SemaChecking.cpp revision b7c21018ec1049580cf6df88db09e606550a7baa
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/FormatString.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 return SemaBuiltinAtomicOverloaded(move(TheCallResult)); 206 } 207 208 // Since the target specific builtins for each arch overlap, only check those 209 // of the arch we are compiling for. 210 if (BuiltinID >= Builtin::FirstTSBuiltin) { 211 switch (Context.Target.getTriple().getArch()) { 212 case llvm::Triple::arm: 213 case llvm::Triple::thumb: 214 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 215 return ExprError(); 216 break; 217 case llvm::Triple::x86: 218 case llvm::Triple::x86_64: 219 if (CheckX86BuiltinFunctionCall(BuiltinID, TheCall)) 220 return ExprError(); 221 break; 222 default: 223 break; 224 } 225 } 226 227 return move(TheCallResult); 228} 229 230bool Sema::CheckX86BuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 231 switch (BuiltinID) { 232 case X86::BI__builtin_ia32_palignr128: 233 case X86::BI__builtin_ia32_palignr: { 234 llvm::APSInt Result; 235 if (SemaBuiltinConstantArg(TheCall, 2, Result)) 236 return true; 237 break; 238 } 239 } 240 return false; 241} 242 243// Get the valid immediate range for the specified NEON type code. 244static unsigned RFT(unsigned t, bool shift = false) { 245 bool quad = t & 0x10; 246 247 switch (t & 0x7) { 248 case 0: // i8 249 return shift ? 7 : (8 << (int)quad) - 1; 250 case 1: // i16 251 return shift ? 15 : (4 << (int)quad) - 1; 252 case 2: // i32 253 return shift ? 31 : (2 << (int)quad) - 1; 254 case 3: // i64 255 return shift ? 63 : (1 << (int)quad) - 1; 256 case 4: // f32 257 assert(!shift && "cannot shift float types!"); 258 return (2 << (int)quad) - 1; 259 case 5: // poly8 260 assert(!shift && "cannot shift polynomial types!"); 261 return (8 << (int)quad) - 1; 262 case 6: // poly16 263 assert(!shift && "cannot shift polynomial types!"); 264 return (4 << (int)quad) - 1; 265 case 7: // float16 266 assert(!shift && "cannot shift float types!"); 267 return (4 << (int)quad) - 1; 268 } 269 return 0; 270} 271 272bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 273 llvm::APSInt Result; 274 275 unsigned mask = 0; 276 unsigned TV = 0; 277 switch (BuiltinID) { 278#define GET_NEON_OVERLOAD_CHECK 279#include "clang/Basic/arm_neon.inc" 280#undef GET_NEON_OVERLOAD_CHECK 281 } 282 283 // For NEON intrinsics which are overloaded on vector element type, validate 284 // the immediate which specifies which variant to emit. 285 if (mask) { 286 unsigned ArgNo = TheCall->getNumArgs()-1; 287 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 288 return true; 289 290 TV = Result.getLimitedValue(32); 291 if ((TV > 31) || (mask & (1 << TV)) == 0) 292 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 293 << TheCall->getArg(ArgNo)->getSourceRange(); 294 } 295 296 // For NEON intrinsics which take an immediate value as part of the 297 // instruction, range check them here. 298 unsigned i = 0, l = 0, u = 0; 299 switch (BuiltinID) { 300 default: return false; 301#define GET_NEON_IMMEDIATE_CHECK 302#include "clang/Basic/arm_neon.inc" 303#undef GET_NEON_IMMEDIATE_CHECK 304 }; 305 306 // Check that the immediate argument is actually a constant. 307 if (SemaBuiltinConstantArg(TheCall, i, Result)) 308 return true; 309 310 // Range check against the upper/lower values for this isntruction. 311 unsigned Val = Result.getZExtValue(); 312 if (Val < l || Val > (u + l)) 313 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 314 << llvm::utostr(l) << llvm::utostr(u+l) 315 << TheCall->getArg(i)->getSourceRange(); 316 317 return false; 318} 319 320/// CheckFunctionCall - Check a direct function call for various correctness 321/// and safety properties not strictly enforced by the C type system. 322bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 323 // Get the IdentifierInfo* for the called function. 324 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 325 326 // None of the checks below are needed for functions that don't have 327 // simple names (e.g., C++ conversion functions). 328 if (!FnInfo) 329 return false; 330 331 // FIXME: This mechanism should be abstracted to be less fragile and 332 // more efficient. For example, just map function ids to custom 333 // handlers. 334 335 // Printf checking. 336 if (const FormatAttr *Format = FDecl->getAttr<FormatAttr>()) { 337 const bool b = Format->getType() == "scanf"; 338 if (b || CheckablePrintfAttr(Format, TheCall)) { 339 bool HasVAListArg = Format->getFirstArg() == 0; 340 CheckPrintfScanfArguments(TheCall, HasVAListArg, 341 Format->getFormatIdx() - 1, 342 HasVAListArg ? 0 : Format->getFirstArg() - 1, 343 !b); 344 } 345 } 346 347 for (const NonNullAttr *NonNull = FDecl->getAttr<NonNullAttr>(); NonNull; 348 NonNull = NonNull->getNext<NonNullAttr>()) 349 CheckNonNullArguments(NonNull, TheCall); 350 351 return false; 352} 353 354bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 355 // Printf checking. 356 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 357 if (!Format) 358 return false; 359 360 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 361 if (!V) 362 return false; 363 364 QualType Ty = V->getType(); 365 if (!Ty->isBlockPointerType()) 366 return false; 367 368 const bool b = Format->getType() == "scanf"; 369 if (!b && !CheckablePrintfAttr(Format, TheCall)) 370 return false; 371 372 bool HasVAListArg = Format->getFirstArg() == 0; 373 CheckPrintfScanfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 374 HasVAListArg ? 0 : Format->getFirstArg() - 1, !b); 375 376 return false; 377} 378 379/// SemaBuiltinAtomicOverloaded - We have a call to a function like 380/// __sync_fetch_and_add, which is an overloaded function based on the pointer 381/// type of its first argument. The main ActOnCallExpr routines have already 382/// promoted the types of arguments because all of these calls are prototyped as 383/// void(...). 384/// 385/// This function goes through and does final semantic checking for these 386/// builtins, 387Sema::OwningExprResult 388Sema::SemaBuiltinAtomicOverloaded(OwningExprResult TheCallResult) { 389 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 390 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 391 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 392 393 // Ensure that we have at least one argument to do type inference from. 394 if (TheCall->getNumArgs() < 1) { 395 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 396 << 0 << 1 << TheCall->getNumArgs() 397 << TheCall->getCallee()->getSourceRange(); 398 return ExprError(); 399 } 400 401 // Inspect the first argument of the atomic builtin. This should always be 402 // a pointer type, whose element is an integral scalar or pointer type. 403 // Because it is a pointer type, we don't have to worry about any implicit 404 // casts here. 405 // FIXME: We don't allow floating point scalars as input. 406 Expr *FirstArg = TheCall->getArg(0); 407 if (!FirstArg->getType()->isPointerType()) { 408 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 409 << FirstArg->getType() << FirstArg->getSourceRange(); 410 return ExprError(); 411 } 412 413 QualType ValType = 414 FirstArg->getType()->getAs<PointerType>()->getPointeeType(); 415 if (!ValType->isIntegerType() && !ValType->isPointerType() && 416 !ValType->isBlockPointerType()) { 417 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 418 << FirstArg->getType() << FirstArg->getSourceRange(); 419 return ExprError(); 420 } 421 422 // We need to figure out which concrete builtin this maps onto. For example, 423 // __sync_fetch_and_add with a 2 byte object turns into 424 // __sync_fetch_and_add_2. 425#define BUILTIN_ROW(x) \ 426 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 427 Builtin::BI##x##_8, Builtin::BI##x##_16 } 428 429 static const unsigned BuiltinIndices[][5] = { 430 BUILTIN_ROW(__sync_fetch_and_add), 431 BUILTIN_ROW(__sync_fetch_and_sub), 432 BUILTIN_ROW(__sync_fetch_and_or), 433 BUILTIN_ROW(__sync_fetch_and_and), 434 BUILTIN_ROW(__sync_fetch_and_xor), 435 436 BUILTIN_ROW(__sync_add_and_fetch), 437 BUILTIN_ROW(__sync_sub_and_fetch), 438 BUILTIN_ROW(__sync_and_and_fetch), 439 BUILTIN_ROW(__sync_or_and_fetch), 440 BUILTIN_ROW(__sync_xor_and_fetch), 441 442 BUILTIN_ROW(__sync_val_compare_and_swap), 443 BUILTIN_ROW(__sync_bool_compare_and_swap), 444 BUILTIN_ROW(__sync_lock_test_and_set), 445 BUILTIN_ROW(__sync_lock_release) 446 }; 447#undef BUILTIN_ROW 448 449 // Determine the index of the size. 450 unsigned SizeIndex; 451 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 452 case 1: SizeIndex = 0; break; 453 case 2: SizeIndex = 1; break; 454 case 4: SizeIndex = 2; break; 455 case 8: SizeIndex = 3; break; 456 case 16: SizeIndex = 4; break; 457 default: 458 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 459 << FirstArg->getType() << FirstArg->getSourceRange(); 460 return ExprError(); 461 } 462 463 // Each of these builtins has one pointer argument, followed by some number of 464 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 465 // that we ignore. Find out which row of BuiltinIndices to read from as well 466 // as the number of fixed args. 467 unsigned BuiltinID = FDecl->getBuiltinID(); 468 unsigned BuiltinIndex, NumFixed = 1; 469 switch (BuiltinID) { 470 default: assert(0 && "Unknown overloaded atomic builtin!"); 471 case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break; 472 case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break; 473 case Builtin::BI__sync_fetch_and_or: BuiltinIndex = 2; break; 474 case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break; 475 case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break; 476 477 case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break; 478 case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break; 479 case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break; 480 case Builtin::BI__sync_or_and_fetch: BuiltinIndex = 8; break; 481 case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break; 482 483 case Builtin::BI__sync_val_compare_and_swap: 484 BuiltinIndex = 10; 485 NumFixed = 2; 486 break; 487 case Builtin::BI__sync_bool_compare_and_swap: 488 BuiltinIndex = 11; 489 NumFixed = 2; 490 break; 491 case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break; 492 case Builtin::BI__sync_lock_release: 493 BuiltinIndex = 13; 494 NumFixed = 0; 495 break; 496 } 497 498 // Now that we know how many fixed arguments we expect, first check that we 499 // have at least that many. 500 if (TheCall->getNumArgs() < 1+NumFixed) { 501 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 502 << 0 << 1+NumFixed << TheCall->getNumArgs() 503 << TheCall->getCallee()->getSourceRange(); 504 return ExprError(); 505 } 506 507 // Get the decl for the concrete builtin from this, we can tell what the 508 // concrete integer type we should convert to is. 509 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 510 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 511 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 512 FunctionDecl *NewBuiltinDecl = 513 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 514 TUScope, false, DRE->getLocStart())); 515 const FunctionProtoType *BuiltinFT = 516 NewBuiltinDecl->getType()->getAs<FunctionProtoType>(); 517 518 QualType OrigValType = ValType; 519 ValType = BuiltinFT->getArgType(0)->getAs<PointerType>()->getPointeeType(); 520 521 // If the first type needs to be converted (e.g. void** -> int*), do it now. 522 if (BuiltinFT->getArgType(0) != FirstArg->getType()) { 523 ImpCastExprToType(FirstArg, BuiltinFT->getArgType(0), CastExpr::CK_BitCast); 524 TheCall->setArg(0, FirstArg); 525 } 526 527 // Next, walk the valid ones promoting to the right type. 528 for (unsigned i = 0; i != NumFixed; ++i) { 529 Expr *Arg = TheCall->getArg(i+1); 530 531 // If the argument is an implicit cast, then there was a promotion due to 532 // "...", just remove it now. 533 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) { 534 Arg = ICE->getSubExpr(); 535 ICE->setSubExpr(0); 536 ICE->Destroy(Context); 537 TheCall->setArg(i+1, Arg); 538 } 539 540 // GCC does an implicit conversion to the pointer or integer ValType. This 541 // can fail in some cases (1i -> int**), check for this error case now. 542 CastExpr::CastKind Kind = CastExpr::CK_Unknown; 543 CXXBaseSpecifierArray BasePath; 544 if (CheckCastTypes(Arg->getSourceRange(), ValType, Arg, Kind, BasePath)) 545 return ExprError(); 546 547 // Okay, we have something that *can* be converted to the right type. Check 548 // to see if there is a potentially weird extension going on here. This can 549 // happen when you do an atomic operation on something like an char* and 550 // pass in 42. The 42 gets converted to char. This is even more strange 551 // for things like 45.123 -> char, etc. 552 // FIXME: Do this check. 553 ImpCastExprToType(Arg, ValType, Kind); 554 TheCall->setArg(i+1, Arg); 555 } 556 557 // Switch the DeclRefExpr to refer to the new decl. 558 DRE->setDecl(NewBuiltinDecl); 559 DRE->setType(NewBuiltinDecl->getType()); 560 561 // Set the callee in the CallExpr. 562 // FIXME: This leaks the original parens and implicit casts. 563 Expr *PromotedCall = DRE; 564 UsualUnaryConversions(PromotedCall); 565 TheCall->setCallee(PromotedCall); 566 567 // Change the result type of the call to match the result type of the decl. 568 TheCall->setType(NewBuiltinDecl->getCallResultType()); 569 570 // If the value type was converted to an integer when processing the 571 // arguments (e.g. void* -> int), we need to convert the result back. 572 if (!Context.hasSameUnqualifiedType(ValType, OrigValType)) { 573 Expr *E = TheCallResult.takeAs<Expr>(); 574 575 assert(ValType->isIntegerType() && 576 "We always convert atomic operation values to integers."); 577 // FIXME: Handle floating point value type here too. 578 CastExpr::CastKind Kind; 579 if (OrigValType->isIntegerType()) 580 Kind = CastExpr::CK_IntegralCast; 581 else if (OrigValType->hasPointerRepresentation()) 582 Kind = CastExpr::CK_IntegralToPointer; 583 else 584 llvm_unreachable("Unhandled original value type!"); 585 586 ImpCastExprToType(E, OrigValType, Kind); 587 return Owned(E); 588 } 589 590 return move(TheCallResult); 591} 592 593 594/// CheckObjCString - Checks that the argument to the builtin 595/// CFString constructor is correct 596/// FIXME: GCC currently emits the following warning: 597/// "warning: input conversion stopped due to an input byte that does not 598/// belong to the input codeset UTF-8" 599/// Note: It might also make sense to do the UTF-16 conversion here (would 600/// simplify the backend). 601bool Sema::CheckObjCString(Expr *Arg) { 602 Arg = Arg->IgnoreParenCasts(); 603 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 604 605 if (!Literal || Literal->isWide()) { 606 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 607 << Arg->getSourceRange(); 608 return true; 609 } 610 611 const char *Data = Literal->getStrData(); 612 unsigned Length = Literal->getByteLength(); 613 614 for (unsigned i = 0; i < Length; ++i) { 615 if (!Data[i]) { 616 Diag(getLocationOfStringLiteralByte(Literal, i), 617 diag::warn_cfstring_literal_contains_nul_character) 618 << Arg->getSourceRange(); 619 break; 620 } 621 } 622 623 return false; 624} 625 626/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 627/// Emit an error and return true on failure, return false on success. 628bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 629 Expr *Fn = TheCall->getCallee(); 630 if (TheCall->getNumArgs() > 2) { 631 Diag(TheCall->getArg(2)->getLocStart(), 632 diag::err_typecheck_call_too_many_args) 633 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 634 << Fn->getSourceRange() 635 << SourceRange(TheCall->getArg(2)->getLocStart(), 636 (*(TheCall->arg_end()-1))->getLocEnd()); 637 return true; 638 } 639 640 if (TheCall->getNumArgs() < 2) { 641 return Diag(TheCall->getLocEnd(), 642 diag::err_typecheck_call_too_few_args_at_least) 643 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 644 } 645 646 // Determine whether the current function is variadic or not. 647 BlockScopeInfo *CurBlock = getCurBlock(); 648 bool isVariadic; 649 if (CurBlock) 650 isVariadic = CurBlock->TheDecl->isVariadic(); 651 else if (FunctionDecl *FD = getCurFunctionDecl()) 652 isVariadic = FD->isVariadic(); 653 else 654 isVariadic = getCurMethodDecl()->isVariadic(); 655 656 if (!isVariadic) { 657 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 658 return true; 659 } 660 661 // Verify that the second argument to the builtin is the last argument of the 662 // current function or method. 663 bool SecondArgIsLastNamedArgument = false; 664 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 665 666 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 667 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 668 // FIXME: This isn't correct for methods (results in bogus warning). 669 // Get the last formal in the current function. 670 const ParmVarDecl *LastArg; 671 if (CurBlock) 672 LastArg = *(CurBlock->TheDecl->param_end()-1); 673 else if (FunctionDecl *FD = getCurFunctionDecl()) 674 LastArg = *(FD->param_end()-1); 675 else 676 LastArg = *(getCurMethodDecl()->param_end()-1); 677 SecondArgIsLastNamedArgument = PV == LastArg; 678 } 679 } 680 681 if (!SecondArgIsLastNamedArgument) 682 Diag(TheCall->getArg(1)->getLocStart(), 683 diag::warn_second_parameter_of_va_start_not_last_named_argument); 684 return false; 685} 686 687/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 688/// friends. This is declared to take (...), so we have to check everything. 689bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 690 if (TheCall->getNumArgs() < 2) 691 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 692 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 693 if (TheCall->getNumArgs() > 2) 694 return Diag(TheCall->getArg(2)->getLocStart(), 695 diag::err_typecheck_call_too_many_args) 696 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 697 << SourceRange(TheCall->getArg(2)->getLocStart(), 698 (*(TheCall->arg_end()-1))->getLocEnd()); 699 700 Expr *OrigArg0 = TheCall->getArg(0); 701 Expr *OrigArg1 = TheCall->getArg(1); 702 703 // Do standard promotions between the two arguments, returning their common 704 // type. 705 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 706 707 // Make sure any conversions are pushed back into the call; this is 708 // type safe since unordered compare builtins are declared as "_Bool 709 // foo(...)". 710 TheCall->setArg(0, OrigArg0); 711 TheCall->setArg(1, OrigArg1); 712 713 if (OrigArg0->isTypeDependent() || OrigArg1->isTypeDependent()) 714 return false; 715 716 // If the common type isn't a real floating type, then the arguments were 717 // invalid for this operation. 718 if (!Res->isRealFloatingType()) 719 return Diag(OrigArg0->getLocStart(), 720 diag::err_typecheck_call_invalid_ordered_compare) 721 << OrigArg0->getType() << OrigArg1->getType() 722 << SourceRange(OrigArg0->getLocStart(), OrigArg1->getLocEnd()); 723 724 return false; 725} 726 727/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 728/// __builtin_isnan and friends. This is declared to take (...), so we have 729/// to check everything. We expect the last argument to be a floating point 730/// value. 731bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 732 if (TheCall->getNumArgs() < NumArgs) 733 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 734 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 735 if (TheCall->getNumArgs() > NumArgs) 736 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 737 diag::err_typecheck_call_too_many_args) 738 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 739 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 740 (*(TheCall->arg_end()-1))->getLocEnd()); 741 742 Expr *OrigArg = TheCall->getArg(NumArgs-1); 743 744 if (OrigArg->isTypeDependent()) 745 return false; 746 747 // This operation requires a non-_Complex floating-point number. 748 if (!OrigArg->getType()->isRealFloatingType()) 749 return Diag(OrigArg->getLocStart(), 750 diag::err_typecheck_call_invalid_unary_fp) 751 << OrigArg->getType() << OrigArg->getSourceRange(); 752 753 // If this is an implicit conversion from float -> double, remove it. 754 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 755 Expr *CastArg = Cast->getSubExpr(); 756 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 757 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 758 "promotion from float to double is the only expected cast here"); 759 Cast->setSubExpr(0); 760 Cast->Destroy(Context); 761 TheCall->setArg(NumArgs-1, CastArg); 762 OrigArg = CastArg; 763 } 764 } 765 766 return false; 767} 768 769/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 770// This is declared to take (...), so we have to check everything. 771Action::OwningExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 772 if (TheCall->getNumArgs() < 2) 773 return ExprError(Diag(TheCall->getLocEnd(), 774 diag::err_typecheck_call_too_few_args_at_least) 775 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 776 << TheCall->getSourceRange()); 777 778 // Determine which of the following types of shufflevector we're checking: 779 // 1) unary, vector mask: (lhs, mask) 780 // 2) binary, vector mask: (lhs, rhs, mask) 781 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 782 QualType resType = TheCall->getArg(0)->getType(); 783 unsigned numElements = 0; 784 785 if (!TheCall->getArg(0)->isTypeDependent() && 786 !TheCall->getArg(1)->isTypeDependent()) { 787 QualType LHSType = TheCall->getArg(0)->getType(); 788 QualType RHSType = TheCall->getArg(1)->getType(); 789 790 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 791 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 792 << SourceRange(TheCall->getArg(0)->getLocStart(), 793 TheCall->getArg(1)->getLocEnd()); 794 return ExprError(); 795 } 796 797 numElements = LHSType->getAs<VectorType>()->getNumElements(); 798 unsigned numResElements = TheCall->getNumArgs() - 2; 799 800 // Check to see if we have a call with 2 vector arguments, the unary shuffle 801 // with mask. If so, verify that RHS is an integer vector type with the 802 // same number of elts as lhs. 803 if (TheCall->getNumArgs() == 2) { 804 if (!RHSType->isIntegerType() || 805 RHSType->getAs<VectorType>()->getNumElements() != numElements) 806 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 807 << SourceRange(TheCall->getArg(1)->getLocStart(), 808 TheCall->getArg(1)->getLocEnd()); 809 numResElements = numElements; 810 } 811 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 812 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 813 << SourceRange(TheCall->getArg(0)->getLocStart(), 814 TheCall->getArg(1)->getLocEnd()); 815 return ExprError(); 816 } else if (numElements != numResElements) { 817 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 818 resType = Context.getVectorType(eltType, numResElements, 819 VectorType::NotAltiVec); 820 } 821 } 822 823 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 824 if (TheCall->getArg(i)->isTypeDependent() || 825 TheCall->getArg(i)->isValueDependent()) 826 continue; 827 828 llvm::APSInt Result(32); 829 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 830 return ExprError(Diag(TheCall->getLocStart(), 831 diag::err_shufflevector_nonconstant_argument) 832 << TheCall->getArg(i)->getSourceRange()); 833 834 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 835 return ExprError(Diag(TheCall->getLocStart(), 836 diag::err_shufflevector_argument_too_large) 837 << TheCall->getArg(i)->getSourceRange()); 838 } 839 840 llvm::SmallVector<Expr*, 32> exprs; 841 842 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 843 exprs.push_back(TheCall->getArg(i)); 844 TheCall->setArg(i, 0); 845 } 846 847 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 848 exprs.size(), resType, 849 TheCall->getCallee()->getLocStart(), 850 TheCall->getRParenLoc())); 851} 852 853/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 854// This is declared to take (const void*, ...) and can take two 855// optional constant int args. 856bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 857 unsigned NumArgs = TheCall->getNumArgs(); 858 859 if (NumArgs > 3) 860 return Diag(TheCall->getLocEnd(), 861 diag::err_typecheck_call_too_many_args_at_most) 862 << 0 /*function call*/ << 3 << NumArgs 863 << TheCall->getSourceRange(); 864 865 // Argument 0 is checked for us and the remaining arguments must be 866 // constant integers. 867 for (unsigned i = 1; i != NumArgs; ++i) { 868 Expr *Arg = TheCall->getArg(i); 869 870 llvm::APSInt Result; 871 if (SemaBuiltinConstantArg(TheCall, i, Result)) 872 return true; 873 874 // FIXME: gcc issues a warning and rewrites these to 0. These 875 // seems especially odd for the third argument since the default 876 // is 3. 877 if (i == 1) { 878 if (Result.getLimitedValue() > 1) 879 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 880 << "0" << "1" << Arg->getSourceRange(); 881 } else { 882 if (Result.getLimitedValue() > 3) 883 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 884 << "0" << "3" << Arg->getSourceRange(); 885 } 886 } 887 888 return false; 889} 890 891/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 892/// TheCall is a constant expression. 893bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 894 llvm::APSInt &Result) { 895 Expr *Arg = TheCall->getArg(ArgNum); 896 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 897 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 898 899 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 900 901 if (!Arg->isIntegerConstantExpr(Result, Context)) 902 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 903 << FDecl->getDeclName() << Arg->getSourceRange(); 904 905 return false; 906} 907 908/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 909/// int type). This simply type checks that type is one of the defined 910/// constants (0-3). 911// For compatability check 0-3, llvm only handles 0 and 2. 912bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 913 llvm::APSInt Result; 914 915 // Check constant-ness first. 916 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 917 return true; 918 919 Expr *Arg = TheCall->getArg(1); 920 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 921 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 922 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 923 } 924 925 return false; 926} 927 928/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 929/// This checks that val is a constant 1. 930bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 931 Expr *Arg = TheCall->getArg(1); 932 llvm::APSInt Result; 933 934 // TODO: This is less than ideal. Overload this to take a value. 935 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 936 return true; 937 938 if (Result != 1) 939 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 940 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 941 942 return false; 943} 944 945// Handle i > 1 ? "x" : "y", recursivelly 946bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 947 bool HasVAListArg, 948 unsigned format_idx, unsigned firstDataArg, 949 bool isPrintf) { 950 951 if (E->isTypeDependent() || E->isValueDependent()) 952 return false; 953 954 switch (E->getStmtClass()) { 955 case Stmt::ConditionalOperatorClass: { 956 const ConditionalOperator *C = cast<ConditionalOperator>(E); 957 return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, HasVAListArg, 958 format_idx, firstDataArg, isPrintf) 959 && SemaCheckStringLiteral(C->getRHS(), TheCall, HasVAListArg, 960 format_idx, firstDataArg, isPrintf); 961 } 962 963 case Stmt::ImplicitCastExprClass: { 964 const ImplicitCastExpr *Expr = cast<ImplicitCastExpr>(E); 965 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 966 format_idx, firstDataArg, isPrintf); 967 } 968 969 case Stmt::ParenExprClass: { 970 const ParenExpr *Expr = cast<ParenExpr>(E); 971 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 972 format_idx, firstDataArg, isPrintf); 973 } 974 975 case Stmt::DeclRefExprClass: { 976 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 977 978 // As an exception, do not flag errors for variables binding to 979 // const string literals. 980 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 981 bool isConstant = false; 982 QualType T = DR->getType(); 983 984 if (const ArrayType *AT = Context.getAsArrayType(T)) { 985 isConstant = AT->getElementType().isConstant(Context); 986 } else if (const PointerType *PT = T->getAs<PointerType>()) { 987 isConstant = T.isConstant(Context) && 988 PT->getPointeeType().isConstant(Context); 989 } 990 991 if (isConstant) { 992 if (const Expr *Init = VD->getAnyInitializer()) 993 return SemaCheckStringLiteral(Init, TheCall, 994 HasVAListArg, format_idx, firstDataArg, 995 isPrintf); 996 } 997 998 // For vprintf* functions (i.e., HasVAListArg==true), we add a 999 // special check to see if the format string is a function parameter 1000 // of the function calling the printf function. If the function 1001 // has an attribute indicating it is a printf-like function, then we 1002 // should suppress warnings concerning non-literals being used in a call 1003 // to a vprintf function. For example: 1004 // 1005 // void 1006 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1007 // va_list ap; 1008 // va_start(ap, fmt); 1009 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1010 // ... 1011 // 1012 // 1013 // FIXME: We don't have full attribute support yet, so just check to see 1014 // if the argument is a DeclRefExpr that references a parameter. We'll 1015 // add proper support for checking the attribute later. 1016 if (HasVAListArg) 1017 if (isa<ParmVarDecl>(VD)) 1018 return true; 1019 } 1020 1021 return false; 1022 } 1023 1024 case Stmt::CallExprClass: { 1025 const CallExpr *CE = cast<CallExpr>(E); 1026 if (const ImplicitCastExpr *ICE 1027 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 1028 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 1029 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 1030 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 1031 unsigned ArgIndex = FA->getFormatIdx(); 1032 const Expr *Arg = CE->getArg(ArgIndex - 1); 1033 1034 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 1035 format_idx, firstDataArg, isPrintf); 1036 } 1037 } 1038 } 1039 } 1040 1041 return false; 1042 } 1043 case Stmt::ObjCStringLiteralClass: 1044 case Stmt::StringLiteralClass: { 1045 const StringLiteral *StrE = NULL; 1046 1047 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1048 StrE = ObjCFExpr->getString(); 1049 else 1050 StrE = cast<StringLiteral>(E); 1051 1052 if (StrE) { 1053 CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx, 1054 firstDataArg, isPrintf); 1055 return true; 1056 } 1057 1058 return false; 1059 } 1060 1061 default: 1062 return false; 1063 } 1064} 1065 1066void 1067Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1068 const CallExpr *TheCall) { 1069 for (NonNullAttr::iterator i = NonNull->begin(), e = NonNull->end(); 1070 i != e; ++i) { 1071 const Expr *ArgExpr = TheCall->getArg(*i); 1072 if (ArgExpr->isNullPointerConstant(Context, 1073 Expr::NPC_ValueDependentIsNotNull)) 1074 Diag(TheCall->getCallee()->getLocStart(), diag::warn_null_arg) 1075 << ArgExpr->getSourceRange(); 1076 } 1077} 1078 1079/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar 1080/// functions) for correct use of format strings. 1081void 1082Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg, 1083 unsigned format_idx, unsigned firstDataArg, 1084 bool isPrintf) { 1085 1086 const Expr *Fn = TheCall->getCallee(); 1087 1088 // The way the format attribute works in GCC, the implicit this argument 1089 // of member functions is counted. However, it doesn't appear in our own 1090 // lists, so decrement format_idx in that case. 1091 if (isa<CXXMemberCallExpr>(TheCall)) { 1092 // Catch a format attribute mistakenly referring to the object argument. 1093 if (format_idx == 0) 1094 return; 1095 --format_idx; 1096 if(firstDataArg != 0) 1097 --firstDataArg; 1098 } 1099 1100 // CHECK: printf/scanf-like function is called with no format string. 1101 if (format_idx >= TheCall->getNumArgs()) { 1102 Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string) 1103 << Fn->getSourceRange(); 1104 return; 1105 } 1106 1107 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 1108 1109 // CHECK: format string is not a string literal. 1110 // 1111 // Dynamically generated format strings are difficult to 1112 // automatically vet at compile time. Requiring that format strings 1113 // are string literals: (1) permits the checking of format strings by 1114 // the compiler and thereby (2) can practically remove the source of 1115 // many format string exploits. 1116 1117 // Format string can be either ObjC string (e.g. @"%d") or 1118 // C string (e.g. "%d") 1119 // ObjC string uses the same format specifiers as C string, so we can use 1120 // the same format string checking logic for both ObjC and C strings. 1121 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 1122 firstDataArg, isPrintf)) 1123 return; // Literal format string found, check done! 1124 1125 // If there are no arguments specified, warn with -Wformat-security, otherwise 1126 // warn only with -Wformat-nonliteral. 1127 if (TheCall->getNumArgs() == format_idx+1) 1128 Diag(TheCall->getArg(format_idx)->getLocStart(), 1129 diag::warn_format_nonliteral_noargs) 1130 << OrigFormatExpr->getSourceRange(); 1131 else 1132 Diag(TheCall->getArg(format_idx)->getLocStart(), 1133 diag::warn_format_nonliteral) 1134 << OrigFormatExpr->getSourceRange(); 1135} 1136 1137namespace { 1138class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1139protected: 1140 Sema &S; 1141 const StringLiteral *FExpr; 1142 const Expr *OrigFormatExpr; 1143 const unsigned FirstDataArg; 1144 const unsigned NumDataArgs; 1145 const bool IsObjCLiteral; 1146 const char *Beg; // Start of format string. 1147 const bool HasVAListArg; 1148 const CallExpr *TheCall; 1149 unsigned FormatIdx; 1150 llvm::BitVector CoveredArgs; 1151 bool usesPositionalArgs; 1152 bool atFirstArg; 1153public: 1154 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1155 const Expr *origFormatExpr, unsigned firstDataArg, 1156 unsigned numDataArgs, bool isObjCLiteral, 1157 const char *beg, bool hasVAListArg, 1158 const CallExpr *theCall, unsigned formatIdx) 1159 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1160 FirstDataArg(firstDataArg), 1161 NumDataArgs(numDataArgs), 1162 IsObjCLiteral(isObjCLiteral), Beg(beg), 1163 HasVAListArg(hasVAListArg), 1164 TheCall(theCall), FormatIdx(formatIdx), 1165 usesPositionalArgs(false), atFirstArg(true) { 1166 CoveredArgs.resize(numDataArgs); 1167 CoveredArgs.reset(); 1168 } 1169 1170 void DoneProcessing(); 1171 1172 void HandleIncompleteSpecifier(const char *startSpecifier, 1173 unsigned specifierLen); 1174 1175 virtual void HandleInvalidPosition(const char *startSpecifier, 1176 unsigned specifierLen, 1177 analyze_format_string::PositionContext p); 1178 1179 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1180 1181 void HandleNullChar(const char *nullCharacter); 1182 1183protected: 1184 SourceRange getFormatStringRange(); 1185 CharSourceRange getSpecifierRange(const char *startSpecifier, 1186 unsigned specifierLen); 1187 SourceLocation getLocationOfByte(const char *x); 1188 1189 const Expr *getDataArg(unsigned i) const; 1190}; 1191} 1192 1193SourceRange CheckFormatHandler::getFormatStringRange() { 1194 return OrigFormatExpr->getSourceRange(); 1195} 1196 1197CharSourceRange CheckFormatHandler:: 1198getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1199 SourceLocation Start = getLocationOfByte(startSpecifier); 1200 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1201 1202 // Advance the end SourceLocation by one due to half-open ranges. 1203 End = End.getFileLocWithOffset(1); 1204 1205 return CharSourceRange::getCharRange(Start, End); 1206} 1207 1208SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 1209 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1210} 1211 1212void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 1213 unsigned specifierLen){ 1214 SourceLocation Loc = getLocationOfByte(startSpecifier); 1215 S.Diag(Loc, diag::warn_printf_incomplete_specifier) 1216 << getSpecifierRange(startSpecifier, specifierLen); 1217} 1218 1219void 1220CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1221 analyze_format_string::PositionContext p) { 1222 SourceLocation Loc = getLocationOfByte(startPos); 1223 S.Diag(Loc, diag::warn_format_invalid_positional_specifier) 1224 << (unsigned) p << getSpecifierRange(startPos, posLen); 1225} 1226 1227void CheckFormatHandler::HandleZeroPosition(const char *startPos, 1228 unsigned posLen) { 1229 SourceLocation Loc = getLocationOfByte(startPos); 1230 S.Diag(Loc, diag::warn_format_zero_positional_specifier) 1231 << getSpecifierRange(startPos, posLen); 1232} 1233 1234void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 1235 // The presence of a null character is likely an error. 1236 S.Diag(getLocationOfByte(nullCharacter), 1237 diag::warn_printf_format_string_contains_null_char) 1238 << getFormatStringRange(); 1239} 1240 1241const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 1242 return TheCall->getArg(FirstDataArg + i); 1243} 1244 1245void CheckFormatHandler::DoneProcessing() { 1246 // Does the number of data arguments exceed the number of 1247 // format conversions in the format string? 1248 if (!HasVAListArg) { 1249 // Find any arguments that weren't covered. 1250 CoveredArgs.flip(); 1251 signed notCoveredArg = CoveredArgs.find_first(); 1252 if (notCoveredArg >= 0) { 1253 assert((unsigned)notCoveredArg < NumDataArgs); 1254 S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(), 1255 diag::warn_printf_data_arg_not_used) 1256 << getFormatStringRange(); 1257 } 1258 } 1259} 1260 1261//===--- CHECK: Printf format string checking ------------------------------===// 1262 1263namespace { 1264class CheckPrintfHandler : public CheckFormatHandler { 1265public: 1266 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1267 const Expr *origFormatExpr, unsigned firstDataArg, 1268 unsigned numDataArgs, bool isObjCLiteral, 1269 const char *beg, bool hasVAListArg, 1270 const CallExpr *theCall, unsigned formatIdx) 1271 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1272 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1273 theCall, formatIdx) {} 1274 1275 1276 bool HandleInvalidPrintfConversionSpecifier( 1277 const analyze_printf::PrintfSpecifier &FS, 1278 const char *startSpecifier, 1279 unsigned specifierLen); 1280 1281 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 1282 const char *startSpecifier, 1283 unsigned specifierLen); 1284 1285 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 1286 const char *startSpecifier, unsigned specifierLen); 1287 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 1288 const analyze_printf::OptionalAmount &Amt, 1289 unsigned type, 1290 const char *startSpecifier, unsigned specifierLen); 1291 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1292 const analyze_printf::OptionalFlag &flag, 1293 const char *startSpecifier, unsigned specifierLen); 1294 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 1295 const analyze_printf::OptionalFlag &ignoredFlag, 1296 const analyze_printf::OptionalFlag &flag, 1297 const char *startSpecifier, unsigned specifierLen); 1298}; 1299} 1300 1301bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 1302 const analyze_printf::PrintfSpecifier &FS, 1303 const char *startSpecifier, 1304 unsigned specifierLen) { 1305 1306 unsigned argIndex = FS.getArgIndex(); 1307 bool keepGoing = true; 1308 if (argIndex < NumDataArgs) { 1309 // Consider the argument coverered, even though the specifier doesn't 1310 // make sense. 1311 CoveredArgs.set(argIndex); 1312 } 1313 else { 1314 // If argIndex exceeds the number of data arguments we 1315 // don't issue a warning because that is just a cascade of warnings (and 1316 // they may have intended '%%' anyway). We don't want to continue processing 1317 // the format string after this point, however, as we will like just get 1318 // gibberish when trying to match arguments. 1319 keepGoing = false; 1320 } 1321 1322 const analyze_printf::ConversionSpecifier &CS = 1323 FS.getConversionSpecifier(); 1324 SourceLocation Loc = getLocationOfByte(CS.getStart()); 1325 S.Diag(Loc, diag::warn_printf_invalid_conversion) 1326 << llvm::StringRef(CS.getStart(), CS.getLength()) 1327 << getSpecifierRange(startSpecifier, specifierLen); 1328 1329 return keepGoing; 1330} 1331 1332bool CheckPrintfHandler::HandleAmount( 1333 const analyze_format_string::OptionalAmount &Amt, 1334 unsigned k, const char *startSpecifier, 1335 unsigned specifierLen) { 1336 1337 if (Amt.hasDataArgument()) { 1338 if (!HasVAListArg) { 1339 unsigned argIndex = Amt.getArgIndex(); 1340 if (argIndex >= NumDataArgs) { 1341 S.Diag(getLocationOfByte(Amt.getStart()), 1342 diag::warn_printf_asterisk_missing_arg) 1343 << k << getSpecifierRange(startSpecifier, specifierLen); 1344 // Don't do any more checking. We will just emit 1345 // spurious errors. 1346 return false; 1347 } 1348 1349 // Type check the data argument. It should be an 'int'. 1350 // Although not in conformance with C99, we also allow the argument to be 1351 // an 'unsigned int' as that is a reasonably safe case. GCC also 1352 // doesn't emit a warning for that case. 1353 CoveredArgs.set(argIndex); 1354 const Expr *Arg = getDataArg(argIndex); 1355 QualType T = Arg->getType(); 1356 1357 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1358 assert(ATR.isValid()); 1359 1360 if (!ATR.matchesType(S.Context, T)) { 1361 S.Diag(getLocationOfByte(Amt.getStart()), 1362 diag::warn_printf_asterisk_wrong_type) 1363 << k 1364 << ATR.getRepresentativeType(S.Context) << T 1365 << getSpecifierRange(startSpecifier, specifierLen) 1366 << Arg->getSourceRange(); 1367 // Don't do any more checking. We will just emit 1368 // spurious errors. 1369 return false; 1370 } 1371 } 1372 } 1373 return true; 1374} 1375 1376void CheckPrintfHandler::HandleInvalidAmount( 1377 const analyze_printf::PrintfSpecifier &FS, 1378 const analyze_printf::OptionalAmount &Amt, 1379 unsigned type, 1380 const char *startSpecifier, 1381 unsigned specifierLen) { 1382 const analyze_printf::ConversionSpecifier &CS = FS.getConversionSpecifier(); 1383 switch (Amt.getHowSpecified()) { 1384 case analyze_printf::OptionalAmount::Constant: 1385 S.Diag(getLocationOfByte(Amt.getStart()), 1386 diag::warn_printf_nonsensical_optional_amount) 1387 << type 1388 << CS.toString() 1389 << getSpecifierRange(startSpecifier, specifierLen) 1390 << FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 1391 Amt.getConstantLength())); 1392 break; 1393 1394 default: 1395 S.Diag(getLocationOfByte(Amt.getStart()), 1396 diag::warn_printf_nonsensical_optional_amount) 1397 << type 1398 << CS.toString() 1399 << getSpecifierRange(startSpecifier, specifierLen); 1400 break; 1401 } 1402} 1403 1404void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1405 const analyze_printf::OptionalFlag &flag, 1406 const char *startSpecifier, 1407 unsigned specifierLen) { 1408 // Warn about pointless flag with a fixit removal. 1409 const analyze_printf::ConversionSpecifier &CS = FS.getConversionSpecifier(); 1410 S.Diag(getLocationOfByte(flag.getPosition()), 1411 diag::warn_printf_nonsensical_flag) 1412 << flag.toString() << CS.toString() 1413 << getSpecifierRange(startSpecifier, specifierLen) 1414 << FixItHint::CreateRemoval(getSpecifierRange(flag.getPosition(), 1)); 1415} 1416 1417void CheckPrintfHandler::HandleIgnoredFlag( 1418 const analyze_printf::PrintfSpecifier &FS, 1419 const analyze_printf::OptionalFlag &ignoredFlag, 1420 const analyze_printf::OptionalFlag &flag, 1421 const char *startSpecifier, 1422 unsigned specifierLen) { 1423 // Warn about ignored flag with a fixit removal. 1424 S.Diag(getLocationOfByte(ignoredFlag.getPosition()), 1425 diag::warn_printf_ignored_flag) 1426 << ignoredFlag.toString() << flag.toString() 1427 << getSpecifierRange(startSpecifier, specifierLen) 1428 << FixItHint::CreateRemoval(getSpecifierRange( 1429 ignoredFlag.getPosition(), 1)); 1430} 1431 1432bool 1433CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 1434 &FS, 1435 const char *startSpecifier, 1436 unsigned specifierLen) { 1437 1438 using namespace analyze_printf; 1439 const ConversionSpecifier &CS = FS.getConversionSpecifier(); 1440 1441 if (atFirstArg) { 1442 atFirstArg = false; 1443 usesPositionalArgs = FS.usesPositionalArg(); 1444 } 1445 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1446 // Cannot mix-and-match positional and non-positional arguments. 1447 S.Diag(getLocationOfByte(CS.getStart()), 1448 diag::warn_format_mix_positional_nonpositional_args) 1449 << getSpecifierRange(startSpecifier, specifierLen); 1450 return false; 1451 } 1452 1453 // First check if the field width, precision, and conversion specifier 1454 // have matching data arguments. 1455 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 1456 startSpecifier, specifierLen)) { 1457 return false; 1458 } 1459 1460 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 1461 startSpecifier, specifierLen)) { 1462 return false; 1463 } 1464 1465 if (!CS.consumesDataArgument()) { 1466 // FIXME: Technically specifying a precision or field width here 1467 // makes no sense. Worth issuing a warning at some point. 1468 return true; 1469 } 1470 1471 // Consume the argument. 1472 unsigned argIndex = FS.getArgIndex(); 1473 if (argIndex < NumDataArgs) { 1474 // The check to see if the argIndex is valid will come later. 1475 // We set the bit here because we may exit early from this 1476 // function if we encounter some other error. 1477 CoveredArgs.set(argIndex); 1478 } 1479 1480 // Check for using an Objective-C specific conversion specifier 1481 // in a non-ObjC literal. 1482 if (!IsObjCLiteral && CS.isObjCArg()) { 1483 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 1484 specifierLen); 1485 } 1486 1487 // Check for invalid use of field width 1488 if (!FS.hasValidFieldWidth()) { 1489 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 1490 startSpecifier, specifierLen); 1491 } 1492 1493 // Check for invalid use of precision 1494 if (!FS.hasValidPrecision()) { 1495 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 1496 startSpecifier, specifierLen); 1497 } 1498 1499 // Check each flag does not conflict with any other component. 1500 if (!FS.hasValidLeadingZeros()) 1501 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 1502 if (!FS.hasValidPlusPrefix()) 1503 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 1504 if (!FS.hasValidSpacePrefix()) 1505 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 1506 if (!FS.hasValidAlternativeForm()) 1507 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 1508 if (!FS.hasValidLeftJustified()) 1509 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 1510 1511 // Check that flags are not ignored by another flag 1512 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 1513 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 1514 startSpecifier, specifierLen); 1515 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 1516 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 1517 startSpecifier, specifierLen); 1518 1519 // Check the length modifier is valid with the given conversion specifier. 1520 const LengthModifier &LM = FS.getLengthModifier(); 1521 if (!FS.hasValidLengthModifier()) 1522 S.Diag(getLocationOfByte(LM.getStart()), 1523 diag::warn_printf_nonsensical_length) 1524 << LM.toString() << CS.toString() 1525 << getSpecifierRange(startSpecifier, specifierLen) 1526 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 1527 LM.getLength())); 1528 1529 // Are we using '%n'? 1530 if (CS.getKind() == ConversionSpecifier::OutIntPtrArg) { 1531 // Issue a warning about this being a possible security issue. 1532 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back) 1533 << getSpecifierRange(startSpecifier, specifierLen); 1534 // Continue checking the other format specifiers. 1535 return true; 1536 } 1537 1538 // The remaining checks depend on the data arguments. 1539 if (HasVAListArg) 1540 return true; 1541 1542 if (argIndex >= NumDataArgs) { 1543 if (FS.usesPositionalArg()) { 1544 S.Diag(getLocationOfByte(CS.getStart()), 1545 diag::warn_printf_positional_arg_exceeds_data_args) 1546 << (argIndex+1) << NumDataArgs 1547 << getSpecifierRange(startSpecifier, specifierLen); 1548 } 1549 else { 1550 S.Diag(getLocationOfByte(CS.getStart()), 1551 diag::warn_printf_insufficient_data_args) 1552 << getSpecifierRange(startSpecifier, specifierLen); 1553 } 1554 1555 // Don't do any more checking. 1556 return false; 1557 } 1558 1559 // Now type check the data expression that matches the 1560 // format specifier. 1561 const Expr *Ex = getDataArg(argIndex); 1562 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context); 1563 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 1564 // Check if we didn't match because of an implicit cast from a 'char' 1565 // or 'short' to an 'int'. This is done because printf is a varargs 1566 // function. 1567 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 1568 if (ICE->getType() == S.Context.IntTy) 1569 if (ATR.matchesType(S.Context, ICE->getSubExpr()->getType())) 1570 return true; 1571 1572 // We may be able to offer a FixItHint if it is a supported type. 1573 PrintfSpecifier fixedFS = FS; 1574 bool success = fixedFS.fixType(Ex->getType()); 1575 1576 if (success) { 1577 // Get the fix string from the fixed format specifier 1578 llvm::SmallString<128> buf; 1579 llvm::raw_svector_ostream os(buf); 1580 fixedFS.toString(os); 1581 1582 S.Diag(getLocationOfByte(CS.getStart()), 1583 diag::warn_printf_conversion_argument_type_mismatch) 1584 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1585 << getSpecifierRange(startSpecifier, specifierLen) 1586 << Ex->getSourceRange() 1587 << FixItHint::CreateReplacement( 1588 getSpecifierRange(startSpecifier, specifierLen), 1589 os.str()); 1590 } 1591 else { 1592 S.Diag(getLocationOfByte(CS.getStart()), 1593 diag::warn_printf_conversion_argument_type_mismatch) 1594 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1595 << getSpecifierRange(startSpecifier, specifierLen) 1596 << Ex->getSourceRange(); 1597 } 1598 } 1599 1600 return true; 1601} 1602 1603//===--- CHECK: Scanf format string checking ------------------------------===// 1604 1605namespace { 1606class CheckScanfHandler : public CheckFormatHandler { 1607public: 1608 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 1609 const Expr *origFormatExpr, unsigned firstDataArg, 1610 unsigned numDataArgs, bool isObjCLiteral, 1611 const char *beg, bool hasVAListArg, 1612 const CallExpr *theCall, unsigned formatIdx) 1613 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1614 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1615 theCall, formatIdx) {} 1616 1617 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 1618 const char *startSpecifier, 1619 unsigned specifierLen); 1620 1621 void HandleIncompleteScanList(const char *start, const char *end); 1622}; 1623} 1624 1625void CheckScanfHandler::HandleIncompleteScanList(const char *start, 1626 const char *end) { 1627 S.Diag(getLocationOfByte(end), diag::warn_scanf_scanlist_incomplete) 1628 << getSpecifierRange(start, end - start); 1629} 1630 1631bool CheckScanfHandler::HandleScanfSpecifier( 1632 const analyze_scanf::ScanfSpecifier &FS, 1633 const char *startSpecifier, 1634 unsigned specifierLen) { 1635 1636 using namespace analyze_scanf; 1637 using namespace analyze_format_string; 1638 1639 const ConversionSpecifier &CS = FS.getConversionSpecifier(); 1640 1641 // FIXME: Handle case where '%' and '*' don't consume an argument. 1642 // This needs to be done for the printf case as well. 1643 if (atFirstArg) { 1644 atFirstArg = false; 1645 usesPositionalArgs = FS.usesPositionalArg(); 1646 } 1647 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1648 // Cannot mix-and-match positional and non-positional arguments. 1649 S.Diag(getLocationOfByte(CS.getStart()), 1650 diag::warn_format_mix_positional_nonpositional_args) 1651 << getSpecifierRange(startSpecifier, specifierLen); 1652 return false; 1653 } 1654 1655 // Check if the field with is non-zero. 1656 const OptionalAmount &Amt = FS.getFieldWidth(); 1657 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 1658 if (Amt.getConstantAmount() == 0) { 1659 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 1660 Amt.getConstantLength()); 1661 S.Diag(getLocationOfByte(Amt.getStart()), 1662 diag::warn_scanf_nonzero_width) 1663 << R << FixItHint::CreateRemoval(R); 1664 } 1665 } 1666 1667 if (!FS.consumesDataArgument()) { 1668 // FIXME: Technically specifying a precision or field width here 1669 // makes no sense. Worth issuing a warning at some point. 1670 return true; 1671 } 1672 1673 // Consume the argument. 1674 unsigned argIndex = FS.getArgIndex(); 1675 if (argIndex < NumDataArgs) { 1676 // The check to see if the argIndex is valid will come later. 1677 // We set the bit here because we may exit early from this 1678 // function if we encounter some other error. 1679 CoveredArgs.set(argIndex); 1680 } 1681 1682 // FIXME: Check that the length modifier is valid with the given 1683 // conversion specifier. 1684 1685 // The remaining checks depend on the data arguments. 1686 if (HasVAListArg) 1687 return true; 1688 1689 if (argIndex >= NumDataArgs) { 1690 if (FS.usesPositionalArg()) { 1691 S.Diag(getLocationOfByte(CS.getStart()), 1692 diag::warn_printf_positional_arg_exceeds_data_args) 1693 << (argIndex+1) << NumDataArgs 1694 << getSpecifierRange(startSpecifier, specifierLen); 1695 } 1696 else { 1697 S.Diag(getLocationOfByte(CS.getStart()), 1698 diag::warn_printf_insufficient_data_args) 1699 << getSpecifierRange(startSpecifier, specifierLen); 1700 } 1701 1702 // Don't do any more checking. 1703 return false; 1704 } 1705 1706 // FIXME: Check that the argument type matches the format specifier. 1707 1708 return true; 1709} 1710 1711void Sema::CheckFormatString(const StringLiteral *FExpr, 1712 const Expr *OrigFormatExpr, 1713 const CallExpr *TheCall, bool HasVAListArg, 1714 unsigned format_idx, unsigned firstDataArg, 1715 bool isPrintf) { 1716 1717 // CHECK: is the format string a wide literal? 1718 if (FExpr->isWide()) { 1719 Diag(FExpr->getLocStart(), 1720 diag::warn_format_string_is_wide_literal) 1721 << OrigFormatExpr->getSourceRange(); 1722 return; 1723 } 1724 1725 // Str - The format string. NOTE: this is NOT null-terminated! 1726 const char *Str = FExpr->getStrData(); 1727 1728 // CHECK: empty format string? 1729 unsigned StrLen = FExpr->getByteLength(); 1730 1731 if (StrLen == 0) { 1732 Diag(FExpr->getLocStart(), diag::warn_empty_format_string) 1733 << OrigFormatExpr->getSourceRange(); 1734 return; 1735 } 1736 1737 if (isPrintf) { 1738 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1739 TheCall->getNumArgs() - firstDataArg, 1740 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1741 HasVAListArg, TheCall, format_idx); 1742 1743 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen)) 1744 H.DoneProcessing(); 1745 } 1746 else { 1747 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1748 TheCall->getNumArgs() - firstDataArg, 1749 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1750 HasVAListArg, TheCall, format_idx); 1751 1752 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen)) 1753 H.DoneProcessing(); 1754 } 1755} 1756 1757//===--- CHECK: Return Address of Stack Variable --------------------------===// 1758 1759static DeclRefExpr* EvalVal(Expr *E); 1760static DeclRefExpr* EvalAddr(Expr* E); 1761 1762/// CheckReturnStackAddr - Check if a return statement returns the address 1763/// of a stack variable. 1764void 1765Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 1766 SourceLocation ReturnLoc) { 1767 1768 // Perform checking for returned stack addresses. 1769 if (lhsType->isPointerType() || lhsType->isBlockPointerType()) { 1770 if (DeclRefExpr *DR = EvalAddr(RetValExp)) 1771 Diag(DR->getLocStart(), diag::warn_ret_stack_addr) 1772 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1773 1774 // Skip over implicit cast expressions when checking for block expressions. 1775 RetValExp = RetValExp->IgnoreParenCasts(); 1776 1777 if (BlockExpr *C = dyn_cast<BlockExpr>(RetValExp)) 1778 if (C->hasBlockDeclRefExprs()) 1779 Diag(C->getLocStart(), diag::err_ret_local_block) 1780 << C->getSourceRange(); 1781 1782 if (AddrLabelExpr *ALE = dyn_cast<AddrLabelExpr>(RetValExp)) 1783 Diag(ALE->getLocStart(), diag::warn_ret_addr_label) 1784 << ALE->getSourceRange(); 1785 1786 } else if (lhsType->isReferenceType()) { 1787 // Perform checking for stack values returned by reference. 1788 // Check for a reference to the stack 1789 if (DeclRefExpr *DR = EvalVal(RetValExp)) 1790 Diag(DR->getLocStart(), diag::warn_ret_stack_ref) 1791 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1792 } 1793} 1794 1795/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 1796/// check if the expression in a return statement evaluates to an address 1797/// to a location on the stack. The recursion is used to traverse the 1798/// AST of the return expression, with recursion backtracking when we 1799/// encounter a subexpression that (1) clearly does not lead to the address 1800/// of a stack variable or (2) is something we cannot determine leads to 1801/// the address of a stack variable based on such local checking. 1802/// 1803/// EvalAddr processes expressions that are pointers that are used as 1804/// references (and not L-values). EvalVal handles all other values. 1805/// At the base case of the recursion is a check for a DeclRefExpr* in 1806/// the refers to a stack variable. 1807/// 1808/// This implementation handles: 1809/// 1810/// * pointer-to-pointer casts 1811/// * implicit conversions from array references to pointers 1812/// * taking the address of fields 1813/// * arbitrary interplay between "&" and "*" operators 1814/// * pointer arithmetic from an address of a stack variable 1815/// * taking the address of an array element where the array is on the stack 1816static DeclRefExpr* EvalAddr(Expr *E) { 1817 // We should only be called for evaluating pointer expressions. 1818 assert((E->getType()->isAnyPointerType() || 1819 E->getType()->isBlockPointerType() || 1820 E->getType()->isObjCQualifiedIdType()) && 1821 "EvalAddr only works on pointers"); 1822 1823 // Our "symbolic interpreter" is just a dispatch off the currently 1824 // viewed AST node. We then recursively traverse the AST by calling 1825 // EvalAddr and EvalVal appropriately. 1826 switch (E->getStmtClass()) { 1827 case Stmt::ParenExprClass: 1828 // Ignore parentheses. 1829 return EvalAddr(cast<ParenExpr>(E)->getSubExpr()); 1830 1831 case Stmt::UnaryOperatorClass: { 1832 // The only unary operator that make sense to handle here 1833 // is AddrOf. All others don't make sense as pointers. 1834 UnaryOperator *U = cast<UnaryOperator>(E); 1835 1836 if (U->getOpcode() == UnaryOperator::AddrOf) 1837 return EvalVal(U->getSubExpr()); 1838 else 1839 return NULL; 1840 } 1841 1842 case Stmt::BinaryOperatorClass: { 1843 // Handle pointer arithmetic. All other binary operators are not valid 1844 // in this context. 1845 BinaryOperator *B = cast<BinaryOperator>(E); 1846 BinaryOperator::Opcode op = B->getOpcode(); 1847 1848 if (op != BinaryOperator::Add && op != BinaryOperator::Sub) 1849 return NULL; 1850 1851 Expr *Base = B->getLHS(); 1852 1853 // Determine which argument is the real pointer base. It could be 1854 // the RHS argument instead of the LHS. 1855 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 1856 1857 assert (Base->getType()->isPointerType()); 1858 return EvalAddr(Base); 1859 } 1860 1861 // For conditional operators we need to see if either the LHS or RHS are 1862 // valid DeclRefExpr*s. If one of them is valid, we return it. 1863 case Stmt::ConditionalOperatorClass: { 1864 ConditionalOperator *C = cast<ConditionalOperator>(E); 1865 1866 // Handle the GNU extension for missing LHS. 1867 if (Expr *lhsExpr = C->getLHS()) 1868 if (DeclRefExpr* LHS = EvalAddr(lhsExpr)) 1869 return LHS; 1870 1871 return EvalAddr(C->getRHS()); 1872 } 1873 1874 // For casts, we need to handle conversions from arrays to 1875 // pointer values, and pointer-to-pointer conversions. 1876 case Stmt::ImplicitCastExprClass: 1877 case Stmt::CStyleCastExprClass: 1878 case Stmt::CXXFunctionalCastExprClass: { 1879 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 1880 QualType T = SubExpr->getType(); 1881 1882 if (SubExpr->getType()->isPointerType() || 1883 SubExpr->getType()->isBlockPointerType() || 1884 SubExpr->getType()->isObjCQualifiedIdType()) 1885 return EvalAddr(SubExpr); 1886 else if (T->isArrayType()) 1887 return EvalVal(SubExpr); 1888 else 1889 return 0; 1890 } 1891 1892 // C++ casts. For dynamic casts, static casts, and const casts, we 1893 // are always converting from a pointer-to-pointer, so we just blow 1894 // through the cast. In the case the dynamic cast doesn't fail (and 1895 // return NULL), we take the conservative route and report cases 1896 // where we return the address of a stack variable. For Reinterpre 1897 // FIXME: The comment about is wrong; we're not always converting 1898 // from pointer to pointer. I'm guessing that this code should also 1899 // handle references to objects. 1900 case Stmt::CXXStaticCastExprClass: 1901 case Stmt::CXXDynamicCastExprClass: 1902 case Stmt::CXXConstCastExprClass: 1903 case Stmt::CXXReinterpretCastExprClass: { 1904 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 1905 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 1906 return EvalAddr(S); 1907 else 1908 return NULL; 1909 } 1910 1911 // Everything else: we simply don't reason about them. 1912 default: 1913 return NULL; 1914 } 1915} 1916 1917 1918/// EvalVal - This function is complements EvalAddr in the mutual recursion. 1919/// See the comments for EvalAddr for more details. 1920static DeclRefExpr* EvalVal(Expr *E) { 1921 1922 // We should only be called for evaluating non-pointer expressions, or 1923 // expressions with a pointer type that are not used as references but instead 1924 // are l-values (e.g., DeclRefExpr with a pointer type). 1925 1926 // Our "symbolic interpreter" is just a dispatch off the currently 1927 // viewed AST node. We then recursively traverse the AST by calling 1928 // EvalAddr and EvalVal appropriately. 1929 switch (E->getStmtClass()) { 1930 case Stmt::DeclRefExprClass: { 1931 // DeclRefExpr: the base case. When we hit a DeclRefExpr we are looking 1932 // at code that refers to a variable's name. We check if it has local 1933 // storage within the function, and if so, return the expression. 1934 DeclRefExpr *DR = cast<DeclRefExpr>(E); 1935 1936 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 1937 if (V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR; 1938 1939 return NULL; 1940 } 1941 1942 case Stmt::ParenExprClass: 1943 // Ignore parentheses. 1944 return EvalVal(cast<ParenExpr>(E)->getSubExpr()); 1945 1946 case Stmt::UnaryOperatorClass: { 1947 // The only unary operator that make sense to handle here 1948 // is Deref. All others don't resolve to a "name." This includes 1949 // handling all sorts of rvalues passed to a unary operator. 1950 UnaryOperator *U = cast<UnaryOperator>(E); 1951 1952 if (U->getOpcode() == UnaryOperator::Deref) 1953 return EvalAddr(U->getSubExpr()); 1954 1955 return NULL; 1956 } 1957 1958 case Stmt::ArraySubscriptExprClass: { 1959 // Array subscripts are potential references to data on the stack. We 1960 // retrieve the DeclRefExpr* for the array variable if it indeed 1961 // has local storage. 1962 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase()); 1963 } 1964 1965 case Stmt::ConditionalOperatorClass: { 1966 // For conditional operators we need to see if either the LHS or RHS are 1967 // non-NULL DeclRefExpr's. If one is non-NULL, we return it. 1968 ConditionalOperator *C = cast<ConditionalOperator>(E); 1969 1970 // Handle the GNU extension for missing LHS. 1971 if (Expr *lhsExpr = C->getLHS()) 1972 if (DeclRefExpr *LHS = EvalVal(lhsExpr)) 1973 return LHS; 1974 1975 return EvalVal(C->getRHS()); 1976 } 1977 1978 // Accesses to members are potential references to data on the stack. 1979 case Stmt::MemberExprClass: { 1980 MemberExpr *M = cast<MemberExpr>(E); 1981 1982 // Check for indirect access. We only want direct field accesses. 1983 if (!M->isArrow()) 1984 return EvalVal(M->getBase()); 1985 else 1986 return NULL; 1987 } 1988 1989 // Everything else: we simply don't reason about them. 1990 default: 1991 return NULL; 1992 } 1993} 1994 1995//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 1996 1997/// Check for comparisons of floating point operands using != and ==. 1998/// Issue a warning if these are no self-comparisons, as they are not likely 1999/// to do what the programmer intended. 2000void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 2001 bool EmitWarning = true; 2002 2003 Expr* LeftExprSansParen = lex->IgnoreParens(); 2004 Expr* RightExprSansParen = rex->IgnoreParens(); 2005 2006 // Special case: check for x == x (which is OK). 2007 // Do not emit warnings for such cases. 2008 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 2009 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 2010 if (DRL->getDecl() == DRR->getDecl()) 2011 EmitWarning = false; 2012 2013 2014 // Special case: check for comparisons against literals that can be exactly 2015 // represented by APFloat. In such cases, do not emit a warning. This 2016 // is a heuristic: often comparison against such literals are used to 2017 // detect if a value in a variable has not changed. This clearly can 2018 // lead to false negatives. 2019 if (EmitWarning) { 2020 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 2021 if (FLL->isExact()) 2022 EmitWarning = false; 2023 } else 2024 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 2025 if (FLR->isExact()) 2026 EmitWarning = false; 2027 } 2028 } 2029 2030 // Check for comparisons with builtin types. 2031 if (EmitWarning) 2032 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 2033 if (CL->isBuiltinCall(Context)) 2034 EmitWarning = false; 2035 2036 if (EmitWarning) 2037 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 2038 if (CR->isBuiltinCall(Context)) 2039 EmitWarning = false; 2040 2041 // Emit the diagnostic. 2042 if (EmitWarning) 2043 Diag(loc, diag::warn_floatingpoint_eq) 2044 << lex->getSourceRange() << rex->getSourceRange(); 2045} 2046 2047//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 2048//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 2049 2050namespace { 2051 2052/// Structure recording the 'active' range of an integer-valued 2053/// expression. 2054struct IntRange { 2055 /// The number of bits active in the int. 2056 unsigned Width; 2057 2058 /// True if the int is known not to have negative values. 2059 bool NonNegative; 2060 2061 IntRange() {} 2062 IntRange(unsigned Width, bool NonNegative) 2063 : Width(Width), NonNegative(NonNegative) 2064 {} 2065 2066 // Returns the range of the bool type. 2067 static IntRange forBoolType() { 2068 return IntRange(1, true); 2069 } 2070 2071 // Returns the range of an integral type. 2072 static IntRange forType(ASTContext &C, QualType T) { 2073 return forCanonicalType(C, T->getCanonicalTypeInternal().getTypePtr()); 2074 } 2075 2076 // Returns the range of an integeral type based on its canonical 2077 // representation. 2078 static IntRange forCanonicalType(ASTContext &C, const Type *T) { 2079 assert(T->isCanonicalUnqualified()); 2080 2081 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2082 T = VT->getElementType().getTypePtr(); 2083 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2084 T = CT->getElementType().getTypePtr(); 2085 2086 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 2087 EnumDecl *Enum = ET->getDecl(); 2088 unsigned NumPositive = Enum->getNumPositiveBits(); 2089 unsigned NumNegative = Enum->getNumNegativeBits(); 2090 2091 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 2092 } 2093 2094 const BuiltinType *BT = cast<BuiltinType>(T); 2095 assert(BT->isInteger()); 2096 2097 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2098 } 2099 2100 // Returns the supremum of two ranges: i.e. their conservative merge. 2101 static IntRange join(IntRange L, IntRange R) { 2102 return IntRange(std::max(L.Width, R.Width), 2103 L.NonNegative && R.NonNegative); 2104 } 2105 2106 // Returns the infinum of two ranges: i.e. their aggressive merge. 2107 static IntRange meet(IntRange L, IntRange R) { 2108 return IntRange(std::min(L.Width, R.Width), 2109 L.NonNegative || R.NonNegative); 2110 } 2111}; 2112 2113IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 2114 if (value.isSigned() && value.isNegative()) 2115 return IntRange(value.getMinSignedBits(), false); 2116 2117 if (value.getBitWidth() > MaxWidth) 2118 value.trunc(MaxWidth); 2119 2120 // isNonNegative() just checks the sign bit without considering 2121 // signedness. 2122 return IntRange(value.getActiveBits(), true); 2123} 2124 2125IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 2126 unsigned MaxWidth) { 2127 if (result.isInt()) 2128 return GetValueRange(C, result.getInt(), MaxWidth); 2129 2130 if (result.isVector()) { 2131 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 2132 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 2133 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 2134 R = IntRange::join(R, El); 2135 } 2136 return R; 2137 } 2138 2139 if (result.isComplexInt()) { 2140 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 2141 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 2142 return IntRange::join(R, I); 2143 } 2144 2145 // This can happen with lossless casts to intptr_t of "based" lvalues. 2146 // Assume it might use arbitrary bits. 2147 // FIXME: The only reason we need to pass the type in here is to get 2148 // the sign right on this one case. It would be nice if APValue 2149 // preserved this. 2150 assert(result.isLValue()); 2151 return IntRange(MaxWidth, Ty->isUnsignedIntegerType()); 2152} 2153 2154/// Pseudo-evaluate the given integer expression, estimating the 2155/// range of values it might take. 2156/// 2157/// \param MaxWidth - the width to which the value will be truncated 2158IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 2159 E = E->IgnoreParens(); 2160 2161 // Try a full evaluation first. 2162 Expr::EvalResult result; 2163 if (E->Evaluate(result, C)) 2164 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 2165 2166 // I think we only want to look through implicit casts here; if the 2167 // user has an explicit widening cast, we should treat the value as 2168 // being of the new, wider type. 2169 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 2170 if (CE->getCastKind() == CastExpr::CK_NoOp) 2171 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 2172 2173 IntRange OutputTypeRange = IntRange::forType(C, CE->getType()); 2174 2175 bool isIntegerCast = (CE->getCastKind() == CastExpr::CK_IntegralCast); 2176 if (!isIntegerCast && CE->getCastKind() == CastExpr::CK_Unknown) 2177 isIntegerCast = CE->getSubExpr()->getType()->isIntegerType(); 2178 2179 // Assume that non-integer casts can span the full range of the type. 2180 if (!isIntegerCast) 2181 return OutputTypeRange; 2182 2183 IntRange SubRange 2184 = GetExprRange(C, CE->getSubExpr(), 2185 std::min(MaxWidth, OutputTypeRange.Width)); 2186 2187 // Bail out if the subexpr's range is as wide as the cast type. 2188 if (SubRange.Width >= OutputTypeRange.Width) 2189 return OutputTypeRange; 2190 2191 // Otherwise, we take the smaller width, and we're non-negative if 2192 // either the output type or the subexpr is. 2193 return IntRange(SubRange.Width, 2194 SubRange.NonNegative || OutputTypeRange.NonNegative); 2195 } 2196 2197 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 2198 // If we can fold the condition, just take that operand. 2199 bool CondResult; 2200 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 2201 return GetExprRange(C, CondResult ? CO->getTrueExpr() 2202 : CO->getFalseExpr(), 2203 MaxWidth); 2204 2205 // Otherwise, conservatively merge. 2206 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 2207 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 2208 return IntRange::join(L, R); 2209 } 2210 2211 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 2212 switch (BO->getOpcode()) { 2213 2214 // Boolean-valued operations are single-bit and positive. 2215 case BinaryOperator::LAnd: 2216 case BinaryOperator::LOr: 2217 case BinaryOperator::LT: 2218 case BinaryOperator::GT: 2219 case BinaryOperator::LE: 2220 case BinaryOperator::GE: 2221 case BinaryOperator::EQ: 2222 case BinaryOperator::NE: 2223 return IntRange::forBoolType(); 2224 2225 // The type of these compound assignments is the type of the LHS, 2226 // so the RHS is not necessarily an integer. 2227 case BinaryOperator::MulAssign: 2228 case BinaryOperator::DivAssign: 2229 case BinaryOperator::RemAssign: 2230 case BinaryOperator::AddAssign: 2231 case BinaryOperator::SubAssign: 2232 return IntRange::forType(C, E->getType()); 2233 2234 // Operations with opaque sources are black-listed. 2235 case BinaryOperator::PtrMemD: 2236 case BinaryOperator::PtrMemI: 2237 return IntRange::forType(C, E->getType()); 2238 2239 // Bitwise-and uses the *infinum* of the two source ranges. 2240 case BinaryOperator::And: 2241 case BinaryOperator::AndAssign: 2242 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 2243 GetExprRange(C, BO->getRHS(), MaxWidth)); 2244 2245 // Left shift gets black-listed based on a judgement call. 2246 case BinaryOperator::Shl: 2247 // ...except that we want to treat '1 << (blah)' as logically 2248 // positive. It's an important idiom. 2249 if (IntegerLiteral *I 2250 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 2251 if (I->getValue() == 1) { 2252 IntRange R = IntRange::forType(C, E->getType()); 2253 return IntRange(R.Width, /*NonNegative*/ true); 2254 } 2255 } 2256 // fallthrough 2257 2258 case BinaryOperator::ShlAssign: 2259 return IntRange::forType(C, E->getType()); 2260 2261 // Right shift by a constant can narrow its left argument. 2262 case BinaryOperator::Shr: 2263 case BinaryOperator::ShrAssign: { 2264 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2265 2266 // If the shift amount is a positive constant, drop the width by 2267 // that much. 2268 llvm::APSInt shift; 2269 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 2270 shift.isNonNegative()) { 2271 unsigned zext = shift.getZExtValue(); 2272 if (zext >= L.Width) 2273 L.Width = (L.NonNegative ? 0 : 1); 2274 else 2275 L.Width -= zext; 2276 } 2277 2278 return L; 2279 } 2280 2281 // Comma acts as its right operand. 2282 case BinaryOperator::Comma: 2283 return GetExprRange(C, BO->getRHS(), MaxWidth); 2284 2285 // Black-list pointer subtractions. 2286 case BinaryOperator::Sub: 2287 if (BO->getLHS()->getType()->isPointerType()) 2288 return IntRange::forType(C, E->getType()); 2289 // fallthrough 2290 2291 default: 2292 break; 2293 } 2294 2295 // Treat every other operator as if it were closed on the 2296 // narrowest type that encompasses both operands. 2297 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2298 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 2299 return IntRange::join(L, R); 2300 } 2301 2302 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 2303 switch (UO->getOpcode()) { 2304 // Boolean-valued operations are white-listed. 2305 case UnaryOperator::LNot: 2306 return IntRange::forBoolType(); 2307 2308 // Operations with opaque sources are black-listed. 2309 case UnaryOperator::Deref: 2310 case UnaryOperator::AddrOf: // should be impossible 2311 case UnaryOperator::OffsetOf: 2312 return IntRange::forType(C, E->getType()); 2313 2314 default: 2315 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 2316 } 2317 } 2318 2319 if (dyn_cast<OffsetOfExpr>(E)) { 2320 IntRange::forType(C, E->getType()); 2321 } 2322 2323 FieldDecl *BitField = E->getBitField(); 2324 if (BitField) { 2325 llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C); 2326 unsigned BitWidth = BitWidthAP.getZExtValue(); 2327 2328 return IntRange(BitWidth, BitField->getType()->isUnsignedIntegerType()); 2329 } 2330 2331 return IntRange::forType(C, E->getType()); 2332} 2333 2334IntRange GetExprRange(ASTContext &C, Expr *E) { 2335 return GetExprRange(C, E, C.getIntWidth(E->getType())); 2336} 2337 2338/// Checks whether the given value, which currently has the given 2339/// source semantics, has the same value when coerced through the 2340/// target semantics. 2341bool IsSameFloatAfterCast(const llvm::APFloat &value, 2342 const llvm::fltSemantics &Src, 2343 const llvm::fltSemantics &Tgt) { 2344 llvm::APFloat truncated = value; 2345 2346 bool ignored; 2347 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 2348 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 2349 2350 return truncated.bitwiseIsEqual(value); 2351} 2352 2353/// Checks whether the given value, which currently has the given 2354/// source semantics, has the same value when coerced through the 2355/// target semantics. 2356/// 2357/// The value might be a vector of floats (or a complex number). 2358bool IsSameFloatAfterCast(const APValue &value, 2359 const llvm::fltSemantics &Src, 2360 const llvm::fltSemantics &Tgt) { 2361 if (value.isFloat()) 2362 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 2363 2364 if (value.isVector()) { 2365 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 2366 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 2367 return false; 2368 return true; 2369 } 2370 2371 assert(value.isComplexFloat()); 2372 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 2373 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 2374} 2375 2376void AnalyzeImplicitConversions(Sema &S, Expr *E); 2377 2378bool IsZero(Sema &S, Expr *E) { 2379 llvm::APSInt Value; 2380 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 2381} 2382 2383void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 2384 BinaryOperator::Opcode op = E->getOpcode(); 2385 if (op == BinaryOperator::LT && IsZero(S, E->getRHS())) { 2386 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2387 << "< 0" << "false" 2388 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2389 } else if (op == BinaryOperator::GE && IsZero(S, E->getRHS())) { 2390 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2391 << ">= 0" << "true" 2392 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2393 } else if (op == BinaryOperator::GT && IsZero(S, E->getLHS())) { 2394 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2395 << "0 >" << "false" 2396 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2397 } else if (op == BinaryOperator::LE && IsZero(S, E->getLHS())) { 2398 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2399 << "0 <=" << "true" 2400 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2401 } 2402} 2403 2404/// Analyze the operands of the given comparison. Implements the 2405/// fallback case from AnalyzeComparison. 2406void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 2407 AnalyzeImplicitConversions(S, E->getLHS()); 2408 AnalyzeImplicitConversions(S, E->getRHS()); 2409} 2410 2411/// \brief Implements -Wsign-compare. 2412/// 2413/// \param lex the left-hand expression 2414/// \param rex the right-hand expression 2415/// \param OpLoc the location of the joining operator 2416/// \param BinOpc binary opcode or 0 2417void AnalyzeComparison(Sema &S, BinaryOperator *E) { 2418 // The type the comparison is being performed in. 2419 QualType T = E->getLHS()->getType(); 2420 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 2421 && "comparison with mismatched types"); 2422 2423 // We don't do anything special if this isn't an unsigned integral 2424 // comparison: we're only interested in integral comparisons, and 2425 // signed comparisons only happen in cases we don't care to warn about. 2426 if (!T->isUnsignedIntegerType()) 2427 return AnalyzeImpConvsInComparison(S, E); 2428 2429 Expr *lex = E->getLHS()->IgnoreParenImpCasts(); 2430 Expr *rex = E->getRHS()->IgnoreParenImpCasts(); 2431 2432 // Check to see if one of the (unmodified) operands is of different 2433 // signedness. 2434 Expr *signedOperand, *unsignedOperand; 2435 if (lex->getType()->isSignedIntegerType()) { 2436 assert(!rex->getType()->isSignedIntegerType() && 2437 "unsigned comparison between two signed integer expressions?"); 2438 signedOperand = lex; 2439 unsignedOperand = rex; 2440 } else if (rex->getType()->isSignedIntegerType()) { 2441 signedOperand = rex; 2442 unsignedOperand = lex; 2443 } else { 2444 CheckTrivialUnsignedComparison(S, E); 2445 return AnalyzeImpConvsInComparison(S, E); 2446 } 2447 2448 // Otherwise, calculate the effective range of the signed operand. 2449 IntRange signedRange = GetExprRange(S.Context, signedOperand); 2450 2451 // Go ahead and analyze implicit conversions in the operands. Note 2452 // that we skip the implicit conversions on both sides. 2453 AnalyzeImplicitConversions(S, lex); 2454 AnalyzeImplicitConversions(S, rex); 2455 2456 // If the signed range is non-negative, -Wsign-compare won't fire, 2457 // but we should still check for comparisons which are always true 2458 // or false. 2459 if (signedRange.NonNegative) 2460 return CheckTrivialUnsignedComparison(S, E); 2461 2462 // For (in)equality comparisons, if the unsigned operand is a 2463 // constant which cannot collide with a overflowed signed operand, 2464 // then reinterpreting the signed operand as unsigned will not 2465 // change the result of the comparison. 2466 if (E->isEqualityOp()) { 2467 unsigned comparisonWidth = S.Context.getIntWidth(T); 2468 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 2469 2470 // We should never be unable to prove that the unsigned operand is 2471 // non-negative. 2472 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 2473 2474 if (unsignedRange.Width < comparisonWidth) 2475 return; 2476 } 2477 2478 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison) 2479 << lex->getType() << rex->getType() 2480 << lex->getSourceRange() << rex->getSourceRange(); 2481} 2482 2483/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 2484void DiagnoseImpCast(Sema &S, Expr *E, QualType T, unsigned diag) { 2485 S.Diag(E->getExprLoc(), diag) << E->getType() << T << E->getSourceRange(); 2486} 2487 2488void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 2489 bool *ICContext = 0) { 2490 if (E->isTypeDependent() || E->isValueDependent()) return; 2491 2492 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 2493 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 2494 if (Source == Target) return; 2495 if (Target->isDependentType()) return; 2496 2497 // Never diagnose implicit casts to bool. 2498 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 2499 return; 2500 2501 // Strip vector types. 2502 if (isa<VectorType>(Source)) { 2503 if (!isa<VectorType>(Target)) 2504 return DiagnoseImpCast(S, E, T, diag::warn_impcast_vector_scalar); 2505 2506 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 2507 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 2508 } 2509 2510 // Strip complex types. 2511 if (isa<ComplexType>(Source)) { 2512 if (!isa<ComplexType>(Target)) 2513 return DiagnoseImpCast(S, E, T, diag::warn_impcast_complex_scalar); 2514 2515 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 2516 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 2517 } 2518 2519 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 2520 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 2521 2522 // If the source is floating point... 2523 if (SourceBT && SourceBT->isFloatingPoint()) { 2524 // ...and the target is floating point... 2525 if (TargetBT && TargetBT->isFloatingPoint()) { 2526 // ...then warn if we're dropping FP rank. 2527 2528 // Builtin FP kinds are ordered by increasing FP rank. 2529 if (SourceBT->getKind() > TargetBT->getKind()) { 2530 // Don't warn about float constants that are precisely 2531 // representable in the target type. 2532 Expr::EvalResult result; 2533 if (E->Evaluate(result, S.Context)) { 2534 // Value might be a float, a float vector, or a float complex. 2535 if (IsSameFloatAfterCast(result.Val, 2536 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 2537 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 2538 return; 2539 } 2540 2541 DiagnoseImpCast(S, E, T, diag::warn_impcast_float_precision); 2542 } 2543 return; 2544 } 2545 2546 // If the target is integral, always warn. 2547 if ((TargetBT && TargetBT->isInteger())) 2548 // TODO: don't warn for integer values? 2549 DiagnoseImpCast(S, E, T, diag::warn_impcast_float_integer); 2550 2551 return; 2552 } 2553 2554 if (!Source->isIntegerType() || !Target->isIntegerType()) 2555 return; 2556 2557 IntRange SourceRange = GetExprRange(S.Context, E); 2558 IntRange TargetRange = IntRange::forCanonicalType(S.Context, Target); 2559 2560 if (SourceRange.Width > TargetRange.Width) { 2561 // People want to build with -Wshorten-64-to-32 and not -Wconversion 2562 // and by god we'll let them. 2563 if (SourceRange.Width == 64 && TargetRange.Width == 32) 2564 return DiagnoseImpCast(S, E, T, diag::warn_impcast_integer_64_32); 2565 return DiagnoseImpCast(S, E, T, diag::warn_impcast_integer_precision); 2566 } 2567 2568 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 2569 (!TargetRange.NonNegative && SourceRange.NonNegative && 2570 SourceRange.Width == TargetRange.Width)) { 2571 unsigned DiagID = diag::warn_impcast_integer_sign; 2572 2573 // Traditionally, gcc has warned about this under -Wsign-compare. 2574 // We also want to warn about it in -Wconversion. 2575 // So if -Wconversion is off, use a completely identical diagnostic 2576 // in the sign-compare group. 2577 // The conditional-checking code will 2578 if (ICContext) { 2579 DiagID = diag::warn_impcast_integer_sign_conditional; 2580 *ICContext = true; 2581 } 2582 2583 return DiagnoseImpCast(S, E, T, DiagID); 2584 } 2585 2586 return; 2587} 2588 2589void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 2590 2591void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 2592 bool &ICContext) { 2593 E = E->IgnoreParenImpCasts(); 2594 2595 if (isa<ConditionalOperator>(E)) 2596 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 2597 2598 AnalyzeImplicitConversions(S, E); 2599 if (E->getType() != T) 2600 return CheckImplicitConversion(S, E, T, &ICContext); 2601 return; 2602} 2603 2604void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 2605 AnalyzeImplicitConversions(S, E->getCond()); 2606 2607 bool Suspicious = false; 2608 CheckConditionalOperand(S, E->getTrueExpr(), T, Suspicious); 2609 CheckConditionalOperand(S, E->getFalseExpr(), T, Suspicious); 2610 2611 // If -Wconversion would have warned about either of the candidates 2612 // for a signedness conversion to the context type... 2613 if (!Suspicious) return; 2614 2615 // ...but it's currently ignored... 2616 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional)) 2617 return; 2618 2619 // ...and -Wsign-compare isn't... 2620 if (!S.Diags.getDiagnosticLevel(diag::warn_mixed_sign_conditional)) 2621 return; 2622 2623 // ...then check whether it would have warned about either of the 2624 // candidates for a signedness conversion to the condition type. 2625 if (E->getType() != T) { 2626 Suspicious = false; 2627 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 2628 E->getType(), &Suspicious); 2629 if (!Suspicious) 2630 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 2631 E->getType(), &Suspicious); 2632 if (!Suspicious) 2633 return; 2634 } 2635 2636 // If so, emit a diagnostic under -Wsign-compare. 2637 Expr *lex = E->getTrueExpr()->IgnoreParenImpCasts(); 2638 Expr *rex = E->getFalseExpr()->IgnoreParenImpCasts(); 2639 S.Diag(E->getQuestionLoc(), diag::warn_mixed_sign_conditional) 2640 << lex->getType() << rex->getType() 2641 << lex->getSourceRange() << rex->getSourceRange(); 2642} 2643 2644/// AnalyzeImplicitConversions - Find and report any interesting 2645/// implicit conversions in the given expression. There are a couple 2646/// of competing diagnostics here, -Wconversion and -Wsign-compare. 2647void AnalyzeImplicitConversions(Sema &S, Expr *OrigE) { 2648 QualType T = OrigE->getType(); 2649 Expr *E = OrigE->IgnoreParenImpCasts(); 2650 2651 // For conditional operators, we analyze the arguments as if they 2652 // were being fed directly into the output. 2653 if (isa<ConditionalOperator>(E)) { 2654 ConditionalOperator *CO = cast<ConditionalOperator>(E); 2655 CheckConditionalOperator(S, CO, T); 2656 return; 2657 } 2658 2659 // Go ahead and check any implicit conversions we might have skipped. 2660 // The non-canonical typecheck is just an optimization; 2661 // CheckImplicitConversion will filter out dead implicit conversions. 2662 if (E->getType() != T) 2663 CheckImplicitConversion(S, E, T); 2664 2665 // Now continue drilling into this expression. 2666 2667 // Skip past explicit casts. 2668 if (isa<ExplicitCastExpr>(E)) { 2669 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 2670 return AnalyzeImplicitConversions(S, E); 2671 } 2672 2673 // Do a somewhat different check with comparison operators. 2674 if (isa<BinaryOperator>(E) && cast<BinaryOperator>(E)->isComparisonOp()) 2675 return AnalyzeComparison(S, cast<BinaryOperator>(E)); 2676 2677 // These break the otherwise-useful invariant below. Fortunately, 2678 // we don't really need to recurse into them, because any internal 2679 // expressions should have been analyzed already when they were 2680 // built into statements. 2681 if (isa<StmtExpr>(E)) return; 2682 2683 // Don't descend into unevaluated contexts. 2684 if (isa<SizeOfAlignOfExpr>(E)) return; 2685 2686 // Now just recurse over the expression's children. 2687 for (Stmt::child_iterator I = E->child_begin(), IE = E->child_end(); 2688 I != IE; ++I) 2689 AnalyzeImplicitConversions(S, cast<Expr>(*I)); 2690} 2691 2692} // end anonymous namespace 2693 2694/// Diagnoses "dangerous" implicit conversions within the given 2695/// expression (which is a full expression). Implements -Wconversion 2696/// and -Wsign-compare. 2697void Sema::CheckImplicitConversions(Expr *E) { 2698 // Don't diagnose in unevaluated contexts. 2699 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 2700 return; 2701 2702 // Don't diagnose for value- or type-dependent expressions. 2703 if (E->isTypeDependent() || E->isValueDependent()) 2704 return; 2705 2706 AnalyzeImplicitConversions(*this, E); 2707} 2708 2709/// CheckParmsForFunctionDef - Check that the parameters of the given 2710/// function are appropriate for the definition of a function. This 2711/// takes care of any checks that cannot be performed on the 2712/// declaration itself, e.g., that the types of each of the function 2713/// parameters are complete. 2714bool Sema::CheckParmsForFunctionDef(FunctionDecl *FD) { 2715 bool HasInvalidParm = false; 2716 for (unsigned p = 0, NumParams = FD->getNumParams(); p < NumParams; ++p) { 2717 ParmVarDecl *Param = FD->getParamDecl(p); 2718 2719 // C99 6.7.5.3p4: the parameters in a parameter type list in a 2720 // function declarator that is part of a function definition of 2721 // that function shall not have incomplete type. 2722 // 2723 // This is also C++ [dcl.fct]p6. 2724 if (!Param->isInvalidDecl() && 2725 RequireCompleteType(Param->getLocation(), Param->getType(), 2726 diag::err_typecheck_decl_incomplete_type)) { 2727 Param->setInvalidDecl(); 2728 HasInvalidParm = true; 2729 } 2730 2731 // C99 6.9.1p5: If the declarator includes a parameter type list, the 2732 // declaration of each parameter shall include an identifier. 2733 if (Param->getIdentifier() == 0 && 2734 !Param->isImplicit() && 2735 !getLangOptions().CPlusPlus) 2736 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 2737 2738 // C99 6.7.5.3p12: 2739 // If the function declarator is not part of a definition of that 2740 // function, parameters may have incomplete type and may use the [*] 2741 // notation in their sequences of declarator specifiers to specify 2742 // variable length array types. 2743 QualType PType = Param->getOriginalType(); 2744 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 2745 if (AT->getSizeModifier() == ArrayType::Star) { 2746 // FIXME: This diagnosic should point the the '[*]' if source-location 2747 // information is added for it. 2748 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 2749 } 2750 } 2751 } 2752 2753 return HasInvalidParm; 2754} 2755