SemaChecking.cpp revision a1f3dba77b7418575c1ff539ffa74ebaa068280c
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 "Sema.h" 16#include "clang/AST/ASTContext.h" 17#include "clang/AST/DeclObjC.h" 18#include "clang/AST/ExprCXX.h" 19#include "clang/AST/ExprObjC.h" 20#include "clang/Lex/LiteralSupport.h" 21#include "clang/Lex/Preprocessor.h" 22#include <limits> 23using namespace clang; 24 25/// getLocationOfStringLiteralByte - Return a source location that points to the 26/// specified byte of the specified string literal. 27/// 28/// Strings are amazingly complex. They can be formed from multiple tokens and 29/// can have escape sequences in them in addition to the usual trigraph and 30/// escaped newline business. This routine handles this complexity. 31/// 32SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 33 unsigned ByteNo) const { 34 assert(!SL->isWide() && "This doesn't work for wide strings yet"); 35 36 // Loop over all of the tokens in this string until we find the one that 37 // contains the byte we're looking for. 38 unsigned TokNo = 0; 39 while (1) { 40 assert(TokNo < SL->getNumConcatenated() && "Invalid byte number!"); 41 SourceLocation StrTokLoc = SL->getStrTokenLoc(TokNo); 42 43 // Get the spelling of the string so that we can get the data that makes up 44 // the string literal, not the identifier for the macro it is potentially 45 // expanded through. 46 SourceLocation StrTokSpellingLoc = SourceMgr.getSpellingLoc(StrTokLoc); 47 48 // Re-lex the token to get its length and original spelling. 49 std::pair<FileID, unsigned> LocInfo = 50 SourceMgr.getDecomposedLoc(StrTokSpellingLoc); 51 std::pair<const char *,const char *> Buffer = 52 SourceMgr.getBufferData(LocInfo.first); 53 const char *StrData = Buffer.first+LocInfo.second; 54 55 // Create a langops struct and enable trigraphs. This is sufficient for 56 // relexing tokens. 57 LangOptions LangOpts; 58 LangOpts.Trigraphs = true; 59 60 // Create a lexer starting at the beginning of this token. 61 Lexer TheLexer(StrTokSpellingLoc, LangOpts, Buffer.first, StrData, 62 Buffer.second); 63 Token TheTok; 64 TheLexer.LexFromRawLexer(TheTok); 65 66 // Use the StringLiteralParser to compute the length of the string in bytes. 67 StringLiteralParser SLP(&TheTok, 1, PP); 68 unsigned TokNumBytes = SLP.GetStringLength(); 69 70 // If the byte is in this token, return the location of the byte. 71 if (ByteNo < TokNumBytes || 72 (ByteNo == TokNumBytes && TokNo == SL->getNumConcatenated())) { 73 unsigned Offset = 74 StringLiteralParser::getOffsetOfStringByte(TheTok, ByteNo, PP); 75 76 // Now that we know the offset of the token in the spelling, use the 77 // preprocessor to get the offset in the original source. 78 return PP.AdvanceToTokenCharacter(StrTokLoc, Offset); 79 } 80 81 // Move to the next string token. 82 ++TokNo; 83 ByteNo -= TokNumBytes; 84 } 85} 86 87 88/// CheckFunctionCall - Check a direct function call for various correctness 89/// and safety properties not strictly enforced by the C type system. 90Action::OwningExprResult 91Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 92 OwningExprResult TheCallResult(Owned(TheCall)); 93 // Get the IdentifierInfo* for the called function. 94 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 95 96 // None of the checks below are needed for functions that don't have 97 // simple names (e.g., C++ conversion functions). 98 if (!FnInfo) 99 return move(TheCallResult); 100 101 switch (FDecl->getBuiltinID(Context)) { 102 case Builtin::BI__builtin___CFStringMakeConstantString: 103 assert(TheCall->getNumArgs() == 1 && 104 "Wrong # arguments to builtin CFStringMakeConstantString"); 105 if (CheckObjCString(TheCall->getArg(0))) 106 return ExprError(); 107 return move(TheCallResult); 108 case Builtin::BI__builtin_stdarg_start: 109 case Builtin::BI__builtin_va_start: 110 if (SemaBuiltinVAStart(TheCall)) 111 return ExprError(); 112 return move(TheCallResult); 113 case Builtin::BI__builtin_isgreater: 114 case Builtin::BI__builtin_isgreaterequal: 115 case Builtin::BI__builtin_isless: 116 case Builtin::BI__builtin_islessequal: 117 case Builtin::BI__builtin_islessgreater: 118 case Builtin::BI__builtin_isunordered: 119 if (SemaBuiltinUnorderedCompare(TheCall)) 120 return ExprError(); 121 return move(TheCallResult); 122 case Builtin::BI__builtin_return_address: 123 case Builtin::BI__builtin_frame_address: 124 if (SemaBuiltinStackAddress(TheCall)) 125 return ExprError(); 126 return move(TheCallResult); 127 case Builtin::BI__builtin_shufflevector: 128 return SemaBuiltinShuffleVector(TheCall); 129 // TheCall will be freed by the smart pointer here, but that's fine, since 130 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 131 case Builtin::BI__builtin_prefetch: 132 if (SemaBuiltinPrefetch(TheCall)) 133 return ExprError(); 134 return move(TheCallResult); 135 case Builtin::BI__builtin_object_size: 136 if (SemaBuiltinObjectSize(TheCall)) 137 return ExprError(); 138 return move(TheCallResult); 139 case Builtin::BI__builtin_longjmp: 140 if (SemaBuiltinLongjmp(TheCall)) 141 return ExprError(); 142 return move(TheCallResult); 143 case Builtin::BI__sync_fetch_and_add: 144 case Builtin::BI__sync_fetch_and_sub: 145 case Builtin::BI__sync_fetch_and_or: 146 case Builtin::BI__sync_fetch_and_and: 147 case Builtin::BI__sync_fetch_and_xor: 148 case Builtin::BI__sync_fetch_and_nand: 149 case Builtin::BI__sync_add_and_fetch: 150 case Builtin::BI__sync_sub_and_fetch: 151 case Builtin::BI__sync_and_and_fetch: 152 case Builtin::BI__sync_or_and_fetch: 153 case Builtin::BI__sync_xor_and_fetch: 154 case Builtin::BI__sync_nand_and_fetch: 155 case Builtin::BI__sync_val_compare_and_swap: 156 case Builtin::BI__sync_bool_compare_and_swap: 157 case Builtin::BI__sync_lock_test_and_set: 158 case Builtin::BI__sync_lock_release: 159 if (SemaBuiltinAtomicOverloaded(TheCall)) 160 return ExprError(); 161 return move(TheCallResult); 162 } 163 164 // FIXME: This mechanism should be abstracted to be less fragile and 165 // more efficient. For example, just map function ids to custom 166 // handlers. 167 168 // Printf checking. 169 if (const FormatAttr *Format = FDecl->getAttr<FormatAttr>()) { 170 if (Format->getType() == "printf") { 171 bool HasVAListArg = Format->getFirstArg() == 0; 172 if (!HasVAListArg) { 173 if (const FunctionProtoType *Proto 174 = FDecl->getType()->getAsFunctionProtoType()) 175 HasVAListArg = !Proto->isVariadic(); 176 } 177 CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 178 HasVAListArg ? 0 : Format->getFirstArg() - 1); 179 } 180 } 181 182 return move(TheCallResult); 183} 184 185Action::OwningExprResult 186Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 187 188 OwningExprResult TheCallResult(Owned(TheCall)); 189 // Printf checking. 190 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 191 if (!Format) 192 return move(TheCallResult); 193 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 194 if (!V) 195 return move(TheCallResult); 196 QualType Ty = V->getType(); 197 if (!Ty->isBlockPointerType()) 198 return move(TheCallResult); 199 if (Format->getType() == "printf") { 200 bool HasVAListArg = Format->getFirstArg() == 0; 201 if (!HasVAListArg) { 202 const FunctionType *FT = 203 Ty->getAsBlockPointerType()->getPointeeType()->getAsFunctionType(); 204 if (const FunctionProtoType *Proto = dyn_cast<FunctionProtoType>(FT)) 205 HasVAListArg = !Proto->isVariadic(); 206 } 207 CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 208 HasVAListArg ? 0 : Format->getFirstArg() - 1); 209 } 210 return move(TheCallResult); 211} 212 213/// SemaBuiltinAtomicOverloaded - We have a call to a function like 214/// __sync_fetch_and_add, which is an overloaded function based on the pointer 215/// type of its first argument. The main ActOnCallExpr routines have already 216/// promoted the types of arguments because all of these calls are prototyped as 217/// void(...). 218/// 219/// This function goes through and does final semantic checking for these 220/// builtins, 221bool Sema::SemaBuiltinAtomicOverloaded(CallExpr *TheCall) { 222 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 223 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 224 225 // Ensure that we have at least one argument to do type inference from. 226 if (TheCall->getNumArgs() < 1) 227 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 228 << 0 << TheCall->getCallee()->getSourceRange(); 229 230 // Inspect the first argument of the atomic builtin. This should always be 231 // a pointer type, whose element is an integral scalar or pointer type. 232 // Because it is a pointer type, we don't have to worry about any implicit 233 // casts here. 234 Expr *FirstArg = TheCall->getArg(0); 235 if (!FirstArg->getType()->isPointerType()) 236 return Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 237 << FirstArg->getType() << FirstArg->getSourceRange(); 238 239 QualType ValType = FirstArg->getType()->getAsPointerType()->getPointeeType(); 240 if (!ValType->isIntegerType() && !ValType->isPointerType() && 241 !ValType->isBlockPointerType()) 242 return Diag(DRE->getLocStart(), 243 diag::err_atomic_builtin_must_be_pointer_intptr) 244 << FirstArg->getType() << FirstArg->getSourceRange(); 245 246 // We need to figure out which concrete builtin this maps onto. For example, 247 // __sync_fetch_and_add with a 2 byte object turns into 248 // __sync_fetch_and_add_2. 249#define BUILTIN_ROW(x) \ 250 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 251 Builtin::BI##x##_8, Builtin::BI##x##_16 } 252 253 static const unsigned BuiltinIndices[][5] = { 254 BUILTIN_ROW(__sync_fetch_and_add), 255 BUILTIN_ROW(__sync_fetch_and_sub), 256 BUILTIN_ROW(__sync_fetch_and_or), 257 BUILTIN_ROW(__sync_fetch_and_and), 258 BUILTIN_ROW(__sync_fetch_and_xor), 259 BUILTIN_ROW(__sync_fetch_and_nand), 260 261 BUILTIN_ROW(__sync_add_and_fetch), 262 BUILTIN_ROW(__sync_sub_and_fetch), 263 BUILTIN_ROW(__sync_and_and_fetch), 264 BUILTIN_ROW(__sync_or_and_fetch), 265 BUILTIN_ROW(__sync_xor_and_fetch), 266 BUILTIN_ROW(__sync_nand_and_fetch), 267 268 BUILTIN_ROW(__sync_val_compare_and_swap), 269 BUILTIN_ROW(__sync_bool_compare_and_swap), 270 BUILTIN_ROW(__sync_lock_test_and_set), 271 BUILTIN_ROW(__sync_lock_release) 272 }; 273#undef BUILTIN_ROW 274 275 // Determine the index of the size. 276 unsigned SizeIndex; 277 switch (Context.getTypeSize(ValType)/8) { 278 case 1: SizeIndex = 0; break; 279 case 2: SizeIndex = 1; break; 280 case 4: SizeIndex = 2; break; 281 case 8: SizeIndex = 3; break; 282 case 16: SizeIndex = 4; break; 283 default: 284 return Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 285 << FirstArg->getType() << FirstArg->getSourceRange(); 286 } 287 288 // Each of these builtins has one pointer argument, followed by some number of 289 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 290 // that we ignore. Find out which row of BuiltinIndices to read from as well 291 // as the number of fixed args. 292 unsigned BuiltinID = FDecl->getBuiltinID(Context); 293 unsigned BuiltinIndex, NumFixed = 1; 294 switch (BuiltinID) { 295 default: assert(0 && "Unknown overloaded atomic builtin!"); 296 case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break; 297 case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break; 298 case Builtin::BI__sync_fetch_and_or: BuiltinIndex = 2; break; 299 case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break; 300 case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break; 301 case Builtin::BI__sync_fetch_and_nand:BuiltinIndex = 5; break; 302 303 case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 6; break; 304 case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 7; break; 305 case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 8; break; 306 case Builtin::BI__sync_or_and_fetch: BuiltinIndex = 9; break; 307 case Builtin::BI__sync_xor_and_fetch: BuiltinIndex =10; break; 308 case Builtin::BI__sync_nand_and_fetch:BuiltinIndex =11; break; 309 310 case Builtin::BI__sync_val_compare_and_swap: 311 BuiltinIndex = 12; 312 NumFixed = 2; 313 break; 314 case Builtin::BI__sync_bool_compare_and_swap: 315 BuiltinIndex = 13; 316 NumFixed = 2; 317 break; 318 case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 14; break; 319 case Builtin::BI__sync_lock_release: 320 BuiltinIndex = 15; 321 NumFixed = 0; 322 break; 323 } 324 325 // Now that we know how many fixed arguments we expect, first check that we 326 // have at least that many. 327 if (TheCall->getNumArgs() < 1+NumFixed) 328 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 329 << 0 << TheCall->getCallee()->getSourceRange(); 330 331 332 // Get the decl for the concrete builtin from this, we can tell what the 333 // concrete integer type we should convert to is. 334 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 335 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 336 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 337 FunctionDecl *NewBuiltinDecl = 338 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 339 TUScope, false, DRE->getLocStart())); 340 const FunctionProtoType *BuiltinFT = 341 NewBuiltinDecl->getType()->getAsFunctionProtoType(); 342 ValType = BuiltinFT->getArgType(0)->getAsPointerType()->getPointeeType(); 343 344 // If the first type needs to be converted (e.g. void** -> int*), do it now. 345 if (BuiltinFT->getArgType(0) != FirstArg->getType()) { 346 ImpCastExprToType(FirstArg, BuiltinFT->getArgType(0), false); 347 TheCall->setArg(0, FirstArg); 348 } 349 350 // Next, walk the valid ones promoting to the right type. 351 for (unsigned i = 0; i != NumFixed; ++i) { 352 Expr *Arg = TheCall->getArg(i+1); 353 354 // If the argument is an implicit cast, then there was a promotion due to 355 // "...", just remove it now. 356 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) { 357 Arg = ICE->getSubExpr(); 358 ICE->setSubExpr(0); 359 ICE->Destroy(Context); 360 TheCall->setArg(i+1, Arg); 361 } 362 363 // GCC does an implicit conversion to the pointer or integer ValType. This 364 // can fail in some cases (1i -> int**), check for this error case now. 365 if (CheckCastTypes(Arg->getSourceRange(), ValType, Arg)) 366 return true; 367 368 // Okay, we have something that *can* be converted to the right type. Check 369 // to see if there is a potentially weird extension going on here. This can 370 // happen when you do an atomic operation on something like an char* and 371 // pass in 42. The 42 gets converted to char. This is even more strange 372 // for things like 45.123 -> char, etc. 373 // FIXME: Do this check. 374 ImpCastExprToType(Arg, ValType, false); 375 TheCall->setArg(i+1, Arg); 376 } 377 378 // Switch the DeclRefExpr to refer to the new decl. 379 DRE->setDecl(NewBuiltinDecl); 380 DRE->setType(NewBuiltinDecl->getType()); 381 382 // Set the callee in the CallExpr. 383 // FIXME: This leaks the original parens and implicit casts. 384 Expr *PromotedCall = DRE; 385 UsualUnaryConversions(PromotedCall); 386 TheCall->setCallee(PromotedCall); 387 388 389 // Change the result type of the call to match the result type of the decl. 390 TheCall->setType(NewBuiltinDecl->getResultType()); 391 return false; 392} 393 394 395/// CheckObjCString - Checks that the argument to the builtin 396/// CFString constructor is correct 397/// FIXME: GCC currently emits the following warning: 398/// "warning: input conversion stopped due to an input byte that does not 399/// belong to the input codeset UTF-8" 400/// Note: It might also make sense to do the UTF-16 conversion here (would 401/// simplify the backend). 402bool Sema::CheckObjCString(Expr *Arg) { 403 Arg = Arg->IgnoreParenCasts(); 404 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 405 406 if (!Literal || Literal->isWide()) { 407 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 408 << Arg->getSourceRange(); 409 return true; 410 } 411 412 const char *Data = Literal->getStrData(); 413 unsigned Length = Literal->getByteLength(); 414 415 for (unsigned i = 0; i < Length; ++i) { 416 if (!Data[i]) { 417 Diag(getLocationOfStringLiteralByte(Literal, i), 418 diag::warn_cfstring_literal_contains_nul_character) 419 << Arg->getSourceRange(); 420 break; 421 } 422 } 423 424 return false; 425} 426 427/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 428/// Emit an error and return true on failure, return false on success. 429bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 430 Expr *Fn = TheCall->getCallee(); 431 if (TheCall->getNumArgs() > 2) { 432 Diag(TheCall->getArg(2)->getLocStart(), 433 diag::err_typecheck_call_too_many_args) 434 << 0 /*function call*/ << Fn->getSourceRange() 435 << SourceRange(TheCall->getArg(2)->getLocStart(), 436 (*(TheCall->arg_end()-1))->getLocEnd()); 437 return true; 438 } 439 440 if (TheCall->getNumArgs() < 2) { 441 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 442 << 0 /*function call*/; 443 } 444 445 // Determine whether the current function is variadic or not. 446 bool isVariadic; 447 if (CurBlock) 448 isVariadic = CurBlock->isVariadic; 449 else if (getCurFunctionDecl()) { 450 if (FunctionProtoType* FTP = 451 dyn_cast<FunctionProtoType>(getCurFunctionDecl()->getType())) 452 isVariadic = FTP->isVariadic(); 453 else 454 isVariadic = false; 455 } else { 456 isVariadic = getCurMethodDecl()->isVariadic(); 457 } 458 459 if (!isVariadic) { 460 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 461 return true; 462 } 463 464 // Verify that the second argument to the builtin is the last argument of the 465 // current function or method. 466 bool SecondArgIsLastNamedArgument = false; 467 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 468 469 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 470 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 471 // FIXME: This isn't correct for methods (results in bogus warning). 472 // Get the last formal in the current function. 473 const ParmVarDecl *LastArg; 474 if (CurBlock) 475 LastArg = *(CurBlock->TheDecl->param_end()-1); 476 else if (FunctionDecl *FD = getCurFunctionDecl()) 477 LastArg = *(FD->param_end()-1); 478 else 479 LastArg = *(getCurMethodDecl()->param_end()-1); 480 SecondArgIsLastNamedArgument = PV == LastArg; 481 } 482 } 483 484 if (!SecondArgIsLastNamedArgument) 485 Diag(TheCall->getArg(1)->getLocStart(), 486 diag::warn_second_parameter_of_va_start_not_last_named_argument); 487 return false; 488} 489 490/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 491/// friends. This is declared to take (...), so we have to check everything. 492bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 493 if (TheCall->getNumArgs() < 2) 494 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 495 << 0 /*function call*/; 496 if (TheCall->getNumArgs() > 2) 497 return Diag(TheCall->getArg(2)->getLocStart(), 498 diag::err_typecheck_call_too_many_args) 499 << 0 /*function call*/ 500 << SourceRange(TheCall->getArg(2)->getLocStart(), 501 (*(TheCall->arg_end()-1))->getLocEnd()); 502 503 Expr *OrigArg0 = TheCall->getArg(0); 504 Expr *OrigArg1 = TheCall->getArg(1); 505 506 // Do standard promotions between the two arguments, returning their common 507 // type. 508 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 509 510 // Make sure any conversions are pushed back into the call; this is 511 // type safe since unordered compare builtins are declared as "_Bool 512 // foo(...)". 513 TheCall->setArg(0, OrigArg0); 514 TheCall->setArg(1, OrigArg1); 515 516 if (OrigArg0->isTypeDependent() || OrigArg1->isTypeDependent()) 517 return false; 518 519 // If the common type isn't a real floating type, then the arguments were 520 // invalid for this operation. 521 if (!Res->isRealFloatingType()) 522 return Diag(OrigArg0->getLocStart(), 523 diag::err_typecheck_call_invalid_ordered_compare) 524 << OrigArg0->getType() << OrigArg1->getType() 525 << SourceRange(OrigArg0->getLocStart(), OrigArg1->getLocEnd()); 526 527 return false; 528} 529 530bool Sema::SemaBuiltinStackAddress(CallExpr *TheCall) { 531 // The signature for these builtins is exact; the only thing we need 532 // to check is that the argument is a constant. 533 SourceLocation Loc; 534 if (!TheCall->getArg(0)->isTypeDependent() && 535 !TheCall->getArg(0)->isValueDependent() && 536 !TheCall->getArg(0)->isIntegerConstantExpr(Context, &Loc)) 537 return Diag(Loc, diag::err_stack_const_level) << TheCall->getSourceRange(); 538 539 return false; 540} 541 542/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 543// This is declared to take (...), so we have to check everything. 544Action::OwningExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 545 if (TheCall->getNumArgs() < 3) 546 return ExprError(Diag(TheCall->getLocEnd(), 547 diag::err_typecheck_call_too_few_args) 548 << 0 /*function call*/ << TheCall->getSourceRange()); 549 550 unsigned numElements = std::numeric_limits<unsigned>::max(); 551 if (!TheCall->getArg(0)->isTypeDependent() && 552 !TheCall->getArg(1)->isTypeDependent()) { 553 QualType FAType = TheCall->getArg(0)->getType(); 554 QualType SAType = TheCall->getArg(1)->getType(); 555 556 if (!FAType->isVectorType() || !SAType->isVectorType()) { 557 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 558 << SourceRange(TheCall->getArg(0)->getLocStart(), 559 TheCall->getArg(1)->getLocEnd()); 560 return ExprError(); 561 } 562 563 if (Context.getCanonicalType(FAType).getUnqualifiedType() != 564 Context.getCanonicalType(SAType).getUnqualifiedType()) { 565 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 566 << SourceRange(TheCall->getArg(0)->getLocStart(), 567 TheCall->getArg(1)->getLocEnd()); 568 return ExprError(); 569 } 570 571 numElements = FAType->getAsVectorType()->getNumElements(); 572 if (TheCall->getNumArgs() != numElements+2) { 573 if (TheCall->getNumArgs() < numElements+2) 574 return ExprError(Diag(TheCall->getLocEnd(), 575 diag::err_typecheck_call_too_few_args) 576 << 0 /*function call*/ << TheCall->getSourceRange()); 577 return ExprError(Diag(TheCall->getLocEnd(), 578 diag::err_typecheck_call_too_many_args) 579 << 0 /*function call*/ << TheCall->getSourceRange()); 580 } 581 } 582 583 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 584 if (TheCall->getArg(i)->isTypeDependent() || 585 TheCall->getArg(i)->isValueDependent()) 586 continue; 587 588 llvm::APSInt Result(32); 589 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 590 return ExprError(Diag(TheCall->getLocStart(), 591 diag::err_shufflevector_nonconstant_argument) 592 << TheCall->getArg(i)->getSourceRange()); 593 594 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 595 return ExprError(Diag(TheCall->getLocStart(), 596 diag::err_shufflevector_argument_too_large) 597 << TheCall->getArg(i)->getSourceRange()); 598 } 599 600 llvm::SmallVector<Expr*, 32> exprs; 601 602 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 603 exprs.push_back(TheCall->getArg(i)); 604 TheCall->setArg(i, 0); 605 } 606 607 return Owned(new (Context) ShuffleVectorExpr(exprs.begin(), exprs.size(), 608 exprs[0]->getType(), 609 TheCall->getCallee()->getLocStart(), 610 TheCall->getRParenLoc())); 611} 612 613/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 614// This is declared to take (const void*, ...) and can take two 615// optional constant int args. 616bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 617 unsigned NumArgs = TheCall->getNumArgs(); 618 619 if (NumArgs > 3) 620 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_many_args) 621 << 0 /*function call*/ << TheCall->getSourceRange(); 622 623 // Argument 0 is checked for us and the remaining arguments must be 624 // constant integers. 625 for (unsigned i = 1; i != NumArgs; ++i) { 626 Expr *Arg = TheCall->getArg(i); 627 if (Arg->isTypeDependent()) 628 continue; 629 630 QualType RWType = Arg->getType(); 631 632 const BuiltinType *BT = RWType->getAsBuiltinType(); 633 llvm::APSInt Result; 634 if (!BT || BT->getKind() != BuiltinType::Int) 635 return Diag(TheCall->getLocStart(), diag::err_prefetch_invalid_argument) 636 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 637 638 if (Arg->isValueDependent()) 639 continue; 640 641 if (!Arg->isIntegerConstantExpr(Result, Context)) 642 return Diag(TheCall->getLocStart(), diag::err_prefetch_invalid_argument) 643 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 644 645 // FIXME: gcc issues a warning and rewrites these to 0. These 646 // seems especially odd for the third argument since the default 647 // is 3. 648 if (i == 1) { 649 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 1) 650 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 651 << "0" << "1" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 652 } else { 653 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) 654 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 655 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 656 } 657 } 658 659 return false; 660} 661 662/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 663/// int type). This simply type checks that type is one of the defined 664/// constants (0-3). 665bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 666 Expr *Arg = TheCall->getArg(1); 667 if (Arg->isTypeDependent()) 668 return false; 669 670 QualType ArgType = Arg->getType(); 671 const BuiltinType *BT = ArgType->getAsBuiltinType(); 672 llvm::APSInt Result(32); 673 if (!BT || BT->getKind() != BuiltinType::Int) 674 return Diag(TheCall->getLocStart(), diag::err_object_size_invalid_argument) 675 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 676 677 if (Arg->isValueDependent()) 678 return false; 679 680 if (!Arg->isIntegerConstantExpr(Result, Context)) { 681 return Diag(TheCall->getLocStart(), diag::err_object_size_invalid_argument) 682 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 683 } 684 685 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 686 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 687 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 688 } 689 690 return false; 691} 692 693/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 694/// This checks that val is a constant 1. 695bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 696 Expr *Arg = TheCall->getArg(1); 697 if (Arg->isTypeDependent() || Arg->isValueDependent()) 698 return false; 699 700 llvm::APSInt Result(32); 701 if (!Arg->isIntegerConstantExpr(Result, Context) || Result != 1) 702 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 703 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 704 705 return false; 706} 707 708// Handle i > 1 ? "x" : "y", recursivelly 709bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 710 bool HasVAListArg, 711 unsigned format_idx, unsigned firstDataArg) { 712 if (E->isTypeDependent() || E->isValueDependent()) 713 return false; 714 715 switch (E->getStmtClass()) { 716 case Stmt::ConditionalOperatorClass: { 717 const ConditionalOperator *C = cast<ConditionalOperator>(E); 718 return SemaCheckStringLiteral(C->getLHS(), TheCall, 719 HasVAListArg, format_idx, firstDataArg) 720 && SemaCheckStringLiteral(C->getRHS(), TheCall, 721 HasVAListArg, format_idx, firstDataArg); 722 } 723 724 case Stmt::ImplicitCastExprClass: { 725 const ImplicitCastExpr *Expr = cast<ImplicitCastExpr>(E); 726 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 727 format_idx, firstDataArg); 728 } 729 730 case Stmt::ParenExprClass: { 731 const ParenExpr *Expr = cast<ParenExpr>(E); 732 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 733 format_idx, firstDataArg); 734 } 735 736 case Stmt::DeclRefExprClass: { 737 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 738 739 // As an exception, do not flag errors for variables binding to 740 // const string literals. 741 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 742 bool isConstant = false; 743 QualType T = DR->getType(); 744 745 if (const ArrayType *AT = Context.getAsArrayType(T)) { 746 isConstant = AT->getElementType().isConstant(Context); 747 } 748 else if (const PointerType *PT = T->getAsPointerType()) { 749 isConstant = T.isConstant(Context) && 750 PT->getPointeeType().isConstant(Context); 751 } 752 753 if (isConstant) { 754 const VarDecl *Def = 0; 755 if (const Expr *Init = VD->getDefinition(Def)) 756 return SemaCheckStringLiteral(Init, TheCall, 757 HasVAListArg, format_idx, firstDataArg); 758 } 759 } 760 761 return false; 762 } 763 764 case Stmt::ObjCStringLiteralClass: 765 case Stmt::StringLiteralClass: { 766 const StringLiteral *StrE = NULL; 767 768 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 769 StrE = ObjCFExpr->getString(); 770 else 771 StrE = cast<StringLiteral>(E); 772 773 if (StrE) { 774 CheckPrintfString(StrE, E, TheCall, HasVAListArg, format_idx, 775 firstDataArg); 776 return true; 777 } 778 779 return false; 780 } 781 782 default: 783 return false; 784 } 785} 786 787 788/// CheckPrintfArguments - Check calls to printf (and similar functions) for 789/// correct use of format strings. 790/// 791/// HasVAListArg - A predicate indicating whether the printf-like 792/// function is passed an explicit va_arg argument (e.g., vprintf) 793/// 794/// format_idx - The index into Args for the format string. 795/// 796/// Improper format strings to functions in the printf family can be 797/// the source of bizarre bugs and very serious security holes. A 798/// good source of information is available in the following paper 799/// (which includes additional references): 800/// 801/// FormatGuard: Automatic Protection From printf Format String 802/// Vulnerabilities, Proceedings of the 10th USENIX Security Symposium, 2001. 803/// 804/// Functionality implemented: 805/// 806/// We can statically check the following properties for string 807/// literal format strings for non v.*printf functions (where the 808/// arguments are passed directly): 809// 810/// (1) Are the number of format conversions equal to the number of 811/// data arguments? 812/// 813/// (2) Does each format conversion correctly match the type of the 814/// corresponding data argument? (TODO) 815/// 816/// Moreover, for all printf functions we can: 817/// 818/// (3) Check for a missing format string (when not caught by type checking). 819/// 820/// (4) Check for no-operation flags; e.g. using "#" with format 821/// conversion 'c' (TODO) 822/// 823/// (5) Check the use of '%n', a major source of security holes. 824/// 825/// (6) Check for malformed format conversions that don't specify anything. 826/// 827/// (7) Check for empty format strings. e.g: printf(""); 828/// 829/// (8) Check that the format string is a wide literal. 830/// 831/// (9) Also check the arguments of functions with the __format__ attribute. 832/// (TODO). 833/// 834/// All of these checks can be done by parsing the format string. 835/// 836/// For now, we ONLY do (1), (3), (5), (6), (7), and (8). 837void 838Sema::CheckPrintfArguments(const CallExpr *TheCall, bool HasVAListArg, 839 unsigned format_idx, unsigned firstDataArg) { 840 const Expr *Fn = TheCall->getCallee(); 841 842 // CHECK: printf-like function is called with no format string. 843 if (format_idx >= TheCall->getNumArgs()) { 844 Diag(TheCall->getRParenLoc(), diag::warn_printf_missing_format_string) 845 << Fn->getSourceRange(); 846 return; 847 } 848 849 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 850 851 // CHECK: format string is not a string literal. 852 // 853 // Dynamically generated format strings are difficult to 854 // automatically vet at compile time. Requiring that format strings 855 // are string literals: (1) permits the checking of format strings by 856 // the compiler and thereby (2) can practically remove the source of 857 // many format string exploits. 858 859 // Format string can be either ObjC string (e.g. @"%d") or 860 // C string (e.g. "%d") 861 // ObjC string uses the same format specifiers as C string, so we can use 862 // the same format string checking logic for both ObjC and C strings. 863 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 864 firstDataArg)) 865 return; // Literal format string found, check done! 866 867 // For vprintf* functions (i.e., HasVAListArg==true), we add a 868 // special check to see if the format string is a function parameter 869 // of the function calling the printf function. If the function 870 // has an attribute indicating it is a printf-like function, then we 871 // should suppress warnings concerning non-literals being used in a call 872 // to a vprintf function. For example: 873 // 874 // void 875 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...) { 876 // va_list ap; 877 // va_start(ap, fmt); 878 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 879 // ... 880 // 881 // 882 // FIXME: We don't have full attribute support yet, so just check to see 883 // if the argument is a DeclRefExpr that references a parameter. We'll 884 // add proper support for checking the attribute later. 885 if (HasVAListArg) 886 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(OrigFormatExpr)) 887 if (isa<ParmVarDecl>(DR->getDecl())) 888 return; 889 890 // If there are no arguments specified, warn with -Wformat-security, otherwise 891 // warn only with -Wformat-nonliteral. 892 if (TheCall->getNumArgs() == format_idx+1) 893 Diag(TheCall->getArg(format_idx)->getLocStart(), 894 diag::warn_printf_nonliteral_noargs) 895 << OrigFormatExpr->getSourceRange(); 896 else 897 Diag(TheCall->getArg(format_idx)->getLocStart(), 898 diag::warn_printf_nonliteral) 899 << OrigFormatExpr->getSourceRange(); 900} 901 902void Sema::CheckPrintfString(const StringLiteral *FExpr, 903 const Expr *OrigFormatExpr, 904 const CallExpr *TheCall, bool HasVAListArg, 905 unsigned format_idx, unsigned firstDataArg) { 906 907 const ObjCStringLiteral *ObjCFExpr = 908 dyn_cast<ObjCStringLiteral>(OrigFormatExpr); 909 910 // CHECK: is the format string a wide literal? 911 if (FExpr->isWide()) { 912 Diag(FExpr->getLocStart(), 913 diag::warn_printf_format_string_is_wide_literal) 914 << OrigFormatExpr->getSourceRange(); 915 return; 916 } 917 918 // Str - The format string. NOTE: this is NOT null-terminated! 919 const char *Str = FExpr->getStrData(); 920 921 // CHECK: empty format string? 922 unsigned StrLen = FExpr->getByteLength(); 923 924 if (StrLen == 0) { 925 Diag(FExpr->getLocStart(), diag::warn_printf_empty_format_string) 926 << OrigFormatExpr->getSourceRange(); 927 return; 928 } 929 930 // We process the format string using a binary state machine. The 931 // current state is stored in CurrentState. 932 enum { 933 state_OrdChr, 934 state_Conversion 935 } CurrentState = state_OrdChr; 936 937 // numConversions - The number of conversions seen so far. This is 938 // incremented as we traverse the format string. 939 unsigned numConversions = 0; 940 941 // numDataArgs - The number of data arguments after the format 942 // string. This can only be determined for non vprintf-like 943 // functions. For those functions, this value is 1 (the sole 944 // va_arg argument). 945 unsigned numDataArgs = TheCall->getNumArgs()-firstDataArg; 946 947 // Inspect the format string. 948 unsigned StrIdx = 0; 949 950 // LastConversionIdx - Index within the format string where we last saw 951 // a '%' character that starts a new format conversion. 952 unsigned LastConversionIdx = 0; 953 954 for (; StrIdx < StrLen; ++StrIdx) { 955 956 // Is the number of detected conversion conversions greater than 957 // the number of matching data arguments? If so, stop. 958 if (!HasVAListArg && numConversions > numDataArgs) break; 959 960 // Handle "\0" 961 if (Str[StrIdx] == '\0') { 962 // The string returned by getStrData() is not null-terminated, 963 // so the presence of a null character is likely an error. 964 Diag(getLocationOfStringLiteralByte(FExpr, StrIdx), 965 diag::warn_printf_format_string_contains_null_char) 966 << OrigFormatExpr->getSourceRange(); 967 return; 968 } 969 970 // Ordinary characters (not processing a format conversion). 971 if (CurrentState == state_OrdChr) { 972 if (Str[StrIdx] == '%') { 973 CurrentState = state_Conversion; 974 LastConversionIdx = StrIdx; 975 } 976 continue; 977 } 978 979 // Seen '%'. Now processing a format conversion. 980 switch (Str[StrIdx]) { 981 // Handle dynamic precision or width specifier. 982 case '*': { 983 ++numConversions; 984 985 if (!HasVAListArg) { 986 if (numConversions > numDataArgs) { 987 SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, StrIdx); 988 989 if (Str[StrIdx-1] == '.') 990 Diag(Loc, diag::warn_printf_asterisk_precision_missing_arg) 991 << OrigFormatExpr->getSourceRange(); 992 else 993 Diag(Loc, diag::warn_printf_asterisk_width_missing_arg) 994 << OrigFormatExpr->getSourceRange(); 995 996 // Don't do any more checking. We'll just emit spurious errors. 997 return; 998 } 999 1000 // Perform type checking on width/precision specifier. 1001 const Expr *E = TheCall->getArg(format_idx+numConversions); 1002 if (const BuiltinType *BT = E->getType()->getAsBuiltinType()) 1003 if (BT->getKind() == BuiltinType::Int) 1004 break; 1005 1006 SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, StrIdx); 1007 1008 if (Str[StrIdx-1] == '.') 1009 Diag(Loc, diag::warn_printf_asterisk_precision_wrong_type) 1010 << E->getType() << E->getSourceRange(); 1011 else 1012 Diag(Loc, diag::warn_printf_asterisk_width_wrong_type) 1013 << E->getType() << E->getSourceRange(); 1014 1015 break; 1016 } 1017 } 1018 1019 // Characters which can terminate a format conversion 1020 // (e.g. "%d"). Characters that specify length modifiers or 1021 // other flags are handled by the default case below. 1022 // 1023 // FIXME: additional checks will go into the following cases. 1024 case 'i': 1025 case 'd': 1026 case 'o': 1027 case 'u': 1028 case 'x': 1029 case 'X': 1030 case 'D': 1031 case 'O': 1032 case 'U': 1033 case 'e': 1034 case 'E': 1035 case 'f': 1036 case 'F': 1037 case 'g': 1038 case 'G': 1039 case 'a': 1040 case 'A': 1041 case 'c': 1042 case 'C': 1043 case 'S': 1044 case 's': 1045 case 'p': 1046 ++numConversions; 1047 CurrentState = state_OrdChr; 1048 break; 1049 1050 // CHECK: Are we using "%n"? Issue a warning. 1051 case 'n': { 1052 ++numConversions; 1053 CurrentState = state_OrdChr; 1054 SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, 1055 LastConversionIdx); 1056 1057 Diag(Loc, diag::warn_printf_write_back)<<OrigFormatExpr->getSourceRange(); 1058 break; 1059 } 1060 1061 // Handle "%@" 1062 case '@': 1063 // %@ is allowed in ObjC format strings only. 1064 if(ObjCFExpr != NULL) 1065 CurrentState = state_OrdChr; 1066 else { 1067 // Issue a warning: invalid format conversion. 1068 SourceLocation Loc = 1069 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 1070 1071 Diag(Loc, diag::warn_printf_invalid_conversion) 1072 << std::string(Str+LastConversionIdx, 1073 Str+std::min(LastConversionIdx+2, StrLen)) 1074 << OrigFormatExpr->getSourceRange(); 1075 } 1076 ++numConversions; 1077 break; 1078 1079 // Handle "%%" 1080 case '%': 1081 // Sanity check: Was the first "%" character the previous one? 1082 // If not, we will assume that we have a malformed format 1083 // conversion, and that the current "%" character is the start 1084 // of a new conversion. 1085 if (StrIdx - LastConversionIdx == 1) 1086 CurrentState = state_OrdChr; 1087 else { 1088 // Issue a warning: invalid format conversion. 1089 SourceLocation Loc = 1090 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 1091 1092 Diag(Loc, diag::warn_printf_invalid_conversion) 1093 << std::string(Str+LastConversionIdx, Str+StrIdx) 1094 << OrigFormatExpr->getSourceRange(); 1095 1096 // This conversion is broken. Advance to the next format 1097 // conversion. 1098 LastConversionIdx = StrIdx; 1099 ++numConversions; 1100 } 1101 break; 1102 1103 default: 1104 // This case catches all other characters: flags, widths, etc. 1105 // We should eventually process those as well. 1106 break; 1107 } 1108 } 1109 1110 if (CurrentState == state_Conversion) { 1111 // Issue a warning: invalid format conversion. 1112 SourceLocation Loc = 1113 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 1114 1115 Diag(Loc, diag::warn_printf_invalid_conversion) 1116 << std::string(Str+LastConversionIdx, 1117 Str+std::min(LastConversionIdx+2, StrLen)) 1118 << OrigFormatExpr->getSourceRange(); 1119 return; 1120 } 1121 1122 if (!HasVAListArg) { 1123 // CHECK: Does the number of format conversions exceed the number 1124 // of data arguments? 1125 if (numConversions > numDataArgs) { 1126 SourceLocation Loc = 1127 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 1128 1129 Diag(Loc, diag::warn_printf_insufficient_data_args) 1130 << OrigFormatExpr->getSourceRange(); 1131 } 1132 // CHECK: Does the number of data arguments exceed the number of 1133 // format conversions in the format string? 1134 else if (numConversions < numDataArgs) 1135 Diag(TheCall->getArg(format_idx+numConversions+1)->getLocStart(), 1136 diag::warn_printf_too_many_data_args) 1137 << OrigFormatExpr->getSourceRange(); 1138 } 1139} 1140 1141//===--- CHECK: Return Address of Stack Variable --------------------------===// 1142 1143static DeclRefExpr* EvalVal(Expr *E); 1144static DeclRefExpr* EvalAddr(Expr* E); 1145 1146/// CheckReturnStackAddr - Check if a return statement returns the address 1147/// of a stack variable. 1148void 1149Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 1150 SourceLocation ReturnLoc) { 1151 1152 // Perform checking for returned stack addresses. 1153 if (lhsType->isPointerType() || lhsType->isBlockPointerType()) { 1154 if (DeclRefExpr *DR = EvalAddr(RetValExp)) 1155 Diag(DR->getLocStart(), diag::warn_ret_stack_addr) 1156 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1157 1158 // Skip over implicit cast expressions when checking for block expressions. 1159 if (ImplicitCastExpr *IcExpr = 1160 dyn_cast_or_null<ImplicitCastExpr>(RetValExp)) 1161 RetValExp = IcExpr->getSubExpr(); 1162 1163 if (BlockExpr *C = dyn_cast_or_null<BlockExpr>(RetValExp)) 1164 if (C->hasBlockDeclRefExprs()) 1165 Diag(C->getLocStart(), diag::err_ret_local_block) 1166 << C->getSourceRange(); 1167 } 1168 // Perform checking for stack values returned by reference. 1169 else if (lhsType->isReferenceType()) { 1170 // Check for a reference to the stack 1171 if (DeclRefExpr *DR = EvalVal(RetValExp)) 1172 Diag(DR->getLocStart(), diag::warn_ret_stack_ref) 1173 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1174 } 1175} 1176 1177/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 1178/// check if the expression in a return statement evaluates to an address 1179/// to a location on the stack. The recursion is used to traverse the 1180/// AST of the return expression, with recursion backtracking when we 1181/// encounter a subexpression that (1) clearly does not lead to the address 1182/// of a stack variable or (2) is something we cannot determine leads to 1183/// the address of a stack variable based on such local checking. 1184/// 1185/// EvalAddr processes expressions that are pointers that are used as 1186/// references (and not L-values). EvalVal handles all other values. 1187/// At the base case of the recursion is a check for a DeclRefExpr* in 1188/// the refers to a stack variable. 1189/// 1190/// This implementation handles: 1191/// 1192/// * pointer-to-pointer casts 1193/// * implicit conversions from array references to pointers 1194/// * taking the address of fields 1195/// * arbitrary interplay between "&" and "*" operators 1196/// * pointer arithmetic from an address of a stack variable 1197/// * taking the address of an array element where the array is on the stack 1198static DeclRefExpr* EvalAddr(Expr *E) { 1199 // We should only be called for evaluating pointer expressions. 1200 assert((E->getType()->isPointerType() || 1201 E->getType()->isBlockPointerType() || 1202 E->getType()->isObjCQualifiedIdType()) && 1203 "EvalAddr only works on pointers"); 1204 1205 // Our "symbolic interpreter" is just a dispatch off the currently 1206 // viewed AST node. We then recursively traverse the AST by calling 1207 // EvalAddr and EvalVal appropriately. 1208 switch (E->getStmtClass()) { 1209 case Stmt::ParenExprClass: 1210 // Ignore parentheses. 1211 return EvalAddr(cast<ParenExpr>(E)->getSubExpr()); 1212 1213 case Stmt::UnaryOperatorClass: { 1214 // The only unary operator that make sense to handle here 1215 // is AddrOf. All others don't make sense as pointers. 1216 UnaryOperator *U = cast<UnaryOperator>(E); 1217 1218 if (U->getOpcode() == UnaryOperator::AddrOf) 1219 return EvalVal(U->getSubExpr()); 1220 else 1221 return NULL; 1222 } 1223 1224 case Stmt::BinaryOperatorClass: { 1225 // Handle pointer arithmetic. All other binary operators are not valid 1226 // in this context. 1227 BinaryOperator *B = cast<BinaryOperator>(E); 1228 BinaryOperator::Opcode op = B->getOpcode(); 1229 1230 if (op != BinaryOperator::Add && op != BinaryOperator::Sub) 1231 return NULL; 1232 1233 Expr *Base = B->getLHS(); 1234 1235 // Determine which argument is the real pointer base. It could be 1236 // the RHS argument instead of the LHS. 1237 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 1238 1239 assert (Base->getType()->isPointerType()); 1240 return EvalAddr(Base); 1241 } 1242 1243 // For conditional operators we need to see if either the LHS or RHS are 1244 // valid DeclRefExpr*s. If one of them is valid, we return it. 1245 case Stmt::ConditionalOperatorClass: { 1246 ConditionalOperator *C = cast<ConditionalOperator>(E); 1247 1248 // Handle the GNU extension for missing LHS. 1249 if (Expr *lhsExpr = C->getLHS()) 1250 if (DeclRefExpr* LHS = EvalAddr(lhsExpr)) 1251 return LHS; 1252 1253 return EvalAddr(C->getRHS()); 1254 } 1255 1256 // For casts, we need to handle conversions from arrays to 1257 // pointer values, and pointer-to-pointer conversions. 1258 case Stmt::ImplicitCastExprClass: 1259 case Stmt::CStyleCastExprClass: 1260 case Stmt::CXXFunctionalCastExprClass: { 1261 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 1262 QualType T = SubExpr->getType(); 1263 1264 if (SubExpr->getType()->isPointerType() || 1265 SubExpr->getType()->isBlockPointerType() || 1266 SubExpr->getType()->isObjCQualifiedIdType()) 1267 return EvalAddr(SubExpr); 1268 else if (T->isArrayType()) 1269 return EvalVal(SubExpr); 1270 else 1271 return 0; 1272 } 1273 1274 // C++ casts. For dynamic casts, static casts, and const casts, we 1275 // are always converting from a pointer-to-pointer, so we just blow 1276 // through the cast. In the case the dynamic cast doesn't fail (and 1277 // return NULL), we take the conservative route and report cases 1278 // where we return the address of a stack variable. For Reinterpre 1279 // FIXME: The comment about is wrong; we're not always converting 1280 // from pointer to pointer. I'm guessing that this code should also 1281 // handle references to objects. 1282 case Stmt::CXXStaticCastExprClass: 1283 case Stmt::CXXDynamicCastExprClass: 1284 case Stmt::CXXConstCastExprClass: 1285 case Stmt::CXXReinterpretCastExprClass: { 1286 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 1287 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 1288 return EvalAddr(S); 1289 else 1290 return NULL; 1291 } 1292 1293 // Everything else: we simply don't reason about them. 1294 default: 1295 return NULL; 1296 } 1297} 1298 1299 1300/// EvalVal - This function is complements EvalAddr in the mutual recursion. 1301/// See the comments for EvalAddr for more details. 1302static DeclRefExpr* EvalVal(Expr *E) { 1303 1304 // We should only be called for evaluating non-pointer expressions, or 1305 // expressions with a pointer type that are not used as references but instead 1306 // are l-values (e.g., DeclRefExpr with a pointer type). 1307 1308 // Our "symbolic interpreter" is just a dispatch off the currently 1309 // viewed AST node. We then recursively traverse the AST by calling 1310 // EvalAddr and EvalVal appropriately. 1311 switch (E->getStmtClass()) { 1312 case Stmt::DeclRefExprClass: 1313 case Stmt::QualifiedDeclRefExprClass: { 1314 // DeclRefExpr: the base case. When we hit a DeclRefExpr we are looking 1315 // at code that refers to a variable's name. We check if it has local 1316 // storage within the function, and if so, return the expression. 1317 DeclRefExpr *DR = cast<DeclRefExpr>(E); 1318 1319 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 1320 if(V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR; 1321 1322 return NULL; 1323 } 1324 1325 case Stmt::ParenExprClass: 1326 // Ignore parentheses. 1327 return EvalVal(cast<ParenExpr>(E)->getSubExpr()); 1328 1329 case Stmt::UnaryOperatorClass: { 1330 // The only unary operator that make sense to handle here 1331 // is Deref. All others don't resolve to a "name." This includes 1332 // handling all sorts of rvalues passed to a unary operator. 1333 UnaryOperator *U = cast<UnaryOperator>(E); 1334 1335 if (U->getOpcode() == UnaryOperator::Deref) 1336 return EvalAddr(U->getSubExpr()); 1337 1338 return NULL; 1339 } 1340 1341 case Stmt::ArraySubscriptExprClass: { 1342 // Array subscripts are potential references to data on the stack. We 1343 // retrieve the DeclRefExpr* for the array variable if it indeed 1344 // has local storage. 1345 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase()); 1346 } 1347 1348 case Stmt::ConditionalOperatorClass: { 1349 // For conditional operators we need to see if either the LHS or RHS are 1350 // non-NULL DeclRefExpr's. If one is non-NULL, we return it. 1351 ConditionalOperator *C = cast<ConditionalOperator>(E); 1352 1353 // Handle the GNU extension for missing LHS. 1354 if (Expr *lhsExpr = C->getLHS()) 1355 if (DeclRefExpr *LHS = EvalVal(lhsExpr)) 1356 return LHS; 1357 1358 return EvalVal(C->getRHS()); 1359 } 1360 1361 // Accesses to members are potential references to data on the stack. 1362 case Stmt::MemberExprClass: { 1363 MemberExpr *M = cast<MemberExpr>(E); 1364 1365 // Check for indirect access. We only want direct field accesses. 1366 if (!M->isArrow()) 1367 return EvalVal(M->getBase()); 1368 else 1369 return NULL; 1370 } 1371 1372 // Everything else: we simply don't reason about them. 1373 default: 1374 return NULL; 1375 } 1376} 1377 1378//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 1379 1380/// Check for comparisons of floating point operands using != and ==. 1381/// Issue a warning if these are no self-comparisons, as they are not likely 1382/// to do what the programmer intended. 1383void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 1384 bool EmitWarning = true; 1385 1386 Expr* LeftExprSansParen = lex->IgnoreParens(); 1387 Expr* RightExprSansParen = rex->IgnoreParens(); 1388 1389 // Special case: check for x == x (which is OK). 1390 // Do not emit warnings for such cases. 1391 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 1392 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 1393 if (DRL->getDecl() == DRR->getDecl()) 1394 EmitWarning = false; 1395 1396 1397 // Special case: check for comparisons against literals that can be exactly 1398 // represented by APFloat. In such cases, do not emit a warning. This 1399 // is a heuristic: often comparison against such literals are used to 1400 // detect if a value in a variable has not changed. This clearly can 1401 // lead to false negatives. 1402 if (EmitWarning) { 1403 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 1404 if (FLL->isExact()) 1405 EmitWarning = false; 1406 } 1407 else 1408 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 1409 if (FLR->isExact()) 1410 EmitWarning = false; 1411 } 1412 } 1413 1414 // Check for comparisons with builtin types. 1415 if (EmitWarning) 1416 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 1417 if (CL->isBuiltinCall(Context)) 1418 EmitWarning = false; 1419 1420 if (EmitWarning) 1421 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 1422 if (CR->isBuiltinCall(Context)) 1423 EmitWarning = false; 1424 1425 // Emit the diagnostic. 1426 if (EmitWarning) 1427 Diag(loc, diag::warn_floatingpoint_eq) 1428 << lex->getSourceRange() << rex->getSourceRange(); 1429} 1430