SemaChecking.cpp revision 1a1a6e2bd4c5aefd7fd643cf25915f9623a02e59
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(); 182 attr; attr = attr->getNext()) { 183 if (const NonNullAttr *NonNull = dyn_cast<NonNullAttr>(attr)) 184 CheckNonNullArguments(NonNull, TheCall); 185 } 186 187 return move(TheCallResult); 188} 189 190Action::OwningExprResult 191Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 192 193 OwningExprResult TheCallResult(Owned(TheCall)); 194 // Printf checking. 195 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 196 if (!Format) 197 return move(TheCallResult); 198 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 199 if (!V) 200 return move(TheCallResult); 201 QualType Ty = V->getType(); 202 if (!Ty->isBlockPointerType()) 203 return move(TheCallResult); 204 if (Format->getType() == "printf") { 205 bool HasVAListArg = Format->getFirstArg() == 0; 206 if (!HasVAListArg) { 207 const FunctionType *FT = 208 Ty->getAs<BlockPointerType>()->getPointeeType()->getAsFunctionType(); 209 if (const FunctionProtoType *Proto = dyn_cast<FunctionProtoType>(FT)) 210 HasVAListArg = !Proto->isVariadic(); 211 } 212 CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 213 HasVAListArg ? 0 : Format->getFirstArg() - 1); 214 } 215 return move(TheCallResult); 216} 217 218/// SemaBuiltinAtomicOverloaded - We have a call to a function like 219/// __sync_fetch_and_add, which is an overloaded function based on the pointer 220/// type of its first argument. The main ActOnCallExpr routines have already 221/// promoted the types of arguments because all of these calls are prototyped as 222/// void(...). 223/// 224/// This function goes through and does final semantic checking for these 225/// builtins, 226bool Sema::SemaBuiltinAtomicOverloaded(CallExpr *TheCall) { 227 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 228 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 229 230 // Ensure that we have at least one argument to do type inference from. 231 if (TheCall->getNumArgs() < 1) 232 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 233 << 0 << TheCall->getCallee()->getSourceRange(); 234 235 // Inspect the first argument of the atomic builtin. This should always be 236 // a pointer type, whose element is an integral scalar or pointer type. 237 // Because it is a pointer type, we don't have to worry about any implicit 238 // casts here. 239 Expr *FirstArg = TheCall->getArg(0); 240 if (!FirstArg->getType()->isPointerType()) 241 return Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 242 << FirstArg->getType() << FirstArg->getSourceRange(); 243 244 QualType ValType = FirstArg->getType()->getAs<PointerType>()->getPointeeType(); 245 if (!ValType->isIntegerType() && !ValType->isPointerType() && 246 !ValType->isBlockPointerType()) 247 return Diag(DRE->getLocStart(), 248 diag::err_atomic_builtin_must_be_pointer_intptr) 249 << FirstArg->getType() << FirstArg->getSourceRange(); 250 251 // We need to figure out which concrete builtin this maps onto. For example, 252 // __sync_fetch_and_add with a 2 byte object turns into 253 // __sync_fetch_and_add_2. 254#define BUILTIN_ROW(x) \ 255 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 256 Builtin::BI##x##_8, Builtin::BI##x##_16 } 257 258 static const unsigned BuiltinIndices[][5] = { 259 BUILTIN_ROW(__sync_fetch_and_add), 260 BUILTIN_ROW(__sync_fetch_and_sub), 261 BUILTIN_ROW(__sync_fetch_and_or), 262 BUILTIN_ROW(__sync_fetch_and_and), 263 BUILTIN_ROW(__sync_fetch_and_xor), 264 BUILTIN_ROW(__sync_fetch_and_nand), 265 266 BUILTIN_ROW(__sync_add_and_fetch), 267 BUILTIN_ROW(__sync_sub_and_fetch), 268 BUILTIN_ROW(__sync_and_and_fetch), 269 BUILTIN_ROW(__sync_or_and_fetch), 270 BUILTIN_ROW(__sync_xor_and_fetch), 271 BUILTIN_ROW(__sync_nand_and_fetch), 272 273 BUILTIN_ROW(__sync_val_compare_and_swap), 274 BUILTIN_ROW(__sync_bool_compare_and_swap), 275 BUILTIN_ROW(__sync_lock_test_and_set), 276 BUILTIN_ROW(__sync_lock_release) 277 }; 278#undef BUILTIN_ROW 279 280 // Determine the index of the size. 281 unsigned SizeIndex; 282 switch (Context.getTypeSize(ValType)/8) { 283 case 1: SizeIndex = 0; break; 284 case 2: SizeIndex = 1; break; 285 case 4: SizeIndex = 2; break; 286 case 8: SizeIndex = 3; break; 287 case 16: SizeIndex = 4; break; 288 default: 289 return Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 290 << FirstArg->getType() << FirstArg->getSourceRange(); 291 } 292 293 // Each of these builtins has one pointer argument, followed by some number of 294 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 295 // that we ignore. Find out which row of BuiltinIndices to read from as well 296 // as the number of fixed args. 297 unsigned BuiltinID = FDecl->getBuiltinID(Context); 298 unsigned BuiltinIndex, NumFixed = 1; 299 switch (BuiltinID) { 300 default: assert(0 && "Unknown overloaded atomic builtin!"); 301 case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break; 302 case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break; 303 case Builtin::BI__sync_fetch_and_or: BuiltinIndex = 2; break; 304 case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break; 305 case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break; 306 case Builtin::BI__sync_fetch_and_nand:BuiltinIndex = 5; break; 307 308 case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 6; break; 309 case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 7; break; 310 case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 8; break; 311 case Builtin::BI__sync_or_and_fetch: BuiltinIndex = 9; break; 312 case Builtin::BI__sync_xor_and_fetch: BuiltinIndex =10; break; 313 case Builtin::BI__sync_nand_and_fetch:BuiltinIndex =11; break; 314 315 case Builtin::BI__sync_val_compare_and_swap: 316 BuiltinIndex = 12; 317 NumFixed = 2; 318 break; 319 case Builtin::BI__sync_bool_compare_and_swap: 320 BuiltinIndex = 13; 321 NumFixed = 2; 322 break; 323 case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 14; break; 324 case Builtin::BI__sync_lock_release: 325 BuiltinIndex = 15; 326 NumFixed = 0; 327 break; 328 } 329 330 // Now that we know how many fixed arguments we expect, first check that we 331 // have at least that many. 332 if (TheCall->getNumArgs() < 1+NumFixed) 333 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 334 << 0 << TheCall->getCallee()->getSourceRange(); 335 336 337 // Get the decl for the concrete builtin from this, we can tell what the 338 // concrete integer type we should convert to is. 339 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 340 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 341 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 342 FunctionDecl *NewBuiltinDecl = 343 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 344 TUScope, false, DRE->getLocStart())); 345 const FunctionProtoType *BuiltinFT = 346 NewBuiltinDecl->getType()->getAsFunctionProtoType(); 347 ValType = BuiltinFT->getArgType(0)->getAs<PointerType>()->getPointeeType(); 348 349 // If the first type needs to be converted (e.g. void** -> int*), do it now. 350 if (BuiltinFT->getArgType(0) != FirstArg->getType()) { 351 ImpCastExprToType(FirstArg, BuiltinFT->getArgType(0), false); 352 TheCall->setArg(0, FirstArg); 353 } 354 355 // Next, walk the valid ones promoting to the right type. 356 for (unsigned i = 0; i != NumFixed; ++i) { 357 Expr *Arg = TheCall->getArg(i+1); 358 359 // If the argument is an implicit cast, then there was a promotion due to 360 // "...", just remove it now. 361 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) { 362 Arg = ICE->getSubExpr(); 363 ICE->setSubExpr(0); 364 ICE->Destroy(Context); 365 TheCall->setArg(i+1, Arg); 366 } 367 368 // GCC does an implicit conversion to the pointer or integer ValType. This 369 // can fail in some cases (1i -> int**), check for this error case now. 370 if (CheckCastTypes(Arg->getSourceRange(), ValType, Arg)) 371 return true; 372 373 // Okay, we have something that *can* be converted to the right type. Check 374 // to see if there is a potentially weird extension going on here. This can 375 // happen when you do an atomic operation on something like an char* and 376 // pass in 42. The 42 gets converted to char. This is even more strange 377 // for things like 45.123 -> char, etc. 378 // FIXME: Do this check. 379 ImpCastExprToType(Arg, ValType, false); 380 TheCall->setArg(i+1, Arg); 381 } 382 383 // Switch the DeclRefExpr to refer to the new decl. 384 DRE->setDecl(NewBuiltinDecl); 385 DRE->setType(NewBuiltinDecl->getType()); 386 387 // Set the callee in the CallExpr. 388 // FIXME: This leaks the original parens and implicit casts. 389 Expr *PromotedCall = DRE; 390 UsualUnaryConversions(PromotedCall); 391 TheCall->setCallee(PromotedCall); 392 393 394 // Change the result type of the call to match the result type of the decl. 395 TheCall->setType(NewBuiltinDecl->getResultType()); 396 return false; 397} 398 399 400/// CheckObjCString - Checks that the argument to the builtin 401/// CFString constructor is correct 402/// FIXME: GCC currently emits the following warning: 403/// "warning: input conversion stopped due to an input byte that does not 404/// belong to the input codeset UTF-8" 405/// Note: It might also make sense to do the UTF-16 conversion here (would 406/// simplify the backend). 407bool Sema::CheckObjCString(Expr *Arg) { 408 Arg = Arg->IgnoreParenCasts(); 409 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 410 411 if (!Literal || Literal->isWide()) { 412 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 413 << Arg->getSourceRange(); 414 return true; 415 } 416 417 const char *Data = Literal->getStrData(); 418 unsigned Length = Literal->getByteLength(); 419 420 for (unsigned i = 0; i < Length; ++i) { 421 if (!Data[i]) { 422 Diag(getLocationOfStringLiteralByte(Literal, i), 423 diag::warn_cfstring_literal_contains_nul_character) 424 << Arg->getSourceRange(); 425 break; 426 } 427 } 428 429 return false; 430} 431 432/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 433/// Emit an error and return true on failure, return false on success. 434bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 435 Expr *Fn = TheCall->getCallee(); 436 if (TheCall->getNumArgs() > 2) { 437 Diag(TheCall->getArg(2)->getLocStart(), 438 diag::err_typecheck_call_too_many_args) 439 << 0 /*function call*/ << Fn->getSourceRange() 440 << SourceRange(TheCall->getArg(2)->getLocStart(), 441 (*(TheCall->arg_end()-1))->getLocEnd()); 442 return true; 443 } 444 445 if (TheCall->getNumArgs() < 2) { 446 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 447 << 0 /*function call*/; 448 } 449 450 // Determine whether the current function is variadic or not. 451 bool isVariadic; 452 if (CurBlock) 453 isVariadic = CurBlock->isVariadic; 454 else if (getCurFunctionDecl()) { 455 if (FunctionProtoType* FTP = 456 dyn_cast<FunctionProtoType>(getCurFunctionDecl()->getType())) 457 isVariadic = FTP->isVariadic(); 458 else 459 isVariadic = false; 460 } else { 461 isVariadic = getCurMethodDecl()->isVariadic(); 462 } 463 464 if (!isVariadic) { 465 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 466 return true; 467 } 468 469 // Verify that the second argument to the builtin is the last argument of the 470 // current function or method. 471 bool SecondArgIsLastNamedArgument = false; 472 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 473 474 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 475 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 476 // FIXME: This isn't correct for methods (results in bogus warning). 477 // Get the last formal in the current function. 478 const ParmVarDecl *LastArg; 479 if (CurBlock) 480 LastArg = *(CurBlock->TheDecl->param_end()-1); 481 else if (FunctionDecl *FD = getCurFunctionDecl()) 482 LastArg = *(FD->param_end()-1); 483 else 484 LastArg = *(getCurMethodDecl()->param_end()-1); 485 SecondArgIsLastNamedArgument = PV == LastArg; 486 } 487 } 488 489 if (!SecondArgIsLastNamedArgument) 490 Diag(TheCall->getArg(1)->getLocStart(), 491 diag::warn_second_parameter_of_va_start_not_last_named_argument); 492 return false; 493} 494 495/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 496/// friends. This is declared to take (...), so we have to check everything. 497bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 498 if (TheCall->getNumArgs() < 2) 499 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 500 << 0 /*function call*/; 501 if (TheCall->getNumArgs() > 2) 502 return Diag(TheCall->getArg(2)->getLocStart(), 503 diag::err_typecheck_call_too_many_args) 504 << 0 /*function call*/ 505 << SourceRange(TheCall->getArg(2)->getLocStart(), 506 (*(TheCall->arg_end()-1))->getLocEnd()); 507 508 Expr *OrigArg0 = TheCall->getArg(0); 509 Expr *OrigArg1 = TheCall->getArg(1); 510 511 // Do standard promotions between the two arguments, returning their common 512 // type. 513 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 514 515 // Make sure any conversions are pushed back into the call; this is 516 // type safe since unordered compare builtins are declared as "_Bool 517 // foo(...)". 518 TheCall->setArg(0, OrigArg0); 519 TheCall->setArg(1, OrigArg1); 520 521 if (OrigArg0->isTypeDependent() || OrigArg1->isTypeDependent()) 522 return false; 523 524 // If the common type isn't a real floating type, then the arguments were 525 // invalid for this operation. 526 if (!Res->isRealFloatingType()) 527 return Diag(OrigArg0->getLocStart(), 528 diag::err_typecheck_call_invalid_ordered_compare) 529 << OrigArg0->getType() << OrigArg1->getType() 530 << SourceRange(OrigArg0->getLocStart(), OrigArg1->getLocEnd()); 531 532 return false; 533} 534 535bool Sema::SemaBuiltinStackAddress(CallExpr *TheCall) { 536 // The signature for these builtins is exact; the only thing we need 537 // to check is that the argument is a constant. 538 SourceLocation Loc; 539 if (!TheCall->getArg(0)->isTypeDependent() && 540 !TheCall->getArg(0)->isValueDependent() && 541 !TheCall->getArg(0)->isIntegerConstantExpr(Context, &Loc)) 542 return Diag(Loc, diag::err_stack_const_level) << TheCall->getSourceRange(); 543 544 return false; 545} 546 547/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 548// This is declared to take (...), so we have to check everything. 549Action::OwningExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 550 if (TheCall->getNumArgs() < 3) 551 return ExprError(Diag(TheCall->getLocEnd(), 552 diag::err_typecheck_call_too_few_args) 553 << 0 /*function call*/ << TheCall->getSourceRange()); 554 555 unsigned numElements = std::numeric_limits<unsigned>::max(); 556 if (!TheCall->getArg(0)->isTypeDependent() && 557 !TheCall->getArg(1)->isTypeDependent()) { 558 QualType FAType = TheCall->getArg(0)->getType(); 559 QualType SAType = TheCall->getArg(1)->getType(); 560 561 if (!FAType->isVectorType() || !SAType->isVectorType()) { 562 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 563 << SourceRange(TheCall->getArg(0)->getLocStart(), 564 TheCall->getArg(1)->getLocEnd()); 565 return ExprError(); 566 } 567 568 if (Context.getCanonicalType(FAType).getUnqualifiedType() != 569 Context.getCanonicalType(SAType).getUnqualifiedType()) { 570 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 571 << SourceRange(TheCall->getArg(0)->getLocStart(), 572 TheCall->getArg(1)->getLocEnd()); 573 return ExprError(); 574 } 575 576 numElements = FAType->getAsVectorType()->getNumElements(); 577 if (TheCall->getNumArgs() != numElements+2) { 578 if (TheCall->getNumArgs() < numElements+2) 579 return ExprError(Diag(TheCall->getLocEnd(), 580 diag::err_typecheck_call_too_few_args) 581 << 0 /*function call*/ << TheCall->getSourceRange()); 582 return ExprError(Diag(TheCall->getLocEnd(), 583 diag::err_typecheck_call_too_many_args) 584 << 0 /*function call*/ << TheCall->getSourceRange()); 585 } 586 } 587 588 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 589 if (TheCall->getArg(i)->isTypeDependent() || 590 TheCall->getArg(i)->isValueDependent()) 591 continue; 592 593 llvm::APSInt Result(32); 594 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 595 return ExprError(Diag(TheCall->getLocStart(), 596 diag::err_shufflevector_nonconstant_argument) 597 << TheCall->getArg(i)->getSourceRange()); 598 599 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 600 return ExprError(Diag(TheCall->getLocStart(), 601 diag::err_shufflevector_argument_too_large) 602 << TheCall->getArg(i)->getSourceRange()); 603 } 604 605 llvm::SmallVector<Expr*, 32> exprs; 606 607 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 608 exprs.push_back(TheCall->getArg(i)); 609 TheCall->setArg(i, 0); 610 } 611 612 return Owned(new (Context) ShuffleVectorExpr(exprs.begin(), exprs.size(), 613 exprs[0]->getType(), 614 TheCall->getCallee()->getLocStart(), 615 TheCall->getRParenLoc())); 616} 617 618/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 619// This is declared to take (const void*, ...) and can take two 620// optional constant int args. 621bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 622 unsigned NumArgs = TheCall->getNumArgs(); 623 624 if (NumArgs > 3) 625 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_many_args) 626 << 0 /*function call*/ << TheCall->getSourceRange(); 627 628 // Argument 0 is checked for us and the remaining arguments must be 629 // constant integers. 630 for (unsigned i = 1; i != NumArgs; ++i) { 631 Expr *Arg = TheCall->getArg(i); 632 if (Arg->isTypeDependent()) 633 continue; 634 635 QualType RWType = Arg->getType(); 636 637 const BuiltinType *BT = RWType->getAsBuiltinType(); 638 llvm::APSInt Result; 639 if (!BT || BT->getKind() != BuiltinType::Int) 640 return Diag(TheCall->getLocStart(), diag::err_prefetch_invalid_argument) 641 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 642 643 if (Arg->isValueDependent()) 644 continue; 645 646 if (!Arg->isIntegerConstantExpr(Result, Context)) 647 return Diag(TheCall->getLocStart(), diag::err_prefetch_invalid_argument) 648 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 649 650 // FIXME: gcc issues a warning and rewrites these to 0. These 651 // seems especially odd for the third argument since the default 652 // is 3. 653 if (i == 1) { 654 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 1) 655 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 656 << "0" << "1" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 657 } else { 658 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) 659 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 660 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 661 } 662 } 663 664 return false; 665} 666 667/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 668/// int type). This simply type checks that type is one of the defined 669/// constants (0-3). 670bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 671 Expr *Arg = TheCall->getArg(1); 672 if (Arg->isTypeDependent()) 673 return false; 674 675 QualType ArgType = Arg->getType(); 676 const BuiltinType *BT = ArgType->getAsBuiltinType(); 677 llvm::APSInt Result(32); 678 if (!BT || BT->getKind() != BuiltinType::Int) 679 return Diag(TheCall->getLocStart(), diag::err_object_size_invalid_argument) 680 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 681 682 if (Arg->isValueDependent()) 683 return false; 684 685 if (!Arg->isIntegerConstantExpr(Result, Context)) { 686 return Diag(TheCall->getLocStart(), diag::err_object_size_invalid_argument) 687 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 688 } 689 690 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 691 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 692 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 693 } 694 695 return false; 696} 697 698/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 699/// This checks that val is a constant 1. 700bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 701 Expr *Arg = TheCall->getArg(1); 702 if (Arg->isTypeDependent() || Arg->isValueDependent()) 703 return false; 704 705 llvm::APSInt Result(32); 706 if (!Arg->isIntegerConstantExpr(Result, Context) || Result != 1) 707 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 708 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 709 710 return false; 711} 712 713// Handle i > 1 ? "x" : "y", recursivelly 714bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 715 bool HasVAListArg, 716 unsigned format_idx, unsigned firstDataArg) { 717 if (E->isTypeDependent() || E->isValueDependent()) 718 return false; 719 720 switch (E->getStmtClass()) { 721 case Stmt::ConditionalOperatorClass: { 722 const ConditionalOperator *C = cast<ConditionalOperator>(E); 723 return SemaCheckStringLiteral(C->getLHS(), TheCall, 724 HasVAListArg, format_idx, firstDataArg) 725 && SemaCheckStringLiteral(C->getRHS(), TheCall, 726 HasVAListArg, format_idx, firstDataArg); 727 } 728 729 case Stmt::ImplicitCastExprClass: { 730 const ImplicitCastExpr *Expr = cast<ImplicitCastExpr>(E); 731 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 732 format_idx, firstDataArg); 733 } 734 735 case Stmt::ParenExprClass: { 736 const ParenExpr *Expr = cast<ParenExpr>(E); 737 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 738 format_idx, firstDataArg); 739 } 740 741 case Stmt::DeclRefExprClass: { 742 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 743 744 // As an exception, do not flag errors for variables binding to 745 // const string literals. 746 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 747 bool isConstant = false; 748 QualType T = DR->getType(); 749 750 if (const ArrayType *AT = Context.getAsArrayType(T)) { 751 isConstant = AT->getElementType().isConstant(Context); 752 } 753 else if (const PointerType *PT = T->getAs<PointerType>()) { 754 isConstant = T.isConstant(Context) && 755 PT->getPointeeType().isConstant(Context); 756 } 757 758 if (isConstant) { 759 const VarDecl *Def = 0; 760 if (const Expr *Init = VD->getDefinition(Def)) 761 return SemaCheckStringLiteral(Init, TheCall, 762 HasVAListArg, format_idx, firstDataArg); 763 } 764 765 // For vprintf* functions (i.e., HasVAListArg==true), we add a 766 // special check to see if the format string is a function parameter 767 // of the function calling the printf function. If the function 768 // has an attribute indicating it is a printf-like function, then we 769 // should suppress warnings concerning non-literals being used in a call 770 // to a vprintf function. For example: 771 // 772 // void 773 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 774 // va_list ap; 775 // va_start(ap, fmt); 776 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 777 // ... 778 // 779 // 780 // FIXME: We don't have full attribute support yet, so just check to see 781 // if the argument is a DeclRefExpr that references a parameter. We'll 782 // add proper support for checking the attribute later. 783 if (HasVAListArg) 784 if (isa<ParmVarDecl>(VD)) 785 return true; 786 } 787 788 return false; 789 } 790 791 case Stmt::CallExprClass: { 792 const CallExpr *CE = cast<CallExpr>(E); 793 if (const ImplicitCastExpr *ICE 794 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 795 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 796 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 797 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 798 unsigned ArgIndex = FA->getFormatIdx(); 799 const Expr *Arg = CE->getArg(ArgIndex - 1); 800 801 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 802 format_idx, firstDataArg); 803 } 804 } 805 } 806 } 807 808 return false; 809 } 810 case Stmt::ObjCStringLiteralClass: 811 case Stmt::StringLiteralClass: { 812 const StringLiteral *StrE = NULL; 813 814 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 815 StrE = ObjCFExpr->getString(); 816 else 817 StrE = cast<StringLiteral>(E); 818 819 if (StrE) { 820 CheckPrintfString(StrE, E, TheCall, HasVAListArg, format_idx, 821 firstDataArg); 822 return true; 823 } 824 825 return false; 826 } 827 828 default: 829 return false; 830 } 831} 832 833void 834Sema::CheckNonNullArguments(const NonNullAttr *NonNull, const CallExpr *TheCall) 835{ 836 for (NonNullAttr::iterator i = NonNull->begin(), e = NonNull->end(); 837 i != e; ++i) { 838 const Expr *ArgExpr = TheCall->getArg(*i); 839 if (ArgExpr->isNullPointerConstant(Context)) 840 Diag(TheCall->getCallee()->getLocStart(), diag::warn_null_arg) 841 << ArgExpr->getSourceRange(); 842 } 843} 844 845/// CheckPrintfArguments - Check calls to printf (and similar functions) for 846/// correct use of format strings. 847/// 848/// HasVAListArg - A predicate indicating whether the printf-like 849/// function is passed an explicit va_arg argument (e.g., vprintf) 850/// 851/// format_idx - The index into Args for the format string. 852/// 853/// Improper format strings to functions in the printf family can be 854/// the source of bizarre bugs and very serious security holes. A 855/// good source of information is available in the following paper 856/// (which includes additional references): 857/// 858/// FormatGuard: Automatic Protection From printf Format String 859/// Vulnerabilities, Proceedings of the 10th USENIX Security Symposium, 2001. 860/// 861/// Functionality implemented: 862/// 863/// We can statically check the following properties for string 864/// literal format strings for non v.*printf functions (where the 865/// arguments are passed directly): 866// 867/// (1) Are the number of format conversions equal to the number of 868/// data arguments? 869/// 870/// (2) Does each format conversion correctly match the type of the 871/// corresponding data argument? (TODO) 872/// 873/// Moreover, for all printf functions we can: 874/// 875/// (3) Check for a missing format string (when not caught by type checking). 876/// 877/// (4) Check for no-operation flags; e.g. using "#" with format 878/// conversion 'c' (TODO) 879/// 880/// (5) Check the use of '%n', a major source of security holes. 881/// 882/// (6) Check for malformed format conversions that don't specify anything. 883/// 884/// (7) Check for empty format strings. e.g: printf(""); 885/// 886/// (8) Check that the format string is a wide literal. 887/// 888/// (9) Also check the arguments of functions with the __format__ attribute. 889/// (TODO). 890/// 891/// All of these checks can be done by parsing the format string. 892/// 893/// For now, we ONLY do (1), (3), (5), (6), (7), and (8). 894void 895Sema::CheckPrintfArguments(const CallExpr *TheCall, bool HasVAListArg, 896 unsigned format_idx, unsigned firstDataArg) { 897 const Expr *Fn = TheCall->getCallee(); 898 899 // CHECK: printf-like function is called with no format string. 900 if (format_idx >= TheCall->getNumArgs()) { 901 Diag(TheCall->getRParenLoc(), diag::warn_printf_missing_format_string) 902 << Fn->getSourceRange(); 903 return; 904 } 905 906 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 907 908 // CHECK: format string is not a string literal. 909 // 910 // Dynamically generated format strings are difficult to 911 // automatically vet at compile time. Requiring that format strings 912 // are string literals: (1) permits the checking of format strings by 913 // the compiler and thereby (2) can practically remove the source of 914 // many format string exploits. 915 916 // Format string can be either ObjC string (e.g. @"%d") or 917 // C string (e.g. "%d") 918 // ObjC string uses the same format specifiers as C string, so we can use 919 // the same format string checking logic for both ObjC and C strings. 920 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 921 firstDataArg)) 922 return; // Literal format string found, check done! 923 924 // If there are no arguments specified, warn with -Wformat-security, otherwise 925 // warn only with -Wformat-nonliteral. 926 if (TheCall->getNumArgs() == format_idx+1) 927 Diag(TheCall->getArg(format_idx)->getLocStart(), 928 diag::warn_printf_nonliteral_noargs) 929 << OrigFormatExpr->getSourceRange(); 930 else 931 Diag(TheCall->getArg(format_idx)->getLocStart(), 932 diag::warn_printf_nonliteral) 933 << OrigFormatExpr->getSourceRange(); 934} 935 936void Sema::CheckPrintfString(const StringLiteral *FExpr, 937 const Expr *OrigFormatExpr, 938 const CallExpr *TheCall, bool HasVAListArg, 939 unsigned format_idx, unsigned firstDataArg) { 940 941 const ObjCStringLiteral *ObjCFExpr = 942 dyn_cast<ObjCStringLiteral>(OrigFormatExpr); 943 944 // CHECK: is the format string a wide literal? 945 if (FExpr->isWide()) { 946 Diag(FExpr->getLocStart(), 947 diag::warn_printf_format_string_is_wide_literal) 948 << OrigFormatExpr->getSourceRange(); 949 return; 950 } 951 952 // Str - The format string. NOTE: this is NOT null-terminated! 953 const char *Str = FExpr->getStrData(); 954 955 // CHECK: empty format string? 956 unsigned StrLen = FExpr->getByteLength(); 957 958 if (StrLen == 0) { 959 Diag(FExpr->getLocStart(), diag::warn_printf_empty_format_string) 960 << OrigFormatExpr->getSourceRange(); 961 return; 962 } 963 964 // We process the format string using a binary state machine. The 965 // current state is stored in CurrentState. 966 enum { 967 state_OrdChr, 968 state_Conversion 969 } CurrentState = state_OrdChr; 970 971 // numConversions - The number of conversions seen so far. This is 972 // incremented as we traverse the format string. 973 unsigned numConversions = 0; 974 975 // numDataArgs - The number of data arguments after the format 976 // string. This can only be determined for non vprintf-like 977 // functions. For those functions, this value is 1 (the sole 978 // va_arg argument). 979 unsigned numDataArgs = TheCall->getNumArgs()-firstDataArg; 980 981 // Inspect the format string. 982 unsigned StrIdx = 0; 983 984 // LastConversionIdx - Index within the format string where we last saw 985 // a '%' character that starts a new format conversion. 986 unsigned LastConversionIdx = 0; 987 988 for (; StrIdx < StrLen; ++StrIdx) { 989 990 // Is the number of detected conversion conversions greater than 991 // the number of matching data arguments? If so, stop. 992 if (!HasVAListArg && numConversions > numDataArgs) break; 993 994 // Handle "\0" 995 if (Str[StrIdx] == '\0') { 996 // The string returned by getStrData() is not null-terminated, 997 // so the presence of a null character is likely an error. 998 Diag(getLocationOfStringLiteralByte(FExpr, StrIdx), 999 diag::warn_printf_format_string_contains_null_char) 1000 << OrigFormatExpr->getSourceRange(); 1001 return; 1002 } 1003 1004 // Ordinary characters (not processing a format conversion). 1005 if (CurrentState == state_OrdChr) { 1006 if (Str[StrIdx] == '%') { 1007 CurrentState = state_Conversion; 1008 LastConversionIdx = StrIdx; 1009 } 1010 continue; 1011 } 1012 1013 // Seen '%'. Now processing a format conversion. 1014 switch (Str[StrIdx]) { 1015 // Handle dynamic precision or width specifier. 1016 case '*': { 1017 ++numConversions; 1018 1019 if (!HasVAListArg) { 1020 if (numConversions > numDataArgs) { 1021 SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, StrIdx); 1022 1023 if (Str[StrIdx-1] == '.') 1024 Diag(Loc, diag::warn_printf_asterisk_precision_missing_arg) 1025 << OrigFormatExpr->getSourceRange(); 1026 else 1027 Diag(Loc, diag::warn_printf_asterisk_width_missing_arg) 1028 << OrigFormatExpr->getSourceRange(); 1029 1030 // Don't do any more checking. We'll just emit spurious errors. 1031 return; 1032 } 1033 1034 // Perform type checking on width/precision specifier. 1035 const Expr *E = TheCall->getArg(format_idx+numConversions); 1036 if (const BuiltinType *BT = E->getType()->getAsBuiltinType()) 1037 if (BT->getKind() == BuiltinType::Int) 1038 break; 1039 1040 SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, StrIdx); 1041 1042 if (Str[StrIdx-1] == '.') 1043 Diag(Loc, diag::warn_printf_asterisk_precision_wrong_type) 1044 << E->getType() << E->getSourceRange(); 1045 else 1046 Diag(Loc, diag::warn_printf_asterisk_width_wrong_type) 1047 << E->getType() << E->getSourceRange(); 1048 1049 break; 1050 } 1051 } 1052 1053 // Characters which can terminate a format conversion 1054 // (e.g. "%d"). Characters that specify length modifiers or 1055 // other flags are handled by the default case below. 1056 // 1057 // FIXME: additional checks will go into the following cases. 1058 case 'i': 1059 case 'd': 1060 case 'o': 1061 case 'u': 1062 case 'x': 1063 case 'X': 1064 case 'D': 1065 case 'O': 1066 case 'U': 1067 case 'e': 1068 case 'E': 1069 case 'f': 1070 case 'F': 1071 case 'g': 1072 case 'G': 1073 case 'a': 1074 case 'A': 1075 case 'c': 1076 case 'C': 1077 case 'S': 1078 case 's': 1079 case 'p': 1080 ++numConversions; 1081 CurrentState = state_OrdChr; 1082 break; 1083 1084 case 'm': 1085 // FIXME: Warn in situations where this isn't supported! 1086 CurrentState = state_OrdChr; 1087 break; 1088 1089 // CHECK: Are we using "%n"? Issue a warning. 1090 case 'n': { 1091 ++numConversions; 1092 CurrentState = state_OrdChr; 1093 SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, 1094 LastConversionIdx); 1095 1096 Diag(Loc, diag::warn_printf_write_back)<<OrigFormatExpr->getSourceRange(); 1097 break; 1098 } 1099 1100 // Handle "%@" 1101 case '@': 1102 // %@ is allowed in ObjC format strings only. 1103 if(ObjCFExpr != NULL) 1104 CurrentState = state_OrdChr; 1105 else { 1106 // Issue a warning: invalid format conversion. 1107 SourceLocation Loc = 1108 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 1109 1110 Diag(Loc, diag::warn_printf_invalid_conversion) 1111 << std::string(Str+LastConversionIdx, 1112 Str+std::min(LastConversionIdx+2, StrLen)) 1113 << OrigFormatExpr->getSourceRange(); 1114 } 1115 ++numConversions; 1116 break; 1117 1118 // Handle "%%" 1119 case '%': 1120 // Sanity check: Was the first "%" character the previous one? 1121 // If not, we will assume that we have a malformed format 1122 // conversion, and that the current "%" character is the start 1123 // of a new conversion. 1124 if (StrIdx - LastConversionIdx == 1) 1125 CurrentState = state_OrdChr; 1126 else { 1127 // Issue a warning: invalid format conversion. 1128 SourceLocation Loc = 1129 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 1130 1131 Diag(Loc, diag::warn_printf_invalid_conversion) 1132 << std::string(Str+LastConversionIdx, Str+StrIdx) 1133 << OrigFormatExpr->getSourceRange(); 1134 1135 // This conversion is broken. Advance to the next format 1136 // conversion. 1137 LastConversionIdx = StrIdx; 1138 ++numConversions; 1139 } 1140 break; 1141 1142 default: 1143 // This case catches all other characters: flags, widths, etc. 1144 // We should eventually process those as well. 1145 break; 1146 } 1147 } 1148 1149 if (CurrentState == state_Conversion) { 1150 // Issue a warning: invalid format conversion. 1151 SourceLocation Loc = 1152 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 1153 1154 Diag(Loc, diag::warn_printf_invalid_conversion) 1155 << std::string(Str+LastConversionIdx, 1156 Str+std::min(LastConversionIdx+2, StrLen)) 1157 << OrigFormatExpr->getSourceRange(); 1158 return; 1159 } 1160 1161 if (!HasVAListArg) { 1162 // CHECK: Does the number of format conversions exceed the number 1163 // of data arguments? 1164 if (numConversions > numDataArgs) { 1165 SourceLocation Loc = 1166 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 1167 1168 Diag(Loc, diag::warn_printf_insufficient_data_args) 1169 << OrigFormatExpr->getSourceRange(); 1170 } 1171 // CHECK: Does the number of data arguments exceed the number of 1172 // format conversions in the format string? 1173 else if (numConversions < numDataArgs) 1174 Diag(TheCall->getArg(format_idx+numConversions+1)->getLocStart(), 1175 diag::warn_printf_too_many_data_args) 1176 << OrigFormatExpr->getSourceRange(); 1177 } 1178} 1179 1180//===--- CHECK: Return Address of Stack Variable --------------------------===// 1181 1182static DeclRefExpr* EvalVal(Expr *E); 1183static DeclRefExpr* EvalAddr(Expr* E); 1184 1185/// CheckReturnStackAddr - Check if a return statement returns the address 1186/// of a stack variable. 1187void 1188Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 1189 SourceLocation ReturnLoc) { 1190 1191 // Perform checking for returned stack addresses. 1192 if (lhsType->isPointerType() || lhsType->isBlockPointerType()) { 1193 if (DeclRefExpr *DR = EvalAddr(RetValExp)) 1194 Diag(DR->getLocStart(), diag::warn_ret_stack_addr) 1195 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1196 1197 // Skip over implicit cast expressions when checking for block expressions. 1198 if (ImplicitCastExpr *IcExpr = 1199 dyn_cast_or_null<ImplicitCastExpr>(RetValExp)) 1200 RetValExp = IcExpr->getSubExpr(); 1201 1202 if (BlockExpr *C = dyn_cast_or_null<BlockExpr>(RetValExp)) 1203 if (C->hasBlockDeclRefExprs()) 1204 Diag(C->getLocStart(), diag::err_ret_local_block) 1205 << C->getSourceRange(); 1206 } 1207 // Perform checking for stack values returned by reference. 1208 else if (lhsType->isReferenceType()) { 1209 // Check for a reference to the stack 1210 if (DeclRefExpr *DR = EvalVal(RetValExp)) 1211 Diag(DR->getLocStart(), diag::warn_ret_stack_ref) 1212 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1213 } 1214} 1215 1216/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 1217/// check if the expression in a return statement evaluates to an address 1218/// to a location on the stack. The recursion is used to traverse the 1219/// AST of the return expression, with recursion backtracking when we 1220/// encounter a subexpression that (1) clearly does not lead to the address 1221/// of a stack variable or (2) is something we cannot determine leads to 1222/// the address of a stack variable based on such local checking. 1223/// 1224/// EvalAddr processes expressions that are pointers that are used as 1225/// references (and not L-values). EvalVal handles all other values. 1226/// At the base case of the recursion is a check for a DeclRefExpr* in 1227/// the refers to a stack variable. 1228/// 1229/// This implementation handles: 1230/// 1231/// * pointer-to-pointer casts 1232/// * implicit conversions from array references to pointers 1233/// * taking the address of fields 1234/// * arbitrary interplay between "&" and "*" operators 1235/// * pointer arithmetic from an address of a stack variable 1236/// * taking the address of an array element where the array is on the stack 1237static DeclRefExpr* EvalAddr(Expr *E) { 1238 // We should only be called for evaluating pointer expressions. 1239 assert((E->getType()->isPointerType() || 1240 E->getType()->isBlockPointerType() || 1241 E->getType()->isObjCQualifiedIdType()) && 1242 "EvalAddr only works on pointers"); 1243 1244 // Our "symbolic interpreter" is just a dispatch off the currently 1245 // viewed AST node. We then recursively traverse the AST by calling 1246 // EvalAddr and EvalVal appropriately. 1247 switch (E->getStmtClass()) { 1248 case Stmt::ParenExprClass: 1249 // Ignore parentheses. 1250 return EvalAddr(cast<ParenExpr>(E)->getSubExpr()); 1251 1252 case Stmt::UnaryOperatorClass: { 1253 // The only unary operator that make sense to handle here 1254 // is AddrOf. All others don't make sense as pointers. 1255 UnaryOperator *U = cast<UnaryOperator>(E); 1256 1257 if (U->getOpcode() == UnaryOperator::AddrOf) 1258 return EvalVal(U->getSubExpr()); 1259 else 1260 return NULL; 1261 } 1262 1263 case Stmt::BinaryOperatorClass: { 1264 // Handle pointer arithmetic. All other binary operators are not valid 1265 // in this context. 1266 BinaryOperator *B = cast<BinaryOperator>(E); 1267 BinaryOperator::Opcode op = B->getOpcode(); 1268 1269 if (op != BinaryOperator::Add && op != BinaryOperator::Sub) 1270 return NULL; 1271 1272 Expr *Base = B->getLHS(); 1273 1274 // Determine which argument is the real pointer base. It could be 1275 // the RHS argument instead of the LHS. 1276 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 1277 1278 assert (Base->getType()->isPointerType()); 1279 return EvalAddr(Base); 1280 } 1281 1282 // For conditional operators we need to see if either the LHS or RHS are 1283 // valid DeclRefExpr*s. If one of them is valid, we return it. 1284 case Stmt::ConditionalOperatorClass: { 1285 ConditionalOperator *C = cast<ConditionalOperator>(E); 1286 1287 // Handle the GNU extension for missing LHS. 1288 if (Expr *lhsExpr = C->getLHS()) 1289 if (DeclRefExpr* LHS = EvalAddr(lhsExpr)) 1290 return LHS; 1291 1292 return EvalAddr(C->getRHS()); 1293 } 1294 1295 // For casts, we need to handle conversions from arrays to 1296 // pointer values, and pointer-to-pointer conversions. 1297 case Stmt::ImplicitCastExprClass: 1298 case Stmt::CStyleCastExprClass: 1299 case Stmt::CXXFunctionalCastExprClass: { 1300 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 1301 QualType T = SubExpr->getType(); 1302 1303 if (SubExpr->getType()->isPointerType() || 1304 SubExpr->getType()->isBlockPointerType() || 1305 SubExpr->getType()->isObjCQualifiedIdType()) 1306 return EvalAddr(SubExpr); 1307 else if (T->isArrayType()) 1308 return EvalVal(SubExpr); 1309 else 1310 return 0; 1311 } 1312 1313 // C++ casts. For dynamic casts, static casts, and const casts, we 1314 // are always converting from a pointer-to-pointer, so we just blow 1315 // through the cast. In the case the dynamic cast doesn't fail (and 1316 // return NULL), we take the conservative route and report cases 1317 // where we return the address of a stack variable. For Reinterpre 1318 // FIXME: The comment about is wrong; we're not always converting 1319 // from pointer to pointer. I'm guessing that this code should also 1320 // handle references to objects. 1321 case Stmt::CXXStaticCastExprClass: 1322 case Stmt::CXXDynamicCastExprClass: 1323 case Stmt::CXXConstCastExprClass: 1324 case Stmt::CXXReinterpretCastExprClass: { 1325 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 1326 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 1327 return EvalAddr(S); 1328 else 1329 return NULL; 1330 } 1331 1332 // Everything else: we simply don't reason about them. 1333 default: 1334 return NULL; 1335 } 1336} 1337 1338 1339/// EvalVal - This function is complements EvalAddr in the mutual recursion. 1340/// See the comments for EvalAddr for more details. 1341static DeclRefExpr* EvalVal(Expr *E) { 1342 1343 // We should only be called for evaluating non-pointer expressions, or 1344 // expressions with a pointer type that are not used as references but instead 1345 // are l-values (e.g., DeclRefExpr with a pointer type). 1346 1347 // Our "symbolic interpreter" is just a dispatch off the currently 1348 // viewed AST node. We then recursively traverse the AST by calling 1349 // EvalAddr and EvalVal appropriately. 1350 switch (E->getStmtClass()) { 1351 case Stmt::DeclRefExprClass: 1352 case Stmt::QualifiedDeclRefExprClass: { 1353 // DeclRefExpr: the base case. When we hit a DeclRefExpr we are looking 1354 // at code that refers to a variable's name. We check if it has local 1355 // storage within the function, and if so, return the expression. 1356 DeclRefExpr *DR = cast<DeclRefExpr>(E); 1357 1358 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 1359 if(V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR; 1360 1361 return NULL; 1362 } 1363 1364 case Stmt::ParenExprClass: 1365 // Ignore parentheses. 1366 return EvalVal(cast<ParenExpr>(E)->getSubExpr()); 1367 1368 case Stmt::UnaryOperatorClass: { 1369 // The only unary operator that make sense to handle here 1370 // is Deref. All others don't resolve to a "name." This includes 1371 // handling all sorts of rvalues passed to a unary operator. 1372 UnaryOperator *U = cast<UnaryOperator>(E); 1373 1374 if (U->getOpcode() == UnaryOperator::Deref) 1375 return EvalAddr(U->getSubExpr()); 1376 1377 return NULL; 1378 } 1379 1380 case Stmt::ArraySubscriptExprClass: { 1381 // Array subscripts are potential references to data on the stack. We 1382 // retrieve the DeclRefExpr* for the array variable if it indeed 1383 // has local storage. 1384 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase()); 1385 } 1386 1387 case Stmt::ConditionalOperatorClass: { 1388 // For conditional operators we need to see if either the LHS or RHS are 1389 // non-NULL DeclRefExpr's. If one is non-NULL, we return it. 1390 ConditionalOperator *C = cast<ConditionalOperator>(E); 1391 1392 // Handle the GNU extension for missing LHS. 1393 if (Expr *lhsExpr = C->getLHS()) 1394 if (DeclRefExpr *LHS = EvalVal(lhsExpr)) 1395 return LHS; 1396 1397 return EvalVal(C->getRHS()); 1398 } 1399 1400 // Accesses to members are potential references to data on the stack. 1401 case Stmt::MemberExprClass: { 1402 MemberExpr *M = cast<MemberExpr>(E); 1403 1404 // Check for indirect access. We only want direct field accesses. 1405 if (!M->isArrow()) 1406 return EvalVal(M->getBase()); 1407 else 1408 return NULL; 1409 } 1410 1411 // Everything else: we simply don't reason about them. 1412 default: 1413 return NULL; 1414 } 1415} 1416 1417//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 1418 1419/// Check for comparisons of floating point operands using != and ==. 1420/// Issue a warning if these are no self-comparisons, as they are not likely 1421/// to do what the programmer intended. 1422void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 1423 bool EmitWarning = true; 1424 1425 Expr* LeftExprSansParen = lex->IgnoreParens(); 1426 Expr* RightExprSansParen = rex->IgnoreParens(); 1427 1428 // Special case: check for x == x (which is OK). 1429 // Do not emit warnings for such cases. 1430 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 1431 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 1432 if (DRL->getDecl() == DRR->getDecl()) 1433 EmitWarning = false; 1434 1435 1436 // Special case: check for comparisons against literals that can be exactly 1437 // represented by APFloat. In such cases, do not emit a warning. This 1438 // is a heuristic: often comparison against such literals are used to 1439 // detect if a value in a variable has not changed. This clearly can 1440 // lead to false negatives. 1441 if (EmitWarning) { 1442 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 1443 if (FLL->isExact()) 1444 EmitWarning = false; 1445 } 1446 else 1447 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 1448 if (FLR->isExact()) 1449 EmitWarning = false; 1450 } 1451 } 1452 1453 // Check for comparisons with builtin types. 1454 if (EmitWarning) 1455 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 1456 if (CL->isBuiltinCall(Context)) 1457 EmitWarning = false; 1458 1459 if (EmitWarning) 1460 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 1461 if (CR->isBuiltinCall(Context)) 1462 EmitWarning = false; 1463 1464 // Emit the diagnostic. 1465 if (EmitWarning) 1466 Diag(loc, diag::warn_floatingpoint_eq) 1467 << lex->getSourceRange() << rex->getSourceRange(); 1468} 1469