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