SemaChecking.cpp revision 8ecdb65716cd7914ffb2eeee993fa9039fcd31e8
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/Analysis/Analyses/PrintfFormatString.h" 17#include "clang/AST/ASTContext.h" 18#include "clang/AST/CharUnits.h" 19#include "clang/AST/DeclObjC.h" 20#include "clang/AST/ExprCXX.h" 21#include "clang/AST/ExprObjC.h" 22#include "clang/AST/DeclObjC.h" 23#include "clang/AST/StmtCXX.h" 24#include "clang/AST/StmtObjC.h" 25#include "clang/Lex/LiteralSupport.h" 26#include "clang/Lex/Preprocessor.h" 27#include "llvm/ADT/BitVector.h" 28#include "llvm/ADT/STLExtras.h" 29#include "clang/Basic/TargetBuiltins.h" 30#include <limits> 31using namespace clang; 32 33/// getLocationOfStringLiteralByte - Return a source location that points to the 34/// specified byte of the specified string literal. 35/// 36/// Strings are amazingly complex. They can be formed from multiple tokens and 37/// can have escape sequences in them in addition to the usual trigraph and 38/// escaped newline business. This routine handles this complexity. 39/// 40SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 41 unsigned ByteNo) const { 42 assert(!SL->isWide() && "This doesn't work for wide strings yet"); 43 44 // Loop over all of the tokens in this string until we find the one that 45 // contains the byte we're looking for. 46 unsigned TokNo = 0; 47 while (1) { 48 assert(TokNo < SL->getNumConcatenated() && "Invalid byte number!"); 49 SourceLocation StrTokLoc = SL->getStrTokenLoc(TokNo); 50 51 // Get the spelling of the string so that we can get the data that makes up 52 // the string literal, not the identifier for the macro it is potentially 53 // expanded through. 54 SourceLocation StrTokSpellingLoc = SourceMgr.getSpellingLoc(StrTokLoc); 55 56 // Re-lex the token to get its length and original spelling. 57 std::pair<FileID, unsigned> LocInfo = 58 SourceMgr.getDecomposedLoc(StrTokSpellingLoc); 59 bool Invalid = false; 60 llvm::StringRef Buffer = SourceMgr.getBufferData(LocInfo.first, &Invalid); 61 if (Invalid) 62 return StrTokSpellingLoc; 63 64 const char *StrData = Buffer.data()+LocInfo.second; 65 66 // Create a langops struct and enable trigraphs. This is sufficient for 67 // relexing tokens. 68 LangOptions LangOpts; 69 LangOpts.Trigraphs = true; 70 71 // Create a lexer starting at the beginning of this token. 72 Lexer TheLexer(StrTokSpellingLoc, LangOpts, Buffer.begin(), StrData, 73 Buffer.end()); 74 Token TheTok; 75 TheLexer.LexFromRawLexer(TheTok); 76 77 // Use the StringLiteralParser to compute the length of the string in bytes. 78 StringLiteralParser SLP(&TheTok, 1, PP); 79 unsigned TokNumBytes = SLP.GetStringLength(); 80 81 // If the byte is in this token, return the location of the byte. 82 if (ByteNo < TokNumBytes || 83 (ByteNo == TokNumBytes && TokNo == SL->getNumConcatenated())) { 84 unsigned Offset = 85 StringLiteralParser::getOffsetOfStringByte(TheTok, ByteNo, PP); 86 87 // Now that we know the offset of the token in the spelling, use the 88 // preprocessor to get the offset in the original source. 89 return PP.AdvanceToTokenCharacter(StrTokLoc, Offset); 90 } 91 92 // Move to the next string token. 93 ++TokNo; 94 ByteNo -= TokNumBytes; 95 } 96} 97 98/// CheckablePrintfAttr - does a function call have a "printf" attribute 99/// and arguments that merit checking? 100bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) { 101 if (Format->getType() == "printf") return true; 102 if (Format->getType() == "printf0") { 103 // printf0 allows null "format" string; if so don't check format/args 104 unsigned format_idx = Format->getFormatIdx() - 1; 105 // Does the index refer to the implicit object argument? 106 if (isa<CXXMemberCallExpr>(TheCall)) { 107 if (format_idx == 0) 108 return false; 109 --format_idx; 110 } 111 if (format_idx < TheCall->getNumArgs()) { 112 Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts(); 113 if (!Format->isNullPointerConstant(Context, 114 Expr::NPC_ValueDependentIsNull)) 115 return true; 116 } 117 } 118 return false; 119} 120 121Action::OwningExprResult 122Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 123 OwningExprResult TheCallResult(Owned(TheCall)); 124 125 switch (BuiltinID) { 126 case Builtin::BI__builtin___CFStringMakeConstantString: 127 assert(TheCall->getNumArgs() == 1 && 128 "Wrong # arguments to builtin CFStringMakeConstantString"); 129 if (CheckObjCString(TheCall->getArg(0))) 130 return ExprError(); 131 break; 132 case Builtin::BI__builtin_stdarg_start: 133 case Builtin::BI__builtin_va_start: 134 if (SemaBuiltinVAStart(TheCall)) 135 return ExprError(); 136 break; 137 case Builtin::BI__builtin_isgreater: 138 case Builtin::BI__builtin_isgreaterequal: 139 case Builtin::BI__builtin_isless: 140 case Builtin::BI__builtin_islessequal: 141 case Builtin::BI__builtin_islessgreater: 142 case Builtin::BI__builtin_isunordered: 143 if (SemaBuiltinUnorderedCompare(TheCall)) 144 return ExprError(); 145 break; 146 case Builtin::BI__builtin_fpclassify: 147 if (SemaBuiltinFPClassification(TheCall, 6)) 148 return ExprError(); 149 break; 150 case Builtin::BI__builtin_isfinite: 151 case Builtin::BI__builtin_isinf: 152 case Builtin::BI__builtin_isinf_sign: 153 case Builtin::BI__builtin_isnan: 154 case Builtin::BI__builtin_isnormal: 155 if (SemaBuiltinFPClassification(TheCall, 1)) 156 return ExprError(); 157 break; 158 case Builtin::BI__builtin_return_address: 159 case Builtin::BI__builtin_frame_address: { 160 llvm::APSInt Result; 161 if (SemaBuiltinConstantArg(TheCall, 0, Result)) 162 return ExprError(); 163 break; 164 } 165 case Builtin::BI__builtin_eh_return_data_regno: { 166 llvm::APSInt Result; 167 if (SemaBuiltinConstantArg(TheCall, 0, Result)) 168 return ExprError(); 169 break; 170 } 171 case Builtin::BI__builtin_shufflevector: 172 return SemaBuiltinShuffleVector(TheCall); 173 // TheCall will be freed by the smart pointer here, but that's fine, since 174 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 175 case Builtin::BI__builtin_prefetch: 176 if (SemaBuiltinPrefetch(TheCall)) 177 return ExprError(); 178 break; 179 case Builtin::BI__builtin_object_size: 180 if (SemaBuiltinObjectSize(TheCall)) 181 return ExprError(); 182 break; 183 case Builtin::BI__builtin_longjmp: 184 if (SemaBuiltinLongjmp(TheCall)) 185 return ExprError(); 186 break; 187 case Builtin::BI__sync_fetch_and_add: 188 case Builtin::BI__sync_fetch_and_sub: 189 case Builtin::BI__sync_fetch_and_or: 190 case Builtin::BI__sync_fetch_and_and: 191 case Builtin::BI__sync_fetch_and_xor: 192 case Builtin::BI__sync_add_and_fetch: 193 case Builtin::BI__sync_sub_and_fetch: 194 case Builtin::BI__sync_and_and_fetch: 195 case Builtin::BI__sync_or_and_fetch: 196 case Builtin::BI__sync_xor_and_fetch: 197 case Builtin::BI__sync_val_compare_and_swap: 198 case Builtin::BI__sync_bool_compare_and_swap: 199 case Builtin::BI__sync_lock_test_and_set: 200 case Builtin::BI__sync_lock_release: 201 if (SemaBuiltinAtomicOverloaded(TheCall)) 202 return ExprError(); 203 break; 204 205 // Target specific builtins start here. 206 case X86::BI__builtin_ia32_palignr128: 207 case X86::BI__builtin_ia32_palignr: { 208 llvm::APSInt Result; 209 if (SemaBuiltinConstantArg(TheCall, 2, Result)) 210 return ExprError(); 211 break; 212 } 213 } 214 215 return move(TheCallResult); 216} 217 218/// CheckFunctionCall - Check a direct function call for various correctness 219/// and safety properties not strictly enforced by the C type system. 220bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 221 // Get the IdentifierInfo* for the called function. 222 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 223 224 // None of the checks below are needed for functions that don't have 225 // simple names (e.g., C++ conversion functions). 226 if (!FnInfo) 227 return false; 228 229 // FIXME: This mechanism should be abstracted to be less fragile and 230 // more efficient. For example, just map function ids to custom 231 // handlers. 232 233 // Printf checking. 234 if (const FormatAttr *Format = FDecl->getAttr<FormatAttr>()) { 235 if (CheckablePrintfAttr(Format, TheCall)) { 236 bool HasVAListArg = Format->getFirstArg() == 0; 237 CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 238 HasVAListArg ? 0 : Format->getFirstArg() - 1); 239 } 240 } 241 242 for (const NonNullAttr *NonNull = FDecl->getAttr<NonNullAttr>(); NonNull; 243 NonNull = NonNull->getNext<NonNullAttr>()) 244 CheckNonNullArguments(NonNull, TheCall); 245 246 return false; 247} 248 249bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 250 // Printf checking. 251 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 252 if (!Format) 253 return false; 254 255 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 256 if (!V) 257 return false; 258 259 QualType Ty = V->getType(); 260 if (!Ty->isBlockPointerType()) 261 return false; 262 263 if (!CheckablePrintfAttr(Format, TheCall)) 264 return false; 265 266 bool HasVAListArg = Format->getFirstArg() == 0; 267 CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 268 HasVAListArg ? 0 : Format->getFirstArg() - 1); 269 270 return false; 271} 272 273/// SemaBuiltinAtomicOverloaded - We have a call to a function like 274/// __sync_fetch_and_add, which is an overloaded function based on the pointer 275/// type of its first argument. The main ActOnCallExpr routines have already 276/// promoted the types of arguments because all of these calls are prototyped as 277/// void(...). 278/// 279/// This function goes through and does final semantic checking for these 280/// builtins, 281bool Sema::SemaBuiltinAtomicOverloaded(CallExpr *TheCall) { 282 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 283 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 284 285 // Ensure that we have at least one argument to do type inference from. 286 if (TheCall->getNumArgs() < 1) 287 return Diag(TheCall->getLocEnd(), 288 diag::err_typecheck_call_too_few_args_at_least) 289 << 0 << 1 << TheCall->getNumArgs() 290 << TheCall->getCallee()->getSourceRange(); 291 292 // Inspect the first argument of the atomic builtin. This should always be 293 // a pointer type, whose element is an integral scalar or pointer type. 294 // Because it is a pointer type, we don't have to worry about any implicit 295 // casts here. 296 Expr *FirstArg = TheCall->getArg(0); 297 if (!FirstArg->getType()->isPointerType()) 298 return Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 299 << FirstArg->getType() << FirstArg->getSourceRange(); 300 301 QualType ValType = FirstArg->getType()->getAs<PointerType>()->getPointeeType(); 302 if (!ValType->isIntegerType() && !ValType->isPointerType() && 303 !ValType->isBlockPointerType()) 304 return Diag(DRE->getLocStart(), 305 diag::err_atomic_builtin_must_be_pointer_intptr) 306 << FirstArg->getType() << FirstArg->getSourceRange(); 307 308 // We need to figure out which concrete builtin this maps onto. For example, 309 // __sync_fetch_and_add with a 2 byte object turns into 310 // __sync_fetch_and_add_2. 311#define BUILTIN_ROW(x) \ 312 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 313 Builtin::BI##x##_8, Builtin::BI##x##_16 } 314 315 static const unsigned BuiltinIndices[][5] = { 316 BUILTIN_ROW(__sync_fetch_and_add), 317 BUILTIN_ROW(__sync_fetch_and_sub), 318 BUILTIN_ROW(__sync_fetch_and_or), 319 BUILTIN_ROW(__sync_fetch_and_and), 320 BUILTIN_ROW(__sync_fetch_and_xor), 321 322 BUILTIN_ROW(__sync_add_and_fetch), 323 BUILTIN_ROW(__sync_sub_and_fetch), 324 BUILTIN_ROW(__sync_and_and_fetch), 325 BUILTIN_ROW(__sync_or_and_fetch), 326 BUILTIN_ROW(__sync_xor_and_fetch), 327 328 BUILTIN_ROW(__sync_val_compare_and_swap), 329 BUILTIN_ROW(__sync_bool_compare_and_swap), 330 BUILTIN_ROW(__sync_lock_test_and_set), 331 BUILTIN_ROW(__sync_lock_release) 332 }; 333#undef BUILTIN_ROW 334 335 // Determine the index of the size. 336 unsigned SizeIndex; 337 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 338 case 1: SizeIndex = 0; break; 339 case 2: SizeIndex = 1; break; 340 case 4: SizeIndex = 2; break; 341 case 8: SizeIndex = 3; break; 342 case 16: SizeIndex = 4; break; 343 default: 344 return Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 345 << FirstArg->getType() << FirstArg->getSourceRange(); 346 } 347 348 // Each of these builtins has one pointer argument, followed by some number of 349 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 350 // that we ignore. Find out which row of BuiltinIndices to read from as well 351 // as the number of fixed args. 352 unsigned BuiltinID = FDecl->getBuiltinID(); 353 unsigned BuiltinIndex, NumFixed = 1; 354 switch (BuiltinID) { 355 default: assert(0 && "Unknown overloaded atomic builtin!"); 356 case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break; 357 case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break; 358 case Builtin::BI__sync_fetch_and_or: BuiltinIndex = 2; break; 359 case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break; 360 case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break; 361 362 case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break; 363 case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break; 364 case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break; 365 case Builtin::BI__sync_or_and_fetch: BuiltinIndex = 8; break; 366 case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break; 367 368 case Builtin::BI__sync_val_compare_and_swap: 369 BuiltinIndex = 10; 370 NumFixed = 2; 371 break; 372 case Builtin::BI__sync_bool_compare_and_swap: 373 BuiltinIndex = 11; 374 NumFixed = 2; 375 break; 376 case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break; 377 case Builtin::BI__sync_lock_release: 378 BuiltinIndex = 13; 379 NumFixed = 0; 380 break; 381 } 382 383 // Now that we know how many fixed arguments we expect, first check that we 384 // have at least that many. 385 if (TheCall->getNumArgs() < 1+NumFixed) 386 return Diag(TheCall->getLocEnd(), 387 diag::err_typecheck_call_too_few_args_at_least) 388 << 0 << 1+NumFixed << TheCall->getNumArgs() 389 << TheCall->getCallee()->getSourceRange(); 390 391 392 // Get the decl for the concrete builtin from this, we can tell what the 393 // concrete integer type we should convert to is. 394 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 395 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 396 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 397 FunctionDecl *NewBuiltinDecl = 398 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 399 TUScope, false, DRE->getLocStart())); 400 const FunctionProtoType *BuiltinFT = 401 NewBuiltinDecl->getType()->getAs<FunctionProtoType>(); 402 ValType = BuiltinFT->getArgType(0)->getAs<PointerType>()->getPointeeType(); 403 404 // If the first type needs to be converted (e.g. void** -> int*), do it now. 405 if (BuiltinFT->getArgType(0) != FirstArg->getType()) { 406 ImpCastExprToType(FirstArg, BuiltinFT->getArgType(0), CastExpr::CK_BitCast); 407 TheCall->setArg(0, FirstArg); 408 } 409 410 // Next, walk the valid ones promoting to the right type. 411 for (unsigned i = 0; i != NumFixed; ++i) { 412 Expr *Arg = TheCall->getArg(i+1); 413 414 // If the argument is an implicit cast, then there was a promotion due to 415 // "...", just remove it now. 416 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) { 417 Arg = ICE->getSubExpr(); 418 ICE->setSubExpr(0); 419 ICE->Destroy(Context); 420 TheCall->setArg(i+1, Arg); 421 } 422 423 // GCC does an implicit conversion to the pointer or integer ValType. This 424 // can fail in some cases (1i -> int**), check for this error case now. 425 CastExpr::CastKind Kind = CastExpr::CK_Unknown; 426 CXXBaseSpecifierArray BasePath; 427 if (CheckCastTypes(Arg->getSourceRange(), ValType, Arg, Kind, BasePath)) 428 return true; 429 430 // Okay, we have something that *can* be converted to the right type. Check 431 // to see if there is a potentially weird extension going on here. This can 432 // happen when you do an atomic operation on something like an char* and 433 // pass in 42. The 42 gets converted to char. This is even more strange 434 // for things like 45.123 -> char, etc. 435 // FIXME: Do this check. 436 ImpCastExprToType(Arg, ValType, Kind); 437 TheCall->setArg(i+1, Arg); 438 } 439 440 // Switch the DeclRefExpr to refer to the new decl. 441 DRE->setDecl(NewBuiltinDecl); 442 DRE->setType(NewBuiltinDecl->getType()); 443 444 // Set the callee in the CallExpr. 445 // FIXME: This leaks the original parens and implicit casts. 446 Expr *PromotedCall = DRE; 447 UsualUnaryConversions(PromotedCall); 448 TheCall->setCallee(PromotedCall); 449 450 451 // Change the result type of the call to match the result type of the decl. 452 TheCall->setType(NewBuiltinDecl->getResultType()); 453 return false; 454} 455 456 457/// CheckObjCString - Checks that the argument to the builtin 458/// CFString constructor is correct 459/// FIXME: GCC currently emits the following warning: 460/// "warning: input conversion stopped due to an input byte that does not 461/// belong to the input codeset UTF-8" 462/// Note: It might also make sense to do the UTF-16 conversion here (would 463/// simplify the backend). 464bool Sema::CheckObjCString(Expr *Arg) { 465 Arg = Arg->IgnoreParenCasts(); 466 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 467 468 if (!Literal || Literal->isWide()) { 469 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 470 << Arg->getSourceRange(); 471 return true; 472 } 473 474 const char *Data = Literal->getStrData(); 475 unsigned Length = Literal->getByteLength(); 476 477 for (unsigned i = 0; i < Length; ++i) { 478 if (!Data[i]) { 479 Diag(getLocationOfStringLiteralByte(Literal, i), 480 diag::warn_cfstring_literal_contains_nul_character) 481 << Arg->getSourceRange(); 482 break; 483 } 484 } 485 486 return false; 487} 488 489/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 490/// Emit an error and return true on failure, return false on success. 491bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 492 Expr *Fn = TheCall->getCallee(); 493 if (TheCall->getNumArgs() > 2) { 494 Diag(TheCall->getArg(2)->getLocStart(), 495 diag::err_typecheck_call_too_many_args) 496 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 497 << Fn->getSourceRange() 498 << SourceRange(TheCall->getArg(2)->getLocStart(), 499 (*(TheCall->arg_end()-1))->getLocEnd()); 500 return true; 501 } 502 503 if (TheCall->getNumArgs() < 2) { 504 return Diag(TheCall->getLocEnd(), 505 diag::err_typecheck_call_too_few_args_at_least) 506 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 507 } 508 509 // Determine whether the current function is variadic or not. 510 BlockScopeInfo *CurBlock = getCurBlock(); 511 bool isVariadic; 512 if (CurBlock) 513 isVariadic = CurBlock->isVariadic; 514 else if (getCurFunctionDecl()) { 515 if (FunctionProtoType* FTP = 516 dyn_cast<FunctionProtoType>(getCurFunctionDecl()->getType())) 517 isVariadic = FTP->isVariadic(); 518 else 519 isVariadic = false; 520 } else { 521 isVariadic = getCurMethodDecl()->isVariadic(); 522 } 523 524 if (!isVariadic) { 525 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 526 return true; 527 } 528 529 // Verify that the second argument to the builtin is the last argument of the 530 // current function or method. 531 bool SecondArgIsLastNamedArgument = false; 532 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 533 534 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 535 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 536 // FIXME: This isn't correct for methods (results in bogus warning). 537 // Get the last formal in the current function. 538 const ParmVarDecl *LastArg; 539 if (CurBlock) 540 LastArg = *(CurBlock->TheDecl->param_end()-1); 541 else if (FunctionDecl *FD = getCurFunctionDecl()) 542 LastArg = *(FD->param_end()-1); 543 else 544 LastArg = *(getCurMethodDecl()->param_end()-1); 545 SecondArgIsLastNamedArgument = PV == LastArg; 546 } 547 } 548 549 if (!SecondArgIsLastNamedArgument) 550 Diag(TheCall->getArg(1)->getLocStart(), 551 diag::warn_second_parameter_of_va_start_not_last_named_argument); 552 return false; 553} 554 555/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 556/// friends. This is declared to take (...), so we have to check everything. 557bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 558 if (TheCall->getNumArgs() < 2) 559 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 560 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 561 if (TheCall->getNumArgs() > 2) 562 return Diag(TheCall->getArg(2)->getLocStart(), 563 diag::err_typecheck_call_too_many_args) 564 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 565 << SourceRange(TheCall->getArg(2)->getLocStart(), 566 (*(TheCall->arg_end()-1))->getLocEnd()); 567 568 Expr *OrigArg0 = TheCall->getArg(0); 569 Expr *OrigArg1 = TheCall->getArg(1); 570 571 // Do standard promotions between the two arguments, returning their common 572 // type. 573 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 574 575 // Make sure any conversions are pushed back into the call; this is 576 // type safe since unordered compare builtins are declared as "_Bool 577 // foo(...)". 578 TheCall->setArg(0, OrigArg0); 579 TheCall->setArg(1, OrigArg1); 580 581 if (OrigArg0->isTypeDependent() || OrigArg1->isTypeDependent()) 582 return false; 583 584 // If the common type isn't a real floating type, then the arguments were 585 // invalid for this operation. 586 if (!Res->isRealFloatingType()) 587 return Diag(OrigArg0->getLocStart(), 588 diag::err_typecheck_call_invalid_ordered_compare) 589 << OrigArg0->getType() << OrigArg1->getType() 590 << SourceRange(OrigArg0->getLocStart(), OrigArg1->getLocEnd()); 591 592 return false; 593} 594 595/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 596/// __builtin_isnan and friends. This is declared to take (...), so we have 597/// to check everything. We expect the last argument to be a floating point 598/// value. 599bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 600 if (TheCall->getNumArgs() < NumArgs) 601 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 602 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 603 if (TheCall->getNumArgs() > NumArgs) 604 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 605 diag::err_typecheck_call_too_many_args) 606 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 607 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 608 (*(TheCall->arg_end()-1))->getLocEnd()); 609 610 Expr *OrigArg = TheCall->getArg(NumArgs-1); 611 612 if (OrigArg->isTypeDependent()) 613 return false; 614 615 // This operation requires a floating-point number 616 if (!OrigArg->getType()->isRealFloatingType()) 617 return Diag(OrigArg->getLocStart(), 618 diag::err_typecheck_call_invalid_unary_fp) 619 << OrigArg->getType() << OrigArg->getSourceRange(); 620 621 return false; 622} 623 624/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 625// This is declared to take (...), so we have to check everything. 626Action::OwningExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 627 if (TheCall->getNumArgs() < 3) 628 return ExprError(Diag(TheCall->getLocEnd(), 629 diag::err_typecheck_call_too_few_args_at_least) 630 << 0 /*function call*/ << 3 << TheCall->getNumArgs() 631 << TheCall->getSourceRange()); 632 633 unsigned numElements = std::numeric_limits<unsigned>::max(); 634 if (!TheCall->getArg(0)->isTypeDependent() && 635 !TheCall->getArg(1)->isTypeDependent()) { 636 QualType FAType = TheCall->getArg(0)->getType(); 637 QualType SAType = TheCall->getArg(1)->getType(); 638 639 if (!FAType->isVectorType() || !SAType->isVectorType()) { 640 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 641 << SourceRange(TheCall->getArg(0)->getLocStart(), 642 TheCall->getArg(1)->getLocEnd()); 643 return ExprError(); 644 } 645 646 if (!Context.hasSameUnqualifiedType(FAType, SAType)) { 647 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 648 << SourceRange(TheCall->getArg(0)->getLocStart(), 649 TheCall->getArg(1)->getLocEnd()); 650 return ExprError(); 651 } 652 653 numElements = FAType->getAs<VectorType>()->getNumElements(); 654 if (TheCall->getNumArgs() != numElements+2) { 655 if (TheCall->getNumArgs() < numElements+2) 656 return ExprError(Diag(TheCall->getLocEnd(), 657 diag::err_typecheck_call_too_few_args) 658 << 0 /*function call*/ 659 << numElements+2 << TheCall->getNumArgs() 660 << TheCall->getSourceRange()); 661 return ExprError(Diag(TheCall->getLocEnd(), 662 diag::err_typecheck_call_too_many_args) 663 << 0 /*function call*/ 664 << numElements+2 << TheCall->getNumArgs() 665 << TheCall->getSourceRange()); 666 } 667 } 668 669 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 670 if (TheCall->getArg(i)->isTypeDependent() || 671 TheCall->getArg(i)->isValueDependent()) 672 continue; 673 674 llvm::APSInt Result; 675 if (SemaBuiltinConstantArg(TheCall, i, Result)) 676 return ExprError(); 677 678 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 679 return ExprError(Diag(TheCall->getLocStart(), 680 diag::err_shufflevector_argument_too_large) 681 << TheCall->getArg(i)->getSourceRange()); 682 } 683 684 llvm::SmallVector<Expr*, 32> exprs; 685 686 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 687 exprs.push_back(TheCall->getArg(i)); 688 TheCall->setArg(i, 0); 689 } 690 691 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 692 exprs.size(), exprs[0]->getType(), 693 TheCall->getCallee()->getLocStart(), 694 TheCall->getRParenLoc())); 695} 696 697/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 698// This is declared to take (const void*, ...) and can take two 699// optional constant int args. 700bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 701 unsigned NumArgs = TheCall->getNumArgs(); 702 703 if (NumArgs > 3) 704 return Diag(TheCall->getLocEnd(), 705 diag::err_typecheck_call_too_many_args_at_most) 706 << 0 /*function call*/ << 3 << NumArgs 707 << TheCall->getSourceRange(); 708 709 // Argument 0 is checked for us and the remaining arguments must be 710 // constant integers. 711 for (unsigned i = 1; i != NumArgs; ++i) { 712 Expr *Arg = TheCall->getArg(i); 713 714 llvm::APSInt Result; 715 if (SemaBuiltinConstantArg(TheCall, i, Result)) 716 return true; 717 718 // FIXME: gcc issues a warning and rewrites these to 0. These 719 // seems especially odd for the third argument since the default 720 // is 3. 721 if (i == 1) { 722 if (Result.getLimitedValue() > 1) 723 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 724 << "0" << "1" << Arg->getSourceRange(); 725 } else { 726 if (Result.getLimitedValue() > 3) 727 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 728 << "0" << "3" << Arg->getSourceRange(); 729 } 730 } 731 732 return false; 733} 734 735/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 736/// TheCall is a constant expression. 737bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 738 llvm::APSInt &Result) { 739 Expr *Arg = TheCall->getArg(ArgNum); 740 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 741 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 742 743 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 744 745 if (!Arg->isIntegerConstantExpr(Result, Context)) 746 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 747 << FDecl->getDeclName() << Arg->getSourceRange(); 748 749 return false; 750} 751 752/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 753/// int type). This simply type checks that type is one of the defined 754/// constants (0-3). 755// For compatability check 0-3, llvm only handles 0 and 2. 756bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 757 llvm::APSInt Result; 758 759 // Check constant-ness first. 760 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 761 return true; 762 763 Expr *Arg = TheCall->getArg(1); 764 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 765 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 766 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 767 } 768 769 return false; 770} 771 772/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 773/// This checks that val is a constant 1. 774bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 775 Expr *Arg = TheCall->getArg(1); 776 llvm::APSInt Result; 777 778 // TODO: This is less than ideal. Overload this to take a value. 779 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 780 return true; 781 782 if (Result != 1) 783 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 784 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 785 786 return false; 787} 788 789// Handle i > 1 ? "x" : "y", recursivelly 790bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 791 bool HasVAListArg, 792 unsigned format_idx, unsigned firstDataArg) { 793 if (E->isTypeDependent() || E->isValueDependent()) 794 return false; 795 796 switch (E->getStmtClass()) { 797 case Stmt::ConditionalOperatorClass: { 798 const ConditionalOperator *C = cast<ConditionalOperator>(E); 799 return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, 800 HasVAListArg, format_idx, firstDataArg) 801 && SemaCheckStringLiteral(C->getRHS(), TheCall, 802 HasVAListArg, format_idx, firstDataArg); 803 } 804 805 case Stmt::ImplicitCastExprClass: { 806 const ImplicitCastExpr *Expr = cast<ImplicitCastExpr>(E); 807 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 808 format_idx, firstDataArg); 809 } 810 811 case Stmt::ParenExprClass: { 812 const ParenExpr *Expr = cast<ParenExpr>(E); 813 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 814 format_idx, firstDataArg); 815 } 816 817 case Stmt::DeclRefExprClass: { 818 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 819 820 // As an exception, do not flag errors for variables binding to 821 // const string literals. 822 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 823 bool isConstant = false; 824 QualType T = DR->getType(); 825 826 if (const ArrayType *AT = Context.getAsArrayType(T)) { 827 isConstant = AT->getElementType().isConstant(Context); 828 } else if (const PointerType *PT = T->getAs<PointerType>()) { 829 isConstant = T.isConstant(Context) && 830 PT->getPointeeType().isConstant(Context); 831 } 832 833 if (isConstant) { 834 if (const Expr *Init = VD->getAnyInitializer()) 835 return SemaCheckStringLiteral(Init, TheCall, 836 HasVAListArg, format_idx, firstDataArg); 837 } 838 839 // For vprintf* functions (i.e., HasVAListArg==true), we add a 840 // special check to see if the format string is a function parameter 841 // of the function calling the printf function. If the function 842 // has an attribute indicating it is a printf-like function, then we 843 // should suppress warnings concerning non-literals being used in a call 844 // to a vprintf function. For example: 845 // 846 // void 847 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 848 // va_list ap; 849 // va_start(ap, fmt); 850 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 851 // ... 852 // 853 // 854 // FIXME: We don't have full attribute support yet, so just check to see 855 // if the argument is a DeclRefExpr that references a parameter. We'll 856 // add proper support for checking the attribute later. 857 if (HasVAListArg) 858 if (isa<ParmVarDecl>(VD)) 859 return true; 860 } 861 862 return false; 863 } 864 865 case Stmt::CallExprClass: { 866 const CallExpr *CE = cast<CallExpr>(E); 867 if (const ImplicitCastExpr *ICE 868 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 869 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 870 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 871 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 872 unsigned ArgIndex = FA->getFormatIdx(); 873 const Expr *Arg = CE->getArg(ArgIndex - 1); 874 875 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 876 format_idx, firstDataArg); 877 } 878 } 879 } 880 } 881 882 return false; 883 } 884 case Stmt::ObjCStringLiteralClass: 885 case Stmt::StringLiteralClass: { 886 const StringLiteral *StrE = NULL; 887 888 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 889 StrE = ObjCFExpr->getString(); 890 else 891 StrE = cast<StringLiteral>(E); 892 893 if (StrE) { 894 CheckPrintfString(StrE, E, TheCall, HasVAListArg, format_idx, 895 firstDataArg); 896 return true; 897 } 898 899 return false; 900 } 901 902 default: 903 return false; 904 } 905} 906 907void 908Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 909 const CallExpr *TheCall) { 910 for (NonNullAttr::iterator i = NonNull->begin(), e = NonNull->end(); 911 i != e; ++i) { 912 const Expr *ArgExpr = TheCall->getArg(*i); 913 if (ArgExpr->isNullPointerConstant(Context, 914 Expr::NPC_ValueDependentIsNotNull)) 915 Diag(TheCall->getCallee()->getLocStart(), diag::warn_null_arg) 916 << ArgExpr->getSourceRange(); 917 } 918} 919 920/// CheckPrintfArguments - Check calls to printf (and similar functions) for 921/// correct use of format strings. 922/// 923/// HasVAListArg - A predicate indicating whether the printf-like 924/// function is passed an explicit va_arg argument (e.g., vprintf) 925/// 926/// format_idx - The index into Args for the format string. 927/// 928/// Improper format strings to functions in the printf family can be 929/// the source of bizarre bugs and very serious security holes. A 930/// good source of information is available in the following paper 931/// (which includes additional references): 932/// 933/// FormatGuard: Automatic Protection From printf Format String 934/// Vulnerabilities, Proceedings of the 10th USENIX Security Symposium, 2001. 935/// 936/// TODO: 937/// Functionality implemented: 938/// 939/// We can statically check the following properties for string 940/// literal format strings for non v.*printf functions (where the 941/// arguments are passed directly): 942// 943/// (1) Are the number of format conversions equal to the number of 944/// data arguments? 945/// 946/// (2) Does each format conversion correctly match the type of the 947/// corresponding data argument? 948/// 949/// Moreover, for all printf functions we can: 950/// 951/// (3) Check for a missing format string (when not caught by type checking). 952/// 953/// (4) Check for no-operation flags; e.g. using "#" with format 954/// conversion 'c' (TODO) 955/// 956/// (5) Check the use of '%n', a major source of security holes. 957/// 958/// (6) Check for malformed format conversions that don't specify anything. 959/// 960/// (7) Check for empty format strings. e.g: printf(""); 961/// 962/// (8) Check that the format string is a wide literal. 963/// 964/// All of these checks can be done by parsing the format string. 965/// 966void 967Sema::CheckPrintfArguments(const CallExpr *TheCall, bool HasVAListArg, 968 unsigned format_idx, unsigned firstDataArg) { 969 const Expr *Fn = TheCall->getCallee(); 970 971 // The way the format attribute works in GCC, the implicit this argument 972 // of member functions is counted. However, it doesn't appear in our own 973 // lists, so decrement format_idx in that case. 974 if (isa<CXXMemberCallExpr>(TheCall)) { 975 // Catch a format attribute mistakenly referring to the object argument. 976 if (format_idx == 0) 977 return; 978 --format_idx; 979 if(firstDataArg != 0) 980 --firstDataArg; 981 } 982 983 // CHECK: printf-like function is called with no format string. 984 if (format_idx >= TheCall->getNumArgs()) { 985 Diag(TheCall->getRParenLoc(), diag::warn_printf_missing_format_string) 986 << Fn->getSourceRange(); 987 return; 988 } 989 990 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 991 992 // CHECK: format string is not a string literal. 993 // 994 // Dynamically generated format strings are difficult to 995 // automatically vet at compile time. Requiring that format strings 996 // are string literals: (1) permits the checking of format strings by 997 // the compiler and thereby (2) can practically remove the source of 998 // many format string exploits. 999 1000 // Format string can be either ObjC string (e.g. @"%d") or 1001 // C string (e.g. "%d") 1002 // ObjC string uses the same format specifiers as C string, so we can use 1003 // the same format string checking logic for both ObjC and C strings. 1004 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 1005 firstDataArg)) 1006 return; // Literal format string found, check done! 1007 1008 // If there are no arguments specified, warn with -Wformat-security, otherwise 1009 // warn only with -Wformat-nonliteral. 1010 if (TheCall->getNumArgs() == format_idx+1) 1011 Diag(TheCall->getArg(format_idx)->getLocStart(), 1012 diag::warn_printf_nonliteral_noargs) 1013 << OrigFormatExpr->getSourceRange(); 1014 else 1015 Diag(TheCall->getArg(format_idx)->getLocStart(), 1016 diag::warn_printf_nonliteral) 1017 << OrigFormatExpr->getSourceRange(); 1018} 1019 1020namespace { 1021class CheckPrintfHandler : public analyze_printf::FormatStringHandler { 1022 Sema &S; 1023 const StringLiteral *FExpr; 1024 const Expr *OrigFormatExpr; 1025 const unsigned FirstDataArg; 1026 const unsigned NumDataArgs; 1027 const bool IsObjCLiteral; 1028 const char *Beg; // Start of format string. 1029 const bool HasVAListArg; 1030 const CallExpr *TheCall; 1031 unsigned FormatIdx; 1032 llvm::BitVector CoveredArgs; 1033 bool usesPositionalArgs; 1034 bool atFirstArg; 1035public: 1036 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1037 const Expr *origFormatExpr, unsigned firstDataArg, 1038 unsigned numDataArgs, bool isObjCLiteral, 1039 const char *beg, bool hasVAListArg, 1040 const CallExpr *theCall, unsigned formatIdx) 1041 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1042 FirstDataArg(firstDataArg), 1043 NumDataArgs(numDataArgs), 1044 IsObjCLiteral(isObjCLiteral), Beg(beg), 1045 HasVAListArg(hasVAListArg), 1046 TheCall(theCall), FormatIdx(formatIdx), 1047 usesPositionalArgs(false), atFirstArg(true) { 1048 CoveredArgs.resize(numDataArgs); 1049 CoveredArgs.reset(); 1050 } 1051 1052 void DoneProcessing(); 1053 1054 void HandleIncompleteFormatSpecifier(const char *startSpecifier, 1055 unsigned specifierLen); 1056 1057 bool 1058 HandleInvalidConversionSpecifier(const analyze_printf::FormatSpecifier &FS, 1059 const char *startSpecifier, 1060 unsigned specifierLen); 1061 1062 virtual void HandleInvalidPosition(const char *startSpecifier, 1063 unsigned specifierLen, 1064 analyze_printf::PositionContext p); 1065 1066 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1067 1068 void HandleNullChar(const char *nullCharacter); 1069 1070 bool HandleFormatSpecifier(const analyze_printf::FormatSpecifier &FS, 1071 const char *startSpecifier, 1072 unsigned specifierLen); 1073private: 1074 SourceRange getFormatStringRange(); 1075 SourceRange getFormatSpecifierRange(const char *startSpecifier, 1076 unsigned specifierLen); 1077 SourceLocation getLocationOfByte(const char *x); 1078 1079 bool HandleAmount(const analyze_printf::OptionalAmount &Amt, unsigned k, 1080 const char *startSpecifier, unsigned specifierLen); 1081 void HandleFlags(const analyze_printf::FormatSpecifier &FS, 1082 llvm::StringRef flag, llvm::StringRef cspec, 1083 const char *startSpecifier, unsigned specifierLen); 1084 1085 const Expr *getDataArg(unsigned i) const; 1086}; 1087} 1088 1089SourceRange CheckPrintfHandler::getFormatStringRange() { 1090 return OrigFormatExpr->getSourceRange(); 1091} 1092 1093SourceRange CheckPrintfHandler:: 1094getFormatSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1095 return SourceRange(getLocationOfByte(startSpecifier), 1096 getLocationOfByte(startSpecifier+specifierLen-1)); 1097} 1098 1099SourceLocation CheckPrintfHandler::getLocationOfByte(const char *x) { 1100 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1101} 1102 1103void CheckPrintfHandler:: 1104HandleIncompleteFormatSpecifier(const char *startSpecifier, 1105 unsigned specifierLen) { 1106 SourceLocation Loc = getLocationOfByte(startSpecifier); 1107 S.Diag(Loc, diag::warn_printf_incomplete_specifier) 1108 << getFormatSpecifierRange(startSpecifier, specifierLen); 1109} 1110 1111void 1112CheckPrintfHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1113 analyze_printf::PositionContext p) { 1114 SourceLocation Loc = getLocationOfByte(startPos); 1115 S.Diag(Loc, diag::warn_printf_invalid_positional_specifier) 1116 << (unsigned) p << getFormatSpecifierRange(startPos, posLen); 1117} 1118 1119void CheckPrintfHandler::HandleZeroPosition(const char *startPos, 1120 unsigned posLen) { 1121 SourceLocation Loc = getLocationOfByte(startPos); 1122 S.Diag(Loc, diag::warn_printf_zero_positional_specifier) 1123 << getFormatSpecifierRange(startPos, posLen); 1124} 1125 1126bool CheckPrintfHandler:: 1127HandleInvalidConversionSpecifier(const analyze_printf::FormatSpecifier &FS, 1128 const char *startSpecifier, 1129 unsigned specifierLen) { 1130 1131 unsigned argIndex = FS.getArgIndex(); 1132 bool keepGoing = true; 1133 if (argIndex < NumDataArgs) { 1134 // Consider the argument coverered, even though the specifier doesn't 1135 // make sense. 1136 CoveredArgs.set(argIndex); 1137 } 1138 else { 1139 // If argIndex exceeds the number of data arguments we 1140 // don't issue a warning because that is just a cascade of warnings (and 1141 // they may have intended '%%' anyway). We don't want to continue processing 1142 // the format string after this point, however, as we will like just get 1143 // gibberish when trying to match arguments. 1144 keepGoing = false; 1145 } 1146 1147 const analyze_printf::ConversionSpecifier &CS = 1148 FS.getConversionSpecifier(); 1149 SourceLocation Loc = getLocationOfByte(CS.getStart()); 1150 S.Diag(Loc, diag::warn_printf_invalid_conversion) 1151 << llvm::StringRef(CS.getStart(), CS.getLength()) 1152 << getFormatSpecifierRange(startSpecifier, specifierLen); 1153 1154 return keepGoing; 1155} 1156 1157void CheckPrintfHandler::HandleNullChar(const char *nullCharacter) { 1158 // The presence of a null character is likely an error. 1159 S.Diag(getLocationOfByte(nullCharacter), 1160 diag::warn_printf_format_string_contains_null_char) 1161 << getFormatStringRange(); 1162} 1163 1164const Expr *CheckPrintfHandler::getDataArg(unsigned i) const { 1165 return TheCall->getArg(FirstDataArg + i); 1166} 1167 1168void CheckPrintfHandler::HandleFlags(const analyze_printf::FormatSpecifier &FS, 1169 llvm::StringRef flag, 1170 llvm::StringRef cspec, 1171 const char *startSpecifier, 1172 unsigned specifierLen) { 1173 const analyze_printf::ConversionSpecifier &CS = FS.getConversionSpecifier(); 1174 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_nonsensical_flag) 1175 << flag << cspec << getFormatSpecifierRange(startSpecifier, specifierLen); 1176} 1177 1178bool 1179CheckPrintfHandler::HandleAmount(const analyze_printf::OptionalAmount &Amt, 1180 unsigned k, const char *startSpecifier, 1181 unsigned specifierLen) { 1182 1183 if (Amt.hasDataArgument()) { 1184 if (!HasVAListArg) { 1185 unsigned argIndex = Amt.getArgIndex(); 1186 if (argIndex >= NumDataArgs) { 1187 S.Diag(getLocationOfByte(Amt.getStart()), 1188 diag::warn_printf_asterisk_missing_arg) 1189 << k << getFormatSpecifierRange(startSpecifier, specifierLen); 1190 // Don't do any more checking. We will just emit 1191 // spurious errors. 1192 return false; 1193 } 1194 1195 // Type check the data argument. It should be an 'int'. 1196 // Although not in conformance with C99, we also allow the argument to be 1197 // an 'unsigned int' as that is a reasonably safe case. GCC also 1198 // doesn't emit a warning for that case. 1199 CoveredArgs.set(argIndex); 1200 const Expr *Arg = getDataArg(argIndex); 1201 QualType T = Arg->getType(); 1202 1203 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1204 assert(ATR.isValid()); 1205 1206 if (!ATR.matchesType(S.Context, T)) { 1207 S.Diag(getLocationOfByte(Amt.getStart()), 1208 diag::warn_printf_asterisk_wrong_type) 1209 << k 1210 << ATR.getRepresentativeType(S.Context) << T 1211 << getFormatSpecifierRange(startSpecifier, specifierLen) 1212 << Arg->getSourceRange(); 1213 // Don't do any more checking. We will just emit 1214 // spurious errors. 1215 return false; 1216 } 1217 } 1218 } 1219 return true; 1220} 1221 1222bool 1223CheckPrintfHandler::HandleFormatSpecifier(const analyze_printf::FormatSpecifier 1224 &FS, 1225 const char *startSpecifier, 1226 unsigned specifierLen) { 1227 1228 using namespace analyze_printf; 1229 const ConversionSpecifier &CS = FS.getConversionSpecifier(); 1230 1231 if (atFirstArg) { 1232 atFirstArg = false; 1233 usesPositionalArgs = FS.usesPositionalArg(); 1234 } 1235 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1236 // Cannot mix-and-match positional and non-positional arguments. 1237 S.Diag(getLocationOfByte(CS.getStart()), 1238 diag::warn_printf_mix_positional_nonpositional_args) 1239 << getFormatSpecifierRange(startSpecifier, specifierLen); 1240 return false; 1241 } 1242 1243 // First check if the field width, precision, and conversion specifier 1244 // have matching data arguments. 1245 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 1246 startSpecifier, specifierLen)) { 1247 return false; 1248 } 1249 1250 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 1251 startSpecifier, specifierLen)) { 1252 return false; 1253 } 1254 1255 if (!CS.consumesDataArgument()) { 1256 // FIXME: Technically specifying a precision or field width here 1257 // makes no sense. Worth issuing a warning at some point. 1258 return true; 1259 } 1260 1261 // Consume the argument. 1262 unsigned argIndex = FS.getArgIndex(); 1263 if (argIndex < NumDataArgs) { 1264 // The check to see if the argIndex is valid will come later. 1265 // We set the bit here because we may exit early from this 1266 // function if we encounter some other error. 1267 CoveredArgs.set(argIndex); 1268 } 1269 1270 // Check for using an Objective-C specific conversion specifier 1271 // in a non-ObjC literal. 1272 if (!IsObjCLiteral && CS.isObjCArg()) { 1273 return HandleInvalidConversionSpecifier(FS, startSpecifier, specifierLen); 1274 } 1275 1276 // Are we using '%n'? Issue a warning about this being 1277 // a possible security issue. 1278 if (CS.getKind() == ConversionSpecifier::OutIntPtrArg) { 1279 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back) 1280 << getFormatSpecifierRange(startSpecifier, specifierLen); 1281 // Continue checking the other format specifiers. 1282 return true; 1283 } 1284 1285 if (CS.getKind() == ConversionSpecifier::VoidPtrArg) { 1286 if (FS.getPrecision().getHowSpecified() != OptionalAmount::NotSpecified) 1287 S.Diag(getLocationOfByte(CS.getStart()), 1288 diag::warn_printf_nonsensical_precision) 1289 << CS.getCharacters() 1290 << getFormatSpecifierRange(startSpecifier, specifierLen); 1291 } 1292 if (CS.getKind() == ConversionSpecifier::VoidPtrArg || 1293 CS.getKind() == ConversionSpecifier::CStrArg) { 1294 // FIXME: Instead of using "0", "+", etc., eventually get them from 1295 // the FormatSpecifier. 1296 if (FS.hasLeadingZeros()) 1297 HandleFlags(FS, "0", CS.getCharacters(), startSpecifier, specifierLen); 1298 if (FS.hasPlusPrefix()) 1299 HandleFlags(FS, "+", CS.getCharacters(), startSpecifier, specifierLen); 1300 if (FS.hasSpacePrefix()) 1301 HandleFlags(FS, " ", CS.getCharacters(), startSpecifier, specifierLen); 1302 } 1303 1304 // The remaining checks depend on the data arguments. 1305 if (HasVAListArg) 1306 return true; 1307 1308 if (argIndex >= NumDataArgs) { 1309 if (FS.usesPositionalArg()) { 1310 S.Diag(getLocationOfByte(CS.getStart()), 1311 diag::warn_printf_positional_arg_exceeds_data_args) 1312 << (argIndex+1) << NumDataArgs 1313 << getFormatSpecifierRange(startSpecifier, specifierLen); 1314 } 1315 else { 1316 S.Diag(getLocationOfByte(CS.getStart()), 1317 diag::warn_printf_insufficient_data_args) 1318 << getFormatSpecifierRange(startSpecifier, specifierLen); 1319 } 1320 1321 // Don't do any more checking. 1322 return false; 1323 } 1324 1325 // Now type check the data expression that matches the 1326 // format specifier. 1327 const Expr *Ex = getDataArg(argIndex); 1328 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context); 1329 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 1330 // Check if we didn't match because of an implicit cast from a 'char' 1331 // or 'short' to an 'int'. This is done because printf is a varargs 1332 // function. 1333 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 1334 if (ICE->getType() == S.Context.IntTy) 1335 if (ATR.matchesType(S.Context, ICE->getSubExpr()->getType())) 1336 return true; 1337 1338 S.Diag(getLocationOfByte(CS.getStart()), 1339 diag::warn_printf_conversion_argument_type_mismatch) 1340 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1341 << getFormatSpecifierRange(startSpecifier, specifierLen) 1342 << Ex->getSourceRange(); 1343 } 1344 1345 return true; 1346} 1347 1348void CheckPrintfHandler::DoneProcessing() { 1349 // Does the number of data arguments exceed the number of 1350 // format conversions in the format string? 1351 if (!HasVAListArg) { 1352 // Find any arguments that weren't covered. 1353 CoveredArgs.flip(); 1354 signed notCoveredArg = CoveredArgs.find_first(); 1355 if (notCoveredArg >= 0) { 1356 assert((unsigned)notCoveredArg < NumDataArgs); 1357 S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(), 1358 diag::warn_printf_data_arg_not_used) 1359 << getFormatStringRange(); 1360 } 1361 } 1362} 1363 1364void Sema::CheckPrintfString(const StringLiteral *FExpr, 1365 const Expr *OrigFormatExpr, 1366 const CallExpr *TheCall, bool HasVAListArg, 1367 unsigned format_idx, unsigned firstDataArg) { 1368 1369 // CHECK: is the format string a wide literal? 1370 if (FExpr->isWide()) { 1371 Diag(FExpr->getLocStart(), 1372 diag::warn_printf_format_string_is_wide_literal) 1373 << OrigFormatExpr->getSourceRange(); 1374 return; 1375 } 1376 1377 // Str - The format string. NOTE: this is NOT null-terminated! 1378 const char *Str = FExpr->getStrData(); 1379 1380 // CHECK: empty format string? 1381 unsigned StrLen = FExpr->getByteLength(); 1382 1383 if (StrLen == 0) { 1384 Diag(FExpr->getLocStart(), diag::warn_printf_empty_format_string) 1385 << OrigFormatExpr->getSourceRange(); 1386 return; 1387 } 1388 1389 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1390 TheCall->getNumArgs() - firstDataArg, 1391 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1392 HasVAListArg, TheCall, format_idx); 1393 1394 if (!analyze_printf::ParseFormatString(H, Str, Str + StrLen)) 1395 H.DoneProcessing(); 1396} 1397 1398//===--- CHECK: Return Address of Stack Variable --------------------------===// 1399 1400static DeclRefExpr* EvalVal(Expr *E); 1401static DeclRefExpr* EvalAddr(Expr* E); 1402 1403/// CheckReturnStackAddr - Check if a return statement returns the address 1404/// of a stack variable. 1405void 1406Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 1407 SourceLocation ReturnLoc) { 1408 1409 // Perform checking for returned stack addresses. 1410 if (lhsType->isPointerType() || lhsType->isBlockPointerType()) { 1411 if (DeclRefExpr *DR = EvalAddr(RetValExp)) 1412 Diag(DR->getLocStart(), diag::warn_ret_stack_addr) 1413 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1414 1415 // Skip over implicit cast expressions when checking for block expressions. 1416 RetValExp = RetValExp->IgnoreParenCasts(); 1417 1418 if (BlockExpr *C = dyn_cast<BlockExpr>(RetValExp)) 1419 if (C->hasBlockDeclRefExprs()) 1420 Diag(C->getLocStart(), diag::err_ret_local_block) 1421 << C->getSourceRange(); 1422 1423 if (AddrLabelExpr *ALE = dyn_cast<AddrLabelExpr>(RetValExp)) 1424 Diag(ALE->getLocStart(), diag::warn_ret_addr_label) 1425 << ALE->getSourceRange(); 1426 1427 } else if (lhsType->isReferenceType()) { 1428 // Perform checking for stack values returned by reference. 1429 // Check for a reference to the stack 1430 if (DeclRefExpr *DR = EvalVal(RetValExp)) 1431 Diag(DR->getLocStart(), diag::warn_ret_stack_ref) 1432 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1433 } 1434} 1435 1436/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 1437/// check if the expression in a return statement evaluates to an address 1438/// to a location on the stack. The recursion is used to traverse the 1439/// AST of the return expression, with recursion backtracking when we 1440/// encounter a subexpression that (1) clearly does not lead to the address 1441/// of a stack variable or (2) is something we cannot determine leads to 1442/// the address of a stack variable based on such local checking. 1443/// 1444/// EvalAddr processes expressions that are pointers that are used as 1445/// references (and not L-values). EvalVal handles all other values. 1446/// At the base case of the recursion is a check for a DeclRefExpr* in 1447/// the refers to a stack variable. 1448/// 1449/// This implementation handles: 1450/// 1451/// * pointer-to-pointer casts 1452/// * implicit conversions from array references to pointers 1453/// * taking the address of fields 1454/// * arbitrary interplay between "&" and "*" operators 1455/// * pointer arithmetic from an address of a stack variable 1456/// * taking the address of an array element where the array is on the stack 1457static DeclRefExpr* EvalAddr(Expr *E) { 1458 // We should only be called for evaluating pointer expressions. 1459 assert((E->getType()->isAnyPointerType() || 1460 E->getType()->isBlockPointerType() || 1461 E->getType()->isObjCQualifiedIdType()) && 1462 "EvalAddr only works on pointers"); 1463 1464 // Our "symbolic interpreter" is just a dispatch off the currently 1465 // viewed AST node. We then recursively traverse the AST by calling 1466 // EvalAddr and EvalVal appropriately. 1467 switch (E->getStmtClass()) { 1468 case Stmt::ParenExprClass: 1469 // Ignore parentheses. 1470 return EvalAddr(cast<ParenExpr>(E)->getSubExpr()); 1471 1472 case Stmt::UnaryOperatorClass: { 1473 // The only unary operator that make sense to handle here 1474 // is AddrOf. All others don't make sense as pointers. 1475 UnaryOperator *U = cast<UnaryOperator>(E); 1476 1477 if (U->getOpcode() == UnaryOperator::AddrOf) 1478 return EvalVal(U->getSubExpr()); 1479 else 1480 return NULL; 1481 } 1482 1483 case Stmt::BinaryOperatorClass: { 1484 // Handle pointer arithmetic. All other binary operators are not valid 1485 // in this context. 1486 BinaryOperator *B = cast<BinaryOperator>(E); 1487 BinaryOperator::Opcode op = B->getOpcode(); 1488 1489 if (op != BinaryOperator::Add && op != BinaryOperator::Sub) 1490 return NULL; 1491 1492 Expr *Base = B->getLHS(); 1493 1494 // Determine which argument is the real pointer base. It could be 1495 // the RHS argument instead of the LHS. 1496 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 1497 1498 assert (Base->getType()->isPointerType()); 1499 return EvalAddr(Base); 1500 } 1501 1502 // For conditional operators we need to see if either the LHS or RHS are 1503 // valid DeclRefExpr*s. If one of them is valid, we return it. 1504 case Stmt::ConditionalOperatorClass: { 1505 ConditionalOperator *C = cast<ConditionalOperator>(E); 1506 1507 // Handle the GNU extension for missing LHS. 1508 if (Expr *lhsExpr = C->getLHS()) 1509 if (DeclRefExpr* LHS = EvalAddr(lhsExpr)) 1510 return LHS; 1511 1512 return EvalAddr(C->getRHS()); 1513 } 1514 1515 // For casts, we need to handle conversions from arrays to 1516 // pointer values, and pointer-to-pointer conversions. 1517 case Stmt::ImplicitCastExprClass: 1518 case Stmt::CStyleCastExprClass: 1519 case Stmt::CXXFunctionalCastExprClass: { 1520 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 1521 QualType T = SubExpr->getType(); 1522 1523 if (SubExpr->getType()->isPointerType() || 1524 SubExpr->getType()->isBlockPointerType() || 1525 SubExpr->getType()->isObjCQualifiedIdType()) 1526 return EvalAddr(SubExpr); 1527 else if (T->isArrayType()) 1528 return EvalVal(SubExpr); 1529 else 1530 return 0; 1531 } 1532 1533 // C++ casts. For dynamic casts, static casts, and const casts, we 1534 // are always converting from a pointer-to-pointer, so we just blow 1535 // through the cast. In the case the dynamic cast doesn't fail (and 1536 // return NULL), we take the conservative route and report cases 1537 // where we return the address of a stack variable. For Reinterpre 1538 // FIXME: The comment about is wrong; we're not always converting 1539 // from pointer to pointer. I'm guessing that this code should also 1540 // handle references to objects. 1541 case Stmt::CXXStaticCastExprClass: 1542 case Stmt::CXXDynamicCastExprClass: 1543 case Stmt::CXXConstCastExprClass: 1544 case Stmt::CXXReinterpretCastExprClass: { 1545 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 1546 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 1547 return EvalAddr(S); 1548 else 1549 return NULL; 1550 } 1551 1552 // Everything else: we simply don't reason about them. 1553 default: 1554 return NULL; 1555 } 1556} 1557 1558 1559/// EvalVal - This function is complements EvalAddr in the mutual recursion. 1560/// See the comments for EvalAddr for more details. 1561static DeclRefExpr* EvalVal(Expr *E) { 1562 1563 // We should only be called for evaluating non-pointer expressions, or 1564 // expressions with a pointer type that are not used as references but instead 1565 // are l-values (e.g., DeclRefExpr with a pointer type). 1566 1567 // Our "symbolic interpreter" is just a dispatch off the currently 1568 // viewed AST node. We then recursively traverse the AST by calling 1569 // EvalAddr and EvalVal appropriately. 1570 switch (E->getStmtClass()) { 1571 case Stmt::DeclRefExprClass: { 1572 // DeclRefExpr: the base case. When we hit a DeclRefExpr we are looking 1573 // at code that refers to a variable's name. We check if it has local 1574 // storage within the function, and if so, return the expression. 1575 DeclRefExpr *DR = cast<DeclRefExpr>(E); 1576 1577 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 1578 if (V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR; 1579 1580 return NULL; 1581 } 1582 1583 case Stmt::ParenExprClass: 1584 // Ignore parentheses. 1585 return EvalVal(cast<ParenExpr>(E)->getSubExpr()); 1586 1587 case Stmt::UnaryOperatorClass: { 1588 // The only unary operator that make sense to handle here 1589 // is Deref. All others don't resolve to a "name." This includes 1590 // handling all sorts of rvalues passed to a unary operator. 1591 UnaryOperator *U = cast<UnaryOperator>(E); 1592 1593 if (U->getOpcode() == UnaryOperator::Deref) 1594 return EvalAddr(U->getSubExpr()); 1595 1596 return NULL; 1597 } 1598 1599 case Stmt::ArraySubscriptExprClass: { 1600 // Array subscripts are potential references to data on the stack. We 1601 // retrieve the DeclRefExpr* for the array variable if it indeed 1602 // has local storage. 1603 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase()); 1604 } 1605 1606 case Stmt::ConditionalOperatorClass: { 1607 // For conditional operators we need to see if either the LHS or RHS are 1608 // non-NULL DeclRefExpr's. If one is non-NULL, we return it. 1609 ConditionalOperator *C = cast<ConditionalOperator>(E); 1610 1611 // Handle the GNU extension for missing LHS. 1612 if (Expr *lhsExpr = C->getLHS()) 1613 if (DeclRefExpr *LHS = EvalVal(lhsExpr)) 1614 return LHS; 1615 1616 return EvalVal(C->getRHS()); 1617 } 1618 1619 // Accesses to members are potential references to data on the stack. 1620 case Stmt::MemberExprClass: { 1621 MemberExpr *M = cast<MemberExpr>(E); 1622 1623 // Check for indirect access. We only want direct field accesses. 1624 if (!M->isArrow()) 1625 return EvalVal(M->getBase()); 1626 else 1627 return NULL; 1628 } 1629 1630 // Everything else: we simply don't reason about them. 1631 default: 1632 return NULL; 1633 } 1634} 1635 1636//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 1637 1638/// Check for comparisons of floating point operands using != and ==. 1639/// Issue a warning if these are no self-comparisons, as they are not likely 1640/// to do what the programmer intended. 1641void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 1642 bool EmitWarning = true; 1643 1644 Expr* LeftExprSansParen = lex->IgnoreParens(); 1645 Expr* RightExprSansParen = rex->IgnoreParens(); 1646 1647 // Special case: check for x == x (which is OK). 1648 // Do not emit warnings for such cases. 1649 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 1650 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 1651 if (DRL->getDecl() == DRR->getDecl()) 1652 EmitWarning = false; 1653 1654 1655 // Special case: check for comparisons against literals that can be exactly 1656 // represented by APFloat. In such cases, do not emit a warning. This 1657 // is a heuristic: often comparison against such literals are used to 1658 // detect if a value in a variable has not changed. This clearly can 1659 // lead to false negatives. 1660 if (EmitWarning) { 1661 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 1662 if (FLL->isExact()) 1663 EmitWarning = false; 1664 } else 1665 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 1666 if (FLR->isExact()) 1667 EmitWarning = false; 1668 } 1669 } 1670 1671 // Check for comparisons with builtin types. 1672 if (EmitWarning) 1673 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 1674 if (CL->isBuiltinCall(Context)) 1675 EmitWarning = false; 1676 1677 if (EmitWarning) 1678 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 1679 if (CR->isBuiltinCall(Context)) 1680 EmitWarning = false; 1681 1682 // Emit the diagnostic. 1683 if (EmitWarning) 1684 Diag(loc, diag::warn_floatingpoint_eq) 1685 << lex->getSourceRange() << rex->getSourceRange(); 1686} 1687 1688//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 1689//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 1690 1691namespace { 1692 1693/// Structure recording the 'active' range of an integer-valued 1694/// expression. 1695struct IntRange { 1696 /// The number of bits active in the int. 1697 unsigned Width; 1698 1699 /// True if the int is known not to have negative values. 1700 bool NonNegative; 1701 1702 IntRange() {} 1703 IntRange(unsigned Width, bool NonNegative) 1704 : Width(Width), NonNegative(NonNegative) 1705 {} 1706 1707 // Returns the range of the bool type. 1708 static IntRange forBoolType() { 1709 return IntRange(1, true); 1710 } 1711 1712 // Returns the range of an integral type. 1713 static IntRange forType(ASTContext &C, QualType T) { 1714 return forCanonicalType(C, T->getCanonicalTypeInternal().getTypePtr()); 1715 } 1716 1717 // Returns the range of an integeral type based on its canonical 1718 // representation. 1719 static IntRange forCanonicalType(ASTContext &C, const Type *T) { 1720 assert(T->isCanonicalUnqualified()); 1721 1722 if (const VectorType *VT = dyn_cast<VectorType>(T)) 1723 T = VT->getElementType().getTypePtr(); 1724 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 1725 T = CT->getElementType().getTypePtr(); 1726 if (const EnumType *ET = dyn_cast<EnumType>(T)) 1727 T = ET->getDecl()->getIntegerType().getTypePtr(); 1728 1729 const BuiltinType *BT = cast<BuiltinType>(T); 1730 assert(BT->isInteger()); 1731 1732 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 1733 } 1734 1735 // Returns the supremum of two ranges: i.e. their conservative merge. 1736 static IntRange join(IntRange L, IntRange R) { 1737 return IntRange(std::max(L.Width, R.Width), 1738 L.NonNegative && R.NonNegative); 1739 } 1740 1741 // Returns the infinum of two ranges: i.e. their aggressive merge. 1742 static IntRange meet(IntRange L, IntRange R) { 1743 return IntRange(std::min(L.Width, R.Width), 1744 L.NonNegative || R.NonNegative); 1745 } 1746}; 1747 1748IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 1749 if (value.isSigned() && value.isNegative()) 1750 return IntRange(value.getMinSignedBits(), false); 1751 1752 if (value.getBitWidth() > MaxWidth) 1753 value.trunc(MaxWidth); 1754 1755 // isNonNegative() just checks the sign bit without considering 1756 // signedness. 1757 return IntRange(value.getActiveBits(), true); 1758} 1759 1760IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 1761 unsigned MaxWidth) { 1762 if (result.isInt()) 1763 return GetValueRange(C, result.getInt(), MaxWidth); 1764 1765 if (result.isVector()) { 1766 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 1767 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 1768 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 1769 R = IntRange::join(R, El); 1770 } 1771 return R; 1772 } 1773 1774 if (result.isComplexInt()) { 1775 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 1776 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 1777 return IntRange::join(R, I); 1778 } 1779 1780 // This can happen with lossless casts to intptr_t of "based" lvalues. 1781 // Assume it might use arbitrary bits. 1782 // FIXME: The only reason we need to pass the type in here is to get 1783 // the sign right on this one case. It would be nice if APValue 1784 // preserved this. 1785 assert(result.isLValue()); 1786 return IntRange(MaxWidth, Ty->isUnsignedIntegerType()); 1787} 1788 1789/// Pseudo-evaluate the given integer expression, estimating the 1790/// range of values it might take. 1791/// 1792/// \param MaxWidth - the width to which the value will be truncated 1793IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 1794 E = E->IgnoreParens(); 1795 1796 // Try a full evaluation first. 1797 Expr::EvalResult result; 1798 if (E->Evaluate(result, C)) 1799 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 1800 1801 // I think we only want to look through implicit casts here; if the 1802 // user has an explicit widening cast, we should treat the value as 1803 // being of the new, wider type. 1804 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 1805 if (CE->getCastKind() == CastExpr::CK_NoOp) 1806 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 1807 1808 IntRange OutputTypeRange = IntRange::forType(C, CE->getType()); 1809 1810 bool isIntegerCast = (CE->getCastKind() == CastExpr::CK_IntegralCast); 1811 if (!isIntegerCast && CE->getCastKind() == CastExpr::CK_Unknown) 1812 isIntegerCast = CE->getSubExpr()->getType()->isIntegerType(); 1813 1814 // Assume that non-integer casts can span the full range of the type. 1815 if (!isIntegerCast) 1816 return OutputTypeRange; 1817 1818 IntRange SubRange 1819 = GetExprRange(C, CE->getSubExpr(), 1820 std::min(MaxWidth, OutputTypeRange.Width)); 1821 1822 // Bail out if the subexpr's range is as wide as the cast type. 1823 if (SubRange.Width >= OutputTypeRange.Width) 1824 return OutputTypeRange; 1825 1826 // Otherwise, we take the smaller width, and we're non-negative if 1827 // either the output type or the subexpr is. 1828 return IntRange(SubRange.Width, 1829 SubRange.NonNegative || OutputTypeRange.NonNegative); 1830 } 1831 1832 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 1833 // If we can fold the condition, just take that operand. 1834 bool CondResult; 1835 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 1836 return GetExprRange(C, CondResult ? CO->getTrueExpr() 1837 : CO->getFalseExpr(), 1838 MaxWidth); 1839 1840 // Otherwise, conservatively merge. 1841 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 1842 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 1843 return IntRange::join(L, R); 1844 } 1845 1846 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 1847 switch (BO->getOpcode()) { 1848 1849 // Boolean-valued operations are single-bit and positive. 1850 case BinaryOperator::LAnd: 1851 case BinaryOperator::LOr: 1852 case BinaryOperator::LT: 1853 case BinaryOperator::GT: 1854 case BinaryOperator::LE: 1855 case BinaryOperator::GE: 1856 case BinaryOperator::EQ: 1857 case BinaryOperator::NE: 1858 return IntRange::forBoolType(); 1859 1860 // The type of these compound assignments is the type of the LHS, 1861 // so the RHS is not necessarily an integer. 1862 case BinaryOperator::MulAssign: 1863 case BinaryOperator::DivAssign: 1864 case BinaryOperator::RemAssign: 1865 case BinaryOperator::AddAssign: 1866 case BinaryOperator::SubAssign: 1867 return IntRange::forType(C, E->getType()); 1868 1869 // Operations with opaque sources are black-listed. 1870 case BinaryOperator::PtrMemD: 1871 case BinaryOperator::PtrMemI: 1872 return IntRange::forType(C, E->getType()); 1873 1874 // Bitwise-and uses the *infinum* of the two source ranges. 1875 case BinaryOperator::And: 1876 case BinaryOperator::AndAssign: 1877 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 1878 GetExprRange(C, BO->getRHS(), MaxWidth)); 1879 1880 // Left shift gets black-listed based on a judgement call. 1881 case BinaryOperator::Shl: 1882 // ...except that we want to treat '1 << (blah)' as logically 1883 // positive. It's an important idiom. 1884 if (IntegerLiteral *I 1885 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 1886 if (I->getValue() == 1) { 1887 IntRange R = IntRange::forType(C, E->getType()); 1888 return IntRange(R.Width, /*NonNegative*/ true); 1889 } 1890 } 1891 // fallthrough 1892 1893 case BinaryOperator::ShlAssign: 1894 return IntRange::forType(C, E->getType()); 1895 1896 // Right shift by a constant can narrow its left argument. 1897 case BinaryOperator::Shr: 1898 case BinaryOperator::ShrAssign: { 1899 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 1900 1901 // If the shift amount is a positive constant, drop the width by 1902 // that much. 1903 llvm::APSInt shift; 1904 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 1905 shift.isNonNegative()) { 1906 unsigned zext = shift.getZExtValue(); 1907 if (zext >= L.Width) 1908 L.Width = (L.NonNegative ? 0 : 1); 1909 else 1910 L.Width -= zext; 1911 } 1912 1913 return L; 1914 } 1915 1916 // Comma acts as its right operand. 1917 case BinaryOperator::Comma: 1918 return GetExprRange(C, BO->getRHS(), MaxWidth); 1919 1920 // Black-list pointer subtractions. 1921 case BinaryOperator::Sub: 1922 if (BO->getLHS()->getType()->isPointerType()) 1923 return IntRange::forType(C, E->getType()); 1924 // fallthrough 1925 1926 default: 1927 break; 1928 } 1929 1930 // Treat every other operator as if it were closed on the 1931 // narrowest type that encompasses both operands. 1932 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 1933 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 1934 return IntRange::join(L, R); 1935 } 1936 1937 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 1938 switch (UO->getOpcode()) { 1939 // Boolean-valued operations are white-listed. 1940 case UnaryOperator::LNot: 1941 return IntRange::forBoolType(); 1942 1943 // Operations with opaque sources are black-listed. 1944 case UnaryOperator::Deref: 1945 case UnaryOperator::AddrOf: // should be impossible 1946 case UnaryOperator::OffsetOf: 1947 return IntRange::forType(C, E->getType()); 1948 1949 default: 1950 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 1951 } 1952 } 1953 1954 if (dyn_cast<OffsetOfExpr>(E)) { 1955 IntRange::forType(C, E->getType()); 1956 } 1957 1958 FieldDecl *BitField = E->getBitField(); 1959 if (BitField) { 1960 llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C); 1961 unsigned BitWidth = BitWidthAP.getZExtValue(); 1962 1963 return IntRange(BitWidth, BitField->getType()->isUnsignedIntegerType()); 1964 } 1965 1966 return IntRange::forType(C, E->getType()); 1967} 1968 1969/// Checks whether the given value, which currently has the given 1970/// source semantics, has the same value when coerced through the 1971/// target semantics. 1972bool IsSameFloatAfterCast(const llvm::APFloat &value, 1973 const llvm::fltSemantics &Src, 1974 const llvm::fltSemantics &Tgt) { 1975 llvm::APFloat truncated = value; 1976 1977 bool ignored; 1978 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 1979 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 1980 1981 return truncated.bitwiseIsEqual(value); 1982} 1983 1984/// Checks whether the given value, which currently has the given 1985/// source semantics, has the same value when coerced through the 1986/// target semantics. 1987/// 1988/// The value might be a vector of floats (or a complex number). 1989bool IsSameFloatAfterCast(const APValue &value, 1990 const llvm::fltSemantics &Src, 1991 const llvm::fltSemantics &Tgt) { 1992 if (value.isFloat()) 1993 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 1994 1995 if (value.isVector()) { 1996 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 1997 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 1998 return false; 1999 return true; 2000 } 2001 2002 assert(value.isComplexFloat()); 2003 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 2004 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 2005} 2006 2007} // end anonymous namespace 2008 2009/// \brief Implements -Wsign-compare. 2010/// 2011/// \param lex the left-hand expression 2012/// \param rex the right-hand expression 2013/// \param OpLoc the location of the joining operator 2014/// \param BinOpc binary opcode or 0 2015void Sema::CheckSignCompare(Expr *lex, Expr *rex, SourceLocation OpLoc, 2016 const BinaryOperator::Opcode* BinOpc) { 2017 // Don't warn if we're in an unevaluated context. 2018 if (ExprEvalContexts.back().Context == Unevaluated) 2019 return; 2020 2021 // If either expression is value-dependent, don't warn. We'll get another 2022 // chance at instantiation time. 2023 if (lex->isValueDependent() || rex->isValueDependent()) 2024 return; 2025 2026 QualType lt = lex->getType(), rt = rex->getType(); 2027 2028 // Only warn if both operands are integral. 2029 if (!lt->isIntegerType() || !rt->isIntegerType()) 2030 return; 2031 2032 // In C, the width of a bitfield determines its type, and the 2033 // declared type only contributes the signedness. This duplicates 2034 // the work that will later be done by UsualUnaryConversions. 2035 // Eventually, this check will be reorganized in a way that avoids 2036 // this duplication. 2037 if (!getLangOptions().CPlusPlus) { 2038 QualType tmp; 2039 tmp = Context.isPromotableBitField(lex); 2040 if (!tmp.isNull()) lt = tmp; 2041 tmp = Context.isPromotableBitField(rex); 2042 if (!tmp.isNull()) rt = tmp; 2043 } 2044 2045 if (const EnumType *E = lt->getAs<EnumType>()) 2046 lt = E->getDecl()->getPromotionType(); 2047 if (const EnumType *E = rt->getAs<EnumType>()) 2048 rt = E->getDecl()->getPromotionType(); 2049 2050 // The rule is that the signed operand becomes unsigned, so isolate the 2051 // signed operand. 2052 Expr *signedOperand = lex, *unsignedOperand = rex; 2053 QualType signedType = lt, unsignedType = rt; 2054 if (lt->isSignedIntegerType()) { 2055 if (rt->isSignedIntegerType()) return; 2056 } else { 2057 if (!rt->isSignedIntegerType()) return; 2058 std::swap(signedOperand, unsignedOperand); 2059 std::swap(signedType, unsignedType); 2060 } 2061 2062 unsigned unsignedWidth = Context.getIntWidth(unsignedType); 2063 unsigned signedWidth = Context.getIntWidth(signedType); 2064 2065 // If the unsigned type is strictly smaller than the signed type, 2066 // then (1) the result type will be signed and (2) the unsigned 2067 // value will fit fully within the signed type, and thus the result 2068 // of the comparison will be exact. 2069 if (signedWidth > unsignedWidth) 2070 return; 2071 2072 // Otherwise, calculate the effective ranges. 2073 IntRange signedRange = GetExprRange(Context, signedOperand, signedWidth); 2074 IntRange unsignedRange = GetExprRange(Context, unsignedOperand, unsignedWidth); 2075 2076 // We should never be unable to prove that the unsigned operand is 2077 // non-negative. 2078 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 2079 2080 // If the signed operand is non-negative, then the signed->unsigned 2081 // conversion won't change it. 2082 if (signedRange.NonNegative) { 2083 // Emit warnings for comparisons of unsigned to integer constant 0. 2084 // always false: x < 0 (or 0 > x) 2085 // always true: x >= 0 (or 0 <= x) 2086 llvm::APSInt X; 2087 if (BinOpc && signedOperand->isIntegerConstantExpr(X, Context) && X == 0) { 2088 if (signedOperand != lex) { 2089 if (*BinOpc == BinaryOperator::LT) { 2090 Diag(OpLoc, diag::warn_lunsigned_always_true_comparison) 2091 << "< 0" << "false" 2092 << lex->getSourceRange() << rex->getSourceRange(); 2093 } 2094 else if (*BinOpc == BinaryOperator::GE) { 2095 Diag(OpLoc, diag::warn_lunsigned_always_true_comparison) 2096 << ">= 0" << "true" 2097 << lex->getSourceRange() << rex->getSourceRange(); 2098 } 2099 } 2100 else { 2101 if (*BinOpc == BinaryOperator::GT) { 2102 Diag(OpLoc, diag::warn_runsigned_always_true_comparison) 2103 << "0 >" << "false" 2104 << lex->getSourceRange() << rex->getSourceRange(); 2105 } 2106 else if (*BinOpc == BinaryOperator::LE) { 2107 Diag(OpLoc, diag::warn_runsigned_always_true_comparison) 2108 << "0 <=" << "true" 2109 << lex->getSourceRange() << rex->getSourceRange(); 2110 } 2111 } 2112 } 2113 return; 2114 } 2115 2116 // For (in)equality comparisons, if the unsigned operand is a 2117 // constant which cannot collide with a overflowed signed operand, 2118 // then reinterpreting the signed operand as unsigned will not 2119 // change the result of the comparison. 2120 if (BinOpc && 2121 (*BinOpc == BinaryOperator::EQ || *BinOpc == BinaryOperator::NE) && 2122 unsignedRange.Width < unsignedWidth) 2123 return; 2124 2125 Diag(OpLoc, BinOpc ? diag::warn_mixed_sign_comparison 2126 : diag::warn_mixed_sign_conditional) 2127 << lt << rt << lex->getSourceRange() << rex->getSourceRange(); 2128} 2129 2130/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 2131static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, unsigned diag) { 2132 S.Diag(E->getExprLoc(), diag) << E->getType() << T << E->getSourceRange(); 2133} 2134 2135/// Implements -Wconversion. 2136void Sema::CheckImplicitConversion(Expr *E, QualType T) { 2137 // Don't diagnose in unevaluated contexts. 2138 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 2139 return; 2140 2141 // Don't diagnose for value-dependent expressions. 2142 if (E->isValueDependent()) 2143 return; 2144 2145 const Type *Source = Context.getCanonicalType(E->getType()).getTypePtr(); 2146 const Type *Target = Context.getCanonicalType(T).getTypePtr(); 2147 2148 // Never diagnose implicit casts to bool. 2149 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 2150 return; 2151 2152 // Strip vector types. 2153 if (isa<VectorType>(Source)) { 2154 if (!isa<VectorType>(Target)) 2155 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_vector_scalar); 2156 2157 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 2158 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 2159 } 2160 2161 // Strip complex types. 2162 if (isa<ComplexType>(Source)) { 2163 if (!isa<ComplexType>(Target)) 2164 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_complex_scalar); 2165 2166 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 2167 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 2168 } 2169 2170 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 2171 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 2172 2173 // If the source is floating point... 2174 if (SourceBT && SourceBT->isFloatingPoint()) { 2175 // ...and the target is floating point... 2176 if (TargetBT && TargetBT->isFloatingPoint()) { 2177 // ...then warn if we're dropping FP rank. 2178 2179 // Builtin FP kinds are ordered by increasing FP rank. 2180 if (SourceBT->getKind() > TargetBT->getKind()) { 2181 // Don't warn about float constants that are precisely 2182 // representable in the target type. 2183 Expr::EvalResult result; 2184 if (E->Evaluate(result, Context)) { 2185 // Value might be a float, a float vector, or a float complex. 2186 if (IsSameFloatAfterCast(result.Val, 2187 Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 2188 Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 2189 return; 2190 } 2191 2192 DiagnoseImpCast(*this, E, T, diag::warn_impcast_float_precision); 2193 } 2194 return; 2195 } 2196 2197 // If the target is integral, always warn. 2198 if ((TargetBT && TargetBT->isInteger())) 2199 // TODO: don't warn for integer values? 2200 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_float_integer); 2201 2202 return; 2203 } 2204 2205 if (!Source->isIntegerType() || !Target->isIntegerType()) 2206 return; 2207 2208 IntRange SourceRange = GetExprRange(Context, E, Context.getIntWidth(E->getType())); 2209 IntRange TargetRange = IntRange::forCanonicalType(Context, Target); 2210 2211 // FIXME: also signed<->unsigned? 2212 2213 if (SourceRange.Width > TargetRange.Width) { 2214 // People want to build with -Wshorten-64-to-32 and not -Wconversion 2215 // and by god we'll let them. 2216 if (SourceRange.Width == 64 && TargetRange.Width == 32) 2217 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_integer_64_32); 2218 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_integer_precision); 2219 } 2220 2221 return; 2222} 2223 2224/// CheckParmsForFunctionDef - Check that the parameters of the given 2225/// function are appropriate for the definition of a function. This 2226/// takes care of any checks that cannot be performed on the 2227/// declaration itself, e.g., that the types of each of the function 2228/// parameters are complete. 2229bool Sema::CheckParmsForFunctionDef(FunctionDecl *FD) { 2230 bool HasInvalidParm = false; 2231 for (unsigned p = 0, NumParams = FD->getNumParams(); p < NumParams; ++p) { 2232 ParmVarDecl *Param = FD->getParamDecl(p); 2233 2234 // C99 6.7.5.3p4: the parameters in a parameter type list in a 2235 // function declarator that is part of a function definition of 2236 // that function shall not have incomplete type. 2237 // 2238 // This is also C++ [dcl.fct]p6. 2239 if (!Param->isInvalidDecl() && 2240 RequireCompleteType(Param->getLocation(), Param->getType(), 2241 diag::err_typecheck_decl_incomplete_type)) { 2242 Param->setInvalidDecl(); 2243 HasInvalidParm = true; 2244 } 2245 2246 // C99 6.9.1p5: If the declarator includes a parameter type list, the 2247 // declaration of each parameter shall include an identifier. 2248 if (Param->getIdentifier() == 0 && 2249 !Param->isImplicit() && 2250 !getLangOptions().CPlusPlus) 2251 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 2252 2253 // C99 6.7.5.3p12: 2254 // If the function declarator is not part of a definition of that 2255 // function, parameters may have incomplete type and may use the [*] 2256 // notation in their sequences of declarator specifiers to specify 2257 // variable length array types. 2258 QualType PType = Param->getOriginalType(); 2259 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 2260 if (AT->getSizeModifier() == ArrayType::Star) { 2261 // FIXME: This diagnosic should point the the '[*]' if source-location 2262 // information is added for it. 2263 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 2264 } 2265 } 2266 } 2267 2268 return HasInvalidParm; 2269} 2270