SemaChecking.cpp revision 0979c805475d1ba49b5d6ef93c4d2ce6d2eab6ed
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/// CheckablePrintfAttr - does a function call have a "printf" attribute 88/// and arguments that merit checking? 89bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) { 90 if (Format->getType() == "printf") return true; 91 if (Format->getType() == "printf0") { 92 // printf0 allows null "format" string; if so don't check format/args 93 unsigned format_idx = Format->getFormatIdx() - 1; 94 if (format_idx < TheCall->getNumArgs()) { 95 Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts(); 96 if (!Format->isNullPointerConstant(Context)) 97 return true; 98 } 99 } 100 return false; 101} 102 103Action::OwningExprResult 104Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 105 OwningExprResult TheCallResult(Owned(TheCall)); 106 107 switch (BuiltinID) { 108 case Builtin::BI__builtin___CFStringMakeConstantString: 109 assert(TheCall->getNumArgs() == 1 && 110 "Wrong # arguments to builtin CFStringMakeConstantString"); 111 if (CheckObjCString(TheCall->getArg(0))) 112 return ExprError(); 113 break; 114 case Builtin::BI__builtin_stdarg_start: 115 case Builtin::BI__builtin_va_start: 116 if (SemaBuiltinVAStart(TheCall)) 117 return ExprError(); 118 break; 119 case Builtin::BI__builtin_isgreater: 120 case Builtin::BI__builtin_isgreaterequal: 121 case Builtin::BI__builtin_isless: 122 case Builtin::BI__builtin_islessequal: 123 case Builtin::BI__builtin_islessgreater: 124 case Builtin::BI__builtin_isunordered: 125 if (SemaBuiltinUnorderedCompare(TheCall)) 126 return ExprError(); 127 break; 128 case Builtin::BI__builtin_isfinite: 129 case Builtin::BI__builtin_isinf: 130 case Builtin::BI__builtin_isinf_sign: 131 case Builtin::BI__builtin_isnan: 132 case Builtin::BI__builtin_isnormal: 133 if (SemaBuiltinUnaryFP(TheCall)) 134 return ExprError(); 135 break; 136 case Builtin::BI__builtin_return_address: 137 case Builtin::BI__builtin_frame_address: 138 if (SemaBuiltinStackAddress(TheCall)) 139 return ExprError(); 140 break; 141 case Builtin::BI__builtin_shufflevector: 142 return SemaBuiltinShuffleVector(TheCall); 143 // TheCall will be freed by the smart pointer here, but that's fine, since 144 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 145 case Builtin::BI__builtin_prefetch: 146 if (SemaBuiltinPrefetch(TheCall)) 147 return ExprError(); 148 break; 149 case Builtin::BI__builtin_object_size: 150 if (SemaBuiltinObjectSize(TheCall)) 151 return ExprError(); 152 break; 153 case Builtin::BI__builtin_longjmp: 154 if (SemaBuiltinLongjmp(TheCall)) 155 return ExprError(); 156 break; 157 case Builtin::BI__sync_fetch_and_add: 158 case Builtin::BI__sync_fetch_and_sub: 159 case Builtin::BI__sync_fetch_and_or: 160 case Builtin::BI__sync_fetch_and_and: 161 case Builtin::BI__sync_fetch_and_xor: 162 case Builtin::BI__sync_fetch_and_nand: 163 case Builtin::BI__sync_add_and_fetch: 164 case Builtin::BI__sync_sub_and_fetch: 165 case Builtin::BI__sync_and_and_fetch: 166 case Builtin::BI__sync_or_and_fetch: 167 case Builtin::BI__sync_xor_and_fetch: 168 case Builtin::BI__sync_nand_and_fetch: 169 case Builtin::BI__sync_val_compare_and_swap: 170 case Builtin::BI__sync_bool_compare_and_swap: 171 case Builtin::BI__sync_lock_test_and_set: 172 case Builtin::BI__sync_lock_release: 173 if (SemaBuiltinAtomicOverloaded(TheCall)) 174 return ExprError(); 175 break; 176 } 177 178 return move(TheCallResult); 179} 180 181/// CheckFunctionCall - Check a direct function call for various correctness 182/// and safety properties not strictly enforced by the C type system. 183bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 184 // Get the IdentifierInfo* for the called function. 185 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 186 187 // None of the checks below are needed for functions that don't have 188 // simple names (e.g., C++ conversion functions). 189 if (!FnInfo) 190 return false; 191 192 // FIXME: This mechanism should be abstracted to be less fragile and 193 // more efficient. For example, just map function ids to custom 194 // handlers. 195 196 // Printf checking. 197 if (const FormatAttr *Format = FDecl->getAttr<FormatAttr>()) { 198 if (CheckablePrintfAttr(Format, TheCall)) { 199 bool HasVAListArg = Format->getFirstArg() == 0; 200 if (!HasVAListArg) { 201 if (const FunctionProtoType *Proto 202 = FDecl->getType()->getAsFunctionProtoType()) 203 HasVAListArg = !Proto->isVariadic(); 204 } 205 CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 206 HasVAListArg ? 0 : Format->getFirstArg() - 1); 207 } 208 } 209 210 for (const NonNullAttr *NonNull = FDecl->getAttr<NonNullAttr>(); NonNull; 211 NonNull = NonNull->getNext<NonNullAttr>()) 212 CheckNonNullArguments(NonNull, TheCall); 213 214 return false; 215} 216 217bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 218 // Printf checking. 219 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 220 if (!Format) 221 return false; 222 223 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 224 if (!V) 225 return false; 226 227 QualType Ty = V->getType(); 228 if (!Ty->isBlockPointerType()) 229 return false; 230 231 if (!CheckablePrintfAttr(Format, TheCall)) 232 return false; 233 234 bool HasVAListArg = Format->getFirstArg() == 0; 235 if (!HasVAListArg) { 236 const FunctionType *FT = 237 Ty->getAs<BlockPointerType>()->getPointeeType()->getAsFunctionType(); 238 if (const FunctionProtoType *Proto = dyn_cast<FunctionProtoType>(FT)) 239 HasVAListArg = !Proto->isVariadic(); 240 } 241 CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 242 HasVAListArg ? 0 : Format->getFirstArg() - 1); 243 244 return false; 245} 246 247/// SemaBuiltinAtomicOverloaded - We have a call to a function like 248/// __sync_fetch_and_add, which is an overloaded function based on the pointer 249/// type of its first argument. The main ActOnCallExpr routines have already 250/// promoted the types of arguments because all of these calls are prototyped as 251/// void(...). 252/// 253/// This function goes through and does final semantic checking for these 254/// builtins, 255bool Sema::SemaBuiltinAtomicOverloaded(CallExpr *TheCall) { 256 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 257 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 258 259 // Ensure that we have at least one argument to do type inference from. 260 if (TheCall->getNumArgs() < 1) 261 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 262 << 0 << TheCall->getCallee()->getSourceRange(); 263 264 // Inspect the first argument of the atomic builtin. This should always be 265 // a pointer type, whose element is an integral scalar or pointer type. 266 // Because it is a pointer type, we don't have to worry about any implicit 267 // casts here. 268 Expr *FirstArg = TheCall->getArg(0); 269 if (!FirstArg->getType()->isPointerType()) 270 return Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 271 << FirstArg->getType() << FirstArg->getSourceRange(); 272 273 QualType ValType = FirstArg->getType()->getAs<PointerType>()->getPointeeType(); 274 if (!ValType->isIntegerType() && !ValType->isPointerType() && 275 !ValType->isBlockPointerType()) 276 return Diag(DRE->getLocStart(), 277 diag::err_atomic_builtin_must_be_pointer_intptr) 278 << FirstArg->getType() << FirstArg->getSourceRange(); 279 280 // We need to figure out which concrete builtin this maps onto. For example, 281 // __sync_fetch_and_add with a 2 byte object turns into 282 // __sync_fetch_and_add_2. 283#define BUILTIN_ROW(x) \ 284 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 285 Builtin::BI##x##_8, Builtin::BI##x##_16 } 286 287 static const unsigned BuiltinIndices[][5] = { 288 BUILTIN_ROW(__sync_fetch_and_add), 289 BUILTIN_ROW(__sync_fetch_and_sub), 290 BUILTIN_ROW(__sync_fetch_and_or), 291 BUILTIN_ROW(__sync_fetch_and_and), 292 BUILTIN_ROW(__sync_fetch_and_xor), 293 BUILTIN_ROW(__sync_fetch_and_nand), 294 295 BUILTIN_ROW(__sync_add_and_fetch), 296 BUILTIN_ROW(__sync_sub_and_fetch), 297 BUILTIN_ROW(__sync_and_and_fetch), 298 BUILTIN_ROW(__sync_or_and_fetch), 299 BUILTIN_ROW(__sync_xor_and_fetch), 300 BUILTIN_ROW(__sync_nand_and_fetch), 301 302 BUILTIN_ROW(__sync_val_compare_and_swap), 303 BUILTIN_ROW(__sync_bool_compare_and_swap), 304 BUILTIN_ROW(__sync_lock_test_and_set), 305 BUILTIN_ROW(__sync_lock_release) 306 }; 307#undef BUILTIN_ROW 308 309 // Determine the index of the size. 310 unsigned SizeIndex; 311 switch (Context.getTypeSize(ValType)/8) { 312 case 1: SizeIndex = 0; break; 313 case 2: SizeIndex = 1; break; 314 case 4: SizeIndex = 2; break; 315 case 8: SizeIndex = 3; break; 316 case 16: SizeIndex = 4; break; 317 default: 318 return Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 319 << FirstArg->getType() << FirstArg->getSourceRange(); 320 } 321 322 // Each of these builtins has one pointer argument, followed by some number of 323 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 324 // that we ignore. Find out which row of BuiltinIndices to read from as well 325 // as the number of fixed args. 326 unsigned BuiltinID = FDecl->getBuiltinID(Context); 327 unsigned BuiltinIndex, NumFixed = 1; 328 switch (BuiltinID) { 329 default: assert(0 && "Unknown overloaded atomic builtin!"); 330 case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break; 331 case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break; 332 case Builtin::BI__sync_fetch_and_or: BuiltinIndex = 2; break; 333 case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break; 334 case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break; 335 case Builtin::BI__sync_fetch_and_nand:BuiltinIndex = 5; break; 336 337 case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 6; break; 338 case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 7; break; 339 case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 8; break; 340 case Builtin::BI__sync_or_and_fetch: BuiltinIndex = 9; break; 341 case Builtin::BI__sync_xor_and_fetch: BuiltinIndex =10; break; 342 case Builtin::BI__sync_nand_and_fetch:BuiltinIndex =11; break; 343 344 case Builtin::BI__sync_val_compare_and_swap: 345 BuiltinIndex = 12; 346 NumFixed = 2; 347 break; 348 case Builtin::BI__sync_bool_compare_and_swap: 349 BuiltinIndex = 13; 350 NumFixed = 2; 351 break; 352 case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 14; break; 353 case Builtin::BI__sync_lock_release: 354 BuiltinIndex = 15; 355 NumFixed = 0; 356 break; 357 } 358 359 // Now that we know how many fixed arguments we expect, first check that we 360 // have at least that many. 361 if (TheCall->getNumArgs() < 1+NumFixed) 362 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 363 << 0 << TheCall->getCallee()->getSourceRange(); 364 365 366 // Get the decl for the concrete builtin from this, we can tell what the 367 // concrete integer type we should convert to is. 368 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 369 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 370 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 371 FunctionDecl *NewBuiltinDecl = 372 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 373 TUScope, false, DRE->getLocStart())); 374 const FunctionProtoType *BuiltinFT = 375 NewBuiltinDecl->getType()->getAsFunctionProtoType(); 376 ValType = BuiltinFT->getArgType(0)->getAs<PointerType>()->getPointeeType(); 377 378 // If the first type needs to be converted (e.g. void** -> int*), do it now. 379 if (BuiltinFT->getArgType(0) != FirstArg->getType()) { 380 ImpCastExprToType(FirstArg, BuiltinFT->getArgType(0), CastExpr::CK_Unknown, 381 /*isLvalue=*/false); 382 TheCall->setArg(0, FirstArg); 383 } 384 385 // Next, walk the valid ones promoting to the right type. 386 for (unsigned i = 0; i != NumFixed; ++i) { 387 Expr *Arg = TheCall->getArg(i+1); 388 389 // If the argument is an implicit cast, then there was a promotion due to 390 // "...", just remove it now. 391 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) { 392 Arg = ICE->getSubExpr(); 393 ICE->setSubExpr(0); 394 ICE->Destroy(Context); 395 TheCall->setArg(i+1, Arg); 396 } 397 398 // GCC does an implicit conversion to the pointer or integer ValType. This 399 // can fail in some cases (1i -> int**), check for this error case now. 400 CastExpr::CastKind Kind = CastExpr::CK_Unknown; 401 CXXMethodDecl *ConversionDecl = 0; 402 if (CheckCastTypes(Arg->getSourceRange(), ValType, Arg, Kind, 403 ConversionDecl)) 404 return true; 405 406 // Okay, we have something that *can* be converted to the right type. Check 407 // to see if there is a potentially weird extension going on here. This can 408 // happen when you do an atomic operation on something like an char* and 409 // pass in 42. The 42 gets converted to char. This is even more strange 410 // for things like 45.123 -> char, etc. 411 // FIXME: Do this check. 412 ImpCastExprToType(Arg, ValType, Kind, /*isLvalue=*/false); 413 TheCall->setArg(i+1, Arg); 414 } 415 416 // Switch the DeclRefExpr to refer to the new decl. 417 DRE->setDecl(NewBuiltinDecl); 418 DRE->setType(NewBuiltinDecl->getType()); 419 420 // Set the callee in the CallExpr. 421 // FIXME: This leaks the original parens and implicit casts. 422 Expr *PromotedCall = DRE; 423 UsualUnaryConversions(PromotedCall); 424 TheCall->setCallee(PromotedCall); 425 426 427 // Change the result type of the call to match the result type of the decl. 428 TheCall->setType(NewBuiltinDecl->getResultType()); 429 return false; 430} 431 432 433/// CheckObjCString - Checks that the argument to the builtin 434/// CFString constructor is correct 435/// FIXME: GCC currently emits the following warning: 436/// "warning: input conversion stopped due to an input byte that does not 437/// belong to the input codeset UTF-8" 438/// Note: It might also make sense to do the UTF-16 conversion here (would 439/// simplify the backend). 440bool Sema::CheckObjCString(Expr *Arg) { 441 Arg = Arg->IgnoreParenCasts(); 442 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 443 444 if (!Literal || Literal->isWide()) { 445 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 446 << Arg->getSourceRange(); 447 return true; 448 } 449 450 const char *Data = Literal->getStrData(); 451 unsigned Length = Literal->getByteLength(); 452 453 for (unsigned i = 0; i < Length; ++i) { 454 if (!Data[i]) { 455 Diag(getLocationOfStringLiteralByte(Literal, i), 456 diag::warn_cfstring_literal_contains_nul_character) 457 << Arg->getSourceRange(); 458 break; 459 } 460 } 461 462 return false; 463} 464 465/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 466/// Emit an error and return true on failure, return false on success. 467bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 468 Expr *Fn = TheCall->getCallee(); 469 if (TheCall->getNumArgs() > 2) { 470 Diag(TheCall->getArg(2)->getLocStart(), 471 diag::err_typecheck_call_too_many_args) 472 << 0 /*function call*/ << Fn->getSourceRange() 473 << SourceRange(TheCall->getArg(2)->getLocStart(), 474 (*(TheCall->arg_end()-1))->getLocEnd()); 475 return true; 476 } 477 478 if (TheCall->getNumArgs() < 2) { 479 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 480 << 0 /*function call*/; 481 } 482 483 // Determine whether the current function is variadic or not. 484 bool isVariadic; 485 if (CurBlock) 486 isVariadic = CurBlock->isVariadic; 487 else if (getCurFunctionDecl()) { 488 if (FunctionProtoType* FTP = 489 dyn_cast<FunctionProtoType>(getCurFunctionDecl()->getType())) 490 isVariadic = FTP->isVariadic(); 491 else 492 isVariadic = false; 493 } else { 494 isVariadic = getCurMethodDecl()->isVariadic(); 495 } 496 497 if (!isVariadic) { 498 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 499 return true; 500 } 501 502 // Verify that the second argument to the builtin is the last argument of the 503 // current function or method. 504 bool SecondArgIsLastNamedArgument = false; 505 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 506 507 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 508 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 509 // FIXME: This isn't correct for methods (results in bogus warning). 510 // Get the last formal in the current function. 511 const ParmVarDecl *LastArg; 512 if (CurBlock) 513 LastArg = *(CurBlock->TheDecl->param_end()-1); 514 else if (FunctionDecl *FD = getCurFunctionDecl()) 515 LastArg = *(FD->param_end()-1); 516 else 517 LastArg = *(getCurMethodDecl()->param_end()-1); 518 SecondArgIsLastNamedArgument = PV == LastArg; 519 } 520 } 521 522 if (!SecondArgIsLastNamedArgument) 523 Diag(TheCall->getArg(1)->getLocStart(), 524 diag::warn_second_parameter_of_va_start_not_last_named_argument); 525 return false; 526} 527 528/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 529/// friends. This is declared to take (...), so we have to check everything. 530bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 531 if (TheCall->getNumArgs() < 2) 532 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 533 << 0 /*function call*/; 534 if (TheCall->getNumArgs() > 2) 535 return Diag(TheCall->getArg(2)->getLocStart(), 536 diag::err_typecheck_call_too_many_args) 537 << 0 /*function call*/ 538 << SourceRange(TheCall->getArg(2)->getLocStart(), 539 (*(TheCall->arg_end()-1))->getLocEnd()); 540 541 Expr *OrigArg0 = TheCall->getArg(0); 542 Expr *OrigArg1 = TheCall->getArg(1); 543 544 // Do standard promotions between the two arguments, returning their common 545 // type. 546 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 547 548 // Make sure any conversions are pushed back into the call; this is 549 // type safe since unordered compare builtins are declared as "_Bool 550 // foo(...)". 551 TheCall->setArg(0, OrigArg0); 552 TheCall->setArg(1, OrigArg1); 553 554 if (OrigArg0->isTypeDependent() || OrigArg1->isTypeDependent()) 555 return false; 556 557 // If the common type isn't a real floating type, then the arguments were 558 // invalid for this operation. 559 if (!Res->isRealFloatingType()) 560 return Diag(OrigArg0->getLocStart(), 561 diag::err_typecheck_call_invalid_ordered_compare) 562 << OrigArg0->getType() << OrigArg1->getType() 563 << SourceRange(OrigArg0->getLocStart(), OrigArg1->getLocEnd()); 564 565 return false; 566} 567 568/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isnan and 569/// friends. This is declared to take (...), so we have to check everything. 570bool Sema::SemaBuiltinUnaryFP(CallExpr *TheCall) { 571 if (TheCall->getNumArgs() < 1) 572 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 573 << 0 /*function call*/; 574 if (TheCall->getNumArgs() > 1) 575 return Diag(TheCall->getArg(1)->getLocStart(), 576 diag::err_typecheck_call_too_many_args) 577 << 0 /*function call*/ 578 << SourceRange(TheCall->getArg(1)->getLocStart(), 579 (*(TheCall->arg_end()-1))->getLocEnd()); 580 581 Expr *OrigArg = TheCall->getArg(0); 582 583 if (OrigArg->isTypeDependent()) 584 return false; 585 586 // This operation requires a floating-point number 587 if (!OrigArg->getType()->isRealFloatingType()) 588 return Diag(OrigArg->getLocStart(), 589 diag::err_typecheck_call_invalid_unary_fp) 590 << OrigArg->getType() << OrigArg->getSourceRange(); 591 592 return false; 593} 594 595bool Sema::SemaBuiltinStackAddress(CallExpr *TheCall) { 596 // The signature for these builtins is exact; the only thing we need 597 // to check is that the argument is a constant. 598 SourceLocation Loc; 599 if (!TheCall->getArg(0)->isTypeDependent() && 600 !TheCall->getArg(0)->isValueDependent() && 601 !TheCall->getArg(0)->isIntegerConstantExpr(Context, &Loc)) 602 return Diag(Loc, diag::err_stack_const_level) << TheCall->getSourceRange(); 603 604 return false; 605} 606 607/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 608// This is declared to take (...), so we have to check everything. 609Action::OwningExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 610 if (TheCall->getNumArgs() < 3) 611 return ExprError(Diag(TheCall->getLocEnd(), 612 diag::err_typecheck_call_too_few_args) 613 << 0 /*function call*/ << TheCall->getSourceRange()); 614 615 unsigned numElements = std::numeric_limits<unsigned>::max(); 616 if (!TheCall->getArg(0)->isTypeDependent() && 617 !TheCall->getArg(1)->isTypeDependent()) { 618 QualType FAType = TheCall->getArg(0)->getType(); 619 QualType SAType = TheCall->getArg(1)->getType(); 620 621 if (!FAType->isVectorType() || !SAType->isVectorType()) { 622 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 623 << SourceRange(TheCall->getArg(0)->getLocStart(), 624 TheCall->getArg(1)->getLocEnd()); 625 return ExprError(); 626 } 627 628 if (Context.getCanonicalType(FAType).getUnqualifiedType() != 629 Context.getCanonicalType(SAType).getUnqualifiedType()) { 630 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 631 << SourceRange(TheCall->getArg(0)->getLocStart(), 632 TheCall->getArg(1)->getLocEnd()); 633 return ExprError(); 634 } 635 636 numElements = FAType->getAsVectorType()->getNumElements(); 637 if (TheCall->getNumArgs() != numElements+2) { 638 if (TheCall->getNumArgs() < numElements+2) 639 return ExprError(Diag(TheCall->getLocEnd(), 640 diag::err_typecheck_call_too_few_args) 641 << 0 /*function call*/ << TheCall->getSourceRange()); 642 return ExprError(Diag(TheCall->getLocEnd(), 643 diag::err_typecheck_call_too_many_args) 644 << 0 /*function call*/ << TheCall->getSourceRange()); 645 } 646 } 647 648 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 649 if (TheCall->getArg(i)->isTypeDependent() || 650 TheCall->getArg(i)->isValueDependent()) 651 continue; 652 653 llvm::APSInt Result(32); 654 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 655 return ExprError(Diag(TheCall->getLocStart(), 656 diag::err_shufflevector_nonconstant_argument) 657 << TheCall->getArg(i)->getSourceRange()); 658 659 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 660 return ExprError(Diag(TheCall->getLocStart(), 661 diag::err_shufflevector_argument_too_large) 662 << TheCall->getArg(i)->getSourceRange()); 663 } 664 665 llvm::SmallVector<Expr*, 32> exprs; 666 667 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 668 exprs.push_back(TheCall->getArg(i)); 669 TheCall->setArg(i, 0); 670 } 671 672 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 673 exprs.size(), exprs[0]->getType(), 674 TheCall->getCallee()->getLocStart(), 675 TheCall->getRParenLoc())); 676} 677 678/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 679// This is declared to take (const void*, ...) and can take two 680// optional constant int args. 681bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 682 unsigned NumArgs = TheCall->getNumArgs(); 683 684 if (NumArgs > 3) 685 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_many_args) 686 << 0 /*function call*/ << TheCall->getSourceRange(); 687 688 // Argument 0 is checked for us and the remaining arguments must be 689 // constant integers. 690 for (unsigned i = 1; i != NumArgs; ++i) { 691 Expr *Arg = TheCall->getArg(i); 692 if (Arg->isTypeDependent()) 693 continue; 694 695 QualType RWType = Arg->getType(); 696 697 const BuiltinType *BT = RWType->getAsBuiltinType(); 698 llvm::APSInt Result; 699 if (!BT || BT->getKind() != BuiltinType::Int) 700 return Diag(TheCall->getLocStart(), diag::err_prefetch_invalid_argument) 701 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 702 703 if (Arg->isValueDependent()) 704 continue; 705 706 if (!Arg->isIntegerConstantExpr(Result, Context)) 707 return Diag(TheCall->getLocStart(), diag::err_prefetch_invalid_argument) 708 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 709 710 // FIXME: gcc issues a warning and rewrites these to 0. These 711 // seems especially odd for the third argument since the default 712 // is 3. 713 if (i == 1) { 714 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 1) 715 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 716 << "0" << "1" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 717 } else { 718 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) 719 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 720 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 721 } 722 } 723 724 return false; 725} 726 727/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 728/// int type). This simply type checks that type is one of the defined 729/// constants (0-3). 730bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 731 Expr *Arg = TheCall->getArg(1); 732 if (Arg->isTypeDependent()) 733 return false; 734 735 QualType ArgType = Arg->getType(); 736 const BuiltinType *BT = ArgType->getAsBuiltinType(); 737 llvm::APSInt Result(32); 738 if (!BT || BT->getKind() != BuiltinType::Int) 739 return Diag(TheCall->getLocStart(), diag::err_object_size_invalid_argument) 740 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 741 742 if (Arg->isValueDependent()) 743 return false; 744 745 if (!Arg->isIntegerConstantExpr(Result, Context)) { 746 return Diag(TheCall->getLocStart(), diag::err_object_size_invalid_argument) 747 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 748 } 749 750 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 751 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 752 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 753 } 754 755 return false; 756} 757 758/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 759/// This checks that val is a constant 1. 760bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 761 Expr *Arg = TheCall->getArg(1); 762 if (Arg->isTypeDependent() || Arg->isValueDependent()) 763 return false; 764 765 llvm::APSInt Result(32); 766 if (!Arg->isIntegerConstantExpr(Result, Context) || Result != 1) 767 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 768 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 769 770 return false; 771} 772 773// Handle i > 1 ? "x" : "y", recursivelly 774bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 775 bool HasVAListArg, 776 unsigned format_idx, unsigned firstDataArg) { 777 if (E->isTypeDependent() || E->isValueDependent()) 778 return false; 779 780 switch (E->getStmtClass()) { 781 case Stmt::ConditionalOperatorClass: { 782 const ConditionalOperator *C = cast<ConditionalOperator>(E); 783 return SemaCheckStringLiteral(C->getLHS(), TheCall, 784 HasVAListArg, format_idx, firstDataArg) 785 && SemaCheckStringLiteral(C->getRHS(), TheCall, 786 HasVAListArg, format_idx, firstDataArg); 787 } 788 789 case Stmt::ImplicitCastExprClass: { 790 const ImplicitCastExpr *Expr = cast<ImplicitCastExpr>(E); 791 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 792 format_idx, firstDataArg); 793 } 794 795 case Stmt::ParenExprClass: { 796 const ParenExpr *Expr = cast<ParenExpr>(E); 797 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 798 format_idx, firstDataArg); 799 } 800 801 case Stmt::DeclRefExprClass: { 802 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 803 804 // As an exception, do not flag errors for variables binding to 805 // const string literals. 806 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 807 bool isConstant = false; 808 QualType T = DR->getType(); 809 810 if (const ArrayType *AT = Context.getAsArrayType(T)) { 811 isConstant = AT->getElementType().isConstant(Context); 812 } else if (const PointerType *PT = T->getAs<PointerType>()) { 813 isConstant = T.isConstant(Context) && 814 PT->getPointeeType().isConstant(Context); 815 } 816 817 if (isConstant) { 818 const VarDecl *Def = 0; 819 if (const Expr *Init = VD->getDefinition(Def)) 820 return SemaCheckStringLiteral(Init, TheCall, 821 HasVAListArg, format_idx, firstDataArg); 822 } 823 824 // For vprintf* functions (i.e., HasVAListArg==true), we add a 825 // special check to see if the format string is a function parameter 826 // of the function calling the printf function. If the function 827 // has an attribute indicating it is a printf-like function, then we 828 // should suppress warnings concerning non-literals being used in a call 829 // to a vprintf function. For example: 830 // 831 // void 832 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 833 // va_list ap; 834 // va_start(ap, fmt); 835 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 836 // ... 837 // 838 // 839 // FIXME: We don't have full attribute support yet, so just check to see 840 // if the argument is a DeclRefExpr that references a parameter. We'll 841 // add proper support for checking the attribute later. 842 if (HasVAListArg) 843 if (isa<ParmVarDecl>(VD)) 844 return true; 845 } 846 847 return false; 848 } 849 850 case Stmt::CallExprClass: { 851 const CallExpr *CE = cast<CallExpr>(E); 852 if (const ImplicitCastExpr *ICE 853 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 854 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 855 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 856 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 857 unsigned ArgIndex = FA->getFormatIdx(); 858 const Expr *Arg = CE->getArg(ArgIndex - 1); 859 860 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 861 format_idx, firstDataArg); 862 } 863 } 864 } 865 } 866 867 return false; 868 } 869 case Stmt::ObjCStringLiteralClass: 870 case Stmt::StringLiteralClass: { 871 const StringLiteral *StrE = NULL; 872 873 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 874 StrE = ObjCFExpr->getString(); 875 else 876 StrE = cast<StringLiteral>(E); 877 878 if (StrE) { 879 CheckPrintfString(StrE, E, TheCall, HasVAListArg, format_idx, 880 firstDataArg); 881 return true; 882 } 883 884 return false; 885 } 886 887 default: 888 return false; 889 } 890} 891 892void 893Sema::CheckNonNullArguments(const NonNullAttr *NonNull, const CallExpr *TheCall) 894{ 895 for (NonNullAttr::iterator i = NonNull->begin(), e = NonNull->end(); 896 i != e; ++i) { 897 const Expr *ArgExpr = TheCall->getArg(*i); 898 if (ArgExpr->isNullPointerConstant(Context)) 899 Diag(TheCall->getCallee()->getLocStart(), diag::warn_null_arg) 900 << ArgExpr->getSourceRange(); 901 } 902} 903 904/// CheckPrintfArguments - Check calls to printf (and similar functions) for 905/// correct use of format strings. 906/// 907/// HasVAListArg - A predicate indicating whether the printf-like 908/// function is passed an explicit va_arg argument (e.g., vprintf) 909/// 910/// format_idx - The index into Args for the format string. 911/// 912/// Improper format strings to functions in the printf family can be 913/// the source of bizarre bugs and very serious security holes. A 914/// good source of information is available in the following paper 915/// (which includes additional references): 916/// 917/// FormatGuard: Automatic Protection From printf Format String 918/// Vulnerabilities, Proceedings of the 10th USENIX Security Symposium, 2001. 919/// 920/// Functionality implemented: 921/// 922/// We can statically check the following properties for string 923/// literal format strings for non v.*printf functions (where the 924/// arguments are passed directly): 925// 926/// (1) Are the number of format conversions equal to the number of 927/// data arguments? 928/// 929/// (2) Does each format conversion correctly match the type of the 930/// corresponding data argument? (TODO) 931/// 932/// Moreover, for all printf functions we can: 933/// 934/// (3) Check for a missing format string (when not caught by type checking). 935/// 936/// (4) Check for no-operation flags; e.g. using "#" with format 937/// conversion 'c' (TODO) 938/// 939/// (5) Check the use of '%n', a major source of security holes. 940/// 941/// (6) Check for malformed format conversions that don't specify anything. 942/// 943/// (7) Check for empty format strings. e.g: printf(""); 944/// 945/// (8) Check that the format string is a wide literal. 946/// 947/// (9) Also check the arguments of functions with the __format__ attribute. 948/// (TODO). 949/// 950/// All of these checks can be done by parsing the format string. 951/// 952/// For now, we ONLY do (1), (3), (5), (6), (7), and (8). 953void 954Sema::CheckPrintfArguments(const CallExpr *TheCall, bool HasVAListArg, 955 unsigned format_idx, unsigned firstDataArg) { 956 const Expr *Fn = TheCall->getCallee(); 957 958 // CHECK: printf-like function is called with no format string. 959 if (format_idx >= TheCall->getNumArgs()) { 960 Diag(TheCall->getRParenLoc(), diag::warn_printf_missing_format_string) 961 << Fn->getSourceRange(); 962 return; 963 } 964 965 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 966 967 // CHECK: format string is not a string literal. 968 // 969 // Dynamically generated format strings are difficult to 970 // automatically vet at compile time. Requiring that format strings 971 // are string literals: (1) permits the checking of format strings by 972 // the compiler and thereby (2) can practically remove the source of 973 // many format string exploits. 974 975 // Format string can be either ObjC string (e.g. @"%d") or 976 // C string (e.g. "%d") 977 // ObjC string uses the same format specifiers as C string, so we can use 978 // the same format string checking logic for both ObjC and C strings. 979 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 980 firstDataArg)) 981 return; // Literal format string found, check done! 982 983 // If there are no arguments specified, warn with -Wformat-security, otherwise 984 // warn only with -Wformat-nonliteral. 985 if (TheCall->getNumArgs() == format_idx+1) 986 Diag(TheCall->getArg(format_idx)->getLocStart(), 987 diag::warn_printf_nonliteral_noargs) 988 << OrigFormatExpr->getSourceRange(); 989 else 990 Diag(TheCall->getArg(format_idx)->getLocStart(), 991 diag::warn_printf_nonliteral) 992 << OrigFormatExpr->getSourceRange(); 993} 994 995void Sema::CheckPrintfString(const StringLiteral *FExpr, 996 const Expr *OrigFormatExpr, 997 const CallExpr *TheCall, bool HasVAListArg, 998 unsigned format_idx, unsigned firstDataArg) { 999 1000 const ObjCStringLiteral *ObjCFExpr = 1001 dyn_cast<ObjCStringLiteral>(OrigFormatExpr); 1002 1003 // CHECK: is the format string a wide literal? 1004 if (FExpr->isWide()) { 1005 Diag(FExpr->getLocStart(), 1006 diag::warn_printf_format_string_is_wide_literal) 1007 << OrigFormatExpr->getSourceRange(); 1008 return; 1009 } 1010 1011 // Str - The format string. NOTE: this is NOT null-terminated! 1012 const char *Str = FExpr->getStrData(); 1013 1014 // CHECK: empty format string? 1015 unsigned StrLen = FExpr->getByteLength(); 1016 1017 if (StrLen == 0) { 1018 Diag(FExpr->getLocStart(), diag::warn_printf_empty_format_string) 1019 << OrigFormatExpr->getSourceRange(); 1020 return; 1021 } 1022 1023 // We process the format string using a binary state machine. The 1024 // current state is stored in CurrentState. 1025 enum { 1026 state_OrdChr, 1027 state_Conversion 1028 } CurrentState = state_OrdChr; 1029 1030 // numConversions - The number of conversions seen so far. This is 1031 // incremented as we traverse the format string. 1032 unsigned numConversions = 0; 1033 1034 // numDataArgs - The number of data arguments after the format 1035 // string. This can only be determined for non vprintf-like 1036 // functions. For those functions, this value is 1 (the sole 1037 // va_arg argument). 1038 unsigned numDataArgs = TheCall->getNumArgs()-firstDataArg; 1039 1040 // Inspect the format string. 1041 unsigned StrIdx = 0; 1042 1043 // LastConversionIdx - Index within the format string where we last saw 1044 // a '%' character that starts a new format conversion. 1045 unsigned LastConversionIdx = 0; 1046 1047 for (; StrIdx < StrLen; ++StrIdx) { 1048 1049 // Is the number of detected conversion conversions greater than 1050 // the number of matching data arguments? If so, stop. 1051 if (!HasVAListArg && numConversions > numDataArgs) break; 1052 1053 // Handle "\0" 1054 if (Str[StrIdx] == '\0') { 1055 // The string returned by getStrData() is not null-terminated, 1056 // so the presence of a null character is likely an error. 1057 Diag(getLocationOfStringLiteralByte(FExpr, StrIdx), 1058 diag::warn_printf_format_string_contains_null_char) 1059 << OrigFormatExpr->getSourceRange(); 1060 return; 1061 } 1062 1063 // Ordinary characters (not processing a format conversion). 1064 if (CurrentState == state_OrdChr) { 1065 if (Str[StrIdx] == '%') { 1066 CurrentState = state_Conversion; 1067 LastConversionIdx = StrIdx; 1068 } 1069 continue; 1070 } 1071 1072 // Seen '%'. Now processing a format conversion. 1073 switch (Str[StrIdx]) { 1074 // Handle dynamic precision or width specifier. 1075 case '*': { 1076 ++numConversions; 1077 1078 if (!HasVAListArg) { 1079 if (numConversions > numDataArgs) { 1080 SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, StrIdx); 1081 1082 if (Str[StrIdx-1] == '.') 1083 Diag(Loc, diag::warn_printf_asterisk_precision_missing_arg) 1084 << OrigFormatExpr->getSourceRange(); 1085 else 1086 Diag(Loc, diag::warn_printf_asterisk_width_missing_arg) 1087 << OrigFormatExpr->getSourceRange(); 1088 1089 // Don't do any more checking. We'll just emit spurious errors. 1090 return; 1091 } 1092 1093 // Perform type checking on width/precision specifier. 1094 const Expr *E = TheCall->getArg(format_idx+numConversions); 1095 if (const BuiltinType *BT = E->getType()->getAsBuiltinType()) 1096 if (BT->getKind() == BuiltinType::Int) 1097 break; 1098 1099 SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, StrIdx); 1100 1101 if (Str[StrIdx-1] == '.') 1102 Diag(Loc, diag::warn_printf_asterisk_precision_wrong_type) 1103 << E->getType() << E->getSourceRange(); 1104 else 1105 Diag(Loc, diag::warn_printf_asterisk_width_wrong_type) 1106 << E->getType() << E->getSourceRange(); 1107 1108 break; 1109 } 1110 } 1111 1112 // Characters which can terminate a format conversion 1113 // (e.g. "%d"). Characters that specify length modifiers or 1114 // other flags are handled by the default case below. 1115 // 1116 // FIXME: additional checks will go into the following cases. 1117 case 'i': 1118 case 'd': 1119 case 'o': 1120 case 'u': 1121 case 'x': 1122 case 'X': 1123 case 'D': 1124 case 'O': 1125 case 'U': 1126 case 'e': 1127 case 'E': 1128 case 'f': 1129 case 'F': 1130 case 'g': 1131 case 'G': 1132 case 'a': 1133 case 'A': 1134 case 'c': 1135 case 'C': 1136 case 'S': 1137 case 's': 1138 case 'p': 1139 ++numConversions; 1140 CurrentState = state_OrdChr; 1141 break; 1142 1143 case 'm': 1144 // FIXME: Warn in situations where this isn't supported! 1145 CurrentState = state_OrdChr; 1146 break; 1147 1148 // CHECK: Are we using "%n"? Issue a warning. 1149 case 'n': { 1150 ++numConversions; 1151 CurrentState = state_OrdChr; 1152 SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, 1153 LastConversionIdx); 1154 1155 Diag(Loc, diag::warn_printf_write_back)<<OrigFormatExpr->getSourceRange(); 1156 break; 1157 } 1158 1159 // Handle "%@" 1160 case '@': 1161 // %@ is allowed in ObjC format strings only. 1162 if(ObjCFExpr != NULL) 1163 CurrentState = state_OrdChr; 1164 else { 1165 // Issue a warning: invalid format conversion. 1166 SourceLocation Loc = 1167 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 1168 1169 Diag(Loc, diag::warn_printf_invalid_conversion) 1170 << std::string(Str+LastConversionIdx, 1171 Str+std::min(LastConversionIdx+2, StrLen)) 1172 << OrigFormatExpr->getSourceRange(); 1173 } 1174 ++numConversions; 1175 break; 1176 1177 // Handle "%%" 1178 case '%': 1179 // Sanity check: Was the first "%" character the previous one? 1180 // If not, we will assume that we have a malformed format 1181 // conversion, and that the current "%" character is the start 1182 // of a new conversion. 1183 if (StrIdx - LastConversionIdx == 1) 1184 CurrentState = state_OrdChr; 1185 else { 1186 // Issue a warning: invalid format conversion. 1187 SourceLocation Loc = 1188 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 1189 1190 Diag(Loc, diag::warn_printf_invalid_conversion) 1191 << std::string(Str+LastConversionIdx, Str+StrIdx) 1192 << OrigFormatExpr->getSourceRange(); 1193 1194 // This conversion is broken. Advance to the next format 1195 // conversion. 1196 LastConversionIdx = StrIdx; 1197 ++numConversions; 1198 } 1199 break; 1200 1201 default: 1202 // This case catches all other characters: flags, widths, etc. 1203 // We should eventually process those as well. 1204 break; 1205 } 1206 } 1207 1208 if (CurrentState == state_Conversion) { 1209 // Issue a warning: invalid format conversion. 1210 SourceLocation Loc = 1211 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 1212 1213 Diag(Loc, diag::warn_printf_invalid_conversion) 1214 << std::string(Str+LastConversionIdx, 1215 Str+std::min(LastConversionIdx+2, StrLen)) 1216 << OrigFormatExpr->getSourceRange(); 1217 return; 1218 } 1219 1220 if (!HasVAListArg) { 1221 // CHECK: Does the number of format conversions exceed the number 1222 // of data arguments? 1223 if (numConversions > numDataArgs) { 1224 SourceLocation Loc = 1225 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 1226 1227 Diag(Loc, diag::warn_printf_insufficient_data_args) 1228 << OrigFormatExpr->getSourceRange(); 1229 } 1230 // CHECK: Does the number of data arguments exceed the number of 1231 // format conversions in the format string? 1232 else if (numConversions < numDataArgs) 1233 Diag(TheCall->getArg(format_idx+numConversions+1)->getLocStart(), 1234 diag::warn_printf_too_many_data_args) 1235 << OrigFormatExpr->getSourceRange(); 1236 } 1237} 1238 1239//===--- CHECK: Return Address of Stack Variable --------------------------===// 1240 1241static DeclRefExpr* EvalVal(Expr *E); 1242static DeclRefExpr* EvalAddr(Expr* E); 1243 1244/// CheckReturnStackAddr - Check if a return statement returns the address 1245/// of a stack variable. 1246void 1247Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 1248 SourceLocation ReturnLoc) { 1249 1250 // Perform checking for returned stack addresses. 1251 if (lhsType->isPointerType() || lhsType->isBlockPointerType()) { 1252 if (DeclRefExpr *DR = EvalAddr(RetValExp)) 1253 Diag(DR->getLocStart(), diag::warn_ret_stack_addr) 1254 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1255 1256 // Skip over implicit cast expressions when checking for block expressions. 1257 if (ImplicitCastExpr *IcExpr = 1258 dyn_cast_or_null<ImplicitCastExpr>(RetValExp)) 1259 RetValExp = IcExpr->getSubExpr(); 1260 1261 if (BlockExpr *C = dyn_cast_or_null<BlockExpr>(RetValExp)) 1262 if (C->hasBlockDeclRefExprs()) 1263 Diag(C->getLocStart(), diag::err_ret_local_block) 1264 << C->getSourceRange(); 1265 } else if (lhsType->isReferenceType()) { 1266 // Perform checking for stack values returned by reference. 1267 // Check for a reference to the stack 1268 if (DeclRefExpr *DR = EvalVal(RetValExp)) 1269 Diag(DR->getLocStart(), diag::warn_ret_stack_ref) 1270 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1271 } 1272} 1273 1274/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 1275/// check if the expression in a return statement evaluates to an address 1276/// to a location on the stack. The recursion is used to traverse the 1277/// AST of the return expression, with recursion backtracking when we 1278/// encounter a subexpression that (1) clearly does not lead to the address 1279/// of a stack variable or (2) is something we cannot determine leads to 1280/// the address of a stack variable based on such local checking. 1281/// 1282/// EvalAddr processes expressions that are pointers that are used as 1283/// references (and not L-values). EvalVal handles all other values. 1284/// At the base case of the recursion is a check for a DeclRefExpr* in 1285/// the refers to a stack variable. 1286/// 1287/// This implementation handles: 1288/// 1289/// * pointer-to-pointer casts 1290/// * implicit conversions from array references to pointers 1291/// * taking the address of fields 1292/// * arbitrary interplay between "&" and "*" operators 1293/// * pointer arithmetic from an address of a stack variable 1294/// * taking the address of an array element where the array is on the stack 1295static DeclRefExpr* EvalAddr(Expr *E) { 1296 // We should only be called for evaluating pointer expressions. 1297 assert((E->getType()->isAnyPointerType() || 1298 E->getType()->isBlockPointerType() || 1299 E->getType()->isObjCQualifiedIdType()) && 1300 "EvalAddr only works on pointers"); 1301 1302 // Our "symbolic interpreter" is just a dispatch off the currently 1303 // viewed AST node. We then recursively traverse the AST by calling 1304 // EvalAddr and EvalVal appropriately. 1305 switch (E->getStmtClass()) { 1306 case Stmt::ParenExprClass: 1307 // Ignore parentheses. 1308 return EvalAddr(cast<ParenExpr>(E)->getSubExpr()); 1309 1310 case Stmt::UnaryOperatorClass: { 1311 // The only unary operator that make sense to handle here 1312 // is AddrOf. All others don't make sense as pointers. 1313 UnaryOperator *U = cast<UnaryOperator>(E); 1314 1315 if (U->getOpcode() == UnaryOperator::AddrOf) 1316 return EvalVal(U->getSubExpr()); 1317 else 1318 return NULL; 1319 } 1320 1321 case Stmt::BinaryOperatorClass: { 1322 // Handle pointer arithmetic. All other binary operators are not valid 1323 // in this context. 1324 BinaryOperator *B = cast<BinaryOperator>(E); 1325 BinaryOperator::Opcode op = B->getOpcode(); 1326 1327 if (op != BinaryOperator::Add && op != BinaryOperator::Sub) 1328 return NULL; 1329 1330 Expr *Base = B->getLHS(); 1331 1332 // Determine which argument is the real pointer base. It could be 1333 // the RHS argument instead of the LHS. 1334 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 1335 1336 assert (Base->getType()->isPointerType()); 1337 return EvalAddr(Base); 1338 } 1339 1340 // For conditional operators we need to see if either the LHS or RHS are 1341 // valid DeclRefExpr*s. If one of them is valid, we return it. 1342 case Stmt::ConditionalOperatorClass: { 1343 ConditionalOperator *C = cast<ConditionalOperator>(E); 1344 1345 // Handle the GNU extension for missing LHS. 1346 if (Expr *lhsExpr = C->getLHS()) 1347 if (DeclRefExpr* LHS = EvalAddr(lhsExpr)) 1348 return LHS; 1349 1350 return EvalAddr(C->getRHS()); 1351 } 1352 1353 // For casts, we need to handle conversions from arrays to 1354 // pointer values, and pointer-to-pointer conversions. 1355 case Stmt::ImplicitCastExprClass: 1356 case Stmt::CStyleCastExprClass: 1357 case Stmt::CXXFunctionalCastExprClass: { 1358 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 1359 QualType T = SubExpr->getType(); 1360 1361 if (SubExpr->getType()->isPointerType() || 1362 SubExpr->getType()->isBlockPointerType() || 1363 SubExpr->getType()->isObjCQualifiedIdType()) 1364 return EvalAddr(SubExpr); 1365 else if (T->isArrayType()) 1366 return EvalVal(SubExpr); 1367 else 1368 return 0; 1369 } 1370 1371 // C++ casts. For dynamic casts, static casts, and const casts, we 1372 // are always converting from a pointer-to-pointer, so we just blow 1373 // through the cast. In the case the dynamic cast doesn't fail (and 1374 // return NULL), we take the conservative route and report cases 1375 // where we return the address of a stack variable. For Reinterpre 1376 // FIXME: The comment about is wrong; we're not always converting 1377 // from pointer to pointer. I'm guessing that this code should also 1378 // handle references to objects. 1379 case Stmt::CXXStaticCastExprClass: 1380 case Stmt::CXXDynamicCastExprClass: 1381 case Stmt::CXXConstCastExprClass: 1382 case Stmt::CXXReinterpretCastExprClass: { 1383 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 1384 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 1385 return EvalAddr(S); 1386 else 1387 return NULL; 1388 } 1389 1390 // Everything else: we simply don't reason about them. 1391 default: 1392 return NULL; 1393 } 1394} 1395 1396 1397/// EvalVal - This function is complements EvalAddr in the mutual recursion. 1398/// See the comments for EvalAddr for more details. 1399static DeclRefExpr* EvalVal(Expr *E) { 1400 1401 // We should only be called for evaluating non-pointer expressions, or 1402 // expressions with a pointer type that are not used as references but instead 1403 // are l-values (e.g., DeclRefExpr with a pointer type). 1404 1405 // Our "symbolic interpreter" is just a dispatch off the currently 1406 // viewed AST node. We then recursively traverse the AST by calling 1407 // EvalAddr and EvalVal appropriately. 1408 switch (E->getStmtClass()) { 1409 case Stmt::DeclRefExprClass: 1410 case Stmt::QualifiedDeclRefExprClass: { 1411 // DeclRefExpr: the base case. When we hit a DeclRefExpr we are looking 1412 // at code that refers to a variable's name. We check if it has local 1413 // storage within the function, and if so, return the expression. 1414 DeclRefExpr *DR = cast<DeclRefExpr>(E); 1415 1416 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 1417 if(V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR; 1418 1419 return NULL; 1420 } 1421 1422 case Stmt::ParenExprClass: 1423 // Ignore parentheses. 1424 return EvalVal(cast<ParenExpr>(E)->getSubExpr()); 1425 1426 case Stmt::UnaryOperatorClass: { 1427 // The only unary operator that make sense to handle here 1428 // is Deref. All others don't resolve to a "name." This includes 1429 // handling all sorts of rvalues passed to a unary operator. 1430 UnaryOperator *U = cast<UnaryOperator>(E); 1431 1432 if (U->getOpcode() == UnaryOperator::Deref) 1433 return EvalAddr(U->getSubExpr()); 1434 1435 return NULL; 1436 } 1437 1438 case Stmt::ArraySubscriptExprClass: { 1439 // Array subscripts are potential references to data on the stack. We 1440 // retrieve the DeclRefExpr* for the array variable if it indeed 1441 // has local storage. 1442 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase()); 1443 } 1444 1445 case Stmt::ConditionalOperatorClass: { 1446 // For conditional operators we need to see if either the LHS or RHS are 1447 // non-NULL DeclRefExpr's. If one is non-NULL, we return it. 1448 ConditionalOperator *C = cast<ConditionalOperator>(E); 1449 1450 // Handle the GNU extension for missing LHS. 1451 if (Expr *lhsExpr = C->getLHS()) 1452 if (DeclRefExpr *LHS = EvalVal(lhsExpr)) 1453 return LHS; 1454 1455 return EvalVal(C->getRHS()); 1456 } 1457 1458 // Accesses to members are potential references to data on the stack. 1459 case Stmt::MemberExprClass: 1460 case Stmt::CXXAdornedMemberExprClass: { 1461 MemberExpr *M = cast<MemberExpr>(E); 1462 1463 // Check for indirect access. We only want direct field accesses. 1464 if (!M->isArrow()) 1465 return EvalVal(M->getBase()); 1466 else 1467 return NULL; 1468 } 1469 1470 // Everything else: we simply don't reason about them. 1471 default: 1472 return NULL; 1473 } 1474} 1475 1476//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 1477 1478/// Check for comparisons of floating point operands using != and ==. 1479/// Issue a warning if these are no self-comparisons, as they are not likely 1480/// to do what the programmer intended. 1481void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 1482 bool EmitWarning = true; 1483 1484 Expr* LeftExprSansParen = lex->IgnoreParens(); 1485 Expr* RightExprSansParen = rex->IgnoreParens(); 1486 1487 // Special case: check for x == x (which is OK). 1488 // Do not emit warnings for such cases. 1489 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 1490 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 1491 if (DRL->getDecl() == DRR->getDecl()) 1492 EmitWarning = false; 1493 1494 1495 // Special case: check for comparisons against literals that can be exactly 1496 // represented by APFloat. In such cases, do not emit a warning. This 1497 // is a heuristic: often comparison against such literals are used to 1498 // detect if a value in a variable has not changed. This clearly can 1499 // lead to false negatives. 1500 if (EmitWarning) { 1501 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 1502 if (FLL->isExact()) 1503 EmitWarning = false; 1504 } else 1505 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 1506 if (FLR->isExact()) 1507 EmitWarning = false; 1508 } 1509 } 1510 1511 // Check for comparisons with builtin types. 1512 if (EmitWarning) 1513 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 1514 if (CL->isBuiltinCall(Context)) 1515 EmitWarning = false; 1516 1517 if (EmitWarning) 1518 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 1519 if (CR->isBuiltinCall(Context)) 1520 EmitWarning = false; 1521 1522 // Emit the diagnostic. 1523 if (EmitWarning) 1524 Diag(loc, diag::warn_floatingpoint_eq) 1525 << lex->getSourceRange() << rex->getSourceRange(); 1526} 1527