SemaChecking.cpp revision f85e193739c953358c865005855253af4f68a497
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 "clang/Sema/Sema.h" 16#include "clang/Sema/SemaInternal.h" 17#include "clang/Sema/ScopeInfo.h" 18#include "clang/Analysis/Analyses/FormatString.h" 19#include "clang/AST/ASTContext.h" 20#include "clang/AST/CharUnits.h" 21#include "clang/AST/DeclCXX.h" 22#include "clang/AST/DeclObjC.h" 23#include "clang/AST/ExprCXX.h" 24#include "clang/AST/ExprObjC.h" 25#include "clang/AST/EvaluatedExprVisitor.h" 26#include "clang/AST/DeclObjC.h" 27#include "clang/AST/StmtCXX.h" 28#include "clang/AST/StmtObjC.h" 29#include "clang/Lex/Preprocessor.h" 30#include "llvm/ADT/BitVector.h" 31#include "llvm/ADT/STLExtras.h" 32#include "llvm/Support/raw_ostream.h" 33#include "clang/Basic/TargetBuiltins.h" 34#include "clang/Basic/TargetInfo.h" 35#include "clang/Basic/ConvertUTF.h" 36#include <limits> 37using namespace clang; 38using namespace sema; 39 40SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 41 unsigned ByteNo) const { 42 return SL->getLocationOfByte(ByteNo, PP.getSourceManager(), 43 PP.getLangOptions(), PP.getTargetInfo()); 44} 45 46 47/// CheckablePrintfAttr - does a function call have a "printf" attribute 48/// and arguments that merit checking? 49bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) { 50 if (Format->getType() == "printf") return true; 51 if (Format->getType() == "printf0") { 52 // printf0 allows null "format" string; if so don't check format/args 53 unsigned format_idx = Format->getFormatIdx() - 1; 54 // Does the index refer to the implicit object argument? 55 if (isa<CXXMemberCallExpr>(TheCall)) { 56 if (format_idx == 0) 57 return false; 58 --format_idx; 59 } 60 if (format_idx < TheCall->getNumArgs()) { 61 Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts(); 62 if (!Format->isNullPointerConstant(Context, 63 Expr::NPC_ValueDependentIsNull)) 64 return true; 65 } 66 } 67 return false; 68} 69 70/// Checks that a call expression's argument count is the desired number. 71/// This is useful when doing custom type-checking. Returns true on error. 72static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) { 73 unsigned argCount = call->getNumArgs(); 74 if (argCount == desiredArgCount) return false; 75 76 if (argCount < desiredArgCount) 77 return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args) 78 << 0 /*function call*/ << desiredArgCount << argCount 79 << call->getSourceRange(); 80 81 // Highlight all the excess arguments. 82 SourceRange range(call->getArg(desiredArgCount)->getLocStart(), 83 call->getArg(argCount - 1)->getLocEnd()); 84 85 return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args) 86 << 0 /*function call*/ << desiredArgCount << argCount 87 << call->getArg(1)->getSourceRange(); 88} 89 90ExprResult 91Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 92 ExprResult TheCallResult(Owned(TheCall)); 93 94 // Find out if any arguments are required to be integer constant expressions. 95 unsigned ICEArguments = 0; 96 ASTContext::GetBuiltinTypeError Error; 97 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments); 98 if (Error != ASTContext::GE_None) 99 ICEArguments = 0; // Don't diagnose previously diagnosed errors. 100 101 // If any arguments are required to be ICE's, check and diagnose. 102 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) { 103 // Skip arguments not required to be ICE's. 104 if ((ICEArguments & (1 << ArgNo)) == 0) continue; 105 106 llvm::APSInt Result; 107 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 108 return true; 109 ICEArguments &= ~(1 << ArgNo); 110 } 111 112 switch (BuiltinID) { 113 case Builtin::BI__builtin___CFStringMakeConstantString: 114 assert(TheCall->getNumArgs() == 1 && 115 "Wrong # arguments to builtin CFStringMakeConstantString"); 116 if (CheckObjCString(TheCall->getArg(0))) 117 return ExprError(); 118 break; 119 case Builtin::BI__builtin_stdarg_start: 120 case Builtin::BI__builtin_va_start: 121 if (SemaBuiltinVAStart(TheCall)) 122 return ExprError(); 123 break; 124 case Builtin::BI__builtin_isgreater: 125 case Builtin::BI__builtin_isgreaterequal: 126 case Builtin::BI__builtin_isless: 127 case Builtin::BI__builtin_islessequal: 128 case Builtin::BI__builtin_islessgreater: 129 case Builtin::BI__builtin_isunordered: 130 if (SemaBuiltinUnorderedCompare(TheCall)) 131 return ExprError(); 132 break; 133 case Builtin::BI__builtin_fpclassify: 134 if (SemaBuiltinFPClassification(TheCall, 6)) 135 return ExprError(); 136 break; 137 case Builtin::BI__builtin_isfinite: 138 case Builtin::BI__builtin_isinf: 139 case Builtin::BI__builtin_isinf_sign: 140 case Builtin::BI__builtin_isnan: 141 case Builtin::BI__builtin_isnormal: 142 if (SemaBuiltinFPClassification(TheCall, 1)) 143 return ExprError(); 144 break; 145 case Builtin::BI__builtin_shufflevector: 146 return SemaBuiltinShuffleVector(TheCall); 147 // TheCall will be freed by the smart pointer here, but that's fine, since 148 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 149 case Builtin::BI__builtin_prefetch: 150 if (SemaBuiltinPrefetch(TheCall)) 151 return ExprError(); 152 break; 153 case Builtin::BI__builtin_object_size: 154 if (SemaBuiltinObjectSize(TheCall)) 155 return ExprError(); 156 break; 157 case Builtin::BI__builtin_longjmp: 158 if (SemaBuiltinLongjmp(TheCall)) 159 return ExprError(); 160 break; 161 162 case Builtin::BI__builtin_classify_type: 163 if (checkArgCount(*this, TheCall, 1)) return true; 164 TheCall->setType(Context.IntTy); 165 break; 166 case Builtin::BI__builtin_constant_p: 167 if (checkArgCount(*this, TheCall, 1)) return true; 168 TheCall->setType(Context.IntTy); 169 break; 170 case Builtin::BI__sync_fetch_and_add: 171 case Builtin::BI__sync_fetch_and_sub: 172 case Builtin::BI__sync_fetch_and_or: 173 case Builtin::BI__sync_fetch_and_and: 174 case Builtin::BI__sync_fetch_and_xor: 175 case Builtin::BI__sync_add_and_fetch: 176 case Builtin::BI__sync_sub_and_fetch: 177 case Builtin::BI__sync_and_and_fetch: 178 case Builtin::BI__sync_or_and_fetch: 179 case Builtin::BI__sync_xor_and_fetch: 180 case Builtin::BI__sync_val_compare_and_swap: 181 case Builtin::BI__sync_bool_compare_and_swap: 182 case Builtin::BI__sync_lock_test_and_set: 183 case Builtin::BI__sync_lock_release: 184 case Builtin::BI__sync_swap: 185 return SemaBuiltinAtomicOverloaded(move(TheCallResult)); 186 } 187 188 // Since the target specific builtins for each arch overlap, only check those 189 // of the arch we are compiling for. 190 if (BuiltinID >= Builtin::FirstTSBuiltin) { 191 switch (Context.Target.getTriple().getArch()) { 192 case llvm::Triple::arm: 193 case llvm::Triple::thumb: 194 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 195 return ExprError(); 196 break; 197 default: 198 break; 199 } 200 } 201 202 return move(TheCallResult); 203} 204 205// Get the valid immediate range for the specified NEON type code. 206static unsigned RFT(unsigned t, bool shift = false) { 207 bool quad = t & 0x10; 208 209 switch (t & 0x7) { 210 case 0: // i8 211 return shift ? 7 : (8 << (int)quad) - 1; 212 case 1: // i16 213 return shift ? 15 : (4 << (int)quad) - 1; 214 case 2: // i32 215 return shift ? 31 : (2 << (int)quad) - 1; 216 case 3: // i64 217 return shift ? 63 : (1 << (int)quad) - 1; 218 case 4: // f32 219 assert(!shift && "cannot shift float types!"); 220 return (2 << (int)quad) - 1; 221 case 5: // poly8 222 return shift ? 7 : (8 << (int)quad) - 1; 223 case 6: // poly16 224 return shift ? 15 : (4 << (int)quad) - 1; 225 case 7: // float16 226 assert(!shift && "cannot shift float types!"); 227 return (4 << (int)quad) - 1; 228 } 229 return 0; 230} 231 232bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 233 llvm::APSInt Result; 234 235 unsigned mask = 0; 236 unsigned TV = 0; 237 switch (BuiltinID) { 238#define GET_NEON_OVERLOAD_CHECK 239#include "clang/Basic/arm_neon.inc" 240#undef GET_NEON_OVERLOAD_CHECK 241 } 242 243 // For NEON intrinsics which are overloaded on vector element type, validate 244 // the immediate which specifies which variant to emit. 245 if (mask) { 246 unsigned ArgNo = TheCall->getNumArgs()-1; 247 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 248 return true; 249 250 TV = Result.getLimitedValue(32); 251 if ((TV > 31) || (mask & (1 << TV)) == 0) 252 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 253 << TheCall->getArg(ArgNo)->getSourceRange(); 254 } 255 256 // For NEON intrinsics which take an immediate value as part of the 257 // instruction, range check them here. 258 unsigned i = 0, l = 0, u = 0; 259 switch (BuiltinID) { 260 default: return false; 261 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 262 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 263 case ARM::BI__builtin_arm_vcvtr_f: 264 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 265#define GET_NEON_IMMEDIATE_CHECK 266#include "clang/Basic/arm_neon.inc" 267#undef GET_NEON_IMMEDIATE_CHECK 268 }; 269 270 // Check that the immediate argument is actually a constant. 271 if (SemaBuiltinConstantArg(TheCall, i, Result)) 272 return true; 273 274 // Range check against the upper/lower values for this isntruction. 275 unsigned Val = Result.getZExtValue(); 276 if (Val < l || Val > (u + l)) 277 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 278 << l << u+l << TheCall->getArg(i)->getSourceRange(); 279 280 // FIXME: VFP Intrinsics should error if VFP not present. 281 return false; 282} 283 284/// CheckFunctionCall - Check a direct function call for various correctness 285/// and safety properties not strictly enforced by the C type system. 286bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 287 // Get the IdentifierInfo* for the called function. 288 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 289 290 // None of the checks below are needed for functions that don't have 291 // simple names (e.g., C++ conversion functions). 292 if (!FnInfo) 293 return false; 294 295 // FIXME: This mechanism should be abstracted to be less fragile and 296 // more efficient. For example, just map function ids to custom 297 // handlers. 298 299 // Printf and scanf checking. 300 for (specific_attr_iterator<FormatAttr> 301 i = FDecl->specific_attr_begin<FormatAttr>(), 302 e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { 303 304 const FormatAttr *Format = *i; 305 const bool b = Format->getType() == "scanf"; 306 if (b || CheckablePrintfAttr(Format, TheCall)) { 307 bool HasVAListArg = Format->getFirstArg() == 0; 308 CheckPrintfScanfArguments(TheCall, HasVAListArg, 309 Format->getFormatIdx() - 1, 310 HasVAListArg ? 0 : Format->getFirstArg() - 1, 311 !b); 312 } 313 } 314 315 for (specific_attr_iterator<NonNullAttr> 316 i = FDecl->specific_attr_begin<NonNullAttr>(), 317 e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) { 318 CheckNonNullArguments(*i, TheCall->getArgs(), 319 TheCall->getCallee()->getLocStart()); 320 } 321 322 // Memset/memcpy/memmove handling 323 if (FDecl->getLinkage() == ExternalLinkage && 324 (!getLangOptions().CPlusPlus || FDecl->isExternC())) { 325 if (FnInfo->isStr("memset") || FnInfo->isStr("memcpy") || 326 FnInfo->isStr("memmove")) 327 CheckMemsetcpymoveArguments(TheCall, FnInfo); 328 } 329 330 return false; 331} 332 333bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 334 // Printf checking. 335 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 336 if (!Format) 337 return false; 338 339 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 340 if (!V) 341 return false; 342 343 QualType Ty = V->getType(); 344 if (!Ty->isBlockPointerType()) 345 return false; 346 347 const bool b = Format->getType() == "scanf"; 348 if (!b && !CheckablePrintfAttr(Format, TheCall)) 349 return false; 350 351 bool HasVAListArg = Format->getFirstArg() == 0; 352 CheckPrintfScanfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 353 HasVAListArg ? 0 : Format->getFirstArg() - 1, !b); 354 355 return false; 356} 357 358/// SemaBuiltinAtomicOverloaded - We have a call to a function like 359/// __sync_fetch_and_add, which is an overloaded function based on the pointer 360/// type of its first argument. The main ActOnCallExpr routines have already 361/// promoted the types of arguments because all of these calls are prototyped as 362/// void(...). 363/// 364/// This function goes through and does final semantic checking for these 365/// builtins, 366ExprResult 367Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 368 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 369 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 370 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 371 372 // Ensure that we have at least one argument to do type inference from. 373 if (TheCall->getNumArgs() < 1) { 374 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 375 << 0 << 1 << TheCall->getNumArgs() 376 << TheCall->getCallee()->getSourceRange(); 377 return ExprError(); 378 } 379 380 // Inspect the first argument of the atomic builtin. This should always be 381 // a pointer type, whose element is an integral scalar or pointer type. 382 // Because it is a pointer type, we don't have to worry about any implicit 383 // casts here. 384 // FIXME: We don't allow floating point scalars as input. 385 Expr *FirstArg = TheCall->getArg(0); 386 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 387 if (!pointerType) { 388 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 389 << FirstArg->getType() << FirstArg->getSourceRange(); 390 return ExprError(); 391 } 392 393 QualType ValType = pointerType->getPointeeType(); 394 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 395 !ValType->isBlockPointerType()) { 396 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 397 << FirstArg->getType() << FirstArg->getSourceRange(); 398 return ExprError(); 399 } 400 401 switch (ValType.getObjCLifetime()) { 402 case Qualifiers::OCL_None: 403 case Qualifiers::OCL_ExplicitNone: 404 // okay 405 break; 406 407 case Qualifiers::OCL_Weak: 408 case Qualifiers::OCL_Strong: 409 case Qualifiers::OCL_Autoreleasing: 410 Diag(DRE->getLocStart(), diag::err_arc_atomic_lifetime) 411 << ValType << FirstArg->getSourceRange(); 412 return ExprError(); 413 } 414 415 // The majority of builtins return a value, but a few have special return 416 // types, so allow them to override appropriately below. 417 QualType ResultType = ValType; 418 419 // We need to figure out which concrete builtin this maps onto. For example, 420 // __sync_fetch_and_add with a 2 byte object turns into 421 // __sync_fetch_and_add_2. 422#define BUILTIN_ROW(x) \ 423 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 424 Builtin::BI##x##_8, Builtin::BI##x##_16 } 425 426 static const unsigned BuiltinIndices[][5] = { 427 BUILTIN_ROW(__sync_fetch_and_add), 428 BUILTIN_ROW(__sync_fetch_and_sub), 429 BUILTIN_ROW(__sync_fetch_and_or), 430 BUILTIN_ROW(__sync_fetch_and_and), 431 BUILTIN_ROW(__sync_fetch_and_xor), 432 433 BUILTIN_ROW(__sync_add_and_fetch), 434 BUILTIN_ROW(__sync_sub_and_fetch), 435 BUILTIN_ROW(__sync_and_and_fetch), 436 BUILTIN_ROW(__sync_or_and_fetch), 437 BUILTIN_ROW(__sync_xor_and_fetch), 438 439 BUILTIN_ROW(__sync_val_compare_and_swap), 440 BUILTIN_ROW(__sync_bool_compare_and_swap), 441 BUILTIN_ROW(__sync_lock_test_and_set), 442 BUILTIN_ROW(__sync_lock_release), 443 BUILTIN_ROW(__sync_swap) 444 }; 445#undef BUILTIN_ROW 446 447 // Determine the index of the size. 448 unsigned SizeIndex; 449 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 450 case 1: SizeIndex = 0; break; 451 case 2: SizeIndex = 1; break; 452 case 4: SizeIndex = 2; break; 453 case 8: SizeIndex = 3; break; 454 case 16: SizeIndex = 4; break; 455 default: 456 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 457 << FirstArg->getType() << FirstArg->getSourceRange(); 458 return ExprError(); 459 } 460 461 // Each of these builtins has one pointer argument, followed by some number of 462 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 463 // that we ignore. Find out which row of BuiltinIndices to read from as well 464 // as the number of fixed args. 465 unsigned BuiltinID = FDecl->getBuiltinID(); 466 unsigned BuiltinIndex, NumFixed = 1; 467 switch (BuiltinID) { 468 default: assert(0 && "Unknown overloaded atomic builtin!"); 469 case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break; 470 case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break; 471 case Builtin::BI__sync_fetch_and_or: BuiltinIndex = 2; break; 472 case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break; 473 case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break; 474 475 case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break; 476 case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break; 477 case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break; 478 case Builtin::BI__sync_or_and_fetch: BuiltinIndex = 8; break; 479 case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break; 480 481 case Builtin::BI__sync_val_compare_and_swap: 482 BuiltinIndex = 10; 483 NumFixed = 2; 484 break; 485 case Builtin::BI__sync_bool_compare_and_swap: 486 BuiltinIndex = 11; 487 NumFixed = 2; 488 ResultType = Context.BoolTy; 489 break; 490 case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break; 491 case Builtin::BI__sync_lock_release: 492 BuiltinIndex = 13; 493 NumFixed = 0; 494 ResultType = Context.VoidTy; 495 break; 496 case Builtin::BI__sync_swap: BuiltinIndex = 14; break; 497 } 498 499 // Now that we know how many fixed arguments we expect, first check that we 500 // have at least that many. 501 if (TheCall->getNumArgs() < 1+NumFixed) { 502 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 503 << 0 << 1+NumFixed << TheCall->getNumArgs() 504 << TheCall->getCallee()->getSourceRange(); 505 return ExprError(); 506 } 507 508 // Get the decl for the concrete builtin from this, we can tell what the 509 // concrete integer type we should convert to is. 510 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 511 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 512 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 513 FunctionDecl *NewBuiltinDecl = 514 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 515 TUScope, false, DRE->getLocStart())); 516 517 // The first argument --- the pointer --- has a fixed type; we 518 // deduce the types of the rest of the arguments accordingly. Walk 519 // the remaining arguments, converting them to the deduced value type. 520 for (unsigned i = 0; i != NumFixed; ++i) { 521 ExprResult Arg = TheCall->getArg(i+1); 522 523 // If the argument is an implicit cast, then there was a promotion due to 524 // "...", just remove it now. 525 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg.get())) { 526 Arg = ICE->getSubExpr(); 527 ICE->setSubExpr(0); 528 TheCall->setArg(i+1, Arg.get()); 529 } 530 531 // GCC does an implicit conversion to the pointer or integer ValType. This 532 // can fail in some cases (1i -> int**), check for this error case now. 533 CastKind Kind = CK_Invalid; 534 ExprValueKind VK = VK_RValue; 535 CXXCastPath BasePath; 536 Arg = CheckCastTypes(Arg.get()->getLocStart(), Arg.get()->getSourceRange(), 537 ValType, Arg.take(), Kind, VK, BasePath); 538 if (Arg.isInvalid()) 539 return ExprError(); 540 541 // Okay, we have something that *can* be converted to the right type. Check 542 // to see if there is a potentially weird extension going on here. This can 543 // happen when you do an atomic operation on something like an char* and 544 // pass in 42. The 42 gets converted to char. This is even more strange 545 // for things like 45.123 -> char, etc. 546 // FIXME: Do this check. 547 Arg = ImpCastExprToType(Arg.take(), ValType, Kind, VK, &BasePath); 548 TheCall->setArg(i+1, Arg.get()); 549 } 550 551 // Switch the DeclRefExpr to refer to the new decl. 552 DRE->setDecl(NewBuiltinDecl); 553 DRE->setType(NewBuiltinDecl->getType()); 554 555 // Set the callee in the CallExpr. 556 // FIXME: This leaks the original parens and implicit casts. 557 ExprResult PromotedCall = UsualUnaryConversions(DRE); 558 if (PromotedCall.isInvalid()) 559 return ExprError(); 560 TheCall->setCallee(PromotedCall.take()); 561 562 // Change the result type of the call to match the original value type. This 563 // is arbitrary, but the codegen for these builtins ins design to handle it 564 // gracefully. 565 TheCall->setType(ResultType); 566 567 return move(TheCallResult); 568} 569 570 571/// CheckObjCString - Checks that the argument to the builtin 572/// CFString constructor is correct 573/// Note: It might also make sense to do the UTF-16 conversion here (would 574/// simplify the backend). 575bool Sema::CheckObjCString(Expr *Arg) { 576 Arg = Arg->IgnoreParenCasts(); 577 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 578 579 if (!Literal || Literal->isWide()) { 580 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 581 << Arg->getSourceRange(); 582 return true; 583 } 584 585 if (Literal->containsNonAsciiOrNull()) { 586 llvm::StringRef String = Literal->getString(); 587 unsigned NumBytes = String.size(); 588 llvm::SmallVector<UTF16, 128> ToBuf(NumBytes); 589 const UTF8 *FromPtr = (UTF8 *)String.data(); 590 UTF16 *ToPtr = &ToBuf[0]; 591 592 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 593 &ToPtr, ToPtr + NumBytes, 594 strictConversion); 595 // Check for conversion failure. 596 if (Result != conversionOK) 597 Diag(Arg->getLocStart(), 598 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 599 } 600 return false; 601} 602 603/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 604/// Emit an error and return true on failure, return false on success. 605bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 606 Expr *Fn = TheCall->getCallee(); 607 if (TheCall->getNumArgs() > 2) { 608 Diag(TheCall->getArg(2)->getLocStart(), 609 diag::err_typecheck_call_too_many_args) 610 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 611 << Fn->getSourceRange() 612 << SourceRange(TheCall->getArg(2)->getLocStart(), 613 (*(TheCall->arg_end()-1))->getLocEnd()); 614 return true; 615 } 616 617 if (TheCall->getNumArgs() < 2) { 618 return Diag(TheCall->getLocEnd(), 619 diag::err_typecheck_call_too_few_args_at_least) 620 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 621 } 622 623 // Determine whether the current function is variadic or not. 624 BlockScopeInfo *CurBlock = getCurBlock(); 625 bool isVariadic; 626 if (CurBlock) 627 isVariadic = CurBlock->TheDecl->isVariadic(); 628 else if (FunctionDecl *FD = getCurFunctionDecl()) 629 isVariadic = FD->isVariadic(); 630 else 631 isVariadic = getCurMethodDecl()->isVariadic(); 632 633 if (!isVariadic) { 634 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 635 return true; 636 } 637 638 // Verify that the second argument to the builtin is the last argument of the 639 // current function or method. 640 bool SecondArgIsLastNamedArgument = false; 641 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 642 643 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 644 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 645 // FIXME: This isn't correct for methods (results in bogus warning). 646 // Get the last formal in the current function. 647 const ParmVarDecl *LastArg; 648 if (CurBlock) 649 LastArg = *(CurBlock->TheDecl->param_end()-1); 650 else if (FunctionDecl *FD = getCurFunctionDecl()) 651 LastArg = *(FD->param_end()-1); 652 else 653 LastArg = *(getCurMethodDecl()->param_end()-1); 654 SecondArgIsLastNamedArgument = PV == LastArg; 655 } 656 } 657 658 if (!SecondArgIsLastNamedArgument) 659 Diag(TheCall->getArg(1)->getLocStart(), 660 diag::warn_second_parameter_of_va_start_not_last_named_argument); 661 return false; 662} 663 664/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 665/// friends. This is declared to take (...), so we have to check everything. 666bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 667 if (TheCall->getNumArgs() < 2) 668 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 669 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 670 if (TheCall->getNumArgs() > 2) 671 return Diag(TheCall->getArg(2)->getLocStart(), 672 diag::err_typecheck_call_too_many_args) 673 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 674 << SourceRange(TheCall->getArg(2)->getLocStart(), 675 (*(TheCall->arg_end()-1))->getLocEnd()); 676 677 ExprResult OrigArg0 = TheCall->getArg(0); 678 ExprResult OrigArg1 = TheCall->getArg(1); 679 680 // Do standard promotions between the two arguments, returning their common 681 // type. 682 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 683 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 684 return true; 685 686 // Make sure any conversions are pushed back into the call; this is 687 // type safe since unordered compare builtins are declared as "_Bool 688 // foo(...)". 689 TheCall->setArg(0, OrigArg0.get()); 690 TheCall->setArg(1, OrigArg1.get()); 691 692 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 693 return false; 694 695 // If the common type isn't a real floating type, then the arguments were 696 // invalid for this operation. 697 if (!Res->isRealFloatingType()) 698 return Diag(OrigArg0.get()->getLocStart(), 699 diag::err_typecheck_call_invalid_ordered_compare) 700 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 701 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 702 703 return false; 704} 705 706/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 707/// __builtin_isnan and friends. This is declared to take (...), so we have 708/// to check everything. We expect the last argument to be a floating point 709/// value. 710bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 711 if (TheCall->getNumArgs() < NumArgs) 712 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 713 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 714 if (TheCall->getNumArgs() > NumArgs) 715 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 716 diag::err_typecheck_call_too_many_args) 717 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 718 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 719 (*(TheCall->arg_end()-1))->getLocEnd()); 720 721 Expr *OrigArg = TheCall->getArg(NumArgs-1); 722 723 if (OrigArg->isTypeDependent()) 724 return false; 725 726 // This operation requires a non-_Complex floating-point number. 727 if (!OrigArg->getType()->isRealFloatingType()) 728 return Diag(OrigArg->getLocStart(), 729 diag::err_typecheck_call_invalid_unary_fp) 730 << OrigArg->getType() << OrigArg->getSourceRange(); 731 732 // If this is an implicit conversion from float -> double, remove it. 733 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 734 Expr *CastArg = Cast->getSubExpr(); 735 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 736 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 737 "promotion from float to double is the only expected cast here"); 738 Cast->setSubExpr(0); 739 TheCall->setArg(NumArgs-1, CastArg); 740 OrigArg = CastArg; 741 } 742 } 743 744 return false; 745} 746 747/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 748// This is declared to take (...), so we have to check everything. 749ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 750 if (TheCall->getNumArgs() < 2) 751 return ExprError(Diag(TheCall->getLocEnd(), 752 diag::err_typecheck_call_too_few_args_at_least) 753 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 754 << TheCall->getSourceRange()); 755 756 // Determine which of the following types of shufflevector we're checking: 757 // 1) unary, vector mask: (lhs, mask) 758 // 2) binary, vector mask: (lhs, rhs, mask) 759 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 760 QualType resType = TheCall->getArg(0)->getType(); 761 unsigned numElements = 0; 762 763 if (!TheCall->getArg(0)->isTypeDependent() && 764 !TheCall->getArg(1)->isTypeDependent()) { 765 QualType LHSType = TheCall->getArg(0)->getType(); 766 QualType RHSType = TheCall->getArg(1)->getType(); 767 768 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 769 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 770 << SourceRange(TheCall->getArg(0)->getLocStart(), 771 TheCall->getArg(1)->getLocEnd()); 772 return ExprError(); 773 } 774 775 numElements = LHSType->getAs<VectorType>()->getNumElements(); 776 unsigned numResElements = TheCall->getNumArgs() - 2; 777 778 // Check to see if we have a call with 2 vector arguments, the unary shuffle 779 // with mask. If so, verify that RHS is an integer vector type with the 780 // same number of elts as lhs. 781 if (TheCall->getNumArgs() == 2) { 782 if (!RHSType->hasIntegerRepresentation() || 783 RHSType->getAs<VectorType>()->getNumElements() != numElements) 784 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 785 << SourceRange(TheCall->getArg(1)->getLocStart(), 786 TheCall->getArg(1)->getLocEnd()); 787 numResElements = numElements; 788 } 789 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 790 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 791 << SourceRange(TheCall->getArg(0)->getLocStart(), 792 TheCall->getArg(1)->getLocEnd()); 793 return ExprError(); 794 } else if (numElements != numResElements) { 795 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 796 resType = Context.getVectorType(eltType, numResElements, 797 VectorType::GenericVector); 798 } 799 } 800 801 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 802 if (TheCall->getArg(i)->isTypeDependent() || 803 TheCall->getArg(i)->isValueDependent()) 804 continue; 805 806 llvm::APSInt Result(32); 807 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 808 return ExprError(Diag(TheCall->getLocStart(), 809 diag::err_shufflevector_nonconstant_argument) 810 << TheCall->getArg(i)->getSourceRange()); 811 812 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 813 return ExprError(Diag(TheCall->getLocStart(), 814 diag::err_shufflevector_argument_too_large) 815 << TheCall->getArg(i)->getSourceRange()); 816 } 817 818 llvm::SmallVector<Expr*, 32> exprs; 819 820 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 821 exprs.push_back(TheCall->getArg(i)); 822 TheCall->setArg(i, 0); 823 } 824 825 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 826 exprs.size(), resType, 827 TheCall->getCallee()->getLocStart(), 828 TheCall->getRParenLoc())); 829} 830 831/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 832// This is declared to take (const void*, ...) and can take two 833// optional constant int args. 834bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 835 unsigned NumArgs = TheCall->getNumArgs(); 836 837 if (NumArgs > 3) 838 return Diag(TheCall->getLocEnd(), 839 diag::err_typecheck_call_too_many_args_at_most) 840 << 0 /*function call*/ << 3 << NumArgs 841 << TheCall->getSourceRange(); 842 843 // Argument 0 is checked for us and the remaining arguments must be 844 // constant integers. 845 for (unsigned i = 1; i != NumArgs; ++i) { 846 Expr *Arg = TheCall->getArg(i); 847 848 llvm::APSInt Result; 849 if (SemaBuiltinConstantArg(TheCall, i, Result)) 850 return true; 851 852 // FIXME: gcc issues a warning and rewrites these to 0. These 853 // seems especially odd for the third argument since the default 854 // is 3. 855 if (i == 1) { 856 if (Result.getLimitedValue() > 1) 857 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 858 << "0" << "1" << Arg->getSourceRange(); 859 } else { 860 if (Result.getLimitedValue() > 3) 861 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 862 << "0" << "3" << Arg->getSourceRange(); 863 } 864 } 865 866 return false; 867} 868 869/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 870/// TheCall is a constant expression. 871bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 872 llvm::APSInt &Result) { 873 Expr *Arg = TheCall->getArg(ArgNum); 874 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 875 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 876 877 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 878 879 if (!Arg->isIntegerConstantExpr(Result, Context)) 880 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 881 << FDecl->getDeclName() << Arg->getSourceRange(); 882 883 return false; 884} 885 886/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 887/// int type). This simply type checks that type is one of the defined 888/// constants (0-3). 889// For compatibility check 0-3, llvm only handles 0 and 2. 890bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 891 llvm::APSInt Result; 892 893 // Check constant-ness first. 894 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 895 return true; 896 897 Expr *Arg = TheCall->getArg(1); 898 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 899 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 900 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 901 } 902 903 return false; 904} 905 906/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 907/// This checks that val is a constant 1. 908bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 909 Expr *Arg = TheCall->getArg(1); 910 llvm::APSInt Result; 911 912 // TODO: This is less than ideal. Overload this to take a value. 913 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 914 return true; 915 916 if (Result != 1) 917 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 918 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 919 920 return false; 921} 922 923// Handle i > 1 ? "x" : "y", recursively. 924bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 925 bool HasVAListArg, 926 unsigned format_idx, unsigned firstDataArg, 927 bool isPrintf) { 928 tryAgain: 929 if (E->isTypeDependent() || E->isValueDependent()) 930 return false; 931 932 E = E->IgnoreParens(); 933 934 switch (E->getStmtClass()) { 935 case Stmt::BinaryConditionalOperatorClass: 936 case Stmt::ConditionalOperatorClass: { 937 const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E); 938 return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, HasVAListArg, 939 format_idx, firstDataArg, isPrintf) 940 && SemaCheckStringLiteral(C->getFalseExpr(), TheCall, HasVAListArg, 941 format_idx, firstDataArg, isPrintf); 942 } 943 944 case Stmt::IntegerLiteralClass: 945 // Technically -Wformat-nonliteral does not warn about this case. 946 // The behavior of printf and friends in this case is implementation 947 // dependent. Ideally if the format string cannot be null then 948 // it should have a 'nonnull' attribute in the function prototype. 949 return true; 950 951 case Stmt::ImplicitCastExprClass: { 952 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 953 goto tryAgain; 954 } 955 956 case Stmt::OpaqueValueExprClass: 957 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 958 E = src; 959 goto tryAgain; 960 } 961 return false; 962 963 case Stmt::PredefinedExprClass: 964 // While __func__, etc., are technically not string literals, they 965 // cannot contain format specifiers and thus are not a security 966 // liability. 967 return true; 968 969 case Stmt::DeclRefExprClass: { 970 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 971 972 // As an exception, do not flag errors for variables binding to 973 // const string literals. 974 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 975 bool isConstant = false; 976 QualType T = DR->getType(); 977 978 if (const ArrayType *AT = Context.getAsArrayType(T)) { 979 isConstant = AT->getElementType().isConstant(Context); 980 } else if (const PointerType *PT = T->getAs<PointerType>()) { 981 isConstant = T.isConstant(Context) && 982 PT->getPointeeType().isConstant(Context); 983 } 984 985 if (isConstant) { 986 if (const Expr *Init = VD->getAnyInitializer()) 987 return SemaCheckStringLiteral(Init, TheCall, 988 HasVAListArg, format_idx, firstDataArg, 989 isPrintf); 990 } 991 992 // For vprintf* functions (i.e., HasVAListArg==true), we add a 993 // special check to see if the format string is a function parameter 994 // of the function calling the printf function. If the function 995 // has an attribute indicating it is a printf-like function, then we 996 // should suppress warnings concerning non-literals being used in a call 997 // to a vprintf function. For example: 998 // 999 // void 1000 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1001 // va_list ap; 1002 // va_start(ap, fmt); 1003 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1004 // ... 1005 // 1006 // 1007 // FIXME: We don't have full attribute support yet, so just check to see 1008 // if the argument is a DeclRefExpr that references a parameter. We'll 1009 // add proper support for checking the attribute later. 1010 if (HasVAListArg) 1011 if (isa<ParmVarDecl>(VD)) 1012 return true; 1013 } 1014 1015 return false; 1016 } 1017 1018 case Stmt::CallExprClass: { 1019 const CallExpr *CE = cast<CallExpr>(E); 1020 if (const ImplicitCastExpr *ICE 1021 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 1022 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 1023 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 1024 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 1025 unsigned ArgIndex = FA->getFormatIdx(); 1026 const Expr *Arg = CE->getArg(ArgIndex - 1); 1027 1028 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 1029 format_idx, firstDataArg, isPrintf); 1030 } 1031 } 1032 } 1033 } 1034 1035 return false; 1036 } 1037 case Stmt::ObjCStringLiteralClass: 1038 case Stmt::StringLiteralClass: { 1039 const StringLiteral *StrE = NULL; 1040 1041 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1042 StrE = ObjCFExpr->getString(); 1043 else 1044 StrE = cast<StringLiteral>(E); 1045 1046 if (StrE) { 1047 CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx, 1048 firstDataArg, isPrintf); 1049 return true; 1050 } 1051 1052 return false; 1053 } 1054 1055 default: 1056 return false; 1057 } 1058} 1059 1060void 1061Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1062 const Expr * const *ExprArgs, 1063 SourceLocation CallSiteLoc) { 1064 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1065 e = NonNull->args_end(); 1066 i != e; ++i) { 1067 const Expr *ArgExpr = ExprArgs[*i]; 1068 if (ArgExpr->isNullPointerConstant(Context, 1069 Expr::NPC_ValueDependentIsNotNull)) 1070 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 1071 } 1072} 1073 1074/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar 1075/// functions) for correct use of format strings. 1076void 1077Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg, 1078 unsigned format_idx, unsigned firstDataArg, 1079 bool isPrintf) { 1080 1081 const Expr *Fn = TheCall->getCallee(); 1082 1083 // The way the format attribute works in GCC, the implicit this argument 1084 // of member functions is counted. However, it doesn't appear in our own 1085 // lists, so decrement format_idx in that case. 1086 if (isa<CXXMemberCallExpr>(TheCall)) { 1087 const CXXMethodDecl *method_decl = 1088 dyn_cast<CXXMethodDecl>(TheCall->getCalleeDecl()); 1089 if (method_decl && method_decl->isInstance()) { 1090 // Catch a format attribute mistakenly referring to the object argument. 1091 if (format_idx == 0) 1092 return; 1093 --format_idx; 1094 if(firstDataArg != 0) 1095 --firstDataArg; 1096 } 1097 } 1098 1099 // CHECK: printf/scanf-like function is called with no format string. 1100 if (format_idx >= TheCall->getNumArgs()) { 1101 Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string) 1102 << Fn->getSourceRange(); 1103 return; 1104 } 1105 1106 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 1107 1108 // CHECK: format string is not a string literal. 1109 // 1110 // Dynamically generated format strings are difficult to 1111 // automatically vet at compile time. Requiring that format strings 1112 // are string literals: (1) permits the checking of format strings by 1113 // the compiler and thereby (2) can practically remove the source of 1114 // many format string exploits. 1115 1116 // Format string can be either ObjC string (e.g. @"%d") or 1117 // C string (e.g. "%d") 1118 // ObjC string uses the same format specifiers as C string, so we can use 1119 // the same format string checking logic for both ObjC and C strings. 1120 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 1121 firstDataArg, isPrintf)) 1122 return; // Literal format string found, check done! 1123 1124 // If there are no arguments specified, warn with -Wformat-security, otherwise 1125 // warn only with -Wformat-nonliteral. 1126 if (TheCall->getNumArgs() == format_idx+1) 1127 Diag(TheCall->getArg(format_idx)->getLocStart(), 1128 diag::warn_format_nonliteral_noargs) 1129 << OrigFormatExpr->getSourceRange(); 1130 else 1131 Diag(TheCall->getArg(format_idx)->getLocStart(), 1132 diag::warn_format_nonliteral) 1133 << OrigFormatExpr->getSourceRange(); 1134} 1135 1136namespace { 1137class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1138protected: 1139 Sema &S; 1140 const StringLiteral *FExpr; 1141 const Expr *OrigFormatExpr; 1142 const unsigned FirstDataArg; 1143 const unsigned NumDataArgs; 1144 const bool IsObjCLiteral; 1145 const char *Beg; // Start of format string. 1146 const bool HasVAListArg; 1147 const CallExpr *TheCall; 1148 unsigned FormatIdx; 1149 llvm::BitVector CoveredArgs; 1150 bool usesPositionalArgs; 1151 bool atFirstArg; 1152public: 1153 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1154 const Expr *origFormatExpr, unsigned firstDataArg, 1155 unsigned numDataArgs, bool isObjCLiteral, 1156 const char *beg, bool hasVAListArg, 1157 const CallExpr *theCall, unsigned formatIdx) 1158 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1159 FirstDataArg(firstDataArg), 1160 NumDataArgs(numDataArgs), 1161 IsObjCLiteral(isObjCLiteral), Beg(beg), 1162 HasVAListArg(hasVAListArg), 1163 TheCall(theCall), FormatIdx(formatIdx), 1164 usesPositionalArgs(false), atFirstArg(true) { 1165 CoveredArgs.resize(numDataArgs); 1166 CoveredArgs.reset(); 1167 } 1168 1169 void DoneProcessing(); 1170 1171 void HandleIncompleteSpecifier(const char *startSpecifier, 1172 unsigned specifierLen); 1173 1174 virtual void HandleInvalidPosition(const char *startSpecifier, 1175 unsigned specifierLen, 1176 analyze_format_string::PositionContext p); 1177 1178 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1179 1180 void HandleNullChar(const char *nullCharacter); 1181 1182protected: 1183 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1184 const char *startSpec, 1185 unsigned specifierLen, 1186 const char *csStart, unsigned csLen); 1187 1188 SourceRange getFormatStringRange(); 1189 CharSourceRange getSpecifierRange(const char *startSpecifier, 1190 unsigned specifierLen); 1191 SourceLocation getLocationOfByte(const char *x); 1192 1193 const Expr *getDataArg(unsigned i) const; 1194 1195 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1196 const analyze_format_string::ConversionSpecifier &CS, 1197 const char *startSpecifier, unsigned specifierLen, 1198 unsigned argIndex); 1199}; 1200} 1201 1202SourceRange CheckFormatHandler::getFormatStringRange() { 1203 return OrigFormatExpr->getSourceRange(); 1204} 1205 1206CharSourceRange CheckFormatHandler:: 1207getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1208 SourceLocation Start = getLocationOfByte(startSpecifier); 1209 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1210 1211 // Advance the end SourceLocation by one due to half-open ranges. 1212 End = End.getFileLocWithOffset(1); 1213 1214 return CharSourceRange::getCharRange(Start, End); 1215} 1216 1217SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 1218 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1219} 1220 1221void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 1222 unsigned specifierLen){ 1223 SourceLocation Loc = getLocationOfByte(startSpecifier); 1224 S.Diag(Loc, diag::warn_printf_incomplete_specifier) 1225 << getSpecifierRange(startSpecifier, specifierLen); 1226} 1227 1228void 1229CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1230 analyze_format_string::PositionContext p) { 1231 SourceLocation Loc = getLocationOfByte(startPos); 1232 S.Diag(Loc, diag::warn_format_invalid_positional_specifier) 1233 << (unsigned) p << getSpecifierRange(startPos, posLen); 1234} 1235 1236void CheckFormatHandler::HandleZeroPosition(const char *startPos, 1237 unsigned posLen) { 1238 SourceLocation Loc = getLocationOfByte(startPos); 1239 S.Diag(Loc, diag::warn_format_zero_positional_specifier) 1240 << getSpecifierRange(startPos, posLen); 1241} 1242 1243void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 1244 if (!IsObjCLiteral) { 1245 // The presence of a null character is likely an error. 1246 S.Diag(getLocationOfByte(nullCharacter), 1247 diag::warn_printf_format_string_contains_null_char) 1248 << getFormatStringRange(); 1249 } 1250} 1251 1252const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 1253 return TheCall->getArg(FirstDataArg + i); 1254} 1255 1256void CheckFormatHandler::DoneProcessing() { 1257 // Does the number of data arguments exceed the number of 1258 // format conversions in the format string? 1259 if (!HasVAListArg) { 1260 // Find any arguments that weren't covered. 1261 CoveredArgs.flip(); 1262 signed notCoveredArg = CoveredArgs.find_first(); 1263 if (notCoveredArg >= 0) { 1264 assert((unsigned)notCoveredArg < NumDataArgs); 1265 S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(), 1266 diag::warn_printf_data_arg_not_used) 1267 << getFormatStringRange(); 1268 } 1269 } 1270} 1271 1272bool 1273CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 1274 SourceLocation Loc, 1275 const char *startSpec, 1276 unsigned specifierLen, 1277 const char *csStart, 1278 unsigned csLen) { 1279 1280 bool keepGoing = true; 1281 if (argIndex < NumDataArgs) { 1282 // Consider the argument coverered, even though the specifier doesn't 1283 // make sense. 1284 CoveredArgs.set(argIndex); 1285 } 1286 else { 1287 // If argIndex exceeds the number of data arguments we 1288 // don't issue a warning because that is just a cascade of warnings (and 1289 // they may have intended '%%' anyway). We don't want to continue processing 1290 // the format string after this point, however, as we will like just get 1291 // gibberish when trying to match arguments. 1292 keepGoing = false; 1293 } 1294 1295 S.Diag(Loc, diag::warn_format_invalid_conversion) 1296 << llvm::StringRef(csStart, csLen) 1297 << getSpecifierRange(startSpec, specifierLen); 1298 1299 return keepGoing; 1300} 1301 1302bool 1303CheckFormatHandler::CheckNumArgs( 1304 const analyze_format_string::FormatSpecifier &FS, 1305 const analyze_format_string::ConversionSpecifier &CS, 1306 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 1307 1308 if (argIndex >= NumDataArgs) { 1309 if (FS.usesPositionalArg()) { 1310 S.Diag(getLocationOfByte(CS.getStart()), 1311 diag::warn_printf_positional_arg_exceeds_data_args) 1312 << (argIndex+1) << NumDataArgs 1313 << getSpecifierRange(startSpecifier, specifierLen); 1314 } 1315 else { 1316 S.Diag(getLocationOfByte(CS.getStart()), 1317 diag::warn_printf_insufficient_data_args) 1318 << getSpecifierRange(startSpecifier, specifierLen); 1319 } 1320 1321 return false; 1322 } 1323 return true; 1324} 1325 1326//===--- CHECK: Printf format string checking ------------------------------===// 1327 1328namespace { 1329class CheckPrintfHandler : public CheckFormatHandler { 1330public: 1331 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1332 const Expr *origFormatExpr, unsigned firstDataArg, 1333 unsigned numDataArgs, bool isObjCLiteral, 1334 const char *beg, bool hasVAListArg, 1335 const CallExpr *theCall, unsigned formatIdx) 1336 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1337 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1338 theCall, formatIdx) {} 1339 1340 1341 bool HandleInvalidPrintfConversionSpecifier( 1342 const analyze_printf::PrintfSpecifier &FS, 1343 const char *startSpecifier, 1344 unsigned specifierLen); 1345 1346 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 1347 const char *startSpecifier, 1348 unsigned specifierLen); 1349 1350 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 1351 const char *startSpecifier, unsigned specifierLen); 1352 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 1353 const analyze_printf::OptionalAmount &Amt, 1354 unsigned type, 1355 const char *startSpecifier, unsigned specifierLen); 1356 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1357 const analyze_printf::OptionalFlag &flag, 1358 const char *startSpecifier, unsigned specifierLen); 1359 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 1360 const analyze_printf::OptionalFlag &ignoredFlag, 1361 const analyze_printf::OptionalFlag &flag, 1362 const char *startSpecifier, unsigned specifierLen); 1363}; 1364} 1365 1366bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 1367 const analyze_printf::PrintfSpecifier &FS, 1368 const char *startSpecifier, 1369 unsigned specifierLen) { 1370 const analyze_printf::PrintfConversionSpecifier &CS = 1371 FS.getConversionSpecifier(); 1372 1373 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1374 getLocationOfByte(CS.getStart()), 1375 startSpecifier, specifierLen, 1376 CS.getStart(), CS.getLength()); 1377} 1378 1379bool CheckPrintfHandler::HandleAmount( 1380 const analyze_format_string::OptionalAmount &Amt, 1381 unsigned k, const char *startSpecifier, 1382 unsigned specifierLen) { 1383 1384 if (Amt.hasDataArgument()) { 1385 if (!HasVAListArg) { 1386 unsigned argIndex = Amt.getArgIndex(); 1387 if (argIndex >= NumDataArgs) { 1388 S.Diag(getLocationOfByte(Amt.getStart()), 1389 diag::warn_printf_asterisk_missing_arg) 1390 << k << getSpecifierRange(startSpecifier, specifierLen); 1391 // Don't do any more checking. We will just emit 1392 // spurious errors. 1393 return false; 1394 } 1395 1396 // Type check the data argument. It should be an 'int'. 1397 // Although not in conformance with C99, we also allow the argument to be 1398 // an 'unsigned int' as that is a reasonably safe case. GCC also 1399 // doesn't emit a warning for that case. 1400 CoveredArgs.set(argIndex); 1401 const Expr *Arg = getDataArg(argIndex); 1402 QualType T = Arg->getType(); 1403 1404 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1405 assert(ATR.isValid()); 1406 1407 if (!ATR.matchesType(S.Context, T)) { 1408 S.Diag(getLocationOfByte(Amt.getStart()), 1409 diag::warn_printf_asterisk_wrong_type) 1410 << k 1411 << ATR.getRepresentativeType(S.Context) << T 1412 << getSpecifierRange(startSpecifier, specifierLen) 1413 << Arg->getSourceRange(); 1414 // Don't do any more checking. We will just emit 1415 // spurious errors. 1416 return false; 1417 } 1418 } 1419 } 1420 return true; 1421} 1422 1423void CheckPrintfHandler::HandleInvalidAmount( 1424 const analyze_printf::PrintfSpecifier &FS, 1425 const analyze_printf::OptionalAmount &Amt, 1426 unsigned type, 1427 const char *startSpecifier, 1428 unsigned specifierLen) { 1429 const analyze_printf::PrintfConversionSpecifier &CS = 1430 FS.getConversionSpecifier(); 1431 switch (Amt.getHowSpecified()) { 1432 case analyze_printf::OptionalAmount::Constant: 1433 S.Diag(getLocationOfByte(Amt.getStart()), 1434 diag::warn_printf_nonsensical_optional_amount) 1435 << type 1436 << CS.toString() 1437 << getSpecifierRange(startSpecifier, specifierLen) 1438 << FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 1439 Amt.getConstantLength())); 1440 break; 1441 1442 default: 1443 S.Diag(getLocationOfByte(Amt.getStart()), 1444 diag::warn_printf_nonsensical_optional_amount) 1445 << type 1446 << CS.toString() 1447 << getSpecifierRange(startSpecifier, specifierLen); 1448 break; 1449 } 1450} 1451 1452void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1453 const analyze_printf::OptionalFlag &flag, 1454 const char *startSpecifier, 1455 unsigned specifierLen) { 1456 // Warn about pointless flag with a fixit removal. 1457 const analyze_printf::PrintfConversionSpecifier &CS = 1458 FS.getConversionSpecifier(); 1459 S.Diag(getLocationOfByte(flag.getPosition()), 1460 diag::warn_printf_nonsensical_flag) 1461 << flag.toString() << CS.toString() 1462 << getSpecifierRange(startSpecifier, specifierLen) 1463 << FixItHint::CreateRemoval(getSpecifierRange(flag.getPosition(), 1)); 1464} 1465 1466void CheckPrintfHandler::HandleIgnoredFlag( 1467 const analyze_printf::PrintfSpecifier &FS, 1468 const analyze_printf::OptionalFlag &ignoredFlag, 1469 const analyze_printf::OptionalFlag &flag, 1470 const char *startSpecifier, 1471 unsigned specifierLen) { 1472 // Warn about ignored flag with a fixit removal. 1473 S.Diag(getLocationOfByte(ignoredFlag.getPosition()), 1474 diag::warn_printf_ignored_flag) 1475 << ignoredFlag.toString() << flag.toString() 1476 << getSpecifierRange(startSpecifier, specifierLen) 1477 << FixItHint::CreateRemoval(getSpecifierRange( 1478 ignoredFlag.getPosition(), 1)); 1479} 1480 1481bool 1482CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 1483 &FS, 1484 const char *startSpecifier, 1485 unsigned specifierLen) { 1486 1487 using namespace analyze_format_string; 1488 using namespace analyze_printf; 1489 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 1490 1491 if (FS.consumesDataArgument()) { 1492 if (atFirstArg) { 1493 atFirstArg = false; 1494 usesPositionalArgs = FS.usesPositionalArg(); 1495 } 1496 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1497 // Cannot mix-and-match positional and non-positional arguments. 1498 S.Diag(getLocationOfByte(CS.getStart()), 1499 diag::warn_format_mix_positional_nonpositional_args) 1500 << getSpecifierRange(startSpecifier, specifierLen); 1501 return false; 1502 } 1503 } 1504 1505 // First check if the field width, precision, and conversion specifier 1506 // have matching data arguments. 1507 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 1508 startSpecifier, specifierLen)) { 1509 return false; 1510 } 1511 1512 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 1513 startSpecifier, specifierLen)) { 1514 return false; 1515 } 1516 1517 if (!CS.consumesDataArgument()) { 1518 // FIXME: Technically specifying a precision or field width here 1519 // makes no sense. Worth issuing a warning at some point. 1520 return true; 1521 } 1522 1523 // Consume the argument. 1524 unsigned argIndex = FS.getArgIndex(); 1525 if (argIndex < NumDataArgs) { 1526 // The check to see if the argIndex is valid will come later. 1527 // We set the bit here because we may exit early from this 1528 // function if we encounter some other error. 1529 CoveredArgs.set(argIndex); 1530 } 1531 1532 // Check for using an Objective-C specific conversion specifier 1533 // in a non-ObjC literal. 1534 if (!IsObjCLiteral && CS.isObjCArg()) { 1535 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 1536 specifierLen); 1537 } 1538 1539 // Check for invalid use of field width 1540 if (!FS.hasValidFieldWidth()) { 1541 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 1542 startSpecifier, specifierLen); 1543 } 1544 1545 // Check for invalid use of precision 1546 if (!FS.hasValidPrecision()) { 1547 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 1548 startSpecifier, specifierLen); 1549 } 1550 1551 // Check each flag does not conflict with any other component. 1552 if (!FS.hasValidThousandsGroupingPrefix()) 1553 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 1554 if (!FS.hasValidLeadingZeros()) 1555 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 1556 if (!FS.hasValidPlusPrefix()) 1557 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 1558 if (!FS.hasValidSpacePrefix()) 1559 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 1560 if (!FS.hasValidAlternativeForm()) 1561 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 1562 if (!FS.hasValidLeftJustified()) 1563 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 1564 1565 // Check that flags are not ignored by another flag 1566 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 1567 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 1568 startSpecifier, specifierLen); 1569 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 1570 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 1571 startSpecifier, specifierLen); 1572 1573 // Check the length modifier is valid with the given conversion specifier. 1574 const LengthModifier &LM = FS.getLengthModifier(); 1575 if (!FS.hasValidLengthModifier()) 1576 S.Diag(getLocationOfByte(LM.getStart()), 1577 diag::warn_format_nonsensical_length) 1578 << LM.toString() << CS.toString() 1579 << getSpecifierRange(startSpecifier, specifierLen) 1580 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 1581 LM.getLength())); 1582 1583 // Are we using '%n'? 1584 if (CS.getKind() == ConversionSpecifier::nArg) { 1585 // Issue a warning about this being a possible security issue. 1586 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back) 1587 << getSpecifierRange(startSpecifier, specifierLen); 1588 // Continue checking the other format specifiers. 1589 return true; 1590 } 1591 1592 // The remaining checks depend on the data arguments. 1593 if (HasVAListArg) 1594 return true; 1595 1596 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 1597 return false; 1598 1599 // Now type check the data expression that matches the 1600 // format specifier. 1601 const Expr *Ex = getDataArg(argIndex); 1602 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context); 1603 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 1604 // Check if we didn't match because of an implicit cast from a 'char' 1605 // or 'short' to an 'int'. This is done because printf is a varargs 1606 // function. 1607 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 1608 if (ICE->getType() == S.Context.IntTy) { 1609 // All further checking is done on the subexpression. 1610 Ex = ICE->getSubExpr(); 1611 if (ATR.matchesType(S.Context, Ex->getType())) 1612 return true; 1613 } 1614 1615 // We may be able to offer a FixItHint if it is a supported type. 1616 PrintfSpecifier fixedFS = FS; 1617 bool success = fixedFS.fixType(Ex->getType()); 1618 1619 if (success) { 1620 // Get the fix string from the fixed format specifier 1621 llvm::SmallString<128> buf; 1622 llvm::raw_svector_ostream os(buf); 1623 fixedFS.toString(os); 1624 1625 // FIXME: getRepresentativeType() perhaps should return a string 1626 // instead of a QualType to better handle when the representative 1627 // type is 'wint_t' (which is defined in the system headers). 1628 S.Diag(getLocationOfByte(CS.getStart()), 1629 diag::warn_printf_conversion_argument_type_mismatch) 1630 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1631 << getSpecifierRange(startSpecifier, specifierLen) 1632 << Ex->getSourceRange() 1633 << FixItHint::CreateReplacement( 1634 getSpecifierRange(startSpecifier, specifierLen), 1635 os.str()); 1636 } 1637 else { 1638 S.Diag(getLocationOfByte(CS.getStart()), 1639 diag::warn_printf_conversion_argument_type_mismatch) 1640 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1641 << getSpecifierRange(startSpecifier, specifierLen) 1642 << Ex->getSourceRange(); 1643 } 1644 } 1645 1646 return true; 1647} 1648 1649//===--- CHECK: Scanf format string checking ------------------------------===// 1650 1651namespace { 1652class CheckScanfHandler : public CheckFormatHandler { 1653public: 1654 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 1655 const Expr *origFormatExpr, unsigned firstDataArg, 1656 unsigned numDataArgs, bool isObjCLiteral, 1657 const char *beg, bool hasVAListArg, 1658 const CallExpr *theCall, unsigned formatIdx) 1659 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1660 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1661 theCall, formatIdx) {} 1662 1663 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 1664 const char *startSpecifier, 1665 unsigned specifierLen); 1666 1667 bool HandleInvalidScanfConversionSpecifier( 1668 const analyze_scanf::ScanfSpecifier &FS, 1669 const char *startSpecifier, 1670 unsigned specifierLen); 1671 1672 void HandleIncompleteScanList(const char *start, const char *end); 1673}; 1674} 1675 1676void CheckScanfHandler::HandleIncompleteScanList(const char *start, 1677 const char *end) { 1678 S.Diag(getLocationOfByte(end), diag::warn_scanf_scanlist_incomplete) 1679 << getSpecifierRange(start, end - start); 1680} 1681 1682bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 1683 const analyze_scanf::ScanfSpecifier &FS, 1684 const char *startSpecifier, 1685 unsigned specifierLen) { 1686 1687 const analyze_scanf::ScanfConversionSpecifier &CS = 1688 FS.getConversionSpecifier(); 1689 1690 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1691 getLocationOfByte(CS.getStart()), 1692 startSpecifier, specifierLen, 1693 CS.getStart(), CS.getLength()); 1694} 1695 1696bool CheckScanfHandler::HandleScanfSpecifier( 1697 const analyze_scanf::ScanfSpecifier &FS, 1698 const char *startSpecifier, 1699 unsigned specifierLen) { 1700 1701 using namespace analyze_scanf; 1702 using namespace analyze_format_string; 1703 1704 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 1705 1706 // Handle case where '%' and '*' don't consume an argument. These shouldn't 1707 // be used to decide if we are using positional arguments consistently. 1708 if (FS.consumesDataArgument()) { 1709 if (atFirstArg) { 1710 atFirstArg = false; 1711 usesPositionalArgs = FS.usesPositionalArg(); 1712 } 1713 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1714 // Cannot mix-and-match positional and non-positional arguments. 1715 S.Diag(getLocationOfByte(CS.getStart()), 1716 diag::warn_format_mix_positional_nonpositional_args) 1717 << getSpecifierRange(startSpecifier, specifierLen); 1718 return false; 1719 } 1720 } 1721 1722 // Check if the field with is non-zero. 1723 const OptionalAmount &Amt = FS.getFieldWidth(); 1724 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 1725 if (Amt.getConstantAmount() == 0) { 1726 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 1727 Amt.getConstantLength()); 1728 S.Diag(getLocationOfByte(Amt.getStart()), 1729 diag::warn_scanf_nonzero_width) 1730 << R << FixItHint::CreateRemoval(R); 1731 } 1732 } 1733 1734 if (!FS.consumesDataArgument()) { 1735 // FIXME: Technically specifying a precision or field width here 1736 // makes no sense. Worth issuing a warning at some point. 1737 return true; 1738 } 1739 1740 // Consume the argument. 1741 unsigned argIndex = FS.getArgIndex(); 1742 if (argIndex < NumDataArgs) { 1743 // The check to see if the argIndex is valid will come later. 1744 // We set the bit here because we may exit early from this 1745 // function if we encounter some other error. 1746 CoveredArgs.set(argIndex); 1747 } 1748 1749 // Check the length modifier is valid with the given conversion specifier. 1750 const LengthModifier &LM = FS.getLengthModifier(); 1751 if (!FS.hasValidLengthModifier()) { 1752 S.Diag(getLocationOfByte(LM.getStart()), 1753 diag::warn_format_nonsensical_length) 1754 << LM.toString() << CS.toString() 1755 << getSpecifierRange(startSpecifier, specifierLen) 1756 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 1757 LM.getLength())); 1758 } 1759 1760 // The remaining checks depend on the data arguments. 1761 if (HasVAListArg) 1762 return true; 1763 1764 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 1765 return false; 1766 1767 // FIXME: Check that the argument type matches the format specifier. 1768 1769 return true; 1770} 1771 1772void Sema::CheckFormatString(const StringLiteral *FExpr, 1773 const Expr *OrigFormatExpr, 1774 const CallExpr *TheCall, bool HasVAListArg, 1775 unsigned format_idx, unsigned firstDataArg, 1776 bool isPrintf) { 1777 1778 // CHECK: is the format string a wide literal? 1779 if (FExpr->isWide()) { 1780 Diag(FExpr->getLocStart(), 1781 diag::warn_format_string_is_wide_literal) 1782 << OrigFormatExpr->getSourceRange(); 1783 return; 1784 } 1785 1786 // Str - The format string. NOTE: this is NOT null-terminated! 1787 llvm::StringRef StrRef = FExpr->getString(); 1788 const char *Str = StrRef.data(); 1789 unsigned StrLen = StrRef.size(); 1790 1791 // CHECK: empty format string? 1792 if (StrLen == 0) { 1793 Diag(FExpr->getLocStart(), diag::warn_empty_format_string) 1794 << OrigFormatExpr->getSourceRange(); 1795 return; 1796 } 1797 1798 if (isPrintf) { 1799 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1800 TheCall->getNumArgs() - firstDataArg, 1801 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1802 HasVAListArg, TheCall, format_idx); 1803 1804 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen)) 1805 H.DoneProcessing(); 1806 } 1807 else { 1808 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1809 TheCall->getNumArgs() - firstDataArg, 1810 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1811 HasVAListArg, TheCall, format_idx); 1812 1813 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen)) 1814 H.DoneProcessing(); 1815 } 1816} 1817 1818//===--- CHECK: Standard memory functions ---------------------------------===// 1819 1820/// \brief Determine whether the given type is a dynamic class type (e.g., 1821/// whether it has a vtable). 1822static bool isDynamicClassType(QualType T) { 1823 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 1824 if (CXXRecordDecl *Definition = Record->getDefinition()) 1825 if (Definition->isDynamicClass()) 1826 return true; 1827 1828 return false; 1829} 1830 1831/// \brief If E is a sizeof expression, returns the expression's type in 1832/// OutType. 1833static bool sizeofExprType(const Expr* E, QualType *OutType) { 1834 if (const UnaryExprOrTypeTraitExpr *SizeOf = 1835 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) { 1836 if (SizeOf->getKind() != clang::UETT_SizeOf) 1837 return false; 1838 1839 *OutType = SizeOf->getTypeOfArgument(); 1840 return true; 1841 } 1842 return false; 1843} 1844 1845/// \brief Check for dangerous or invalid arguments to memset(). 1846/// 1847/// This issues warnings on known problematic, dangerous or unspecified 1848/// arguments to the standard 'memset', 'memcpy', and 'memmove' function calls. 1849/// 1850/// \param Call The call expression to diagnose. 1851void Sema::CheckMemsetcpymoveArguments(const CallExpr *Call, 1852 const IdentifierInfo *FnName) { 1853 // It is possible to have a non-standard definition of memset. Validate 1854 // we have the proper number of arguments, and if not, abort further 1855 // checking. 1856 if (Call->getNumArgs() != 3) 1857 return; 1858 1859 unsigned LastArg = FnName->isStr("memset")? 1 : 2; 1860 const Expr *LenExpr = Call->getArg(2)->IgnoreParenImpCasts(); 1861 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 1862 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 1863 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 1864 1865 QualType DestTy = Dest->getType(); 1866 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 1867 QualType PointeeTy = DestPtrTy->getPointeeType(); 1868 1869 if (PointeeTy->isVoidType()) 1870 continue; 1871 1872 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). 1873 QualType SizeofTy; 1874 if (sizeofExprType(LenExpr, &SizeofTy) && 1875 Context.typesAreCompatible(SizeofTy, DestTy)) { 1876 // Note: This complains about sizeof(typeof(p)) as well. 1877 SourceLocation loc = LenExpr->getSourceRange().getBegin(); 1878 Diag(loc, diag::warn_sizeof_pointer) 1879 << SizeofTy << PointeeTy << ArgIdx << FnName; 1880 break; 1881 } 1882 1883 unsigned DiagID; 1884 1885 // Always complain about dynamic classes. 1886 if (isDynamicClassType(PointeeTy)) 1887 DiagID = diag::warn_dyn_class_memaccess; 1888 else if (PointeeTy.hasNonTrivialObjCLifetime() && 1889 !FnName->isStr("memset")) 1890 DiagID = diag::warn_arc_object_memaccess; 1891 else 1892 continue; 1893 1894 DiagRuntimeBehavior( 1895 Dest->getExprLoc(), Dest, 1896 PDiag(DiagID) 1897 << ArgIdx << FnName << PointeeTy 1898 << Call->getCallee()->getSourceRange()); 1899 1900 DiagRuntimeBehavior( 1901 Dest->getExprLoc(), Dest, 1902 PDiag(diag::note_bad_memaccess_silence) 1903 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 1904 break; 1905 } 1906 } 1907} 1908 1909//===--- CHECK: Return Address of Stack Variable --------------------------===// 1910 1911static Expr *EvalVal(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars); 1912static Expr *EvalAddr(Expr* E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars); 1913 1914/// CheckReturnStackAddr - Check if a return statement returns the address 1915/// of a stack variable. 1916void 1917Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 1918 SourceLocation ReturnLoc) { 1919 1920 Expr *stackE = 0; 1921 llvm::SmallVector<DeclRefExpr *, 8> refVars; 1922 1923 // Perform checking for returned stack addresses, local blocks, 1924 // label addresses or references to temporaries. 1925 if (lhsType->isPointerType() || 1926 (!getLangOptions().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 1927 stackE = EvalAddr(RetValExp, refVars); 1928 } else if (lhsType->isReferenceType()) { 1929 stackE = EvalVal(RetValExp, refVars); 1930 } 1931 1932 if (stackE == 0) 1933 return; // Nothing suspicious was found. 1934 1935 SourceLocation diagLoc; 1936 SourceRange diagRange; 1937 if (refVars.empty()) { 1938 diagLoc = stackE->getLocStart(); 1939 diagRange = stackE->getSourceRange(); 1940 } else { 1941 // We followed through a reference variable. 'stackE' contains the 1942 // problematic expression but we will warn at the return statement pointing 1943 // at the reference variable. We will later display the "trail" of 1944 // reference variables using notes. 1945 diagLoc = refVars[0]->getLocStart(); 1946 diagRange = refVars[0]->getSourceRange(); 1947 } 1948 1949 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 1950 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 1951 : diag::warn_ret_stack_addr) 1952 << DR->getDecl()->getDeclName() << diagRange; 1953 } else if (isa<BlockExpr>(stackE)) { // local block. 1954 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 1955 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 1956 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 1957 } else { // local temporary. 1958 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 1959 : diag::warn_ret_local_temp_addr) 1960 << diagRange; 1961 } 1962 1963 // Display the "trail" of reference variables that we followed until we 1964 // found the problematic expression using notes. 1965 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 1966 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 1967 // If this var binds to another reference var, show the range of the next 1968 // var, otherwise the var binds to the problematic expression, in which case 1969 // show the range of the expression. 1970 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 1971 : stackE->getSourceRange(); 1972 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 1973 << VD->getDeclName() << range; 1974 } 1975} 1976 1977/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 1978/// check if the expression in a return statement evaluates to an address 1979/// to a location on the stack, a local block, an address of a label, or a 1980/// reference to local temporary. The recursion is used to traverse the 1981/// AST of the return expression, with recursion backtracking when we 1982/// encounter a subexpression that (1) clearly does not lead to one of the 1983/// above problematic expressions (2) is something we cannot determine leads to 1984/// a problematic expression based on such local checking. 1985/// 1986/// Both EvalAddr and EvalVal follow through reference variables to evaluate 1987/// the expression that they point to. Such variables are added to the 1988/// 'refVars' vector so that we know what the reference variable "trail" was. 1989/// 1990/// EvalAddr processes expressions that are pointers that are used as 1991/// references (and not L-values). EvalVal handles all other values. 1992/// At the base case of the recursion is a check for the above problematic 1993/// expressions. 1994/// 1995/// This implementation handles: 1996/// 1997/// * pointer-to-pointer casts 1998/// * implicit conversions from array references to pointers 1999/// * taking the address of fields 2000/// * arbitrary interplay between "&" and "*" operators 2001/// * pointer arithmetic from an address of a stack variable 2002/// * taking the address of an array element where the array is on the stack 2003static Expr *EvalAddr(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars) { 2004 if (E->isTypeDependent()) 2005 return NULL; 2006 2007 // We should only be called for evaluating pointer expressions. 2008 assert((E->getType()->isAnyPointerType() || 2009 E->getType()->isBlockPointerType() || 2010 E->getType()->isObjCQualifiedIdType()) && 2011 "EvalAddr only works on pointers"); 2012 2013 E = E->IgnoreParens(); 2014 2015 // Our "symbolic interpreter" is just a dispatch off the currently 2016 // viewed AST node. We then recursively traverse the AST by calling 2017 // EvalAddr and EvalVal appropriately. 2018 switch (E->getStmtClass()) { 2019 case Stmt::DeclRefExprClass: { 2020 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2021 2022 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2023 // If this is a reference variable, follow through to the expression that 2024 // it points to. 2025 if (V->hasLocalStorage() && 2026 V->getType()->isReferenceType() && V->hasInit()) { 2027 // Add the reference variable to the "trail". 2028 refVars.push_back(DR); 2029 return EvalAddr(V->getInit(), refVars); 2030 } 2031 2032 return NULL; 2033 } 2034 2035 case Stmt::UnaryOperatorClass: { 2036 // The only unary operator that make sense to handle here 2037 // is AddrOf. All others don't make sense as pointers. 2038 UnaryOperator *U = cast<UnaryOperator>(E); 2039 2040 if (U->getOpcode() == UO_AddrOf) 2041 return EvalVal(U->getSubExpr(), refVars); 2042 else 2043 return NULL; 2044 } 2045 2046 case Stmt::BinaryOperatorClass: { 2047 // Handle pointer arithmetic. All other binary operators are not valid 2048 // in this context. 2049 BinaryOperator *B = cast<BinaryOperator>(E); 2050 BinaryOperatorKind op = B->getOpcode(); 2051 2052 if (op != BO_Add && op != BO_Sub) 2053 return NULL; 2054 2055 Expr *Base = B->getLHS(); 2056 2057 // Determine which argument is the real pointer base. It could be 2058 // the RHS argument instead of the LHS. 2059 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 2060 2061 assert (Base->getType()->isPointerType()); 2062 return EvalAddr(Base, refVars); 2063 } 2064 2065 // For conditional operators we need to see if either the LHS or RHS are 2066 // valid DeclRefExpr*s. If one of them is valid, we return it. 2067 case Stmt::ConditionalOperatorClass: { 2068 ConditionalOperator *C = cast<ConditionalOperator>(E); 2069 2070 // Handle the GNU extension for missing LHS. 2071 if (Expr *lhsExpr = C->getLHS()) { 2072 // In C++, we can have a throw-expression, which has 'void' type. 2073 if (!lhsExpr->getType()->isVoidType()) 2074 if (Expr* LHS = EvalAddr(lhsExpr, refVars)) 2075 return LHS; 2076 } 2077 2078 // In C++, we can have a throw-expression, which has 'void' type. 2079 if (C->getRHS()->getType()->isVoidType()) 2080 return NULL; 2081 2082 return EvalAddr(C->getRHS(), refVars); 2083 } 2084 2085 case Stmt::BlockExprClass: 2086 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 2087 return E; // local block. 2088 return NULL; 2089 2090 case Stmt::AddrLabelExprClass: 2091 return E; // address of label. 2092 2093 // For casts, we need to handle conversions from arrays to 2094 // pointer values, and pointer-to-pointer conversions. 2095 case Stmt::ImplicitCastExprClass: 2096 case Stmt::CStyleCastExprClass: 2097 case Stmt::CXXFunctionalCastExprClass: 2098 case Stmt::ObjCBridgedCastExprClass: { 2099 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 2100 QualType T = SubExpr->getType(); 2101 2102 if (SubExpr->getType()->isPointerType() || 2103 SubExpr->getType()->isBlockPointerType() || 2104 SubExpr->getType()->isObjCQualifiedIdType()) 2105 return EvalAddr(SubExpr, refVars); 2106 else if (T->isArrayType()) 2107 return EvalVal(SubExpr, refVars); 2108 else 2109 return 0; 2110 } 2111 2112 // C++ casts. For dynamic casts, static casts, and const casts, we 2113 // are always converting from a pointer-to-pointer, so we just blow 2114 // through the cast. In the case the dynamic cast doesn't fail (and 2115 // return NULL), we take the conservative route and report cases 2116 // where we return the address of a stack variable. For Reinterpre 2117 // FIXME: The comment about is wrong; we're not always converting 2118 // from pointer to pointer. I'm guessing that this code should also 2119 // handle references to objects. 2120 case Stmt::CXXStaticCastExprClass: 2121 case Stmt::CXXDynamicCastExprClass: 2122 case Stmt::CXXConstCastExprClass: 2123 case Stmt::CXXReinterpretCastExprClass: { 2124 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 2125 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 2126 return EvalAddr(S, refVars); 2127 else 2128 return NULL; 2129 } 2130 2131 // Everything else: we simply don't reason about them. 2132 default: 2133 return NULL; 2134 } 2135} 2136 2137 2138/// EvalVal - This function is complements EvalAddr in the mutual recursion. 2139/// See the comments for EvalAddr for more details. 2140static Expr *EvalVal(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars) { 2141do { 2142 // We should only be called for evaluating non-pointer expressions, or 2143 // expressions with a pointer type that are not used as references but instead 2144 // are l-values (e.g., DeclRefExpr with a pointer type). 2145 2146 // Our "symbolic interpreter" is just a dispatch off the currently 2147 // viewed AST node. We then recursively traverse the AST by calling 2148 // EvalAddr and EvalVal appropriately. 2149 2150 E = E->IgnoreParens(); 2151 switch (E->getStmtClass()) { 2152 case Stmt::ImplicitCastExprClass: { 2153 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 2154 if (IE->getValueKind() == VK_LValue) { 2155 E = IE->getSubExpr(); 2156 continue; 2157 } 2158 return NULL; 2159 } 2160 2161 case Stmt::DeclRefExprClass: { 2162 // When we hit a DeclRefExpr we are looking at code that refers to a 2163 // variable's name. If it's not a reference variable we check if it has 2164 // local storage within the function, and if so, return the expression. 2165 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2166 2167 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2168 if (V->hasLocalStorage()) { 2169 if (!V->getType()->isReferenceType()) 2170 return DR; 2171 2172 // Reference variable, follow through to the expression that 2173 // it points to. 2174 if (V->hasInit()) { 2175 // Add the reference variable to the "trail". 2176 refVars.push_back(DR); 2177 return EvalVal(V->getInit(), refVars); 2178 } 2179 } 2180 2181 return NULL; 2182 } 2183 2184 case Stmt::UnaryOperatorClass: { 2185 // The only unary operator that make sense to handle here 2186 // is Deref. All others don't resolve to a "name." This includes 2187 // handling all sorts of rvalues passed to a unary operator. 2188 UnaryOperator *U = cast<UnaryOperator>(E); 2189 2190 if (U->getOpcode() == UO_Deref) 2191 return EvalAddr(U->getSubExpr(), refVars); 2192 2193 return NULL; 2194 } 2195 2196 case Stmt::ArraySubscriptExprClass: { 2197 // Array subscripts are potential references to data on the stack. We 2198 // retrieve the DeclRefExpr* for the array variable if it indeed 2199 // has local storage. 2200 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars); 2201 } 2202 2203 case Stmt::ConditionalOperatorClass: { 2204 // For conditional operators we need to see if either the LHS or RHS are 2205 // non-NULL Expr's. If one is non-NULL, we return it. 2206 ConditionalOperator *C = cast<ConditionalOperator>(E); 2207 2208 // Handle the GNU extension for missing LHS. 2209 if (Expr *lhsExpr = C->getLHS()) 2210 if (Expr *LHS = EvalVal(lhsExpr, refVars)) 2211 return LHS; 2212 2213 return EvalVal(C->getRHS(), refVars); 2214 } 2215 2216 // Accesses to members are potential references to data on the stack. 2217 case Stmt::MemberExprClass: { 2218 MemberExpr *M = cast<MemberExpr>(E); 2219 2220 // Check for indirect access. We only want direct field accesses. 2221 if (M->isArrow()) 2222 return NULL; 2223 2224 // Check whether the member type is itself a reference, in which case 2225 // we're not going to refer to the member, but to what the member refers to. 2226 if (M->getMemberDecl()->getType()->isReferenceType()) 2227 return NULL; 2228 2229 return EvalVal(M->getBase(), refVars); 2230 } 2231 2232 default: 2233 // Check that we don't return or take the address of a reference to a 2234 // temporary. This is only useful in C++. 2235 if (!E->isTypeDependent() && E->isRValue()) 2236 return E; 2237 2238 // Everything else: we simply don't reason about them. 2239 return NULL; 2240 } 2241} while (true); 2242} 2243 2244//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 2245 2246/// Check for comparisons of floating point operands using != and ==. 2247/// Issue a warning if these are no self-comparisons, as they are not likely 2248/// to do what the programmer intended. 2249void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 2250 bool EmitWarning = true; 2251 2252 Expr* LeftExprSansParen = lex->IgnoreParenImpCasts(); 2253 Expr* RightExprSansParen = rex->IgnoreParenImpCasts(); 2254 2255 // Special case: check for x == x (which is OK). 2256 // Do not emit warnings for such cases. 2257 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 2258 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 2259 if (DRL->getDecl() == DRR->getDecl()) 2260 EmitWarning = false; 2261 2262 2263 // Special case: check for comparisons against literals that can be exactly 2264 // represented by APFloat. In such cases, do not emit a warning. This 2265 // is a heuristic: often comparison against such literals are used to 2266 // detect if a value in a variable has not changed. This clearly can 2267 // lead to false negatives. 2268 if (EmitWarning) { 2269 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 2270 if (FLL->isExact()) 2271 EmitWarning = false; 2272 } else 2273 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 2274 if (FLR->isExact()) 2275 EmitWarning = false; 2276 } 2277 } 2278 2279 // Check for comparisons with builtin types. 2280 if (EmitWarning) 2281 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 2282 if (CL->isBuiltinCall(Context)) 2283 EmitWarning = false; 2284 2285 if (EmitWarning) 2286 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 2287 if (CR->isBuiltinCall(Context)) 2288 EmitWarning = false; 2289 2290 // Emit the diagnostic. 2291 if (EmitWarning) 2292 Diag(loc, diag::warn_floatingpoint_eq) 2293 << lex->getSourceRange() << rex->getSourceRange(); 2294} 2295 2296//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 2297//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 2298 2299namespace { 2300 2301/// Structure recording the 'active' range of an integer-valued 2302/// expression. 2303struct IntRange { 2304 /// The number of bits active in the int. 2305 unsigned Width; 2306 2307 /// True if the int is known not to have negative values. 2308 bool NonNegative; 2309 2310 IntRange(unsigned Width, bool NonNegative) 2311 : Width(Width), NonNegative(NonNegative) 2312 {} 2313 2314 /// Returns the range of the bool type. 2315 static IntRange forBoolType() { 2316 return IntRange(1, true); 2317 } 2318 2319 /// Returns the range of an opaque value of the given integral type. 2320 static IntRange forValueOfType(ASTContext &C, QualType T) { 2321 return forValueOfCanonicalType(C, 2322 T->getCanonicalTypeInternal().getTypePtr()); 2323 } 2324 2325 /// Returns the range of an opaque value of a canonical integral type. 2326 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 2327 assert(T->isCanonicalUnqualified()); 2328 2329 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2330 T = VT->getElementType().getTypePtr(); 2331 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2332 T = CT->getElementType().getTypePtr(); 2333 2334 // For enum types, use the known bit width of the enumerators. 2335 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 2336 EnumDecl *Enum = ET->getDecl(); 2337 if (!Enum->isDefinition()) 2338 return IntRange(C.getIntWidth(QualType(T, 0)), false); 2339 2340 unsigned NumPositive = Enum->getNumPositiveBits(); 2341 unsigned NumNegative = Enum->getNumNegativeBits(); 2342 2343 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 2344 } 2345 2346 const BuiltinType *BT = cast<BuiltinType>(T); 2347 assert(BT->isInteger()); 2348 2349 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2350 } 2351 2352 /// Returns the "target" range of a canonical integral type, i.e. 2353 /// the range of values expressible in the type. 2354 /// 2355 /// This matches forValueOfCanonicalType except that enums have the 2356 /// full range of their type, not the range of their enumerators. 2357 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 2358 assert(T->isCanonicalUnqualified()); 2359 2360 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2361 T = VT->getElementType().getTypePtr(); 2362 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2363 T = CT->getElementType().getTypePtr(); 2364 if (const EnumType *ET = dyn_cast<EnumType>(T)) 2365 T = ET->getDecl()->getIntegerType().getTypePtr(); 2366 2367 const BuiltinType *BT = cast<BuiltinType>(T); 2368 assert(BT->isInteger()); 2369 2370 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2371 } 2372 2373 /// Returns the supremum of two ranges: i.e. their conservative merge. 2374 static IntRange join(IntRange L, IntRange R) { 2375 return IntRange(std::max(L.Width, R.Width), 2376 L.NonNegative && R.NonNegative); 2377 } 2378 2379 /// Returns the infinum of two ranges: i.e. their aggressive merge. 2380 static IntRange meet(IntRange L, IntRange R) { 2381 return IntRange(std::min(L.Width, R.Width), 2382 L.NonNegative || R.NonNegative); 2383 } 2384}; 2385 2386IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 2387 if (value.isSigned() && value.isNegative()) 2388 return IntRange(value.getMinSignedBits(), false); 2389 2390 if (value.getBitWidth() > MaxWidth) 2391 value = value.trunc(MaxWidth); 2392 2393 // isNonNegative() just checks the sign bit without considering 2394 // signedness. 2395 return IntRange(value.getActiveBits(), true); 2396} 2397 2398IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 2399 unsigned MaxWidth) { 2400 if (result.isInt()) 2401 return GetValueRange(C, result.getInt(), MaxWidth); 2402 2403 if (result.isVector()) { 2404 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 2405 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 2406 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 2407 R = IntRange::join(R, El); 2408 } 2409 return R; 2410 } 2411 2412 if (result.isComplexInt()) { 2413 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 2414 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 2415 return IntRange::join(R, I); 2416 } 2417 2418 // This can happen with lossless casts to intptr_t of "based" lvalues. 2419 // Assume it might use arbitrary bits. 2420 // FIXME: The only reason we need to pass the type in here is to get 2421 // the sign right on this one case. It would be nice if APValue 2422 // preserved this. 2423 assert(result.isLValue()); 2424 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 2425} 2426 2427/// Pseudo-evaluate the given integer expression, estimating the 2428/// range of values it might take. 2429/// 2430/// \param MaxWidth - the width to which the value will be truncated 2431IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 2432 E = E->IgnoreParens(); 2433 2434 // Try a full evaluation first. 2435 Expr::EvalResult result; 2436 if (E->Evaluate(result, C)) 2437 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 2438 2439 // I think we only want to look through implicit casts here; if the 2440 // user has an explicit widening cast, we should treat the value as 2441 // being of the new, wider type. 2442 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 2443 if (CE->getCastKind() == CK_NoOp) 2444 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 2445 2446 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 2447 2448 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 2449 2450 // Assume that non-integer casts can span the full range of the type. 2451 if (!isIntegerCast) 2452 return OutputTypeRange; 2453 2454 IntRange SubRange 2455 = GetExprRange(C, CE->getSubExpr(), 2456 std::min(MaxWidth, OutputTypeRange.Width)); 2457 2458 // Bail out if the subexpr's range is as wide as the cast type. 2459 if (SubRange.Width >= OutputTypeRange.Width) 2460 return OutputTypeRange; 2461 2462 // Otherwise, we take the smaller width, and we're non-negative if 2463 // either the output type or the subexpr is. 2464 return IntRange(SubRange.Width, 2465 SubRange.NonNegative || OutputTypeRange.NonNegative); 2466 } 2467 2468 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 2469 // If we can fold the condition, just take that operand. 2470 bool CondResult; 2471 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 2472 return GetExprRange(C, CondResult ? CO->getTrueExpr() 2473 : CO->getFalseExpr(), 2474 MaxWidth); 2475 2476 // Otherwise, conservatively merge. 2477 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 2478 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 2479 return IntRange::join(L, R); 2480 } 2481 2482 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 2483 switch (BO->getOpcode()) { 2484 2485 // Boolean-valued operations are single-bit and positive. 2486 case BO_LAnd: 2487 case BO_LOr: 2488 case BO_LT: 2489 case BO_GT: 2490 case BO_LE: 2491 case BO_GE: 2492 case BO_EQ: 2493 case BO_NE: 2494 return IntRange::forBoolType(); 2495 2496 // The type of these compound assignments is the type of the LHS, 2497 // so the RHS is not necessarily an integer. 2498 case BO_MulAssign: 2499 case BO_DivAssign: 2500 case BO_RemAssign: 2501 case BO_AddAssign: 2502 case BO_SubAssign: 2503 return IntRange::forValueOfType(C, E->getType()); 2504 2505 // Operations with opaque sources are black-listed. 2506 case BO_PtrMemD: 2507 case BO_PtrMemI: 2508 return IntRange::forValueOfType(C, E->getType()); 2509 2510 // Bitwise-and uses the *infinum* of the two source ranges. 2511 case BO_And: 2512 case BO_AndAssign: 2513 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 2514 GetExprRange(C, BO->getRHS(), MaxWidth)); 2515 2516 // Left shift gets black-listed based on a judgement call. 2517 case BO_Shl: 2518 // ...except that we want to treat '1 << (blah)' as logically 2519 // positive. It's an important idiom. 2520 if (IntegerLiteral *I 2521 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 2522 if (I->getValue() == 1) { 2523 IntRange R = IntRange::forValueOfType(C, E->getType()); 2524 return IntRange(R.Width, /*NonNegative*/ true); 2525 } 2526 } 2527 // fallthrough 2528 2529 case BO_ShlAssign: 2530 return IntRange::forValueOfType(C, E->getType()); 2531 2532 // Right shift by a constant can narrow its left argument. 2533 case BO_Shr: 2534 case BO_ShrAssign: { 2535 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2536 2537 // If the shift amount is a positive constant, drop the width by 2538 // that much. 2539 llvm::APSInt shift; 2540 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 2541 shift.isNonNegative()) { 2542 unsigned zext = shift.getZExtValue(); 2543 if (zext >= L.Width) 2544 L.Width = (L.NonNegative ? 0 : 1); 2545 else 2546 L.Width -= zext; 2547 } 2548 2549 return L; 2550 } 2551 2552 // Comma acts as its right operand. 2553 case BO_Comma: 2554 return GetExprRange(C, BO->getRHS(), MaxWidth); 2555 2556 // Black-list pointer subtractions. 2557 case BO_Sub: 2558 if (BO->getLHS()->getType()->isPointerType()) 2559 return IntRange::forValueOfType(C, E->getType()); 2560 // fallthrough 2561 2562 default: 2563 break; 2564 } 2565 2566 // Treat every other operator as if it were closed on the 2567 // narrowest type that encompasses both operands. 2568 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2569 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 2570 return IntRange::join(L, R); 2571 } 2572 2573 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 2574 switch (UO->getOpcode()) { 2575 // Boolean-valued operations are white-listed. 2576 case UO_LNot: 2577 return IntRange::forBoolType(); 2578 2579 // Operations with opaque sources are black-listed. 2580 case UO_Deref: 2581 case UO_AddrOf: // should be impossible 2582 return IntRange::forValueOfType(C, E->getType()); 2583 2584 default: 2585 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 2586 } 2587 } 2588 2589 if (dyn_cast<OffsetOfExpr>(E)) { 2590 IntRange::forValueOfType(C, E->getType()); 2591 } 2592 2593 FieldDecl *BitField = E->getBitField(); 2594 if (BitField) { 2595 llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C); 2596 unsigned BitWidth = BitWidthAP.getZExtValue(); 2597 2598 return IntRange(BitWidth, 2599 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 2600 } 2601 2602 return IntRange::forValueOfType(C, E->getType()); 2603} 2604 2605IntRange GetExprRange(ASTContext &C, Expr *E) { 2606 return GetExprRange(C, E, C.getIntWidth(E->getType())); 2607} 2608 2609/// Checks whether the given value, which currently has the given 2610/// source semantics, has the same value when coerced through the 2611/// target semantics. 2612bool IsSameFloatAfterCast(const llvm::APFloat &value, 2613 const llvm::fltSemantics &Src, 2614 const llvm::fltSemantics &Tgt) { 2615 llvm::APFloat truncated = value; 2616 2617 bool ignored; 2618 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 2619 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 2620 2621 return truncated.bitwiseIsEqual(value); 2622} 2623 2624/// Checks whether the given value, which currently has the given 2625/// source semantics, has the same value when coerced through the 2626/// target semantics. 2627/// 2628/// The value might be a vector of floats (or a complex number). 2629bool IsSameFloatAfterCast(const APValue &value, 2630 const llvm::fltSemantics &Src, 2631 const llvm::fltSemantics &Tgt) { 2632 if (value.isFloat()) 2633 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 2634 2635 if (value.isVector()) { 2636 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 2637 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 2638 return false; 2639 return true; 2640 } 2641 2642 assert(value.isComplexFloat()); 2643 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 2644 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 2645} 2646 2647void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 2648 2649static bool IsZero(Sema &S, Expr *E) { 2650 // Suppress cases where we are comparing against an enum constant. 2651 if (const DeclRefExpr *DR = 2652 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 2653 if (isa<EnumConstantDecl>(DR->getDecl())) 2654 return false; 2655 2656 // Suppress cases where the '0' value is expanded from a macro. 2657 if (E->getLocStart().isMacroID()) 2658 return false; 2659 2660 llvm::APSInt Value; 2661 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 2662} 2663 2664static bool HasEnumType(Expr *E) { 2665 // Strip off implicit integral promotions. 2666 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 2667 if (ICE->getCastKind() != CK_IntegralCast && 2668 ICE->getCastKind() != CK_NoOp) 2669 break; 2670 E = ICE->getSubExpr(); 2671 } 2672 2673 return E->getType()->isEnumeralType(); 2674} 2675 2676void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 2677 BinaryOperatorKind op = E->getOpcode(); 2678 if (E->isValueDependent()) 2679 return; 2680 2681 if (op == BO_LT && IsZero(S, E->getRHS())) { 2682 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2683 << "< 0" << "false" << HasEnumType(E->getLHS()) 2684 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2685 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 2686 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2687 << ">= 0" << "true" << HasEnumType(E->getLHS()) 2688 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2689 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 2690 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2691 << "0 >" << "false" << HasEnumType(E->getRHS()) 2692 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2693 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 2694 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2695 << "0 <=" << "true" << HasEnumType(E->getRHS()) 2696 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2697 } 2698} 2699 2700/// Analyze the operands of the given comparison. Implements the 2701/// fallback case from AnalyzeComparison. 2702void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 2703 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 2704 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 2705} 2706 2707/// \brief Implements -Wsign-compare. 2708/// 2709/// \param lex the left-hand expression 2710/// \param rex the right-hand expression 2711/// \param OpLoc the location of the joining operator 2712/// \param BinOpc binary opcode or 0 2713void AnalyzeComparison(Sema &S, BinaryOperator *E) { 2714 // The type the comparison is being performed in. 2715 QualType T = E->getLHS()->getType(); 2716 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 2717 && "comparison with mismatched types"); 2718 2719 // We don't do anything special if this isn't an unsigned integral 2720 // comparison: we're only interested in integral comparisons, and 2721 // signed comparisons only happen in cases we don't care to warn about. 2722 // 2723 // We also don't care about value-dependent expressions or expressions 2724 // whose result is a constant. 2725 if (!T->hasUnsignedIntegerRepresentation() 2726 || E->isValueDependent() || E->isIntegerConstantExpr(S.Context)) 2727 return AnalyzeImpConvsInComparison(S, E); 2728 2729 Expr *lex = E->getLHS()->IgnoreParenImpCasts(); 2730 Expr *rex = E->getRHS()->IgnoreParenImpCasts(); 2731 2732 // Check to see if one of the (unmodified) operands is of different 2733 // signedness. 2734 Expr *signedOperand, *unsignedOperand; 2735 if (lex->getType()->hasSignedIntegerRepresentation()) { 2736 assert(!rex->getType()->hasSignedIntegerRepresentation() && 2737 "unsigned comparison between two signed integer expressions?"); 2738 signedOperand = lex; 2739 unsignedOperand = rex; 2740 } else if (rex->getType()->hasSignedIntegerRepresentation()) { 2741 signedOperand = rex; 2742 unsignedOperand = lex; 2743 } else { 2744 CheckTrivialUnsignedComparison(S, E); 2745 return AnalyzeImpConvsInComparison(S, E); 2746 } 2747 2748 // Otherwise, calculate the effective range of the signed operand. 2749 IntRange signedRange = GetExprRange(S.Context, signedOperand); 2750 2751 // Go ahead and analyze implicit conversions in the operands. Note 2752 // that we skip the implicit conversions on both sides. 2753 AnalyzeImplicitConversions(S, lex, E->getOperatorLoc()); 2754 AnalyzeImplicitConversions(S, rex, E->getOperatorLoc()); 2755 2756 // If the signed range is non-negative, -Wsign-compare won't fire, 2757 // but we should still check for comparisons which are always true 2758 // or false. 2759 if (signedRange.NonNegative) 2760 return CheckTrivialUnsignedComparison(S, E); 2761 2762 // For (in)equality comparisons, if the unsigned operand is a 2763 // constant which cannot collide with a overflowed signed operand, 2764 // then reinterpreting the signed operand as unsigned will not 2765 // change the result of the comparison. 2766 if (E->isEqualityOp()) { 2767 unsigned comparisonWidth = S.Context.getIntWidth(T); 2768 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 2769 2770 // We should never be unable to prove that the unsigned operand is 2771 // non-negative. 2772 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 2773 2774 if (unsignedRange.Width < comparisonWidth) 2775 return; 2776 } 2777 2778 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison) 2779 << lex->getType() << rex->getType() 2780 << lex->getSourceRange() << rex->getSourceRange(); 2781} 2782 2783/// Analyzes an attempt to assign the given value to a bitfield. 2784/// 2785/// Returns true if there was something fishy about the attempt. 2786bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 2787 SourceLocation InitLoc) { 2788 assert(Bitfield->isBitField()); 2789 if (Bitfield->isInvalidDecl()) 2790 return false; 2791 2792 // White-list bool bitfields. 2793 if (Bitfield->getType()->isBooleanType()) 2794 return false; 2795 2796 // Ignore value- or type-dependent expressions. 2797 if (Bitfield->getBitWidth()->isValueDependent() || 2798 Bitfield->getBitWidth()->isTypeDependent() || 2799 Init->isValueDependent() || 2800 Init->isTypeDependent()) 2801 return false; 2802 2803 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 2804 2805 llvm::APSInt Width(32); 2806 Expr::EvalResult InitValue; 2807 if (!Bitfield->getBitWidth()->isIntegerConstantExpr(Width, S.Context) || 2808 !OriginalInit->Evaluate(InitValue, S.Context) || 2809 !InitValue.Val.isInt()) 2810 return false; 2811 2812 const llvm::APSInt &Value = InitValue.Val.getInt(); 2813 unsigned OriginalWidth = Value.getBitWidth(); 2814 unsigned FieldWidth = Width.getZExtValue(); 2815 2816 if (OriginalWidth <= FieldWidth) 2817 return false; 2818 2819 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 2820 2821 // It's fairly common to write values into signed bitfields 2822 // that, if sign-extended, would end up becoming a different 2823 // value. We don't want to warn about that. 2824 if (Value.isSigned() && Value.isNegative()) 2825 TruncatedValue = TruncatedValue.sext(OriginalWidth); 2826 else 2827 TruncatedValue = TruncatedValue.zext(OriginalWidth); 2828 2829 if (Value == TruncatedValue) 2830 return false; 2831 2832 std::string PrettyValue = Value.toString(10); 2833 std::string PrettyTrunc = TruncatedValue.toString(10); 2834 2835 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 2836 << PrettyValue << PrettyTrunc << OriginalInit->getType() 2837 << Init->getSourceRange(); 2838 2839 return true; 2840} 2841 2842/// Analyze the given simple or compound assignment for warning-worthy 2843/// operations. 2844void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 2845 // Just recurse on the LHS. 2846 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 2847 2848 // We want to recurse on the RHS as normal unless we're assigning to 2849 // a bitfield. 2850 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 2851 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 2852 E->getOperatorLoc())) { 2853 // Recurse, ignoring any implicit conversions on the RHS. 2854 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 2855 E->getOperatorLoc()); 2856 } 2857 } 2858 2859 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 2860} 2861 2862/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 2863void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 2864 SourceLocation CContext, unsigned diag) { 2865 S.Diag(E->getExprLoc(), diag) 2866 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 2867} 2868 2869/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 2870void DiagnoseImpCast(Sema &S, Expr *E, QualType T, SourceLocation CContext, 2871 unsigned diag) { 2872 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag); 2873} 2874 2875/// Diagnose an implicit cast from a literal expression. Also attemps to supply 2876/// fixit hints when the cast wouldn't lose information to simply write the 2877/// expression with the expected type. 2878void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 2879 SourceLocation CContext) { 2880 // Emit the primary warning first, then try to emit a fixit hint note if 2881 // reasonable. 2882 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 2883 << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext); 2884 2885 const llvm::APFloat &Value = FL->getValue(); 2886 2887 // Don't attempt to fix PPC double double literals. 2888 if (&Value.getSemantics() == &llvm::APFloat::PPCDoubleDouble) 2889 return; 2890 2891 // Try to convert this exactly to an 64-bit integer. FIXME: It would be 2892 // nice to support arbitrarily large integers here. 2893 bool isExact = false; 2894 uint64_t IntegerPart; 2895 if (Value.convertToInteger(&IntegerPart, 64, /*isSigned=*/true, 2896 llvm::APFloat::rmTowardZero, &isExact) 2897 != llvm::APFloat::opOK || !isExact) 2898 return; 2899 2900 llvm::APInt IntegerValue(64, IntegerPart, /*isSigned=*/true); 2901 2902 std::string LiteralValue = IntegerValue.toString(10, /*isSigned=*/true); 2903 S.Diag(FL->getExprLoc(), diag::note_fix_integral_float_as_integer) 2904 << FixItHint::CreateReplacement(FL->getSourceRange(), LiteralValue); 2905} 2906 2907std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 2908 if (!Range.Width) return "0"; 2909 2910 llvm::APSInt ValueInRange = Value; 2911 ValueInRange.setIsSigned(!Range.NonNegative); 2912 ValueInRange = ValueInRange.trunc(Range.Width); 2913 return ValueInRange.toString(10); 2914} 2915 2916static bool isFromSystemMacro(Sema &S, SourceLocation loc) { 2917 SourceManager &smgr = S.Context.getSourceManager(); 2918 return loc.isMacroID() && smgr.isInSystemHeader(smgr.getSpellingLoc(loc)); 2919} 2920 2921void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 2922 SourceLocation CC, bool *ICContext = 0) { 2923 if (E->isTypeDependent() || E->isValueDependent()) return; 2924 2925 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 2926 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 2927 if (Source == Target) return; 2928 if (Target->isDependentType()) return; 2929 2930 // If the conversion context location is invalid don't complain. 2931 // We also don't want to emit a warning if the issue occurs from the 2932 // instantiation of a system macro. The problem is that 'getSpellingLoc()' 2933 // is slow, so we delay this check as long as possible. Once we detect 2934 // we are in that scenario, we just return. 2935 if (CC.isInvalid()) 2936 return; 2937 2938 // Never diagnose implicit casts to bool. 2939 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 2940 return; 2941 2942 // Strip vector types. 2943 if (isa<VectorType>(Source)) { 2944 if (!isa<VectorType>(Target)) { 2945 if (isFromSystemMacro(S, CC)) 2946 return; 2947 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 2948 } 2949 2950 // If the vector cast is cast between two vectors of the same size, it is 2951 // a bitcast, not a conversion. 2952 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 2953 return; 2954 2955 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 2956 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 2957 } 2958 2959 // Strip complex types. 2960 if (isa<ComplexType>(Source)) { 2961 if (!isa<ComplexType>(Target)) { 2962 if (isFromSystemMacro(S, CC)) 2963 return; 2964 2965 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 2966 } 2967 2968 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 2969 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 2970 } 2971 2972 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 2973 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 2974 2975 // If the source is floating point... 2976 if (SourceBT && SourceBT->isFloatingPoint()) { 2977 // ...and the target is floating point... 2978 if (TargetBT && TargetBT->isFloatingPoint()) { 2979 // ...then warn if we're dropping FP rank. 2980 2981 // Builtin FP kinds are ordered by increasing FP rank. 2982 if (SourceBT->getKind() > TargetBT->getKind()) { 2983 // Don't warn about float constants that are precisely 2984 // representable in the target type. 2985 Expr::EvalResult result; 2986 if (E->Evaluate(result, S.Context)) { 2987 // Value might be a float, a float vector, or a float complex. 2988 if (IsSameFloatAfterCast(result.Val, 2989 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 2990 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 2991 return; 2992 } 2993 2994 if (isFromSystemMacro(S, CC)) 2995 return; 2996 2997 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 2998 } 2999 return; 3000 } 3001 3002 // If the target is integral, always warn. 3003 if ((TargetBT && TargetBT->isInteger())) { 3004 if (isFromSystemMacro(S, CC)) 3005 return; 3006 3007 Expr *InnerE = E->IgnoreParenImpCasts(); 3008 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 3009 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 3010 } else { 3011 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 3012 } 3013 } 3014 3015 return; 3016 } 3017 3018 if (!Source->isIntegerType() || !Target->isIntegerType()) 3019 return; 3020 3021 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 3022 == Expr::NPCK_GNUNull) && Target->isIntegerType()) { 3023 S.Diag(E->getExprLoc(), diag::warn_impcast_null_pointer_to_integer) 3024 << E->getSourceRange() << clang::SourceRange(CC); 3025 return; 3026 } 3027 3028 IntRange SourceRange = GetExprRange(S.Context, E); 3029 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 3030 3031 if (SourceRange.Width > TargetRange.Width) { 3032 // If the source is a constant, use a default-on diagnostic. 3033 // TODO: this should happen for bitfield stores, too. 3034 llvm::APSInt Value(32); 3035 if (E->isIntegerConstantExpr(Value, S.Context)) { 3036 if (isFromSystemMacro(S, CC)) 3037 return; 3038 3039 std::string PrettySourceValue = Value.toString(10); 3040 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 3041 3042 S.Diag(E->getExprLoc(), diag::warn_impcast_integer_precision_constant) 3043 << PrettySourceValue << PrettyTargetValue 3044 << E->getType() << T << E->getSourceRange() << clang::SourceRange(CC); 3045 return; 3046 } 3047 3048 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 3049 if (isFromSystemMacro(S, CC)) 3050 return; 3051 3052 if (SourceRange.Width == 64 && TargetRange.Width == 32) 3053 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32); 3054 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 3055 } 3056 3057 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 3058 (!TargetRange.NonNegative && SourceRange.NonNegative && 3059 SourceRange.Width == TargetRange.Width)) { 3060 3061 if (isFromSystemMacro(S, CC)) 3062 return; 3063 3064 unsigned DiagID = diag::warn_impcast_integer_sign; 3065 3066 // Traditionally, gcc has warned about this under -Wsign-compare. 3067 // We also want to warn about it in -Wconversion. 3068 // So if -Wconversion is off, use a completely identical diagnostic 3069 // in the sign-compare group. 3070 // The conditional-checking code will 3071 if (ICContext) { 3072 DiagID = diag::warn_impcast_integer_sign_conditional; 3073 *ICContext = true; 3074 } 3075 3076 return DiagnoseImpCast(S, E, T, CC, DiagID); 3077 } 3078 3079 // Diagnose conversions between different enumeration types. 3080 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 3081 // type, to give us better diagnostics. 3082 QualType SourceType = E->getType(); 3083 if (!S.getLangOptions().CPlusPlus) { 3084 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 3085 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 3086 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 3087 SourceType = S.Context.getTypeDeclType(Enum); 3088 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 3089 } 3090 } 3091 3092 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 3093 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 3094 if ((SourceEnum->getDecl()->getIdentifier() || 3095 SourceEnum->getDecl()->getTypedefNameForAnonDecl()) && 3096 (TargetEnum->getDecl()->getIdentifier() || 3097 TargetEnum->getDecl()->getTypedefNameForAnonDecl()) && 3098 SourceEnum != TargetEnum) { 3099 if (isFromSystemMacro(S, CC)) 3100 return; 3101 3102 return DiagnoseImpCast(S, E, SourceType, T, CC, 3103 diag::warn_impcast_different_enum_types); 3104 } 3105 3106 return; 3107} 3108 3109void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 3110 3111void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 3112 SourceLocation CC, bool &ICContext) { 3113 E = E->IgnoreParenImpCasts(); 3114 3115 if (isa<ConditionalOperator>(E)) 3116 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 3117 3118 AnalyzeImplicitConversions(S, E, CC); 3119 if (E->getType() != T) 3120 return CheckImplicitConversion(S, E, T, CC, &ICContext); 3121 return; 3122} 3123 3124void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 3125 SourceLocation CC = E->getQuestionLoc(); 3126 3127 AnalyzeImplicitConversions(S, E->getCond(), CC); 3128 3129 bool Suspicious = false; 3130 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 3131 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 3132 3133 // If -Wconversion would have warned about either of the candidates 3134 // for a signedness conversion to the context type... 3135 if (!Suspicious) return; 3136 3137 // ...but it's currently ignored... 3138 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 3139 CC)) 3140 return; 3141 3142 // ...and -Wsign-compare isn't... 3143 if (!S.Diags.getDiagnosticLevel(diag::warn_mixed_sign_conditional, CC)) 3144 return; 3145 3146 // ...then check whether it would have warned about either of the 3147 // candidates for a signedness conversion to the condition type. 3148 if (E->getType() != T) { 3149 Suspicious = false; 3150 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 3151 E->getType(), CC, &Suspicious); 3152 if (!Suspicious) 3153 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 3154 E->getType(), CC, &Suspicious); 3155 if (!Suspicious) 3156 return; 3157 } 3158 3159 // If so, emit a diagnostic under -Wsign-compare. 3160 Expr *lex = E->getTrueExpr()->IgnoreParenImpCasts(); 3161 Expr *rex = E->getFalseExpr()->IgnoreParenImpCasts(); 3162 S.Diag(E->getQuestionLoc(), diag::warn_mixed_sign_conditional) 3163 << lex->getType() << rex->getType() 3164 << lex->getSourceRange() << rex->getSourceRange(); 3165} 3166 3167/// AnalyzeImplicitConversions - Find and report any interesting 3168/// implicit conversions in the given expression. There are a couple 3169/// of competing diagnostics here, -Wconversion and -Wsign-compare. 3170void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 3171 QualType T = OrigE->getType(); 3172 Expr *E = OrigE->IgnoreParenImpCasts(); 3173 3174 // For conditional operators, we analyze the arguments as if they 3175 // were being fed directly into the output. 3176 if (isa<ConditionalOperator>(E)) { 3177 ConditionalOperator *CO = cast<ConditionalOperator>(E); 3178 CheckConditionalOperator(S, CO, T); 3179 return; 3180 } 3181 3182 // Go ahead and check any implicit conversions we might have skipped. 3183 // The non-canonical typecheck is just an optimization; 3184 // CheckImplicitConversion will filter out dead implicit conversions. 3185 if (E->getType() != T) 3186 CheckImplicitConversion(S, E, T, CC); 3187 3188 // Now continue drilling into this expression. 3189 3190 // Skip past explicit casts. 3191 if (isa<ExplicitCastExpr>(E)) { 3192 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 3193 return AnalyzeImplicitConversions(S, E, CC); 3194 } 3195 3196 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 3197 // Do a somewhat different check with comparison operators. 3198 if (BO->isComparisonOp()) 3199 return AnalyzeComparison(S, BO); 3200 3201 // And with assignments and compound assignments. 3202 if (BO->isAssignmentOp()) 3203 return AnalyzeAssignment(S, BO); 3204 } 3205 3206 // These break the otherwise-useful invariant below. Fortunately, 3207 // we don't really need to recurse into them, because any internal 3208 // expressions should have been analyzed already when they were 3209 // built into statements. 3210 if (isa<StmtExpr>(E)) return; 3211 3212 // Don't descend into unevaluated contexts. 3213 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 3214 3215 // Now just recurse over the expression's children. 3216 CC = E->getExprLoc(); 3217 for (Stmt::child_range I = E->children(); I; ++I) 3218 AnalyzeImplicitConversions(S, cast<Expr>(*I), CC); 3219} 3220 3221} // end anonymous namespace 3222 3223/// Diagnoses "dangerous" implicit conversions within the given 3224/// expression (which is a full expression). Implements -Wconversion 3225/// and -Wsign-compare. 3226/// 3227/// \param CC the "context" location of the implicit conversion, i.e. 3228/// the most location of the syntactic entity requiring the implicit 3229/// conversion 3230void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 3231 // Don't diagnose in unevaluated contexts. 3232 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 3233 return; 3234 3235 // Don't diagnose for value- or type-dependent expressions. 3236 if (E->isTypeDependent() || E->isValueDependent()) 3237 return; 3238 3239 // This is not the right CC for (e.g.) a variable initialization. 3240 AnalyzeImplicitConversions(*this, E, CC); 3241} 3242 3243void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 3244 FieldDecl *BitField, 3245 Expr *Init) { 3246 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 3247} 3248 3249/// CheckParmsForFunctionDef - Check that the parameters of the given 3250/// function are appropriate for the definition of a function. This 3251/// takes care of any checks that cannot be performed on the 3252/// declaration itself, e.g., that the types of each of the function 3253/// parameters are complete. 3254bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 3255 bool CheckParameterNames) { 3256 bool HasInvalidParm = false; 3257 for (; P != PEnd; ++P) { 3258 ParmVarDecl *Param = *P; 3259 3260 // C99 6.7.5.3p4: the parameters in a parameter type list in a 3261 // function declarator that is part of a function definition of 3262 // that function shall not have incomplete type. 3263 // 3264 // This is also C++ [dcl.fct]p6. 3265 if (!Param->isInvalidDecl() && 3266 RequireCompleteType(Param->getLocation(), Param->getType(), 3267 diag::err_typecheck_decl_incomplete_type)) { 3268 Param->setInvalidDecl(); 3269 HasInvalidParm = true; 3270 } 3271 3272 // C99 6.9.1p5: If the declarator includes a parameter type list, the 3273 // declaration of each parameter shall include an identifier. 3274 if (CheckParameterNames && 3275 Param->getIdentifier() == 0 && 3276 !Param->isImplicit() && 3277 !getLangOptions().CPlusPlus) 3278 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 3279 3280 // C99 6.7.5.3p12: 3281 // If the function declarator is not part of a definition of that 3282 // function, parameters may have incomplete type and may use the [*] 3283 // notation in their sequences of declarator specifiers to specify 3284 // variable length array types. 3285 QualType PType = Param->getOriginalType(); 3286 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 3287 if (AT->getSizeModifier() == ArrayType::Star) { 3288 // FIXME: This diagnosic should point the the '[*]' if source-location 3289 // information is added for it. 3290 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 3291 } 3292 } 3293 } 3294 3295 return HasInvalidParm; 3296} 3297 3298/// CheckCastAlign - Implements -Wcast-align, which warns when a 3299/// pointer cast increases the alignment requirements. 3300void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 3301 // This is actually a lot of work to potentially be doing on every 3302 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 3303 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 3304 TRange.getBegin()) 3305 == Diagnostic::Ignored) 3306 return; 3307 3308 // Ignore dependent types. 3309 if (T->isDependentType() || Op->getType()->isDependentType()) 3310 return; 3311 3312 // Require that the destination be a pointer type. 3313 const PointerType *DestPtr = T->getAs<PointerType>(); 3314 if (!DestPtr) return; 3315 3316 // If the destination has alignment 1, we're done. 3317 QualType DestPointee = DestPtr->getPointeeType(); 3318 if (DestPointee->isIncompleteType()) return; 3319 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 3320 if (DestAlign.isOne()) return; 3321 3322 // Require that the source be a pointer type. 3323 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 3324 if (!SrcPtr) return; 3325 QualType SrcPointee = SrcPtr->getPointeeType(); 3326 3327 // Whitelist casts from cv void*. We already implicitly 3328 // whitelisted casts to cv void*, since they have alignment 1. 3329 // Also whitelist casts involving incomplete types, which implicitly 3330 // includes 'void'. 3331 if (SrcPointee->isIncompleteType()) return; 3332 3333 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 3334 if (SrcAlign >= DestAlign) return; 3335 3336 Diag(TRange.getBegin(), diag::warn_cast_align) 3337 << Op->getType() << T 3338 << static_cast<unsigned>(SrcAlign.getQuantity()) 3339 << static_cast<unsigned>(DestAlign.getQuantity()) 3340 << TRange << Op->getSourceRange(); 3341} 3342 3343static void CheckArrayAccess_Check(Sema &S, 3344 const clang::ArraySubscriptExpr *E) { 3345 const Expr *BaseExpr = E->getBase()->IgnoreParenImpCasts(); 3346 const ConstantArrayType *ArrayTy = 3347 S.Context.getAsConstantArrayType(BaseExpr->getType()); 3348 if (!ArrayTy) 3349 return; 3350 3351 const Expr *IndexExpr = E->getIdx(); 3352 if (IndexExpr->isValueDependent()) 3353 return; 3354 llvm::APSInt index; 3355 if (!IndexExpr->isIntegerConstantExpr(index, S.Context)) 3356 return; 3357 3358 if (index.isUnsigned() || !index.isNegative()) { 3359 llvm::APInt size = ArrayTy->getSize(); 3360 if (!size.isStrictlyPositive()) 3361 return; 3362 if (size.getBitWidth() > index.getBitWidth()) 3363 index = index.sext(size.getBitWidth()); 3364 else if (size.getBitWidth() < index.getBitWidth()) 3365 size = size.sext(index.getBitWidth()); 3366 3367 if (index.slt(size)) 3368 return; 3369 3370 S.DiagRuntimeBehavior(E->getBase()->getLocStart(), BaseExpr, 3371 S.PDiag(diag::warn_array_index_exceeds_bounds) 3372 << index.toString(10, true) 3373 << size.toString(10, true) 3374 << IndexExpr->getSourceRange()); 3375 } else { 3376 S.DiagRuntimeBehavior(E->getBase()->getLocStart(), BaseExpr, 3377 S.PDiag(diag::warn_array_index_precedes_bounds) 3378 << index.toString(10, true) 3379 << IndexExpr->getSourceRange()); 3380 } 3381 3382 const NamedDecl *ND = NULL; 3383 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 3384 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 3385 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 3386 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 3387 if (ND) 3388 S.DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 3389 S.PDiag(diag::note_array_index_out_of_bounds) 3390 << ND->getDeclName()); 3391} 3392 3393void Sema::CheckArrayAccess(const Expr *expr) { 3394 while (true) { 3395 expr = expr->IgnoreParens(); 3396 switch (expr->getStmtClass()) { 3397 case Stmt::ArraySubscriptExprClass: 3398 CheckArrayAccess_Check(*this, cast<ArraySubscriptExpr>(expr)); 3399 return; 3400 case Stmt::ConditionalOperatorClass: { 3401 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 3402 if (const Expr *lhs = cond->getLHS()) 3403 CheckArrayAccess(lhs); 3404 if (const Expr *rhs = cond->getRHS()) 3405 CheckArrayAccess(rhs); 3406 return; 3407 } 3408 default: 3409 return; 3410 } 3411 } 3412} 3413 3414//===--- CHECK: Objective-C retain cycles ----------------------------------// 3415 3416namespace { 3417 struct RetainCycleOwner { 3418 RetainCycleOwner() : Variable(0), Indirect(false) {} 3419 VarDecl *Variable; 3420 SourceRange Range; 3421 SourceLocation Loc; 3422 bool Indirect; 3423 3424 void setLocsFrom(Expr *e) { 3425 Loc = e->getExprLoc(); 3426 Range = e->getSourceRange(); 3427 } 3428 }; 3429} 3430 3431/// Consider whether capturing the given variable can possibly lead to 3432/// a retain cycle. 3433static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 3434 // In ARC, it's captured strongly iff the variable has __strong 3435 // lifetime. In MRR, it's captured strongly if the variable is 3436 // __block and has an appropriate type. 3437 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 3438 return false; 3439 3440 owner.Variable = var; 3441 owner.setLocsFrom(ref); 3442 return true; 3443} 3444 3445static bool findRetainCycleOwner(Expr *e, RetainCycleOwner &owner) { 3446 while (true) { 3447 e = e->IgnoreParens(); 3448 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 3449 switch (cast->getCastKind()) { 3450 case CK_BitCast: 3451 case CK_LValueBitCast: 3452 case CK_LValueToRValue: 3453 e = cast->getSubExpr(); 3454 continue; 3455 3456 case CK_GetObjCProperty: { 3457 // Bail out if this isn't a strong explicit property. 3458 const ObjCPropertyRefExpr *pre = cast->getSubExpr()->getObjCProperty(); 3459 if (pre->isImplicitProperty()) return false; 3460 ObjCPropertyDecl *property = pre->getExplicitProperty(); 3461 if (!(property->getPropertyAttributes() & 3462 (ObjCPropertyDecl::OBJC_PR_retain | 3463 ObjCPropertyDecl::OBJC_PR_copy | 3464 ObjCPropertyDecl::OBJC_PR_strong)) && 3465 !(property->getPropertyIvarDecl() && 3466 property->getPropertyIvarDecl()->getType() 3467 .getObjCLifetime() == Qualifiers::OCL_Strong)) 3468 return false; 3469 3470 owner.Indirect = true; 3471 e = const_cast<Expr*>(pre->getBase()); 3472 continue; 3473 } 3474 3475 default: 3476 return false; 3477 } 3478 } 3479 3480 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 3481 ObjCIvarDecl *ivar = ref->getDecl(); 3482 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 3483 return false; 3484 3485 // Try to find a retain cycle in the base. 3486 if (!findRetainCycleOwner(ref->getBase(), owner)) 3487 return false; 3488 3489 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 3490 owner.Indirect = true; 3491 return true; 3492 } 3493 3494 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 3495 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 3496 if (!var) return false; 3497 return considerVariable(var, ref, owner); 3498 } 3499 3500 if (BlockDeclRefExpr *ref = dyn_cast<BlockDeclRefExpr>(e)) { 3501 owner.Variable = ref->getDecl(); 3502 owner.setLocsFrom(ref); 3503 return true; 3504 } 3505 3506 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 3507 if (member->isArrow()) return false; 3508 3509 // Don't count this as an indirect ownership. 3510 e = member->getBase(); 3511 continue; 3512 } 3513 3514 // Array ivars? 3515 3516 return false; 3517 } 3518} 3519 3520namespace { 3521 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 3522 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 3523 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 3524 Variable(variable), Capturer(0) {} 3525 3526 VarDecl *Variable; 3527 Expr *Capturer; 3528 3529 void VisitDeclRefExpr(DeclRefExpr *ref) { 3530 if (ref->getDecl() == Variable && !Capturer) 3531 Capturer = ref; 3532 } 3533 3534 void VisitBlockDeclRefExpr(BlockDeclRefExpr *ref) { 3535 if (ref->getDecl() == Variable && !Capturer) 3536 Capturer = ref; 3537 } 3538 3539 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 3540 if (Capturer) return; 3541 Visit(ref->getBase()); 3542 if (Capturer && ref->isFreeIvar()) 3543 Capturer = ref; 3544 } 3545 3546 void VisitBlockExpr(BlockExpr *block) { 3547 // Look inside nested blocks 3548 if (block->getBlockDecl()->capturesVariable(Variable)) 3549 Visit(block->getBlockDecl()->getBody()); 3550 } 3551 }; 3552} 3553 3554/// Check whether the given argument is a block which captures a 3555/// variable. 3556static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 3557 assert(owner.Variable && owner.Loc.isValid()); 3558 3559 e = e->IgnoreParenCasts(); 3560 BlockExpr *block = dyn_cast<BlockExpr>(e); 3561 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 3562 return 0; 3563 3564 FindCaptureVisitor visitor(S.Context, owner.Variable); 3565 visitor.Visit(block->getBlockDecl()->getBody()); 3566 return visitor.Capturer; 3567} 3568 3569static void diagnoseRetainCycle(Sema &S, Expr *capturer, 3570 RetainCycleOwner &owner) { 3571 assert(capturer); 3572 assert(owner.Variable && owner.Loc.isValid()); 3573 3574 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 3575 << owner.Variable << capturer->getSourceRange(); 3576 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 3577 << owner.Indirect << owner.Range; 3578} 3579 3580/// Check for a keyword selector that starts with the word 'add' or 3581/// 'set'. 3582static bool isSetterLikeSelector(Selector sel) { 3583 if (sel.isUnarySelector()) return false; 3584 3585 llvm::StringRef str = sel.getNameForSlot(0); 3586 while (!str.empty() && str.front() == '_') str = str.substr(1); 3587 if (str.startswith("set") || str.startswith("add")) 3588 str = str.substr(3); 3589 else 3590 return false; 3591 3592 if (str.empty()) return true; 3593 return !islower(str.front()); 3594} 3595 3596/// Check a message send to see if it's likely to cause a retain cycle. 3597void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 3598 // Only check instance methods whose selector looks like a setter. 3599 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 3600 return; 3601 3602 // Try to find a variable that the receiver is strongly owned by. 3603 RetainCycleOwner owner; 3604 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 3605 if (!findRetainCycleOwner(msg->getInstanceReceiver(), owner)) 3606 return; 3607 } else { 3608 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 3609 owner.Variable = getCurMethodDecl()->getSelfDecl(); 3610 owner.Loc = msg->getSuperLoc(); 3611 owner.Range = msg->getSuperLoc(); 3612 } 3613 3614 // Check whether the receiver is captured by any of the arguments. 3615 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 3616 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 3617 return diagnoseRetainCycle(*this, capturer, owner); 3618} 3619 3620/// Check a property assign to see if it's likely to cause a retain cycle. 3621void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 3622 RetainCycleOwner owner; 3623 if (!findRetainCycleOwner(receiver, owner)) 3624 return; 3625 3626 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 3627 diagnoseRetainCycle(*this, capturer, owner); 3628} 3629 3630void Sema::checkUnsafeAssigns(SourceLocation Loc, 3631 QualType LHS, Expr *RHS) { 3632 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 3633 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 3634 return; 3635 if (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) 3636 if (cast->getCastKind() == CK_ObjCConsumeObject) 3637 Diag(Loc, diag::warn_arc_retained_assign) 3638 << (LT == Qualifiers::OCL_ExplicitNone) 3639 << RHS->getSourceRange(); 3640} 3641 3642