SemaChecking.cpp revision e898f8a94947c6074d76ff83943b47d5bbdf210d
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 for (const Attr *attr = FDecl->getAttrs(); attr; attr = attr->getNext()) { 182 if (const NonNullAttr *NonNull = dyn_cast<NonNullAttr>(attr)) 183 CheckNonNullArguments(NonNull, TheCall); 184 } 185 186 return move(TheCallResult); 187} 188 189Action::OwningExprResult 190Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 191 192 OwningExprResult TheCallResult(Owned(TheCall)); 193 // Printf checking. 194 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 195 if (!Format) 196 return move(TheCallResult); 197 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 198 if (!V) 199 return move(TheCallResult); 200 QualType Ty = V->getType(); 201 if (!Ty->isBlockPointerType()) 202 return move(TheCallResult); 203 if (Format->getType() == "printf") { 204 bool HasVAListArg = Format->getFirstArg() == 0; 205 if (!HasVAListArg) { 206 const FunctionType *FT = 207 Ty->getAsBlockPointerType()->getPointeeType()->getAsFunctionType(); 208 if (const FunctionProtoType *Proto = dyn_cast<FunctionProtoType>(FT)) 209 HasVAListArg = !Proto->isVariadic(); 210 } 211 CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 212 HasVAListArg ? 0 : Format->getFirstArg() - 1); 213 } 214 return move(TheCallResult); 215} 216 217/// SemaBuiltinAtomicOverloaded - We have a call to a function like 218/// __sync_fetch_and_add, which is an overloaded function based on the pointer 219/// type of its first argument. The main ActOnCallExpr routines have already 220/// promoted the types of arguments because all of these calls are prototyped as 221/// void(...). 222/// 223/// This function goes through and does final semantic checking for these 224/// builtins, 225bool Sema::SemaBuiltinAtomicOverloaded(CallExpr *TheCall) { 226 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 227 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 228 229 // Ensure that we have at least one argument to do type inference from. 230 if (TheCall->getNumArgs() < 1) 231 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 232 << 0 << TheCall->getCallee()->getSourceRange(); 233 234 // Inspect the first argument of the atomic builtin. This should always be 235 // a pointer type, whose element is an integral scalar or pointer type. 236 // Because it is a pointer type, we don't have to worry about any implicit 237 // casts here. 238 Expr *FirstArg = TheCall->getArg(0); 239 if (!FirstArg->getType()->isPointerType()) 240 return Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 241 << FirstArg->getType() << FirstArg->getSourceRange(); 242 243 QualType ValType = FirstArg->getType()->getAsPointerType()->getPointeeType(); 244 if (!ValType->isIntegerType() && !ValType->isPointerType() && 245 !ValType->isBlockPointerType()) 246 return Diag(DRE->getLocStart(), 247 diag::err_atomic_builtin_must_be_pointer_intptr) 248 << FirstArg->getType() << FirstArg->getSourceRange(); 249 250 // We need to figure out which concrete builtin this maps onto. For example, 251 // __sync_fetch_and_add with a 2 byte object turns into 252 // __sync_fetch_and_add_2. 253#define BUILTIN_ROW(x) \ 254 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 255 Builtin::BI##x##_8, Builtin::BI##x##_16 } 256 257 static const unsigned BuiltinIndices[][5] = { 258 BUILTIN_ROW(__sync_fetch_and_add), 259 BUILTIN_ROW(__sync_fetch_and_sub), 260 BUILTIN_ROW(__sync_fetch_and_or), 261 BUILTIN_ROW(__sync_fetch_and_and), 262 BUILTIN_ROW(__sync_fetch_and_xor), 263 BUILTIN_ROW(__sync_fetch_and_nand), 264 265 BUILTIN_ROW(__sync_add_and_fetch), 266 BUILTIN_ROW(__sync_sub_and_fetch), 267 BUILTIN_ROW(__sync_and_and_fetch), 268 BUILTIN_ROW(__sync_or_and_fetch), 269 BUILTIN_ROW(__sync_xor_and_fetch), 270 BUILTIN_ROW(__sync_nand_and_fetch), 271 272 BUILTIN_ROW(__sync_val_compare_and_swap), 273 BUILTIN_ROW(__sync_bool_compare_and_swap), 274 BUILTIN_ROW(__sync_lock_test_and_set), 275 BUILTIN_ROW(__sync_lock_release) 276 }; 277#undef BUILTIN_ROW 278 279 // Determine the index of the size. 280 unsigned SizeIndex; 281 switch (Context.getTypeSize(ValType)/8) { 282 case 1: SizeIndex = 0; break; 283 case 2: SizeIndex = 1; break; 284 case 4: SizeIndex = 2; break; 285 case 8: SizeIndex = 3; break; 286 case 16: SizeIndex = 4; break; 287 default: 288 return Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 289 << FirstArg->getType() << FirstArg->getSourceRange(); 290 } 291 292 // Each of these builtins has one pointer argument, followed by some number of 293 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 294 // that we ignore. Find out which row of BuiltinIndices to read from as well 295 // as the number of fixed args. 296 unsigned BuiltinID = FDecl->getBuiltinID(Context); 297 unsigned BuiltinIndex, NumFixed = 1; 298 switch (BuiltinID) { 299 default: assert(0 && "Unknown overloaded atomic builtin!"); 300 case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break; 301 case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break; 302 case Builtin::BI__sync_fetch_and_or: BuiltinIndex = 2; break; 303 case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break; 304 case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break; 305 case Builtin::BI__sync_fetch_and_nand:BuiltinIndex = 5; break; 306 307 case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 6; break; 308 case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 7; break; 309 case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 8; break; 310 case Builtin::BI__sync_or_and_fetch: BuiltinIndex = 9; break; 311 case Builtin::BI__sync_xor_and_fetch: BuiltinIndex =10; break; 312 case Builtin::BI__sync_nand_and_fetch:BuiltinIndex =11; break; 313 314 case Builtin::BI__sync_val_compare_and_swap: 315 BuiltinIndex = 12; 316 NumFixed = 2; 317 break; 318 case Builtin::BI__sync_bool_compare_and_swap: 319 BuiltinIndex = 13; 320 NumFixed = 2; 321 break; 322 case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 14; break; 323 case Builtin::BI__sync_lock_release: 324 BuiltinIndex = 15; 325 NumFixed = 0; 326 break; 327 } 328 329 // Now that we know how many fixed arguments we expect, first check that we 330 // have at least that many. 331 if (TheCall->getNumArgs() < 1+NumFixed) 332 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 333 << 0 << TheCall->getCallee()->getSourceRange(); 334 335 336 // Get the decl for the concrete builtin from this, we can tell what the 337 // concrete integer type we should convert to is. 338 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 339 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 340 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 341 FunctionDecl *NewBuiltinDecl = 342 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 343 TUScope, false, DRE->getLocStart())); 344 const FunctionProtoType *BuiltinFT = 345 NewBuiltinDecl->getType()->getAsFunctionProtoType(); 346 ValType = BuiltinFT->getArgType(0)->getAsPointerType()->getPointeeType(); 347 348 // If the first type needs to be converted (e.g. void** -> int*), do it now. 349 if (BuiltinFT->getArgType(0) != FirstArg->getType()) { 350 ImpCastExprToType(FirstArg, BuiltinFT->getArgType(0), false); 351 TheCall->setArg(0, FirstArg); 352 } 353 354 // Next, walk the valid ones promoting to the right type. 355 for (unsigned i = 0; i != NumFixed; ++i) { 356 Expr *Arg = TheCall->getArg(i+1); 357 358 // If the argument is an implicit cast, then there was a promotion due to 359 // "...", just remove it now. 360 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) { 361 Arg = ICE->getSubExpr(); 362 ICE->setSubExpr(0); 363 ICE->Destroy(Context); 364 TheCall->setArg(i+1, Arg); 365 } 366 367 // GCC does an implicit conversion to the pointer or integer ValType. This 368 // can fail in some cases (1i -> int**), check for this error case now. 369 if (CheckCastTypes(Arg->getSourceRange(), ValType, Arg)) 370 return true; 371 372 // Okay, we have something that *can* be converted to the right type. Check 373 // to see if there is a potentially weird extension going on here. This can 374 // happen when you do an atomic operation on something like an char* and 375 // pass in 42. The 42 gets converted to char. This is even more strange 376 // for things like 45.123 -> char, etc. 377 // FIXME: Do this check. 378 ImpCastExprToType(Arg, ValType, false); 379 TheCall->setArg(i+1, Arg); 380 } 381 382 // Switch the DeclRefExpr to refer to the new decl. 383 DRE->setDecl(NewBuiltinDecl); 384 DRE->setType(NewBuiltinDecl->getType()); 385 386 // Set the callee in the CallExpr. 387 // FIXME: This leaks the original parens and implicit casts. 388 Expr *PromotedCall = DRE; 389 UsualUnaryConversions(PromotedCall); 390 TheCall->setCallee(PromotedCall); 391 392 393 // Change the result type of the call to match the result type of the decl. 394 TheCall->setType(NewBuiltinDecl->getResultType()); 395 return false; 396} 397 398 399/// CheckObjCString - Checks that the argument to the builtin 400/// CFString constructor is correct 401/// FIXME: GCC currently emits the following warning: 402/// "warning: input conversion stopped due to an input byte that does not 403/// belong to the input codeset UTF-8" 404/// Note: It might also make sense to do the UTF-16 conversion here (would 405/// simplify the backend). 406bool Sema::CheckObjCString(Expr *Arg) { 407 Arg = Arg->IgnoreParenCasts(); 408 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 409 410 if (!Literal || Literal->isWide()) { 411 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 412 << Arg->getSourceRange(); 413 return true; 414 } 415 416 const char *Data = Literal->getStrData(); 417 unsigned Length = Literal->getByteLength(); 418 419 for (unsigned i = 0; i < Length; ++i) { 420 if (!Data[i]) { 421 Diag(getLocationOfStringLiteralByte(Literal, i), 422 diag::warn_cfstring_literal_contains_nul_character) 423 << Arg->getSourceRange(); 424 break; 425 } 426 } 427 428 return false; 429} 430 431/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 432/// Emit an error and return true on failure, return false on success. 433bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 434 Expr *Fn = TheCall->getCallee(); 435 if (TheCall->getNumArgs() > 2) { 436 Diag(TheCall->getArg(2)->getLocStart(), 437 diag::err_typecheck_call_too_many_args) 438 << 0 /*function call*/ << Fn->getSourceRange() 439 << SourceRange(TheCall->getArg(2)->getLocStart(), 440 (*(TheCall->arg_end()-1))->getLocEnd()); 441 return true; 442 } 443 444 if (TheCall->getNumArgs() < 2) { 445 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 446 << 0 /*function call*/; 447 } 448 449 // Determine whether the current function is variadic or not. 450 bool isVariadic; 451 if (CurBlock) 452 isVariadic = CurBlock->isVariadic; 453 else if (getCurFunctionDecl()) { 454 if (FunctionProtoType* FTP = 455 dyn_cast<FunctionProtoType>(getCurFunctionDecl()->getType())) 456 isVariadic = FTP->isVariadic(); 457 else 458 isVariadic = false; 459 } else { 460 isVariadic = getCurMethodDecl()->isVariadic(); 461 } 462 463 if (!isVariadic) { 464 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 465 return true; 466 } 467 468 // Verify that the second argument to the builtin is the last argument of the 469 // current function or method. 470 bool SecondArgIsLastNamedArgument = false; 471 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 472 473 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 474 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 475 // FIXME: This isn't correct for methods (results in bogus warning). 476 // Get the last formal in the current function. 477 const ParmVarDecl *LastArg; 478 if (CurBlock) 479 LastArg = *(CurBlock->TheDecl->param_end()-1); 480 else if (FunctionDecl *FD = getCurFunctionDecl()) 481 LastArg = *(FD->param_end()-1); 482 else 483 LastArg = *(getCurMethodDecl()->param_end()-1); 484 SecondArgIsLastNamedArgument = PV == LastArg; 485 } 486 } 487 488 if (!SecondArgIsLastNamedArgument) 489 Diag(TheCall->getArg(1)->getLocStart(), 490 diag::warn_second_parameter_of_va_start_not_last_named_argument); 491 return false; 492} 493 494/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 495/// friends. This is declared to take (...), so we have to check everything. 496bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 497 if (TheCall->getNumArgs() < 2) 498 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 499 << 0 /*function call*/; 500 if (TheCall->getNumArgs() > 2) 501 return Diag(TheCall->getArg(2)->getLocStart(), 502 diag::err_typecheck_call_too_many_args) 503 << 0 /*function call*/ 504 << SourceRange(TheCall->getArg(2)->getLocStart(), 505 (*(TheCall->arg_end()-1))->getLocEnd()); 506 507 Expr *OrigArg0 = TheCall->getArg(0); 508 Expr *OrigArg1 = TheCall->getArg(1); 509 510 // Do standard promotions between the two arguments, returning their common 511 // type. 512 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 513 514 // Make sure any conversions are pushed back into the call; this is 515 // type safe since unordered compare builtins are declared as "_Bool 516 // foo(...)". 517 TheCall->setArg(0, OrigArg0); 518 TheCall->setArg(1, OrigArg1); 519 520 if (OrigArg0->isTypeDependent() || OrigArg1->isTypeDependent()) 521 return false; 522 523 // If the common type isn't a real floating type, then the arguments were 524 // invalid for this operation. 525 if (!Res->isRealFloatingType()) 526 return Diag(OrigArg0->getLocStart(), 527 diag::err_typecheck_call_invalid_ordered_compare) 528 << OrigArg0->getType() << OrigArg1->getType() 529 << SourceRange(OrigArg0->getLocStart(), OrigArg1->getLocEnd()); 530 531 return false; 532} 533 534bool Sema::SemaBuiltinStackAddress(CallExpr *TheCall) { 535 // The signature for these builtins is exact; the only thing we need 536 // to check is that the argument is a constant. 537 SourceLocation Loc; 538 if (!TheCall->getArg(0)->isTypeDependent() && 539 !TheCall->getArg(0)->isValueDependent() && 540 !TheCall->getArg(0)->isIntegerConstantExpr(Context, &Loc)) 541 return Diag(Loc, diag::err_stack_const_level) << TheCall->getSourceRange(); 542 543 return false; 544} 545 546/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 547// This is declared to take (...), so we have to check everything. 548Action::OwningExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 549 if (TheCall->getNumArgs() < 3) 550 return ExprError(Diag(TheCall->getLocEnd(), 551 diag::err_typecheck_call_too_few_args) 552 << 0 /*function call*/ << TheCall->getSourceRange()); 553 554 unsigned numElements = std::numeric_limits<unsigned>::max(); 555 if (!TheCall->getArg(0)->isTypeDependent() && 556 !TheCall->getArg(1)->isTypeDependent()) { 557 QualType FAType = TheCall->getArg(0)->getType(); 558 QualType SAType = TheCall->getArg(1)->getType(); 559 560 if (!FAType->isVectorType() || !SAType->isVectorType()) { 561 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 562 << SourceRange(TheCall->getArg(0)->getLocStart(), 563 TheCall->getArg(1)->getLocEnd()); 564 return ExprError(); 565 } 566 567 if (Context.getCanonicalType(FAType).getUnqualifiedType() != 568 Context.getCanonicalType(SAType).getUnqualifiedType()) { 569 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 570 << SourceRange(TheCall->getArg(0)->getLocStart(), 571 TheCall->getArg(1)->getLocEnd()); 572 return ExprError(); 573 } 574 575 numElements = FAType->getAsVectorType()->getNumElements(); 576 if (TheCall->getNumArgs() != numElements+2) { 577 if (TheCall->getNumArgs() < numElements+2) 578 return ExprError(Diag(TheCall->getLocEnd(), 579 diag::err_typecheck_call_too_few_args) 580 << 0 /*function call*/ << TheCall->getSourceRange()); 581 return ExprError(Diag(TheCall->getLocEnd(), 582 diag::err_typecheck_call_too_many_args) 583 << 0 /*function call*/ << TheCall->getSourceRange()); 584 } 585 } 586 587 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 588 if (TheCall->getArg(i)->isTypeDependent() || 589 TheCall->getArg(i)->isValueDependent()) 590 continue; 591 592 llvm::APSInt Result(32); 593 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 594 return ExprError(Diag(TheCall->getLocStart(), 595 diag::err_shufflevector_nonconstant_argument) 596 << TheCall->getArg(i)->getSourceRange()); 597 598 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 599 return ExprError(Diag(TheCall->getLocStart(), 600 diag::err_shufflevector_argument_too_large) 601 << TheCall->getArg(i)->getSourceRange()); 602 } 603 604 llvm::SmallVector<Expr*, 32> exprs; 605 606 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 607 exprs.push_back(TheCall->getArg(i)); 608 TheCall->setArg(i, 0); 609 } 610 611 return Owned(new (Context) ShuffleVectorExpr(exprs.begin(), exprs.size(), 612 exprs[0]->getType(), 613 TheCall->getCallee()->getLocStart(), 614 TheCall->getRParenLoc())); 615} 616 617/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 618// This is declared to take (const void*, ...) and can take two 619// optional constant int args. 620bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 621 unsigned NumArgs = TheCall->getNumArgs(); 622 623 if (NumArgs > 3) 624 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_many_args) 625 << 0 /*function call*/ << TheCall->getSourceRange(); 626 627 // Argument 0 is checked for us and the remaining arguments must be 628 // constant integers. 629 for (unsigned i = 1; i != NumArgs; ++i) { 630 Expr *Arg = TheCall->getArg(i); 631 if (Arg->isTypeDependent()) 632 continue; 633 634 QualType RWType = Arg->getType(); 635 636 const BuiltinType *BT = RWType->getAsBuiltinType(); 637 llvm::APSInt Result; 638 if (!BT || BT->getKind() != BuiltinType::Int) 639 return Diag(TheCall->getLocStart(), diag::err_prefetch_invalid_argument) 640 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 641 642 if (Arg->isValueDependent()) 643 continue; 644 645 if (!Arg->isIntegerConstantExpr(Result, Context)) 646 return Diag(TheCall->getLocStart(), diag::err_prefetch_invalid_argument) 647 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 648 649 // FIXME: gcc issues a warning and rewrites these to 0. These 650 // seems especially odd for the third argument since the default 651 // is 3. 652 if (i == 1) { 653 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 1) 654 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 655 << "0" << "1" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 656 } else { 657 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) 658 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 659 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 660 } 661 } 662 663 return false; 664} 665 666/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 667/// int type). This simply type checks that type is one of the defined 668/// constants (0-3). 669bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 670 Expr *Arg = TheCall->getArg(1); 671 if (Arg->isTypeDependent()) 672 return false; 673 674 QualType ArgType = Arg->getType(); 675 const BuiltinType *BT = ArgType->getAsBuiltinType(); 676 llvm::APSInt Result(32); 677 if (!BT || BT->getKind() != BuiltinType::Int) 678 return Diag(TheCall->getLocStart(), diag::err_object_size_invalid_argument) 679 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 680 681 if (Arg->isValueDependent()) 682 return false; 683 684 if (!Arg->isIntegerConstantExpr(Result, Context)) { 685 return Diag(TheCall->getLocStart(), diag::err_object_size_invalid_argument) 686 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 687 } 688 689 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 690 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 691 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 692 } 693 694 return false; 695} 696 697/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 698/// This checks that val is a constant 1. 699bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 700 Expr *Arg = TheCall->getArg(1); 701 if (Arg->isTypeDependent() || Arg->isValueDependent()) 702 return false; 703 704 llvm::APSInt Result(32); 705 if (!Arg->isIntegerConstantExpr(Result, Context) || Result != 1) 706 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 707 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 708 709 return false; 710} 711 712// Handle i > 1 ? "x" : "y", recursivelly 713bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 714 bool HasVAListArg, 715 unsigned format_idx, unsigned firstDataArg) { 716 if (E->isTypeDependent() || E->isValueDependent()) 717 return false; 718 719 switch (E->getStmtClass()) { 720 case Stmt::ConditionalOperatorClass: { 721 const ConditionalOperator *C = cast<ConditionalOperator>(E); 722 return SemaCheckStringLiteral(C->getLHS(), TheCall, 723 HasVAListArg, format_idx, firstDataArg) 724 && SemaCheckStringLiteral(C->getRHS(), TheCall, 725 HasVAListArg, format_idx, firstDataArg); 726 } 727 728 case Stmt::ImplicitCastExprClass: { 729 const ImplicitCastExpr *Expr = cast<ImplicitCastExpr>(E); 730 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 731 format_idx, firstDataArg); 732 } 733 734 case Stmt::ParenExprClass: { 735 const ParenExpr *Expr = cast<ParenExpr>(E); 736 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 737 format_idx, firstDataArg); 738 } 739 740 case Stmt::DeclRefExprClass: { 741 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 742 743 // As an exception, do not flag errors for variables binding to 744 // const string literals. 745 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 746 bool isConstant = false; 747 QualType T = DR->getType(); 748 749 if (const ArrayType *AT = Context.getAsArrayType(T)) { 750 isConstant = AT->getElementType().isConstant(Context); 751 } 752 else if (const PointerType *PT = T->getAsPointerType()) { 753 isConstant = T.isConstant(Context) && 754 PT->getPointeeType().isConstant(Context); 755 } 756 757 if (isConstant) { 758 const VarDecl *Def = 0; 759 if (const Expr *Init = VD->getDefinition(Def)) 760 return SemaCheckStringLiteral(Init, TheCall, 761 HasVAListArg, format_idx, firstDataArg); 762 } 763 } 764 765 return false; 766 } 767 768 case Stmt::ObjCStringLiteralClass: 769 case Stmt::StringLiteralClass: { 770 const StringLiteral *StrE = NULL; 771 772 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 773 StrE = ObjCFExpr->getString(); 774 else 775 StrE = cast<StringLiteral>(E); 776 777 if (StrE) { 778 CheckPrintfString(StrE, E, TheCall, HasVAListArg, format_idx, 779 firstDataArg); 780 return true; 781 } 782 783 return false; 784 } 785 786 default: 787 return false; 788 } 789} 790 791void 792Sema::CheckNonNullArguments(const NonNullAttr *NonNull, const CallExpr *TheCall) 793{ 794 for (NonNullAttr::iterator i = NonNull->begin(), e = NonNull->end(); 795 i != e; ++i) { 796 const Expr *ArgExpr = TheCall->getArg(*i)->IgnoreParenCasts(); 797 if (ArgExpr->isNullPointerConstant(Context)) 798 Diag(ArgExpr->getLocStart(), diag::warn_null_arg); 799 } 800} 801 802/// CheckPrintfArguments - Check calls to printf (and similar functions) for 803/// correct use of format strings. 804/// 805/// HasVAListArg - A predicate indicating whether the printf-like 806/// function is passed an explicit va_arg argument (e.g., vprintf) 807/// 808/// format_idx - The index into Args for the format string. 809/// 810/// Improper format strings to functions in the printf family can be 811/// the source of bizarre bugs and very serious security holes. A 812/// good source of information is available in the following paper 813/// (which includes additional references): 814/// 815/// FormatGuard: Automatic Protection From printf Format String 816/// Vulnerabilities, Proceedings of the 10th USENIX Security Symposium, 2001. 817/// 818/// Functionality implemented: 819/// 820/// We can statically check the following properties for string 821/// literal format strings for non v.*printf functions (where the 822/// arguments are passed directly): 823// 824/// (1) Are the number of format conversions equal to the number of 825/// data arguments? 826/// 827/// (2) Does each format conversion correctly match the type of the 828/// corresponding data argument? (TODO) 829/// 830/// Moreover, for all printf functions we can: 831/// 832/// (3) Check for a missing format string (when not caught by type checking). 833/// 834/// (4) Check for no-operation flags; e.g. using "#" with format 835/// conversion 'c' (TODO) 836/// 837/// (5) Check the use of '%n', a major source of security holes. 838/// 839/// (6) Check for malformed format conversions that don't specify anything. 840/// 841/// (7) Check for empty format strings. e.g: printf(""); 842/// 843/// (8) Check that the format string is a wide literal. 844/// 845/// (9) Also check the arguments of functions with the __format__ attribute. 846/// (TODO). 847/// 848/// All of these checks can be done by parsing the format string. 849/// 850/// For now, we ONLY do (1), (3), (5), (6), (7), and (8). 851void 852Sema::CheckPrintfArguments(const CallExpr *TheCall, bool HasVAListArg, 853 unsigned format_idx, unsigned firstDataArg) { 854 const Expr *Fn = TheCall->getCallee(); 855 856 // CHECK: printf-like function is called with no format string. 857 if (format_idx >= TheCall->getNumArgs()) { 858 Diag(TheCall->getRParenLoc(), diag::warn_printf_missing_format_string) 859 << Fn->getSourceRange(); 860 return; 861 } 862 863 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 864 865 // CHECK: format string is not a string literal. 866 // 867 // Dynamically generated format strings are difficult to 868 // automatically vet at compile time. Requiring that format strings 869 // are string literals: (1) permits the checking of format strings by 870 // the compiler and thereby (2) can practically remove the source of 871 // many format string exploits. 872 873 // Format string can be either ObjC string (e.g. @"%d") or 874 // C string (e.g. "%d") 875 // ObjC string uses the same format specifiers as C string, so we can use 876 // the same format string checking logic for both ObjC and C strings. 877 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 878 firstDataArg)) 879 return; // Literal format string found, check done! 880 881 // For vprintf* functions (i.e., HasVAListArg==true), we add a 882 // special check to see if the format string is a function parameter 883 // of the function calling the printf function. If the function 884 // has an attribute indicating it is a printf-like function, then we 885 // should suppress warnings concerning non-literals being used in a call 886 // to a vprintf function. For example: 887 // 888 // void 889 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...) { 890 // va_list ap; 891 // va_start(ap, fmt); 892 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 893 // ... 894 // 895 // 896 // FIXME: We don't have full attribute support yet, so just check to see 897 // if the argument is a DeclRefExpr that references a parameter. We'll 898 // add proper support for checking the attribute later. 899 if (HasVAListArg) 900 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(OrigFormatExpr)) 901 if (isa<ParmVarDecl>(DR->getDecl())) 902 return; 903 904 // If there are no arguments specified, warn with -Wformat-security, otherwise 905 // warn only with -Wformat-nonliteral. 906 if (TheCall->getNumArgs() == format_idx+1) 907 Diag(TheCall->getArg(format_idx)->getLocStart(), 908 diag::warn_printf_nonliteral_noargs) 909 << OrigFormatExpr->getSourceRange(); 910 else 911 Diag(TheCall->getArg(format_idx)->getLocStart(), 912 diag::warn_printf_nonliteral) 913 << OrigFormatExpr->getSourceRange(); 914} 915 916void Sema::CheckPrintfString(const StringLiteral *FExpr, 917 const Expr *OrigFormatExpr, 918 const CallExpr *TheCall, bool HasVAListArg, 919 unsigned format_idx, unsigned firstDataArg) { 920 921 const ObjCStringLiteral *ObjCFExpr = 922 dyn_cast<ObjCStringLiteral>(OrigFormatExpr); 923 924 // CHECK: is the format string a wide literal? 925 if (FExpr->isWide()) { 926 Diag(FExpr->getLocStart(), 927 diag::warn_printf_format_string_is_wide_literal) 928 << OrigFormatExpr->getSourceRange(); 929 return; 930 } 931 932 // Str - The format string. NOTE: this is NOT null-terminated! 933 const char *Str = FExpr->getStrData(); 934 935 // CHECK: empty format string? 936 unsigned StrLen = FExpr->getByteLength(); 937 938 if (StrLen == 0) { 939 Diag(FExpr->getLocStart(), diag::warn_printf_empty_format_string) 940 << OrigFormatExpr->getSourceRange(); 941 return; 942 } 943 944 // We process the format string using a binary state machine. The 945 // current state is stored in CurrentState. 946 enum { 947 state_OrdChr, 948 state_Conversion 949 } CurrentState = state_OrdChr; 950 951 // numConversions - The number of conversions seen so far. This is 952 // incremented as we traverse the format string. 953 unsigned numConversions = 0; 954 955 // numDataArgs - The number of data arguments after the format 956 // string. This can only be determined for non vprintf-like 957 // functions. For those functions, this value is 1 (the sole 958 // va_arg argument). 959 unsigned numDataArgs = TheCall->getNumArgs()-firstDataArg; 960 961 // Inspect the format string. 962 unsigned StrIdx = 0; 963 964 // LastConversionIdx - Index within the format string where we last saw 965 // a '%' character that starts a new format conversion. 966 unsigned LastConversionIdx = 0; 967 968 for (; StrIdx < StrLen; ++StrIdx) { 969 970 // Is the number of detected conversion conversions greater than 971 // the number of matching data arguments? If so, stop. 972 if (!HasVAListArg && numConversions > numDataArgs) break; 973 974 // Handle "\0" 975 if (Str[StrIdx] == '\0') { 976 // The string returned by getStrData() is not null-terminated, 977 // so the presence of a null character is likely an error. 978 Diag(getLocationOfStringLiteralByte(FExpr, StrIdx), 979 diag::warn_printf_format_string_contains_null_char) 980 << OrigFormatExpr->getSourceRange(); 981 return; 982 } 983 984 // Ordinary characters (not processing a format conversion). 985 if (CurrentState == state_OrdChr) { 986 if (Str[StrIdx] == '%') { 987 CurrentState = state_Conversion; 988 LastConversionIdx = StrIdx; 989 } 990 continue; 991 } 992 993 // Seen '%'. Now processing a format conversion. 994 switch (Str[StrIdx]) { 995 // Handle dynamic precision or width specifier. 996 case '*': { 997 ++numConversions; 998 999 if (!HasVAListArg) { 1000 if (numConversions > numDataArgs) { 1001 SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, StrIdx); 1002 1003 if (Str[StrIdx-1] == '.') 1004 Diag(Loc, diag::warn_printf_asterisk_precision_missing_arg) 1005 << OrigFormatExpr->getSourceRange(); 1006 else 1007 Diag(Loc, diag::warn_printf_asterisk_width_missing_arg) 1008 << OrigFormatExpr->getSourceRange(); 1009 1010 // Don't do any more checking. We'll just emit spurious errors. 1011 return; 1012 } 1013 1014 // Perform type checking on width/precision specifier. 1015 const Expr *E = TheCall->getArg(format_idx+numConversions); 1016 if (const BuiltinType *BT = E->getType()->getAsBuiltinType()) 1017 if (BT->getKind() == BuiltinType::Int) 1018 break; 1019 1020 SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, StrIdx); 1021 1022 if (Str[StrIdx-1] == '.') 1023 Diag(Loc, diag::warn_printf_asterisk_precision_wrong_type) 1024 << E->getType() << E->getSourceRange(); 1025 else 1026 Diag(Loc, diag::warn_printf_asterisk_width_wrong_type) 1027 << E->getType() << E->getSourceRange(); 1028 1029 break; 1030 } 1031 } 1032 1033 // Characters which can terminate a format conversion 1034 // (e.g. "%d"). Characters that specify length modifiers or 1035 // other flags are handled by the default case below. 1036 // 1037 // FIXME: additional checks will go into the following cases. 1038 case 'i': 1039 case 'd': 1040 case 'o': 1041 case 'u': 1042 case 'x': 1043 case 'X': 1044 case 'D': 1045 case 'O': 1046 case 'U': 1047 case 'e': 1048 case 'E': 1049 case 'f': 1050 case 'F': 1051 case 'g': 1052 case 'G': 1053 case 'a': 1054 case 'A': 1055 case 'c': 1056 case 'C': 1057 case 'S': 1058 case 's': 1059 case 'p': 1060 ++numConversions; 1061 CurrentState = state_OrdChr; 1062 break; 1063 1064 // CHECK: Are we using "%n"? Issue a warning. 1065 case 'n': { 1066 ++numConversions; 1067 CurrentState = state_OrdChr; 1068 SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, 1069 LastConversionIdx); 1070 1071 Diag(Loc, diag::warn_printf_write_back)<<OrigFormatExpr->getSourceRange(); 1072 break; 1073 } 1074 1075 // Handle "%@" 1076 case '@': 1077 // %@ is allowed in ObjC format strings only. 1078 if(ObjCFExpr != NULL) 1079 CurrentState = state_OrdChr; 1080 else { 1081 // Issue a warning: invalid format conversion. 1082 SourceLocation Loc = 1083 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 1084 1085 Diag(Loc, diag::warn_printf_invalid_conversion) 1086 << std::string(Str+LastConversionIdx, 1087 Str+std::min(LastConversionIdx+2, StrLen)) 1088 << OrigFormatExpr->getSourceRange(); 1089 } 1090 ++numConversions; 1091 break; 1092 1093 // Handle "%%" 1094 case '%': 1095 // Sanity check: Was the first "%" character the previous one? 1096 // If not, we will assume that we have a malformed format 1097 // conversion, and that the current "%" character is the start 1098 // of a new conversion. 1099 if (StrIdx - LastConversionIdx == 1) 1100 CurrentState = state_OrdChr; 1101 else { 1102 // Issue a warning: invalid format conversion. 1103 SourceLocation Loc = 1104 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 1105 1106 Diag(Loc, diag::warn_printf_invalid_conversion) 1107 << std::string(Str+LastConversionIdx, Str+StrIdx) 1108 << OrigFormatExpr->getSourceRange(); 1109 1110 // This conversion is broken. Advance to the next format 1111 // conversion. 1112 LastConversionIdx = StrIdx; 1113 ++numConversions; 1114 } 1115 break; 1116 1117 default: 1118 // This case catches all other characters: flags, widths, etc. 1119 // We should eventually process those as well. 1120 break; 1121 } 1122 } 1123 1124 if (CurrentState == state_Conversion) { 1125 // Issue a warning: invalid format conversion. 1126 SourceLocation Loc = 1127 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 1128 1129 Diag(Loc, diag::warn_printf_invalid_conversion) 1130 << std::string(Str+LastConversionIdx, 1131 Str+std::min(LastConversionIdx+2, StrLen)) 1132 << OrigFormatExpr->getSourceRange(); 1133 return; 1134 } 1135 1136 if (!HasVAListArg) { 1137 // CHECK: Does the number of format conversions exceed the number 1138 // of data arguments? 1139 if (numConversions > numDataArgs) { 1140 SourceLocation Loc = 1141 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 1142 1143 Diag(Loc, diag::warn_printf_insufficient_data_args) 1144 << OrigFormatExpr->getSourceRange(); 1145 } 1146 // CHECK: Does the number of data arguments exceed the number of 1147 // format conversions in the format string? 1148 else if (numConversions < numDataArgs) 1149 Diag(TheCall->getArg(format_idx+numConversions+1)->getLocStart(), 1150 diag::warn_printf_too_many_data_args) 1151 << OrigFormatExpr->getSourceRange(); 1152 } 1153} 1154 1155//===--- CHECK: Return Address of Stack Variable --------------------------===// 1156 1157static DeclRefExpr* EvalVal(Expr *E); 1158static DeclRefExpr* EvalAddr(Expr* E); 1159 1160/// CheckReturnStackAddr - Check if a return statement returns the address 1161/// of a stack variable. 1162void 1163Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 1164 SourceLocation ReturnLoc) { 1165 1166 // Perform checking for returned stack addresses. 1167 if (lhsType->isPointerType() || lhsType->isBlockPointerType()) { 1168 if (DeclRefExpr *DR = EvalAddr(RetValExp)) 1169 Diag(DR->getLocStart(), diag::warn_ret_stack_addr) 1170 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1171 1172 // Skip over implicit cast expressions when checking for block expressions. 1173 if (ImplicitCastExpr *IcExpr = 1174 dyn_cast_or_null<ImplicitCastExpr>(RetValExp)) 1175 RetValExp = IcExpr->getSubExpr(); 1176 1177 if (BlockExpr *C = dyn_cast_or_null<BlockExpr>(RetValExp)) 1178 if (C->hasBlockDeclRefExprs()) 1179 Diag(C->getLocStart(), diag::err_ret_local_block) 1180 << C->getSourceRange(); 1181 } 1182 // Perform checking for stack values returned by reference. 1183 else if (lhsType->isReferenceType()) { 1184 // Check for a reference to the stack 1185 if (DeclRefExpr *DR = EvalVal(RetValExp)) 1186 Diag(DR->getLocStart(), diag::warn_ret_stack_ref) 1187 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1188 } 1189} 1190 1191/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 1192/// check if the expression in a return statement evaluates to an address 1193/// to a location on the stack. The recursion is used to traverse the 1194/// AST of the return expression, with recursion backtracking when we 1195/// encounter a subexpression that (1) clearly does not lead to the address 1196/// of a stack variable or (2) is something we cannot determine leads to 1197/// the address of a stack variable based on such local checking. 1198/// 1199/// EvalAddr processes expressions that are pointers that are used as 1200/// references (and not L-values). EvalVal handles all other values. 1201/// At the base case of the recursion is a check for a DeclRefExpr* in 1202/// the refers to a stack variable. 1203/// 1204/// This implementation handles: 1205/// 1206/// * pointer-to-pointer casts 1207/// * implicit conversions from array references to pointers 1208/// * taking the address of fields 1209/// * arbitrary interplay between "&" and "*" operators 1210/// * pointer arithmetic from an address of a stack variable 1211/// * taking the address of an array element where the array is on the stack 1212static DeclRefExpr* EvalAddr(Expr *E) { 1213 // We should only be called for evaluating pointer expressions. 1214 assert((E->getType()->isPointerType() || 1215 E->getType()->isBlockPointerType() || 1216 E->getType()->isObjCQualifiedIdType()) && 1217 "EvalAddr only works on pointers"); 1218 1219 // Our "symbolic interpreter" is just a dispatch off the currently 1220 // viewed AST node. We then recursively traverse the AST by calling 1221 // EvalAddr and EvalVal appropriately. 1222 switch (E->getStmtClass()) { 1223 case Stmt::ParenExprClass: 1224 // Ignore parentheses. 1225 return EvalAddr(cast<ParenExpr>(E)->getSubExpr()); 1226 1227 case Stmt::UnaryOperatorClass: { 1228 // The only unary operator that make sense to handle here 1229 // is AddrOf. All others don't make sense as pointers. 1230 UnaryOperator *U = cast<UnaryOperator>(E); 1231 1232 if (U->getOpcode() == UnaryOperator::AddrOf) 1233 return EvalVal(U->getSubExpr()); 1234 else 1235 return NULL; 1236 } 1237 1238 case Stmt::BinaryOperatorClass: { 1239 // Handle pointer arithmetic. All other binary operators are not valid 1240 // in this context. 1241 BinaryOperator *B = cast<BinaryOperator>(E); 1242 BinaryOperator::Opcode op = B->getOpcode(); 1243 1244 if (op != BinaryOperator::Add && op != BinaryOperator::Sub) 1245 return NULL; 1246 1247 Expr *Base = B->getLHS(); 1248 1249 // Determine which argument is the real pointer base. It could be 1250 // the RHS argument instead of the LHS. 1251 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 1252 1253 assert (Base->getType()->isPointerType()); 1254 return EvalAddr(Base); 1255 } 1256 1257 // For conditional operators we need to see if either the LHS or RHS are 1258 // valid DeclRefExpr*s. If one of them is valid, we return it. 1259 case Stmt::ConditionalOperatorClass: { 1260 ConditionalOperator *C = cast<ConditionalOperator>(E); 1261 1262 // Handle the GNU extension for missing LHS. 1263 if (Expr *lhsExpr = C->getLHS()) 1264 if (DeclRefExpr* LHS = EvalAddr(lhsExpr)) 1265 return LHS; 1266 1267 return EvalAddr(C->getRHS()); 1268 } 1269 1270 // For casts, we need to handle conversions from arrays to 1271 // pointer values, and pointer-to-pointer conversions. 1272 case Stmt::ImplicitCastExprClass: 1273 case Stmt::CStyleCastExprClass: 1274 case Stmt::CXXFunctionalCastExprClass: { 1275 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 1276 QualType T = SubExpr->getType(); 1277 1278 if (SubExpr->getType()->isPointerType() || 1279 SubExpr->getType()->isBlockPointerType() || 1280 SubExpr->getType()->isObjCQualifiedIdType()) 1281 return EvalAddr(SubExpr); 1282 else if (T->isArrayType()) 1283 return EvalVal(SubExpr); 1284 else 1285 return 0; 1286 } 1287 1288 // C++ casts. For dynamic casts, static casts, and const casts, we 1289 // are always converting from a pointer-to-pointer, so we just blow 1290 // through the cast. In the case the dynamic cast doesn't fail (and 1291 // return NULL), we take the conservative route and report cases 1292 // where we return the address of a stack variable. For Reinterpre 1293 // FIXME: The comment about is wrong; we're not always converting 1294 // from pointer to pointer. I'm guessing that this code should also 1295 // handle references to objects. 1296 case Stmt::CXXStaticCastExprClass: 1297 case Stmt::CXXDynamicCastExprClass: 1298 case Stmt::CXXConstCastExprClass: 1299 case Stmt::CXXReinterpretCastExprClass: { 1300 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 1301 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 1302 return EvalAddr(S); 1303 else 1304 return NULL; 1305 } 1306 1307 // Everything else: we simply don't reason about them. 1308 default: 1309 return NULL; 1310 } 1311} 1312 1313 1314/// EvalVal - This function is complements EvalAddr in the mutual recursion. 1315/// See the comments for EvalAddr for more details. 1316static DeclRefExpr* EvalVal(Expr *E) { 1317 1318 // We should only be called for evaluating non-pointer expressions, or 1319 // expressions with a pointer type that are not used as references but instead 1320 // are l-values (e.g., DeclRefExpr with a pointer type). 1321 1322 // Our "symbolic interpreter" is just a dispatch off the currently 1323 // viewed AST node. We then recursively traverse the AST by calling 1324 // EvalAddr and EvalVal appropriately. 1325 switch (E->getStmtClass()) { 1326 case Stmt::DeclRefExprClass: 1327 case Stmt::QualifiedDeclRefExprClass: { 1328 // DeclRefExpr: the base case. When we hit a DeclRefExpr we are looking 1329 // at code that refers to a variable's name. We check if it has local 1330 // storage within the function, and if so, return the expression. 1331 DeclRefExpr *DR = cast<DeclRefExpr>(E); 1332 1333 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 1334 if(V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR; 1335 1336 return NULL; 1337 } 1338 1339 case Stmt::ParenExprClass: 1340 // Ignore parentheses. 1341 return EvalVal(cast<ParenExpr>(E)->getSubExpr()); 1342 1343 case Stmt::UnaryOperatorClass: { 1344 // The only unary operator that make sense to handle here 1345 // is Deref. All others don't resolve to a "name." This includes 1346 // handling all sorts of rvalues passed to a unary operator. 1347 UnaryOperator *U = cast<UnaryOperator>(E); 1348 1349 if (U->getOpcode() == UnaryOperator::Deref) 1350 return EvalAddr(U->getSubExpr()); 1351 1352 return NULL; 1353 } 1354 1355 case Stmt::ArraySubscriptExprClass: { 1356 // Array subscripts are potential references to data on the stack. We 1357 // retrieve the DeclRefExpr* for the array variable if it indeed 1358 // has local storage. 1359 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase()); 1360 } 1361 1362 case Stmt::ConditionalOperatorClass: { 1363 // For conditional operators we need to see if either the LHS or RHS are 1364 // non-NULL DeclRefExpr's. If one is non-NULL, we return it. 1365 ConditionalOperator *C = cast<ConditionalOperator>(E); 1366 1367 // Handle the GNU extension for missing LHS. 1368 if (Expr *lhsExpr = C->getLHS()) 1369 if (DeclRefExpr *LHS = EvalVal(lhsExpr)) 1370 return LHS; 1371 1372 return EvalVal(C->getRHS()); 1373 } 1374 1375 // Accesses to members are potential references to data on the stack. 1376 case Stmt::MemberExprClass: { 1377 MemberExpr *M = cast<MemberExpr>(E); 1378 1379 // Check for indirect access. We only want direct field accesses. 1380 if (!M->isArrow()) 1381 return EvalVal(M->getBase()); 1382 else 1383 return NULL; 1384 } 1385 1386 // Everything else: we simply don't reason about them. 1387 default: 1388 return NULL; 1389 } 1390} 1391 1392//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 1393 1394/// Check for comparisons of floating point operands using != and ==. 1395/// Issue a warning if these are no self-comparisons, as they are not likely 1396/// to do what the programmer intended. 1397void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 1398 bool EmitWarning = true; 1399 1400 Expr* LeftExprSansParen = lex->IgnoreParens(); 1401 Expr* RightExprSansParen = rex->IgnoreParens(); 1402 1403 // Special case: check for x == x (which is OK). 1404 // Do not emit warnings for such cases. 1405 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 1406 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 1407 if (DRL->getDecl() == DRR->getDecl()) 1408 EmitWarning = false; 1409 1410 1411 // Special case: check for comparisons against literals that can be exactly 1412 // represented by APFloat. In such cases, do not emit a warning. This 1413 // is a heuristic: often comparison against such literals are used to 1414 // detect if a value in a variable has not changed. This clearly can 1415 // lead to false negatives. 1416 if (EmitWarning) { 1417 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 1418 if (FLL->isExact()) 1419 EmitWarning = false; 1420 } 1421 else 1422 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 1423 if (FLR->isExact()) 1424 EmitWarning = false; 1425 } 1426 } 1427 1428 // Check for comparisons with builtin types. 1429 if (EmitWarning) 1430 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 1431 if (CL->isBuiltinCall(Context)) 1432 EmitWarning = false; 1433 1434 if (EmitWarning) 1435 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 1436 if (CR->isBuiltinCall(Context)) 1437 EmitWarning = false; 1438 1439 // Emit the diagnostic. 1440 if (EmitWarning) 1441 Diag(loc, diag::warn_floatingpoint_eq) 1442 << lex->getSourceRange() << rex->getSourceRange(); 1443} 1444