return NULL; 2110 } 2111 2112 case Stmt::UnaryOperatorClass: { 2113 // The only unary operator that make sense to handle here 2114 // is AddrOf. All others don't make sense as pointers. 2115 UnaryOperator *U = cast<UnaryOperator>(E); 2116 2117 if (U->getOpcode() == UO_AddrOf) 2118 return EvalVal(U->getSubExpr(), refVars); 2119 else 2120 return NULL; 2121 } 2122 2123 case Stmt::BinaryOperatorClass: { 2124 // Handle pointer arithmetic. All other binary operators are not valid 2125 // in this context. 2126 BinaryOperator *B = cast<BinaryOperator>(E); 2127 BinaryOperatorKind op = B->getOpcode(); 2128 2129 if (op != BO_Add && op != BO_Sub) 2130 return NULL; 2131 2132 Expr *Base = B->getLHS(); 2133 2134 // Determine which argument is the real pointer base. It could be 2135 // the RHS argument instead of the LHS. 2136 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 2137 2138 assert (Base->getType()->isPointerType()); 2139 return EvalAddr(Base, refVars); 2140 } 2141 2142 // For conditional operators we need to see if either the LHS or RHS are 2143 // valid DeclRefExpr*s. If one of them is valid, we return it. 2144 case Stmt::ConditionalOperatorClass: { 2145 ConditionalOperator *C = cast<ConditionalOperator>(E); 2146 2147 // Handle the GNU extension for missing LHS. 2148 if (Expr *lhsExpr = C->getLHS()) { 2149 // In C++, we can have a throw-expression, which has 'void' type. 2150 if (!lhsExpr->getType()->isVoidType()) 2151 if (Expr* LHS = EvalAddr(lhsExpr, refVars)) 2152 return LHS; 2153 } 2154 2155 // In C++, we can have a throw-expression, which has 'void' type. 2156 if (C->getRHS()->getType()->isVoidType()) 2157 return NULL; 2158 2159 return EvalAddr(C->getRHS(), refVars); 2160 } 2161 2162 case Stmt::BlockExprClass: 2163 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 2164 return E; // local block. 2165 return NULL; 2166 2167 case Stmt::AddrLabelExprClass: 2168 return E; // address of label. 2169 2170 // For casts, we need to handle conversions from arrays to 2171 // pointer values, and pointer-to-pointer conversions. 2172 case Stmt::ImplicitCastExprClass: 2173 case Stmt::CStyleCastExprClass: 2174 case Stmt::CXXFunctionalCastExprClass: 2175 case Stmt::ObjCBridgedCastExprClass: { 2176 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 2177 QualType T = SubExpr->getType(); 2178 2179 if (SubExpr->getType()->isPointerType() || 2180 SubExpr->getType()->isBlockPointerType() || 2181 SubExpr->getType()->isObjCQualifiedIdType()) 2182 return EvalAddr(SubExpr, refVars); 2183 else if (T->isArrayType()) 2184 return EvalVal(SubExpr, refVars); 2185 else 2186 return 0; 2187 } 2188 2189 // C++ casts. For dynamic casts, static casts, and const casts, we 2190 // are always converting from a pointer-to-pointer, so we just blow 2191 // through the cast. In the case the dynamic cast doesn't fail (and 2192 // return NULL), we take the conservative route and report cases 2193 // where we return the address of a stack variable. For Reinterpre 2194 // FIXME: The comment about is wrong; we're not always converting 2195 // from pointer to pointer. I'm guessing that this code should also 2196 // handle references to objects. 2197 case Stmt::CXXStaticCastExprClass: 2198 case Stmt::CXXDynamicCastExprClass: 2199 case Stmt::CXXConstCastExprClass: 2200 case Stmt::CXXReinterpretCastExprClass: { 2201 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 2202 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 2203 return EvalAddr(S, refVars); 2204 else 2205 return NULL; 2206 } 2207 2208 case Stmt::MaterializeTemporaryExprClass: 2209 if (Expr *Result = EvalAddr( 2210 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 2211 refVars)) 2212 return Result; 2213 2214 return E; 2215 2216 // Everything else: we simply don't reason about them. 2217 default: 2218 return NULL; 2219 } 2220 } 2221 2222 2223 /// EvalVal - This function is complements EvalAddr in the mutual recursion. 2224 /// See the comments for EvalAddr for more details. 2225 static Expr *EvalVal(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars) { 2226 do { 2227 // We should only be called for evaluating non-pointer expressions, or 2228 // expressions with a pointer type that are not used as references but instead 2229 // are l-values (e.g., DeclRefExpr with a pointer type). 2230 2231 // Our "symbolic interpreter" is just a dispatch off the currently 2232 // viewed AST node. We then recursively traverse the AST by calling 2233 // EvalAddr and EvalVal appropriately. 2234 2235 E = E->IgnoreParens(); 2236 switch (E->getStmtClass()) { 2237 case Stmt::ImplicitCastExprClass: { 2238 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 2239 if (IE->getValueKind() == VK_LValue) { 2240 E = IE->getSubExpr(); 2241 continue; 2242 } 2243 return NULL; 2244 } 2245 2246 case Stmt::DeclRefExprClass: { 2247 // When we hit a DeclRefExpr we are looking at code that refers to a 2248 // variable's name. If it's not a reference variable we check if it has 2249 // local storage within the function, and if so, return the expression. 2250 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2251 2252 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2253 if (V->hasLocalStorage()) { 2254 if (!V->getType()->isReferenceType()) 2255 return DR; 2256 2257 // Reference variable, follow through to the expression that 2258 // it points to. 2259 if (V->hasInit()) { 2260 // Add the reference variable to the "trail". 2261 refVars.push_back(DR); 2262 return EvalVal(V->getInit(), refVars); 2263 } 2264 } 2265 2266 return NULL; 2267 } 2268 2269 case Stmt::UnaryOperatorClass: { 2270 // The only unary operator that make sense to handle here 2271 // is Deref. All others don't resolve to a "name." This includes 2272 // handling all sorts of rvalues passed to a unary operator. 2273 UnaryOperator *U = cast<UnaryOperator>(E); 2274 2275 if (U->getOpcode() == UO_Deref) 2276 return EvalAddr(U->getSubExpr(), refVars); 2277 2278 return NULL; 2279 } 2280 2281 case Stmt::ArraySubscriptExprClass: { 2282 // Array subscripts are potential references to data on the stack. We 2283 // retrieve the DeclRefExpr* for the array variable if it indeed 2284 // has local storage. 2285 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars); 2286 } 2287 2288 case Stmt::ConditionalOperatorClass: { 2289 // For conditional operators we need to see if either the LHS or RHS are 2290 // non-NULL Expr's. If one is non-NULL, we return it. 2291 ConditionalOperator *C = cast<ConditionalOperator>(E); 2292 2293 // Handle the GNU extension for missing LHS. 2294 if (Expr *lhsExpr = C->getLHS()) 2295 if (Expr *LHS = EvalVal(lhsExpr, refVars)) 2296 return LHS; 2297 2298 return EvalVal(C->getRHS(), refVars); 2299 } 2300 2301 // Accesses to members are potential references to data on the stack. 2302 case Stmt::MemberExprClass: { 2303 MemberExpr *M = cast<MemberExpr>(E); 2304 2305 // Check for indirect access. We only want direct field accesses. 2306 if (M->isArrow()) 2307 return NULL; 2308 2309 // Check whether the member type is itself a reference, in which case 2310 // we're not going to refer to the member, but to what the member refers to. 2311 if (M->getMemberDecl()->getType()->isReferenceType()) 2312 return NULL; 2313 2314 return EvalVal(M->getBase(), refVars); 2315 } 2316 2317 case Stmt::MaterializeTemporaryExprClass: 2318 if (Expr *Result = EvalVal( 2319 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 2320 refVars)) 2321 return Result; 2322 2323 return E; 2324 2325 default: 2326 // Check that we don't return or take the address of a reference to a 2327 // temporary. This is only useful in C++. 2328 if (!E->isTypeDependent() && E->isRValue()) 2329 return E; 2330 2331 // Everything else: we simply don't reason about them. 2332 return NULL; 2333 } 2334 } while (true); 2335 } 2336 2337 //===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 2338 2339 /// Check for comparisons of floating point operands using != and ==. 2340 /// Issue a warning if these are no self-comparisons, as they are not likely 2341 /// to do what the programmer intended. 2342 void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 2343 bool EmitWarning = true; 2344 2345 Expr* LeftExprSansParen = lex->IgnoreParenImpCasts(); 2346 Expr* RightExprSansParen = rex->IgnoreParenImpCasts(); 2347 2348 // Special case: check for x == x (which is OK). 2349 // Do not emit warnings for such cases. 2350 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 2351 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 2352 if (DRL->getDecl() == DRR->getDecl()) 2353 EmitWarning = false; 2354 2355 2356 // Special case: check for comparisons against literals that can be exactly 2357 // represented by APFloat. In such cases, do not emit a warning. This 2358 // is a heuristic: often comparison against such literals are used to 2359 // detect if a value in a variable has not changed. This clearly can 2360 // lead to false negatives. 2361 if (EmitWarning) { 2362 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 2363 if (FLL->isExact()) 2364 EmitWarning = false; 2365 } else 2366 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 2367 if (FLR->isExact()) 2368 EmitWarning = false; 2369 } 2370 } 2371 2372 // Check for comparisons with builtin types. 2373 if (EmitWarning) 2374 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 2375 if (CL->isBuiltinCall(Context)) 2376 EmitWarning = false; 2377 2378 if (EmitWarning) 2379 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 2380 if (CR->isBuiltinCall(Context)) 2381 EmitWarning = false; 2382 2383 // Emit the diagnostic. 2384 if (EmitWarning) 2385 Diag(loc, diag::warn_floatingpoint_eq) 2386 << lex->getSourceRange() << rex->getSourceRange(); 2387 } 2388 2389 //===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 2390 //===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 2391 2392 namespace { 2393 2394 /// Structure recording the 'active' range of an integer-valued 2395 /// expression. 2396 struct IntRange { 2397 /// The number of bits active in the int. 2398 unsigned Width; 2399 2400 /// True if the int is known not to have negative values. 2401 bool NonNegative; 2402 2403 IntRange(unsigned Width, bool NonNegative) 2404 : Width(Width), NonNegative(NonNegative) 2405 {} 2406 2407 /// Returns the range of the bool type. 2408 static IntRange forBoolType() { 2409 return IntRange(1, true); 2410 } 2411 2412 /// Returns the range of an opaque value of the given integral type. 2413 static IntRange forValueOfType(ASTContext &C, QualType T) { 2414 return forValueOfCanonicalType(C, 2415 T->getCanonicalTypeInternal().getTypePtr()); 2416 } 2417 2418 /// Returns the range of an opaque value of a canonical integral type. 2419 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 2420 assert(T->isCanonicalUnqualified()); 2421 2422 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2423 T = VT->getElementType().getTypePtr(); 2424 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2425 T = CT->getElementType().getTypePtr(); 2426 2427 // For enum types, use the known bit width of the enumerators. 2428 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 2429 EnumDecl *Enum = ET->getDecl(); 2430 if (!Enum->isDefinition()) 2431 return IntRange(C.getIntWidth(QualType(T, 0)), false); 2432 2433 unsigned NumPositive = Enum->getNumPositiveBits(); 2434 unsigned NumNegative = Enum->getNumNegativeBits(); 2435 2436 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 2437 } 2438 2439 const BuiltinType *BT = cast<BuiltinType>(T); 2440 assert(BT->isInteger()); 2441 2442 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2443 } 2444 2445 /// Returns the "target" range of a canonical integral type, i.e. 2446 /// the range of values expressible in the type. 2447 /// 2448 /// This matches forValueOfCanonicalType except that enums have the 2449 /// full range of their type, not the range of their enumerators. 2450 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 2451 assert(T->isCanonicalUnqualified()); 2452 2453 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2454 T = VT->getElementType().getTypePtr(); 2455 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2456 T = CT->getElementType().getTypePtr(); 2457 if (const EnumType *ET = dyn_cast<EnumType>(T)) 2458 T = ET->getDecl()->getIntegerType().getTypePtr(); 2459 2460 const BuiltinType *BT = cast<BuiltinType>(T); 2461 assert(BT->isInteger()); 2462 2463 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2464 } 2465 2466 /// Returns the supremum of two ranges: i.e. their conservative merge. 2467 static IntRange join(IntRange L, IntRange R) { 2468 return IntRange(std::max(L.Width, R.Width), 2469 L.NonNegative && R.NonNegative); 2470 } 2471 2472 /// Returns the infinum of two ranges: i.e. their aggressive merge. 2473 static IntRange meet(IntRange L, IntRange R) { 2474 return IntRange(std::min(L.Width, R.Width), 2475 L.NonNegative || R.NonNegative); 2476 } 2477 }; 2478 2479 IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 2480 if (value.isSigned() && value.isNegative()) 2481 return IntRange(value.getMinSignedBits(), false); 2482 2483 if (value.getBitWidth() > MaxWidth) 2484 value = value.trunc(MaxWidth); 2485 2486 // isNonNegative() just checks the sign bit without considering 2487 // signedness. 2488 return IntRange(value.getActiveBits(), true); 2489 } 2490 2491 IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 2492 unsigned MaxWidth) { 2493 if (result.isInt()) 2494 return GetValueRange(C, result.getInt(), MaxWidth); 2495 2496 if (result.isVector()) { 2497 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 2498 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 2499 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 2500 R = IntRange::join(R, El); 2501 } 2502 return R; 2503 } 2504 2505 if (result.isComplexInt()) { 2506 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 2507 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 2508 return IntRange::join(R, I); 2509 } 2510 2511 // This can happen with lossless casts to intptr_t of "based" lvalues. 2512 // Assume it might use arbitrary bits. 2513 // FIXME: The only reason we need to pass the type in here is to get 2514 // the sign right on this one case. It would be nice if APValue 2515 // preserved this. 2516 assert(result.isLValue()); 2517 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 2518 } 2519 2520 /// Pseudo-evaluate the given integer expression, estimating the 2521 /// range of values it might take. 2522 /// 2523 /// \param MaxWidth - the width to which the value will be truncated 2524 IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 2525 E = E->IgnoreParens(); 2526 2527 // Try a full evaluation first. 2528 Expr::EvalResult result; 2529 if (E->Evaluate(result, C)) 2530 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 2531 2532 // I think we only want to look through implicit casts here; if the 2533 // user has an explicit widening cast, we should treat the value as 2534 // being of the new, wider type. 2535 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 2536 if (CE->getCastKind() == CK_NoOp) 2537 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 2538 2539 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 2540 2541 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 2542 2543 // Assume that non-integer casts can span the full range of the type. 2544 if (!isIntegerCast) 2545 return OutputTypeRange; 2546 2547 IntRange SubRange 2548 = GetExprRange(C, CE->getSubExpr(), 2549 std::min(MaxWidth, OutputTypeRange.Width)); 2550 2551 // Bail out if the subexpr's range is as wide as the cast type. 2552 if (SubRange.Width >= OutputTypeRange.Width) 2553 return OutputTypeRange; 2554 2555 // Otherwise, we take the smaller width, and we're non-negative if 2556 // either the output type or the subexpr is. 2557 return IntRange(SubRange.Width, 2558 SubRange.NonNegative || OutputTypeRange.NonNegative); 2559 } 2560 2561 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 2562 // If we can fold the condition, just take that operand. 2563 bool CondResult; 2564 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 2565 return GetExprRange(C, CondResult ? CO->getTrueExpr() 2566 : CO->getFalseExpr(), 2567 MaxWidth); 2568 2569 // Otherwise, conservatively merge. 2570 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 2571 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 2572 return IntRange::join(L, R); 2573 } 2574 2575 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 2576 switch (BO->getOpcode()) { 2577 2578 // Boolean-valued operations are single-bit and positive. 2579 case BO_LAnd: 2580 case BO_LOr: 2581 case BO_LT: 2582 case BO_GT: 2583 case BO_LE: 2584 case BO_GE: 2585 case BO_EQ: 2586 case BO_NE: 2587 return IntRange::forBoolType(); 2588 2589 // The type of the assignments is the type of the LHS, so the RHS 2590 // is not necessarily the same type. 2591 case BO_MulAssign: 2592 case BO_DivAssign: 2593 case BO_RemAssign: 2594 case BO_AddAssign: 2595 case BO_SubAssign: 2596 case BO_XorAssign: 2597 case BO_OrAssign: 2598 // TODO: bitfields? 2599 return IntRange::forValueOfType(C, E->getType()); 2600 2601 // Simple assignments just pass through the RHS, which will have 2602 // been coerced to the LHS type. 2603 case BO_Assign: 2604 // TODO: bitfields? 2605 return GetExprRange(C, BO->getRHS(), MaxWidth); 2606 2607 // Operations with opaque sources are black-listed. 2608 case BO_PtrMemD: 2609 case BO_PtrMemI: 2610 return IntRange::forValueOfType(C, E->getType()); 2611 2612 // Bitwise-and uses the *infinum* of the two source ranges. 2613 case BO_And: 2614 case BO_AndAssign: 2615 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 2616 GetExprRange(C, BO->getRHS(), MaxWidth)); 2617 2618 // Left shift gets black-listed based on a judgement call. 2619 case BO_Shl: 2620 // ...except that we want to treat '1 << (blah)' as logically 2621 // positive. It's an important idiom. 2622 if (IntegerLiteral *I 2623 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 2624 if (I->getValue() == 1) { 2625 IntRange R = IntRange::forValueOfType(C, E->getType()); 2626 return IntRange(R.Width, /*NonNegative*/ true); 2627 } 2628 } 2629 // fallthrough 2630 2631 case BO_ShlAssign: 2632 return IntRange::forValueOfType(C, E->getType()); 2633 2634 // Right shift by a constant can narrow its left argument. 2635 case BO_Shr: 2636 case BO_ShrAssign: { 2637 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2638 2639 // If the shift amount is a positive constant, drop the width by 2640 // that much. 2641 llvm::APSInt shift; 2642 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 2643 shift.isNonNegative()) { 2644 unsigned zext = shift.getZExtValue(); 2645 if (zext >= L.Width) 2646 L.Width = (L.NonNegative ? 0 : 1); 2647 else 2648 L.Width -= zext; 2649 } 2650 2651 return L; 2652 } 2653 2654 // Comma acts as its right operand. 2655 case BO_Comma: 2656 return GetExprRange(C, BO->getRHS(), MaxWidth); 2657 2658 // Black-list pointer subtractions. 2659 case BO_Sub: 2660 if (BO->getLHS()->getType()->isPointerType()) 2661 return IntRange::forValueOfType(C, E->getType()); 2662 break; 2663 2664 // The width of a division result is mostly determined by the size 2665 // of the LHS. 2666 case BO_Div: { 2667 // Don't 'pre-truncate' the operands. 2668 unsigned opWidth = C.getIntWidth(E->getType()); 2669 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 2670 2671 // If the divisor is constant, use that. 2672 llvm::APSInt divisor; 2673 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 2674 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 2675 if (log2 >= L.Width) 2676 L.Width = (L.NonNegative ? 0 : 1); 2677 else 2678 L.Width = std::min(L.Width - log2, MaxWidth); 2679 return L; 2680 } 2681 2682 // Otherwise, just use the LHS's width. 2683 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 2684 return IntRange(L.Width, L.NonNegative && R.NonNegative); 2685 } 2686 2687 // The result of a remainder can't be larger than the result of 2688 // either side. 2689 case BO_Rem: { 2690 // Don't 'pre-truncate' the operands. 2691 unsigned opWidth = C.getIntWidth(E->getType()); 2692 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 2693 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 2694 2695 IntRange meet = IntRange::meet(L, R); 2696 meet.Width = std::min(meet.Width, MaxWidth); 2697 return meet; 2698 } 2699 2700 // The default behavior is okay for these. 2701 case BO_Mul: 2702 case BO_Add: 2703 case BO_Xor: 2704 case BO_Or: 2705 break; 2706 } 2707 2708 // The default case is to treat the operation as if it were closed 2709 // on the narrowest type that encompasses both operands. 2710 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2711 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 2712 return IntRange::join(L, R); 2713 } 2714 2715 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 2716 switch (UO->getOpcode()) { 2717 // Boolean-valued operations are white-listed. 2718 case UO_LNot: 2719 return IntRange::forBoolType(); 2720 2721 // Operations with opaque sources are black-listed. 2722 case UO_Deref: 2723 case UO_AddrOf: // should be impossible 2724 return IntRange::forValueOfType(C, E->getType()); 2725 2726 default: 2727 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 2728 } 2729 } 2730 2731 if (dyn_cast<OffsetOfExpr>(E)) { 2732 IntRange::forValueOfType(C, E->getType()); 2733 } 2734 2735 FieldDecl *BitField = E->getBitField(); 2736 if (BitField) { 2737 llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C); 2738 unsigned BitWidth = BitWidthAP.getZExtValue(); 2739 2740 return IntRange(BitWidth, 2741 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 2742 } 2743 2744 return IntRange::forValueOfType(C, E->getType()); 2745 } 2746 2747 IntRange GetExprRange(ASTContext &C, Expr *E) { 2748 return GetExprRange(C, E, C.getIntWidth(E->getType())); 2749 } 2750 2751 /// Checks whether the given value, which currently has the given 2752 /// source semantics, has the same value when coerced through the 2753 /// target semantics. 2754 bool IsSameFloatAfterCast(const llvm::APFloat &value, 2755 const llvm::fltSemantics &Src, 2756 const llvm::fltSemantics &Tgt) { 2757 llvm::APFloat truncated = value; 2758 2759 bool ignored; 2760 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 2761 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 2762 2763 return truncated.bitwiseIsEqual(value); 2764 } 2765 2766 /// Checks whether the given value, which currently has the given 2767 /// source semantics, has the same value when coerced through the 2768 /// target semantics. 2769 /// 2770 /// The value might be a vector of floats (or a complex number). 2771 bool IsSameFloatAfterCast(const APValue &value, 2772 const llvm::fltSemantics &Src, 2773 const llvm::fltSemantics &Tgt) { 2774 if (value.isFloat()) 2775 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 2776 2777 if (value.isVector()) { 2778 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 2779 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 2780 return false; 2781 return true; 2782 } 2783 2784 assert(value.isComplexFloat()); 2785 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 2786 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 2787 } 2788 2789 void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 2790 2791 static bool IsZero(Sema &S, Expr *E) { 2792 // Suppress cases where we are comparing against an enum constant. 2793 if (const DeclRefExpr *DR = 2794 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 2795 if (isa<EnumConstantDecl>(DR->getDecl())) 2796 return false; 2797 2798 // Suppress cases where the '0' value is expanded from a macro. 2799 if (E->getLocStart().isMacroID()) 2800 return false; 2801 2802 llvm::APSInt Value; 2803 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 2804 } 2805 2806 static bool HasEnumType(Expr *E) { 2807 // Strip off implicit integral promotions. 2808 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 2809 if (ICE->getCastKind() != CK_IntegralCast && 2810 ICE->getCastKind() != CK_NoOp) 2811 break; 2812 E = ICE->getSubExpr(); 2813 } 2814 2815 return E->getType()->isEnumeralType(); 2816 } 2817 2818 void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 2819 BinaryOperatorKind op = E->getOpcode(); 2820 if (E->isValueDependent()) 2821 return; 2822 2823 if (op == BO_LT && IsZero(S, E->getRHS())) { 2824 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2825 << "< 0" << "false" << HasEnumType(E->getLHS()) 2826 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2827 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 2828 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2829 << ">= 0" << "true" << HasEnumType(E->getLHS()) 2830 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2831 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 2832 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2833 << "0 >" << "false" << HasEnumType(E->getRHS()) 2834 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2835 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 2836 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2837 << "0 <=" << "true" << HasEnumType(E->getRHS()) 2838 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2839 } 2840 } 2841 2842 /// Analyze the operands of the given comparison. Implements the 2843 /// fallback case from AnalyzeComparison. 2844 void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 2845 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 2846 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 2847 } 2848 2849 /// \brief Implements -Wsign-compare. 2850 /// 2851 /// \param lex the left-hand expression 2852 /// \param rex the right-hand expression 2853 /// \param OpLoc the location of the joining operator 2854 /// \param BinOpc binary opcode or 0 2855 void AnalyzeComparison(Sema &S, BinaryOperator *E) { 2856 // The type the comparison is being performed in. 2857 QualType T = E->getLHS()->getType(); 2858 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 2859 && "comparison with mismatched types"); 2860 2861 // We don't do anything special if this isn't an unsigned integral 2862 // comparison: we're only interested in integral comparisons, and 2863 // signed comparisons only happen in cases we don't care to warn about. 2864 // 2865 // We also don't care about value-dependent expressions or expressions 2866 // whose result is a constant. 2867 if (!T->hasUnsignedIntegerRepresentation() 2868 || E->isValueDependent() || E->isIntegerConstantExpr(S.Context)) 2869 return AnalyzeImpConvsInComparison(S, E); 2870 2871 Expr *lex = E->getLHS()->IgnoreParenImpCasts(); 2872 Expr *rex = E->getRHS()->IgnoreParenImpCasts(); 2873 2874 // Check to see if one of the (unmodified) operands is of different 2875 // signedness. 2876 Expr *signedOperand, *unsignedOperand; 2877 if (lex->getType()->hasSignedIntegerRepresentation()) { 2878 assert(!rex->getType()->hasSignedIntegerRepresentation() && 2879 "unsigned comparison between two signed integer expressions?"); 2880 signedOperand = lex; 2881 unsignedOperand = rex; 2882 } else if (rex->getType()->hasSignedIntegerRepresentation()) { 2883 signedOperand = rex; 2884 unsignedOperand = lex; 2885 } else { 2886 CheckTrivialUnsignedComparison(S, E); 2887 return AnalyzeImpConvsInComparison(S, E); 2888 } 2889 2890 // Otherwise, calculate the effective range of the signed operand. 2891 IntRange signedRange = GetExprRange(S.Context, signedOperand); 2892 2893 // Go ahead and analyze implicit conversions in the operands. Note 2894 // that we skip the implicit conversions on both sides. 2895 AnalyzeImplicitConversions(S, lex, E->getOperatorLoc()); 2896 AnalyzeImplicitConversions(S, rex, E->getOperatorLoc()); 2897 2898 // If the signed range is non-negative, -Wsign-compare won't fire, 2899 // but we should still check for comparisons which are always true 2900 // or false. 2901 if (signedRange.NonNegative) 2902 return CheckTrivialUnsignedComparison(S, E); 2903 2904 // For (in)equality comparisons, if the unsigned operand is a 2905 // constant which cannot collide with a overflowed signed operand, 2906 // then reinterpreting the signed operand as unsigned will not 2907 // change the result of the comparison. 2908 if (E->isEqualityOp()) { 2909 unsigned comparisonWidth = S.Context.getIntWidth(T); 2910 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 2911 2912 // We should never be unable to prove that the unsigned operand is 2913 // non-negative. 2914 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 2915 2916 if (unsignedRange.Width < comparisonWidth) 2917 return; 2918 } 2919 2920 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison) 2921 << lex->getType() << rex->getType() 2922 << lex->getSourceRange() << rex->getSourceRange(); 2923 } 2924 2925 /// Analyzes an attempt to assign the given value to a bitfield. 2926 /// 2927 /// Returns true if there was something fishy about the attempt. 2928 bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 2929 SourceLocation InitLoc) { 2930 assert(Bitfield->isBitField()); 2931 if (Bitfield->isInvalidDecl()) 2932 return false; 2933 2934 // White-list bool bitfields. 2935 if (Bitfield->getType()->isBooleanType()) 2936 return false; 2937 2938 // Ignore value- or type-dependent expressions. 2939 if (Bitfield->getBitWidth()->isValueDependent() || 2940 Bitfield->getBitWidth()->isTypeDependent() || 2941 Init->isValueDependent() || 2942 Init->isTypeDependent()) 2943 return false; 2944 2945 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 2946 2947 llvm::APSInt Width(32); 2948 Expr::EvalResult InitValue; 2949 if (!Bitfield->getBitWidth()->isIntegerConstantExpr(Width, S.Context) || 2950 !OriginalInit->Evaluate(InitValue, S.Context) || 2951 !InitValue.Val.isInt()) 2952 return false; 2953 2954 const llvm::APSInt &Value = InitValue.Val.getInt(); 2955 unsigned OriginalWidth = Value.getBitWidth(); 2956 unsigned FieldWidth = Width.getZExtValue(); 2957 2958 if (OriginalWidth <= FieldWidth) 2959 return false; 2960 2961 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 2962 2963 // It's fairly common to write values into signed bitfields 2964 // that, if sign-extended, would end up becoming a different 2965 // value. We don't want to warn about that. 2966 if (Value.isSigned() && Value.isNegative()) 2967 TruncatedValue = TruncatedValue.sext(OriginalWidth); 2968 else 2969 TruncatedValue = TruncatedValue.zext(OriginalWidth); 2970 2971 if (Value == TruncatedValue) 2972 return false; 2973 2974 std::string PrettyValue = Value.toString(10); 2975 std::string PrettyTrunc = TruncatedValue.toString(10); 2976 2977 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 2978 << PrettyValue << PrettyTrunc << OriginalInit->getType() 2979 << Init->getSourceRange(); 2980 2981 return true; 2982 } 2983 2984 /// Analyze the given simple or compound assignment for warning-worthy 2985 /// operations. 2986 void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 2987 // Just recurse on the LHS. 2988 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 2989 2990 // We want to recurse on the RHS as normal unless we're assigning to 2991 // a bitfield. 2992 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 2993 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 2994 E->getOperatorLoc())) { 2995 // Recurse, ignoring any implicit conversions on the RHS. 2996 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 2997 E->getOperatorLoc()); 2998 } 2999 } 3000 3001 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3002 } 3003 3004 /// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3005 void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 3006 SourceLocation CContext, unsigned diag) { 3007 S.Diag(E->getExprLoc(), diag) 3008 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 3009 } 3010 3011 /// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3012 void DiagnoseImpCast(Sema &S, Expr *E, QualType T, SourceLocation CContext, 3013 unsigned diag) { 3014 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag); 3015 } 3016 3017 /// Diagnose an implicit cast from a literal expression. Also attemps to supply 3018 /// fixit hints when the cast wouldn't lose information to simply write the 3019 /// expression with the expected type. 3020 void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 3021 SourceLocation CContext) { 3022 // Emit the primary warning first, then try to emit a fixit hint note if 3023 // reasonable. 3024 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 3025 << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext); 3026 3027 const llvm::APFloat &Value = FL->getValue(); 3028 3029 // Don't attempt to fix PPC double double literals. 3030 if (&Value.getSemantics() == &llvm::APFloat::PPCDoubleDouble) 3031 return; 3032 3033 // Try to convert this exactly to an integer. 3034 bool isExact = false; 3035 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 3036 T->hasUnsignedIntegerRepresentation()); 3037 if (Value.convertToInteger(IntegerValue, 3038 llvm::APFloat::rmTowardZero, &isExact) 3039 != llvm::APFloat::opOK || !isExact) 3040 return; 3041 3042 std::string LiteralValue = IntegerValue.toString(10); 3043 S.Diag(FL->getExprLoc(), diag::note_fix_integral_float_as_integer) 3044 << FixItHint::CreateReplacement(FL->getSourceRange(), LiteralValue); 3045 } 3046 3047 std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 3048 if (!Range.Width) return "0"; 3049 3050 llvm::APSInt ValueInRange = Value; 3051 ValueInRange.setIsSigned(!Range.NonNegative); 3052 ValueInRange = ValueInRange.trunc(Range.Width); 3053 return ValueInRange.toString(10); 3054 } 3055 3056 static bool isFromSystemMacro(Sema &S, SourceLocation loc) { 3057 SourceManager &smgr = S.Context.getSourceManager(); 3058 return loc.isMacroID() && smgr.isInSystemHeader(smgr.getSpellingLoc(loc)); 3059 } 3060 3061 void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 3062 SourceLocation CC, bool *ICContext = 0) { 3063 if (E->isTypeDependent() || E->isValueDependent()) return; 3064 3065 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 3066 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 3067 if (Source == Target) return; 3068 if (Target->isDependentType()) return; 3069 3070 // If the conversion context location is invalid don't complain. 3071 // We also don't want to emit a warning if the issue occurs from the 3072 // instantiation of a system macro. The problem is that 'getSpellingLoc()' 3073 // is slow, so we delay this check as long as possible. Once we detect 3074 // we are in that scenario, we just return. 3075 if (CC.isInvalid()) 3076 return; 3077 3078 // Never diagnose implicit casts to bool. 3079 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 3080 return; 3081 3082 // Strip vector types. 3083 if (isa<VectorType>(Source)) { 3084 if (!isa<VectorType>(Target)) { 3085 if (isFromSystemMacro(S, CC)) 3086 return; 3087 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 3088 } 3089 3090 // If the vector cast is cast between two vectors of the same size, it is 3091 // a bitcast, not a conversion. 3092 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 3093 return; 3094 3095 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 3096 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 3097 } 3098 3099 // Strip complex types. 3100 if (isa<ComplexType>(Source)) { 3101 if (!isa<ComplexType>(Target)) { 3102 if (isFromSystemMacro(S, CC)) 3103 return; 3104 3105 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 3106 } 3107 3108 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 3109 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 3110 } 3111 3112 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 3113 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 3114 3115 // If the source is floating point... 3116 if (SourceBT && SourceBT->isFloatingPoint()) { 3117 // ...and the target is floating point... 3118 if (TargetBT && TargetBT->isFloatingPoint()) { 3119 // ...then warn if we're dropping FP rank. 3120 3121 // Builtin FP kinds are ordered by increasing FP rank. 3122 if (SourceBT->getKind() > TargetBT->getKind()) { 3123 // Don't warn about float constants that are precisely 3124 // representable in the target type. 3125 Expr::EvalResult result; 3126 if (E->Evaluate(result, S.Context)) { 3127 // Value might be a float, a float vector, or a float complex. 3128 if (IsSameFloatAfterCast(result.Val, 3129 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 3130 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 3131 return; 3132 } 3133 3134 if (isFromSystemMacro(S, CC)) 3135 return; 3136 3137 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 3138 } 3139 return; 3140 } 3141 3142 // If the target is integral, always warn. 3143 if ((TargetBT && TargetBT->isInteger())) { 3144 if (isFromSystemMacro(S, CC)) 3145 return; 3146 3147 Expr *InnerE = E->IgnoreParenImpCasts(); 3148 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 3149 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 3150 } else { 3151 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 3152 } 3153 } 3154 3155 return; 3156 } 3157 3158 if (!Source->isIntegerType() || !Target->isIntegerType()) 3159 return; 3160 3161 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 3162 == Expr::NPCK_GNUNull) && Target->isIntegerType()) { 3163 S.Diag(E->getExprLoc(), diag::warn_impcast_null_pointer_to_integer) 3164 << E->getSourceRange() << clang::SourceRange(CC); 3165 return; 3166 } 3167 3168 IntRange SourceRange = GetExprRange(S.Context, E); 3169 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 3170 3171 if (SourceRange.Width > TargetRange.Width) { 3172 // If the source is a constant, use a default-on diagnostic. 3173 // TODO: this should happen for bitfield stores, too. 3174 llvm::APSInt Value(32); 3175 if (E->isIntegerConstantExpr(Value, S.Context)) { 3176 if (isFromSystemMacro(S, CC)) 3177 return; 3178 3179 std::string PrettySourceValue = Value.toString(10); 3180 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 3181 3182 S.Diag(E->getExprLoc(), diag::warn_impcast_integer_precision_constant) 3183 << PrettySourceValue << PrettyTargetValue 3184 << E->getType() << T << E->getSourceRange() << clang::SourceRange(CC); 3185 return; 3186 } 3187 3188 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 3189 if (isFromSystemMacro(S, CC)) 3190 return; 3191 3192 if (SourceRange.Width == 64 && TargetRange.Width == 32) 3193 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32); 3194 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 3195 } 3196 3197 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 3198 (!TargetRange.NonNegative && SourceRange.NonNegative && 3199 SourceRange.Width == TargetRange.Width)) { 3200 3201 if (isFromSystemMacro(S, CC)) 3202 return; 3203 3204 unsigned DiagID = diag::warn_impcast_integer_sign; 3205 3206 // Traditionally, gcc has warned about this under -Wsign-compare. 3207 // We also want to warn about it in -Wconversion. 3208 // So if -Wconversion is off, use a completely identical diagnostic 3209 // in the sign-compare group. 3210 // The conditional-checking code will 3211 if (ICContext) { 3212 DiagID = diag::warn_impcast_integer_sign_conditional; 3213 *ICContext = true; 3214 } 3215 3216 return DiagnoseImpCast(S, E, T, CC, DiagID); 3217 } 3218 3219 // Diagnose conversions between different enumeration types. 3220 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 3221 // type, to give us better diagnostics. 3222 QualType SourceType = E->getType(); 3223 if (!S.getLangOptions().CPlusPlus) { 3224 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 3225 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 3226 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 3227 SourceType = S.Context.getTypeDeclType(Enum); 3228 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 3229 } 3230 } 3231 3232 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 3233 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 3234 if ((SourceEnum->getDecl()->getIdentifier() || 3235 SourceEnum->getDecl()->getTypedefNameForAnonDecl()) && 3236 (TargetEnum->getDecl()->getIdentifier() || 3237 TargetEnum->getDecl()->getTypedefNameForAnonDecl()) && 3238 SourceEnum != TargetEnum) { 3239 if (isFromSystemMacro(S, CC)) 3240 return; 3241 3242 return DiagnoseImpCast(S, E, SourceType, T, CC, 3243 diag::warn_impcast_different_enum_types); 3244 } 3245 3246 return; 3247 } 3248 3249 void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 3250 3251 void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 3252 SourceLocation CC, bool &ICContext) { 3253 E = E->IgnoreParenImpCasts(); 3254 3255 if (isa<ConditionalOperator>(E)) 3256 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 3257 3258 AnalyzeImplicitConversions(S, E, CC); 3259 if (E->getType() != T) 3260 return CheckImplicitConversion(S, E, T, CC, &ICContext); 3261 return; 3262 } 3263 3264 void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 3265 SourceLocation CC = E->getQuestionLoc(); 3266 3267 AnalyzeImplicitConversions(S, E->getCond(), CC); 3268 3269 bool Suspicious = false; 3270 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 3271 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 3272 3273 // If -Wconversion would have warned about either of the candidates 3274 // for a signedness conversion to the context type... 3275 if (!Suspicious) return; 3276 3277 // ...but it's currently ignored... 3278 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 3279 CC)) 3280 return; 3281 3282 // ...and -Wsign-compare isn't... 3283 if (!S.Diags.getDiagnosticLevel(diag::warn_mixed_sign_conditional, CC)) 3284 return; 3285 3286 // ...then check whether it would have warned about either of the 3287 // candidates for a signedness conversion to the condition type. 3288 if (E->getType() != T) { 3289 Suspicious = false; 3290 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 3291 E->getType(), CC, &Suspicious); 3292 if (!Suspicious) 3293 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 3294 E->getType(), CC, &Suspicious); 3295 if (!Suspicious) 3296 return; 3297 } 3298 3299 // If so, emit a diagnostic under -Wsign-compare. 3300 Expr *lex = E->getTrueExpr()->IgnoreParenImpCasts(); 3301 Expr *rex = E->getFalseExpr()->IgnoreParenImpCasts(); 3302 S.Diag(E->getQuestionLoc(), diag::warn_mixed_sign_conditional) 3303 << lex->getType() << rex->getType() 3304 << lex->getSourceRange() << rex->getSourceRange(); 3305 } 3306 3307 /// AnalyzeImplicitConversions - Find and report any interesting 3308 /// implicit conversions in the given expression. There are a couple 3309 /// of competing diagnostics here, -Wconversion and -Wsign-compare. 3310 void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 3311 QualType T = OrigE->getType(); 3312 Expr *E = OrigE->IgnoreParenImpCasts(); 3313 3314 // For conditional operators, we analyze the arguments as if they 3315 // were being fed directly into the output. 3316 if (isa<ConditionalOperator>(E)) { 3317 ConditionalOperator *CO = cast<ConditionalOperator>(E); 3318 CheckConditionalOperator(S, CO, T); 3319 return; 3320 } 3321 3322 // Go ahead and check any implicit conversions we might have skipped. 3323 // The non-canonical typecheck is just an optimization; 3324 // CheckImplicitConversion will filter out dead implicit conversions. 3325 if (E->getType() != T) 3326 CheckImplicitConversion(S, E, T, CC); 3327 3328 // Now continue drilling into this expression. 3329 3330 // Skip past explicit casts. 3331 if (isa<ExplicitCastExpr>(E)) { 3332 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 3333 return AnalyzeImplicitConversions(S, E, CC); 3334 } 3335 3336 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 3337 // Do a somewhat different check with comparison operators. 3338 if (BO->isComparisonOp()) 3339 return AnalyzeComparison(S, BO); 3340 3341 // And with assignments and compound assignments. 3342 if (BO->isAssignmentOp()) 3343 return AnalyzeAssignment(S, BO); 3344 } 3345 3346 // These break the otherwise-useful invariant below. Fortunately, 3347 // we don't really need to recurse into them, because any internal 3348 // expressions should have been analyzed already when they were 3349 // built into statements. 3350 if (isa<StmtExpr>(E)) return; 3351 3352 // Don't descend into unevaluated contexts. 3353 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 3354 3355 // Now just recurse over the expression's children. 3356 CC = E->getExprLoc(); 3357 for (Stmt::child_range I = E->children(); I; ++I) 3358 AnalyzeImplicitConversions(S, cast<Expr>(*I), CC); 3359 } 3360 3361 } // end anonymous namespace 3362 3363 /// Diagnoses "dangerous" implicit conversions within the given 3364 /// expression (which is a full expression). Implements -Wconversion 3365 /// and -Wsign-compare. 3366 /// 3367 /// \param CC the "context" location of the implicit conversion, i.e. 3368 /// the most location of the syntactic entity requiring the implicit 3369 /// conversion 3370 void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 3371 // Don't diagnose in unevaluated contexts. 3372 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 3373 return; 3374 3375 // Don't diagnose for value- or type-dependent expressions. 3376 if (E->isTypeDependent() || E->isValueDependent()) 3377 return; 3378 3379 // This is not the right CC for (e.g.) a variable initialization. 3380 AnalyzeImplicitConversions(*this, E, CC); 3381 } 3382 3383 void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 3384 FieldDecl *BitField, 3385 Expr *Init) { 3386 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 3387 } 3388 3389 /// CheckParmsForFunctionDef - Check that the parameters of the given 3390 /// function are appropriate for the definition of a function. This 3391 /// takes care of any checks that cannot be performed on the 3392 /// declaration itself, e.g., that the types of each of the function 3393 /// parameters are complete. 3394 bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 3395 bool CheckParameterNames) { 3396 bool HasInvalidParm = false; 3397 for (; P != PEnd; ++P) { 3398 ParmVarDecl *Param = *P; 3399 3400 // C99 6.7.5.3p4: the parameters in a parameter type list in a 3401 // function declarator that is part of a function definition of 3402 // that function shall not have incomplete type. 3403 // 3404 // This is also C++ [dcl.fct]p6. 3405 if (!Param->isInvalidDecl() && 3406 RequireCompleteType(Param->getLocation(), Param->getType(), 3407 diag::err_typecheck_decl_incomplete_type)) { 3408 Param->setInvalidDecl(); 3409 HasInvalidParm = true; 3410 } 3411 3412 // C99 6.9.1p5: If the declarator includes a parameter type list, the 3413 // declaration of each parameter shall include an identifier. 3414 if (CheckParameterNames && 3415 Param->getIdentifier() == 0 && 3416 !Param->isImplicit() && 3417 !getLangOptions().CPlusPlus) 3418 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 3419 3420 // C99 6.7.5.3p12: 3421 // If the function declarator is not part of a definition of that 3422 // function, parameters may have incomplete type and may use the [*] 3423 // notation in their sequences of declarator specifiers to specify 3424 // variable length array types. 3425 QualType PType = Param->getOriginalType(); 3426 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 3427 if (AT->getSizeModifier() == ArrayType::Star) { 3428 // FIXME: This diagnosic should point the the '[*]' if source-location 3429 // information is added for it. 3430 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 3431 } 3432 } 3433 } 3434 3435 return HasInvalidParm; 3436 } 3437 3438 /// CheckCastAlign - Implements -Wcast-align, which warns when a 3439 /// pointer cast increases the alignment requirements. 3440 void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 3441 // This is actually a lot of work to potentially be doing on every 3442 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 3443 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 3444 TRange.getBegin()) 3445 == Diagnostic::Ignored) 3446 return; 3447 3448 // Ignore dependent types. 3449 if (T->isDependentType() || Op->getType()->isDependentType()) 3450 return; 3451 3452 // Require that the destination be a pointer type. 3453 const PointerType *DestPtr = T->getAs<PointerType>(); 3454 if (!DestPtr) return; 3455 3456 // If the destination has alignment 1, we're done. 3457 QualType DestPointee = DestPtr->getPointeeType(); 3458 if (DestPointee->isIncompleteType()) return; 3459 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 3460 if (DestAlign.isOne()) return; 3461 3462 // Require that the source be a pointer type. 3463 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 3464 if (!SrcPtr) return; 3465 QualType SrcPointee = SrcPtr->getPointeeType(); 3466 3467 // Whitelist casts from cv void*. We already implicitly 3468 // whitelisted casts to cv void*, since they have alignment 1. 3469 // Also whitelist casts involving incomplete types, which implicitly 3470 // includes 'void'. 3471 if (SrcPointee->isIncompleteType()) return; 3472 3473 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 3474 if (SrcAlign >= DestAlign) return; 3475 3476 Diag(TRange.getBegin(), diag::warn_cast_align) 3477 << Op->getType() << T 3478 << static_cast<unsigned>(SrcAlign.getQuantity()) 3479 << static_cast<unsigned>(DestAlign.getQuantity()) 3480 << TRange << Op->getSourceRange(); 3481 } 3482 3483 static void CheckArrayAccess_Check(Sema &S, 3484 const clang::ArraySubscriptExpr *E) { 3485 const Expr *BaseExpr = E->getBase()->IgnoreParenImpCasts(); 3486 const ConstantArrayType *ArrayTy = 3487 S.Context.getAsConstantArrayType(BaseExpr->getType()); 3488 if (!ArrayTy) 3489 return; 3490 3491 const Expr *IndexExpr = E->getIdx(); 3492 if (IndexExpr->isValueDependent()) 3493 return; 3494 llvm::APSInt index; 3495 if (!IndexExpr->isIntegerConstantExpr(index, S.Context)) 3496 return; 3497 3498 if (index.isUnsigned() || !index.isNegative()) { 3499 llvm::APInt size = ArrayTy->getSize(); 3500 if (!size.isStrictlyPositive()) 3501 return; 3502 if (size.getBitWidth() > index.getBitWidth()) 3503 index = index.sext(size.getBitWidth()); 3504 else if (size.getBitWidth() < index.getBitWidth()) 3505 size = size.sext(index.getBitWidth()); 3506 3507 if (index.slt(size)) 3508 return; 3509 3510 S.DiagRuntimeBehavior(E->getBase()->getLocStart(), BaseExpr, 3511 S.PDiag(diag::warn_array_index_exceeds_bounds) 3512 << index.toString(10, true) 3513 << size.toString(10, true) 3514 << IndexExpr->getSourceRange()); 3515 } else { 3516 S.DiagRuntimeBehavior(E->getBase()->getLocStart(), BaseExpr, 3517 S.PDiag(diag::warn_array_index_precedes_bounds) 3518 << index.toString(10, true) 3519 << IndexExpr->getSourceRange()); 3520 } 3521 3522 const NamedDecl *ND = NULL; 3523 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 3524 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 3525 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 3526 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 3527 if (ND) 3528 S.DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 3529 S.PDiag(diag::note_array_index_out_of_bounds) 3530 << ND->getDeclName()); 3531 } 3532 3533 void Sema::CheckArrayAccess(const Expr *expr) { 3534 while (true) { 3535 expr = expr->IgnoreParens(); 3536 switch (expr->getStmtClass()) { 3537 case Stmt::ArraySubscriptExprClass: 3538 CheckArrayAccess_Check(*this, cast<ArraySubscriptExpr>(expr)); 3539 return; 3540 case Stmt::ConditionalOperatorClass: { 3541 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 3542 if (const Expr *lhs = cond->getLHS()) 3543 CheckArrayAccess(lhs); 3544 if (const Expr *rhs = cond->getRHS()) 3545 CheckArrayAccess(rhs); 3546 return; 3547 } 3548 default: 3549 return; 3550 } 3551 } 3552 } 3553 3554 //===--- CHECK: Objective-C retain cycles ----------------------------------// 3555 3556 namespace { 3557 struct RetainCycleOwner { 3558 RetainCycleOwner() : Variable(0), Indirect(false) {} 3559 VarDecl *Variable; 3560 SourceRange Range; 3561 SourceLocation Loc; 3562 bool Indirect; 3563 3564 void setLocsFrom(Expr *e) { 3565 Loc = e->getExprLoc(); 3566 Range = e->getSourceRange(); 3567 } 3568 }; 3569 } 3570 3571 /// Consider whether capturing the given variable can possibly lead to 3572 /// a retain cycle. 3573 static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 3574 // In ARC, it's captured strongly iff the variable has __strong 3575 // lifetime. In MRR, it's captured strongly if the variable is 3576 // __block and has an appropriate type. 3577 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 3578 return false; 3579 3580 owner.Variable = var; 3581 owner.setLocsFrom(ref); 3582 return true; 3583 } 3584 3585 static bool findRetainCycleOwner(Expr *e, RetainCycleOwner &owner) { 3586 while (true) { 3587 e = e->IgnoreParens(); 3588 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 3589 switch (cast->getCastKind()) { 3590 case CK_BitCast: 3591 case CK_LValueBitCast: 3592 case CK_LValueToRValue: 3593 case CK_ObjCReclaimReturnedObject: 3594 e = cast->getSubExpr(); 3595 continue; 3596 3597 case CK_GetObjCProperty: { 3598 // Bail out if this isn't a strong explicit property. 3599 const ObjCPropertyRefExpr *pre = cast->getSubExpr()->getObjCProperty(); 3600 if (pre->isImplicitProperty()) return false; 3601 ObjCPropertyDecl *property = pre->getExplicitProperty(); 3602 if (!(property->getPropertyAttributes() & 3603 (ObjCPropertyDecl::OBJC_PR_retain | 3604 ObjCPropertyDecl::OBJC_PR_copy | 3605 ObjCPropertyDecl::OBJC_PR_strong)) && 3606 !(property->getPropertyIvarDecl() && 3607 property->getPropertyIvarDecl()->getType() 3608 .getObjCLifetime() == Qualifiers::OCL_Strong)) 3609 return false; 3610 3611 owner.Indirect = true; 3612 e = const_cast<Expr*>(pre->getBase()); 3613 continue; 3614 } 3615 3616 default: 3617 return false; 3618 } 3619 } 3620 3621 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 3622 ObjCIvarDecl *ivar = ref->getDecl(); 3623 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 3624 return false; 3625 3626 // Try to find a retain cycle in the base. 3627 if (!findRetainCycleOwner(ref->getBase(), owner)) 3628 return false; 3629 3630 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 3631 owner.Indirect = true; 3632 return true; 3633 } 3634 3635 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 3636 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 3637 if (!var) return false; 3638 return considerVariable(var, ref, owner); 3639 } 3640 3641 if (BlockDeclRefExpr *ref = dyn_cast<BlockDeclRefExpr>(e)) { 3642 owner.Variable = ref->getDecl(); 3643 owner.setLocsFrom(ref); 3644 return true; 3645 } 3646 3647 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 3648 if (member->isArrow()) return false; 3649 3650 // Don't count this as an indirect ownership. 3651 e = member->getBase(); 3652 continue; 3653 } 3654 3655 // Array ivars? 3656 3657 return false; 3658 } 3659 } 3660 3661 namespace { 3662 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 3663 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 3664 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 3665 Variable(variable), Capturer(0) {} 3666 3667 VarDecl *Variable; 3668 Expr *Capturer; 3669 3670 void VisitDeclRefExpr(DeclRefExpr *ref) { 3671 if (ref->getDecl() == Variable && !Capturer) 3672 Capturer = ref; 3673 } 3674 3675 void VisitBlockDeclRefExpr(BlockDeclRefExpr *ref) { 3676 if (ref->getDecl() == Variable && !Capturer) 3677 Capturer = ref; 3678 } 3679 3680 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 3681 if (Capturer) return; 3682 Visit(ref->getBase()); 3683 if (Capturer && ref->isFreeIvar()) 3684 Capturer = ref; 3685 } 3686 3687 void VisitBlockExpr(BlockExpr *block) { 3688 // Look inside nested blocks 3689 if (block->getBlockDecl()->capturesVariable(Variable)) 3690 Visit(block->getBlockDecl()->getBody()); 3691 } 3692 }; 3693 } 3694 3695 /// Check whether the given argument is a block which captures a 3696 /// variable. 3697 static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 3698 assert(owner.Variable && owner.Loc.isValid()); 3699 3700 e = e->IgnoreParenCasts(); 3701 BlockExpr *block = dyn_cast<BlockExpr>(e); 3702 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 3703 return 0; 3704 3705 FindCaptureVisitor visitor(S.Context, owner.Variable); 3706 visitor.Visit(block->getBlockDecl()->getBody()); 3707 return visitor.Capturer; 3708 } 3709 3710 static void diagnoseRetainCycle(Sema &S, Expr *capturer, 3711 RetainCycleOwner &owner) { 3712 assert(capturer); 3713 assert(owner.Variable && owner.Loc.isValid()); 3714 3715 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 3716 << owner.Variable << capturer->getSourceRange(); 3717 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 3718 << owner.Indirect << owner.Range; 3719 } 3720 3721 /// Check for a keyword selector that starts with the word 'add' or 3722 /// 'set'. 3723 static bool isSetterLikeSelector(Selector sel) { 3724 if (sel.isUnarySelector()) return false; 3725 3726 llvm::StringRef str = sel.getNameForSlot(0); 3727 while (!str.empty() && str.front() == '_') str = str.substr(1); 3728 if (str.startswith("set") || str.startswith("add")) 3729 str = str.substr(3); 3730 else 3731 return false; 3732 3733 if (str.empty()) return true; 3734 return !islower(str.front()); 3735 } 3736 3737 /// Check a message send to see if it's likely to cause a retain cycle. 3738 void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 3739 // Only check instance methods whose selector looks like a setter. 3740 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 3741 return; 3742 3743 // Try to find a variable that the receiver is strongly owned by. 3744 RetainCycleOwner owner; 3745 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 3746 if (!findRetainCycleOwner(msg->getInstanceReceiver(), owner)) 3747 return; 3748 } else { 3749 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 3750 owner.Variable = getCurMethodDecl()->getSelfDecl(); 3751 owner.Loc = msg->getSuperLoc(); 3752 owner.Range = msg->getSuperLoc(); 3753 } 3754 3755 // Check whether the receiver is captured by any of the arguments. 3756 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 3757 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 3758 return diagnoseRetainCycle(*this, capturer, owner); 3759 } 3760 3761 /// Check a property assign to see if it's likely to cause a retain cycle. 3762 void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 3763 RetainCycleOwner owner; 3764 if (!findRetainCycleOwner(receiver, owner)) 3765 return; 3766 3767 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 3768 diagnoseRetainCycle(*this, capturer, owner); 3769 } 3770 3771 bool Sema::checkUnsafeAssigns(SourceLocation Loc, 3772 QualType LHS, Expr *RHS) { 3773 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 3774 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 3775 return false; 3776 // strip off any implicit cast added to get to the one arc-specific 3777 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 3778 if (cast->getCastKind() == CK_ObjCConsumeObject) { 3779 Diag(Loc, diag::warn_arc_retained_assign) 3780 << (LT == Qualifiers::OCL_ExplicitNone) 3781 << RHS->getSourceRange(); 3782 return true; 3783 } 3784 RHS = cast->getSubExpr(); 3785 } 3786 return false; 3787 } 3788 3789 void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 3790 Expr *LHS, Expr *RHS) { 3791 QualType LHSType = LHS->getType(); 3792 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 3793 return; 3794 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 3795 // FIXME. Check for other life times. 3796 if (LT != Qualifiers::OCL_None) 3797 return; 3798 3799 if (ObjCPropertyRefExpr *PRE = dyn_cast<ObjCPropertyRefExpr>(LHS)) { 3800 if (PRE->isImplicitProperty()) 3801 return; 3802 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 3803 if (!PD) 3804 return; 3805 3806 unsigned Attributes = PD->getPropertyAttributes(); 3807 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) 3808 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 3809 if (cast->getCastKind() == CK_ObjCConsumeObject) { 3810 Diag(Loc, diag::warn_arc_retained_property_assign) 3811 << RHS->getSourceRange(); 3812 return; 3813 } 3814 RHS = cast->getSubExpr(); 3815 } 3816 } 3817 } 3818