SemaChecking.cpp revision 0692a1991bdb8b3112887fce90d75f2103a0080b
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/// Checks that a call expression's argument count is the desired number. 49/// This is useful when doing custom type-checking. Returns true on error. 50static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) { 51 unsigned argCount = call->getNumArgs(); 52 if (argCount == desiredArgCount) return false; 53 54 if (argCount < desiredArgCount) 55 return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args) 56 << 0 /*function call*/ << desiredArgCount << argCount 57 << call->getSourceRange(); 58 59 // Highlight all the excess arguments. 60 SourceRange range(call->getArg(desiredArgCount)->getLocStart(), 61 call->getArg(argCount - 1)->getLocEnd()); 62 63 return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args) 64 << 0 /*function call*/ << desiredArgCount << argCount 65 << call->getArg(1)->getSourceRange(); 66} 67 68/// CheckBuiltinAnnotationString - Checks that string argument to the builtin 69/// annotation is a non wide string literal. 70static bool CheckBuiltinAnnotationString(Sema &S, Expr *Arg) { 71 Arg = Arg->IgnoreParenCasts(); 72 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 73 if (!Literal || !Literal->isAscii()) { 74 S.Diag(Arg->getLocStart(), diag::err_builtin_annotation_not_string_constant) 75 << Arg->getSourceRange(); 76 return true; 77 } 78 return false; 79} 80 81ExprResult 82Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 83 ExprResult TheCallResult(Owned(TheCall)); 84 85 // Find out if any arguments are required to be integer constant expressions. 86 unsigned ICEArguments = 0; 87 ASTContext::GetBuiltinTypeError Error; 88 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments); 89 if (Error != ASTContext::GE_None) 90 ICEArguments = 0; // Don't diagnose previously diagnosed errors. 91 92 // If any arguments are required to be ICE's, check and diagnose. 93 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) { 94 // Skip arguments not required to be ICE's. 95 if ((ICEArguments & (1 << ArgNo)) == 0) continue; 96 97 llvm::APSInt Result; 98 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 99 return true; 100 ICEArguments &= ~(1 << ArgNo); 101 } 102 103 switch (BuiltinID) { 104 case Builtin::BI__builtin___CFStringMakeConstantString: 105 assert(TheCall->getNumArgs() == 1 && 106 "Wrong # arguments to builtin CFStringMakeConstantString"); 107 if (CheckObjCString(TheCall->getArg(0))) 108 return ExprError(); 109 break; 110 case Builtin::BI__builtin_stdarg_start: 111 case Builtin::BI__builtin_va_start: 112 if (SemaBuiltinVAStart(TheCall)) 113 return ExprError(); 114 break; 115 case Builtin::BI__builtin_isgreater: 116 case Builtin::BI__builtin_isgreaterequal: 117 case Builtin::BI__builtin_isless: 118 case Builtin::BI__builtin_islessequal: 119 case Builtin::BI__builtin_islessgreater: 120 case Builtin::BI__builtin_isunordered: 121 if (SemaBuiltinUnorderedCompare(TheCall)) 122 return ExprError(); 123 break; 124 case Builtin::BI__builtin_fpclassify: 125 if (SemaBuiltinFPClassification(TheCall, 6)) 126 return ExprError(); 127 break; 128 case Builtin::BI__builtin_isfinite: 129 case Builtin::BI__builtin_isinf: 130 case Builtin::BI__builtin_isinf_sign: 131 case Builtin::BI__builtin_isnan: 132 case Builtin::BI__builtin_isnormal: 133 if (SemaBuiltinFPClassification(TheCall, 1)) 134 return ExprError(); 135 break; 136 case Builtin::BI__builtin_shufflevector: 137 return SemaBuiltinShuffleVector(TheCall); 138 // TheCall will be freed by the smart pointer here, but that's fine, since 139 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 140 case Builtin::BI__builtin_prefetch: 141 if (SemaBuiltinPrefetch(TheCall)) 142 return ExprError(); 143 break; 144 case Builtin::BI__builtin_object_size: 145 if (SemaBuiltinObjectSize(TheCall)) 146 return ExprError(); 147 break; 148 case Builtin::BI__builtin_longjmp: 149 if (SemaBuiltinLongjmp(TheCall)) 150 return ExprError(); 151 break; 152 153 case Builtin::BI__builtin_classify_type: 154 if (checkArgCount(*this, TheCall, 1)) return true; 155 TheCall->setType(Context.IntTy); 156 break; 157 case Builtin::BI__builtin_constant_p: 158 if (checkArgCount(*this, TheCall, 1)) return true; 159 TheCall->setType(Context.IntTy); 160 break; 161 case Builtin::BI__sync_fetch_and_add: 162 case Builtin::BI__sync_fetch_and_add_1: 163 case Builtin::BI__sync_fetch_and_add_2: 164 case Builtin::BI__sync_fetch_and_add_4: 165 case Builtin::BI__sync_fetch_and_add_8: 166 case Builtin::BI__sync_fetch_and_add_16: 167 case Builtin::BI__sync_fetch_and_sub: 168 case Builtin::BI__sync_fetch_and_sub_1: 169 case Builtin::BI__sync_fetch_and_sub_2: 170 case Builtin::BI__sync_fetch_and_sub_4: 171 case Builtin::BI__sync_fetch_and_sub_8: 172 case Builtin::BI__sync_fetch_and_sub_16: 173 case Builtin::BI__sync_fetch_and_or: 174 case Builtin::BI__sync_fetch_and_or_1: 175 case Builtin::BI__sync_fetch_and_or_2: 176 case Builtin::BI__sync_fetch_and_or_4: 177 case Builtin::BI__sync_fetch_and_or_8: 178 case Builtin::BI__sync_fetch_and_or_16: 179 case Builtin::BI__sync_fetch_and_and: 180 case Builtin::BI__sync_fetch_and_and_1: 181 case Builtin::BI__sync_fetch_and_and_2: 182 case Builtin::BI__sync_fetch_and_and_4: 183 case Builtin::BI__sync_fetch_and_and_8: 184 case Builtin::BI__sync_fetch_and_and_16: 185 case Builtin::BI__sync_fetch_and_xor: 186 case Builtin::BI__sync_fetch_and_xor_1: 187 case Builtin::BI__sync_fetch_and_xor_2: 188 case Builtin::BI__sync_fetch_and_xor_4: 189 case Builtin::BI__sync_fetch_and_xor_8: 190 case Builtin::BI__sync_fetch_and_xor_16: 191 case Builtin::BI__sync_add_and_fetch: 192 case Builtin::BI__sync_add_and_fetch_1: 193 case Builtin::BI__sync_add_and_fetch_2: 194 case Builtin::BI__sync_add_and_fetch_4: 195 case Builtin::BI__sync_add_and_fetch_8: 196 case Builtin::BI__sync_add_and_fetch_16: 197 case Builtin::BI__sync_sub_and_fetch: 198 case Builtin::BI__sync_sub_and_fetch_1: 199 case Builtin::BI__sync_sub_and_fetch_2: 200 case Builtin::BI__sync_sub_and_fetch_4: 201 case Builtin::BI__sync_sub_and_fetch_8: 202 case Builtin::BI__sync_sub_and_fetch_16: 203 case Builtin::BI__sync_and_and_fetch: 204 case Builtin::BI__sync_and_and_fetch_1: 205 case Builtin::BI__sync_and_and_fetch_2: 206 case Builtin::BI__sync_and_and_fetch_4: 207 case Builtin::BI__sync_and_and_fetch_8: 208 case Builtin::BI__sync_and_and_fetch_16: 209 case Builtin::BI__sync_or_and_fetch: 210 case Builtin::BI__sync_or_and_fetch_1: 211 case Builtin::BI__sync_or_and_fetch_2: 212 case Builtin::BI__sync_or_and_fetch_4: 213 case Builtin::BI__sync_or_and_fetch_8: 214 case Builtin::BI__sync_or_and_fetch_16: 215 case Builtin::BI__sync_xor_and_fetch: 216 case Builtin::BI__sync_xor_and_fetch_1: 217 case Builtin::BI__sync_xor_and_fetch_2: 218 case Builtin::BI__sync_xor_and_fetch_4: 219 case Builtin::BI__sync_xor_and_fetch_8: 220 case Builtin::BI__sync_xor_and_fetch_16: 221 case Builtin::BI__sync_val_compare_and_swap: 222 case Builtin::BI__sync_val_compare_and_swap_1: 223 case Builtin::BI__sync_val_compare_and_swap_2: 224 case Builtin::BI__sync_val_compare_and_swap_4: 225 case Builtin::BI__sync_val_compare_and_swap_8: 226 case Builtin::BI__sync_val_compare_and_swap_16: 227 case Builtin::BI__sync_bool_compare_and_swap: 228 case Builtin::BI__sync_bool_compare_and_swap_1: 229 case Builtin::BI__sync_bool_compare_and_swap_2: 230 case Builtin::BI__sync_bool_compare_and_swap_4: 231 case Builtin::BI__sync_bool_compare_and_swap_8: 232 case Builtin::BI__sync_bool_compare_and_swap_16: 233 case Builtin::BI__sync_lock_test_and_set: 234 case Builtin::BI__sync_lock_test_and_set_1: 235 case Builtin::BI__sync_lock_test_and_set_2: 236 case Builtin::BI__sync_lock_test_and_set_4: 237 case Builtin::BI__sync_lock_test_and_set_8: 238 case Builtin::BI__sync_lock_test_and_set_16: 239 case Builtin::BI__sync_lock_release: 240 case Builtin::BI__sync_lock_release_1: 241 case Builtin::BI__sync_lock_release_2: 242 case Builtin::BI__sync_lock_release_4: 243 case Builtin::BI__sync_lock_release_8: 244 case Builtin::BI__sync_lock_release_16: 245 case Builtin::BI__sync_swap: 246 case Builtin::BI__sync_swap_1: 247 case Builtin::BI__sync_swap_2: 248 case Builtin::BI__sync_swap_4: 249 case Builtin::BI__sync_swap_8: 250 case Builtin::BI__sync_swap_16: 251 return SemaBuiltinAtomicOverloaded(move(TheCallResult)); 252 case Builtin::BI__atomic_load: 253 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Load); 254 case Builtin::BI__atomic_store: 255 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Store); 256 case Builtin::BI__atomic_init: 257 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Init); 258 case Builtin::BI__atomic_exchange: 259 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Xchg); 260 case Builtin::BI__atomic_compare_exchange_strong: 261 return SemaAtomicOpsOverloaded(move(TheCallResult), 262 AtomicExpr::CmpXchgStrong); 263 case Builtin::BI__atomic_compare_exchange_weak: 264 return SemaAtomicOpsOverloaded(move(TheCallResult), 265 AtomicExpr::CmpXchgWeak); 266 case Builtin::BI__atomic_fetch_add: 267 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Add); 268 case Builtin::BI__atomic_fetch_sub: 269 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Sub); 270 case Builtin::BI__atomic_fetch_and: 271 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::And); 272 case Builtin::BI__atomic_fetch_or: 273 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Or); 274 case Builtin::BI__atomic_fetch_xor: 275 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Xor); 276 case Builtin::BI__builtin_annotation: 277 if (CheckBuiltinAnnotationString(*this, TheCall->getArg(1))) 278 return ExprError(); 279 break; 280 } 281 282 // Since the target specific builtins for each arch overlap, only check those 283 // of the arch we are compiling for. 284 if (BuiltinID >= Builtin::FirstTSBuiltin) { 285 switch (Context.getTargetInfo().getTriple().getArch()) { 286 case llvm::Triple::arm: 287 case llvm::Triple::thumb: 288 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 289 return ExprError(); 290 break; 291 default: 292 break; 293 } 294 } 295 296 return move(TheCallResult); 297} 298 299// Get the valid immediate range for the specified NEON type code. 300static unsigned RFT(unsigned t, bool shift = false) { 301 NeonTypeFlags Type(t); 302 int IsQuad = Type.isQuad(); 303 switch (Type.getEltType()) { 304 case NeonTypeFlags::Int8: 305 case NeonTypeFlags::Poly8: 306 return shift ? 7 : (8 << IsQuad) - 1; 307 case NeonTypeFlags::Int16: 308 case NeonTypeFlags::Poly16: 309 return shift ? 15 : (4 << IsQuad) - 1; 310 case NeonTypeFlags::Int32: 311 return shift ? 31 : (2 << IsQuad) - 1; 312 case NeonTypeFlags::Int64: 313 return shift ? 63 : (1 << IsQuad) - 1; 314 case NeonTypeFlags::Float16: 315 assert(!shift && "cannot shift float types!"); 316 return (4 << IsQuad) - 1; 317 case NeonTypeFlags::Float32: 318 assert(!shift && "cannot shift float types!"); 319 return (2 << IsQuad) - 1; 320 } 321 llvm_unreachable("Invalid NeonTypeFlag!"); 322} 323 324/// getNeonEltType - Return the QualType corresponding to the elements of 325/// the vector type specified by the NeonTypeFlags. This is used to check 326/// the pointer arguments for Neon load/store intrinsics. 327static QualType getNeonEltType(NeonTypeFlags Flags, ASTContext &Context) { 328 switch (Flags.getEltType()) { 329 case NeonTypeFlags::Int8: 330 return Flags.isUnsigned() ? Context.UnsignedCharTy : Context.SignedCharTy; 331 case NeonTypeFlags::Int16: 332 return Flags.isUnsigned() ? Context.UnsignedShortTy : Context.ShortTy; 333 case NeonTypeFlags::Int32: 334 return Flags.isUnsigned() ? Context.UnsignedIntTy : Context.IntTy; 335 case NeonTypeFlags::Int64: 336 return Flags.isUnsigned() ? Context.UnsignedLongLongTy : Context.LongLongTy; 337 case NeonTypeFlags::Poly8: 338 return Context.SignedCharTy; 339 case NeonTypeFlags::Poly16: 340 return Context.ShortTy; 341 case NeonTypeFlags::Float16: 342 return Context.UnsignedShortTy; 343 case NeonTypeFlags::Float32: 344 return Context.FloatTy; 345 } 346 llvm_unreachable("Invalid NeonTypeFlag!"); 347} 348 349bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 350 llvm::APSInt Result; 351 352 unsigned mask = 0; 353 unsigned TV = 0; 354 int PtrArgNum = -1; 355 bool HasConstPtr = false; 356 switch (BuiltinID) { 357#define GET_NEON_OVERLOAD_CHECK 358#include "clang/Basic/arm_neon.inc" 359#undef GET_NEON_OVERLOAD_CHECK 360 } 361 362 // For NEON intrinsics which are overloaded on vector element type, validate 363 // the immediate which specifies which variant to emit. 364 unsigned ImmArg = TheCall->getNumArgs()-1; 365 if (mask) { 366 if (SemaBuiltinConstantArg(TheCall, ImmArg, Result)) 367 return true; 368 369 TV = Result.getLimitedValue(64); 370 if ((TV > 63) || (mask & (1 << TV)) == 0) 371 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 372 << TheCall->getArg(ImmArg)->getSourceRange(); 373 } 374 375 if (PtrArgNum >= 0) { 376 // Check that pointer arguments have the specified type. 377 Expr *Arg = TheCall->getArg(PtrArgNum); 378 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) 379 Arg = ICE->getSubExpr(); 380 ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg); 381 QualType RHSTy = RHS.get()->getType(); 382 QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context); 383 if (HasConstPtr) 384 EltTy = EltTy.withConst(); 385 QualType LHSTy = Context.getPointerType(EltTy); 386 AssignConvertType ConvTy; 387 ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS); 388 if (RHS.isInvalid()) 389 return true; 390 if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy, 391 RHS.get(), AA_Assigning)) 392 return true; 393 } 394 395 // For NEON intrinsics which take an immediate value as part of the 396 // instruction, range check them here. 397 unsigned i = 0, l = 0, u = 0; 398 switch (BuiltinID) { 399 default: return false; 400 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 401 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 402 case ARM::BI__builtin_arm_vcvtr_f: 403 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 404#define GET_NEON_IMMEDIATE_CHECK 405#include "clang/Basic/arm_neon.inc" 406#undef GET_NEON_IMMEDIATE_CHECK 407 }; 408 409 // Check that the immediate argument is actually a constant. 410 if (SemaBuiltinConstantArg(TheCall, i, Result)) 411 return true; 412 413 // Range check against the upper/lower values for this isntruction. 414 unsigned Val = Result.getZExtValue(); 415 if (Val < l || Val > (u + l)) 416 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 417 << l << u+l << TheCall->getArg(i)->getSourceRange(); 418 419 // FIXME: VFP Intrinsics should error if VFP not present. 420 return false; 421} 422 423/// CheckFunctionCall - Check a direct function call for various correctness 424/// and safety properties not strictly enforced by the C type system. 425bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 426 // Get the IdentifierInfo* for the called function. 427 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 428 429 // None of the checks below are needed for functions that don't have 430 // simple names (e.g., C++ conversion functions). 431 if (!FnInfo) 432 return false; 433 434 // FIXME: This mechanism should be abstracted to be less fragile and 435 // more efficient. For example, just map function ids to custom 436 // handlers. 437 438 // Printf and scanf checking. 439 for (specific_attr_iterator<FormatAttr> 440 i = FDecl->specific_attr_begin<FormatAttr>(), 441 e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { 442 CheckFormatArguments(*i, TheCall); 443 } 444 445 for (specific_attr_iterator<NonNullAttr> 446 i = FDecl->specific_attr_begin<NonNullAttr>(), 447 e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) { 448 CheckNonNullArguments(*i, TheCall->getArgs(), 449 TheCall->getCallee()->getLocStart()); 450 } 451 452 unsigned CMId = FDecl->getMemoryFunctionKind(); 453 if (CMId == 0) 454 return false; 455 456 // Handle memory setting and copying functions. 457 if (CMId == Builtin::BIstrlcpy || CMId == Builtin::BIstrlcat) 458 CheckStrlcpycatArguments(TheCall, FnInfo); 459 else 460 CheckMemaccessArguments(TheCall, CMId, FnInfo); 461 462 return false; 463} 464 465bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac, 466 Expr **Args, unsigned NumArgs) { 467 for (specific_attr_iterator<FormatAttr> 468 i = Method->specific_attr_begin<FormatAttr>(), 469 e = Method->specific_attr_end<FormatAttr>(); i != e ; ++i) { 470 471 CheckFormatArguments(*i, Args, NumArgs, false, lbrac, 472 Method->getSourceRange()); 473 } 474 475 // diagnose nonnull arguments. 476 for (specific_attr_iterator<NonNullAttr> 477 i = Method->specific_attr_begin<NonNullAttr>(), 478 e = Method->specific_attr_end<NonNullAttr>(); i != e; ++i) { 479 CheckNonNullArguments(*i, Args, lbrac); 480 } 481 482 return false; 483} 484 485bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 486 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 487 if (!V) 488 return false; 489 490 QualType Ty = V->getType(); 491 if (!Ty->isBlockPointerType()) 492 return false; 493 494 // format string checking. 495 for (specific_attr_iterator<FormatAttr> 496 i = NDecl->specific_attr_begin<FormatAttr>(), 497 e = NDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { 498 CheckFormatArguments(*i, TheCall); 499 } 500 501 return false; 502} 503 504ExprResult 505Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, AtomicExpr::AtomicOp Op) { 506 CallExpr *TheCall = cast<CallExpr>(TheCallResult.get()); 507 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 508 509 // All these operations take one of the following four forms: 510 // T __atomic_load(_Atomic(T)*, int) (loads) 511 // T* __atomic_add(_Atomic(T*)*, ptrdiff_t, int) (pointer add/sub) 512 // int __atomic_compare_exchange_strong(_Atomic(T)*, T*, T, int, int) 513 // (cmpxchg) 514 // T __atomic_exchange(_Atomic(T)*, T, int) (everything else) 515 // where T is an appropriate type, and the int paremeterss are for orderings. 516 unsigned NumVals = 1; 517 unsigned NumOrders = 1; 518 if (Op == AtomicExpr::Load) { 519 NumVals = 0; 520 } else if (Op == AtomicExpr::CmpXchgWeak || Op == AtomicExpr::CmpXchgStrong) { 521 NumVals = 2; 522 NumOrders = 2; 523 } 524 if (Op == AtomicExpr::Init) 525 NumOrders = 0; 526 527 if (TheCall->getNumArgs() < NumVals+NumOrders+1) { 528 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 529 << 0 << NumVals+NumOrders+1 << TheCall->getNumArgs() 530 << TheCall->getCallee()->getSourceRange(); 531 return ExprError(); 532 } else if (TheCall->getNumArgs() > NumVals+NumOrders+1) { 533 Diag(TheCall->getArg(NumVals+NumOrders+1)->getLocStart(), 534 diag::err_typecheck_call_too_many_args) 535 << 0 << NumVals+NumOrders+1 << TheCall->getNumArgs() 536 << TheCall->getCallee()->getSourceRange(); 537 return ExprError(); 538 } 539 540 // Inspect the first argument of the atomic operation. This should always be 541 // a pointer to an _Atomic type. 542 Expr *Ptr = TheCall->getArg(0); 543 Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get(); 544 const PointerType *pointerType = Ptr->getType()->getAs<PointerType>(); 545 if (!pointerType) { 546 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 547 << Ptr->getType() << Ptr->getSourceRange(); 548 return ExprError(); 549 } 550 551 QualType AtomTy = pointerType->getPointeeType(); 552 if (!AtomTy->isAtomicType()) { 553 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 554 << Ptr->getType() << Ptr->getSourceRange(); 555 return ExprError(); 556 } 557 QualType ValType = AtomTy->getAs<AtomicType>()->getValueType(); 558 559 if ((Op == AtomicExpr::Add || Op == AtomicExpr::Sub) && 560 !ValType->isIntegerType() && !ValType->isPointerType()) { 561 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 562 << Ptr->getType() << Ptr->getSourceRange(); 563 return ExprError(); 564 } 565 566 if (!ValType->isIntegerType() && 567 (Op == AtomicExpr::And || Op == AtomicExpr::Or || Op == AtomicExpr::Xor)){ 568 Diag(DRE->getLocStart(), diag::err_atomic_op_logical_needs_atomic_int) 569 << Ptr->getType() << Ptr->getSourceRange(); 570 return ExprError(); 571 } 572 573 switch (ValType.getObjCLifetime()) { 574 case Qualifiers::OCL_None: 575 case Qualifiers::OCL_ExplicitNone: 576 // okay 577 break; 578 579 case Qualifiers::OCL_Weak: 580 case Qualifiers::OCL_Strong: 581 case Qualifiers::OCL_Autoreleasing: 582 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 583 << ValType << Ptr->getSourceRange(); 584 return ExprError(); 585 } 586 587 QualType ResultType = ValType; 588 if (Op == AtomicExpr::Store || Op == AtomicExpr::Init) 589 ResultType = Context.VoidTy; 590 else if (Op == AtomicExpr::CmpXchgWeak || Op == AtomicExpr::CmpXchgStrong) 591 ResultType = Context.BoolTy; 592 593 // The first argument --- the pointer --- has a fixed type; we 594 // deduce the types of the rest of the arguments accordingly. Walk 595 // the remaining arguments, converting them to the deduced value type. 596 for (unsigned i = 1; i != NumVals+NumOrders+1; ++i) { 597 ExprResult Arg = TheCall->getArg(i); 598 QualType Ty; 599 if (i < NumVals+1) { 600 // The second argument to a cmpxchg is a pointer to the data which will 601 // be exchanged. The second argument to a pointer add/subtract is the 602 // amount to add/subtract, which must be a ptrdiff_t. The third 603 // argument to a cmpxchg and the second argument in all other cases 604 // is the type of the value. 605 if (i == 1 && (Op == AtomicExpr::CmpXchgWeak || 606 Op == AtomicExpr::CmpXchgStrong)) 607 Ty = Context.getPointerType(ValType.getUnqualifiedType()); 608 else if (!ValType->isIntegerType() && 609 (Op == AtomicExpr::Add || Op == AtomicExpr::Sub)) 610 Ty = Context.getPointerDiffType(); 611 else 612 Ty = ValType; 613 } else { 614 // The order(s) are always converted to int. 615 Ty = Context.IntTy; 616 } 617 InitializedEntity Entity = 618 InitializedEntity::InitializeParameter(Context, Ty, false); 619 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 620 if (Arg.isInvalid()) 621 return true; 622 TheCall->setArg(i, Arg.get()); 623 } 624 625 SmallVector<Expr*, 5> SubExprs; 626 SubExprs.push_back(Ptr); 627 if (Op == AtomicExpr::Load) { 628 SubExprs.push_back(TheCall->getArg(1)); // Order 629 } else if (Op == AtomicExpr::Init) { 630 SubExprs.push_back(TheCall->getArg(1)); // Val1 631 } else if (Op != AtomicExpr::CmpXchgWeak && Op != AtomicExpr::CmpXchgStrong) { 632 SubExprs.push_back(TheCall->getArg(2)); // Order 633 SubExprs.push_back(TheCall->getArg(1)); // Val1 634 } else { 635 SubExprs.push_back(TheCall->getArg(3)); // Order 636 SubExprs.push_back(TheCall->getArg(1)); // Val1 637 SubExprs.push_back(TheCall->getArg(2)); // Val2 638 SubExprs.push_back(TheCall->getArg(4)); // OrderFail 639 } 640 641 return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 642 SubExprs.data(), SubExprs.size(), 643 ResultType, Op, 644 TheCall->getRParenLoc())); 645} 646 647 648/// checkBuiltinArgument - Given a call to a builtin function, perform 649/// normal type-checking on the given argument, updating the call in 650/// place. This is useful when a builtin function requires custom 651/// type-checking for some of its arguments but not necessarily all of 652/// them. 653/// 654/// Returns true on error. 655static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 656 FunctionDecl *Fn = E->getDirectCallee(); 657 assert(Fn && "builtin call without direct callee!"); 658 659 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 660 InitializedEntity Entity = 661 InitializedEntity::InitializeParameter(S.Context, Param); 662 663 ExprResult Arg = E->getArg(0); 664 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 665 if (Arg.isInvalid()) 666 return true; 667 668 E->setArg(ArgIndex, Arg.take()); 669 return false; 670} 671 672/// SemaBuiltinAtomicOverloaded - We have a call to a function like 673/// __sync_fetch_and_add, which is an overloaded function based on the pointer 674/// type of its first argument. The main ActOnCallExpr routines have already 675/// promoted the types of arguments because all of these calls are prototyped as 676/// void(...). 677/// 678/// This function goes through and does final semantic checking for these 679/// builtins, 680ExprResult 681Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 682 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 683 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 684 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 685 686 // Ensure that we have at least one argument to do type inference from. 687 if (TheCall->getNumArgs() < 1) { 688 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 689 << 0 << 1 << TheCall->getNumArgs() 690 << TheCall->getCallee()->getSourceRange(); 691 return ExprError(); 692 } 693 694 // Inspect the first argument of the atomic builtin. This should always be 695 // a pointer type, whose element is an integral scalar or pointer type. 696 // Because it is a pointer type, we don't have to worry about any implicit 697 // casts here. 698 // FIXME: We don't allow floating point scalars as input. 699 Expr *FirstArg = TheCall->getArg(0); 700 ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg); 701 if (FirstArgResult.isInvalid()) 702 return ExprError(); 703 FirstArg = FirstArgResult.take(); 704 TheCall->setArg(0, FirstArg); 705 706 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 707 if (!pointerType) { 708 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 709 << FirstArg->getType() << FirstArg->getSourceRange(); 710 return ExprError(); 711 } 712 713 QualType ValType = pointerType->getPointeeType(); 714 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 715 !ValType->isBlockPointerType()) { 716 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 717 << FirstArg->getType() << FirstArg->getSourceRange(); 718 return ExprError(); 719 } 720 721 switch (ValType.getObjCLifetime()) { 722 case Qualifiers::OCL_None: 723 case Qualifiers::OCL_ExplicitNone: 724 // okay 725 break; 726 727 case Qualifiers::OCL_Weak: 728 case Qualifiers::OCL_Strong: 729 case Qualifiers::OCL_Autoreleasing: 730 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 731 << ValType << FirstArg->getSourceRange(); 732 return ExprError(); 733 } 734 735 // Strip any qualifiers off ValType. 736 ValType = ValType.getUnqualifiedType(); 737 738 // The majority of builtins return a value, but a few have special return 739 // types, so allow them to override appropriately below. 740 QualType ResultType = ValType; 741 742 // We need to figure out which concrete builtin this maps onto. For example, 743 // __sync_fetch_and_add with a 2 byte object turns into 744 // __sync_fetch_and_add_2. 745#define BUILTIN_ROW(x) \ 746 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 747 Builtin::BI##x##_8, Builtin::BI##x##_16 } 748 749 static const unsigned BuiltinIndices[][5] = { 750 BUILTIN_ROW(__sync_fetch_and_add), 751 BUILTIN_ROW(__sync_fetch_and_sub), 752 BUILTIN_ROW(__sync_fetch_and_or), 753 BUILTIN_ROW(__sync_fetch_and_and), 754 BUILTIN_ROW(__sync_fetch_and_xor), 755 756 BUILTIN_ROW(__sync_add_and_fetch), 757 BUILTIN_ROW(__sync_sub_and_fetch), 758 BUILTIN_ROW(__sync_and_and_fetch), 759 BUILTIN_ROW(__sync_or_and_fetch), 760 BUILTIN_ROW(__sync_xor_and_fetch), 761 762 BUILTIN_ROW(__sync_val_compare_and_swap), 763 BUILTIN_ROW(__sync_bool_compare_and_swap), 764 BUILTIN_ROW(__sync_lock_test_and_set), 765 BUILTIN_ROW(__sync_lock_release), 766 BUILTIN_ROW(__sync_swap) 767 }; 768#undef BUILTIN_ROW 769 770 // Determine the index of the size. 771 unsigned SizeIndex; 772 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 773 case 1: SizeIndex = 0; break; 774 case 2: SizeIndex = 1; break; 775 case 4: SizeIndex = 2; break; 776 case 8: SizeIndex = 3; break; 777 case 16: SizeIndex = 4; break; 778 default: 779 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 780 << FirstArg->getType() << FirstArg->getSourceRange(); 781 return ExprError(); 782 } 783 784 // Each of these builtins has one pointer argument, followed by some number of 785 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 786 // that we ignore. Find out which row of BuiltinIndices to read from as well 787 // as the number of fixed args. 788 unsigned BuiltinID = FDecl->getBuiltinID(); 789 unsigned BuiltinIndex, NumFixed = 1; 790 switch (BuiltinID) { 791 default: llvm_unreachable("Unknown overloaded atomic builtin!"); 792 case Builtin::BI__sync_fetch_and_add: 793 case Builtin::BI__sync_fetch_and_add_1: 794 case Builtin::BI__sync_fetch_and_add_2: 795 case Builtin::BI__sync_fetch_and_add_4: 796 case Builtin::BI__sync_fetch_and_add_8: 797 case Builtin::BI__sync_fetch_and_add_16: 798 BuiltinIndex = 0; 799 break; 800 801 case Builtin::BI__sync_fetch_and_sub: 802 case Builtin::BI__sync_fetch_and_sub_1: 803 case Builtin::BI__sync_fetch_and_sub_2: 804 case Builtin::BI__sync_fetch_and_sub_4: 805 case Builtin::BI__sync_fetch_and_sub_8: 806 case Builtin::BI__sync_fetch_and_sub_16: 807 BuiltinIndex = 1; 808 break; 809 810 case Builtin::BI__sync_fetch_and_or: 811 case Builtin::BI__sync_fetch_and_or_1: 812 case Builtin::BI__sync_fetch_and_or_2: 813 case Builtin::BI__sync_fetch_and_or_4: 814 case Builtin::BI__sync_fetch_and_or_8: 815 case Builtin::BI__sync_fetch_and_or_16: 816 BuiltinIndex = 2; 817 break; 818 819 case Builtin::BI__sync_fetch_and_and: 820 case Builtin::BI__sync_fetch_and_and_1: 821 case Builtin::BI__sync_fetch_and_and_2: 822 case Builtin::BI__sync_fetch_and_and_4: 823 case Builtin::BI__sync_fetch_and_and_8: 824 case Builtin::BI__sync_fetch_and_and_16: 825 BuiltinIndex = 3; 826 break; 827 828 case Builtin::BI__sync_fetch_and_xor: 829 case Builtin::BI__sync_fetch_and_xor_1: 830 case Builtin::BI__sync_fetch_and_xor_2: 831 case Builtin::BI__sync_fetch_and_xor_4: 832 case Builtin::BI__sync_fetch_and_xor_8: 833 case Builtin::BI__sync_fetch_and_xor_16: 834 BuiltinIndex = 4; 835 break; 836 837 case Builtin::BI__sync_add_and_fetch: 838 case Builtin::BI__sync_add_and_fetch_1: 839 case Builtin::BI__sync_add_and_fetch_2: 840 case Builtin::BI__sync_add_and_fetch_4: 841 case Builtin::BI__sync_add_and_fetch_8: 842 case Builtin::BI__sync_add_and_fetch_16: 843 BuiltinIndex = 5; 844 break; 845 846 case Builtin::BI__sync_sub_and_fetch: 847 case Builtin::BI__sync_sub_and_fetch_1: 848 case Builtin::BI__sync_sub_and_fetch_2: 849 case Builtin::BI__sync_sub_and_fetch_4: 850 case Builtin::BI__sync_sub_and_fetch_8: 851 case Builtin::BI__sync_sub_and_fetch_16: 852 BuiltinIndex = 6; 853 break; 854 855 case Builtin::BI__sync_and_and_fetch: 856 case Builtin::BI__sync_and_and_fetch_1: 857 case Builtin::BI__sync_and_and_fetch_2: 858 case Builtin::BI__sync_and_and_fetch_4: 859 case Builtin::BI__sync_and_and_fetch_8: 860 case Builtin::BI__sync_and_and_fetch_16: 861 BuiltinIndex = 7; 862 break; 863 864 case Builtin::BI__sync_or_and_fetch: 865 case Builtin::BI__sync_or_and_fetch_1: 866 case Builtin::BI__sync_or_and_fetch_2: 867 case Builtin::BI__sync_or_and_fetch_4: 868 case Builtin::BI__sync_or_and_fetch_8: 869 case Builtin::BI__sync_or_and_fetch_16: 870 BuiltinIndex = 8; 871 break; 872 873 case Builtin::BI__sync_xor_and_fetch: 874 case Builtin::BI__sync_xor_and_fetch_1: 875 case Builtin::BI__sync_xor_and_fetch_2: 876 case Builtin::BI__sync_xor_and_fetch_4: 877 case Builtin::BI__sync_xor_and_fetch_8: 878 case Builtin::BI__sync_xor_and_fetch_16: 879 BuiltinIndex = 9; 880 break; 881 882 case Builtin::BI__sync_val_compare_and_swap: 883 case Builtin::BI__sync_val_compare_and_swap_1: 884 case Builtin::BI__sync_val_compare_and_swap_2: 885 case Builtin::BI__sync_val_compare_and_swap_4: 886 case Builtin::BI__sync_val_compare_and_swap_8: 887 case Builtin::BI__sync_val_compare_and_swap_16: 888 BuiltinIndex = 10; 889 NumFixed = 2; 890 break; 891 892 case Builtin::BI__sync_bool_compare_and_swap: 893 case Builtin::BI__sync_bool_compare_and_swap_1: 894 case Builtin::BI__sync_bool_compare_and_swap_2: 895 case Builtin::BI__sync_bool_compare_and_swap_4: 896 case Builtin::BI__sync_bool_compare_and_swap_8: 897 case Builtin::BI__sync_bool_compare_and_swap_16: 898 BuiltinIndex = 11; 899 NumFixed = 2; 900 ResultType = Context.BoolTy; 901 break; 902 903 case Builtin::BI__sync_lock_test_and_set: 904 case Builtin::BI__sync_lock_test_and_set_1: 905 case Builtin::BI__sync_lock_test_and_set_2: 906 case Builtin::BI__sync_lock_test_and_set_4: 907 case Builtin::BI__sync_lock_test_and_set_8: 908 case Builtin::BI__sync_lock_test_and_set_16: 909 BuiltinIndex = 12; 910 break; 911 912 case Builtin::BI__sync_lock_release: 913 case Builtin::BI__sync_lock_release_1: 914 case Builtin::BI__sync_lock_release_2: 915 case Builtin::BI__sync_lock_release_4: 916 case Builtin::BI__sync_lock_release_8: 917 case Builtin::BI__sync_lock_release_16: 918 BuiltinIndex = 13; 919 NumFixed = 0; 920 ResultType = Context.VoidTy; 921 break; 922 923 case Builtin::BI__sync_swap: 924 case Builtin::BI__sync_swap_1: 925 case Builtin::BI__sync_swap_2: 926 case Builtin::BI__sync_swap_4: 927 case Builtin::BI__sync_swap_8: 928 case Builtin::BI__sync_swap_16: 929 BuiltinIndex = 14; 930 break; 931 } 932 933 // Now that we know how many fixed arguments we expect, first check that we 934 // have at least that many. 935 if (TheCall->getNumArgs() < 1+NumFixed) { 936 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 937 << 0 << 1+NumFixed << TheCall->getNumArgs() 938 << TheCall->getCallee()->getSourceRange(); 939 return ExprError(); 940 } 941 942 // Get the decl for the concrete builtin from this, we can tell what the 943 // concrete integer type we should convert to is. 944 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 945 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 946 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 947 FunctionDecl *NewBuiltinDecl = 948 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 949 TUScope, false, DRE->getLocStart())); 950 951 // The first argument --- the pointer --- has a fixed type; we 952 // deduce the types of the rest of the arguments accordingly. Walk 953 // the remaining arguments, converting them to the deduced value type. 954 for (unsigned i = 0; i != NumFixed; ++i) { 955 ExprResult Arg = TheCall->getArg(i+1); 956 957 // GCC does an implicit conversion to the pointer or integer ValType. This 958 // can fail in some cases (1i -> int**), check for this error case now. 959 // Initialize the argument. 960 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context, 961 ValType, /*consume*/ false); 962 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 963 if (Arg.isInvalid()) 964 return ExprError(); 965 966 // Okay, we have something that *can* be converted to the right type. Check 967 // to see if there is a potentially weird extension going on here. This can 968 // happen when you do an atomic operation on something like an char* and 969 // pass in 42. The 42 gets converted to char. This is even more strange 970 // for things like 45.123 -> char, etc. 971 // FIXME: Do this check. 972 TheCall->setArg(i+1, Arg.take()); 973 } 974 975 ASTContext& Context = this->getASTContext(); 976 977 // Create a new DeclRefExpr to refer to the new decl. 978 DeclRefExpr* NewDRE = DeclRefExpr::Create( 979 Context, 980 DRE->getQualifierLoc(), 981 SourceLocation(), 982 NewBuiltinDecl, 983 DRE->getLocation(), 984 NewBuiltinDecl->getType(), 985 DRE->getValueKind()); 986 987 // Set the callee in the CallExpr. 988 // FIXME: This leaks the original parens and implicit casts. 989 ExprResult PromotedCall = UsualUnaryConversions(NewDRE); 990 if (PromotedCall.isInvalid()) 991 return ExprError(); 992 TheCall->setCallee(PromotedCall.take()); 993 994 // Change the result type of the call to match the original value type. This 995 // is arbitrary, but the codegen for these builtins ins design to handle it 996 // gracefully. 997 TheCall->setType(ResultType); 998 999 return move(TheCallResult); 1000} 1001 1002/// CheckObjCString - Checks that the argument to the builtin 1003/// CFString constructor is correct 1004/// Note: It might also make sense to do the UTF-16 conversion here (would 1005/// simplify the backend). 1006bool Sema::CheckObjCString(Expr *Arg) { 1007 Arg = Arg->IgnoreParenCasts(); 1008 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 1009 1010 if (!Literal || !Literal->isAscii()) { 1011 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 1012 << Arg->getSourceRange(); 1013 return true; 1014 } 1015 1016 if (Literal->containsNonAsciiOrNull()) { 1017 StringRef String = Literal->getString(); 1018 unsigned NumBytes = String.size(); 1019 SmallVector<UTF16, 128> ToBuf(NumBytes); 1020 const UTF8 *FromPtr = (UTF8 *)String.data(); 1021 UTF16 *ToPtr = &ToBuf[0]; 1022 1023 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 1024 &ToPtr, ToPtr + NumBytes, 1025 strictConversion); 1026 // Check for conversion failure. 1027 if (Result != conversionOK) 1028 Diag(Arg->getLocStart(), 1029 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 1030 } 1031 return false; 1032} 1033 1034/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 1035/// Emit an error and return true on failure, return false on success. 1036bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 1037 Expr *Fn = TheCall->getCallee(); 1038 if (TheCall->getNumArgs() > 2) { 1039 Diag(TheCall->getArg(2)->getLocStart(), 1040 diag::err_typecheck_call_too_many_args) 1041 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1042 << Fn->getSourceRange() 1043 << SourceRange(TheCall->getArg(2)->getLocStart(), 1044 (*(TheCall->arg_end()-1))->getLocEnd()); 1045 return true; 1046 } 1047 1048 if (TheCall->getNumArgs() < 2) { 1049 return Diag(TheCall->getLocEnd(), 1050 diag::err_typecheck_call_too_few_args_at_least) 1051 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 1052 } 1053 1054 // Type-check the first argument normally. 1055 if (checkBuiltinArgument(*this, TheCall, 0)) 1056 return true; 1057 1058 // Determine whether the current function is variadic or not. 1059 BlockScopeInfo *CurBlock = getCurBlock(); 1060 bool isVariadic; 1061 if (CurBlock) 1062 isVariadic = CurBlock->TheDecl->isVariadic(); 1063 else if (FunctionDecl *FD = getCurFunctionDecl()) 1064 isVariadic = FD->isVariadic(); 1065 else 1066 isVariadic = getCurMethodDecl()->isVariadic(); 1067 1068 if (!isVariadic) { 1069 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 1070 return true; 1071 } 1072 1073 // Verify that the second argument to the builtin is the last argument of the 1074 // current function or method. 1075 bool SecondArgIsLastNamedArgument = false; 1076 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 1077 1078 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 1079 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 1080 // FIXME: This isn't correct for methods (results in bogus warning). 1081 // Get the last formal in the current function. 1082 const ParmVarDecl *LastArg; 1083 if (CurBlock) 1084 LastArg = *(CurBlock->TheDecl->param_end()-1); 1085 else if (FunctionDecl *FD = getCurFunctionDecl()) 1086 LastArg = *(FD->param_end()-1); 1087 else 1088 LastArg = *(getCurMethodDecl()->param_end()-1); 1089 SecondArgIsLastNamedArgument = PV == LastArg; 1090 } 1091 } 1092 1093 if (!SecondArgIsLastNamedArgument) 1094 Diag(TheCall->getArg(1)->getLocStart(), 1095 diag::warn_second_parameter_of_va_start_not_last_named_argument); 1096 return false; 1097} 1098 1099/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 1100/// friends. This is declared to take (...), so we have to check everything. 1101bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 1102 if (TheCall->getNumArgs() < 2) 1103 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1104 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 1105 if (TheCall->getNumArgs() > 2) 1106 return Diag(TheCall->getArg(2)->getLocStart(), 1107 diag::err_typecheck_call_too_many_args) 1108 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1109 << SourceRange(TheCall->getArg(2)->getLocStart(), 1110 (*(TheCall->arg_end()-1))->getLocEnd()); 1111 1112 ExprResult OrigArg0 = TheCall->getArg(0); 1113 ExprResult OrigArg1 = TheCall->getArg(1); 1114 1115 // Do standard promotions between the two arguments, returning their common 1116 // type. 1117 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 1118 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 1119 return true; 1120 1121 // Make sure any conversions are pushed back into the call; this is 1122 // type safe since unordered compare builtins are declared as "_Bool 1123 // foo(...)". 1124 TheCall->setArg(0, OrigArg0.get()); 1125 TheCall->setArg(1, OrigArg1.get()); 1126 1127 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 1128 return false; 1129 1130 // If the common type isn't a real floating type, then the arguments were 1131 // invalid for this operation. 1132 if (!Res->isRealFloatingType()) 1133 return Diag(OrigArg0.get()->getLocStart(), 1134 diag::err_typecheck_call_invalid_ordered_compare) 1135 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 1136 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 1137 1138 return false; 1139} 1140 1141/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 1142/// __builtin_isnan and friends. This is declared to take (...), so we have 1143/// to check everything. We expect the last argument to be a floating point 1144/// value. 1145bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 1146 if (TheCall->getNumArgs() < NumArgs) 1147 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1148 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 1149 if (TheCall->getNumArgs() > NumArgs) 1150 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 1151 diag::err_typecheck_call_too_many_args) 1152 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 1153 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 1154 (*(TheCall->arg_end()-1))->getLocEnd()); 1155 1156 Expr *OrigArg = TheCall->getArg(NumArgs-1); 1157 1158 if (OrigArg->isTypeDependent()) 1159 return false; 1160 1161 // This operation requires a non-_Complex floating-point number. 1162 if (!OrigArg->getType()->isRealFloatingType()) 1163 return Diag(OrigArg->getLocStart(), 1164 diag::err_typecheck_call_invalid_unary_fp) 1165 << OrigArg->getType() << OrigArg->getSourceRange(); 1166 1167 // If this is an implicit conversion from float -> double, remove it. 1168 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 1169 Expr *CastArg = Cast->getSubExpr(); 1170 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 1171 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 1172 "promotion from float to double is the only expected cast here"); 1173 Cast->setSubExpr(0); 1174 TheCall->setArg(NumArgs-1, CastArg); 1175 OrigArg = CastArg; 1176 } 1177 } 1178 1179 return false; 1180} 1181 1182/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1183// This is declared to take (...), so we have to check everything. 1184ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1185 if (TheCall->getNumArgs() < 2) 1186 return ExprError(Diag(TheCall->getLocEnd(), 1187 diag::err_typecheck_call_too_few_args_at_least) 1188 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1189 << TheCall->getSourceRange()); 1190 1191 // Determine which of the following types of shufflevector we're checking: 1192 // 1) unary, vector mask: (lhs, mask) 1193 // 2) binary, vector mask: (lhs, rhs, mask) 1194 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1195 QualType resType = TheCall->getArg(0)->getType(); 1196 unsigned numElements = 0; 1197 1198 if (!TheCall->getArg(0)->isTypeDependent() && 1199 !TheCall->getArg(1)->isTypeDependent()) { 1200 QualType LHSType = TheCall->getArg(0)->getType(); 1201 QualType RHSType = TheCall->getArg(1)->getType(); 1202 1203 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 1204 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 1205 << SourceRange(TheCall->getArg(0)->getLocStart(), 1206 TheCall->getArg(1)->getLocEnd()); 1207 return ExprError(); 1208 } 1209 1210 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1211 unsigned numResElements = TheCall->getNumArgs() - 2; 1212 1213 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1214 // with mask. If so, verify that RHS is an integer vector type with the 1215 // same number of elts as lhs. 1216 if (TheCall->getNumArgs() == 2) { 1217 if (!RHSType->hasIntegerRepresentation() || 1218 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1219 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1220 << SourceRange(TheCall->getArg(1)->getLocStart(), 1221 TheCall->getArg(1)->getLocEnd()); 1222 numResElements = numElements; 1223 } 1224 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1225 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1226 << SourceRange(TheCall->getArg(0)->getLocStart(), 1227 TheCall->getArg(1)->getLocEnd()); 1228 return ExprError(); 1229 } else if (numElements != numResElements) { 1230 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1231 resType = Context.getVectorType(eltType, numResElements, 1232 VectorType::GenericVector); 1233 } 1234 } 1235 1236 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1237 if (TheCall->getArg(i)->isTypeDependent() || 1238 TheCall->getArg(i)->isValueDependent()) 1239 continue; 1240 1241 llvm::APSInt Result(32); 1242 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1243 return ExprError(Diag(TheCall->getLocStart(), 1244 diag::err_shufflevector_nonconstant_argument) 1245 << TheCall->getArg(i)->getSourceRange()); 1246 1247 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1248 return ExprError(Diag(TheCall->getLocStart(), 1249 diag::err_shufflevector_argument_too_large) 1250 << TheCall->getArg(i)->getSourceRange()); 1251 } 1252 1253 SmallVector<Expr*, 32> exprs; 1254 1255 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1256 exprs.push_back(TheCall->getArg(i)); 1257 TheCall->setArg(i, 0); 1258 } 1259 1260 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 1261 exprs.size(), resType, 1262 TheCall->getCallee()->getLocStart(), 1263 TheCall->getRParenLoc())); 1264} 1265 1266/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1267// This is declared to take (const void*, ...) and can take two 1268// optional constant int args. 1269bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1270 unsigned NumArgs = TheCall->getNumArgs(); 1271 1272 if (NumArgs > 3) 1273 return Diag(TheCall->getLocEnd(), 1274 diag::err_typecheck_call_too_many_args_at_most) 1275 << 0 /*function call*/ << 3 << NumArgs 1276 << TheCall->getSourceRange(); 1277 1278 // Argument 0 is checked for us and the remaining arguments must be 1279 // constant integers. 1280 for (unsigned i = 1; i != NumArgs; ++i) { 1281 Expr *Arg = TheCall->getArg(i); 1282 1283 llvm::APSInt Result; 1284 if (SemaBuiltinConstantArg(TheCall, i, Result)) 1285 return true; 1286 1287 // FIXME: gcc issues a warning and rewrites these to 0. These 1288 // seems especially odd for the third argument since the default 1289 // is 3. 1290 if (i == 1) { 1291 if (Result.getLimitedValue() > 1) 1292 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1293 << "0" << "1" << Arg->getSourceRange(); 1294 } else { 1295 if (Result.getLimitedValue() > 3) 1296 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1297 << "0" << "3" << Arg->getSourceRange(); 1298 } 1299 } 1300 1301 return false; 1302} 1303 1304/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1305/// TheCall is a constant expression. 1306bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1307 llvm::APSInt &Result) { 1308 Expr *Arg = TheCall->getArg(ArgNum); 1309 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1310 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1311 1312 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 1313 1314 if (!Arg->isIntegerConstantExpr(Result, Context)) 1315 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 1316 << FDecl->getDeclName() << Arg->getSourceRange(); 1317 1318 return false; 1319} 1320 1321/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 1322/// int type). This simply type checks that type is one of the defined 1323/// constants (0-3). 1324// For compatibility check 0-3, llvm only handles 0 and 2. 1325bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 1326 llvm::APSInt Result; 1327 1328 // Check constant-ness first. 1329 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1330 return true; 1331 1332 Expr *Arg = TheCall->getArg(1); 1333 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 1334 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1335 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1336 } 1337 1338 return false; 1339} 1340 1341/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1342/// This checks that val is a constant 1. 1343bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1344 Expr *Arg = TheCall->getArg(1); 1345 llvm::APSInt Result; 1346 1347 // TODO: This is less than ideal. Overload this to take a value. 1348 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1349 return true; 1350 1351 if (Result != 1) 1352 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1353 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1354 1355 return false; 1356} 1357 1358// Handle i > 1 ? "x" : "y", recursively. 1359bool Sema::SemaCheckStringLiteral(const Expr *E, Expr **Args, 1360 unsigned NumArgs, bool HasVAListArg, 1361 unsigned format_idx, unsigned firstDataArg, 1362 FormatStringType Type, bool inFunctionCall) { 1363 tryAgain: 1364 if (E->isTypeDependent() || E->isValueDependent()) 1365 return false; 1366 1367 E = E->IgnoreParenCasts(); 1368 1369 switch (E->getStmtClass()) { 1370 case Stmt::BinaryConditionalOperatorClass: 1371 case Stmt::ConditionalOperatorClass: { 1372 const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E); 1373 return SemaCheckStringLiteral(C->getTrueExpr(), Args, NumArgs, HasVAListArg, 1374 format_idx, firstDataArg, Type, 1375 inFunctionCall) 1376 && SemaCheckStringLiteral(C->getFalseExpr(), Args, NumArgs, HasVAListArg, 1377 format_idx, firstDataArg, Type, 1378 inFunctionCall); 1379 } 1380 1381 case Stmt::IntegerLiteralClass: 1382 // Technically -Wformat-nonliteral does not warn about this case. 1383 // The behavior of printf and friends in this case is implementation 1384 // dependent. Ideally if the format string cannot be null then 1385 // it should have a 'nonnull' attribute in the function prototype. 1386 return true; 1387 1388 case Stmt::ImplicitCastExprClass: { 1389 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 1390 goto tryAgain; 1391 } 1392 1393 case Stmt::OpaqueValueExprClass: 1394 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 1395 E = src; 1396 goto tryAgain; 1397 } 1398 return false; 1399 1400 case Stmt::PredefinedExprClass: 1401 // While __func__, etc., are technically not string literals, they 1402 // cannot contain format specifiers and thus are not a security 1403 // liability. 1404 return true; 1405 1406 case Stmt::DeclRefExprClass: { 1407 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1408 1409 // As an exception, do not flag errors for variables binding to 1410 // const string literals. 1411 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1412 bool isConstant = false; 1413 QualType T = DR->getType(); 1414 1415 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1416 isConstant = AT->getElementType().isConstant(Context); 1417 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1418 isConstant = T.isConstant(Context) && 1419 PT->getPointeeType().isConstant(Context); 1420 } else if (T->isObjCObjectPointerType()) { 1421 // In ObjC, there is usually no "const ObjectPointer" type, 1422 // so don't check if the pointee type is constant. 1423 isConstant = T.isConstant(Context); 1424 } 1425 1426 if (isConstant) { 1427 if (const Expr *Init = VD->getAnyInitializer()) 1428 return SemaCheckStringLiteral(Init, Args, NumArgs, 1429 HasVAListArg, format_idx, firstDataArg, 1430 Type, /*inFunctionCall*/false); 1431 } 1432 1433 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1434 // special check to see if the format string is a function parameter 1435 // of the function calling the printf function. If the function 1436 // has an attribute indicating it is a printf-like function, then we 1437 // should suppress warnings concerning non-literals being used in a call 1438 // to a vprintf function. For example: 1439 // 1440 // void 1441 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1442 // va_list ap; 1443 // va_start(ap, fmt); 1444 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1445 // ... 1446 // 1447 // 1448 // FIXME: We don't have full attribute support yet, so just check to see 1449 // if the argument is a DeclRefExpr that references a parameter. We'll 1450 // add proper support for checking the attribute later. 1451 if (HasVAListArg) 1452 if (isa<ParmVarDecl>(VD)) 1453 return true; 1454 } 1455 1456 return false; 1457 } 1458 1459 case Stmt::CallExprClass: { 1460 const CallExpr *CE = cast<CallExpr>(E); 1461 if (const ImplicitCastExpr *ICE 1462 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 1463 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 1464 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 1465 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 1466 unsigned ArgIndex = FA->getFormatIdx(); 1467 const Expr *Arg = CE->getArg(ArgIndex - 1); 1468 1469 return SemaCheckStringLiteral(Arg, Args, NumArgs, HasVAListArg, 1470 format_idx, firstDataArg, Type, 1471 inFunctionCall); 1472 } 1473 } 1474 } 1475 } 1476 1477 return false; 1478 } 1479 case Stmt::ObjCStringLiteralClass: 1480 case Stmt::StringLiteralClass: { 1481 const StringLiteral *StrE = NULL; 1482 1483 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1484 StrE = ObjCFExpr->getString(); 1485 else 1486 StrE = cast<StringLiteral>(E); 1487 1488 if (StrE) { 1489 CheckFormatString(StrE, E, Args, NumArgs, HasVAListArg, format_idx, 1490 firstDataArg, Type, inFunctionCall); 1491 return true; 1492 } 1493 1494 return false; 1495 } 1496 1497 default: 1498 return false; 1499 } 1500} 1501 1502void 1503Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1504 const Expr * const *ExprArgs, 1505 SourceLocation CallSiteLoc) { 1506 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1507 e = NonNull->args_end(); 1508 i != e; ++i) { 1509 const Expr *ArgExpr = ExprArgs[*i]; 1510 if (ArgExpr->isNullPointerConstant(Context, 1511 Expr::NPC_ValueDependentIsNotNull)) 1512 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 1513 } 1514} 1515 1516Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) { 1517 return llvm::StringSwitch<FormatStringType>(Format->getType()) 1518 .Case("scanf", FST_Scanf) 1519 .Cases("printf", "printf0", FST_Printf) 1520 .Cases("NSString", "CFString", FST_NSString) 1521 .Case("strftime", FST_Strftime) 1522 .Case("strfmon", FST_Strfmon) 1523 .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf) 1524 .Default(FST_Unknown); 1525} 1526 1527/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar 1528/// functions) for correct use of format strings. 1529void Sema::CheckFormatArguments(const FormatAttr *Format, CallExpr *TheCall) { 1530 bool IsCXXMember = false; 1531 // The way the format attribute works in GCC, the implicit this argument 1532 // of member functions is counted. However, it doesn't appear in our own 1533 // lists, so decrement format_idx in that case. 1534 if (isa<CXXMemberCallExpr>(TheCall)) { 1535 const CXXMethodDecl *method_decl = 1536 dyn_cast<CXXMethodDecl>(TheCall->getCalleeDecl()); 1537 IsCXXMember = method_decl && method_decl->isInstance(); 1538 } 1539 CheckFormatArguments(Format, TheCall->getArgs(), TheCall->getNumArgs(), 1540 IsCXXMember, TheCall->getRParenLoc(), 1541 TheCall->getCallee()->getSourceRange()); 1542} 1543 1544void Sema::CheckFormatArguments(const FormatAttr *Format, Expr **Args, 1545 unsigned NumArgs, bool IsCXXMember, 1546 SourceLocation Loc, SourceRange Range) { 1547 bool HasVAListArg = Format->getFirstArg() == 0; 1548 unsigned format_idx = Format->getFormatIdx() - 1; 1549 unsigned firstDataArg = HasVAListArg ? 0 : Format->getFirstArg() - 1; 1550 if (IsCXXMember) { 1551 if (format_idx == 0) 1552 return; 1553 --format_idx; 1554 if(firstDataArg != 0) 1555 --firstDataArg; 1556 } 1557 CheckFormatArguments(Args, NumArgs, HasVAListArg, format_idx, 1558 firstDataArg, GetFormatStringType(Format), Loc, Range); 1559} 1560 1561void Sema::CheckFormatArguments(Expr **Args, unsigned NumArgs, 1562 bool HasVAListArg, unsigned format_idx, 1563 unsigned firstDataArg, FormatStringType Type, 1564 SourceLocation Loc, SourceRange Range) { 1565 // CHECK: printf/scanf-like function is called with no format string. 1566 if (format_idx >= NumArgs) { 1567 Diag(Loc, diag::warn_missing_format_string) << Range; 1568 return; 1569 } 1570 1571 const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts(); 1572 1573 // CHECK: format string is not a string literal. 1574 // 1575 // Dynamically generated format strings are difficult to 1576 // automatically vet at compile time. Requiring that format strings 1577 // are string literals: (1) permits the checking of format strings by 1578 // the compiler and thereby (2) can practically remove the source of 1579 // many format string exploits. 1580 1581 // Format string can be either ObjC string (e.g. @"%d") or 1582 // C string (e.g. "%d") 1583 // ObjC string uses the same format specifiers as C string, so we can use 1584 // the same format string checking logic for both ObjC and C strings. 1585 if (SemaCheckStringLiteral(OrigFormatExpr, Args, NumArgs, HasVAListArg, 1586 format_idx, firstDataArg, Type)) 1587 return; // Literal format string found, check done! 1588 1589 // Do not emit diag when the string param is a macro expansion and the 1590 // format is either NSString or CFString. This is a hack to prevent 1591 // diag when using the NSLocalizedString and CFCopyLocalizedString macros 1592 // which are usually used in place of NS and CF string literals. 1593 if (Type == FST_NSString && Args[format_idx]->getLocStart().isMacroID()) 1594 return; 1595 1596 // If there are no arguments specified, warn with -Wformat-security, otherwise 1597 // warn only with -Wformat-nonliteral. 1598 if (NumArgs == format_idx+1) 1599 Diag(Args[format_idx]->getLocStart(), 1600 diag::warn_format_nonliteral_noargs) 1601 << OrigFormatExpr->getSourceRange(); 1602 else 1603 Diag(Args[format_idx]->getLocStart(), 1604 diag::warn_format_nonliteral) 1605 << OrigFormatExpr->getSourceRange(); 1606} 1607 1608namespace { 1609class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1610protected: 1611 Sema &S; 1612 const StringLiteral *FExpr; 1613 const Expr *OrigFormatExpr; 1614 const unsigned FirstDataArg; 1615 const unsigned NumDataArgs; 1616 const bool IsObjCLiteral; 1617 const char *Beg; // Start of format string. 1618 const bool HasVAListArg; 1619 const Expr * const *Args; 1620 const unsigned NumArgs; 1621 unsigned FormatIdx; 1622 llvm::BitVector CoveredArgs; 1623 bool usesPositionalArgs; 1624 bool atFirstArg; 1625 bool inFunctionCall; 1626public: 1627 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1628 const Expr *origFormatExpr, unsigned firstDataArg, 1629 unsigned numDataArgs, bool isObjCLiteral, 1630 const char *beg, bool hasVAListArg, 1631 Expr **args, unsigned numArgs, 1632 unsigned formatIdx, bool inFunctionCall) 1633 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1634 FirstDataArg(firstDataArg), 1635 NumDataArgs(numDataArgs), 1636 IsObjCLiteral(isObjCLiteral), Beg(beg), 1637 HasVAListArg(hasVAListArg), 1638 Args(args), NumArgs(numArgs), FormatIdx(formatIdx), 1639 usesPositionalArgs(false), atFirstArg(true), 1640 inFunctionCall(inFunctionCall) { 1641 CoveredArgs.resize(numDataArgs); 1642 CoveredArgs.reset(); 1643 } 1644 1645 void DoneProcessing(); 1646 1647 void HandleIncompleteSpecifier(const char *startSpecifier, 1648 unsigned specifierLen); 1649 1650 virtual void HandleInvalidPosition(const char *startSpecifier, 1651 unsigned specifierLen, 1652 analyze_format_string::PositionContext p); 1653 1654 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1655 1656 void HandleNullChar(const char *nullCharacter); 1657 1658 template <typename Range> 1659 static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall, 1660 const Expr *ArgumentExpr, 1661 PartialDiagnostic PDiag, 1662 SourceLocation StringLoc, 1663 bool IsStringLocation, Range StringRange, 1664 FixItHint Fixit = FixItHint()); 1665 1666protected: 1667 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1668 const char *startSpec, 1669 unsigned specifierLen, 1670 const char *csStart, unsigned csLen); 1671 1672 void HandlePositionalNonpositionalArgs(SourceLocation Loc, 1673 const char *startSpec, 1674 unsigned specifierLen); 1675 1676 SourceRange getFormatStringRange(); 1677 CharSourceRange getSpecifierRange(const char *startSpecifier, 1678 unsigned specifierLen); 1679 SourceLocation getLocationOfByte(const char *x); 1680 1681 const Expr *getDataArg(unsigned i) const; 1682 1683 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1684 const analyze_format_string::ConversionSpecifier &CS, 1685 const char *startSpecifier, unsigned specifierLen, 1686 unsigned argIndex); 1687 1688 template <typename Range> 1689 void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc, 1690 bool IsStringLocation, Range StringRange, 1691 FixItHint Fixit = FixItHint()); 1692 1693 void CheckPositionalAndNonpositionalArgs( 1694 const analyze_format_string::FormatSpecifier *FS); 1695}; 1696} 1697 1698SourceRange CheckFormatHandler::getFormatStringRange() { 1699 return OrigFormatExpr->getSourceRange(); 1700} 1701 1702CharSourceRange CheckFormatHandler:: 1703getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1704 SourceLocation Start = getLocationOfByte(startSpecifier); 1705 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1706 1707 // Advance the end SourceLocation by one due to half-open ranges. 1708 End = End.getLocWithOffset(1); 1709 1710 return CharSourceRange::getCharRange(Start, End); 1711} 1712 1713SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 1714 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1715} 1716 1717void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 1718 unsigned specifierLen){ 1719 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier), 1720 getLocationOfByte(startSpecifier), 1721 /*IsStringLocation*/true, 1722 getSpecifierRange(startSpecifier, specifierLen)); 1723} 1724 1725void 1726CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1727 analyze_format_string::PositionContext p) { 1728 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier) 1729 << (unsigned) p, 1730 getLocationOfByte(startPos), /*IsStringLocation*/true, 1731 getSpecifierRange(startPos, posLen)); 1732} 1733 1734void CheckFormatHandler::HandleZeroPosition(const char *startPos, 1735 unsigned posLen) { 1736 EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier), 1737 getLocationOfByte(startPos), 1738 /*IsStringLocation*/true, 1739 getSpecifierRange(startPos, posLen)); 1740} 1741 1742void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 1743 if (!IsObjCLiteral) { 1744 // The presence of a null character is likely an error. 1745 EmitFormatDiagnostic( 1746 S.PDiag(diag::warn_printf_format_string_contains_null_char), 1747 getLocationOfByte(nullCharacter), /*IsStringLocation*/true, 1748 getFormatStringRange()); 1749 } 1750} 1751 1752const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 1753 return Args[FirstDataArg + i]; 1754} 1755 1756void CheckFormatHandler::DoneProcessing() { 1757 // Does the number of data arguments exceed the number of 1758 // format conversions in the format string? 1759 if (!HasVAListArg) { 1760 // Find any arguments that weren't covered. 1761 CoveredArgs.flip(); 1762 signed notCoveredArg = CoveredArgs.find_first(); 1763 if (notCoveredArg >= 0) { 1764 assert((unsigned)notCoveredArg < NumDataArgs); 1765 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used), 1766 getDataArg((unsigned) notCoveredArg)->getLocStart(), 1767 /*IsStringLocation*/false, getFormatStringRange()); 1768 } 1769 } 1770} 1771 1772bool 1773CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 1774 SourceLocation Loc, 1775 const char *startSpec, 1776 unsigned specifierLen, 1777 const char *csStart, 1778 unsigned csLen) { 1779 1780 bool keepGoing = true; 1781 if (argIndex < NumDataArgs) { 1782 // Consider the argument coverered, even though the specifier doesn't 1783 // make sense. 1784 CoveredArgs.set(argIndex); 1785 } 1786 else { 1787 // If argIndex exceeds the number of data arguments we 1788 // don't issue a warning because that is just a cascade of warnings (and 1789 // they may have intended '%%' anyway). We don't want to continue processing 1790 // the format string after this point, however, as we will like just get 1791 // gibberish when trying to match arguments. 1792 keepGoing = false; 1793 } 1794 1795 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion) 1796 << StringRef(csStart, csLen), 1797 Loc, /*IsStringLocation*/true, 1798 getSpecifierRange(startSpec, specifierLen)); 1799 1800 return keepGoing; 1801} 1802 1803void 1804CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc, 1805 const char *startSpec, 1806 unsigned specifierLen) { 1807 EmitFormatDiagnostic( 1808 S.PDiag(diag::warn_format_mix_positional_nonpositional_args), 1809 Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen)); 1810} 1811 1812bool 1813CheckFormatHandler::CheckNumArgs( 1814 const analyze_format_string::FormatSpecifier &FS, 1815 const analyze_format_string::ConversionSpecifier &CS, 1816 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 1817 1818 if (argIndex >= NumDataArgs) { 1819 PartialDiagnostic PDiag = FS.usesPositionalArg() 1820 ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args) 1821 << (argIndex+1) << NumDataArgs) 1822 : S.PDiag(diag::warn_printf_insufficient_data_args); 1823 EmitFormatDiagnostic( 1824 PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true, 1825 getSpecifierRange(startSpecifier, specifierLen)); 1826 return false; 1827 } 1828 return true; 1829} 1830 1831template<typename Range> 1832void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag, 1833 SourceLocation Loc, 1834 bool IsStringLocation, 1835 Range StringRange, 1836 FixItHint FixIt) { 1837 EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag, 1838 Loc, IsStringLocation, StringRange, FixIt); 1839} 1840 1841/// \brief If the format string is not within the funcion call, emit a note 1842/// so that the function call and string are in diagnostic messages. 1843/// 1844/// \param inFunctionCall if true, the format string is within the function 1845/// call and only one diagnostic message will be produced. Otherwise, an 1846/// extra note will be emitted pointing to location of the format string. 1847/// 1848/// \param ArgumentExpr the expression that is passed as the format string 1849/// argument in the function call. Used for getting locations when two 1850/// diagnostics are emitted. 1851/// 1852/// \param PDiag the callee should already have provided any strings for the 1853/// diagnostic message. This function only adds locations and fixits 1854/// to diagnostics. 1855/// 1856/// \param Loc primary location for diagnostic. If two diagnostics are 1857/// required, one will be at Loc and a new SourceLocation will be created for 1858/// the other one. 1859/// 1860/// \param IsStringLocation if true, Loc points to the format string should be 1861/// used for the note. Otherwise, Loc points to the argument list and will 1862/// be used with PDiag. 1863/// 1864/// \param StringRange some or all of the string to highlight. This is 1865/// templated so it can accept either a CharSourceRange or a SourceRange. 1866/// 1867/// \param Fixit optional fix it hint for the format string. 1868template<typename Range> 1869void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall, 1870 const Expr *ArgumentExpr, 1871 PartialDiagnostic PDiag, 1872 SourceLocation Loc, 1873 bool IsStringLocation, 1874 Range StringRange, 1875 FixItHint FixIt) { 1876 if (InFunctionCall) 1877 S.Diag(Loc, PDiag) << StringRange << FixIt; 1878 else { 1879 S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag) 1880 << ArgumentExpr->getSourceRange(); 1881 S.Diag(IsStringLocation ? Loc : StringRange.getBegin(), 1882 diag::note_format_string_defined) 1883 << StringRange << FixIt; 1884 } 1885} 1886 1887//===--- CHECK: Printf format string checking ------------------------------===// 1888 1889namespace { 1890class CheckPrintfHandler : public CheckFormatHandler { 1891public: 1892 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1893 const Expr *origFormatExpr, unsigned firstDataArg, 1894 unsigned numDataArgs, bool isObjCLiteral, 1895 const char *beg, bool hasVAListArg, 1896 Expr **Args, unsigned NumArgs, 1897 unsigned formatIdx, bool inFunctionCall) 1898 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1899 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1900 Args, NumArgs, formatIdx, inFunctionCall) {} 1901 1902 1903 bool HandleInvalidPrintfConversionSpecifier( 1904 const analyze_printf::PrintfSpecifier &FS, 1905 const char *startSpecifier, 1906 unsigned specifierLen); 1907 1908 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 1909 const char *startSpecifier, 1910 unsigned specifierLen); 1911 1912 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 1913 const char *startSpecifier, unsigned specifierLen); 1914 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 1915 const analyze_printf::OptionalAmount &Amt, 1916 unsigned type, 1917 const char *startSpecifier, unsigned specifierLen); 1918 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1919 const analyze_printf::OptionalFlag &flag, 1920 const char *startSpecifier, unsigned specifierLen); 1921 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 1922 const analyze_printf::OptionalFlag &ignoredFlag, 1923 const analyze_printf::OptionalFlag &flag, 1924 const char *startSpecifier, unsigned specifierLen); 1925}; 1926} 1927 1928bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 1929 const analyze_printf::PrintfSpecifier &FS, 1930 const char *startSpecifier, 1931 unsigned specifierLen) { 1932 const analyze_printf::PrintfConversionSpecifier &CS = 1933 FS.getConversionSpecifier(); 1934 1935 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1936 getLocationOfByte(CS.getStart()), 1937 startSpecifier, specifierLen, 1938 CS.getStart(), CS.getLength()); 1939} 1940 1941bool CheckPrintfHandler::HandleAmount( 1942 const analyze_format_string::OptionalAmount &Amt, 1943 unsigned k, const char *startSpecifier, 1944 unsigned specifierLen) { 1945 1946 if (Amt.hasDataArgument()) { 1947 if (!HasVAListArg) { 1948 unsigned argIndex = Amt.getArgIndex(); 1949 if (argIndex >= NumDataArgs) { 1950 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg) 1951 << k, 1952 getLocationOfByte(Amt.getStart()), 1953 /*IsStringLocation*/true, 1954 getSpecifierRange(startSpecifier, specifierLen)); 1955 // Don't do any more checking. We will just emit 1956 // spurious errors. 1957 return false; 1958 } 1959 1960 // Type check the data argument. It should be an 'int'. 1961 // Although not in conformance with C99, we also allow the argument to be 1962 // an 'unsigned int' as that is a reasonably safe case. GCC also 1963 // doesn't emit a warning for that case. 1964 CoveredArgs.set(argIndex); 1965 const Expr *Arg = getDataArg(argIndex); 1966 QualType T = Arg->getType(); 1967 1968 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1969 assert(ATR.isValid()); 1970 1971 if (!ATR.matchesType(S.Context, T)) { 1972 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type) 1973 << k << ATR.getRepresentativeTypeName(S.Context) 1974 << T << Arg->getSourceRange(), 1975 getLocationOfByte(Amt.getStart()), 1976 /*IsStringLocation*/true, 1977 getSpecifierRange(startSpecifier, specifierLen)); 1978 // Don't do any more checking. We will just emit 1979 // spurious errors. 1980 return false; 1981 } 1982 } 1983 } 1984 return true; 1985} 1986 1987void CheckPrintfHandler::HandleInvalidAmount( 1988 const analyze_printf::PrintfSpecifier &FS, 1989 const analyze_printf::OptionalAmount &Amt, 1990 unsigned type, 1991 const char *startSpecifier, 1992 unsigned specifierLen) { 1993 const analyze_printf::PrintfConversionSpecifier &CS = 1994 FS.getConversionSpecifier(); 1995 1996 FixItHint fixit = 1997 Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant 1998 ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 1999 Amt.getConstantLength())) 2000 : FixItHint(); 2001 2002 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount) 2003 << type << CS.toString(), 2004 getLocationOfByte(Amt.getStart()), 2005 /*IsStringLocation*/true, 2006 getSpecifierRange(startSpecifier, specifierLen), 2007 fixit); 2008} 2009 2010void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2011 const analyze_printf::OptionalFlag &flag, 2012 const char *startSpecifier, 2013 unsigned specifierLen) { 2014 // Warn about pointless flag with a fixit removal. 2015 const analyze_printf::PrintfConversionSpecifier &CS = 2016 FS.getConversionSpecifier(); 2017 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag) 2018 << flag.toString() << CS.toString(), 2019 getLocationOfByte(flag.getPosition()), 2020 /*IsStringLocation*/true, 2021 getSpecifierRange(startSpecifier, specifierLen), 2022 FixItHint::CreateRemoval( 2023 getSpecifierRange(flag.getPosition(), 1))); 2024} 2025 2026void CheckPrintfHandler::HandleIgnoredFlag( 2027 const analyze_printf::PrintfSpecifier &FS, 2028 const analyze_printf::OptionalFlag &ignoredFlag, 2029 const analyze_printf::OptionalFlag &flag, 2030 const char *startSpecifier, 2031 unsigned specifierLen) { 2032 // Warn about ignored flag with a fixit removal. 2033 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag) 2034 << ignoredFlag.toString() << flag.toString(), 2035 getLocationOfByte(ignoredFlag.getPosition()), 2036 /*IsStringLocation*/true, 2037 getSpecifierRange(startSpecifier, specifierLen), 2038 FixItHint::CreateRemoval( 2039 getSpecifierRange(ignoredFlag.getPosition(), 1))); 2040} 2041 2042bool 2043CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 2044 &FS, 2045 const char *startSpecifier, 2046 unsigned specifierLen) { 2047 2048 using namespace analyze_format_string; 2049 using namespace analyze_printf; 2050 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 2051 2052 if (FS.consumesDataArgument()) { 2053 if (atFirstArg) { 2054 atFirstArg = false; 2055 usesPositionalArgs = FS.usesPositionalArg(); 2056 } 2057 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2058 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2059 startSpecifier, specifierLen); 2060 return false; 2061 } 2062 } 2063 2064 // First check if the field width, precision, and conversion specifier 2065 // have matching data arguments. 2066 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 2067 startSpecifier, specifierLen)) { 2068 return false; 2069 } 2070 2071 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 2072 startSpecifier, specifierLen)) { 2073 return false; 2074 } 2075 2076 if (!CS.consumesDataArgument()) { 2077 // FIXME: Technically specifying a precision or field width here 2078 // makes no sense. Worth issuing a warning at some point. 2079 return true; 2080 } 2081 2082 // Consume the argument. 2083 unsigned argIndex = FS.getArgIndex(); 2084 if (argIndex < NumDataArgs) { 2085 // The check to see if the argIndex is valid will come later. 2086 // We set the bit here because we may exit early from this 2087 // function if we encounter some other error. 2088 CoveredArgs.set(argIndex); 2089 } 2090 2091 // Check for using an Objective-C specific conversion specifier 2092 // in a non-ObjC literal. 2093 if (!IsObjCLiteral && CS.isObjCArg()) { 2094 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 2095 specifierLen); 2096 } 2097 2098 // Check for invalid use of field width 2099 if (!FS.hasValidFieldWidth()) { 2100 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 2101 startSpecifier, specifierLen); 2102 } 2103 2104 // Check for invalid use of precision 2105 if (!FS.hasValidPrecision()) { 2106 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 2107 startSpecifier, specifierLen); 2108 } 2109 2110 // Check each flag does not conflict with any other component. 2111 if (!FS.hasValidThousandsGroupingPrefix()) 2112 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 2113 if (!FS.hasValidLeadingZeros()) 2114 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 2115 if (!FS.hasValidPlusPrefix()) 2116 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 2117 if (!FS.hasValidSpacePrefix()) 2118 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 2119 if (!FS.hasValidAlternativeForm()) 2120 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 2121 if (!FS.hasValidLeftJustified()) 2122 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 2123 2124 // Check that flags are not ignored by another flag 2125 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 2126 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 2127 startSpecifier, specifierLen); 2128 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 2129 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 2130 startSpecifier, specifierLen); 2131 2132 // Check the length modifier is valid with the given conversion specifier. 2133 const LengthModifier &LM = FS.getLengthModifier(); 2134 if (!FS.hasValidLengthModifier()) 2135 EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length) 2136 << LM.toString() << CS.toString(), 2137 getLocationOfByte(LM.getStart()), 2138 /*IsStringLocation*/true, 2139 getSpecifierRange(startSpecifier, specifierLen), 2140 FixItHint::CreateRemoval( 2141 getSpecifierRange(LM.getStart(), 2142 LM.getLength()))); 2143 2144 // Are we using '%n'? 2145 if (CS.getKind() == ConversionSpecifier::nArg) { 2146 // Issue a warning about this being a possible security issue. 2147 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_write_back), 2148 getLocationOfByte(CS.getStart()), 2149 /*IsStringLocation*/true, 2150 getSpecifierRange(startSpecifier, specifierLen)); 2151 // Continue checking the other format specifiers. 2152 return true; 2153 } 2154 2155 // The remaining checks depend on the data arguments. 2156 if (HasVAListArg) 2157 return true; 2158 2159 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2160 return false; 2161 2162 // Now type check the data expression that matches the 2163 // format specifier. 2164 const Expr *Ex = getDataArg(argIndex); 2165 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context, 2166 IsObjCLiteral); 2167 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 2168 // Check if we didn't match because of an implicit cast from a 'char' 2169 // or 'short' to an 'int'. This is done because printf is a varargs 2170 // function. 2171 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 2172 if (ICE->getType() == S.Context.IntTy) { 2173 // All further checking is done on the subexpression. 2174 Ex = ICE->getSubExpr(); 2175 if (ATR.matchesType(S.Context, Ex->getType())) 2176 return true; 2177 } 2178 2179 // We may be able to offer a FixItHint if it is a supported type. 2180 PrintfSpecifier fixedFS = FS; 2181 bool success = fixedFS.fixType(Ex->getType(), S.getLangOptions()); 2182 2183 if (success) { 2184 // Get the fix string from the fixed format specifier 2185 llvm::SmallString<128> buf; 2186 llvm::raw_svector_ostream os(buf); 2187 fixedFS.toString(os); 2188 2189 EmitFormatDiagnostic( 2190 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2191 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2192 << Ex->getSourceRange(), 2193 getLocationOfByte(CS.getStart()), 2194 /*IsStringLocation*/true, 2195 getSpecifierRange(startSpecifier, specifierLen), 2196 FixItHint::CreateReplacement( 2197 getSpecifierRange(startSpecifier, specifierLen), 2198 os.str())); 2199 } 2200 else { 2201 S.Diag(getLocationOfByte(CS.getStart()), 2202 diag::warn_printf_conversion_argument_type_mismatch) 2203 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2204 << getSpecifierRange(startSpecifier, specifierLen) 2205 << Ex->getSourceRange(); 2206 } 2207 } 2208 2209 return true; 2210} 2211 2212//===--- CHECK: Scanf format string checking ------------------------------===// 2213 2214namespace { 2215class CheckScanfHandler : public CheckFormatHandler { 2216public: 2217 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 2218 const Expr *origFormatExpr, unsigned firstDataArg, 2219 unsigned numDataArgs, bool isObjCLiteral, 2220 const char *beg, bool hasVAListArg, 2221 Expr **Args, unsigned NumArgs, 2222 unsigned formatIdx, bool inFunctionCall) 2223 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2224 numDataArgs, isObjCLiteral, beg, hasVAListArg, 2225 Args, NumArgs, formatIdx, inFunctionCall) {} 2226 2227 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 2228 const char *startSpecifier, 2229 unsigned specifierLen); 2230 2231 bool HandleInvalidScanfConversionSpecifier( 2232 const analyze_scanf::ScanfSpecifier &FS, 2233 const char *startSpecifier, 2234 unsigned specifierLen); 2235 2236 void HandleIncompleteScanList(const char *start, const char *end); 2237}; 2238} 2239 2240void CheckScanfHandler::HandleIncompleteScanList(const char *start, 2241 const char *end) { 2242 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 2243 getLocationOfByte(end), /*IsStringLocation*/true, 2244 getSpecifierRange(start, end - start)); 2245} 2246 2247bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 2248 const analyze_scanf::ScanfSpecifier &FS, 2249 const char *startSpecifier, 2250 unsigned specifierLen) { 2251 2252 const analyze_scanf::ScanfConversionSpecifier &CS = 2253 FS.getConversionSpecifier(); 2254 2255 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2256 getLocationOfByte(CS.getStart()), 2257 startSpecifier, specifierLen, 2258 CS.getStart(), CS.getLength()); 2259} 2260 2261bool CheckScanfHandler::HandleScanfSpecifier( 2262 const analyze_scanf::ScanfSpecifier &FS, 2263 const char *startSpecifier, 2264 unsigned specifierLen) { 2265 2266 using namespace analyze_scanf; 2267 using namespace analyze_format_string; 2268 2269 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 2270 2271 // Handle case where '%' and '*' don't consume an argument. These shouldn't 2272 // be used to decide if we are using positional arguments consistently. 2273 if (FS.consumesDataArgument()) { 2274 if (atFirstArg) { 2275 atFirstArg = false; 2276 usesPositionalArgs = FS.usesPositionalArg(); 2277 } 2278 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2279 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2280 startSpecifier, specifierLen); 2281 return false; 2282 } 2283 } 2284 2285 // Check if the field with is non-zero. 2286 const OptionalAmount &Amt = FS.getFieldWidth(); 2287 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 2288 if (Amt.getConstantAmount() == 0) { 2289 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 2290 Amt.getConstantLength()); 2291 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 2292 getLocationOfByte(Amt.getStart()), 2293 /*IsStringLocation*/true, R, 2294 FixItHint::CreateRemoval(R)); 2295 } 2296 } 2297 2298 if (!FS.consumesDataArgument()) { 2299 // FIXME: Technically specifying a precision or field width here 2300 // makes no sense. Worth issuing a warning at some point. 2301 return true; 2302 } 2303 2304 // Consume the argument. 2305 unsigned argIndex = FS.getArgIndex(); 2306 if (argIndex < NumDataArgs) { 2307 // The check to see if the argIndex is valid will come later. 2308 // We set the bit here because we may exit early from this 2309 // function if we encounter some other error. 2310 CoveredArgs.set(argIndex); 2311 } 2312 2313 // Check the length modifier is valid with the given conversion specifier. 2314 const LengthModifier &LM = FS.getLengthModifier(); 2315 if (!FS.hasValidLengthModifier()) { 2316 S.Diag(getLocationOfByte(LM.getStart()), 2317 diag::warn_format_nonsensical_length) 2318 << LM.toString() << CS.toString() 2319 << getSpecifierRange(startSpecifier, specifierLen) 2320 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 2321 LM.getLength())); 2322 } 2323 2324 // The remaining checks depend on the data arguments. 2325 if (HasVAListArg) 2326 return true; 2327 2328 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2329 return false; 2330 2331 // Check that the argument type matches the format specifier. 2332 const Expr *Ex = getDataArg(argIndex); 2333 const analyze_scanf::ScanfArgTypeResult &ATR = FS.getArgType(S.Context); 2334 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 2335 ScanfSpecifier fixedFS = FS; 2336 bool success = fixedFS.fixType(Ex->getType(), S.getLangOptions()); 2337 2338 if (success) { 2339 // Get the fix string from the fixed format specifier. 2340 llvm::SmallString<128> buf; 2341 llvm::raw_svector_ostream os(buf); 2342 fixedFS.toString(os); 2343 2344 EmitFormatDiagnostic( 2345 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2346 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2347 << Ex->getSourceRange(), 2348 getLocationOfByte(CS.getStart()), 2349 /*IsStringLocation*/true, 2350 getSpecifierRange(startSpecifier, specifierLen), 2351 FixItHint::CreateReplacement( 2352 getSpecifierRange(startSpecifier, specifierLen), 2353 os.str())); 2354 } else { 2355 S.Diag(getLocationOfByte(CS.getStart()), 2356 diag::warn_printf_conversion_argument_type_mismatch) 2357 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2358 << getSpecifierRange(startSpecifier, specifierLen) 2359 << Ex->getSourceRange(); 2360 } 2361 } 2362 2363 return true; 2364} 2365 2366void Sema::CheckFormatString(const StringLiteral *FExpr, 2367 const Expr *OrigFormatExpr, 2368 Expr **Args, unsigned NumArgs, 2369 bool HasVAListArg, unsigned format_idx, 2370 unsigned firstDataArg, FormatStringType Type, 2371 bool inFunctionCall) { 2372 2373 // CHECK: is the format string a wide literal? 2374 if (!FExpr->isAscii()) { 2375 CheckFormatHandler::EmitFormatDiagnostic( 2376 *this, inFunctionCall, Args[format_idx], 2377 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 2378 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2379 return; 2380 } 2381 2382 // Str - The format string. NOTE: this is NOT null-terminated! 2383 StringRef StrRef = FExpr->getString(); 2384 const char *Str = StrRef.data(); 2385 unsigned StrLen = StrRef.size(); 2386 const unsigned numDataArgs = NumArgs - firstDataArg; 2387 2388 // CHECK: empty format string? 2389 if (StrLen == 0 && numDataArgs > 0) { 2390 CheckFormatHandler::EmitFormatDiagnostic( 2391 *this, inFunctionCall, Args[format_idx], 2392 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 2393 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2394 return; 2395 } 2396 2397 if (Type == FST_Printf || Type == FST_NSString) { 2398 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2399 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2400 Str, HasVAListArg, Args, NumArgs, format_idx, 2401 inFunctionCall); 2402 2403 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen, 2404 getLangOptions())) 2405 H.DoneProcessing(); 2406 } else if (Type == FST_Scanf) { 2407 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2408 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2409 Str, HasVAListArg, Args, NumArgs, format_idx, 2410 inFunctionCall); 2411 2412 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen, 2413 getLangOptions())) 2414 H.DoneProcessing(); 2415 } // TODO: handle other formats 2416} 2417 2418//===--- CHECK: Standard memory functions ---------------------------------===// 2419 2420/// \brief Determine whether the given type is a dynamic class type (e.g., 2421/// whether it has a vtable). 2422static bool isDynamicClassType(QualType T) { 2423 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 2424 if (CXXRecordDecl *Definition = Record->getDefinition()) 2425 if (Definition->isDynamicClass()) 2426 return true; 2427 2428 return false; 2429} 2430 2431/// \brief If E is a sizeof expression, returns its argument expression, 2432/// otherwise returns NULL. 2433static const Expr *getSizeOfExprArg(const Expr* E) { 2434 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2435 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2436 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 2437 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 2438 2439 return 0; 2440} 2441 2442/// \brief If E is a sizeof expression, returns its argument type. 2443static QualType getSizeOfArgType(const Expr* E) { 2444 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2445 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2446 if (SizeOf->getKind() == clang::UETT_SizeOf) 2447 return SizeOf->getTypeOfArgument(); 2448 2449 return QualType(); 2450} 2451 2452/// \brief Check for dangerous or invalid arguments to memset(). 2453/// 2454/// This issues warnings on known problematic, dangerous or unspecified 2455/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 2456/// function calls. 2457/// 2458/// \param Call The call expression to diagnose. 2459void Sema::CheckMemaccessArguments(const CallExpr *Call, 2460 unsigned BId, 2461 IdentifierInfo *FnName) { 2462 assert(BId != 0); 2463 2464 // It is possible to have a non-standard definition of memset. Validate 2465 // we have enough arguments, and if not, abort further checking. 2466 unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3); 2467 if (Call->getNumArgs() < ExpectedNumArgs) 2468 return; 2469 2470 unsigned LastArg = (BId == Builtin::BImemset || 2471 BId == Builtin::BIstrndup ? 1 : 2); 2472 unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2); 2473 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 2474 2475 // We have special checking when the length is a sizeof expression. 2476 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 2477 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 2478 llvm::FoldingSetNodeID SizeOfArgID; 2479 2480 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 2481 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 2482 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 2483 2484 QualType DestTy = Dest->getType(); 2485 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 2486 QualType PointeeTy = DestPtrTy->getPointeeType(); 2487 2488 // Never warn about void type pointers. This can be used to suppress 2489 // false positives. 2490 if (PointeeTy->isVoidType()) 2491 continue; 2492 2493 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 2494 // actually comparing the expressions for equality. Because computing the 2495 // expression IDs can be expensive, we only do this if the diagnostic is 2496 // enabled. 2497 if (SizeOfArg && 2498 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 2499 SizeOfArg->getExprLoc())) { 2500 // We only compute IDs for expressions if the warning is enabled, and 2501 // cache the sizeof arg's ID. 2502 if (SizeOfArgID == llvm::FoldingSetNodeID()) 2503 SizeOfArg->Profile(SizeOfArgID, Context, true); 2504 llvm::FoldingSetNodeID DestID; 2505 Dest->Profile(DestID, Context, true); 2506 if (DestID == SizeOfArgID) { 2507 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 2508 // over sizeof(src) as well. 2509 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 2510 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 2511 if (UnaryOp->getOpcode() == UO_AddrOf) 2512 ActionIdx = 1; // If its an address-of operator, just remove it. 2513 if (Context.getTypeSize(PointeeTy) == Context.getCharWidth()) 2514 ActionIdx = 2; // If the pointee's size is sizeof(char), 2515 // suggest an explicit length. 2516 unsigned DestSrcSelect = 2517 (BId == Builtin::BIstrndup ? 1 : ArgIdx); 2518 DiagRuntimeBehavior(SizeOfArg->getExprLoc(), Dest, 2519 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 2520 << FnName << DestSrcSelect << ActionIdx 2521 << Dest->getSourceRange() 2522 << SizeOfArg->getSourceRange()); 2523 break; 2524 } 2525 } 2526 2527 // Also check for cases where the sizeof argument is the exact same 2528 // type as the memory argument, and where it points to a user-defined 2529 // record type. 2530 if (SizeOfArgTy != QualType()) { 2531 if (PointeeTy->isRecordType() && 2532 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 2533 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 2534 PDiag(diag::warn_sizeof_pointer_type_memaccess) 2535 << FnName << SizeOfArgTy << ArgIdx 2536 << PointeeTy << Dest->getSourceRange() 2537 << LenExpr->getSourceRange()); 2538 break; 2539 } 2540 } 2541 2542 // Always complain about dynamic classes. 2543 if (isDynamicClassType(PointeeTy)) { 2544 2545 unsigned OperationType = 0; 2546 // "overwritten" if we're warning about the destination for any call 2547 // but memcmp; otherwise a verb appropriate to the call. 2548 if (ArgIdx != 0 || BId == Builtin::BImemcmp) { 2549 if (BId == Builtin::BImemcpy) 2550 OperationType = 1; 2551 else if(BId == Builtin::BImemmove) 2552 OperationType = 2; 2553 else if (BId == Builtin::BImemcmp) 2554 OperationType = 3; 2555 } 2556 2557 DiagRuntimeBehavior( 2558 Dest->getExprLoc(), Dest, 2559 PDiag(diag::warn_dyn_class_memaccess) 2560 << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx) 2561 << FnName << PointeeTy 2562 << OperationType 2563 << Call->getCallee()->getSourceRange()); 2564 } else if (PointeeTy.hasNonTrivialObjCLifetime() && 2565 BId != Builtin::BImemset) 2566 DiagRuntimeBehavior( 2567 Dest->getExprLoc(), Dest, 2568 PDiag(diag::warn_arc_object_memaccess) 2569 << ArgIdx << FnName << PointeeTy 2570 << Call->getCallee()->getSourceRange()); 2571 else 2572 continue; 2573 2574 DiagRuntimeBehavior( 2575 Dest->getExprLoc(), Dest, 2576 PDiag(diag::note_bad_memaccess_silence) 2577 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 2578 break; 2579 } 2580 } 2581} 2582 2583// A little helper routine: ignore addition and subtraction of integer literals. 2584// This intentionally does not ignore all integer constant expressions because 2585// we don't want to remove sizeof(). 2586static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 2587 Ex = Ex->IgnoreParenCasts(); 2588 2589 for (;;) { 2590 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 2591 if (!BO || !BO->isAdditiveOp()) 2592 break; 2593 2594 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 2595 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 2596 2597 if (isa<IntegerLiteral>(RHS)) 2598 Ex = LHS; 2599 else if (isa<IntegerLiteral>(LHS)) 2600 Ex = RHS; 2601 else 2602 break; 2603 } 2604 2605 return Ex; 2606} 2607 2608// Warn if the user has made the 'size' argument to strlcpy or strlcat 2609// be the size of the source, instead of the destination. 2610void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 2611 IdentifierInfo *FnName) { 2612 2613 // Don't crash if the user has the wrong number of arguments 2614 if (Call->getNumArgs() != 3) 2615 return; 2616 2617 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 2618 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 2619 const Expr *CompareWithSrc = NULL; 2620 2621 // Look for 'strlcpy(dst, x, sizeof(x))' 2622 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 2623 CompareWithSrc = Ex; 2624 else { 2625 // Look for 'strlcpy(dst, x, strlen(x))' 2626 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 2627 if (SizeCall->isBuiltinCall() == Builtin::BIstrlen 2628 && SizeCall->getNumArgs() == 1) 2629 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 2630 } 2631 } 2632 2633 if (!CompareWithSrc) 2634 return; 2635 2636 // Determine if the argument to sizeof/strlen is equal to the source 2637 // argument. In principle there's all kinds of things you could do 2638 // here, for instance creating an == expression and evaluating it with 2639 // EvaluateAsBooleanCondition, but this uses a more direct technique: 2640 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 2641 if (!SrcArgDRE) 2642 return; 2643 2644 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 2645 if (!CompareWithSrcDRE || 2646 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 2647 return; 2648 2649 const Expr *OriginalSizeArg = Call->getArg(2); 2650 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 2651 << OriginalSizeArg->getSourceRange() << FnName; 2652 2653 // Output a FIXIT hint if the destination is an array (rather than a 2654 // pointer to an array). This could be enhanced to handle some 2655 // pointers if we know the actual size, like if DstArg is 'array+2' 2656 // we could say 'sizeof(array)-2'. 2657 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 2658 QualType DstArgTy = DstArg->getType(); 2659 2660 // Only handle constant-sized or VLAs, but not flexible members. 2661 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 2662 // Only issue the FIXIT for arrays of size > 1. 2663 if (CAT->getSize().getSExtValue() <= 1) 2664 return; 2665 } else if (!DstArgTy->isVariableArrayType()) { 2666 return; 2667 } 2668 2669 llvm::SmallString<128> sizeString; 2670 llvm::raw_svector_ostream OS(sizeString); 2671 OS << "sizeof("; 2672 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 2673 OS << ")"; 2674 2675 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 2676 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 2677 OS.str()); 2678} 2679 2680//===--- CHECK: Return Address of Stack Variable --------------------------===// 2681 2682static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars); 2683static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars); 2684 2685/// CheckReturnStackAddr - Check if a return statement returns the address 2686/// of a stack variable. 2687void 2688Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 2689 SourceLocation ReturnLoc) { 2690 2691 Expr *stackE = 0; 2692 SmallVector<DeclRefExpr *, 8> refVars; 2693 2694 // Perform checking for returned stack addresses, local blocks, 2695 // label addresses or references to temporaries. 2696 if (lhsType->isPointerType() || 2697 (!getLangOptions().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 2698 stackE = EvalAddr(RetValExp, refVars); 2699 } else if (lhsType->isReferenceType()) { 2700 stackE = EvalVal(RetValExp, refVars); 2701 } 2702 2703 if (stackE == 0) 2704 return; // Nothing suspicious was found. 2705 2706 SourceLocation diagLoc; 2707 SourceRange diagRange; 2708 if (refVars.empty()) { 2709 diagLoc = stackE->getLocStart(); 2710 diagRange = stackE->getSourceRange(); 2711 } else { 2712 // We followed through a reference variable. 'stackE' contains the 2713 // problematic expression but we will warn at the return statement pointing 2714 // at the reference variable. We will later display the "trail" of 2715 // reference variables using notes. 2716 diagLoc = refVars[0]->getLocStart(); 2717 diagRange = refVars[0]->getSourceRange(); 2718 } 2719 2720 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 2721 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 2722 : diag::warn_ret_stack_addr) 2723 << DR->getDecl()->getDeclName() << diagRange; 2724 } else if (isa<BlockExpr>(stackE)) { // local block. 2725 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 2726 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 2727 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 2728 } else { // local temporary. 2729 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 2730 : diag::warn_ret_local_temp_addr) 2731 << diagRange; 2732 } 2733 2734 // Display the "trail" of reference variables that we followed until we 2735 // found the problematic expression using notes. 2736 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 2737 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 2738 // If this var binds to another reference var, show the range of the next 2739 // var, otherwise the var binds to the problematic expression, in which case 2740 // show the range of the expression. 2741 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 2742 : stackE->getSourceRange(); 2743 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 2744 << VD->getDeclName() << range; 2745 } 2746} 2747 2748/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 2749/// check if the expression in a return statement evaluates to an address 2750/// to a location on the stack, a local block, an address of a label, or a 2751/// reference to local temporary. The recursion is used to traverse the 2752/// AST of the return expression, with recursion backtracking when we 2753/// encounter a subexpression that (1) clearly does not lead to one of the 2754/// above problematic expressions (2) is something we cannot determine leads to 2755/// a problematic expression based on such local checking. 2756/// 2757/// Both EvalAddr and EvalVal follow through reference variables to evaluate 2758/// the expression that they point to. Such variables are added to the 2759/// 'refVars' vector so that we know what the reference variable "trail" was. 2760/// 2761/// EvalAddr processes expressions that are pointers that are used as 2762/// references (and not L-values). EvalVal handles all other values. 2763/// At the base case of the recursion is a check for the above problematic 2764/// expressions. 2765/// 2766/// This implementation handles: 2767/// 2768/// * pointer-to-pointer casts 2769/// * implicit conversions from array references to pointers 2770/// * taking the address of fields 2771/// * arbitrary interplay between "&" and "*" operators 2772/// * pointer arithmetic from an address of a stack variable 2773/// * taking the address of an array element where the array is on the stack 2774static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 2775 if (E->isTypeDependent()) 2776 return NULL; 2777 2778 // We should only be called for evaluating pointer expressions. 2779 assert((E->getType()->isAnyPointerType() || 2780 E->getType()->isBlockPointerType() || 2781 E->getType()->isObjCQualifiedIdType()) && 2782 "EvalAddr only works on pointers"); 2783 2784 E = E->IgnoreParens(); 2785 2786 // Our "symbolic interpreter" is just a dispatch off the currently 2787 // viewed AST node. We then recursively traverse the AST by calling 2788 // EvalAddr and EvalVal appropriately. 2789 switch (E->getStmtClass()) { 2790 case Stmt::DeclRefExprClass: { 2791 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2792 2793 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2794 // If this is a reference variable, follow through to the expression that 2795 // it points to. 2796 if (V->hasLocalStorage() && 2797 V->getType()->isReferenceType() && V->hasInit()) { 2798 // Add the reference variable to the "trail". 2799 refVars.push_back(DR); 2800 return EvalAddr(V->getInit(), refVars); 2801 } 2802 2803 return NULL; 2804 } 2805 2806 case Stmt::UnaryOperatorClass: { 2807 // The only unary operator that make sense to handle here 2808 // is AddrOf. All others don't make sense as pointers. 2809 UnaryOperator *U = cast<UnaryOperator>(E); 2810 2811 if (U->getOpcode() == UO_AddrOf) 2812 return EvalVal(U->getSubExpr(), refVars); 2813 else 2814 return NULL; 2815 } 2816 2817 case Stmt::BinaryOperatorClass: { 2818 // Handle pointer arithmetic. All other binary operators are not valid 2819 // in this context. 2820 BinaryOperator *B = cast<BinaryOperator>(E); 2821 BinaryOperatorKind op = B->getOpcode(); 2822 2823 if (op != BO_Add && op != BO_Sub) 2824 return NULL; 2825 2826 Expr *Base = B->getLHS(); 2827 2828 // Determine which argument is the real pointer base. It could be 2829 // the RHS argument instead of the LHS. 2830 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 2831 2832 assert (Base->getType()->isPointerType()); 2833 return EvalAddr(Base, refVars); 2834 } 2835 2836 // For conditional operators we need to see if either the LHS or RHS are 2837 // valid DeclRefExpr*s. If one of them is valid, we return it. 2838 case Stmt::ConditionalOperatorClass: { 2839 ConditionalOperator *C = cast<ConditionalOperator>(E); 2840 2841 // Handle the GNU extension for missing LHS. 2842 if (Expr *lhsExpr = C->getLHS()) { 2843 // In C++, we can have a throw-expression, which has 'void' type. 2844 if (!lhsExpr->getType()->isVoidType()) 2845 if (Expr* LHS = EvalAddr(lhsExpr, refVars)) 2846 return LHS; 2847 } 2848 2849 // In C++, we can have a throw-expression, which has 'void' type. 2850 if (C->getRHS()->getType()->isVoidType()) 2851 return NULL; 2852 2853 return EvalAddr(C->getRHS(), refVars); 2854 } 2855 2856 case Stmt::BlockExprClass: 2857 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 2858 return E; // local block. 2859 return NULL; 2860 2861 case Stmt::AddrLabelExprClass: 2862 return E; // address of label. 2863 2864 case Stmt::ExprWithCleanupsClass: 2865 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars); 2866 2867 // For casts, we need to handle conversions from arrays to 2868 // pointer values, and pointer-to-pointer conversions. 2869 case Stmt::ImplicitCastExprClass: 2870 case Stmt::CStyleCastExprClass: 2871 case Stmt::CXXFunctionalCastExprClass: 2872 case Stmt::ObjCBridgedCastExprClass: { 2873 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 2874 QualType T = SubExpr->getType(); 2875 2876 if (SubExpr->getType()->isPointerType() || 2877 SubExpr->getType()->isBlockPointerType() || 2878 SubExpr->getType()->isObjCQualifiedIdType()) 2879 return EvalAddr(SubExpr, refVars); 2880 else if (T->isArrayType()) 2881 return EvalVal(SubExpr, refVars); 2882 else 2883 return 0; 2884 } 2885 2886 // C++ casts. For dynamic casts, static casts, and const casts, we 2887 // are always converting from a pointer-to-pointer, so we just blow 2888 // through the cast. In the case the dynamic cast doesn't fail (and 2889 // return NULL), we take the conservative route and report cases 2890 // where we return the address of a stack variable. For Reinterpre 2891 // FIXME: The comment about is wrong; we're not always converting 2892 // from pointer to pointer. I'm guessing that this code should also 2893 // handle references to objects. 2894 case Stmt::CXXStaticCastExprClass: 2895 case Stmt::CXXDynamicCastExprClass: 2896 case Stmt::CXXConstCastExprClass: 2897 case Stmt::CXXReinterpretCastExprClass: { 2898 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 2899 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 2900 return EvalAddr(S, refVars); 2901 else 2902 return NULL; 2903 } 2904 2905 case Stmt::MaterializeTemporaryExprClass: 2906 if (Expr *Result = EvalAddr( 2907 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 2908 refVars)) 2909 return Result; 2910 2911 return E; 2912 2913 // Everything else: we simply don't reason about them. 2914 default: 2915 return NULL; 2916 } 2917} 2918 2919 2920/// EvalVal - This function is complements EvalAddr in the mutual recursion. 2921/// See the comments for EvalAddr for more details. 2922static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 2923do { 2924 // We should only be called for evaluating non-pointer expressions, or 2925 // expressions with a pointer type that are not used as references but instead 2926 // are l-values (e.g., DeclRefExpr with a pointer type). 2927 2928 // Our "symbolic interpreter" is just a dispatch off the currently 2929 // viewed AST node. We then recursively traverse the AST by calling 2930 // EvalAddr and EvalVal appropriately. 2931 2932 E = E->IgnoreParens(); 2933 switch (E->getStmtClass()) { 2934 case Stmt::ImplicitCastExprClass: { 2935 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 2936 if (IE->getValueKind() == VK_LValue) { 2937 E = IE->getSubExpr(); 2938 continue; 2939 } 2940 return NULL; 2941 } 2942 2943 case Stmt::ExprWithCleanupsClass: 2944 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars); 2945 2946 case Stmt::DeclRefExprClass: { 2947 // When we hit a DeclRefExpr we are looking at code that refers to a 2948 // variable's name. If it's not a reference variable we check if it has 2949 // local storage within the function, and if so, return the expression. 2950 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2951 2952 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2953 if (V->hasLocalStorage()) { 2954 if (!V->getType()->isReferenceType()) 2955 return DR; 2956 2957 // Reference variable, follow through to the expression that 2958 // it points to. 2959 if (V->hasInit()) { 2960 // Add the reference variable to the "trail". 2961 refVars.push_back(DR); 2962 return EvalVal(V->getInit(), refVars); 2963 } 2964 } 2965 2966 return NULL; 2967 } 2968 2969 case Stmt::UnaryOperatorClass: { 2970 // The only unary operator that make sense to handle here 2971 // is Deref. All others don't resolve to a "name." This includes 2972 // handling all sorts of rvalues passed to a unary operator. 2973 UnaryOperator *U = cast<UnaryOperator>(E); 2974 2975 if (U->getOpcode() == UO_Deref) 2976 return EvalAddr(U->getSubExpr(), refVars); 2977 2978 return NULL; 2979 } 2980 2981 case Stmt::ArraySubscriptExprClass: { 2982 // Array subscripts are potential references to data on the stack. We 2983 // retrieve the DeclRefExpr* for the array variable if it indeed 2984 // has local storage. 2985 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars); 2986 } 2987 2988 case Stmt::ConditionalOperatorClass: { 2989 // For conditional operators we need to see if either the LHS or RHS are 2990 // non-NULL Expr's. If one is non-NULL, we return it. 2991 ConditionalOperator *C = cast<ConditionalOperator>(E); 2992 2993 // Handle the GNU extension for missing LHS. 2994 if (Expr *lhsExpr = C->getLHS()) 2995 if (Expr *LHS = EvalVal(lhsExpr, refVars)) 2996 return LHS; 2997 2998 return EvalVal(C->getRHS(), refVars); 2999 } 3000 3001 // Accesses to members are potential references to data on the stack. 3002 case Stmt::MemberExprClass: { 3003 MemberExpr *M = cast<MemberExpr>(E); 3004 3005 // Check for indirect access. We only want direct field accesses. 3006 if (M->isArrow()) 3007 return NULL; 3008 3009 // Check whether the member type is itself a reference, in which case 3010 // we're not going to refer to the member, but to what the member refers to. 3011 if (M->getMemberDecl()->getType()->isReferenceType()) 3012 return NULL; 3013 3014 return EvalVal(M->getBase(), refVars); 3015 } 3016 3017 case Stmt::MaterializeTemporaryExprClass: 3018 if (Expr *Result = EvalVal( 3019 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3020 refVars)) 3021 return Result; 3022 3023 return E; 3024 3025 default: 3026 // Check that we don't return or take the address of a reference to a 3027 // temporary. This is only useful in C++. 3028 if (!E->isTypeDependent() && E->isRValue()) 3029 return E; 3030 3031 // Everything else: we simply don't reason about them. 3032 return NULL; 3033 } 3034} while (true); 3035} 3036 3037//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 3038 3039/// Check for comparisons of floating point operands using != and ==. 3040/// Issue a warning if these are no self-comparisons, as they are not likely 3041/// to do what the programmer intended. 3042void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 3043 bool EmitWarning = true; 3044 3045 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 3046 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 3047 3048 // Special case: check for x == x (which is OK). 3049 // Do not emit warnings for such cases. 3050 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 3051 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 3052 if (DRL->getDecl() == DRR->getDecl()) 3053 EmitWarning = false; 3054 3055 3056 // Special case: check for comparisons against literals that can be exactly 3057 // represented by APFloat. In such cases, do not emit a warning. This 3058 // is a heuristic: often comparison against such literals are used to 3059 // detect if a value in a variable has not changed. This clearly can 3060 // lead to false negatives. 3061 if (EmitWarning) { 3062 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 3063 if (FLL->isExact()) 3064 EmitWarning = false; 3065 } else 3066 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 3067 if (FLR->isExact()) 3068 EmitWarning = false; 3069 } 3070 } 3071 3072 // Check for comparisons with builtin types. 3073 if (EmitWarning) 3074 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 3075 if (CL->isBuiltinCall()) 3076 EmitWarning = false; 3077 3078 if (EmitWarning) 3079 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 3080 if (CR->isBuiltinCall()) 3081 EmitWarning = false; 3082 3083 // Emit the diagnostic. 3084 if (EmitWarning) 3085 Diag(Loc, diag::warn_floatingpoint_eq) 3086 << LHS->getSourceRange() << RHS->getSourceRange(); 3087} 3088 3089//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 3090//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 3091 3092namespace { 3093 3094/// Structure recording the 'active' range of an integer-valued 3095/// expression. 3096struct IntRange { 3097 /// The number of bits active in the int. 3098 unsigned Width; 3099 3100 /// True if the int is known not to have negative values. 3101 bool NonNegative; 3102 3103 IntRange(unsigned Width, bool NonNegative) 3104 : Width(Width), NonNegative(NonNegative) 3105 {} 3106 3107 /// Returns the range of the bool type. 3108 static IntRange forBoolType() { 3109 return IntRange(1, true); 3110 } 3111 3112 /// Returns the range of an opaque value of the given integral type. 3113 static IntRange forValueOfType(ASTContext &C, QualType T) { 3114 return forValueOfCanonicalType(C, 3115 T->getCanonicalTypeInternal().getTypePtr()); 3116 } 3117 3118 /// Returns the range of an opaque value of a canonical integral type. 3119 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 3120 assert(T->isCanonicalUnqualified()); 3121 3122 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3123 T = VT->getElementType().getTypePtr(); 3124 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3125 T = CT->getElementType().getTypePtr(); 3126 3127 // For enum types, use the known bit width of the enumerators. 3128 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 3129 EnumDecl *Enum = ET->getDecl(); 3130 if (!Enum->isCompleteDefinition()) 3131 return IntRange(C.getIntWidth(QualType(T, 0)), false); 3132 3133 unsigned NumPositive = Enum->getNumPositiveBits(); 3134 unsigned NumNegative = Enum->getNumNegativeBits(); 3135 3136 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 3137 } 3138 3139 const BuiltinType *BT = cast<BuiltinType>(T); 3140 assert(BT->isInteger()); 3141 3142 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3143 } 3144 3145 /// Returns the "target" range of a canonical integral type, i.e. 3146 /// the range of values expressible in the type. 3147 /// 3148 /// This matches forValueOfCanonicalType except that enums have the 3149 /// full range of their type, not the range of their enumerators. 3150 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 3151 assert(T->isCanonicalUnqualified()); 3152 3153 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3154 T = VT->getElementType().getTypePtr(); 3155 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3156 T = CT->getElementType().getTypePtr(); 3157 if (const EnumType *ET = dyn_cast<EnumType>(T)) 3158 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 3159 3160 const BuiltinType *BT = cast<BuiltinType>(T); 3161 assert(BT->isInteger()); 3162 3163 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3164 } 3165 3166 /// Returns the supremum of two ranges: i.e. their conservative merge. 3167 static IntRange join(IntRange L, IntRange R) { 3168 return IntRange(std::max(L.Width, R.Width), 3169 L.NonNegative && R.NonNegative); 3170 } 3171 3172 /// Returns the infinum of two ranges: i.e. their aggressive merge. 3173 static IntRange meet(IntRange L, IntRange R) { 3174 return IntRange(std::min(L.Width, R.Width), 3175 L.NonNegative || R.NonNegative); 3176 } 3177}; 3178 3179static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, 3180 unsigned MaxWidth) { 3181 if (value.isSigned() && value.isNegative()) 3182 return IntRange(value.getMinSignedBits(), false); 3183 3184 if (value.getBitWidth() > MaxWidth) 3185 value = value.trunc(MaxWidth); 3186 3187 // isNonNegative() just checks the sign bit without considering 3188 // signedness. 3189 return IntRange(value.getActiveBits(), true); 3190} 3191 3192static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 3193 unsigned MaxWidth) { 3194 if (result.isInt()) 3195 return GetValueRange(C, result.getInt(), MaxWidth); 3196 3197 if (result.isVector()) { 3198 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 3199 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 3200 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 3201 R = IntRange::join(R, El); 3202 } 3203 return R; 3204 } 3205 3206 if (result.isComplexInt()) { 3207 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 3208 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 3209 return IntRange::join(R, I); 3210 } 3211 3212 // This can happen with lossless casts to intptr_t of "based" lvalues. 3213 // Assume it might use arbitrary bits. 3214 // FIXME: The only reason we need to pass the type in here is to get 3215 // the sign right on this one case. It would be nice if APValue 3216 // preserved this. 3217 assert(result.isLValue() || result.isAddrLabelDiff()); 3218 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 3219} 3220 3221/// Pseudo-evaluate the given integer expression, estimating the 3222/// range of values it might take. 3223/// 3224/// \param MaxWidth - the width to which the value will be truncated 3225static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 3226 E = E->IgnoreParens(); 3227 3228 // Try a full evaluation first. 3229 Expr::EvalResult result; 3230 if (E->EvaluateAsRValue(result, C)) 3231 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 3232 3233 // I think we only want to look through implicit casts here; if the 3234 // user has an explicit widening cast, we should treat the value as 3235 // being of the new, wider type. 3236 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 3237 if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue) 3238 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 3239 3240 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 3241 3242 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 3243 3244 // Assume that non-integer casts can span the full range of the type. 3245 if (!isIntegerCast) 3246 return OutputTypeRange; 3247 3248 IntRange SubRange 3249 = GetExprRange(C, CE->getSubExpr(), 3250 std::min(MaxWidth, OutputTypeRange.Width)); 3251 3252 // Bail out if the subexpr's range is as wide as the cast type. 3253 if (SubRange.Width >= OutputTypeRange.Width) 3254 return OutputTypeRange; 3255 3256 // Otherwise, we take the smaller width, and we're non-negative if 3257 // either the output type or the subexpr is. 3258 return IntRange(SubRange.Width, 3259 SubRange.NonNegative || OutputTypeRange.NonNegative); 3260 } 3261 3262 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 3263 // If we can fold the condition, just take that operand. 3264 bool CondResult; 3265 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 3266 return GetExprRange(C, CondResult ? CO->getTrueExpr() 3267 : CO->getFalseExpr(), 3268 MaxWidth); 3269 3270 // Otherwise, conservatively merge. 3271 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 3272 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 3273 return IntRange::join(L, R); 3274 } 3275 3276 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 3277 switch (BO->getOpcode()) { 3278 3279 // Boolean-valued operations are single-bit and positive. 3280 case BO_LAnd: 3281 case BO_LOr: 3282 case BO_LT: 3283 case BO_GT: 3284 case BO_LE: 3285 case BO_GE: 3286 case BO_EQ: 3287 case BO_NE: 3288 return IntRange::forBoolType(); 3289 3290 // The type of the assignments is the type of the LHS, so the RHS 3291 // is not necessarily the same type. 3292 case BO_MulAssign: 3293 case BO_DivAssign: 3294 case BO_RemAssign: 3295 case BO_AddAssign: 3296 case BO_SubAssign: 3297 case BO_XorAssign: 3298 case BO_OrAssign: 3299 // TODO: bitfields? 3300 return IntRange::forValueOfType(C, E->getType()); 3301 3302 // Simple assignments just pass through the RHS, which will have 3303 // been coerced to the LHS type. 3304 case BO_Assign: 3305 // TODO: bitfields? 3306 return GetExprRange(C, BO->getRHS(), MaxWidth); 3307 3308 // Operations with opaque sources are black-listed. 3309 case BO_PtrMemD: 3310 case BO_PtrMemI: 3311 return IntRange::forValueOfType(C, E->getType()); 3312 3313 // Bitwise-and uses the *infinum* of the two source ranges. 3314 case BO_And: 3315 case BO_AndAssign: 3316 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 3317 GetExprRange(C, BO->getRHS(), MaxWidth)); 3318 3319 // Left shift gets black-listed based on a judgement call. 3320 case BO_Shl: 3321 // ...except that we want to treat '1 << (blah)' as logically 3322 // positive. It's an important idiom. 3323 if (IntegerLiteral *I 3324 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 3325 if (I->getValue() == 1) { 3326 IntRange R = IntRange::forValueOfType(C, E->getType()); 3327 return IntRange(R.Width, /*NonNegative*/ true); 3328 } 3329 } 3330 // fallthrough 3331 3332 case BO_ShlAssign: 3333 return IntRange::forValueOfType(C, E->getType()); 3334 3335 // Right shift by a constant can narrow its left argument. 3336 case BO_Shr: 3337 case BO_ShrAssign: { 3338 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3339 3340 // If the shift amount is a positive constant, drop the width by 3341 // that much. 3342 llvm::APSInt shift; 3343 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 3344 shift.isNonNegative()) { 3345 unsigned zext = shift.getZExtValue(); 3346 if (zext >= L.Width) 3347 L.Width = (L.NonNegative ? 0 : 1); 3348 else 3349 L.Width -= zext; 3350 } 3351 3352 return L; 3353 } 3354 3355 // Comma acts as its right operand. 3356 case BO_Comma: 3357 return GetExprRange(C, BO->getRHS(), MaxWidth); 3358 3359 // Black-list pointer subtractions. 3360 case BO_Sub: 3361 if (BO->getLHS()->getType()->isPointerType()) 3362 return IntRange::forValueOfType(C, E->getType()); 3363 break; 3364 3365 // The width of a division result is mostly determined by the size 3366 // of the LHS. 3367 case BO_Div: { 3368 // Don't 'pre-truncate' the operands. 3369 unsigned opWidth = C.getIntWidth(E->getType()); 3370 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3371 3372 // If the divisor is constant, use that. 3373 llvm::APSInt divisor; 3374 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 3375 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 3376 if (log2 >= L.Width) 3377 L.Width = (L.NonNegative ? 0 : 1); 3378 else 3379 L.Width = std::min(L.Width - log2, MaxWidth); 3380 return L; 3381 } 3382 3383 // Otherwise, just use the LHS's width. 3384 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3385 return IntRange(L.Width, L.NonNegative && R.NonNegative); 3386 } 3387 3388 // The result of a remainder can't be larger than the result of 3389 // either side. 3390 case BO_Rem: { 3391 // Don't 'pre-truncate' the operands. 3392 unsigned opWidth = C.getIntWidth(E->getType()); 3393 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3394 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3395 3396 IntRange meet = IntRange::meet(L, R); 3397 meet.Width = std::min(meet.Width, MaxWidth); 3398 return meet; 3399 } 3400 3401 // The default behavior is okay for these. 3402 case BO_Mul: 3403 case BO_Add: 3404 case BO_Xor: 3405 case BO_Or: 3406 break; 3407 } 3408 3409 // The default case is to treat the operation as if it were closed 3410 // on the narrowest type that encompasses both operands. 3411 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3412 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 3413 return IntRange::join(L, R); 3414 } 3415 3416 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 3417 switch (UO->getOpcode()) { 3418 // Boolean-valued operations are white-listed. 3419 case UO_LNot: 3420 return IntRange::forBoolType(); 3421 3422 // Operations with opaque sources are black-listed. 3423 case UO_Deref: 3424 case UO_AddrOf: // should be impossible 3425 return IntRange::forValueOfType(C, E->getType()); 3426 3427 default: 3428 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 3429 } 3430 } 3431 3432 if (dyn_cast<OffsetOfExpr>(E)) { 3433 IntRange::forValueOfType(C, E->getType()); 3434 } 3435 3436 if (FieldDecl *BitField = E->getBitField()) 3437 return IntRange(BitField->getBitWidthValue(C), 3438 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 3439 3440 return IntRange::forValueOfType(C, E->getType()); 3441} 3442 3443static IntRange GetExprRange(ASTContext &C, Expr *E) { 3444 return GetExprRange(C, E, C.getIntWidth(E->getType())); 3445} 3446 3447/// Checks whether the given value, which currently has the given 3448/// source semantics, has the same value when coerced through the 3449/// target semantics. 3450static bool IsSameFloatAfterCast(const llvm::APFloat &value, 3451 const llvm::fltSemantics &Src, 3452 const llvm::fltSemantics &Tgt) { 3453 llvm::APFloat truncated = value; 3454 3455 bool ignored; 3456 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 3457 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 3458 3459 return truncated.bitwiseIsEqual(value); 3460} 3461 3462/// Checks whether the given value, which currently has the given 3463/// source semantics, has the same value when coerced through the 3464/// target semantics. 3465/// 3466/// The value might be a vector of floats (or a complex number). 3467static bool IsSameFloatAfterCast(const APValue &value, 3468 const llvm::fltSemantics &Src, 3469 const llvm::fltSemantics &Tgt) { 3470 if (value.isFloat()) 3471 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 3472 3473 if (value.isVector()) { 3474 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 3475 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 3476 return false; 3477 return true; 3478 } 3479 3480 assert(value.isComplexFloat()); 3481 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 3482 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 3483} 3484 3485static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 3486 3487static bool IsZero(Sema &S, Expr *E) { 3488 // Suppress cases where we are comparing against an enum constant. 3489 if (const DeclRefExpr *DR = 3490 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 3491 if (isa<EnumConstantDecl>(DR->getDecl())) 3492 return false; 3493 3494 // Suppress cases where the '0' value is expanded from a macro. 3495 if (E->getLocStart().isMacroID()) 3496 return false; 3497 3498 llvm::APSInt Value; 3499 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 3500} 3501 3502static bool HasEnumType(Expr *E) { 3503 // Strip off implicit integral promotions. 3504 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 3505 if (ICE->getCastKind() != CK_IntegralCast && 3506 ICE->getCastKind() != CK_NoOp) 3507 break; 3508 E = ICE->getSubExpr(); 3509 } 3510 3511 return E->getType()->isEnumeralType(); 3512} 3513 3514static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 3515 BinaryOperatorKind op = E->getOpcode(); 3516 if (E->isValueDependent()) 3517 return; 3518 3519 if (op == BO_LT && IsZero(S, E->getRHS())) { 3520 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3521 << "< 0" << "false" << HasEnumType(E->getLHS()) 3522 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3523 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 3524 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3525 << ">= 0" << "true" << HasEnumType(E->getLHS()) 3526 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3527 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 3528 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3529 << "0 >" << "false" << HasEnumType(E->getRHS()) 3530 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3531 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 3532 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3533 << "0 <=" << "true" << HasEnumType(E->getRHS()) 3534 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3535 } 3536} 3537 3538/// Analyze the operands of the given comparison. Implements the 3539/// fallback case from AnalyzeComparison. 3540static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 3541 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3542 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3543} 3544 3545/// \brief Implements -Wsign-compare. 3546/// 3547/// \param E the binary operator to check for warnings 3548static void AnalyzeComparison(Sema &S, BinaryOperator *E) { 3549 // The type the comparison is being performed in. 3550 QualType T = E->getLHS()->getType(); 3551 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 3552 && "comparison with mismatched types"); 3553 3554 // We don't do anything special if this isn't an unsigned integral 3555 // comparison: we're only interested in integral comparisons, and 3556 // signed comparisons only happen in cases we don't care to warn about. 3557 // 3558 // We also don't care about value-dependent expressions or expressions 3559 // whose result is a constant. 3560 if (!T->hasUnsignedIntegerRepresentation() 3561 || E->isValueDependent() || E->isIntegerConstantExpr(S.Context)) 3562 return AnalyzeImpConvsInComparison(S, E); 3563 3564 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 3565 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 3566 3567 // Check to see if one of the (unmodified) operands is of different 3568 // signedness. 3569 Expr *signedOperand, *unsignedOperand; 3570 if (LHS->getType()->hasSignedIntegerRepresentation()) { 3571 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 3572 "unsigned comparison between two signed integer expressions?"); 3573 signedOperand = LHS; 3574 unsignedOperand = RHS; 3575 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 3576 signedOperand = RHS; 3577 unsignedOperand = LHS; 3578 } else { 3579 CheckTrivialUnsignedComparison(S, E); 3580 return AnalyzeImpConvsInComparison(S, E); 3581 } 3582 3583 // Otherwise, calculate the effective range of the signed operand. 3584 IntRange signedRange = GetExprRange(S.Context, signedOperand); 3585 3586 // Go ahead and analyze implicit conversions in the operands. Note 3587 // that we skip the implicit conversions on both sides. 3588 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 3589 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 3590 3591 // If the signed range is non-negative, -Wsign-compare won't fire, 3592 // but we should still check for comparisons which are always true 3593 // or false. 3594 if (signedRange.NonNegative) 3595 return CheckTrivialUnsignedComparison(S, E); 3596 3597 // For (in)equality comparisons, if the unsigned operand is a 3598 // constant which cannot collide with a overflowed signed operand, 3599 // then reinterpreting the signed operand as unsigned will not 3600 // change the result of the comparison. 3601 if (E->isEqualityOp()) { 3602 unsigned comparisonWidth = S.Context.getIntWidth(T); 3603 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 3604 3605 // We should never be unable to prove that the unsigned operand is 3606 // non-negative. 3607 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 3608 3609 if (unsignedRange.Width < comparisonWidth) 3610 return; 3611 } 3612 3613 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison) 3614 << LHS->getType() << RHS->getType() 3615 << LHS->getSourceRange() << RHS->getSourceRange(); 3616} 3617 3618/// Analyzes an attempt to assign the given value to a bitfield. 3619/// 3620/// Returns true if there was something fishy about the attempt. 3621static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 3622 SourceLocation InitLoc) { 3623 assert(Bitfield->isBitField()); 3624 if (Bitfield->isInvalidDecl()) 3625 return false; 3626 3627 // White-list bool bitfields. 3628 if (Bitfield->getType()->isBooleanType()) 3629 return false; 3630 3631 // Ignore value- or type-dependent expressions. 3632 if (Bitfield->getBitWidth()->isValueDependent() || 3633 Bitfield->getBitWidth()->isTypeDependent() || 3634 Init->isValueDependent() || 3635 Init->isTypeDependent()) 3636 return false; 3637 3638 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 3639 3640 llvm::APSInt Value; 3641 if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects)) 3642 return false; 3643 3644 unsigned OriginalWidth = Value.getBitWidth(); 3645 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 3646 3647 if (OriginalWidth <= FieldWidth) 3648 return false; 3649 3650 // Compute the value which the bitfield will contain. 3651 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 3652 TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType()); 3653 3654 // Check whether the stored value is equal to the original value. 3655 TruncatedValue = TruncatedValue.extend(OriginalWidth); 3656 if (Value == TruncatedValue) 3657 return false; 3658 3659 // Special-case bitfields of width 1: booleans are naturally 0/1, and 3660 // therefore don't strictly fit into a bitfield of width 1. 3661 if (FieldWidth == 1 && Value.getBoolValue() == TruncatedValue.getBoolValue()) 3662 return false; 3663 3664 std::string PrettyValue = Value.toString(10); 3665 std::string PrettyTrunc = TruncatedValue.toString(10); 3666 3667 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 3668 << PrettyValue << PrettyTrunc << OriginalInit->getType() 3669 << Init->getSourceRange(); 3670 3671 return true; 3672} 3673 3674/// Analyze the given simple or compound assignment for warning-worthy 3675/// operations. 3676static void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 3677 // Just recurse on the LHS. 3678 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3679 3680 // We want to recurse on the RHS as normal unless we're assigning to 3681 // a bitfield. 3682 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 3683 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 3684 E->getOperatorLoc())) { 3685 // Recurse, ignoring any implicit conversions on the RHS. 3686 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 3687 E->getOperatorLoc()); 3688 } 3689 } 3690 3691 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3692} 3693 3694/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3695static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 3696 SourceLocation CContext, unsigned diag) { 3697 S.Diag(E->getExprLoc(), diag) 3698 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 3699} 3700 3701/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3702static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, 3703 SourceLocation CContext, unsigned diag) { 3704 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag); 3705} 3706 3707/// Diagnose an implicit cast from a literal expression. Does not warn when the 3708/// cast wouldn't lose information. 3709void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 3710 SourceLocation CContext) { 3711 // Try to convert the literal exactly to an integer. If we can, don't warn. 3712 bool isExact = false; 3713 const llvm::APFloat &Value = FL->getValue(); 3714 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 3715 T->hasUnsignedIntegerRepresentation()); 3716 if (Value.convertToInteger(IntegerValue, 3717 llvm::APFloat::rmTowardZero, &isExact) 3718 == llvm::APFloat::opOK && isExact) 3719 return; 3720 3721 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 3722 << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext); 3723} 3724 3725std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 3726 if (!Range.Width) return "0"; 3727 3728 llvm::APSInt ValueInRange = Value; 3729 ValueInRange.setIsSigned(!Range.NonNegative); 3730 ValueInRange = ValueInRange.trunc(Range.Width); 3731 return ValueInRange.toString(10); 3732} 3733 3734void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 3735 SourceLocation CC, bool *ICContext = 0) { 3736 if (E->isTypeDependent() || E->isValueDependent()) return; 3737 3738 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 3739 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 3740 if (Source == Target) return; 3741 if (Target->isDependentType()) return; 3742 3743 // If the conversion context location is invalid don't complain. We also 3744 // don't want to emit a warning if the issue occurs from the expansion of 3745 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 3746 // delay this check as long as possible. Once we detect we are in that 3747 // scenario, we just return. 3748 if (CC.isInvalid()) 3749 return; 3750 3751 // Diagnose implicit casts to bool. 3752 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 3753 if (isa<StringLiteral>(E)) 3754 // Warn on string literal to bool. Checks for string literals in logical 3755 // expressions, for instances, assert(0 && "error here"), is prevented 3756 // by a check in AnalyzeImplicitConversions(). 3757 return DiagnoseImpCast(S, E, T, CC, 3758 diag::warn_impcast_string_literal_to_bool); 3759 if (Source->isFunctionType()) { 3760 // Warn on function to bool. Checks free functions and static member 3761 // functions. Weakly imported functions are excluded from the check, 3762 // since it's common to test their value to check whether the linker 3763 // found a definition for them. 3764 ValueDecl *D = 0; 3765 if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) { 3766 D = R->getDecl(); 3767 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 3768 D = M->getMemberDecl(); 3769 } 3770 3771 if (D && !D->isWeak()) { 3772 if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) { 3773 S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool) 3774 << F << E->getSourceRange() << SourceRange(CC); 3775 S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence) 3776 << FixItHint::CreateInsertion(E->getExprLoc(), "&"); 3777 QualType ReturnType; 3778 UnresolvedSet<4> NonTemplateOverloads; 3779 S.isExprCallable(*E, ReturnType, NonTemplateOverloads); 3780 if (!ReturnType.isNull() 3781 && ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 3782 S.Diag(E->getExprLoc(), diag::note_function_to_bool_call) 3783 << FixItHint::CreateInsertion( 3784 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()"); 3785 return; 3786 } 3787 } 3788 } 3789 return; // Other casts to bool are not checked. 3790 } 3791 3792 // Strip vector types. 3793 if (isa<VectorType>(Source)) { 3794 if (!isa<VectorType>(Target)) { 3795 if (S.SourceMgr.isInSystemMacro(CC)) 3796 return; 3797 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 3798 } 3799 3800 // If the vector cast is cast between two vectors of the same size, it is 3801 // a bitcast, not a conversion. 3802 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 3803 return; 3804 3805 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 3806 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 3807 } 3808 3809 // Strip complex types. 3810 if (isa<ComplexType>(Source)) { 3811 if (!isa<ComplexType>(Target)) { 3812 if (S.SourceMgr.isInSystemMacro(CC)) 3813 return; 3814 3815 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 3816 } 3817 3818 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 3819 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 3820 } 3821 3822 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 3823 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 3824 3825 // If the source is floating point... 3826 if (SourceBT && SourceBT->isFloatingPoint()) { 3827 // ...and the target is floating point... 3828 if (TargetBT && TargetBT->isFloatingPoint()) { 3829 // ...then warn if we're dropping FP rank. 3830 3831 // Builtin FP kinds are ordered by increasing FP rank. 3832 if (SourceBT->getKind() > TargetBT->getKind()) { 3833 // Don't warn about float constants that are precisely 3834 // representable in the target type. 3835 Expr::EvalResult result; 3836 if (E->EvaluateAsRValue(result, S.Context)) { 3837 // Value might be a float, a float vector, or a float complex. 3838 if (IsSameFloatAfterCast(result.Val, 3839 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 3840 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 3841 return; 3842 } 3843 3844 if (S.SourceMgr.isInSystemMacro(CC)) 3845 return; 3846 3847 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 3848 } 3849 return; 3850 } 3851 3852 // If the target is integral, always warn. 3853 if ((TargetBT && TargetBT->isInteger())) { 3854 if (S.SourceMgr.isInSystemMacro(CC)) 3855 return; 3856 3857 Expr *InnerE = E->IgnoreParenImpCasts(); 3858 // We also want to warn on, e.g., "int i = -1.234" 3859 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 3860 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 3861 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 3862 3863 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 3864 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 3865 } else { 3866 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 3867 } 3868 } 3869 3870 return; 3871 } 3872 3873 if (!Source->isIntegerType() || !Target->isIntegerType()) 3874 return; 3875 3876 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 3877 == Expr::NPCK_GNUNull) && Target->isIntegerType()) { 3878 S.Diag(E->getExprLoc(), diag::warn_impcast_null_pointer_to_integer) 3879 << E->getSourceRange() << clang::SourceRange(CC); 3880 return; 3881 } 3882 3883 IntRange SourceRange = GetExprRange(S.Context, E); 3884 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 3885 3886 if (SourceRange.Width > TargetRange.Width) { 3887 // If the source is a constant, use a default-on diagnostic. 3888 // TODO: this should happen for bitfield stores, too. 3889 llvm::APSInt Value(32); 3890 if (E->isIntegerConstantExpr(Value, S.Context)) { 3891 if (S.SourceMgr.isInSystemMacro(CC)) 3892 return; 3893 3894 std::string PrettySourceValue = Value.toString(10); 3895 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 3896 3897 S.DiagRuntimeBehavior(E->getExprLoc(), E, 3898 S.PDiag(diag::warn_impcast_integer_precision_constant) 3899 << PrettySourceValue << PrettyTargetValue 3900 << E->getType() << T << E->getSourceRange() 3901 << clang::SourceRange(CC)); 3902 return; 3903 } 3904 3905 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 3906 if (S.SourceMgr.isInSystemMacro(CC)) 3907 return; 3908 3909 if (SourceRange.Width == 64 && TargetRange.Width == 32) 3910 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32); 3911 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 3912 } 3913 3914 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 3915 (!TargetRange.NonNegative && SourceRange.NonNegative && 3916 SourceRange.Width == TargetRange.Width)) { 3917 3918 if (S.SourceMgr.isInSystemMacro(CC)) 3919 return; 3920 3921 unsigned DiagID = diag::warn_impcast_integer_sign; 3922 3923 // Traditionally, gcc has warned about this under -Wsign-compare. 3924 // We also want to warn about it in -Wconversion. 3925 // So if -Wconversion is off, use a completely identical diagnostic 3926 // in the sign-compare group. 3927 // The conditional-checking code will 3928 if (ICContext) { 3929 DiagID = diag::warn_impcast_integer_sign_conditional; 3930 *ICContext = true; 3931 } 3932 3933 return DiagnoseImpCast(S, E, T, CC, DiagID); 3934 } 3935 3936 // Diagnose conversions between different enumeration types. 3937 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 3938 // type, to give us better diagnostics. 3939 QualType SourceType = E->getType(); 3940 if (!S.getLangOptions().CPlusPlus) { 3941 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 3942 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 3943 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 3944 SourceType = S.Context.getTypeDeclType(Enum); 3945 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 3946 } 3947 } 3948 3949 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 3950 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 3951 if ((SourceEnum->getDecl()->getIdentifier() || 3952 SourceEnum->getDecl()->getTypedefNameForAnonDecl()) && 3953 (TargetEnum->getDecl()->getIdentifier() || 3954 TargetEnum->getDecl()->getTypedefNameForAnonDecl()) && 3955 SourceEnum != TargetEnum) { 3956 if (S.SourceMgr.isInSystemMacro(CC)) 3957 return; 3958 3959 return DiagnoseImpCast(S, E, SourceType, T, CC, 3960 diag::warn_impcast_different_enum_types); 3961 } 3962 3963 return; 3964} 3965 3966void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 3967 3968void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 3969 SourceLocation CC, bool &ICContext) { 3970 E = E->IgnoreParenImpCasts(); 3971 3972 if (isa<ConditionalOperator>(E)) 3973 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 3974 3975 AnalyzeImplicitConversions(S, E, CC); 3976 if (E->getType() != T) 3977 return CheckImplicitConversion(S, E, T, CC, &ICContext); 3978 return; 3979} 3980 3981void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 3982 SourceLocation CC = E->getQuestionLoc(); 3983 3984 AnalyzeImplicitConversions(S, E->getCond(), CC); 3985 3986 bool Suspicious = false; 3987 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 3988 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 3989 3990 // If -Wconversion would have warned about either of the candidates 3991 // for a signedness conversion to the context type... 3992 if (!Suspicious) return; 3993 3994 // ...but it's currently ignored... 3995 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 3996 CC)) 3997 return; 3998 3999 // ...then check whether it would have warned about either of the 4000 // candidates for a signedness conversion to the condition type. 4001 if (E->getType() == T) return; 4002 4003 Suspicious = false; 4004 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 4005 E->getType(), CC, &Suspicious); 4006 if (!Suspicious) 4007 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 4008 E->getType(), CC, &Suspicious); 4009} 4010 4011/// AnalyzeImplicitConversions - Find and report any interesting 4012/// implicit conversions in the given expression. There are a couple 4013/// of competing diagnostics here, -Wconversion and -Wsign-compare. 4014void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 4015 QualType T = OrigE->getType(); 4016 Expr *E = OrigE->IgnoreParenImpCasts(); 4017 4018 if (E->isTypeDependent() || E->isValueDependent()) 4019 return; 4020 4021 // For conditional operators, we analyze the arguments as if they 4022 // were being fed directly into the output. 4023 if (isa<ConditionalOperator>(E)) { 4024 ConditionalOperator *CO = cast<ConditionalOperator>(E); 4025 CheckConditionalOperator(S, CO, T); 4026 return; 4027 } 4028 4029 // Go ahead and check any implicit conversions we might have skipped. 4030 // The non-canonical typecheck is just an optimization; 4031 // CheckImplicitConversion will filter out dead implicit conversions. 4032 if (E->getType() != T) 4033 CheckImplicitConversion(S, E, T, CC); 4034 4035 // Now continue drilling into this expression. 4036 4037 // Skip past explicit casts. 4038 if (isa<ExplicitCastExpr>(E)) { 4039 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 4040 return AnalyzeImplicitConversions(S, E, CC); 4041 } 4042 4043 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 4044 // Do a somewhat different check with comparison operators. 4045 if (BO->isComparisonOp()) 4046 return AnalyzeComparison(S, BO); 4047 4048 // And with simple assignments. 4049 if (BO->getOpcode() == BO_Assign) 4050 return AnalyzeAssignment(S, BO); 4051 } 4052 4053 // These break the otherwise-useful invariant below. Fortunately, 4054 // we don't really need to recurse into them, because any internal 4055 // expressions should have been analyzed already when they were 4056 // built into statements. 4057 if (isa<StmtExpr>(E)) return; 4058 4059 // Don't descend into unevaluated contexts. 4060 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 4061 4062 // Now just recurse over the expression's children. 4063 CC = E->getExprLoc(); 4064 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 4065 bool IsLogicalOperator = BO && BO->isLogicalOp(); 4066 for (Stmt::child_range I = E->children(); I; ++I) { 4067 Expr *ChildExpr = cast<Expr>(*I); 4068 if (IsLogicalOperator && 4069 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 4070 // Ignore checking string literals that are in logical operators. 4071 continue; 4072 AnalyzeImplicitConversions(S, ChildExpr, CC); 4073 } 4074} 4075 4076} // end anonymous namespace 4077 4078/// Diagnoses "dangerous" implicit conversions within the given 4079/// expression (which is a full expression). Implements -Wconversion 4080/// and -Wsign-compare. 4081/// 4082/// \param CC the "context" location of the implicit conversion, i.e. 4083/// the most location of the syntactic entity requiring the implicit 4084/// conversion 4085void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 4086 // Don't diagnose in unevaluated contexts. 4087 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 4088 return; 4089 4090 // Don't diagnose for value- or type-dependent expressions. 4091 if (E->isTypeDependent() || E->isValueDependent()) 4092 return; 4093 4094 // Check for array bounds violations in cases where the check isn't triggered 4095 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 4096 // ArraySubscriptExpr is on the RHS of a variable initialization. 4097 CheckArrayAccess(E); 4098 4099 // This is not the right CC for (e.g.) a variable initialization. 4100 AnalyzeImplicitConversions(*this, E, CC); 4101} 4102 4103void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 4104 FieldDecl *BitField, 4105 Expr *Init) { 4106 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 4107} 4108 4109/// CheckParmsForFunctionDef - Check that the parameters of the given 4110/// function are appropriate for the definition of a function. This 4111/// takes care of any checks that cannot be performed on the 4112/// declaration itself, e.g., that the types of each of the function 4113/// parameters are complete. 4114bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 4115 bool CheckParameterNames) { 4116 bool HasInvalidParm = false; 4117 for (; P != PEnd; ++P) { 4118 ParmVarDecl *Param = *P; 4119 4120 // C99 6.7.5.3p4: the parameters in a parameter type list in a 4121 // function declarator that is part of a function definition of 4122 // that function shall not have incomplete type. 4123 // 4124 // This is also C++ [dcl.fct]p6. 4125 if (!Param->isInvalidDecl() && 4126 RequireCompleteType(Param->getLocation(), Param->getType(), 4127 diag::err_typecheck_decl_incomplete_type)) { 4128 Param->setInvalidDecl(); 4129 HasInvalidParm = true; 4130 } 4131 4132 // C99 6.9.1p5: If the declarator includes a parameter type list, the 4133 // declaration of each parameter shall include an identifier. 4134 if (CheckParameterNames && 4135 Param->getIdentifier() == 0 && 4136 !Param->isImplicit() && 4137 !getLangOptions().CPlusPlus) 4138 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 4139 4140 // C99 6.7.5.3p12: 4141 // If the function declarator is not part of a definition of that 4142 // function, parameters may have incomplete type and may use the [*] 4143 // notation in their sequences of declarator specifiers to specify 4144 // variable length array types. 4145 QualType PType = Param->getOriginalType(); 4146 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 4147 if (AT->getSizeModifier() == ArrayType::Star) { 4148 // FIXME: This diagnosic should point the the '[*]' if source-location 4149 // information is added for it. 4150 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 4151 } 4152 } 4153 } 4154 4155 return HasInvalidParm; 4156} 4157 4158/// CheckCastAlign - Implements -Wcast-align, which warns when a 4159/// pointer cast increases the alignment requirements. 4160void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 4161 // This is actually a lot of work to potentially be doing on every 4162 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 4163 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 4164 TRange.getBegin()) 4165 == DiagnosticsEngine::Ignored) 4166 return; 4167 4168 // Ignore dependent types. 4169 if (T->isDependentType() || Op->getType()->isDependentType()) 4170 return; 4171 4172 // Require that the destination be a pointer type. 4173 const PointerType *DestPtr = T->getAs<PointerType>(); 4174 if (!DestPtr) return; 4175 4176 // If the destination has alignment 1, we're done. 4177 QualType DestPointee = DestPtr->getPointeeType(); 4178 if (DestPointee->isIncompleteType()) return; 4179 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 4180 if (DestAlign.isOne()) return; 4181 4182 // Require that the source be a pointer type. 4183 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 4184 if (!SrcPtr) return; 4185 QualType SrcPointee = SrcPtr->getPointeeType(); 4186 4187 // Whitelist casts from cv void*. We already implicitly 4188 // whitelisted casts to cv void*, since they have alignment 1. 4189 // Also whitelist casts involving incomplete types, which implicitly 4190 // includes 'void'. 4191 if (SrcPointee->isIncompleteType()) return; 4192 4193 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 4194 if (SrcAlign >= DestAlign) return; 4195 4196 Diag(TRange.getBegin(), diag::warn_cast_align) 4197 << Op->getType() << T 4198 << static_cast<unsigned>(SrcAlign.getQuantity()) 4199 << static_cast<unsigned>(DestAlign.getQuantity()) 4200 << TRange << Op->getSourceRange(); 4201} 4202 4203static const Type* getElementType(const Expr *BaseExpr) { 4204 const Type* EltType = BaseExpr->getType().getTypePtr(); 4205 if (EltType->isAnyPointerType()) 4206 return EltType->getPointeeType().getTypePtr(); 4207 else if (EltType->isArrayType()) 4208 return EltType->getBaseElementTypeUnsafe(); 4209 return EltType; 4210} 4211 4212/// \brief Check whether this array fits the idiom of a size-one tail padded 4213/// array member of a struct. 4214/// 4215/// We avoid emitting out-of-bounds access warnings for such arrays as they are 4216/// commonly used to emulate flexible arrays in C89 code. 4217static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 4218 const NamedDecl *ND) { 4219 if (Size != 1 || !ND) return false; 4220 4221 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 4222 if (!FD) return false; 4223 4224 // Don't consider sizes resulting from macro expansions or template argument 4225 // substitution to form C89 tail-padded arrays. 4226 ConstantArrayTypeLoc TL = 4227 cast<ConstantArrayTypeLoc>(FD->getTypeSourceInfo()->getTypeLoc()); 4228 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(TL.getSizeExpr()); 4229 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 4230 return false; 4231 4232 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 4233 if (!RD) return false; 4234 if (RD->isUnion()) return false; 4235 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 4236 if (!CRD->isStandardLayout()) return false; 4237 } 4238 4239 // See if this is the last field decl in the record. 4240 const Decl *D = FD; 4241 while ((D = D->getNextDeclInContext())) 4242 if (isa<FieldDecl>(D)) 4243 return false; 4244 return true; 4245} 4246 4247void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 4248 const ArraySubscriptExpr *ASE, 4249 bool AllowOnePastEnd, bool IndexNegated) { 4250 IndexExpr = IndexExpr->IgnoreParenCasts(); 4251 if (IndexExpr->isValueDependent()) 4252 return; 4253 4254 const Type *EffectiveType = getElementType(BaseExpr); 4255 BaseExpr = BaseExpr->IgnoreParenCasts(); 4256 const ConstantArrayType *ArrayTy = 4257 Context.getAsConstantArrayType(BaseExpr->getType()); 4258 if (!ArrayTy) 4259 return; 4260 4261 llvm::APSInt index; 4262 if (!IndexExpr->EvaluateAsInt(index, Context)) 4263 return; 4264 if (IndexNegated) 4265 index = -index; 4266 4267 const NamedDecl *ND = NULL; 4268 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4269 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 4270 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 4271 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 4272 4273 if (index.isUnsigned() || !index.isNegative()) { 4274 llvm::APInt size = ArrayTy->getSize(); 4275 if (!size.isStrictlyPositive()) 4276 return; 4277 4278 const Type* BaseType = getElementType(BaseExpr); 4279 if (BaseType != EffectiveType) { 4280 // Make sure we're comparing apples to apples when comparing index to size 4281 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 4282 uint64_t array_typesize = Context.getTypeSize(BaseType); 4283 // Handle ptrarith_typesize being zero, such as when casting to void* 4284 if (!ptrarith_typesize) ptrarith_typesize = 1; 4285 if (ptrarith_typesize != array_typesize) { 4286 // There's a cast to a different size type involved 4287 uint64_t ratio = array_typesize / ptrarith_typesize; 4288 // TODO: Be smarter about handling cases where array_typesize is not a 4289 // multiple of ptrarith_typesize 4290 if (ptrarith_typesize * ratio == array_typesize) 4291 size *= llvm::APInt(size.getBitWidth(), ratio); 4292 } 4293 } 4294 4295 if (size.getBitWidth() > index.getBitWidth()) 4296 index = index.sext(size.getBitWidth()); 4297 else if (size.getBitWidth() < index.getBitWidth()) 4298 size = size.sext(index.getBitWidth()); 4299 4300 // For array subscripting the index must be less than size, but for pointer 4301 // arithmetic also allow the index (offset) to be equal to size since 4302 // computing the next address after the end of the array is legal and 4303 // commonly done e.g. in C++ iterators and range-based for loops. 4304 if (AllowOnePastEnd ? index.sle(size) : index.slt(size)) 4305 return; 4306 4307 // Also don't warn for arrays of size 1 which are members of some 4308 // structure. These are often used to approximate flexible arrays in C89 4309 // code. 4310 if (IsTailPaddedMemberArray(*this, size, ND)) 4311 return; 4312 4313 // Suppress the warning if the subscript expression (as identified by the 4314 // ']' location) and the index expression are both from macro expansions 4315 // within a system header. 4316 if (ASE) { 4317 SourceLocation RBracketLoc = SourceMgr.getSpellingLoc( 4318 ASE->getRBracketLoc()); 4319 if (SourceMgr.isInSystemHeader(RBracketLoc)) { 4320 SourceLocation IndexLoc = SourceMgr.getSpellingLoc( 4321 IndexExpr->getLocStart()); 4322 if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc)) 4323 return; 4324 } 4325 } 4326 4327 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 4328 if (ASE) 4329 DiagID = diag::warn_array_index_exceeds_bounds; 4330 4331 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4332 PDiag(DiagID) << index.toString(10, true) 4333 << size.toString(10, true) 4334 << (unsigned)size.getLimitedValue(~0U) 4335 << IndexExpr->getSourceRange()); 4336 } else { 4337 unsigned DiagID = diag::warn_array_index_precedes_bounds; 4338 if (!ASE) { 4339 DiagID = diag::warn_ptr_arith_precedes_bounds; 4340 if (index.isNegative()) index = -index; 4341 } 4342 4343 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4344 PDiag(DiagID) << index.toString(10, true) 4345 << IndexExpr->getSourceRange()); 4346 } 4347 4348 if (!ND) { 4349 // Try harder to find a NamedDecl to point at in the note. 4350 while (const ArraySubscriptExpr *ASE = 4351 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 4352 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 4353 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4354 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 4355 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 4356 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 4357 } 4358 4359 if (ND) 4360 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 4361 PDiag(diag::note_array_index_out_of_bounds) 4362 << ND->getDeclName()); 4363} 4364 4365void Sema::CheckArrayAccess(const Expr *expr) { 4366 int AllowOnePastEnd = 0; 4367 while (expr) { 4368 expr = expr->IgnoreParenImpCasts(); 4369 switch (expr->getStmtClass()) { 4370 case Stmt::ArraySubscriptExprClass: { 4371 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 4372 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE, 4373 AllowOnePastEnd > 0); 4374 return; 4375 } 4376 case Stmt::UnaryOperatorClass: { 4377 // Only unwrap the * and & unary operators 4378 const UnaryOperator *UO = cast<UnaryOperator>(expr); 4379 expr = UO->getSubExpr(); 4380 switch (UO->getOpcode()) { 4381 case UO_AddrOf: 4382 AllowOnePastEnd++; 4383 break; 4384 case UO_Deref: 4385 AllowOnePastEnd--; 4386 break; 4387 default: 4388 return; 4389 } 4390 break; 4391 } 4392 case Stmt::ConditionalOperatorClass: { 4393 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 4394 if (const Expr *lhs = cond->getLHS()) 4395 CheckArrayAccess(lhs); 4396 if (const Expr *rhs = cond->getRHS()) 4397 CheckArrayAccess(rhs); 4398 return; 4399 } 4400 default: 4401 return; 4402 } 4403 } 4404} 4405 4406//===--- CHECK: Objective-C retain cycles ----------------------------------// 4407 4408namespace { 4409 struct RetainCycleOwner { 4410 RetainCycleOwner() : Variable(0), Indirect(false) {} 4411 VarDecl *Variable; 4412 SourceRange Range; 4413 SourceLocation Loc; 4414 bool Indirect; 4415 4416 void setLocsFrom(Expr *e) { 4417 Loc = e->getExprLoc(); 4418 Range = e->getSourceRange(); 4419 } 4420 }; 4421} 4422 4423/// Consider whether capturing the given variable can possibly lead to 4424/// a retain cycle. 4425static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 4426 // In ARC, it's captured strongly iff the variable has __strong 4427 // lifetime. In MRR, it's captured strongly if the variable is 4428 // __block and has an appropriate type. 4429 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4430 return false; 4431 4432 owner.Variable = var; 4433 owner.setLocsFrom(ref); 4434 return true; 4435} 4436 4437static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) { 4438 while (true) { 4439 e = e->IgnoreParens(); 4440 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 4441 switch (cast->getCastKind()) { 4442 case CK_BitCast: 4443 case CK_LValueBitCast: 4444 case CK_LValueToRValue: 4445 case CK_ARCReclaimReturnedObject: 4446 e = cast->getSubExpr(); 4447 continue; 4448 4449 default: 4450 return false; 4451 } 4452 } 4453 4454 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 4455 ObjCIvarDecl *ivar = ref->getDecl(); 4456 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4457 return false; 4458 4459 // Try to find a retain cycle in the base. 4460 if (!findRetainCycleOwner(S, ref->getBase(), owner)) 4461 return false; 4462 4463 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 4464 owner.Indirect = true; 4465 return true; 4466 } 4467 4468 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 4469 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 4470 if (!var) return false; 4471 return considerVariable(var, ref, owner); 4472 } 4473 4474 if (BlockDeclRefExpr *ref = dyn_cast<BlockDeclRefExpr>(e)) { 4475 owner.Variable = ref->getDecl(); 4476 owner.setLocsFrom(ref); 4477 return true; 4478 } 4479 4480 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 4481 if (member->isArrow()) return false; 4482 4483 // Don't count this as an indirect ownership. 4484 e = member->getBase(); 4485 continue; 4486 } 4487 4488 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 4489 // Only pay attention to pseudo-objects on property references. 4490 ObjCPropertyRefExpr *pre 4491 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 4492 ->IgnoreParens()); 4493 if (!pre) return false; 4494 if (pre->isImplicitProperty()) return false; 4495 ObjCPropertyDecl *property = pre->getExplicitProperty(); 4496 if (!property->isRetaining() && 4497 !(property->getPropertyIvarDecl() && 4498 property->getPropertyIvarDecl()->getType() 4499 .getObjCLifetime() == Qualifiers::OCL_Strong)) 4500 return false; 4501 4502 owner.Indirect = true; 4503 if (pre->isSuperReceiver()) { 4504 owner.Variable = S.getCurMethodDecl()->getSelfDecl(); 4505 if (!owner.Variable) 4506 return false; 4507 owner.Loc = pre->getLocation(); 4508 owner.Range = pre->getSourceRange(); 4509 return true; 4510 } 4511 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 4512 ->getSourceExpr()); 4513 continue; 4514 } 4515 4516 // Array ivars? 4517 4518 return false; 4519 } 4520} 4521 4522namespace { 4523 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 4524 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 4525 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 4526 Variable(variable), Capturer(0) {} 4527 4528 VarDecl *Variable; 4529 Expr *Capturer; 4530 4531 void VisitDeclRefExpr(DeclRefExpr *ref) { 4532 if (ref->getDecl() == Variable && !Capturer) 4533 Capturer = ref; 4534 } 4535 4536 void VisitBlockDeclRefExpr(BlockDeclRefExpr *ref) { 4537 if (ref->getDecl() == Variable && !Capturer) 4538 Capturer = ref; 4539 } 4540 4541 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 4542 if (Capturer) return; 4543 Visit(ref->getBase()); 4544 if (Capturer && ref->isFreeIvar()) 4545 Capturer = ref; 4546 } 4547 4548 void VisitBlockExpr(BlockExpr *block) { 4549 // Look inside nested blocks 4550 if (block->getBlockDecl()->capturesVariable(Variable)) 4551 Visit(block->getBlockDecl()->getBody()); 4552 } 4553 }; 4554} 4555 4556/// Check whether the given argument is a block which captures a 4557/// variable. 4558static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 4559 assert(owner.Variable && owner.Loc.isValid()); 4560 4561 e = e->IgnoreParenCasts(); 4562 BlockExpr *block = dyn_cast<BlockExpr>(e); 4563 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 4564 return 0; 4565 4566 FindCaptureVisitor visitor(S.Context, owner.Variable); 4567 visitor.Visit(block->getBlockDecl()->getBody()); 4568 return visitor.Capturer; 4569} 4570 4571static void diagnoseRetainCycle(Sema &S, Expr *capturer, 4572 RetainCycleOwner &owner) { 4573 assert(capturer); 4574 assert(owner.Variable && owner.Loc.isValid()); 4575 4576 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 4577 << owner.Variable << capturer->getSourceRange(); 4578 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 4579 << owner.Indirect << owner.Range; 4580} 4581 4582/// Check for a keyword selector that starts with the word 'add' or 4583/// 'set'. 4584static bool isSetterLikeSelector(Selector sel) { 4585 if (sel.isUnarySelector()) return false; 4586 4587 StringRef str = sel.getNameForSlot(0); 4588 while (!str.empty() && str.front() == '_') str = str.substr(1); 4589 if (str.startswith("set")) 4590 str = str.substr(3); 4591 else if (str.startswith("add")) { 4592 // Specially whitelist 'addOperationWithBlock:'. 4593 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 4594 return false; 4595 str = str.substr(3); 4596 } 4597 else 4598 return false; 4599 4600 if (str.empty()) return true; 4601 return !islower(str.front()); 4602} 4603 4604/// Check a message send to see if it's likely to cause a retain cycle. 4605void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 4606 // Only check instance methods whose selector looks like a setter. 4607 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 4608 return; 4609 4610 // Try to find a variable that the receiver is strongly owned by. 4611 RetainCycleOwner owner; 4612 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 4613 if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner)) 4614 return; 4615 } else { 4616 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 4617 owner.Variable = getCurMethodDecl()->getSelfDecl(); 4618 owner.Loc = msg->getSuperLoc(); 4619 owner.Range = msg->getSuperLoc(); 4620 } 4621 4622 // Check whether the receiver is captured by any of the arguments. 4623 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 4624 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 4625 return diagnoseRetainCycle(*this, capturer, owner); 4626} 4627 4628/// Check a property assign to see if it's likely to cause a retain cycle. 4629void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 4630 RetainCycleOwner owner; 4631 if (!findRetainCycleOwner(*this, receiver, owner)) 4632 return; 4633 4634 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 4635 diagnoseRetainCycle(*this, capturer, owner); 4636} 4637 4638bool Sema::checkUnsafeAssigns(SourceLocation Loc, 4639 QualType LHS, Expr *RHS) { 4640 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 4641 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 4642 return false; 4643 // strip off any implicit cast added to get to the one arc-specific 4644 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4645 if (cast->getCastKind() == CK_ARCConsumeObject) { 4646 Diag(Loc, diag::warn_arc_retained_assign) 4647 << (LT == Qualifiers::OCL_ExplicitNone) 4648 << RHS->getSourceRange(); 4649 return true; 4650 } 4651 RHS = cast->getSubExpr(); 4652 } 4653 return false; 4654} 4655 4656void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 4657 Expr *LHS, Expr *RHS) { 4658 QualType LHSType; 4659 // PropertyRef on LHS type need be directly obtained from 4660 // its declaration as it has a PsuedoType. 4661 ObjCPropertyRefExpr *PRE 4662 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens()); 4663 if (PRE && !PRE->isImplicitProperty()) { 4664 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 4665 if (PD) 4666 LHSType = PD->getType(); 4667 } 4668 4669 if (LHSType.isNull()) 4670 LHSType = LHS->getType(); 4671 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 4672 return; 4673 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 4674 // FIXME. Check for other life times. 4675 if (LT != Qualifiers::OCL_None) 4676 return; 4677 4678 if (PRE) { 4679 if (PRE->isImplicitProperty()) 4680 return; 4681 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 4682 if (!PD) 4683 return; 4684 4685 unsigned Attributes = PD->getPropertyAttributes(); 4686 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) { 4687 // when 'assign' attribute was not explicitly specified 4688 // by user, ignore it and rely on property type itself 4689 // for lifetime info. 4690 unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten(); 4691 if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) && 4692 LHSType->isObjCRetainableType()) 4693 return; 4694 4695 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4696 if (cast->getCastKind() == CK_ARCConsumeObject) { 4697 Diag(Loc, diag::warn_arc_retained_property_assign) 4698 << RHS->getSourceRange(); 4699 return; 4700 } 4701 RHS = cast->getSubExpr(); 4702 } 4703 } 4704 } 4705} 4706