SemaChecking.cpp revision c03f2df2a2b097bf2686bf32411d5022323988b2
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/Initialization.h" 19#include "clang/Sema/ScopeInfo.h" 20#include "clang/Analysis/Analyses/FormatString.h" 21#include "clang/AST/ASTContext.h" 22#include "clang/AST/CharUnits.h" 23#include "clang/AST/DeclCXX.h" 24#include "clang/AST/DeclObjC.h" 25#include "clang/AST/ExprCXX.h" 26#include "clang/AST/ExprObjC.h" 27#include "clang/AST/EvaluatedExprVisitor.h" 28#include "clang/AST/DeclObjC.h" 29#include "clang/AST/StmtCXX.h" 30#include "clang/AST/StmtObjC.h" 31#include "clang/Lex/Preprocessor.h" 32#include "llvm/ADT/BitVector.h" 33#include "llvm/ADT/SmallString.h" 34#include "llvm/ADT/STLExtras.h" 35#include "llvm/Support/raw_ostream.h" 36#include "clang/Basic/TargetBuiltins.h" 37#include "clang/Basic/TargetInfo.h" 38#include "clang/Basic/ConvertUTF.h" 39#include <limits> 40using namespace clang; 41using namespace sema; 42 43SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 44 unsigned ByteNo) const { 45 return SL->getLocationOfByte(ByteNo, PP.getSourceManager(), 46 PP.getLangOpts(), PP.getTargetInfo()); 47} 48 49/// Checks that a call expression's argument count is the desired number. 50/// This is useful when doing custom type-checking. Returns true on error. 51static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) { 52 unsigned argCount = call->getNumArgs(); 53 if (argCount == desiredArgCount) return false; 54 55 if (argCount < desiredArgCount) 56 return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args) 57 << 0 /*function call*/ << desiredArgCount << argCount 58 << call->getSourceRange(); 59 60 // Highlight all the excess arguments. 61 SourceRange range(call->getArg(desiredArgCount)->getLocStart(), 62 call->getArg(argCount - 1)->getLocEnd()); 63 64 return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args) 65 << 0 /*function call*/ << desiredArgCount << argCount 66 << call->getArg(1)->getSourceRange(); 67} 68 69/// Check that the first argument to __builtin_annotation is an integer 70/// and the second argument is a non-wide string literal. 71static bool SemaBuiltinAnnotation(Sema &S, CallExpr *TheCall) { 72 if (checkArgCount(S, TheCall, 2)) 73 return true; 74 75 // First argument should be an integer. 76 Expr *ValArg = TheCall->getArg(0); 77 QualType Ty = ValArg->getType(); 78 if (!Ty->isIntegerType()) { 79 S.Diag(ValArg->getLocStart(), diag::err_builtin_annotation_first_arg) 80 << ValArg->getSourceRange(); 81 return true; 82 } 83 84 // Second argument should be a constant string. 85 Expr *StrArg = TheCall->getArg(1)->IgnoreParenCasts(); 86 StringLiteral *Literal = dyn_cast<StringLiteral>(StrArg); 87 if (!Literal || !Literal->isAscii()) { 88 S.Diag(StrArg->getLocStart(), diag::err_builtin_annotation_second_arg) 89 << StrArg->getSourceRange(); 90 return true; 91 } 92 93 TheCall->setType(Ty); 94 return false; 95} 96 97ExprResult 98Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 99 ExprResult TheCallResult(Owned(TheCall)); 100 101 // Find out if any arguments are required to be integer constant expressions. 102 unsigned ICEArguments = 0; 103 ASTContext::GetBuiltinTypeError Error; 104 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments); 105 if (Error != ASTContext::GE_None) 106 ICEArguments = 0; // Don't diagnose previously diagnosed errors. 107 108 // If any arguments are required to be ICE's, check and diagnose. 109 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) { 110 // Skip arguments not required to be ICE's. 111 if ((ICEArguments & (1 << ArgNo)) == 0) continue; 112 113 llvm::APSInt Result; 114 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 115 return true; 116 ICEArguments &= ~(1 << ArgNo); 117 } 118 119 switch (BuiltinID) { 120 case Builtin::BI__builtin___CFStringMakeConstantString: 121 assert(TheCall->getNumArgs() == 1 && 122 "Wrong # arguments to builtin CFStringMakeConstantString"); 123 if (CheckObjCString(TheCall->getArg(0))) 124 return ExprError(); 125 break; 126 case Builtin::BI__builtin_stdarg_start: 127 case Builtin::BI__builtin_va_start: 128 if (SemaBuiltinVAStart(TheCall)) 129 return ExprError(); 130 break; 131 case Builtin::BI__builtin_isgreater: 132 case Builtin::BI__builtin_isgreaterequal: 133 case Builtin::BI__builtin_isless: 134 case Builtin::BI__builtin_islessequal: 135 case Builtin::BI__builtin_islessgreater: 136 case Builtin::BI__builtin_isunordered: 137 if (SemaBuiltinUnorderedCompare(TheCall)) 138 return ExprError(); 139 break; 140 case Builtin::BI__builtin_fpclassify: 141 if (SemaBuiltinFPClassification(TheCall, 6)) 142 return ExprError(); 143 break; 144 case Builtin::BI__builtin_isfinite: 145 case Builtin::BI__builtin_isinf: 146 case Builtin::BI__builtin_isinf_sign: 147 case Builtin::BI__builtin_isnan: 148 case Builtin::BI__builtin_isnormal: 149 if (SemaBuiltinFPClassification(TheCall, 1)) 150 return ExprError(); 151 break; 152 case Builtin::BI__builtin_shufflevector: 153 return SemaBuiltinShuffleVector(TheCall); 154 // TheCall will be freed by the smart pointer here, but that's fine, since 155 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 156 case Builtin::BI__builtin_prefetch: 157 if (SemaBuiltinPrefetch(TheCall)) 158 return ExprError(); 159 break; 160 case Builtin::BI__builtin_object_size: 161 if (SemaBuiltinObjectSize(TheCall)) 162 return ExprError(); 163 break; 164 case Builtin::BI__builtin_longjmp: 165 if (SemaBuiltinLongjmp(TheCall)) 166 return ExprError(); 167 break; 168 169 case Builtin::BI__builtin_classify_type: 170 if (checkArgCount(*this, TheCall, 1)) return true; 171 TheCall->setType(Context.IntTy); 172 break; 173 case Builtin::BI__builtin_constant_p: 174 if (checkArgCount(*this, TheCall, 1)) return true; 175 TheCall->setType(Context.IntTy); 176 break; 177 case Builtin::BI__sync_fetch_and_add: 178 case Builtin::BI__sync_fetch_and_add_1: 179 case Builtin::BI__sync_fetch_and_add_2: 180 case Builtin::BI__sync_fetch_and_add_4: 181 case Builtin::BI__sync_fetch_and_add_8: 182 case Builtin::BI__sync_fetch_and_add_16: 183 case Builtin::BI__sync_fetch_and_sub: 184 case Builtin::BI__sync_fetch_and_sub_1: 185 case Builtin::BI__sync_fetch_and_sub_2: 186 case Builtin::BI__sync_fetch_and_sub_4: 187 case Builtin::BI__sync_fetch_and_sub_8: 188 case Builtin::BI__sync_fetch_and_sub_16: 189 case Builtin::BI__sync_fetch_and_or: 190 case Builtin::BI__sync_fetch_and_or_1: 191 case Builtin::BI__sync_fetch_and_or_2: 192 case Builtin::BI__sync_fetch_and_or_4: 193 case Builtin::BI__sync_fetch_and_or_8: 194 case Builtin::BI__sync_fetch_and_or_16: 195 case Builtin::BI__sync_fetch_and_and: 196 case Builtin::BI__sync_fetch_and_and_1: 197 case Builtin::BI__sync_fetch_and_and_2: 198 case Builtin::BI__sync_fetch_and_and_4: 199 case Builtin::BI__sync_fetch_and_and_8: 200 case Builtin::BI__sync_fetch_and_and_16: 201 case Builtin::BI__sync_fetch_and_xor: 202 case Builtin::BI__sync_fetch_and_xor_1: 203 case Builtin::BI__sync_fetch_and_xor_2: 204 case Builtin::BI__sync_fetch_and_xor_4: 205 case Builtin::BI__sync_fetch_and_xor_8: 206 case Builtin::BI__sync_fetch_and_xor_16: 207 case Builtin::BI__sync_add_and_fetch: 208 case Builtin::BI__sync_add_and_fetch_1: 209 case Builtin::BI__sync_add_and_fetch_2: 210 case Builtin::BI__sync_add_and_fetch_4: 211 case Builtin::BI__sync_add_and_fetch_8: 212 case Builtin::BI__sync_add_and_fetch_16: 213 case Builtin::BI__sync_sub_and_fetch: 214 case Builtin::BI__sync_sub_and_fetch_1: 215 case Builtin::BI__sync_sub_and_fetch_2: 216 case Builtin::BI__sync_sub_and_fetch_4: 217 case Builtin::BI__sync_sub_and_fetch_8: 218 case Builtin::BI__sync_sub_and_fetch_16: 219 case Builtin::BI__sync_and_and_fetch: 220 case Builtin::BI__sync_and_and_fetch_1: 221 case Builtin::BI__sync_and_and_fetch_2: 222 case Builtin::BI__sync_and_and_fetch_4: 223 case Builtin::BI__sync_and_and_fetch_8: 224 case Builtin::BI__sync_and_and_fetch_16: 225 case Builtin::BI__sync_or_and_fetch: 226 case Builtin::BI__sync_or_and_fetch_1: 227 case Builtin::BI__sync_or_and_fetch_2: 228 case Builtin::BI__sync_or_and_fetch_4: 229 case Builtin::BI__sync_or_and_fetch_8: 230 case Builtin::BI__sync_or_and_fetch_16: 231 case Builtin::BI__sync_xor_and_fetch: 232 case Builtin::BI__sync_xor_and_fetch_1: 233 case Builtin::BI__sync_xor_and_fetch_2: 234 case Builtin::BI__sync_xor_and_fetch_4: 235 case Builtin::BI__sync_xor_and_fetch_8: 236 case Builtin::BI__sync_xor_and_fetch_16: 237 case Builtin::BI__sync_val_compare_and_swap: 238 case Builtin::BI__sync_val_compare_and_swap_1: 239 case Builtin::BI__sync_val_compare_and_swap_2: 240 case Builtin::BI__sync_val_compare_and_swap_4: 241 case Builtin::BI__sync_val_compare_and_swap_8: 242 case Builtin::BI__sync_val_compare_and_swap_16: 243 case Builtin::BI__sync_bool_compare_and_swap: 244 case Builtin::BI__sync_bool_compare_and_swap_1: 245 case Builtin::BI__sync_bool_compare_and_swap_2: 246 case Builtin::BI__sync_bool_compare_and_swap_4: 247 case Builtin::BI__sync_bool_compare_and_swap_8: 248 case Builtin::BI__sync_bool_compare_and_swap_16: 249 case Builtin::BI__sync_lock_test_and_set: 250 case Builtin::BI__sync_lock_test_and_set_1: 251 case Builtin::BI__sync_lock_test_and_set_2: 252 case Builtin::BI__sync_lock_test_and_set_4: 253 case Builtin::BI__sync_lock_test_and_set_8: 254 case Builtin::BI__sync_lock_test_and_set_16: 255 case Builtin::BI__sync_lock_release: 256 case Builtin::BI__sync_lock_release_1: 257 case Builtin::BI__sync_lock_release_2: 258 case Builtin::BI__sync_lock_release_4: 259 case Builtin::BI__sync_lock_release_8: 260 case Builtin::BI__sync_lock_release_16: 261 case Builtin::BI__sync_swap: 262 case Builtin::BI__sync_swap_1: 263 case Builtin::BI__sync_swap_2: 264 case Builtin::BI__sync_swap_4: 265 case Builtin::BI__sync_swap_8: 266 case Builtin::BI__sync_swap_16: 267 return SemaBuiltinAtomicOverloaded(move(TheCallResult)); 268#define BUILTIN(ID, TYPE, ATTRS) 269#define ATOMIC_BUILTIN(ID, TYPE, ATTRS) \ 270 case Builtin::BI##ID: \ 271 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::AO##ID); 272#include "clang/Basic/Builtins.def" 273 case Builtin::BI__builtin_annotation: 274 if (SemaBuiltinAnnotation(*this, TheCall)) 275 return ExprError(); 276 break; 277 } 278 279 // Since the target specific builtins for each arch overlap, only check those 280 // of the arch we are compiling for. 281 if (BuiltinID >= Builtin::FirstTSBuiltin) { 282 switch (Context.getTargetInfo().getTriple().getArch()) { 283 case llvm::Triple::arm: 284 case llvm::Triple::thumb: 285 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 286 return ExprError(); 287 break; 288 default: 289 break; 290 } 291 } 292 293 return move(TheCallResult); 294} 295 296// Get the valid immediate range for the specified NEON type code. 297static unsigned RFT(unsigned t, bool shift = false) { 298 NeonTypeFlags Type(t); 299 int IsQuad = Type.isQuad(); 300 switch (Type.getEltType()) { 301 case NeonTypeFlags::Int8: 302 case NeonTypeFlags::Poly8: 303 return shift ? 7 : (8 << IsQuad) - 1; 304 case NeonTypeFlags::Int16: 305 case NeonTypeFlags::Poly16: 306 return shift ? 15 : (4 << IsQuad) - 1; 307 case NeonTypeFlags::Int32: 308 return shift ? 31 : (2 << IsQuad) - 1; 309 case NeonTypeFlags::Int64: 310 return shift ? 63 : (1 << IsQuad) - 1; 311 case NeonTypeFlags::Float16: 312 assert(!shift && "cannot shift float types!"); 313 return (4 << IsQuad) - 1; 314 case NeonTypeFlags::Float32: 315 assert(!shift && "cannot shift float types!"); 316 return (2 << IsQuad) - 1; 317 } 318 llvm_unreachable("Invalid NeonTypeFlag!"); 319} 320 321/// getNeonEltType - Return the QualType corresponding to the elements of 322/// the vector type specified by the NeonTypeFlags. This is used to check 323/// the pointer arguments for Neon load/store intrinsics. 324static QualType getNeonEltType(NeonTypeFlags Flags, ASTContext &Context) { 325 switch (Flags.getEltType()) { 326 case NeonTypeFlags::Int8: 327 return Flags.isUnsigned() ? Context.UnsignedCharTy : Context.SignedCharTy; 328 case NeonTypeFlags::Int16: 329 return Flags.isUnsigned() ? Context.UnsignedShortTy : Context.ShortTy; 330 case NeonTypeFlags::Int32: 331 return Flags.isUnsigned() ? Context.UnsignedIntTy : Context.IntTy; 332 case NeonTypeFlags::Int64: 333 return Flags.isUnsigned() ? Context.UnsignedLongLongTy : Context.LongLongTy; 334 case NeonTypeFlags::Poly8: 335 return Context.SignedCharTy; 336 case NeonTypeFlags::Poly16: 337 return Context.ShortTy; 338 case NeonTypeFlags::Float16: 339 return Context.UnsignedShortTy; 340 case NeonTypeFlags::Float32: 341 return Context.FloatTy; 342 } 343 llvm_unreachable("Invalid NeonTypeFlag!"); 344} 345 346bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 347 llvm::APSInt Result; 348 349 unsigned mask = 0; 350 unsigned TV = 0; 351 int PtrArgNum = -1; 352 bool HasConstPtr = false; 353 switch (BuiltinID) { 354#define GET_NEON_OVERLOAD_CHECK 355#include "clang/Basic/arm_neon.inc" 356#undef GET_NEON_OVERLOAD_CHECK 357 } 358 359 // For NEON intrinsics which are overloaded on vector element type, validate 360 // the immediate which specifies which variant to emit. 361 unsigned ImmArg = TheCall->getNumArgs()-1; 362 if (mask) { 363 if (SemaBuiltinConstantArg(TheCall, ImmArg, Result)) 364 return true; 365 366 TV = Result.getLimitedValue(64); 367 if ((TV > 63) || (mask & (1 << TV)) == 0) 368 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 369 << TheCall->getArg(ImmArg)->getSourceRange(); 370 } 371 372 if (PtrArgNum >= 0) { 373 // Check that pointer arguments have the specified type. 374 Expr *Arg = TheCall->getArg(PtrArgNum); 375 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) 376 Arg = ICE->getSubExpr(); 377 ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg); 378 QualType RHSTy = RHS.get()->getType(); 379 QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context); 380 if (HasConstPtr) 381 EltTy = EltTy.withConst(); 382 QualType LHSTy = Context.getPointerType(EltTy); 383 AssignConvertType ConvTy; 384 ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS); 385 if (RHS.isInvalid()) 386 return true; 387 if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy, 388 RHS.get(), AA_Assigning)) 389 return true; 390 } 391 392 // For NEON intrinsics which take an immediate value as part of the 393 // instruction, range check them here. 394 unsigned i = 0, l = 0, u = 0; 395 switch (BuiltinID) { 396 default: return false; 397 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 398 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 399 case ARM::BI__builtin_arm_vcvtr_f: 400 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 401#define GET_NEON_IMMEDIATE_CHECK 402#include "clang/Basic/arm_neon.inc" 403#undef GET_NEON_IMMEDIATE_CHECK 404 }; 405 406 // Check that the immediate argument is actually a constant. 407 if (SemaBuiltinConstantArg(TheCall, i, Result)) 408 return true; 409 410 // Range check against the upper/lower values for this isntruction. 411 unsigned Val = Result.getZExtValue(); 412 if (Val < l || Val > (u + l)) 413 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 414 << l << u+l << TheCall->getArg(i)->getSourceRange(); 415 416 // FIXME: VFP Intrinsics should error if VFP not present. 417 return false; 418} 419 420/// CheckFunctionCall - Check a direct function call for various correctness 421/// and safety properties not strictly enforced by the C type system. 422bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 423 // Get the IdentifierInfo* for the called function. 424 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 425 426 // None of the checks below are needed for functions that don't have 427 // simple names (e.g., C++ conversion functions). 428 if (!FnInfo) 429 return false; 430 431 // FIXME: This mechanism should be abstracted to be less fragile and 432 // more efficient. For example, just map function ids to custom 433 // handlers. 434 435 // Printf and scanf checking. 436 for (specific_attr_iterator<FormatAttr> 437 i = FDecl->specific_attr_begin<FormatAttr>(), 438 e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { 439 CheckFormatArguments(*i, TheCall); 440 } 441 442 for (specific_attr_iterator<NonNullAttr> 443 i = FDecl->specific_attr_begin<NonNullAttr>(), 444 e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) { 445 CheckNonNullArguments(*i, TheCall->getArgs(), 446 TheCall->getCallee()->getLocStart()); 447 } 448 449 unsigned CMId = FDecl->getMemoryFunctionKind(); 450 if (CMId == 0) 451 return false; 452 453 // Handle memory setting and copying functions. 454 if (CMId == Builtin::BIstrlcpy || CMId == Builtin::BIstrlcat) 455 CheckStrlcpycatArguments(TheCall, FnInfo); 456 else if (CMId == Builtin::BIstrncat) 457 CheckStrncatArguments(TheCall, FnInfo); 458 else 459 CheckMemaccessArguments(TheCall, CMId, FnInfo); 460 461 return false; 462} 463 464bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac, 465 Expr **Args, unsigned NumArgs) { 466 for (specific_attr_iterator<FormatAttr> 467 i = Method->specific_attr_begin<FormatAttr>(), 468 e = Method->specific_attr_end<FormatAttr>(); i != e ; ++i) { 469 470 CheckFormatArguments(*i, Args, NumArgs, false, lbrac, 471 Method->getSourceRange()); 472 } 473 474 // diagnose nonnull arguments. 475 for (specific_attr_iterator<NonNullAttr> 476 i = Method->specific_attr_begin<NonNullAttr>(), 477 e = Method->specific_attr_end<NonNullAttr>(); i != e; ++i) { 478 CheckNonNullArguments(*i, Args, lbrac); 479 } 480 481 return false; 482} 483 484bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 485 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 486 if (!V) 487 return false; 488 489 QualType Ty = V->getType(); 490 if (!Ty->isBlockPointerType()) 491 return false; 492 493 // format string checking. 494 for (specific_attr_iterator<FormatAttr> 495 i = NDecl->specific_attr_begin<FormatAttr>(), 496 e = NDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { 497 CheckFormatArguments(*i, TheCall); 498 } 499 500 return false; 501} 502 503ExprResult Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, 504 AtomicExpr::AtomicOp Op) { 505 CallExpr *TheCall = cast<CallExpr>(TheCallResult.get()); 506 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 507 508 // All these operations take one of the following forms: 509 enum { 510 // C __c11_atomic_init(A *, C) 511 Init, 512 // C __c11_atomic_load(A *, int) 513 Load, 514 // void __atomic_load(A *, CP, int) 515 Copy, 516 // C __c11_atomic_add(A *, M, int) 517 Arithmetic, 518 // C __atomic_exchange_n(A *, CP, int) 519 Xchg, 520 // void __atomic_exchange(A *, C *, CP, int) 521 GNUXchg, 522 // bool __c11_atomic_compare_exchange_strong(A *, C *, CP, int, int) 523 C11CmpXchg, 524 // bool __atomic_compare_exchange(A *, C *, CP, bool, int, int) 525 GNUCmpXchg 526 } Form = Init; 527 const unsigned NumArgs[] = { 2, 2, 3, 3, 3, 4, 5, 6 }; 528 const unsigned NumVals[] = { 1, 0, 1, 1, 1, 2, 2, 3 }; 529 // where: 530 // C is an appropriate type, 531 // A is volatile _Atomic(C) for __c11 builtins and is C for GNU builtins, 532 // CP is C for __c11 builtins and GNU _n builtins and is C * otherwise, 533 // M is C if C is an integer, and ptrdiff_t if C is a pointer, and 534 // the int parameters are for orderings. 535 536 assert(AtomicExpr::AO__c11_atomic_init == 0 && 537 AtomicExpr::AO__c11_atomic_fetch_xor + 1 == AtomicExpr::AO__atomic_load 538 && "need to update code for modified C11 atomics"); 539 bool IsC11 = Op >= AtomicExpr::AO__c11_atomic_init && 540 Op <= AtomicExpr::AO__c11_atomic_fetch_xor; 541 bool IsN = Op == AtomicExpr::AO__atomic_load_n || 542 Op == AtomicExpr::AO__atomic_store_n || 543 Op == AtomicExpr::AO__atomic_exchange_n || 544 Op == AtomicExpr::AO__atomic_compare_exchange_n; 545 bool IsAddSub = false; 546 547 switch (Op) { 548 case AtomicExpr::AO__c11_atomic_init: 549 Form = Init; 550 break; 551 552 case AtomicExpr::AO__c11_atomic_load: 553 case AtomicExpr::AO__atomic_load_n: 554 Form = Load; 555 break; 556 557 case AtomicExpr::AO__c11_atomic_store: 558 case AtomicExpr::AO__atomic_load: 559 case AtomicExpr::AO__atomic_store: 560 case AtomicExpr::AO__atomic_store_n: 561 Form = Copy; 562 break; 563 564 case AtomicExpr::AO__c11_atomic_fetch_add: 565 case AtomicExpr::AO__c11_atomic_fetch_sub: 566 case AtomicExpr::AO__atomic_fetch_add: 567 case AtomicExpr::AO__atomic_fetch_sub: 568 case AtomicExpr::AO__atomic_add_fetch: 569 case AtomicExpr::AO__atomic_sub_fetch: 570 IsAddSub = true; 571 // Fall through. 572 case AtomicExpr::AO__c11_atomic_fetch_and: 573 case AtomicExpr::AO__c11_atomic_fetch_or: 574 case AtomicExpr::AO__c11_atomic_fetch_xor: 575 case AtomicExpr::AO__atomic_fetch_and: 576 case AtomicExpr::AO__atomic_fetch_or: 577 case AtomicExpr::AO__atomic_fetch_xor: 578 case AtomicExpr::AO__atomic_fetch_nand: 579 case AtomicExpr::AO__atomic_and_fetch: 580 case AtomicExpr::AO__atomic_or_fetch: 581 case AtomicExpr::AO__atomic_xor_fetch: 582 case AtomicExpr::AO__atomic_nand_fetch: 583 Form = Arithmetic; 584 break; 585 586 case AtomicExpr::AO__c11_atomic_exchange: 587 case AtomicExpr::AO__atomic_exchange_n: 588 Form = Xchg; 589 break; 590 591 case AtomicExpr::AO__atomic_exchange: 592 Form = GNUXchg; 593 break; 594 595 case AtomicExpr::AO__c11_atomic_compare_exchange_strong: 596 case AtomicExpr::AO__c11_atomic_compare_exchange_weak: 597 Form = C11CmpXchg; 598 break; 599 600 case AtomicExpr::AO__atomic_compare_exchange: 601 case AtomicExpr::AO__atomic_compare_exchange_n: 602 Form = GNUCmpXchg; 603 break; 604 } 605 606 // Check we have the right number of arguments. 607 if (TheCall->getNumArgs() < NumArgs[Form]) { 608 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 609 << 0 << NumArgs[Form] << TheCall->getNumArgs() 610 << TheCall->getCallee()->getSourceRange(); 611 return ExprError(); 612 } else if (TheCall->getNumArgs() > NumArgs[Form]) { 613 Diag(TheCall->getArg(NumArgs[Form])->getLocStart(), 614 diag::err_typecheck_call_too_many_args) 615 << 0 << NumArgs[Form] << TheCall->getNumArgs() 616 << TheCall->getCallee()->getSourceRange(); 617 return ExprError(); 618 } 619 620 // Inspect the first argument of the atomic operation. 621 Expr *Ptr = TheCall->getArg(0); 622 Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get(); 623 const PointerType *pointerType = Ptr->getType()->getAs<PointerType>(); 624 if (!pointerType) { 625 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 626 << Ptr->getType() << Ptr->getSourceRange(); 627 return ExprError(); 628 } 629 630 // For a __c11 builtin, this should be a pointer to an _Atomic type. 631 QualType AtomTy = pointerType->getPointeeType(); // 'A' 632 QualType ValType = AtomTy; // 'C' 633 if (IsC11) { 634 if (!AtomTy->isAtomicType()) { 635 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 636 << Ptr->getType() << Ptr->getSourceRange(); 637 return ExprError(); 638 } 639 ValType = AtomTy->getAs<AtomicType>()->getValueType(); 640 } 641 642 // For an arithmetic operation, the implied arithmetic must be well-formed. 643 if (Form == Arithmetic) { 644 // gcc does not enforce these rules for GNU atomics, but we do so for sanity. 645 if (IsAddSub && !ValType->isIntegerType() && !ValType->isPointerType()) { 646 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 647 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 648 return ExprError(); 649 } 650 if (!IsAddSub && !ValType->isIntegerType()) { 651 Diag(DRE->getLocStart(), diag::err_atomic_op_bitwise_needs_atomic_int) 652 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 653 return ExprError(); 654 } 655 } else if (IsN && !ValType->isIntegerType() && !ValType->isPointerType()) { 656 // For __atomic_*_n operations, the value type must be a scalar integral or 657 // pointer type which is 1, 2, 4, 8 or 16 bytes in length. 658 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 659 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 660 return ExprError(); 661 } 662 663 if (!IsC11 && !AtomTy.isTriviallyCopyableType(Context)) { 664 // For GNU atomics, require a trivially-copyable type. This is not part of 665 // the GNU atomics specification, but we enforce it for sanity. 666 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_trivial_copy) 667 << Ptr->getType() << Ptr->getSourceRange(); 668 return ExprError(); 669 } 670 671 // FIXME: For any builtin other than a load, the ValType must not be 672 // const-qualified. 673 674 switch (ValType.getObjCLifetime()) { 675 case Qualifiers::OCL_None: 676 case Qualifiers::OCL_ExplicitNone: 677 // okay 678 break; 679 680 case Qualifiers::OCL_Weak: 681 case Qualifiers::OCL_Strong: 682 case Qualifiers::OCL_Autoreleasing: 683 // FIXME: Can this happen? By this point, ValType should be known 684 // to be trivially copyable. 685 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 686 << ValType << Ptr->getSourceRange(); 687 return ExprError(); 688 } 689 690 QualType ResultType = ValType; 691 if (Form == Copy || Form == GNUXchg || Form == Init) 692 ResultType = Context.VoidTy; 693 else if (Form == C11CmpXchg || Form == GNUCmpXchg) 694 ResultType = Context.BoolTy; 695 696 // The type of a parameter passed 'by value'. In the GNU atomics, such 697 // arguments are actually passed as pointers. 698 QualType ByValType = ValType; // 'CP' 699 if (!IsC11 && !IsN) 700 ByValType = Ptr->getType(); 701 702 // The first argument --- the pointer --- has a fixed type; we 703 // deduce the types of the rest of the arguments accordingly. Walk 704 // the remaining arguments, converting them to the deduced value type. 705 for (unsigned i = 1; i != NumArgs[Form]; ++i) { 706 QualType Ty; 707 if (i < NumVals[Form] + 1) { 708 switch (i) { 709 case 1: 710 // The second argument is the non-atomic operand. For arithmetic, this 711 // is always passed by value, and for a compare_exchange it is always 712 // passed by address. For the rest, GNU uses by-address and C11 uses 713 // by-value. 714 assert(Form != Load); 715 if (Form == Init || (Form == Arithmetic && ValType->isIntegerType())) 716 Ty = ValType; 717 else if (Form == Copy || Form == Xchg) 718 Ty = ByValType; 719 else if (Form == Arithmetic) 720 Ty = Context.getPointerDiffType(); 721 else 722 Ty = Context.getPointerType(ValType.getUnqualifiedType()); 723 break; 724 case 2: 725 // The third argument to compare_exchange / GNU exchange is a 726 // (pointer to a) desired value. 727 Ty = ByValType; 728 break; 729 case 3: 730 // The fourth argument to GNU compare_exchange is a 'weak' flag. 731 Ty = Context.BoolTy; 732 break; 733 } 734 } else { 735 // The order(s) are always converted to int. 736 Ty = Context.IntTy; 737 } 738 739 InitializedEntity Entity = 740 InitializedEntity::InitializeParameter(Context, Ty, false); 741 ExprResult Arg = TheCall->getArg(i); 742 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 743 if (Arg.isInvalid()) 744 return true; 745 TheCall->setArg(i, Arg.get()); 746 } 747 748 // Permute the arguments into a 'consistent' order. 749 SmallVector<Expr*, 5> SubExprs; 750 SubExprs.push_back(Ptr); 751 switch (Form) { 752 case Init: 753 // Note, AtomicExpr::getVal1() has a special case for this atomic. 754 SubExprs.push_back(TheCall->getArg(1)); // Val1 755 break; 756 case Load: 757 SubExprs.push_back(TheCall->getArg(1)); // Order 758 break; 759 case Copy: 760 case Arithmetic: 761 case Xchg: 762 SubExprs.push_back(TheCall->getArg(2)); // Order 763 SubExprs.push_back(TheCall->getArg(1)); // Val1 764 break; 765 case GNUXchg: 766 // Note, AtomicExpr::getVal2() has a special case for this atomic. 767 SubExprs.push_back(TheCall->getArg(3)); // Order 768 SubExprs.push_back(TheCall->getArg(1)); // Val1 769 SubExprs.push_back(TheCall->getArg(2)); // Val2 770 break; 771 case C11CmpXchg: 772 SubExprs.push_back(TheCall->getArg(3)); // Order 773 SubExprs.push_back(TheCall->getArg(1)); // Val1 774 SubExprs.push_back(TheCall->getArg(4)); // OrderFail 775 SubExprs.push_back(TheCall->getArg(2)); // Val2 776 break; 777 case GNUCmpXchg: 778 SubExprs.push_back(TheCall->getArg(4)); // Order 779 SubExprs.push_back(TheCall->getArg(1)); // Val1 780 SubExprs.push_back(TheCall->getArg(5)); // OrderFail 781 SubExprs.push_back(TheCall->getArg(2)); // Val2 782 SubExprs.push_back(TheCall->getArg(3)); // Weak 783 break; 784 } 785 786 return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 787 SubExprs.data(), SubExprs.size(), 788 ResultType, Op, 789 TheCall->getRParenLoc())); 790} 791 792 793/// checkBuiltinArgument - Given a call to a builtin function, perform 794/// normal type-checking on the given argument, updating the call in 795/// place. This is useful when a builtin function requires custom 796/// type-checking for some of its arguments but not necessarily all of 797/// them. 798/// 799/// Returns true on error. 800static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 801 FunctionDecl *Fn = E->getDirectCallee(); 802 assert(Fn && "builtin call without direct callee!"); 803 804 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 805 InitializedEntity Entity = 806 InitializedEntity::InitializeParameter(S.Context, Param); 807 808 ExprResult Arg = E->getArg(0); 809 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 810 if (Arg.isInvalid()) 811 return true; 812 813 E->setArg(ArgIndex, Arg.take()); 814 return false; 815} 816 817/// SemaBuiltinAtomicOverloaded - We have a call to a function like 818/// __sync_fetch_and_add, which is an overloaded function based on the pointer 819/// type of its first argument. The main ActOnCallExpr routines have already 820/// promoted the types of arguments because all of these calls are prototyped as 821/// void(...). 822/// 823/// This function goes through and does final semantic checking for these 824/// builtins, 825ExprResult 826Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 827 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 828 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 829 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 830 831 // Ensure that we have at least one argument to do type inference from. 832 if (TheCall->getNumArgs() < 1) { 833 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 834 << 0 << 1 << TheCall->getNumArgs() 835 << TheCall->getCallee()->getSourceRange(); 836 return ExprError(); 837 } 838 839 // Inspect the first argument of the atomic builtin. This should always be 840 // a pointer type, whose element is an integral scalar or pointer type. 841 // Because it is a pointer type, we don't have to worry about any implicit 842 // casts here. 843 // FIXME: We don't allow floating point scalars as input. 844 Expr *FirstArg = TheCall->getArg(0); 845 ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg); 846 if (FirstArgResult.isInvalid()) 847 return ExprError(); 848 FirstArg = FirstArgResult.take(); 849 TheCall->setArg(0, FirstArg); 850 851 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 852 if (!pointerType) { 853 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 854 << FirstArg->getType() << FirstArg->getSourceRange(); 855 return ExprError(); 856 } 857 858 QualType ValType = pointerType->getPointeeType(); 859 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 860 !ValType->isBlockPointerType()) { 861 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 862 << FirstArg->getType() << FirstArg->getSourceRange(); 863 return ExprError(); 864 } 865 866 switch (ValType.getObjCLifetime()) { 867 case Qualifiers::OCL_None: 868 case Qualifiers::OCL_ExplicitNone: 869 // okay 870 break; 871 872 case Qualifiers::OCL_Weak: 873 case Qualifiers::OCL_Strong: 874 case Qualifiers::OCL_Autoreleasing: 875 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 876 << ValType << FirstArg->getSourceRange(); 877 return ExprError(); 878 } 879 880 // Strip any qualifiers off ValType. 881 ValType = ValType.getUnqualifiedType(); 882 883 // The majority of builtins return a value, but a few have special return 884 // types, so allow them to override appropriately below. 885 QualType ResultType = ValType; 886 887 // We need to figure out which concrete builtin this maps onto. For example, 888 // __sync_fetch_and_add with a 2 byte object turns into 889 // __sync_fetch_and_add_2. 890#define BUILTIN_ROW(x) \ 891 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 892 Builtin::BI##x##_8, Builtin::BI##x##_16 } 893 894 static const unsigned BuiltinIndices[][5] = { 895 BUILTIN_ROW(__sync_fetch_and_add), 896 BUILTIN_ROW(__sync_fetch_and_sub), 897 BUILTIN_ROW(__sync_fetch_and_or), 898 BUILTIN_ROW(__sync_fetch_and_and), 899 BUILTIN_ROW(__sync_fetch_and_xor), 900 901 BUILTIN_ROW(__sync_add_and_fetch), 902 BUILTIN_ROW(__sync_sub_and_fetch), 903 BUILTIN_ROW(__sync_and_and_fetch), 904 BUILTIN_ROW(__sync_or_and_fetch), 905 BUILTIN_ROW(__sync_xor_and_fetch), 906 907 BUILTIN_ROW(__sync_val_compare_and_swap), 908 BUILTIN_ROW(__sync_bool_compare_and_swap), 909 BUILTIN_ROW(__sync_lock_test_and_set), 910 BUILTIN_ROW(__sync_lock_release), 911 BUILTIN_ROW(__sync_swap) 912 }; 913#undef BUILTIN_ROW 914 915 // Determine the index of the size. 916 unsigned SizeIndex; 917 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 918 case 1: SizeIndex = 0; break; 919 case 2: SizeIndex = 1; break; 920 case 4: SizeIndex = 2; break; 921 case 8: SizeIndex = 3; break; 922 case 16: SizeIndex = 4; break; 923 default: 924 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 925 << FirstArg->getType() << FirstArg->getSourceRange(); 926 return ExprError(); 927 } 928 929 // Each of these builtins has one pointer argument, followed by some number of 930 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 931 // that we ignore. Find out which row of BuiltinIndices to read from as well 932 // as the number of fixed args. 933 unsigned BuiltinID = FDecl->getBuiltinID(); 934 unsigned BuiltinIndex, NumFixed = 1; 935 switch (BuiltinID) { 936 default: llvm_unreachable("Unknown overloaded atomic builtin!"); 937 case Builtin::BI__sync_fetch_and_add: 938 case Builtin::BI__sync_fetch_and_add_1: 939 case Builtin::BI__sync_fetch_and_add_2: 940 case Builtin::BI__sync_fetch_and_add_4: 941 case Builtin::BI__sync_fetch_and_add_8: 942 case Builtin::BI__sync_fetch_and_add_16: 943 BuiltinIndex = 0; 944 break; 945 946 case Builtin::BI__sync_fetch_and_sub: 947 case Builtin::BI__sync_fetch_and_sub_1: 948 case Builtin::BI__sync_fetch_and_sub_2: 949 case Builtin::BI__sync_fetch_and_sub_4: 950 case Builtin::BI__sync_fetch_and_sub_8: 951 case Builtin::BI__sync_fetch_and_sub_16: 952 BuiltinIndex = 1; 953 break; 954 955 case Builtin::BI__sync_fetch_and_or: 956 case Builtin::BI__sync_fetch_and_or_1: 957 case Builtin::BI__sync_fetch_and_or_2: 958 case Builtin::BI__sync_fetch_and_or_4: 959 case Builtin::BI__sync_fetch_and_or_8: 960 case Builtin::BI__sync_fetch_and_or_16: 961 BuiltinIndex = 2; 962 break; 963 964 case Builtin::BI__sync_fetch_and_and: 965 case Builtin::BI__sync_fetch_and_and_1: 966 case Builtin::BI__sync_fetch_and_and_2: 967 case Builtin::BI__sync_fetch_and_and_4: 968 case Builtin::BI__sync_fetch_and_and_8: 969 case Builtin::BI__sync_fetch_and_and_16: 970 BuiltinIndex = 3; 971 break; 972 973 case Builtin::BI__sync_fetch_and_xor: 974 case Builtin::BI__sync_fetch_and_xor_1: 975 case Builtin::BI__sync_fetch_and_xor_2: 976 case Builtin::BI__sync_fetch_and_xor_4: 977 case Builtin::BI__sync_fetch_and_xor_8: 978 case Builtin::BI__sync_fetch_and_xor_16: 979 BuiltinIndex = 4; 980 break; 981 982 case Builtin::BI__sync_add_and_fetch: 983 case Builtin::BI__sync_add_and_fetch_1: 984 case Builtin::BI__sync_add_and_fetch_2: 985 case Builtin::BI__sync_add_and_fetch_4: 986 case Builtin::BI__sync_add_and_fetch_8: 987 case Builtin::BI__sync_add_and_fetch_16: 988 BuiltinIndex = 5; 989 break; 990 991 case Builtin::BI__sync_sub_and_fetch: 992 case Builtin::BI__sync_sub_and_fetch_1: 993 case Builtin::BI__sync_sub_and_fetch_2: 994 case Builtin::BI__sync_sub_and_fetch_4: 995 case Builtin::BI__sync_sub_and_fetch_8: 996 case Builtin::BI__sync_sub_and_fetch_16: 997 BuiltinIndex = 6; 998 break; 999 1000 case Builtin::BI__sync_and_and_fetch: 1001 case Builtin::BI__sync_and_and_fetch_1: 1002 case Builtin::BI__sync_and_and_fetch_2: 1003 case Builtin::BI__sync_and_and_fetch_4: 1004 case Builtin::BI__sync_and_and_fetch_8: 1005 case Builtin::BI__sync_and_and_fetch_16: 1006 BuiltinIndex = 7; 1007 break; 1008 1009 case Builtin::BI__sync_or_and_fetch: 1010 case Builtin::BI__sync_or_and_fetch_1: 1011 case Builtin::BI__sync_or_and_fetch_2: 1012 case Builtin::BI__sync_or_and_fetch_4: 1013 case Builtin::BI__sync_or_and_fetch_8: 1014 case Builtin::BI__sync_or_and_fetch_16: 1015 BuiltinIndex = 8; 1016 break; 1017 1018 case Builtin::BI__sync_xor_and_fetch: 1019 case Builtin::BI__sync_xor_and_fetch_1: 1020 case Builtin::BI__sync_xor_and_fetch_2: 1021 case Builtin::BI__sync_xor_and_fetch_4: 1022 case Builtin::BI__sync_xor_and_fetch_8: 1023 case Builtin::BI__sync_xor_and_fetch_16: 1024 BuiltinIndex = 9; 1025 break; 1026 1027 case Builtin::BI__sync_val_compare_and_swap: 1028 case Builtin::BI__sync_val_compare_and_swap_1: 1029 case Builtin::BI__sync_val_compare_and_swap_2: 1030 case Builtin::BI__sync_val_compare_and_swap_4: 1031 case Builtin::BI__sync_val_compare_and_swap_8: 1032 case Builtin::BI__sync_val_compare_and_swap_16: 1033 BuiltinIndex = 10; 1034 NumFixed = 2; 1035 break; 1036 1037 case Builtin::BI__sync_bool_compare_and_swap: 1038 case Builtin::BI__sync_bool_compare_and_swap_1: 1039 case Builtin::BI__sync_bool_compare_and_swap_2: 1040 case Builtin::BI__sync_bool_compare_and_swap_4: 1041 case Builtin::BI__sync_bool_compare_and_swap_8: 1042 case Builtin::BI__sync_bool_compare_and_swap_16: 1043 BuiltinIndex = 11; 1044 NumFixed = 2; 1045 ResultType = Context.BoolTy; 1046 break; 1047 1048 case Builtin::BI__sync_lock_test_and_set: 1049 case Builtin::BI__sync_lock_test_and_set_1: 1050 case Builtin::BI__sync_lock_test_and_set_2: 1051 case Builtin::BI__sync_lock_test_and_set_4: 1052 case Builtin::BI__sync_lock_test_and_set_8: 1053 case Builtin::BI__sync_lock_test_and_set_16: 1054 BuiltinIndex = 12; 1055 break; 1056 1057 case Builtin::BI__sync_lock_release: 1058 case Builtin::BI__sync_lock_release_1: 1059 case Builtin::BI__sync_lock_release_2: 1060 case Builtin::BI__sync_lock_release_4: 1061 case Builtin::BI__sync_lock_release_8: 1062 case Builtin::BI__sync_lock_release_16: 1063 BuiltinIndex = 13; 1064 NumFixed = 0; 1065 ResultType = Context.VoidTy; 1066 break; 1067 1068 case Builtin::BI__sync_swap: 1069 case Builtin::BI__sync_swap_1: 1070 case Builtin::BI__sync_swap_2: 1071 case Builtin::BI__sync_swap_4: 1072 case Builtin::BI__sync_swap_8: 1073 case Builtin::BI__sync_swap_16: 1074 BuiltinIndex = 14; 1075 break; 1076 } 1077 1078 // Now that we know how many fixed arguments we expect, first check that we 1079 // have at least that many. 1080 if (TheCall->getNumArgs() < 1+NumFixed) { 1081 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 1082 << 0 << 1+NumFixed << TheCall->getNumArgs() 1083 << TheCall->getCallee()->getSourceRange(); 1084 return ExprError(); 1085 } 1086 1087 // Get the decl for the concrete builtin from this, we can tell what the 1088 // concrete integer type we should convert to is. 1089 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 1090 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 1091 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 1092 FunctionDecl *NewBuiltinDecl = 1093 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 1094 TUScope, false, DRE->getLocStart())); 1095 1096 // The first argument --- the pointer --- has a fixed type; we 1097 // deduce the types of the rest of the arguments accordingly. Walk 1098 // the remaining arguments, converting them to the deduced value type. 1099 for (unsigned i = 0; i != NumFixed; ++i) { 1100 ExprResult Arg = TheCall->getArg(i+1); 1101 1102 // GCC does an implicit conversion to the pointer or integer ValType. This 1103 // can fail in some cases (1i -> int**), check for this error case now. 1104 // Initialize the argument. 1105 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context, 1106 ValType, /*consume*/ false); 1107 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 1108 if (Arg.isInvalid()) 1109 return ExprError(); 1110 1111 // Okay, we have something that *can* be converted to the right type. Check 1112 // to see if there is a potentially weird extension going on here. This can 1113 // happen when you do an atomic operation on something like an char* and 1114 // pass in 42. The 42 gets converted to char. This is even more strange 1115 // for things like 45.123 -> char, etc. 1116 // FIXME: Do this check. 1117 TheCall->setArg(i+1, Arg.take()); 1118 } 1119 1120 ASTContext& Context = this->getASTContext(); 1121 1122 // Create a new DeclRefExpr to refer to the new decl. 1123 DeclRefExpr* NewDRE = DeclRefExpr::Create( 1124 Context, 1125 DRE->getQualifierLoc(), 1126 SourceLocation(), 1127 NewBuiltinDecl, 1128 /*enclosing*/ false, 1129 DRE->getLocation(), 1130 NewBuiltinDecl->getType(), 1131 DRE->getValueKind()); 1132 1133 // Set the callee in the CallExpr. 1134 // FIXME: This leaks the original parens and implicit casts. 1135 ExprResult PromotedCall = UsualUnaryConversions(NewDRE); 1136 if (PromotedCall.isInvalid()) 1137 return ExprError(); 1138 TheCall->setCallee(PromotedCall.take()); 1139 1140 // Change the result type of the call to match the original value type. This 1141 // is arbitrary, but the codegen for these builtins ins design to handle it 1142 // gracefully. 1143 TheCall->setType(ResultType); 1144 1145 return move(TheCallResult); 1146} 1147 1148/// CheckObjCString - Checks that the argument to the builtin 1149/// CFString constructor is correct 1150/// Note: It might also make sense to do the UTF-16 conversion here (would 1151/// simplify the backend). 1152bool Sema::CheckObjCString(Expr *Arg) { 1153 Arg = Arg->IgnoreParenCasts(); 1154 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 1155 1156 if (!Literal || !Literal->isAscii()) { 1157 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 1158 << Arg->getSourceRange(); 1159 return true; 1160 } 1161 1162 if (Literal->containsNonAsciiOrNull()) { 1163 StringRef String = Literal->getString(); 1164 unsigned NumBytes = String.size(); 1165 SmallVector<UTF16, 128> ToBuf(NumBytes); 1166 const UTF8 *FromPtr = (UTF8 *)String.data(); 1167 UTF16 *ToPtr = &ToBuf[0]; 1168 1169 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 1170 &ToPtr, ToPtr + NumBytes, 1171 strictConversion); 1172 // Check for conversion failure. 1173 if (Result != conversionOK) 1174 Diag(Arg->getLocStart(), 1175 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 1176 } 1177 return false; 1178} 1179 1180/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 1181/// Emit an error and return true on failure, return false on success. 1182bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 1183 Expr *Fn = TheCall->getCallee(); 1184 if (TheCall->getNumArgs() > 2) { 1185 Diag(TheCall->getArg(2)->getLocStart(), 1186 diag::err_typecheck_call_too_many_args) 1187 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1188 << Fn->getSourceRange() 1189 << SourceRange(TheCall->getArg(2)->getLocStart(), 1190 (*(TheCall->arg_end()-1))->getLocEnd()); 1191 return true; 1192 } 1193 1194 if (TheCall->getNumArgs() < 2) { 1195 return Diag(TheCall->getLocEnd(), 1196 diag::err_typecheck_call_too_few_args_at_least) 1197 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 1198 } 1199 1200 // Type-check the first argument normally. 1201 if (checkBuiltinArgument(*this, TheCall, 0)) 1202 return true; 1203 1204 // Determine whether the current function is variadic or not. 1205 BlockScopeInfo *CurBlock = getCurBlock(); 1206 bool isVariadic; 1207 if (CurBlock) 1208 isVariadic = CurBlock->TheDecl->isVariadic(); 1209 else if (FunctionDecl *FD = getCurFunctionDecl()) 1210 isVariadic = FD->isVariadic(); 1211 else 1212 isVariadic = getCurMethodDecl()->isVariadic(); 1213 1214 if (!isVariadic) { 1215 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 1216 return true; 1217 } 1218 1219 // Verify that the second argument to the builtin is the last argument of the 1220 // current function or method. 1221 bool SecondArgIsLastNamedArgument = false; 1222 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 1223 1224 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 1225 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 1226 // FIXME: This isn't correct for methods (results in bogus warning). 1227 // Get the last formal in the current function. 1228 const ParmVarDecl *LastArg; 1229 if (CurBlock) 1230 LastArg = *(CurBlock->TheDecl->param_end()-1); 1231 else if (FunctionDecl *FD = getCurFunctionDecl()) 1232 LastArg = *(FD->param_end()-1); 1233 else 1234 LastArg = *(getCurMethodDecl()->param_end()-1); 1235 SecondArgIsLastNamedArgument = PV == LastArg; 1236 } 1237 } 1238 1239 if (!SecondArgIsLastNamedArgument) 1240 Diag(TheCall->getArg(1)->getLocStart(), 1241 diag::warn_second_parameter_of_va_start_not_last_named_argument); 1242 return false; 1243} 1244 1245/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 1246/// friends. This is declared to take (...), so we have to check everything. 1247bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 1248 if (TheCall->getNumArgs() < 2) 1249 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1250 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 1251 if (TheCall->getNumArgs() > 2) 1252 return Diag(TheCall->getArg(2)->getLocStart(), 1253 diag::err_typecheck_call_too_many_args) 1254 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1255 << SourceRange(TheCall->getArg(2)->getLocStart(), 1256 (*(TheCall->arg_end()-1))->getLocEnd()); 1257 1258 ExprResult OrigArg0 = TheCall->getArg(0); 1259 ExprResult OrigArg1 = TheCall->getArg(1); 1260 1261 // Do standard promotions between the two arguments, returning their common 1262 // type. 1263 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 1264 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 1265 return true; 1266 1267 // Make sure any conversions are pushed back into the call; this is 1268 // type safe since unordered compare builtins are declared as "_Bool 1269 // foo(...)". 1270 TheCall->setArg(0, OrigArg0.get()); 1271 TheCall->setArg(1, OrigArg1.get()); 1272 1273 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 1274 return false; 1275 1276 // If the common type isn't a real floating type, then the arguments were 1277 // invalid for this operation. 1278 if (!Res->isRealFloatingType()) 1279 return Diag(OrigArg0.get()->getLocStart(), 1280 diag::err_typecheck_call_invalid_ordered_compare) 1281 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 1282 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 1283 1284 return false; 1285} 1286 1287/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 1288/// __builtin_isnan and friends. This is declared to take (...), so we have 1289/// to check everything. We expect the last argument to be a floating point 1290/// value. 1291bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 1292 if (TheCall->getNumArgs() < NumArgs) 1293 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1294 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 1295 if (TheCall->getNumArgs() > NumArgs) 1296 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 1297 diag::err_typecheck_call_too_many_args) 1298 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 1299 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 1300 (*(TheCall->arg_end()-1))->getLocEnd()); 1301 1302 Expr *OrigArg = TheCall->getArg(NumArgs-1); 1303 1304 if (OrigArg->isTypeDependent()) 1305 return false; 1306 1307 // This operation requires a non-_Complex floating-point number. 1308 if (!OrigArg->getType()->isRealFloatingType()) 1309 return Diag(OrigArg->getLocStart(), 1310 diag::err_typecheck_call_invalid_unary_fp) 1311 << OrigArg->getType() << OrigArg->getSourceRange(); 1312 1313 // If this is an implicit conversion from float -> double, remove it. 1314 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 1315 Expr *CastArg = Cast->getSubExpr(); 1316 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 1317 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 1318 "promotion from float to double is the only expected cast here"); 1319 Cast->setSubExpr(0); 1320 TheCall->setArg(NumArgs-1, CastArg); 1321 } 1322 } 1323 1324 return false; 1325} 1326 1327/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1328// This is declared to take (...), so we have to check everything. 1329ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1330 if (TheCall->getNumArgs() < 2) 1331 return ExprError(Diag(TheCall->getLocEnd(), 1332 diag::err_typecheck_call_too_few_args_at_least) 1333 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1334 << TheCall->getSourceRange()); 1335 1336 // Determine which of the following types of shufflevector we're checking: 1337 // 1) unary, vector mask: (lhs, mask) 1338 // 2) binary, vector mask: (lhs, rhs, mask) 1339 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1340 QualType resType = TheCall->getArg(0)->getType(); 1341 unsigned numElements = 0; 1342 1343 if (!TheCall->getArg(0)->isTypeDependent() && 1344 !TheCall->getArg(1)->isTypeDependent()) { 1345 QualType LHSType = TheCall->getArg(0)->getType(); 1346 QualType RHSType = TheCall->getArg(1)->getType(); 1347 1348 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 1349 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 1350 << SourceRange(TheCall->getArg(0)->getLocStart(), 1351 TheCall->getArg(1)->getLocEnd()); 1352 return ExprError(); 1353 } 1354 1355 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1356 unsigned numResElements = TheCall->getNumArgs() - 2; 1357 1358 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1359 // with mask. If so, verify that RHS is an integer vector type with the 1360 // same number of elts as lhs. 1361 if (TheCall->getNumArgs() == 2) { 1362 if (!RHSType->hasIntegerRepresentation() || 1363 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1364 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1365 << SourceRange(TheCall->getArg(1)->getLocStart(), 1366 TheCall->getArg(1)->getLocEnd()); 1367 numResElements = numElements; 1368 } 1369 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1370 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1371 << SourceRange(TheCall->getArg(0)->getLocStart(), 1372 TheCall->getArg(1)->getLocEnd()); 1373 return ExprError(); 1374 } else if (numElements != numResElements) { 1375 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1376 resType = Context.getVectorType(eltType, numResElements, 1377 VectorType::GenericVector); 1378 } 1379 } 1380 1381 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1382 if (TheCall->getArg(i)->isTypeDependent() || 1383 TheCall->getArg(i)->isValueDependent()) 1384 continue; 1385 1386 llvm::APSInt Result(32); 1387 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1388 return ExprError(Diag(TheCall->getLocStart(), 1389 diag::err_shufflevector_nonconstant_argument) 1390 << TheCall->getArg(i)->getSourceRange()); 1391 1392 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1393 return ExprError(Diag(TheCall->getLocStart(), 1394 diag::err_shufflevector_argument_too_large) 1395 << TheCall->getArg(i)->getSourceRange()); 1396 } 1397 1398 SmallVector<Expr*, 32> exprs; 1399 1400 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1401 exprs.push_back(TheCall->getArg(i)); 1402 TheCall->setArg(i, 0); 1403 } 1404 1405 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 1406 exprs.size(), resType, 1407 TheCall->getCallee()->getLocStart(), 1408 TheCall->getRParenLoc())); 1409} 1410 1411/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1412// This is declared to take (const void*, ...) and can take two 1413// optional constant int args. 1414bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1415 unsigned NumArgs = TheCall->getNumArgs(); 1416 1417 if (NumArgs > 3) 1418 return Diag(TheCall->getLocEnd(), 1419 diag::err_typecheck_call_too_many_args_at_most) 1420 << 0 /*function call*/ << 3 << NumArgs 1421 << TheCall->getSourceRange(); 1422 1423 // Argument 0 is checked for us and the remaining arguments must be 1424 // constant integers. 1425 for (unsigned i = 1; i != NumArgs; ++i) { 1426 Expr *Arg = TheCall->getArg(i); 1427 1428 llvm::APSInt Result; 1429 if (SemaBuiltinConstantArg(TheCall, i, Result)) 1430 return true; 1431 1432 // FIXME: gcc issues a warning and rewrites these to 0. These 1433 // seems especially odd for the third argument since the default 1434 // is 3. 1435 if (i == 1) { 1436 if (Result.getLimitedValue() > 1) 1437 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1438 << "0" << "1" << Arg->getSourceRange(); 1439 } else { 1440 if (Result.getLimitedValue() > 3) 1441 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1442 << "0" << "3" << Arg->getSourceRange(); 1443 } 1444 } 1445 1446 return false; 1447} 1448 1449/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1450/// TheCall is a constant expression. 1451bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1452 llvm::APSInt &Result) { 1453 Expr *Arg = TheCall->getArg(ArgNum); 1454 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1455 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1456 1457 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 1458 1459 if (!Arg->isIntegerConstantExpr(Result, Context)) 1460 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 1461 << FDecl->getDeclName() << Arg->getSourceRange(); 1462 1463 return false; 1464} 1465 1466/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 1467/// int type). This simply type checks that type is one of the defined 1468/// constants (0-3). 1469// For compatibility check 0-3, llvm only handles 0 and 2. 1470bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 1471 llvm::APSInt Result; 1472 1473 // Check constant-ness first. 1474 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1475 return true; 1476 1477 Expr *Arg = TheCall->getArg(1); 1478 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 1479 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1480 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1481 } 1482 1483 return false; 1484} 1485 1486/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1487/// This checks that val is a constant 1. 1488bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1489 Expr *Arg = TheCall->getArg(1); 1490 llvm::APSInt Result; 1491 1492 // TODO: This is less than ideal. Overload this to take a value. 1493 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1494 return true; 1495 1496 if (Result != 1) 1497 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1498 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1499 1500 return false; 1501} 1502 1503// Handle i > 1 ? "x" : "y", recursively. 1504bool Sema::SemaCheckStringLiteral(const Expr *E, Expr **Args, 1505 unsigned NumArgs, bool HasVAListArg, 1506 unsigned format_idx, unsigned firstDataArg, 1507 FormatStringType Type, bool inFunctionCall) { 1508 tryAgain: 1509 if (E->isTypeDependent() || E->isValueDependent()) 1510 return false; 1511 1512 E = E->IgnoreParenCasts(); 1513 1514 if (E->isNullPointerConstant(Context, Expr::NPC_ValueDependentIsNotNull)) 1515 // Technically -Wformat-nonliteral does not warn about this case. 1516 // The behavior of printf and friends in this case is implementation 1517 // dependent. Ideally if the format string cannot be null then 1518 // it should have a 'nonnull' attribute in the function prototype. 1519 return true; 1520 1521 switch (E->getStmtClass()) { 1522 case Stmt::BinaryConditionalOperatorClass: 1523 case Stmt::ConditionalOperatorClass: { 1524 const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E); 1525 return SemaCheckStringLiteral(C->getTrueExpr(), Args, NumArgs, HasVAListArg, 1526 format_idx, firstDataArg, Type, 1527 inFunctionCall) 1528 && SemaCheckStringLiteral(C->getFalseExpr(), Args, NumArgs, HasVAListArg, 1529 format_idx, firstDataArg, Type, 1530 inFunctionCall); 1531 } 1532 1533 case Stmt::ImplicitCastExprClass: { 1534 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 1535 goto tryAgain; 1536 } 1537 1538 case Stmt::OpaqueValueExprClass: 1539 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 1540 E = src; 1541 goto tryAgain; 1542 } 1543 return false; 1544 1545 case Stmt::PredefinedExprClass: 1546 // While __func__, etc., are technically not string literals, they 1547 // cannot contain format specifiers and thus are not a security 1548 // liability. 1549 return true; 1550 1551 case Stmt::DeclRefExprClass: { 1552 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1553 1554 // As an exception, do not flag errors for variables binding to 1555 // const string literals. 1556 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1557 bool isConstant = false; 1558 QualType T = DR->getType(); 1559 1560 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1561 isConstant = AT->getElementType().isConstant(Context); 1562 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1563 isConstant = T.isConstant(Context) && 1564 PT->getPointeeType().isConstant(Context); 1565 } else if (T->isObjCObjectPointerType()) { 1566 // In ObjC, there is usually no "const ObjectPointer" type, 1567 // so don't check if the pointee type is constant. 1568 isConstant = T.isConstant(Context); 1569 } 1570 1571 if (isConstant) { 1572 if (const Expr *Init = VD->getAnyInitializer()) 1573 return SemaCheckStringLiteral(Init, Args, NumArgs, 1574 HasVAListArg, format_idx, firstDataArg, 1575 Type, /*inFunctionCall*/false); 1576 } 1577 1578 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1579 // special check to see if the format string is a function parameter 1580 // of the function calling the printf function. If the function 1581 // has an attribute indicating it is a printf-like function, then we 1582 // should suppress warnings concerning non-literals being used in a call 1583 // to a vprintf function. For example: 1584 // 1585 // void 1586 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1587 // va_list ap; 1588 // va_start(ap, fmt); 1589 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1590 // ... 1591 // 1592 if (HasVAListArg) { 1593 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(VD)) { 1594 if (const NamedDecl *ND = dyn_cast<NamedDecl>(PV->getDeclContext())) { 1595 int PVIndex = PV->getFunctionScopeIndex() + 1; 1596 for (specific_attr_iterator<FormatAttr> 1597 i = ND->specific_attr_begin<FormatAttr>(), 1598 e = ND->specific_attr_end<FormatAttr>(); i != e ; ++i) { 1599 FormatAttr *PVFormat = *i; 1600 // adjust for implicit parameter 1601 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1602 if (MD->isInstance()) 1603 ++PVIndex; 1604 // We also check if the formats are compatible. 1605 // We can't pass a 'scanf' string to a 'printf' function. 1606 if (PVIndex == PVFormat->getFormatIdx() && 1607 Type == GetFormatStringType(PVFormat)) 1608 return true; 1609 } 1610 } 1611 } 1612 } 1613 } 1614 1615 return false; 1616 } 1617 1618 case Stmt::CallExprClass: 1619 case Stmt::CXXMemberCallExprClass: { 1620 const CallExpr *CE = cast<CallExpr>(E); 1621 if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(CE->getCalleeDecl())) { 1622 if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) { 1623 unsigned ArgIndex = FA->getFormatIdx(); 1624 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1625 if (MD->isInstance()) 1626 --ArgIndex; 1627 const Expr *Arg = CE->getArg(ArgIndex - 1); 1628 1629 return SemaCheckStringLiteral(Arg, Args, NumArgs, HasVAListArg, 1630 format_idx, firstDataArg, Type, 1631 inFunctionCall); 1632 } 1633 } 1634 1635 return false; 1636 } 1637 case Stmt::ObjCStringLiteralClass: 1638 case Stmt::StringLiteralClass: { 1639 const StringLiteral *StrE = NULL; 1640 1641 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1642 StrE = ObjCFExpr->getString(); 1643 else 1644 StrE = cast<StringLiteral>(E); 1645 1646 if (StrE) { 1647 CheckFormatString(StrE, E, Args, NumArgs, HasVAListArg, format_idx, 1648 firstDataArg, Type, inFunctionCall); 1649 return true; 1650 } 1651 1652 return false; 1653 } 1654 1655 default: 1656 return false; 1657 } 1658} 1659 1660void 1661Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1662 const Expr * const *ExprArgs, 1663 SourceLocation CallSiteLoc) { 1664 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1665 e = NonNull->args_end(); 1666 i != e; ++i) { 1667 const Expr *ArgExpr = ExprArgs[*i]; 1668 if (ArgExpr->isNullPointerConstant(Context, 1669 Expr::NPC_ValueDependentIsNotNull)) 1670 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 1671 } 1672} 1673 1674Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) { 1675 return llvm::StringSwitch<FormatStringType>(Format->getType()) 1676 .Case("scanf", FST_Scanf) 1677 .Cases("printf", "printf0", FST_Printf) 1678 .Cases("NSString", "CFString", FST_NSString) 1679 .Case("strftime", FST_Strftime) 1680 .Case("strfmon", FST_Strfmon) 1681 .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf) 1682 .Default(FST_Unknown); 1683} 1684 1685/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar 1686/// functions) for correct use of format strings. 1687void Sema::CheckFormatArguments(const FormatAttr *Format, CallExpr *TheCall) { 1688 bool IsCXXMember = false; 1689 // The way the format attribute works in GCC, the implicit this argument 1690 // of member functions is counted. However, it doesn't appear in our own 1691 // lists, so decrement format_idx in that case. 1692 IsCXXMember = isa<CXXMemberCallExpr>(TheCall); 1693 CheckFormatArguments(Format, TheCall->getArgs(), TheCall->getNumArgs(), 1694 IsCXXMember, TheCall->getRParenLoc(), 1695 TheCall->getCallee()->getSourceRange()); 1696} 1697 1698void Sema::CheckFormatArguments(const FormatAttr *Format, Expr **Args, 1699 unsigned NumArgs, bool IsCXXMember, 1700 SourceLocation Loc, SourceRange Range) { 1701 bool HasVAListArg = Format->getFirstArg() == 0; 1702 unsigned format_idx = Format->getFormatIdx() - 1; 1703 unsigned firstDataArg = HasVAListArg ? 0 : Format->getFirstArg() - 1; 1704 if (IsCXXMember) { 1705 if (format_idx == 0) 1706 return; 1707 --format_idx; 1708 if(firstDataArg != 0) 1709 --firstDataArg; 1710 } 1711 CheckFormatArguments(Args, NumArgs, HasVAListArg, format_idx, 1712 firstDataArg, GetFormatStringType(Format), Loc, Range); 1713} 1714 1715void Sema::CheckFormatArguments(Expr **Args, unsigned NumArgs, 1716 bool HasVAListArg, unsigned format_idx, 1717 unsigned firstDataArg, FormatStringType Type, 1718 SourceLocation Loc, SourceRange Range) { 1719 // CHECK: printf/scanf-like function is called with no format string. 1720 if (format_idx >= NumArgs) { 1721 Diag(Loc, diag::warn_missing_format_string) << Range; 1722 return; 1723 } 1724 1725 const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts(); 1726 1727 // CHECK: format string is not a string literal. 1728 // 1729 // Dynamically generated format strings are difficult to 1730 // automatically vet at compile time. Requiring that format strings 1731 // are string literals: (1) permits the checking of format strings by 1732 // the compiler and thereby (2) can practically remove the source of 1733 // many format string exploits. 1734 1735 // Format string can be either ObjC string (e.g. @"%d") or 1736 // C string (e.g. "%d") 1737 // ObjC string uses the same format specifiers as C string, so we can use 1738 // the same format string checking logic for both ObjC and C strings. 1739 if (SemaCheckStringLiteral(OrigFormatExpr, Args, NumArgs, HasVAListArg, 1740 format_idx, firstDataArg, Type)) 1741 return; // Literal format string found, check done! 1742 1743 // Strftime is particular as it always uses a single 'time' argument, 1744 // so it is safe to pass a non-literal string. 1745 if (Type == FST_Strftime) 1746 return; 1747 1748 // Do not emit diag when the string param is a macro expansion and the 1749 // format is either NSString or CFString. This is a hack to prevent 1750 // diag when using the NSLocalizedString and CFCopyLocalizedString macros 1751 // which are usually used in place of NS and CF string literals. 1752 if (Type == FST_NSString && Args[format_idx]->getLocStart().isMacroID()) 1753 return; 1754 1755 // If there are no arguments specified, warn with -Wformat-security, otherwise 1756 // warn only with -Wformat-nonliteral. 1757 if (NumArgs == format_idx+1) 1758 Diag(Args[format_idx]->getLocStart(), 1759 diag::warn_format_nonliteral_noargs) 1760 << OrigFormatExpr->getSourceRange(); 1761 else 1762 Diag(Args[format_idx]->getLocStart(), 1763 diag::warn_format_nonliteral) 1764 << OrigFormatExpr->getSourceRange(); 1765} 1766 1767namespace { 1768class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1769protected: 1770 Sema &S; 1771 const StringLiteral *FExpr; 1772 const Expr *OrigFormatExpr; 1773 const unsigned FirstDataArg; 1774 const unsigned NumDataArgs; 1775 const bool IsObjCLiteral; 1776 const char *Beg; // Start of format string. 1777 const bool HasVAListArg; 1778 const Expr * const *Args; 1779 const unsigned NumArgs; 1780 unsigned FormatIdx; 1781 llvm::BitVector CoveredArgs; 1782 bool usesPositionalArgs; 1783 bool atFirstArg; 1784 bool inFunctionCall; 1785public: 1786 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1787 const Expr *origFormatExpr, unsigned firstDataArg, 1788 unsigned numDataArgs, bool isObjCLiteral, 1789 const char *beg, bool hasVAListArg, 1790 Expr **args, unsigned numArgs, 1791 unsigned formatIdx, bool inFunctionCall) 1792 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1793 FirstDataArg(firstDataArg), 1794 NumDataArgs(numDataArgs), 1795 IsObjCLiteral(isObjCLiteral), Beg(beg), 1796 HasVAListArg(hasVAListArg), 1797 Args(args), NumArgs(numArgs), FormatIdx(formatIdx), 1798 usesPositionalArgs(false), atFirstArg(true), 1799 inFunctionCall(inFunctionCall) { 1800 CoveredArgs.resize(numDataArgs); 1801 CoveredArgs.reset(); 1802 } 1803 1804 void DoneProcessing(); 1805 1806 void HandleIncompleteSpecifier(const char *startSpecifier, 1807 unsigned specifierLen); 1808 1809 void HandleNonStandardLengthModifier( 1810 const analyze_format_string::LengthModifier &LM, 1811 const char *startSpecifier, unsigned specifierLen); 1812 1813 void HandleNonStandardConversionSpecifier( 1814 const analyze_format_string::ConversionSpecifier &CS, 1815 const char *startSpecifier, unsigned specifierLen); 1816 1817 void HandleNonStandardConversionSpecification( 1818 const analyze_format_string::LengthModifier &LM, 1819 const analyze_format_string::ConversionSpecifier &CS, 1820 const char *startSpecifier, unsigned specifierLen); 1821 1822 virtual void HandlePosition(const char *startPos, unsigned posLen); 1823 1824 virtual void HandleInvalidPosition(const char *startSpecifier, 1825 unsigned specifierLen, 1826 analyze_format_string::PositionContext p); 1827 1828 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1829 1830 void HandleNullChar(const char *nullCharacter); 1831 1832 template <typename Range> 1833 static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall, 1834 const Expr *ArgumentExpr, 1835 PartialDiagnostic PDiag, 1836 SourceLocation StringLoc, 1837 bool IsStringLocation, Range StringRange, 1838 FixItHint Fixit = FixItHint()); 1839 1840protected: 1841 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1842 const char *startSpec, 1843 unsigned specifierLen, 1844 const char *csStart, unsigned csLen); 1845 1846 void HandlePositionalNonpositionalArgs(SourceLocation Loc, 1847 const char *startSpec, 1848 unsigned specifierLen); 1849 1850 SourceRange getFormatStringRange(); 1851 CharSourceRange getSpecifierRange(const char *startSpecifier, 1852 unsigned specifierLen); 1853 SourceLocation getLocationOfByte(const char *x); 1854 1855 const Expr *getDataArg(unsigned i) const; 1856 1857 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1858 const analyze_format_string::ConversionSpecifier &CS, 1859 const char *startSpecifier, unsigned specifierLen, 1860 unsigned argIndex); 1861 1862 template <typename Range> 1863 void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc, 1864 bool IsStringLocation, Range StringRange, 1865 FixItHint Fixit = FixItHint()); 1866 1867 void CheckPositionalAndNonpositionalArgs( 1868 const analyze_format_string::FormatSpecifier *FS); 1869}; 1870} 1871 1872SourceRange CheckFormatHandler::getFormatStringRange() { 1873 return OrigFormatExpr->getSourceRange(); 1874} 1875 1876CharSourceRange CheckFormatHandler:: 1877getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1878 SourceLocation Start = getLocationOfByte(startSpecifier); 1879 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1880 1881 // Advance the end SourceLocation by one due to half-open ranges. 1882 End = End.getLocWithOffset(1); 1883 1884 return CharSourceRange::getCharRange(Start, End); 1885} 1886 1887SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 1888 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1889} 1890 1891void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 1892 unsigned specifierLen){ 1893 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier), 1894 getLocationOfByte(startSpecifier), 1895 /*IsStringLocation*/true, 1896 getSpecifierRange(startSpecifier, specifierLen)); 1897} 1898 1899void CheckFormatHandler::HandleNonStandardLengthModifier( 1900 const analyze_format_string::LengthModifier &LM, 1901 const char *startSpecifier, unsigned specifierLen) { 1902 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) << LM.toString() 1903 << 0, 1904 getLocationOfByte(LM.getStart()), 1905 /*IsStringLocation*/true, 1906 getSpecifierRange(startSpecifier, specifierLen)); 1907} 1908 1909void CheckFormatHandler::HandleNonStandardConversionSpecifier( 1910 const analyze_format_string::ConversionSpecifier &CS, 1911 const char *startSpecifier, unsigned specifierLen) { 1912 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) << CS.toString() 1913 << 1, 1914 getLocationOfByte(CS.getStart()), 1915 /*IsStringLocation*/true, 1916 getSpecifierRange(startSpecifier, specifierLen)); 1917} 1918 1919void CheckFormatHandler::HandleNonStandardConversionSpecification( 1920 const analyze_format_string::LengthModifier &LM, 1921 const analyze_format_string::ConversionSpecifier &CS, 1922 const char *startSpecifier, unsigned specifierLen) { 1923 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_conversion_spec) 1924 << LM.toString() << CS.toString(), 1925 getLocationOfByte(LM.getStart()), 1926 /*IsStringLocation*/true, 1927 getSpecifierRange(startSpecifier, specifierLen)); 1928} 1929 1930void CheckFormatHandler::HandlePosition(const char *startPos, 1931 unsigned posLen) { 1932 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_positional_arg), 1933 getLocationOfByte(startPos), 1934 /*IsStringLocation*/true, 1935 getSpecifierRange(startPos, posLen)); 1936} 1937 1938void 1939CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1940 analyze_format_string::PositionContext p) { 1941 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier) 1942 << (unsigned) p, 1943 getLocationOfByte(startPos), /*IsStringLocation*/true, 1944 getSpecifierRange(startPos, posLen)); 1945} 1946 1947void CheckFormatHandler::HandleZeroPosition(const char *startPos, 1948 unsigned posLen) { 1949 EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier), 1950 getLocationOfByte(startPos), 1951 /*IsStringLocation*/true, 1952 getSpecifierRange(startPos, posLen)); 1953} 1954 1955void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 1956 if (!IsObjCLiteral) { 1957 // The presence of a null character is likely an error. 1958 EmitFormatDiagnostic( 1959 S.PDiag(diag::warn_printf_format_string_contains_null_char), 1960 getLocationOfByte(nullCharacter), /*IsStringLocation*/true, 1961 getFormatStringRange()); 1962 } 1963} 1964 1965const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 1966 return Args[FirstDataArg + i]; 1967} 1968 1969void CheckFormatHandler::DoneProcessing() { 1970 // Does the number of data arguments exceed the number of 1971 // format conversions in the format string? 1972 if (!HasVAListArg) { 1973 // Find any arguments that weren't covered. 1974 CoveredArgs.flip(); 1975 signed notCoveredArg = CoveredArgs.find_first(); 1976 if (notCoveredArg >= 0) { 1977 assert((unsigned)notCoveredArg < NumDataArgs); 1978 SourceLocation Loc = getDataArg((unsigned) notCoveredArg)->getLocStart(); 1979 if (!S.getSourceManager().isInSystemMacro(Loc)) { 1980 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used), 1981 Loc, 1982 /*IsStringLocation*/false, getFormatStringRange()); 1983 } 1984 } 1985 } 1986} 1987 1988bool 1989CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 1990 SourceLocation Loc, 1991 const char *startSpec, 1992 unsigned specifierLen, 1993 const char *csStart, 1994 unsigned csLen) { 1995 1996 bool keepGoing = true; 1997 if (argIndex < NumDataArgs) { 1998 // Consider the argument coverered, even though the specifier doesn't 1999 // make sense. 2000 CoveredArgs.set(argIndex); 2001 } 2002 else { 2003 // If argIndex exceeds the number of data arguments we 2004 // don't issue a warning because that is just a cascade of warnings (and 2005 // they may have intended '%%' anyway). We don't want to continue processing 2006 // the format string after this point, however, as we will like just get 2007 // gibberish when trying to match arguments. 2008 keepGoing = false; 2009 } 2010 2011 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion) 2012 << StringRef(csStart, csLen), 2013 Loc, /*IsStringLocation*/true, 2014 getSpecifierRange(startSpec, specifierLen)); 2015 2016 return keepGoing; 2017} 2018 2019void 2020CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc, 2021 const char *startSpec, 2022 unsigned specifierLen) { 2023 EmitFormatDiagnostic( 2024 S.PDiag(diag::warn_format_mix_positional_nonpositional_args), 2025 Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen)); 2026} 2027 2028bool 2029CheckFormatHandler::CheckNumArgs( 2030 const analyze_format_string::FormatSpecifier &FS, 2031 const analyze_format_string::ConversionSpecifier &CS, 2032 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 2033 2034 if (argIndex >= NumDataArgs) { 2035 PartialDiagnostic PDiag = FS.usesPositionalArg() 2036 ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args) 2037 << (argIndex+1) << NumDataArgs) 2038 : S.PDiag(diag::warn_printf_insufficient_data_args); 2039 EmitFormatDiagnostic( 2040 PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true, 2041 getSpecifierRange(startSpecifier, specifierLen)); 2042 return false; 2043 } 2044 return true; 2045} 2046 2047template<typename Range> 2048void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag, 2049 SourceLocation Loc, 2050 bool IsStringLocation, 2051 Range StringRange, 2052 FixItHint FixIt) { 2053 EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag, 2054 Loc, IsStringLocation, StringRange, FixIt); 2055} 2056 2057/// \brief If the format string is not within the funcion call, emit a note 2058/// so that the function call and string are in diagnostic messages. 2059/// 2060/// \param inFunctionCall if true, the format string is within the function 2061/// call and only one diagnostic message will be produced. Otherwise, an 2062/// extra note will be emitted pointing to location of the format string. 2063/// 2064/// \param ArgumentExpr the expression that is passed as the format string 2065/// argument in the function call. Used for getting locations when two 2066/// diagnostics are emitted. 2067/// 2068/// \param PDiag the callee should already have provided any strings for the 2069/// diagnostic message. This function only adds locations and fixits 2070/// to diagnostics. 2071/// 2072/// \param Loc primary location for diagnostic. If two diagnostics are 2073/// required, one will be at Loc and a new SourceLocation will be created for 2074/// the other one. 2075/// 2076/// \param IsStringLocation if true, Loc points to the format string should be 2077/// used for the note. Otherwise, Loc points to the argument list and will 2078/// be used with PDiag. 2079/// 2080/// \param StringRange some or all of the string to highlight. This is 2081/// templated so it can accept either a CharSourceRange or a SourceRange. 2082/// 2083/// \param Fixit optional fix it hint for the format string. 2084template<typename Range> 2085void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall, 2086 const Expr *ArgumentExpr, 2087 PartialDiagnostic PDiag, 2088 SourceLocation Loc, 2089 bool IsStringLocation, 2090 Range StringRange, 2091 FixItHint FixIt) { 2092 if (InFunctionCall) 2093 S.Diag(Loc, PDiag) << StringRange << FixIt; 2094 else { 2095 S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag) 2096 << ArgumentExpr->getSourceRange(); 2097 S.Diag(IsStringLocation ? Loc : StringRange.getBegin(), 2098 diag::note_format_string_defined) 2099 << StringRange << FixIt; 2100 } 2101} 2102 2103//===--- CHECK: Printf format string checking ------------------------------===// 2104 2105namespace { 2106class CheckPrintfHandler : public CheckFormatHandler { 2107public: 2108 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 2109 const Expr *origFormatExpr, unsigned firstDataArg, 2110 unsigned numDataArgs, bool isObjCLiteral, 2111 const char *beg, bool hasVAListArg, 2112 Expr **Args, unsigned NumArgs, 2113 unsigned formatIdx, bool inFunctionCall) 2114 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2115 numDataArgs, isObjCLiteral, beg, hasVAListArg, 2116 Args, NumArgs, formatIdx, inFunctionCall) {} 2117 2118 2119 bool HandleInvalidPrintfConversionSpecifier( 2120 const analyze_printf::PrintfSpecifier &FS, 2121 const char *startSpecifier, 2122 unsigned specifierLen); 2123 2124 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 2125 const char *startSpecifier, 2126 unsigned specifierLen); 2127 2128 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 2129 const char *startSpecifier, unsigned specifierLen); 2130 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 2131 const analyze_printf::OptionalAmount &Amt, 2132 unsigned type, 2133 const char *startSpecifier, unsigned specifierLen); 2134 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2135 const analyze_printf::OptionalFlag &flag, 2136 const char *startSpecifier, unsigned specifierLen); 2137 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 2138 const analyze_printf::OptionalFlag &ignoredFlag, 2139 const analyze_printf::OptionalFlag &flag, 2140 const char *startSpecifier, unsigned specifierLen); 2141}; 2142} 2143 2144bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 2145 const analyze_printf::PrintfSpecifier &FS, 2146 const char *startSpecifier, 2147 unsigned specifierLen) { 2148 const analyze_printf::PrintfConversionSpecifier &CS = 2149 FS.getConversionSpecifier(); 2150 2151 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2152 getLocationOfByte(CS.getStart()), 2153 startSpecifier, specifierLen, 2154 CS.getStart(), CS.getLength()); 2155} 2156 2157bool CheckPrintfHandler::HandleAmount( 2158 const analyze_format_string::OptionalAmount &Amt, 2159 unsigned k, const char *startSpecifier, 2160 unsigned specifierLen) { 2161 2162 if (Amt.hasDataArgument()) { 2163 if (!HasVAListArg) { 2164 unsigned argIndex = Amt.getArgIndex(); 2165 if (argIndex >= NumDataArgs) { 2166 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg) 2167 << k, 2168 getLocationOfByte(Amt.getStart()), 2169 /*IsStringLocation*/true, 2170 getSpecifierRange(startSpecifier, specifierLen)); 2171 // Don't do any more checking. We will just emit 2172 // spurious errors. 2173 return false; 2174 } 2175 2176 // Type check the data argument. It should be an 'int'. 2177 // Although not in conformance with C99, we also allow the argument to be 2178 // an 'unsigned int' as that is a reasonably safe case. GCC also 2179 // doesn't emit a warning for that case. 2180 CoveredArgs.set(argIndex); 2181 const Expr *Arg = getDataArg(argIndex); 2182 QualType T = Arg->getType(); 2183 2184 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 2185 assert(ATR.isValid()); 2186 2187 if (!ATR.matchesType(S.Context, T)) { 2188 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type) 2189 << k << ATR.getRepresentativeTypeName(S.Context) 2190 << T << Arg->getSourceRange(), 2191 getLocationOfByte(Amt.getStart()), 2192 /*IsStringLocation*/true, 2193 getSpecifierRange(startSpecifier, specifierLen)); 2194 // Don't do any more checking. We will just emit 2195 // spurious errors. 2196 return false; 2197 } 2198 } 2199 } 2200 return true; 2201} 2202 2203void CheckPrintfHandler::HandleInvalidAmount( 2204 const analyze_printf::PrintfSpecifier &FS, 2205 const analyze_printf::OptionalAmount &Amt, 2206 unsigned type, 2207 const char *startSpecifier, 2208 unsigned specifierLen) { 2209 const analyze_printf::PrintfConversionSpecifier &CS = 2210 FS.getConversionSpecifier(); 2211 2212 FixItHint fixit = 2213 Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant 2214 ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 2215 Amt.getConstantLength())) 2216 : FixItHint(); 2217 2218 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount) 2219 << type << CS.toString(), 2220 getLocationOfByte(Amt.getStart()), 2221 /*IsStringLocation*/true, 2222 getSpecifierRange(startSpecifier, specifierLen), 2223 fixit); 2224} 2225 2226void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2227 const analyze_printf::OptionalFlag &flag, 2228 const char *startSpecifier, 2229 unsigned specifierLen) { 2230 // Warn about pointless flag with a fixit removal. 2231 const analyze_printf::PrintfConversionSpecifier &CS = 2232 FS.getConversionSpecifier(); 2233 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag) 2234 << flag.toString() << CS.toString(), 2235 getLocationOfByte(flag.getPosition()), 2236 /*IsStringLocation*/true, 2237 getSpecifierRange(startSpecifier, specifierLen), 2238 FixItHint::CreateRemoval( 2239 getSpecifierRange(flag.getPosition(), 1))); 2240} 2241 2242void CheckPrintfHandler::HandleIgnoredFlag( 2243 const analyze_printf::PrintfSpecifier &FS, 2244 const analyze_printf::OptionalFlag &ignoredFlag, 2245 const analyze_printf::OptionalFlag &flag, 2246 const char *startSpecifier, 2247 unsigned specifierLen) { 2248 // Warn about ignored flag with a fixit removal. 2249 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag) 2250 << ignoredFlag.toString() << flag.toString(), 2251 getLocationOfByte(ignoredFlag.getPosition()), 2252 /*IsStringLocation*/true, 2253 getSpecifierRange(startSpecifier, specifierLen), 2254 FixItHint::CreateRemoval( 2255 getSpecifierRange(ignoredFlag.getPosition(), 1))); 2256} 2257 2258bool 2259CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 2260 &FS, 2261 const char *startSpecifier, 2262 unsigned specifierLen) { 2263 2264 using namespace analyze_format_string; 2265 using namespace analyze_printf; 2266 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 2267 2268 if (FS.consumesDataArgument()) { 2269 if (atFirstArg) { 2270 atFirstArg = false; 2271 usesPositionalArgs = FS.usesPositionalArg(); 2272 } 2273 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2274 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2275 startSpecifier, specifierLen); 2276 return false; 2277 } 2278 } 2279 2280 // First check if the field width, precision, and conversion specifier 2281 // have matching data arguments. 2282 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 2283 startSpecifier, specifierLen)) { 2284 return false; 2285 } 2286 2287 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 2288 startSpecifier, specifierLen)) { 2289 return false; 2290 } 2291 2292 if (!CS.consumesDataArgument()) { 2293 // FIXME: Technically specifying a precision or field width here 2294 // makes no sense. Worth issuing a warning at some point. 2295 return true; 2296 } 2297 2298 // Consume the argument. 2299 unsigned argIndex = FS.getArgIndex(); 2300 if (argIndex < NumDataArgs) { 2301 // The check to see if the argIndex is valid will come later. 2302 // We set the bit here because we may exit early from this 2303 // function if we encounter some other error. 2304 CoveredArgs.set(argIndex); 2305 } 2306 2307 // Check for using an Objective-C specific conversion specifier 2308 // in a non-ObjC literal. 2309 if (!IsObjCLiteral && CS.isObjCArg()) { 2310 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 2311 specifierLen); 2312 } 2313 2314 // Check for invalid use of field width 2315 if (!FS.hasValidFieldWidth()) { 2316 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 2317 startSpecifier, specifierLen); 2318 } 2319 2320 // Check for invalid use of precision 2321 if (!FS.hasValidPrecision()) { 2322 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 2323 startSpecifier, specifierLen); 2324 } 2325 2326 // Check each flag does not conflict with any other component. 2327 if (!FS.hasValidThousandsGroupingPrefix()) 2328 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 2329 if (!FS.hasValidLeadingZeros()) 2330 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 2331 if (!FS.hasValidPlusPrefix()) 2332 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 2333 if (!FS.hasValidSpacePrefix()) 2334 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 2335 if (!FS.hasValidAlternativeForm()) 2336 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 2337 if (!FS.hasValidLeftJustified()) 2338 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 2339 2340 // Check that flags are not ignored by another flag 2341 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 2342 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 2343 startSpecifier, specifierLen); 2344 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 2345 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 2346 startSpecifier, specifierLen); 2347 2348 // Check the length modifier is valid with the given conversion specifier. 2349 const LengthModifier &LM = FS.getLengthModifier(); 2350 if (!FS.hasValidLengthModifier()) 2351 EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length) 2352 << LM.toString() << CS.toString(), 2353 getLocationOfByte(LM.getStart()), 2354 /*IsStringLocation*/true, 2355 getSpecifierRange(startSpecifier, specifierLen), 2356 FixItHint::CreateRemoval( 2357 getSpecifierRange(LM.getStart(), 2358 LM.getLength()))); 2359 if (!FS.hasStandardLengthModifier()) 2360 HandleNonStandardLengthModifier(LM, startSpecifier, specifierLen); 2361 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 2362 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 2363 if (!FS.hasStandardLengthConversionCombination()) 2364 HandleNonStandardConversionSpecification(LM, CS, startSpecifier, 2365 specifierLen); 2366 2367 // Are we using '%n'? 2368 if (CS.getKind() == ConversionSpecifier::nArg) { 2369 // Issue a warning about this being a possible security issue. 2370 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_write_back), 2371 getLocationOfByte(CS.getStart()), 2372 /*IsStringLocation*/true, 2373 getSpecifierRange(startSpecifier, specifierLen)); 2374 // Continue checking the other format specifiers. 2375 return true; 2376 } 2377 2378 // The remaining checks depend on the data arguments. 2379 if (HasVAListArg) 2380 return true; 2381 2382 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2383 return false; 2384 2385 // Now type check the data expression that matches the 2386 // format specifier. 2387 const Expr *Ex = getDataArg(argIndex); 2388 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context, 2389 IsObjCLiteral); 2390 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 2391 // Check if we didn't match because of an implicit cast from a 'char' 2392 // or 'short' to an 'int'. This is done because printf is a varargs 2393 // function. 2394 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 2395 if (ICE->getType() == S.Context.IntTy) { 2396 // All further checking is done on the subexpression. 2397 Ex = ICE->getSubExpr(); 2398 if (ATR.matchesType(S.Context, Ex->getType())) 2399 return true; 2400 } 2401 2402 // We may be able to offer a FixItHint if it is a supported type. 2403 PrintfSpecifier fixedFS = FS; 2404 bool success = fixedFS.fixType(Ex->getType(), S.getLangOpts(), 2405 S.Context, IsObjCLiteral); 2406 2407 if (success) { 2408 // Get the fix string from the fixed format specifier 2409 SmallString<128> buf; 2410 llvm::raw_svector_ostream os(buf); 2411 fixedFS.toString(os); 2412 2413 EmitFormatDiagnostic( 2414 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2415 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2416 << Ex->getSourceRange(), 2417 getLocationOfByte(CS.getStart()), 2418 /*IsStringLocation*/true, 2419 getSpecifierRange(startSpecifier, specifierLen), 2420 FixItHint::CreateReplacement( 2421 getSpecifierRange(startSpecifier, specifierLen), 2422 os.str())); 2423 } 2424 else { 2425 EmitFormatDiagnostic( 2426 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2427 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2428 << getSpecifierRange(startSpecifier, specifierLen) 2429 << Ex->getSourceRange(), 2430 getLocationOfByte(CS.getStart()), 2431 true, 2432 getSpecifierRange(startSpecifier, specifierLen)); 2433 } 2434 } 2435 2436 return true; 2437} 2438 2439//===--- CHECK: Scanf format string checking ------------------------------===// 2440 2441namespace { 2442class CheckScanfHandler : public CheckFormatHandler { 2443public: 2444 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 2445 const Expr *origFormatExpr, unsigned firstDataArg, 2446 unsigned numDataArgs, bool isObjCLiteral, 2447 const char *beg, bool hasVAListArg, 2448 Expr **Args, unsigned NumArgs, 2449 unsigned formatIdx, bool inFunctionCall) 2450 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2451 numDataArgs, isObjCLiteral, beg, hasVAListArg, 2452 Args, NumArgs, formatIdx, inFunctionCall) {} 2453 2454 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 2455 const char *startSpecifier, 2456 unsigned specifierLen); 2457 2458 bool HandleInvalidScanfConversionSpecifier( 2459 const analyze_scanf::ScanfSpecifier &FS, 2460 const char *startSpecifier, 2461 unsigned specifierLen); 2462 2463 void HandleIncompleteScanList(const char *start, const char *end); 2464}; 2465} 2466 2467void CheckScanfHandler::HandleIncompleteScanList(const char *start, 2468 const char *end) { 2469 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 2470 getLocationOfByte(end), /*IsStringLocation*/true, 2471 getSpecifierRange(start, end - start)); 2472} 2473 2474bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 2475 const analyze_scanf::ScanfSpecifier &FS, 2476 const char *startSpecifier, 2477 unsigned specifierLen) { 2478 2479 const analyze_scanf::ScanfConversionSpecifier &CS = 2480 FS.getConversionSpecifier(); 2481 2482 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2483 getLocationOfByte(CS.getStart()), 2484 startSpecifier, specifierLen, 2485 CS.getStart(), CS.getLength()); 2486} 2487 2488bool CheckScanfHandler::HandleScanfSpecifier( 2489 const analyze_scanf::ScanfSpecifier &FS, 2490 const char *startSpecifier, 2491 unsigned specifierLen) { 2492 2493 using namespace analyze_scanf; 2494 using namespace analyze_format_string; 2495 2496 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 2497 2498 // Handle case where '%' and '*' don't consume an argument. These shouldn't 2499 // be used to decide if we are using positional arguments consistently. 2500 if (FS.consumesDataArgument()) { 2501 if (atFirstArg) { 2502 atFirstArg = false; 2503 usesPositionalArgs = FS.usesPositionalArg(); 2504 } 2505 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2506 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2507 startSpecifier, specifierLen); 2508 return false; 2509 } 2510 } 2511 2512 // Check if the field with is non-zero. 2513 const OptionalAmount &Amt = FS.getFieldWidth(); 2514 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 2515 if (Amt.getConstantAmount() == 0) { 2516 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 2517 Amt.getConstantLength()); 2518 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 2519 getLocationOfByte(Amt.getStart()), 2520 /*IsStringLocation*/true, R, 2521 FixItHint::CreateRemoval(R)); 2522 } 2523 } 2524 2525 if (!FS.consumesDataArgument()) { 2526 // FIXME: Technically specifying a precision or field width here 2527 // makes no sense. Worth issuing a warning at some point. 2528 return true; 2529 } 2530 2531 // Consume the argument. 2532 unsigned argIndex = FS.getArgIndex(); 2533 if (argIndex < NumDataArgs) { 2534 // The check to see if the argIndex is valid will come later. 2535 // We set the bit here because we may exit early from this 2536 // function if we encounter some other error. 2537 CoveredArgs.set(argIndex); 2538 } 2539 2540 // Check the length modifier is valid with the given conversion specifier. 2541 const LengthModifier &LM = FS.getLengthModifier(); 2542 if (!FS.hasValidLengthModifier()) { 2543 const CharSourceRange &R = getSpecifierRange(LM.getStart(), LM.getLength()); 2544 EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length) 2545 << LM.toString() << CS.toString() 2546 << getSpecifierRange(startSpecifier, specifierLen), 2547 getLocationOfByte(LM.getStart()), 2548 /*IsStringLocation*/true, R, 2549 FixItHint::CreateRemoval(R)); 2550 } 2551 2552 if (!FS.hasStandardLengthModifier()) 2553 HandleNonStandardLengthModifier(LM, startSpecifier, specifierLen); 2554 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 2555 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 2556 if (!FS.hasStandardLengthConversionCombination()) 2557 HandleNonStandardConversionSpecification(LM, CS, startSpecifier, 2558 specifierLen); 2559 2560 // The remaining checks depend on the data arguments. 2561 if (HasVAListArg) 2562 return true; 2563 2564 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2565 return false; 2566 2567 // Check that the argument type matches the format specifier. 2568 const Expr *Ex = getDataArg(argIndex); 2569 const analyze_scanf::ScanfArgTypeResult &ATR = FS.getArgType(S.Context); 2570 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 2571 ScanfSpecifier fixedFS = FS; 2572 bool success = fixedFS.fixType(Ex->getType(), S.getLangOpts(), 2573 S.Context); 2574 2575 if (success) { 2576 // Get the fix string from the fixed format specifier. 2577 SmallString<128> buf; 2578 llvm::raw_svector_ostream os(buf); 2579 fixedFS.toString(os); 2580 2581 EmitFormatDiagnostic( 2582 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2583 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2584 << Ex->getSourceRange(), 2585 getLocationOfByte(CS.getStart()), 2586 /*IsStringLocation*/true, 2587 getSpecifierRange(startSpecifier, specifierLen), 2588 FixItHint::CreateReplacement( 2589 getSpecifierRange(startSpecifier, specifierLen), 2590 os.str())); 2591 } else { 2592 EmitFormatDiagnostic( 2593 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2594 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2595 << Ex->getSourceRange(), 2596 getLocationOfByte(CS.getStart()), 2597 /*IsStringLocation*/true, 2598 getSpecifierRange(startSpecifier, specifierLen)); 2599 } 2600 } 2601 2602 return true; 2603} 2604 2605void Sema::CheckFormatString(const StringLiteral *FExpr, 2606 const Expr *OrigFormatExpr, 2607 Expr **Args, unsigned NumArgs, 2608 bool HasVAListArg, unsigned format_idx, 2609 unsigned firstDataArg, FormatStringType Type, 2610 bool inFunctionCall) { 2611 2612 // CHECK: is the format string a wide literal? 2613 if (!FExpr->isAscii()) { 2614 CheckFormatHandler::EmitFormatDiagnostic( 2615 *this, inFunctionCall, Args[format_idx], 2616 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 2617 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2618 return; 2619 } 2620 2621 // Str - The format string. NOTE: this is NOT null-terminated! 2622 StringRef StrRef = FExpr->getString(); 2623 const char *Str = StrRef.data(); 2624 unsigned StrLen = StrRef.size(); 2625 const unsigned numDataArgs = NumArgs - firstDataArg; 2626 2627 // CHECK: empty format string? 2628 if (StrLen == 0 && numDataArgs > 0) { 2629 CheckFormatHandler::EmitFormatDiagnostic( 2630 *this, inFunctionCall, Args[format_idx], 2631 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 2632 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2633 return; 2634 } 2635 2636 if (Type == FST_Printf || Type == FST_NSString) { 2637 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2638 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2639 Str, HasVAListArg, Args, NumArgs, format_idx, 2640 inFunctionCall); 2641 2642 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen, 2643 getLangOpts())) 2644 H.DoneProcessing(); 2645 } else if (Type == FST_Scanf) { 2646 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2647 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2648 Str, HasVAListArg, Args, NumArgs, format_idx, 2649 inFunctionCall); 2650 2651 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen, 2652 getLangOpts())) 2653 H.DoneProcessing(); 2654 } // TODO: handle other formats 2655} 2656 2657//===--- CHECK: Standard memory functions ---------------------------------===// 2658 2659/// \brief Determine whether the given type is a dynamic class type (e.g., 2660/// whether it has a vtable). 2661static bool isDynamicClassType(QualType T) { 2662 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 2663 if (CXXRecordDecl *Definition = Record->getDefinition()) 2664 if (Definition->isDynamicClass()) 2665 return true; 2666 2667 return false; 2668} 2669 2670/// \brief If E is a sizeof expression, returns its argument expression, 2671/// otherwise returns NULL. 2672static const Expr *getSizeOfExprArg(const Expr* E) { 2673 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2674 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2675 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 2676 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 2677 2678 return 0; 2679} 2680 2681/// \brief If E is a sizeof expression, returns its argument type. 2682static QualType getSizeOfArgType(const Expr* E) { 2683 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2684 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2685 if (SizeOf->getKind() == clang::UETT_SizeOf) 2686 return SizeOf->getTypeOfArgument(); 2687 2688 return QualType(); 2689} 2690 2691/// \brief Check for dangerous or invalid arguments to memset(). 2692/// 2693/// This issues warnings on known problematic, dangerous or unspecified 2694/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 2695/// function calls. 2696/// 2697/// \param Call The call expression to diagnose. 2698void Sema::CheckMemaccessArguments(const CallExpr *Call, 2699 unsigned BId, 2700 IdentifierInfo *FnName) { 2701 assert(BId != 0); 2702 2703 // It is possible to have a non-standard definition of memset. Validate 2704 // we have enough arguments, and if not, abort further checking. 2705 unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3); 2706 if (Call->getNumArgs() < ExpectedNumArgs) 2707 return; 2708 2709 unsigned LastArg = (BId == Builtin::BImemset || 2710 BId == Builtin::BIstrndup ? 1 : 2); 2711 unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2); 2712 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 2713 2714 // We have special checking when the length is a sizeof expression. 2715 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 2716 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 2717 llvm::FoldingSetNodeID SizeOfArgID; 2718 2719 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 2720 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 2721 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 2722 2723 QualType DestTy = Dest->getType(); 2724 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 2725 QualType PointeeTy = DestPtrTy->getPointeeType(); 2726 2727 // Never warn about void type pointers. This can be used to suppress 2728 // false positives. 2729 if (PointeeTy->isVoidType()) 2730 continue; 2731 2732 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 2733 // actually comparing the expressions for equality. Because computing the 2734 // expression IDs can be expensive, we only do this if the diagnostic is 2735 // enabled. 2736 if (SizeOfArg && 2737 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 2738 SizeOfArg->getExprLoc())) { 2739 // We only compute IDs for expressions if the warning is enabled, and 2740 // cache the sizeof arg's ID. 2741 if (SizeOfArgID == llvm::FoldingSetNodeID()) 2742 SizeOfArg->Profile(SizeOfArgID, Context, true); 2743 llvm::FoldingSetNodeID DestID; 2744 Dest->Profile(DestID, Context, true); 2745 if (DestID == SizeOfArgID) { 2746 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 2747 // over sizeof(src) as well. 2748 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 2749 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 2750 if (UnaryOp->getOpcode() == UO_AddrOf) 2751 ActionIdx = 1; // If its an address-of operator, just remove it. 2752 if (Context.getTypeSize(PointeeTy) == Context.getCharWidth()) 2753 ActionIdx = 2; // If the pointee's size is sizeof(char), 2754 // suggest an explicit length. 2755 unsigned DestSrcSelect = 2756 (BId == Builtin::BIstrndup ? 1 : ArgIdx); 2757 DiagRuntimeBehavior(SizeOfArg->getExprLoc(), Dest, 2758 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 2759 << FnName << DestSrcSelect << ActionIdx 2760 << Dest->getSourceRange() 2761 << SizeOfArg->getSourceRange()); 2762 break; 2763 } 2764 } 2765 2766 // Also check for cases where the sizeof argument is the exact same 2767 // type as the memory argument, and where it points to a user-defined 2768 // record type. 2769 if (SizeOfArgTy != QualType()) { 2770 if (PointeeTy->isRecordType() && 2771 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 2772 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 2773 PDiag(diag::warn_sizeof_pointer_type_memaccess) 2774 << FnName << SizeOfArgTy << ArgIdx 2775 << PointeeTy << Dest->getSourceRange() 2776 << LenExpr->getSourceRange()); 2777 break; 2778 } 2779 } 2780 2781 // Always complain about dynamic classes. 2782 if (isDynamicClassType(PointeeTy)) { 2783 2784 unsigned OperationType = 0; 2785 // "overwritten" if we're warning about the destination for any call 2786 // but memcmp; otherwise a verb appropriate to the call. 2787 if (ArgIdx != 0 || BId == Builtin::BImemcmp) { 2788 if (BId == Builtin::BImemcpy) 2789 OperationType = 1; 2790 else if(BId == Builtin::BImemmove) 2791 OperationType = 2; 2792 else if (BId == Builtin::BImemcmp) 2793 OperationType = 3; 2794 } 2795 2796 DiagRuntimeBehavior( 2797 Dest->getExprLoc(), Dest, 2798 PDiag(diag::warn_dyn_class_memaccess) 2799 << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx) 2800 << FnName << PointeeTy 2801 << OperationType 2802 << Call->getCallee()->getSourceRange()); 2803 } else if (PointeeTy.hasNonTrivialObjCLifetime() && 2804 BId != Builtin::BImemset) 2805 DiagRuntimeBehavior( 2806 Dest->getExprLoc(), Dest, 2807 PDiag(diag::warn_arc_object_memaccess) 2808 << ArgIdx << FnName << PointeeTy 2809 << Call->getCallee()->getSourceRange()); 2810 else 2811 continue; 2812 2813 DiagRuntimeBehavior( 2814 Dest->getExprLoc(), Dest, 2815 PDiag(diag::note_bad_memaccess_silence) 2816 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 2817 break; 2818 } 2819 } 2820} 2821 2822// A little helper routine: ignore addition and subtraction of integer literals. 2823// This intentionally does not ignore all integer constant expressions because 2824// we don't want to remove sizeof(). 2825static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 2826 Ex = Ex->IgnoreParenCasts(); 2827 2828 for (;;) { 2829 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 2830 if (!BO || !BO->isAdditiveOp()) 2831 break; 2832 2833 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 2834 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 2835 2836 if (isa<IntegerLiteral>(RHS)) 2837 Ex = LHS; 2838 else if (isa<IntegerLiteral>(LHS)) 2839 Ex = RHS; 2840 else 2841 break; 2842 } 2843 2844 return Ex; 2845} 2846 2847// Warn if the user has made the 'size' argument to strlcpy or strlcat 2848// be the size of the source, instead of the destination. 2849void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 2850 IdentifierInfo *FnName) { 2851 2852 // Don't crash if the user has the wrong number of arguments 2853 if (Call->getNumArgs() != 3) 2854 return; 2855 2856 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 2857 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 2858 const Expr *CompareWithSrc = NULL; 2859 2860 // Look for 'strlcpy(dst, x, sizeof(x))' 2861 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 2862 CompareWithSrc = Ex; 2863 else { 2864 // Look for 'strlcpy(dst, x, strlen(x))' 2865 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 2866 if (SizeCall->isBuiltinCall() == Builtin::BIstrlen 2867 && SizeCall->getNumArgs() == 1) 2868 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 2869 } 2870 } 2871 2872 if (!CompareWithSrc) 2873 return; 2874 2875 // Determine if the argument to sizeof/strlen is equal to the source 2876 // argument. In principle there's all kinds of things you could do 2877 // here, for instance creating an == expression and evaluating it with 2878 // EvaluateAsBooleanCondition, but this uses a more direct technique: 2879 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 2880 if (!SrcArgDRE) 2881 return; 2882 2883 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 2884 if (!CompareWithSrcDRE || 2885 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 2886 return; 2887 2888 const Expr *OriginalSizeArg = Call->getArg(2); 2889 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 2890 << OriginalSizeArg->getSourceRange() << FnName; 2891 2892 // Output a FIXIT hint if the destination is an array (rather than a 2893 // pointer to an array). This could be enhanced to handle some 2894 // pointers if we know the actual size, like if DstArg is 'array+2' 2895 // we could say 'sizeof(array)-2'. 2896 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 2897 QualType DstArgTy = DstArg->getType(); 2898 2899 // Only handle constant-sized or VLAs, but not flexible members. 2900 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 2901 // Only issue the FIXIT for arrays of size > 1. 2902 if (CAT->getSize().getSExtValue() <= 1) 2903 return; 2904 } else if (!DstArgTy->isVariableArrayType()) { 2905 return; 2906 } 2907 2908 SmallString<128> sizeString; 2909 llvm::raw_svector_ostream OS(sizeString); 2910 OS << "sizeof("; 2911 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 2912 OS << ")"; 2913 2914 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 2915 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 2916 OS.str()); 2917} 2918 2919/// Check if two expressions refer to the same declaration. 2920static bool referToTheSameDecl(const Expr *E1, const Expr *E2) { 2921 if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1)) 2922 if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2)) 2923 return D1->getDecl() == D2->getDecl(); 2924 return false; 2925} 2926 2927static const Expr *getStrlenExprArg(const Expr *E) { 2928 if (const CallExpr *CE = dyn_cast<CallExpr>(E)) { 2929 const FunctionDecl *FD = CE->getDirectCallee(); 2930 if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen) 2931 return 0; 2932 return CE->getArg(0)->IgnoreParenCasts(); 2933 } 2934 return 0; 2935} 2936 2937// Warn on anti-patterns as the 'size' argument to strncat. 2938// The correct size argument should look like following: 2939// strncat(dst, src, sizeof(dst) - strlen(dest) - 1); 2940void Sema::CheckStrncatArguments(const CallExpr *CE, 2941 IdentifierInfo *FnName) { 2942 // Don't crash if the user has the wrong number of arguments. 2943 if (CE->getNumArgs() < 3) 2944 return; 2945 const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts(); 2946 const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts(); 2947 const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts(); 2948 2949 // Identify common expressions, which are wrongly used as the size argument 2950 // to strncat and may lead to buffer overflows. 2951 unsigned PatternType = 0; 2952 if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) { 2953 // - sizeof(dst) 2954 if (referToTheSameDecl(SizeOfArg, DstArg)) 2955 PatternType = 1; 2956 // - sizeof(src) 2957 else if (referToTheSameDecl(SizeOfArg, SrcArg)) 2958 PatternType = 2; 2959 } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) { 2960 if (BE->getOpcode() == BO_Sub) { 2961 const Expr *L = BE->getLHS()->IgnoreParenCasts(); 2962 const Expr *R = BE->getRHS()->IgnoreParenCasts(); 2963 // - sizeof(dst) - strlen(dst) 2964 if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) && 2965 referToTheSameDecl(DstArg, getStrlenExprArg(R))) 2966 PatternType = 1; 2967 // - sizeof(src) - (anything) 2968 else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L))) 2969 PatternType = 2; 2970 } 2971 } 2972 2973 if (PatternType == 0) 2974 return; 2975 2976 // Generate the diagnostic. 2977 SourceLocation SL = LenArg->getLocStart(); 2978 SourceRange SR = LenArg->getSourceRange(); 2979 SourceManager &SM = PP.getSourceManager(); 2980 2981 // If the function is defined as a builtin macro, do not show macro expansion. 2982 if (SM.isMacroArgExpansion(SL)) { 2983 SL = SM.getSpellingLoc(SL); 2984 SR = SourceRange(SM.getSpellingLoc(SR.getBegin()), 2985 SM.getSpellingLoc(SR.getEnd())); 2986 } 2987 2988 if (PatternType == 1) 2989 Diag(SL, diag::warn_strncat_large_size) << SR; 2990 else 2991 Diag(SL, diag::warn_strncat_src_size) << SR; 2992 2993 // Output a FIXIT hint if the destination is an array (rather than a 2994 // pointer to an array). This could be enhanced to handle some 2995 // pointers if we know the actual size, like if DstArg is 'array+2' 2996 // we could say 'sizeof(array)-2'. 2997 QualType DstArgTy = DstArg->getType(); 2998 2999 // Only handle constant-sized or VLAs, but not flexible members. 3000 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 3001 // Only issue the FIXIT for arrays of size > 1. 3002 if (CAT->getSize().getSExtValue() <= 1) 3003 return; 3004 } else if (!DstArgTy->isVariableArrayType()) { 3005 return; 3006 } 3007 3008 SmallString<128> sizeString; 3009 llvm::raw_svector_ostream OS(sizeString); 3010 OS << "sizeof("; 3011 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 3012 OS << ") - "; 3013 OS << "strlen("; 3014 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 3015 OS << ") - 1"; 3016 3017 Diag(SL, diag::note_strncat_wrong_size) 3018 << FixItHint::CreateReplacement(SR, OS.str()); 3019} 3020 3021//===--- CHECK: Return Address of Stack Variable --------------------------===// 3022 3023static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3024 Decl *ParentDecl); 3025static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars, 3026 Decl *ParentDecl); 3027 3028/// CheckReturnStackAddr - Check if a return statement returns the address 3029/// of a stack variable. 3030void 3031Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 3032 SourceLocation ReturnLoc) { 3033 3034 Expr *stackE = 0; 3035 SmallVector<DeclRefExpr *, 8> refVars; 3036 3037 // Perform checking for returned stack addresses, local blocks, 3038 // label addresses or references to temporaries. 3039 if (lhsType->isPointerType() || 3040 (!getLangOpts().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 3041 stackE = EvalAddr(RetValExp, refVars, /*ParentDecl=*/0); 3042 } else if (lhsType->isReferenceType()) { 3043 stackE = EvalVal(RetValExp, refVars, /*ParentDecl=*/0); 3044 } 3045 3046 if (stackE == 0) 3047 return; // Nothing suspicious was found. 3048 3049 SourceLocation diagLoc; 3050 SourceRange diagRange; 3051 if (refVars.empty()) { 3052 diagLoc = stackE->getLocStart(); 3053 diagRange = stackE->getSourceRange(); 3054 } else { 3055 // We followed through a reference variable. 'stackE' contains the 3056 // problematic expression but we will warn at the return statement pointing 3057 // at the reference variable. We will later display the "trail" of 3058 // reference variables using notes. 3059 diagLoc = refVars[0]->getLocStart(); 3060 diagRange = refVars[0]->getSourceRange(); 3061 } 3062 3063 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 3064 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 3065 : diag::warn_ret_stack_addr) 3066 << DR->getDecl()->getDeclName() << diagRange; 3067 } else if (isa<BlockExpr>(stackE)) { // local block. 3068 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 3069 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 3070 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 3071 } else { // local temporary. 3072 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 3073 : diag::warn_ret_local_temp_addr) 3074 << diagRange; 3075 } 3076 3077 // Display the "trail" of reference variables that we followed until we 3078 // found the problematic expression using notes. 3079 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 3080 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 3081 // If this var binds to another reference var, show the range of the next 3082 // var, otherwise the var binds to the problematic expression, in which case 3083 // show the range of the expression. 3084 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 3085 : stackE->getSourceRange(); 3086 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 3087 << VD->getDeclName() << range; 3088 } 3089} 3090 3091/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 3092/// check if the expression in a return statement evaluates to an address 3093/// to a location on the stack, a local block, an address of a label, or a 3094/// reference to local temporary. The recursion is used to traverse the 3095/// AST of the return expression, with recursion backtracking when we 3096/// encounter a subexpression that (1) clearly does not lead to one of the 3097/// above problematic expressions (2) is something we cannot determine leads to 3098/// a problematic expression based on such local checking. 3099/// 3100/// Both EvalAddr and EvalVal follow through reference variables to evaluate 3101/// the expression that they point to. Such variables are added to the 3102/// 'refVars' vector so that we know what the reference variable "trail" was. 3103/// 3104/// EvalAddr processes expressions that are pointers that are used as 3105/// references (and not L-values). EvalVal handles all other values. 3106/// At the base case of the recursion is a check for the above problematic 3107/// expressions. 3108/// 3109/// This implementation handles: 3110/// 3111/// * pointer-to-pointer casts 3112/// * implicit conversions from array references to pointers 3113/// * taking the address of fields 3114/// * arbitrary interplay between "&" and "*" operators 3115/// * pointer arithmetic from an address of a stack variable 3116/// * taking the address of an array element where the array is on the stack 3117static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3118 Decl *ParentDecl) { 3119 if (E->isTypeDependent()) 3120 return NULL; 3121 3122 // We should only be called for evaluating pointer expressions. 3123 assert((E->getType()->isAnyPointerType() || 3124 E->getType()->isBlockPointerType() || 3125 E->getType()->isObjCQualifiedIdType()) && 3126 "EvalAddr only works on pointers"); 3127 3128 E = E->IgnoreParens(); 3129 3130 // Our "symbolic interpreter" is just a dispatch off the currently 3131 // viewed AST node. We then recursively traverse the AST by calling 3132 // EvalAddr and EvalVal appropriately. 3133 switch (E->getStmtClass()) { 3134 case Stmt::DeclRefExprClass: { 3135 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3136 3137 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 3138 // If this is a reference variable, follow through to the expression that 3139 // it points to. 3140 if (V->hasLocalStorage() && 3141 V->getType()->isReferenceType() && V->hasInit()) { 3142 // Add the reference variable to the "trail". 3143 refVars.push_back(DR); 3144 return EvalAddr(V->getInit(), refVars, ParentDecl); 3145 } 3146 3147 return NULL; 3148 } 3149 3150 case Stmt::UnaryOperatorClass: { 3151 // The only unary operator that make sense to handle here 3152 // is AddrOf. All others don't make sense as pointers. 3153 UnaryOperator *U = cast<UnaryOperator>(E); 3154 3155 if (U->getOpcode() == UO_AddrOf) 3156 return EvalVal(U->getSubExpr(), refVars, ParentDecl); 3157 else 3158 return NULL; 3159 } 3160 3161 case Stmt::BinaryOperatorClass: { 3162 // Handle pointer arithmetic. All other binary operators are not valid 3163 // in this context. 3164 BinaryOperator *B = cast<BinaryOperator>(E); 3165 BinaryOperatorKind op = B->getOpcode(); 3166 3167 if (op != BO_Add && op != BO_Sub) 3168 return NULL; 3169 3170 Expr *Base = B->getLHS(); 3171 3172 // Determine which argument is the real pointer base. It could be 3173 // the RHS argument instead of the LHS. 3174 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 3175 3176 assert (Base->getType()->isPointerType()); 3177 return EvalAddr(Base, refVars, ParentDecl); 3178 } 3179 3180 // For conditional operators we need to see if either the LHS or RHS are 3181 // valid DeclRefExpr*s. If one of them is valid, we return it. 3182 case Stmt::ConditionalOperatorClass: { 3183 ConditionalOperator *C = cast<ConditionalOperator>(E); 3184 3185 // Handle the GNU extension for missing LHS. 3186 if (Expr *lhsExpr = C->getLHS()) { 3187 // In C++, we can have a throw-expression, which has 'void' type. 3188 if (!lhsExpr->getType()->isVoidType()) 3189 if (Expr* LHS = EvalAddr(lhsExpr, refVars, ParentDecl)) 3190 return LHS; 3191 } 3192 3193 // In C++, we can have a throw-expression, which has 'void' type. 3194 if (C->getRHS()->getType()->isVoidType()) 3195 return NULL; 3196 3197 return EvalAddr(C->getRHS(), refVars, ParentDecl); 3198 } 3199 3200 case Stmt::BlockExprClass: 3201 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 3202 return E; // local block. 3203 return NULL; 3204 3205 case Stmt::AddrLabelExprClass: 3206 return E; // address of label. 3207 3208 case Stmt::ExprWithCleanupsClass: 3209 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars, 3210 ParentDecl); 3211 3212 // For casts, we need to handle conversions from arrays to 3213 // pointer values, and pointer-to-pointer conversions. 3214 case Stmt::ImplicitCastExprClass: 3215 case Stmt::CStyleCastExprClass: 3216 case Stmt::CXXFunctionalCastExprClass: 3217 case Stmt::ObjCBridgedCastExprClass: 3218 case Stmt::CXXStaticCastExprClass: 3219 case Stmt::CXXDynamicCastExprClass: 3220 case Stmt::CXXConstCastExprClass: 3221 case Stmt::CXXReinterpretCastExprClass: { 3222 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 3223 switch (cast<CastExpr>(E)->getCastKind()) { 3224 case CK_BitCast: 3225 case CK_LValueToRValue: 3226 case CK_NoOp: 3227 case CK_BaseToDerived: 3228 case CK_DerivedToBase: 3229 case CK_UncheckedDerivedToBase: 3230 case CK_Dynamic: 3231 case CK_CPointerToObjCPointerCast: 3232 case CK_BlockPointerToObjCPointerCast: 3233 case CK_AnyPointerToBlockPointerCast: 3234 return EvalAddr(SubExpr, refVars, ParentDecl); 3235 3236 case CK_ArrayToPointerDecay: 3237 return EvalVal(SubExpr, refVars, ParentDecl); 3238 3239 default: 3240 return 0; 3241 } 3242 } 3243 3244 case Stmt::MaterializeTemporaryExprClass: 3245 if (Expr *Result = EvalAddr( 3246 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3247 refVars, ParentDecl)) 3248 return Result; 3249 3250 return E; 3251 3252 // Everything else: we simply don't reason about them. 3253 default: 3254 return NULL; 3255 } 3256} 3257 3258 3259/// EvalVal - This function is complements EvalAddr in the mutual recursion. 3260/// See the comments for EvalAddr for more details. 3261static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3262 Decl *ParentDecl) { 3263do { 3264 // We should only be called for evaluating non-pointer expressions, or 3265 // expressions with a pointer type that are not used as references but instead 3266 // are l-values (e.g., DeclRefExpr with a pointer type). 3267 3268 // Our "symbolic interpreter" is just a dispatch off the currently 3269 // viewed AST node. We then recursively traverse the AST by calling 3270 // EvalAddr and EvalVal appropriately. 3271 3272 E = E->IgnoreParens(); 3273 switch (E->getStmtClass()) { 3274 case Stmt::ImplicitCastExprClass: { 3275 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 3276 if (IE->getValueKind() == VK_LValue) { 3277 E = IE->getSubExpr(); 3278 continue; 3279 } 3280 return NULL; 3281 } 3282 3283 case Stmt::ExprWithCleanupsClass: 3284 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,ParentDecl); 3285 3286 case Stmt::DeclRefExprClass: { 3287 // When we hit a DeclRefExpr we are looking at code that refers to a 3288 // variable's name. If it's not a reference variable we check if it has 3289 // local storage within the function, and if so, return the expression. 3290 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3291 3292 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) { 3293 // Check if it refers to itself, e.g. "int& i = i;". 3294 if (V == ParentDecl) 3295 return DR; 3296 3297 if (V->hasLocalStorage()) { 3298 if (!V->getType()->isReferenceType()) 3299 return DR; 3300 3301 // Reference variable, follow through to the expression that 3302 // it points to. 3303 if (V->hasInit()) { 3304 // Add the reference variable to the "trail". 3305 refVars.push_back(DR); 3306 return EvalVal(V->getInit(), refVars, V); 3307 } 3308 } 3309 } 3310 3311 return NULL; 3312 } 3313 3314 case Stmt::UnaryOperatorClass: { 3315 // The only unary operator that make sense to handle here 3316 // is Deref. All others don't resolve to a "name." This includes 3317 // handling all sorts of rvalues passed to a unary operator. 3318 UnaryOperator *U = cast<UnaryOperator>(E); 3319 3320 if (U->getOpcode() == UO_Deref) 3321 return EvalAddr(U->getSubExpr(), refVars, ParentDecl); 3322 3323 return NULL; 3324 } 3325 3326 case Stmt::ArraySubscriptExprClass: { 3327 // Array subscripts are potential references to data on the stack. We 3328 // retrieve the DeclRefExpr* for the array variable if it indeed 3329 // has local storage. 3330 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars,ParentDecl); 3331 } 3332 3333 case Stmt::ConditionalOperatorClass: { 3334 // For conditional operators we need to see if either the LHS or RHS are 3335 // non-NULL Expr's. If one is non-NULL, we return it. 3336 ConditionalOperator *C = cast<ConditionalOperator>(E); 3337 3338 // Handle the GNU extension for missing LHS. 3339 if (Expr *lhsExpr = C->getLHS()) 3340 if (Expr *LHS = EvalVal(lhsExpr, refVars, ParentDecl)) 3341 return LHS; 3342 3343 return EvalVal(C->getRHS(), refVars, ParentDecl); 3344 } 3345 3346 // Accesses to members are potential references to data on the stack. 3347 case Stmt::MemberExprClass: { 3348 MemberExpr *M = cast<MemberExpr>(E); 3349 3350 // Check for indirect access. We only want direct field accesses. 3351 if (M->isArrow()) 3352 return NULL; 3353 3354 // Check whether the member type is itself a reference, in which case 3355 // we're not going to refer to the member, but to what the member refers to. 3356 if (M->getMemberDecl()->getType()->isReferenceType()) 3357 return NULL; 3358 3359 return EvalVal(M->getBase(), refVars, ParentDecl); 3360 } 3361 3362 case Stmt::MaterializeTemporaryExprClass: 3363 if (Expr *Result = EvalVal( 3364 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3365 refVars, ParentDecl)) 3366 return Result; 3367 3368 return E; 3369 3370 default: 3371 // Check that we don't return or take the address of a reference to a 3372 // temporary. This is only useful in C++. 3373 if (!E->isTypeDependent() && E->isRValue()) 3374 return E; 3375 3376 // Everything else: we simply don't reason about them. 3377 return NULL; 3378 } 3379} while (true); 3380} 3381 3382//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 3383 3384/// Check for comparisons of floating point operands using != and ==. 3385/// Issue a warning if these are no self-comparisons, as they are not likely 3386/// to do what the programmer intended. 3387void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 3388 bool EmitWarning = true; 3389 3390 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 3391 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 3392 3393 // Special case: check for x == x (which is OK). 3394 // Do not emit warnings for such cases. 3395 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 3396 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 3397 if (DRL->getDecl() == DRR->getDecl()) 3398 EmitWarning = false; 3399 3400 3401 // Special case: check for comparisons against literals that can be exactly 3402 // represented by APFloat. In such cases, do not emit a warning. This 3403 // is a heuristic: often comparison against such literals are used to 3404 // detect if a value in a variable has not changed. This clearly can 3405 // lead to false negatives. 3406 if (EmitWarning) { 3407 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 3408 if (FLL->isExact()) 3409 EmitWarning = false; 3410 } else 3411 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 3412 if (FLR->isExact()) 3413 EmitWarning = false; 3414 } 3415 } 3416 3417 // Check for comparisons with builtin types. 3418 if (EmitWarning) 3419 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 3420 if (CL->isBuiltinCall()) 3421 EmitWarning = false; 3422 3423 if (EmitWarning) 3424 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 3425 if (CR->isBuiltinCall()) 3426 EmitWarning = false; 3427 3428 // Emit the diagnostic. 3429 if (EmitWarning) 3430 Diag(Loc, diag::warn_floatingpoint_eq) 3431 << LHS->getSourceRange() << RHS->getSourceRange(); 3432} 3433 3434//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 3435//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 3436 3437namespace { 3438 3439/// Structure recording the 'active' range of an integer-valued 3440/// expression. 3441struct IntRange { 3442 /// The number of bits active in the int. 3443 unsigned Width; 3444 3445 /// True if the int is known not to have negative values. 3446 bool NonNegative; 3447 3448 IntRange(unsigned Width, bool NonNegative) 3449 : Width(Width), NonNegative(NonNegative) 3450 {} 3451 3452 /// Returns the range of the bool type. 3453 static IntRange forBoolType() { 3454 return IntRange(1, true); 3455 } 3456 3457 /// Returns the range of an opaque value of the given integral type. 3458 static IntRange forValueOfType(ASTContext &C, QualType T) { 3459 return forValueOfCanonicalType(C, 3460 T->getCanonicalTypeInternal().getTypePtr()); 3461 } 3462 3463 /// Returns the range of an opaque value of a canonical integral type. 3464 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 3465 assert(T->isCanonicalUnqualified()); 3466 3467 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3468 T = VT->getElementType().getTypePtr(); 3469 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3470 T = CT->getElementType().getTypePtr(); 3471 3472 // For enum types, use the known bit width of the enumerators. 3473 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 3474 EnumDecl *Enum = ET->getDecl(); 3475 if (!Enum->isCompleteDefinition()) 3476 return IntRange(C.getIntWidth(QualType(T, 0)), false); 3477 3478 unsigned NumPositive = Enum->getNumPositiveBits(); 3479 unsigned NumNegative = Enum->getNumNegativeBits(); 3480 3481 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 3482 } 3483 3484 const BuiltinType *BT = cast<BuiltinType>(T); 3485 assert(BT->isInteger()); 3486 3487 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3488 } 3489 3490 /// Returns the "target" range of a canonical integral type, i.e. 3491 /// the range of values expressible in the type. 3492 /// 3493 /// This matches forValueOfCanonicalType except that enums have the 3494 /// full range of their type, not the range of their enumerators. 3495 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 3496 assert(T->isCanonicalUnqualified()); 3497 3498 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3499 T = VT->getElementType().getTypePtr(); 3500 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3501 T = CT->getElementType().getTypePtr(); 3502 if (const EnumType *ET = dyn_cast<EnumType>(T)) 3503 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 3504 3505 const BuiltinType *BT = cast<BuiltinType>(T); 3506 assert(BT->isInteger()); 3507 3508 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3509 } 3510 3511 /// Returns the supremum of two ranges: i.e. their conservative merge. 3512 static IntRange join(IntRange L, IntRange R) { 3513 return IntRange(std::max(L.Width, R.Width), 3514 L.NonNegative && R.NonNegative); 3515 } 3516 3517 /// Returns the infinum of two ranges: i.e. their aggressive merge. 3518 static IntRange meet(IntRange L, IntRange R) { 3519 return IntRange(std::min(L.Width, R.Width), 3520 L.NonNegative || R.NonNegative); 3521 } 3522}; 3523 3524static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, 3525 unsigned MaxWidth) { 3526 if (value.isSigned() && value.isNegative()) 3527 return IntRange(value.getMinSignedBits(), false); 3528 3529 if (value.getBitWidth() > MaxWidth) 3530 value = value.trunc(MaxWidth); 3531 3532 // isNonNegative() just checks the sign bit without considering 3533 // signedness. 3534 return IntRange(value.getActiveBits(), true); 3535} 3536 3537static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 3538 unsigned MaxWidth) { 3539 if (result.isInt()) 3540 return GetValueRange(C, result.getInt(), MaxWidth); 3541 3542 if (result.isVector()) { 3543 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 3544 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 3545 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 3546 R = IntRange::join(R, El); 3547 } 3548 return R; 3549 } 3550 3551 if (result.isComplexInt()) { 3552 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 3553 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 3554 return IntRange::join(R, I); 3555 } 3556 3557 // This can happen with lossless casts to intptr_t of "based" lvalues. 3558 // Assume it might use arbitrary bits. 3559 // FIXME: The only reason we need to pass the type in here is to get 3560 // the sign right on this one case. It would be nice if APValue 3561 // preserved this. 3562 assert(result.isLValue() || result.isAddrLabelDiff()); 3563 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 3564} 3565 3566/// Pseudo-evaluate the given integer expression, estimating the 3567/// range of values it might take. 3568/// 3569/// \param MaxWidth - the width to which the value will be truncated 3570static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 3571 E = E->IgnoreParens(); 3572 3573 // Try a full evaluation first. 3574 Expr::EvalResult result; 3575 if (E->EvaluateAsRValue(result, C)) 3576 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 3577 3578 // I think we only want to look through implicit casts here; if the 3579 // user has an explicit widening cast, we should treat the value as 3580 // being of the new, wider type. 3581 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 3582 if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue) 3583 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 3584 3585 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 3586 3587 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 3588 3589 // Assume that non-integer casts can span the full range of the type. 3590 if (!isIntegerCast) 3591 return OutputTypeRange; 3592 3593 IntRange SubRange 3594 = GetExprRange(C, CE->getSubExpr(), 3595 std::min(MaxWidth, OutputTypeRange.Width)); 3596 3597 // Bail out if the subexpr's range is as wide as the cast type. 3598 if (SubRange.Width >= OutputTypeRange.Width) 3599 return OutputTypeRange; 3600 3601 // Otherwise, we take the smaller width, and we're non-negative if 3602 // either the output type or the subexpr is. 3603 return IntRange(SubRange.Width, 3604 SubRange.NonNegative || OutputTypeRange.NonNegative); 3605 } 3606 3607 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 3608 // If we can fold the condition, just take that operand. 3609 bool CondResult; 3610 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 3611 return GetExprRange(C, CondResult ? CO->getTrueExpr() 3612 : CO->getFalseExpr(), 3613 MaxWidth); 3614 3615 // Otherwise, conservatively merge. 3616 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 3617 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 3618 return IntRange::join(L, R); 3619 } 3620 3621 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 3622 switch (BO->getOpcode()) { 3623 3624 // Boolean-valued operations are single-bit and positive. 3625 case BO_LAnd: 3626 case BO_LOr: 3627 case BO_LT: 3628 case BO_GT: 3629 case BO_LE: 3630 case BO_GE: 3631 case BO_EQ: 3632 case BO_NE: 3633 return IntRange::forBoolType(); 3634 3635 // The type of the assignments is the type of the LHS, so the RHS 3636 // is not necessarily the same type. 3637 case BO_MulAssign: 3638 case BO_DivAssign: 3639 case BO_RemAssign: 3640 case BO_AddAssign: 3641 case BO_SubAssign: 3642 case BO_XorAssign: 3643 case BO_OrAssign: 3644 // TODO: bitfields? 3645 return IntRange::forValueOfType(C, E->getType()); 3646 3647 // Simple assignments just pass through the RHS, which will have 3648 // been coerced to the LHS type. 3649 case BO_Assign: 3650 // TODO: bitfields? 3651 return GetExprRange(C, BO->getRHS(), MaxWidth); 3652 3653 // Operations with opaque sources are black-listed. 3654 case BO_PtrMemD: 3655 case BO_PtrMemI: 3656 return IntRange::forValueOfType(C, E->getType()); 3657 3658 // Bitwise-and uses the *infinum* of the two source ranges. 3659 case BO_And: 3660 case BO_AndAssign: 3661 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 3662 GetExprRange(C, BO->getRHS(), MaxWidth)); 3663 3664 // Left shift gets black-listed based on a judgement call. 3665 case BO_Shl: 3666 // ...except that we want to treat '1 << (blah)' as logically 3667 // positive. It's an important idiom. 3668 if (IntegerLiteral *I 3669 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 3670 if (I->getValue() == 1) { 3671 IntRange R = IntRange::forValueOfType(C, E->getType()); 3672 return IntRange(R.Width, /*NonNegative*/ true); 3673 } 3674 } 3675 // fallthrough 3676 3677 case BO_ShlAssign: 3678 return IntRange::forValueOfType(C, E->getType()); 3679 3680 // Right shift by a constant can narrow its left argument. 3681 case BO_Shr: 3682 case BO_ShrAssign: { 3683 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3684 3685 // If the shift amount is a positive constant, drop the width by 3686 // that much. 3687 llvm::APSInt shift; 3688 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 3689 shift.isNonNegative()) { 3690 unsigned zext = shift.getZExtValue(); 3691 if (zext >= L.Width) 3692 L.Width = (L.NonNegative ? 0 : 1); 3693 else 3694 L.Width -= zext; 3695 } 3696 3697 return L; 3698 } 3699 3700 // Comma acts as its right operand. 3701 case BO_Comma: 3702 return GetExprRange(C, BO->getRHS(), MaxWidth); 3703 3704 // Black-list pointer subtractions. 3705 case BO_Sub: 3706 if (BO->getLHS()->getType()->isPointerType()) 3707 return IntRange::forValueOfType(C, E->getType()); 3708 break; 3709 3710 // The width of a division result is mostly determined by the size 3711 // of the LHS. 3712 case BO_Div: { 3713 // Don't 'pre-truncate' the operands. 3714 unsigned opWidth = C.getIntWidth(E->getType()); 3715 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3716 3717 // If the divisor is constant, use that. 3718 llvm::APSInt divisor; 3719 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 3720 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 3721 if (log2 >= L.Width) 3722 L.Width = (L.NonNegative ? 0 : 1); 3723 else 3724 L.Width = std::min(L.Width - log2, MaxWidth); 3725 return L; 3726 } 3727 3728 // Otherwise, just use the LHS's width. 3729 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3730 return IntRange(L.Width, L.NonNegative && R.NonNegative); 3731 } 3732 3733 // The result of a remainder can't be larger than the result of 3734 // either side. 3735 case BO_Rem: { 3736 // Don't 'pre-truncate' the operands. 3737 unsigned opWidth = C.getIntWidth(E->getType()); 3738 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3739 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3740 3741 IntRange meet = IntRange::meet(L, R); 3742 meet.Width = std::min(meet.Width, MaxWidth); 3743 return meet; 3744 } 3745 3746 // The default behavior is okay for these. 3747 case BO_Mul: 3748 case BO_Add: 3749 case BO_Xor: 3750 case BO_Or: 3751 break; 3752 } 3753 3754 // The default case is to treat the operation as if it were closed 3755 // on the narrowest type that encompasses both operands. 3756 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3757 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 3758 return IntRange::join(L, R); 3759 } 3760 3761 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 3762 switch (UO->getOpcode()) { 3763 // Boolean-valued operations are white-listed. 3764 case UO_LNot: 3765 return IntRange::forBoolType(); 3766 3767 // Operations with opaque sources are black-listed. 3768 case UO_Deref: 3769 case UO_AddrOf: // should be impossible 3770 return IntRange::forValueOfType(C, E->getType()); 3771 3772 default: 3773 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 3774 } 3775 } 3776 3777 if (dyn_cast<OffsetOfExpr>(E)) { 3778 IntRange::forValueOfType(C, E->getType()); 3779 } 3780 3781 if (FieldDecl *BitField = E->getBitField()) 3782 return IntRange(BitField->getBitWidthValue(C), 3783 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 3784 3785 return IntRange::forValueOfType(C, E->getType()); 3786} 3787 3788static IntRange GetExprRange(ASTContext &C, Expr *E) { 3789 return GetExprRange(C, E, C.getIntWidth(E->getType())); 3790} 3791 3792/// Checks whether the given value, which currently has the given 3793/// source semantics, has the same value when coerced through the 3794/// target semantics. 3795static bool IsSameFloatAfterCast(const llvm::APFloat &value, 3796 const llvm::fltSemantics &Src, 3797 const llvm::fltSemantics &Tgt) { 3798 llvm::APFloat truncated = value; 3799 3800 bool ignored; 3801 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 3802 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 3803 3804 return truncated.bitwiseIsEqual(value); 3805} 3806 3807/// Checks whether the given value, which currently has the given 3808/// source semantics, has the same value when coerced through the 3809/// target semantics. 3810/// 3811/// The value might be a vector of floats (or a complex number). 3812static bool IsSameFloatAfterCast(const APValue &value, 3813 const llvm::fltSemantics &Src, 3814 const llvm::fltSemantics &Tgt) { 3815 if (value.isFloat()) 3816 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 3817 3818 if (value.isVector()) { 3819 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 3820 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 3821 return false; 3822 return true; 3823 } 3824 3825 assert(value.isComplexFloat()); 3826 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 3827 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 3828} 3829 3830static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 3831 3832static bool IsZero(Sema &S, Expr *E) { 3833 // Suppress cases where we are comparing against an enum constant. 3834 if (const DeclRefExpr *DR = 3835 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 3836 if (isa<EnumConstantDecl>(DR->getDecl())) 3837 return false; 3838 3839 // Suppress cases where the '0' value is expanded from a macro. 3840 if (E->getLocStart().isMacroID()) 3841 return false; 3842 3843 llvm::APSInt Value; 3844 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 3845} 3846 3847static bool HasEnumType(Expr *E) { 3848 // Strip off implicit integral promotions. 3849 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 3850 if (ICE->getCastKind() != CK_IntegralCast && 3851 ICE->getCastKind() != CK_NoOp) 3852 break; 3853 E = ICE->getSubExpr(); 3854 } 3855 3856 return E->getType()->isEnumeralType(); 3857} 3858 3859static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 3860 BinaryOperatorKind op = E->getOpcode(); 3861 if (E->isValueDependent()) 3862 return; 3863 3864 if (op == BO_LT && IsZero(S, E->getRHS())) { 3865 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3866 << "< 0" << "false" << HasEnumType(E->getLHS()) 3867 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3868 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 3869 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3870 << ">= 0" << "true" << HasEnumType(E->getLHS()) 3871 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3872 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 3873 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3874 << "0 >" << "false" << HasEnumType(E->getRHS()) 3875 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3876 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 3877 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3878 << "0 <=" << "true" << HasEnumType(E->getRHS()) 3879 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3880 } 3881} 3882 3883/// Analyze the operands of the given comparison. Implements the 3884/// fallback case from AnalyzeComparison. 3885static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 3886 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3887 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3888} 3889 3890/// \brief Implements -Wsign-compare. 3891/// 3892/// \param E the binary operator to check for warnings 3893static void AnalyzeComparison(Sema &S, BinaryOperator *E) { 3894 // The type the comparison is being performed in. 3895 QualType T = E->getLHS()->getType(); 3896 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 3897 && "comparison with mismatched types"); 3898 3899 // We don't do anything special if this isn't an unsigned integral 3900 // comparison: we're only interested in integral comparisons, and 3901 // signed comparisons only happen in cases we don't care to warn about. 3902 // 3903 // We also don't care about value-dependent expressions or expressions 3904 // whose result is a constant. 3905 if (!T->hasUnsignedIntegerRepresentation() 3906 || E->isValueDependent() || E->isIntegerConstantExpr(S.Context)) 3907 return AnalyzeImpConvsInComparison(S, E); 3908 3909 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 3910 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 3911 3912 // Check to see if one of the (unmodified) operands is of different 3913 // signedness. 3914 Expr *signedOperand, *unsignedOperand; 3915 if (LHS->getType()->hasSignedIntegerRepresentation()) { 3916 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 3917 "unsigned comparison between two signed integer expressions?"); 3918 signedOperand = LHS; 3919 unsignedOperand = RHS; 3920 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 3921 signedOperand = RHS; 3922 unsignedOperand = LHS; 3923 } else { 3924 CheckTrivialUnsignedComparison(S, E); 3925 return AnalyzeImpConvsInComparison(S, E); 3926 } 3927 3928 // Otherwise, calculate the effective range of the signed operand. 3929 IntRange signedRange = GetExprRange(S.Context, signedOperand); 3930 3931 // Go ahead and analyze implicit conversions in the operands. Note 3932 // that we skip the implicit conversions on both sides. 3933 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 3934 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 3935 3936 // If the signed range is non-negative, -Wsign-compare won't fire, 3937 // but we should still check for comparisons which are always true 3938 // or false. 3939 if (signedRange.NonNegative) 3940 return CheckTrivialUnsignedComparison(S, E); 3941 3942 // For (in)equality comparisons, if the unsigned operand is a 3943 // constant which cannot collide with a overflowed signed operand, 3944 // then reinterpreting the signed operand as unsigned will not 3945 // change the result of the comparison. 3946 if (E->isEqualityOp()) { 3947 unsigned comparisonWidth = S.Context.getIntWidth(T); 3948 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 3949 3950 // We should never be unable to prove that the unsigned operand is 3951 // non-negative. 3952 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 3953 3954 if (unsignedRange.Width < comparisonWidth) 3955 return; 3956 } 3957 3958 S.DiagRuntimeBehavior(E->getOperatorLoc(), E, 3959 S.PDiag(diag::warn_mixed_sign_comparison) 3960 << LHS->getType() << RHS->getType() 3961 << LHS->getSourceRange() << RHS->getSourceRange()); 3962} 3963 3964/// Analyzes an attempt to assign the given value to a bitfield. 3965/// 3966/// Returns true if there was something fishy about the attempt. 3967static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 3968 SourceLocation InitLoc) { 3969 assert(Bitfield->isBitField()); 3970 if (Bitfield->isInvalidDecl()) 3971 return false; 3972 3973 // White-list bool bitfields. 3974 if (Bitfield->getType()->isBooleanType()) 3975 return false; 3976 3977 // Ignore value- or type-dependent expressions. 3978 if (Bitfield->getBitWidth()->isValueDependent() || 3979 Bitfield->getBitWidth()->isTypeDependent() || 3980 Init->isValueDependent() || 3981 Init->isTypeDependent()) 3982 return false; 3983 3984 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 3985 3986 llvm::APSInt Value; 3987 if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects)) 3988 return false; 3989 3990 unsigned OriginalWidth = Value.getBitWidth(); 3991 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 3992 3993 if (OriginalWidth <= FieldWidth) 3994 return false; 3995 3996 // Compute the value which the bitfield will contain. 3997 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 3998 TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType()); 3999 4000 // Check whether the stored value is equal to the original value. 4001 TruncatedValue = TruncatedValue.extend(OriginalWidth); 4002 if (Value == TruncatedValue) 4003 return false; 4004 4005 // Special-case bitfields of width 1: booleans are naturally 0/1, and 4006 // therefore don't strictly fit into a signed bitfield of width 1. 4007 if (FieldWidth == 1 && Value == 1) 4008 return false; 4009 4010 std::string PrettyValue = Value.toString(10); 4011 std::string PrettyTrunc = TruncatedValue.toString(10); 4012 4013 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 4014 << PrettyValue << PrettyTrunc << OriginalInit->getType() 4015 << Init->getSourceRange(); 4016 4017 return true; 4018} 4019 4020/// Analyze the given simple or compound assignment for warning-worthy 4021/// operations. 4022static void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 4023 // Just recurse on the LHS. 4024 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4025 4026 // We want to recurse on the RHS as normal unless we're assigning to 4027 // a bitfield. 4028 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 4029 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 4030 E->getOperatorLoc())) { 4031 // Recurse, ignoring any implicit conversions on the RHS. 4032 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 4033 E->getOperatorLoc()); 4034 } 4035 } 4036 4037 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4038} 4039 4040/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4041static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 4042 SourceLocation CContext, unsigned diag, 4043 bool pruneControlFlow = false) { 4044 if (pruneControlFlow) { 4045 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4046 S.PDiag(diag) 4047 << SourceType << T << E->getSourceRange() 4048 << SourceRange(CContext)); 4049 return; 4050 } 4051 S.Diag(E->getExprLoc(), diag) 4052 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 4053} 4054 4055/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4056static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, 4057 SourceLocation CContext, unsigned diag, 4058 bool pruneControlFlow = false) { 4059 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow); 4060} 4061 4062/// Diagnose an implicit cast from a literal expression. Does not warn when the 4063/// cast wouldn't lose information. 4064void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 4065 SourceLocation CContext) { 4066 // Try to convert the literal exactly to an integer. If we can, don't warn. 4067 bool isExact = false; 4068 const llvm::APFloat &Value = FL->getValue(); 4069 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 4070 T->hasUnsignedIntegerRepresentation()); 4071 if (Value.convertToInteger(IntegerValue, 4072 llvm::APFloat::rmTowardZero, &isExact) 4073 == llvm::APFloat::opOK && isExact) 4074 return; 4075 4076 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 4077 << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext); 4078} 4079 4080std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 4081 if (!Range.Width) return "0"; 4082 4083 llvm::APSInt ValueInRange = Value; 4084 ValueInRange.setIsSigned(!Range.NonNegative); 4085 ValueInRange = ValueInRange.trunc(Range.Width); 4086 return ValueInRange.toString(10); 4087} 4088 4089void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 4090 SourceLocation CC, bool *ICContext = 0) { 4091 if (E->isTypeDependent() || E->isValueDependent()) return; 4092 4093 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 4094 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 4095 if (Source == Target) return; 4096 if (Target->isDependentType()) return; 4097 4098 // If the conversion context location is invalid don't complain. We also 4099 // don't want to emit a warning if the issue occurs from the expansion of 4100 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 4101 // delay this check as long as possible. Once we detect we are in that 4102 // scenario, we just return. 4103 if (CC.isInvalid()) 4104 return; 4105 4106 // Diagnose implicit casts to bool. 4107 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 4108 if (isa<StringLiteral>(E)) 4109 // Warn on string literal to bool. Checks for string literals in logical 4110 // expressions, for instances, assert(0 && "error here"), is prevented 4111 // by a check in AnalyzeImplicitConversions(). 4112 return DiagnoseImpCast(S, E, T, CC, 4113 diag::warn_impcast_string_literal_to_bool); 4114 if (Source->isFunctionType()) { 4115 // Warn on function to bool. Checks free functions and static member 4116 // functions. Weakly imported functions are excluded from the check, 4117 // since it's common to test their value to check whether the linker 4118 // found a definition for them. 4119 ValueDecl *D = 0; 4120 if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) { 4121 D = R->getDecl(); 4122 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 4123 D = M->getMemberDecl(); 4124 } 4125 4126 if (D && !D->isWeak()) { 4127 if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) { 4128 S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool) 4129 << F << E->getSourceRange() << SourceRange(CC); 4130 S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence) 4131 << FixItHint::CreateInsertion(E->getExprLoc(), "&"); 4132 QualType ReturnType; 4133 UnresolvedSet<4> NonTemplateOverloads; 4134 S.isExprCallable(*E, ReturnType, NonTemplateOverloads); 4135 if (!ReturnType.isNull() 4136 && ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 4137 S.Diag(E->getExprLoc(), diag::note_function_to_bool_call) 4138 << FixItHint::CreateInsertion( 4139 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()"); 4140 return; 4141 } 4142 } 4143 } 4144 return; // Other casts to bool are not checked. 4145 } 4146 4147 // Strip vector types. 4148 if (isa<VectorType>(Source)) { 4149 if (!isa<VectorType>(Target)) { 4150 if (S.SourceMgr.isInSystemMacro(CC)) 4151 return; 4152 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 4153 } 4154 4155 // If the vector cast is cast between two vectors of the same size, it is 4156 // a bitcast, not a conversion. 4157 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 4158 return; 4159 4160 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 4161 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 4162 } 4163 4164 // Strip complex types. 4165 if (isa<ComplexType>(Source)) { 4166 if (!isa<ComplexType>(Target)) { 4167 if (S.SourceMgr.isInSystemMacro(CC)) 4168 return; 4169 4170 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 4171 } 4172 4173 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 4174 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 4175 } 4176 4177 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 4178 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 4179 4180 // If the source is floating point... 4181 if (SourceBT && SourceBT->isFloatingPoint()) { 4182 // ...and the target is floating point... 4183 if (TargetBT && TargetBT->isFloatingPoint()) { 4184 // ...then warn if we're dropping FP rank. 4185 4186 // Builtin FP kinds are ordered by increasing FP rank. 4187 if (SourceBT->getKind() > TargetBT->getKind()) { 4188 // Don't warn about float constants that are precisely 4189 // representable in the target type. 4190 Expr::EvalResult result; 4191 if (E->EvaluateAsRValue(result, S.Context)) { 4192 // Value might be a float, a float vector, or a float complex. 4193 if (IsSameFloatAfterCast(result.Val, 4194 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 4195 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 4196 return; 4197 } 4198 4199 if (S.SourceMgr.isInSystemMacro(CC)) 4200 return; 4201 4202 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 4203 } 4204 return; 4205 } 4206 4207 // If the target is integral, always warn. 4208 if ((TargetBT && TargetBT->isInteger())) { 4209 if (S.SourceMgr.isInSystemMacro(CC)) 4210 return; 4211 4212 Expr *InnerE = E->IgnoreParenImpCasts(); 4213 // We also want to warn on, e.g., "int i = -1.234" 4214 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 4215 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 4216 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 4217 4218 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 4219 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 4220 } else { 4221 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 4222 } 4223 } 4224 4225 return; 4226 } 4227 4228 if (!Source->isIntegerType() || !Target->isIntegerType()) 4229 return; 4230 4231 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 4232 == Expr::NPCK_GNUNull) && Target->isIntegerType()) { 4233 SourceLocation Loc = E->getSourceRange().getBegin(); 4234 if (Loc.isMacroID()) 4235 Loc = S.SourceMgr.getImmediateExpansionRange(Loc).first; 4236 S.Diag(Loc, diag::warn_impcast_null_pointer_to_integer) 4237 << T << clang::SourceRange(CC) 4238 << FixItHint::CreateReplacement(Loc, S.getFixItZeroLiteralForType(T)); 4239 return; 4240 } 4241 4242 IntRange SourceRange = GetExprRange(S.Context, E); 4243 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 4244 4245 if (SourceRange.Width > TargetRange.Width) { 4246 // If the source is a constant, use a default-on diagnostic. 4247 // TODO: this should happen for bitfield stores, too. 4248 llvm::APSInt Value(32); 4249 if (E->isIntegerConstantExpr(Value, S.Context)) { 4250 if (S.SourceMgr.isInSystemMacro(CC)) 4251 return; 4252 4253 std::string PrettySourceValue = Value.toString(10); 4254 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 4255 4256 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4257 S.PDiag(diag::warn_impcast_integer_precision_constant) 4258 << PrettySourceValue << PrettyTargetValue 4259 << E->getType() << T << E->getSourceRange() 4260 << clang::SourceRange(CC)); 4261 return; 4262 } 4263 4264 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 4265 if (S.SourceMgr.isInSystemMacro(CC)) 4266 return; 4267 4268 if (TargetRange.Width == 32 && S.Context.getIntWidth(E->getType()) == 64) 4269 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32, 4270 /* pruneControlFlow */ true); 4271 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 4272 } 4273 4274 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 4275 (!TargetRange.NonNegative && SourceRange.NonNegative && 4276 SourceRange.Width == TargetRange.Width)) { 4277 4278 if (S.SourceMgr.isInSystemMacro(CC)) 4279 return; 4280 4281 unsigned DiagID = diag::warn_impcast_integer_sign; 4282 4283 // Traditionally, gcc has warned about this under -Wsign-compare. 4284 // We also want to warn about it in -Wconversion. 4285 // So if -Wconversion is off, use a completely identical diagnostic 4286 // in the sign-compare group. 4287 // The conditional-checking code will 4288 if (ICContext) { 4289 DiagID = diag::warn_impcast_integer_sign_conditional; 4290 *ICContext = true; 4291 } 4292 4293 return DiagnoseImpCast(S, E, T, CC, DiagID); 4294 } 4295 4296 // Diagnose conversions between different enumeration types. 4297 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 4298 // type, to give us better diagnostics. 4299 QualType SourceType = E->getType(); 4300 if (!S.getLangOpts().CPlusPlus) { 4301 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 4302 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 4303 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 4304 SourceType = S.Context.getTypeDeclType(Enum); 4305 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 4306 } 4307 } 4308 4309 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 4310 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 4311 if ((SourceEnum->getDecl()->getIdentifier() || 4312 SourceEnum->getDecl()->getTypedefNameForAnonDecl()) && 4313 (TargetEnum->getDecl()->getIdentifier() || 4314 TargetEnum->getDecl()->getTypedefNameForAnonDecl()) && 4315 SourceEnum != TargetEnum) { 4316 if (S.SourceMgr.isInSystemMacro(CC)) 4317 return; 4318 4319 return DiagnoseImpCast(S, E, SourceType, T, CC, 4320 diag::warn_impcast_different_enum_types); 4321 } 4322 4323 return; 4324} 4325 4326void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 4327 4328void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 4329 SourceLocation CC, bool &ICContext) { 4330 E = E->IgnoreParenImpCasts(); 4331 4332 if (isa<ConditionalOperator>(E)) 4333 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 4334 4335 AnalyzeImplicitConversions(S, E, CC); 4336 if (E->getType() != T) 4337 return CheckImplicitConversion(S, E, T, CC, &ICContext); 4338 return; 4339} 4340 4341void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 4342 SourceLocation CC = E->getQuestionLoc(); 4343 4344 AnalyzeImplicitConversions(S, E->getCond(), CC); 4345 4346 bool Suspicious = false; 4347 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 4348 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 4349 4350 // If -Wconversion would have warned about either of the candidates 4351 // for a signedness conversion to the context type... 4352 if (!Suspicious) return; 4353 4354 // ...but it's currently ignored... 4355 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 4356 CC)) 4357 return; 4358 4359 // ...then check whether it would have warned about either of the 4360 // candidates for a signedness conversion to the condition type. 4361 if (E->getType() == T) return; 4362 4363 Suspicious = false; 4364 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 4365 E->getType(), CC, &Suspicious); 4366 if (!Suspicious) 4367 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 4368 E->getType(), CC, &Suspicious); 4369} 4370 4371/// AnalyzeImplicitConversions - Find and report any interesting 4372/// implicit conversions in the given expression. There are a couple 4373/// of competing diagnostics here, -Wconversion and -Wsign-compare. 4374void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 4375 QualType T = OrigE->getType(); 4376 Expr *E = OrigE->IgnoreParenImpCasts(); 4377 4378 if (E->isTypeDependent() || E->isValueDependent()) 4379 return; 4380 4381 // For conditional operators, we analyze the arguments as if they 4382 // were being fed directly into the output. 4383 if (isa<ConditionalOperator>(E)) { 4384 ConditionalOperator *CO = cast<ConditionalOperator>(E); 4385 CheckConditionalOperator(S, CO, T); 4386 return; 4387 } 4388 4389 // Go ahead and check any implicit conversions we might have skipped. 4390 // The non-canonical typecheck is just an optimization; 4391 // CheckImplicitConversion will filter out dead implicit conversions. 4392 if (E->getType() != T) 4393 CheckImplicitConversion(S, E, T, CC); 4394 4395 // Now continue drilling into this expression. 4396 4397 // Skip past explicit casts. 4398 if (isa<ExplicitCastExpr>(E)) { 4399 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 4400 return AnalyzeImplicitConversions(S, E, CC); 4401 } 4402 4403 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 4404 // Do a somewhat different check with comparison operators. 4405 if (BO->isComparisonOp()) 4406 return AnalyzeComparison(S, BO); 4407 4408 // And with simple assignments. 4409 if (BO->getOpcode() == BO_Assign) 4410 return AnalyzeAssignment(S, BO); 4411 } 4412 4413 // These break the otherwise-useful invariant below. Fortunately, 4414 // we don't really need to recurse into them, because any internal 4415 // expressions should have been analyzed already when they were 4416 // built into statements. 4417 if (isa<StmtExpr>(E)) return; 4418 4419 // Don't descend into unevaluated contexts. 4420 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 4421 4422 // Now just recurse over the expression's children. 4423 CC = E->getExprLoc(); 4424 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 4425 bool IsLogicalOperator = BO && BO->isLogicalOp(); 4426 for (Stmt::child_range I = E->children(); I; ++I) { 4427 Expr *ChildExpr = dyn_cast_or_null<Expr>(*I); 4428 if (!ChildExpr) 4429 continue; 4430 4431 if (IsLogicalOperator && 4432 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 4433 // Ignore checking string literals that are in logical operators. 4434 continue; 4435 AnalyzeImplicitConversions(S, ChildExpr, CC); 4436 } 4437} 4438 4439} // end anonymous namespace 4440 4441/// Diagnoses "dangerous" implicit conversions within the given 4442/// expression (which is a full expression). Implements -Wconversion 4443/// and -Wsign-compare. 4444/// 4445/// \param CC the "context" location of the implicit conversion, i.e. 4446/// the most location of the syntactic entity requiring the implicit 4447/// conversion 4448void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 4449 // Don't diagnose in unevaluated contexts. 4450 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 4451 return; 4452 4453 // Don't diagnose for value- or type-dependent expressions. 4454 if (E->isTypeDependent() || E->isValueDependent()) 4455 return; 4456 4457 // Check for array bounds violations in cases where the check isn't triggered 4458 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 4459 // ArraySubscriptExpr is on the RHS of a variable initialization. 4460 CheckArrayAccess(E); 4461 4462 // This is not the right CC for (e.g.) a variable initialization. 4463 AnalyzeImplicitConversions(*this, E, CC); 4464} 4465 4466void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 4467 FieldDecl *BitField, 4468 Expr *Init) { 4469 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 4470} 4471 4472/// CheckParmsForFunctionDef - Check that the parameters of the given 4473/// function are appropriate for the definition of a function. This 4474/// takes care of any checks that cannot be performed on the 4475/// declaration itself, e.g., that the types of each of the function 4476/// parameters are complete. 4477bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 4478 bool CheckParameterNames) { 4479 bool HasInvalidParm = false; 4480 for (; P != PEnd; ++P) { 4481 ParmVarDecl *Param = *P; 4482 4483 // C99 6.7.5.3p4: the parameters in a parameter type list in a 4484 // function declarator that is part of a function definition of 4485 // that function shall not have incomplete type. 4486 // 4487 // This is also C++ [dcl.fct]p6. 4488 if (!Param->isInvalidDecl() && 4489 RequireCompleteType(Param->getLocation(), Param->getType(), 4490 diag::err_typecheck_decl_incomplete_type)) { 4491 Param->setInvalidDecl(); 4492 HasInvalidParm = true; 4493 } 4494 4495 // C99 6.9.1p5: If the declarator includes a parameter type list, the 4496 // declaration of each parameter shall include an identifier. 4497 if (CheckParameterNames && 4498 Param->getIdentifier() == 0 && 4499 !Param->isImplicit() && 4500 !getLangOpts().CPlusPlus) 4501 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 4502 4503 // C99 6.7.5.3p12: 4504 // If the function declarator is not part of a definition of that 4505 // function, parameters may have incomplete type and may use the [*] 4506 // notation in their sequences of declarator specifiers to specify 4507 // variable length array types. 4508 QualType PType = Param->getOriginalType(); 4509 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 4510 if (AT->getSizeModifier() == ArrayType::Star) { 4511 // FIXME: This diagnosic should point the the '[*]' if source-location 4512 // information is added for it. 4513 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 4514 } 4515 } 4516 } 4517 4518 return HasInvalidParm; 4519} 4520 4521/// CheckCastAlign - Implements -Wcast-align, which warns when a 4522/// pointer cast increases the alignment requirements. 4523void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 4524 // This is actually a lot of work to potentially be doing on every 4525 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 4526 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 4527 TRange.getBegin()) 4528 == DiagnosticsEngine::Ignored) 4529 return; 4530 4531 // Ignore dependent types. 4532 if (T->isDependentType() || Op->getType()->isDependentType()) 4533 return; 4534 4535 // Require that the destination be a pointer type. 4536 const PointerType *DestPtr = T->getAs<PointerType>(); 4537 if (!DestPtr) return; 4538 4539 // If the destination has alignment 1, we're done. 4540 QualType DestPointee = DestPtr->getPointeeType(); 4541 if (DestPointee->isIncompleteType()) return; 4542 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 4543 if (DestAlign.isOne()) return; 4544 4545 // Require that the source be a pointer type. 4546 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 4547 if (!SrcPtr) return; 4548 QualType SrcPointee = SrcPtr->getPointeeType(); 4549 4550 // Whitelist casts from cv void*. We already implicitly 4551 // whitelisted casts to cv void*, since they have alignment 1. 4552 // Also whitelist casts involving incomplete types, which implicitly 4553 // includes 'void'. 4554 if (SrcPointee->isIncompleteType()) return; 4555 4556 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 4557 if (SrcAlign >= DestAlign) return; 4558 4559 Diag(TRange.getBegin(), diag::warn_cast_align) 4560 << Op->getType() << T 4561 << static_cast<unsigned>(SrcAlign.getQuantity()) 4562 << static_cast<unsigned>(DestAlign.getQuantity()) 4563 << TRange << Op->getSourceRange(); 4564} 4565 4566static const Type* getElementType(const Expr *BaseExpr) { 4567 const Type* EltType = BaseExpr->getType().getTypePtr(); 4568 if (EltType->isAnyPointerType()) 4569 return EltType->getPointeeType().getTypePtr(); 4570 else if (EltType->isArrayType()) 4571 return EltType->getBaseElementTypeUnsafe(); 4572 return EltType; 4573} 4574 4575/// \brief Check whether this array fits the idiom of a size-one tail padded 4576/// array member of a struct. 4577/// 4578/// We avoid emitting out-of-bounds access warnings for such arrays as they are 4579/// commonly used to emulate flexible arrays in C89 code. 4580static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 4581 const NamedDecl *ND) { 4582 if (Size != 1 || !ND) return false; 4583 4584 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 4585 if (!FD) return false; 4586 4587 // Don't consider sizes resulting from macro expansions or template argument 4588 // substitution to form C89 tail-padded arrays. 4589 ConstantArrayTypeLoc TL = 4590 cast<ConstantArrayTypeLoc>(FD->getTypeSourceInfo()->getTypeLoc()); 4591 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(TL.getSizeExpr()); 4592 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 4593 return false; 4594 4595 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 4596 if (!RD) return false; 4597 if (RD->isUnion()) return false; 4598 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 4599 if (!CRD->isStandardLayout()) return false; 4600 } 4601 4602 // See if this is the last field decl in the record. 4603 const Decl *D = FD; 4604 while ((D = D->getNextDeclInContext())) 4605 if (isa<FieldDecl>(D)) 4606 return false; 4607 return true; 4608} 4609 4610void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 4611 const ArraySubscriptExpr *ASE, 4612 bool AllowOnePastEnd, bool IndexNegated) { 4613 IndexExpr = IndexExpr->IgnoreParenImpCasts(); 4614 if (IndexExpr->isValueDependent()) 4615 return; 4616 4617 const Type *EffectiveType = getElementType(BaseExpr); 4618 BaseExpr = BaseExpr->IgnoreParenCasts(); 4619 const ConstantArrayType *ArrayTy = 4620 Context.getAsConstantArrayType(BaseExpr->getType()); 4621 if (!ArrayTy) 4622 return; 4623 4624 llvm::APSInt index; 4625 if (!IndexExpr->EvaluateAsInt(index, Context)) 4626 return; 4627 if (IndexNegated) 4628 index = -index; 4629 4630 const NamedDecl *ND = NULL; 4631 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4632 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 4633 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 4634 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 4635 4636 if (index.isUnsigned() || !index.isNegative()) { 4637 llvm::APInt size = ArrayTy->getSize(); 4638 if (!size.isStrictlyPositive()) 4639 return; 4640 4641 const Type* BaseType = getElementType(BaseExpr); 4642 if (BaseType != EffectiveType) { 4643 // Make sure we're comparing apples to apples when comparing index to size 4644 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 4645 uint64_t array_typesize = Context.getTypeSize(BaseType); 4646 // Handle ptrarith_typesize being zero, such as when casting to void* 4647 if (!ptrarith_typesize) ptrarith_typesize = 1; 4648 if (ptrarith_typesize != array_typesize) { 4649 // There's a cast to a different size type involved 4650 uint64_t ratio = array_typesize / ptrarith_typesize; 4651 // TODO: Be smarter about handling cases where array_typesize is not a 4652 // multiple of ptrarith_typesize 4653 if (ptrarith_typesize * ratio == array_typesize) 4654 size *= llvm::APInt(size.getBitWidth(), ratio); 4655 } 4656 } 4657 4658 if (size.getBitWidth() > index.getBitWidth()) 4659 index = index.zext(size.getBitWidth()); 4660 else if (size.getBitWidth() < index.getBitWidth()) 4661 size = size.zext(index.getBitWidth()); 4662 4663 // For array subscripting the index must be less than size, but for pointer 4664 // arithmetic also allow the index (offset) to be equal to size since 4665 // computing the next address after the end of the array is legal and 4666 // commonly done e.g. in C++ iterators and range-based for loops. 4667 if (AllowOnePastEnd ? index.ule(size) : index.ult(size)) 4668 return; 4669 4670 // Also don't warn for arrays of size 1 which are members of some 4671 // structure. These are often used to approximate flexible arrays in C89 4672 // code. 4673 if (IsTailPaddedMemberArray(*this, size, ND)) 4674 return; 4675 4676 // Suppress the warning if the subscript expression (as identified by the 4677 // ']' location) and the index expression are both from macro expansions 4678 // within a system header. 4679 if (ASE) { 4680 SourceLocation RBracketLoc = SourceMgr.getSpellingLoc( 4681 ASE->getRBracketLoc()); 4682 if (SourceMgr.isInSystemHeader(RBracketLoc)) { 4683 SourceLocation IndexLoc = SourceMgr.getSpellingLoc( 4684 IndexExpr->getLocStart()); 4685 if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc)) 4686 return; 4687 } 4688 } 4689 4690 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 4691 if (ASE) 4692 DiagID = diag::warn_array_index_exceeds_bounds; 4693 4694 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4695 PDiag(DiagID) << index.toString(10, true) 4696 << size.toString(10, true) 4697 << (unsigned)size.getLimitedValue(~0U) 4698 << IndexExpr->getSourceRange()); 4699 } else { 4700 unsigned DiagID = diag::warn_array_index_precedes_bounds; 4701 if (!ASE) { 4702 DiagID = diag::warn_ptr_arith_precedes_bounds; 4703 if (index.isNegative()) index = -index; 4704 } 4705 4706 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4707 PDiag(DiagID) << index.toString(10, true) 4708 << IndexExpr->getSourceRange()); 4709 } 4710 4711 if (!ND) { 4712 // Try harder to find a NamedDecl to point at in the note. 4713 while (const ArraySubscriptExpr *ASE = 4714 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 4715 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 4716 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4717 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 4718 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 4719 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 4720 } 4721 4722 if (ND) 4723 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 4724 PDiag(diag::note_array_index_out_of_bounds) 4725 << ND->getDeclName()); 4726} 4727 4728void Sema::CheckArrayAccess(const Expr *expr) { 4729 int AllowOnePastEnd = 0; 4730 while (expr) { 4731 expr = expr->IgnoreParenImpCasts(); 4732 switch (expr->getStmtClass()) { 4733 case Stmt::ArraySubscriptExprClass: { 4734 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 4735 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE, 4736 AllowOnePastEnd > 0); 4737 return; 4738 } 4739 case Stmt::UnaryOperatorClass: { 4740 // Only unwrap the * and & unary operators 4741 const UnaryOperator *UO = cast<UnaryOperator>(expr); 4742 expr = UO->getSubExpr(); 4743 switch (UO->getOpcode()) { 4744 case UO_AddrOf: 4745 AllowOnePastEnd++; 4746 break; 4747 case UO_Deref: 4748 AllowOnePastEnd--; 4749 break; 4750 default: 4751 return; 4752 } 4753 break; 4754 } 4755 case Stmt::ConditionalOperatorClass: { 4756 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 4757 if (const Expr *lhs = cond->getLHS()) 4758 CheckArrayAccess(lhs); 4759 if (const Expr *rhs = cond->getRHS()) 4760 CheckArrayAccess(rhs); 4761 return; 4762 } 4763 default: 4764 return; 4765 } 4766 } 4767} 4768 4769//===--- CHECK: Objective-C retain cycles ----------------------------------// 4770 4771namespace { 4772 struct RetainCycleOwner { 4773 RetainCycleOwner() : Variable(0), Indirect(false) {} 4774 VarDecl *Variable; 4775 SourceRange Range; 4776 SourceLocation Loc; 4777 bool Indirect; 4778 4779 void setLocsFrom(Expr *e) { 4780 Loc = e->getExprLoc(); 4781 Range = e->getSourceRange(); 4782 } 4783 }; 4784} 4785 4786/// Consider whether capturing the given variable can possibly lead to 4787/// a retain cycle. 4788static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 4789 // In ARC, it's captured strongly iff the variable has __strong 4790 // lifetime. In MRR, it's captured strongly if the variable is 4791 // __block and has an appropriate type. 4792 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4793 return false; 4794 4795 owner.Variable = var; 4796 owner.setLocsFrom(ref); 4797 return true; 4798} 4799 4800static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) { 4801 while (true) { 4802 e = e->IgnoreParens(); 4803 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 4804 switch (cast->getCastKind()) { 4805 case CK_BitCast: 4806 case CK_LValueBitCast: 4807 case CK_LValueToRValue: 4808 case CK_ARCReclaimReturnedObject: 4809 e = cast->getSubExpr(); 4810 continue; 4811 4812 default: 4813 return false; 4814 } 4815 } 4816 4817 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 4818 ObjCIvarDecl *ivar = ref->getDecl(); 4819 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4820 return false; 4821 4822 // Try to find a retain cycle in the base. 4823 if (!findRetainCycleOwner(S, ref->getBase(), owner)) 4824 return false; 4825 4826 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 4827 owner.Indirect = true; 4828 return true; 4829 } 4830 4831 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 4832 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 4833 if (!var) return false; 4834 return considerVariable(var, ref, owner); 4835 } 4836 4837 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 4838 if (member->isArrow()) return false; 4839 4840 // Don't count this as an indirect ownership. 4841 e = member->getBase(); 4842 continue; 4843 } 4844 4845 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 4846 // Only pay attention to pseudo-objects on property references. 4847 ObjCPropertyRefExpr *pre 4848 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 4849 ->IgnoreParens()); 4850 if (!pre) return false; 4851 if (pre->isImplicitProperty()) return false; 4852 ObjCPropertyDecl *property = pre->getExplicitProperty(); 4853 if (!property->isRetaining() && 4854 !(property->getPropertyIvarDecl() && 4855 property->getPropertyIvarDecl()->getType() 4856 .getObjCLifetime() == Qualifiers::OCL_Strong)) 4857 return false; 4858 4859 owner.Indirect = true; 4860 if (pre->isSuperReceiver()) { 4861 owner.Variable = S.getCurMethodDecl()->getSelfDecl(); 4862 if (!owner.Variable) 4863 return false; 4864 owner.Loc = pre->getLocation(); 4865 owner.Range = pre->getSourceRange(); 4866 return true; 4867 } 4868 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 4869 ->getSourceExpr()); 4870 continue; 4871 } 4872 4873 // Array ivars? 4874 4875 return false; 4876 } 4877} 4878 4879namespace { 4880 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 4881 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 4882 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 4883 Variable(variable), Capturer(0) {} 4884 4885 VarDecl *Variable; 4886 Expr *Capturer; 4887 4888 void VisitDeclRefExpr(DeclRefExpr *ref) { 4889 if (ref->getDecl() == Variable && !Capturer) 4890 Capturer = ref; 4891 } 4892 4893 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 4894 if (Capturer) return; 4895 Visit(ref->getBase()); 4896 if (Capturer && ref->isFreeIvar()) 4897 Capturer = ref; 4898 } 4899 4900 void VisitBlockExpr(BlockExpr *block) { 4901 // Look inside nested blocks 4902 if (block->getBlockDecl()->capturesVariable(Variable)) 4903 Visit(block->getBlockDecl()->getBody()); 4904 } 4905 }; 4906} 4907 4908/// Check whether the given argument is a block which captures a 4909/// variable. 4910static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 4911 assert(owner.Variable && owner.Loc.isValid()); 4912 4913 e = e->IgnoreParenCasts(); 4914 BlockExpr *block = dyn_cast<BlockExpr>(e); 4915 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 4916 return 0; 4917 4918 FindCaptureVisitor visitor(S.Context, owner.Variable); 4919 visitor.Visit(block->getBlockDecl()->getBody()); 4920 return visitor.Capturer; 4921} 4922 4923static void diagnoseRetainCycle(Sema &S, Expr *capturer, 4924 RetainCycleOwner &owner) { 4925 assert(capturer); 4926 assert(owner.Variable && owner.Loc.isValid()); 4927 4928 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 4929 << owner.Variable << capturer->getSourceRange(); 4930 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 4931 << owner.Indirect << owner.Range; 4932} 4933 4934/// Check for a keyword selector that starts with the word 'add' or 4935/// 'set'. 4936static bool isSetterLikeSelector(Selector sel) { 4937 if (sel.isUnarySelector()) return false; 4938 4939 StringRef str = sel.getNameForSlot(0); 4940 while (!str.empty() && str.front() == '_') str = str.substr(1); 4941 if (str.startswith("set")) 4942 str = str.substr(3); 4943 else if (str.startswith("add")) { 4944 // Specially whitelist 'addOperationWithBlock:'. 4945 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 4946 return false; 4947 str = str.substr(3); 4948 } 4949 else 4950 return false; 4951 4952 if (str.empty()) return true; 4953 return !islower(str.front()); 4954} 4955 4956/// Check a message send to see if it's likely to cause a retain cycle. 4957void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 4958 // Only check instance methods whose selector looks like a setter. 4959 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 4960 return; 4961 4962 // Try to find a variable that the receiver is strongly owned by. 4963 RetainCycleOwner owner; 4964 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 4965 if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner)) 4966 return; 4967 } else { 4968 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 4969 owner.Variable = getCurMethodDecl()->getSelfDecl(); 4970 owner.Loc = msg->getSuperLoc(); 4971 owner.Range = msg->getSuperLoc(); 4972 } 4973 4974 // Check whether the receiver is captured by any of the arguments. 4975 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 4976 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 4977 return diagnoseRetainCycle(*this, capturer, owner); 4978} 4979 4980/// Check a property assign to see if it's likely to cause a retain cycle. 4981void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 4982 RetainCycleOwner owner; 4983 if (!findRetainCycleOwner(*this, receiver, owner)) 4984 return; 4985 4986 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 4987 diagnoseRetainCycle(*this, capturer, owner); 4988} 4989 4990bool Sema::checkUnsafeAssigns(SourceLocation Loc, 4991 QualType LHS, Expr *RHS) { 4992 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 4993 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 4994 return false; 4995 // strip off any implicit cast added to get to the one arc-specific 4996 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4997 if (cast->getCastKind() == CK_ARCConsumeObject) { 4998 Diag(Loc, diag::warn_arc_retained_assign) 4999 << (LT == Qualifiers::OCL_ExplicitNone) 5000 << RHS->getSourceRange(); 5001 return true; 5002 } 5003 RHS = cast->getSubExpr(); 5004 } 5005 return false; 5006} 5007 5008void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 5009 Expr *LHS, Expr *RHS) { 5010 QualType LHSType; 5011 // PropertyRef on LHS type need be directly obtained from 5012 // its declaration as it has a PsuedoType. 5013 ObjCPropertyRefExpr *PRE 5014 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens()); 5015 if (PRE && !PRE->isImplicitProperty()) { 5016 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 5017 if (PD) 5018 LHSType = PD->getType(); 5019 } 5020 5021 if (LHSType.isNull()) 5022 LHSType = LHS->getType(); 5023 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 5024 return; 5025 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 5026 // FIXME. Check for other life times. 5027 if (LT != Qualifiers::OCL_None) 5028 return; 5029 5030 if (PRE) { 5031 if (PRE->isImplicitProperty()) 5032 return; 5033 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 5034 if (!PD) 5035 return; 5036 5037 unsigned Attributes = PD->getPropertyAttributes(); 5038 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) { 5039 // when 'assign' attribute was not explicitly specified 5040 // by user, ignore it and rely on property type itself 5041 // for lifetime info. 5042 unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten(); 5043 if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) && 5044 LHSType->isObjCRetainableType()) 5045 return; 5046 5047 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 5048 if (cast->getCastKind() == CK_ARCConsumeObject) { 5049 Diag(Loc, diag::warn_arc_retained_property_assign) 5050 << RHS->getSourceRange(); 5051 return; 5052 } 5053 RHS = cast->getSubExpr(); 5054 } 5055 } 5056 } 5057} 5058 5059//===--- CHECK: Empty statement body (-Wempty-body) ---------------------===// 5060 5061namespace { 5062bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr, 5063 SourceLocation StmtLoc, 5064 const NullStmt *Body) { 5065 // Do not warn if the body is a macro that expands to nothing, e.g: 5066 // 5067 // #define CALL(x) 5068 // if (condition) 5069 // CALL(0); 5070 // 5071 if (Body->hasLeadingEmptyMacro()) 5072 return false; 5073 5074 // Get line numbers of statement and body. 5075 bool StmtLineInvalid; 5076 unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc, 5077 &StmtLineInvalid); 5078 if (StmtLineInvalid) 5079 return false; 5080 5081 bool BodyLineInvalid; 5082 unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(), 5083 &BodyLineInvalid); 5084 if (BodyLineInvalid) 5085 return false; 5086 5087 // Warn if null statement and body are on the same line. 5088 if (StmtLine != BodyLine) 5089 return false; 5090 5091 return true; 5092} 5093} // Unnamed namespace 5094 5095void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc, 5096 const Stmt *Body, 5097 unsigned DiagID) { 5098 // Since this is a syntactic check, don't emit diagnostic for template 5099 // instantiations, this just adds noise. 5100 if (CurrentInstantiationScope) 5101 return; 5102 5103 // The body should be a null statement. 5104 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 5105 if (!NBody) 5106 return; 5107 5108 // Do the usual checks. 5109 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 5110 return; 5111 5112 Diag(NBody->getSemiLoc(), DiagID); 5113 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 5114} 5115 5116void Sema::DiagnoseEmptyLoopBody(const Stmt *S, 5117 const Stmt *PossibleBody) { 5118 assert(!CurrentInstantiationScope); // Ensured by caller 5119 5120 SourceLocation StmtLoc; 5121 const Stmt *Body; 5122 unsigned DiagID; 5123 if (const ForStmt *FS = dyn_cast<ForStmt>(S)) { 5124 StmtLoc = FS->getRParenLoc(); 5125 Body = FS->getBody(); 5126 DiagID = diag::warn_empty_for_body; 5127 } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) { 5128 StmtLoc = WS->getCond()->getSourceRange().getEnd(); 5129 Body = WS->getBody(); 5130 DiagID = diag::warn_empty_while_body; 5131 } else 5132 return; // Neither `for' nor `while'. 5133 5134 // The body should be a null statement. 5135 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 5136 if (!NBody) 5137 return; 5138 5139 // Skip expensive checks if diagnostic is disabled. 5140 if (Diags.getDiagnosticLevel(DiagID, NBody->getSemiLoc()) == 5141 DiagnosticsEngine::Ignored) 5142 return; 5143 5144 // Do the usual checks. 5145 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 5146 return; 5147 5148 // `for(...);' and `while(...);' are popular idioms, so in order to keep 5149 // noise level low, emit diagnostics only if for/while is followed by a 5150 // CompoundStmt, e.g.: 5151 // for (int i = 0; i < n; i++); 5152 // { 5153 // a(i); 5154 // } 5155 // or if for/while is followed by a statement with more indentation 5156 // than for/while itself: 5157 // for (int i = 0; i < n; i++); 5158 // a(i); 5159 bool ProbableTypo = isa<CompoundStmt>(PossibleBody); 5160 if (!ProbableTypo) { 5161 bool BodyColInvalid; 5162 unsigned BodyCol = SourceMgr.getPresumedColumnNumber( 5163 PossibleBody->getLocStart(), 5164 &BodyColInvalid); 5165 if (BodyColInvalid) 5166 return; 5167 5168 bool StmtColInvalid; 5169 unsigned StmtCol = SourceMgr.getPresumedColumnNumber( 5170 S->getLocStart(), 5171 &StmtColInvalid); 5172 if (StmtColInvalid) 5173 return; 5174 5175 if (BodyCol > StmtCol) 5176 ProbableTypo = true; 5177 } 5178 5179 if (ProbableTypo) { 5180 Diag(NBody->getSemiLoc(), DiagID); 5181 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 5182 } 5183} 5184