SemaChecking.cpp revision aea67dbd653a2dd6dd5cc2159279e81e855b2482
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/AnalysisContext.h" 17#include "clang/Analysis/CFG.h" 18#include "clang/Analysis/Analyses/ReachableCode.h" 19#include "clang/Analysis/Analyses/PrintfFormatString.h" 20#include "clang/AST/ASTContext.h" 21#include "clang/AST/CharUnits.h" 22#include "clang/AST/DeclObjC.h" 23#include "clang/AST/ExprCXX.h" 24#include "clang/AST/ExprObjC.h" 25#include "clang/AST/DeclObjC.h" 26#include "clang/AST/StmtCXX.h" 27#include "clang/AST/StmtObjC.h" 28#include "clang/Lex/LiteralSupport.h" 29#include "clang/Lex/Preprocessor.h" 30#include "llvm/ADT/BitVector.h" 31#include "llvm/ADT/STLExtras.h" 32#include <limits> 33#include <queue> 34using namespace clang; 35 36/// getLocationOfStringLiteralByte - Return a source location that points to the 37/// specified byte of the specified string literal. 38/// 39/// Strings are amazingly complex. They can be formed from multiple tokens and 40/// can have escape sequences in them in addition to the usual trigraph and 41/// escaped newline business. This routine handles this complexity. 42/// 43SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 44 unsigned ByteNo) const { 45 assert(!SL->isWide() && "This doesn't work for wide strings yet"); 46 47 // Loop over all of the tokens in this string until we find the one that 48 // contains the byte we're looking for. 49 unsigned TokNo = 0; 50 while (1) { 51 assert(TokNo < SL->getNumConcatenated() && "Invalid byte number!"); 52 SourceLocation StrTokLoc = SL->getStrTokenLoc(TokNo); 53 54 // Get the spelling of the string so that we can get the data that makes up 55 // the string literal, not the identifier for the macro it is potentially 56 // expanded through. 57 SourceLocation StrTokSpellingLoc = SourceMgr.getSpellingLoc(StrTokLoc); 58 59 // Re-lex the token to get its length and original spelling. 60 std::pair<FileID, unsigned> LocInfo = 61 SourceMgr.getDecomposedLoc(StrTokSpellingLoc); 62 std::pair<const char *,const char *> Buffer = 63 SourceMgr.getBufferData(LocInfo.first, Diags); 64 if (!Buffer.first) 65 return StrTokSpellingLoc; 66 67 const char *StrData = Buffer.first+LocInfo.second; 68 69 // Create a langops struct and enable trigraphs. This is sufficient for 70 // relexing tokens. 71 LangOptions LangOpts; 72 LangOpts.Trigraphs = true; 73 74 // Create a lexer starting at the beginning of this token. 75 Lexer TheLexer(StrTokSpellingLoc, LangOpts, Buffer.first, StrData, 76 Buffer.second); 77 Token TheTok; 78 TheLexer.LexFromRawLexer(TheTok); 79 80 // Use the StringLiteralParser to compute the length of the string in bytes. 81 StringLiteralParser SLP(&TheTok, 1, PP); 82 unsigned TokNumBytes = SLP.GetStringLength(); 83 84 // If the byte is in this token, return the location of the byte. 85 if (ByteNo < TokNumBytes || 86 (ByteNo == TokNumBytes && TokNo == SL->getNumConcatenated())) { 87 unsigned Offset = 88 StringLiteralParser::getOffsetOfStringByte(TheTok, ByteNo, PP); 89 90 // Now that we know the offset of the token in the spelling, use the 91 // preprocessor to get the offset in the original source. 92 return PP.AdvanceToTokenCharacter(StrTokLoc, Offset); 93 } 94 95 // Move to the next string token. 96 ++TokNo; 97 ByteNo -= TokNumBytes; 98 } 99} 100 101/// CheckablePrintfAttr - does a function call have a "printf" attribute 102/// and arguments that merit checking? 103bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) { 104 if (Format->getType() == "printf") return true; 105 if (Format->getType() == "printf0") { 106 // printf0 allows null "format" string; if so don't check format/args 107 unsigned format_idx = Format->getFormatIdx() - 1; 108 // Does the index refer to the implicit object argument? 109 if (isa<CXXMemberCallExpr>(TheCall)) { 110 if (format_idx == 0) 111 return false; 112 --format_idx; 113 } 114 if (format_idx < TheCall->getNumArgs()) { 115 Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts(); 116 if (!Format->isNullPointerConstant(Context, 117 Expr::NPC_ValueDependentIsNull)) 118 return true; 119 } 120 } 121 return false; 122} 123 124Action::OwningExprResult 125Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 126 OwningExprResult TheCallResult(Owned(TheCall)); 127 128 switch (BuiltinID) { 129 case Builtin::BI__builtin___CFStringMakeConstantString: 130 assert(TheCall->getNumArgs() == 1 && 131 "Wrong # arguments to builtin CFStringMakeConstantString"); 132 if (CheckObjCString(TheCall->getArg(0))) 133 return ExprError(); 134 break; 135 case Builtin::BI__builtin_stdarg_start: 136 case Builtin::BI__builtin_va_start: 137 if (SemaBuiltinVAStart(TheCall)) 138 return ExprError(); 139 break; 140 case Builtin::BI__builtin_isgreater: 141 case Builtin::BI__builtin_isgreaterequal: 142 case Builtin::BI__builtin_isless: 143 case Builtin::BI__builtin_islessequal: 144 case Builtin::BI__builtin_islessgreater: 145 case Builtin::BI__builtin_isunordered: 146 if (SemaBuiltinUnorderedCompare(TheCall)) 147 return ExprError(); 148 break; 149 case Builtin::BI__builtin_fpclassify: 150 if (SemaBuiltinFPClassification(TheCall, 6)) 151 return ExprError(); 152 break; 153 case Builtin::BI__builtin_isfinite: 154 case Builtin::BI__builtin_isinf: 155 case Builtin::BI__builtin_isinf_sign: 156 case Builtin::BI__builtin_isnan: 157 case Builtin::BI__builtin_isnormal: 158 if (SemaBuiltinFPClassification(TheCall, 1)) 159 return ExprError(); 160 break; 161 case Builtin::BI__builtin_return_address: 162 case Builtin::BI__builtin_frame_address: 163 if (SemaBuiltinStackAddress(TheCall)) 164 return ExprError(); 165 break; 166 case Builtin::BI__builtin_eh_return_data_regno: 167 if (SemaBuiltinEHReturnDataRegNo(TheCall)) 168 return ExprError(); 169 break; 170 case Builtin::BI__builtin_shufflevector: 171 return SemaBuiltinShuffleVector(TheCall); 172 // TheCall will be freed by the smart pointer here, but that's fine, since 173 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 174 case Builtin::BI__builtin_prefetch: 175 if (SemaBuiltinPrefetch(TheCall)) 176 return ExprError(); 177 break; 178 case Builtin::BI__builtin_object_size: 179 if (SemaBuiltinObjectSize(TheCall)) 180 return ExprError(); 181 break; 182 case Builtin::BI__builtin_longjmp: 183 if (SemaBuiltinLongjmp(TheCall)) 184 return ExprError(); 185 break; 186 case Builtin::BI__sync_fetch_and_add: 187 case Builtin::BI__sync_fetch_and_sub: 188 case Builtin::BI__sync_fetch_and_or: 189 case Builtin::BI__sync_fetch_and_and: 190 case Builtin::BI__sync_fetch_and_xor: 191 case Builtin::BI__sync_fetch_and_nand: 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_nand_and_fetch: 198 case Builtin::BI__sync_val_compare_and_swap: 199 case Builtin::BI__sync_bool_compare_and_swap: 200 case Builtin::BI__sync_lock_test_and_set: 201 case Builtin::BI__sync_lock_release: 202 if (SemaBuiltinAtomicOverloaded(TheCall)) 203 return ExprError(); 204 break; 205 } 206 207 return move(TheCallResult); 208} 209 210/// CheckFunctionCall - Check a direct function call for various correctness 211/// and safety properties not strictly enforced by the C type system. 212bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 213 // Get the IdentifierInfo* for the called function. 214 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 215 216 // None of the checks below are needed for functions that don't have 217 // simple names (e.g., C++ conversion functions). 218 if (!FnInfo) 219 return false; 220 221 // FIXME: This mechanism should be abstracted to be less fragile and 222 // more efficient. For example, just map function ids to custom 223 // handlers. 224 225 // Printf checking. 226 if (const FormatAttr *Format = FDecl->getAttr<FormatAttr>()) { 227 if (CheckablePrintfAttr(Format, TheCall)) { 228 bool HasVAListArg = Format->getFirstArg() == 0; 229 if (!HasVAListArg) { 230 if (const FunctionProtoType *Proto 231 = FDecl->getType()->getAs<FunctionProtoType>()) 232 HasVAListArg = !Proto->isVariadic(); 233 } 234 CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 235 HasVAListArg ? 0 : Format->getFirstArg() - 1); 236 } 237 } 238 239 for (const NonNullAttr *NonNull = FDecl->getAttr<NonNullAttr>(); NonNull; 240 NonNull = NonNull->getNext<NonNullAttr>()) 241 CheckNonNullArguments(NonNull, TheCall); 242 243 return false; 244} 245 246bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 247 // Printf checking. 248 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 249 if (!Format) 250 return false; 251 252 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 253 if (!V) 254 return false; 255 256 QualType Ty = V->getType(); 257 if (!Ty->isBlockPointerType()) 258 return false; 259 260 if (!CheckablePrintfAttr(Format, TheCall)) 261 return false; 262 263 bool HasVAListArg = Format->getFirstArg() == 0; 264 if (!HasVAListArg) { 265 const FunctionType *FT = 266 Ty->getAs<BlockPointerType>()->getPointeeType()->getAs<FunctionType>(); 267 if (const FunctionProtoType *Proto = dyn_cast<FunctionProtoType>(FT)) 268 HasVAListArg = !Proto->isVariadic(); 269 } 270 CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 271 HasVAListArg ? 0 : Format->getFirstArg() - 1); 272 273 return false; 274} 275 276/// SemaBuiltinAtomicOverloaded - We have a call to a function like 277/// __sync_fetch_and_add, which is an overloaded function based on the pointer 278/// type of its first argument. The main ActOnCallExpr routines have already 279/// promoted the types of arguments because all of these calls are prototyped as 280/// void(...). 281/// 282/// This function goes through and does final semantic checking for these 283/// builtins, 284bool Sema::SemaBuiltinAtomicOverloaded(CallExpr *TheCall) { 285 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 286 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 287 288 // Ensure that we have at least one argument to do type inference from. 289 if (TheCall->getNumArgs() < 1) 290 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 291 << 0 << TheCall->getCallee()->getSourceRange(); 292 293 // Inspect the first argument of the atomic builtin. This should always be 294 // a pointer type, whose element is an integral scalar or pointer type. 295 // Because it is a pointer type, we don't have to worry about any implicit 296 // casts here. 297 Expr *FirstArg = TheCall->getArg(0); 298 if (!FirstArg->getType()->isPointerType()) 299 return Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 300 << FirstArg->getType() << FirstArg->getSourceRange(); 301 302 QualType ValType = FirstArg->getType()->getAs<PointerType>()->getPointeeType(); 303 if (!ValType->isIntegerType() && !ValType->isPointerType() && 304 !ValType->isBlockPointerType()) 305 return Diag(DRE->getLocStart(), 306 diag::err_atomic_builtin_must_be_pointer_intptr) 307 << FirstArg->getType() << FirstArg->getSourceRange(); 308 309 // We need to figure out which concrete builtin this maps onto. For example, 310 // __sync_fetch_and_add with a 2 byte object turns into 311 // __sync_fetch_and_add_2. 312#define BUILTIN_ROW(x) \ 313 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 314 Builtin::BI##x##_8, Builtin::BI##x##_16 } 315 316 static const unsigned BuiltinIndices[][5] = { 317 BUILTIN_ROW(__sync_fetch_and_add), 318 BUILTIN_ROW(__sync_fetch_and_sub), 319 BUILTIN_ROW(__sync_fetch_and_or), 320 BUILTIN_ROW(__sync_fetch_and_and), 321 BUILTIN_ROW(__sync_fetch_and_xor), 322 BUILTIN_ROW(__sync_fetch_and_nand), 323 324 BUILTIN_ROW(__sync_add_and_fetch), 325 BUILTIN_ROW(__sync_sub_and_fetch), 326 BUILTIN_ROW(__sync_and_and_fetch), 327 BUILTIN_ROW(__sync_or_and_fetch), 328 BUILTIN_ROW(__sync_xor_and_fetch), 329 BUILTIN_ROW(__sync_nand_and_fetch), 330 331 BUILTIN_ROW(__sync_val_compare_and_swap), 332 BUILTIN_ROW(__sync_bool_compare_and_swap), 333 BUILTIN_ROW(__sync_lock_test_and_set), 334 BUILTIN_ROW(__sync_lock_release) 335 }; 336#undef BUILTIN_ROW 337 338 // Determine the index of the size. 339 unsigned SizeIndex; 340 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 341 case 1: SizeIndex = 0; break; 342 case 2: SizeIndex = 1; break; 343 case 4: SizeIndex = 2; break; 344 case 8: SizeIndex = 3; break; 345 case 16: SizeIndex = 4; break; 346 default: 347 return Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 348 << FirstArg->getType() << FirstArg->getSourceRange(); 349 } 350 351 // Each of these builtins has one pointer argument, followed by some number of 352 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 353 // that we ignore. Find out which row of BuiltinIndices to read from as well 354 // as the number of fixed args. 355 unsigned BuiltinID = FDecl->getBuiltinID(); 356 unsigned BuiltinIndex, NumFixed = 1; 357 switch (BuiltinID) { 358 default: assert(0 && "Unknown overloaded atomic builtin!"); 359 case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break; 360 case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break; 361 case Builtin::BI__sync_fetch_and_or: BuiltinIndex = 2; break; 362 case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break; 363 case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break; 364 case Builtin::BI__sync_fetch_and_nand:BuiltinIndex = 5; break; 365 366 case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 6; break; 367 case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 7; break; 368 case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 8; break; 369 case Builtin::BI__sync_or_and_fetch: BuiltinIndex = 9; break; 370 case Builtin::BI__sync_xor_and_fetch: BuiltinIndex =10; break; 371 case Builtin::BI__sync_nand_and_fetch:BuiltinIndex =11; break; 372 373 case Builtin::BI__sync_val_compare_and_swap: 374 BuiltinIndex = 12; 375 NumFixed = 2; 376 break; 377 case Builtin::BI__sync_bool_compare_and_swap: 378 BuiltinIndex = 13; 379 NumFixed = 2; 380 break; 381 case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 14; break; 382 case Builtin::BI__sync_lock_release: 383 BuiltinIndex = 15; 384 NumFixed = 0; 385 break; 386 } 387 388 // Now that we know how many fixed arguments we expect, first check that we 389 // have at least that many. 390 if (TheCall->getNumArgs() < 1+NumFixed) 391 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 392 << 0 << TheCall->getCallee()->getSourceRange(); 393 394 395 // Get the decl for the concrete builtin from this, we can tell what the 396 // concrete integer type we should convert to is. 397 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 398 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 399 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 400 FunctionDecl *NewBuiltinDecl = 401 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 402 TUScope, false, DRE->getLocStart())); 403 const FunctionProtoType *BuiltinFT = 404 NewBuiltinDecl->getType()->getAs<FunctionProtoType>(); 405 ValType = BuiltinFT->getArgType(0)->getAs<PointerType>()->getPointeeType(); 406 407 // If the first type needs to be converted (e.g. void** -> int*), do it now. 408 if (BuiltinFT->getArgType(0) != FirstArg->getType()) { 409 ImpCastExprToType(FirstArg, BuiltinFT->getArgType(0), CastExpr::CK_BitCast); 410 TheCall->setArg(0, FirstArg); 411 } 412 413 // Next, walk the valid ones promoting to the right type. 414 for (unsigned i = 0; i != NumFixed; ++i) { 415 Expr *Arg = TheCall->getArg(i+1); 416 417 // If the argument is an implicit cast, then there was a promotion due to 418 // "...", just remove it now. 419 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) { 420 Arg = ICE->getSubExpr(); 421 ICE->setSubExpr(0); 422 ICE->Destroy(Context); 423 TheCall->setArg(i+1, Arg); 424 } 425 426 // GCC does an implicit conversion to the pointer or integer ValType. This 427 // can fail in some cases (1i -> int**), check for this error case now. 428 CastExpr::CastKind Kind = CastExpr::CK_Unknown; 429 CXXMethodDecl *ConversionDecl = 0; 430 if (CheckCastTypes(Arg->getSourceRange(), ValType, Arg, Kind, 431 ConversionDecl)) 432 return true; 433 434 // Okay, we have something that *can* be converted to the right type. Check 435 // to see if there is a potentially weird extension going on here. This can 436 // happen when you do an atomic operation on something like an char* and 437 // pass in 42. The 42 gets converted to char. This is even more strange 438 // for things like 45.123 -> char, etc. 439 // FIXME: Do this check. 440 ImpCastExprToType(Arg, ValType, Kind, /*isLvalue=*/false); 441 TheCall->setArg(i+1, Arg); 442 } 443 444 // Switch the DeclRefExpr to refer to the new decl. 445 DRE->setDecl(NewBuiltinDecl); 446 DRE->setType(NewBuiltinDecl->getType()); 447 448 // Set the callee in the CallExpr. 449 // FIXME: This leaks the original parens and implicit casts. 450 Expr *PromotedCall = DRE; 451 UsualUnaryConversions(PromotedCall); 452 TheCall->setCallee(PromotedCall); 453 454 455 // Change the result type of the call to match the result type of the decl. 456 TheCall->setType(NewBuiltinDecl->getResultType()); 457 return false; 458} 459 460 461/// CheckObjCString - Checks that the argument to the builtin 462/// CFString constructor is correct 463/// FIXME: GCC currently emits the following warning: 464/// "warning: input conversion stopped due to an input byte that does not 465/// belong to the input codeset UTF-8" 466/// Note: It might also make sense to do the UTF-16 conversion here (would 467/// simplify the backend). 468bool Sema::CheckObjCString(Expr *Arg) { 469 Arg = Arg->IgnoreParenCasts(); 470 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 471 472 if (!Literal || Literal->isWide()) { 473 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 474 << Arg->getSourceRange(); 475 return true; 476 } 477 478 const char *Data = Literal->getStrData(); 479 unsigned Length = Literal->getByteLength(); 480 481 for (unsigned i = 0; i < Length; ++i) { 482 if (!Data[i]) { 483 Diag(getLocationOfStringLiteralByte(Literal, i), 484 diag::warn_cfstring_literal_contains_nul_character) 485 << Arg->getSourceRange(); 486 break; 487 } 488 } 489 490 return false; 491} 492 493/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 494/// Emit an error and return true on failure, return false on success. 495bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 496 Expr *Fn = TheCall->getCallee(); 497 if (TheCall->getNumArgs() > 2) { 498 Diag(TheCall->getArg(2)->getLocStart(), 499 diag::err_typecheck_call_too_many_args) 500 << 0 /*function call*/ << Fn->getSourceRange() 501 << SourceRange(TheCall->getArg(2)->getLocStart(), 502 (*(TheCall->arg_end()-1))->getLocEnd()); 503 return true; 504 } 505 506 if (TheCall->getNumArgs() < 2) { 507 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 508 << 0 /*function call*/; 509 } 510 511 // Determine whether the current function is variadic or not. 512 BlockScopeInfo *CurBlock = getCurBlock(); 513 bool isVariadic; 514 if (CurBlock) 515 isVariadic = CurBlock->isVariadic; 516 else if (getCurFunctionDecl()) { 517 if (FunctionProtoType* FTP = 518 dyn_cast<FunctionProtoType>(getCurFunctionDecl()->getType())) 519 isVariadic = FTP->isVariadic(); 520 else 521 isVariadic = false; 522 } else { 523 isVariadic = getCurMethodDecl()->isVariadic(); 524 } 525 526 if (!isVariadic) { 527 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 528 return true; 529 } 530 531 // Verify that the second argument to the builtin is the last argument of the 532 // current function or method. 533 bool SecondArgIsLastNamedArgument = false; 534 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 535 536 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 537 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 538 // FIXME: This isn't correct for methods (results in bogus warning). 539 // Get the last formal in the current function. 540 const ParmVarDecl *LastArg; 541 if (CurBlock) 542 LastArg = *(CurBlock->TheDecl->param_end()-1); 543 else if (FunctionDecl *FD = getCurFunctionDecl()) 544 LastArg = *(FD->param_end()-1); 545 else 546 LastArg = *(getCurMethodDecl()->param_end()-1); 547 SecondArgIsLastNamedArgument = PV == LastArg; 548 } 549 } 550 551 if (!SecondArgIsLastNamedArgument) 552 Diag(TheCall->getArg(1)->getLocStart(), 553 diag::warn_second_parameter_of_va_start_not_last_named_argument); 554 return false; 555} 556 557/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 558/// friends. This is declared to take (...), so we have to check everything. 559bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 560 if (TheCall->getNumArgs() < 2) 561 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 562 << 0 /*function call*/; 563 if (TheCall->getNumArgs() > 2) 564 return Diag(TheCall->getArg(2)->getLocStart(), 565 diag::err_typecheck_call_too_many_args) 566 << 0 /*function call*/ 567 << SourceRange(TheCall->getArg(2)->getLocStart(), 568 (*(TheCall->arg_end()-1))->getLocEnd()); 569 570 Expr *OrigArg0 = TheCall->getArg(0); 571 Expr *OrigArg1 = TheCall->getArg(1); 572 573 // Do standard promotions between the two arguments, returning their common 574 // type. 575 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 576 577 // Make sure any conversions are pushed back into the call; this is 578 // type safe since unordered compare builtins are declared as "_Bool 579 // foo(...)". 580 TheCall->setArg(0, OrigArg0); 581 TheCall->setArg(1, OrigArg1); 582 583 if (OrigArg0->isTypeDependent() || OrigArg1->isTypeDependent()) 584 return false; 585 586 // If the common type isn't a real floating type, then the arguments were 587 // invalid for this operation. 588 if (!Res->isRealFloatingType()) 589 return Diag(OrigArg0->getLocStart(), 590 diag::err_typecheck_call_invalid_ordered_compare) 591 << OrigArg0->getType() << OrigArg1->getType() 592 << SourceRange(OrigArg0->getLocStart(), OrigArg1->getLocEnd()); 593 594 return false; 595} 596 597/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 598/// __builtin_isnan and friends. This is declared to take (...), so we have 599/// to check everything. We expect the last argument to be a floating point 600/// value. 601bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 602 if (TheCall->getNumArgs() < NumArgs) 603 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 604 << 0 /*function call*/; 605 if (TheCall->getNumArgs() > NumArgs) 606 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 607 diag::err_typecheck_call_too_many_args) 608 << 0 /*function call*/ 609 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 610 (*(TheCall->arg_end()-1))->getLocEnd()); 611 612 Expr *OrigArg = TheCall->getArg(NumArgs-1); 613 614 if (OrigArg->isTypeDependent()) 615 return false; 616 617 // This operation requires a floating-point number 618 if (!OrigArg->getType()->isRealFloatingType()) 619 return Diag(OrigArg->getLocStart(), 620 diag::err_typecheck_call_invalid_unary_fp) 621 << OrigArg->getType() << OrigArg->getSourceRange(); 622 623 return false; 624} 625 626bool Sema::SemaBuiltinStackAddress(CallExpr *TheCall) { 627 // The signature for these builtins is exact; the only thing we need 628 // to check is that the argument is a constant. 629 SourceLocation Loc; 630 if (!TheCall->getArg(0)->isTypeDependent() && 631 !TheCall->getArg(0)->isValueDependent() && 632 !TheCall->getArg(0)->isIntegerConstantExpr(Context, &Loc)) 633 return Diag(Loc, diag::err_stack_const_level) << TheCall->getSourceRange(); 634 635 return false; 636} 637 638/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 639// This is declared to take (...), so we have to check everything. 640Action::OwningExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 641 if (TheCall->getNumArgs() < 3) 642 return ExprError(Diag(TheCall->getLocEnd(), 643 diag::err_typecheck_call_too_few_args) 644 << 0 /*function call*/ << TheCall->getSourceRange()); 645 646 unsigned numElements = std::numeric_limits<unsigned>::max(); 647 if (!TheCall->getArg(0)->isTypeDependent() && 648 !TheCall->getArg(1)->isTypeDependent()) { 649 QualType FAType = TheCall->getArg(0)->getType(); 650 QualType SAType = TheCall->getArg(1)->getType(); 651 652 if (!FAType->isVectorType() || !SAType->isVectorType()) { 653 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 654 << SourceRange(TheCall->getArg(0)->getLocStart(), 655 TheCall->getArg(1)->getLocEnd()); 656 return ExprError(); 657 } 658 659 if (!Context.hasSameUnqualifiedType(FAType, SAType)) { 660 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 661 << SourceRange(TheCall->getArg(0)->getLocStart(), 662 TheCall->getArg(1)->getLocEnd()); 663 return ExprError(); 664 } 665 666 numElements = FAType->getAs<VectorType>()->getNumElements(); 667 if (TheCall->getNumArgs() != numElements+2) { 668 if (TheCall->getNumArgs() < numElements+2) 669 return ExprError(Diag(TheCall->getLocEnd(), 670 diag::err_typecheck_call_too_few_args) 671 << 0 /*function call*/ << TheCall->getSourceRange()); 672 return ExprError(Diag(TheCall->getLocEnd(), 673 diag::err_typecheck_call_too_many_args) 674 << 0 /*function call*/ << TheCall->getSourceRange()); 675 } 676 } 677 678 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 679 if (TheCall->getArg(i)->isTypeDependent() || 680 TheCall->getArg(i)->isValueDependent()) 681 continue; 682 683 llvm::APSInt Result(32); 684 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 685 return ExprError(Diag(TheCall->getLocStart(), 686 diag::err_shufflevector_nonconstant_argument) 687 << TheCall->getArg(i)->getSourceRange()); 688 689 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 690 return ExprError(Diag(TheCall->getLocStart(), 691 diag::err_shufflevector_argument_too_large) 692 << TheCall->getArg(i)->getSourceRange()); 693 } 694 695 llvm::SmallVector<Expr*, 32> exprs; 696 697 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 698 exprs.push_back(TheCall->getArg(i)); 699 TheCall->setArg(i, 0); 700 } 701 702 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 703 exprs.size(), exprs[0]->getType(), 704 TheCall->getCallee()->getLocStart(), 705 TheCall->getRParenLoc())); 706} 707 708/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 709// This is declared to take (const void*, ...) and can take two 710// optional constant int args. 711bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 712 unsigned NumArgs = TheCall->getNumArgs(); 713 714 if (NumArgs > 3) 715 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_many_args) 716 << 0 /*function call*/ << TheCall->getSourceRange(); 717 718 // Argument 0 is checked for us and the remaining arguments must be 719 // constant integers. 720 for (unsigned i = 1; i != NumArgs; ++i) { 721 Expr *Arg = TheCall->getArg(i); 722 if (Arg->isTypeDependent()) 723 continue; 724 725 if (!Arg->getType()->isIntegralType()) 726 return Diag(TheCall->getLocStart(), diag::err_prefetch_invalid_arg_type) 727 << Arg->getSourceRange(); 728 729 ImpCastExprToType(Arg, Context.IntTy, CastExpr::CK_IntegralCast); 730 TheCall->setArg(i, Arg); 731 732 if (Arg->isValueDependent()) 733 continue; 734 735 llvm::APSInt Result; 736 if (!Arg->isIntegerConstantExpr(Result, Context)) 737 return Diag(TheCall->getLocStart(), diag::err_prefetch_invalid_arg_ice) 738 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 739 740 // FIXME: gcc issues a warning and rewrites these to 0. These 741 // seems especially odd for the third argument since the default 742 // is 3. 743 if (i == 1) { 744 if (Result.getLimitedValue() > 1) 745 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 746 << "0" << "1" << Arg->getSourceRange(); 747 } else { 748 if (Result.getLimitedValue() > 3) 749 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 750 << "0" << "3" << Arg->getSourceRange(); 751 } 752 } 753 754 return false; 755} 756 757/// SemaBuiltinEHReturnDataRegNo - Handle __builtin_eh_return_data_regno, the 758/// operand must be an integer constant. 759bool Sema::SemaBuiltinEHReturnDataRegNo(CallExpr *TheCall) { 760 llvm::APSInt Result; 761 if (!TheCall->getArg(0)->isIntegerConstantExpr(Result, Context)) 762 return Diag(TheCall->getLocStart(), diag::err_expr_not_ice) 763 << TheCall->getArg(0)->getSourceRange(); 764 765 return false; 766} 767 768 769/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 770/// int type). This simply type checks that type is one of the defined 771/// constants (0-3). 772// For compatability check 0-3, llvm only handles 0 and 2. 773bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 774 Expr *Arg = TheCall->getArg(1); 775 if (Arg->isTypeDependent()) 776 return false; 777 778 QualType ArgType = Arg->getType(); 779 const BuiltinType *BT = ArgType->getAs<BuiltinType>(); 780 llvm::APSInt Result(32); 781 if (!BT || BT->getKind() != BuiltinType::Int) 782 return Diag(TheCall->getLocStart(), diag::err_object_size_invalid_argument) 783 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 784 785 if (Arg->isValueDependent()) 786 return false; 787 788 if (!Arg->isIntegerConstantExpr(Result, Context)) { 789 return Diag(TheCall->getLocStart(), diag::err_object_size_invalid_argument) 790 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 791 } 792 793 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 794 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 795 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 796 } 797 798 return false; 799} 800 801/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 802/// This checks that val is a constant 1. 803bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 804 Expr *Arg = TheCall->getArg(1); 805 if (Arg->isTypeDependent() || Arg->isValueDependent()) 806 return false; 807 808 llvm::APSInt Result(32); 809 if (!Arg->isIntegerConstantExpr(Result, Context) || Result != 1) 810 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 811 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 812 813 return false; 814} 815 816// Handle i > 1 ? "x" : "y", recursivelly 817bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 818 bool HasVAListArg, 819 unsigned format_idx, unsigned firstDataArg) { 820 if (E->isTypeDependent() || E->isValueDependent()) 821 return false; 822 823 switch (E->getStmtClass()) { 824 case Stmt::ConditionalOperatorClass: { 825 const ConditionalOperator *C = cast<ConditionalOperator>(E); 826 return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, 827 HasVAListArg, format_idx, firstDataArg) 828 && SemaCheckStringLiteral(C->getRHS(), TheCall, 829 HasVAListArg, format_idx, firstDataArg); 830 } 831 832 case Stmt::ImplicitCastExprClass: { 833 const ImplicitCastExpr *Expr = cast<ImplicitCastExpr>(E); 834 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 835 format_idx, firstDataArg); 836 } 837 838 case Stmt::ParenExprClass: { 839 const ParenExpr *Expr = cast<ParenExpr>(E); 840 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 841 format_idx, firstDataArg); 842 } 843 844 case Stmt::DeclRefExprClass: { 845 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 846 847 // As an exception, do not flag errors for variables binding to 848 // const string literals. 849 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 850 bool isConstant = false; 851 QualType T = DR->getType(); 852 853 if (const ArrayType *AT = Context.getAsArrayType(T)) { 854 isConstant = AT->getElementType().isConstant(Context); 855 } else if (const PointerType *PT = T->getAs<PointerType>()) { 856 isConstant = T.isConstant(Context) && 857 PT->getPointeeType().isConstant(Context); 858 } 859 860 if (isConstant) { 861 if (const Expr *Init = VD->getAnyInitializer()) 862 return SemaCheckStringLiteral(Init, TheCall, 863 HasVAListArg, format_idx, firstDataArg); 864 } 865 866 // For vprintf* functions (i.e., HasVAListArg==true), we add a 867 // special check to see if the format string is a function parameter 868 // of the function calling the printf function. If the function 869 // has an attribute indicating it is a printf-like function, then we 870 // should suppress warnings concerning non-literals being used in a call 871 // to a vprintf function. For example: 872 // 873 // void 874 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 875 // va_list ap; 876 // va_start(ap, fmt); 877 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 878 // ... 879 // 880 // 881 // FIXME: We don't have full attribute support yet, so just check to see 882 // if the argument is a DeclRefExpr that references a parameter. We'll 883 // add proper support for checking the attribute later. 884 if (HasVAListArg) 885 if (isa<ParmVarDecl>(VD)) 886 return true; 887 } 888 889 return false; 890 } 891 892 case Stmt::CallExprClass: { 893 const CallExpr *CE = cast<CallExpr>(E); 894 if (const ImplicitCastExpr *ICE 895 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 896 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 897 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 898 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 899 unsigned ArgIndex = FA->getFormatIdx(); 900 const Expr *Arg = CE->getArg(ArgIndex - 1); 901 902 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 903 format_idx, firstDataArg); 904 } 905 } 906 } 907 } 908 909 return false; 910 } 911 case Stmt::ObjCStringLiteralClass: 912 case Stmt::StringLiteralClass: { 913 const StringLiteral *StrE = NULL; 914 915 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 916 StrE = ObjCFExpr->getString(); 917 else 918 StrE = cast<StringLiteral>(E); 919 920 if (StrE) { 921 CheckPrintfString(StrE, E, TheCall, HasVAListArg, format_idx, 922 firstDataArg); 923 return true; 924 } 925 926 return false; 927 } 928 929 default: 930 return false; 931 } 932} 933 934void 935Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 936 const CallExpr *TheCall) { 937 for (NonNullAttr::iterator i = NonNull->begin(), e = NonNull->end(); 938 i != e; ++i) { 939 const Expr *ArgExpr = TheCall->getArg(*i); 940 if (ArgExpr->isNullPointerConstant(Context, 941 Expr::NPC_ValueDependentIsNotNull)) 942 Diag(TheCall->getCallee()->getLocStart(), diag::warn_null_arg) 943 << ArgExpr->getSourceRange(); 944 } 945} 946 947/// CheckPrintfArguments - Check calls to printf (and similar functions) for 948/// correct use of format strings. 949/// 950/// HasVAListArg - A predicate indicating whether the printf-like 951/// function is passed an explicit va_arg argument (e.g., vprintf) 952/// 953/// format_idx - The index into Args for the format string. 954/// 955/// Improper format strings to functions in the printf family can be 956/// the source of bizarre bugs and very serious security holes. A 957/// good source of information is available in the following paper 958/// (which includes additional references): 959/// 960/// FormatGuard: Automatic Protection From printf Format String 961/// Vulnerabilities, Proceedings of the 10th USENIX Security Symposium, 2001. 962/// 963/// TODO: 964/// Functionality implemented: 965/// 966/// We can statically check the following properties for string 967/// literal format strings for non v.*printf functions (where the 968/// arguments are passed directly): 969// 970/// (1) Are the number of format conversions equal to the number of 971/// data arguments? 972/// 973/// (2) Does each format conversion correctly match the type of the 974/// corresponding data argument? 975/// 976/// Moreover, for all printf functions we can: 977/// 978/// (3) Check for a missing format string (when not caught by type checking). 979/// 980/// (4) Check for no-operation flags; e.g. using "#" with format 981/// conversion 'c' (TODO) 982/// 983/// (5) Check the use of '%n', a major source of security holes. 984/// 985/// (6) Check for malformed format conversions that don't specify anything. 986/// 987/// (7) Check for empty format strings. e.g: printf(""); 988/// 989/// (8) Check that the format string is a wide literal. 990/// 991/// All of these checks can be done by parsing the format string. 992/// 993void 994Sema::CheckPrintfArguments(const CallExpr *TheCall, bool HasVAListArg, 995 unsigned format_idx, unsigned firstDataArg) { 996 const Expr *Fn = TheCall->getCallee(); 997 998 // The way the format attribute works in GCC, the implicit this argument 999 // of member functions is counted. However, it doesn't appear in our own 1000 // lists, so decrement format_idx in that case. 1001 if (isa<CXXMemberCallExpr>(TheCall)) { 1002 // Catch a format attribute mistakenly referring to the object argument. 1003 if (format_idx == 0) 1004 return; 1005 --format_idx; 1006 if(firstDataArg != 0) 1007 --firstDataArg; 1008 } 1009 1010 // CHECK: printf-like function is called with no format string. 1011 if (format_idx >= TheCall->getNumArgs()) { 1012 Diag(TheCall->getRParenLoc(), diag::warn_printf_missing_format_string) 1013 << Fn->getSourceRange(); 1014 return; 1015 } 1016 1017 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 1018 1019 // CHECK: format string is not a string literal. 1020 // 1021 // Dynamically generated format strings are difficult to 1022 // automatically vet at compile time. Requiring that format strings 1023 // are string literals: (1) permits the checking of format strings by 1024 // the compiler and thereby (2) can practically remove the source of 1025 // many format string exploits. 1026 1027 // Format string can be either ObjC string (e.g. @"%d") or 1028 // C string (e.g. "%d") 1029 // ObjC string uses the same format specifiers as C string, so we can use 1030 // the same format string checking logic for both ObjC and C strings. 1031 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 1032 firstDataArg)) 1033 return; // Literal format string found, check done! 1034 1035 // If there are no arguments specified, warn with -Wformat-security, otherwise 1036 // warn only with -Wformat-nonliteral. 1037 if (TheCall->getNumArgs() == format_idx+1) 1038 Diag(TheCall->getArg(format_idx)->getLocStart(), 1039 diag::warn_printf_nonliteral_noargs) 1040 << OrigFormatExpr->getSourceRange(); 1041 else 1042 Diag(TheCall->getArg(format_idx)->getLocStart(), 1043 diag::warn_printf_nonliteral) 1044 << OrigFormatExpr->getSourceRange(); 1045} 1046 1047namespace { 1048class CheckPrintfHandler : public analyze_printf::FormatStringHandler { 1049 Sema &S; 1050 const StringLiteral *FExpr; 1051 const Expr *OrigFormatExpr; 1052 const unsigned NumDataArgs; 1053 const bool IsObjCLiteral; 1054 const char *Beg; // Start of format string. 1055 const bool HasVAListArg; 1056 const CallExpr *TheCall; 1057 unsigned FormatIdx; 1058 llvm::BitVector CoveredArgs; 1059 bool usesPositionalArgs; 1060 bool atFirstArg; 1061public: 1062 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1063 const Expr *origFormatExpr, 1064 unsigned numDataArgs, bool isObjCLiteral, 1065 const char *beg, bool hasVAListArg, 1066 const CallExpr *theCall, unsigned formatIdx) 1067 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1068 NumDataArgs(numDataArgs), 1069 IsObjCLiteral(isObjCLiteral), Beg(beg), 1070 HasVAListArg(hasVAListArg), 1071 TheCall(theCall), FormatIdx(formatIdx), 1072 usesPositionalArgs(false), atFirstArg(true) { 1073 CoveredArgs.resize(numDataArgs); 1074 CoveredArgs.reset(); 1075 } 1076 1077 void DoneProcessing(); 1078 1079 void HandleIncompleteFormatSpecifier(const char *startSpecifier, 1080 unsigned specifierLen); 1081 1082 bool 1083 HandleInvalidConversionSpecifier(const analyze_printf::FormatSpecifier &FS, 1084 const char *startSpecifier, 1085 unsigned specifierLen); 1086 1087 virtual void HandleInvalidPosition(const char *startSpecifier, 1088 unsigned specifierLen, 1089 analyze_printf::PositionContext p); 1090 1091 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1092 1093 void HandleNullChar(const char *nullCharacter); 1094 1095 bool HandleFormatSpecifier(const analyze_printf::FormatSpecifier &FS, 1096 const char *startSpecifier, 1097 unsigned specifierLen); 1098private: 1099 SourceRange getFormatStringRange(); 1100 SourceRange getFormatSpecifierRange(const char *startSpecifier, 1101 unsigned specifierLen); 1102 SourceLocation getLocationOfByte(const char *x); 1103 1104 bool HandleAmount(const analyze_printf::OptionalAmount &Amt, unsigned k, 1105 const char *startSpecifier, unsigned specifierLen); 1106 void HandleFlags(const analyze_printf::FormatSpecifier &FS, 1107 llvm::StringRef flag, llvm::StringRef cspec, 1108 const char *startSpecifier, unsigned specifierLen); 1109 1110 const Expr *getDataArg(unsigned i) const; 1111}; 1112} 1113 1114SourceRange CheckPrintfHandler::getFormatStringRange() { 1115 return OrigFormatExpr->getSourceRange(); 1116} 1117 1118SourceRange CheckPrintfHandler:: 1119getFormatSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1120 return SourceRange(getLocationOfByte(startSpecifier), 1121 getLocationOfByte(startSpecifier+specifierLen-1)); 1122} 1123 1124SourceLocation CheckPrintfHandler::getLocationOfByte(const char *x) { 1125 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1126} 1127 1128void CheckPrintfHandler:: 1129HandleIncompleteFormatSpecifier(const char *startSpecifier, 1130 unsigned specifierLen) { 1131 SourceLocation Loc = getLocationOfByte(startSpecifier); 1132 S.Diag(Loc, diag::warn_printf_incomplete_specifier) 1133 << getFormatSpecifierRange(startSpecifier, specifierLen); 1134} 1135 1136void 1137CheckPrintfHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1138 analyze_printf::PositionContext p) { 1139 SourceLocation Loc = getLocationOfByte(startPos); 1140 S.Diag(Loc, diag::warn_printf_invalid_positional_specifier) 1141 << (unsigned) p << getFormatSpecifierRange(startPos, posLen); 1142} 1143 1144void CheckPrintfHandler::HandleZeroPosition(const char *startPos, 1145 unsigned posLen) { 1146 SourceLocation Loc = getLocationOfByte(startPos); 1147 S.Diag(Loc, diag::warn_printf_zero_positional_specifier) 1148 << getFormatSpecifierRange(startPos, posLen); 1149} 1150 1151bool CheckPrintfHandler:: 1152HandleInvalidConversionSpecifier(const analyze_printf::FormatSpecifier &FS, 1153 const char *startSpecifier, 1154 unsigned specifierLen) { 1155 1156 unsigned argIndex = FS.getArgIndex(); 1157 bool keepGoing = true; 1158 if (argIndex < NumDataArgs) { 1159 // Consider the argument coverered, even though the specifier doesn't 1160 // make sense. 1161 CoveredArgs.set(argIndex); 1162 } 1163 else { 1164 // If argIndex exceeds the number of data arguments we 1165 // don't issue a warning because that is just a cascade of warnings (and 1166 // they may have intended '%%' anyway). We don't want to continue processing 1167 // the format string after this point, however, as we will like just get 1168 // gibberish when trying to match arguments. 1169 keepGoing = false; 1170 } 1171 1172 const analyze_printf::ConversionSpecifier &CS = 1173 FS.getConversionSpecifier(); 1174 SourceLocation Loc = getLocationOfByte(CS.getStart()); 1175 S.Diag(Loc, diag::warn_printf_invalid_conversion) 1176 << llvm::StringRef(CS.getStart(), CS.getLength()) 1177 << getFormatSpecifierRange(startSpecifier, specifierLen); 1178 1179 return keepGoing; 1180} 1181 1182void CheckPrintfHandler::HandleNullChar(const char *nullCharacter) { 1183 // The presence of a null character is likely an error. 1184 S.Diag(getLocationOfByte(nullCharacter), 1185 diag::warn_printf_format_string_contains_null_char) 1186 << getFormatStringRange(); 1187} 1188 1189const Expr *CheckPrintfHandler::getDataArg(unsigned i) const { 1190 return TheCall->getArg(FormatIdx + i + 1); 1191} 1192 1193 1194 1195void CheckPrintfHandler::HandleFlags(const analyze_printf::FormatSpecifier &FS, 1196 llvm::StringRef flag, 1197 llvm::StringRef cspec, 1198 const char *startSpecifier, 1199 unsigned specifierLen) { 1200 const analyze_printf::ConversionSpecifier &CS = FS.getConversionSpecifier(); 1201 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_nonsensical_flag) 1202 << flag << cspec << getFormatSpecifierRange(startSpecifier, specifierLen); 1203} 1204 1205bool 1206CheckPrintfHandler::HandleAmount(const analyze_printf::OptionalAmount &Amt, 1207 unsigned k, const char *startSpecifier, 1208 unsigned specifierLen) { 1209 1210 if (Amt.hasDataArgument()) { 1211 if (!HasVAListArg) { 1212 unsigned argIndex = Amt.getArgIndex(); 1213 if (argIndex >= NumDataArgs) { 1214 S.Diag(getLocationOfByte(Amt.getStart()), 1215 diag::warn_printf_asterisk_missing_arg) 1216 << k << getFormatSpecifierRange(startSpecifier, specifierLen); 1217 // Don't do any more checking. We will just emit 1218 // spurious errors. 1219 return false; 1220 } 1221 1222 // Type check the data argument. It should be an 'int'. 1223 // Although not in conformance with C99, we also allow the argument to be 1224 // an 'unsigned int' as that is a reasonably safe case. GCC also 1225 // doesn't emit a warning for that case. 1226 CoveredArgs.set(argIndex); 1227 const Expr *Arg = getDataArg(argIndex); 1228 QualType T = Arg->getType(); 1229 1230 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1231 assert(ATR.isValid()); 1232 1233 if (!ATR.matchesType(S.Context, T)) { 1234 S.Diag(getLocationOfByte(Amt.getStart()), 1235 diag::warn_printf_asterisk_wrong_type) 1236 << k 1237 << ATR.getRepresentativeType(S.Context) << T 1238 << getFormatSpecifierRange(startSpecifier, specifierLen) 1239 << Arg->getSourceRange(); 1240 // Don't do any more checking. We will just emit 1241 // spurious errors. 1242 return false; 1243 } 1244 } 1245 } 1246 return true; 1247} 1248 1249bool 1250CheckPrintfHandler::HandleFormatSpecifier(const analyze_printf::FormatSpecifier 1251 &FS, 1252 const char *startSpecifier, 1253 unsigned specifierLen) { 1254 1255 using namespace analyze_printf; 1256 const ConversionSpecifier &CS = FS.getConversionSpecifier(); 1257 1258 if (atFirstArg) { 1259 atFirstArg = false; 1260 usesPositionalArgs = FS.usesPositionalArg(); 1261 } 1262 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1263 // Cannot mix-and-match positional and non-positional arguments. 1264 S.Diag(getLocationOfByte(CS.getStart()), 1265 diag::warn_printf_mix_positional_nonpositional_args) 1266 << getFormatSpecifierRange(startSpecifier, specifierLen); 1267 return false; 1268 } 1269 1270 // First check if the field width, precision, and conversion specifier 1271 // have matching data arguments. 1272 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 1273 startSpecifier, specifierLen)) { 1274 return false; 1275 } 1276 1277 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 1278 startSpecifier, specifierLen)) { 1279 return false; 1280 } 1281 1282 if (!CS.consumesDataArgument()) { 1283 // FIXME: Technically specifying a precision or field width here 1284 // makes no sense. Worth issuing a warning at some point. 1285 return true; 1286 } 1287 1288 // Consume the argument. 1289 unsigned argIndex = FS.getArgIndex(); 1290 if (argIndex < NumDataArgs) { 1291 // The check to see if the argIndex is valid will come later. 1292 // We set the bit here because we may exit early from this 1293 // function if we encounter some other error. 1294 CoveredArgs.set(argIndex); 1295 } 1296 1297 // Check for using an Objective-C specific conversion specifier 1298 // in a non-ObjC literal. 1299 if (!IsObjCLiteral && CS.isObjCArg()) { 1300 return HandleInvalidConversionSpecifier(FS, startSpecifier, specifierLen); 1301 } 1302 1303 // Are we using '%n'? Issue a warning about this being 1304 // a possible security issue. 1305 if (CS.getKind() == ConversionSpecifier::OutIntPtrArg) { 1306 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back) 1307 << getFormatSpecifierRange(startSpecifier, specifierLen); 1308 // Continue checking the other format specifiers. 1309 return true; 1310 } 1311 1312 if (CS.getKind() == ConversionSpecifier::VoidPtrArg) { 1313 if (FS.getPrecision().getHowSpecified() != OptionalAmount::NotSpecified) 1314 S.Diag(getLocationOfByte(CS.getStart()), 1315 diag::warn_printf_nonsensical_precision) 1316 << CS.getCharacters() 1317 << getFormatSpecifierRange(startSpecifier, specifierLen); 1318 } 1319 if (CS.getKind() == ConversionSpecifier::VoidPtrArg || 1320 CS.getKind() == ConversionSpecifier::CStrArg) { 1321 // FIXME: Instead of using "0", "+", etc., eventually get them from 1322 // the FormatSpecifier. 1323 if (FS.hasLeadingZeros()) 1324 HandleFlags(FS, "0", CS.getCharacters(), startSpecifier, specifierLen); 1325 if (FS.hasPlusPrefix()) 1326 HandleFlags(FS, "+", CS.getCharacters(), startSpecifier, specifierLen); 1327 if (FS.hasSpacePrefix()) 1328 HandleFlags(FS, " ", CS.getCharacters(), startSpecifier, specifierLen); 1329 } 1330 1331 // The remaining checks depend on the data arguments. 1332 if (HasVAListArg) 1333 return true; 1334 1335 if (argIndex >= NumDataArgs) { 1336 S.Diag(getLocationOfByte(CS.getStart()), 1337 diag::warn_printf_insufficient_data_args) 1338 << getFormatSpecifierRange(startSpecifier, specifierLen); 1339 // Don't do any more checking. 1340 return false; 1341 } 1342 1343 // Now type check the data expression that matches the 1344 // format specifier. 1345 const Expr *Ex = getDataArg(argIndex); 1346 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context); 1347 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 1348 // Check if we didn't match because of an implicit cast from a 'char' 1349 // or 'short' to an 'int'. This is done because printf is a varargs 1350 // function. 1351 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 1352 if (ICE->getType() == S.Context.IntTy) 1353 if (ATR.matchesType(S.Context, ICE->getSubExpr()->getType())) 1354 return true; 1355 1356 S.Diag(getLocationOfByte(CS.getStart()), 1357 diag::warn_printf_conversion_argument_type_mismatch) 1358 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1359 << getFormatSpecifierRange(startSpecifier, specifierLen) 1360 << Ex->getSourceRange(); 1361 } 1362 1363 return true; 1364} 1365 1366void CheckPrintfHandler::DoneProcessing() { 1367 // Does the number of data arguments exceed the number of 1368 // format conversions in the format string? 1369 if (!HasVAListArg) { 1370 // Find any arguments that weren't covered. 1371 CoveredArgs.flip(); 1372 signed notCoveredArg = CoveredArgs.find_first(); 1373 if (notCoveredArg >= 0) { 1374 assert((unsigned)notCoveredArg < NumDataArgs); 1375 S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(), 1376 diag::warn_printf_data_arg_not_used) 1377 << getFormatStringRange(); 1378 } 1379 } 1380} 1381 1382void Sema::CheckPrintfString(const StringLiteral *FExpr, 1383 const Expr *OrigFormatExpr, 1384 const CallExpr *TheCall, bool HasVAListArg, 1385 unsigned format_idx, unsigned firstDataArg) { 1386 1387 // CHECK: is the format string a wide literal? 1388 if (FExpr->isWide()) { 1389 Diag(FExpr->getLocStart(), 1390 diag::warn_printf_format_string_is_wide_literal) 1391 << OrigFormatExpr->getSourceRange(); 1392 return; 1393 } 1394 1395 // Str - The format string. NOTE: this is NOT null-terminated! 1396 const char *Str = FExpr->getStrData(); 1397 1398 // CHECK: empty format string? 1399 unsigned StrLen = FExpr->getByteLength(); 1400 1401 if (StrLen == 0) { 1402 Diag(FExpr->getLocStart(), diag::warn_printf_empty_format_string) 1403 << OrigFormatExpr->getSourceRange(); 1404 return; 1405 } 1406 1407 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, 1408 TheCall->getNumArgs() - firstDataArg, 1409 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1410 HasVAListArg, TheCall, format_idx); 1411 1412 if (!analyze_printf::ParseFormatString(H, Str, Str + StrLen)) 1413 H.DoneProcessing(); 1414} 1415 1416//===--- CHECK: Return Address of Stack Variable --------------------------===// 1417 1418static DeclRefExpr* EvalVal(Expr *E); 1419static DeclRefExpr* EvalAddr(Expr* E); 1420 1421/// CheckReturnStackAddr - Check if a return statement returns the address 1422/// of a stack variable. 1423void 1424Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 1425 SourceLocation ReturnLoc) { 1426 1427 // Perform checking for returned stack addresses. 1428 if (lhsType->isPointerType() || lhsType->isBlockPointerType()) { 1429 if (DeclRefExpr *DR = EvalAddr(RetValExp)) 1430 Diag(DR->getLocStart(), diag::warn_ret_stack_addr) 1431 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1432 1433 // Skip over implicit cast expressions when checking for block expressions. 1434 RetValExp = RetValExp->IgnoreParenCasts(); 1435 1436 if (BlockExpr *C = dyn_cast<BlockExpr>(RetValExp)) 1437 if (C->hasBlockDeclRefExprs()) 1438 Diag(C->getLocStart(), diag::err_ret_local_block) 1439 << C->getSourceRange(); 1440 1441 if (AddrLabelExpr *ALE = dyn_cast<AddrLabelExpr>(RetValExp)) 1442 Diag(ALE->getLocStart(), diag::warn_ret_addr_label) 1443 << ALE->getSourceRange(); 1444 1445 } else if (lhsType->isReferenceType()) { 1446 // Perform checking for stack values returned by reference. 1447 // Check for a reference to the stack 1448 if (DeclRefExpr *DR = EvalVal(RetValExp)) 1449 Diag(DR->getLocStart(), diag::warn_ret_stack_ref) 1450 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1451 } 1452} 1453 1454/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 1455/// check if the expression in a return statement evaluates to an address 1456/// to a location on the stack. The recursion is used to traverse the 1457/// AST of the return expression, with recursion backtracking when we 1458/// encounter a subexpression that (1) clearly does not lead to the address 1459/// of a stack variable or (2) is something we cannot determine leads to 1460/// the address of a stack variable based on such local checking. 1461/// 1462/// EvalAddr processes expressions that are pointers that are used as 1463/// references (and not L-values). EvalVal handles all other values. 1464/// At the base case of the recursion is a check for a DeclRefExpr* in 1465/// the refers to a stack variable. 1466/// 1467/// This implementation handles: 1468/// 1469/// * pointer-to-pointer casts 1470/// * implicit conversions from array references to pointers 1471/// * taking the address of fields 1472/// * arbitrary interplay between "&" and "*" operators 1473/// * pointer arithmetic from an address of a stack variable 1474/// * taking the address of an array element where the array is on the stack 1475static DeclRefExpr* EvalAddr(Expr *E) { 1476 // We should only be called for evaluating pointer expressions. 1477 assert((E->getType()->isAnyPointerType() || 1478 E->getType()->isBlockPointerType() || 1479 E->getType()->isObjCQualifiedIdType()) && 1480 "EvalAddr only works on pointers"); 1481 1482 // Our "symbolic interpreter" is just a dispatch off the currently 1483 // viewed AST node. We then recursively traverse the AST by calling 1484 // EvalAddr and EvalVal appropriately. 1485 switch (E->getStmtClass()) { 1486 case Stmt::ParenExprClass: 1487 // Ignore parentheses. 1488 return EvalAddr(cast<ParenExpr>(E)->getSubExpr()); 1489 1490 case Stmt::UnaryOperatorClass: { 1491 // The only unary operator that make sense to handle here 1492 // is AddrOf. All others don't make sense as pointers. 1493 UnaryOperator *U = cast<UnaryOperator>(E); 1494 1495 if (U->getOpcode() == UnaryOperator::AddrOf) 1496 return EvalVal(U->getSubExpr()); 1497 else 1498 return NULL; 1499 } 1500 1501 case Stmt::BinaryOperatorClass: { 1502 // Handle pointer arithmetic. All other binary operators are not valid 1503 // in this context. 1504 BinaryOperator *B = cast<BinaryOperator>(E); 1505 BinaryOperator::Opcode op = B->getOpcode(); 1506 1507 if (op != BinaryOperator::Add && op != BinaryOperator::Sub) 1508 return NULL; 1509 1510 Expr *Base = B->getLHS(); 1511 1512 // Determine which argument is the real pointer base. It could be 1513 // the RHS argument instead of the LHS. 1514 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 1515 1516 assert (Base->getType()->isPointerType()); 1517 return EvalAddr(Base); 1518 } 1519 1520 // For conditional operators we need to see if either the LHS or RHS are 1521 // valid DeclRefExpr*s. If one of them is valid, we return it. 1522 case Stmt::ConditionalOperatorClass: { 1523 ConditionalOperator *C = cast<ConditionalOperator>(E); 1524 1525 // Handle the GNU extension for missing LHS. 1526 if (Expr *lhsExpr = C->getLHS()) 1527 if (DeclRefExpr* LHS = EvalAddr(lhsExpr)) 1528 return LHS; 1529 1530 return EvalAddr(C->getRHS()); 1531 } 1532 1533 // For casts, we need to handle conversions from arrays to 1534 // pointer values, and pointer-to-pointer conversions. 1535 case Stmt::ImplicitCastExprClass: 1536 case Stmt::CStyleCastExprClass: 1537 case Stmt::CXXFunctionalCastExprClass: { 1538 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 1539 QualType T = SubExpr->getType(); 1540 1541 if (SubExpr->getType()->isPointerType() || 1542 SubExpr->getType()->isBlockPointerType() || 1543 SubExpr->getType()->isObjCQualifiedIdType()) 1544 return EvalAddr(SubExpr); 1545 else if (T->isArrayType()) 1546 return EvalVal(SubExpr); 1547 else 1548 return 0; 1549 } 1550 1551 // C++ casts. For dynamic casts, static casts, and const casts, we 1552 // are always converting from a pointer-to-pointer, so we just blow 1553 // through the cast. In the case the dynamic cast doesn't fail (and 1554 // return NULL), we take the conservative route and report cases 1555 // where we return the address of a stack variable. For Reinterpre 1556 // FIXME: The comment about is wrong; we're not always converting 1557 // from pointer to pointer. I'm guessing that this code should also 1558 // handle references to objects. 1559 case Stmt::CXXStaticCastExprClass: 1560 case Stmt::CXXDynamicCastExprClass: 1561 case Stmt::CXXConstCastExprClass: 1562 case Stmt::CXXReinterpretCastExprClass: { 1563 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 1564 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 1565 return EvalAddr(S); 1566 else 1567 return NULL; 1568 } 1569 1570 // Everything else: we simply don't reason about them. 1571 default: 1572 return NULL; 1573 } 1574} 1575 1576 1577/// EvalVal - This function is complements EvalAddr in the mutual recursion. 1578/// See the comments for EvalAddr for more details. 1579static DeclRefExpr* EvalVal(Expr *E) { 1580 1581 // We should only be called for evaluating non-pointer expressions, or 1582 // expressions with a pointer type that are not used as references but instead 1583 // are l-values (e.g., DeclRefExpr with a pointer type). 1584 1585 // Our "symbolic interpreter" is just a dispatch off the currently 1586 // viewed AST node. We then recursively traverse the AST by calling 1587 // EvalAddr and EvalVal appropriately. 1588 switch (E->getStmtClass()) { 1589 case Stmt::DeclRefExprClass: { 1590 // DeclRefExpr: the base case. When we hit a DeclRefExpr we are looking 1591 // at code that refers to a variable's name. We check if it has local 1592 // storage within the function, and if so, return the expression. 1593 DeclRefExpr *DR = cast<DeclRefExpr>(E); 1594 1595 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 1596 if (V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR; 1597 1598 return NULL; 1599 } 1600 1601 case Stmt::ParenExprClass: 1602 // Ignore parentheses. 1603 return EvalVal(cast<ParenExpr>(E)->getSubExpr()); 1604 1605 case Stmt::UnaryOperatorClass: { 1606 // The only unary operator that make sense to handle here 1607 // is Deref. All others don't resolve to a "name." This includes 1608 // handling all sorts of rvalues passed to a unary operator. 1609 UnaryOperator *U = cast<UnaryOperator>(E); 1610 1611 if (U->getOpcode() == UnaryOperator::Deref) 1612 return EvalAddr(U->getSubExpr()); 1613 1614 return NULL; 1615 } 1616 1617 case Stmt::ArraySubscriptExprClass: { 1618 // Array subscripts are potential references to data on the stack. We 1619 // retrieve the DeclRefExpr* for the array variable if it indeed 1620 // has local storage. 1621 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase()); 1622 } 1623 1624 case Stmt::ConditionalOperatorClass: { 1625 // For conditional operators we need to see if either the LHS or RHS are 1626 // non-NULL DeclRefExpr's. If one is non-NULL, we return it. 1627 ConditionalOperator *C = cast<ConditionalOperator>(E); 1628 1629 // Handle the GNU extension for missing LHS. 1630 if (Expr *lhsExpr = C->getLHS()) 1631 if (DeclRefExpr *LHS = EvalVal(lhsExpr)) 1632 return LHS; 1633 1634 return EvalVal(C->getRHS()); 1635 } 1636 1637 // Accesses to members are potential references to data on the stack. 1638 case Stmt::MemberExprClass: { 1639 MemberExpr *M = cast<MemberExpr>(E); 1640 1641 // Check for indirect access. We only want direct field accesses. 1642 if (!M->isArrow()) 1643 return EvalVal(M->getBase()); 1644 else 1645 return NULL; 1646 } 1647 1648 // Everything else: we simply don't reason about them. 1649 default: 1650 return NULL; 1651 } 1652} 1653 1654//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 1655 1656/// Check for comparisons of floating point operands using != and ==. 1657/// Issue a warning if these are no self-comparisons, as they are not likely 1658/// to do what the programmer intended. 1659void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 1660 bool EmitWarning = true; 1661 1662 Expr* LeftExprSansParen = lex->IgnoreParens(); 1663 Expr* RightExprSansParen = rex->IgnoreParens(); 1664 1665 // Special case: check for x == x (which is OK). 1666 // Do not emit warnings for such cases. 1667 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 1668 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 1669 if (DRL->getDecl() == DRR->getDecl()) 1670 EmitWarning = false; 1671 1672 1673 // Special case: check for comparisons against literals that can be exactly 1674 // represented by APFloat. In such cases, do not emit a warning. This 1675 // is a heuristic: often comparison against such literals are used to 1676 // detect if a value in a variable has not changed. This clearly can 1677 // lead to false negatives. 1678 if (EmitWarning) { 1679 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 1680 if (FLL->isExact()) 1681 EmitWarning = false; 1682 } else 1683 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 1684 if (FLR->isExact()) 1685 EmitWarning = false; 1686 } 1687 } 1688 1689 // Check for comparisons with builtin types. 1690 if (EmitWarning) 1691 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 1692 if (CL->isBuiltinCall(Context)) 1693 EmitWarning = false; 1694 1695 if (EmitWarning) 1696 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 1697 if (CR->isBuiltinCall(Context)) 1698 EmitWarning = false; 1699 1700 // Emit the diagnostic. 1701 if (EmitWarning) 1702 Diag(loc, diag::warn_floatingpoint_eq) 1703 << lex->getSourceRange() << rex->getSourceRange(); 1704} 1705 1706//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 1707//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 1708 1709namespace { 1710 1711/// Structure recording the 'active' range of an integer-valued 1712/// expression. 1713struct IntRange { 1714 /// The number of bits active in the int. 1715 unsigned Width; 1716 1717 /// True if the int is known not to have negative values. 1718 bool NonNegative; 1719 1720 IntRange() {} 1721 IntRange(unsigned Width, bool NonNegative) 1722 : Width(Width), NonNegative(NonNegative) 1723 {} 1724 1725 // Returns the range of the bool type. 1726 static IntRange forBoolType() { 1727 return IntRange(1, true); 1728 } 1729 1730 // Returns the range of an integral type. 1731 static IntRange forType(ASTContext &C, QualType T) { 1732 return forCanonicalType(C, T->getCanonicalTypeInternal().getTypePtr()); 1733 } 1734 1735 // Returns the range of an integeral type based on its canonical 1736 // representation. 1737 static IntRange forCanonicalType(ASTContext &C, const Type *T) { 1738 assert(T->isCanonicalUnqualified()); 1739 1740 if (const VectorType *VT = dyn_cast<VectorType>(T)) 1741 T = VT->getElementType().getTypePtr(); 1742 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 1743 T = CT->getElementType().getTypePtr(); 1744 if (const EnumType *ET = dyn_cast<EnumType>(T)) 1745 T = ET->getDecl()->getIntegerType().getTypePtr(); 1746 1747 const BuiltinType *BT = cast<BuiltinType>(T); 1748 assert(BT->isInteger()); 1749 1750 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 1751 } 1752 1753 // Returns the supremum of two ranges: i.e. their conservative merge. 1754 static IntRange join(IntRange L, IntRange R) { 1755 return IntRange(std::max(L.Width, R.Width), 1756 L.NonNegative && R.NonNegative); 1757 } 1758 1759 // Returns the infinum of two ranges: i.e. their aggressive merge. 1760 static IntRange meet(IntRange L, IntRange R) { 1761 return IntRange(std::min(L.Width, R.Width), 1762 L.NonNegative || R.NonNegative); 1763 } 1764}; 1765 1766IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 1767 if (value.isSigned() && value.isNegative()) 1768 return IntRange(value.getMinSignedBits(), false); 1769 1770 if (value.getBitWidth() > MaxWidth) 1771 value.trunc(MaxWidth); 1772 1773 // isNonNegative() just checks the sign bit without considering 1774 // signedness. 1775 return IntRange(value.getActiveBits(), true); 1776} 1777 1778IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 1779 unsigned MaxWidth) { 1780 if (result.isInt()) 1781 return GetValueRange(C, result.getInt(), MaxWidth); 1782 1783 if (result.isVector()) { 1784 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 1785 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 1786 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 1787 R = IntRange::join(R, El); 1788 } 1789 return R; 1790 } 1791 1792 if (result.isComplexInt()) { 1793 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 1794 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 1795 return IntRange::join(R, I); 1796 } 1797 1798 // This can happen with lossless casts to intptr_t of "based" lvalues. 1799 // Assume it might use arbitrary bits. 1800 // FIXME: The only reason we need to pass the type in here is to get 1801 // the sign right on this one case. It would be nice if APValue 1802 // preserved this. 1803 assert(result.isLValue()); 1804 return IntRange(MaxWidth, Ty->isUnsignedIntegerType()); 1805} 1806 1807/// Pseudo-evaluate the given integer expression, estimating the 1808/// range of values it might take. 1809/// 1810/// \param MaxWidth - the width to which the value will be truncated 1811IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 1812 E = E->IgnoreParens(); 1813 1814 // Try a full evaluation first. 1815 Expr::EvalResult result; 1816 if (E->Evaluate(result, C)) 1817 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 1818 1819 // I think we only want to look through implicit casts here; if the 1820 // user has an explicit widening cast, we should treat the value as 1821 // being of the new, wider type. 1822 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 1823 if (CE->getCastKind() == CastExpr::CK_NoOp) 1824 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 1825 1826 IntRange OutputTypeRange = IntRange::forType(C, CE->getType()); 1827 1828 bool isIntegerCast = (CE->getCastKind() == CastExpr::CK_IntegralCast); 1829 if (!isIntegerCast && CE->getCastKind() == CastExpr::CK_Unknown) 1830 isIntegerCast = CE->getSubExpr()->getType()->isIntegerType(); 1831 1832 // Assume that non-integer casts can span the full range of the type. 1833 if (!isIntegerCast) 1834 return OutputTypeRange; 1835 1836 IntRange SubRange 1837 = GetExprRange(C, CE->getSubExpr(), 1838 std::min(MaxWidth, OutputTypeRange.Width)); 1839 1840 // Bail out if the subexpr's range is as wide as the cast type. 1841 if (SubRange.Width >= OutputTypeRange.Width) 1842 return OutputTypeRange; 1843 1844 // Otherwise, we take the smaller width, and we're non-negative if 1845 // either the output type or the subexpr is. 1846 return IntRange(SubRange.Width, 1847 SubRange.NonNegative || OutputTypeRange.NonNegative); 1848 } 1849 1850 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 1851 // If we can fold the condition, just take that operand. 1852 bool CondResult; 1853 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 1854 return GetExprRange(C, CondResult ? CO->getTrueExpr() 1855 : CO->getFalseExpr(), 1856 MaxWidth); 1857 1858 // Otherwise, conservatively merge. 1859 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 1860 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 1861 return IntRange::join(L, R); 1862 } 1863 1864 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 1865 switch (BO->getOpcode()) { 1866 1867 // Boolean-valued operations are single-bit and positive. 1868 case BinaryOperator::LAnd: 1869 case BinaryOperator::LOr: 1870 case BinaryOperator::LT: 1871 case BinaryOperator::GT: 1872 case BinaryOperator::LE: 1873 case BinaryOperator::GE: 1874 case BinaryOperator::EQ: 1875 case BinaryOperator::NE: 1876 return IntRange::forBoolType(); 1877 1878 // The type of these compound assignments is the type of the LHS, 1879 // so the RHS is not necessarily an integer. 1880 case BinaryOperator::MulAssign: 1881 case BinaryOperator::DivAssign: 1882 case BinaryOperator::RemAssign: 1883 case BinaryOperator::AddAssign: 1884 case BinaryOperator::SubAssign: 1885 return IntRange::forType(C, E->getType()); 1886 1887 // Operations with opaque sources are black-listed. 1888 case BinaryOperator::PtrMemD: 1889 case BinaryOperator::PtrMemI: 1890 return IntRange::forType(C, E->getType()); 1891 1892 // Bitwise-and uses the *infinum* of the two source ranges. 1893 case BinaryOperator::And: 1894 case BinaryOperator::AndAssign: 1895 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 1896 GetExprRange(C, BO->getRHS(), MaxWidth)); 1897 1898 // Left shift gets black-listed based on a judgement call. 1899 case BinaryOperator::Shl: 1900 case BinaryOperator::ShlAssign: 1901 return IntRange::forType(C, E->getType()); 1902 1903 // Right shift by a constant can narrow its left argument. 1904 case BinaryOperator::Shr: 1905 case BinaryOperator::ShrAssign: { 1906 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 1907 1908 // If the shift amount is a positive constant, drop the width by 1909 // that much. 1910 llvm::APSInt shift; 1911 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 1912 shift.isNonNegative()) { 1913 unsigned zext = shift.getZExtValue(); 1914 if (zext >= L.Width) 1915 L.Width = (L.NonNegative ? 0 : 1); 1916 else 1917 L.Width -= zext; 1918 } 1919 1920 return L; 1921 } 1922 1923 // Comma acts as its right operand. 1924 case BinaryOperator::Comma: 1925 return GetExprRange(C, BO->getRHS(), MaxWidth); 1926 1927 // Black-list pointer subtractions. 1928 case BinaryOperator::Sub: 1929 if (BO->getLHS()->getType()->isPointerType()) 1930 return IntRange::forType(C, E->getType()); 1931 // fallthrough 1932 1933 default: 1934 break; 1935 } 1936 1937 // Treat every other operator as if it were closed on the 1938 // narrowest type that encompasses both operands. 1939 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 1940 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 1941 return IntRange::join(L, R); 1942 } 1943 1944 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 1945 switch (UO->getOpcode()) { 1946 // Boolean-valued operations are white-listed. 1947 case UnaryOperator::LNot: 1948 return IntRange::forBoolType(); 1949 1950 // Operations with opaque sources are black-listed. 1951 case UnaryOperator::Deref: 1952 case UnaryOperator::AddrOf: // should be impossible 1953 case UnaryOperator::OffsetOf: 1954 return IntRange::forType(C, E->getType()); 1955 1956 default: 1957 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 1958 } 1959 } 1960 1961 FieldDecl *BitField = E->getBitField(); 1962 if (BitField) { 1963 llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C); 1964 unsigned BitWidth = BitWidthAP.getZExtValue(); 1965 1966 return IntRange(BitWidth, BitField->getType()->isUnsignedIntegerType()); 1967 } 1968 1969 return IntRange::forType(C, E->getType()); 1970} 1971 1972/// Checks whether the given value, which currently has the given 1973/// source semantics, has the same value when coerced through the 1974/// target semantics. 1975bool IsSameFloatAfterCast(const llvm::APFloat &value, 1976 const llvm::fltSemantics &Src, 1977 const llvm::fltSemantics &Tgt) { 1978 llvm::APFloat truncated = value; 1979 1980 bool ignored; 1981 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 1982 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 1983 1984 return truncated.bitwiseIsEqual(value); 1985} 1986 1987/// Checks whether the given value, which currently has the given 1988/// source semantics, has the same value when coerced through the 1989/// target semantics. 1990/// 1991/// The value might be a vector of floats (or a complex number). 1992bool IsSameFloatAfterCast(const APValue &value, 1993 const llvm::fltSemantics &Src, 1994 const llvm::fltSemantics &Tgt) { 1995 if (value.isFloat()) 1996 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 1997 1998 if (value.isVector()) { 1999 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 2000 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 2001 return false; 2002 return true; 2003 } 2004 2005 assert(value.isComplexFloat()); 2006 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 2007 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 2008} 2009 2010} // end anonymous namespace 2011 2012/// \brief Implements -Wsign-compare. 2013/// 2014/// \param lex the left-hand expression 2015/// \param rex the right-hand expression 2016/// \param OpLoc the location of the joining operator 2017/// \param BinOpc binary opcode or 0 2018void Sema::CheckSignCompare(Expr *lex, Expr *rex, SourceLocation OpLoc, 2019 const BinaryOperator::Opcode* BinOpc) { 2020 // Don't warn if we're in an unevaluated context. 2021 if (ExprEvalContexts.back().Context == Unevaluated) 2022 return; 2023 2024 // If either expression is value-dependent, don't warn. We'll get another 2025 // chance at instantiation time. 2026 if (lex->isValueDependent() || rex->isValueDependent()) 2027 return; 2028 2029 QualType lt = lex->getType(), rt = rex->getType(); 2030 2031 // Only warn if both operands are integral. 2032 if (!lt->isIntegerType() || !rt->isIntegerType()) 2033 return; 2034 2035 // In C, the width of a bitfield determines its type, and the 2036 // declared type only contributes the signedness. This duplicates 2037 // the work that will later be done by UsualUnaryConversions. 2038 // Eventually, this check will be reorganized in a way that avoids 2039 // this duplication. 2040 if (!getLangOptions().CPlusPlus) { 2041 QualType tmp; 2042 tmp = Context.isPromotableBitField(lex); 2043 if (!tmp.isNull()) lt = tmp; 2044 tmp = Context.isPromotableBitField(rex); 2045 if (!tmp.isNull()) rt = tmp; 2046 } 2047 2048 // The rule is that the signed operand becomes unsigned, so isolate the 2049 // signed operand. 2050 Expr *signedOperand = lex, *unsignedOperand = rex; 2051 QualType signedType = lt, unsignedType = rt; 2052 if (lt->isSignedIntegerType()) { 2053 if (rt->isSignedIntegerType()) return; 2054 } else { 2055 if (!rt->isSignedIntegerType()) return; 2056 std::swap(signedOperand, unsignedOperand); 2057 std::swap(signedType, unsignedType); 2058 } 2059 2060 unsigned unsignedWidth = Context.getIntWidth(unsignedType); 2061 unsigned signedWidth = Context.getIntWidth(signedType); 2062 2063 // If the unsigned type is strictly smaller than the signed type, 2064 // then (1) the result type will be signed and (2) the unsigned 2065 // value will fit fully within the signed type, and thus the result 2066 // of the comparison will be exact. 2067 if (signedWidth > unsignedWidth) 2068 return; 2069 2070 // Otherwise, calculate the effective ranges. 2071 IntRange signedRange = GetExprRange(Context, signedOperand, signedWidth); 2072 IntRange unsignedRange = GetExprRange(Context, unsignedOperand, unsignedWidth); 2073 2074 // We should never be unable to prove that the unsigned operand is 2075 // non-negative. 2076 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 2077 2078 // If the signed operand is non-negative, then the signed->unsigned 2079 // conversion won't change it. 2080 if (signedRange.NonNegative) { 2081 // Emit warnings for comparisons of unsigned to integer constant 0. 2082 // always false: x < 0 (or 0 > x) 2083 // always true: x >= 0 (or 0 <= x) 2084 llvm::APSInt X; 2085 if (BinOpc && signedOperand->isIntegerConstantExpr(X, Context) && X == 0) { 2086 if (signedOperand != lex) { 2087 if (*BinOpc == BinaryOperator::LT) { 2088 Diag(OpLoc, diag::warn_lunsigned_always_true_comparison) 2089 << "< 0" << "false" 2090 << lex->getSourceRange() << rex->getSourceRange(); 2091 } 2092 else if (*BinOpc == BinaryOperator::GE) { 2093 Diag(OpLoc, diag::warn_lunsigned_always_true_comparison) 2094 << ">= 0" << "true" 2095 << lex->getSourceRange() << rex->getSourceRange(); 2096 } 2097 } 2098 else { 2099 if (*BinOpc == BinaryOperator::GT) { 2100 Diag(OpLoc, diag::warn_runsigned_always_true_comparison) 2101 << "0 >" << "false" 2102 << lex->getSourceRange() << rex->getSourceRange(); 2103 } 2104 else if (*BinOpc == BinaryOperator::LE) { 2105 Diag(OpLoc, diag::warn_runsigned_always_true_comparison) 2106 << "0 <=" << "true" 2107 << lex->getSourceRange() << rex->getSourceRange(); 2108 } 2109 } 2110 } 2111 return; 2112 } 2113 2114 // For (in)equality comparisons, if the unsigned operand is a 2115 // constant which cannot collide with a overflowed signed operand, 2116 // then reinterpreting the signed operand as unsigned will not 2117 // change the result of the comparison. 2118 if (BinOpc && 2119 (*BinOpc == BinaryOperator::EQ || *BinOpc == BinaryOperator::NE) && 2120 unsignedRange.Width < unsignedWidth) 2121 return; 2122 2123 Diag(OpLoc, BinOpc ? diag::warn_mixed_sign_comparison 2124 : diag::warn_mixed_sign_conditional) 2125 << lt << rt << lex->getSourceRange() << rex->getSourceRange(); 2126} 2127 2128/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 2129static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, unsigned diag) { 2130 S.Diag(E->getExprLoc(), diag) << E->getType() << T << E->getSourceRange(); 2131} 2132 2133/// Implements -Wconversion. 2134void Sema::CheckImplicitConversion(Expr *E, QualType T) { 2135 // Don't diagnose in unevaluated contexts. 2136 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 2137 return; 2138 2139 // Don't diagnose for value-dependent expressions. 2140 if (E->isValueDependent()) 2141 return; 2142 2143 const Type *Source = Context.getCanonicalType(E->getType()).getTypePtr(); 2144 const Type *Target = Context.getCanonicalType(T).getTypePtr(); 2145 2146 // Never diagnose implicit casts to bool. 2147 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 2148 return; 2149 2150 // Strip vector types. 2151 if (isa<VectorType>(Source)) { 2152 if (!isa<VectorType>(Target)) 2153 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_vector_scalar); 2154 2155 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 2156 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 2157 } 2158 2159 // Strip complex types. 2160 if (isa<ComplexType>(Source)) { 2161 if (!isa<ComplexType>(Target)) 2162 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_complex_scalar); 2163 2164 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 2165 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 2166 } 2167 2168 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 2169 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 2170 2171 // If the source is floating point... 2172 if (SourceBT && SourceBT->isFloatingPoint()) { 2173 // ...and the target is floating point... 2174 if (TargetBT && TargetBT->isFloatingPoint()) { 2175 // ...then warn if we're dropping FP rank. 2176 2177 // Builtin FP kinds are ordered by increasing FP rank. 2178 if (SourceBT->getKind() > TargetBT->getKind()) { 2179 // Don't warn about float constants that are precisely 2180 // representable in the target type. 2181 Expr::EvalResult result; 2182 if (E->Evaluate(result, Context)) { 2183 // Value might be a float, a float vector, or a float complex. 2184 if (IsSameFloatAfterCast(result.Val, 2185 Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 2186 Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 2187 return; 2188 } 2189 2190 DiagnoseImpCast(*this, E, T, diag::warn_impcast_float_precision); 2191 } 2192 return; 2193 } 2194 2195 // If the target is integral, always warn. 2196 if ((TargetBT && TargetBT->isInteger())) 2197 // TODO: don't warn for integer values? 2198 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_float_integer); 2199 2200 return; 2201 } 2202 2203 if (!Source->isIntegerType() || !Target->isIntegerType()) 2204 return; 2205 2206 IntRange SourceRange = GetExprRange(Context, E, Context.getIntWidth(E->getType())); 2207 IntRange TargetRange = IntRange::forCanonicalType(Context, Target); 2208 2209 // FIXME: also signed<->unsigned? 2210 2211 if (SourceRange.Width > TargetRange.Width) { 2212 // People want to build with -Wshorten-64-to-32 and not -Wconversion 2213 // and by god we'll let them. 2214 if (SourceRange.Width == 64 && TargetRange.Width == 32) 2215 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_integer_64_32); 2216 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_integer_precision); 2217 } 2218 2219 return; 2220} 2221 2222 2223 2224namespace { 2225class UnreachableCodeHandler : public reachable_code::Callback { 2226 Sema &S; 2227public: 2228 UnreachableCodeHandler(Sema *s) : S(*s) {} 2229 2230 void HandleUnreachable(SourceLocation L, SourceRange R1, SourceRange R2) { 2231 S.Diag(L, diag::warn_unreachable) << R1 << R2; 2232 } 2233}; 2234} 2235 2236/// CheckUnreachable - Check for unreachable code. 2237void Sema::CheckUnreachable(AnalysisContext &AC) { 2238 // We avoid checking when there are errors, as the CFG won't faithfully match 2239 // the user's code. 2240 if (getDiagnostics().hasErrorOccurred() || 2241 Diags.getDiagnosticLevel(diag::warn_unreachable) == Diagnostic::Ignored) 2242 return; 2243 2244 UnreachableCodeHandler UC(this); 2245 reachable_code::FindUnreachableCode(AC, UC); 2246} 2247 2248/// CheckFallThrough - Check that we don't fall off the end of a 2249/// Statement that should return a value. 2250/// 2251/// \returns AlwaysFallThrough iff we always fall off the end of the statement, 2252/// MaybeFallThrough iff we might or might not fall off the end, 2253/// NeverFallThroughOrReturn iff we never fall off the end of the statement or 2254/// return. We assume NeverFallThrough iff we never fall off the end of the 2255/// statement but we may return. We assume that functions not marked noreturn 2256/// will return. 2257Sema::ControlFlowKind Sema::CheckFallThrough(AnalysisContext &AC) { 2258 CFG *cfg = AC.getCFG(); 2259 if (cfg == 0) 2260 // FIXME: This should be NeverFallThrough 2261 return NeverFallThroughOrReturn; 2262 2263 // The CFG leaves in dead things, and we don't want the dead code paths to 2264 // confuse us, so we mark all live things first. 2265 std::queue<CFGBlock*> workq; 2266 llvm::BitVector live(cfg->getNumBlockIDs()); 2267 unsigned count = reachable_code::ScanReachableFromBlock(cfg->getEntry(), 2268 live); 2269 2270 bool AddEHEdges = AC.getAddEHEdges(); 2271 if (!AddEHEdges && count != cfg->getNumBlockIDs()) 2272 // When there are things remaining dead, and we didn't add EH edges 2273 // from CallExprs to the catch clauses, we have to go back and 2274 // mark them as live. 2275 for (CFG::iterator I = cfg->begin(), E = cfg->end(); I != E; ++I) { 2276 CFGBlock &b = **I; 2277 if (!live[b.getBlockID()]) { 2278 if (b.pred_begin() == b.pred_end()) { 2279 if (b.getTerminator() && isa<CXXTryStmt>(b.getTerminator())) 2280 // When not adding EH edges from calls, catch clauses 2281 // can otherwise seem dead. Avoid noting them as dead. 2282 count += reachable_code::ScanReachableFromBlock(b, live); 2283 continue; 2284 } 2285 } 2286 } 2287 2288 // Now we know what is live, we check the live precessors of the exit block 2289 // and look for fall through paths, being careful to ignore normal returns, 2290 // and exceptional paths. 2291 bool HasLiveReturn = false; 2292 bool HasFakeEdge = false; 2293 bool HasPlainEdge = false; 2294 bool HasAbnormalEdge = false; 2295 for (CFGBlock::pred_iterator I=cfg->getExit().pred_begin(), 2296 E = cfg->getExit().pred_end(); 2297 I != E; 2298 ++I) { 2299 CFGBlock& B = **I; 2300 if (!live[B.getBlockID()]) 2301 continue; 2302 if (B.size() == 0) { 2303 if (B.getTerminator() && isa<CXXTryStmt>(B.getTerminator())) { 2304 HasAbnormalEdge = true; 2305 continue; 2306 } 2307 2308 // A labeled empty statement, or the entry block... 2309 HasPlainEdge = true; 2310 continue; 2311 } 2312 Stmt *S = B[B.size()-1]; 2313 if (isa<ReturnStmt>(S)) { 2314 HasLiveReturn = true; 2315 continue; 2316 } 2317 if (isa<ObjCAtThrowStmt>(S)) { 2318 HasFakeEdge = true; 2319 continue; 2320 } 2321 if (isa<CXXThrowExpr>(S)) { 2322 HasFakeEdge = true; 2323 continue; 2324 } 2325 if (const AsmStmt *AS = dyn_cast<AsmStmt>(S)) { 2326 if (AS->isMSAsm()) { 2327 HasFakeEdge = true; 2328 HasLiveReturn = true; 2329 continue; 2330 } 2331 } 2332 if (isa<CXXTryStmt>(S)) { 2333 HasAbnormalEdge = true; 2334 continue; 2335 } 2336 2337 bool NoReturnEdge = false; 2338 if (CallExpr *C = dyn_cast<CallExpr>(S)) { 2339 if (B.succ_begin()[0] != &cfg->getExit()) { 2340 HasAbnormalEdge = true; 2341 continue; 2342 } 2343 Expr *CEE = C->getCallee()->IgnoreParenCasts(); 2344 if (CEE->getType().getNoReturnAttr()) { 2345 NoReturnEdge = true; 2346 HasFakeEdge = true; 2347 } else if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(CEE)) { 2348 ValueDecl *VD = DRE->getDecl(); 2349 if (VD->hasAttr<NoReturnAttr>()) { 2350 NoReturnEdge = true; 2351 HasFakeEdge = true; 2352 } 2353 } 2354 } 2355 // FIXME: Add noreturn message sends. 2356 if (NoReturnEdge == false) 2357 HasPlainEdge = true; 2358 } 2359 if (!HasPlainEdge) { 2360 if (HasLiveReturn) 2361 return NeverFallThrough; 2362 return NeverFallThroughOrReturn; 2363 } 2364 if (HasAbnormalEdge || HasFakeEdge || HasLiveReturn) 2365 return MaybeFallThrough; 2366 // This says AlwaysFallThrough for calls to functions that are not marked 2367 // noreturn, that don't return. If people would like this warning to be more 2368 // accurate, such functions should be marked as noreturn. 2369 return AlwaysFallThrough; 2370} 2371 2372/// CheckFallThroughForFunctionDef - Check that we don't fall off the end of a 2373/// function that should return a value. Check that we don't fall off the end 2374/// of a noreturn function. We assume that functions and blocks not marked 2375/// noreturn will return. 2376void Sema::CheckFallThroughForFunctionDef(Decl *D, Stmt *Body, 2377 AnalysisContext &AC) { 2378 // FIXME: Would be nice if we had a better way to control cascading errors, 2379 // but for now, avoid them. The problem is that when Parse sees: 2380 // int foo() { return a; } 2381 // The return is eaten and the Sema code sees just: 2382 // int foo() { } 2383 // which this code would then warn about. 2384 if (getDiagnostics().hasErrorOccurred()) 2385 return; 2386 2387 bool ReturnsVoid = false; 2388 bool HasNoReturn = false; 2389 2390 if (FunctionDecl *FD = dyn_cast<FunctionDecl>(D)) { 2391 // For function templates, class templates and member function templates 2392 // we'll do the analysis at instantiation time. 2393 if (FD->isDependentContext()) 2394 return; 2395 2396 ReturnsVoid = FD->getResultType()->isVoidType(); 2397 HasNoReturn = FD->hasAttr<NoReturnAttr>() || 2398 FD->getType()->getAs<FunctionType>()->getNoReturnAttr(); 2399 2400 } else if (ObjCMethodDecl *MD = dyn_cast<ObjCMethodDecl>(D)) { 2401 ReturnsVoid = MD->getResultType()->isVoidType(); 2402 HasNoReturn = MD->hasAttr<NoReturnAttr>(); 2403 } 2404 2405 // Short circuit for compilation speed. 2406 if ((Diags.getDiagnosticLevel(diag::warn_maybe_falloff_nonvoid_function) 2407 == Diagnostic::Ignored || ReturnsVoid) 2408 && (Diags.getDiagnosticLevel(diag::warn_noreturn_function_has_return_expr) 2409 == Diagnostic::Ignored || !HasNoReturn) 2410 && (Diags.getDiagnosticLevel(diag::warn_suggest_noreturn_block) 2411 == Diagnostic::Ignored || !ReturnsVoid)) 2412 return; 2413 // FIXME: Function try block 2414 if (CompoundStmt *Compound = dyn_cast<CompoundStmt>(Body)) { 2415 switch (CheckFallThrough(AC)) { 2416 case MaybeFallThrough: 2417 if (HasNoReturn) 2418 Diag(Compound->getRBracLoc(), diag::warn_falloff_noreturn_function); 2419 else if (!ReturnsVoid) 2420 Diag(Compound->getRBracLoc(),diag::warn_maybe_falloff_nonvoid_function); 2421 break; 2422 case AlwaysFallThrough: 2423 if (HasNoReturn) 2424 Diag(Compound->getRBracLoc(), diag::warn_falloff_noreturn_function); 2425 else if (!ReturnsVoid) 2426 Diag(Compound->getRBracLoc(), diag::warn_falloff_nonvoid_function); 2427 break; 2428 case NeverFallThroughOrReturn: 2429 if (ReturnsVoid && !HasNoReturn) 2430 Diag(Compound->getLBracLoc(), diag::warn_suggest_noreturn_function); 2431 break; 2432 case NeverFallThrough: 2433 break; 2434 } 2435 } 2436} 2437 2438/// CheckFallThroughForBlock - Check that we don't fall off the end of a block 2439/// that should return a value. Check that we don't fall off the end of a 2440/// noreturn block. We assume that functions and blocks not marked noreturn 2441/// will return. 2442void Sema::CheckFallThroughForBlock(QualType BlockTy, Stmt *Body, 2443 AnalysisContext &AC) { 2444 // FIXME: Would be nice if we had a better way to control cascading errors, 2445 // but for now, avoid them. The problem is that when Parse sees: 2446 // int foo() { return a; } 2447 // The return is eaten and the Sema code sees just: 2448 // int foo() { } 2449 // which this code would then warn about. 2450 if (getDiagnostics().hasErrorOccurred()) 2451 return; 2452 bool ReturnsVoid = false; 2453 bool HasNoReturn = false; 2454 if (const FunctionType *FT =BlockTy->getPointeeType()->getAs<FunctionType>()){ 2455 if (FT->getResultType()->isVoidType()) 2456 ReturnsVoid = true; 2457 if (FT->getNoReturnAttr()) 2458 HasNoReturn = true; 2459 } 2460 2461 // Short circuit for compilation speed. 2462 if (ReturnsVoid 2463 && !HasNoReturn 2464 && (Diags.getDiagnosticLevel(diag::warn_suggest_noreturn_block) 2465 == Diagnostic::Ignored || !ReturnsVoid)) 2466 return; 2467 // FIXME: Funtion try block 2468 if (CompoundStmt *Compound = dyn_cast<CompoundStmt>(Body)) { 2469 switch (CheckFallThrough(AC)) { 2470 case MaybeFallThrough: 2471 if (HasNoReturn) 2472 Diag(Compound->getRBracLoc(), diag::err_noreturn_block_has_return_expr); 2473 else if (!ReturnsVoid) 2474 Diag(Compound->getRBracLoc(), diag::err_maybe_falloff_nonvoid_block); 2475 break; 2476 case AlwaysFallThrough: 2477 if (HasNoReturn) 2478 Diag(Compound->getRBracLoc(), diag::err_noreturn_block_has_return_expr); 2479 else if (!ReturnsVoid) 2480 Diag(Compound->getRBracLoc(), diag::err_falloff_nonvoid_block); 2481 break; 2482 case NeverFallThroughOrReturn: 2483 if (ReturnsVoid) 2484 Diag(Compound->getLBracLoc(), diag::warn_suggest_noreturn_block); 2485 break; 2486 case NeverFallThrough: 2487 break; 2488 } 2489 } 2490} 2491 2492/// CheckParmsForFunctionDef - Check that the parameters of the given 2493/// function are appropriate for the definition of a function. This 2494/// takes care of any checks that cannot be performed on the 2495/// declaration itself, e.g., that the types of each of the function 2496/// parameters are complete. 2497bool Sema::CheckParmsForFunctionDef(FunctionDecl *FD) { 2498 bool HasInvalidParm = false; 2499 for (unsigned p = 0, NumParams = FD->getNumParams(); p < NumParams; ++p) { 2500 ParmVarDecl *Param = FD->getParamDecl(p); 2501 2502 // C99 6.7.5.3p4: the parameters in a parameter type list in a 2503 // function declarator that is part of a function definition of 2504 // that function shall not have incomplete type. 2505 // 2506 // This is also C++ [dcl.fct]p6. 2507 if (!Param->isInvalidDecl() && 2508 RequireCompleteType(Param->getLocation(), Param->getType(), 2509 diag::err_typecheck_decl_incomplete_type)) { 2510 Param->setInvalidDecl(); 2511 HasInvalidParm = true; 2512 } 2513 2514 // C99 6.9.1p5: If the declarator includes a parameter type list, the 2515 // declaration of each parameter shall include an identifier. 2516 if (Param->getIdentifier() == 0 && 2517 !Param->isImplicit() && 2518 !getLangOptions().CPlusPlus) 2519 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 2520 2521 // C99 6.7.5.3p12: 2522 // If the function declarator is not part of a definition of that 2523 // function, parameters may have incomplete type and may use the [*] 2524 // notation in their sequences of declarator specifiers to specify 2525 // variable length array types. 2526 QualType PType = Param->getOriginalType(); 2527 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 2528 if (AT->getSizeModifier() == ArrayType::Star) { 2529 // FIXME: This diagnosic should point the the '[*]' if source-location 2530 // information is added for it. 2531 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 2532 } 2533 } 2534 2535 if (getLangOptions().CPlusPlus) 2536 if (const RecordType *RT = Param->getType()->getAs<RecordType>()) 2537 FinalizeVarWithDestructor(Param, RT); 2538 } 2539 2540 return HasInvalidParm; 2541} 2542