SemaChecking.cpp revision 788b0fd67e1992f23555454efcdb16a19dfefac3
1//===--- SemaChecking.cpp - Extra Semantic Checking -----------------------===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// This file implements extra semantic analysis beyond what is enforced 11// by the C type system. 12// 13//===----------------------------------------------------------------------===// 14 15#include "Sema.h" 16#include "clang/Analysis/Analyses/PrintfFormatString.h" 17#include "clang/AST/ASTContext.h" 18#include "clang/AST/CharUnits.h" 19#include "clang/AST/DeclObjC.h" 20#include "clang/AST/ExprCXX.h" 21#include "clang/AST/ExprObjC.h" 22#include "clang/AST/DeclObjC.h" 23#include "clang/AST/StmtCXX.h" 24#include "clang/AST/StmtObjC.h" 25#include "clang/Lex/LiteralSupport.h" 26#include "clang/Lex/Preprocessor.h" 27#include "llvm/ADT/BitVector.h" 28#include "llvm/ADT/STLExtras.h" 29#include "llvm/ADT/StringExtras.h" 30#include "llvm/Support/raw_ostream.h" 31#include "clang/Basic/TargetBuiltins.h" 32#include "clang/Basic/TargetInfo.h" 33#include <limits> 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 bool Invalid = false; 63 llvm::StringRef Buffer = SourceMgr.getBufferData(LocInfo.first, &Invalid); 64 if (Invalid) 65 return StrTokSpellingLoc; 66 67 const char *StrData = Buffer.data()+LocInfo.second; 68 69 // Create a langops struct and enable trigraphs. This is sufficient for 70 // relexing tokens. 71 LangOptions LangOpts; 72 LangOpts.Trigraphs = true; 73 74 // Create a lexer starting at the beginning of this token. 75 Lexer TheLexer(StrTokSpellingLoc, LangOpts, Buffer.begin(), StrData, 76 Buffer.end()); 77 Token TheTok; 78 TheLexer.LexFromRawLexer(TheTok); 79 80 // Use the StringLiteralParser to compute the length of the string in bytes. 81 StringLiteralParser SLP(&TheTok, 1, PP, /*Complain=*/false); 82 unsigned TokNumBytes = SLP.GetStringLength(); 83 84 // If the byte is in this token, return the location of the byte. 85 if (ByteNo < TokNumBytes || 86 (ByteNo == TokNumBytes && TokNo == SL->getNumConcatenated())) { 87 unsigned Offset = 88 StringLiteralParser::getOffsetOfStringByte(TheTok, ByteNo, PP, 89 /*Complain=*/false); 90 91 // Now that we know the offset of the token in the spelling, use the 92 // preprocessor to get the offset in the original source. 93 return PP.AdvanceToTokenCharacter(StrTokLoc, Offset); 94 } 95 96 // Move to the next string token. 97 ++TokNo; 98 ByteNo -= TokNumBytes; 99 } 100} 101 102/// CheckablePrintfAttr - does a function call have a "printf" attribute 103/// and arguments that merit checking? 104bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) { 105 if (Format->getType() == "printf") return true; 106 if (Format->getType() == "printf0") { 107 // printf0 allows null "format" string; if so don't check format/args 108 unsigned format_idx = Format->getFormatIdx() - 1; 109 // Does the index refer to the implicit object argument? 110 if (isa<CXXMemberCallExpr>(TheCall)) { 111 if (format_idx == 0) 112 return false; 113 --format_idx; 114 } 115 if (format_idx < TheCall->getNumArgs()) { 116 Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts(); 117 if (!Format->isNullPointerConstant(Context, 118 Expr::NPC_ValueDependentIsNull)) 119 return true; 120 } 121 } 122 return false; 123} 124 125Action::OwningExprResult 126Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 127 OwningExprResult TheCallResult(Owned(TheCall)); 128 129 switch (BuiltinID) { 130 case Builtin::BI__builtin___CFStringMakeConstantString: 131 assert(TheCall->getNumArgs() == 1 && 132 "Wrong # arguments to builtin CFStringMakeConstantString"); 133 if (CheckObjCString(TheCall->getArg(0))) 134 return ExprError(); 135 break; 136 case Builtin::BI__builtin_stdarg_start: 137 case Builtin::BI__builtin_va_start: 138 if (SemaBuiltinVAStart(TheCall)) 139 return ExprError(); 140 break; 141 case Builtin::BI__builtin_isgreater: 142 case Builtin::BI__builtin_isgreaterequal: 143 case Builtin::BI__builtin_isless: 144 case Builtin::BI__builtin_islessequal: 145 case Builtin::BI__builtin_islessgreater: 146 case Builtin::BI__builtin_isunordered: 147 if (SemaBuiltinUnorderedCompare(TheCall)) 148 return ExprError(); 149 break; 150 case Builtin::BI__builtin_fpclassify: 151 if (SemaBuiltinFPClassification(TheCall, 6)) 152 return ExprError(); 153 break; 154 case Builtin::BI__builtin_isfinite: 155 case Builtin::BI__builtin_isinf: 156 case Builtin::BI__builtin_isinf_sign: 157 case Builtin::BI__builtin_isnan: 158 case Builtin::BI__builtin_isnormal: 159 if (SemaBuiltinFPClassification(TheCall, 1)) 160 return ExprError(); 161 break; 162 case Builtin::BI__builtin_return_address: 163 case Builtin::BI__builtin_frame_address: { 164 llvm::APSInt Result; 165 if (SemaBuiltinConstantArg(TheCall, 0, Result)) 166 return ExprError(); 167 break; 168 } 169 case Builtin::BI__builtin_eh_return_data_regno: { 170 llvm::APSInt Result; 171 if (SemaBuiltinConstantArg(TheCall, 0, Result)) 172 return ExprError(); 173 break; 174 } 175 case Builtin::BI__builtin_shufflevector: 176 return SemaBuiltinShuffleVector(TheCall); 177 // TheCall will be freed by the smart pointer here, but that's fine, since 178 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 179 case Builtin::BI__builtin_prefetch: 180 if (SemaBuiltinPrefetch(TheCall)) 181 return ExprError(); 182 break; 183 case Builtin::BI__builtin_object_size: 184 if (SemaBuiltinObjectSize(TheCall)) 185 return ExprError(); 186 break; 187 case Builtin::BI__builtin_longjmp: 188 if (SemaBuiltinLongjmp(TheCall)) 189 return ExprError(); 190 break; 191 case Builtin::BI__sync_fetch_and_add: 192 case Builtin::BI__sync_fetch_and_sub: 193 case Builtin::BI__sync_fetch_and_or: 194 case Builtin::BI__sync_fetch_and_and: 195 case Builtin::BI__sync_fetch_and_xor: 196 case Builtin::BI__sync_add_and_fetch: 197 case Builtin::BI__sync_sub_and_fetch: 198 case Builtin::BI__sync_and_and_fetch: 199 case Builtin::BI__sync_or_and_fetch: 200 case Builtin::BI__sync_xor_and_fetch: 201 case Builtin::BI__sync_val_compare_and_swap: 202 case Builtin::BI__sync_bool_compare_and_swap: 203 case Builtin::BI__sync_lock_test_and_set: 204 case Builtin::BI__sync_lock_release: 205 if (SemaBuiltinAtomicOverloaded(TheCall)) 206 return ExprError(); 207 break; 208 } 209 210 // Since the target specific builtins for each arch overlap, only check those 211 // of the arch we are compiling for. 212 if (BuiltinID >= Builtin::FirstTSBuiltin) { 213 switch (Context.Target.getTriple().getArch()) { 214 case llvm::Triple::arm: 215 case llvm::Triple::thumb: 216 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 217 return ExprError(); 218 break; 219 case llvm::Triple::x86: 220 case llvm::Triple::x86_64: 221 if (CheckX86BuiltinFunctionCall(BuiltinID, TheCall)) 222 return ExprError(); 223 break; 224 default: 225 break; 226 } 227 } 228 229 return move(TheCallResult); 230} 231 232bool Sema::CheckX86BuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 233 switch (BuiltinID) { 234 case X86::BI__builtin_ia32_palignr128: 235 case X86::BI__builtin_ia32_palignr: { 236 llvm::APSInt Result; 237 if (SemaBuiltinConstantArg(TheCall, 2, Result)) 238 return true; 239 break; 240 } 241 } 242 return false; 243} 244 245// Get the valid immediate range for the specified NEON type code. 246static unsigned RFT(unsigned t, bool shift = false) { 247 bool quad = t & 0x10; 248 249 switch (t & 0x7) { 250 case 0: // i8 251 return shift ? 7 : (8 << (int)quad) - 1; 252 case 1: // i16 253 return shift ? 15 : (4 << (int)quad) - 1; 254 case 2: // i32 255 return shift ? 31 : (2 << (int)quad) - 1; 256 case 3: // i64 257 return shift ? 63 : (1 << (int)quad) - 1; 258 case 4: // f32 259 assert(!shift && "cannot shift float types!"); 260 return (2 << (int)quad) - 1; 261 case 5: // poly8 262 assert(!shift && "cannot shift polynomial types!"); 263 return (8 << (int)quad) - 1; 264 case 6: // poly16 265 assert(!shift && "cannot shift polynomial types!"); 266 return (4 << (int)quad) - 1; 267 case 7: // float16 268 assert(!shift && "cannot shift float types!"); 269 return (4 << (int)quad) - 1; 270 } 271 return 0; 272} 273 274bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 275 llvm::APSInt Result; 276 277 unsigned mask = 0; 278 unsigned TV = 0; 279 switch (BuiltinID) { 280#define GET_NEON_OVERLOAD_CHECK 281#include "clang/Basic/arm_neon.inc" 282#undef GET_NEON_OVERLOAD_CHECK 283 } 284 285 // For NEON intrinsics which are overloaded on vector element type, validate 286 // the immediate which specifies which variant to emit. 287 if (mask) { 288 unsigned ArgNo = TheCall->getNumArgs()-1; 289 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 290 return true; 291 292 TV = Result.getLimitedValue(32); 293 if ((TV > 31) || (mask & (1 << TV)) == 0) 294 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 295 << TheCall->getArg(ArgNo)->getSourceRange(); 296 } 297 298 // For NEON intrinsics which take an immediate value as part of the 299 // instruction, range check them here. 300 unsigned i = 0, l = 0, u = 0; 301 switch (BuiltinID) { 302 default: return false; 303#define GET_NEON_IMMEDIATE_CHECK 304#include "clang/Basic/arm_neon.inc" 305#undef GET_NEON_IMMEDIATE_CHECK 306 }; 307 308 // Check that the immediate argument is actually a constant. 309 if (SemaBuiltinConstantArg(TheCall, i, Result)) 310 return true; 311 312 // Range check against the upper/lower values for this isntruction. 313 unsigned Val = Result.getZExtValue(); 314 if (Val < l || Val > (u + l)) 315 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 316 << llvm::utostr(l) << llvm::utostr(u+l) 317 << TheCall->getArg(i)->getSourceRange(); 318 319 return false; 320} 321 322/// CheckFunctionCall - Check a direct function call for various correctness 323/// and safety properties not strictly enforced by the C type system. 324bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 325 // Get the IdentifierInfo* for the called function. 326 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 327 328 // None of the checks below are needed for functions that don't have 329 // simple names (e.g., C++ conversion functions). 330 if (!FnInfo) 331 return false; 332 333 // FIXME: This mechanism should be abstracted to be less fragile and 334 // more efficient. For example, just map function ids to custom 335 // handlers. 336 337 // Printf checking. 338 if (const FormatAttr *Format = FDecl->getAttr<FormatAttr>()) { 339 if (CheckablePrintfAttr(Format, TheCall)) { 340 bool HasVAListArg = Format->getFirstArg() == 0; 341 CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 342 HasVAListArg ? 0 : Format->getFirstArg() - 1); 343 } 344 } 345 346 for (const NonNullAttr *NonNull = FDecl->getAttr<NonNullAttr>(); NonNull; 347 NonNull = NonNull->getNext<NonNullAttr>()) 348 CheckNonNullArguments(NonNull, TheCall); 349 350 return false; 351} 352 353bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 354 // Printf checking. 355 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 356 if (!Format) 357 return false; 358 359 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 360 if (!V) 361 return false; 362 363 QualType Ty = V->getType(); 364 if (!Ty->isBlockPointerType()) 365 return false; 366 367 if (!CheckablePrintfAttr(Format, TheCall)) 368 return false; 369 370 bool HasVAListArg = Format->getFirstArg() == 0; 371 CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 372 HasVAListArg ? 0 : Format->getFirstArg() - 1); 373 374 return false; 375} 376 377/// SemaBuiltinAtomicOverloaded - We have a call to a function like 378/// __sync_fetch_and_add, which is an overloaded function based on the pointer 379/// type of its first argument. The main ActOnCallExpr routines have already 380/// promoted the types of arguments because all of these calls are prototyped as 381/// void(...). 382/// 383/// This function goes through and does final semantic checking for these 384/// builtins, 385bool Sema::SemaBuiltinAtomicOverloaded(CallExpr *TheCall) { 386 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 387 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 388 389 // Ensure that we have at least one argument to do type inference from. 390 if (TheCall->getNumArgs() < 1) 391 return Diag(TheCall->getLocEnd(), 392 diag::err_typecheck_call_too_few_args_at_least) 393 << 0 << 1 << TheCall->getNumArgs() 394 << TheCall->getCallee()->getSourceRange(); 395 396 // Inspect the first argument of the atomic builtin. This should always be 397 // a pointer type, whose element is an integral scalar or pointer type. 398 // Because it is a pointer type, we don't have to worry about any implicit 399 // casts here. 400 Expr *FirstArg = TheCall->getArg(0); 401 if (!FirstArg->getType()->isPointerType()) 402 return Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 403 << FirstArg->getType() << FirstArg->getSourceRange(); 404 405 QualType ValType = FirstArg->getType()->getAs<PointerType>()->getPointeeType(); 406 if (!ValType->isIntegerType() && !ValType->isPointerType() && 407 !ValType->isBlockPointerType()) 408 return Diag(DRE->getLocStart(), 409 diag::err_atomic_builtin_must_be_pointer_intptr) 410 << FirstArg->getType() << FirstArg->getSourceRange(); 411 412 // We need to figure out which concrete builtin this maps onto. For example, 413 // __sync_fetch_and_add with a 2 byte object turns into 414 // __sync_fetch_and_add_2. 415#define BUILTIN_ROW(x) \ 416 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 417 Builtin::BI##x##_8, Builtin::BI##x##_16 } 418 419 static const unsigned BuiltinIndices[][5] = { 420 BUILTIN_ROW(__sync_fetch_and_add), 421 BUILTIN_ROW(__sync_fetch_and_sub), 422 BUILTIN_ROW(__sync_fetch_and_or), 423 BUILTIN_ROW(__sync_fetch_and_and), 424 BUILTIN_ROW(__sync_fetch_and_xor), 425 426 BUILTIN_ROW(__sync_add_and_fetch), 427 BUILTIN_ROW(__sync_sub_and_fetch), 428 BUILTIN_ROW(__sync_and_and_fetch), 429 BUILTIN_ROW(__sync_or_and_fetch), 430 BUILTIN_ROW(__sync_xor_and_fetch), 431 432 BUILTIN_ROW(__sync_val_compare_and_swap), 433 BUILTIN_ROW(__sync_bool_compare_and_swap), 434 BUILTIN_ROW(__sync_lock_test_and_set), 435 BUILTIN_ROW(__sync_lock_release) 436 }; 437#undef BUILTIN_ROW 438 439 // Determine the index of the size. 440 unsigned SizeIndex; 441 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 442 case 1: SizeIndex = 0; break; 443 case 2: SizeIndex = 1; break; 444 case 4: SizeIndex = 2; break; 445 case 8: SizeIndex = 3; break; 446 case 16: SizeIndex = 4; break; 447 default: 448 return Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 449 << FirstArg->getType() << FirstArg->getSourceRange(); 450 } 451 452 // Each of these builtins has one pointer argument, followed by some number of 453 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 454 // that we ignore. Find out which row of BuiltinIndices to read from as well 455 // as the number of fixed args. 456 unsigned BuiltinID = FDecl->getBuiltinID(); 457 unsigned BuiltinIndex, NumFixed = 1; 458 switch (BuiltinID) { 459 default: assert(0 && "Unknown overloaded atomic builtin!"); 460 case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break; 461 case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break; 462 case Builtin::BI__sync_fetch_and_or: BuiltinIndex = 2; break; 463 case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break; 464 case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break; 465 466 case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break; 467 case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break; 468 case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break; 469 case Builtin::BI__sync_or_and_fetch: BuiltinIndex = 8; break; 470 case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break; 471 472 case Builtin::BI__sync_val_compare_and_swap: 473 BuiltinIndex = 10; 474 NumFixed = 2; 475 break; 476 case Builtin::BI__sync_bool_compare_and_swap: 477 BuiltinIndex = 11; 478 NumFixed = 2; 479 break; 480 case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break; 481 case Builtin::BI__sync_lock_release: 482 BuiltinIndex = 13; 483 NumFixed = 0; 484 break; 485 } 486 487 // Now that we know how many fixed arguments we expect, first check that we 488 // have at least that many. 489 if (TheCall->getNumArgs() < 1+NumFixed) 490 return Diag(TheCall->getLocEnd(), 491 diag::err_typecheck_call_too_few_args_at_least) 492 << 0 << 1+NumFixed << TheCall->getNumArgs() 493 << TheCall->getCallee()->getSourceRange(); 494 495 496 // Get the decl for the concrete builtin from this, we can tell what the 497 // concrete integer type we should convert to is. 498 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 499 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 500 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 501 FunctionDecl *NewBuiltinDecl = 502 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 503 TUScope, false, DRE->getLocStart())); 504 const FunctionProtoType *BuiltinFT = 505 NewBuiltinDecl->getType()->getAs<FunctionProtoType>(); 506 ValType = BuiltinFT->getArgType(0)->getAs<PointerType>()->getPointeeType(); 507 508 // If the first type needs to be converted (e.g. void** -> int*), do it now. 509 if (BuiltinFT->getArgType(0) != FirstArg->getType()) { 510 ImpCastExprToType(FirstArg, BuiltinFT->getArgType(0), CastExpr::CK_BitCast); 511 TheCall->setArg(0, FirstArg); 512 } 513 514 // Next, walk the valid ones promoting to the right type. 515 for (unsigned i = 0; i != NumFixed; ++i) { 516 Expr *Arg = TheCall->getArg(i+1); 517 518 // If the argument is an implicit cast, then there was a promotion due to 519 // "...", just remove it now. 520 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) { 521 Arg = ICE->getSubExpr(); 522 ICE->setSubExpr(0); 523 ICE->Destroy(Context); 524 TheCall->setArg(i+1, Arg); 525 } 526 527 // GCC does an implicit conversion to the pointer or integer ValType. This 528 // can fail in some cases (1i -> int**), check for this error case now. 529 CastExpr::CastKind Kind = CastExpr::CK_Unknown; 530 CXXBaseSpecifierArray BasePath; 531 if (CheckCastTypes(Arg->getSourceRange(), ValType, Arg, Kind, BasePath)) 532 return true; 533 534 // Okay, we have something that *can* be converted to the right type. Check 535 // to see if there is a potentially weird extension going on here. This can 536 // happen when you do an atomic operation on something like an char* and 537 // pass in 42. The 42 gets converted to char. This is even more strange 538 // for things like 45.123 -> char, etc. 539 // FIXME: Do this check. 540 ImpCastExprToType(Arg, ValType, Kind); 541 TheCall->setArg(i+1, Arg); 542 } 543 544 // Switch the DeclRefExpr to refer to the new decl. 545 DRE->setDecl(NewBuiltinDecl); 546 DRE->setType(NewBuiltinDecl->getType()); 547 548 // Set the callee in the CallExpr. 549 // FIXME: This leaks the original parens and implicit casts. 550 Expr *PromotedCall = DRE; 551 UsualUnaryConversions(PromotedCall); 552 TheCall->setCallee(PromotedCall); 553 554 555 // Change the result type of the call to match the result type of the decl. 556 TheCall->setType(NewBuiltinDecl->getResultType()); 557 return false; 558} 559 560 561/// CheckObjCString - Checks that the argument to the builtin 562/// CFString constructor is correct 563/// FIXME: GCC currently emits the following warning: 564/// "warning: input conversion stopped due to an input byte that does not 565/// belong to the input codeset UTF-8" 566/// Note: It might also make sense to do the UTF-16 conversion here (would 567/// simplify the backend). 568bool Sema::CheckObjCString(Expr *Arg) { 569 Arg = Arg->IgnoreParenCasts(); 570 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 571 572 if (!Literal || Literal->isWide()) { 573 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 574 << Arg->getSourceRange(); 575 return true; 576 } 577 578 const char *Data = Literal->getStrData(); 579 unsigned Length = Literal->getByteLength(); 580 581 for (unsigned i = 0; i < Length; ++i) { 582 if (!Data[i]) { 583 Diag(getLocationOfStringLiteralByte(Literal, i), 584 diag::warn_cfstring_literal_contains_nul_character) 585 << Arg->getSourceRange(); 586 break; 587 } 588 } 589 590 return false; 591} 592 593/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 594/// Emit an error and return true on failure, return false on success. 595bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 596 Expr *Fn = TheCall->getCallee(); 597 if (TheCall->getNumArgs() > 2) { 598 Diag(TheCall->getArg(2)->getLocStart(), 599 diag::err_typecheck_call_too_many_args) 600 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 601 << Fn->getSourceRange() 602 << SourceRange(TheCall->getArg(2)->getLocStart(), 603 (*(TheCall->arg_end()-1))->getLocEnd()); 604 return true; 605 } 606 607 if (TheCall->getNumArgs() < 2) { 608 return Diag(TheCall->getLocEnd(), 609 diag::err_typecheck_call_too_few_args_at_least) 610 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 611 } 612 613 // Determine whether the current function is variadic or not. 614 BlockScopeInfo *CurBlock = getCurBlock(); 615 bool isVariadic; 616 if (CurBlock) 617 isVariadic = CurBlock->TheDecl->isVariadic(); 618 else if (FunctionDecl *FD = getCurFunctionDecl()) 619 isVariadic = FD->isVariadic(); 620 else 621 isVariadic = getCurMethodDecl()->isVariadic(); 622 623 if (!isVariadic) { 624 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 625 return true; 626 } 627 628 // Verify that the second argument to the builtin is the last argument of the 629 // current function or method. 630 bool SecondArgIsLastNamedArgument = false; 631 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 632 633 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 634 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 635 // FIXME: This isn't correct for methods (results in bogus warning). 636 // Get the last formal in the current function. 637 const ParmVarDecl *LastArg; 638 if (CurBlock) 639 LastArg = *(CurBlock->TheDecl->param_end()-1); 640 else if (FunctionDecl *FD = getCurFunctionDecl()) 641 LastArg = *(FD->param_end()-1); 642 else 643 LastArg = *(getCurMethodDecl()->param_end()-1); 644 SecondArgIsLastNamedArgument = PV == LastArg; 645 } 646 } 647 648 if (!SecondArgIsLastNamedArgument) 649 Diag(TheCall->getArg(1)->getLocStart(), 650 diag::warn_second_parameter_of_va_start_not_last_named_argument); 651 return false; 652} 653 654/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 655/// friends. This is declared to take (...), so we have to check everything. 656bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 657 if (TheCall->getNumArgs() < 2) 658 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 659 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 660 if (TheCall->getNumArgs() > 2) 661 return Diag(TheCall->getArg(2)->getLocStart(), 662 diag::err_typecheck_call_too_many_args) 663 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 664 << SourceRange(TheCall->getArg(2)->getLocStart(), 665 (*(TheCall->arg_end()-1))->getLocEnd()); 666 667 Expr *OrigArg0 = TheCall->getArg(0); 668 Expr *OrigArg1 = TheCall->getArg(1); 669 670 // Do standard promotions between the two arguments, returning their common 671 // type. 672 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 673 674 // Make sure any conversions are pushed back into the call; this is 675 // type safe since unordered compare builtins are declared as "_Bool 676 // foo(...)". 677 TheCall->setArg(0, OrigArg0); 678 TheCall->setArg(1, OrigArg1); 679 680 if (OrigArg0->isTypeDependent() || OrigArg1->isTypeDependent()) 681 return false; 682 683 // If the common type isn't a real floating type, then the arguments were 684 // invalid for this operation. 685 if (!Res->isRealFloatingType()) 686 return Diag(OrigArg0->getLocStart(), 687 diag::err_typecheck_call_invalid_ordered_compare) 688 << OrigArg0->getType() << OrigArg1->getType() 689 << SourceRange(OrigArg0->getLocStart(), OrigArg1->getLocEnd()); 690 691 return false; 692} 693 694/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 695/// __builtin_isnan and friends. This is declared to take (...), so we have 696/// to check everything. We expect the last argument to be a floating point 697/// value. 698bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 699 if (TheCall->getNumArgs() < NumArgs) 700 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 701 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 702 if (TheCall->getNumArgs() > NumArgs) 703 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 704 diag::err_typecheck_call_too_many_args) 705 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 706 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 707 (*(TheCall->arg_end()-1))->getLocEnd()); 708 709 Expr *OrigArg = TheCall->getArg(NumArgs-1); 710 711 if (OrigArg->isTypeDependent()) 712 return false; 713 714 // This operation requires a non-_Complex floating-point number. 715 if (!OrigArg->getType()->isRealFloatingType()) 716 return Diag(OrigArg->getLocStart(), 717 diag::err_typecheck_call_invalid_unary_fp) 718 << OrigArg->getType() << OrigArg->getSourceRange(); 719 720 // If this is an implicit conversion from float -> double, remove it. 721 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 722 Expr *CastArg = Cast->getSubExpr(); 723 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 724 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 725 "promotion from float to double is the only expected cast here"); 726 Cast->setSubExpr(0); 727 Cast->Destroy(Context); 728 TheCall->setArg(NumArgs-1, CastArg); 729 OrigArg = CastArg; 730 } 731 } 732 733 return false; 734} 735 736/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 737// This is declared to take (...), so we have to check everything. 738Action::OwningExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 739 if (TheCall->getNumArgs() < 2) 740 return ExprError(Diag(TheCall->getLocEnd(), 741 diag::err_typecheck_call_too_few_args_at_least) 742 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 743 << TheCall->getSourceRange()); 744 745 // Determine which of the following types of shufflevector we're checking: 746 // 1) unary, vector mask: (lhs, mask) 747 // 2) binary, vector mask: (lhs, rhs, mask) 748 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 749 QualType resType = TheCall->getArg(0)->getType(); 750 unsigned numElements = 0; 751 752 if (!TheCall->getArg(0)->isTypeDependent() && 753 !TheCall->getArg(1)->isTypeDependent()) { 754 QualType LHSType = TheCall->getArg(0)->getType(); 755 QualType RHSType = TheCall->getArg(1)->getType(); 756 757 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 758 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 759 << SourceRange(TheCall->getArg(0)->getLocStart(), 760 TheCall->getArg(1)->getLocEnd()); 761 return ExprError(); 762 } 763 764 numElements = LHSType->getAs<VectorType>()->getNumElements(); 765 unsigned numResElements = TheCall->getNumArgs() - 2; 766 767 // Check to see if we have a call with 2 vector arguments, the unary shuffle 768 // with mask. If so, verify that RHS is an integer vector type with the 769 // same number of elts as lhs. 770 if (TheCall->getNumArgs() == 2) { 771 if (!RHSType->isIntegerType() || 772 RHSType->getAs<VectorType>()->getNumElements() != numElements) 773 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 774 << SourceRange(TheCall->getArg(1)->getLocStart(), 775 TheCall->getArg(1)->getLocEnd()); 776 numResElements = numElements; 777 } 778 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 779 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 780 << SourceRange(TheCall->getArg(0)->getLocStart(), 781 TheCall->getArg(1)->getLocEnd()); 782 return ExprError(); 783 } else if (numElements != numResElements) { 784 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 785 resType = Context.getVectorType(eltType, numResElements, 786 VectorType::NotAltiVec); 787 } 788 } 789 790 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 791 if (TheCall->getArg(i)->isTypeDependent() || 792 TheCall->getArg(i)->isValueDependent()) 793 continue; 794 795 llvm::APSInt Result(32); 796 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 797 return ExprError(Diag(TheCall->getLocStart(), 798 diag::err_shufflevector_nonconstant_argument) 799 << TheCall->getArg(i)->getSourceRange()); 800 801 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 802 return ExprError(Diag(TheCall->getLocStart(), 803 diag::err_shufflevector_argument_too_large) 804 << TheCall->getArg(i)->getSourceRange()); 805 } 806 807 llvm::SmallVector<Expr*, 32> exprs; 808 809 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 810 exprs.push_back(TheCall->getArg(i)); 811 TheCall->setArg(i, 0); 812 } 813 814 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 815 exprs.size(), resType, 816 TheCall->getCallee()->getLocStart(), 817 TheCall->getRParenLoc())); 818} 819 820/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 821// This is declared to take (const void*, ...) and can take two 822// optional constant int args. 823bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 824 unsigned NumArgs = TheCall->getNumArgs(); 825 826 if (NumArgs > 3) 827 return Diag(TheCall->getLocEnd(), 828 diag::err_typecheck_call_too_many_args_at_most) 829 << 0 /*function call*/ << 3 << NumArgs 830 << TheCall->getSourceRange(); 831 832 // Argument 0 is checked for us and the remaining arguments must be 833 // constant integers. 834 for (unsigned i = 1; i != NumArgs; ++i) { 835 Expr *Arg = TheCall->getArg(i); 836 837 llvm::APSInt Result; 838 if (SemaBuiltinConstantArg(TheCall, i, Result)) 839 return true; 840 841 // FIXME: gcc issues a warning and rewrites these to 0. These 842 // seems especially odd for the third argument since the default 843 // is 3. 844 if (i == 1) { 845 if (Result.getLimitedValue() > 1) 846 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 847 << "0" << "1" << Arg->getSourceRange(); 848 } else { 849 if (Result.getLimitedValue() > 3) 850 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 851 << "0" << "3" << Arg->getSourceRange(); 852 } 853 } 854 855 return false; 856} 857 858/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 859/// TheCall is a constant expression. 860bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 861 llvm::APSInt &Result) { 862 Expr *Arg = TheCall->getArg(ArgNum); 863 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 864 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 865 866 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 867 868 if (!Arg->isIntegerConstantExpr(Result, Context)) 869 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 870 << FDecl->getDeclName() << Arg->getSourceRange(); 871 872 return false; 873} 874 875/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 876/// int type). This simply type checks that type is one of the defined 877/// constants (0-3). 878// For compatability check 0-3, llvm only handles 0 and 2. 879bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 880 llvm::APSInt Result; 881 882 // Check constant-ness first. 883 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 884 return true; 885 886 Expr *Arg = TheCall->getArg(1); 887 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 888 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 889 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 890 } 891 892 return false; 893} 894 895/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 896/// This checks that val is a constant 1. 897bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 898 Expr *Arg = TheCall->getArg(1); 899 llvm::APSInt Result; 900 901 // TODO: This is less than ideal. Overload this to take a value. 902 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 903 return true; 904 905 if (Result != 1) 906 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 907 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 908 909 return false; 910} 911 912// Handle i > 1 ? "x" : "y", recursivelly 913bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 914 bool HasVAListArg, 915 unsigned format_idx, unsigned firstDataArg) { 916 if (E->isTypeDependent() || E->isValueDependent()) 917 return false; 918 919 switch (E->getStmtClass()) { 920 case Stmt::ConditionalOperatorClass: { 921 const ConditionalOperator *C = cast<ConditionalOperator>(E); 922 return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, 923 HasVAListArg, format_idx, firstDataArg) 924 && SemaCheckStringLiteral(C->getRHS(), TheCall, 925 HasVAListArg, format_idx, firstDataArg); 926 } 927 928 case Stmt::ImplicitCastExprClass: { 929 const ImplicitCastExpr *Expr = cast<ImplicitCastExpr>(E); 930 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 931 format_idx, firstDataArg); 932 } 933 934 case Stmt::ParenExprClass: { 935 const ParenExpr *Expr = cast<ParenExpr>(E); 936 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 937 format_idx, firstDataArg); 938 } 939 940 case Stmt::DeclRefExprClass: { 941 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 942 943 // As an exception, do not flag errors for variables binding to 944 // const string literals. 945 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 946 bool isConstant = false; 947 QualType T = DR->getType(); 948 949 if (const ArrayType *AT = Context.getAsArrayType(T)) { 950 isConstant = AT->getElementType().isConstant(Context); 951 } else if (const PointerType *PT = T->getAs<PointerType>()) { 952 isConstant = T.isConstant(Context) && 953 PT->getPointeeType().isConstant(Context); 954 } 955 956 if (isConstant) { 957 if (const Expr *Init = VD->getAnyInitializer()) 958 return SemaCheckStringLiteral(Init, TheCall, 959 HasVAListArg, format_idx, firstDataArg); 960 } 961 962 // For vprintf* functions (i.e., HasVAListArg==true), we add a 963 // special check to see if the format string is a function parameter 964 // of the function calling the printf function. If the function 965 // has an attribute indicating it is a printf-like function, then we 966 // should suppress warnings concerning non-literals being used in a call 967 // to a vprintf function. For example: 968 // 969 // void 970 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 971 // va_list ap; 972 // va_start(ap, fmt); 973 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 974 // ... 975 // 976 // 977 // FIXME: We don't have full attribute support yet, so just check to see 978 // if the argument is a DeclRefExpr that references a parameter. We'll 979 // add proper support for checking the attribute later. 980 if (HasVAListArg) 981 if (isa<ParmVarDecl>(VD)) 982 return true; 983 } 984 985 return false; 986 } 987 988 case Stmt::CallExprClass: { 989 const CallExpr *CE = cast<CallExpr>(E); 990 if (const ImplicitCastExpr *ICE 991 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 992 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 993 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 994 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 995 unsigned ArgIndex = FA->getFormatIdx(); 996 const Expr *Arg = CE->getArg(ArgIndex - 1); 997 998 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 999 format_idx, firstDataArg); 1000 } 1001 } 1002 } 1003 } 1004 1005 return false; 1006 } 1007 case Stmt::ObjCStringLiteralClass: 1008 case Stmt::StringLiteralClass: { 1009 const StringLiteral *StrE = NULL; 1010 1011 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1012 StrE = ObjCFExpr->getString(); 1013 else 1014 StrE = cast<StringLiteral>(E); 1015 1016 if (StrE) { 1017 CheckPrintfString(StrE, E, TheCall, HasVAListArg, format_idx, 1018 firstDataArg); 1019 return true; 1020 } 1021 1022 return false; 1023 } 1024 1025 default: 1026 return false; 1027 } 1028} 1029 1030void 1031Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1032 const CallExpr *TheCall) { 1033 for (NonNullAttr::iterator i = NonNull->begin(), e = NonNull->end(); 1034 i != e; ++i) { 1035 const Expr *ArgExpr = TheCall->getArg(*i); 1036 if (ArgExpr->isNullPointerConstant(Context, 1037 Expr::NPC_ValueDependentIsNotNull)) 1038 Diag(TheCall->getCallee()->getLocStart(), diag::warn_null_arg) 1039 << ArgExpr->getSourceRange(); 1040 } 1041} 1042 1043/// CheckPrintfArguments - Check calls to printf (and similar functions) for 1044/// correct use of format strings. 1045/// 1046/// HasVAListArg - A predicate indicating whether the printf-like 1047/// function is passed an explicit va_arg argument (e.g., vprintf) 1048/// 1049/// format_idx - The index into Args for the format string. 1050/// 1051/// Improper format strings to functions in the printf family can be 1052/// the source of bizarre bugs and very serious security holes. A 1053/// good source of information is available in the following paper 1054/// (which includes additional references): 1055/// 1056/// FormatGuard: Automatic Protection From printf Format String 1057/// Vulnerabilities, Proceedings of the 10th USENIX Security Symposium, 2001. 1058/// 1059/// TODO: 1060/// Functionality implemented: 1061/// 1062/// We can statically check the following properties for string 1063/// literal format strings for non v.*printf functions (where the 1064/// arguments are passed directly): 1065// 1066/// (1) Are the number of format conversions equal to the number of 1067/// data arguments? 1068/// 1069/// (2) Does each format conversion correctly match the type of the 1070/// corresponding data argument? 1071/// 1072/// Moreover, for all printf functions we can: 1073/// 1074/// (3) Check for a missing format string (when not caught by type checking). 1075/// 1076/// (4) Check for no-operation flags; e.g. using "#" with format 1077/// conversion 'c' (TODO) 1078/// 1079/// (5) Check the use of '%n', a major source of security holes. 1080/// 1081/// (6) Check for malformed format conversions that don't specify anything. 1082/// 1083/// (7) Check for empty format strings. e.g: printf(""); 1084/// 1085/// (8) Check that the format string is a wide literal. 1086/// 1087/// All of these checks can be done by parsing the format string. 1088/// 1089void 1090Sema::CheckPrintfArguments(const CallExpr *TheCall, bool HasVAListArg, 1091 unsigned format_idx, unsigned firstDataArg) { 1092 const Expr *Fn = TheCall->getCallee(); 1093 1094 // The way the format attribute works in GCC, the implicit this argument 1095 // of member functions is counted. However, it doesn't appear in our own 1096 // lists, so decrement format_idx in that case. 1097 if (isa<CXXMemberCallExpr>(TheCall)) { 1098 // Catch a format attribute mistakenly referring to the object argument. 1099 if (format_idx == 0) 1100 return; 1101 --format_idx; 1102 if(firstDataArg != 0) 1103 --firstDataArg; 1104 } 1105 1106 // CHECK: printf-like function is called with no format string. 1107 if (format_idx >= TheCall->getNumArgs()) { 1108 Diag(TheCall->getRParenLoc(), diag::warn_printf_missing_format_string) 1109 << Fn->getSourceRange(); 1110 return; 1111 } 1112 1113 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 1114 1115 // CHECK: format string is not a string literal. 1116 // 1117 // Dynamically generated format strings are difficult to 1118 // automatically vet at compile time. Requiring that format strings 1119 // are string literals: (1) permits the checking of format strings by 1120 // the compiler and thereby (2) can practically remove the source of 1121 // many format string exploits. 1122 1123 // Format string can be either ObjC string (e.g. @"%d") or 1124 // C string (e.g. "%d") 1125 // ObjC string uses the same format specifiers as C string, so we can use 1126 // the same format string checking logic for both ObjC and C strings. 1127 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 1128 firstDataArg)) 1129 return; // Literal format string found, check done! 1130 1131 // If there are no arguments specified, warn with -Wformat-security, otherwise 1132 // warn only with -Wformat-nonliteral. 1133 if (TheCall->getNumArgs() == format_idx+1) 1134 Diag(TheCall->getArg(format_idx)->getLocStart(), 1135 diag::warn_printf_nonliteral_noargs) 1136 << OrigFormatExpr->getSourceRange(); 1137 else 1138 Diag(TheCall->getArg(format_idx)->getLocStart(), 1139 diag::warn_printf_nonliteral) 1140 << OrigFormatExpr->getSourceRange(); 1141} 1142 1143namespace { 1144class CheckPrintfHandler : public analyze_printf::FormatStringHandler { 1145 Sema &S; 1146 const StringLiteral *FExpr; 1147 const Expr *OrigFormatExpr; 1148 const unsigned FirstDataArg; 1149 const unsigned NumDataArgs; 1150 const bool IsObjCLiteral; 1151 const char *Beg; // Start of format string. 1152 const bool HasVAListArg; 1153 const CallExpr *TheCall; 1154 unsigned FormatIdx; 1155 llvm::BitVector CoveredArgs; 1156 bool usesPositionalArgs; 1157 bool atFirstArg; 1158public: 1159 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1160 const Expr *origFormatExpr, unsigned firstDataArg, 1161 unsigned numDataArgs, bool isObjCLiteral, 1162 const char *beg, bool hasVAListArg, 1163 const CallExpr *theCall, unsigned formatIdx) 1164 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1165 FirstDataArg(firstDataArg), 1166 NumDataArgs(numDataArgs), 1167 IsObjCLiteral(isObjCLiteral), Beg(beg), 1168 HasVAListArg(hasVAListArg), 1169 TheCall(theCall), FormatIdx(formatIdx), 1170 usesPositionalArgs(false), atFirstArg(true) { 1171 CoveredArgs.resize(numDataArgs); 1172 CoveredArgs.reset(); 1173 } 1174 1175 void DoneProcessing(); 1176 1177 void HandleIncompleteFormatSpecifier(const char *startSpecifier, 1178 unsigned specifierLen); 1179 1180 bool 1181 HandleInvalidConversionSpecifier(const analyze_printf::FormatSpecifier &FS, 1182 const char *startSpecifier, 1183 unsigned specifierLen); 1184 1185 virtual void HandleInvalidPosition(const char *startSpecifier, 1186 unsigned specifierLen, 1187 analyze_printf::PositionContext p); 1188 1189 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1190 1191 void HandleNullChar(const char *nullCharacter); 1192 1193 bool HandleFormatSpecifier(const analyze_printf::FormatSpecifier &FS, 1194 const char *startSpecifier, 1195 unsigned specifierLen); 1196private: 1197 SourceRange getFormatStringRange(); 1198 CharSourceRange getFormatSpecifierRange(const char *startSpecifier, 1199 unsigned specifierLen); 1200 SourceLocation getLocationOfByte(const char *x); 1201 1202 bool HandleAmount(const analyze_printf::OptionalAmount &Amt, unsigned k, 1203 const char *startSpecifier, unsigned specifierLen); 1204 void HandleInvalidAmount(const analyze_printf::FormatSpecifier &FS, 1205 const analyze_printf::OptionalAmount &Amt, 1206 unsigned type, 1207 const char *startSpecifier, unsigned specifierLen); 1208 void HandleFlag(const analyze_printf::FormatSpecifier &FS, 1209 const analyze_printf::OptionalFlag &flag, 1210 const char *startSpecifier, unsigned specifierLen); 1211 void HandleIgnoredFlag(const analyze_printf::FormatSpecifier &FS, 1212 const analyze_printf::OptionalFlag &ignoredFlag, 1213 const analyze_printf::OptionalFlag &flag, 1214 const char *startSpecifier, unsigned specifierLen); 1215 1216 const Expr *getDataArg(unsigned i) const; 1217}; 1218} 1219 1220SourceRange CheckPrintfHandler::getFormatStringRange() { 1221 return OrigFormatExpr->getSourceRange(); 1222} 1223 1224CharSourceRange CheckPrintfHandler:: 1225getFormatSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1226 SourceLocation Start = getLocationOfByte(startSpecifier); 1227 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1228 1229 // Advance the end SourceLocation by one due to half-open ranges. 1230 End = End.getFileLocWithOffset(1); 1231 1232 return CharSourceRange::getCharRange(Start, End); 1233} 1234 1235SourceLocation CheckPrintfHandler::getLocationOfByte(const char *x) { 1236 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1237} 1238 1239void CheckPrintfHandler:: 1240HandleIncompleteFormatSpecifier(const char *startSpecifier, 1241 unsigned specifierLen) { 1242 SourceLocation Loc = getLocationOfByte(startSpecifier); 1243 S.Diag(Loc, diag::warn_printf_incomplete_specifier) 1244 << getFormatSpecifierRange(startSpecifier, specifierLen); 1245} 1246 1247void 1248CheckPrintfHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1249 analyze_printf::PositionContext p) { 1250 SourceLocation Loc = getLocationOfByte(startPos); 1251 S.Diag(Loc, diag::warn_printf_invalid_positional_specifier) 1252 << (unsigned) p << getFormatSpecifierRange(startPos, posLen); 1253} 1254 1255void CheckPrintfHandler::HandleZeroPosition(const char *startPos, 1256 unsigned posLen) { 1257 SourceLocation Loc = getLocationOfByte(startPos); 1258 S.Diag(Loc, diag::warn_printf_zero_positional_specifier) 1259 << getFormatSpecifierRange(startPos, posLen); 1260} 1261 1262bool CheckPrintfHandler:: 1263HandleInvalidConversionSpecifier(const analyze_printf::FormatSpecifier &FS, 1264 const char *startSpecifier, 1265 unsigned specifierLen) { 1266 1267 unsigned argIndex = FS.getArgIndex(); 1268 bool keepGoing = true; 1269 if (argIndex < NumDataArgs) { 1270 // Consider the argument coverered, even though the specifier doesn't 1271 // make sense. 1272 CoveredArgs.set(argIndex); 1273 } 1274 else { 1275 // If argIndex exceeds the number of data arguments we 1276 // don't issue a warning because that is just a cascade of warnings (and 1277 // they may have intended '%%' anyway). We don't want to continue processing 1278 // the format string after this point, however, as we will like just get 1279 // gibberish when trying to match arguments. 1280 keepGoing = false; 1281 } 1282 1283 const analyze_printf::ConversionSpecifier &CS = 1284 FS.getConversionSpecifier(); 1285 SourceLocation Loc = getLocationOfByte(CS.getStart()); 1286 S.Diag(Loc, diag::warn_printf_invalid_conversion) 1287 << llvm::StringRef(CS.getStart(), CS.getLength()) 1288 << getFormatSpecifierRange(startSpecifier, specifierLen); 1289 1290 return keepGoing; 1291} 1292 1293void CheckPrintfHandler::HandleNullChar(const char *nullCharacter) { 1294 // The presence of a null character is likely an error. 1295 S.Diag(getLocationOfByte(nullCharacter), 1296 diag::warn_printf_format_string_contains_null_char) 1297 << getFormatStringRange(); 1298} 1299 1300const Expr *CheckPrintfHandler::getDataArg(unsigned i) const { 1301 return TheCall->getArg(FirstDataArg + i); 1302} 1303 1304bool 1305CheckPrintfHandler::HandleAmount(const analyze_printf::OptionalAmount &Amt, 1306 unsigned k, const char *startSpecifier, 1307 unsigned specifierLen) { 1308 1309 if (Amt.hasDataArgument()) { 1310 if (!HasVAListArg) { 1311 unsigned argIndex = Amt.getArgIndex(); 1312 if (argIndex >= NumDataArgs) { 1313 S.Diag(getLocationOfByte(Amt.getStart()), 1314 diag::warn_printf_asterisk_missing_arg) 1315 << k << getFormatSpecifierRange(startSpecifier, specifierLen); 1316 // Don't do any more checking. We will just emit 1317 // spurious errors. 1318 return false; 1319 } 1320 1321 // Type check the data argument. It should be an 'int'. 1322 // Although not in conformance with C99, we also allow the argument to be 1323 // an 'unsigned int' as that is a reasonably safe case. GCC also 1324 // doesn't emit a warning for that case. 1325 CoveredArgs.set(argIndex); 1326 const Expr *Arg = getDataArg(argIndex); 1327 QualType T = Arg->getType(); 1328 1329 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1330 assert(ATR.isValid()); 1331 1332 if (!ATR.matchesType(S.Context, T)) { 1333 S.Diag(getLocationOfByte(Amt.getStart()), 1334 diag::warn_printf_asterisk_wrong_type) 1335 << k 1336 << ATR.getRepresentativeType(S.Context) << T 1337 << getFormatSpecifierRange(startSpecifier, specifierLen) 1338 << Arg->getSourceRange(); 1339 // Don't do any more checking. We will just emit 1340 // spurious errors. 1341 return false; 1342 } 1343 } 1344 } 1345 return true; 1346} 1347 1348void CheckPrintfHandler::HandleInvalidAmount( 1349 const analyze_printf::FormatSpecifier &FS, 1350 const analyze_printf::OptionalAmount &Amt, 1351 unsigned type, 1352 const char *startSpecifier, 1353 unsigned specifierLen) { 1354 const analyze_printf::ConversionSpecifier &CS = FS.getConversionSpecifier(); 1355 switch (Amt.getHowSpecified()) { 1356 case analyze_printf::OptionalAmount::Constant: 1357 S.Diag(getLocationOfByte(Amt.getStart()), 1358 diag::warn_printf_nonsensical_optional_amount) 1359 << type 1360 << CS.toString() 1361 << getFormatSpecifierRange(startSpecifier, specifierLen) 1362 << FixItHint::CreateRemoval(getFormatSpecifierRange(Amt.getStart(), 1363 Amt.getConstantLength())); 1364 break; 1365 1366 default: 1367 S.Diag(getLocationOfByte(Amt.getStart()), 1368 diag::warn_printf_nonsensical_optional_amount) 1369 << type 1370 << CS.toString() 1371 << getFormatSpecifierRange(startSpecifier, specifierLen); 1372 break; 1373 } 1374} 1375 1376void CheckPrintfHandler::HandleFlag(const analyze_printf::FormatSpecifier &FS, 1377 const analyze_printf::OptionalFlag &flag, 1378 const char *startSpecifier, 1379 unsigned specifierLen) { 1380 // Warn about pointless flag with a fixit removal. 1381 const analyze_printf::ConversionSpecifier &CS = FS.getConversionSpecifier(); 1382 S.Diag(getLocationOfByte(flag.getPosition()), 1383 diag::warn_printf_nonsensical_flag) 1384 << flag.toString() << CS.toString() 1385 << getFormatSpecifierRange(startSpecifier, specifierLen) 1386 << FixItHint::CreateRemoval(getFormatSpecifierRange(flag.getPosition(), 1)); 1387} 1388 1389void CheckPrintfHandler::HandleIgnoredFlag( 1390 const analyze_printf::FormatSpecifier &FS, 1391 const analyze_printf::OptionalFlag &ignoredFlag, 1392 const analyze_printf::OptionalFlag &flag, 1393 const char *startSpecifier, 1394 unsigned specifierLen) { 1395 // Warn about ignored flag with a fixit removal. 1396 S.Diag(getLocationOfByte(ignoredFlag.getPosition()), 1397 diag::warn_printf_ignored_flag) 1398 << ignoredFlag.toString() << flag.toString() 1399 << getFormatSpecifierRange(startSpecifier, specifierLen) 1400 << FixItHint::CreateRemoval(getFormatSpecifierRange( 1401 ignoredFlag.getPosition(), 1)); 1402} 1403 1404bool 1405CheckPrintfHandler::HandleFormatSpecifier(const analyze_printf::FormatSpecifier 1406 &FS, 1407 const char *startSpecifier, 1408 unsigned specifierLen) { 1409 1410 using namespace analyze_printf; 1411 const ConversionSpecifier &CS = FS.getConversionSpecifier(); 1412 1413 if (atFirstArg) { 1414 atFirstArg = false; 1415 usesPositionalArgs = FS.usesPositionalArg(); 1416 } 1417 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1418 // Cannot mix-and-match positional and non-positional arguments. 1419 S.Diag(getLocationOfByte(CS.getStart()), 1420 diag::warn_printf_mix_positional_nonpositional_args) 1421 << getFormatSpecifierRange(startSpecifier, specifierLen); 1422 return false; 1423 } 1424 1425 // First check if the field width, precision, and conversion specifier 1426 // have matching data arguments. 1427 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 1428 startSpecifier, specifierLen)) { 1429 return false; 1430 } 1431 1432 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 1433 startSpecifier, specifierLen)) { 1434 return false; 1435 } 1436 1437 if (!CS.consumesDataArgument()) { 1438 // FIXME: Technically specifying a precision or field width here 1439 // makes no sense. Worth issuing a warning at some point. 1440 return true; 1441 } 1442 1443 // Consume the argument. 1444 unsigned argIndex = FS.getArgIndex(); 1445 if (argIndex < NumDataArgs) { 1446 // The check to see if the argIndex is valid will come later. 1447 // We set the bit here because we may exit early from this 1448 // function if we encounter some other error. 1449 CoveredArgs.set(argIndex); 1450 } 1451 1452 // Check for using an Objective-C specific conversion specifier 1453 // in a non-ObjC literal. 1454 if (!IsObjCLiteral && CS.isObjCArg()) { 1455 return HandleInvalidConversionSpecifier(FS, startSpecifier, specifierLen); 1456 } 1457 1458 // Check for invalid use of field width 1459 if (!FS.hasValidFieldWidth()) { 1460 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 1461 startSpecifier, specifierLen); 1462 } 1463 1464 // Check for invalid use of precision 1465 if (!FS.hasValidPrecision()) { 1466 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 1467 startSpecifier, specifierLen); 1468 } 1469 1470 // Check each flag does not conflict with any other component. 1471 if (!FS.hasValidLeadingZeros()) 1472 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 1473 if (!FS.hasValidPlusPrefix()) 1474 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 1475 if (!FS.hasValidSpacePrefix()) 1476 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 1477 if (!FS.hasValidAlternativeForm()) 1478 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 1479 if (!FS.hasValidLeftJustified()) 1480 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 1481 1482 // Check that flags are not ignored by another flag 1483 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 1484 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 1485 startSpecifier, specifierLen); 1486 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 1487 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 1488 startSpecifier, specifierLen); 1489 1490 // Check the length modifier is valid with the given conversion specifier. 1491 const LengthModifier &LM = FS.getLengthModifier(); 1492 if (!FS.hasValidLengthModifier()) 1493 S.Diag(getLocationOfByte(LM.getStart()), 1494 diag::warn_printf_nonsensical_length) 1495 << LM.toString() << CS.toString() 1496 << getFormatSpecifierRange(startSpecifier, specifierLen) 1497 << FixItHint::CreateRemoval(getFormatSpecifierRange(LM.getStart(), 1498 LM.getLength())); 1499 1500 // Are we using '%n'? 1501 if (CS.getKind() == ConversionSpecifier::OutIntPtrArg) { 1502 // Issue a warning about this being a possible security issue. 1503 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back) 1504 << getFormatSpecifierRange(startSpecifier, specifierLen); 1505 // Continue checking the other format specifiers. 1506 return true; 1507 } 1508 1509 // The remaining checks depend on the data arguments. 1510 if (HasVAListArg) 1511 return true; 1512 1513 if (argIndex >= NumDataArgs) { 1514 if (FS.usesPositionalArg()) { 1515 S.Diag(getLocationOfByte(CS.getStart()), 1516 diag::warn_printf_positional_arg_exceeds_data_args) 1517 << (argIndex+1) << NumDataArgs 1518 << getFormatSpecifierRange(startSpecifier, specifierLen); 1519 } 1520 else { 1521 S.Diag(getLocationOfByte(CS.getStart()), 1522 diag::warn_printf_insufficient_data_args) 1523 << getFormatSpecifierRange(startSpecifier, specifierLen); 1524 } 1525 1526 // Don't do any more checking. 1527 return false; 1528 } 1529 1530 // Now type check the data expression that matches the 1531 // format specifier. 1532 const Expr *Ex = getDataArg(argIndex); 1533 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context); 1534 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 1535 // Check if we didn't match because of an implicit cast from a 'char' 1536 // or 'short' to an 'int'. This is done because printf is a varargs 1537 // function. 1538 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 1539 if (ICE->getType() == S.Context.IntTy) 1540 if (ATR.matchesType(S.Context, ICE->getSubExpr()->getType())) 1541 return true; 1542 1543 // We may be able to offer a FixItHint if it is a supported type. 1544 FormatSpecifier fixedFS = FS; 1545 bool success = fixedFS.fixType(Ex->getType()); 1546 1547 if (success) { 1548 // Get the fix string from the fixed format specifier 1549 llvm::SmallString<128> buf; 1550 llvm::raw_svector_ostream os(buf); 1551 fixedFS.toString(os); 1552 1553 S.Diag(getLocationOfByte(CS.getStart()), 1554 diag::warn_printf_conversion_argument_type_mismatch) 1555 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1556 << getFormatSpecifierRange(startSpecifier, specifierLen) 1557 << Ex->getSourceRange() 1558 << FixItHint::CreateReplacement( 1559 getFormatSpecifierRange(startSpecifier, specifierLen), 1560 os.str()); 1561 } 1562 else { 1563 S.Diag(getLocationOfByte(CS.getStart()), 1564 diag::warn_printf_conversion_argument_type_mismatch) 1565 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1566 << getFormatSpecifierRange(startSpecifier, specifierLen) 1567 << Ex->getSourceRange(); 1568 } 1569 } 1570 1571 return true; 1572} 1573 1574void CheckPrintfHandler::DoneProcessing() { 1575 // Does the number of data arguments exceed the number of 1576 // format conversions in the format string? 1577 if (!HasVAListArg) { 1578 // Find any arguments that weren't covered. 1579 CoveredArgs.flip(); 1580 signed notCoveredArg = CoveredArgs.find_first(); 1581 if (notCoveredArg >= 0) { 1582 assert((unsigned)notCoveredArg < NumDataArgs); 1583 S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(), 1584 diag::warn_printf_data_arg_not_used) 1585 << getFormatStringRange(); 1586 } 1587 } 1588} 1589 1590void Sema::CheckPrintfString(const StringLiteral *FExpr, 1591 const Expr *OrigFormatExpr, 1592 const CallExpr *TheCall, bool HasVAListArg, 1593 unsigned format_idx, unsigned firstDataArg) { 1594 1595 // CHECK: is the format string a wide literal? 1596 if (FExpr->isWide()) { 1597 Diag(FExpr->getLocStart(), 1598 diag::warn_printf_format_string_is_wide_literal) 1599 << OrigFormatExpr->getSourceRange(); 1600 return; 1601 } 1602 1603 // Str - The format string. NOTE: this is NOT null-terminated! 1604 const char *Str = FExpr->getStrData(); 1605 1606 // CHECK: empty format string? 1607 unsigned StrLen = FExpr->getByteLength(); 1608 1609 if (StrLen == 0) { 1610 Diag(FExpr->getLocStart(), diag::warn_printf_empty_format_string) 1611 << OrigFormatExpr->getSourceRange(); 1612 return; 1613 } 1614 1615 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1616 TheCall->getNumArgs() - firstDataArg, 1617 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1618 HasVAListArg, TheCall, format_idx); 1619 1620 if (!analyze_printf::ParseFormatString(H, Str, Str + StrLen)) 1621 H.DoneProcessing(); 1622} 1623 1624//===--- CHECK: Return Address of Stack Variable --------------------------===// 1625 1626static DeclRefExpr* EvalVal(Expr *E); 1627static DeclRefExpr* EvalAddr(Expr* E); 1628 1629/// CheckReturnStackAddr - Check if a return statement returns the address 1630/// of a stack variable. 1631void 1632Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 1633 SourceLocation ReturnLoc) { 1634 1635 // Perform checking for returned stack addresses. 1636 if (lhsType->isPointerType() || lhsType->isBlockPointerType()) { 1637 if (DeclRefExpr *DR = EvalAddr(RetValExp)) 1638 Diag(DR->getLocStart(), diag::warn_ret_stack_addr) 1639 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1640 1641 // Skip over implicit cast expressions when checking for block expressions. 1642 RetValExp = RetValExp->IgnoreParenCasts(); 1643 1644 if (BlockExpr *C = dyn_cast<BlockExpr>(RetValExp)) 1645 if (C->hasBlockDeclRefExprs()) 1646 Diag(C->getLocStart(), diag::err_ret_local_block) 1647 << C->getSourceRange(); 1648 1649 if (AddrLabelExpr *ALE = dyn_cast<AddrLabelExpr>(RetValExp)) 1650 Diag(ALE->getLocStart(), diag::warn_ret_addr_label) 1651 << ALE->getSourceRange(); 1652 1653 } else if (lhsType->isReferenceType()) { 1654 // Perform checking for stack values returned by reference. 1655 // Check for a reference to the stack 1656 if (DeclRefExpr *DR = EvalVal(RetValExp)) 1657 Diag(DR->getLocStart(), diag::warn_ret_stack_ref) 1658 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1659 } 1660} 1661 1662/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 1663/// check if the expression in a return statement evaluates to an address 1664/// to a location on the stack. The recursion is used to traverse the 1665/// AST of the return expression, with recursion backtracking when we 1666/// encounter a subexpression that (1) clearly does not lead to the address 1667/// of a stack variable or (2) is something we cannot determine leads to 1668/// the address of a stack variable based on such local checking. 1669/// 1670/// EvalAddr processes expressions that are pointers that are used as 1671/// references (and not L-values). EvalVal handles all other values. 1672/// At the base case of the recursion is a check for a DeclRefExpr* in 1673/// the refers to a stack variable. 1674/// 1675/// This implementation handles: 1676/// 1677/// * pointer-to-pointer casts 1678/// * implicit conversions from array references to pointers 1679/// * taking the address of fields 1680/// * arbitrary interplay between "&" and "*" operators 1681/// * pointer arithmetic from an address of a stack variable 1682/// * taking the address of an array element where the array is on the stack 1683static DeclRefExpr* EvalAddr(Expr *E) { 1684 // We should only be called for evaluating pointer expressions. 1685 assert((E->getType()->isAnyPointerType() || 1686 E->getType()->isBlockPointerType() || 1687 E->getType()->isObjCQualifiedIdType()) && 1688 "EvalAddr only works on pointers"); 1689 1690 // Our "symbolic interpreter" is just a dispatch off the currently 1691 // viewed AST node. We then recursively traverse the AST by calling 1692 // EvalAddr and EvalVal appropriately. 1693 switch (E->getStmtClass()) { 1694 case Stmt::ParenExprClass: 1695 // Ignore parentheses. 1696 return EvalAddr(cast<ParenExpr>(E)->getSubExpr()); 1697 1698 case Stmt::UnaryOperatorClass: { 1699 // The only unary operator that make sense to handle here 1700 // is AddrOf. All others don't make sense as pointers. 1701 UnaryOperator *U = cast<UnaryOperator>(E); 1702 1703 if (U->getOpcode() == UnaryOperator::AddrOf) 1704 return EvalVal(U->getSubExpr()); 1705 else 1706 return NULL; 1707 } 1708 1709 case Stmt::BinaryOperatorClass: { 1710 // Handle pointer arithmetic. All other binary operators are not valid 1711 // in this context. 1712 BinaryOperator *B = cast<BinaryOperator>(E); 1713 BinaryOperator::Opcode op = B->getOpcode(); 1714 1715 if (op != BinaryOperator::Add && op != BinaryOperator::Sub) 1716 return NULL; 1717 1718 Expr *Base = B->getLHS(); 1719 1720 // Determine which argument is the real pointer base. It could be 1721 // the RHS argument instead of the LHS. 1722 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 1723 1724 assert (Base->getType()->isPointerType()); 1725 return EvalAddr(Base); 1726 } 1727 1728 // For conditional operators we need to see if either the LHS or RHS are 1729 // valid DeclRefExpr*s. If one of them is valid, we return it. 1730 case Stmt::ConditionalOperatorClass: { 1731 ConditionalOperator *C = cast<ConditionalOperator>(E); 1732 1733 // Handle the GNU extension for missing LHS. 1734 if (Expr *lhsExpr = C->getLHS()) 1735 if (DeclRefExpr* LHS = EvalAddr(lhsExpr)) 1736 return LHS; 1737 1738 return EvalAddr(C->getRHS()); 1739 } 1740 1741 // For casts, we need to handle conversions from arrays to 1742 // pointer values, and pointer-to-pointer conversions. 1743 case Stmt::ImplicitCastExprClass: 1744 case Stmt::CStyleCastExprClass: 1745 case Stmt::CXXFunctionalCastExprClass: { 1746 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 1747 QualType T = SubExpr->getType(); 1748 1749 if (SubExpr->getType()->isPointerType() || 1750 SubExpr->getType()->isBlockPointerType() || 1751 SubExpr->getType()->isObjCQualifiedIdType()) 1752 return EvalAddr(SubExpr); 1753 else if (T->isArrayType()) 1754 return EvalVal(SubExpr); 1755 else 1756 return 0; 1757 } 1758 1759 // C++ casts. For dynamic casts, static casts, and const casts, we 1760 // are always converting from a pointer-to-pointer, so we just blow 1761 // through the cast. In the case the dynamic cast doesn't fail (and 1762 // return NULL), we take the conservative route and report cases 1763 // where we return the address of a stack variable. For Reinterpre 1764 // FIXME: The comment about is wrong; we're not always converting 1765 // from pointer to pointer. I'm guessing that this code should also 1766 // handle references to objects. 1767 case Stmt::CXXStaticCastExprClass: 1768 case Stmt::CXXDynamicCastExprClass: 1769 case Stmt::CXXConstCastExprClass: 1770 case Stmt::CXXReinterpretCastExprClass: { 1771 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 1772 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 1773 return EvalAddr(S); 1774 else 1775 return NULL; 1776 } 1777 1778 // Everything else: we simply don't reason about them. 1779 default: 1780 return NULL; 1781 } 1782} 1783 1784 1785/// EvalVal - This function is complements EvalAddr in the mutual recursion. 1786/// See the comments for EvalAddr for more details. 1787static DeclRefExpr* EvalVal(Expr *E) { 1788 1789 // We should only be called for evaluating non-pointer expressions, or 1790 // expressions with a pointer type that are not used as references but instead 1791 // are l-values (e.g., DeclRefExpr with a pointer type). 1792 1793 // Our "symbolic interpreter" is just a dispatch off the currently 1794 // viewed AST node. We then recursively traverse the AST by calling 1795 // EvalAddr and EvalVal appropriately. 1796 switch (E->getStmtClass()) { 1797 case Stmt::DeclRefExprClass: { 1798 // DeclRefExpr: the base case. When we hit a DeclRefExpr we are looking 1799 // at code that refers to a variable's name. We check if it has local 1800 // storage within the function, and if so, return the expression. 1801 DeclRefExpr *DR = cast<DeclRefExpr>(E); 1802 1803 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 1804 if (V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR; 1805 1806 return NULL; 1807 } 1808 1809 case Stmt::ParenExprClass: 1810 // Ignore parentheses. 1811 return EvalVal(cast<ParenExpr>(E)->getSubExpr()); 1812 1813 case Stmt::UnaryOperatorClass: { 1814 // The only unary operator that make sense to handle here 1815 // is Deref. All others don't resolve to a "name." This includes 1816 // handling all sorts of rvalues passed to a unary operator. 1817 UnaryOperator *U = cast<UnaryOperator>(E); 1818 1819 if (U->getOpcode() == UnaryOperator::Deref) 1820 return EvalAddr(U->getSubExpr()); 1821 1822 return NULL; 1823 } 1824 1825 case Stmt::ArraySubscriptExprClass: { 1826 // Array subscripts are potential references to data on the stack. We 1827 // retrieve the DeclRefExpr* for the array variable if it indeed 1828 // has local storage. 1829 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase()); 1830 } 1831 1832 case Stmt::ConditionalOperatorClass: { 1833 // For conditional operators we need to see if either the LHS or RHS are 1834 // non-NULL DeclRefExpr's. If one is non-NULL, we return it. 1835 ConditionalOperator *C = cast<ConditionalOperator>(E); 1836 1837 // Handle the GNU extension for missing LHS. 1838 if (Expr *lhsExpr = C->getLHS()) 1839 if (DeclRefExpr *LHS = EvalVal(lhsExpr)) 1840 return LHS; 1841 1842 return EvalVal(C->getRHS()); 1843 } 1844 1845 // Accesses to members are potential references to data on the stack. 1846 case Stmt::MemberExprClass: { 1847 MemberExpr *M = cast<MemberExpr>(E); 1848 1849 // Check for indirect access. We only want direct field accesses. 1850 if (!M->isArrow()) 1851 return EvalVal(M->getBase()); 1852 else 1853 return NULL; 1854 } 1855 1856 // Everything else: we simply don't reason about them. 1857 default: 1858 return NULL; 1859 } 1860} 1861 1862//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 1863 1864/// Check for comparisons of floating point operands using != and ==. 1865/// Issue a warning if these are no self-comparisons, as they are not likely 1866/// to do what the programmer intended. 1867void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 1868 bool EmitWarning = true; 1869 1870 Expr* LeftExprSansParen = lex->IgnoreParens(); 1871 Expr* RightExprSansParen = rex->IgnoreParens(); 1872 1873 // Special case: check for x == x (which is OK). 1874 // Do not emit warnings for such cases. 1875 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 1876 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 1877 if (DRL->getDecl() == DRR->getDecl()) 1878 EmitWarning = false; 1879 1880 1881 // Special case: check for comparisons against literals that can be exactly 1882 // represented by APFloat. In such cases, do not emit a warning. This 1883 // is a heuristic: often comparison against such literals are used to 1884 // detect if a value in a variable has not changed. This clearly can 1885 // lead to false negatives. 1886 if (EmitWarning) { 1887 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 1888 if (FLL->isExact()) 1889 EmitWarning = false; 1890 } else 1891 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 1892 if (FLR->isExact()) 1893 EmitWarning = false; 1894 } 1895 } 1896 1897 // Check for comparisons with builtin types. 1898 if (EmitWarning) 1899 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 1900 if (CL->isBuiltinCall(Context)) 1901 EmitWarning = false; 1902 1903 if (EmitWarning) 1904 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 1905 if (CR->isBuiltinCall(Context)) 1906 EmitWarning = false; 1907 1908 // Emit the diagnostic. 1909 if (EmitWarning) 1910 Diag(loc, diag::warn_floatingpoint_eq) 1911 << lex->getSourceRange() << rex->getSourceRange(); 1912} 1913 1914//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 1915//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 1916 1917namespace { 1918 1919/// Structure recording the 'active' range of an integer-valued 1920/// expression. 1921struct IntRange { 1922 /// The number of bits active in the int. 1923 unsigned Width; 1924 1925 /// True if the int is known not to have negative values. 1926 bool NonNegative; 1927 1928 IntRange() {} 1929 IntRange(unsigned Width, bool NonNegative) 1930 : Width(Width), NonNegative(NonNegative) 1931 {} 1932 1933 // Returns the range of the bool type. 1934 static IntRange forBoolType() { 1935 return IntRange(1, true); 1936 } 1937 1938 // Returns the range of an integral type. 1939 static IntRange forType(ASTContext &C, QualType T) { 1940 return forCanonicalType(C, T->getCanonicalTypeInternal().getTypePtr()); 1941 } 1942 1943 // Returns the range of an integeral type based on its canonical 1944 // representation. 1945 static IntRange forCanonicalType(ASTContext &C, const Type *T) { 1946 assert(T->isCanonicalUnqualified()); 1947 1948 if (const VectorType *VT = dyn_cast<VectorType>(T)) 1949 T = VT->getElementType().getTypePtr(); 1950 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 1951 T = CT->getElementType().getTypePtr(); 1952 1953 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 1954 EnumDecl *Enum = ET->getDecl(); 1955 unsigned NumPositive = Enum->getNumPositiveBits(); 1956 unsigned NumNegative = Enum->getNumNegativeBits(); 1957 1958 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 1959 } 1960 1961 const BuiltinType *BT = cast<BuiltinType>(T); 1962 assert(BT->isInteger()); 1963 1964 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 1965 } 1966 1967 // Returns the supremum of two ranges: i.e. their conservative merge. 1968 static IntRange join(IntRange L, IntRange R) { 1969 return IntRange(std::max(L.Width, R.Width), 1970 L.NonNegative && R.NonNegative); 1971 } 1972 1973 // Returns the infinum of two ranges: i.e. their aggressive merge. 1974 static IntRange meet(IntRange L, IntRange R) { 1975 return IntRange(std::min(L.Width, R.Width), 1976 L.NonNegative || R.NonNegative); 1977 } 1978}; 1979 1980IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 1981 if (value.isSigned() && value.isNegative()) 1982 return IntRange(value.getMinSignedBits(), false); 1983 1984 if (value.getBitWidth() > MaxWidth) 1985 value.trunc(MaxWidth); 1986 1987 // isNonNegative() just checks the sign bit without considering 1988 // signedness. 1989 return IntRange(value.getActiveBits(), true); 1990} 1991 1992IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 1993 unsigned MaxWidth) { 1994 if (result.isInt()) 1995 return GetValueRange(C, result.getInt(), MaxWidth); 1996 1997 if (result.isVector()) { 1998 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 1999 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 2000 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 2001 R = IntRange::join(R, El); 2002 } 2003 return R; 2004 } 2005 2006 if (result.isComplexInt()) { 2007 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 2008 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 2009 return IntRange::join(R, I); 2010 } 2011 2012 // This can happen with lossless casts to intptr_t of "based" lvalues. 2013 // Assume it might use arbitrary bits. 2014 // FIXME: The only reason we need to pass the type in here is to get 2015 // the sign right on this one case. It would be nice if APValue 2016 // preserved this. 2017 assert(result.isLValue()); 2018 return IntRange(MaxWidth, Ty->isUnsignedIntegerType()); 2019} 2020 2021/// Pseudo-evaluate the given integer expression, estimating the 2022/// range of values it might take. 2023/// 2024/// \param MaxWidth - the width to which the value will be truncated 2025IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 2026 E = E->IgnoreParens(); 2027 2028 // Try a full evaluation first. 2029 Expr::EvalResult result; 2030 if (E->Evaluate(result, C)) 2031 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 2032 2033 // I think we only want to look through implicit casts here; if the 2034 // user has an explicit widening cast, we should treat the value as 2035 // being of the new, wider type. 2036 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 2037 if (CE->getCastKind() == CastExpr::CK_NoOp) 2038 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 2039 2040 IntRange OutputTypeRange = IntRange::forType(C, CE->getType()); 2041 2042 bool isIntegerCast = (CE->getCastKind() == CastExpr::CK_IntegralCast); 2043 if (!isIntegerCast && CE->getCastKind() == CastExpr::CK_Unknown) 2044 isIntegerCast = CE->getSubExpr()->getType()->isIntegerType(); 2045 2046 // Assume that non-integer casts can span the full range of the type. 2047 if (!isIntegerCast) 2048 return OutputTypeRange; 2049 2050 IntRange SubRange 2051 = GetExprRange(C, CE->getSubExpr(), 2052 std::min(MaxWidth, OutputTypeRange.Width)); 2053 2054 // Bail out if the subexpr's range is as wide as the cast type. 2055 if (SubRange.Width >= OutputTypeRange.Width) 2056 return OutputTypeRange; 2057 2058 // Otherwise, we take the smaller width, and we're non-negative if 2059 // either the output type or the subexpr is. 2060 return IntRange(SubRange.Width, 2061 SubRange.NonNegative || OutputTypeRange.NonNegative); 2062 } 2063 2064 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 2065 // If we can fold the condition, just take that operand. 2066 bool CondResult; 2067 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 2068 return GetExprRange(C, CondResult ? CO->getTrueExpr() 2069 : CO->getFalseExpr(), 2070 MaxWidth); 2071 2072 // Otherwise, conservatively merge. 2073 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 2074 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 2075 return IntRange::join(L, R); 2076 } 2077 2078 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 2079 switch (BO->getOpcode()) { 2080 2081 // Boolean-valued operations are single-bit and positive. 2082 case BinaryOperator::LAnd: 2083 case BinaryOperator::LOr: 2084 case BinaryOperator::LT: 2085 case BinaryOperator::GT: 2086 case BinaryOperator::LE: 2087 case BinaryOperator::GE: 2088 case BinaryOperator::EQ: 2089 case BinaryOperator::NE: 2090 return IntRange::forBoolType(); 2091 2092 // The type of these compound assignments is the type of the LHS, 2093 // so the RHS is not necessarily an integer. 2094 case BinaryOperator::MulAssign: 2095 case BinaryOperator::DivAssign: 2096 case BinaryOperator::RemAssign: 2097 case BinaryOperator::AddAssign: 2098 case BinaryOperator::SubAssign: 2099 return IntRange::forType(C, E->getType()); 2100 2101 // Operations with opaque sources are black-listed. 2102 case BinaryOperator::PtrMemD: 2103 case BinaryOperator::PtrMemI: 2104 return IntRange::forType(C, E->getType()); 2105 2106 // Bitwise-and uses the *infinum* of the two source ranges. 2107 case BinaryOperator::And: 2108 case BinaryOperator::AndAssign: 2109 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 2110 GetExprRange(C, BO->getRHS(), MaxWidth)); 2111 2112 // Left shift gets black-listed based on a judgement call. 2113 case BinaryOperator::Shl: 2114 // ...except that we want to treat '1 << (blah)' as logically 2115 // positive. It's an important idiom. 2116 if (IntegerLiteral *I 2117 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 2118 if (I->getValue() == 1) { 2119 IntRange R = IntRange::forType(C, E->getType()); 2120 return IntRange(R.Width, /*NonNegative*/ true); 2121 } 2122 } 2123 // fallthrough 2124 2125 case BinaryOperator::ShlAssign: 2126 return IntRange::forType(C, E->getType()); 2127 2128 // Right shift by a constant can narrow its left argument. 2129 case BinaryOperator::Shr: 2130 case BinaryOperator::ShrAssign: { 2131 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2132 2133 // If the shift amount is a positive constant, drop the width by 2134 // that much. 2135 llvm::APSInt shift; 2136 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 2137 shift.isNonNegative()) { 2138 unsigned zext = shift.getZExtValue(); 2139 if (zext >= L.Width) 2140 L.Width = (L.NonNegative ? 0 : 1); 2141 else 2142 L.Width -= zext; 2143 } 2144 2145 return L; 2146 } 2147 2148 // Comma acts as its right operand. 2149 case BinaryOperator::Comma: 2150 return GetExprRange(C, BO->getRHS(), MaxWidth); 2151 2152 // Black-list pointer subtractions. 2153 case BinaryOperator::Sub: 2154 if (BO->getLHS()->getType()->isPointerType()) 2155 return IntRange::forType(C, E->getType()); 2156 // fallthrough 2157 2158 default: 2159 break; 2160 } 2161 2162 // Treat every other operator as if it were closed on the 2163 // narrowest type that encompasses both operands. 2164 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2165 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 2166 return IntRange::join(L, R); 2167 } 2168 2169 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 2170 switch (UO->getOpcode()) { 2171 // Boolean-valued operations are white-listed. 2172 case UnaryOperator::LNot: 2173 return IntRange::forBoolType(); 2174 2175 // Operations with opaque sources are black-listed. 2176 case UnaryOperator::Deref: 2177 case UnaryOperator::AddrOf: // should be impossible 2178 case UnaryOperator::OffsetOf: 2179 return IntRange::forType(C, E->getType()); 2180 2181 default: 2182 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 2183 } 2184 } 2185 2186 if (dyn_cast<OffsetOfExpr>(E)) { 2187 IntRange::forType(C, E->getType()); 2188 } 2189 2190 FieldDecl *BitField = E->getBitField(); 2191 if (BitField) { 2192 llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C); 2193 unsigned BitWidth = BitWidthAP.getZExtValue(); 2194 2195 return IntRange(BitWidth, BitField->getType()->isUnsignedIntegerType()); 2196 } 2197 2198 return IntRange::forType(C, E->getType()); 2199} 2200 2201IntRange GetExprRange(ASTContext &C, Expr *E) { 2202 return GetExprRange(C, E, C.getIntWidth(E->getType())); 2203} 2204 2205/// Checks whether the given value, which currently has the given 2206/// source semantics, has the same value when coerced through the 2207/// target semantics. 2208bool IsSameFloatAfterCast(const llvm::APFloat &value, 2209 const llvm::fltSemantics &Src, 2210 const llvm::fltSemantics &Tgt) { 2211 llvm::APFloat truncated = value; 2212 2213 bool ignored; 2214 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 2215 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 2216 2217 return truncated.bitwiseIsEqual(value); 2218} 2219 2220/// Checks whether the given value, which currently has the given 2221/// source semantics, has the same value when coerced through the 2222/// target semantics. 2223/// 2224/// The value might be a vector of floats (or a complex number). 2225bool IsSameFloatAfterCast(const APValue &value, 2226 const llvm::fltSemantics &Src, 2227 const llvm::fltSemantics &Tgt) { 2228 if (value.isFloat()) 2229 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 2230 2231 if (value.isVector()) { 2232 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 2233 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 2234 return false; 2235 return true; 2236 } 2237 2238 assert(value.isComplexFloat()); 2239 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 2240 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 2241} 2242 2243void AnalyzeImplicitConversions(Sema &S, Expr *E); 2244 2245bool IsZero(Sema &S, Expr *E) { 2246 llvm::APSInt Value; 2247 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 2248} 2249 2250void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 2251 BinaryOperator::Opcode op = E->getOpcode(); 2252 if (op == BinaryOperator::LT && IsZero(S, E->getRHS())) { 2253 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2254 << "< 0" << "false" 2255 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2256 } else if (op == BinaryOperator::GE && IsZero(S, E->getRHS())) { 2257 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2258 << ">= 0" << "true" 2259 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2260 } else if (op == BinaryOperator::GT && IsZero(S, E->getLHS())) { 2261 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2262 << "0 >" << "false" 2263 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2264 } else if (op == BinaryOperator::LE && IsZero(S, E->getLHS())) { 2265 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2266 << "0 <=" << "true" 2267 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2268 } 2269} 2270 2271/// Analyze the operands of the given comparison. Implements the 2272/// fallback case from AnalyzeComparison. 2273void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 2274 AnalyzeImplicitConversions(S, E->getLHS()); 2275 AnalyzeImplicitConversions(S, E->getRHS()); 2276} 2277 2278/// \brief Implements -Wsign-compare. 2279/// 2280/// \param lex the left-hand expression 2281/// \param rex the right-hand expression 2282/// \param OpLoc the location of the joining operator 2283/// \param BinOpc binary opcode or 0 2284void AnalyzeComparison(Sema &S, BinaryOperator *E) { 2285 // The type the comparison is being performed in. 2286 QualType T = E->getLHS()->getType(); 2287 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 2288 && "comparison with mismatched types"); 2289 2290 // We don't do anything special if this isn't an unsigned integral 2291 // comparison: we're only interested in integral comparisons, and 2292 // signed comparisons only happen in cases we don't care to warn about. 2293 if (!T->isUnsignedIntegerType()) 2294 return AnalyzeImpConvsInComparison(S, E); 2295 2296 Expr *lex = E->getLHS()->IgnoreParenImpCasts(); 2297 Expr *rex = E->getRHS()->IgnoreParenImpCasts(); 2298 2299 // Check to see if one of the (unmodified) operands is of different 2300 // signedness. 2301 Expr *signedOperand, *unsignedOperand; 2302 if (lex->getType()->isSignedIntegerType()) { 2303 assert(!rex->getType()->isSignedIntegerType() && 2304 "unsigned comparison between two signed integer expressions?"); 2305 signedOperand = lex; 2306 unsignedOperand = rex; 2307 } else if (rex->getType()->isSignedIntegerType()) { 2308 signedOperand = rex; 2309 unsignedOperand = lex; 2310 } else { 2311 CheckTrivialUnsignedComparison(S, E); 2312 return AnalyzeImpConvsInComparison(S, E); 2313 } 2314 2315 // Otherwise, calculate the effective range of the signed operand. 2316 IntRange signedRange = GetExprRange(S.Context, signedOperand); 2317 2318 // Go ahead and analyze implicit conversions in the operands. Note 2319 // that we skip the implicit conversions on both sides. 2320 AnalyzeImplicitConversions(S, lex); 2321 AnalyzeImplicitConversions(S, rex); 2322 2323 // If the signed range is non-negative, -Wsign-compare won't fire, 2324 // but we should still check for comparisons which are always true 2325 // or false. 2326 if (signedRange.NonNegative) 2327 return CheckTrivialUnsignedComparison(S, E); 2328 2329 // For (in)equality comparisons, if the unsigned operand is a 2330 // constant which cannot collide with a overflowed signed operand, 2331 // then reinterpreting the signed operand as unsigned will not 2332 // change the result of the comparison. 2333 if (E->isEqualityOp()) { 2334 unsigned comparisonWidth = S.Context.getIntWidth(T); 2335 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 2336 2337 // We should never be unable to prove that the unsigned operand is 2338 // non-negative. 2339 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 2340 2341 if (unsignedRange.Width < comparisonWidth) 2342 return; 2343 } 2344 2345 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison) 2346 << lex->getType() << rex->getType() 2347 << lex->getSourceRange() << rex->getSourceRange(); 2348} 2349 2350/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 2351void DiagnoseImpCast(Sema &S, Expr *E, QualType T, unsigned diag) { 2352 S.Diag(E->getExprLoc(), diag) << E->getType() << T << E->getSourceRange(); 2353} 2354 2355void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 2356 bool *ICContext = 0) { 2357 if (E->isTypeDependent() || E->isValueDependent()) return; 2358 2359 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 2360 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 2361 if (Source == Target) return; 2362 if (Target->isDependentType()) return; 2363 2364 // Never diagnose implicit casts to bool. 2365 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 2366 return; 2367 2368 // Strip vector types. 2369 if (isa<VectorType>(Source)) { 2370 if (!isa<VectorType>(Target)) 2371 return DiagnoseImpCast(S, E, T, diag::warn_impcast_vector_scalar); 2372 2373 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 2374 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 2375 } 2376 2377 // Strip complex types. 2378 if (isa<ComplexType>(Source)) { 2379 if (!isa<ComplexType>(Target)) 2380 return DiagnoseImpCast(S, E, T, diag::warn_impcast_complex_scalar); 2381 2382 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 2383 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 2384 } 2385 2386 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 2387 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 2388 2389 // If the source is floating point... 2390 if (SourceBT && SourceBT->isFloatingPoint()) { 2391 // ...and the target is floating point... 2392 if (TargetBT && TargetBT->isFloatingPoint()) { 2393 // ...then warn if we're dropping FP rank. 2394 2395 // Builtin FP kinds are ordered by increasing FP rank. 2396 if (SourceBT->getKind() > TargetBT->getKind()) { 2397 // Don't warn about float constants that are precisely 2398 // representable in the target type. 2399 Expr::EvalResult result; 2400 if (E->Evaluate(result, S.Context)) { 2401 // Value might be a float, a float vector, or a float complex. 2402 if (IsSameFloatAfterCast(result.Val, 2403 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 2404 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 2405 return; 2406 } 2407 2408 DiagnoseImpCast(S, E, T, diag::warn_impcast_float_precision); 2409 } 2410 return; 2411 } 2412 2413 // If the target is integral, always warn. 2414 if ((TargetBT && TargetBT->isInteger())) 2415 // TODO: don't warn for integer values? 2416 DiagnoseImpCast(S, E, T, diag::warn_impcast_float_integer); 2417 2418 return; 2419 } 2420 2421 if (!Source->isIntegerType() || !Target->isIntegerType()) 2422 return; 2423 2424 IntRange SourceRange = GetExprRange(S.Context, E); 2425 IntRange TargetRange = IntRange::forCanonicalType(S.Context, Target); 2426 2427 if (SourceRange.Width > TargetRange.Width) { 2428 // People want to build with -Wshorten-64-to-32 and not -Wconversion 2429 // and by god we'll let them. 2430 if (SourceRange.Width == 64 && TargetRange.Width == 32) 2431 return DiagnoseImpCast(S, E, T, diag::warn_impcast_integer_64_32); 2432 return DiagnoseImpCast(S, E, T, diag::warn_impcast_integer_precision); 2433 } 2434 2435 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 2436 (!TargetRange.NonNegative && SourceRange.NonNegative && 2437 SourceRange.Width == TargetRange.Width)) { 2438 unsigned DiagID = diag::warn_impcast_integer_sign; 2439 2440 // Traditionally, gcc has warned about this under -Wsign-compare. 2441 // We also want to warn about it in -Wconversion. 2442 // So if -Wconversion is off, use a completely identical diagnostic 2443 // in the sign-compare group. 2444 // The conditional-checking code will 2445 if (ICContext) { 2446 DiagID = diag::warn_impcast_integer_sign_conditional; 2447 *ICContext = true; 2448 } 2449 2450 return DiagnoseImpCast(S, E, T, DiagID); 2451 } 2452 2453 return; 2454} 2455 2456void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 2457 2458void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 2459 bool &ICContext) { 2460 E = E->IgnoreParenImpCasts(); 2461 2462 if (isa<ConditionalOperator>(E)) 2463 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 2464 2465 AnalyzeImplicitConversions(S, E); 2466 if (E->getType() != T) 2467 return CheckImplicitConversion(S, E, T, &ICContext); 2468 return; 2469} 2470 2471void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 2472 AnalyzeImplicitConversions(S, E->getCond()); 2473 2474 bool Suspicious = false; 2475 CheckConditionalOperand(S, E->getTrueExpr(), T, Suspicious); 2476 CheckConditionalOperand(S, E->getFalseExpr(), T, Suspicious); 2477 2478 // If -Wconversion would have warned about either of the candidates 2479 // for a signedness conversion to the context type... 2480 if (!Suspicious) return; 2481 2482 // ...but it's currently ignored... 2483 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional)) 2484 return; 2485 2486 // ...and -Wsign-compare isn't... 2487 if (!S.Diags.getDiagnosticLevel(diag::warn_mixed_sign_conditional)) 2488 return; 2489 2490 // ...then check whether it would have warned about either of the 2491 // candidates for a signedness conversion to the condition type. 2492 if (E->getType() != T) { 2493 Suspicious = false; 2494 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 2495 E->getType(), &Suspicious); 2496 if (!Suspicious) 2497 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 2498 E->getType(), &Suspicious); 2499 if (!Suspicious) 2500 return; 2501 } 2502 2503 // If so, emit a diagnostic under -Wsign-compare. 2504 Expr *lex = E->getTrueExpr()->IgnoreParenImpCasts(); 2505 Expr *rex = E->getFalseExpr()->IgnoreParenImpCasts(); 2506 S.Diag(E->getQuestionLoc(), diag::warn_mixed_sign_conditional) 2507 << lex->getType() << rex->getType() 2508 << lex->getSourceRange() << rex->getSourceRange(); 2509} 2510 2511/// AnalyzeImplicitConversions - Find and report any interesting 2512/// implicit conversions in the given expression. There are a couple 2513/// of competing diagnostics here, -Wconversion and -Wsign-compare. 2514void AnalyzeImplicitConversions(Sema &S, Expr *OrigE) { 2515 QualType T = OrigE->getType(); 2516 Expr *E = OrigE->IgnoreParenImpCasts(); 2517 2518 // For conditional operators, we analyze the arguments as if they 2519 // were being fed directly into the output. 2520 if (isa<ConditionalOperator>(E)) { 2521 ConditionalOperator *CO = cast<ConditionalOperator>(E); 2522 CheckConditionalOperator(S, CO, T); 2523 return; 2524 } 2525 2526 // Go ahead and check any implicit conversions we might have skipped. 2527 // The non-canonical typecheck is just an optimization; 2528 // CheckImplicitConversion will filter out dead implicit conversions. 2529 if (E->getType() != T) 2530 CheckImplicitConversion(S, E, T); 2531 2532 // Now continue drilling into this expression. 2533 2534 // Skip past explicit casts. 2535 if (isa<ExplicitCastExpr>(E)) { 2536 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 2537 return AnalyzeImplicitConversions(S, E); 2538 } 2539 2540 // Do a somewhat different check with comparison operators. 2541 if (isa<BinaryOperator>(E) && cast<BinaryOperator>(E)->isComparisonOp()) 2542 return AnalyzeComparison(S, cast<BinaryOperator>(E)); 2543 2544 // These break the otherwise-useful invariant below. Fortunately, 2545 // we don't really need to recurse into them, because any internal 2546 // expressions should have been analyzed already when they were 2547 // built into statements. 2548 if (isa<StmtExpr>(E)) return; 2549 2550 // Don't descend into unevaluated contexts. 2551 if (isa<SizeOfAlignOfExpr>(E)) return; 2552 2553 // Now just recurse over the expression's children. 2554 for (Stmt::child_iterator I = E->child_begin(), IE = E->child_end(); 2555 I != IE; ++I) 2556 AnalyzeImplicitConversions(S, cast<Expr>(*I)); 2557} 2558 2559} // end anonymous namespace 2560 2561/// Diagnoses "dangerous" implicit conversions within the given 2562/// expression (which is a full expression). Implements -Wconversion 2563/// and -Wsign-compare. 2564void Sema::CheckImplicitConversions(Expr *E) { 2565 // Don't diagnose in unevaluated contexts. 2566 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 2567 return; 2568 2569 // Don't diagnose for value- or type-dependent expressions. 2570 if (E->isTypeDependent() || E->isValueDependent()) 2571 return; 2572 2573 AnalyzeImplicitConversions(*this, E); 2574} 2575 2576/// CheckParmsForFunctionDef - Check that the parameters of the given 2577/// function are appropriate for the definition of a function. This 2578/// takes care of any checks that cannot be performed on the 2579/// declaration itself, e.g., that the types of each of the function 2580/// parameters are complete. 2581bool Sema::CheckParmsForFunctionDef(FunctionDecl *FD) { 2582 bool HasInvalidParm = false; 2583 for (unsigned p = 0, NumParams = FD->getNumParams(); p < NumParams; ++p) { 2584 ParmVarDecl *Param = FD->getParamDecl(p); 2585 2586 // C99 6.7.5.3p4: the parameters in a parameter type list in a 2587 // function declarator that is part of a function definition of 2588 // that function shall not have incomplete type. 2589 // 2590 // This is also C++ [dcl.fct]p6. 2591 if (!Param->isInvalidDecl() && 2592 RequireCompleteType(Param->getLocation(), Param->getType(), 2593 diag::err_typecheck_decl_incomplete_type)) { 2594 Param->setInvalidDecl(); 2595 HasInvalidParm = true; 2596 } 2597 2598 // C99 6.9.1p5: If the declarator includes a parameter type list, the 2599 // declaration of each parameter shall include an identifier. 2600 if (Param->getIdentifier() == 0 && 2601 !Param->isImplicit() && 2602 !getLangOptions().CPlusPlus) 2603 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 2604 2605 // C99 6.7.5.3p12: 2606 // If the function declarator is not part of a definition of that 2607 // function, parameters may have incomplete type and may use the [*] 2608 // notation in their sequences of declarator specifiers to specify 2609 // variable length array types. 2610 QualType PType = Param->getOriginalType(); 2611 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 2612 if (AT->getSizeModifier() == ArrayType::Star) { 2613 // FIXME: This diagnosic should point the the '[*]' if source-location 2614 // information is added for it. 2615 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 2616 } 2617 } 2618 } 2619 2620 return HasInvalidParm; 2621} 2622