SemaChecking.cpp revision 5e1cdac63c3d9c9b32fa41fa0b2d242a58a20d49
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/Initialization.h" 16#include "clang/Sema/Sema.h" 17#include "clang/Sema/SemaInternal.h" 18#include "clang/Sema/ScopeInfo.h" 19#include "clang/Analysis/Analyses/FormatString.h" 20#include "clang/AST/ASTContext.h" 21#include "clang/AST/CharUnits.h" 22#include "clang/AST/DeclCXX.h" 23#include "clang/AST/DeclObjC.h" 24#include "clang/AST/ExprCXX.h" 25#include "clang/AST/ExprObjC.h" 26#include "clang/AST/EvaluatedExprVisitor.h" 27#include "clang/AST/DeclObjC.h" 28#include "clang/AST/StmtCXX.h" 29#include "clang/AST/StmtObjC.h" 30#include "clang/Lex/Preprocessor.h" 31#include "llvm/ADT/BitVector.h" 32#include "llvm/ADT/STLExtras.h" 33#include "llvm/Support/raw_ostream.h" 34#include "clang/Basic/TargetBuiltins.h" 35#include "clang/Basic/TargetInfo.h" 36#include "clang/Basic/ConvertUTF.h" 37#include <limits> 38using namespace clang; 39using namespace sema; 40 41SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 42 unsigned ByteNo) const { 43 return SL->getLocationOfByte(ByteNo, PP.getSourceManager(), 44 PP.getLangOptions(), PP.getTargetInfo()); 45} 46 47 48/// CheckablePrintfAttr - does a function call have a "printf" attribute 49/// and arguments that merit checking? 50bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) { 51 if (Format->getType() == "printf") return true; 52 if (Format->getType() == "printf0") { 53 // printf0 allows null "format" string; if so don't check format/args 54 unsigned format_idx = Format->getFormatIdx() - 1; 55 // Does the index refer to the implicit object argument? 56 if (isa<CXXMemberCallExpr>(TheCall)) { 57 if (format_idx == 0) 58 return false; 59 --format_idx; 60 } 61 if (format_idx < TheCall->getNumArgs()) { 62 Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts(); 63 if (!Format->isNullPointerConstant(Context, 64 Expr::NPC_ValueDependentIsNull)) 65 return true; 66 } 67 } 68 return false; 69} 70 71/// Checks that a call expression's argument count is the desired number. 72/// This is useful when doing custom type-checking. Returns true on error. 73static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) { 74 unsigned argCount = call->getNumArgs(); 75 if (argCount == desiredArgCount) return false; 76 77 if (argCount < desiredArgCount) 78 return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args) 79 << 0 /*function call*/ << desiredArgCount << argCount 80 << call->getSourceRange(); 81 82 // Highlight all the excess arguments. 83 SourceRange range(call->getArg(desiredArgCount)->getLocStart(), 84 call->getArg(argCount - 1)->getLocEnd()); 85 86 return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args) 87 << 0 /*function call*/ << desiredArgCount << argCount 88 << call->getArg(1)->getSourceRange(); 89} 90 91/// CheckBuiltinAnnotationString - Checks that string argument to the builtin 92/// annotation is a non wide string literal. 93static bool CheckBuiltinAnnotationString(Sema &S, Expr *Arg) { 94 Arg = Arg->IgnoreParenCasts(); 95 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 96 if (!Literal || !Literal->isAscii()) { 97 S.Diag(Arg->getLocStart(), diag::err_builtin_annotation_not_string_constant) 98 << Arg->getSourceRange(); 99 return true; 100 } 101 return false; 102} 103 104ExprResult 105Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 106 ExprResult TheCallResult(Owned(TheCall)); 107 108 // Find out if any arguments are required to be integer constant expressions. 109 unsigned ICEArguments = 0; 110 ASTContext::GetBuiltinTypeError Error; 111 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments); 112 if (Error != ASTContext::GE_None) 113 ICEArguments = 0; // Don't diagnose previously diagnosed errors. 114 115 // If any arguments are required to be ICE's, check and diagnose. 116 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) { 117 // Skip arguments not required to be ICE's. 118 if ((ICEArguments & (1 << ArgNo)) == 0) continue; 119 120 llvm::APSInt Result; 121 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 122 return true; 123 ICEArguments &= ~(1 << ArgNo); 124 } 125 126 switch (BuiltinID) { 127 case Builtin::BI__builtin___CFStringMakeConstantString: 128 assert(TheCall->getNumArgs() == 1 && 129 "Wrong # arguments to builtin CFStringMakeConstantString"); 130 if (CheckObjCString(TheCall->getArg(0))) 131 return ExprError(); 132 break; 133 case Builtin::BI__builtin_stdarg_start: 134 case Builtin::BI__builtin_va_start: 135 if (SemaBuiltinVAStart(TheCall)) 136 return ExprError(); 137 break; 138 case Builtin::BI__builtin_isgreater: 139 case Builtin::BI__builtin_isgreaterequal: 140 case Builtin::BI__builtin_isless: 141 case Builtin::BI__builtin_islessequal: 142 case Builtin::BI__builtin_islessgreater: 143 case Builtin::BI__builtin_isunordered: 144 if (SemaBuiltinUnorderedCompare(TheCall)) 145 return ExprError(); 146 break; 147 case Builtin::BI__builtin_fpclassify: 148 if (SemaBuiltinFPClassification(TheCall, 6)) 149 return ExprError(); 150 break; 151 case Builtin::BI__builtin_isfinite: 152 case Builtin::BI__builtin_isinf: 153 case Builtin::BI__builtin_isinf_sign: 154 case Builtin::BI__builtin_isnan: 155 case Builtin::BI__builtin_isnormal: 156 if (SemaBuiltinFPClassification(TheCall, 1)) 157 return ExprError(); 158 break; 159 case Builtin::BI__builtin_shufflevector: 160 return SemaBuiltinShuffleVector(TheCall); 161 // TheCall will be freed by the smart pointer here, but that's fine, since 162 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 163 case Builtin::BI__builtin_prefetch: 164 if (SemaBuiltinPrefetch(TheCall)) 165 return ExprError(); 166 break; 167 case Builtin::BI__builtin_object_size: 168 if (SemaBuiltinObjectSize(TheCall)) 169 return ExprError(); 170 break; 171 case Builtin::BI__builtin_longjmp: 172 if (SemaBuiltinLongjmp(TheCall)) 173 return ExprError(); 174 break; 175 176 case Builtin::BI__builtin_classify_type: 177 if (checkArgCount(*this, TheCall, 1)) return true; 178 TheCall->setType(Context.IntTy); 179 break; 180 case Builtin::BI__builtin_constant_p: 181 if (checkArgCount(*this, TheCall, 1)) return true; 182 TheCall->setType(Context.IntTy); 183 break; 184 case Builtin::BI__sync_fetch_and_add: 185 case Builtin::BI__sync_fetch_and_sub: 186 case Builtin::BI__sync_fetch_and_or: 187 case Builtin::BI__sync_fetch_and_and: 188 case Builtin::BI__sync_fetch_and_xor: 189 case Builtin::BI__sync_add_and_fetch: 190 case Builtin::BI__sync_sub_and_fetch: 191 case Builtin::BI__sync_and_and_fetch: 192 case Builtin::BI__sync_or_and_fetch: 193 case Builtin::BI__sync_xor_and_fetch: 194 case Builtin::BI__sync_val_compare_and_swap: 195 case Builtin::BI__sync_bool_compare_and_swap: 196 case Builtin::BI__sync_lock_test_and_set: 197 case Builtin::BI__sync_lock_release: 198 case Builtin::BI__sync_swap: 199 return SemaBuiltinAtomicOverloaded(move(TheCallResult)); 200 case Builtin::BI__builtin_annotation: 201 if (CheckBuiltinAnnotationString(*this, TheCall->getArg(1))) 202 return ExprError(); 203 break; 204 } 205 206 // Since the target specific builtins for each arch overlap, only check those 207 // of the arch we are compiling for. 208 if (BuiltinID >= Builtin::FirstTSBuiltin) { 209 switch (Context.getTargetInfo().getTriple().getArch()) { 210 case llvm::Triple::arm: 211 case llvm::Triple::thumb: 212 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 213 return ExprError(); 214 break; 215 default: 216 break; 217 } 218 } 219 220 return move(TheCallResult); 221} 222 223// Get the valid immediate range for the specified NEON type code. 224static unsigned RFT(unsigned t, bool shift = false) { 225 bool quad = t & 0x10; 226 227 switch (t & 0x7) { 228 case 0: // i8 229 return shift ? 7 : (8 << (int)quad) - 1; 230 case 1: // i16 231 return shift ? 15 : (4 << (int)quad) - 1; 232 case 2: // i32 233 return shift ? 31 : (2 << (int)quad) - 1; 234 case 3: // i64 235 return shift ? 63 : (1 << (int)quad) - 1; 236 case 4: // f32 237 assert(!shift && "cannot shift float types!"); 238 return (2 << (int)quad) - 1; 239 case 5: // poly8 240 return shift ? 7 : (8 << (int)quad) - 1; 241 case 6: // poly16 242 return shift ? 15 : (4 << (int)quad) - 1; 243 case 7: // float16 244 assert(!shift && "cannot shift float types!"); 245 return (4 << (int)quad) - 1; 246 } 247 return 0; 248} 249 250bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 251 llvm::APSInt Result; 252 253 unsigned mask = 0; 254 unsigned TV = 0; 255 switch (BuiltinID) { 256#define GET_NEON_OVERLOAD_CHECK 257#include "clang/Basic/arm_neon.inc" 258#undef GET_NEON_OVERLOAD_CHECK 259 } 260 261 // For NEON intrinsics which are overloaded on vector element type, validate 262 // the immediate which specifies which variant to emit. 263 if (mask) { 264 unsigned ArgNo = TheCall->getNumArgs()-1; 265 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 266 return true; 267 268 TV = Result.getLimitedValue(32); 269 if ((TV > 31) || (mask & (1 << TV)) == 0) 270 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 271 << TheCall->getArg(ArgNo)->getSourceRange(); 272 } 273 274 // For NEON intrinsics which take an immediate value as part of the 275 // instruction, range check them here. 276 unsigned i = 0, l = 0, u = 0; 277 switch (BuiltinID) { 278 default: return false; 279 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 280 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 281 case ARM::BI__builtin_arm_vcvtr_f: 282 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 283#define GET_NEON_IMMEDIATE_CHECK 284#include "clang/Basic/arm_neon.inc" 285#undef GET_NEON_IMMEDIATE_CHECK 286 }; 287 288 // Check that the immediate argument is actually a constant. 289 if (SemaBuiltinConstantArg(TheCall, i, Result)) 290 return true; 291 292 // Range check against the upper/lower values for this isntruction. 293 unsigned Val = Result.getZExtValue(); 294 if (Val < l || Val > (u + l)) 295 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 296 << l << u+l << TheCall->getArg(i)->getSourceRange(); 297 298 // FIXME: VFP Intrinsics should error if VFP not present. 299 return false; 300} 301 302/// CheckFunctionCall - Check a direct function call for various correctness 303/// and safety properties not strictly enforced by the C type system. 304bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 305 // Get the IdentifierInfo* for the called function. 306 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 307 308 // None of the checks below are needed for functions that don't have 309 // simple names (e.g., C++ conversion functions). 310 if (!FnInfo) 311 return false; 312 313 // FIXME: This mechanism should be abstracted to be less fragile and 314 // more efficient. For example, just map function ids to custom 315 // handlers. 316 317 // Printf and scanf checking. 318 for (specific_attr_iterator<FormatAttr> 319 i = FDecl->specific_attr_begin<FormatAttr>(), 320 e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { 321 322 const FormatAttr *Format = *i; 323 const bool b = Format->getType() == "scanf"; 324 if (b || CheckablePrintfAttr(Format, TheCall)) { 325 bool HasVAListArg = Format->getFirstArg() == 0; 326 CheckPrintfScanfArguments(TheCall, HasVAListArg, 327 Format->getFormatIdx() - 1, 328 HasVAListArg ? 0 : Format->getFirstArg() - 1, 329 !b); 330 } 331 } 332 333 for (specific_attr_iterator<NonNullAttr> 334 i = FDecl->specific_attr_begin<NonNullAttr>(), 335 e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) { 336 CheckNonNullArguments(*i, TheCall->getArgs(), 337 TheCall->getCallee()->getLocStart()); 338 } 339 340 // Builtin handling 341 int CMF = -1; 342 switch (FDecl->getBuiltinID()) { 343 case Builtin::BI__builtin_memset: 344 case Builtin::BI__builtin___memset_chk: 345 case Builtin::BImemset: 346 CMF = CMF_Memset; 347 break; 348 349 case Builtin::BI__builtin_memcpy: 350 case Builtin::BI__builtin___memcpy_chk: 351 case Builtin::BImemcpy: 352 CMF = CMF_Memcpy; 353 break; 354 355 case Builtin::BI__builtin_memmove: 356 case Builtin::BI__builtin___memmove_chk: 357 case Builtin::BImemmove: 358 CMF = CMF_Memmove; 359 break; 360 361 case Builtin::BIstrlcpy: 362 case Builtin::BIstrlcat: 363 CheckStrlcpycatArguments(TheCall, FnInfo); 364 break; 365 366 case Builtin::BI__builtin_memcmp: 367 CMF = CMF_Memcmp; 368 break; 369 370 default: 371 if (FDecl->getLinkage() == ExternalLinkage && 372 (!getLangOptions().CPlusPlus || FDecl->isExternC())) { 373 if (FnInfo->isStr("memset")) 374 CMF = CMF_Memset; 375 else if (FnInfo->isStr("memcpy")) 376 CMF = CMF_Memcpy; 377 else if (FnInfo->isStr("memmove")) 378 CMF = CMF_Memmove; 379 else if (FnInfo->isStr("memcmp")) 380 CMF = CMF_Memcmp; 381 } 382 break; 383 } 384 385 // Memset/memcpy/memmove handling 386 if (CMF != -1) 387 CheckMemaccessArguments(TheCall, CheckedMemoryFunction(CMF), FnInfo); 388 389 return false; 390} 391 392bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 393 // Printf checking. 394 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 395 if (!Format) 396 return false; 397 398 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 399 if (!V) 400 return false; 401 402 QualType Ty = V->getType(); 403 if (!Ty->isBlockPointerType()) 404 return false; 405 406 const bool b = Format->getType() == "scanf"; 407 if (!b && !CheckablePrintfAttr(Format, TheCall)) 408 return false; 409 410 bool HasVAListArg = Format->getFirstArg() == 0; 411 CheckPrintfScanfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 412 HasVAListArg ? 0 : Format->getFirstArg() - 1, !b); 413 414 return false; 415} 416 417/// checkBuiltinArgument - Given a call to a builtin function, perform 418/// normal type-checking on the given argument, updating the call in 419/// place. This is useful when a builtin function requires custom 420/// type-checking for some of its arguments but not necessarily all of 421/// them. 422/// 423/// Returns true on error. 424static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 425 FunctionDecl *Fn = E->getDirectCallee(); 426 assert(Fn && "builtin call without direct callee!"); 427 428 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 429 InitializedEntity Entity = 430 InitializedEntity::InitializeParameter(S.Context, Param); 431 432 ExprResult Arg = E->getArg(0); 433 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 434 if (Arg.isInvalid()) 435 return true; 436 437 E->setArg(ArgIndex, Arg.take()); 438 return false; 439} 440 441/// SemaBuiltinAtomicOverloaded - We have a call to a function like 442/// __sync_fetch_and_add, which is an overloaded function based on the pointer 443/// type of its first argument. The main ActOnCallExpr routines have already 444/// promoted the types of arguments because all of these calls are prototyped as 445/// void(...). 446/// 447/// This function goes through and does final semantic checking for these 448/// builtins, 449ExprResult 450Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 451 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 452 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 453 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 454 455 // Ensure that we have at least one argument to do type inference from. 456 if (TheCall->getNumArgs() < 1) { 457 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 458 << 0 << 1 << TheCall->getNumArgs() 459 << TheCall->getCallee()->getSourceRange(); 460 return ExprError(); 461 } 462 463 // Inspect the first argument of the atomic builtin. This should always be 464 // a pointer type, whose element is an integral scalar or pointer type. 465 // Because it is a pointer type, we don't have to worry about any implicit 466 // casts here. 467 // FIXME: We don't allow floating point scalars as input. 468 Expr *FirstArg = TheCall->getArg(0); 469 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 470 if (!pointerType) { 471 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 472 << FirstArg->getType() << FirstArg->getSourceRange(); 473 return ExprError(); 474 } 475 476 QualType ValType = pointerType->getPointeeType(); 477 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 478 !ValType->isBlockPointerType()) { 479 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 480 << FirstArg->getType() << FirstArg->getSourceRange(); 481 return ExprError(); 482 } 483 484 switch (ValType.getObjCLifetime()) { 485 case Qualifiers::OCL_None: 486 case Qualifiers::OCL_ExplicitNone: 487 // okay 488 break; 489 490 case Qualifiers::OCL_Weak: 491 case Qualifiers::OCL_Strong: 492 case Qualifiers::OCL_Autoreleasing: 493 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 494 << ValType << FirstArg->getSourceRange(); 495 return ExprError(); 496 } 497 498 // Strip any qualifiers off ValType. 499 ValType = ValType.getUnqualifiedType(); 500 501 // The majority of builtins return a value, but a few have special return 502 // types, so allow them to override appropriately below. 503 QualType ResultType = ValType; 504 505 // We need to figure out which concrete builtin this maps onto. For example, 506 // __sync_fetch_and_add with a 2 byte object turns into 507 // __sync_fetch_and_add_2. 508#define BUILTIN_ROW(x) \ 509 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 510 Builtin::BI##x##_8, Builtin::BI##x##_16 } 511 512 static const unsigned BuiltinIndices[][5] = { 513 BUILTIN_ROW(__sync_fetch_and_add), 514 BUILTIN_ROW(__sync_fetch_and_sub), 515 BUILTIN_ROW(__sync_fetch_and_or), 516 BUILTIN_ROW(__sync_fetch_and_and), 517 BUILTIN_ROW(__sync_fetch_and_xor), 518 519 BUILTIN_ROW(__sync_add_and_fetch), 520 BUILTIN_ROW(__sync_sub_and_fetch), 521 BUILTIN_ROW(__sync_and_and_fetch), 522 BUILTIN_ROW(__sync_or_and_fetch), 523 BUILTIN_ROW(__sync_xor_and_fetch), 524 525 BUILTIN_ROW(__sync_val_compare_and_swap), 526 BUILTIN_ROW(__sync_bool_compare_and_swap), 527 BUILTIN_ROW(__sync_lock_test_and_set), 528 BUILTIN_ROW(__sync_lock_release), 529 BUILTIN_ROW(__sync_swap) 530 }; 531#undef BUILTIN_ROW 532 533 // Determine the index of the size. 534 unsigned SizeIndex; 535 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 536 case 1: SizeIndex = 0; break; 537 case 2: SizeIndex = 1; break; 538 case 4: SizeIndex = 2; break; 539 case 8: SizeIndex = 3; break; 540 case 16: SizeIndex = 4; break; 541 default: 542 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 543 << FirstArg->getType() << FirstArg->getSourceRange(); 544 return ExprError(); 545 } 546 547 // Each of these builtins has one pointer argument, followed by some number of 548 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 549 // that we ignore. Find out which row of BuiltinIndices to read from as well 550 // as the number of fixed args. 551 unsigned BuiltinID = FDecl->getBuiltinID(); 552 unsigned BuiltinIndex, NumFixed = 1; 553 switch (BuiltinID) { 554 default: llvm_unreachable("Unknown overloaded atomic builtin!"); 555 case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break; 556 case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break; 557 case Builtin::BI__sync_fetch_and_or: BuiltinIndex = 2; break; 558 case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break; 559 case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break; 560 561 case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break; 562 case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break; 563 case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break; 564 case Builtin::BI__sync_or_and_fetch: BuiltinIndex = 8; break; 565 case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break; 566 567 case Builtin::BI__sync_val_compare_and_swap: 568 BuiltinIndex = 10; 569 NumFixed = 2; 570 break; 571 case Builtin::BI__sync_bool_compare_and_swap: 572 BuiltinIndex = 11; 573 NumFixed = 2; 574 ResultType = Context.BoolTy; 575 break; 576 case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break; 577 case Builtin::BI__sync_lock_release: 578 BuiltinIndex = 13; 579 NumFixed = 0; 580 ResultType = Context.VoidTy; 581 break; 582 case Builtin::BI__sync_swap: BuiltinIndex = 14; break; 583 } 584 585 // Now that we know how many fixed arguments we expect, first check that we 586 // have at least that many. 587 if (TheCall->getNumArgs() < 1+NumFixed) { 588 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 589 << 0 << 1+NumFixed << TheCall->getNumArgs() 590 << TheCall->getCallee()->getSourceRange(); 591 return ExprError(); 592 } 593 594 // Get the decl for the concrete builtin from this, we can tell what the 595 // concrete integer type we should convert to is. 596 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 597 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 598 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 599 FunctionDecl *NewBuiltinDecl = 600 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 601 TUScope, false, DRE->getLocStart())); 602 603 // The first argument --- the pointer --- has a fixed type; we 604 // deduce the types of the rest of the arguments accordingly. Walk 605 // the remaining arguments, converting them to the deduced value type. 606 for (unsigned i = 0; i != NumFixed; ++i) { 607 ExprResult Arg = TheCall->getArg(i+1); 608 609 // If the argument is an implicit cast, then there was a promotion due to 610 // "...", just remove it now. 611 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg.get())) { 612 Arg = ICE->getSubExpr(); 613 ICE->setSubExpr(0); 614 TheCall->setArg(i+1, Arg.get()); 615 } 616 617 // GCC does an implicit conversion to the pointer or integer ValType. This 618 // can fail in some cases (1i -> int**), check for this error case now. 619 // Initialize the argument. 620 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context, 621 ValType, /*consume*/ false); 622 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 623 if (Arg.isInvalid()) 624 return ExprError(); 625 626 // Okay, we have something that *can* be converted to the right type. Check 627 // to see if there is a potentially weird extension going on here. This can 628 // happen when you do an atomic operation on something like an char* and 629 // pass in 42. The 42 gets converted to char. This is even more strange 630 // for things like 45.123 -> char, etc. 631 // FIXME: Do this check. 632 TheCall->setArg(i+1, Arg.take()); 633 } 634 635 ASTContext& Context = this->getASTContext(); 636 637 // Create a new DeclRefExpr to refer to the new decl. 638 DeclRefExpr* NewDRE = DeclRefExpr::Create( 639 Context, 640 DRE->getQualifierLoc(), 641 NewBuiltinDecl, 642 DRE->getLocation(), 643 NewBuiltinDecl->getType(), 644 DRE->getValueKind()); 645 646 // Set the callee in the CallExpr. 647 // FIXME: This leaks the original parens and implicit casts. 648 ExprResult PromotedCall = UsualUnaryConversions(NewDRE); 649 if (PromotedCall.isInvalid()) 650 return ExprError(); 651 TheCall->setCallee(PromotedCall.take()); 652 653 // Change the result type of the call to match the original value type. This 654 // is arbitrary, but the codegen for these builtins ins design to handle it 655 // gracefully. 656 TheCall->setType(ResultType); 657 658 return move(TheCallResult); 659} 660 661/// CheckObjCString - Checks that the argument to the builtin 662/// CFString constructor is correct 663/// Note: It might also make sense to do the UTF-16 conversion here (would 664/// simplify the backend). 665bool Sema::CheckObjCString(Expr *Arg) { 666 Arg = Arg->IgnoreParenCasts(); 667 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 668 669 if (!Literal || !Literal->isAscii()) { 670 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 671 << Arg->getSourceRange(); 672 return true; 673 } 674 675 if (Literal->containsNonAsciiOrNull()) { 676 StringRef String = Literal->getString(); 677 unsigned NumBytes = String.size(); 678 SmallVector<UTF16, 128> ToBuf(NumBytes); 679 const UTF8 *FromPtr = (UTF8 *)String.data(); 680 UTF16 *ToPtr = &ToBuf[0]; 681 682 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 683 &ToPtr, ToPtr + NumBytes, 684 strictConversion); 685 // Check for conversion failure. 686 if (Result != conversionOK) 687 Diag(Arg->getLocStart(), 688 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 689 } 690 return false; 691} 692 693/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 694/// Emit an error and return true on failure, return false on success. 695bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 696 Expr *Fn = TheCall->getCallee(); 697 if (TheCall->getNumArgs() > 2) { 698 Diag(TheCall->getArg(2)->getLocStart(), 699 diag::err_typecheck_call_too_many_args) 700 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 701 << Fn->getSourceRange() 702 << SourceRange(TheCall->getArg(2)->getLocStart(), 703 (*(TheCall->arg_end()-1))->getLocEnd()); 704 return true; 705 } 706 707 if (TheCall->getNumArgs() < 2) { 708 return Diag(TheCall->getLocEnd(), 709 diag::err_typecheck_call_too_few_args_at_least) 710 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 711 } 712 713 // Type-check the first argument normally. 714 if (checkBuiltinArgument(*this, TheCall, 0)) 715 return true; 716 717 // Determine whether the current function is variadic or not. 718 BlockScopeInfo *CurBlock = getCurBlock(); 719 bool isVariadic; 720 if (CurBlock) 721 isVariadic = CurBlock->TheDecl->isVariadic(); 722 else if (FunctionDecl *FD = getCurFunctionDecl()) 723 isVariadic = FD->isVariadic(); 724 else 725 isVariadic = getCurMethodDecl()->isVariadic(); 726 727 if (!isVariadic) { 728 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 729 return true; 730 } 731 732 // Verify that the second argument to the builtin is the last argument of the 733 // current function or method. 734 bool SecondArgIsLastNamedArgument = false; 735 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 736 737 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 738 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 739 // FIXME: This isn't correct for methods (results in bogus warning). 740 // Get the last formal in the current function. 741 const ParmVarDecl *LastArg; 742 if (CurBlock) 743 LastArg = *(CurBlock->TheDecl->param_end()-1); 744 else if (FunctionDecl *FD = getCurFunctionDecl()) 745 LastArg = *(FD->param_end()-1); 746 else 747 LastArg = *(getCurMethodDecl()->param_end()-1); 748 SecondArgIsLastNamedArgument = PV == LastArg; 749 } 750 } 751 752 if (!SecondArgIsLastNamedArgument) 753 Diag(TheCall->getArg(1)->getLocStart(), 754 diag::warn_second_parameter_of_va_start_not_last_named_argument); 755 return false; 756} 757 758/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 759/// friends. This is declared to take (...), so we have to check everything. 760bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 761 if (TheCall->getNumArgs() < 2) 762 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 763 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 764 if (TheCall->getNumArgs() > 2) 765 return Diag(TheCall->getArg(2)->getLocStart(), 766 diag::err_typecheck_call_too_many_args) 767 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 768 << SourceRange(TheCall->getArg(2)->getLocStart(), 769 (*(TheCall->arg_end()-1))->getLocEnd()); 770 771 ExprResult OrigArg0 = TheCall->getArg(0); 772 ExprResult OrigArg1 = TheCall->getArg(1); 773 774 // Do standard promotions between the two arguments, returning their common 775 // type. 776 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 777 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 778 return true; 779 780 // Make sure any conversions are pushed back into the call; this is 781 // type safe since unordered compare builtins are declared as "_Bool 782 // foo(...)". 783 TheCall->setArg(0, OrigArg0.get()); 784 TheCall->setArg(1, OrigArg1.get()); 785 786 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 787 return false; 788 789 // If the common type isn't a real floating type, then the arguments were 790 // invalid for this operation. 791 if (!Res->isRealFloatingType()) 792 return Diag(OrigArg0.get()->getLocStart(), 793 diag::err_typecheck_call_invalid_ordered_compare) 794 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 795 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 796 797 return false; 798} 799 800/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 801/// __builtin_isnan and friends. This is declared to take (...), so we have 802/// to check everything. We expect the last argument to be a floating point 803/// value. 804bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 805 if (TheCall->getNumArgs() < NumArgs) 806 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 807 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 808 if (TheCall->getNumArgs() > NumArgs) 809 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 810 diag::err_typecheck_call_too_many_args) 811 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 812 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 813 (*(TheCall->arg_end()-1))->getLocEnd()); 814 815 Expr *OrigArg = TheCall->getArg(NumArgs-1); 816 817 if (OrigArg->isTypeDependent()) 818 return false; 819 820 // This operation requires a non-_Complex floating-point number. 821 if (!OrigArg->getType()->isRealFloatingType()) 822 return Diag(OrigArg->getLocStart(), 823 diag::err_typecheck_call_invalid_unary_fp) 824 << OrigArg->getType() << OrigArg->getSourceRange(); 825 826 // If this is an implicit conversion from float -> double, remove it. 827 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 828 Expr *CastArg = Cast->getSubExpr(); 829 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 830 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 831 "promotion from float to double is the only expected cast here"); 832 Cast->setSubExpr(0); 833 TheCall->setArg(NumArgs-1, CastArg); 834 OrigArg = CastArg; 835 } 836 } 837 838 return false; 839} 840 841/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 842// This is declared to take (...), so we have to check everything. 843ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 844 if (TheCall->getNumArgs() < 2) 845 return ExprError(Diag(TheCall->getLocEnd(), 846 diag::err_typecheck_call_too_few_args_at_least) 847 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 848 << TheCall->getSourceRange()); 849 850 // Determine which of the following types of shufflevector we're checking: 851 // 1) unary, vector mask: (lhs, mask) 852 // 2) binary, vector mask: (lhs, rhs, mask) 853 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 854 QualType resType = TheCall->getArg(0)->getType(); 855 unsigned numElements = 0; 856 857 if (!TheCall->getArg(0)->isTypeDependent() && 858 !TheCall->getArg(1)->isTypeDependent()) { 859 QualType LHSType = TheCall->getArg(0)->getType(); 860 QualType RHSType = TheCall->getArg(1)->getType(); 861 862 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 863 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 864 << SourceRange(TheCall->getArg(0)->getLocStart(), 865 TheCall->getArg(1)->getLocEnd()); 866 return ExprError(); 867 } 868 869 numElements = LHSType->getAs<VectorType>()->getNumElements(); 870 unsigned numResElements = TheCall->getNumArgs() - 2; 871 872 // Check to see if we have a call with 2 vector arguments, the unary shuffle 873 // with mask. If so, verify that RHS is an integer vector type with the 874 // same number of elts as lhs. 875 if (TheCall->getNumArgs() == 2) { 876 if (!RHSType->hasIntegerRepresentation() || 877 RHSType->getAs<VectorType>()->getNumElements() != numElements) 878 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 879 << SourceRange(TheCall->getArg(1)->getLocStart(), 880 TheCall->getArg(1)->getLocEnd()); 881 numResElements = numElements; 882 } 883 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 884 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 885 << SourceRange(TheCall->getArg(0)->getLocStart(), 886 TheCall->getArg(1)->getLocEnd()); 887 return ExprError(); 888 } else if (numElements != numResElements) { 889 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 890 resType = Context.getVectorType(eltType, numResElements, 891 VectorType::GenericVector); 892 } 893 } 894 895 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 896 if (TheCall->getArg(i)->isTypeDependent() || 897 TheCall->getArg(i)->isValueDependent()) 898 continue; 899 900 llvm::APSInt Result(32); 901 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 902 return ExprError(Diag(TheCall->getLocStart(), 903 diag::err_shufflevector_nonconstant_argument) 904 << TheCall->getArg(i)->getSourceRange()); 905 906 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 907 return ExprError(Diag(TheCall->getLocStart(), 908 diag::err_shufflevector_argument_too_large) 909 << TheCall->getArg(i)->getSourceRange()); 910 } 911 912 SmallVector<Expr*, 32> exprs; 913 914 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 915 exprs.push_back(TheCall->getArg(i)); 916 TheCall->setArg(i, 0); 917 } 918 919 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 920 exprs.size(), resType, 921 TheCall->getCallee()->getLocStart(), 922 TheCall->getRParenLoc())); 923} 924 925/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 926// This is declared to take (const void*, ...) and can take two 927// optional constant int args. 928bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 929 unsigned NumArgs = TheCall->getNumArgs(); 930 931 if (NumArgs > 3) 932 return Diag(TheCall->getLocEnd(), 933 diag::err_typecheck_call_too_many_args_at_most) 934 << 0 /*function call*/ << 3 << NumArgs 935 << TheCall->getSourceRange(); 936 937 // Argument 0 is checked for us and the remaining arguments must be 938 // constant integers. 939 for (unsigned i = 1; i != NumArgs; ++i) { 940 Expr *Arg = TheCall->getArg(i); 941 942 llvm::APSInt Result; 943 if (SemaBuiltinConstantArg(TheCall, i, Result)) 944 return true; 945 946 // FIXME: gcc issues a warning and rewrites these to 0. These 947 // seems especially odd for the third argument since the default 948 // is 3. 949 if (i == 1) { 950 if (Result.getLimitedValue() > 1) 951 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 952 << "0" << "1" << Arg->getSourceRange(); 953 } else { 954 if (Result.getLimitedValue() > 3) 955 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 956 << "0" << "3" << Arg->getSourceRange(); 957 } 958 } 959 960 return false; 961} 962 963/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 964/// TheCall is a constant expression. 965bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 966 llvm::APSInt &Result) { 967 Expr *Arg = TheCall->getArg(ArgNum); 968 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 969 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 970 971 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 972 973 if (!Arg->isIntegerConstantExpr(Result, Context)) 974 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 975 << FDecl->getDeclName() << Arg->getSourceRange(); 976 977 return false; 978} 979 980/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 981/// int type). This simply type checks that type is one of the defined 982/// constants (0-3). 983// For compatibility check 0-3, llvm only handles 0 and 2. 984bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 985 llvm::APSInt Result; 986 987 // Check constant-ness first. 988 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 989 return true; 990 991 Expr *Arg = TheCall->getArg(1); 992 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 993 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 994 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 995 } 996 997 return false; 998} 999 1000/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1001/// This checks that val is a constant 1. 1002bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1003 Expr *Arg = TheCall->getArg(1); 1004 llvm::APSInt Result; 1005 1006 // TODO: This is less than ideal. Overload this to take a value. 1007 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1008 return true; 1009 1010 if (Result != 1) 1011 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1012 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1013 1014 return false; 1015} 1016 1017// Handle i > 1 ? "x" : "y", recursively. 1018bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 1019 bool HasVAListArg, 1020 unsigned format_idx, unsigned firstDataArg, 1021 bool isPrintf) { 1022 tryAgain: 1023 if (E->isTypeDependent() || E->isValueDependent()) 1024 return false; 1025 1026 E = E->IgnoreParens(); 1027 1028 switch (E->getStmtClass()) { 1029 case Stmt::BinaryConditionalOperatorClass: 1030 case Stmt::ConditionalOperatorClass: { 1031 const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E); 1032 return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, HasVAListArg, 1033 format_idx, firstDataArg, isPrintf) 1034 && SemaCheckStringLiteral(C->getFalseExpr(), TheCall, HasVAListArg, 1035 format_idx, firstDataArg, isPrintf); 1036 } 1037 1038 case Stmt::IntegerLiteralClass: 1039 // Technically -Wformat-nonliteral does not warn about this case. 1040 // The behavior of printf and friends in this case is implementation 1041 // dependent. Ideally if the format string cannot be null then 1042 // it should have a 'nonnull' attribute in the function prototype. 1043 return true; 1044 1045 case Stmt::ImplicitCastExprClass: { 1046 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 1047 goto tryAgain; 1048 } 1049 1050 case Stmt::OpaqueValueExprClass: 1051 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 1052 E = src; 1053 goto tryAgain; 1054 } 1055 return false; 1056 1057 case Stmt::PredefinedExprClass: 1058 // While __func__, etc., are technically not string literals, they 1059 // cannot contain format specifiers and thus are not a security 1060 // liability. 1061 return true; 1062 1063 case Stmt::DeclRefExprClass: { 1064 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1065 1066 // As an exception, do not flag errors for variables binding to 1067 // const string literals. 1068 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1069 bool isConstant = false; 1070 QualType T = DR->getType(); 1071 1072 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1073 isConstant = AT->getElementType().isConstant(Context); 1074 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1075 isConstant = T.isConstant(Context) && 1076 PT->getPointeeType().isConstant(Context); 1077 } 1078 1079 if (isConstant) { 1080 if (const Expr *Init = VD->getAnyInitializer()) 1081 return SemaCheckStringLiteral(Init, TheCall, 1082 HasVAListArg, format_idx, firstDataArg, 1083 isPrintf); 1084 } 1085 1086 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1087 // special check to see if the format string is a function parameter 1088 // of the function calling the printf function. If the function 1089 // has an attribute indicating it is a printf-like function, then we 1090 // should suppress warnings concerning non-literals being used in a call 1091 // to a vprintf function. For example: 1092 // 1093 // void 1094 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1095 // va_list ap; 1096 // va_start(ap, fmt); 1097 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1098 // ... 1099 // 1100 // 1101 // FIXME: We don't have full attribute support yet, so just check to see 1102 // if the argument is a DeclRefExpr that references a parameter. We'll 1103 // add proper support for checking the attribute later. 1104 if (HasVAListArg) 1105 if (isa<ParmVarDecl>(VD)) 1106 return true; 1107 } 1108 1109 return false; 1110 } 1111 1112 case Stmt::CallExprClass: { 1113 const CallExpr *CE = cast<CallExpr>(E); 1114 if (const ImplicitCastExpr *ICE 1115 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 1116 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 1117 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 1118 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 1119 unsigned ArgIndex = FA->getFormatIdx(); 1120 const Expr *Arg = CE->getArg(ArgIndex - 1); 1121 1122 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 1123 format_idx, firstDataArg, isPrintf); 1124 } 1125 } 1126 } 1127 } 1128 1129 return false; 1130 } 1131 case Stmt::ObjCStringLiteralClass: 1132 case Stmt::StringLiteralClass: { 1133 const StringLiteral *StrE = NULL; 1134 1135 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1136 StrE = ObjCFExpr->getString(); 1137 else 1138 StrE = cast<StringLiteral>(E); 1139 1140 if (StrE) { 1141 CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx, 1142 firstDataArg, isPrintf); 1143 return true; 1144 } 1145 1146 return false; 1147 } 1148 1149 default: 1150 return false; 1151 } 1152} 1153 1154void 1155Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1156 const Expr * const *ExprArgs, 1157 SourceLocation CallSiteLoc) { 1158 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1159 e = NonNull->args_end(); 1160 i != e; ++i) { 1161 const Expr *ArgExpr = ExprArgs[*i]; 1162 if (ArgExpr->isNullPointerConstant(Context, 1163 Expr::NPC_ValueDependentIsNotNull)) 1164 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 1165 } 1166} 1167 1168/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar 1169/// functions) for correct use of format strings. 1170void 1171Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg, 1172 unsigned format_idx, unsigned firstDataArg, 1173 bool isPrintf) { 1174 1175 const Expr *Fn = TheCall->getCallee(); 1176 1177 // The way the format attribute works in GCC, the implicit this argument 1178 // of member functions is counted. However, it doesn't appear in our own 1179 // lists, so decrement format_idx in that case. 1180 if (isa<CXXMemberCallExpr>(TheCall)) { 1181 const CXXMethodDecl *method_decl = 1182 dyn_cast<CXXMethodDecl>(TheCall->getCalleeDecl()); 1183 if (method_decl && method_decl->isInstance()) { 1184 // Catch a format attribute mistakenly referring to the object argument. 1185 if (format_idx == 0) 1186 return; 1187 --format_idx; 1188 if(firstDataArg != 0) 1189 --firstDataArg; 1190 } 1191 } 1192 1193 // CHECK: printf/scanf-like function is called with no format string. 1194 if (format_idx >= TheCall->getNumArgs()) { 1195 Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string) 1196 << Fn->getSourceRange(); 1197 return; 1198 } 1199 1200 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 1201 1202 // CHECK: format string is not a string literal. 1203 // 1204 // Dynamically generated format strings are difficult to 1205 // automatically vet at compile time. Requiring that format strings 1206 // are string literals: (1) permits the checking of format strings by 1207 // the compiler and thereby (2) can practically remove the source of 1208 // many format string exploits. 1209 1210 // Format string can be either ObjC string (e.g. @"%d") or 1211 // C string (e.g. "%d") 1212 // ObjC string uses the same format specifiers as C string, so we can use 1213 // the same format string checking logic for both ObjC and C strings. 1214 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 1215 firstDataArg, isPrintf)) 1216 return; // Literal format string found, check done! 1217 1218 // If there are no arguments specified, warn with -Wformat-security, otherwise 1219 // warn only with -Wformat-nonliteral. 1220 if (TheCall->getNumArgs() == format_idx+1) 1221 Diag(TheCall->getArg(format_idx)->getLocStart(), 1222 diag::warn_format_nonliteral_noargs) 1223 << OrigFormatExpr->getSourceRange(); 1224 else 1225 Diag(TheCall->getArg(format_idx)->getLocStart(), 1226 diag::warn_format_nonliteral) 1227 << OrigFormatExpr->getSourceRange(); 1228} 1229 1230namespace { 1231class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1232protected: 1233 Sema &S; 1234 const StringLiteral *FExpr; 1235 const Expr *OrigFormatExpr; 1236 const unsigned FirstDataArg; 1237 const unsigned NumDataArgs; 1238 const bool IsObjCLiteral; 1239 const char *Beg; // Start of format string. 1240 const bool HasVAListArg; 1241 const CallExpr *TheCall; 1242 unsigned FormatIdx; 1243 llvm::BitVector CoveredArgs; 1244 bool usesPositionalArgs; 1245 bool atFirstArg; 1246public: 1247 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1248 const Expr *origFormatExpr, unsigned firstDataArg, 1249 unsigned numDataArgs, bool isObjCLiteral, 1250 const char *beg, bool hasVAListArg, 1251 const CallExpr *theCall, unsigned formatIdx) 1252 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1253 FirstDataArg(firstDataArg), 1254 NumDataArgs(numDataArgs), 1255 IsObjCLiteral(isObjCLiteral), Beg(beg), 1256 HasVAListArg(hasVAListArg), 1257 TheCall(theCall), FormatIdx(formatIdx), 1258 usesPositionalArgs(false), atFirstArg(true) { 1259 CoveredArgs.resize(numDataArgs); 1260 CoveredArgs.reset(); 1261 } 1262 1263 void DoneProcessing(); 1264 1265 void HandleIncompleteSpecifier(const char *startSpecifier, 1266 unsigned specifierLen); 1267 1268 virtual void HandleInvalidPosition(const char *startSpecifier, 1269 unsigned specifierLen, 1270 analyze_format_string::PositionContext p); 1271 1272 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1273 1274 void HandleNullChar(const char *nullCharacter); 1275 1276protected: 1277 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1278 const char *startSpec, 1279 unsigned specifierLen, 1280 const char *csStart, unsigned csLen); 1281 1282 SourceRange getFormatStringRange(); 1283 CharSourceRange getSpecifierRange(const char *startSpecifier, 1284 unsigned specifierLen); 1285 SourceLocation getLocationOfByte(const char *x); 1286 1287 const Expr *getDataArg(unsigned i) const; 1288 1289 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1290 const analyze_format_string::ConversionSpecifier &CS, 1291 const char *startSpecifier, unsigned specifierLen, 1292 unsigned argIndex); 1293}; 1294} 1295 1296SourceRange CheckFormatHandler::getFormatStringRange() { 1297 return OrigFormatExpr->getSourceRange(); 1298} 1299 1300CharSourceRange CheckFormatHandler:: 1301getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1302 SourceLocation Start = getLocationOfByte(startSpecifier); 1303 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1304 1305 // Advance the end SourceLocation by one due to half-open ranges. 1306 End = End.getLocWithOffset(1); 1307 1308 return CharSourceRange::getCharRange(Start, End); 1309} 1310 1311SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 1312 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1313} 1314 1315void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 1316 unsigned specifierLen){ 1317 SourceLocation Loc = getLocationOfByte(startSpecifier); 1318 S.Diag(Loc, diag::warn_printf_incomplete_specifier) 1319 << getSpecifierRange(startSpecifier, specifierLen); 1320} 1321 1322void 1323CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1324 analyze_format_string::PositionContext p) { 1325 SourceLocation Loc = getLocationOfByte(startPos); 1326 S.Diag(Loc, diag::warn_format_invalid_positional_specifier) 1327 << (unsigned) p << getSpecifierRange(startPos, posLen); 1328} 1329 1330void CheckFormatHandler::HandleZeroPosition(const char *startPos, 1331 unsigned posLen) { 1332 SourceLocation Loc = getLocationOfByte(startPos); 1333 S.Diag(Loc, diag::warn_format_zero_positional_specifier) 1334 << getSpecifierRange(startPos, posLen); 1335} 1336 1337void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 1338 if (!IsObjCLiteral) { 1339 // The presence of a null character is likely an error. 1340 S.Diag(getLocationOfByte(nullCharacter), 1341 diag::warn_printf_format_string_contains_null_char) 1342 << getFormatStringRange(); 1343 } 1344} 1345 1346const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 1347 return TheCall->getArg(FirstDataArg + i); 1348} 1349 1350void CheckFormatHandler::DoneProcessing() { 1351 // Does the number of data arguments exceed the number of 1352 // format conversions in the format string? 1353 if (!HasVAListArg) { 1354 // Find any arguments that weren't covered. 1355 CoveredArgs.flip(); 1356 signed notCoveredArg = CoveredArgs.find_first(); 1357 if (notCoveredArg >= 0) { 1358 assert((unsigned)notCoveredArg < NumDataArgs); 1359 S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(), 1360 diag::warn_printf_data_arg_not_used) 1361 << getFormatStringRange(); 1362 } 1363 } 1364} 1365 1366bool 1367CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 1368 SourceLocation Loc, 1369 const char *startSpec, 1370 unsigned specifierLen, 1371 const char *csStart, 1372 unsigned csLen) { 1373 1374 bool keepGoing = true; 1375 if (argIndex < NumDataArgs) { 1376 // Consider the argument coverered, even though the specifier doesn't 1377 // make sense. 1378 CoveredArgs.set(argIndex); 1379 } 1380 else { 1381 // If argIndex exceeds the number of data arguments we 1382 // don't issue a warning because that is just a cascade of warnings (and 1383 // they may have intended '%%' anyway). We don't want to continue processing 1384 // the format string after this point, however, as we will like just get 1385 // gibberish when trying to match arguments. 1386 keepGoing = false; 1387 } 1388 1389 S.Diag(Loc, diag::warn_format_invalid_conversion) 1390 << StringRef(csStart, csLen) 1391 << getSpecifierRange(startSpec, specifierLen); 1392 1393 return keepGoing; 1394} 1395 1396bool 1397CheckFormatHandler::CheckNumArgs( 1398 const analyze_format_string::FormatSpecifier &FS, 1399 const analyze_format_string::ConversionSpecifier &CS, 1400 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 1401 1402 if (argIndex >= NumDataArgs) { 1403 if (FS.usesPositionalArg()) { 1404 S.Diag(getLocationOfByte(CS.getStart()), 1405 diag::warn_printf_positional_arg_exceeds_data_args) 1406 << (argIndex+1) << NumDataArgs 1407 << getSpecifierRange(startSpecifier, specifierLen); 1408 } 1409 else { 1410 S.Diag(getLocationOfByte(CS.getStart()), 1411 diag::warn_printf_insufficient_data_args) 1412 << getSpecifierRange(startSpecifier, specifierLen); 1413 } 1414 1415 return false; 1416 } 1417 return true; 1418} 1419 1420//===--- CHECK: Printf format string checking ------------------------------===// 1421 1422namespace { 1423class CheckPrintfHandler : public CheckFormatHandler { 1424public: 1425 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1426 const Expr *origFormatExpr, unsigned firstDataArg, 1427 unsigned numDataArgs, bool isObjCLiteral, 1428 const char *beg, bool hasVAListArg, 1429 const CallExpr *theCall, unsigned formatIdx) 1430 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1431 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1432 theCall, formatIdx) {} 1433 1434 1435 bool HandleInvalidPrintfConversionSpecifier( 1436 const analyze_printf::PrintfSpecifier &FS, 1437 const char *startSpecifier, 1438 unsigned specifierLen); 1439 1440 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 1441 const char *startSpecifier, 1442 unsigned specifierLen); 1443 1444 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 1445 const char *startSpecifier, unsigned specifierLen); 1446 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 1447 const analyze_printf::OptionalAmount &Amt, 1448 unsigned type, 1449 const char *startSpecifier, unsigned specifierLen); 1450 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1451 const analyze_printf::OptionalFlag &flag, 1452 const char *startSpecifier, unsigned specifierLen); 1453 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 1454 const analyze_printf::OptionalFlag &ignoredFlag, 1455 const analyze_printf::OptionalFlag &flag, 1456 const char *startSpecifier, unsigned specifierLen); 1457}; 1458} 1459 1460bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 1461 const analyze_printf::PrintfSpecifier &FS, 1462 const char *startSpecifier, 1463 unsigned specifierLen) { 1464 const analyze_printf::PrintfConversionSpecifier &CS = 1465 FS.getConversionSpecifier(); 1466 1467 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1468 getLocationOfByte(CS.getStart()), 1469 startSpecifier, specifierLen, 1470 CS.getStart(), CS.getLength()); 1471} 1472 1473bool CheckPrintfHandler::HandleAmount( 1474 const analyze_format_string::OptionalAmount &Amt, 1475 unsigned k, const char *startSpecifier, 1476 unsigned specifierLen) { 1477 1478 if (Amt.hasDataArgument()) { 1479 if (!HasVAListArg) { 1480 unsigned argIndex = Amt.getArgIndex(); 1481 if (argIndex >= NumDataArgs) { 1482 S.Diag(getLocationOfByte(Amt.getStart()), 1483 diag::warn_printf_asterisk_missing_arg) 1484 << k << getSpecifierRange(startSpecifier, specifierLen); 1485 // Don't do any more checking. We will just emit 1486 // spurious errors. 1487 return false; 1488 } 1489 1490 // Type check the data argument. It should be an 'int'. 1491 // Although not in conformance with C99, we also allow the argument to be 1492 // an 'unsigned int' as that is a reasonably safe case. GCC also 1493 // doesn't emit a warning for that case. 1494 CoveredArgs.set(argIndex); 1495 const Expr *Arg = getDataArg(argIndex); 1496 QualType T = Arg->getType(); 1497 1498 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1499 assert(ATR.isValid()); 1500 1501 if (!ATR.matchesType(S.Context, T)) { 1502 S.Diag(getLocationOfByte(Amt.getStart()), 1503 diag::warn_printf_asterisk_wrong_type) 1504 << k 1505 << ATR.getRepresentativeType(S.Context) << T 1506 << getSpecifierRange(startSpecifier, specifierLen) 1507 << Arg->getSourceRange(); 1508 // Don't do any more checking. We will just emit 1509 // spurious errors. 1510 return false; 1511 } 1512 } 1513 } 1514 return true; 1515} 1516 1517void CheckPrintfHandler::HandleInvalidAmount( 1518 const analyze_printf::PrintfSpecifier &FS, 1519 const analyze_printf::OptionalAmount &Amt, 1520 unsigned type, 1521 const char *startSpecifier, 1522 unsigned specifierLen) { 1523 const analyze_printf::PrintfConversionSpecifier &CS = 1524 FS.getConversionSpecifier(); 1525 switch (Amt.getHowSpecified()) { 1526 case analyze_printf::OptionalAmount::Constant: 1527 S.Diag(getLocationOfByte(Amt.getStart()), 1528 diag::warn_printf_nonsensical_optional_amount) 1529 << type 1530 << CS.toString() 1531 << getSpecifierRange(startSpecifier, specifierLen) 1532 << FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 1533 Amt.getConstantLength())); 1534 break; 1535 1536 default: 1537 S.Diag(getLocationOfByte(Amt.getStart()), 1538 diag::warn_printf_nonsensical_optional_amount) 1539 << type 1540 << CS.toString() 1541 << getSpecifierRange(startSpecifier, specifierLen); 1542 break; 1543 } 1544} 1545 1546void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1547 const analyze_printf::OptionalFlag &flag, 1548 const char *startSpecifier, 1549 unsigned specifierLen) { 1550 // Warn about pointless flag with a fixit removal. 1551 const analyze_printf::PrintfConversionSpecifier &CS = 1552 FS.getConversionSpecifier(); 1553 S.Diag(getLocationOfByte(flag.getPosition()), 1554 diag::warn_printf_nonsensical_flag) 1555 << flag.toString() << CS.toString() 1556 << getSpecifierRange(startSpecifier, specifierLen) 1557 << FixItHint::CreateRemoval(getSpecifierRange(flag.getPosition(), 1)); 1558} 1559 1560void CheckPrintfHandler::HandleIgnoredFlag( 1561 const analyze_printf::PrintfSpecifier &FS, 1562 const analyze_printf::OptionalFlag &ignoredFlag, 1563 const analyze_printf::OptionalFlag &flag, 1564 const char *startSpecifier, 1565 unsigned specifierLen) { 1566 // Warn about ignored flag with a fixit removal. 1567 S.Diag(getLocationOfByte(ignoredFlag.getPosition()), 1568 diag::warn_printf_ignored_flag) 1569 << ignoredFlag.toString() << flag.toString() 1570 << getSpecifierRange(startSpecifier, specifierLen) 1571 << FixItHint::CreateRemoval(getSpecifierRange( 1572 ignoredFlag.getPosition(), 1)); 1573} 1574 1575bool 1576CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 1577 &FS, 1578 const char *startSpecifier, 1579 unsigned specifierLen) { 1580 1581 using namespace analyze_format_string; 1582 using namespace analyze_printf; 1583 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 1584 1585 if (FS.consumesDataArgument()) { 1586 if (atFirstArg) { 1587 atFirstArg = false; 1588 usesPositionalArgs = FS.usesPositionalArg(); 1589 } 1590 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1591 // Cannot mix-and-match positional and non-positional arguments. 1592 S.Diag(getLocationOfByte(CS.getStart()), 1593 diag::warn_format_mix_positional_nonpositional_args) 1594 << getSpecifierRange(startSpecifier, specifierLen); 1595 return false; 1596 } 1597 } 1598 1599 // First check if the field width, precision, and conversion specifier 1600 // have matching data arguments. 1601 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 1602 startSpecifier, specifierLen)) { 1603 return false; 1604 } 1605 1606 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 1607 startSpecifier, specifierLen)) { 1608 return false; 1609 } 1610 1611 if (!CS.consumesDataArgument()) { 1612 // FIXME: Technically specifying a precision or field width here 1613 // makes no sense. Worth issuing a warning at some point. 1614 return true; 1615 } 1616 1617 // Consume the argument. 1618 unsigned argIndex = FS.getArgIndex(); 1619 if (argIndex < NumDataArgs) { 1620 // The check to see if the argIndex is valid will come later. 1621 // We set the bit here because we may exit early from this 1622 // function if we encounter some other error. 1623 CoveredArgs.set(argIndex); 1624 } 1625 1626 // Check for using an Objective-C specific conversion specifier 1627 // in a non-ObjC literal. 1628 if (!IsObjCLiteral && CS.isObjCArg()) { 1629 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 1630 specifierLen); 1631 } 1632 1633 // Check for invalid use of field width 1634 if (!FS.hasValidFieldWidth()) { 1635 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 1636 startSpecifier, specifierLen); 1637 } 1638 1639 // Check for invalid use of precision 1640 if (!FS.hasValidPrecision()) { 1641 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 1642 startSpecifier, specifierLen); 1643 } 1644 1645 // Check each flag does not conflict with any other component. 1646 if (!FS.hasValidThousandsGroupingPrefix()) 1647 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 1648 if (!FS.hasValidLeadingZeros()) 1649 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 1650 if (!FS.hasValidPlusPrefix()) 1651 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 1652 if (!FS.hasValidSpacePrefix()) 1653 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 1654 if (!FS.hasValidAlternativeForm()) 1655 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 1656 if (!FS.hasValidLeftJustified()) 1657 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 1658 1659 // Check that flags are not ignored by another flag 1660 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 1661 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 1662 startSpecifier, specifierLen); 1663 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 1664 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 1665 startSpecifier, specifierLen); 1666 1667 // Check the length modifier is valid with the given conversion specifier. 1668 const LengthModifier &LM = FS.getLengthModifier(); 1669 if (!FS.hasValidLengthModifier()) 1670 S.Diag(getLocationOfByte(LM.getStart()), 1671 diag::warn_format_nonsensical_length) 1672 << LM.toString() << CS.toString() 1673 << getSpecifierRange(startSpecifier, specifierLen) 1674 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 1675 LM.getLength())); 1676 1677 // Are we using '%n'? 1678 if (CS.getKind() == ConversionSpecifier::nArg) { 1679 // Issue a warning about this being a possible security issue. 1680 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back) 1681 << getSpecifierRange(startSpecifier, specifierLen); 1682 // Continue checking the other format specifiers. 1683 return true; 1684 } 1685 1686 // The remaining checks depend on the data arguments. 1687 if (HasVAListArg) 1688 return true; 1689 1690 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 1691 return false; 1692 1693 // Now type check the data expression that matches the 1694 // format specifier. 1695 const Expr *Ex = getDataArg(argIndex); 1696 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context); 1697 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 1698 // Check if we didn't match because of an implicit cast from a 'char' 1699 // or 'short' to an 'int'. This is done because printf is a varargs 1700 // function. 1701 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 1702 if (ICE->getType() == S.Context.IntTy) { 1703 // All further checking is done on the subexpression. 1704 Ex = ICE->getSubExpr(); 1705 if (ATR.matchesType(S.Context, Ex->getType())) 1706 return true; 1707 } 1708 1709 // We may be able to offer a FixItHint if it is a supported type. 1710 PrintfSpecifier fixedFS = FS; 1711 bool success = fixedFS.fixType(Ex->getType()); 1712 1713 if (success) { 1714 // Get the fix string from the fixed format specifier 1715 llvm::SmallString<128> buf; 1716 llvm::raw_svector_ostream os(buf); 1717 fixedFS.toString(os); 1718 1719 // FIXME: getRepresentativeType() perhaps should return a string 1720 // instead of a QualType to better handle when the representative 1721 // type is 'wint_t' (which is defined in the system headers). 1722 S.Diag(getLocationOfByte(CS.getStart()), 1723 diag::warn_printf_conversion_argument_type_mismatch) 1724 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1725 << getSpecifierRange(startSpecifier, specifierLen) 1726 << Ex->getSourceRange() 1727 << FixItHint::CreateReplacement( 1728 getSpecifierRange(startSpecifier, specifierLen), 1729 os.str()); 1730 } 1731 else { 1732 S.Diag(getLocationOfByte(CS.getStart()), 1733 diag::warn_printf_conversion_argument_type_mismatch) 1734 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1735 << getSpecifierRange(startSpecifier, specifierLen) 1736 << Ex->getSourceRange(); 1737 } 1738 } 1739 1740 return true; 1741} 1742 1743//===--- CHECK: Scanf format string checking ------------------------------===// 1744 1745namespace { 1746class CheckScanfHandler : public CheckFormatHandler { 1747public: 1748 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 1749 const Expr *origFormatExpr, unsigned firstDataArg, 1750 unsigned numDataArgs, bool isObjCLiteral, 1751 const char *beg, bool hasVAListArg, 1752 const CallExpr *theCall, unsigned formatIdx) 1753 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1754 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1755 theCall, formatIdx) {} 1756 1757 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 1758 const char *startSpecifier, 1759 unsigned specifierLen); 1760 1761 bool HandleInvalidScanfConversionSpecifier( 1762 const analyze_scanf::ScanfSpecifier &FS, 1763 const char *startSpecifier, 1764 unsigned specifierLen); 1765 1766 void HandleIncompleteScanList(const char *start, const char *end); 1767}; 1768} 1769 1770void CheckScanfHandler::HandleIncompleteScanList(const char *start, 1771 const char *end) { 1772 S.Diag(getLocationOfByte(end), diag::warn_scanf_scanlist_incomplete) 1773 << getSpecifierRange(start, end - start); 1774} 1775 1776bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 1777 const analyze_scanf::ScanfSpecifier &FS, 1778 const char *startSpecifier, 1779 unsigned specifierLen) { 1780 1781 const analyze_scanf::ScanfConversionSpecifier &CS = 1782 FS.getConversionSpecifier(); 1783 1784 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1785 getLocationOfByte(CS.getStart()), 1786 startSpecifier, specifierLen, 1787 CS.getStart(), CS.getLength()); 1788} 1789 1790bool CheckScanfHandler::HandleScanfSpecifier( 1791 const analyze_scanf::ScanfSpecifier &FS, 1792 const char *startSpecifier, 1793 unsigned specifierLen) { 1794 1795 using namespace analyze_scanf; 1796 using namespace analyze_format_string; 1797 1798 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 1799 1800 // Handle case where '%' and '*' don't consume an argument. These shouldn't 1801 // be used to decide if we are using positional arguments consistently. 1802 if (FS.consumesDataArgument()) { 1803 if (atFirstArg) { 1804 atFirstArg = false; 1805 usesPositionalArgs = FS.usesPositionalArg(); 1806 } 1807 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1808 // Cannot mix-and-match positional and non-positional arguments. 1809 S.Diag(getLocationOfByte(CS.getStart()), 1810 diag::warn_format_mix_positional_nonpositional_args) 1811 << getSpecifierRange(startSpecifier, specifierLen); 1812 return false; 1813 } 1814 } 1815 1816 // Check if the field with is non-zero. 1817 const OptionalAmount &Amt = FS.getFieldWidth(); 1818 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 1819 if (Amt.getConstantAmount() == 0) { 1820 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 1821 Amt.getConstantLength()); 1822 S.Diag(getLocationOfByte(Amt.getStart()), 1823 diag::warn_scanf_nonzero_width) 1824 << R << FixItHint::CreateRemoval(R); 1825 } 1826 } 1827 1828 if (!FS.consumesDataArgument()) { 1829 // FIXME: Technically specifying a precision or field width here 1830 // makes no sense. Worth issuing a warning at some point. 1831 return true; 1832 } 1833 1834 // Consume the argument. 1835 unsigned argIndex = FS.getArgIndex(); 1836 if (argIndex < NumDataArgs) { 1837 // The check to see if the argIndex is valid will come later. 1838 // We set the bit here because we may exit early from this 1839 // function if we encounter some other error. 1840 CoveredArgs.set(argIndex); 1841 } 1842 1843 // Check the length modifier is valid with the given conversion specifier. 1844 const LengthModifier &LM = FS.getLengthModifier(); 1845 if (!FS.hasValidLengthModifier()) { 1846 S.Diag(getLocationOfByte(LM.getStart()), 1847 diag::warn_format_nonsensical_length) 1848 << LM.toString() << CS.toString() 1849 << getSpecifierRange(startSpecifier, specifierLen) 1850 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 1851 LM.getLength())); 1852 } 1853 1854 // The remaining checks depend on the data arguments. 1855 if (HasVAListArg) 1856 return true; 1857 1858 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 1859 return false; 1860 1861 // FIXME: Check that the argument type matches the format specifier. 1862 1863 return true; 1864} 1865 1866void Sema::CheckFormatString(const StringLiteral *FExpr, 1867 const Expr *OrigFormatExpr, 1868 const CallExpr *TheCall, bool HasVAListArg, 1869 unsigned format_idx, unsigned firstDataArg, 1870 bool isPrintf) { 1871 1872 // CHECK: is the format string a wide literal? 1873 if (!FExpr->isAscii()) { 1874 Diag(FExpr->getLocStart(), 1875 diag::warn_format_string_is_wide_literal) 1876 << OrigFormatExpr->getSourceRange(); 1877 return; 1878 } 1879 1880 // Str - The format string. NOTE: this is NOT null-terminated! 1881 StringRef StrRef = FExpr->getString(); 1882 const char *Str = StrRef.data(); 1883 unsigned StrLen = StrRef.size(); 1884 const unsigned numDataArgs = TheCall->getNumArgs() - firstDataArg; 1885 1886 // CHECK: empty format string? 1887 if (StrLen == 0 && numDataArgs > 0) { 1888 Diag(FExpr->getLocStart(), diag::warn_empty_format_string) 1889 << OrigFormatExpr->getSourceRange(); 1890 return; 1891 } 1892 1893 if (isPrintf) { 1894 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1895 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 1896 Str, HasVAListArg, TheCall, format_idx); 1897 1898 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen)) 1899 H.DoneProcessing(); 1900 } 1901 else { 1902 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1903 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 1904 Str, HasVAListArg, TheCall, format_idx); 1905 1906 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen)) 1907 H.DoneProcessing(); 1908 } 1909} 1910 1911//===--- CHECK: Standard memory functions ---------------------------------===// 1912 1913/// \brief Determine whether the given type is a dynamic class type (e.g., 1914/// whether it has a vtable). 1915static bool isDynamicClassType(QualType T) { 1916 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 1917 if (CXXRecordDecl *Definition = Record->getDefinition()) 1918 if (Definition->isDynamicClass()) 1919 return true; 1920 1921 return false; 1922} 1923 1924/// \brief If E is a sizeof expression, returns its argument expression, 1925/// otherwise returns NULL. 1926static const Expr *getSizeOfExprArg(const Expr* E) { 1927 if (const UnaryExprOrTypeTraitExpr *SizeOf = 1928 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 1929 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 1930 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 1931 1932 return 0; 1933} 1934 1935/// \brief If E is a sizeof expression, returns its argument type. 1936static QualType getSizeOfArgType(const Expr* E) { 1937 if (const UnaryExprOrTypeTraitExpr *SizeOf = 1938 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 1939 if (SizeOf->getKind() == clang::UETT_SizeOf) 1940 return SizeOf->getTypeOfArgument(); 1941 1942 return QualType(); 1943} 1944 1945/// \brief Check for dangerous or invalid arguments to memset(). 1946/// 1947/// This issues warnings on known problematic, dangerous or unspecified 1948/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 1949/// function calls. 1950/// 1951/// \param Call The call expression to diagnose. 1952void Sema::CheckMemaccessArguments(const CallExpr *Call, 1953 CheckedMemoryFunction CMF, 1954 IdentifierInfo *FnName) { 1955 // It is possible to have a non-standard definition of memset. Validate 1956 // we have enough arguments, and if not, abort further checking. 1957 if (Call->getNumArgs() < 3) 1958 return; 1959 1960 unsigned LastArg = (CMF == CMF_Memset? 1 : 2); 1961 const Expr *LenExpr = Call->getArg(2)->IgnoreParenImpCasts(); 1962 1963 // We have special checking when the length is a sizeof expression. 1964 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 1965 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 1966 llvm::FoldingSetNodeID SizeOfArgID; 1967 1968 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 1969 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 1970 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 1971 1972 QualType DestTy = Dest->getType(); 1973 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 1974 QualType PointeeTy = DestPtrTy->getPointeeType(); 1975 1976 // Never warn about void type pointers. This can be used to suppress 1977 // false positives. 1978 if (PointeeTy->isVoidType()) 1979 continue; 1980 1981 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 1982 // actually comparing the expressions for equality. Because computing the 1983 // expression IDs can be expensive, we only do this if the diagnostic is 1984 // enabled. 1985 if (SizeOfArg && 1986 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 1987 SizeOfArg->getExprLoc())) { 1988 // We only compute IDs for expressions if the warning is enabled, and 1989 // cache the sizeof arg's ID. 1990 if (SizeOfArgID == llvm::FoldingSetNodeID()) 1991 SizeOfArg->Profile(SizeOfArgID, Context, true); 1992 llvm::FoldingSetNodeID DestID; 1993 Dest->Profile(DestID, Context, true); 1994 if (DestID == SizeOfArgID) { 1995 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 1996 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 1997 if (UnaryOp->getOpcode() == UO_AddrOf) 1998 ActionIdx = 1; // If its an address-of operator, just remove it. 1999 if (Context.getTypeSize(PointeeTy) == Context.getCharWidth()) 2000 ActionIdx = 2; // If the pointee's size is sizeof(char), 2001 // suggest an explicit length. 2002 DiagRuntimeBehavior(SizeOfArg->getExprLoc(), Dest, 2003 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 2004 << FnName << ArgIdx << ActionIdx 2005 << Dest->getSourceRange() 2006 << SizeOfArg->getSourceRange()); 2007 break; 2008 } 2009 } 2010 2011 // Also check for cases where the sizeof argument is the exact same 2012 // type as the memory argument, and where it points to a user-defined 2013 // record type. 2014 if (SizeOfArgTy != QualType()) { 2015 if (PointeeTy->isRecordType() && 2016 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 2017 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 2018 PDiag(diag::warn_sizeof_pointer_type_memaccess) 2019 << FnName << SizeOfArgTy << ArgIdx 2020 << PointeeTy << Dest->getSourceRange() 2021 << LenExpr->getSourceRange()); 2022 break; 2023 } 2024 } 2025 2026 // Always complain about dynamic classes. 2027 if (isDynamicClassType(PointeeTy)) 2028 DiagRuntimeBehavior( 2029 Dest->getExprLoc(), Dest, 2030 PDiag(diag::warn_dyn_class_memaccess) 2031 << (CMF == CMF_Memcmp ? ArgIdx + 2 : ArgIdx) << FnName << PointeeTy 2032 // "overwritten" if we're warning about the destination for any call 2033 // but memcmp; otherwise a verb appropriate to the call. 2034 << (ArgIdx == 0 && CMF != CMF_Memcmp ? 0 : (unsigned)CMF) 2035 << Call->getCallee()->getSourceRange()); 2036 else if (PointeeTy.hasNonTrivialObjCLifetime() && CMF != CMF_Memset) 2037 DiagRuntimeBehavior( 2038 Dest->getExprLoc(), Dest, 2039 PDiag(diag::warn_arc_object_memaccess) 2040 << ArgIdx << FnName << PointeeTy 2041 << Call->getCallee()->getSourceRange()); 2042 else 2043 continue; 2044 2045 DiagRuntimeBehavior( 2046 Dest->getExprLoc(), Dest, 2047 PDiag(diag::note_bad_memaccess_silence) 2048 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 2049 break; 2050 } 2051 } 2052} 2053 2054// A little helper routine: ignore addition and subtraction of integer literals. 2055// This intentionally does not ignore all integer constant expressions because 2056// we don't want to remove sizeof(). 2057static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 2058 Ex = Ex->IgnoreParenCasts(); 2059 2060 for (;;) { 2061 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 2062 if (!BO || !BO->isAdditiveOp()) 2063 break; 2064 2065 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 2066 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 2067 2068 if (isa<IntegerLiteral>(RHS)) 2069 Ex = LHS; 2070 else if (isa<IntegerLiteral>(LHS)) 2071 Ex = RHS; 2072 else 2073 break; 2074 } 2075 2076 return Ex; 2077} 2078 2079// Warn if the user has made the 'size' argument to strlcpy or strlcat 2080// be the size of the source, instead of the destination. 2081void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 2082 IdentifierInfo *FnName) { 2083 2084 // Don't crash if the user has the wrong number of arguments 2085 if (Call->getNumArgs() != 3) 2086 return; 2087 2088 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 2089 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 2090 const Expr *CompareWithSrc = NULL; 2091 2092 // Look for 'strlcpy(dst, x, sizeof(x))' 2093 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 2094 CompareWithSrc = Ex; 2095 else { 2096 // Look for 'strlcpy(dst, x, strlen(x))' 2097 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 2098 if (SizeCall->isBuiltinCall(Context) == Builtin::BIstrlen 2099 && SizeCall->getNumArgs() == 1) 2100 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 2101 } 2102 } 2103 2104 if (!CompareWithSrc) 2105 return; 2106 2107 // Determine if the argument to sizeof/strlen is equal to the source 2108 // argument. In principle there's all kinds of things you could do 2109 // here, for instance creating an == expression and evaluating it with 2110 // EvaluateAsBooleanCondition, but this uses a more direct technique: 2111 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 2112 if (!SrcArgDRE) 2113 return; 2114 2115 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 2116 if (!CompareWithSrcDRE || 2117 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 2118 return; 2119 2120 const Expr *OriginalSizeArg = Call->getArg(2); 2121 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 2122 << OriginalSizeArg->getSourceRange() << FnName; 2123 2124 // Output a FIXIT hint if the destination is an array (rather than a 2125 // pointer to an array). This could be enhanced to handle some 2126 // pointers if we know the actual size, like if DstArg is 'array+2' 2127 // we could say 'sizeof(array)-2'. 2128 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 2129 QualType DstArgTy = DstArg->getType(); 2130 2131 // Only handle constant-sized or VLAs, but not flexible members. 2132 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 2133 // Only issue the FIXIT for arrays of size > 1. 2134 if (CAT->getSize().getSExtValue() <= 1) 2135 return; 2136 } else if (!DstArgTy->isVariableArrayType()) { 2137 return; 2138 } 2139 2140 llvm::SmallString<128> sizeString; 2141 llvm::raw_svector_ostream OS(sizeString); 2142 OS << "sizeof("; 2143 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 2144 OS << ")"; 2145 2146 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 2147 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 2148 OS.str()); 2149} 2150 2151//===--- CHECK: Return Address of Stack Variable --------------------------===// 2152 2153static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars); 2154static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars); 2155 2156/// CheckReturnStackAddr - Check if a return statement returns the address 2157/// of a stack variable. 2158void 2159Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 2160 SourceLocation ReturnLoc) { 2161 2162 Expr *stackE = 0; 2163 SmallVector<DeclRefExpr *, 8> refVars; 2164 2165 // Perform checking for returned stack addresses, local blocks, 2166 // label addresses or references to temporaries. 2167 if (lhsType->isPointerType() || 2168 (!getLangOptions().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 2169 stackE = EvalAddr(RetValExp, refVars); 2170 } else if (lhsType->isReferenceType()) { 2171 stackE = EvalVal(RetValExp, refVars); 2172 } 2173 2174 if (stackE == 0) 2175 return; // Nothing suspicious was found. 2176 2177 SourceLocation diagLoc; 2178 SourceRange diagRange; 2179 if (refVars.empty()) { 2180 diagLoc = stackE->getLocStart(); 2181 diagRange = stackE->getSourceRange(); 2182 } else { 2183 // We followed through a reference variable. 'stackE' contains the 2184 // problematic expression but we will warn at the return statement pointing 2185 // at the reference variable. We will later display the "trail" of 2186 // reference variables using notes. 2187 diagLoc = refVars[0]->getLocStart(); 2188 diagRange = refVars[0]->getSourceRange(); 2189 } 2190 2191 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 2192 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 2193 : diag::warn_ret_stack_addr) 2194 << DR->getDecl()->getDeclName() << diagRange; 2195 } else if (isa<BlockExpr>(stackE)) { // local block. 2196 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 2197 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 2198 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 2199 } else { // local temporary. 2200 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 2201 : diag::warn_ret_local_temp_addr) 2202 << diagRange; 2203 } 2204 2205 // Display the "trail" of reference variables that we followed until we 2206 // found the problematic expression using notes. 2207 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 2208 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 2209 // If this var binds to another reference var, show the range of the next 2210 // var, otherwise the var binds to the problematic expression, in which case 2211 // show the range of the expression. 2212 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 2213 : stackE->getSourceRange(); 2214 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 2215 << VD->getDeclName() << range; 2216 } 2217} 2218 2219/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 2220/// check if the expression in a return statement evaluates to an address 2221/// to a location on the stack, a local block, an address of a label, or a 2222/// reference to local temporary. The recursion is used to traverse the 2223/// AST of the return expression, with recursion backtracking when we 2224/// encounter a subexpression that (1) clearly does not lead to one of the 2225/// above problematic expressions (2) is something we cannot determine leads to 2226/// a problematic expression based on such local checking. 2227/// 2228/// Both EvalAddr and EvalVal follow through reference variables to evaluate 2229/// the expression that they point to. Such variables are added to the 2230/// 'refVars' vector so that we know what the reference variable "trail" was. 2231/// 2232/// EvalAddr processes expressions that are pointers that are used as 2233/// references (and not L-values). EvalVal handles all other values. 2234/// At the base case of the recursion is a check for the above problematic 2235/// expressions. 2236/// 2237/// This implementation handles: 2238/// 2239/// * pointer-to-pointer casts 2240/// * implicit conversions from array references to pointers 2241/// * taking the address of fields 2242/// * arbitrary interplay between "&" and "*" operators 2243/// * pointer arithmetic from an address of a stack variable 2244/// * taking the address of an array element where the array is on the stack 2245static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 2246 if (E->isTypeDependent()) 2247 return NULL; 2248 2249 // We should only be called for evaluating pointer expressions. 2250 assert((E->getType()->isAnyPointerType() || 2251 E->getType()->isBlockPointerType() || 2252 E->getType()->isObjCQualifiedIdType()) && 2253 "EvalAddr only works on pointers"); 2254 2255 E = E->IgnoreParens(); 2256 2257 // Our "symbolic interpreter" is just a dispatch off the currently 2258 // viewed AST node. We then recursively traverse the AST by calling 2259 // EvalAddr and EvalVal appropriately. 2260 switch (E->getStmtClass()) { 2261 case Stmt::DeclRefExprClass: { 2262 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2263 2264 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2265 // If this is a reference variable, follow through to the expression that 2266 // it points to. 2267 if (V->hasLocalStorage() && 2268 V->getType()->isReferenceType() && V->hasInit()) { 2269 // Add the reference variable to the "trail". 2270 refVars.push_back(DR); 2271 return EvalAddr(V->getInit(), refVars); 2272 } 2273 2274 return NULL; 2275 } 2276 2277 case Stmt::UnaryOperatorClass: { 2278 // The only unary operator that make sense to handle here 2279 // is AddrOf. All others don't make sense as pointers. 2280 UnaryOperator *U = cast<UnaryOperator>(E); 2281 2282 if (U->getOpcode() == UO_AddrOf) 2283 return EvalVal(U->getSubExpr(), refVars); 2284 else 2285 return NULL; 2286 } 2287 2288 case Stmt::BinaryOperatorClass: { 2289 // Handle pointer arithmetic. All other binary operators are not valid 2290 // in this context. 2291 BinaryOperator *B = cast<BinaryOperator>(E); 2292 BinaryOperatorKind op = B->getOpcode(); 2293 2294 if (op != BO_Add && op != BO_Sub) 2295 return NULL; 2296 2297 Expr *Base = B->getLHS(); 2298 2299 // Determine which argument is the real pointer base. It could be 2300 // the RHS argument instead of the LHS. 2301 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 2302 2303 assert (Base->getType()->isPointerType()); 2304 return EvalAddr(Base, refVars); 2305 } 2306 2307 // For conditional operators we need to see if either the LHS or RHS are 2308 // valid DeclRefExpr*s. If one of them is valid, we return it. 2309 case Stmt::ConditionalOperatorClass: { 2310 ConditionalOperator *C = cast<ConditionalOperator>(E); 2311 2312 // Handle the GNU extension for missing LHS. 2313 if (Expr *lhsExpr = C->getLHS()) { 2314 // In C++, we can have a throw-expression, which has 'void' type. 2315 if (!lhsExpr->getType()->isVoidType()) 2316 if (Expr* LHS = EvalAddr(lhsExpr, refVars)) 2317 return LHS; 2318 } 2319 2320 // In C++, we can have a throw-expression, which has 'void' type. 2321 if (C->getRHS()->getType()->isVoidType()) 2322 return NULL; 2323 2324 return EvalAddr(C->getRHS(), refVars); 2325 } 2326 2327 case Stmt::BlockExprClass: 2328 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 2329 return E; // local block. 2330 return NULL; 2331 2332 case Stmt::AddrLabelExprClass: 2333 return E; // address of label. 2334 2335 // For casts, we need to handle conversions from arrays to 2336 // pointer values, and pointer-to-pointer conversions. 2337 case Stmt::ImplicitCastExprClass: 2338 case Stmt::CStyleCastExprClass: 2339 case Stmt::CXXFunctionalCastExprClass: 2340 case Stmt::ObjCBridgedCastExprClass: { 2341 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 2342 QualType T = SubExpr->getType(); 2343 2344 if (SubExpr->getType()->isPointerType() || 2345 SubExpr->getType()->isBlockPointerType() || 2346 SubExpr->getType()->isObjCQualifiedIdType()) 2347 return EvalAddr(SubExpr, refVars); 2348 else if (T->isArrayType()) 2349 return EvalVal(SubExpr, refVars); 2350 else 2351 return 0; 2352 } 2353 2354 // C++ casts. For dynamic casts, static casts, and const casts, we 2355 // are always converting from a pointer-to-pointer, so we just blow 2356 // through the cast. In the case the dynamic cast doesn't fail (and 2357 // return NULL), we take the conservative route and report cases 2358 // where we return the address of a stack variable. For Reinterpre 2359 // FIXME: The comment about is wrong; we're not always converting 2360 // from pointer to pointer. I'm guessing that this code should also 2361 // handle references to objects. 2362 case Stmt::CXXStaticCastExprClass: 2363 case Stmt::CXXDynamicCastExprClass: 2364 case Stmt::CXXConstCastExprClass: 2365 case Stmt::CXXReinterpretCastExprClass: { 2366 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 2367 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 2368 return EvalAddr(S, refVars); 2369 else 2370 return NULL; 2371 } 2372 2373 case Stmt::MaterializeTemporaryExprClass: 2374 if (Expr *Result = EvalAddr( 2375 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 2376 refVars)) 2377 return Result; 2378 2379 return E; 2380 2381 // Everything else: we simply don't reason about them. 2382 default: 2383 return NULL; 2384 } 2385} 2386 2387 2388/// EvalVal - This function is complements EvalAddr in the mutual recursion. 2389/// See the comments for EvalAddr for more details. 2390static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 2391do { 2392 // We should only be called for evaluating non-pointer expressions, or 2393 // expressions with a pointer type that are not used as references but instead 2394 // are l-values (e.g., DeclRefExpr with a pointer type). 2395 2396 // Our "symbolic interpreter" is just a dispatch off the currently 2397 // viewed AST node. We then recursively traverse the AST by calling 2398 // EvalAddr and EvalVal appropriately. 2399 2400 E = E->IgnoreParens(); 2401 switch (E->getStmtClass()) { 2402 case Stmt::ImplicitCastExprClass: { 2403 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 2404 if (IE->getValueKind() == VK_LValue) { 2405 E = IE->getSubExpr(); 2406 continue; 2407 } 2408 return NULL; 2409 } 2410 2411 case Stmt::DeclRefExprClass: { 2412 // When we hit a DeclRefExpr we are looking at code that refers to a 2413 // variable's name. If it's not a reference variable we check if it has 2414 // local storage within the function, and if so, return the expression. 2415 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2416 2417 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2418 if (V->hasLocalStorage()) { 2419 if (!V->getType()->isReferenceType()) 2420 return DR; 2421 2422 // Reference variable, follow through to the expression that 2423 // it points to. 2424 if (V->hasInit()) { 2425 // Add the reference variable to the "trail". 2426 refVars.push_back(DR); 2427 return EvalVal(V->getInit(), refVars); 2428 } 2429 } 2430 2431 return NULL; 2432 } 2433 2434 case Stmt::UnaryOperatorClass: { 2435 // The only unary operator that make sense to handle here 2436 // is Deref. All others don't resolve to a "name." This includes 2437 // handling all sorts of rvalues passed to a unary operator. 2438 UnaryOperator *U = cast<UnaryOperator>(E); 2439 2440 if (U->getOpcode() == UO_Deref) 2441 return EvalAddr(U->getSubExpr(), refVars); 2442 2443 return NULL; 2444 } 2445 2446 case Stmt::ArraySubscriptExprClass: { 2447 // Array subscripts are potential references to data on the stack. We 2448 // retrieve the DeclRefExpr* for the array variable if it indeed 2449 // has local storage. 2450 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars); 2451 } 2452 2453 case Stmt::ConditionalOperatorClass: { 2454 // For conditional operators we need to see if either the LHS or RHS are 2455 // non-NULL Expr's. If one is non-NULL, we return it. 2456 ConditionalOperator *C = cast<ConditionalOperator>(E); 2457 2458 // Handle the GNU extension for missing LHS. 2459 if (Expr *lhsExpr = C->getLHS()) 2460 if (Expr *LHS = EvalVal(lhsExpr, refVars)) 2461 return LHS; 2462 2463 return EvalVal(C->getRHS(), refVars); 2464 } 2465 2466 // Accesses to members are potential references to data on the stack. 2467 case Stmt::MemberExprClass: { 2468 MemberExpr *M = cast<MemberExpr>(E); 2469 2470 // Check for indirect access. We only want direct field accesses. 2471 if (M->isArrow()) 2472 return NULL; 2473 2474 // Check whether the member type is itself a reference, in which case 2475 // we're not going to refer to the member, but to what the member refers to. 2476 if (M->getMemberDecl()->getType()->isReferenceType()) 2477 return NULL; 2478 2479 return EvalVal(M->getBase(), refVars); 2480 } 2481 2482 case Stmt::MaterializeTemporaryExprClass: 2483 if (Expr *Result = EvalVal( 2484 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 2485 refVars)) 2486 return Result; 2487 2488 return E; 2489 2490 default: 2491 // Check that we don't return or take the address of a reference to a 2492 // temporary. This is only useful in C++. 2493 if (!E->isTypeDependent() && E->isRValue()) 2494 return E; 2495 2496 // Everything else: we simply don't reason about them. 2497 return NULL; 2498 } 2499} while (true); 2500} 2501 2502//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 2503 2504/// Check for comparisons of floating point operands using != and ==. 2505/// Issue a warning if these are no self-comparisons, as they are not likely 2506/// to do what the programmer intended. 2507void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 2508 bool EmitWarning = true; 2509 2510 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 2511 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 2512 2513 // Special case: check for x == x (which is OK). 2514 // Do not emit warnings for such cases. 2515 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 2516 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 2517 if (DRL->getDecl() == DRR->getDecl()) 2518 EmitWarning = false; 2519 2520 2521 // Special case: check for comparisons against literals that can be exactly 2522 // represented by APFloat. In such cases, do not emit a warning. This 2523 // is a heuristic: often comparison against such literals are used to 2524 // detect if a value in a variable has not changed. This clearly can 2525 // lead to false negatives. 2526 if (EmitWarning) { 2527 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 2528 if (FLL->isExact()) 2529 EmitWarning = false; 2530 } else 2531 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 2532 if (FLR->isExact()) 2533 EmitWarning = false; 2534 } 2535 } 2536 2537 // Check for comparisons with builtin types. 2538 if (EmitWarning) 2539 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 2540 if (CL->isBuiltinCall(Context)) 2541 EmitWarning = false; 2542 2543 if (EmitWarning) 2544 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 2545 if (CR->isBuiltinCall(Context)) 2546 EmitWarning = false; 2547 2548 // Emit the diagnostic. 2549 if (EmitWarning) 2550 Diag(Loc, diag::warn_floatingpoint_eq) 2551 << LHS->getSourceRange() << RHS->getSourceRange(); 2552} 2553 2554//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 2555//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 2556 2557namespace { 2558 2559/// Structure recording the 'active' range of an integer-valued 2560/// expression. 2561struct IntRange { 2562 /// The number of bits active in the int. 2563 unsigned Width; 2564 2565 /// True if the int is known not to have negative values. 2566 bool NonNegative; 2567 2568 IntRange(unsigned Width, bool NonNegative) 2569 : Width(Width), NonNegative(NonNegative) 2570 {} 2571 2572 /// Returns the range of the bool type. 2573 static IntRange forBoolType() { 2574 return IntRange(1, true); 2575 } 2576 2577 /// Returns the range of an opaque value of the given integral type. 2578 static IntRange forValueOfType(ASTContext &C, QualType T) { 2579 return forValueOfCanonicalType(C, 2580 T->getCanonicalTypeInternal().getTypePtr()); 2581 } 2582 2583 /// Returns the range of an opaque value of a canonical integral type. 2584 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 2585 assert(T->isCanonicalUnqualified()); 2586 2587 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2588 T = VT->getElementType().getTypePtr(); 2589 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2590 T = CT->getElementType().getTypePtr(); 2591 2592 // For enum types, use the known bit width of the enumerators. 2593 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 2594 EnumDecl *Enum = ET->getDecl(); 2595 if (!Enum->isCompleteDefinition()) 2596 return IntRange(C.getIntWidth(QualType(T, 0)), false); 2597 2598 unsigned NumPositive = Enum->getNumPositiveBits(); 2599 unsigned NumNegative = Enum->getNumNegativeBits(); 2600 2601 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 2602 } 2603 2604 const BuiltinType *BT = cast<BuiltinType>(T); 2605 assert(BT->isInteger()); 2606 2607 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2608 } 2609 2610 /// Returns the "target" range of a canonical integral type, i.e. 2611 /// the range of values expressible in the type. 2612 /// 2613 /// This matches forValueOfCanonicalType except that enums have the 2614 /// full range of their type, not the range of their enumerators. 2615 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 2616 assert(T->isCanonicalUnqualified()); 2617 2618 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2619 T = VT->getElementType().getTypePtr(); 2620 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2621 T = CT->getElementType().getTypePtr(); 2622 if (const EnumType *ET = dyn_cast<EnumType>(T)) 2623 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 2624 2625 const BuiltinType *BT = cast<BuiltinType>(T); 2626 assert(BT->isInteger()); 2627 2628 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2629 } 2630 2631 /// Returns the supremum of two ranges: i.e. their conservative merge. 2632 static IntRange join(IntRange L, IntRange R) { 2633 return IntRange(std::max(L.Width, R.Width), 2634 L.NonNegative && R.NonNegative); 2635 } 2636 2637 /// Returns the infinum of two ranges: i.e. their aggressive merge. 2638 static IntRange meet(IntRange L, IntRange R) { 2639 return IntRange(std::min(L.Width, R.Width), 2640 L.NonNegative || R.NonNegative); 2641 } 2642}; 2643 2644IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 2645 if (value.isSigned() && value.isNegative()) 2646 return IntRange(value.getMinSignedBits(), false); 2647 2648 if (value.getBitWidth() > MaxWidth) 2649 value = value.trunc(MaxWidth); 2650 2651 // isNonNegative() just checks the sign bit without considering 2652 // signedness. 2653 return IntRange(value.getActiveBits(), true); 2654} 2655 2656IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 2657 unsigned MaxWidth) { 2658 if (result.isInt()) 2659 return GetValueRange(C, result.getInt(), MaxWidth); 2660 2661 if (result.isVector()) { 2662 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 2663 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 2664 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 2665 R = IntRange::join(R, El); 2666 } 2667 return R; 2668 } 2669 2670 if (result.isComplexInt()) { 2671 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 2672 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 2673 return IntRange::join(R, I); 2674 } 2675 2676 // This can happen with lossless casts to intptr_t of "based" lvalues. 2677 // Assume it might use arbitrary bits. 2678 // FIXME: The only reason we need to pass the type in here is to get 2679 // the sign right on this one case. It would be nice if APValue 2680 // preserved this. 2681 assert(result.isLValue()); 2682 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 2683} 2684 2685/// Pseudo-evaluate the given integer expression, estimating the 2686/// range of values it might take. 2687/// 2688/// \param MaxWidth - the width to which the value will be truncated 2689IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 2690 E = E->IgnoreParens(); 2691 2692 // Try a full evaluation first. 2693 Expr::EvalResult result; 2694 if (E->Evaluate(result, C)) 2695 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 2696 2697 // I think we only want to look through implicit casts here; if the 2698 // user has an explicit widening cast, we should treat the value as 2699 // being of the new, wider type. 2700 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 2701 if (CE->getCastKind() == CK_NoOp) 2702 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 2703 2704 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 2705 2706 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 2707 2708 // Assume that non-integer casts can span the full range of the type. 2709 if (!isIntegerCast) 2710 return OutputTypeRange; 2711 2712 IntRange SubRange 2713 = GetExprRange(C, CE->getSubExpr(), 2714 std::min(MaxWidth, OutputTypeRange.Width)); 2715 2716 // Bail out if the subexpr's range is as wide as the cast type. 2717 if (SubRange.Width >= OutputTypeRange.Width) 2718 return OutputTypeRange; 2719 2720 // Otherwise, we take the smaller width, and we're non-negative if 2721 // either the output type or the subexpr is. 2722 return IntRange(SubRange.Width, 2723 SubRange.NonNegative || OutputTypeRange.NonNegative); 2724 } 2725 2726 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 2727 // If we can fold the condition, just take that operand. 2728 bool CondResult; 2729 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 2730 return GetExprRange(C, CondResult ? CO->getTrueExpr() 2731 : CO->getFalseExpr(), 2732 MaxWidth); 2733 2734 // Otherwise, conservatively merge. 2735 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 2736 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 2737 return IntRange::join(L, R); 2738 } 2739 2740 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 2741 switch (BO->getOpcode()) { 2742 2743 // Boolean-valued operations are single-bit and positive. 2744 case BO_LAnd: 2745 case BO_LOr: 2746 case BO_LT: 2747 case BO_GT: 2748 case BO_LE: 2749 case BO_GE: 2750 case BO_EQ: 2751 case BO_NE: 2752 return IntRange::forBoolType(); 2753 2754 // The type of the assignments is the type of the LHS, so the RHS 2755 // is not necessarily the same type. 2756 case BO_MulAssign: 2757 case BO_DivAssign: 2758 case BO_RemAssign: 2759 case BO_AddAssign: 2760 case BO_SubAssign: 2761 case BO_XorAssign: 2762 case BO_OrAssign: 2763 // TODO: bitfields? 2764 return IntRange::forValueOfType(C, E->getType()); 2765 2766 // Simple assignments just pass through the RHS, which will have 2767 // been coerced to the LHS type. 2768 case BO_Assign: 2769 // TODO: bitfields? 2770 return GetExprRange(C, BO->getRHS(), MaxWidth); 2771 2772 // Operations with opaque sources are black-listed. 2773 case BO_PtrMemD: 2774 case BO_PtrMemI: 2775 return IntRange::forValueOfType(C, E->getType()); 2776 2777 // Bitwise-and uses the *infinum* of the two source ranges. 2778 case BO_And: 2779 case BO_AndAssign: 2780 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 2781 GetExprRange(C, BO->getRHS(), MaxWidth)); 2782 2783 // Left shift gets black-listed based on a judgement call. 2784 case BO_Shl: 2785 // ...except that we want to treat '1 << (blah)' as logically 2786 // positive. It's an important idiom. 2787 if (IntegerLiteral *I 2788 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 2789 if (I->getValue() == 1) { 2790 IntRange R = IntRange::forValueOfType(C, E->getType()); 2791 return IntRange(R.Width, /*NonNegative*/ true); 2792 } 2793 } 2794 // fallthrough 2795 2796 case BO_ShlAssign: 2797 return IntRange::forValueOfType(C, E->getType()); 2798 2799 // Right shift by a constant can narrow its left argument. 2800 case BO_Shr: 2801 case BO_ShrAssign: { 2802 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2803 2804 // If the shift amount is a positive constant, drop the width by 2805 // that much. 2806 llvm::APSInt shift; 2807 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 2808 shift.isNonNegative()) { 2809 unsigned zext = shift.getZExtValue(); 2810 if (zext >= L.Width) 2811 L.Width = (L.NonNegative ? 0 : 1); 2812 else 2813 L.Width -= zext; 2814 } 2815 2816 return L; 2817 } 2818 2819 // Comma acts as its right operand. 2820 case BO_Comma: 2821 return GetExprRange(C, BO->getRHS(), MaxWidth); 2822 2823 // Black-list pointer subtractions. 2824 case BO_Sub: 2825 if (BO->getLHS()->getType()->isPointerType()) 2826 return IntRange::forValueOfType(C, E->getType()); 2827 break; 2828 2829 // The width of a division result is mostly determined by the size 2830 // of the LHS. 2831 case BO_Div: { 2832 // Don't 'pre-truncate' the operands. 2833 unsigned opWidth = C.getIntWidth(E->getType()); 2834 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 2835 2836 // If the divisor is constant, use that. 2837 llvm::APSInt divisor; 2838 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 2839 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 2840 if (log2 >= L.Width) 2841 L.Width = (L.NonNegative ? 0 : 1); 2842 else 2843 L.Width = std::min(L.Width - log2, MaxWidth); 2844 return L; 2845 } 2846 2847 // Otherwise, just use the LHS's width. 2848 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 2849 return IntRange(L.Width, L.NonNegative && R.NonNegative); 2850 } 2851 2852 // The result of a remainder can't be larger than the result of 2853 // either side. 2854 case BO_Rem: { 2855 // Don't 'pre-truncate' the operands. 2856 unsigned opWidth = C.getIntWidth(E->getType()); 2857 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 2858 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 2859 2860 IntRange meet = IntRange::meet(L, R); 2861 meet.Width = std::min(meet.Width, MaxWidth); 2862 return meet; 2863 } 2864 2865 // The default behavior is okay for these. 2866 case BO_Mul: 2867 case BO_Add: 2868 case BO_Xor: 2869 case BO_Or: 2870 break; 2871 } 2872 2873 // The default case is to treat the operation as if it were closed 2874 // on the narrowest type that encompasses both operands. 2875 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2876 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 2877 return IntRange::join(L, R); 2878 } 2879 2880 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 2881 switch (UO->getOpcode()) { 2882 // Boolean-valued operations are white-listed. 2883 case UO_LNot: 2884 return IntRange::forBoolType(); 2885 2886 // Operations with opaque sources are black-listed. 2887 case UO_Deref: 2888 case UO_AddrOf: // should be impossible 2889 return IntRange::forValueOfType(C, E->getType()); 2890 2891 default: 2892 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 2893 } 2894 } 2895 2896 if (dyn_cast<OffsetOfExpr>(E)) { 2897 IntRange::forValueOfType(C, E->getType()); 2898 } 2899 2900 FieldDecl *BitField = E->getBitField(); 2901 if (BitField) { 2902 llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C); 2903 unsigned BitWidth = BitWidthAP.getZExtValue(); 2904 2905 return IntRange(BitWidth, 2906 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 2907 } 2908 2909 return IntRange::forValueOfType(C, E->getType()); 2910} 2911 2912IntRange GetExprRange(ASTContext &C, Expr *E) { 2913 return GetExprRange(C, E, C.getIntWidth(E->getType())); 2914} 2915 2916/// Checks whether the given value, which currently has the given 2917/// source semantics, has the same value when coerced through the 2918/// target semantics. 2919bool IsSameFloatAfterCast(const llvm::APFloat &value, 2920 const llvm::fltSemantics &Src, 2921 const llvm::fltSemantics &Tgt) { 2922 llvm::APFloat truncated = value; 2923 2924 bool ignored; 2925 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 2926 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 2927 2928 return truncated.bitwiseIsEqual(value); 2929} 2930 2931/// Checks whether the given value, which currently has the given 2932/// source semantics, has the same value when coerced through the 2933/// target semantics. 2934/// 2935/// The value might be a vector of floats (or a complex number). 2936bool IsSameFloatAfterCast(const APValue &value, 2937 const llvm::fltSemantics &Src, 2938 const llvm::fltSemantics &Tgt) { 2939 if (value.isFloat()) 2940 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 2941 2942 if (value.isVector()) { 2943 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 2944 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 2945 return false; 2946 return true; 2947 } 2948 2949 assert(value.isComplexFloat()); 2950 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 2951 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 2952} 2953 2954void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 2955 2956static bool IsZero(Sema &S, Expr *E) { 2957 // Suppress cases where we are comparing against an enum constant. 2958 if (const DeclRefExpr *DR = 2959 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 2960 if (isa<EnumConstantDecl>(DR->getDecl())) 2961 return false; 2962 2963 // Suppress cases where the '0' value is expanded from a macro. 2964 if (E->getLocStart().isMacroID()) 2965 return false; 2966 2967 llvm::APSInt Value; 2968 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 2969} 2970 2971static bool HasEnumType(Expr *E) { 2972 // Strip off implicit integral promotions. 2973 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 2974 if (ICE->getCastKind() != CK_IntegralCast && 2975 ICE->getCastKind() != CK_NoOp) 2976 break; 2977 E = ICE->getSubExpr(); 2978 } 2979 2980 return E->getType()->isEnumeralType(); 2981} 2982 2983void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 2984 BinaryOperatorKind op = E->getOpcode(); 2985 if (E->isValueDependent()) 2986 return; 2987 2988 if (op == BO_LT && IsZero(S, E->getRHS())) { 2989 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2990 << "< 0" << "false" << HasEnumType(E->getLHS()) 2991 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2992 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 2993 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2994 << ">= 0" << "true" << HasEnumType(E->getLHS()) 2995 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2996 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 2997 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2998 << "0 >" << "false" << HasEnumType(E->getRHS()) 2999 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3000 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 3001 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3002 << "0 <=" << "true" << HasEnumType(E->getRHS()) 3003 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3004 } 3005} 3006 3007/// Analyze the operands of the given comparison. Implements the 3008/// fallback case from AnalyzeComparison. 3009void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 3010 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3011 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3012} 3013 3014/// \brief Implements -Wsign-compare. 3015/// 3016/// \param E the binary operator to check for warnings 3017void AnalyzeComparison(Sema &S, BinaryOperator *E) { 3018 // The type the comparison is being performed in. 3019 QualType T = E->getLHS()->getType(); 3020 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 3021 && "comparison with mismatched types"); 3022 3023 // We don't do anything special if this isn't an unsigned integral 3024 // comparison: we're only interested in integral comparisons, and 3025 // signed comparisons only happen in cases we don't care to warn about. 3026 // 3027 // We also don't care about value-dependent expressions or expressions 3028 // whose result is a constant. 3029 if (!T->hasUnsignedIntegerRepresentation() 3030 || E->isValueDependent() || E->isIntegerConstantExpr(S.Context)) 3031 return AnalyzeImpConvsInComparison(S, E); 3032 3033 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 3034 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 3035 3036 // Check to see if one of the (unmodified) operands is of different 3037 // signedness. 3038 Expr *signedOperand, *unsignedOperand; 3039 if (LHS->getType()->hasSignedIntegerRepresentation()) { 3040 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 3041 "unsigned comparison between two signed integer expressions?"); 3042 signedOperand = LHS; 3043 unsignedOperand = RHS; 3044 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 3045 signedOperand = RHS; 3046 unsignedOperand = LHS; 3047 } else { 3048 CheckTrivialUnsignedComparison(S, E); 3049 return AnalyzeImpConvsInComparison(S, E); 3050 } 3051 3052 // Otherwise, calculate the effective range of the signed operand. 3053 IntRange signedRange = GetExprRange(S.Context, signedOperand); 3054 3055 // Go ahead and analyze implicit conversions in the operands. Note 3056 // that we skip the implicit conversions on both sides. 3057 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 3058 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 3059 3060 // If the signed range is non-negative, -Wsign-compare won't fire, 3061 // but we should still check for comparisons which are always true 3062 // or false. 3063 if (signedRange.NonNegative) 3064 return CheckTrivialUnsignedComparison(S, E); 3065 3066 // For (in)equality comparisons, if the unsigned operand is a 3067 // constant which cannot collide with a overflowed signed operand, 3068 // then reinterpreting the signed operand as unsigned will not 3069 // change the result of the comparison. 3070 if (E->isEqualityOp()) { 3071 unsigned comparisonWidth = S.Context.getIntWidth(T); 3072 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 3073 3074 // We should never be unable to prove that the unsigned operand is 3075 // non-negative. 3076 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 3077 3078 if (unsignedRange.Width < comparisonWidth) 3079 return; 3080 } 3081 3082 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison) 3083 << LHS->getType() << RHS->getType() 3084 << LHS->getSourceRange() << RHS->getSourceRange(); 3085} 3086 3087/// Analyzes an attempt to assign the given value to a bitfield. 3088/// 3089/// Returns true if there was something fishy about the attempt. 3090bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 3091 SourceLocation InitLoc) { 3092 assert(Bitfield->isBitField()); 3093 if (Bitfield->isInvalidDecl()) 3094 return false; 3095 3096 // White-list bool bitfields. 3097 if (Bitfield->getType()->isBooleanType()) 3098 return false; 3099 3100 // Ignore value- or type-dependent expressions. 3101 if (Bitfield->getBitWidth()->isValueDependent() || 3102 Bitfield->getBitWidth()->isTypeDependent() || 3103 Init->isValueDependent() || 3104 Init->isTypeDependent()) 3105 return false; 3106 3107 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 3108 3109 llvm::APSInt Width(32); 3110 Expr::EvalResult InitValue; 3111 if (!Bitfield->getBitWidth()->isIntegerConstantExpr(Width, S.Context) || 3112 !OriginalInit->Evaluate(InitValue, S.Context) || 3113 !InitValue.Val.isInt()) 3114 return false; 3115 3116 const llvm::APSInt &Value = InitValue.Val.getInt(); 3117 unsigned OriginalWidth = Value.getBitWidth(); 3118 unsigned FieldWidth = Width.getZExtValue(); 3119 3120 if (OriginalWidth <= FieldWidth) 3121 return false; 3122 3123 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 3124 3125 // It's fairly common to write values into signed bitfields 3126 // that, if sign-extended, would end up becoming a different 3127 // value. We don't want to warn about that. 3128 if (Value.isSigned() && Value.isNegative()) 3129 TruncatedValue = TruncatedValue.sext(OriginalWidth); 3130 else 3131 TruncatedValue = TruncatedValue.zext(OriginalWidth); 3132 3133 if (Value == TruncatedValue) 3134 return false; 3135 3136 std::string PrettyValue = Value.toString(10); 3137 std::string PrettyTrunc = TruncatedValue.toString(10); 3138 3139 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 3140 << PrettyValue << PrettyTrunc << OriginalInit->getType() 3141 << Init->getSourceRange(); 3142 3143 return true; 3144} 3145 3146/// Analyze the given simple or compound assignment for warning-worthy 3147/// operations. 3148void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 3149 // Just recurse on the LHS. 3150 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3151 3152 // We want to recurse on the RHS as normal unless we're assigning to 3153 // a bitfield. 3154 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 3155 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 3156 E->getOperatorLoc())) { 3157 // Recurse, ignoring any implicit conversions on the RHS. 3158 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 3159 E->getOperatorLoc()); 3160 } 3161 } 3162 3163 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3164} 3165 3166/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3167void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 3168 SourceLocation CContext, unsigned diag) { 3169 S.Diag(E->getExprLoc(), diag) 3170 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 3171} 3172 3173/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3174void DiagnoseImpCast(Sema &S, Expr *E, QualType T, SourceLocation CContext, 3175 unsigned diag) { 3176 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag); 3177} 3178 3179/// Diagnose an implicit cast from a literal expression. Also attemps to supply 3180/// fixit hints when the cast wouldn't lose information to simply write the 3181/// expression with the expected type. 3182void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 3183 SourceLocation CContext) { 3184 // Emit the primary warning first, then try to emit a fixit hint note if 3185 // reasonable. 3186 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 3187 << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext); 3188 3189 const llvm::APFloat &Value = FL->getValue(); 3190 3191 // Don't attempt to fix PPC double double literals. 3192 if (&Value.getSemantics() == &llvm::APFloat::PPCDoubleDouble) 3193 return; 3194 3195 // Try to convert this exactly to an integer. 3196 bool isExact = false; 3197 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 3198 T->hasUnsignedIntegerRepresentation()); 3199 if (Value.convertToInteger(IntegerValue, 3200 llvm::APFloat::rmTowardZero, &isExact) 3201 != llvm::APFloat::opOK || !isExact) 3202 return; 3203 3204 std::string LiteralValue = IntegerValue.toString(10); 3205 S.Diag(FL->getExprLoc(), diag::note_fix_integral_float_as_integer) 3206 << FixItHint::CreateReplacement(FL->getSourceRange(), LiteralValue); 3207} 3208 3209std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 3210 if (!Range.Width) return "0"; 3211 3212 llvm::APSInt ValueInRange = Value; 3213 ValueInRange.setIsSigned(!Range.NonNegative); 3214 ValueInRange = ValueInRange.trunc(Range.Width); 3215 return ValueInRange.toString(10); 3216} 3217 3218static bool isFromSystemMacro(Sema &S, SourceLocation loc) { 3219 SourceManager &smgr = S.Context.getSourceManager(); 3220 return loc.isMacroID() && smgr.isInSystemHeader(smgr.getSpellingLoc(loc)); 3221} 3222 3223void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 3224 SourceLocation CC, bool *ICContext = 0) { 3225 if (E->isTypeDependent() || E->isValueDependent()) return; 3226 3227 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 3228 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 3229 if (Source == Target) return; 3230 if (Target->isDependentType()) return; 3231 3232 // If the conversion context location is invalid don't complain. We also 3233 // don't want to emit a warning if the issue occurs from the expansion of 3234 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 3235 // delay this check as long as possible. Once we detect we are in that 3236 // scenario, we just return. 3237 if (CC.isInvalid()) 3238 return; 3239 3240 // Diagnose implicit casts to bool. 3241 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 3242 if (isa<StringLiteral>(E)) 3243 // Warn on string literal to bool. Checks for string literals in logical 3244 // expressions, for instances, assert(0 && "error here"), is prevented 3245 // by a check in AnalyzeImplicitConversions(). 3246 return DiagnoseImpCast(S, E, T, CC, 3247 diag::warn_impcast_string_literal_to_bool); 3248 return; // Other casts to bool are not checked. 3249 } 3250 3251 // Strip vector types. 3252 if (isa<VectorType>(Source)) { 3253 if (!isa<VectorType>(Target)) { 3254 if (isFromSystemMacro(S, CC)) 3255 return; 3256 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 3257 } 3258 3259 // If the vector cast is cast between two vectors of the same size, it is 3260 // a bitcast, not a conversion. 3261 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 3262 return; 3263 3264 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 3265 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 3266 } 3267 3268 // Strip complex types. 3269 if (isa<ComplexType>(Source)) { 3270 if (!isa<ComplexType>(Target)) { 3271 if (isFromSystemMacro(S, CC)) 3272 return; 3273 3274 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 3275 } 3276 3277 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 3278 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 3279 } 3280 3281 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 3282 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 3283 3284 // If the source is floating point... 3285 if (SourceBT && SourceBT->isFloatingPoint()) { 3286 // ...and the target is floating point... 3287 if (TargetBT && TargetBT->isFloatingPoint()) { 3288 // ...then warn if we're dropping FP rank. 3289 3290 // Builtin FP kinds are ordered by increasing FP rank. 3291 if (SourceBT->getKind() > TargetBT->getKind()) { 3292 // Don't warn about float constants that are precisely 3293 // representable in the target type. 3294 Expr::EvalResult result; 3295 if (E->Evaluate(result, S.Context)) { 3296 // Value might be a float, a float vector, or a float complex. 3297 if (IsSameFloatAfterCast(result.Val, 3298 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 3299 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 3300 return; 3301 } 3302 3303 if (isFromSystemMacro(S, CC)) 3304 return; 3305 3306 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 3307 } 3308 return; 3309 } 3310 3311 // If the target is integral, always warn. 3312 if ((TargetBT && TargetBT->isInteger())) { 3313 if (isFromSystemMacro(S, CC)) 3314 return; 3315 3316 Expr *InnerE = E->IgnoreParenImpCasts(); 3317 // We also want to warn on, e.g., "int i = -1.234" 3318 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 3319 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 3320 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 3321 3322 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 3323 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 3324 } else { 3325 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 3326 } 3327 } 3328 3329 return; 3330 } 3331 3332 if (!Source->isIntegerType() || !Target->isIntegerType()) 3333 return; 3334 3335 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 3336 == Expr::NPCK_GNUNull) && Target->isIntegerType()) { 3337 S.Diag(E->getExprLoc(), diag::warn_impcast_null_pointer_to_integer) 3338 << E->getSourceRange() << clang::SourceRange(CC); 3339 return; 3340 } 3341 3342 IntRange SourceRange = GetExprRange(S.Context, E); 3343 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 3344 3345 if (SourceRange.Width > TargetRange.Width) { 3346 // If the source is a constant, use a default-on diagnostic. 3347 // TODO: this should happen for bitfield stores, too. 3348 llvm::APSInt Value(32); 3349 if (E->isIntegerConstantExpr(Value, S.Context)) { 3350 if (isFromSystemMacro(S, CC)) 3351 return; 3352 3353 std::string PrettySourceValue = Value.toString(10); 3354 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 3355 3356 S.Diag(E->getExprLoc(), diag::warn_impcast_integer_precision_constant) 3357 << PrettySourceValue << PrettyTargetValue 3358 << E->getType() << T << E->getSourceRange() << clang::SourceRange(CC); 3359 return; 3360 } 3361 3362 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 3363 if (isFromSystemMacro(S, CC)) 3364 return; 3365 3366 if (SourceRange.Width == 64 && TargetRange.Width == 32) 3367 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32); 3368 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 3369 } 3370 3371 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 3372 (!TargetRange.NonNegative && SourceRange.NonNegative && 3373 SourceRange.Width == TargetRange.Width)) { 3374 3375 if (isFromSystemMacro(S, CC)) 3376 return; 3377 3378 unsigned DiagID = diag::warn_impcast_integer_sign; 3379 3380 // Traditionally, gcc has warned about this under -Wsign-compare. 3381 // We also want to warn about it in -Wconversion. 3382 // So if -Wconversion is off, use a completely identical diagnostic 3383 // in the sign-compare group. 3384 // The conditional-checking code will 3385 if (ICContext) { 3386 DiagID = diag::warn_impcast_integer_sign_conditional; 3387 *ICContext = true; 3388 } 3389 3390 return DiagnoseImpCast(S, E, T, CC, DiagID); 3391 } 3392 3393 // Diagnose conversions between different enumeration types. 3394 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 3395 // type, to give us better diagnostics. 3396 QualType SourceType = E->getType(); 3397 if (!S.getLangOptions().CPlusPlus) { 3398 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 3399 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 3400 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 3401 SourceType = S.Context.getTypeDeclType(Enum); 3402 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 3403 } 3404 } 3405 3406 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 3407 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 3408 if ((SourceEnum->getDecl()->getIdentifier() || 3409 SourceEnum->getDecl()->getTypedefNameForAnonDecl()) && 3410 (TargetEnum->getDecl()->getIdentifier() || 3411 TargetEnum->getDecl()->getTypedefNameForAnonDecl()) && 3412 SourceEnum != TargetEnum) { 3413 if (isFromSystemMacro(S, CC)) 3414 return; 3415 3416 return DiagnoseImpCast(S, E, SourceType, T, CC, 3417 diag::warn_impcast_different_enum_types); 3418 } 3419 3420 return; 3421} 3422 3423void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 3424 3425void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 3426 SourceLocation CC, bool &ICContext) { 3427 E = E->IgnoreParenImpCasts(); 3428 3429 if (isa<ConditionalOperator>(E)) 3430 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 3431 3432 AnalyzeImplicitConversions(S, E, CC); 3433 if (E->getType() != T) 3434 return CheckImplicitConversion(S, E, T, CC, &ICContext); 3435 return; 3436} 3437 3438void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 3439 SourceLocation CC = E->getQuestionLoc(); 3440 3441 AnalyzeImplicitConversions(S, E->getCond(), CC); 3442 3443 bool Suspicious = false; 3444 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 3445 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 3446 3447 // If -Wconversion would have warned about either of the candidates 3448 // for a signedness conversion to the context type... 3449 if (!Suspicious) return; 3450 3451 // ...but it's currently ignored... 3452 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 3453 CC)) 3454 return; 3455 3456 // ...then check whether it would have warned about either of the 3457 // candidates for a signedness conversion to the condition type. 3458 if (E->getType() == T) return; 3459 3460 Suspicious = false; 3461 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 3462 E->getType(), CC, &Suspicious); 3463 if (!Suspicious) 3464 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 3465 E->getType(), CC, &Suspicious); 3466} 3467 3468/// AnalyzeImplicitConversions - Find and report any interesting 3469/// implicit conversions in the given expression. There are a couple 3470/// of competing diagnostics here, -Wconversion and -Wsign-compare. 3471void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 3472 QualType T = OrigE->getType(); 3473 Expr *E = OrigE->IgnoreParenImpCasts(); 3474 3475 // For conditional operators, we analyze the arguments as if they 3476 // were being fed directly into the output. 3477 if (isa<ConditionalOperator>(E)) { 3478 ConditionalOperator *CO = cast<ConditionalOperator>(E); 3479 CheckConditionalOperator(S, CO, T); 3480 return; 3481 } 3482 3483 // Go ahead and check any implicit conversions we might have skipped. 3484 // The non-canonical typecheck is just an optimization; 3485 // CheckImplicitConversion will filter out dead implicit conversions. 3486 if (E->getType() != T) 3487 CheckImplicitConversion(S, E, T, CC); 3488 3489 // Now continue drilling into this expression. 3490 3491 // Skip past explicit casts. 3492 if (isa<ExplicitCastExpr>(E)) { 3493 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 3494 return AnalyzeImplicitConversions(S, E, CC); 3495 } 3496 3497 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 3498 // Do a somewhat different check with comparison operators. 3499 if (BO->isComparisonOp()) 3500 return AnalyzeComparison(S, BO); 3501 3502 // And with assignments and compound assignments. 3503 if (BO->isAssignmentOp()) 3504 return AnalyzeAssignment(S, BO); 3505 } 3506 3507 // These break the otherwise-useful invariant below. Fortunately, 3508 // we don't really need to recurse into them, because any internal 3509 // expressions should have been analyzed already when they were 3510 // built into statements. 3511 if (isa<StmtExpr>(E)) return; 3512 3513 // Don't descend into unevaluated contexts. 3514 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 3515 3516 // Now just recurse over the expression's children. 3517 CC = E->getExprLoc(); 3518 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 3519 bool IsLogicalOperator = BO && BO->isLogicalOp(); 3520 for (Stmt::child_range I = E->children(); I; ++I) { 3521 Expr *ChildExpr = cast<Expr>(*I); 3522 if (IsLogicalOperator && 3523 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 3524 // Ignore checking string literals that are in logical operators. 3525 continue; 3526 AnalyzeImplicitConversions(S, ChildExpr, CC); 3527 } 3528} 3529 3530} // end anonymous namespace 3531 3532/// Diagnoses "dangerous" implicit conversions within the given 3533/// expression (which is a full expression). Implements -Wconversion 3534/// and -Wsign-compare. 3535/// 3536/// \param CC the "context" location of the implicit conversion, i.e. 3537/// the most location of the syntactic entity requiring the implicit 3538/// conversion 3539void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 3540 // Don't diagnose in unevaluated contexts. 3541 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 3542 return; 3543 3544 // Don't diagnose for value- or type-dependent expressions. 3545 if (E->isTypeDependent() || E->isValueDependent()) 3546 return; 3547 3548 // Check for array bounds violations in cases where the check isn't triggered 3549 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 3550 // ArraySubscriptExpr is on the RHS of a variable initialization. 3551 CheckArrayAccess(E); 3552 3553 // This is not the right CC for (e.g.) a variable initialization. 3554 AnalyzeImplicitConversions(*this, E, CC); 3555} 3556 3557void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 3558 FieldDecl *BitField, 3559 Expr *Init) { 3560 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 3561} 3562 3563/// CheckParmsForFunctionDef - Check that the parameters of the given 3564/// function are appropriate for the definition of a function. This 3565/// takes care of any checks that cannot be performed on the 3566/// declaration itself, e.g., that the types of each of the function 3567/// parameters are complete. 3568bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 3569 bool CheckParameterNames) { 3570 bool HasInvalidParm = false; 3571 for (; P != PEnd; ++P) { 3572 ParmVarDecl *Param = *P; 3573 3574 // C99 6.7.5.3p4: the parameters in a parameter type list in a 3575 // function declarator that is part of a function definition of 3576 // that function shall not have incomplete type. 3577 // 3578 // This is also C++ [dcl.fct]p6. 3579 if (!Param->isInvalidDecl() && 3580 RequireCompleteType(Param->getLocation(), Param->getType(), 3581 diag::err_typecheck_decl_incomplete_type)) { 3582 Param->setInvalidDecl(); 3583 HasInvalidParm = true; 3584 } 3585 3586 // C99 6.9.1p5: If the declarator includes a parameter type list, the 3587 // declaration of each parameter shall include an identifier. 3588 if (CheckParameterNames && 3589 Param->getIdentifier() == 0 && 3590 !Param->isImplicit() && 3591 !getLangOptions().CPlusPlus) 3592 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 3593 3594 // C99 6.7.5.3p12: 3595 // If the function declarator is not part of a definition of that 3596 // function, parameters may have incomplete type and may use the [*] 3597 // notation in their sequences of declarator specifiers to specify 3598 // variable length array types. 3599 QualType PType = Param->getOriginalType(); 3600 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 3601 if (AT->getSizeModifier() == ArrayType::Star) { 3602 // FIXME: This diagnosic should point the the '[*]' if source-location 3603 // information is added for it. 3604 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 3605 } 3606 } 3607 } 3608 3609 return HasInvalidParm; 3610} 3611 3612/// CheckCastAlign - Implements -Wcast-align, which warns when a 3613/// pointer cast increases the alignment requirements. 3614void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 3615 // This is actually a lot of work to potentially be doing on every 3616 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 3617 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 3618 TRange.getBegin()) 3619 == DiagnosticsEngine::Ignored) 3620 return; 3621 3622 // Ignore dependent types. 3623 if (T->isDependentType() || Op->getType()->isDependentType()) 3624 return; 3625 3626 // Require that the destination be a pointer type. 3627 const PointerType *DestPtr = T->getAs<PointerType>(); 3628 if (!DestPtr) return; 3629 3630 // If the destination has alignment 1, we're done. 3631 QualType DestPointee = DestPtr->getPointeeType(); 3632 if (DestPointee->isIncompleteType()) return; 3633 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 3634 if (DestAlign.isOne()) return; 3635 3636 // Require that the source be a pointer type. 3637 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 3638 if (!SrcPtr) return; 3639 QualType SrcPointee = SrcPtr->getPointeeType(); 3640 3641 // Whitelist casts from cv void*. We already implicitly 3642 // whitelisted casts to cv void*, since they have alignment 1. 3643 // Also whitelist casts involving incomplete types, which implicitly 3644 // includes 'void'. 3645 if (SrcPointee->isIncompleteType()) return; 3646 3647 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 3648 if (SrcAlign >= DestAlign) return; 3649 3650 Diag(TRange.getBegin(), diag::warn_cast_align) 3651 << Op->getType() << T 3652 << static_cast<unsigned>(SrcAlign.getQuantity()) 3653 << static_cast<unsigned>(DestAlign.getQuantity()) 3654 << TRange << Op->getSourceRange(); 3655} 3656 3657static const Type* getElementType(const Expr *BaseExpr) { 3658 const Type* EltType = BaseExpr->getType().getTypePtr(); 3659 if (EltType->isAnyPointerType()) 3660 return EltType->getPointeeType().getTypePtr(); 3661 else if (EltType->isArrayType()) 3662 return EltType->getBaseElementTypeUnsafe(); 3663 return EltType; 3664} 3665 3666/// \brief Check whether this array fits the idiom of a size-one tail padded 3667/// array member of a struct. 3668/// 3669/// We avoid emitting out-of-bounds access warnings for such arrays as they are 3670/// commonly used to emulate flexible arrays in C89 code. 3671static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 3672 const NamedDecl *ND) { 3673 if (Size != 1 || !ND) return false; 3674 3675 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 3676 if (!FD) return false; 3677 3678 // Don't consider sizes resulting from macro expansions or template argument 3679 // substitution to form C89 tail-padded arrays. 3680 ConstantArrayTypeLoc TL = 3681 cast<ConstantArrayTypeLoc>(FD->getTypeSourceInfo()->getTypeLoc()); 3682 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(TL.getSizeExpr()); 3683 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 3684 return false; 3685 3686 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 3687 if (!RD || !RD->isStruct()) 3688 return false; 3689 3690 // See if this is the last field decl in the record. 3691 const Decl *D = FD; 3692 while ((D = D->getNextDeclInContext())) 3693 if (isa<FieldDecl>(D)) 3694 return false; 3695 return true; 3696} 3697 3698void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 3699 bool isSubscript, bool AllowOnePastEnd) { 3700 const Type* EffectiveType = getElementType(BaseExpr); 3701 BaseExpr = BaseExpr->IgnoreParenCasts(); 3702 IndexExpr = IndexExpr->IgnoreParenCasts(); 3703 3704 const ConstantArrayType *ArrayTy = 3705 Context.getAsConstantArrayType(BaseExpr->getType()); 3706 if (!ArrayTy) 3707 return; 3708 3709 if (IndexExpr->isValueDependent()) 3710 return; 3711 llvm::APSInt index; 3712 if (!IndexExpr->isIntegerConstantExpr(index, Context)) 3713 return; 3714 3715 const NamedDecl *ND = NULL; 3716 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 3717 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 3718 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 3719 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 3720 3721 if (index.isUnsigned() || !index.isNegative()) { 3722 llvm::APInt size = ArrayTy->getSize(); 3723 if (!size.isStrictlyPositive()) 3724 return; 3725 3726 const Type* BaseType = getElementType(BaseExpr); 3727 if (BaseType != EffectiveType) { 3728 // Make sure we're comparing apples to apples when comparing index to size 3729 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 3730 uint64_t array_typesize = Context.getTypeSize(BaseType); 3731 // Handle ptrarith_typesize being zero, such as when casting to void* 3732 if (!ptrarith_typesize) ptrarith_typesize = 1; 3733 if (ptrarith_typesize != array_typesize) { 3734 // There's a cast to a different size type involved 3735 uint64_t ratio = array_typesize / ptrarith_typesize; 3736 // TODO: Be smarter about handling cases where array_typesize is not a 3737 // multiple of ptrarith_typesize 3738 if (ptrarith_typesize * ratio == array_typesize) 3739 size *= llvm::APInt(size.getBitWidth(), ratio); 3740 } 3741 } 3742 3743 if (size.getBitWidth() > index.getBitWidth()) 3744 index = index.sext(size.getBitWidth()); 3745 else if (size.getBitWidth() < index.getBitWidth()) 3746 size = size.sext(index.getBitWidth()); 3747 3748 // For array subscripting the index must be less than size, but for pointer 3749 // arithmetic also allow the index (offset) to be equal to size since 3750 // computing the next address after the end of the array is legal and 3751 // commonly done e.g. in C++ iterators and range-based for loops. 3752 if (AllowOnePastEnd ? index.sle(size) : index.slt(size)) 3753 return; 3754 3755 // Also don't warn for arrays of size 1 which are members of some 3756 // structure. These are often used to approximate flexible arrays in C89 3757 // code. 3758 if (IsTailPaddedMemberArray(*this, size, ND)) 3759 return; 3760 3761 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 3762 if (isSubscript) 3763 DiagID = diag::warn_array_index_exceeds_bounds; 3764 3765 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 3766 PDiag(DiagID) << index.toString(10, true) 3767 << size.toString(10, true) 3768 << (unsigned)size.getLimitedValue(~0U) 3769 << IndexExpr->getSourceRange()); 3770 } else { 3771 unsigned DiagID = diag::warn_array_index_precedes_bounds; 3772 if (!isSubscript) { 3773 DiagID = diag::warn_ptr_arith_precedes_bounds; 3774 if (index.isNegative()) index = -index; 3775 } 3776 3777 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 3778 PDiag(DiagID) << index.toString(10, true) 3779 << IndexExpr->getSourceRange()); 3780 } 3781 3782 if (ND) 3783 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 3784 PDiag(diag::note_array_index_out_of_bounds) 3785 << ND->getDeclName()); 3786} 3787 3788void Sema::CheckArrayAccess(const Expr *expr) { 3789 int AllowOnePastEnd = 0; 3790 while (expr) { 3791 expr = expr->IgnoreParenImpCasts(); 3792 switch (expr->getStmtClass()) { 3793 case Stmt::ArraySubscriptExprClass: { 3794 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 3795 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), true, 3796 AllowOnePastEnd > 0); 3797 return; 3798 } 3799 case Stmt::UnaryOperatorClass: { 3800 // Only unwrap the * and & unary operators 3801 const UnaryOperator *UO = cast<UnaryOperator>(expr); 3802 expr = UO->getSubExpr(); 3803 switch (UO->getOpcode()) { 3804 case UO_AddrOf: 3805 AllowOnePastEnd++; 3806 break; 3807 case UO_Deref: 3808 AllowOnePastEnd--; 3809 break; 3810 default: 3811 return; 3812 } 3813 break; 3814 } 3815 case Stmt::ConditionalOperatorClass: { 3816 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 3817 if (const Expr *lhs = cond->getLHS()) 3818 CheckArrayAccess(lhs); 3819 if (const Expr *rhs = cond->getRHS()) 3820 CheckArrayAccess(rhs); 3821 return; 3822 } 3823 default: 3824 return; 3825 } 3826 } 3827} 3828 3829//===--- CHECK: Objective-C retain cycles ----------------------------------// 3830 3831namespace { 3832 struct RetainCycleOwner { 3833 RetainCycleOwner() : Variable(0), Indirect(false) {} 3834 VarDecl *Variable; 3835 SourceRange Range; 3836 SourceLocation Loc; 3837 bool Indirect; 3838 3839 void setLocsFrom(Expr *e) { 3840 Loc = e->getExprLoc(); 3841 Range = e->getSourceRange(); 3842 } 3843 }; 3844} 3845 3846/// Consider whether capturing the given variable can possibly lead to 3847/// a retain cycle. 3848static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 3849 // In ARC, it's captured strongly iff the variable has __strong 3850 // lifetime. In MRR, it's captured strongly if the variable is 3851 // __block and has an appropriate type. 3852 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 3853 return false; 3854 3855 owner.Variable = var; 3856 owner.setLocsFrom(ref); 3857 return true; 3858} 3859 3860static bool findRetainCycleOwner(Expr *e, RetainCycleOwner &owner) { 3861 while (true) { 3862 e = e->IgnoreParens(); 3863 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 3864 switch (cast->getCastKind()) { 3865 case CK_BitCast: 3866 case CK_LValueBitCast: 3867 case CK_LValueToRValue: 3868 case CK_ARCReclaimReturnedObject: 3869 e = cast->getSubExpr(); 3870 continue; 3871 3872 case CK_GetObjCProperty: { 3873 // Bail out if this isn't a strong explicit property. 3874 const ObjCPropertyRefExpr *pre = cast->getSubExpr()->getObjCProperty(); 3875 if (pre->isImplicitProperty()) return false; 3876 ObjCPropertyDecl *property = pre->getExplicitProperty(); 3877 if (!property->isRetaining() && 3878 !(property->getPropertyIvarDecl() && 3879 property->getPropertyIvarDecl()->getType() 3880 .getObjCLifetime() == Qualifiers::OCL_Strong)) 3881 return false; 3882 3883 owner.Indirect = true; 3884 e = const_cast<Expr*>(pre->getBase()); 3885 continue; 3886 } 3887 3888 default: 3889 return false; 3890 } 3891 } 3892 3893 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 3894 ObjCIvarDecl *ivar = ref->getDecl(); 3895 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 3896 return false; 3897 3898 // Try to find a retain cycle in the base. 3899 if (!findRetainCycleOwner(ref->getBase(), owner)) 3900 return false; 3901 3902 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 3903 owner.Indirect = true; 3904 return true; 3905 } 3906 3907 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 3908 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 3909 if (!var) return false; 3910 return considerVariable(var, ref, owner); 3911 } 3912 3913 if (BlockDeclRefExpr *ref = dyn_cast<BlockDeclRefExpr>(e)) { 3914 owner.Variable = ref->getDecl(); 3915 owner.setLocsFrom(ref); 3916 return true; 3917 } 3918 3919 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 3920 if (member->isArrow()) return false; 3921 3922 // Don't count this as an indirect ownership. 3923 e = member->getBase(); 3924 continue; 3925 } 3926 3927 // Array ivars? 3928 3929 return false; 3930 } 3931} 3932 3933namespace { 3934 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 3935 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 3936 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 3937 Variable(variable), Capturer(0) {} 3938 3939 VarDecl *Variable; 3940 Expr *Capturer; 3941 3942 void VisitDeclRefExpr(DeclRefExpr *ref) { 3943 if (ref->getDecl() == Variable && !Capturer) 3944 Capturer = ref; 3945 } 3946 3947 void VisitBlockDeclRefExpr(BlockDeclRefExpr *ref) { 3948 if (ref->getDecl() == Variable && !Capturer) 3949 Capturer = ref; 3950 } 3951 3952 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 3953 if (Capturer) return; 3954 Visit(ref->getBase()); 3955 if (Capturer && ref->isFreeIvar()) 3956 Capturer = ref; 3957 } 3958 3959 void VisitBlockExpr(BlockExpr *block) { 3960 // Look inside nested blocks 3961 if (block->getBlockDecl()->capturesVariable(Variable)) 3962 Visit(block->getBlockDecl()->getBody()); 3963 } 3964 }; 3965} 3966 3967/// Check whether the given argument is a block which captures a 3968/// variable. 3969static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 3970 assert(owner.Variable && owner.Loc.isValid()); 3971 3972 e = e->IgnoreParenCasts(); 3973 BlockExpr *block = dyn_cast<BlockExpr>(e); 3974 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 3975 return 0; 3976 3977 FindCaptureVisitor visitor(S.Context, owner.Variable); 3978 visitor.Visit(block->getBlockDecl()->getBody()); 3979 return visitor.Capturer; 3980} 3981 3982static void diagnoseRetainCycle(Sema &S, Expr *capturer, 3983 RetainCycleOwner &owner) { 3984 assert(capturer); 3985 assert(owner.Variable && owner.Loc.isValid()); 3986 3987 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 3988 << owner.Variable << capturer->getSourceRange(); 3989 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 3990 << owner.Indirect << owner.Range; 3991} 3992 3993/// Check for a keyword selector that starts with the word 'add' or 3994/// 'set'. 3995static bool isSetterLikeSelector(Selector sel) { 3996 if (sel.isUnarySelector()) return false; 3997 3998 StringRef str = sel.getNameForSlot(0); 3999 while (!str.empty() && str.front() == '_') str = str.substr(1); 4000 if (str.startswith("set") || str.startswith("add")) 4001 str = str.substr(3); 4002 else 4003 return false; 4004 4005 if (str.empty()) return true; 4006 return !islower(str.front()); 4007} 4008 4009/// Check a message send to see if it's likely to cause a retain cycle. 4010void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 4011 // Only check instance methods whose selector looks like a setter. 4012 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 4013 return; 4014 4015 // Try to find a variable that the receiver is strongly owned by. 4016 RetainCycleOwner owner; 4017 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 4018 if (!findRetainCycleOwner(msg->getInstanceReceiver(), owner)) 4019 return; 4020 } else { 4021 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 4022 owner.Variable = getCurMethodDecl()->getSelfDecl(); 4023 owner.Loc = msg->getSuperLoc(); 4024 owner.Range = msg->getSuperLoc(); 4025 } 4026 4027 // Check whether the receiver is captured by any of the arguments. 4028 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 4029 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 4030 return diagnoseRetainCycle(*this, capturer, owner); 4031} 4032 4033/// Check a property assign to see if it's likely to cause a retain cycle. 4034void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 4035 RetainCycleOwner owner; 4036 if (!findRetainCycleOwner(receiver, owner)) 4037 return; 4038 4039 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 4040 diagnoseRetainCycle(*this, capturer, owner); 4041} 4042 4043bool Sema::checkUnsafeAssigns(SourceLocation Loc, 4044 QualType LHS, Expr *RHS) { 4045 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 4046 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 4047 return false; 4048 // strip off any implicit cast added to get to the one arc-specific 4049 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4050 if (cast->getCastKind() == CK_ARCConsumeObject) { 4051 Diag(Loc, diag::warn_arc_retained_assign) 4052 << (LT == Qualifiers::OCL_ExplicitNone) 4053 << RHS->getSourceRange(); 4054 return true; 4055 } 4056 RHS = cast->getSubExpr(); 4057 } 4058 return false; 4059} 4060 4061void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 4062 Expr *LHS, Expr *RHS) { 4063 QualType LHSType = LHS->getType(); 4064 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 4065 return; 4066 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 4067 // FIXME. Check for other life times. 4068 if (LT != Qualifiers::OCL_None) 4069 return; 4070 4071 if (ObjCPropertyRefExpr *PRE = dyn_cast<ObjCPropertyRefExpr>(LHS)) { 4072 if (PRE->isImplicitProperty()) 4073 return; 4074 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 4075 if (!PD) 4076 return; 4077 4078 unsigned Attributes = PD->getPropertyAttributes(); 4079 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) 4080 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4081 if (cast->getCastKind() == CK_ARCConsumeObject) { 4082 Diag(Loc, diag::warn_arc_retained_property_assign) 4083 << RHS->getSourceRange(); 4084 return; 4085 } 4086 RHS = cast->getSubExpr(); 4087 } 4088 } 4089} 4090