SemaChecking.cpp revision 62f46195bb61f91f9b3b476f4dcfa10126c18ff2
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); 64 const char *StrData = Buffer.first+LocInfo.second; 65 66 // Create a langops struct and enable trigraphs. This is sufficient for 67 // relexing tokens. 68 LangOptions LangOpts; 69 LangOpts.Trigraphs = true; 70 71 // Create a lexer starting at the beginning of this token. 72 Lexer TheLexer(StrTokSpellingLoc, LangOpts, Buffer.first, StrData, 73 Buffer.second); 74 Token TheTok; 75 TheLexer.LexFromRawLexer(TheTok); 76 77 // Use the StringLiteralParser to compute the length of the string in bytes. 78 StringLiteralParser SLP(&TheTok, 1, PP); 79 unsigned TokNumBytes = SLP.GetStringLength(); 80 81 // If the byte is in this token, return the location of the byte. 82 if (ByteNo < TokNumBytes || 83 (ByteNo == TokNumBytes && TokNo == SL->getNumConcatenated())) { 84 unsigned Offset = 85 StringLiteralParser::getOffsetOfStringByte(TheTok, ByteNo, PP); 86 87 // Now that we know the offset of the token in the spelling, use the 88 // preprocessor to get the offset in the original source. 89 return PP.AdvanceToTokenCharacter(StrTokLoc, Offset); 90 } 91 92 // Move to the next string token. 93 ++TokNo; 94 ByteNo -= TokNumBytes; 95 } 96} 97 98/// CheckablePrintfAttr - does a function call have a "printf" attribute 99/// and arguments that merit checking? 100bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) { 101 if (Format->getType() == "printf") return true; 102 if (Format->getType() == "printf0") { 103 // printf0 allows null "format" string; if so don't check format/args 104 unsigned format_idx = Format->getFormatIdx() - 1; 105 // Does the index refer to the implicit object argument? 106 if (isa<CXXMemberCallExpr>(TheCall)) { 107 if (format_idx == 0) 108 return false; 109 --format_idx; 110 } 111 if (format_idx < TheCall->getNumArgs()) { 112 Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts(); 113 if (!Format->isNullPointerConstant(Context, 114 Expr::NPC_ValueDependentIsNull)) 115 return true; 116 } 117 } 118 return false; 119} 120 121Action::OwningExprResult 122Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 123 OwningExprResult TheCallResult(Owned(TheCall)); 124 125 switch (BuiltinID) { 126 case Builtin::BI__builtin___CFStringMakeConstantString: 127 assert(TheCall->getNumArgs() == 1 && 128 "Wrong # arguments to builtin CFStringMakeConstantString"); 129 if (CheckObjCString(TheCall->getArg(0))) 130 return ExprError(); 131 break; 132 case Builtin::BI__builtin_stdarg_start: 133 case Builtin::BI__builtin_va_start: 134 if (SemaBuiltinVAStart(TheCall)) 135 return ExprError(); 136 break; 137 case Builtin::BI__builtin_isgreater: 138 case Builtin::BI__builtin_isgreaterequal: 139 case Builtin::BI__builtin_isless: 140 case Builtin::BI__builtin_islessequal: 141 case Builtin::BI__builtin_islessgreater: 142 case Builtin::BI__builtin_isunordered: 143 if (SemaBuiltinUnorderedCompare(TheCall)) 144 return ExprError(); 145 break; 146 case Builtin::BI__builtin_fpclassify: 147 if (SemaBuiltinFPClassification(TheCall, 6)) 148 return ExprError(); 149 break; 150 case Builtin::BI__builtin_isfinite: 151 case Builtin::BI__builtin_isinf: 152 case Builtin::BI__builtin_isinf_sign: 153 case Builtin::BI__builtin_isnan: 154 case Builtin::BI__builtin_isnormal: 155 if (SemaBuiltinFPClassification(TheCall, 1)) 156 return ExprError(); 157 break; 158 case Builtin::BI__builtin_return_address: 159 case Builtin::BI__builtin_frame_address: 160 if (SemaBuiltinStackAddress(TheCall)) 161 return ExprError(); 162 break; 163 case Builtin::BI__builtin_eh_return_data_regno: 164 if (SemaBuiltinEHReturnDataRegNo(TheCall)) 165 return ExprError(); 166 break; 167 case Builtin::BI__builtin_shufflevector: 168 return SemaBuiltinShuffleVector(TheCall); 169 // TheCall will be freed by the smart pointer here, but that's fine, since 170 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 171 case Builtin::BI__builtin_prefetch: 172 if (SemaBuiltinPrefetch(TheCall)) 173 return ExprError(); 174 break; 175 case Builtin::BI__builtin_object_size: 176 if (SemaBuiltinObjectSize(TheCall)) 177 return ExprError(); 178 break; 179 case Builtin::BI__builtin_longjmp: 180 if (SemaBuiltinLongjmp(TheCall)) 181 return ExprError(); 182 break; 183 case Builtin::BI__sync_fetch_and_add: 184 case Builtin::BI__sync_fetch_and_sub: 185 case Builtin::BI__sync_fetch_and_or: 186 case Builtin::BI__sync_fetch_and_and: 187 case Builtin::BI__sync_fetch_and_xor: 188 case Builtin::BI__sync_fetch_and_nand: 189 case Builtin::BI__sync_add_and_fetch: 190 case Builtin::BI__sync_sub_and_fetch: 191 case Builtin::BI__sync_and_and_fetch: 192 case Builtin::BI__sync_or_and_fetch: 193 case Builtin::BI__sync_xor_and_fetch: 194 case Builtin::BI__sync_nand_and_fetch: 195 case Builtin::BI__sync_val_compare_and_swap: 196 case Builtin::BI__sync_bool_compare_and_swap: 197 case Builtin::BI__sync_lock_test_and_set: 198 case Builtin::BI__sync_lock_release: 199 if (SemaBuiltinAtomicOverloaded(TheCall)) 200 return ExprError(); 201 break; 202 } 203 204 return move(TheCallResult); 205} 206 207/// CheckFunctionCall - Check a direct function call for various correctness 208/// and safety properties not strictly enforced by the C type system. 209bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 210 // Get the IdentifierInfo* for the called function. 211 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 212 213 // None of the checks below are needed for functions that don't have 214 // simple names (e.g., C++ conversion functions). 215 if (!FnInfo) 216 return false; 217 218 // FIXME: This mechanism should be abstracted to be less fragile and 219 // more efficient. For example, just map function ids to custom 220 // handlers. 221 222 // Printf checking. 223 if (const FormatAttr *Format = FDecl->getAttr<FormatAttr>()) { 224 if (CheckablePrintfAttr(Format, TheCall)) { 225 bool HasVAListArg = Format->getFirstArg() == 0; 226 if (!HasVAListArg) { 227 if (const FunctionProtoType *Proto 228 = FDecl->getType()->getAs<FunctionProtoType>()) 229 HasVAListArg = !Proto->isVariadic(); 230 } 231 CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 232 HasVAListArg ? 0 : Format->getFirstArg() - 1); 233 } 234 } 235 236 for (const NonNullAttr *NonNull = FDecl->getAttr<NonNullAttr>(); NonNull; 237 NonNull = NonNull->getNext<NonNullAttr>()) 238 CheckNonNullArguments(NonNull, TheCall); 239 240 return false; 241} 242 243bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 244 // Printf checking. 245 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 246 if (!Format) 247 return false; 248 249 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 250 if (!V) 251 return false; 252 253 QualType Ty = V->getType(); 254 if (!Ty->isBlockPointerType()) 255 return false; 256 257 if (!CheckablePrintfAttr(Format, TheCall)) 258 return false; 259 260 bool HasVAListArg = Format->getFirstArg() == 0; 261 if (!HasVAListArg) { 262 const FunctionType *FT = 263 Ty->getAs<BlockPointerType>()->getPointeeType()->getAs<FunctionType>(); 264 if (const FunctionProtoType *Proto = dyn_cast<FunctionProtoType>(FT)) 265 HasVAListArg = !Proto->isVariadic(); 266 } 267 CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 268 HasVAListArg ? 0 : Format->getFirstArg() - 1); 269 270 return false; 271} 272 273/// SemaBuiltinAtomicOverloaded - We have a call to a function like 274/// __sync_fetch_and_add, which is an overloaded function based on the pointer 275/// type of its first argument. The main ActOnCallExpr routines have already 276/// promoted the types of arguments because all of these calls are prototyped as 277/// void(...). 278/// 279/// This function goes through and does final semantic checking for these 280/// builtins, 281bool Sema::SemaBuiltinAtomicOverloaded(CallExpr *TheCall) { 282 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 283 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 284 285 // Ensure that we have at least one argument to do type inference from. 286 if (TheCall->getNumArgs() < 1) 287 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 288 << 0 << TheCall->getCallee()->getSourceRange(); 289 290 // Inspect the first argument of the atomic builtin. This should always be 291 // a pointer type, whose element is an integral scalar or pointer type. 292 // Because it is a pointer type, we don't have to worry about any implicit 293 // casts here. 294 Expr *FirstArg = TheCall->getArg(0); 295 if (!FirstArg->getType()->isPointerType()) 296 return Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 297 << FirstArg->getType() << FirstArg->getSourceRange(); 298 299 QualType ValType = FirstArg->getType()->getAs<PointerType>()->getPointeeType(); 300 if (!ValType->isIntegerType() && !ValType->isPointerType() && 301 !ValType->isBlockPointerType()) 302 return Diag(DRE->getLocStart(), 303 diag::err_atomic_builtin_must_be_pointer_intptr) 304 << FirstArg->getType() << FirstArg->getSourceRange(); 305 306 // We need to figure out which concrete builtin this maps onto. For example, 307 // __sync_fetch_and_add with a 2 byte object turns into 308 // __sync_fetch_and_add_2. 309#define BUILTIN_ROW(x) \ 310 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 311 Builtin::BI##x##_8, Builtin::BI##x##_16 } 312 313 static const unsigned BuiltinIndices[][5] = { 314 BUILTIN_ROW(__sync_fetch_and_add), 315 BUILTIN_ROW(__sync_fetch_and_sub), 316 BUILTIN_ROW(__sync_fetch_and_or), 317 BUILTIN_ROW(__sync_fetch_and_and), 318 BUILTIN_ROW(__sync_fetch_and_xor), 319 BUILTIN_ROW(__sync_fetch_and_nand), 320 321 BUILTIN_ROW(__sync_add_and_fetch), 322 BUILTIN_ROW(__sync_sub_and_fetch), 323 BUILTIN_ROW(__sync_and_and_fetch), 324 BUILTIN_ROW(__sync_or_and_fetch), 325 BUILTIN_ROW(__sync_xor_and_fetch), 326 BUILTIN_ROW(__sync_nand_and_fetch), 327 328 BUILTIN_ROW(__sync_val_compare_and_swap), 329 BUILTIN_ROW(__sync_bool_compare_and_swap), 330 BUILTIN_ROW(__sync_lock_test_and_set), 331 BUILTIN_ROW(__sync_lock_release) 332 }; 333#undef BUILTIN_ROW 334 335 // Determine the index of the size. 336 unsigned SizeIndex; 337 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 338 case 1: SizeIndex = 0; break; 339 case 2: SizeIndex = 1; break; 340 case 4: SizeIndex = 2; break; 341 case 8: SizeIndex = 3; break; 342 case 16: SizeIndex = 4; break; 343 default: 344 return Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 345 << FirstArg->getType() << FirstArg->getSourceRange(); 346 } 347 348 // Each of these builtins has one pointer argument, followed by some number of 349 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 350 // that we ignore. Find out which row of BuiltinIndices to read from as well 351 // as the number of fixed args. 352 unsigned BuiltinID = FDecl->getBuiltinID(); 353 unsigned BuiltinIndex, NumFixed = 1; 354 switch (BuiltinID) { 355 default: assert(0 && "Unknown overloaded atomic builtin!"); 356 case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break; 357 case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break; 358 case Builtin::BI__sync_fetch_and_or: BuiltinIndex = 2; break; 359 case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break; 360 case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break; 361 case Builtin::BI__sync_fetch_and_nand:BuiltinIndex = 5; break; 362 363 case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 6; break; 364 case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 7; break; 365 case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 8; break; 366 case Builtin::BI__sync_or_and_fetch: BuiltinIndex = 9; break; 367 case Builtin::BI__sync_xor_and_fetch: BuiltinIndex =10; break; 368 case Builtin::BI__sync_nand_and_fetch:BuiltinIndex =11; break; 369 370 case Builtin::BI__sync_val_compare_and_swap: 371 BuiltinIndex = 12; 372 NumFixed = 2; 373 break; 374 case Builtin::BI__sync_bool_compare_and_swap: 375 BuiltinIndex = 13; 376 NumFixed = 2; 377 break; 378 case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 14; break; 379 case Builtin::BI__sync_lock_release: 380 BuiltinIndex = 15; 381 NumFixed = 0; 382 break; 383 } 384 385 // Now that we know how many fixed arguments we expect, first check that we 386 // have at least that many. 387 if (TheCall->getNumArgs() < 1+NumFixed) 388 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 389 << 0 << TheCall->getCallee()->getSourceRange(); 390 391 392 // Get the decl for the concrete builtin from this, we can tell what the 393 // concrete integer type we should convert to is. 394 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 395 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 396 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 397 FunctionDecl *NewBuiltinDecl = 398 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 399 TUScope, false, DRE->getLocStart())); 400 const FunctionProtoType *BuiltinFT = 401 NewBuiltinDecl->getType()->getAs<FunctionProtoType>(); 402 ValType = BuiltinFT->getArgType(0)->getAs<PointerType>()->getPointeeType(); 403 404 // If the first type needs to be converted (e.g. void** -> int*), do it now. 405 if (BuiltinFT->getArgType(0) != FirstArg->getType()) { 406 ImpCastExprToType(FirstArg, BuiltinFT->getArgType(0), CastExpr::CK_BitCast); 407 TheCall->setArg(0, FirstArg); 408 } 409 410 // Next, walk the valid ones promoting to the right type. 411 for (unsigned i = 0; i != NumFixed; ++i) { 412 Expr *Arg = TheCall->getArg(i+1); 413 414 // If the argument is an implicit cast, then there was a promotion due to 415 // "...", just remove it now. 416 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) { 417 Arg = ICE->getSubExpr(); 418 ICE->setSubExpr(0); 419 ICE->Destroy(Context); 420 TheCall->setArg(i+1, Arg); 421 } 422 423 // GCC does an implicit conversion to the pointer or integer ValType. This 424 // can fail in some cases (1i -> int**), check for this error case now. 425 CastExpr::CastKind Kind = CastExpr::CK_Unknown; 426 CXXMethodDecl *ConversionDecl = 0; 427 if (CheckCastTypes(Arg->getSourceRange(), ValType, Arg, Kind, 428 ConversionDecl)) 429 return true; 430 431 // Okay, we have something that *can* be converted to the right type. Check 432 // to see if there is a potentially weird extension going on here. This can 433 // happen when you do an atomic operation on something like an char* and 434 // pass in 42. The 42 gets converted to char. This is even more strange 435 // for things like 45.123 -> char, etc. 436 // FIXME: Do this check. 437 ImpCastExprToType(Arg, ValType, Kind, /*isLvalue=*/false); 438 TheCall->setArg(i+1, Arg); 439 } 440 441 // Switch the DeclRefExpr to refer to the new decl. 442 DRE->setDecl(NewBuiltinDecl); 443 DRE->setType(NewBuiltinDecl->getType()); 444 445 // Set the callee in the CallExpr. 446 // FIXME: This leaks the original parens and implicit casts. 447 Expr *PromotedCall = DRE; 448 UsualUnaryConversions(PromotedCall); 449 TheCall->setCallee(PromotedCall); 450 451 452 // Change the result type of the call to match the result type of the decl. 453 TheCall->setType(NewBuiltinDecl->getResultType()); 454 return false; 455} 456 457 458/// CheckObjCString - Checks that the argument to the builtin 459/// CFString constructor is correct 460/// FIXME: GCC currently emits the following warning: 461/// "warning: input conversion stopped due to an input byte that does not 462/// belong to the input codeset UTF-8" 463/// Note: It might also make sense to do the UTF-16 conversion here (would 464/// simplify the backend). 465bool Sema::CheckObjCString(Expr *Arg) { 466 Arg = Arg->IgnoreParenCasts(); 467 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 468 469 if (!Literal || Literal->isWide()) { 470 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 471 << Arg->getSourceRange(); 472 return true; 473 } 474 475 const char *Data = Literal->getStrData(); 476 unsigned Length = Literal->getByteLength(); 477 478 for (unsigned i = 0; i < Length; ++i) { 479 if (!Data[i]) { 480 Diag(getLocationOfStringLiteralByte(Literal, i), 481 diag::warn_cfstring_literal_contains_nul_character) 482 << Arg->getSourceRange(); 483 break; 484 } 485 } 486 487 return false; 488} 489 490/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 491/// Emit an error and return true on failure, return false on success. 492bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 493 Expr *Fn = TheCall->getCallee(); 494 if (TheCall->getNumArgs() > 2) { 495 Diag(TheCall->getArg(2)->getLocStart(), 496 diag::err_typecheck_call_too_many_args) 497 << 0 /*function call*/ << Fn->getSourceRange() 498 << SourceRange(TheCall->getArg(2)->getLocStart(), 499 (*(TheCall->arg_end()-1))->getLocEnd()); 500 return true; 501 } 502 503 if (TheCall->getNumArgs() < 2) { 504 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 505 << 0 /*function call*/; 506 } 507 508 // Determine whether the current function is variadic or not. 509 bool isVariadic; 510 if (CurBlock) 511 isVariadic = CurBlock->isVariadic; 512 else if (getCurFunctionDecl()) { 513 if (FunctionProtoType* FTP = 514 dyn_cast<FunctionProtoType>(getCurFunctionDecl()->getType())) 515 isVariadic = FTP->isVariadic(); 516 else 517 isVariadic = false; 518 } else { 519 isVariadic = getCurMethodDecl()->isVariadic(); 520 } 521 522 if (!isVariadic) { 523 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 524 return true; 525 } 526 527 // Verify that the second argument to the builtin is the last argument of the 528 // current function or method. 529 bool SecondArgIsLastNamedArgument = false; 530 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 531 532 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 533 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 534 // FIXME: This isn't correct for methods (results in bogus warning). 535 // Get the last formal in the current function. 536 const ParmVarDecl *LastArg; 537 if (CurBlock) 538 LastArg = *(CurBlock->TheDecl->param_end()-1); 539 else if (FunctionDecl *FD = getCurFunctionDecl()) 540 LastArg = *(FD->param_end()-1); 541 else 542 LastArg = *(getCurMethodDecl()->param_end()-1); 543 SecondArgIsLastNamedArgument = PV == LastArg; 544 } 545 } 546 547 if (!SecondArgIsLastNamedArgument) 548 Diag(TheCall->getArg(1)->getLocStart(), 549 diag::warn_second_parameter_of_va_start_not_last_named_argument); 550 return false; 551} 552 553/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 554/// friends. This is declared to take (...), so we have to check everything. 555bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 556 if (TheCall->getNumArgs() < 2) 557 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 558 << 0 /*function call*/; 559 if (TheCall->getNumArgs() > 2) 560 return Diag(TheCall->getArg(2)->getLocStart(), 561 diag::err_typecheck_call_too_many_args) 562 << 0 /*function call*/ 563 << SourceRange(TheCall->getArg(2)->getLocStart(), 564 (*(TheCall->arg_end()-1))->getLocEnd()); 565 566 Expr *OrigArg0 = TheCall->getArg(0); 567 Expr *OrigArg1 = TheCall->getArg(1); 568 569 // Do standard promotions between the two arguments, returning their common 570 // type. 571 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 572 573 // Make sure any conversions are pushed back into the call; this is 574 // type safe since unordered compare builtins are declared as "_Bool 575 // foo(...)". 576 TheCall->setArg(0, OrigArg0); 577 TheCall->setArg(1, OrigArg1); 578 579 if (OrigArg0->isTypeDependent() || OrigArg1->isTypeDependent()) 580 return false; 581 582 // If the common type isn't a real floating type, then the arguments were 583 // invalid for this operation. 584 if (!Res->isRealFloatingType()) 585 return Diag(OrigArg0->getLocStart(), 586 diag::err_typecheck_call_invalid_ordered_compare) 587 << OrigArg0->getType() << OrigArg1->getType() 588 << SourceRange(OrigArg0->getLocStart(), OrigArg1->getLocEnd()); 589 590 return false; 591} 592 593/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 594/// __builtin_isnan and friends. This is declared to take (...), so we have 595/// to check everything. We expect the last argument to be a floating point 596/// value. 597bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 598 if (TheCall->getNumArgs() < NumArgs) 599 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 600 << 0 /*function call*/; 601 if (TheCall->getNumArgs() > NumArgs) 602 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 603 diag::err_typecheck_call_too_many_args) 604 << 0 /*function call*/ 605 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 606 (*(TheCall->arg_end()-1))->getLocEnd()); 607 608 Expr *OrigArg = TheCall->getArg(NumArgs-1); 609 610 if (OrigArg->isTypeDependent()) 611 return false; 612 613 // This operation requires a floating-point number 614 if (!OrigArg->getType()->isRealFloatingType()) 615 return Diag(OrigArg->getLocStart(), 616 diag::err_typecheck_call_invalid_unary_fp) 617 << OrigArg->getType() << OrigArg->getSourceRange(); 618 619 return false; 620} 621 622bool Sema::SemaBuiltinStackAddress(CallExpr *TheCall) { 623 // The signature for these builtins is exact; the only thing we need 624 // to check is that the argument is a constant. 625 SourceLocation Loc; 626 if (!TheCall->getArg(0)->isTypeDependent() && 627 !TheCall->getArg(0)->isValueDependent() && 628 !TheCall->getArg(0)->isIntegerConstantExpr(Context, &Loc)) 629 return Diag(Loc, diag::err_stack_const_level) << TheCall->getSourceRange(); 630 631 return false; 632} 633 634/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 635// This is declared to take (...), so we have to check everything. 636Action::OwningExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 637 if (TheCall->getNumArgs() < 3) 638 return ExprError(Diag(TheCall->getLocEnd(), 639 diag::err_typecheck_call_too_few_args) 640 << 0 /*function call*/ << TheCall->getSourceRange()); 641 642 unsigned numElements = std::numeric_limits<unsigned>::max(); 643 if (!TheCall->getArg(0)->isTypeDependent() && 644 !TheCall->getArg(1)->isTypeDependent()) { 645 QualType FAType = TheCall->getArg(0)->getType(); 646 QualType SAType = TheCall->getArg(1)->getType(); 647 648 if (!FAType->isVectorType() || !SAType->isVectorType()) { 649 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 650 << SourceRange(TheCall->getArg(0)->getLocStart(), 651 TheCall->getArg(1)->getLocEnd()); 652 return ExprError(); 653 } 654 655 if (!Context.hasSameUnqualifiedType(FAType, SAType)) { 656 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 657 << SourceRange(TheCall->getArg(0)->getLocStart(), 658 TheCall->getArg(1)->getLocEnd()); 659 return ExprError(); 660 } 661 662 numElements = FAType->getAs<VectorType>()->getNumElements(); 663 if (TheCall->getNumArgs() != numElements+2) { 664 if (TheCall->getNumArgs() < numElements+2) 665 return ExprError(Diag(TheCall->getLocEnd(), 666 diag::err_typecheck_call_too_few_args) 667 << 0 /*function call*/ << TheCall->getSourceRange()); 668 return ExprError(Diag(TheCall->getLocEnd(), 669 diag::err_typecheck_call_too_many_args) 670 << 0 /*function call*/ << TheCall->getSourceRange()); 671 } 672 } 673 674 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 675 if (TheCall->getArg(i)->isTypeDependent() || 676 TheCall->getArg(i)->isValueDependent()) 677 continue; 678 679 llvm::APSInt Result(32); 680 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 681 return ExprError(Diag(TheCall->getLocStart(), 682 diag::err_shufflevector_nonconstant_argument) 683 << TheCall->getArg(i)->getSourceRange()); 684 685 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 686 return ExprError(Diag(TheCall->getLocStart(), 687 diag::err_shufflevector_argument_too_large) 688 << TheCall->getArg(i)->getSourceRange()); 689 } 690 691 llvm::SmallVector<Expr*, 32> exprs; 692 693 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 694 exprs.push_back(TheCall->getArg(i)); 695 TheCall->setArg(i, 0); 696 } 697 698 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 699 exprs.size(), exprs[0]->getType(), 700 TheCall->getCallee()->getLocStart(), 701 TheCall->getRParenLoc())); 702} 703 704/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 705// This is declared to take (const void*, ...) and can take two 706// optional constant int args. 707bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 708 unsigned NumArgs = TheCall->getNumArgs(); 709 710 if (NumArgs > 3) 711 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_many_args) 712 << 0 /*function call*/ << TheCall->getSourceRange(); 713 714 // Argument 0 is checked for us and the remaining arguments must be 715 // constant integers. 716 for (unsigned i = 1; i != NumArgs; ++i) { 717 Expr *Arg = TheCall->getArg(i); 718 if (Arg->isTypeDependent()) 719 continue; 720 721 if (!Arg->getType()->isIntegralType()) 722 return Diag(TheCall->getLocStart(), diag::err_prefetch_invalid_arg_type) 723 << Arg->getSourceRange(); 724 725 ImpCastExprToType(Arg, Context.IntTy, CastExpr::CK_IntegralCast); 726 TheCall->setArg(i, Arg); 727 728 if (Arg->isValueDependent()) 729 continue; 730 731 llvm::APSInt Result; 732 if (!Arg->isIntegerConstantExpr(Result, Context)) 733 return Diag(TheCall->getLocStart(), diag::err_prefetch_invalid_arg_ice) 734 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 735 736 // FIXME: gcc issues a warning and rewrites these to 0. These 737 // seems especially odd for the third argument since the default 738 // is 3. 739 if (i == 1) { 740 if (Result.getLimitedValue() > 1) 741 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 742 << "0" << "1" << Arg->getSourceRange(); 743 } else { 744 if (Result.getLimitedValue() > 3) 745 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 746 << "0" << "3" << Arg->getSourceRange(); 747 } 748 } 749 750 return false; 751} 752 753/// SemaBuiltinEHReturnDataRegNo - Handle __builtin_eh_return_data_regno, the 754/// operand must be an integer constant. 755bool Sema::SemaBuiltinEHReturnDataRegNo(CallExpr *TheCall) { 756 llvm::APSInt Result; 757 if (!TheCall->getArg(0)->isIntegerConstantExpr(Result, Context)) 758 return Diag(TheCall->getLocStart(), diag::err_expr_not_ice) 759 << TheCall->getArg(0)->getSourceRange(); 760 761 return false; 762} 763 764 765/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 766/// int type). This simply type checks that type is one of the defined 767/// constants (0-3). 768// For compatability check 0-3, llvm only handles 0 and 2. 769bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 770 Expr *Arg = TheCall->getArg(1); 771 if (Arg->isTypeDependent()) 772 return false; 773 774 QualType ArgType = Arg->getType(); 775 const BuiltinType *BT = ArgType->getAs<BuiltinType>(); 776 llvm::APSInt Result(32); 777 if (!BT || BT->getKind() != BuiltinType::Int) 778 return Diag(TheCall->getLocStart(), diag::err_object_size_invalid_argument) 779 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 780 781 if (Arg->isValueDependent()) 782 return false; 783 784 if (!Arg->isIntegerConstantExpr(Result, Context)) { 785 return Diag(TheCall->getLocStart(), diag::err_object_size_invalid_argument) 786 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 787 } 788 789 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 790 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 791 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 792 } 793 794 return false; 795} 796 797/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 798/// This checks that val is a constant 1. 799bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 800 Expr *Arg = TheCall->getArg(1); 801 if (Arg->isTypeDependent() || Arg->isValueDependent()) 802 return false; 803 804 llvm::APSInt Result(32); 805 if (!Arg->isIntegerConstantExpr(Result, Context) || Result != 1) 806 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 807 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 808 809 return false; 810} 811 812// Handle i > 1 ? "x" : "y", recursivelly 813bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 814 bool HasVAListArg, 815 unsigned format_idx, unsigned firstDataArg) { 816 if (E->isTypeDependent() || E->isValueDependent()) 817 return false; 818 819 switch (E->getStmtClass()) { 820 case Stmt::ConditionalOperatorClass: { 821 const ConditionalOperator *C = cast<ConditionalOperator>(E); 822 return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, 823 HasVAListArg, format_idx, firstDataArg) 824 && SemaCheckStringLiteral(C->getRHS(), TheCall, 825 HasVAListArg, format_idx, firstDataArg); 826 } 827 828 case Stmt::ImplicitCastExprClass: { 829 const ImplicitCastExpr *Expr = cast<ImplicitCastExpr>(E); 830 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 831 format_idx, firstDataArg); 832 } 833 834 case Stmt::ParenExprClass: { 835 const ParenExpr *Expr = cast<ParenExpr>(E); 836 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 837 format_idx, firstDataArg); 838 } 839 840 case Stmt::DeclRefExprClass: { 841 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 842 843 // As an exception, do not flag errors for variables binding to 844 // const string literals. 845 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 846 bool isConstant = false; 847 QualType T = DR->getType(); 848 849 if (const ArrayType *AT = Context.getAsArrayType(T)) { 850 isConstant = AT->getElementType().isConstant(Context); 851 } else if (const PointerType *PT = T->getAs<PointerType>()) { 852 isConstant = T.isConstant(Context) && 853 PT->getPointeeType().isConstant(Context); 854 } 855 856 if (isConstant) { 857 if (const Expr *Init = VD->getAnyInitializer()) 858 return SemaCheckStringLiteral(Init, TheCall, 859 HasVAListArg, format_idx, firstDataArg); 860 } 861 862 // For vprintf* functions (i.e., HasVAListArg==true), we add a 863 // special check to see if the format string is a function parameter 864 // of the function calling the printf function. If the function 865 // has an attribute indicating it is a printf-like function, then we 866 // should suppress warnings concerning non-literals being used in a call 867 // to a vprintf function. For example: 868 // 869 // void 870 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 871 // va_list ap; 872 // va_start(ap, fmt); 873 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 874 // ... 875 // 876 // 877 // FIXME: We don't have full attribute support yet, so just check to see 878 // if the argument is a DeclRefExpr that references a parameter. We'll 879 // add proper support for checking the attribute later. 880 if (HasVAListArg) 881 if (isa<ParmVarDecl>(VD)) 882 return true; 883 } 884 885 return false; 886 } 887 888 case Stmt::CallExprClass: { 889 const CallExpr *CE = cast<CallExpr>(E); 890 if (const ImplicitCastExpr *ICE 891 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 892 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 893 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 894 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 895 unsigned ArgIndex = FA->getFormatIdx(); 896 const Expr *Arg = CE->getArg(ArgIndex - 1); 897 898 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 899 format_idx, firstDataArg); 900 } 901 } 902 } 903 } 904 905 return false; 906 } 907 case Stmt::ObjCStringLiteralClass: 908 case Stmt::StringLiteralClass: { 909 const StringLiteral *StrE = NULL; 910 911 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 912 StrE = ObjCFExpr->getString(); 913 else 914 StrE = cast<StringLiteral>(E); 915 916 if (StrE) { 917 CheckPrintfString(StrE, E, TheCall, HasVAListArg, format_idx, 918 firstDataArg); 919 return true; 920 } 921 922 return false; 923 } 924 925 default: 926 return false; 927 } 928} 929 930void 931Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 932 const CallExpr *TheCall) { 933 for (NonNullAttr::iterator i = NonNull->begin(), e = NonNull->end(); 934 i != e; ++i) { 935 const Expr *ArgExpr = TheCall->getArg(*i); 936 if (ArgExpr->isNullPointerConstant(Context, 937 Expr::NPC_ValueDependentIsNotNull)) 938 Diag(TheCall->getCallee()->getLocStart(), diag::warn_null_arg) 939 << ArgExpr->getSourceRange(); 940 } 941} 942 943/// CheckPrintfArguments - Check calls to printf (and similar functions) for 944/// correct use of format strings. 945/// 946/// HasVAListArg - A predicate indicating whether the printf-like 947/// function is passed an explicit va_arg argument (e.g., vprintf) 948/// 949/// format_idx - The index into Args for the format string. 950/// 951/// Improper format strings to functions in the printf family can be 952/// the source of bizarre bugs and very serious security holes. A 953/// good source of information is available in the following paper 954/// (which includes additional references): 955/// 956/// FormatGuard: Automatic Protection From printf Format String 957/// Vulnerabilities, Proceedings of the 10th USENIX Security Symposium, 2001. 958/// 959/// TODO: 960/// Functionality implemented: 961/// 962/// We can statically check the following properties for string 963/// literal format strings for non v.*printf functions (where the 964/// arguments are passed directly): 965// 966/// (1) Are the number of format conversions equal to the number of 967/// data arguments? 968/// 969/// (2) Does each format conversion correctly match the type of the 970/// corresponding data argument? 971/// 972/// Moreover, for all printf functions we can: 973/// 974/// (3) Check for a missing format string (when not caught by type checking). 975/// 976/// (4) Check for no-operation flags; e.g. using "#" with format 977/// conversion 'c' (TODO) 978/// 979/// (5) Check the use of '%n', a major source of security holes. 980/// 981/// (6) Check for malformed format conversions that don't specify anything. 982/// 983/// (7) Check for empty format strings. e.g: printf(""); 984/// 985/// (8) Check that the format string is a wide literal. 986/// 987/// All of these checks can be done by parsing the format string. 988/// 989void 990Sema::CheckPrintfArguments(const CallExpr *TheCall, bool HasVAListArg, 991 unsigned format_idx, unsigned firstDataArg) { 992 const Expr *Fn = TheCall->getCallee(); 993 994 // The way the format attribute works in GCC, the implicit this argument 995 // of member functions is counted. However, it doesn't appear in our own 996 // lists, so decrement format_idx in that case. 997 if (isa<CXXMemberCallExpr>(TheCall)) { 998 // Catch a format attribute mistakenly referring to the object argument. 999 if (format_idx == 0) 1000 return; 1001 --format_idx; 1002 if(firstDataArg != 0) 1003 --firstDataArg; 1004 } 1005 1006 // CHECK: printf-like function is called with no format string. 1007 if (format_idx >= TheCall->getNumArgs()) { 1008 Diag(TheCall->getRParenLoc(), diag::warn_printf_missing_format_string) 1009 << Fn->getSourceRange(); 1010 return; 1011 } 1012 1013 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 1014 1015 // CHECK: format string is not a string literal. 1016 // 1017 // Dynamically generated format strings are difficult to 1018 // automatically vet at compile time. Requiring that format strings 1019 // are string literals: (1) permits the checking of format strings by 1020 // the compiler and thereby (2) can practically remove the source of 1021 // many format string exploits. 1022 1023 // Format string can be either ObjC string (e.g. @"%d") or 1024 // C string (e.g. "%d") 1025 // ObjC string uses the same format specifiers as C string, so we can use 1026 // the same format string checking logic for both ObjC and C strings. 1027 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 1028 firstDataArg)) 1029 return; // Literal format string found, check done! 1030 1031 // If there are no arguments specified, warn with -Wformat-security, otherwise 1032 // warn only with -Wformat-nonliteral. 1033 if (TheCall->getNumArgs() == format_idx+1) 1034 Diag(TheCall->getArg(format_idx)->getLocStart(), 1035 diag::warn_printf_nonliteral_noargs) 1036 << OrigFormatExpr->getSourceRange(); 1037 else 1038 Diag(TheCall->getArg(format_idx)->getLocStart(), 1039 diag::warn_printf_nonliteral) 1040 << OrigFormatExpr->getSourceRange(); 1041} 1042 1043namespace { 1044class CheckPrintfHandler : public analyze_printf::FormatStringHandler { 1045 Sema &S; 1046 const StringLiteral *FExpr; 1047 const Expr *OrigFormatExpr; 1048 const unsigned NumDataArgs; 1049 const bool IsObjCLiteral; 1050 const char *Beg; // Start of format string. 1051 const bool HasVAListArg; 1052 const CallExpr *TheCall; 1053 unsigned FormatIdx; 1054 llvm::BitVector CoveredArgs; 1055 bool usesPositionalArgs; 1056 bool atFirstArg; 1057public: 1058 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1059 const Expr *origFormatExpr, 1060 unsigned numDataArgs, bool isObjCLiteral, 1061 const char *beg, bool hasVAListArg, 1062 const CallExpr *theCall, unsigned formatIdx) 1063 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1064 NumDataArgs(numDataArgs), 1065 IsObjCLiteral(isObjCLiteral), Beg(beg), 1066 HasVAListArg(hasVAListArg), 1067 TheCall(theCall), FormatIdx(formatIdx), 1068 usesPositionalArgs(false), atFirstArg(true) { 1069 CoveredArgs.resize(numDataArgs); 1070 CoveredArgs.reset(); 1071 } 1072 1073 void DoneProcessing(); 1074 1075 void HandleIncompleteFormatSpecifier(const char *startSpecifier, 1076 unsigned specifierLen); 1077 1078 bool 1079 HandleInvalidConversionSpecifier(const analyze_printf::FormatSpecifier &FS, 1080 const char *startSpecifier, 1081 unsigned specifierLen); 1082 1083 virtual void HandleInvalidPosition(const char *startSpecifier, 1084 unsigned specifierLen, 1085 analyze_printf::PositionContext p); 1086 1087 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1088 1089 void HandleNullChar(const char *nullCharacter); 1090 1091 bool HandleFormatSpecifier(const analyze_printf::FormatSpecifier &FS, 1092 const char *startSpecifier, 1093 unsigned specifierLen); 1094private: 1095 SourceRange getFormatStringRange(); 1096 SourceRange getFormatSpecifierRange(const char *startSpecifier, 1097 unsigned specifierLen); 1098 SourceLocation getLocationOfByte(const char *x); 1099 1100 bool HandleAmount(const analyze_printf::OptionalAmount &Amt, unsigned k, 1101 const char *startSpecifier, unsigned specifierLen); 1102 void HandleFlags(const analyze_printf::FormatSpecifier &FS, 1103 llvm::StringRef flag, llvm::StringRef cspec, 1104 const char *startSpecifier, unsigned specifierLen); 1105 1106 const Expr *getDataArg(unsigned i) const; 1107}; 1108} 1109 1110SourceRange CheckPrintfHandler::getFormatStringRange() { 1111 return OrigFormatExpr->getSourceRange(); 1112} 1113 1114SourceRange CheckPrintfHandler:: 1115getFormatSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1116 return SourceRange(getLocationOfByte(startSpecifier), 1117 getLocationOfByte(startSpecifier+specifierLen-1)); 1118} 1119 1120SourceLocation CheckPrintfHandler::getLocationOfByte(const char *x) { 1121 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1122} 1123 1124void CheckPrintfHandler:: 1125HandleIncompleteFormatSpecifier(const char *startSpecifier, 1126 unsigned specifierLen) { 1127 SourceLocation Loc = getLocationOfByte(startSpecifier); 1128 S.Diag(Loc, diag::warn_printf_incomplete_specifier) 1129 << getFormatSpecifierRange(startSpecifier, specifierLen); 1130} 1131 1132void 1133CheckPrintfHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1134 analyze_printf::PositionContext p) { 1135 SourceLocation Loc = getLocationOfByte(startPos); 1136 S.Diag(Loc, diag::warn_printf_invalid_positional_specifier) 1137 << (unsigned) p << getFormatSpecifierRange(startPos, posLen); 1138} 1139 1140void CheckPrintfHandler::HandleZeroPosition(const char *startPos, 1141 unsigned posLen) { 1142 SourceLocation Loc = getLocationOfByte(startPos); 1143 S.Diag(Loc, diag::warn_printf_zero_positional_specifier) 1144 << getFormatSpecifierRange(startPos, posLen); 1145} 1146 1147bool CheckPrintfHandler:: 1148HandleInvalidConversionSpecifier(const analyze_printf::FormatSpecifier &FS, 1149 const char *startSpecifier, 1150 unsigned specifierLen) { 1151 1152 unsigned argIndex = FS.getArgIndex(); 1153 bool keepGoing = true; 1154 if (argIndex < NumDataArgs) { 1155 // Consider the argument coverered, even though the specifier doesn't 1156 // make sense. 1157 CoveredArgs.set(argIndex); 1158 } 1159 else { 1160 // If argIndex exceeds the number of data arguments we 1161 // don't issue a warning because that is just a cascade of warnings (and 1162 // they may have intended '%%' anyway). We don't want to continue processing 1163 // the format string after this point, however, as we will like just get 1164 // gibberish when trying to match arguments. 1165 keepGoing = false; 1166 } 1167 1168 const analyze_printf::ConversionSpecifier &CS = 1169 FS.getConversionSpecifier(); 1170 SourceLocation Loc = getLocationOfByte(CS.getStart()); 1171 S.Diag(Loc, diag::warn_printf_invalid_conversion) 1172 << llvm::StringRef(CS.getStart(), CS.getLength()) 1173 << getFormatSpecifierRange(startSpecifier, specifierLen); 1174 1175 return keepGoing; 1176} 1177 1178void CheckPrintfHandler::HandleNullChar(const char *nullCharacter) { 1179 // The presence of a null character is likely an error. 1180 S.Diag(getLocationOfByte(nullCharacter), 1181 diag::warn_printf_format_string_contains_null_char) 1182 << getFormatStringRange(); 1183} 1184 1185const Expr *CheckPrintfHandler::getDataArg(unsigned i) const { 1186 return TheCall->getArg(FormatIdx + i + 1); 1187} 1188 1189 1190 1191void CheckPrintfHandler::HandleFlags(const analyze_printf::FormatSpecifier &FS, 1192 llvm::StringRef flag, 1193 llvm::StringRef cspec, 1194 const char *startSpecifier, 1195 unsigned specifierLen) { 1196 const analyze_printf::ConversionSpecifier &CS = FS.getConversionSpecifier(); 1197 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_nonsensical_flag) 1198 << flag << cspec << getFormatSpecifierRange(startSpecifier, specifierLen); 1199} 1200 1201bool 1202CheckPrintfHandler::HandleAmount(const analyze_printf::OptionalAmount &Amt, 1203 unsigned k, const char *startSpecifier, 1204 unsigned specifierLen) { 1205 1206 if (Amt.hasDataArgument()) { 1207 if (!HasVAListArg) { 1208 unsigned argIndex = Amt.getArgIndex(); 1209 if (argIndex >= NumDataArgs) { 1210 S.Diag(getLocationOfByte(Amt.getStart()), 1211 diag::warn_printf_asterisk_missing_arg) 1212 << k << getFormatSpecifierRange(startSpecifier, specifierLen); 1213 // Don't do any more checking. We will just emit 1214 // spurious errors. 1215 return false; 1216 } 1217 1218 // Type check the data argument. It should be an 'int'. 1219 // Although not in conformance with C99, we also allow the argument to be 1220 // an 'unsigned int' as that is a reasonably safe case. GCC also 1221 // doesn't emit a warning for that case. 1222 CoveredArgs.set(argIndex); 1223 const Expr *Arg = getDataArg(argIndex); 1224 QualType T = Arg->getType(); 1225 1226 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1227 assert(ATR.isValid()); 1228 1229 if (!ATR.matchesType(S.Context, T)) { 1230 S.Diag(getLocationOfByte(Amt.getStart()), 1231 diag::warn_printf_asterisk_wrong_type) 1232 << k 1233 << ATR.getRepresentativeType(S.Context) << T 1234 << getFormatSpecifierRange(startSpecifier, specifierLen) 1235 << Arg->getSourceRange(); 1236 // Don't do any more checking. We will just emit 1237 // spurious errors. 1238 return false; 1239 } 1240 } 1241 } 1242 return true; 1243} 1244 1245bool 1246CheckPrintfHandler::HandleFormatSpecifier(const analyze_printf::FormatSpecifier 1247 &FS, 1248 const char *startSpecifier, 1249 unsigned specifierLen) { 1250 1251 using namespace analyze_printf; 1252 const ConversionSpecifier &CS = FS.getConversionSpecifier(); 1253 1254 if (atFirstArg) { 1255 atFirstArg = false; 1256 usesPositionalArgs = FS.usesPositionalArg(); 1257 } 1258 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1259 // Cannot mix-and-match positional and non-positional arguments. 1260 S.Diag(getLocationOfByte(CS.getStart()), 1261 diag::warn_printf_mix_positional_nonpositional_args) 1262 << getFormatSpecifierRange(startSpecifier, specifierLen); 1263 return false; 1264 } 1265 1266 // First check if the field width, precision, and conversion specifier 1267 // have matching data arguments. 1268 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 1269 startSpecifier, specifierLen)) { 1270 return false; 1271 } 1272 1273 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 1274 startSpecifier, specifierLen)) { 1275 return false; 1276 } 1277 1278 if (!CS.consumesDataArgument()) { 1279 // FIXME: Technically specifying a precision or field width here 1280 // makes no sense. Worth issuing a warning at some point. 1281 return true; 1282 } 1283 1284 // Consume the argument. 1285 unsigned argIndex = FS.getArgIndex(); 1286 CoveredArgs.set(argIndex); 1287 1288 // Check for using an Objective-C specific conversion specifier 1289 // in a non-ObjC literal. 1290 if (!IsObjCLiteral && CS.isObjCArg()) { 1291 return HandleInvalidConversionSpecifier(FS, startSpecifier, specifierLen); 1292 } 1293 1294 // Are we using '%n'? Issue a warning about this being 1295 // a possible security issue. 1296 if (CS.getKind() == ConversionSpecifier::OutIntPtrArg) { 1297 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back) 1298 << getFormatSpecifierRange(startSpecifier, specifierLen); 1299 // Continue checking the other format specifiers. 1300 return true; 1301 } 1302 1303 if (CS.getKind() == ConversionSpecifier::VoidPtrArg) { 1304 if (FS.getPrecision().getHowSpecified() != OptionalAmount::NotSpecified) 1305 S.Diag(getLocationOfByte(CS.getStart()), 1306 diag::warn_printf_nonsensical_precision) 1307 << CS.getCharacters() 1308 << getFormatSpecifierRange(startSpecifier, specifierLen); 1309 } 1310 if (CS.getKind() == ConversionSpecifier::VoidPtrArg || 1311 CS.getKind() == ConversionSpecifier::CStrArg) { 1312 // FIXME: Instead of using "0", "+", etc., eventually get them from 1313 // the FormatSpecifier. 1314 if (FS.hasLeadingZeros()) 1315 HandleFlags(FS, "0", CS.getCharacters(), startSpecifier, specifierLen); 1316 if (FS.hasPlusPrefix()) 1317 HandleFlags(FS, "+", CS.getCharacters(), startSpecifier, specifierLen); 1318 if (FS.hasSpacePrefix()) 1319 HandleFlags(FS, " ", CS.getCharacters(), startSpecifier, specifierLen); 1320 } 1321 1322 // The remaining checks depend on the data arguments. 1323 if (HasVAListArg) 1324 return true; 1325 1326 if (argIndex >= NumDataArgs) { 1327 S.Diag(getLocationOfByte(CS.getStart()), 1328 diag::warn_printf_insufficient_data_args) 1329 << getFormatSpecifierRange(startSpecifier, specifierLen); 1330 // Don't do any more checking. 1331 return false; 1332 } 1333 1334 // Now type check the data expression that matches the 1335 // format specifier. 1336 const Expr *Ex = getDataArg(argIndex); 1337 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context); 1338 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 1339 // Check if we didn't match because of an implicit cast from a 'char' 1340 // or 'short' to an 'int'. This is done because printf is a varargs 1341 // function. 1342 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 1343 if (ICE->getType() == S.Context.IntTy) 1344 if (ATR.matchesType(S.Context, ICE->getSubExpr()->getType())) 1345 return true; 1346 1347 S.Diag(getLocationOfByte(CS.getStart()), 1348 diag::warn_printf_conversion_argument_type_mismatch) 1349 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1350 << getFormatSpecifierRange(startSpecifier, specifierLen) 1351 << Ex->getSourceRange(); 1352 } 1353 1354 return true; 1355} 1356 1357void CheckPrintfHandler::DoneProcessing() { 1358 // Does the number of data arguments exceed the number of 1359 // format conversions in the format string? 1360 if (!HasVAListArg) { 1361 // Find any arguments that weren't covered. 1362 CoveredArgs.flip(); 1363 signed notCoveredArg = CoveredArgs.find_first(); 1364 if (notCoveredArg >= 0) { 1365 assert((unsigned)notCoveredArg < NumDataArgs); 1366 S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(), 1367 diag::warn_printf_data_arg_not_used) 1368 << getFormatStringRange(); 1369 } 1370 } 1371} 1372 1373void Sema::CheckPrintfString(const StringLiteral *FExpr, 1374 const Expr *OrigFormatExpr, 1375 const CallExpr *TheCall, bool HasVAListArg, 1376 unsigned format_idx, unsigned firstDataArg) { 1377 1378 // CHECK: is the format string a wide literal? 1379 if (FExpr->isWide()) { 1380 Diag(FExpr->getLocStart(), 1381 diag::warn_printf_format_string_is_wide_literal) 1382 << OrigFormatExpr->getSourceRange(); 1383 return; 1384 } 1385 1386 // Str - The format string. NOTE: this is NOT null-terminated! 1387 const char *Str = FExpr->getStrData(); 1388 1389 // CHECK: empty format string? 1390 unsigned StrLen = FExpr->getByteLength(); 1391 1392 if (StrLen == 0) { 1393 Diag(FExpr->getLocStart(), diag::warn_printf_empty_format_string) 1394 << OrigFormatExpr->getSourceRange(); 1395 return; 1396 } 1397 1398 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, 1399 TheCall->getNumArgs() - firstDataArg, 1400 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1401 HasVAListArg, TheCall, format_idx); 1402 1403 if (!analyze_printf::ParseFormatString(H, Str, Str + StrLen)) 1404 H.DoneProcessing(); 1405} 1406 1407//===--- CHECK: Return Address of Stack Variable --------------------------===// 1408 1409static DeclRefExpr* EvalVal(Expr *E); 1410static DeclRefExpr* EvalAddr(Expr* E); 1411 1412/// CheckReturnStackAddr - Check if a return statement returns the address 1413/// of a stack variable. 1414void 1415Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 1416 SourceLocation ReturnLoc) { 1417 1418 // Perform checking for returned stack addresses. 1419 if (lhsType->isPointerType() || lhsType->isBlockPointerType()) { 1420 if (DeclRefExpr *DR = EvalAddr(RetValExp)) 1421 Diag(DR->getLocStart(), diag::warn_ret_stack_addr) 1422 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1423 1424 // Skip over implicit cast expressions when checking for block expressions. 1425 RetValExp = RetValExp->IgnoreParenCasts(); 1426 1427 if (BlockExpr *C = dyn_cast<BlockExpr>(RetValExp)) 1428 if (C->hasBlockDeclRefExprs()) 1429 Diag(C->getLocStart(), diag::err_ret_local_block) 1430 << C->getSourceRange(); 1431 1432 if (AddrLabelExpr *ALE = dyn_cast<AddrLabelExpr>(RetValExp)) 1433 Diag(ALE->getLocStart(), diag::warn_ret_addr_label) 1434 << ALE->getSourceRange(); 1435 1436 } else if (lhsType->isReferenceType()) { 1437 // Perform checking for stack values returned by reference. 1438 // Check for a reference to the stack 1439 if (DeclRefExpr *DR = EvalVal(RetValExp)) 1440 Diag(DR->getLocStart(), diag::warn_ret_stack_ref) 1441 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1442 } 1443} 1444 1445/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 1446/// check if the expression in a return statement evaluates to an address 1447/// to a location on the stack. The recursion is used to traverse the 1448/// AST of the return expression, with recursion backtracking when we 1449/// encounter a subexpression that (1) clearly does not lead to the address 1450/// of a stack variable or (2) is something we cannot determine leads to 1451/// the address of a stack variable based on such local checking. 1452/// 1453/// EvalAddr processes expressions that are pointers that are used as 1454/// references (and not L-values). EvalVal handles all other values. 1455/// At the base case of the recursion is a check for a DeclRefExpr* in 1456/// the refers to a stack variable. 1457/// 1458/// This implementation handles: 1459/// 1460/// * pointer-to-pointer casts 1461/// * implicit conversions from array references to pointers 1462/// * taking the address of fields 1463/// * arbitrary interplay between "&" and "*" operators 1464/// * pointer arithmetic from an address of a stack variable 1465/// * taking the address of an array element where the array is on the stack 1466static DeclRefExpr* EvalAddr(Expr *E) { 1467 // We should only be called for evaluating pointer expressions. 1468 assert((E->getType()->isAnyPointerType() || 1469 E->getType()->isBlockPointerType() || 1470 E->getType()->isObjCQualifiedIdType()) && 1471 "EvalAddr only works on pointers"); 1472 1473 // Our "symbolic interpreter" is just a dispatch off the currently 1474 // viewed AST node. We then recursively traverse the AST by calling 1475 // EvalAddr and EvalVal appropriately. 1476 switch (E->getStmtClass()) { 1477 case Stmt::ParenExprClass: 1478 // Ignore parentheses. 1479 return EvalAddr(cast<ParenExpr>(E)->getSubExpr()); 1480 1481 case Stmt::UnaryOperatorClass: { 1482 // The only unary operator that make sense to handle here 1483 // is AddrOf. All others don't make sense as pointers. 1484 UnaryOperator *U = cast<UnaryOperator>(E); 1485 1486 if (U->getOpcode() == UnaryOperator::AddrOf) 1487 return EvalVal(U->getSubExpr()); 1488 else 1489 return NULL; 1490 } 1491 1492 case Stmt::BinaryOperatorClass: { 1493 // Handle pointer arithmetic. All other binary operators are not valid 1494 // in this context. 1495 BinaryOperator *B = cast<BinaryOperator>(E); 1496 BinaryOperator::Opcode op = B->getOpcode(); 1497 1498 if (op != BinaryOperator::Add && op != BinaryOperator::Sub) 1499 return NULL; 1500 1501 Expr *Base = B->getLHS(); 1502 1503 // Determine which argument is the real pointer base. It could be 1504 // the RHS argument instead of the LHS. 1505 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 1506 1507 assert (Base->getType()->isPointerType()); 1508 return EvalAddr(Base); 1509 } 1510 1511 // For conditional operators we need to see if either the LHS or RHS are 1512 // valid DeclRefExpr*s. If one of them is valid, we return it. 1513 case Stmt::ConditionalOperatorClass: { 1514 ConditionalOperator *C = cast<ConditionalOperator>(E); 1515 1516 // Handle the GNU extension for missing LHS. 1517 if (Expr *lhsExpr = C->getLHS()) 1518 if (DeclRefExpr* LHS = EvalAddr(lhsExpr)) 1519 return LHS; 1520 1521 return EvalAddr(C->getRHS()); 1522 } 1523 1524 // For casts, we need to handle conversions from arrays to 1525 // pointer values, and pointer-to-pointer conversions. 1526 case Stmt::ImplicitCastExprClass: 1527 case Stmt::CStyleCastExprClass: 1528 case Stmt::CXXFunctionalCastExprClass: { 1529 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 1530 QualType T = SubExpr->getType(); 1531 1532 if (SubExpr->getType()->isPointerType() || 1533 SubExpr->getType()->isBlockPointerType() || 1534 SubExpr->getType()->isObjCQualifiedIdType()) 1535 return EvalAddr(SubExpr); 1536 else if (T->isArrayType()) 1537 return EvalVal(SubExpr); 1538 else 1539 return 0; 1540 } 1541 1542 // C++ casts. For dynamic casts, static casts, and const casts, we 1543 // are always converting from a pointer-to-pointer, so we just blow 1544 // through the cast. In the case the dynamic cast doesn't fail (and 1545 // return NULL), we take the conservative route and report cases 1546 // where we return the address of a stack variable. For Reinterpre 1547 // FIXME: The comment about is wrong; we're not always converting 1548 // from pointer to pointer. I'm guessing that this code should also 1549 // handle references to objects. 1550 case Stmt::CXXStaticCastExprClass: 1551 case Stmt::CXXDynamicCastExprClass: 1552 case Stmt::CXXConstCastExprClass: 1553 case Stmt::CXXReinterpretCastExprClass: { 1554 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 1555 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 1556 return EvalAddr(S); 1557 else 1558 return NULL; 1559 } 1560 1561 // Everything else: we simply don't reason about them. 1562 default: 1563 return NULL; 1564 } 1565} 1566 1567 1568/// EvalVal - This function is complements EvalAddr in the mutual recursion. 1569/// See the comments for EvalAddr for more details. 1570static DeclRefExpr* EvalVal(Expr *E) { 1571 1572 // We should only be called for evaluating non-pointer expressions, or 1573 // expressions with a pointer type that are not used as references but instead 1574 // are l-values (e.g., DeclRefExpr with a pointer type). 1575 1576 // Our "symbolic interpreter" is just a dispatch off the currently 1577 // viewed AST node. We then recursively traverse the AST by calling 1578 // EvalAddr and EvalVal appropriately. 1579 switch (E->getStmtClass()) { 1580 case Stmt::DeclRefExprClass: { 1581 // DeclRefExpr: the base case. When we hit a DeclRefExpr we are looking 1582 // at code that refers to a variable's name. We check if it has local 1583 // storage within the function, and if so, return the expression. 1584 DeclRefExpr *DR = cast<DeclRefExpr>(E); 1585 1586 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 1587 if (V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR; 1588 1589 return NULL; 1590 } 1591 1592 case Stmt::ParenExprClass: 1593 // Ignore parentheses. 1594 return EvalVal(cast<ParenExpr>(E)->getSubExpr()); 1595 1596 case Stmt::UnaryOperatorClass: { 1597 // The only unary operator that make sense to handle here 1598 // is Deref. All others don't resolve to a "name." This includes 1599 // handling all sorts of rvalues passed to a unary operator. 1600 UnaryOperator *U = cast<UnaryOperator>(E); 1601 1602 if (U->getOpcode() == UnaryOperator::Deref) 1603 return EvalAddr(U->getSubExpr()); 1604 1605 return NULL; 1606 } 1607 1608 case Stmt::ArraySubscriptExprClass: { 1609 // Array subscripts are potential references to data on the stack. We 1610 // retrieve the DeclRefExpr* for the array variable if it indeed 1611 // has local storage. 1612 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase()); 1613 } 1614 1615 case Stmt::ConditionalOperatorClass: { 1616 // For conditional operators we need to see if either the LHS or RHS are 1617 // non-NULL DeclRefExpr's. If one is non-NULL, we return it. 1618 ConditionalOperator *C = cast<ConditionalOperator>(E); 1619 1620 // Handle the GNU extension for missing LHS. 1621 if (Expr *lhsExpr = C->getLHS()) 1622 if (DeclRefExpr *LHS = EvalVal(lhsExpr)) 1623 return LHS; 1624 1625 return EvalVal(C->getRHS()); 1626 } 1627 1628 // Accesses to members are potential references to data on the stack. 1629 case Stmt::MemberExprClass: { 1630 MemberExpr *M = cast<MemberExpr>(E); 1631 1632 // Check for indirect access. We only want direct field accesses. 1633 if (!M->isArrow()) 1634 return EvalVal(M->getBase()); 1635 else 1636 return NULL; 1637 } 1638 1639 // Everything else: we simply don't reason about them. 1640 default: 1641 return NULL; 1642 } 1643} 1644 1645//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 1646 1647/// Check for comparisons of floating point operands using != and ==. 1648/// Issue a warning if these are no self-comparisons, as they are not likely 1649/// to do what the programmer intended. 1650void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 1651 bool EmitWarning = true; 1652 1653 Expr* LeftExprSansParen = lex->IgnoreParens(); 1654 Expr* RightExprSansParen = rex->IgnoreParens(); 1655 1656 // Special case: check for x == x (which is OK). 1657 // Do not emit warnings for such cases. 1658 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 1659 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 1660 if (DRL->getDecl() == DRR->getDecl()) 1661 EmitWarning = false; 1662 1663 1664 // Special case: check for comparisons against literals that can be exactly 1665 // represented by APFloat. In such cases, do not emit a warning. This 1666 // is a heuristic: often comparison against such literals are used to 1667 // detect if a value in a variable has not changed. This clearly can 1668 // lead to false negatives. 1669 if (EmitWarning) { 1670 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 1671 if (FLL->isExact()) 1672 EmitWarning = false; 1673 } else 1674 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 1675 if (FLR->isExact()) 1676 EmitWarning = false; 1677 } 1678 } 1679 1680 // Check for comparisons with builtin types. 1681 if (EmitWarning) 1682 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 1683 if (CL->isBuiltinCall(Context)) 1684 EmitWarning = false; 1685 1686 if (EmitWarning) 1687 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 1688 if (CR->isBuiltinCall(Context)) 1689 EmitWarning = false; 1690 1691 // Emit the diagnostic. 1692 if (EmitWarning) 1693 Diag(loc, diag::warn_floatingpoint_eq) 1694 << lex->getSourceRange() << rex->getSourceRange(); 1695} 1696 1697//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 1698//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 1699 1700namespace { 1701 1702/// Structure recording the 'active' range of an integer-valued 1703/// expression. 1704struct IntRange { 1705 /// The number of bits active in the int. 1706 unsigned Width; 1707 1708 /// True if the int is known not to have negative values. 1709 bool NonNegative; 1710 1711 IntRange() {} 1712 IntRange(unsigned Width, bool NonNegative) 1713 : Width(Width), NonNegative(NonNegative) 1714 {} 1715 1716 // Returns the range of the bool type. 1717 static IntRange forBoolType() { 1718 return IntRange(1, true); 1719 } 1720 1721 // Returns the range of an integral type. 1722 static IntRange forType(ASTContext &C, QualType T) { 1723 return forCanonicalType(C, T->getCanonicalTypeInternal().getTypePtr()); 1724 } 1725 1726 // Returns the range of an integeral type based on its canonical 1727 // representation. 1728 static IntRange forCanonicalType(ASTContext &C, const Type *T) { 1729 assert(T->isCanonicalUnqualified()); 1730 1731 if (const VectorType *VT = dyn_cast<VectorType>(T)) 1732 T = VT->getElementType().getTypePtr(); 1733 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 1734 T = CT->getElementType().getTypePtr(); 1735 if (const EnumType *ET = dyn_cast<EnumType>(T)) 1736 T = ET->getDecl()->getIntegerType().getTypePtr(); 1737 1738 const BuiltinType *BT = cast<BuiltinType>(T); 1739 assert(BT->isInteger()); 1740 1741 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 1742 } 1743 1744 // Returns the supremum of two ranges: i.e. their conservative merge. 1745 static IntRange join(IntRange L, IntRange R) { 1746 return IntRange(std::max(L.Width, R.Width), 1747 L.NonNegative && R.NonNegative); 1748 } 1749 1750 // Returns the infinum of two ranges: i.e. their aggressive merge. 1751 static IntRange meet(IntRange L, IntRange R) { 1752 return IntRange(std::min(L.Width, R.Width), 1753 L.NonNegative || R.NonNegative); 1754 } 1755}; 1756 1757IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 1758 if (value.isSigned() && value.isNegative()) 1759 return IntRange(value.getMinSignedBits(), false); 1760 1761 if (value.getBitWidth() > MaxWidth) 1762 value.trunc(MaxWidth); 1763 1764 // isNonNegative() just checks the sign bit without considering 1765 // signedness. 1766 return IntRange(value.getActiveBits(), true); 1767} 1768 1769IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 1770 unsigned MaxWidth) { 1771 if (result.isInt()) 1772 return GetValueRange(C, result.getInt(), MaxWidth); 1773 1774 if (result.isVector()) { 1775 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 1776 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 1777 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 1778 R = IntRange::join(R, El); 1779 } 1780 return R; 1781 } 1782 1783 if (result.isComplexInt()) { 1784 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 1785 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 1786 return IntRange::join(R, I); 1787 } 1788 1789 // This can happen with lossless casts to intptr_t of "based" lvalues. 1790 // Assume it might use arbitrary bits. 1791 // FIXME: The only reason we need to pass the type in here is to get 1792 // the sign right on this one case. It would be nice if APValue 1793 // preserved this. 1794 assert(result.isLValue()); 1795 return IntRange(MaxWidth, Ty->isUnsignedIntegerType()); 1796} 1797 1798/// Pseudo-evaluate the given integer expression, estimating the 1799/// range of values it might take. 1800/// 1801/// \param MaxWidth - the width to which the value will be truncated 1802IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 1803 E = E->IgnoreParens(); 1804 1805 // Try a full evaluation first. 1806 Expr::EvalResult result; 1807 if (E->Evaluate(result, C)) 1808 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 1809 1810 // I think we only want to look through implicit casts here; if the 1811 // user has an explicit widening cast, we should treat the value as 1812 // being of the new, wider type. 1813 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 1814 if (CE->getCastKind() == CastExpr::CK_NoOp) 1815 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 1816 1817 IntRange OutputTypeRange = IntRange::forType(C, CE->getType()); 1818 1819 bool isIntegerCast = (CE->getCastKind() == CastExpr::CK_IntegralCast); 1820 if (!isIntegerCast && CE->getCastKind() == CastExpr::CK_Unknown) 1821 isIntegerCast = CE->getSubExpr()->getType()->isIntegerType(); 1822 1823 // Assume that non-integer casts can span the full range of the type. 1824 if (!isIntegerCast) 1825 return OutputTypeRange; 1826 1827 IntRange SubRange 1828 = GetExprRange(C, CE->getSubExpr(), 1829 std::min(MaxWidth, OutputTypeRange.Width)); 1830 1831 // Bail out if the subexpr's range is as wide as the cast type. 1832 if (SubRange.Width >= OutputTypeRange.Width) 1833 return OutputTypeRange; 1834 1835 // Otherwise, we take the smaller width, and we're non-negative if 1836 // either the output type or the subexpr is. 1837 return IntRange(SubRange.Width, 1838 SubRange.NonNegative || OutputTypeRange.NonNegative); 1839 } 1840 1841 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 1842 // If we can fold the condition, just take that operand. 1843 bool CondResult; 1844 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 1845 return GetExprRange(C, CondResult ? CO->getTrueExpr() 1846 : CO->getFalseExpr(), 1847 MaxWidth); 1848 1849 // Otherwise, conservatively merge. 1850 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 1851 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 1852 return IntRange::join(L, R); 1853 } 1854 1855 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 1856 switch (BO->getOpcode()) { 1857 1858 // Boolean-valued operations are single-bit and positive. 1859 case BinaryOperator::LAnd: 1860 case BinaryOperator::LOr: 1861 case BinaryOperator::LT: 1862 case BinaryOperator::GT: 1863 case BinaryOperator::LE: 1864 case BinaryOperator::GE: 1865 case BinaryOperator::EQ: 1866 case BinaryOperator::NE: 1867 return IntRange::forBoolType(); 1868 1869 // The type of these compound assignments is the type of the LHS, 1870 // so the RHS is not necessarily an integer. 1871 case BinaryOperator::MulAssign: 1872 case BinaryOperator::DivAssign: 1873 case BinaryOperator::RemAssign: 1874 case BinaryOperator::AddAssign: 1875 case BinaryOperator::SubAssign: 1876 return IntRange::forType(C, E->getType()); 1877 1878 // Operations with opaque sources are black-listed. 1879 case BinaryOperator::PtrMemD: 1880 case BinaryOperator::PtrMemI: 1881 return IntRange::forType(C, E->getType()); 1882 1883 // Bitwise-and uses the *infinum* of the two source ranges. 1884 case BinaryOperator::And: 1885 case BinaryOperator::AndAssign: 1886 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 1887 GetExprRange(C, BO->getRHS(), MaxWidth)); 1888 1889 // Left shift gets black-listed based on a judgement call. 1890 case BinaryOperator::Shl: 1891 case BinaryOperator::ShlAssign: 1892 return IntRange::forType(C, E->getType()); 1893 1894 // Right shift by a constant can narrow its left argument. 1895 case BinaryOperator::Shr: 1896 case BinaryOperator::ShrAssign: { 1897 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 1898 1899 // If the shift amount is a positive constant, drop the width by 1900 // that much. 1901 llvm::APSInt shift; 1902 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 1903 shift.isNonNegative()) { 1904 unsigned zext = shift.getZExtValue(); 1905 if (zext >= L.Width) 1906 L.Width = (L.NonNegative ? 0 : 1); 1907 else 1908 L.Width -= zext; 1909 } 1910 1911 return L; 1912 } 1913 1914 // Comma acts as its right operand. 1915 case BinaryOperator::Comma: 1916 return GetExprRange(C, BO->getRHS(), MaxWidth); 1917 1918 // Black-list pointer subtractions. 1919 case BinaryOperator::Sub: 1920 if (BO->getLHS()->getType()->isPointerType()) 1921 return IntRange::forType(C, E->getType()); 1922 // fallthrough 1923 1924 default: 1925 break; 1926 } 1927 1928 // Treat every other operator as if it were closed on the 1929 // narrowest type that encompasses both operands. 1930 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 1931 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 1932 return IntRange::join(L, R); 1933 } 1934 1935 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 1936 switch (UO->getOpcode()) { 1937 // Boolean-valued operations are white-listed. 1938 case UnaryOperator::LNot: 1939 return IntRange::forBoolType(); 1940 1941 // Operations with opaque sources are black-listed. 1942 case UnaryOperator::Deref: 1943 case UnaryOperator::AddrOf: // should be impossible 1944 case UnaryOperator::OffsetOf: 1945 return IntRange::forType(C, E->getType()); 1946 1947 default: 1948 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 1949 } 1950 } 1951 1952 FieldDecl *BitField = E->getBitField(); 1953 if (BitField) { 1954 llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C); 1955 unsigned BitWidth = BitWidthAP.getZExtValue(); 1956 1957 return IntRange(BitWidth, BitField->getType()->isUnsignedIntegerType()); 1958 } 1959 1960 return IntRange::forType(C, E->getType()); 1961} 1962 1963/// Checks whether the given value, which currently has the given 1964/// source semantics, has the same value when coerced through the 1965/// target semantics. 1966bool IsSameFloatAfterCast(const llvm::APFloat &value, 1967 const llvm::fltSemantics &Src, 1968 const llvm::fltSemantics &Tgt) { 1969 llvm::APFloat truncated = value; 1970 1971 bool ignored; 1972 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 1973 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 1974 1975 return truncated.bitwiseIsEqual(value); 1976} 1977 1978/// Checks whether the given value, which currently has the given 1979/// source semantics, has the same value when coerced through the 1980/// target semantics. 1981/// 1982/// The value might be a vector of floats (or a complex number). 1983bool IsSameFloatAfterCast(const APValue &value, 1984 const llvm::fltSemantics &Src, 1985 const llvm::fltSemantics &Tgt) { 1986 if (value.isFloat()) 1987 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 1988 1989 if (value.isVector()) { 1990 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 1991 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 1992 return false; 1993 return true; 1994 } 1995 1996 assert(value.isComplexFloat()); 1997 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 1998 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 1999} 2000 2001} // end anonymous namespace 2002 2003/// \brief Implements -Wsign-compare. 2004/// 2005/// \param lex the left-hand expression 2006/// \param rex the right-hand expression 2007/// \param OpLoc the location of the joining operator 2008/// \param Equality whether this is an "equality-like" join, which 2009/// suppresses the warning in some cases 2010void Sema::CheckSignCompare(Expr *lex, Expr *rex, SourceLocation OpLoc, 2011 const PartialDiagnostic &PD, bool Equality) { 2012 // Don't warn if we're in an unevaluated context. 2013 if (ExprEvalContexts.back().Context == Unevaluated) 2014 return; 2015 2016 // If either expression is value-dependent, don't warn. We'll get another 2017 // chance at instantiation time. 2018 if (lex->isValueDependent() || rex->isValueDependent()) 2019 return; 2020 2021 QualType lt = lex->getType(), rt = rex->getType(); 2022 2023 // Only warn if both operands are integral. 2024 if (!lt->isIntegerType() || !rt->isIntegerType()) 2025 return; 2026 2027 // In C, the width of a bitfield determines its type, and the 2028 // declared type only contributes the signedness. This duplicates 2029 // the work that will later be done by UsualUnaryConversions. 2030 // Eventually, this check will be reorganized in a way that avoids 2031 // this duplication. 2032 if (!getLangOptions().CPlusPlus) { 2033 QualType tmp; 2034 tmp = Context.isPromotableBitField(lex); 2035 if (!tmp.isNull()) lt = tmp; 2036 tmp = Context.isPromotableBitField(rex); 2037 if (!tmp.isNull()) rt = tmp; 2038 } 2039 2040 // The rule is that the signed operand becomes unsigned, so isolate the 2041 // signed operand. 2042 Expr *signedOperand = lex, *unsignedOperand = rex; 2043 QualType signedType = lt, unsignedType = rt; 2044 if (lt->isSignedIntegerType()) { 2045 if (rt->isSignedIntegerType()) return; 2046 } else { 2047 if (!rt->isSignedIntegerType()) return; 2048 std::swap(signedOperand, unsignedOperand); 2049 std::swap(signedType, unsignedType); 2050 } 2051 2052 unsigned unsignedWidth = Context.getIntWidth(unsignedType); 2053 unsigned signedWidth = Context.getIntWidth(signedType); 2054 2055 // If the unsigned type is strictly smaller than the signed type, 2056 // then (1) the result type will be signed and (2) the unsigned 2057 // value will fit fully within the signed type, and thus the result 2058 // of the comparison will be exact. 2059 if (signedWidth > unsignedWidth) 2060 return; 2061 2062 // Otherwise, calculate the effective ranges. 2063 IntRange signedRange = GetExprRange(Context, signedOperand, signedWidth); 2064 IntRange unsignedRange = GetExprRange(Context, unsignedOperand, unsignedWidth); 2065 2066 // We should never be unable to prove that the unsigned operand is 2067 // non-negative. 2068 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 2069 2070 // If the signed operand is non-negative, then the signed->unsigned 2071 // conversion won't change it. 2072 if (signedRange.NonNegative) 2073 return; 2074 2075 // For (in)equality comparisons, if the unsigned operand is a 2076 // constant which cannot collide with a overflowed signed operand, 2077 // then reinterpreting the signed operand as unsigned will not 2078 // change the result of the comparison. 2079 if (Equality && unsignedRange.Width < unsignedWidth) 2080 return; 2081 2082 Diag(OpLoc, PD) 2083 << lt << rt << lex->getSourceRange() << rex->getSourceRange(); 2084} 2085 2086/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 2087static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, unsigned diag) { 2088 S.Diag(E->getExprLoc(), diag) << E->getType() << T << E->getSourceRange(); 2089} 2090 2091/// Implements -Wconversion. 2092void Sema::CheckImplicitConversion(Expr *E, QualType T) { 2093 // Don't diagnose in unevaluated contexts. 2094 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 2095 return; 2096 2097 // Don't diagnose for value-dependent expressions. 2098 if (E->isValueDependent()) 2099 return; 2100 2101 const Type *Source = Context.getCanonicalType(E->getType()).getTypePtr(); 2102 const Type *Target = Context.getCanonicalType(T).getTypePtr(); 2103 2104 // Never diagnose implicit casts to bool. 2105 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 2106 return; 2107 2108 // Strip vector types. 2109 if (isa<VectorType>(Source)) { 2110 if (!isa<VectorType>(Target)) 2111 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_vector_scalar); 2112 2113 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 2114 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 2115 } 2116 2117 // Strip complex types. 2118 if (isa<ComplexType>(Source)) { 2119 if (!isa<ComplexType>(Target)) 2120 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_complex_scalar); 2121 2122 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 2123 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 2124 } 2125 2126 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 2127 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 2128 2129 // If the source is floating point... 2130 if (SourceBT && SourceBT->isFloatingPoint()) { 2131 // ...and the target is floating point... 2132 if (TargetBT && TargetBT->isFloatingPoint()) { 2133 // ...then warn if we're dropping FP rank. 2134 2135 // Builtin FP kinds are ordered by increasing FP rank. 2136 if (SourceBT->getKind() > TargetBT->getKind()) { 2137 // Don't warn about float constants that are precisely 2138 // representable in the target type. 2139 Expr::EvalResult result; 2140 if (E->Evaluate(result, Context)) { 2141 // Value might be a float, a float vector, or a float complex. 2142 if (IsSameFloatAfterCast(result.Val, 2143 Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 2144 Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 2145 return; 2146 } 2147 2148 DiagnoseImpCast(*this, E, T, diag::warn_impcast_float_precision); 2149 } 2150 return; 2151 } 2152 2153 // If the target is integral, always warn. 2154 if ((TargetBT && TargetBT->isInteger())) 2155 // TODO: don't warn for integer values? 2156 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_float_integer); 2157 2158 return; 2159 } 2160 2161 if (!Source->isIntegerType() || !Target->isIntegerType()) 2162 return; 2163 2164 IntRange SourceRange = GetExprRange(Context, E, Context.getIntWidth(E->getType())); 2165 IntRange TargetRange = IntRange::forCanonicalType(Context, Target); 2166 2167 // FIXME: also signed<->unsigned? 2168 2169 if (SourceRange.Width > TargetRange.Width) { 2170 // People want to build with -Wshorten-64-to-32 and not -Wconversion 2171 // and by god we'll let them. 2172 if (SourceRange.Width == 64 && TargetRange.Width == 32) 2173 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_integer_64_32); 2174 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_integer_precision); 2175 } 2176 2177 return; 2178} 2179 2180 2181 2182namespace { 2183class UnreachableCodeHandler : public reachable_code::Callback { 2184 Sema &S; 2185public: 2186 UnreachableCodeHandler(Sema *s) : S(*s) {} 2187 2188 void HandleUnreachable(SourceLocation L, SourceRange R1, SourceRange R2) { 2189 S.Diag(L, diag::warn_unreachable) << R1 << R2; 2190 } 2191}; 2192} 2193 2194/// CheckUnreachable - Check for unreachable code. 2195void Sema::CheckUnreachable(AnalysisContext &AC) { 2196 // We avoid checking when there are errors, as the CFG won't faithfully match 2197 // the user's code. 2198 if (getDiagnostics().hasErrorOccurred() || 2199 Diags.getDiagnosticLevel(diag::warn_unreachable) == Diagnostic::Ignored) 2200 return; 2201 2202 UnreachableCodeHandler UC(this); 2203 reachable_code::FindUnreachableCode(AC, UC); 2204} 2205 2206/// CheckFallThrough - Check that we don't fall off the end of a 2207/// Statement that should return a value. 2208/// 2209/// \returns AlwaysFallThrough iff we always fall off the end of the statement, 2210/// MaybeFallThrough iff we might or might not fall off the end, 2211/// NeverFallThroughOrReturn iff we never fall off the end of the statement or 2212/// return. We assume NeverFallThrough iff we never fall off the end of the 2213/// statement but we may return. We assume that functions not marked noreturn 2214/// will return. 2215Sema::ControlFlowKind Sema::CheckFallThrough(AnalysisContext &AC) { 2216 CFG *cfg = AC.getCFG(); 2217 if (cfg == 0) 2218 // FIXME: This should be NeverFallThrough 2219 return NeverFallThroughOrReturn; 2220 2221 // The CFG leaves in dead things, and we don't want the dead code paths to 2222 // confuse us, so we mark all live things first. 2223 std::queue<CFGBlock*> workq; 2224 llvm::BitVector live(cfg->getNumBlockIDs()); 2225 unsigned count = reachable_code::ScanReachableFromBlock(cfg->getEntry(), 2226 live); 2227 2228 bool AddEHEdges = AC.getAddEHEdges(); 2229 if (!AddEHEdges && count != cfg->getNumBlockIDs()) 2230 // When there are things remaining dead, and we didn't add EH edges 2231 // from CallExprs to the catch clauses, we have to go back and 2232 // mark them as live. 2233 for (CFG::iterator I = cfg->begin(), E = cfg->end(); I != E; ++I) { 2234 CFGBlock &b = **I; 2235 if (!live[b.getBlockID()]) { 2236 if (b.pred_begin() == b.pred_end()) { 2237 if (b.getTerminator() && isa<CXXTryStmt>(b.getTerminator())) 2238 // When not adding EH edges from calls, catch clauses 2239 // can otherwise seem dead. Avoid noting them as dead. 2240 count += reachable_code::ScanReachableFromBlock(b, live); 2241 continue; 2242 } 2243 } 2244 } 2245 2246 // Now we know what is live, we check the live precessors of the exit block 2247 // and look for fall through paths, being careful to ignore normal returns, 2248 // and exceptional paths. 2249 bool HasLiveReturn = false; 2250 bool HasFakeEdge = false; 2251 bool HasPlainEdge = false; 2252 bool HasAbnormalEdge = false; 2253 for (CFGBlock::pred_iterator I=cfg->getExit().pred_begin(), 2254 E = cfg->getExit().pred_end(); 2255 I != E; 2256 ++I) { 2257 CFGBlock& B = **I; 2258 if (!live[B.getBlockID()]) 2259 continue; 2260 if (B.size() == 0) { 2261 if (B.getTerminator() && isa<CXXTryStmt>(B.getTerminator())) { 2262 HasAbnormalEdge = true; 2263 continue; 2264 } 2265 2266 // A labeled empty statement, or the entry block... 2267 HasPlainEdge = true; 2268 continue; 2269 } 2270 Stmt *S = B[B.size()-1]; 2271 if (isa<ReturnStmt>(S)) { 2272 HasLiveReturn = true; 2273 continue; 2274 } 2275 if (isa<ObjCAtThrowStmt>(S)) { 2276 HasFakeEdge = true; 2277 continue; 2278 } 2279 if (isa<CXXThrowExpr>(S)) { 2280 HasFakeEdge = true; 2281 continue; 2282 } 2283 if (const AsmStmt *AS = dyn_cast<AsmStmt>(S)) { 2284 if (AS->isMSAsm()) { 2285 HasFakeEdge = true; 2286 HasLiveReturn = true; 2287 continue; 2288 } 2289 } 2290 if (isa<CXXTryStmt>(S)) { 2291 HasAbnormalEdge = true; 2292 continue; 2293 } 2294 2295 bool NoReturnEdge = false; 2296 if (CallExpr *C = dyn_cast<CallExpr>(S)) { 2297 if (B.succ_begin()[0] != &cfg->getExit()) { 2298 HasAbnormalEdge = true; 2299 continue; 2300 } 2301 Expr *CEE = C->getCallee()->IgnoreParenCasts(); 2302 if (CEE->getType().getNoReturnAttr()) { 2303 NoReturnEdge = true; 2304 HasFakeEdge = true; 2305 } else if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(CEE)) { 2306 ValueDecl *VD = DRE->getDecl(); 2307 if (VD->hasAttr<NoReturnAttr>()) { 2308 NoReturnEdge = true; 2309 HasFakeEdge = true; 2310 } 2311 } 2312 } 2313 // FIXME: Add noreturn message sends. 2314 if (NoReturnEdge == false) 2315 HasPlainEdge = true; 2316 } 2317 if (!HasPlainEdge) { 2318 if (HasLiveReturn) 2319 return NeverFallThrough; 2320 return NeverFallThroughOrReturn; 2321 } 2322 if (HasAbnormalEdge || HasFakeEdge || HasLiveReturn) 2323 return MaybeFallThrough; 2324 // This says AlwaysFallThrough for calls to functions that are not marked 2325 // noreturn, that don't return. If people would like this warning to be more 2326 // accurate, such functions should be marked as noreturn. 2327 return AlwaysFallThrough; 2328} 2329 2330/// CheckFallThroughForFunctionDef - Check that we don't fall off the end of a 2331/// function that should return a value. Check that we don't fall off the end 2332/// of a noreturn function. We assume that functions and blocks not marked 2333/// noreturn will return. 2334void Sema::CheckFallThroughForFunctionDef(Decl *D, Stmt *Body, 2335 AnalysisContext &AC) { 2336 // FIXME: Would be nice if we had a better way to control cascading errors, 2337 // but for now, avoid them. The problem is that when Parse sees: 2338 // int foo() { return a; } 2339 // The return is eaten and the Sema code sees just: 2340 // int foo() { } 2341 // which this code would then warn about. 2342 if (getDiagnostics().hasErrorOccurred()) 2343 return; 2344 2345 bool ReturnsVoid = false; 2346 bool HasNoReturn = false; 2347 2348 if (FunctionDecl *FD = dyn_cast<FunctionDecl>(D)) { 2349 // For function templates, class templates and member function templates 2350 // we'll do the analysis at instantiation time. 2351 if (FD->isDependentContext()) 2352 return; 2353 2354 ReturnsVoid = FD->getResultType()->isVoidType(); 2355 HasNoReturn = FD->hasAttr<NoReturnAttr>() || 2356 FD->getType()->getAs<FunctionType>()->getNoReturnAttr(); 2357 2358 } else if (ObjCMethodDecl *MD = dyn_cast<ObjCMethodDecl>(D)) { 2359 ReturnsVoid = MD->getResultType()->isVoidType(); 2360 HasNoReturn = MD->hasAttr<NoReturnAttr>(); 2361 } 2362 2363 // Short circuit for compilation speed. 2364 if ((Diags.getDiagnosticLevel(diag::warn_maybe_falloff_nonvoid_function) 2365 == Diagnostic::Ignored || ReturnsVoid) 2366 && (Diags.getDiagnosticLevel(diag::warn_noreturn_function_has_return_expr) 2367 == Diagnostic::Ignored || !HasNoReturn) 2368 && (Diags.getDiagnosticLevel(diag::warn_suggest_noreturn_block) 2369 == Diagnostic::Ignored || !ReturnsVoid)) 2370 return; 2371 // FIXME: Function try block 2372 if (CompoundStmt *Compound = dyn_cast<CompoundStmt>(Body)) { 2373 switch (CheckFallThrough(AC)) { 2374 case MaybeFallThrough: 2375 if (HasNoReturn) 2376 Diag(Compound->getRBracLoc(), diag::warn_falloff_noreturn_function); 2377 else if (!ReturnsVoid) 2378 Diag(Compound->getRBracLoc(),diag::warn_maybe_falloff_nonvoid_function); 2379 break; 2380 case AlwaysFallThrough: 2381 if (HasNoReturn) 2382 Diag(Compound->getRBracLoc(), diag::warn_falloff_noreturn_function); 2383 else if (!ReturnsVoid) 2384 Diag(Compound->getRBracLoc(), diag::warn_falloff_nonvoid_function); 2385 break; 2386 case NeverFallThroughOrReturn: 2387 if (ReturnsVoid && !HasNoReturn) 2388 Diag(Compound->getLBracLoc(), diag::warn_suggest_noreturn_function); 2389 break; 2390 case NeverFallThrough: 2391 break; 2392 } 2393 } 2394} 2395 2396/// CheckFallThroughForBlock - Check that we don't fall off the end of a block 2397/// that should return a value. Check that we don't fall off the end of a 2398/// noreturn block. We assume that functions and blocks not marked noreturn 2399/// will return. 2400void Sema::CheckFallThroughForBlock(QualType BlockTy, Stmt *Body, 2401 AnalysisContext &AC) { 2402 // FIXME: Would be nice if we had a better way to control cascading errors, 2403 // but for now, avoid them. The problem is that when Parse sees: 2404 // int foo() { return a; } 2405 // The return is eaten and the Sema code sees just: 2406 // int foo() { } 2407 // which this code would then warn about. 2408 if (getDiagnostics().hasErrorOccurred()) 2409 return; 2410 bool ReturnsVoid = false; 2411 bool HasNoReturn = false; 2412 if (const FunctionType *FT =BlockTy->getPointeeType()->getAs<FunctionType>()){ 2413 if (FT->getResultType()->isVoidType()) 2414 ReturnsVoid = true; 2415 if (FT->getNoReturnAttr()) 2416 HasNoReturn = true; 2417 } 2418 2419 // Short circuit for compilation speed. 2420 if (ReturnsVoid 2421 && !HasNoReturn 2422 && (Diags.getDiagnosticLevel(diag::warn_suggest_noreturn_block) 2423 == Diagnostic::Ignored || !ReturnsVoid)) 2424 return; 2425 // FIXME: Funtion try block 2426 if (CompoundStmt *Compound = dyn_cast<CompoundStmt>(Body)) { 2427 switch (CheckFallThrough(AC)) { 2428 case MaybeFallThrough: 2429 if (HasNoReturn) 2430 Diag(Compound->getRBracLoc(), diag::err_noreturn_block_has_return_expr); 2431 else if (!ReturnsVoid) 2432 Diag(Compound->getRBracLoc(), diag::err_maybe_falloff_nonvoid_block); 2433 break; 2434 case AlwaysFallThrough: 2435 if (HasNoReturn) 2436 Diag(Compound->getRBracLoc(), diag::err_noreturn_block_has_return_expr); 2437 else if (!ReturnsVoid) 2438 Diag(Compound->getRBracLoc(), diag::err_falloff_nonvoid_block); 2439 break; 2440 case NeverFallThroughOrReturn: 2441 if (ReturnsVoid) 2442 Diag(Compound->getLBracLoc(), diag::warn_suggest_noreturn_block); 2443 break; 2444 case NeverFallThrough: 2445 break; 2446 } 2447 } 2448} 2449 2450/// CheckParmsForFunctionDef - Check that the parameters of the given 2451/// function are appropriate for the definition of a function. This 2452/// takes care of any checks that cannot be performed on the 2453/// declaration itself, e.g., that the types of each of the function 2454/// parameters are complete. 2455bool Sema::CheckParmsForFunctionDef(FunctionDecl *FD) { 2456 bool HasInvalidParm = false; 2457 for (unsigned p = 0, NumParams = FD->getNumParams(); p < NumParams; ++p) { 2458 ParmVarDecl *Param = FD->getParamDecl(p); 2459 2460 // C99 6.7.5.3p4: the parameters in a parameter type list in a 2461 // function declarator that is part of a function definition of 2462 // that function shall not have incomplete type. 2463 // 2464 // This is also C++ [dcl.fct]p6. 2465 if (!Param->isInvalidDecl() && 2466 RequireCompleteType(Param->getLocation(), Param->getType(), 2467 diag::err_typecheck_decl_incomplete_type)) { 2468 Param->setInvalidDecl(); 2469 HasInvalidParm = true; 2470 } 2471 2472 // C99 6.9.1p5: If the declarator includes a parameter type list, the 2473 // declaration of each parameter shall include an identifier. 2474 if (Param->getIdentifier() == 0 && 2475 !Param->isImplicit() && 2476 !getLangOptions().CPlusPlus) 2477 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 2478 2479 // C99 6.7.5.3p12: 2480 // If the function declarator is not part of a definition of that 2481 // function, parameters may have incomplete type and may use the [*] 2482 // notation in their sequences of declarator specifiers to specify 2483 // variable length array types. 2484 QualType PType = Param->getOriginalType(); 2485 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 2486 if (AT->getSizeModifier() == ArrayType::Star) { 2487 // FIXME: This diagnosic should point the the '[*]' if source-location 2488 // information is added for it. 2489 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 2490 } 2491 } 2492 2493 if (getLangOptions().CPlusPlus) 2494 if (const RecordType *RT = Param->getType()->getAs<RecordType>()) 2495 FinalizeVarWithDestructor(Param, RT); 2496 } 2497 2498 return HasInvalidParm; 2499} 2500