SemaChecking.cpp revision 9498d388810d284d3970aef0d69fa4d069fd6caf
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 (FunctionDecl *FD = getCurFunctionDecl()) 515 isVariadic = FD->isVariadic(); 516 else 517 isVariadic = getCurMethodDecl()->isVariadic(); 518 519 if (!isVariadic) { 520 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 521 return true; 522 } 523 524 // Verify that the second argument to the builtin is the last argument of the 525 // current function or method. 526 bool SecondArgIsLastNamedArgument = false; 527 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 528 529 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 530 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 531 // FIXME: This isn't correct for methods (results in bogus warning). 532 // Get the last formal in the current function. 533 const ParmVarDecl *LastArg; 534 if (CurBlock) 535 LastArg = *(CurBlock->TheDecl->param_end()-1); 536 else if (FunctionDecl *FD = getCurFunctionDecl()) 537 LastArg = *(FD->param_end()-1); 538 else 539 LastArg = *(getCurMethodDecl()->param_end()-1); 540 SecondArgIsLastNamedArgument = PV == LastArg; 541 } 542 } 543 544 if (!SecondArgIsLastNamedArgument) 545 Diag(TheCall->getArg(1)->getLocStart(), 546 diag::warn_second_parameter_of_va_start_not_last_named_argument); 547 return false; 548} 549 550/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 551/// friends. This is declared to take (...), so we have to check everything. 552bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 553 if (TheCall->getNumArgs() < 2) 554 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 555 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 556 if (TheCall->getNumArgs() > 2) 557 return Diag(TheCall->getArg(2)->getLocStart(), 558 diag::err_typecheck_call_too_many_args) 559 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 560 << SourceRange(TheCall->getArg(2)->getLocStart(), 561 (*(TheCall->arg_end()-1))->getLocEnd()); 562 563 Expr *OrigArg0 = TheCall->getArg(0); 564 Expr *OrigArg1 = TheCall->getArg(1); 565 566 // Do standard promotions between the two arguments, returning their common 567 // type. 568 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 569 570 // Make sure any conversions are pushed back into the call; this is 571 // type safe since unordered compare builtins are declared as "_Bool 572 // foo(...)". 573 TheCall->setArg(0, OrigArg0); 574 TheCall->setArg(1, OrigArg1); 575 576 if (OrigArg0->isTypeDependent() || OrigArg1->isTypeDependent()) 577 return false; 578 579 // If the common type isn't a real floating type, then the arguments were 580 // invalid for this operation. 581 if (!Res->isRealFloatingType()) 582 return Diag(OrigArg0->getLocStart(), 583 diag::err_typecheck_call_invalid_ordered_compare) 584 << OrigArg0->getType() << OrigArg1->getType() 585 << SourceRange(OrigArg0->getLocStart(), OrigArg1->getLocEnd()); 586 587 return false; 588} 589 590/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 591/// __builtin_isnan and friends. This is declared to take (...), so we have 592/// to check everything. We expect the last argument to be a floating point 593/// value. 594bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 595 if (TheCall->getNumArgs() < NumArgs) 596 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 597 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 598 if (TheCall->getNumArgs() > NumArgs) 599 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 600 diag::err_typecheck_call_too_many_args) 601 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 602 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 603 (*(TheCall->arg_end()-1))->getLocEnd()); 604 605 Expr *OrigArg = TheCall->getArg(NumArgs-1); 606 607 if (OrigArg->isTypeDependent()) 608 return false; 609 610 // This operation requires a floating-point number 611 if (!OrigArg->getType()->isRealFloatingType()) 612 return Diag(OrigArg->getLocStart(), 613 diag::err_typecheck_call_invalid_unary_fp) 614 << OrigArg->getType() << OrigArg->getSourceRange(); 615 616 return false; 617} 618 619/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 620// This is declared to take (...), so we have to check everything. 621Action::OwningExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 622 if (TheCall->getNumArgs() < 3) 623 return ExprError(Diag(TheCall->getLocEnd(), 624 diag::err_typecheck_call_too_few_args_at_least) 625 << 0 /*function call*/ << 3 << TheCall->getNumArgs() 626 << TheCall->getSourceRange()); 627 628 unsigned numElements = std::numeric_limits<unsigned>::max(); 629 if (!TheCall->getArg(0)->isTypeDependent() && 630 !TheCall->getArg(1)->isTypeDependent()) { 631 QualType FAType = TheCall->getArg(0)->getType(); 632 QualType SAType = TheCall->getArg(1)->getType(); 633 634 if (!FAType->isVectorType() || !SAType->isVectorType()) { 635 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 636 << SourceRange(TheCall->getArg(0)->getLocStart(), 637 TheCall->getArg(1)->getLocEnd()); 638 return ExprError(); 639 } 640 641 if (!Context.hasSameUnqualifiedType(FAType, SAType)) { 642 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 643 << SourceRange(TheCall->getArg(0)->getLocStart(), 644 TheCall->getArg(1)->getLocEnd()); 645 return ExprError(); 646 } 647 648 numElements = FAType->getAs<VectorType>()->getNumElements(); 649 if (TheCall->getNumArgs() != numElements+2) { 650 if (TheCall->getNumArgs() < numElements+2) 651 return ExprError(Diag(TheCall->getLocEnd(), 652 diag::err_typecheck_call_too_few_args) 653 << 0 /*function call*/ 654 << numElements+2 << TheCall->getNumArgs() 655 << TheCall->getSourceRange()); 656 return ExprError(Diag(TheCall->getLocEnd(), 657 diag::err_typecheck_call_too_many_args) 658 << 0 /*function call*/ 659 << numElements+2 << TheCall->getNumArgs() 660 << TheCall->getSourceRange()); 661 } 662 } 663 664 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 665 if (TheCall->getArg(i)->isTypeDependent() || 666 TheCall->getArg(i)->isValueDependent()) 667 continue; 668 669 llvm::APSInt Result; 670 if (SemaBuiltinConstantArg(TheCall, i, Result)) 671 return ExprError(); 672 673 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 674 return ExprError(Diag(TheCall->getLocStart(), 675 diag::err_shufflevector_argument_too_large) 676 << TheCall->getArg(i)->getSourceRange()); 677 } 678 679 llvm::SmallVector<Expr*, 32> exprs; 680 681 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 682 exprs.push_back(TheCall->getArg(i)); 683 TheCall->setArg(i, 0); 684 } 685 686 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 687 exprs.size(), exprs[0]->getType(), 688 TheCall->getCallee()->getLocStart(), 689 TheCall->getRParenLoc())); 690} 691 692/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 693// This is declared to take (const void*, ...) and can take two 694// optional constant int args. 695bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 696 unsigned NumArgs = TheCall->getNumArgs(); 697 698 if (NumArgs > 3) 699 return Diag(TheCall->getLocEnd(), 700 diag::err_typecheck_call_too_many_args_at_most) 701 << 0 /*function call*/ << 3 << NumArgs 702 << TheCall->getSourceRange(); 703 704 // Argument 0 is checked for us and the remaining arguments must be 705 // constant integers. 706 for (unsigned i = 1; i != NumArgs; ++i) { 707 Expr *Arg = TheCall->getArg(i); 708 709 llvm::APSInt Result; 710 if (SemaBuiltinConstantArg(TheCall, i, Result)) 711 return true; 712 713 // FIXME: gcc issues a warning and rewrites these to 0. These 714 // seems especially odd for the third argument since the default 715 // is 3. 716 if (i == 1) { 717 if (Result.getLimitedValue() > 1) 718 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 719 << "0" << "1" << Arg->getSourceRange(); 720 } else { 721 if (Result.getLimitedValue() > 3) 722 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 723 << "0" << "3" << Arg->getSourceRange(); 724 } 725 } 726 727 return false; 728} 729 730/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 731/// TheCall is a constant expression. 732bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 733 llvm::APSInt &Result) { 734 Expr *Arg = TheCall->getArg(ArgNum); 735 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 736 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 737 738 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 739 740 if (!Arg->isIntegerConstantExpr(Result, Context)) 741 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 742 << FDecl->getDeclName() << Arg->getSourceRange(); 743 744 return false; 745} 746 747/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 748/// int type). This simply type checks that type is one of the defined 749/// constants (0-3). 750// For compatability check 0-3, llvm only handles 0 and 2. 751bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 752 llvm::APSInt Result; 753 754 // Check constant-ness first. 755 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 756 return true; 757 758 Expr *Arg = TheCall->getArg(1); 759 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 760 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 761 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 762 } 763 764 return false; 765} 766 767/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 768/// This checks that val is a constant 1. 769bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 770 Expr *Arg = TheCall->getArg(1); 771 llvm::APSInt Result; 772 773 // TODO: This is less than ideal. Overload this to take a value. 774 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 775 return true; 776 777 if (Result != 1) 778 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 779 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 780 781 return false; 782} 783 784// Handle i > 1 ? "x" : "y", recursivelly 785bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 786 bool HasVAListArg, 787 unsigned format_idx, unsigned firstDataArg) { 788 if (E->isTypeDependent() || E->isValueDependent()) 789 return false; 790 791 switch (E->getStmtClass()) { 792 case Stmt::ConditionalOperatorClass: { 793 const ConditionalOperator *C = cast<ConditionalOperator>(E); 794 return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, 795 HasVAListArg, format_idx, firstDataArg) 796 && SemaCheckStringLiteral(C->getRHS(), TheCall, 797 HasVAListArg, format_idx, firstDataArg); 798 } 799 800 case Stmt::ImplicitCastExprClass: { 801 const ImplicitCastExpr *Expr = cast<ImplicitCastExpr>(E); 802 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 803 format_idx, firstDataArg); 804 } 805 806 case Stmt::ParenExprClass: { 807 const ParenExpr *Expr = cast<ParenExpr>(E); 808 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 809 format_idx, firstDataArg); 810 } 811 812 case Stmt::DeclRefExprClass: { 813 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 814 815 // As an exception, do not flag errors for variables binding to 816 // const string literals. 817 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 818 bool isConstant = false; 819 QualType T = DR->getType(); 820 821 if (const ArrayType *AT = Context.getAsArrayType(T)) { 822 isConstant = AT->getElementType().isConstant(Context); 823 } else if (const PointerType *PT = T->getAs<PointerType>()) { 824 isConstant = T.isConstant(Context) && 825 PT->getPointeeType().isConstant(Context); 826 } 827 828 if (isConstant) { 829 if (const Expr *Init = VD->getAnyInitializer()) 830 return SemaCheckStringLiteral(Init, TheCall, 831 HasVAListArg, format_idx, firstDataArg); 832 } 833 834 // For vprintf* functions (i.e., HasVAListArg==true), we add a 835 // special check to see if the format string is a function parameter 836 // of the function calling the printf function. If the function 837 // has an attribute indicating it is a printf-like function, then we 838 // should suppress warnings concerning non-literals being used in a call 839 // to a vprintf function. For example: 840 // 841 // void 842 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 843 // va_list ap; 844 // va_start(ap, fmt); 845 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 846 // ... 847 // 848 // 849 // FIXME: We don't have full attribute support yet, so just check to see 850 // if the argument is a DeclRefExpr that references a parameter. We'll 851 // add proper support for checking the attribute later. 852 if (HasVAListArg) 853 if (isa<ParmVarDecl>(VD)) 854 return true; 855 } 856 857 return false; 858 } 859 860 case Stmt::CallExprClass: { 861 const CallExpr *CE = cast<CallExpr>(E); 862 if (const ImplicitCastExpr *ICE 863 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 864 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 865 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 866 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 867 unsigned ArgIndex = FA->getFormatIdx(); 868 const Expr *Arg = CE->getArg(ArgIndex - 1); 869 870 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 871 format_idx, firstDataArg); 872 } 873 } 874 } 875 } 876 877 return false; 878 } 879 case Stmt::ObjCStringLiteralClass: 880 case Stmt::StringLiteralClass: { 881 const StringLiteral *StrE = NULL; 882 883 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 884 StrE = ObjCFExpr->getString(); 885 else 886 StrE = cast<StringLiteral>(E); 887 888 if (StrE) { 889 CheckPrintfString(StrE, E, TheCall, HasVAListArg, format_idx, 890 firstDataArg); 891 return true; 892 } 893 894 return false; 895 } 896 897 default: 898 return false; 899 } 900} 901 902void 903Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 904 const CallExpr *TheCall) { 905 for (NonNullAttr::iterator i = NonNull->begin(), e = NonNull->end(); 906 i != e; ++i) { 907 const Expr *ArgExpr = TheCall->getArg(*i); 908 if (ArgExpr->isNullPointerConstant(Context, 909 Expr::NPC_ValueDependentIsNotNull)) 910 Diag(TheCall->getCallee()->getLocStart(), diag::warn_null_arg) 911 << ArgExpr->getSourceRange(); 912 } 913} 914 915/// CheckPrintfArguments - Check calls to printf (and similar functions) for 916/// correct use of format strings. 917/// 918/// HasVAListArg - A predicate indicating whether the printf-like 919/// function is passed an explicit va_arg argument (e.g., vprintf) 920/// 921/// format_idx - The index into Args for the format string. 922/// 923/// Improper format strings to functions in the printf family can be 924/// the source of bizarre bugs and very serious security holes. A 925/// good source of information is available in the following paper 926/// (which includes additional references): 927/// 928/// FormatGuard: Automatic Protection From printf Format String 929/// Vulnerabilities, Proceedings of the 10th USENIX Security Symposium, 2001. 930/// 931/// TODO: 932/// Functionality implemented: 933/// 934/// We can statically check the following properties for string 935/// literal format strings for non v.*printf functions (where the 936/// arguments are passed directly): 937// 938/// (1) Are the number of format conversions equal to the number of 939/// data arguments? 940/// 941/// (2) Does each format conversion correctly match the type of the 942/// corresponding data argument? 943/// 944/// Moreover, for all printf functions we can: 945/// 946/// (3) Check for a missing format string (when not caught by type checking). 947/// 948/// (4) Check for no-operation flags; e.g. using "#" with format 949/// conversion 'c' (TODO) 950/// 951/// (5) Check the use of '%n', a major source of security holes. 952/// 953/// (6) Check for malformed format conversions that don't specify anything. 954/// 955/// (7) Check for empty format strings. e.g: printf(""); 956/// 957/// (8) Check that the format string is a wide literal. 958/// 959/// All of these checks can be done by parsing the format string. 960/// 961void 962Sema::CheckPrintfArguments(const CallExpr *TheCall, bool HasVAListArg, 963 unsigned format_idx, unsigned firstDataArg) { 964 const Expr *Fn = TheCall->getCallee(); 965 966 // The way the format attribute works in GCC, the implicit this argument 967 // of member functions is counted. However, it doesn't appear in our own 968 // lists, so decrement format_idx in that case. 969 if (isa<CXXMemberCallExpr>(TheCall)) { 970 // Catch a format attribute mistakenly referring to the object argument. 971 if (format_idx == 0) 972 return; 973 --format_idx; 974 if(firstDataArg != 0) 975 --firstDataArg; 976 } 977 978 // CHECK: printf-like function is called with no format string. 979 if (format_idx >= TheCall->getNumArgs()) { 980 Diag(TheCall->getRParenLoc(), diag::warn_printf_missing_format_string) 981 << Fn->getSourceRange(); 982 return; 983 } 984 985 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 986 987 // CHECK: format string is not a string literal. 988 // 989 // Dynamically generated format strings are difficult to 990 // automatically vet at compile time. Requiring that format strings 991 // are string literals: (1) permits the checking of format strings by 992 // the compiler and thereby (2) can practically remove the source of 993 // many format string exploits. 994 995 // Format string can be either ObjC string (e.g. @"%d") or 996 // C string (e.g. "%d") 997 // ObjC string uses the same format specifiers as C string, so we can use 998 // the same format string checking logic for both ObjC and C strings. 999 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 1000 firstDataArg)) 1001 return; // Literal format string found, check done! 1002 1003 // If there are no arguments specified, warn with -Wformat-security, otherwise 1004 // warn only with -Wformat-nonliteral. 1005 if (TheCall->getNumArgs() == format_idx+1) 1006 Diag(TheCall->getArg(format_idx)->getLocStart(), 1007 diag::warn_printf_nonliteral_noargs) 1008 << OrigFormatExpr->getSourceRange(); 1009 else 1010 Diag(TheCall->getArg(format_idx)->getLocStart(), 1011 diag::warn_printf_nonliteral) 1012 << OrigFormatExpr->getSourceRange(); 1013} 1014 1015namespace { 1016class CheckPrintfHandler : public analyze_printf::FormatStringHandler { 1017 Sema &S; 1018 const StringLiteral *FExpr; 1019 const Expr *OrigFormatExpr; 1020 const unsigned FirstDataArg; 1021 const unsigned NumDataArgs; 1022 const bool IsObjCLiteral; 1023 const char *Beg; // Start of format string. 1024 const bool HasVAListArg; 1025 const CallExpr *TheCall; 1026 unsigned FormatIdx; 1027 llvm::BitVector CoveredArgs; 1028 bool usesPositionalArgs; 1029 bool atFirstArg; 1030public: 1031 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1032 const Expr *origFormatExpr, unsigned firstDataArg, 1033 unsigned numDataArgs, bool isObjCLiteral, 1034 const char *beg, bool hasVAListArg, 1035 const CallExpr *theCall, unsigned formatIdx) 1036 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1037 FirstDataArg(firstDataArg), 1038 NumDataArgs(numDataArgs), 1039 IsObjCLiteral(isObjCLiteral), Beg(beg), 1040 HasVAListArg(hasVAListArg), 1041 TheCall(theCall), FormatIdx(formatIdx), 1042 usesPositionalArgs(false), atFirstArg(true) { 1043 CoveredArgs.resize(numDataArgs); 1044 CoveredArgs.reset(); 1045 } 1046 1047 void DoneProcessing(); 1048 1049 void HandleIncompleteFormatSpecifier(const char *startSpecifier, 1050 unsigned specifierLen); 1051 1052 bool 1053 HandleInvalidConversionSpecifier(const analyze_printf::FormatSpecifier &FS, 1054 const char *startSpecifier, 1055 unsigned specifierLen); 1056 1057 virtual void HandleInvalidPosition(const char *startSpecifier, 1058 unsigned specifierLen, 1059 analyze_printf::PositionContext p); 1060 1061 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1062 1063 void HandleNullChar(const char *nullCharacter); 1064 1065 bool HandleFormatSpecifier(const analyze_printf::FormatSpecifier &FS, 1066 const char *startSpecifier, 1067 unsigned specifierLen); 1068private: 1069 SourceRange getFormatStringRange(); 1070 SourceRange getFormatSpecifierRange(const char *startSpecifier, 1071 unsigned specifierLen); 1072 SourceLocation getLocationOfByte(const char *x); 1073 1074 bool HandleAmount(const analyze_printf::OptionalAmount &Amt, unsigned k, 1075 const char *startSpecifier, unsigned specifierLen); 1076 void HandleFlags(const analyze_printf::FormatSpecifier &FS, 1077 llvm::StringRef flag, llvm::StringRef cspec, 1078 const char *startSpecifier, unsigned specifierLen); 1079 1080 const Expr *getDataArg(unsigned i) const; 1081}; 1082} 1083 1084SourceRange CheckPrintfHandler::getFormatStringRange() { 1085 return OrigFormatExpr->getSourceRange(); 1086} 1087 1088SourceRange CheckPrintfHandler:: 1089getFormatSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1090 return SourceRange(getLocationOfByte(startSpecifier), 1091 getLocationOfByte(startSpecifier+specifierLen-1)); 1092} 1093 1094SourceLocation CheckPrintfHandler::getLocationOfByte(const char *x) { 1095 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1096} 1097 1098void CheckPrintfHandler:: 1099HandleIncompleteFormatSpecifier(const char *startSpecifier, 1100 unsigned specifierLen) { 1101 SourceLocation Loc = getLocationOfByte(startSpecifier); 1102 S.Diag(Loc, diag::warn_printf_incomplete_specifier) 1103 << getFormatSpecifierRange(startSpecifier, specifierLen); 1104} 1105 1106void 1107CheckPrintfHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1108 analyze_printf::PositionContext p) { 1109 SourceLocation Loc = getLocationOfByte(startPos); 1110 S.Diag(Loc, diag::warn_printf_invalid_positional_specifier) 1111 << (unsigned) p << getFormatSpecifierRange(startPos, posLen); 1112} 1113 1114void CheckPrintfHandler::HandleZeroPosition(const char *startPos, 1115 unsigned posLen) { 1116 SourceLocation Loc = getLocationOfByte(startPos); 1117 S.Diag(Loc, diag::warn_printf_zero_positional_specifier) 1118 << getFormatSpecifierRange(startPos, posLen); 1119} 1120 1121bool CheckPrintfHandler:: 1122HandleInvalidConversionSpecifier(const analyze_printf::FormatSpecifier &FS, 1123 const char *startSpecifier, 1124 unsigned specifierLen) { 1125 1126 unsigned argIndex = FS.getArgIndex(); 1127 bool keepGoing = true; 1128 if (argIndex < NumDataArgs) { 1129 // Consider the argument coverered, even though the specifier doesn't 1130 // make sense. 1131 CoveredArgs.set(argIndex); 1132 } 1133 else { 1134 // If argIndex exceeds the number of data arguments we 1135 // don't issue a warning because that is just a cascade of warnings (and 1136 // they may have intended '%%' anyway). We don't want to continue processing 1137 // the format string after this point, however, as we will like just get 1138 // gibberish when trying to match arguments. 1139 keepGoing = false; 1140 } 1141 1142 const analyze_printf::ConversionSpecifier &CS = 1143 FS.getConversionSpecifier(); 1144 SourceLocation Loc = getLocationOfByte(CS.getStart()); 1145 S.Diag(Loc, diag::warn_printf_invalid_conversion) 1146 << llvm::StringRef(CS.getStart(), CS.getLength()) 1147 << getFormatSpecifierRange(startSpecifier, specifierLen); 1148 1149 return keepGoing; 1150} 1151 1152void CheckPrintfHandler::HandleNullChar(const char *nullCharacter) { 1153 // The presence of a null character is likely an error. 1154 S.Diag(getLocationOfByte(nullCharacter), 1155 diag::warn_printf_format_string_contains_null_char) 1156 << getFormatStringRange(); 1157} 1158 1159const Expr *CheckPrintfHandler::getDataArg(unsigned i) const { 1160 return TheCall->getArg(FirstDataArg + i); 1161} 1162 1163void CheckPrintfHandler::HandleFlags(const analyze_printf::FormatSpecifier &FS, 1164 llvm::StringRef flag, 1165 llvm::StringRef cspec, 1166 const char *startSpecifier, 1167 unsigned specifierLen) { 1168 const analyze_printf::ConversionSpecifier &CS = FS.getConversionSpecifier(); 1169 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_nonsensical_flag) 1170 << flag << cspec << getFormatSpecifierRange(startSpecifier, specifierLen); 1171} 1172 1173bool 1174CheckPrintfHandler::HandleAmount(const analyze_printf::OptionalAmount &Amt, 1175 unsigned k, const char *startSpecifier, 1176 unsigned specifierLen) { 1177 1178 if (Amt.hasDataArgument()) { 1179 if (!HasVAListArg) { 1180 unsigned argIndex = Amt.getArgIndex(); 1181 if (argIndex >= NumDataArgs) { 1182 S.Diag(getLocationOfByte(Amt.getStart()), 1183 diag::warn_printf_asterisk_missing_arg) 1184 << k << getFormatSpecifierRange(startSpecifier, specifierLen); 1185 // Don't do any more checking. We will just emit 1186 // spurious errors. 1187 return false; 1188 } 1189 1190 // Type check the data argument. It should be an 'int'. 1191 // Although not in conformance with C99, we also allow the argument to be 1192 // an 'unsigned int' as that is a reasonably safe case. GCC also 1193 // doesn't emit a warning for that case. 1194 CoveredArgs.set(argIndex); 1195 const Expr *Arg = getDataArg(argIndex); 1196 QualType T = Arg->getType(); 1197 1198 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1199 assert(ATR.isValid()); 1200 1201 if (!ATR.matchesType(S.Context, T)) { 1202 S.Diag(getLocationOfByte(Amt.getStart()), 1203 diag::warn_printf_asterisk_wrong_type) 1204 << k 1205 << ATR.getRepresentativeType(S.Context) << T 1206 << getFormatSpecifierRange(startSpecifier, specifierLen) 1207 << Arg->getSourceRange(); 1208 // Don't do any more checking. We will just emit 1209 // spurious errors. 1210 return false; 1211 } 1212 } 1213 } 1214 return true; 1215} 1216 1217bool 1218CheckPrintfHandler::HandleFormatSpecifier(const analyze_printf::FormatSpecifier 1219 &FS, 1220 const char *startSpecifier, 1221 unsigned specifierLen) { 1222 1223 using namespace analyze_printf; 1224 const ConversionSpecifier &CS = FS.getConversionSpecifier(); 1225 1226 if (atFirstArg) { 1227 atFirstArg = false; 1228 usesPositionalArgs = FS.usesPositionalArg(); 1229 } 1230 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1231 // Cannot mix-and-match positional and non-positional arguments. 1232 S.Diag(getLocationOfByte(CS.getStart()), 1233 diag::warn_printf_mix_positional_nonpositional_args) 1234 << getFormatSpecifierRange(startSpecifier, specifierLen); 1235 return false; 1236 } 1237 1238 // First check if the field width, precision, and conversion specifier 1239 // have matching data arguments. 1240 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 1241 startSpecifier, specifierLen)) { 1242 return false; 1243 } 1244 1245 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 1246 startSpecifier, specifierLen)) { 1247 return false; 1248 } 1249 1250 if (!CS.consumesDataArgument()) { 1251 // FIXME: Technically specifying a precision or field width here 1252 // makes no sense. Worth issuing a warning at some point. 1253 return true; 1254 } 1255 1256 // Consume the argument. 1257 unsigned argIndex = FS.getArgIndex(); 1258 if (argIndex < NumDataArgs) { 1259 // The check to see if the argIndex is valid will come later. 1260 // We set the bit here because we may exit early from this 1261 // function if we encounter some other error. 1262 CoveredArgs.set(argIndex); 1263 } 1264 1265 // Check for using an Objective-C specific conversion specifier 1266 // in a non-ObjC literal. 1267 if (!IsObjCLiteral && CS.isObjCArg()) { 1268 return HandleInvalidConversionSpecifier(FS, startSpecifier, specifierLen); 1269 } 1270 1271 // Are we using '%n'? Issue a warning about this being 1272 // a possible security issue. 1273 if (CS.getKind() == ConversionSpecifier::OutIntPtrArg) { 1274 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back) 1275 << getFormatSpecifierRange(startSpecifier, specifierLen); 1276 // Continue checking the other format specifiers. 1277 return true; 1278 } 1279 1280 if (CS.getKind() == ConversionSpecifier::VoidPtrArg) { 1281 if (FS.getPrecision().getHowSpecified() != OptionalAmount::NotSpecified) 1282 S.Diag(getLocationOfByte(CS.getStart()), 1283 diag::warn_printf_nonsensical_precision) 1284 << CS.getCharacters() 1285 << getFormatSpecifierRange(startSpecifier, specifierLen); 1286 } 1287 if (CS.getKind() == ConversionSpecifier::VoidPtrArg || 1288 CS.getKind() == ConversionSpecifier::CStrArg) { 1289 // FIXME: Instead of using "0", "+", etc., eventually get them from 1290 // the FormatSpecifier. 1291 if (FS.hasLeadingZeros()) 1292 HandleFlags(FS, "0", CS.getCharacters(), startSpecifier, specifierLen); 1293 if (FS.hasPlusPrefix()) 1294 HandleFlags(FS, "+", CS.getCharacters(), startSpecifier, specifierLen); 1295 if (FS.hasSpacePrefix()) 1296 HandleFlags(FS, " ", CS.getCharacters(), startSpecifier, specifierLen); 1297 } 1298 1299 // The remaining checks depend on the data arguments. 1300 if (HasVAListArg) 1301 return true; 1302 1303 if (argIndex >= NumDataArgs) { 1304 if (FS.usesPositionalArg()) { 1305 S.Diag(getLocationOfByte(CS.getStart()), 1306 diag::warn_printf_positional_arg_exceeds_data_args) 1307 << (argIndex+1) << NumDataArgs 1308 << getFormatSpecifierRange(startSpecifier, specifierLen); 1309 } 1310 else { 1311 S.Diag(getLocationOfByte(CS.getStart()), 1312 diag::warn_printf_insufficient_data_args) 1313 << getFormatSpecifierRange(startSpecifier, specifierLen); 1314 } 1315 1316 // Don't do any more checking. 1317 return false; 1318 } 1319 1320 // Now type check the data expression that matches the 1321 // format specifier. 1322 const Expr *Ex = getDataArg(argIndex); 1323 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context); 1324 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 1325 // Check if we didn't match because of an implicit cast from a 'char' 1326 // or 'short' to an 'int'. This is done because printf is a varargs 1327 // function. 1328 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 1329 if (ICE->getType() == S.Context.IntTy) 1330 if (ATR.matchesType(S.Context, ICE->getSubExpr()->getType())) 1331 return true; 1332 1333 S.Diag(getLocationOfByte(CS.getStart()), 1334 diag::warn_printf_conversion_argument_type_mismatch) 1335 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1336 << getFormatSpecifierRange(startSpecifier, specifierLen) 1337 << Ex->getSourceRange(); 1338 } 1339 1340 return true; 1341} 1342 1343void CheckPrintfHandler::DoneProcessing() { 1344 // Does the number of data arguments exceed the number of 1345 // format conversions in the format string? 1346 if (!HasVAListArg) { 1347 // Find any arguments that weren't covered. 1348 CoveredArgs.flip(); 1349 signed notCoveredArg = CoveredArgs.find_first(); 1350 if (notCoveredArg >= 0) { 1351 assert((unsigned)notCoveredArg < NumDataArgs); 1352 S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(), 1353 diag::warn_printf_data_arg_not_used) 1354 << getFormatStringRange(); 1355 } 1356 } 1357} 1358 1359void Sema::CheckPrintfString(const StringLiteral *FExpr, 1360 const Expr *OrigFormatExpr, 1361 const CallExpr *TheCall, bool HasVAListArg, 1362 unsigned format_idx, unsigned firstDataArg) { 1363 1364 // CHECK: is the format string a wide literal? 1365 if (FExpr->isWide()) { 1366 Diag(FExpr->getLocStart(), 1367 diag::warn_printf_format_string_is_wide_literal) 1368 << OrigFormatExpr->getSourceRange(); 1369 return; 1370 } 1371 1372 // Str - The format string. NOTE: this is NOT null-terminated! 1373 const char *Str = FExpr->getStrData(); 1374 1375 // CHECK: empty format string? 1376 unsigned StrLen = FExpr->getByteLength(); 1377 1378 if (StrLen == 0) { 1379 Diag(FExpr->getLocStart(), diag::warn_printf_empty_format_string) 1380 << OrigFormatExpr->getSourceRange(); 1381 return; 1382 } 1383 1384 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1385 TheCall->getNumArgs() - firstDataArg, 1386 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1387 HasVAListArg, TheCall, format_idx); 1388 1389 if (!analyze_printf::ParseFormatString(H, Str, Str + StrLen)) 1390 H.DoneProcessing(); 1391} 1392 1393//===--- CHECK: Return Address of Stack Variable --------------------------===// 1394 1395static DeclRefExpr* EvalVal(Expr *E); 1396static DeclRefExpr* EvalAddr(Expr* E); 1397 1398/// CheckReturnStackAddr - Check if a return statement returns the address 1399/// of a stack variable. 1400void 1401Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 1402 SourceLocation ReturnLoc) { 1403 1404 // Perform checking for returned stack addresses. 1405 if (lhsType->isPointerType() || lhsType->isBlockPointerType()) { 1406 if (DeclRefExpr *DR = EvalAddr(RetValExp)) 1407 Diag(DR->getLocStart(), diag::warn_ret_stack_addr) 1408 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1409 1410 // Skip over implicit cast expressions when checking for block expressions. 1411 RetValExp = RetValExp->IgnoreParenCasts(); 1412 1413 if (BlockExpr *C = dyn_cast<BlockExpr>(RetValExp)) 1414 if (C->hasBlockDeclRefExprs()) 1415 Diag(C->getLocStart(), diag::err_ret_local_block) 1416 << C->getSourceRange(); 1417 1418 if (AddrLabelExpr *ALE = dyn_cast<AddrLabelExpr>(RetValExp)) 1419 Diag(ALE->getLocStart(), diag::warn_ret_addr_label) 1420 << ALE->getSourceRange(); 1421 1422 } else if (lhsType->isReferenceType()) { 1423 // Perform checking for stack values returned by reference. 1424 // Check for a reference to the stack 1425 if (DeclRefExpr *DR = EvalVal(RetValExp)) 1426 Diag(DR->getLocStart(), diag::warn_ret_stack_ref) 1427 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1428 } 1429} 1430 1431/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 1432/// check if the expression in a return statement evaluates to an address 1433/// to a location on the stack. The recursion is used to traverse the 1434/// AST of the return expression, with recursion backtracking when we 1435/// encounter a subexpression that (1) clearly does not lead to the address 1436/// of a stack variable or (2) is something we cannot determine leads to 1437/// the address of a stack variable based on such local checking. 1438/// 1439/// EvalAddr processes expressions that are pointers that are used as 1440/// references (and not L-values). EvalVal handles all other values. 1441/// At the base case of the recursion is a check for a DeclRefExpr* in 1442/// the refers to a stack variable. 1443/// 1444/// This implementation handles: 1445/// 1446/// * pointer-to-pointer casts 1447/// * implicit conversions from array references to pointers 1448/// * taking the address of fields 1449/// * arbitrary interplay between "&" and "*" operators 1450/// * pointer arithmetic from an address of a stack variable 1451/// * taking the address of an array element where the array is on the stack 1452static DeclRefExpr* EvalAddr(Expr *E) { 1453 // We should only be called for evaluating pointer expressions. 1454 assert((E->getType()->isAnyPointerType() || 1455 E->getType()->isBlockPointerType() || 1456 E->getType()->isObjCQualifiedIdType()) && 1457 "EvalAddr only works on pointers"); 1458 1459 // Our "symbolic interpreter" is just a dispatch off the currently 1460 // viewed AST node. We then recursively traverse the AST by calling 1461 // EvalAddr and EvalVal appropriately. 1462 switch (E->getStmtClass()) { 1463 case Stmt::ParenExprClass: 1464 // Ignore parentheses. 1465 return EvalAddr(cast<ParenExpr>(E)->getSubExpr()); 1466 1467 case Stmt::UnaryOperatorClass: { 1468 // The only unary operator that make sense to handle here 1469 // is AddrOf. All others don't make sense as pointers. 1470 UnaryOperator *U = cast<UnaryOperator>(E); 1471 1472 if (U->getOpcode() == UnaryOperator::AddrOf) 1473 return EvalVal(U->getSubExpr()); 1474 else 1475 return NULL; 1476 } 1477 1478 case Stmt::BinaryOperatorClass: { 1479 // Handle pointer arithmetic. All other binary operators are not valid 1480 // in this context. 1481 BinaryOperator *B = cast<BinaryOperator>(E); 1482 BinaryOperator::Opcode op = B->getOpcode(); 1483 1484 if (op != BinaryOperator::Add && op != BinaryOperator::Sub) 1485 return NULL; 1486 1487 Expr *Base = B->getLHS(); 1488 1489 // Determine which argument is the real pointer base. It could be 1490 // the RHS argument instead of the LHS. 1491 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 1492 1493 assert (Base->getType()->isPointerType()); 1494 return EvalAddr(Base); 1495 } 1496 1497 // For conditional operators we need to see if either the LHS or RHS are 1498 // valid DeclRefExpr*s. If one of them is valid, we return it. 1499 case Stmt::ConditionalOperatorClass: { 1500 ConditionalOperator *C = cast<ConditionalOperator>(E); 1501 1502 // Handle the GNU extension for missing LHS. 1503 if (Expr *lhsExpr = C->getLHS()) 1504 if (DeclRefExpr* LHS = EvalAddr(lhsExpr)) 1505 return LHS; 1506 1507 return EvalAddr(C->getRHS()); 1508 } 1509 1510 // For casts, we need to handle conversions from arrays to 1511 // pointer values, and pointer-to-pointer conversions. 1512 case Stmt::ImplicitCastExprClass: 1513 case Stmt::CStyleCastExprClass: 1514 case Stmt::CXXFunctionalCastExprClass: { 1515 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 1516 QualType T = SubExpr->getType(); 1517 1518 if (SubExpr->getType()->isPointerType() || 1519 SubExpr->getType()->isBlockPointerType() || 1520 SubExpr->getType()->isObjCQualifiedIdType()) 1521 return EvalAddr(SubExpr); 1522 else if (T->isArrayType()) 1523 return EvalVal(SubExpr); 1524 else 1525 return 0; 1526 } 1527 1528 // C++ casts. For dynamic casts, static casts, and const casts, we 1529 // are always converting from a pointer-to-pointer, so we just blow 1530 // through the cast. In the case the dynamic cast doesn't fail (and 1531 // return NULL), we take the conservative route and report cases 1532 // where we return the address of a stack variable. For Reinterpre 1533 // FIXME: The comment about is wrong; we're not always converting 1534 // from pointer to pointer. I'm guessing that this code should also 1535 // handle references to objects. 1536 case Stmt::CXXStaticCastExprClass: 1537 case Stmt::CXXDynamicCastExprClass: 1538 case Stmt::CXXConstCastExprClass: 1539 case Stmt::CXXReinterpretCastExprClass: { 1540 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 1541 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 1542 return EvalAddr(S); 1543 else 1544 return NULL; 1545 } 1546 1547 // Everything else: we simply don't reason about them. 1548 default: 1549 return NULL; 1550 } 1551} 1552 1553 1554/// EvalVal - This function is complements EvalAddr in the mutual recursion. 1555/// See the comments for EvalAddr for more details. 1556static DeclRefExpr* EvalVal(Expr *E) { 1557 1558 // We should only be called for evaluating non-pointer expressions, or 1559 // expressions with a pointer type that are not used as references but instead 1560 // are l-values (e.g., DeclRefExpr with a pointer type). 1561 1562 // Our "symbolic interpreter" is just a dispatch off the currently 1563 // viewed AST node. We then recursively traverse the AST by calling 1564 // EvalAddr and EvalVal appropriately. 1565 switch (E->getStmtClass()) { 1566 case Stmt::DeclRefExprClass: { 1567 // DeclRefExpr: the base case. When we hit a DeclRefExpr we are looking 1568 // at code that refers to a variable's name. We check if it has local 1569 // storage within the function, and if so, return the expression. 1570 DeclRefExpr *DR = cast<DeclRefExpr>(E); 1571 1572 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 1573 if (V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR; 1574 1575 return NULL; 1576 } 1577 1578 case Stmt::ParenExprClass: 1579 // Ignore parentheses. 1580 return EvalVal(cast<ParenExpr>(E)->getSubExpr()); 1581 1582 case Stmt::UnaryOperatorClass: { 1583 // The only unary operator that make sense to handle here 1584 // is Deref. All others don't resolve to a "name." This includes 1585 // handling all sorts of rvalues passed to a unary operator. 1586 UnaryOperator *U = cast<UnaryOperator>(E); 1587 1588 if (U->getOpcode() == UnaryOperator::Deref) 1589 return EvalAddr(U->getSubExpr()); 1590 1591 return NULL; 1592 } 1593 1594 case Stmt::ArraySubscriptExprClass: { 1595 // Array subscripts are potential references to data on the stack. We 1596 // retrieve the DeclRefExpr* for the array variable if it indeed 1597 // has local storage. 1598 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase()); 1599 } 1600 1601 case Stmt::ConditionalOperatorClass: { 1602 // For conditional operators we need to see if either the LHS or RHS are 1603 // non-NULL DeclRefExpr's. If one is non-NULL, we return it. 1604 ConditionalOperator *C = cast<ConditionalOperator>(E); 1605 1606 // Handle the GNU extension for missing LHS. 1607 if (Expr *lhsExpr = C->getLHS()) 1608 if (DeclRefExpr *LHS = EvalVal(lhsExpr)) 1609 return LHS; 1610 1611 return EvalVal(C->getRHS()); 1612 } 1613 1614 // Accesses to members are potential references to data on the stack. 1615 case Stmt::MemberExprClass: { 1616 MemberExpr *M = cast<MemberExpr>(E); 1617 1618 // Check for indirect access. We only want direct field accesses. 1619 if (!M->isArrow()) 1620 return EvalVal(M->getBase()); 1621 else 1622 return NULL; 1623 } 1624 1625 // Everything else: we simply don't reason about them. 1626 default: 1627 return NULL; 1628 } 1629} 1630 1631//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 1632 1633/// Check for comparisons of floating point operands using != and ==. 1634/// Issue a warning if these are no self-comparisons, as they are not likely 1635/// to do what the programmer intended. 1636void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 1637 bool EmitWarning = true; 1638 1639 Expr* LeftExprSansParen = lex->IgnoreParens(); 1640 Expr* RightExprSansParen = rex->IgnoreParens(); 1641 1642 // Special case: check for x == x (which is OK). 1643 // Do not emit warnings for such cases. 1644 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 1645 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 1646 if (DRL->getDecl() == DRR->getDecl()) 1647 EmitWarning = false; 1648 1649 1650 // Special case: check for comparisons against literals that can be exactly 1651 // represented by APFloat. In such cases, do not emit a warning. This 1652 // is a heuristic: often comparison against such literals are used to 1653 // detect if a value in a variable has not changed. This clearly can 1654 // lead to false negatives. 1655 if (EmitWarning) { 1656 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 1657 if (FLL->isExact()) 1658 EmitWarning = false; 1659 } else 1660 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 1661 if (FLR->isExact()) 1662 EmitWarning = false; 1663 } 1664 } 1665 1666 // Check for comparisons with builtin types. 1667 if (EmitWarning) 1668 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 1669 if (CL->isBuiltinCall(Context)) 1670 EmitWarning = false; 1671 1672 if (EmitWarning) 1673 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 1674 if (CR->isBuiltinCall(Context)) 1675 EmitWarning = false; 1676 1677 // Emit the diagnostic. 1678 if (EmitWarning) 1679 Diag(loc, diag::warn_floatingpoint_eq) 1680 << lex->getSourceRange() << rex->getSourceRange(); 1681} 1682 1683//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 1684//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 1685 1686namespace { 1687 1688/// Structure recording the 'active' range of an integer-valued 1689/// expression. 1690struct IntRange { 1691 /// The number of bits active in the int. 1692 unsigned Width; 1693 1694 /// True if the int is known not to have negative values. 1695 bool NonNegative; 1696 1697 IntRange() {} 1698 IntRange(unsigned Width, bool NonNegative) 1699 : Width(Width), NonNegative(NonNegative) 1700 {} 1701 1702 // Returns the range of the bool type. 1703 static IntRange forBoolType() { 1704 return IntRange(1, true); 1705 } 1706 1707 // Returns the range of an integral type. 1708 static IntRange forType(ASTContext &C, QualType T) { 1709 return forCanonicalType(C, T->getCanonicalTypeInternal().getTypePtr()); 1710 } 1711 1712 // Returns the range of an integeral type based on its canonical 1713 // representation. 1714 static IntRange forCanonicalType(ASTContext &C, const Type *T) { 1715 assert(T->isCanonicalUnqualified()); 1716 1717 if (const VectorType *VT = dyn_cast<VectorType>(T)) 1718 T = VT->getElementType().getTypePtr(); 1719 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 1720 T = CT->getElementType().getTypePtr(); 1721 if (const EnumType *ET = dyn_cast<EnumType>(T)) 1722 T = ET->getDecl()->getIntegerType().getTypePtr(); 1723 1724 const BuiltinType *BT = cast<BuiltinType>(T); 1725 assert(BT->isInteger()); 1726 1727 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 1728 } 1729 1730 // Returns the supremum of two ranges: i.e. their conservative merge. 1731 static IntRange join(IntRange L, IntRange R) { 1732 return IntRange(std::max(L.Width, R.Width), 1733 L.NonNegative && R.NonNegative); 1734 } 1735 1736 // Returns the infinum of two ranges: i.e. their aggressive merge. 1737 static IntRange meet(IntRange L, IntRange R) { 1738 return IntRange(std::min(L.Width, R.Width), 1739 L.NonNegative || R.NonNegative); 1740 } 1741}; 1742 1743IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 1744 if (value.isSigned() && value.isNegative()) 1745 return IntRange(value.getMinSignedBits(), false); 1746 1747 if (value.getBitWidth() > MaxWidth) 1748 value.trunc(MaxWidth); 1749 1750 // isNonNegative() just checks the sign bit without considering 1751 // signedness. 1752 return IntRange(value.getActiveBits(), true); 1753} 1754 1755IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 1756 unsigned MaxWidth) { 1757 if (result.isInt()) 1758 return GetValueRange(C, result.getInt(), MaxWidth); 1759 1760 if (result.isVector()) { 1761 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 1762 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 1763 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 1764 R = IntRange::join(R, El); 1765 } 1766 return R; 1767 } 1768 1769 if (result.isComplexInt()) { 1770 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 1771 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 1772 return IntRange::join(R, I); 1773 } 1774 1775 // This can happen with lossless casts to intptr_t of "based" lvalues. 1776 // Assume it might use arbitrary bits. 1777 // FIXME: The only reason we need to pass the type in here is to get 1778 // the sign right on this one case. It would be nice if APValue 1779 // preserved this. 1780 assert(result.isLValue()); 1781 return IntRange(MaxWidth, Ty->isUnsignedIntegerType()); 1782} 1783 1784/// Pseudo-evaluate the given integer expression, estimating the 1785/// range of values it might take. 1786/// 1787/// \param MaxWidth - the width to which the value will be truncated 1788IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 1789 E = E->IgnoreParens(); 1790 1791 // Try a full evaluation first. 1792 Expr::EvalResult result; 1793 if (E->Evaluate(result, C)) 1794 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 1795 1796 // I think we only want to look through implicit casts here; if the 1797 // user has an explicit widening cast, we should treat the value as 1798 // being of the new, wider type. 1799 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 1800 if (CE->getCastKind() == CastExpr::CK_NoOp) 1801 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 1802 1803 IntRange OutputTypeRange = IntRange::forType(C, CE->getType()); 1804 1805 bool isIntegerCast = (CE->getCastKind() == CastExpr::CK_IntegralCast); 1806 if (!isIntegerCast && CE->getCastKind() == CastExpr::CK_Unknown) 1807 isIntegerCast = CE->getSubExpr()->getType()->isIntegerType(); 1808 1809 // Assume that non-integer casts can span the full range of the type. 1810 if (!isIntegerCast) 1811 return OutputTypeRange; 1812 1813 IntRange SubRange 1814 = GetExprRange(C, CE->getSubExpr(), 1815 std::min(MaxWidth, OutputTypeRange.Width)); 1816 1817 // Bail out if the subexpr's range is as wide as the cast type. 1818 if (SubRange.Width >= OutputTypeRange.Width) 1819 return OutputTypeRange; 1820 1821 // Otherwise, we take the smaller width, and we're non-negative if 1822 // either the output type or the subexpr is. 1823 return IntRange(SubRange.Width, 1824 SubRange.NonNegative || OutputTypeRange.NonNegative); 1825 } 1826 1827 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 1828 // If we can fold the condition, just take that operand. 1829 bool CondResult; 1830 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 1831 return GetExprRange(C, CondResult ? CO->getTrueExpr() 1832 : CO->getFalseExpr(), 1833 MaxWidth); 1834 1835 // Otherwise, conservatively merge. 1836 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 1837 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 1838 return IntRange::join(L, R); 1839 } 1840 1841 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 1842 switch (BO->getOpcode()) { 1843 1844 // Boolean-valued operations are single-bit and positive. 1845 case BinaryOperator::LAnd: 1846 case BinaryOperator::LOr: 1847 case BinaryOperator::LT: 1848 case BinaryOperator::GT: 1849 case BinaryOperator::LE: 1850 case BinaryOperator::GE: 1851 case BinaryOperator::EQ: 1852 case BinaryOperator::NE: 1853 return IntRange::forBoolType(); 1854 1855 // The type of these compound assignments is the type of the LHS, 1856 // so the RHS is not necessarily an integer. 1857 case BinaryOperator::MulAssign: 1858 case BinaryOperator::DivAssign: 1859 case BinaryOperator::RemAssign: 1860 case BinaryOperator::AddAssign: 1861 case BinaryOperator::SubAssign: 1862 return IntRange::forType(C, E->getType()); 1863 1864 // Operations with opaque sources are black-listed. 1865 case BinaryOperator::PtrMemD: 1866 case BinaryOperator::PtrMemI: 1867 return IntRange::forType(C, E->getType()); 1868 1869 // Bitwise-and uses the *infinum* of the two source ranges. 1870 case BinaryOperator::And: 1871 case BinaryOperator::AndAssign: 1872 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 1873 GetExprRange(C, BO->getRHS(), MaxWidth)); 1874 1875 // Left shift gets black-listed based on a judgement call. 1876 case BinaryOperator::Shl: 1877 // ...except that we want to treat '1 << (blah)' as logically 1878 // positive. It's an important idiom. 1879 if (IntegerLiteral *I 1880 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 1881 if (I->getValue() == 1) { 1882 IntRange R = IntRange::forType(C, E->getType()); 1883 return IntRange(R.Width, /*NonNegative*/ true); 1884 } 1885 } 1886 // fallthrough 1887 1888 case BinaryOperator::ShlAssign: 1889 return IntRange::forType(C, E->getType()); 1890 1891 // Right shift by a constant can narrow its left argument. 1892 case BinaryOperator::Shr: 1893 case BinaryOperator::ShrAssign: { 1894 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 1895 1896 // If the shift amount is a positive constant, drop the width by 1897 // that much. 1898 llvm::APSInt shift; 1899 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 1900 shift.isNonNegative()) { 1901 unsigned zext = shift.getZExtValue(); 1902 if (zext >= L.Width) 1903 L.Width = (L.NonNegative ? 0 : 1); 1904 else 1905 L.Width -= zext; 1906 } 1907 1908 return L; 1909 } 1910 1911 // Comma acts as its right operand. 1912 case BinaryOperator::Comma: 1913 return GetExprRange(C, BO->getRHS(), MaxWidth); 1914 1915 // Black-list pointer subtractions. 1916 case BinaryOperator::Sub: 1917 if (BO->getLHS()->getType()->isPointerType()) 1918 return IntRange::forType(C, E->getType()); 1919 // fallthrough 1920 1921 default: 1922 break; 1923 } 1924 1925 // Treat every other operator as if it were closed on the 1926 // narrowest type that encompasses both operands. 1927 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 1928 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 1929 return IntRange::join(L, R); 1930 } 1931 1932 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 1933 switch (UO->getOpcode()) { 1934 // Boolean-valued operations are white-listed. 1935 case UnaryOperator::LNot: 1936 return IntRange::forBoolType(); 1937 1938 // Operations with opaque sources are black-listed. 1939 case UnaryOperator::Deref: 1940 case UnaryOperator::AddrOf: // should be impossible 1941 case UnaryOperator::OffsetOf: 1942 return IntRange::forType(C, E->getType()); 1943 1944 default: 1945 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 1946 } 1947 } 1948 1949 if (dyn_cast<OffsetOfExpr>(E)) { 1950 IntRange::forType(C, E->getType()); 1951 } 1952 1953 FieldDecl *BitField = E->getBitField(); 1954 if (BitField) { 1955 llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C); 1956 unsigned BitWidth = BitWidthAP.getZExtValue(); 1957 1958 return IntRange(BitWidth, BitField->getType()->isUnsignedIntegerType()); 1959 } 1960 1961 return IntRange::forType(C, E->getType()); 1962} 1963 1964/// Checks whether the given value, which currently has the given 1965/// source semantics, has the same value when coerced through the 1966/// target semantics. 1967bool IsSameFloatAfterCast(const llvm::APFloat &value, 1968 const llvm::fltSemantics &Src, 1969 const llvm::fltSemantics &Tgt) { 1970 llvm::APFloat truncated = value; 1971 1972 bool ignored; 1973 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 1974 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 1975 1976 return truncated.bitwiseIsEqual(value); 1977} 1978 1979/// Checks whether the given value, which currently has the given 1980/// source semantics, has the same value when coerced through the 1981/// target semantics. 1982/// 1983/// The value might be a vector of floats (or a complex number). 1984bool IsSameFloatAfterCast(const APValue &value, 1985 const llvm::fltSemantics &Src, 1986 const llvm::fltSemantics &Tgt) { 1987 if (value.isFloat()) 1988 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 1989 1990 if (value.isVector()) { 1991 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 1992 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 1993 return false; 1994 return true; 1995 } 1996 1997 assert(value.isComplexFloat()); 1998 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 1999 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 2000} 2001 2002} // end anonymous namespace 2003 2004/// \brief Implements -Wsign-compare. 2005/// 2006/// \param lex the left-hand expression 2007/// \param rex the right-hand expression 2008/// \param OpLoc the location of the joining operator 2009/// \param BinOpc binary opcode or 0 2010void Sema::CheckSignCompare(Expr *lex, Expr *rex, SourceLocation OpLoc, 2011 const BinaryOperator::Opcode* BinOpc) { 2012 // Don't warn if we're in an unevaluated context. 2013 if (ExprEvalContexts.back().Context == Unevaluated) 2014 return; 2015 2016 // If either expression is value-dependent, don't warn. We'll get another 2017 // chance at instantiation time. 2018 if (lex->isValueDependent() || rex->isValueDependent()) 2019 return; 2020 2021 QualType lt = lex->getType(), rt = rex->getType(); 2022 2023 // Only warn if both operands are integral. 2024 if (!lt->isIntegerType() || !rt->isIntegerType()) 2025 return; 2026 2027 // In C, the width of a bitfield determines its type, and the 2028 // declared type only contributes the signedness. This duplicates 2029 // the work that will later be done by UsualUnaryConversions. 2030 // Eventually, this check will be reorganized in a way that avoids 2031 // this duplication. 2032 if (!getLangOptions().CPlusPlus) { 2033 QualType tmp; 2034 tmp = Context.isPromotableBitField(lex); 2035 if (!tmp.isNull()) lt = tmp; 2036 tmp = Context.isPromotableBitField(rex); 2037 if (!tmp.isNull()) rt = tmp; 2038 } 2039 2040 if (const EnumType *E = lt->getAs<EnumType>()) 2041 lt = E->getDecl()->getPromotionType(); 2042 if (const EnumType *E = rt->getAs<EnumType>()) 2043 rt = E->getDecl()->getPromotionType(); 2044 2045 // The rule is that the signed operand becomes unsigned, so isolate the 2046 // signed operand. 2047 Expr *signedOperand = lex, *unsignedOperand = rex; 2048 QualType signedType = lt, unsignedType = rt; 2049 if (lt->isSignedIntegerType()) { 2050 if (rt->isSignedIntegerType()) return; 2051 } else { 2052 if (!rt->isSignedIntegerType()) return; 2053 std::swap(signedOperand, unsignedOperand); 2054 std::swap(signedType, unsignedType); 2055 } 2056 2057 unsigned unsignedWidth = Context.getIntWidth(unsignedType); 2058 unsigned signedWidth = Context.getIntWidth(signedType); 2059 2060 // If the unsigned type is strictly smaller than the signed type, 2061 // then (1) the result type will be signed and (2) the unsigned 2062 // value will fit fully within the signed type, and thus the result 2063 // of the comparison will be exact. 2064 if (signedWidth > unsignedWidth) 2065 return; 2066 2067 // Otherwise, calculate the effective ranges. 2068 IntRange signedRange = GetExprRange(Context, signedOperand, signedWidth); 2069 IntRange unsignedRange = GetExprRange(Context, unsignedOperand, unsignedWidth); 2070 2071 // We should never be unable to prove that the unsigned operand is 2072 // non-negative. 2073 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 2074 2075 // If the signed operand is non-negative, then the signed->unsigned 2076 // conversion won't change it. 2077 if (signedRange.NonNegative) { 2078 // Emit warnings for comparisons of unsigned to integer constant 0. 2079 // always false: x < 0 (or 0 > x) 2080 // always true: x >= 0 (or 0 <= x) 2081 llvm::APSInt X; 2082 if (BinOpc && signedOperand->isIntegerConstantExpr(X, Context) && X == 0) { 2083 if (signedOperand != lex) { 2084 if (*BinOpc == BinaryOperator::LT) { 2085 Diag(OpLoc, diag::warn_lunsigned_always_true_comparison) 2086 << "< 0" << "false" 2087 << lex->getSourceRange() << rex->getSourceRange(); 2088 } 2089 else if (*BinOpc == BinaryOperator::GE) { 2090 Diag(OpLoc, diag::warn_lunsigned_always_true_comparison) 2091 << ">= 0" << "true" 2092 << lex->getSourceRange() << rex->getSourceRange(); 2093 } 2094 } 2095 else { 2096 if (*BinOpc == BinaryOperator::GT) { 2097 Diag(OpLoc, diag::warn_runsigned_always_true_comparison) 2098 << "0 >" << "false" 2099 << lex->getSourceRange() << rex->getSourceRange(); 2100 } 2101 else if (*BinOpc == BinaryOperator::LE) { 2102 Diag(OpLoc, diag::warn_runsigned_always_true_comparison) 2103 << "0 <=" << "true" 2104 << lex->getSourceRange() << rex->getSourceRange(); 2105 } 2106 } 2107 } 2108 return; 2109 } 2110 2111 // For (in)equality comparisons, if the unsigned operand is a 2112 // constant which cannot collide with a overflowed signed operand, 2113 // then reinterpreting the signed operand as unsigned will not 2114 // change the result of the comparison. 2115 if (BinOpc && 2116 (*BinOpc == BinaryOperator::EQ || *BinOpc == BinaryOperator::NE) && 2117 unsignedRange.Width < unsignedWidth) 2118 return; 2119 2120 Diag(OpLoc, BinOpc ? diag::warn_mixed_sign_comparison 2121 : diag::warn_mixed_sign_conditional) 2122 << lt << rt << lex->getSourceRange() << rex->getSourceRange(); 2123} 2124 2125/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 2126static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, unsigned diag) { 2127 S.Diag(E->getExprLoc(), diag) << E->getType() << T << E->getSourceRange(); 2128} 2129 2130/// Implements -Wconversion. 2131void Sema::CheckImplicitConversion(Expr *E, QualType T) { 2132 // Don't diagnose in unevaluated contexts. 2133 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 2134 return; 2135 2136 // Don't diagnose for value-dependent expressions. 2137 if (E->isValueDependent()) 2138 return; 2139 2140 const Type *Source = Context.getCanonicalType(E->getType()).getTypePtr(); 2141 const Type *Target = Context.getCanonicalType(T).getTypePtr(); 2142 2143 // Never diagnose implicit casts to bool. 2144 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 2145 return; 2146 2147 // Strip vector types. 2148 if (isa<VectorType>(Source)) { 2149 if (!isa<VectorType>(Target)) 2150 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_vector_scalar); 2151 2152 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 2153 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 2154 } 2155 2156 // Strip complex types. 2157 if (isa<ComplexType>(Source)) { 2158 if (!isa<ComplexType>(Target)) 2159 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_complex_scalar); 2160 2161 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 2162 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 2163 } 2164 2165 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 2166 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 2167 2168 // If the source is floating point... 2169 if (SourceBT && SourceBT->isFloatingPoint()) { 2170 // ...and the target is floating point... 2171 if (TargetBT && TargetBT->isFloatingPoint()) { 2172 // ...then warn if we're dropping FP rank. 2173 2174 // Builtin FP kinds are ordered by increasing FP rank. 2175 if (SourceBT->getKind() > TargetBT->getKind()) { 2176 // Don't warn about float constants that are precisely 2177 // representable in the target type. 2178 Expr::EvalResult result; 2179 if (E->Evaluate(result, Context)) { 2180 // Value might be a float, a float vector, or a float complex. 2181 if (IsSameFloatAfterCast(result.Val, 2182 Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 2183 Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 2184 return; 2185 } 2186 2187 DiagnoseImpCast(*this, E, T, diag::warn_impcast_float_precision); 2188 } 2189 return; 2190 } 2191 2192 // If the target is integral, always warn. 2193 if ((TargetBT && TargetBT->isInteger())) 2194 // TODO: don't warn for integer values? 2195 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_float_integer); 2196 2197 return; 2198 } 2199 2200 if (!Source->isIntegerType() || !Target->isIntegerType()) 2201 return; 2202 2203 IntRange SourceRange = GetExprRange(Context, E, Context.getIntWidth(E->getType())); 2204 IntRange TargetRange = IntRange::forCanonicalType(Context, Target); 2205 2206 // FIXME: also signed<->unsigned? 2207 2208 if (SourceRange.Width > TargetRange.Width) { 2209 // People want to build with -Wshorten-64-to-32 and not -Wconversion 2210 // and by god we'll let them. 2211 if (SourceRange.Width == 64 && TargetRange.Width == 32) 2212 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_integer_64_32); 2213 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_integer_precision); 2214 } 2215 2216 return; 2217} 2218 2219/// CheckParmsForFunctionDef - Check that the parameters of the given 2220/// function are appropriate for the definition of a function. This 2221/// takes care of any checks that cannot be performed on the 2222/// declaration itself, e.g., that the types of each of the function 2223/// parameters are complete. 2224bool Sema::CheckParmsForFunctionDef(FunctionDecl *FD) { 2225 bool HasInvalidParm = false; 2226 for (unsigned p = 0, NumParams = FD->getNumParams(); p < NumParams; ++p) { 2227 ParmVarDecl *Param = FD->getParamDecl(p); 2228 2229 // C99 6.7.5.3p4: the parameters in a parameter type list in a 2230 // function declarator that is part of a function definition of 2231 // that function shall not have incomplete type. 2232 // 2233 // This is also C++ [dcl.fct]p6. 2234 if (!Param->isInvalidDecl() && 2235 RequireCompleteType(Param->getLocation(), Param->getType(), 2236 diag::err_typecheck_decl_incomplete_type)) { 2237 Param->setInvalidDecl(); 2238 HasInvalidParm = true; 2239 } 2240 2241 // C99 6.9.1p5: If the declarator includes a parameter type list, the 2242 // declaration of each parameter shall include an identifier. 2243 if (Param->getIdentifier() == 0 && 2244 !Param->isImplicit() && 2245 !getLangOptions().CPlusPlus) 2246 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 2247 2248 // C99 6.7.5.3p12: 2249 // If the function declarator is not part of a definition of that 2250 // function, parameters may have incomplete type and may use the [*] 2251 // notation in their sequences of declarator specifiers to specify 2252 // variable length array types. 2253 QualType PType = Param->getOriginalType(); 2254 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 2255 if (AT->getSizeModifier() == ArrayType::Star) { 2256 // FIXME: This diagnosic should point the the '[*]' if source-location 2257 // information is added for it. 2258 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 2259 } 2260 } 2261 } 2262 2263 return HasInvalidParm; 2264} 2265