SemaChecking.cpp revision 276b061970939293f1abaf694bd3ef05b2cbda79
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> 39using namespace clang; 40using namespace sema; 41 42SourceLocation 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? 51bool 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. 74static 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. 94static 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 105ExprResult 106Sema::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. 247static unsigned RFT(unsigned t, bool shift = false) { 248 bool quad = t & 0x10; 249 250 switch (t & 0x7) { 251 case 0: // i8 252 return shift ? 7 : (8 << (int)quad) - 1; 253 case 1: // i16 254 return shift ? 15 : (4 << (int)quad) - 1; 255 case 2: // i32 256 return shift ? 31 : (2 << (int)quad) - 1; 257 case 3: // i64 258 return shift ? 63 : (1 << (int)quad) - 1; 259 case 4: // f32 260 assert(!shift && "cannot shift float types!"); 261 return (2 << (int)quad) - 1; 262 case 5: // poly8 263 return shift ? 7 : (8 << (int)quad) - 1; 264 case 6: // poly16 265 return shift ? 15 : (4 << (int)quad) - 1; 266 case 7: // float16 267 assert(!shift && "cannot shift float types!"); 268 return (4 << (int)quad) - 1; 269 } 270 return 0; 271} 272 273bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 274 llvm::APSInt Result; 275 276 unsigned mask = 0; 277 unsigned TV = 0; 278 switch (BuiltinID) { 279#define GET_NEON_OVERLOAD_CHECK 280#include "clang/Basic/arm_neon.inc" 281#undef GET_NEON_OVERLOAD_CHECK 282 } 283 284 // For NEON intrinsics which are overloaded on vector element type, validate 285 // the immediate which specifies which variant to emit. 286 if (mask) { 287 unsigned ArgNo = TheCall->getNumArgs()-1; 288 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 289 return true; 290 291 TV = Result.getLimitedValue(32); 292 if ((TV > 31) || (mask & (1 << TV)) == 0) 293 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 294 << TheCall->getArg(ArgNo)->getSourceRange(); 295 } 296 297 // For NEON intrinsics which take an immediate value as part of the 298 // instruction, range check them here. 299 unsigned i = 0, l = 0, u = 0; 300 switch (BuiltinID) { 301 default: return false; 302 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 303 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 304 case ARM::BI__builtin_arm_vcvtr_f: 305 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 306#define GET_NEON_IMMEDIATE_CHECK 307#include "clang/Basic/arm_neon.inc" 308#undef GET_NEON_IMMEDIATE_CHECK 309 }; 310 311 // Check that the immediate argument is actually a constant. 312 if (SemaBuiltinConstantArg(TheCall, i, Result)) 313 return true; 314 315 // Range check against the upper/lower values for this isntruction. 316 unsigned Val = Result.getZExtValue(); 317 if (Val < l || Val > (u + l)) 318 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 319 << l << u+l << TheCall->getArg(i)->getSourceRange(); 320 321 // FIXME: VFP Intrinsics should error if VFP not present. 322 return false; 323} 324 325/// CheckFunctionCall - Check a direct function call for various correctness 326/// and safety properties not strictly enforced by the C type system. 327bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 328 // Get the IdentifierInfo* for the called function. 329 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 330 331 // None of the checks below are needed for functions that don't have 332 // simple names (e.g., C++ conversion functions). 333 if (!FnInfo) 334 return false; 335 336 // FIXME: This mechanism should be abstracted to be less fragile and 337 // more efficient. For example, just map function ids to custom 338 // handlers. 339 340 // Printf and scanf checking. 341 for (specific_attr_iterator<FormatAttr> 342 i = FDecl->specific_attr_begin<FormatAttr>(), 343 e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { 344 345 const FormatAttr *Format = *i; 346 const bool b = Format->getType() == "scanf"; 347 if (b || CheckablePrintfAttr(Format, TheCall)) { 348 bool HasVAListArg = Format->getFirstArg() == 0; 349 CheckPrintfScanfArguments(TheCall, HasVAListArg, 350 Format->getFormatIdx() - 1, 351 HasVAListArg ? 0 : Format->getFirstArg() - 1, 352 !b); 353 } 354 } 355 356 for (specific_attr_iterator<NonNullAttr> 357 i = FDecl->specific_attr_begin<NonNullAttr>(), 358 e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) { 359 CheckNonNullArguments(*i, TheCall->getArgs(), 360 TheCall->getCallee()->getLocStart()); 361 } 362 363 // Builtin handling 364 int CMF = -1; 365 switch (FDecl->getBuiltinID()) { 366 case Builtin::BI__builtin_memset: 367 case Builtin::BI__builtin___memset_chk: 368 case Builtin::BImemset: 369 CMF = CMF_Memset; 370 break; 371 372 case Builtin::BI__builtin_memcpy: 373 case Builtin::BI__builtin___memcpy_chk: 374 case Builtin::BImemcpy: 375 CMF = CMF_Memcpy; 376 break; 377 378 case Builtin::BI__builtin_memmove: 379 case Builtin::BI__builtin___memmove_chk: 380 case Builtin::BImemmove: 381 CMF = CMF_Memmove; 382 break; 383 384 case Builtin::BIstrlcpy: 385 case Builtin::BIstrlcat: 386 CheckStrlcpycatArguments(TheCall, FnInfo); 387 break; 388 389 case Builtin::BI__builtin_memcmp: 390 CMF = CMF_Memcmp; 391 break; 392 393 default: 394 if (FDecl->getLinkage() == ExternalLinkage && 395 (!getLangOptions().CPlusPlus || FDecl->isExternC())) { 396 if (FnInfo->isStr("memset")) 397 CMF = CMF_Memset; 398 else if (FnInfo->isStr("memcpy")) 399 CMF = CMF_Memcpy; 400 else if (FnInfo->isStr("memmove")) 401 CMF = CMF_Memmove; 402 else if (FnInfo->isStr("memcmp")) 403 CMF = CMF_Memcmp; 404 } 405 break; 406 } 407 408 // Memset/memcpy/memmove handling 409 if (CMF != -1) 410 CheckMemaccessArguments(TheCall, CheckedMemoryFunction(CMF), FnInfo); 411 412 return false; 413} 414 415bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 416 // Printf checking. 417 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 418 if (!Format) 419 return false; 420 421 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 422 if (!V) 423 return false; 424 425 QualType Ty = V->getType(); 426 if (!Ty->isBlockPointerType()) 427 return false; 428 429 const bool b = Format->getType() == "scanf"; 430 if (!b && !CheckablePrintfAttr(Format, TheCall)) 431 return false; 432 433 bool HasVAListArg = Format->getFirstArg() == 0; 434 CheckPrintfScanfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 435 HasVAListArg ? 0 : Format->getFirstArg() - 1, !b); 436 437 return false; 438} 439 440ExprResult 441Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, AtomicExpr::AtomicOp Op) { 442 CallExpr *TheCall = cast<CallExpr>(TheCallResult.get()); 443 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 444 Expr *Ptr, *Order, *Val1, *Val2, *OrderFail; 445 446 // All these operations take one of the following four forms: 447 // T __atomic_load(_Atomic(T)*, int) (loads) 448 // T* __atomic_add(_Atomic(T*)*, ptrdiff_t, int) (pointer add/sub) 449 // int __atomic_compare_exchange_strong(_Atomic(T)*, T*, T, int, int) 450 // (cmpxchg) 451 // T __atomic_exchange(_Atomic(T)*, T, int) (everything else) 452 // where T is an appropriate type, and the int paremeterss are for orderings. 453 unsigned NumVals = 1; 454 unsigned NumOrders = 1; 455 if (Op == AtomicExpr::Load) { 456 NumVals = 0; 457 } else if (Op == AtomicExpr::CmpXchgWeak || Op == AtomicExpr::CmpXchgStrong) { 458 NumVals = 2; 459 NumOrders = 2; 460 } 461 462 if (TheCall->getNumArgs() < NumVals+NumOrders+1) { 463 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 464 << 0 << NumVals+NumOrders+1 << TheCall->getNumArgs() 465 << TheCall->getCallee()->getSourceRange(); 466 return ExprError(); 467 } else if (TheCall->getNumArgs() > NumVals+NumOrders+1) { 468 Diag(TheCall->getArg(NumVals+NumOrders+1)->getLocStart(), 469 diag::err_typecheck_call_too_many_args) 470 << 0 << NumVals+NumOrders+1 << TheCall->getNumArgs() 471 << TheCall->getCallee()->getSourceRange(); 472 return ExprError(); 473 } 474 475 // Inspect the first argument of the atomic operation. This should always be 476 // a pointer to an _Atomic type. 477 Ptr = TheCall->getArg(0); 478 Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get(); 479 const PointerType *pointerType = Ptr->getType()->getAs<PointerType>(); 480 if (!pointerType) { 481 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 482 << Ptr->getType() << Ptr->getSourceRange(); 483 return ExprError(); 484 } 485 486 QualType AtomTy = pointerType->getPointeeType(); 487 if (!AtomTy->isAtomicType()) { 488 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 489 << Ptr->getType() << Ptr->getSourceRange(); 490 return ExprError(); 491 } 492 QualType ValType = AtomTy->getAs<AtomicType>()->getValueType(); 493 494 if ((Op == AtomicExpr::Add || Op == AtomicExpr::Sub) && 495 !ValType->isIntegerType() && !ValType->isPointerType()) { 496 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 497 << Ptr->getType() << Ptr->getSourceRange(); 498 return ExprError(); 499 } 500 501 if (!ValType->isIntegerType() && 502 (Op == AtomicExpr::And || Op == AtomicExpr::Or || Op == AtomicExpr::Xor)){ 503 Diag(DRE->getLocStart(), diag::err_atomic_op_logical_needs_atomic_int) 504 << Ptr->getType() << Ptr->getSourceRange(); 505 return ExprError(); 506 } 507 508 switch (ValType.getObjCLifetime()) { 509 case Qualifiers::OCL_None: 510 case Qualifiers::OCL_ExplicitNone: 511 // okay 512 break; 513 514 case Qualifiers::OCL_Weak: 515 case Qualifiers::OCL_Strong: 516 case Qualifiers::OCL_Autoreleasing: 517 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 518 << ValType << Ptr->getSourceRange(); 519 return ExprError(); 520 } 521 522 QualType ResultType = ValType; 523 if (Op == AtomicExpr::Store) 524 ResultType = Context.VoidTy; 525 else if (Op == AtomicExpr::CmpXchgWeak || Op == AtomicExpr::CmpXchgStrong) 526 ResultType = Context.BoolTy; 527 528 // The first argument --- the pointer --- has a fixed type; we 529 // deduce the types of the rest of the arguments accordingly. Walk 530 // the remaining arguments, converting them to the deduced value type. 531 for (unsigned i = 1; i != NumVals+NumOrders+1; ++i) { 532 ExprResult Arg = TheCall->getArg(i); 533 QualType Ty; 534 if (i < NumVals+1) { 535 // The second argument to a cmpxchg is a pointer to the data which will 536 // be exchanged. The second argument to a pointer add/subtract is the 537 // amount to add/subtract, which must be a ptrdiff_t. The third 538 // argument to a cmpxchg and the second argument in all other cases 539 // is the type of the value. 540 if (i == 1 && (Op == AtomicExpr::CmpXchgWeak || 541 Op == AtomicExpr::CmpXchgStrong)) 542 Ty = Context.getPointerType(ValType.getUnqualifiedType()); 543 else if (!ValType->isIntegerType() && 544 (Op == AtomicExpr::Add || Op == AtomicExpr::Sub)) 545 Ty = Context.getPointerDiffType(); 546 else 547 Ty = ValType; 548 } else { 549 // The order(s) are always converted to int. 550 Ty = Context.IntTy; 551 } 552 InitializedEntity Entity = 553 InitializedEntity::InitializeParameter(Context, Ty, false); 554 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 555 if (Arg.isInvalid()) 556 return true; 557 TheCall->setArg(i, Arg.get()); 558 } 559 560 if (Op == AtomicExpr::Load) { 561 Order = TheCall->getArg(1); 562 return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 563 Ptr, Order, ResultType, Op, 564 TheCall->getRParenLoc(), false, 565 false)); 566 } else if (Op != AtomicExpr::CmpXchgWeak && Op != AtomicExpr::CmpXchgStrong) { 567 Val1 = TheCall->getArg(1); 568 Order = TheCall->getArg(2); 569 return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 570 Ptr, Val1, Order, ResultType, Op, 571 TheCall->getRParenLoc(), false, 572 false)); 573 } else { 574 Val1 = TheCall->getArg(1); 575 Val2 = TheCall->getArg(2); 576 Order = TheCall->getArg(3); 577 OrderFail = TheCall->getArg(4); 578 return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 579 Ptr, Val1, Val2, Order, OrderFail, 580 ResultType, Op, 581 TheCall->getRParenLoc(), false, 582 false)); 583 } 584} 585 586 587/// checkBuiltinArgument - Given a call to a builtin function, perform 588/// normal type-checking on the given argument, updating the call in 589/// place. This is useful when a builtin function requires custom 590/// type-checking for some of its arguments but not necessarily all of 591/// them. 592/// 593/// Returns true on error. 594static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 595 FunctionDecl *Fn = E->getDirectCallee(); 596 assert(Fn && "builtin call without direct callee!"); 597 598 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 599 InitializedEntity Entity = 600 InitializedEntity::InitializeParameter(S.Context, Param); 601 602 ExprResult Arg = E->getArg(0); 603 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 604 if (Arg.isInvalid()) 605 return true; 606 607 E->setArg(ArgIndex, Arg.take()); 608 return false; 609} 610 611/// SemaBuiltinAtomicOverloaded - We have a call to a function like 612/// __sync_fetch_and_add, which is an overloaded function based on the pointer 613/// type of its first argument. The main ActOnCallExpr routines have already 614/// promoted the types of arguments because all of these calls are prototyped as 615/// void(...). 616/// 617/// This function goes through and does final semantic checking for these 618/// builtins, 619ExprResult 620Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 621 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 622 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 623 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 624 625 // Ensure that we have at least one argument to do type inference from. 626 if (TheCall->getNumArgs() < 1) { 627 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 628 << 0 << 1 << TheCall->getNumArgs() 629 << TheCall->getCallee()->getSourceRange(); 630 return ExprError(); 631 } 632 633 // Inspect the first argument of the atomic builtin. This should always be 634 // a pointer type, whose element is an integral scalar or pointer type. 635 // Because it is a pointer type, we don't have to worry about any implicit 636 // casts here. 637 // FIXME: We don't allow floating point scalars as input. 638 Expr *FirstArg = TheCall->getArg(0); 639 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 640 if (!pointerType) { 641 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 642 << FirstArg->getType() << FirstArg->getSourceRange(); 643 return ExprError(); 644 } 645 646 QualType ValType = pointerType->getPointeeType(); 647 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 648 !ValType->isBlockPointerType()) { 649 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 650 << FirstArg->getType() << FirstArg->getSourceRange(); 651 return ExprError(); 652 } 653 654 switch (ValType.getObjCLifetime()) { 655 case Qualifiers::OCL_None: 656 case Qualifiers::OCL_ExplicitNone: 657 // okay 658 break; 659 660 case Qualifiers::OCL_Weak: 661 case Qualifiers::OCL_Strong: 662 case Qualifiers::OCL_Autoreleasing: 663 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 664 << ValType << FirstArg->getSourceRange(); 665 return ExprError(); 666 } 667 668 // Strip any qualifiers off ValType. 669 ValType = ValType.getUnqualifiedType(); 670 671 // The majority of builtins return a value, but a few have special return 672 // types, so allow them to override appropriately below. 673 QualType ResultType = ValType; 674 675 // We need to figure out which concrete builtin this maps onto. For example, 676 // __sync_fetch_and_add with a 2 byte object turns into 677 // __sync_fetch_and_add_2. 678#define BUILTIN_ROW(x) \ 679 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 680 Builtin::BI##x##_8, Builtin::BI##x##_16 } 681 682 static const unsigned BuiltinIndices[][5] = { 683 BUILTIN_ROW(__sync_fetch_and_add), 684 BUILTIN_ROW(__sync_fetch_and_sub), 685 BUILTIN_ROW(__sync_fetch_and_or), 686 BUILTIN_ROW(__sync_fetch_and_and), 687 BUILTIN_ROW(__sync_fetch_and_xor), 688 689 BUILTIN_ROW(__sync_add_and_fetch), 690 BUILTIN_ROW(__sync_sub_and_fetch), 691 BUILTIN_ROW(__sync_and_and_fetch), 692 BUILTIN_ROW(__sync_or_and_fetch), 693 BUILTIN_ROW(__sync_xor_and_fetch), 694 695 BUILTIN_ROW(__sync_val_compare_and_swap), 696 BUILTIN_ROW(__sync_bool_compare_and_swap), 697 BUILTIN_ROW(__sync_lock_test_and_set), 698 BUILTIN_ROW(__sync_lock_release), 699 BUILTIN_ROW(__sync_swap) 700 }; 701#undef BUILTIN_ROW 702 703 // Determine the index of the size. 704 unsigned SizeIndex; 705 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 706 case 1: SizeIndex = 0; break; 707 case 2: SizeIndex = 1; break; 708 case 4: SizeIndex = 2; break; 709 case 8: SizeIndex = 3; break; 710 case 16: SizeIndex = 4; break; 711 default: 712 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 713 << FirstArg->getType() << FirstArg->getSourceRange(); 714 return ExprError(); 715 } 716 717 // Each of these builtins has one pointer argument, followed by some number of 718 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 719 // that we ignore. Find out which row of BuiltinIndices to read from as well 720 // as the number of fixed args. 721 unsigned BuiltinID = FDecl->getBuiltinID(); 722 unsigned BuiltinIndex, NumFixed = 1; 723 switch (BuiltinID) { 724 default: llvm_unreachable("Unknown overloaded atomic builtin!"); 725 case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break; 726 case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break; 727 case Builtin::BI__sync_fetch_and_or: BuiltinIndex = 2; break; 728 case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break; 729 case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break; 730 731 case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break; 732 case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break; 733 case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break; 734 case Builtin::BI__sync_or_and_fetch: BuiltinIndex = 8; break; 735 case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break; 736 737 case Builtin::BI__sync_val_compare_and_swap: 738 BuiltinIndex = 10; 739 NumFixed = 2; 740 break; 741 case Builtin::BI__sync_bool_compare_and_swap: 742 BuiltinIndex = 11; 743 NumFixed = 2; 744 ResultType = Context.BoolTy; 745 break; 746 case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break; 747 case Builtin::BI__sync_lock_release: 748 BuiltinIndex = 13; 749 NumFixed = 0; 750 ResultType = Context.VoidTy; 751 break; 752 case Builtin::BI__sync_swap: BuiltinIndex = 14; break; 753 } 754 755 // Now that we know how many fixed arguments we expect, first check that we 756 // have at least that many. 757 if (TheCall->getNumArgs() < 1+NumFixed) { 758 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 759 << 0 << 1+NumFixed << TheCall->getNumArgs() 760 << TheCall->getCallee()->getSourceRange(); 761 return ExprError(); 762 } 763 764 // Get the decl for the concrete builtin from this, we can tell what the 765 // concrete integer type we should convert to is. 766 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 767 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 768 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 769 FunctionDecl *NewBuiltinDecl = 770 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 771 TUScope, false, DRE->getLocStart())); 772 773 // The first argument --- the pointer --- has a fixed type; we 774 // deduce the types of the rest of the arguments accordingly. Walk 775 // the remaining arguments, converting them to the deduced value type. 776 for (unsigned i = 0; i != NumFixed; ++i) { 777 ExprResult Arg = TheCall->getArg(i+1); 778 779 // If the argument is an implicit cast, then there was a promotion due to 780 // "...", just remove it now. 781 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg.get())) { 782 Arg = ICE->getSubExpr(); 783 ICE->setSubExpr(0); 784 TheCall->setArg(i+1, Arg.get()); 785 } 786 787 // GCC does an implicit conversion to the pointer or integer ValType. This 788 // can fail in some cases (1i -> int**), check for this error case now. 789 // Initialize the argument. 790 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context, 791 ValType, /*consume*/ false); 792 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 793 if (Arg.isInvalid()) 794 return ExprError(); 795 796 // Okay, we have something that *can* be converted to the right type. Check 797 // to see if there is a potentially weird extension going on here. This can 798 // happen when you do an atomic operation on something like an char* and 799 // pass in 42. The 42 gets converted to char. This is even more strange 800 // for things like 45.123 -> char, etc. 801 // FIXME: Do this check. 802 TheCall->setArg(i+1, Arg.take()); 803 } 804 805 ASTContext& Context = this->getASTContext(); 806 807 // Create a new DeclRefExpr to refer to the new decl. 808 DeclRefExpr* NewDRE = DeclRefExpr::Create( 809 Context, 810 DRE->getQualifierLoc(), 811 NewBuiltinDecl, 812 DRE->getLocation(), 813 NewBuiltinDecl->getType(), 814 DRE->getValueKind()); 815 816 // Set the callee in the CallExpr. 817 // FIXME: This leaks the original parens and implicit casts. 818 ExprResult PromotedCall = UsualUnaryConversions(NewDRE); 819 if (PromotedCall.isInvalid()) 820 return ExprError(); 821 TheCall->setCallee(PromotedCall.take()); 822 823 // Change the result type of the call to match the original value type. This 824 // is arbitrary, but the codegen for these builtins ins design to handle it 825 // gracefully. 826 TheCall->setType(ResultType); 827 828 return move(TheCallResult); 829} 830 831/// CheckObjCString - Checks that the argument to the builtin 832/// CFString constructor is correct 833/// Note: It might also make sense to do the UTF-16 conversion here (would 834/// simplify the backend). 835bool Sema::CheckObjCString(Expr *Arg) { 836 Arg = Arg->IgnoreParenCasts(); 837 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 838 839 if (!Literal || !Literal->isAscii()) { 840 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 841 << Arg->getSourceRange(); 842 return true; 843 } 844 845 if (Literal->containsNonAsciiOrNull()) { 846 StringRef String = Literal->getString(); 847 unsigned NumBytes = String.size(); 848 SmallVector<UTF16, 128> ToBuf(NumBytes); 849 const UTF8 *FromPtr = (UTF8 *)String.data(); 850 UTF16 *ToPtr = &ToBuf[0]; 851 852 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 853 &ToPtr, ToPtr + NumBytes, 854 strictConversion); 855 // Check for conversion failure. 856 if (Result != conversionOK) 857 Diag(Arg->getLocStart(), 858 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 859 } 860 return false; 861} 862 863/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 864/// Emit an error and return true on failure, return false on success. 865bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 866 Expr *Fn = TheCall->getCallee(); 867 if (TheCall->getNumArgs() > 2) { 868 Diag(TheCall->getArg(2)->getLocStart(), 869 diag::err_typecheck_call_too_many_args) 870 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 871 << Fn->getSourceRange() 872 << SourceRange(TheCall->getArg(2)->getLocStart(), 873 (*(TheCall->arg_end()-1))->getLocEnd()); 874 return true; 875 } 876 877 if (TheCall->getNumArgs() < 2) { 878 return Diag(TheCall->getLocEnd(), 879 diag::err_typecheck_call_too_few_args_at_least) 880 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 881 } 882 883 // Type-check the first argument normally. 884 if (checkBuiltinArgument(*this, TheCall, 0)) 885 return true; 886 887 // Determine whether the current function is variadic or not. 888 BlockScopeInfo *CurBlock = getCurBlock(); 889 bool isVariadic; 890 if (CurBlock) 891 isVariadic = CurBlock->TheDecl->isVariadic(); 892 else if (FunctionDecl *FD = getCurFunctionDecl()) 893 isVariadic = FD->isVariadic(); 894 else 895 isVariadic = getCurMethodDecl()->isVariadic(); 896 897 if (!isVariadic) { 898 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 899 return true; 900 } 901 902 // Verify that the second argument to the builtin is the last argument of the 903 // current function or method. 904 bool SecondArgIsLastNamedArgument = false; 905 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 906 907 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 908 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 909 // FIXME: This isn't correct for methods (results in bogus warning). 910 // Get the last formal in the current function. 911 const ParmVarDecl *LastArg; 912 if (CurBlock) 913 LastArg = *(CurBlock->TheDecl->param_end()-1); 914 else if (FunctionDecl *FD = getCurFunctionDecl()) 915 LastArg = *(FD->param_end()-1); 916 else 917 LastArg = *(getCurMethodDecl()->param_end()-1); 918 SecondArgIsLastNamedArgument = PV == LastArg; 919 } 920 } 921 922 if (!SecondArgIsLastNamedArgument) 923 Diag(TheCall->getArg(1)->getLocStart(), 924 diag::warn_second_parameter_of_va_start_not_last_named_argument); 925 return false; 926} 927 928/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 929/// friends. This is declared to take (...), so we have to check everything. 930bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 931 if (TheCall->getNumArgs() < 2) 932 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 933 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 934 if (TheCall->getNumArgs() > 2) 935 return Diag(TheCall->getArg(2)->getLocStart(), 936 diag::err_typecheck_call_too_many_args) 937 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 938 << SourceRange(TheCall->getArg(2)->getLocStart(), 939 (*(TheCall->arg_end()-1))->getLocEnd()); 940 941 ExprResult OrigArg0 = TheCall->getArg(0); 942 ExprResult OrigArg1 = TheCall->getArg(1); 943 944 // Do standard promotions between the two arguments, returning their common 945 // type. 946 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 947 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 948 return true; 949 950 // Make sure any conversions are pushed back into the call; this is 951 // type safe since unordered compare builtins are declared as "_Bool 952 // foo(...)". 953 TheCall->setArg(0, OrigArg0.get()); 954 TheCall->setArg(1, OrigArg1.get()); 955 956 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 957 return false; 958 959 // If the common type isn't a real floating type, then the arguments were 960 // invalid for this operation. 961 if (!Res->isRealFloatingType()) 962 return Diag(OrigArg0.get()->getLocStart(), 963 diag::err_typecheck_call_invalid_ordered_compare) 964 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 965 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 966 967 return false; 968} 969 970/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 971/// __builtin_isnan and friends. This is declared to take (...), so we have 972/// to check everything. We expect the last argument to be a floating point 973/// value. 974bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 975 if (TheCall->getNumArgs() < NumArgs) 976 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 977 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 978 if (TheCall->getNumArgs() > NumArgs) 979 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 980 diag::err_typecheck_call_too_many_args) 981 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 982 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 983 (*(TheCall->arg_end()-1))->getLocEnd()); 984 985 Expr *OrigArg = TheCall->getArg(NumArgs-1); 986 987 if (OrigArg->isTypeDependent()) 988 return false; 989 990 // This operation requires a non-_Complex floating-point number. 991 if (!OrigArg->getType()->isRealFloatingType()) 992 return Diag(OrigArg->getLocStart(), 993 diag::err_typecheck_call_invalid_unary_fp) 994 << OrigArg->getType() << OrigArg->getSourceRange(); 995 996 // If this is an implicit conversion from float -> double, remove it. 997 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 998 Expr *CastArg = Cast->getSubExpr(); 999 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 1000 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 1001 "promotion from float to double is the only expected cast here"); 1002 Cast->setSubExpr(0); 1003 TheCall->setArg(NumArgs-1, CastArg); 1004 OrigArg = CastArg; 1005 } 1006 } 1007 1008 return false; 1009} 1010 1011/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1012// This is declared to take (...), so we have to check everything. 1013ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1014 if (TheCall->getNumArgs() < 2) 1015 return ExprError(Diag(TheCall->getLocEnd(), 1016 diag::err_typecheck_call_too_few_args_at_least) 1017 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1018 << TheCall->getSourceRange()); 1019 1020 // Determine which of the following types of shufflevector we're checking: 1021 // 1) unary, vector mask: (lhs, mask) 1022 // 2) binary, vector mask: (lhs, rhs, mask) 1023 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1024 QualType resType = TheCall->getArg(0)->getType(); 1025 unsigned numElements = 0; 1026 1027 if (!TheCall->getArg(0)->isTypeDependent() && 1028 !TheCall->getArg(1)->isTypeDependent()) { 1029 QualType LHSType = TheCall->getArg(0)->getType(); 1030 QualType RHSType = TheCall->getArg(1)->getType(); 1031 1032 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 1033 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 1034 << SourceRange(TheCall->getArg(0)->getLocStart(), 1035 TheCall->getArg(1)->getLocEnd()); 1036 return ExprError(); 1037 } 1038 1039 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1040 unsigned numResElements = TheCall->getNumArgs() - 2; 1041 1042 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1043 // with mask. If so, verify that RHS is an integer vector type with the 1044 // same number of elts as lhs. 1045 if (TheCall->getNumArgs() == 2) { 1046 if (!RHSType->hasIntegerRepresentation() || 1047 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1048 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1049 << SourceRange(TheCall->getArg(1)->getLocStart(), 1050 TheCall->getArg(1)->getLocEnd()); 1051 numResElements = numElements; 1052 } 1053 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1054 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1055 << SourceRange(TheCall->getArg(0)->getLocStart(), 1056 TheCall->getArg(1)->getLocEnd()); 1057 return ExprError(); 1058 } else if (numElements != numResElements) { 1059 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1060 resType = Context.getVectorType(eltType, numResElements, 1061 VectorType::GenericVector); 1062 } 1063 } 1064 1065 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1066 if (TheCall->getArg(i)->isTypeDependent() || 1067 TheCall->getArg(i)->isValueDependent()) 1068 continue; 1069 1070 llvm::APSInt Result(32); 1071 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1072 return ExprError(Diag(TheCall->getLocStart(), 1073 diag::err_shufflevector_nonconstant_argument) 1074 << TheCall->getArg(i)->getSourceRange()); 1075 1076 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1077 return ExprError(Diag(TheCall->getLocStart(), 1078 diag::err_shufflevector_argument_too_large) 1079 << TheCall->getArg(i)->getSourceRange()); 1080 } 1081 1082 SmallVector<Expr*, 32> exprs; 1083 1084 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1085 exprs.push_back(TheCall->getArg(i)); 1086 TheCall->setArg(i, 0); 1087 } 1088 1089 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 1090 exprs.size(), resType, 1091 TheCall->getCallee()->getLocStart(), 1092 TheCall->getRParenLoc())); 1093} 1094 1095/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1096// This is declared to take (const void*, ...) and can take two 1097// optional constant int args. 1098bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1099 unsigned NumArgs = TheCall->getNumArgs(); 1100 1101 if (NumArgs > 3) 1102 return Diag(TheCall->getLocEnd(), 1103 diag::err_typecheck_call_too_many_args_at_most) 1104 << 0 /*function call*/ << 3 << NumArgs 1105 << TheCall->getSourceRange(); 1106 1107 // Argument 0 is checked for us and the remaining arguments must be 1108 // constant integers. 1109 for (unsigned i = 1; i != NumArgs; ++i) { 1110 Expr *Arg = TheCall->getArg(i); 1111 1112 llvm::APSInt Result; 1113 if (SemaBuiltinConstantArg(TheCall, i, Result)) 1114 return true; 1115 1116 // FIXME: gcc issues a warning and rewrites these to 0. These 1117 // seems especially odd for the third argument since the default 1118 // is 3. 1119 if (i == 1) { 1120 if (Result.getLimitedValue() > 1) 1121 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1122 << "0" << "1" << Arg->getSourceRange(); 1123 } else { 1124 if (Result.getLimitedValue() > 3) 1125 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1126 << "0" << "3" << Arg->getSourceRange(); 1127 } 1128 } 1129 1130 return false; 1131} 1132 1133/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1134/// TheCall is a constant expression. 1135bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1136 llvm::APSInt &Result) { 1137 Expr *Arg = TheCall->getArg(ArgNum); 1138 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1139 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1140 1141 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 1142 1143 if (!Arg->isIntegerConstantExpr(Result, Context)) 1144 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 1145 << FDecl->getDeclName() << Arg->getSourceRange(); 1146 1147 return false; 1148} 1149 1150/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 1151/// int type). This simply type checks that type is one of the defined 1152/// constants (0-3). 1153// For compatibility check 0-3, llvm only handles 0 and 2. 1154bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 1155 llvm::APSInt Result; 1156 1157 // Check constant-ness first. 1158 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1159 return true; 1160 1161 Expr *Arg = TheCall->getArg(1); 1162 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 1163 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1164 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1165 } 1166 1167 return false; 1168} 1169 1170/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1171/// This checks that val is a constant 1. 1172bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1173 Expr *Arg = TheCall->getArg(1); 1174 llvm::APSInt Result; 1175 1176 // TODO: This is less than ideal. Overload this to take a value. 1177 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1178 return true; 1179 1180 if (Result != 1) 1181 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1182 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1183 1184 return false; 1185} 1186 1187// Handle i > 1 ? "x" : "y", recursively. 1188bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 1189 bool HasVAListArg, 1190 unsigned format_idx, unsigned firstDataArg, 1191 bool isPrintf) { 1192 tryAgain: 1193 if (E->isTypeDependent() || E->isValueDependent()) 1194 return false; 1195 1196 E = E->IgnoreParens(); 1197 1198 switch (E->getStmtClass()) { 1199 case Stmt::BinaryConditionalOperatorClass: 1200 case Stmt::ConditionalOperatorClass: { 1201 const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E); 1202 return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, HasVAListArg, 1203 format_idx, firstDataArg, isPrintf) 1204 && SemaCheckStringLiteral(C->getFalseExpr(), TheCall, HasVAListArg, 1205 format_idx, firstDataArg, isPrintf); 1206 } 1207 1208 case Stmt::IntegerLiteralClass: 1209 // Technically -Wformat-nonliteral does not warn about this case. 1210 // The behavior of printf and friends in this case is implementation 1211 // dependent. Ideally if the format string cannot be null then 1212 // it should have a 'nonnull' attribute in the function prototype. 1213 return true; 1214 1215 case Stmt::ImplicitCastExprClass: { 1216 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 1217 goto tryAgain; 1218 } 1219 1220 case Stmt::OpaqueValueExprClass: 1221 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 1222 E = src; 1223 goto tryAgain; 1224 } 1225 return false; 1226 1227 case Stmt::PredefinedExprClass: 1228 // While __func__, etc., are technically not string literals, they 1229 // cannot contain format specifiers and thus are not a security 1230 // liability. 1231 return true; 1232 1233 case Stmt::DeclRefExprClass: { 1234 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1235 1236 // As an exception, do not flag errors for variables binding to 1237 // const string literals. 1238 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1239 bool isConstant = false; 1240 QualType T = DR->getType(); 1241 1242 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1243 isConstant = AT->getElementType().isConstant(Context); 1244 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1245 isConstant = T.isConstant(Context) && 1246 PT->getPointeeType().isConstant(Context); 1247 } 1248 1249 if (isConstant) { 1250 if (const Expr *Init = VD->getAnyInitializer()) 1251 return SemaCheckStringLiteral(Init, TheCall, 1252 HasVAListArg, format_idx, firstDataArg, 1253 isPrintf); 1254 } 1255 1256 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1257 // special check to see if the format string is a function parameter 1258 // of the function calling the printf function. If the function 1259 // has an attribute indicating it is a printf-like function, then we 1260 // should suppress warnings concerning non-literals being used in a call 1261 // to a vprintf function. For example: 1262 // 1263 // void 1264 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1265 // va_list ap; 1266 // va_start(ap, fmt); 1267 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1268 // ... 1269 // 1270 // 1271 // FIXME: We don't have full attribute support yet, so just check to see 1272 // if the argument is a DeclRefExpr that references a parameter. We'll 1273 // add proper support for checking the attribute later. 1274 if (HasVAListArg) 1275 if (isa<ParmVarDecl>(VD)) 1276 return true; 1277 } 1278 1279 return false; 1280 } 1281 1282 case Stmt::CallExprClass: { 1283 const CallExpr *CE = cast<CallExpr>(E); 1284 if (const ImplicitCastExpr *ICE 1285 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 1286 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 1287 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 1288 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 1289 unsigned ArgIndex = FA->getFormatIdx(); 1290 const Expr *Arg = CE->getArg(ArgIndex - 1); 1291 1292 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 1293 format_idx, firstDataArg, isPrintf); 1294 } 1295 } 1296 } 1297 } 1298 1299 return false; 1300 } 1301 case Stmt::ObjCStringLiteralClass: 1302 case Stmt::StringLiteralClass: { 1303 const StringLiteral *StrE = NULL; 1304 1305 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1306 StrE = ObjCFExpr->getString(); 1307 else 1308 StrE = cast<StringLiteral>(E); 1309 1310 if (StrE) { 1311 CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx, 1312 firstDataArg, isPrintf); 1313 return true; 1314 } 1315 1316 return false; 1317 } 1318 1319 default: 1320 return false; 1321 } 1322} 1323 1324void 1325Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1326 const Expr * const *ExprArgs, 1327 SourceLocation CallSiteLoc) { 1328 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1329 e = NonNull->args_end(); 1330 i != e; ++i) { 1331 const Expr *ArgExpr = ExprArgs[*i]; 1332 if (ArgExpr->isNullPointerConstant(Context, 1333 Expr::NPC_ValueDependentIsNotNull)) 1334 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 1335 } 1336} 1337 1338/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar 1339/// functions) for correct use of format strings. 1340void 1341Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg, 1342 unsigned format_idx, unsigned firstDataArg, 1343 bool isPrintf) { 1344 1345 const Expr *Fn = TheCall->getCallee(); 1346 1347 // The way the format attribute works in GCC, the implicit this argument 1348 // of member functions is counted. However, it doesn't appear in our own 1349 // lists, so decrement format_idx in that case. 1350 if (isa<CXXMemberCallExpr>(TheCall)) { 1351 const CXXMethodDecl *method_decl = 1352 dyn_cast<CXXMethodDecl>(TheCall->getCalleeDecl()); 1353 if (method_decl && method_decl->isInstance()) { 1354 // Catch a format attribute mistakenly referring to the object argument. 1355 if (format_idx == 0) 1356 return; 1357 --format_idx; 1358 if(firstDataArg != 0) 1359 --firstDataArg; 1360 } 1361 } 1362 1363 // CHECK: printf/scanf-like function is called with no format string. 1364 if (format_idx >= TheCall->getNumArgs()) { 1365 Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string) 1366 << Fn->getSourceRange(); 1367 return; 1368 } 1369 1370 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 1371 1372 // CHECK: format string is not a string literal. 1373 // 1374 // Dynamically generated format strings are difficult to 1375 // automatically vet at compile time. Requiring that format strings 1376 // are string literals: (1) permits the checking of format strings by 1377 // the compiler and thereby (2) can practically remove the source of 1378 // many format string exploits. 1379 1380 // Format string can be either ObjC string (e.g. @"%d") or 1381 // C string (e.g. "%d") 1382 // ObjC string uses the same format specifiers as C string, so we can use 1383 // the same format string checking logic for both ObjC and C strings. 1384 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 1385 firstDataArg, isPrintf)) 1386 return; // Literal format string found, check done! 1387 1388 // If there are no arguments specified, warn with -Wformat-security, otherwise 1389 // warn only with -Wformat-nonliteral. 1390 if (TheCall->getNumArgs() == format_idx+1) 1391 Diag(TheCall->getArg(format_idx)->getLocStart(), 1392 diag::warn_format_nonliteral_noargs) 1393 << OrigFormatExpr->getSourceRange(); 1394 else 1395 Diag(TheCall->getArg(format_idx)->getLocStart(), 1396 diag::warn_format_nonliteral) 1397 << OrigFormatExpr->getSourceRange(); 1398} 1399 1400namespace { 1401class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1402protected: 1403 Sema &S; 1404 const StringLiteral *FExpr; 1405 const Expr *OrigFormatExpr; 1406 const unsigned FirstDataArg; 1407 const unsigned NumDataArgs; 1408 const bool IsObjCLiteral; 1409 const char *Beg; // Start of format string. 1410 const bool HasVAListArg; 1411 const CallExpr *TheCall; 1412 unsigned FormatIdx; 1413 llvm::BitVector CoveredArgs; 1414 bool usesPositionalArgs; 1415 bool atFirstArg; 1416public: 1417 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1418 const Expr *origFormatExpr, unsigned firstDataArg, 1419 unsigned numDataArgs, bool isObjCLiteral, 1420 const char *beg, bool hasVAListArg, 1421 const CallExpr *theCall, unsigned formatIdx) 1422 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1423 FirstDataArg(firstDataArg), 1424 NumDataArgs(numDataArgs), 1425 IsObjCLiteral(isObjCLiteral), Beg(beg), 1426 HasVAListArg(hasVAListArg), 1427 TheCall(theCall), FormatIdx(formatIdx), 1428 usesPositionalArgs(false), atFirstArg(true) { 1429 CoveredArgs.resize(numDataArgs); 1430 CoveredArgs.reset(); 1431 } 1432 1433 void DoneProcessing(); 1434 1435 void HandleIncompleteSpecifier(const char *startSpecifier, 1436 unsigned specifierLen); 1437 1438 virtual void HandleInvalidPosition(const char *startSpecifier, 1439 unsigned specifierLen, 1440 analyze_format_string::PositionContext p); 1441 1442 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1443 1444 void HandleNullChar(const char *nullCharacter); 1445 1446protected: 1447 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1448 const char *startSpec, 1449 unsigned specifierLen, 1450 const char *csStart, unsigned csLen); 1451 1452 SourceRange getFormatStringRange(); 1453 CharSourceRange getSpecifierRange(const char *startSpecifier, 1454 unsigned specifierLen); 1455 SourceLocation getLocationOfByte(const char *x); 1456 1457 const Expr *getDataArg(unsigned i) const; 1458 1459 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1460 const analyze_format_string::ConversionSpecifier &CS, 1461 const char *startSpecifier, unsigned specifierLen, 1462 unsigned argIndex); 1463}; 1464} 1465 1466SourceRange CheckFormatHandler::getFormatStringRange() { 1467 return OrigFormatExpr->getSourceRange(); 1468} 1469 1470CharSourceRange CheckFormatHandler:: 1471getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1472 SourceLocation Start = getLocationOfByte(startSpecifier); 1473 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1474 1475 // Advance the end SourceLocation by one due to half-open ranges. 1476 End = End.getLocWithOffset(1); 1477 1478 return CharSourceRange::getCharRange(Start, End); 1479} 1480 1481SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 1482 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1483} 1484 1485void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 1486 unsigned specifierLen){ 1487 SourceLocation Loc = getLocationOfByte(startSpecifier); 1488 S.Diag(Loc, diag::warn_printf_incomplete_specifier) 1489 << getSpecifierRange(startSpecifier, specifierLen); 1490} 1491 1492void 1493CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1494 analyze_format_string::PositionContext p) { 1495 SourceLocation Loc = getLocationOfByte(startPos); 1496 S.Diag(Loc, diag::warn_format_invalid_positional_specifier) 1497 << (unsigned) p << getSpecifierRange(startPos, posLen); 1498} 1499 1500void CheckFormatHandler::HandleZeroPosition(const char *startPos, 1501 unsigned posLen) { 1502 SourceLocation Loc = getLocationOfByte(startPos); 1503 S.Diag(Loc, diag::warn_format_zero_positional_specifier) 1504 << getSpecifierRange(startPos, posLen); 1505} 1506 1507void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 1508 if (!IsObjCLiteral) { 1509 // The presence of a null character is likely an error. 1510 S.Diag(getLocationOfByte(nullCharacter), 1511 diag::warn_printf_format_string_contains_null_char) 1512 << getFormatStringRange(); 1513 } 1514} 1515 1516const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 1517 return TheCall->getArg(FirstDataArg + i); 1518} 1519 1520void CheckFormatHandler::DoneProcessing() { 1521 // Does the number of data arguments exceed the number of 1522 // format conversions in the format string? 1523 if (!HasVAListArg) { 1524 // Find any arguments that weren't covered. 1525 CoveredArgs.flip(); 1526 signed notCoveredArg = CoveredArgs.find_first(); 1527 if (notCoveredArg >= 0) { 1528 assert((unsigned)notCoveredArg < NumDataArgs); 1529 S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(), 1530 diag::warn_printf_data_arg_not_used) 1531 << getFormatStringRange(); 1532 } 1533 } 1534} 1535 1536bool 1537CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 1538 SourceLocation Loc, 1539 const char *startSpec, 1540 unsigned specifierLen, 1541 const char *csStart, 1542 unsigned csLen) { 1543 1544 bool keepGoing = true; 1545 if (argIndex < NumDataArgs) { 1546 // Consider the argument coverered, even though the specifier doesn't 1547 // make sense. 1548 CoveredArgs.set(argIndex); 1549 } 1550 else { 1551 // If argIndex exceeds the number of data arguments we 1552 // don't issue a warning because that is just a cascade of warnings (and 1553 // they may have intended '%%' anyway). We don't want to continue processing 1554 // the format string after this point, however, as we will like just get 1555 // gibberish when trying to match arguments. 1556 keepGoing = false; 1557 } 1558 1559 S.Diag(Loc, diag::warn_format_invalid_conversion) 1560 << StringRef(csStart, csLen) 1561 << getSpecifierRange(startSpec, specifierLen); 1562 1563 return keepGoing; 1564} 1565 1566bool 1567CheckFormatHandler::CheckNumArgs( 1568 const analyze_format_string::FormatSpecifier &FS, 1569 const analyze_format_string::ConversionSpecifier &CS, 1570 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 1571 1572 if (argIndex >= NumDataArgs) { 1573 if (FS.usesPositionalArg()) { 1574 S.Diag(getLocationOfByte(CS.getStart()), 1575 diag::warn_printf_positional_arg_exceeds_data_args) 1576 << (argIndex+1) << NumDataArgs 1577 << getSpecifierRange(startSpecifier, specifierLen); 1578 } 1579 else { 1580 S.Diag(getLocationOfByte(CS.getStart()), 1581 diag::warn_printf_insufficient_data_args) 1582 << getSpecifierRange(startSpecifier, specifierLen); 1583 } 1584 1585 return false; 1586 } 1587 return true; 1588} 1589 1590//===--- CHECK: Printf format string checking ------------------------------===// 1591 1592namespace { 1593class CheckPrintfHandler : public CheckFormatHandler { 1594public: 1595 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1596 const Expr *origFormatExpr, unsigned firstDataArg, 1597 unsigned numDataArgs, bool isObjCLiteral, 1598 const char *beg, bool hasVAListArg, 1599 const CallExpr *theCall, unsigned formatIdx) 1600 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1601 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1602 theCall, formatIdx) {} 1603 1604 1605 bool HandleInvalidPrintfConversionSpecifier( 1606 const analyze_printf::PrintfSpecifier &FS, 1607 const char *startSpecifier, 1608 unsigned specifierLen); 1609 1610 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 1611 const char *startSpecifier, 1612 unsigned specifierLen); 1613 1614 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 1615 const char *startSpecifier, unsigned specifierLen); 1616 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 1617 const analyze_printf::OptionalAmount &Amt, 1618 unsigned type, 1619 const char *startSpecifier, unsigned specifierLen); 1620 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1621 const analyze_printf::OptionalFlag &flag, 1622 const char *startSpecifier, unsigned specifierLen); 1623 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 1624 const analyze_printf::OptionalFlag &ignoredFlag, 1625 const analyze_printf::OptionalFlag &flag, 1626 const char *startSpecifier, unsigned specifierLen); 1627}; 1628} 1629 1630bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 1631 const analyze_printf::PrintfSpecifier &FS, 1632 const char *startSpecifier, 1633 unsigned specifierLen) { 1634 const analyze_printf::PrintfConversionSpecifier &CS = 1635 FS.getConversionSpecifier(); 1636 1637 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1638 getLocationOfByte(CS.getStart()), 1639 startSpecifier, specifierLen, 1640 CS.getStart(), CS.getLength()); 1641} 1642 1643bool CheckPrintfHandler::HandleAmount( 1644 const analyze_format_string::OptionalAmount &Amt, 1645 unsigned k, const char *startSpecifier, 1646 unsigned specifierLen) { 1647 1648 if (Amt.hasDataArgument()) { 1649 if (!HasVAListArg) { 1650 unsigned argIndex = Amt.getArgIndex(); 1651 if (argIndex >= NumDataArgs) { 1652 S.Diag(getLocationOfByte(Amt.getStart()), 1653 diag::warn_printf_asterisk_missing_arg) 1654 << k << getSpecifierRange(startSpecifier, specifierLen); 1655 // Don't do any more checking. We will just emit 1656 // spurious errors. 1657 return false; 1658 } 1659 1660 // Type check the data argument. It should be an 'int'. 1661 // Although not in conformance with C99, we also allow the argument to be 1662 // an 'unsigned int' as that is a reasonably safe case. GCC also 1663 // doesn't emit a warning for that case. 1664 CoveredArgs.set(argIndex); 1665 const Expr *Arg = getDataArg(argIndex); 1666 QualType T = Arg->getType(); 1667 1668 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1669 assert(ATR.isValid()); 1670 1671 if (!ATR.matchesType(S.Context, T)) { 1672 S.Diag(getLocationOfByte(Amt.getStart()), 1673 diag::warn_printf_asterisk_wrong_type) 1674 << k 1675 << ATR.getRepresentativeType(S.Context) << T 1676 << getSpecifierRange(startSpecifier, specifierLen) 1677 << Arg->getSourceRange(); 1678 // Don't do any more checking. We will just emit 1679 // spurious errors. 1680 return false; 1681 } 1682 } 1683 } 1684 return true; 1685} 1686 1687void CheckPrintfHandler::HandleInvalidAmount( 1688 const analyze_printf::PrintfSpecifier &FS, 1689 const analyze_printf::OptionalAmount &Amt, 1690 unsigned type, 1691 const char *startSpecifier, 1692 unsigned specifierLen) { 1693 const analyze_printf::PrintfConversionSpecifier &CS = 1694 FS.getConversionSpecifier(); 1695 switch (Amt.getHowSpecified()) { 1696 case analyze_printf::OptionalAmount::Constant: 1697 S.Diag(getLocationOfByte(Amt.getStart()), 1698 diag::warn_printf_nonsensical_optional_amount) 1699 << type 1700 << CS.toString() 1701 << getSpecifierRange(startSpecifier, specifierLen) 1702 << FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 1703 Amt.getConstantLength())); 1704 break; 1705 1706 default: 1707 S.Diag(getLocationOfByte(Amt.getStart()), 1708 diag::warn_printf_nonsensical_optional_amount) 1709 << type 1710 << CS.toString() 1711 << getSpecifierRange(startSpecifier, specifierLen); 1712 break; 1713 } 1714} 1715 1716void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1717 const analyze_printf::OptionalFlag &flag, 1718 const char *startSpecifier, 1719 unsigned specifierLen) { 1720 // Warn about pointless flag with a fixit removal. 1721 const analyze_printf::PrintfConversionSpecifier &CS = 1722 FS.getConversionSpecifier(); 1723 S.Diag(getLocationOfByte(flag.getPosition()), 1724 diag::warn_printf_nonsensical_flag) 1725 << flag.toString() << CS.toString() 1726 << getSpecifierRange(startSpecifier, specifierLen) 1727 << FixItHint::CreateRemoval(getSpecifierRange(flag.getPosition(), 1)); 1728} 1729 1730void CheckPrintfHandler::HandleIgnoredFlag( 1731 const analyze_printf::PrintfSpecifier &FS, 1732 const analyze_printf::OptionalFlag &ignoredFlag, 1733 const analyze_printf::OptionalFlag &flag, 1734 const char *startSpecifier, 1735 unsigned specifierLen) { 1736 // Warn about ignored flag with a fixit removal. 1737 S.Diag(getLocationOfByte(ignoredFlag.getPosition()), 1738 diag::warn_printf_ignored_flag) 1739 << ignoredFlag.toString() << flag.toString() 1740 << getSpecifierRange(startSpecifier, specifierLen) 1741 << FixItHint::CreateRemoval(getSpecifierRange( 1742 ignoredFlag.getPosition(), 1)); 1743} 1744 1745bool 1746CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 1747 &FS, 1748 const char *startSpecifier, 1749 unsigned specifierLen) { 1750 1751 using namespace analyze_format_string; 1752 using namespace analyze_printf; 1753 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 1754 1755 if (FS.consumesDataArgument()) { 1756 if (atFirstArg) { 1757 atFirstArg = false; 1758 usesPositionalArgs = FS.usesPositionalArg(); 1759 } 1760 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1761 // Cannot mix-and-match positional and non-positional arguments. 1762 S.Diag(getLocationOfByte(CS.getStart()), 1763 diag::warn_format_mix_positional_nonpositional_args) 1764 << getSpecifierRange(startSpecifier, specifierLen); 1765 return false; 1766 } 1767 } 1768 1769 // First check if the field width, precision, and conversion specifier 1770 // have matching data arguments. 1771 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 1772 startSpecifier, specifierLen)) { 1773 return false; 1774 } 1775 1776 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 1777 startSpecifier, specifierLen)) { 1778 return false; 1779 } 1780 1781 if (!CS.consumesDataArgument()) { 1782 // FIXME: Technically specifying a precision or field width here 1783 // makes no sense. Worth issuing a warning at some point. 1784 return true; 1785 } 1786 1787 // Consume the argument. 1788 unsigned argIndex = FS.getArgIndex(); 1789 if (argIndex < NumDataArgs) { 1790 // The check to see if the argIndex is valid will come later. 1791 // We set the bit here because we may exit early from this 1792 // function if we encounter some other error. 1793 CoveredArgs.set(argIndex); 1794 } 1795 1796 // Check for using an Objective-C specific conversion specifier 1797 // in a non-ObjC literal. 1798 if (!IsObjCLiteral && CS.isObjCArg()) { 1799 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 1800 specifierLen); 1801 } 1802 1803 // Check for invalid use of field width 1804 if (!FS.hasValidFieldWidth()) { 1805 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 1806 startSpecifier, specifierLen); 1807 } 1808 1809 // Check for invalid use of precision 1810 if (!FS.hasValidPrecision()) { 1811 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 1812 startSpecifier, specifierLen); 1813 } 1814 1815 // Check each flag does not conflict with any other component. 1816 if (!FS.hasValidThousandsGroupingPrefix()) 1817 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 1818 if (!FS.hasValidLeadingZeros()) 1819 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 1820 if (!FS.hasValidPlusPrefix()) 1821 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 1822 if (!FS.hasValidSpacePrefix()) 1823 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 1824 if (!FS.hasValidAlternativeForm()) 1825 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 1826 if (!FS.hasValidLeftJustified()) 1827 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 1828 1829 // Check that flags are not ignored by another flag 1830 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 1831 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 1832 startSpecifier, specifierLen); 1833 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 1834 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 1835 startSpecifier, specifierLen); 1836 1837 // Check the length modifier is valid with the given conversion specifier. 1838 const LengthModifier &LM = FS.getLengthModifier(); 1839 if (!FS.hasValidLengthModifier()) 1840 S.Diag(getLocationOfByte(LM.getStart()), 1841 diag::warn_format_nonsensical_length) 1842 << LM.toString() << CS.toString() 1843 << getSpecifierRange(startSpecifier, specifierLen) 1844 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 1845 LM.getLength())); 1846 1847 // Are we using '%n'? 1848 if (CS.getKind() == ConversionSpecifier::nArg) { 1849 // Issue a warning about this being a possible security issue. 1850 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back) 1851 << getSpecifierRange(startSpecifier, specifierLen); 1852 // Continue checking the other format specifiers. 1853 return true; 1854 } 1855 1856 // The remaining checks depend on the data arguments. 1857 if (HasVAListArg) 1858 return true; 1859 1860 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 1861 return false; 1862 1863 // Now type check the data expression that matches the 1864 // format specifier. 1865 const Expr *Ex = getDataArg(argIndex); 1866 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context); 1867 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 1868 // Check if we didn't match because of an implicit cast from a 'char' 1869 // or 'short' to an 'int'. This is done because printf is a varargs 1870 // function. 1871 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 1872 if (ICE->getType() == S.Context.IntTy) { 1873 // All further checking is done on the subexpression. 1874 Ex = ICE->getSubExpr(); 1875 if (ATR.matchesType(S.Context, Ex->getType())) 1876 return true; 1877 } 1878 1879 // We may be able to offer a FixItHint if it is a supported type. 1880 PrintfSpecifier fixedFS = FS; 1881 bool success = fixedFS.fixType(Ex->getType()); 1882 1883 if (success) { 1884 // Get the fix string from the fixed format specifier 1885 llvm::SmallString<128> buf; 1886 llvm::raw_svector_ostream os(buf); 1887 fixedFS.toString(os); 1888 1889 // FIXME: getRepresentativeType() perhaps should return a string 1890 // instead of a QualType to better handle when the representative 1891 // type is 'wint_t' (which is defined in the system headers). 1892 S.Diag(getLocationOfByte(CS.getStart()), 1893 diag::warn_printf_conversion_argument_type_mismatch) 1894 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1895 << getSpecifierRange(startSpecifier, specifierLen) 1896 << Ex->getSourceRange() 1897 << FixItHint::CreateReplacement( 1898 getSpecifierRange(startSpecifier, specifierLen), 1899 os.str()); 1900 } 1901 else { 1902 S.Diag(getLocationOfByte(CS.getStart()), 1903 diag::warn_printf_conversion_argument_type_mismatch) 1904 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1905 << getSpecifierRange(startSpecifier, specifierLen) 1906 << Ex->getSourceRange(); 1907 } 1908 } 1909 1910 return true; 1911} 1912 1913//===--- CHECK: Scanf format string checking ------------------------------===// 1914 1915namespace { 1916class CheckScanfHandler : public CheckFormatHandler { 1917public: 1918 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 1919 const Expr *origFormatExpr, unsigned firstDataArg, 1920 unsigned numDataArgs, bool isObjCLiteral, 1921 const char *beg, bool hasVAListArg, 1922 const CallExpr *theCall, unsigned formatIdx) 1923 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1924 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1925 theCall, formatIdx) {} 1926 1927 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 1928 const char *startSpecifier, 1929 unsigned specifierLen); 1930 1931 bool HandleInvalidScanfConversionSpecifier( 1932 const analyze_scanf::ScanfSpecifier &FS, 1933 const char *startSpecifier, 1934 unsigned specifierLen); 1935 1936 void HandleIncompleteScanList(const char *start, const char *end); 1937}; 1938} 1939 1940void CheckScanfHandler::HandleIncompleteScanList(const char *start, 1941 const char *end) { 1942 S.Diag(getLocationOfByte(end), diag::warn_scanf_scanlist_incomplete) 1943 << getSpecifierRange(start, end - start); 1944} 1945 1946bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 1947 const analyze_scanf::ScanfSpecifier &FS, 1948 const char *startSpecifier, 1949 unsigned specifierLen) { 1950 1951 const analyze_scanf::ScanfConversionSpecifier &CS = 1952 FS.getConversionSpecifier(); 1953 1954 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1955 getLocationOfByte(CS.getStart()), 1956 startSpecifier, specifierLen, 1957 CS.getStart(), CS.getLength()); 1958} 1959 1960bool CheckScanfHandler::HandleScanfSpecifier( 1961 const analyze_scanf::ScanfSpecifier &FS, 1962 const char *startSpecifier, 1963 unsigned specifierLen) { 1964 1965 using namespace analyze_scanf; 1966 using namespace analyze_format_string; 1967 1968 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 1969 1970 // Handle case where '%' and '*' don't consume an argument. These shouldn't 1971 // be used to decide if we are using positional arguments consistently. 1972 if (FS.consumesDataArgument()) { 1973 if (atFirstArg) { 1974 atFirstArg = false; 1975 usesPositionalArgs = FS.usesPositionalArg(); 1976 } 1977 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1978 // Cannot mix-and-match positional and non-positional arguments. 1979 S.Diag(getLocationOfByte(CS.getStart()), 1980 diag::warn_format_mix_positional_nonpositional_args) 1981 << getSpecifierRange(startSpecifier, specifierLen); 1982 return false; 1983 } 1984 } 1985 1986 // Check if the field with is non-zero. 1987 const OptionalAmount &Amt = FS.getFieldWidth(); 1988 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 1989 if (Amt.getConstantAmount() == 0) { 1990 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 1991 Amt.getConstantLength()); 1992 S.Diag(getLocationOfByte(Amt.getStart()), 1993 diag::warn_scanf_nonzero_width) 1994 << R << FixItHint::CreateRemoval(R); 1995 } 1996 } 1997 1998 if (!FS.consumesDataArgument()) { 1999 // FIXME: Technically specifying a precision or field width here 2000 // makes no sense. Worth issuing a warning at some point. 2001 return true; 2002 } 2003 2004 // Consume the argument. 2005 unsigned argIndex = FS.getArgIndex(); 2006 if (argIndex < NumDataArgs) { 2007 // The check to see if the argIndex is valid will come later. 2008 // We set the bit here because we may exit early from this 2009 // function if we encounter some other error. 2010 CoveredArgs.set(argIndex); 2011 } 2012 2013 // Check the length modifier is valid with the given conversion specifier. 2014 const LengthModifier &LM = FS.getLengthModifier(); 2015 if (!FS.hasValidLengthModifier()) { 2016 S.Diag(getLocationOfByte(LM.getStart()), 2017 diag::warn_format_nonsensical_length) 2018 << LM.toString() << CS.toString() 2019 << getSpecifierRange(startSpecifier, specifierLen) 2020 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 2021 LM.getLength())); 2022 } 2023 2024 // The remaining checks depend on the data arguments. 2025 if (HasVAListArg) 2026 return true; 2027 2028 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2029 return false; 2030 2031 // FIXME: Check that the argument type matches the format specifier. 2032 2033 return true; 2034} 2035 2036void Sema::CheckFormatString(const StringLiteral *FExpr, 2037 const Expr *OrigFormatExpr, 2038 const CallExpr *TheCall, bool HasVAListArg, 2039 unsigned format_idx, unsigned firstDataArg, 2040 bool isPrintf) { 2041 2042 // CHECK: is the format string a wide literal? 2043 if (!FExpr->isAscii()) { 2044 Diag(FExpr->getLocStart(), 2045 diag::warn_format_string_is_wide_literal) 2046 << OrigFormatExpr->getSourceRange(); 2047 return; 2048 } 2049 2050 // Str - The format string. NOTE: this is NOT null-terminated! 2051 StringRef StrRef = FExpr->getString(); 2052 const char *Str = StrRef.data(); 2053 unsigned StrLen = StrRef.size(); 2054 const unsigned numDataArgs = TheCall->getNumArgs() - firstDataArg; 2055 2056 // CHECK: empty format string? 2057 if (StrLen == 0 && numDataArgs > 0) { 2058 Diag(FExpr->getLocStart(), diag::warn_empty_format_string) 2059 << OrigFormatExpr->getSourceRange(); 2060 return; 2061 } 2062 2063 if (isPrintf) { 2064 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2065 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2066 Str, HasVAListArg, TheCall, format_idx); 2067 2068 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen)) 2069 H.DoneProcessing(); 2070 } 2071 else { 2072 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2073 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2074 Str, HasVAListArg, TheCall, format_idx); 2075 2076 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen)) 2077 H.DoneProcessing(); 2078 } 2079} 2080 2081//===--- CHECK: Standard memory functions ---------------------------------===// 2082 2083/// \brief Determine whether the given type is a dynamic class type (e.g., 2084/// whether it has a vtable). 2085static bool isDynamicClassType(QualType T) { 2086 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 2087 if (CXXRecordDecl *Definition = Record->getDefinition()) 2088 if (Definition->isDynamicClass()) 2089 return true; 2090 2091 return false; 2092} 2093 2094/// \brief If E is a sizeof expression, returns its argument expression, 2095/// otherwise returns NULL. 2096static const Expr *getSizeOfExprArg(const Expr* E) { 2097 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2098 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2099 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 2100 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 2101 2102 return 0; 2103} 2104 2105/// \brief If E is a sizeof expression, returns its argument type. 2106static QualType getSizeOfArgType(const Expr* E) { 2107 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2108 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2109 if (SizeOf->getKind() == clang::UETT_SizeOf) 2110 return SizeOf->getTypeOfArgument(); 2111 2112 return QualType(); 2113} 2114 2115/// \brief Check for dangerous or invalid arguments to memset(). 2116/// 2117/// This issues warnings on known problematic, dangerous or unspecified 2118/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 2119/// function calls. 2120/// 2121/// \param Call The call expression to diagnose. 2122void Sema::CheckMemaccessArguments(const CallExpr *Call, 2123 CheckedMemoryFunction CMF, 2124 IdentifierInfo *FnName) { 2125 // It is possible to have a non-standard definition of memset. Validate 2126 // we have enough arguments, and if not, abort further checking. 2127 if (Call->getNumArgs() < 3) 2128 return; 2129 2130 unsigned LastArg = (CMF == CMF_Memset? 1 : 2); 2131 const Expr *LenExpr = Call->getArg(2)->IgnoreParenImpCasts(); 2132 2133 // We have special checking when the length is a sizeof expression. 2134 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 2135 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 2136 llvm::FoldingSetNodeID SizeOfArgID; 2137 2138 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 2139 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 2140 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 2141 2142 QualType DestTy = Dest->getType(); 2143 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 2144 QualType PointeeTy = DestPtrTy->getPointeeType(); 2145 2146 // Never warn about void type pointers. This can be used to suppress 2147 // false positives. 2148 if (PointeeTy->isVoidType()) 2149 continue; 2150 2151 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 2152 // actually comparing the expressions for equality. Because computing the 2153 // expression IDs can be expensive, we only do this if the diagnostic is 2154 // enabled. 2155 if (SizeOfArg && 2156 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 2157 SizeOfArg->getExprLoc())) { 2158 // We only compute IDs for expressions if the warning is enabled, and 2159 // cache the sizeof arg's ID. 2160 if (SizeOfArgID == llvm::FoldingSetNodeID()) 2161 SizeOfArg->Profile(SizeOfArgID, Context, true); 2162 llvm::FoldingSetNodeID DestID; 2163 Dest->Profile(DestID, Context, true); 2164 if (DestID == SizeOfArgID) { 2165 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 2166 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 2167 if (UnaryOp->getOpcode() == UO_AddrOf) 2168 ActionIdx = 1; // If its an address-of operator, just remove it. 2169 if (Context.getTypeSize(PointeeTy) == Context.getCharWidth()) 2170 ActionIdx = 2; // If the pointee's size is sizeof(char), 2171 // suggest an explicit length. 2172 DiagRuntimeBehavior(SizeOfArg->getExprLoc(), Dest, 2173 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 2174 << FnName << ArgIdx << ActionIdx 2175 << Dest->getSourceRange() 2176 << SizeOfArg->getSourceRange()); 2177 break; 2178 } 2179 } 2180 2181 // Also check for cases where the sizeof argument is the exact same 2182 // type as the memory argument, and where it points to a user-defined 2183 // record type. 2184 if (SizeOfArgTy != QualType()) { 2185 if (PointeeTy->isRecordType() && 2186 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 2187 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 2188 PDiag(diag::warn_sizeof_pointer_type_memaccess) 2189 << FnName << SizeOfArgTy << ArgIdx 2190 << PointeeTy << Dest->getSourceRange() 2191 << LenExpr->getSourceRange()); 2192 break; 2193 } 2194 } 2195 2196 // Always complain about dynamic classes. 2197 if (isDynamicClassType(PointeeTy)) 2198 DiagRuntimeBehavior( 2199 Dest->getExprLoc(), Dest, 2200 PDiag(diag::warn_dyn_class_memaccess) 2201 << (CMF == CMF_Memcmp ? ArgIdx + 2 : ArgIdx) << FnName << PointeeTy 2202 // "overwritten" if we're warning about the destination for any call 2203 // but memcmp; otherwise a verb appropriate to the call. 2204 << (ArgIdx == 0 && CMF != CMF_Memcmp ? 0 : (unsigned)CMF) 2205 << Call->getCallee()->getSourceRange()); 2206 else if (PointeeTy.hasNonTrivialObjCLifetime() && CMF != CMF_Memset) 2207 DiagRuntimeBehavior( 2208 Dest->getExprLoc(), Dest, 2209 PDiag(diag::warn_arc_object_memaccess) 2210 << ArgIdx << FnName << PointeeTy 2211 << Call->getCallee()->getSourceRange()); 2212 else 2213 continue; 2214 2215 DiagRuntimeBehavior( 2216 Dest->getExprLoc(), Dest, 2217 PDiag(diag::note_bad_memaccess_silence) 2218 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 2219 break; 2220 } 2221 } 2222} 2223 2224// A little helper routine: ignore addition and subtraction of integer literals. 2225// This intentionally does not ignore all integer constant expressions because 2226// we don't want to remove sizeof(). 2227static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 2228 Ex = Ex->IgnoreParenCasts(); 2229 2230 for (;;) { 2231 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 2232 if (!BO || !BO->isAdditiveOp()) 2233 break; 2234 2235 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 2236 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 2237 2238 if (isa<IntegerLiteral>(RHS)) 2239 Ex = LHS; 2240 else if (isa<IntegerLiteral>(LHS)) 2241 Ex = RHS; 2242 else 2243 break; 2244 } 2245 2246 return Ex; 2247} 2248 2249// Warn if the user has made the 'size' argument to strlcpy or strlcat 2250// be the size of the source, instead of the destination. 2251void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 2252 IdentifierInfo *FnName) { 2253 2254 // Don't crash if the user has the wrong number of arguments 2255 if (Call->getNumArgs() != 3) 2256 return; 2257 2258 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 2259 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 2260 const Expr *CompareWithSrc = NULL; 2261 2262 // Look for 'strlcpy(dst, x, sizeof(x))' 2263 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 2264 CompareWithSrc = Ex; 2265 else { 2266 // Look for 'strlcpy(dst, x, strlen(x))' 2267 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 2268 if (SizeCall->isBuiltinCall(Context) == Builtin::BIstrlen 2269 && SizeCall->getNumArgs() == 1) 2270 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 2271 } 2272 } 2273 2274 if (!CompareWithSrc) 2275 return; 2276 2277 // Determine if the argument to sizeof/strlen is equal to the source 2278 // argument. In principle there's all kinds of things you could do 2279 // here, for instance creating an == expression and evaluating it with 2280 // EvaluateAsBooleanCondition, but this uses a more direct technique: 2281 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 2282 if (!SrcArgDRE) 2283 return; 2284 2285 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 2286 if (!CompareWithSrcDRE || 2287 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 2288 return; 2289 2290 const Expr *OriginalSizeArg = Call->getArg(2); 2291 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 2292 << OriginalSizeArg->getSourceRange() << FnName; 2293 2294 // Output a FIXIT hint if the destination is an array (rather than a 2295 // pointer to an array). This could be enhanced to handle some 2296 // pointers if we know the actual size, like if DstArg is 'array+2' 2297 // we could say 'sizeof(array)-2'. 2298 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 2299 QualType DstArgTy = DstArg->getType(); 2300 2301 // Only handle constant-sized or VLAs, but not flexible members. 2302 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 2303 // Only issue the FIXIT for arrays of size > 1. 2304 if (CAT->getSize().getSExtValue() <= 1) 2305 return; 2306 } else if (!DstArgTy->isVariableArrayType()) { 2307 return; 2308 } 2309 2310 llvm::SmallString<128> sizeString; 2311 llvm::raw_svector_ostream OS(sizeString); 2312 OS << "sizeof("; 2313 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 2314 OS << ")"; 2315 2316 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 2317 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 2318 OS.str()); 2319} 2320 2321//===--- CHECK: Return Address of Stack Variable --------------------------===// 2322 2323static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars); 2324static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars); 2325 2326/// CheckReturnStackAddr - Check if a return statement returns the address 2327/// of a stack variable. 2328void 2329Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 2330 SourceLocation ReturnLoc) { 2331 2332 Expr *stackE = 0; 2333 SmallVector<DeclRefExpr *, 8> refVars; 2334 2335 // Perform checking for returned stack addresses, local blocks, 2336 // label addresses or references to temporaries. 2337 if (lhsType->isPointerType() || 2338 (!getLangOptions().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 2339 stackE = EvalAddr(RetValExp, refVars); 2340 } else if (lhsType->isReferenceType()) { 2341 stackE = EvalVal(RetValExp, refVars); 2342 } 2343 2344 if (stackE == 0) 2345 return; // Nothing suspicious was found. 2346 2347 SourceLocation diagLoc; 2348 SourceRange diagRange; 2349 if (refVars.empty()) { 2350 diagLoc = stackE->getLocStart(); 2351 diagRange = stackE->getSourceRange(); 2352 } else { 2353 // We followed through a reference variable. 'stackE' contains the 2354 // problematic expression but we will warn at the return statement pointing 2355 // at the reference variable. We will later display the "trail" of 2356 // reference variables using notes. 2357 diagLoc = refVars[0]->getLocStart(); 2358 diagRange = refVars[0]->getSourceRange(); 2359 } 2360 2361 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 2362 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 2363 : diag::warn_ret_stack_addr) 2364 << DR->getDecl()->getDeclName() << diagRange; 2365 } else if (isa<BlockExpr>(stackE)) { // local block. 2366 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 2367 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 2368 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 2369 } else { // local temporary. 2370 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 2371 : diag::warn_ret_local_temp_addr) 2372 << diagRange; 2373 } 2374 2375 // Display the "trail" of reference variables that we followed until we 2376 // found the problematic expression using notes. 2377 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 2378 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 2379 // If this var binds to another reference var, show the range of the next 2380 // var, otherwise the var binds to the problematic expression, in which case 2381 // show the range of the expression. 2382 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 2383 : stackE->getSourceRange(); 2384 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 2385 << VD->getDeclName() << range; 2386 } 2387} 2388 2389/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 2390/// check if the expression in a return statement evaluates to an address 2391/// to a location on the stack, a local block, an address of a label, or a 2392/// reference to local temporary. The recursion is used to traverse the 2393/// AST of the return expression, with recursion backtracking when we 2394/// encounter a subexpression that (1) clearly does not lead to one of the 2395/// above problematic expressions (2) is something we cannot determine leads to 2396/// a problematic expression based on such local checking. 2397/// 2398/// Both EvalAddr and EvalVal follow through reference variables to evaluate 2399/// the expression that they point to. Such variables are added to the 2400/// 'refVars' vector so that we know what the reference variable "trail" was. 2401/// 2402/// EvalAddr processes expressions that are pointers that are used as 2403/// references (and not L-values). EvalVal handles all other values. 2404/// At the base case of the recursion is a check for the above problematic 2405/// expressions. 2406/// 2407/// This implementation handles: 2408/// 2409/// * pointer-to-pointer casts 2410/// * implicit conversions from array references to pointers 2411/// * taking the address of fields 2412/// * arbitrary interplay between "&" and "*" operators 2413/// * pointer arithmetic from an address of a stack variable 2414/// * taking the address of an array element where the array is on the stack 2415static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 2416 if (E->isTypeDependent()) 2417 return NULL; 2418 2419 // We should only be called for evaluating pointer expressions. 2420 assert((E->getType()->isAnyPointerType() || 2421 E->getType()->isBlockPointerType() || 2422 E->getType()->isObjCQualifiedIdType()) && 2423 "EvalAddr only works on pointers"); 2424 2425 E = E->IgnoreParens(); 2426 2427 // Our "symbolic interpreter" is just a dispatch off the currently 2428 // viewed AST node. We then recursively traverse the AST by calling 2429 // EvalAddr and EvalVal appropriately. 2430 switch (E->getStmtClass()) { 2431 case Stmt::DeclRefExprClass: { 2432 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2433 2434 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2435 // If this is a reference variable, follow through to the expression that 2436 // it points to. 2437 if (V->hasLocalStorage() && 2438 V->getType()->isReferenceType() && V->hasInit()) { 2439 // Add the reference variable to the "trail". 2440 refVars.push_back(DR); 2441 return EvalAddr(V->getInit(), refVars); 2442 } 2443 2444 return NULL; 2445 } 2446 2447 case Stmt::UnaryOperatorClass: { 2448 // The only unary operator that make sense to handle here 2449 // is AddrOf. All others don't make sense as pointers. 2450 UnaryOperator *U = cast<UnaryOperator>(E); 2451 2452 if (U->getOpcode() == UO_AddrOf) 2453 return EvalVal(U->getSubExpr(), refVars); 2454 else 2455 return NULL; 2456 } 2457 2458 case Stmt::BinaryOperatorClass: { 2459 // Handle pointer arithmetic. All other binary operators are not valid 2460 // in this context. 2461 BinaryOperator *B = cast<BinaryOperator>(E); 2462 BinaryOperatorKind op = B->getOpcode(); 2463 2464 if (op != BO_Add && op != BO_Sub) 2465 return NULL; 2466 2467 Expr *Base = B->getLHS(); 2468 2469 // Determine which argument is the real pointer base. It could be 2470 // the RHS argument instead of the LHS. 2471 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 2472 2473 assert (Base->getType()->isPointerType()); 2474 return EvalAddr(Base, refVars); 2475 } 2476 2477 // For conditional operators we need to see if either the LHS or RHS are 2478 // valid DeclRefExpr*s. If one of them is valid, we return it. 2479 case Stmt::ConditionalOperatorClass: { 2480 ConditionalOperator *C = cast<ConditionalOperator>(E); 2481 2482 // Handle the GNU extension for missing LHS. 2483 if (Expr *lhsExpr = C->getLHS()) { 2484 // In C++, we can have a throw-expression, which has 'void' type. 2485 if (!lhsExpr->getType()->isVoidType()) 2486 if (Expr* LHS = EvalAddr(lhsExpr, refVars)) 2487 return LHS; 2488 } 2489 2490 // In C++, we can have a throw-expression, which has 'void' type. 2491 if (C->getRHS()->getType()->isVoidType()) 2492 return NULL; 2493 2494 return EvalAddr(C->getRHS(), refVars); 2495 } 2496 2497 case Stmt::BlockExprClass: 2498 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 2499 return E; // local block. 2500 return NULL; 2501 2502 case Stmt::AddrLabelExprClass: 2503 return E; // address of label. 2504 2505 // For casts, we need to handle conversions from arrays to 2506 // pointer values, and pointer-to-pointer conversions. 2507 case Stmt::ImplicitCastExprClass: 2508 case Stmt::CStyleCastExprClass: 2509 case Stmt::CXXFunctionalCastExprClass: 2510 case Stmt::ObjCBridgedCastExprClass: { 2511 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 2512 QualType T = SubExpr->getType(); 2513 2514 if (SubExpr->getType()->isPointerType() || 2515 SubExpr->getType()->isBlockPointerType() || 2516 SubExpr->getType()->isObjCQualifiedIdType()) 2517 return EvalAddr(SubExpr, refVars); 2518 else if (T->isArrayType()) 2519 return EvalVal(SubExpr, refVars); 2520 else 2521 return 0; 2522 } 2523 2524 // C++ casts. For dynamic casts, static casts, and const casts, we 2525 // are always converting from a pointer-to-pointer, so we just blow 2526 // through the cast. In the case the dynamic cast doesn't fail (and 2527 // return NULL), we take the conservative route and report cases 2528 // where we return the address of a stack variable. For Reinterpre 2529 // FIXME: The comment about is wrong; we're not always converting 2530 // from pointer to pointer. I'm guessing that this code should also 2531 // handle references to objects. 2532 case Stmt::CXXStaticCastExprClass: 2533 case Stmt::CXXDynamicCastExprClass: 2534 case Stmt::CXXConstCastExprClass: 2535 case Stmt::CXXReinterpretCastExprClass: { 2536 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 2537 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 2538 return EvalAddr(S, refVars); 2539 else 2540 return NULL; 2541 } 2542 2543 case Stmt::MaterializeTemporaryExprClass: 2544 if (Expr *Result = EvalAddr( 2545 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 2546 refVars)) 2547 return Result; 2548 2549 return E; 2550 2551 // Everything else: we simply don't reason about them. 2552 default: 2553 return NULL; 2554 } 2555} 2556 2557 2558/// EvalVal - This function is complements EvalAddr in the mutual recursion. 2559/// See the comments for EvalAddr for more details. 2560static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 2561do { 2562 // We should only be called for evaluating non-pointer expressions, or 2563 // expressions with a pointer type that are not used as references but instead 2564 // are l-values (e.g., DeclRefExpr with a pointer type). 2565 2566 // Our "symbolic interpreter" is just a dispatch off the currently 2567 // viewed AST node. We then recursively traverse the AST by calling 2568 // EvalAddr and EvalVal appropriately. 2569 2570 E = E->IgnoreParens(); 2571 switch (E->getStmtClass()) { 2572 case Stmt::ImplicitCastExprClass: { 2573 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 2574 if (IE->getValueKind() == VK_LValue) { 2575 E = IE->getSubExpr(); 2576 continue; 2577 } 2578 return NULL; 2579 } 2580 2581 case Stmt::DeclRefExprClass: { 2582 // When we hit a DeclRefExpr we are looking at code that refers to a 2583 // variable's name. If it's not a reference variable we check if it has 2584 // local storage within the function, and if so, return the expression. 2585 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2586 2587 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2588 if (V->hasLocalStorage()) { 2589 if (!V->getType()->isReferenceType()) 2590 return DR; 2591 2592 // Reference variable, follow through to the expression that 2593 // it points to. 2594 if (V->hasInit()) { 2595 // Add the reference variable to the "trail". 2596 refVars.push_back(DR); 2597 return EvalVal(V->getInit(), refVars); 2598 } 2599 } 2600 2601 return NULL; 2602 } 2603 2604 case Stmt::UnaryOperatorClass: { 2605 // The only unary operator that make sense to handle here 2606 // is Deref. All others don't resolve to a "name." This includes 2607 // handling all sorts of rvalues passed to a unary operator. 2608 UnaryOperator *U = cast<UnaryOperator>(E); 2609 2610 if (U->getOpcode() == UO_Deref) 2611 return EvalAddr(U->getSubExpr(), refVars); 2612 2613 return NULL; 2614 } 2615 2616 case Stmt::ArraySubscriptExprClass: { 2617 // Array subscripts are potential references to data on the stack. We 2618 // retrieve the DeclRefExpr* for the array variable if it indeed 2619 // has local storage. 2620 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars); 2621 } 2622 2623 case Stmt::ConditionalOperatorClass: { 2624 // For conditional operators we need to see if either the LHS or RHS are 2625 // non-NULL Expr's. If one is non-NULL, we return it. 2626 ConditionalOperator *C = cast<ConditionalOperator>(E); 2627 2628 // Handle the GNU extension for missing LHS. 2629 if (Expr *lhsExpr = C->getLHS()) 2630 if (Expr *LHS = EvalVal(lhsExpr, refVars)) 2631 return LHS; 2632 2633 return EvalVal(C->getRHS(), refVars); 2634 } 2635 2636 // Accesses to members are potential references to data on the stack. 2637 case Stmt::MemberExprClass: { 2638 MemberExpr *M = cast<MemberExpr>(E); 2639 2640 // Check for indirect access. We only want direct field accesses. 2641 if (M->isArrow()) 2642 return NULL; 2643 2644 // Check whether the member type is itself a reference, in which case 2645 // we're not going to refer to the member, but to what the member refers to. 2646 if (M->getMemberDecl()->getType()->isReferenceType()) 2647 return NULL; 2648 2649 return EvalVal(M->getBase(), refVars); 2650 } 2651 2652 case Stmt::MaterializeTemporaryExprClass: 2653 if (Expr *Result = EvalVal( 2654 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 2655 refVars)) 2656 return Result; 2657 2658 return E; 2659 2660 default: 2661 // Check that we don't return or take the address of a reference to a 2662 // temporary. This is only useful in C++. 2663 if (!E->isTypeDependent() && E->isRValue()) 2664 return E; 2665 2666 // Everything else: we simply don't reason about them. 2667 return NULL; 2668 } 2669} while (true); 2670} 2671 2672//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 2673 2674/// Check for comparisons of floating point operands using != and ==. 2675/// Issue a warning if these are no self-comparisons, as they are not likely 2676/// to do what the programmer intended. 2677void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 2678 bool EmitWarning = true; 2679 2680 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 2681 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 2682 2683 // Special case: check for x == x (which is OK). 2684 // Do not emit warnings for such cases. 2685 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 2686 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 2687 if (DRL->getDecl() == DRR->getDecl()) 2688 EmitWarning = false; 2689 2690 2691 // Special case: check for comparisons against literals that can be exactly 2692 // represented by APFloat. In such cases, do not emit a warning. This 2693 // is a heuristic: often comparison against such literals are used to 2694 // detect if a value in a variable has not changed. This clearly can 2695 // lead to false negatives. 2696 if (EmitWarning) { 2697 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 2698 if (FLL->isExact()) 2699 EmitWarning = false; 2700 } else 2701 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 2702 if (FLR->isExact()) 2703 EmitWarning = false; 2704 } 2705 } 2706 2707 // Check for comparisons with builtin types. 2708 if (EmitWarning) 2709 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 2710 if (CL->isBuiltinCall(Context)) 2711 EmitWarning = false; 2712 2713 if (EmitWarning) 2714 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 2715 if (CR->isBuiltinCall(Context)) 2716 EmitWarning = false; 2717 2718 // Emit the diagnostic. 2719 if (EmitWarning) 2720 Diag(Loc, diag::warn_floatingpoint_eq) 2721 << LHS->getSourceRange() << RHS->getSourceRange(); 2722} 2723 2724//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 2725//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 2726 2727namespace { 2728 2729/// Structure recording the 'active' range of an integer-valued 2730/// expression. 2731struct IntRange { 2732 /// The number of bits active in the int. 2733 unsigned Width; 2734 2735 /// True if the int is known not to have negative values. 2736 bool NonNegative; 2737 2738 IntRange(unsigned Width, bool NonNegative) 2739 : Width(Width), NonNegative(NonNegative) 2740 {} 2741 2742 /// Returns the range of the bool type. 2743 static IntRange forBoolType() { 2744 return IntRange(1, true); 2745 } 2746 2747 /// Returns the range of an opaque value of the given integral type. 2748 static IntRange forValueOfType(ASTContext &C, QualType T) { 2749 return forValueOfCanonicalType(C, 2750 T->getCanonicalTypeInternal().getTypePtr()); 2751 } 2752 2753 /// Returns the range of an opaque value of a canonical integral type. 2754 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 2755 assert(T->isCanonicalUnqualified()); 2756 2757 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2758 T = VT->getElementType().getTypePtr(); 2759 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2760 T = CT->getElementType().getTypePtr(); 2761 2762 // For enum types, use the known bit width of the enumerators. 2763 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 2764 EnumDecl *Enum = ET->getDecl(); 2765 if (!Enum->isCompleteDefinition()) 2766 return IntRange(C.getIntWidth(QualType(T, 0)), false); 2767 2768 unsigned NumPositive = Enum->getNumPositiveBits(); 2769 unsigned NumNegative = Enum->getNumNegativeBits(); 2770 2771 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 2772 } 2773 2774 const BuiltinType *BT = cast<BuiltinType>(T); 2775 assert(BT->isInteger()); 2776 2777 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2778 } 2779 2780 /// Returns the "target" range of a canonical integral type, i.e. 2781 /// the range of values expressible in the type. 2782 /// 2783 /// This matches forValueOfCanonicalType except that enums have the 2784 /// full range of their type, not the range of their enumerators. 2785 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 2786 assert(T->isCanonicalUnqualified()); 2787 2788 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2789 T = VT->getElementType().getTypePtr(); 2790 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2791 T = CT->getElementType().getTypePtr(); 2792 if (const EnumType *ET = dyn_cast<EnumType>(T)) 2793 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 2794 2795 const BuiltinType *BT = cast<BuiltinType>(T); 2796 assert(BT->isInteger()); 2797 2798 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2799 } 2800 2801 /// Returns the supremum of two ranges: i.e. their conservative merge. 2802 static IntRange join(IntRange L, IntRange R) { 2803 return IntRange(std::max(L.Width, R.Width), 2804 L.NonNegative && R.NonNegative); 2805 } 2806 2807 /// Returns the infinum of two ranges: i.e. their aggressive merge. 2808 static IntRange meet(IntRange L, IntRange R) { 2809 return IntRange(std::min(L.Width, R.Width), 2810 L.NonNegative || R.NonNegative); 2811 } 2812}; 2813 2814IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 2815 if (value.isSigned() && value.isNegative()) 2816 return IntRange(value.getMinSignedBits(), false); 2817 2818 if (value.getBitWidth() > MaxWidth) 2819 value = value.trunc(MaxWidth); 2820 2821 // isNonNegative() just checks the sign bit without considering 2822 // signedness. 2823 return IntRange(value.getActiveBits(), true); 2824} 2825 2826IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 2827 unsigned MaxWidth) { 2828 if (result.isInt()) 2829 return GetValueRange(C, result.getInt(), MaxWidth); 2830 2831 if (result.isVector()) { 2832 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 2833 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 2834 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 2835 R = IntRange::join(R, El); 2836 } 2837 return R; 2838 } 2839 2840 if (result.isComplexInt()) { 2841 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 2842 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 2843 return IntRange::join(R, I); 2844 } 2845 2846 // This can happen with lossless casts to intptr_t of "based" lvalues. 2847 // Assume it might use arbitrary bits. 2848 // FIXME: The only reason we need to pass the type in here is to get 2849 // the sign right on this one case. It would be nice if APValue 2850 // preserved this. 2851 assert(result.isLValue()); 2852 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 2853} 2854 2855/// Pseudo-evaluate the given integer expression, estimating the 2856/// range of values it might take. 2857/// 2858/// \param MaxWidth - the width to which the value will be truncated 2859IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 2860 E = E->IgnoreParens(); 2861 2862 // Try a full evaluation first. 2863 Expr::EvalResult result; 2864 if (E->Evaluate(result, C)) 2865 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 2866 2867 // I think we only want to look through implicit casts here; if the 2868 // user has an explicit widening cast, we should treat the value as 2869 // being of the new, wider type. 2870 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 2871 if (CE->getCastKind() == CK_NoOp) 2872 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 2873 2874 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 2875 2876 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 2877 2878 // Assume that non-integer casts can span the full range of the type. 2879 if (!isIntegerCast) 2880 return OutputTypeRange; 2881 2882 IntRange SubRange 2883 = GetExprRange(C, CE->getSubExpr(), 2884 std::min(MaxWidth, OutputTypeRange.Width)); 2885 2886 // Bail out if the subexpr's range is as wide as the cast type. 2887 if (SubRange.Width >= OutputTypeRange.Width) 2888 return OutputTypeRange; 2889 2890 // Otherwise, we take the smaller width, and we're non-negative if 2891 // either the output type or the subexpr is. 2892 return IntRange(SubRange.Width, 2893 SubRange.NonNegative || OutputTypeRange.NonNegative); 2894 } 2895 2896 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 2897 // If we can fold the condition, just take that operand. 2898 bool CondResult; 2899 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 2900 return GetExprRange(C, CondResult ? CO->getTrueExpr() 2901 : CO->getFalseExpr(), 2902 MaxWidth); 2903 2904 // Otherwise, conservatively merge. 2905 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 2906 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 2907 return IntRange::join(L, R); 2908 } 2909 2910 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 2911 switch (BO->getOpcode()) { 2912 2913 // Boolean-valued operations are single-bit and positive. 2914 case BO_LAnd: 2915 case BO_LOr: 2916 case BO_LT: 2917 case BO_GT: 2918 case BO_LE: 2919 case BO_GE: 2920 case BO_EQ: 2921 case BO_NE: 2922 return IntRange::forBoolType(); 2923 2924 // The type of the assignments is the type of the LHS, so the RHS 2925 // is not necessarily the same type. 2926 case BO_MulAssign: 2927 case BO_DivAssign: 2928 case BO_RemAssign: 2929 case BO_AddAssign: 2930 case BO_SubAssign: 2931 case BO_XorAssign: 2932 case BO_OrAssign: 2933 // TODO: bitfields? 2934 return IntRange::forValueOfType(C, E->getType()); 2935 2936 // Simple assignments just pass through the RHS, which will have 2937 // been coerced to the LHS type. 2938 case BO_Assign: 2939 // TODO: bitfields? 2940 return GetExprRange(C, BO->getRHS(), MaxWidth); 2941 2942 // Operations with opaque sources are black-listed. 2943 case BO_PtrMemD: 2944 case BO_PtrMemI: 2945 return IntRange::forValueOfType(C, E->getType()); 2946 2947 // Bitwise-and uses the *infinum* of the two source ranges. 2948 case BO_And: 2949 case BO_AndAssign: 2950 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 2951 GetExprRange(C, BO->getRHS(), MaxWidth)); 2952 2953 // Left shift gets black-listed based on a judgement call. 2954 case BO_Shl: 2955 // ...except that we want to treat '1 << (blah)' as logically 2956 // positive. It's an important idiom. 2957 if (IntegerLiteral *I 2958 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 2959 if (I->getValue() == 1) { 2960 IntRange R = IntRange::forValueOfType(C, E->getType()); 2961 return IntRange(R.Width, /*NonNegative*/ true); 2962 } 2963 } 2964 // fallthrough 2965 2966 case BO_ShlAssign: 2967 return IntRange::forValueOfType(C, E->getType()); 2968 2969 // Right shift by a constant can narrow its left argument. 2970 case BO_Shr: 2971 case BO_ShrAssign: { 2972 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2973 2974 // If the shift amount is a positive constant, drop the width by 2975 // that much. 2976 llvm::APSInt shift; 2977 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 2978 shift.isNonNegative()) { 2979 unsigned zext = shift.getZExtValue(); 2980 if (zext >= L.Width) 2981 L.Width = (L.NonNegative ? 0 : 1); 2982 else 2983 L.Width -= zext; 2984 } 2985 2986 return L; 2987 } 2988 2989 // Comma acts as its right operand. 2990 case BO_Comma: 2991 return GetExprRange(C, BO->getRHS(), MaxWidth); 2992 2993 // Black-list pointer subtractions. 2994 case BO_Sub: 2995 if (BO->getLHS()->getType()->isPointerType()) 2996 return IntRange::forValueOfType(C, E->getType()); 2997 break; 2998 2999 // The width of a division result is mostly determined by the size 3000 // of the LHS. 3001 case BO_Div: { 3002 // Don't 'pre-truncate' the operands. 3003 unsigned opWidth = C.getIntWidth(E->getType()); 3004 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3005 3006 // If the divisor is constant, use that. 3007 llvm::APSInt divisor; 3008 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 3009 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 3010 if (log2 >= L.Width) 3011 L.Width = (L.NonNegative ? 0 : 1); 3012 else 3013 L.Width = std::min(L.Width - log2, MaxWidth); 3014 return L; 3015 } 3016 3017 // Otherwise, just use the LHS's width. 3018 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3019 return IntRange(L.Width, L.NonNegative && R.NonNegative); 3020 } 3021 3022 // The result of a remainder can't be larger than the result of 3023 // either side. 3024 case BO_Rem: { 3025 // Don't 'pre-truncate' the operands. 3026 unsigned opWidth = C.getIntWidth(E->getType()); 3027 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3028 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3029 3030 IntRange meet = IntRange::meet(L, R); 3031 meet.Width = std::min(meet.Width, MaxWidth); 3032 return meet; 3033 } 3034 3035 // The default behavior is okay for these. 3036 case BO_Mul: 3037 case BO_Add: 3038 case BO_Xor: 3039 case BO_Or: 3040 break; 3041 } 3042 3043 // The default case is to treat the operation as if it were closed 3044 // on the narrowest type that encompasses both operands. 3045 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3046 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 3047 return IntRange::join(L, R); 3048 } 3049 3050 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 3051 switch (UO->getOpcode()) { 3052 // Boolean-valued operations are white-listed. 3053 case UO_LNot: 3054 return IntRange::forBoolType(); 3055 3056 // Operations with opaque sources are black-listed. 3057 case UO_Deref: 3058 case UO_AddrOf: // should be impossible 3059 return IntRange::forValueOfType(C, E->getType()); 3060 3061 default: 3062 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 3063 } 3064 } 3065 3066 if (dyn_cast<OffsetOfExpr>(E)) { 3067 IntRange::forValueOfType(C, E->getType()); 3068 } 3069 3070 if (FieldDecl *BitField = E->getBitField()) 3071 return IntRange(BitField->getBitWidthValue(C), 3072 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 3073 3074 return IntRange::forValueOfType(C, E->getType()); 3075} 3076 3077IntRange GetExprRange(ASTContext &C, Expr *E) { 3078 return GetExprRange(C, E, C.getIntWidth(E->getType())); 3079} 3080 3081/// Checks whether the given value, which currently has the given 3082/// source semantics, has the same value when coerced through the 3083/// target semantics. 3084bool IsSameFloatAfterCast(const llvm::APFloat &value, 3085 const llvm::fltSemantics &Src, 3086 const llvm::fltSemantics &Tgt) { 3087 llvm::APFloat truncated = value; 3088 3089 bool ignored; 3090 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 3091 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 3092 3093 return truncated.bitwiseIsEqual(value); 3094} 3095 3096/// Checks whether the given value, which currently has the given 3097/// source semantics, has the same value when coerced through the 3098/// target semantics. 3099/// 3100/// The value might be a vector of floats (or a complex number). 3101bool IsSameFloatAfterCast(const APValue &value, 3102 const llvm::fltSemantics &Src, 3103 const llvm::fltSemantics &Tgt) { 3104 if (value.isFloat()) 3105 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 3106 3107 if (value.isVector()) { 3108 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 3109 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 3110 return false; 3111 return true; 3112 } 3113 3114 assert(value.isComplexFloat()); 3115 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 3116 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 3117} 3118 3119void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 3120 3121static bool IsZero(Sema &S, Expr *E) { 3122 // Suppress cases where we are comparing against an enum constant. 3123 if (const DeclRefExpr *DR = 3124 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 3125 if (isa<EnumConstantDecl>(DR->getDecl())) 3126 return false; 3127 3128 // Suppress cases where the '0' value is expanded from a macro. 3129 if (E->getLocStart().isMacroID()) 3130 return false; 3131 3132 llvm::APSInt Value; 3133 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 3134} 3135 3136static bool HasEnumType(Expr *E) { 3137 // Strip off implicit integral promotions. 3138 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 3139 if (ICE->getCastKind() != CK_IntegralCast && 3140 ICE->getCastKind() != CK_NoOp) 3141 break; 3142 E = ICE->getSubExpr(); 3143 } 3144 3145 return E->getType()->isEnumeralType(); 3146} 3147 3148void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 3149 BinaryOperatorKind op = E->getOpcode(); 3150 if (E->isValueDependent()) 3151 return; 3152 3153 if (op == BO_LT && IsZero(S, E->getRHS())) { 3154 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3155 << "< 0" << "false" << HasEnumType(E->getLHS()) 3156 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3157 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 3158 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3159 << ">= 0" << "true" << HasEnumType(E->getLHS()) 3160 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3161 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 3162 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3163 << "0 >" << "false" << HasEnumType(E->getRHS()) 3164 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3165 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 3166 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3167 << "0 <=" << "true" << HasEnumType(E->getRHS()) 3168 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3169 } 3170} 3171 3172/// Analyze the operands of the given comparison. Implements the 3173/// fallback case from AnalyzeComparison. 3174void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 3175 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3176 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3177} 3178 3179/// \brief Implements -Wsign-compare. 3180/// 3181/// \param E the binary operator to check for warnings 3182void AnalyzeComparison(Sema &S, BinaryOperator *E) { 3183 // The type the comparison is being performed in. 3184 QualType T = E->getLHS()->getType(); 3185 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 3186 && "comparison with mismatched types"); 3187 3188 // We don't do anything special if this isn't an unsigned integral 3189 // comparison: we're only interested in integral comparisons, and 3190 // signed comparisons only happen in cases we don't care to warn about. 3191 // 3192 // We also don't care about value-dependent expressions or expressions 3193 // whose result is a constant. 3194 if (!T->hasUnsignedIntegerRepresentation() 3195 || E->isValueDependent() || E->isIntegerConstantExpr(S.Context)) 3196 return AnalyzeImpConvsInComparison(S, E); 3197 3198 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 3199 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 3200 3201 // Check to see if one of the (unmodified) operands is of different 3202 // signedness. 3203 Expr *signedOperand, *unsignedOperand; 3204 if (LHS->getType()->hasSignedIntegerRepresentation()) { 3205 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 3206 "unsigned comparison between two signed integer expressions?"); 3207 signedOperand = LHS; 3208 unsignedOperand = RHS; 3209 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 3210 signedOperand = RHS; 3211 unsignedOperand = LHS; 3212 } else { 3213 CheckTrivialUnsignedComparison(S, E); 3214 return AnalyzeImpConvsInComparison(S, E); 3215 } 3216 3217 // Otherwise, calculate the effective range of the signed operand. 3218 IntRange signedRange = GetExprRange(S.Context, signedOperand); 3219 3220 // Go ahead and analyze implicit conversions in the operands. Note 3221 // that we skip the implicit conversions on both sides. 3222 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 3223 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 3224 3225 // If the signed range is non-negative, -Wsign-compare won't fire, 3226 // but we should still check for comparisons which are always true 3227 // or false. 3228 if (signedRange.NonNegative) 3229 return CheckTrivialUnsignedComparison(S, E); 3230 3231 // For (in)equality comparisons, if the unsigned operand is a 3232 // constant which cannot collide with a overflowed signed operand, 3233 // then reinterpreting the signed operand as unsigned will not 3234 // change the result of the comparison. 3235 if (E->isEqualityOp()) { 3236 unsigned comparisonWidth = S.Context.getIntWidth(T); 3237 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 3238 3239 // We should never be unable to prove that the unsigned operand is 3240 // non-negative. 3241 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 3242 3243 if (unsignedRange.Width < comparisonWidth) 3244 return; 3245 } 3246 3247 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison) 3248 << LHS->getType() << RHS->getType() 3249 << LHS->getSourceRange() << RHS->getSourceRange(); 3250} 3251 3252/// Analyzes an attempt to assign the given value to a bitfield. 3253/// 3254/// Returns true if there was something fishy about the attempt. 3255bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 3256 SourceLocation InitLoc) { 3257 assert(Bitfield->isBitField()); 3258 if (Bitfield->isInvalidDecl()) 3259 return false; 3260 3261 // White-list bool bitfields. 3262 if (Bitfield->getType()->isBooleanType()) 3263 return false; 3264 3265 // Ignore value- or type-dependent expressions. 3266 if (Bitfield->getBitWidth()->isValueDependent() || 3267 Bitfield->getBitWidth()->isTypeDependent() || 3268 Init->isValueDependent() || 3269 Init->isTypeDependent()) 3270 return false; 3271 3272 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 3273 3274 Expr::EvalResult InitValue; 3275 if (!OriginalInit->Evaluate(InitValue, S.Context) || 3276 !InitValue.Val.isInt()) 3277 return false; 3278 3279 const llvm::APSInt &Value = InitValue.Val.getInt(); 3280 unsigned OriginalWidth = Value.getBitWidth(); 3281 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 3282 3283 if (OriginalWidth <= FieldWidth) 3284 return false; 3285 3286 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 3287 3288 // It's fairly common to write values into signed bitfields 3289 // that, if sign-extended, would end up becoming a different 3290 // value. We don't want to warn about that. 3291 if (Value.isSigned() && Value.isNegative()) 3292 TruncatedValue = TruncatedValue.sext(OriginalWidth); 3293 else 3294 TruncatedValue = TruncatedValue.zext(OriginalWidth); 3295 3296 if (Value == TruncatedValue) 3297 return false; 3298 3299 std::string PrettyValue = Value.toString(10); 3300 std::string PrettyTrunc = TruncatedValue.toString(10); 3301 3302 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 3303 << PrettyValue << PrettyTrunc << OriginalInit->getType() 3304 << Init->getSourceRange(); 3305 3306 return true; 3307} 3308 3309/// Analyze the given simple or compound assignment for warning-worthy 3310/// operations. 3311void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 3312 // Just recurse on the LHS. 3313 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3314 3315 // We want to recurse on the RHS as normal unless we're assigning to 3316 // a bitfield. 3317 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 3318 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 3319 E->getOperatorLoc())) { 3320 // Recurse, ignoring any implicit conversions on the RHS. 3321 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 3322 E->getOperatorLoc()); 3323 } 3324 } 3325 3326 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3327} 3328 3329/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3330void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 3331 SourceLocation CContext, unsigned diag) { 3332 S.Diag(E->getExprLoc(), diag) 3333 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 3334} 3335 3336/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3337void DiagnoseImpCast(Sema &S, Expr *E, QualType T, SourceLocation CContext, 3338 unsigned diag) { 3339 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag); 3340} 3341 3342/// Diagnose an implicit cast from a literal expression. Also attemps to supply 3343/// fixit hints when the cast wouldn't lose information to simply write the 3344/// expression with the expected type. 3345void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 3346 SourceLocation CContext) { 3347 // Emit the primary warning first, then try to emit a fixit hint note if 3348 // reasonable. 3349 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 3350 << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext); 3351 3352 const llvm::APFloat &Value = FL->getValue(); 3353 3354 // Don't attempt to fix PPC double double literals. 3355 if (&Value.getSemantics() == &llvm::APFloat::PPCDoubleDouble) 3356 return; 3357 3358 // Try to convert this exactly to an integer. 3359 bool isExact = false; 3360 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 3361 T->hasUnsignedIntegerRepresentation()); 3362 if (Value.convertToInteger(IntegerValue, 3363 llvm::APFloat::rmTowardZero, &isExact) 3364 != llvm::APFloat::opOK || !isExact) 3365 return; 3366 3367 std::string LiteralValue = IntegerValue.toString(10); 3368 S.Diag(FL->getExprLoc(), diag::note_fix_integral_float_as_integer) 3369 << FixItHint::CreateReplacement(FL->getSourceRange(), LiteralValue); 3370} 3371 3372std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 3373 if (!Range.Width) return "0"; 3374 3375 llvm::APSInt ValueInRange = Value; 3376 ValueInRange.setIsSigned(!Range.NonNegative); 3377 ValueInRange = ValueInRange.trunc(Range.Width); 3378 return ValueInRange.toString(10); 3379} 3380 3381static bool isFromSystemMacro(Sema &S, SourceLocation loc) { 3382 SourceManager &smgr = S.Context.getSourceManager(); 3383 return loc.isMacroID() && smgr.isInSystemHeader(smgr.getSpellingLoc(loc)); 3384} 3385 3386void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 3387 SourceLocation CC, bool *ICContext = 0) { 3388 if (E->isTypeDependent() || E->isValueDependent()) return; 3389 3390 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 3391 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 3392 if (Source == Target) return; 3393 if (Target->isDependentType()) return; 3394 3395 // If the conversion context location is invalid don't complain. We also 3396 // don't want to emit a warning if the issue occurs from the expansion of 3397 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 3398 // delay this check as long as possible. Once we detect we are in that 3399 // scenario, we just return. 3400 if (CC.isInvalid()) 3401 return; 3402 3403 // Diagnose implicit casts to bool. 3404 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 3405 if (isa<StringLiteral>(E)) 3406 // Warn on string literal to bool. Checks for string literals in logical 3407 // expressions, for instances, assert(0 && "error here"), is prevented 3408 // by a check in AnalyzeImplicitConversions(). 3409 return DiagnoseImpCast(S, E, T, CC, 3410 diag::warn_impcast_string_literal_to_bool); 3411 return; // Other casts to bool are not checked. 3412 } 3413 3414 // Strip vector types. 3415 if (isa<VectorType>(Source)) { 3416 if (!isa<VectorType>(Target)) { 3417 if (isFromSystemMacro(S, CC)) 3418 return; 3419 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 3420 } 3421 3422 // If the vector cast is cast between two vectors of the same size, it is 3423 // a bitcast, not a conversion. 3424 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 3425 return; 3426 3427 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 3428 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 3429 } 3430 3431 // Strip complex types. 3432 if (isa<ComplexType>(Source)) { 3433 if (!isa<ComplexType>(Target)) { 3434 if (isFromSystemMacro(S, CC)) 3435 return; 3436 3437 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 3438 } 3439 3440 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 3441 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 3442 } 3443 3444 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 3445 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 3446 3447 // If the source is floating point... 3448 if (SourceBT && SourceBT->isFloatingPoint()) { 3449 // ...and the target is floating point... 3450 if (TargetBT && TargetBT->isFloatingPoint()) { 3451 // ...then warn if we're dropping FP rank. 3452 3453 // Builtin FP kinds are ordered by increasing FP rank. 3454 if (SourceBT->getKind() > TargetBT->getKind()) { 3455 // Don't warn about float constants that are precisely 3456 // representable in the target type. 3457 Expr::EvalResult result; 3458 if (E->Evaluate(result, S.Context)) { 3459 // Value might be a float, a float vector, or a float complex. 3460 if (IsSameFloatAfterCast(result.Val, 3461 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 3462 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 3463 return; 3464 } 3465 3466 if (isFromSystemMacro(S, CC)) 3467 return; 3468 3469 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 3470 } 3471 return; 3472 } 3473 3474 // If the target is integral, always warn. 3475 if ((TargetBT && TargetBT->isInteger())) { 3476 if (isFromSystemMacro(S, CC)) 3477 return; 3478 3479 Expr *InnerE = E->IgnoreParenImpCasts(); 3480 // We also want to warn on, e.g., "int i = -1.234" 3481 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 3482 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 3483 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 3484 3485 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 3486 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 3487 } else { 3488 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 3489 } 3490 } 3491 3492 return; 3493 } 3494 3495 if (!Source->isIntegerType() || !Target->isIntegerType()) 3496 return; 3497 3498 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 3499 == Expr::NPCK_GNUNull) && Target->isIntegerType()) { 3500 S.Diag(E->getExprLoc(), diag::warn_impcast_null_pointer_to_integer) 3501 << E->getSourceRange() << clang::SourceRange(CC); 3502 return; 3503 } 3504 3505 IntRange SourceRange = GetExprRange(S.Context, E); 3506 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 3507 3508 if (SourceRange.Width > TargetRange.Width) { 3509 // If the source is a constant, use a default-on diagnostic. 3510 // TODO: this should happen for bitfield stores, too. 3511 llvm::APSInt Value(32); 3512 if (E->isIntegerConstantExpr(Value, S.Context)) { 3513 if (isFromSystemMacro(S, CC)) 3514 return; 3515 3516 std::string PrettySourceValue = Value.toString(10); 3517 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 3518 3519 S.Diag(E->getExprLoc(), diag::warn_impcast_integer_precision_constant) 3520 << PrettySourceValue << PrettyTargetValue 3521 << E->getType() << T << E->getSourceRange() << clang::SourceRange(CC); 3522 return; 3523 } 3524 3525 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 3526 if (isFromSystemMacro(S, CC)) 3527 return; 3528 3529 if (SourceRange.Width == 64 && TargetRange.Width == 32) 3530 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32); 3531 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 3532 } 3533 3534 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 3535 (!TargetRange.NonNegative && SourceRange.NonNegative && 3536 SourceRange.Width == TargetRange.Width)) { 3537 3538 if (isFromSystemMacro(S, CC)) 3539 return; 3540 3541 unsigned DiagID = diag::warn_impcast_integer_sign; 3542 3543 // Traditionally, gcc has warned about this under -Wsign-compare. 3544 // We also want to warn about it in -Wconversion. 3545 // So if -Wconversion is off, use a completely identical diagnostic 3546 // in the sign-compare group. 3547 // The conditional-checking code will 3548 if (ICContext) { 3549 DiagID = diag::warn_impcast_integer_sign_conditional; 3550 *ICContext = true; 3551 } 3552 3553 return DiagnoseImpCast(S, E, T, CC, DiagID); 3554 } 3555 3556 // Diagnose conversions between different enumeration types. 3557 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 3558 // type, to give us better diagnostics. 3559 QualType SourceType = E->getType(); 3560 if (!S.getLangOptions().CPlusPlus) { 3561 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 3562 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 3563 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 3564 SourceType = S.Context.getTypeDeclType(Enum); 3565 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 3566 } 3567 } 3568 3569 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 3570 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 3571 if ((SourceEnum->getDecl()->getIdentifier() || 3572 SourceEnum->getDecl()->getTypedefNameForAnonDecl()) && 3573 (TargetEnum->getDecl()->getIdentifier() || 3574 TargetEnum->getDecl()->getTypedefNameForAnonDecl()) && 3575 SourceEnum != TargetEnum) { 3576 if (isFromSystemMacro(S, CC)) 3577 return; 3578 3579 return DiagnoseImpCast(S, E, SourceType, T, CC, 3580 diag::warn_impcast_different_enum_types); 3581 } 3582 3583 return; 3584} 3585 3586void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 3587 3588void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 3589 SourceLocation CC, bool &ICContext) { 3590 E = E->IgnoreParenImpCasts(); 3591 3592 if (isa<ConditionalOperator>(E)) 3593 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 3594 3595 AnalyzeImplicitConversions(S, E, CC); 3596 if (E->getType() != T) 3597 return CheckImplicitConversion(S, E, T, CC, &ICContext); 3598 return; 3599} 3600 3601void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 3602 SourceLocation CC = E->getQuestionLoc(); 3603 3604 AnalyzeImplicitConversions(S, E->getCond(), CC); 3605 3606 bool Suspicious = false; 3607 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 3608 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 3609 3610 // If -Wconversion would have warned about either of the candidates 3611 // for a signedness conversion to the context type... 3612 if (!Suspicious) return; 3613 3614 // ...but it's currently ignored... 3615 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 3616 CC)) 3617 return; 3618 3619 // ...then check whether it would have warned about either of the 3620 // candidates for a signedness conversion to the condition type. 3621 if (E->getType() == T) return; 3622 3623 Suspicious = false; 3624 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 3625 E->getType(), CC, &Suspicious); 3626 if (!Suspicious) 3627 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 3628 E->getType(), CC, &Suspicious); 3629} 3630 3631/// AnalyzeImplicitConversions - Find and report any interesting 3632/// implicit conversions in the given expression. There are a couple 3633/// of competing diagnostics here, -Wconversion and -Wsign-compare. 3634void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 3635 QualType T = OrigE->getType(); 3636 Expr *E = OrigE->IgnoreParenImpCasts(); 3637 3638 if (E->isTypeDependent() || E->isValueDependent()) 3639 return; 3640 3641 // For conditional operators, we analyze the arguments as if they 3642 // were being fed directly into the output. 3643 if (isa<ConditionalOperator>(E)) { 3644 ConditionalOperator *CO = cast<ConditionalOperator>(E); 3645 CheckConditionalOperator(S, CO, T); 3646 return; 3647 } 3648 3649 // Go ahead and check any implicit conversions we might have skipped. 3650 // The non-canonical typecheck is just an optimization; 3651 // CheckImplicitConversion will filter out dead implicit conversions. 3652 if (E->getType() != T) 3653 CheckImplicitConversion(S, E, T, CC); 3654 3655 // Now continue drilling into this expression. 3656 3657 // Skip past explicit casts. 3658 if (isa<ExplicitCastExpr>(E)) { 3659 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 3660 return AnalyzeImplicitConversions(S, E, CC); 3661 } 3662 3663 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 3664 // Do a somewhat different check with comparison operators. 3665 if (BO->isComparisonOp()) 3666 return AnalyzeComparison(S, BO); 3667 3668 // And with assignments and compound assignments. 3669 if (BO->isAssignmentOp()) 3670 return AnalyzeAssignment(S, BO); 3671 } 3672 3673 // These break the otherwise-useful invariant below. Fortunately, 3674 // we don't really need to recurse into them, because any internal 3675 // expressions should have been analyzed already when they were 3676 // built into statements. 3677 if (isa<StmtExpr>(E)) return; 3678 3679 // Don't descend into unevaluated contexts. 3680 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 3681 3682 // Now just recurse over the expression's children. 3683 CC = E->getExprLoc(); 3684 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 3685 bool IsLogicalOperator = BO && BO->isLogicalOp(); 3686 for (Stmt::child_range I = E->children(); I; ++I) { 3687 Expr *ChildExpr = cast<Expr>(*I); 3688 if (IsLogicalOperator && 3689 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 3690 // Ignore checking string literals that are in logical operators. 3691 continue; 3692 AnalyzeImplicitConversions(S, ChildExpr, CC); 3693 } 3694} 3695 3696} // end anonymous namespace 3697 3698/// Diagnoses "dangerous" implicit conversions within the given 3699/// expression (which is a full expression). Implements -Wconversion 3700/// and -Wsign-compare. 3701/// 3702/// \param CC the "context" location of the implicit conversion, i.e. 3703/// the most location of the syntactic entity requiring the implicit 3704/// conversion 3705void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 3706 // Don't diagnose in unevaluated contexts. 3707 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 3708 return; 3709 3710 // Don't diagnose for value- or type-dependent expressions. 3711 if (E->isTypeDependent() || E->isValueDependent()) 3712 return; 3713 3714 // Check for array bounds violations in cases where the check isn't triggered 3715 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 3716 // ArraySubscriptExpr is on the RHS of a variable initialization. 3717 CheckArrayAccess(E); 3718 3719 // This is not the right CC for (e.g.) a variable initialization. 3720 AnalyzeImplicitConversions(*this, E, CC); 3721} 3722 3723void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 3724 FieldDecl *BitField, 3725 Expr *Init) { 3726 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 3727} 3728 3729/// CheckParmsForFunctionDef - Check that the parameters of the given 3730/// function are appropriate for the definition of a function. This 3731/// takes care of any checks that cannot be performed on the 3732/// declaration itself, e.g., that the types of each of the function 3733/// parameters are complete. 3734bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 3735 bool CheckParameterNames) { 3736 bool HasInvalidParm = false; 3737 for (; P != PEnd; ++P) { 3738 ParmVarDecl *Param = *P; 3739 3740 // C99 6.7.5.3p4: the parameters in a parameter type list in a 3741 // function declarator that is part of a function definition of 3742 // that function shall not have incomplete type. 3743 // 3744 // This is also C++ [dcl.fct]p6. 3745 if (!Param->isInvalidDecl() && 3746 RequireCompleteType(Param->getLocation(), Param->getType(), 3747 diag::err_typecheck_decl_incomplete_type)) { 3748 Param->setInvalidDecl(); 3749 HasInvalidParm = true; 3750 } 3751 3752 // C99 6.9.1p5: If the declarator includes a parameter type list, the 3753 // declaration of each parameter shall include an identifier. 3754 if (CheckParameterNames && 3755 Param->getIdentifier() == 0 && 3756 !Param->isImplicit() && 3757 !getLangOptions().CPlusPlus) 3758 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 3759 3760 // C99 6.7.5.3p12: 3761 // If the function declarator is not part of a definition of that 3762 // function, parameters may have incomplete type and may use the [*] 3763 // notation in their sequences of declarator specifiers to specify 3764 // variable length array types. 3765 QualType PType = Param->getOriginalType(); 3766 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 3767 if (AT->getSizeModifier() == ArrayType::Star) { 3768 // FIXME: This diagnosic should point the the '[*]' if source-location 3769 // information is added for it. 3770 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 3771 } 3772 } 3773 } 3774 3775 return HasInvalidParm; 3776} 3777 3778/// CheckCastAlign - Implements -Wcast-align, which warns when a 3779/// pointer cast increases the alignment requirements. 3780void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 3781 // This is actually a lot of work to potentially be doing on every 3782 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 3783 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 3784 TRange.getBegin()) 3785 == DiagnosticsEngine::Ignored) 3786 return; 3787 3788 // Ignore dependent types. 3789 if (T->isDependentType() || Op->getType()->isDependentType()) 3790 return; 3791 3792 // Require that the destination be a pointer type. 3793 const PointerType *DestPtr = T->getAs<PointerType>(); 3794 if (!DestPtr) return; 3795 3796 // If the destination has alignment 1, we're done. 3797 QualType DestPointee = DestPtr->getPointeeType(); 3798 if (DestPointee->isIncompleteType()) return; 3799 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 3800 if (DestAlign.isOne()) return; 3801 3802 // Require that the source be a pointer type. 3803 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 3804 if (!SrcPtr) return; 3805 QualType SrcPointee = SrcPtr->getPointeeType(); 3806 3807 // Whitelist casts from cv void*. We already implicitly 3808 // whitelisted casts to cv void*, since they have alignment 1. 3809 // Also whitelist casts involving incomplete types, which implicitly 3810 // includes 'void'. 3811 if (SrcPointee->isIncompleteType()) return; 3812 3813 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 3814 if (SrcAlign >= DestAlign) return; 3815 3816 Diag(TRange.getBegin(), diag::warn_cast_align) 3817 << Op->getType() << T 3818 << static_cast<unsigned>(SrcAlign.getQuantity()) 3819 << static_cast<unsigned>(DestAlign.getQuantity()) 3820 << TRange << Op->getSourceRange(); 3821} 3822 3823static const Type* getElementType(const Expr *BaseExpr) { 3824 const Type* EltType = BaseExpr->getType().getTypePtr(); 3825 if (EltType->isAnyPointerType()) 3826 return EltType->getPointeeType().getTypePtr(); 3827 else if (EltType->isArrayType()) 3828 return EltType->getBaseElementTypeUnsafe(); 3829 return EltType; 3830} 3831 3832/// \brief Check whether this array fits the idiom of a size-one tail padded 3833/// array member of a struct. 3834/// 3835/// We avoid emitting out-of-bounds access warnings for such arrays as they are 3836/// commonly used to emulate flexible arrays in C89 code. 3837static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 3838 const NamedDecl *ND) { 3839 if (Size != 1 || !ND) return false; 3840 3841 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 3842 if (!FD) return false; 3843 3844 // Don't consider sizes resulting from macro expansions or template argument 3845 // substitution to form C89 tail-padded arrays. 3846 ConstantArrayTypeLoc TL = 3847 cast<ConstantArrayTypeLoc>(FD->getTypeSourceInfo()->getTypeLoc()); 3848 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(TL.getSizeExpr()); 3849 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 3850 return false; 3851 3852 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 3853 if (!RD || !RD->isStruct()) 3854 return false; 3855 3856 // See if this is the last field decl in the record. 3857 const Decl *D = FD; 3858 while ((D = D->getNextDeclInContext())) 3859 if (isa<FieldDecl>(D)) 3860 return false; 3861 return true; 3862} 3863 3864void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 3865 bool isSubscript, bool AllowOnePastEnd) { 3866 const Type* EffectiveType = getElementType(BaseExpr); 3867 BaseExpr = BaseExpr->IgnoreParenCasts(); 3868 IndexExpr = IndexExpr->IgnoreParenCasts(); 3869 3870 const ConstantArrayType *ArrayTy = 3871 Context.getAsConstantArrayType(BaseExpr->getType()); 3872 if (!ArrayTy) 3873 return; 3874 3875 if (IndexExpr->isValueDependent()) 3876 return; 3877 llvm::APSInt index; 3878 if (!IndexExpr->isIntegerConstantExpr(index, Context)) 3879 return; 3880 3881 const NamedDecl *ND = NULL; 3882 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 3883 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 3884 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 3885 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 3886 3887 if (index.isUnsigned() || !index.isNegative()) { 3888 llvm::APInt size = ArrayTy->getSize(); 3889 if (!size.isStrictlyPositive()) 3890 return; 3891 3892 const Type* BaseType = getElementType(BaseExpr); 3893 if (BaseType != EffectiveType) { 3894 // Make sure we're comparing apples to apples when comparing index to size 3895 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 3896 uint64_t array_typesize = Context.getTypeSize(BaseType); 3897 // Handle ptrarith_typesize being zero, such as when casting to void* 3898 if (!ptrarith_typesize) ptrarith_typesize = 1; 3899 if (ptrarith_typesize != array_typesize) { 3900 // There's a cast to a different size type involved 3901 uint64_t ratio = array_typesize / ptrarith_typesize; 3902 // TODO: Be smarter about handling cases where array_typesize is not a 3903 // multiple of ptrarith_typesize 3904 if (ptrarith_typesize * ratio == array_typesize) 3905 size *= llvm::APInt(size.getBitWidth(), ratio); 3906 } 3907 } 3908 3909 if (size.getBitWidth() > index.getBitWidth()) 3910 index = index.sext(size.getBitWidth()); 3911 else if (size.getBitWidth() < index.getBitWidth()) 3912 size = size.sext(index.getBitWidth()); 3913 3914 // For array subscripting the index must be less than size, but for pointer 3915 // arithmetic also allow the index (offset) to be equal to size since 3916 // computing the next address after the end of the array is legal and 3917 // commonly done e.g. in C++ iterators and range-based for loops. 3918 if (AllowOnePastEnd ? index.sle(size) : index.slt(size)) 3919 return; 3920 3921 // Also don't warn for arrays of size 1 which are members of some 3922 // structure. These are often used to approximate flexible arrays in C89 3923 // code. 3924 if (IsTailPaddedMemberArray(*this, size, ND)) 3925 return; 3926 3927 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 3928 if (isSubscript) 3929 DiagID = diag::warn_array_index_exceeds_bounds; 3930 3931 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 3932 PDiag(DiagID) << index.toString(10, true) 3933 << size.toString(10, true) 3934 << (unsigned)size.getLimitedValue(~0U) 3935 << IndexExpr->getSourceRange()); 3936 } else { 3937 unsigned DiagID = diag::warn_array_index_precedes_bounds; 3938 if (!isSubscript) { 3939 DiagID = diag::warn_ptr_arith_precedes_bounds; 3940 if (index.isNegative()) index = -index; 3941 } 3942 3943 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 3944 PDiag(DiagID) << index.toString(10, true) 3945 << IndexExpr->getSourceRange()); 3946 } 3947 3948 if (ND) 3949 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 3950 PDiag(diag::note_array_index_out_of_bounds) 3951 << ND->getDeclName()); 3952} 3953 3954void Sema::CheckArrayAccess(const Expr *expr) { 3955 int AllowOnePastEnd = 0; 3956 while (expr) { 3957 expr = expr->IgnoreParenImpCasts(); 3958 switch (expr->getStmtClass()) { 3959 case Stmt::ArraySubscriptExprClass: { 3960 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 3961 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), true, 3962 AllowOnePastEnd > 0); 3963 return; 3964 } 3965 case Stmt::UnaryOperatorClass: { 3966 // Only unwrap the * and & unary operators 3967 const UnaryOperator *UO = cast<UnaryOperator>(expr); 3968 expr = UO->getSubExpr(); 3969 switch (UO->getOpcode()) { 3970 case UO_AddrOf: 3971 AllowOnePastEnd++; 3972 break; 3973 case UO_Deref: 3974 AllowOnePastEnd--; 3975 break; 3976 default: 3977 return; 3978 } 3979 break; 3980 } 3981 case Stmt::ConditionalOperatorClass: { 3982 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 3983 if (const Expr *lhs = cond->getLHS()) 3984 CheckArrayAccess(lhs); 3985 if (const Expr *rhs = cond->getRHS()) 3986 CheckArrayAccess(rhs); 3987 return; 3988 } 3989 default: 3990 return; 3991 } 3992 } 3993} 3994 3995//===--- CHECK: Objective-C retain cycles ----------------------------------// 3996 3997namespace { 3998 struct RetainCycleOwner { 3999 RetainCycleOwner() : Variable(0), Indirect(false) {} 4000 VarDecl *Variable; 4001 SourceRange Range; 4002 SourceLocation Loc; 4003 bool Indirect; 4004 4005 void setLocsFrom(Expr *e) { 4006 Loc = e->getExprLoc(); 4007 Range = e->getSourceRange(); 4008 } 4009 }; 4010} 4011 4012/// Consider whether capturing the given variable can possibly lead to 4013/// a retain cycle. 4014static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 4015 // In ARC, it's captured strongly iff the variable has __strong 4016 // lifetime. In MRR, it's captured strongly if the variable is 4017 // __block and has an appropriate type. 4018 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4019 return false; 4020 4021 owner.Variable = var; 4022 owner.setLocsFrom(ref); 4023 return true; 4024} 4025 4026static bool findRetainCycleOwner(Expr *e, RetainCycleOwner &owner) { 4027 while (true) { 4028 e = e->IgnoreParens(); 4029 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 4030 switch (cast->getCastKind()) { 4031 case CK_BitCast: 4032 case CK_LValueBitCast: 4033 case CK_LValueToRValue: 4034 case CK_ARCReclaimReturnedObject: 4035 e = cast->getSubExpr(); 4036 continue; 4037 4038 case CK_GetObjCProperty: { 4039 // Bail out if this isn't a strong explicit property. 4040 const ObjCPropertyRefExpr *pre = cast->getSubExpr()->getObjCProperty(); 4041 if (pre->isImplicitProperty()) return false; 4042 ObjCPropertyDecl *property = pre->getExplicitProperty(); 4043 if (!property->isRetaining() && 4044 !(property->getPropertyIvarDecl() && 4045 property->getPropertyIvarDecl()->getType() 4046 .getObjCLifetime() == Qualifiers::OCL_Strong)) 4047 return false; 4048 4049 owner.Indirect = true; 4050 e = const_cast<Expr*>(pre->getBase()); 4051 continue; 4052 } 4053 4054 default: 4055 return false; 4056 } 4057 } 4058 4059 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 4060 ObjCIvarDecl *ivar = ref->getDecl(); 4061 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4062 return false; 4063 4064 // Try to find a retain cycle in the base. 4065 if (!findRetainCycleOwner(ref->getBase(), owner)) 4066 return false; 4067 4068 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 4069 owner.Indirect = true; 4070 return true; 4071 } 4072 4073 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 4074 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 4075 if (!var) return false; 4076 return considerVariable(var, ref, owner); 4077 } 4078 4079 if (BlockDeclRefExpr *ref = dyn_cast<BlockDeclRefExpr>(e)) { 4080 owner.Variable = ref->getDecl(); 4081 owner.setLocsFrom(ref); 4082 return true; 4083 } 4084 4085 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 4086 if (member->isArrow()) return false; 4087 4088 // Don't count this as an indirect ownership. 4089 e = member->getBase(); 4090 continue; 4091 } 4092 4093 // Array ivars? 4094 4095 return false; 4096 } 4097} 4098 4099namespace { 4100 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 4101 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 4102 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 4103 Variable(variable), Capturer(0) {} 4104 4105 VarDecl *Variable; 4106 Expr *Capturer; 4107 4108 void VisitDeclRefExpr(DeclRefExpr *ref) { 4109 if (ref->getDecl() == Variable && !Capturer) 4110 Capturer = ref; 4111 } 4112 4113 void VisitBlockDeclRefExpr(BlockDeclRefExpr *ref) { 4114 if (ref->getDecl() == Variable && !Capturer) 4115 Capturer = ref; 4116 } 4117 4118 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 4119 if (Capturer) return; 4120 Visit(ref->getBase()); 4121 if (Capturer && ref->isFreeIvar()) 4122 Capturer = ref; 4123 } 4124 4125 void VisitBlockExpr(BlockExpr *block) { 4126 // Look inside nested blocks 4127 if (block->getBlockDecl()->capturesVariable(Variable)) 4128 Visit(block->getBlockDecl()->getBody()); 4129 } 4130 }; 4131} 4132 4133/// Check whether the given argument is a block which captures a 4134/// variable. 4135static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 4136 assert(owner.Variable && owner.Loc.isValid()); 4137 4138 e = e->IgnoreParenCasts(); 4139 BlockExpr *block = dyn_cast<BlockExpr>(e); 4140 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 4141 return 0; 4142 4143 FindCaptureVisitor visitor(S.Context, owner.Variable); 4144 visitor.Visit(block->getBlockDecl()->getBody()); 4145 return visitor.Capturer; 4146} 4147 4148static void diagnoseRetainCycle(Sema &S, Expr *capturer, 4149 RetainCycleOwner &owner) { 4150 assert(capturer); 4151 assert(owner.Variable && owner.Loc.isValid()); 4152 4153 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 4154 << owner.Variable << capturer->getSourceRange(); 4155 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 4156 << owner.Indirect << owner.Range; 4157} 4158 4159/// Check for a keyword selector that starts with the word 'add' or 4160/// 'set'. 4161static bool isSetterLikeSelector(Selector sel) { 4162 if (sel.isUnarySelector()) return false; 4163 4164 StringRef str = sel.getNameForSlot(0); 4165 while (!str.empty() && str.front() == '_') str = str.substr(1); 4166 if (str.startswith("set") || str.startswith("add")) 4167 str = str.substr(3); 4168 else 4169 return false; 4170 4171 if (str.empty()) return true; 4172 return !islower(str.front()); 4173} 4174 4175/// Check a message send to see if it's likely to cause a retain cycle. 4176void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 4177 // Only check instance methods whose selector looks like a setter. 4178 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 4179 return; 4180 4181 // Try to find a variable that the receiver is strongly owned by. 4182 RetainCycleOwner owner; 4183 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 4184 if (!findRetainCycleOwner(msg->getInstanceReceiver(), owner)) 4185 return; 4186 } else { 4187 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 4188 owner.Variable = getCurMethodDecl()->getSelfDecl(); 4189 owner.Loc = msg->getSuperLoc(); 4190 owner.Range = msg->getSuperLoc(); 4191 } 4192 4193 // Check whether the receiver is captured by any of the arguments. 4194 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 4195 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 4196 return diagnoseRetainCycle(*this, capturer, owner); 4197} 4198 4199/// Check a property assign to see if it's likely to cause a retain cycle. 4200void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 4201 RetainCycleOwner owner; 4202 if (!findRetainCycleOwner(receiver, owner)) 4203 return; 4204 4205 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 4206 diagnoseRetainCycle(*this, capturer, owner); 4207} 4208 4209bool Sema::checkUnsafeAssigns(SourceLocation Loc, 4210 QualType LHS, Expr *RHS) { 4211 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 4212 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 4213 return false; 4214 // strip off any implicit cast added to get to the one arc-specific 4215 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4216 if (cast->getCastKind() == CK_ARCConsumeObject) { 4217 Diag(Loc, diag::warn_arc_retained_assign) 4218 << (LT == Qualifiers::OCL_ExplicitNone) 4219 << RHS->getSourceRange(); 4220 return true; 4221 } 4222 RHS = cast->getSubExpr(); 4223 } 4224 return false; 4225} 4226 4227void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 4228 Expr *LHS, Expr *RHS) { 4229 QualType LHSType = LHS->getType(); 4230 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 4231 return; 4232 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 4233 // FIXME. Check for other life times. 4234 if (LT != Qualifiers::OCL_None) 4235 return; 4236 4237 if (ObjCPropertyRefExpr *PRE = dyn_cast<ObjCPropertyRefExpr>(LHS)) { 4238 if (PRE->isImplicitProperty()) 4239 return; 4240 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 4241 if (!PD) 4242 return; 4243 4244 unsigned Attributes = PD->getPropertyAttributes(); 4245 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) 4246 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4247 if (cast->getCastKind() == CK_ARCConsumeObject) { 4248 Diag(Loc, diag::warn_arc_retained_property_assign) 4249 << RHS->getSourceRange(); 4250 return; 4251 } 4252 RHS = cast->getSubExpr(); 4253 } 4254 } 4255} 4256