SemaChecking.cpp revision e3d8e737e18f0ce95d87be03f74b35413443173c
1//===--- SemaChecking.cpp - Extra Semantic Checking -----------------------===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// This file implements extra semantic analysis beyond what is enforced 11// by the C type system. 12// 13//===----------------------------------------------------------------------===// 14 15#include "clang/Sema/Initialization.h" 16#include "clang/Sema/Sema.h" 17#include "clang/Sema/SemaInternal.h" 18#include "clang/Sema/Initialization.h" 19#include "clang/Sema/ScopeInfo.h" 20#include "clang/Analysis/Analyses/FormatString.h" 21#include "clang/AST/ASTContext.h" 22#include "clang/AST/CharUnits.h" 23#include "clang/AST/DeclCXX.h" 24#include "clang/AST/DeclObjC.h" 25#include "clang/AST/ExprCXX.h" 26#include "clang/AST/ExprObjC.h" 27#include "clang/AST/EvaluatedExprVisitor.h" 28#include "clang/AST/DeclObjC.h" 29#include "clang/AST/StmtCXX.h" 30#include "clang/AST/StmtObjC.h" 31#include "clang/Lex/Preprocessor.h" 32#include "llvm/ADT/BitVector.h" 33#include "llvm/ADT/SmallString.h" 34#include "llvm/ADT/STLExtras.h" 35#include "llvm/Support/raw_ostream.h" 36#include "clang/Basic/TargetBuiltins.h" 37#include "clang/Basic/TargetInfo.h" 38#include "clang/Basic/ConvertUTF.h" 39#include <limits> 40using namespace clang; 41using namespace sema; 42 43SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 44 unsigned ByteNo) const { 45 return SL->getLocationOfByte(ByteNo, PP.getSourceManager(), 46 PP.getLangOptions(), PP.getTargetInfo()); 47} 48 49/// Checks that a call expression's argument count is the desired number. 50/// This is useful when doing custom type-checking. Returns true on error. 51static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) { 52 unsigned argCount = call->getNumArgs(); 53 if (argCount == desiredArgCount) return false; 54 55 if (argCount < desiredArgCount) 56 return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args) 57 << 0 /*function call*/ << desiredArgCount << argCount 58 << call->getSourceRange(); 59 60 // Highlight all the excess arguments. 61 SourceRange range(call->getArg(desiredArgCount)->getLocStart(), 62 call->getArg(argCount - 1)->getLocEnd()); 63 64 return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args) 65 << 0 /*function call*/ << desiredArgCount << argCount 66 << call->getArg(1)->getSourceRange(); 67} 68 69/// CheckBuiltinAnnotationString - Checks that string argument to the builtin 70/// annotation is a non wide string literal. 71static bool CheckBuiltinAnnotationString(Sema &S, Expr *Arg) { 72 Arg = Arg->IgnoreParenCasts(); 73 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 74 if (!Literal || !Literal->isAscii()) { 75 S.Diag(Arg->getLocStart(), diag::err_builtin_annotation_not_string_constant) 76 << Arg->getSourceRange(); 77 return true; 78 } 79 return false; 80} 81 82ExprResult 83Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 84 ExprResult TheCallResult(Owned(TheCall)); 85 86 // Find out if any arguments are required to be integer constant expressions. 87 unsigned ICEArguments = 0; 88 ASTContext::GetBuiltinTypeError Error; 89 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments); 90 if (Error != ASTContext::GE_None) 91 ICEArguments = 0; // Don't diagnose previously diagnosed errors. 92 93 // If any arguments are required to be ICE's, check and diagnose. 94 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) { 95 // Skip arguments not required to be ICE's. 96 if ((ICEArguments & (1 << ArgNo)) == 0) continue; 97 98 llvm::APSInt Result; 99 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 100 return true; 101 ICEArguments &= ~(1 << ArgNo); 102 } 103 104 switch (BuiltinID) { 105 case Builtin::BI__builtin___CFStringMakeConstantString: 106 assert(TheCall->getNumArgs() == 1 && 107 "Wrong # arguments to builtin CFStringMakeConstantString"); 108 if (CheckObjCString(TheCall->getArg(0))) 109 return ExprError(); 110 break; 111 case Builtin::BI__builtin_stdarg_start: 112 case Builtin::BI__builtin_va_start: 113 if (SemaBuiltinVAStart(TheCall)) 114 return ExprError(); 115 break; 116 case Builtin::BI__builtin_isgreater: 117 case Builtin::BI__builtin_isgreaterequal: 118 case Builtin::BI__builtin_isless: 119 case Builtin::BI__builtin_islessequal: 120 case Builtin::BI__builtin_islessgreater: 121 case Builtin::BI__builtin_isunordered: 122 if (SemaBuiltinUnorderedCompare(TheCall)) 123 return ExprError(); 124 break; 125 case Builtin::BI__builtin_fpclassify: 126 if (SemaBuiltinFPClassification(TheCall, 6)) 127 return ExprError(); 128 break; 129 case Builtin::BI__builtin_isfinite: 130 case Builtin::BI__builtin_isinf: 131 case Builtin::BI__builtin_isinf_sign: 132 case Builtin::BI__builtin_isnan: 133 case Builtin::BI__builtin_isnormal: 134 if (SemaBuiltinFPClassification(TheCall, 1)) 135 return ExprError(); 136 break; 137 case Builtin::BI__builtin_shufflevector: 138 return SemaBuiltinShuffleVector(TheCall); 139 // TheCall will be freed by the smart pointer here, but that's fine, since 140 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 141 case Builtin::BI__builtin_prefetch: 142 if (SemaBuiltinPrefetch(TheCall)) 143 return ExprError(); 144 break; 145 case Builtin::BI__builtin_object_size: 146 if (SemaBuiltinObjectSize(TheCall)) 147 return ExprError(); 148 break; 149 case Builtin::BI__builtin_longjmp: 150 if (SemaBuiltinLongjmp(TheCall)) 151 return ExprError(); 152 break; 153 154 case Builtin::BI__builtin_classify_type: 155 if (checkArgCount(*this, TheCall, 1)) return true; 156 TheCall->setType(Context.IntTy); 157 break; 158 case Builtin::BI__builtin_constant_p: 159 if (checkArgCount(*this, TheCall, 1)) return true; 160 TheCall->setType(Context.IntTy); 161 break; 162 case Builtin::BI__sync_fetch_and_add: 163 case Builtin::BI__sync_fetch_and_add_1: 164 case Builtin::BI__sync_fetch_and_add_2: 165 case Builtin::BI__sync_fetch_and_add_4: 166 case Builtin::BI__sync_fetch_and_add_8: 167 case Builtin::BI__sync_fetch_and_add_16: 168 case Builtin::BI__sync_fetch_and_sub: 169 case Builtin::BI__sync_fetch_and_sub_1: 170 case Builtin::BI__sync_fetch_and_sub_2: 171 case Builtin::BI__sync_fetch_and_sub_4: 172 case Builtin::BI__sync_fetch_and_sub_8: 173 case Builtin::BI__sync_fetch_and_sub_16: 174 case Builtin::BI__sync_fetch_and_or: 175 case Builtin::BI__sync_fetch_and_or_1: 176 case Builtin::BI__sync_fetch_and_or_2: 177 case Builtin::BI__sync_fetch_and_or_4: 178 case Builtin::BI__sync_fetch_and_or_8: 179 case Builtin::BI__sync_fetch_and_or_16: 180 case Builtin::BI__sync_fetch_and_and: 181 case Builtin::BI__sync_fetch_and_and_1: 182 case Builtin::BI__sync_fetch_and_and_2: 183 case Builtin::BI__sync_fetch_and_and_4: 184 case Builtin::BI__sync_fetch_and_and_8: 185 case Builtin::BI__sync_fetch_and_and_16: 186 case Builtin::BI__sync_fetch_and_xor: 187 case Builtin::BI__sync_fetch_and_xor_1: 188 case Builtin::BI__sync_fetch_and_xor_2: 189 case Builtin::BI__sync_fetch_and_xor_4: 190 case Builtin::BI__sync_fetch_and_xor_8: 191 case Builtin::BI__sync_fetch_and_xor_16: 192 case Builtin::BI__sync_add_and_fetch: 193 case Builtin::BI__sync_add_and_fetch_1: 194 case Builtin::BI__sync_add_and_fetch_2: 195 case Builtin::BI__sync_add_and_fetch_4: 196 case Builtin::BI__sync_add_and_fetch_8: 197 case Builtin::BI__sync_add_and_fetch_16: 198 case Builtin::BI__sync_sub_and_fetch: 199 case Builtin::BI__sync_sub_and_fetch_1: 200 case Builtin::BI__sync_sub_and_fetch_2: 201 case Builtin::BI__sync_sub_and_fetch_4: 202 case Builtin::BI__sync_sub_and_fetch_8: 203 case Builtin::BI__sync_sub_and_fetch_16: 204 case Builtin::BI__sync_and_and_fetch: 205 case Builtin::BI__sync_and_and_fetch_1: 206 case Builtin::BI__sync_and_and_fetch_2: 207 case Builtin::BI__sync_and_and_fetch_4: 208 case Builtin::BI__sync_and_and_fetch_8: 209 case Builtin::BI__sync_and_and_fetch_16: 210 case Builtin::BI__sync_or_and_fetch: 211 case Builtin::BI__sync_or_and_fetch_1: 212 case Builtin::BI__sync_or_and_fetch_2: 213 case Builtin::BI__sync_or_and_fetch_4: 214 case Builtin::BI__sync_or_and_fetch_8: 215 case Builtin::BI__sync_or_and_fetch_16: 216 case Builtin::BI__sync_xor_and_fetch: 217 case Builtin::BI__sync_xor_and_fetch_1: 218 case Builtin::BI__sync_xor_and_fetch_2: 219 case Builtin::BI__sync_xor_and_fetch_4: 220 case Builtin::BI__sync_xor_and_fetch_8: 221 case Builtin::BI__sync_xor_and_fetch_16: 222 case Builtin::BI__sync_val_compare_and_swap: 223 case Builtin::BI__sync_val_compare_and_swap_1: 224 case Builtin::BI__sync_val_compare_and_swap_2: 225 case Builtin::BI__sync_val_compare_and_swap_4: 226 case Builtin::BI__sync_val_compare_and_swap_8: 227 case Builtin::BI__sync_val_compare_and_swap_16: 228 case Builtin::BI__sync_bool_compare_and_swap: 229 case Builtin::BI__sync_bool_compare_and_swap_1: 230 case Builtin::BI__sync_bool_compare_and_swap_2: 231 case Builtin::BI__sync_bool_compare_and_swap_4: 232 case Builtin::BI__sync_bool_compare_and_swap_8: 233 case Builtin::BI__sync_bool_compare_and_swap_16: 234 case Builtin::BI__sync_lock_test_and_set: 235 case Builtin::BI__sync_lock_test_and_set_1: 236 case Builtin::BI__sync_lock_test_and_set_2: 237 case Builtin::BI__sync_lock_test_and_set_4: 238 case Builtin::BI__sync_lock_test_and_set_8: 239 case Builtin::BI__sync_lock_test_and_set_16: 240 case Builtin::BI__sync_lock_release: 241 case Builtin::BI__sync_lock_release_1: 242 case Builtin::BI__sync_lock_release_2: 243 case Builtin::BI__sync_lock_release_4: 244 case Builtin::BI__sync_lock_release_8: 245 case Builtin::BI__sync_lock_release_16: 246 case Builtin::BI__sync_swap: 247 case Builtin::BI__sync_swap_1: 248 case Builtin::BI__sync_swap_2: 249 case Builtin::BI__sync_swap_4: 250 case Builtin::BI__sync_swap_8: 251 case Builtin::BI__sync_swap_16: 252 return SemaBuiltinAtomicOverloaded(move(TheCallResult)); 253 case Builtin::BI__atomic_load: 254 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Load); 255 case Builtin::BI__atomic_store: 256 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Store); 257 case Builtin::BI__atomic_init: 258 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Init); 259 case Builtin::BI__atomic_exchange: 260 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Xchg); 261 case Builtin::BI__atomic_compare_exchange_strong: 262 return SemaAtomicOpsOverloaded(move(TheCallResult), 263 AtomicExpr::CmpXchgStrong); 264 case Builtin::BI__atomic_compare_exchange_weak: 265 return SemaAtomicOpsOverloaded(move(TheCallResult), 266 AtomicExpr::CmpXchgWeak); 267 case Builtin::BI__atomic_fetch_add: 268 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Add); 269 case Builtin::BI__atomic_fetch_sub: 270 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Sub); 271 case Builtin::BI__atomic_fetch_and: 272 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::And); 273 case Builtin::BI__atomic_fetch_or: 274 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Or); 275 case Builtin::BI__atomic_fetch_xor: 276 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Xor); 277 case Builtin::BI__builtin_annotation: 278 if (CheckBuiltinAnnotationString(*this, TheCall->getArg(1))) 279 return ExprError(); 280 break; 281 } 282 283 // Since the target specific builtins for each arch overlap, only check those 284 // of the arch we are compiling for. 285 if (BuiltinID >= Builtin::FirstTSBuiltin) { 286 switch (Context.getTargetInfo().getTriple().getArch()) { 287 case llvm::Triple::arm: 288 case llvm::Triple::thumb: 289 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 290 return ExprError(); 291 break; 292 default: 293 break; 294 } 295 } 296 297 return move(TheCallResult); 298} 299 300// Get the valid immediate range for the specified NEON type code. 301static unsigned RFT(unsigned t, bool shift = false) { 302 NeonTypeFlags Type(t); 303 int IsQuad = Type.isQuad(); 304 switch (Type.getEltType()) { 305 case NeonTypeFlags::Int8: 306 case NeonTypeFlags::Poly8: 307 return shift ? 7 : (8 << IsQuad) - 1; 308 case NeonTypeFlags::Int16: 309 case NeonTypeFlags::Poly16: 310 return shift ? 15 : (4 << IsQuad) - 1; 311 case NeonTypeFlags::Int32: 312 return shift ? 31 : (2 << IsQuad) - 1; 313 case NeonTypeFlags::Int64: 314 return shift ? 63 : (1 << IsQuad) - 1; 315 case NeonTypeFlags::Float16: 316 assert(!shift && "cannot shift float types!"); 317 return (4 << IsQuad) - 1; 318 case NeonTypeFlags::Float32: 319 assert(!shift && "cannot shift float types!"); 320 return (2 << IsQuad) - 1; 321 } 322 llvm_unreachable("Invalid NeonTypeFlag!"); 323} 324 325/// getNeonEltType - Return the QualType corresponding to the elements of 326/// the vector type specified by the NeonTypeFlags. This is used to check 327/// the pointer arguments for Neon load/store intrinsics. 328static QualType getNeonEltType(NeonTypeFlags Flags, ASTContext &Context) { 329 switch (Flags.getEltType()) { 330 case NeonTypeFlags::Int8: 331 return Flags.isUnsigned() ? Context.UnsignedCharTy : Context.SignedCharTy; 332 case NeonTypeFlags::Int16: 333 return Flags.isUnsigned() ? Context.UnsignedShortTy : Context.ShortTy; 334 case NeonTypeFlags::Int32: 335 return Flags.isUnsigned() ? Context.UnsignedIntTy : Context.IntTy; 336 case NeonTypeFlags::Int64: 337 return Flags.isUnsigned() ? Context.UnsignedLongLongTy : Context.LongLongTy; 338 case NeonTypeFlags::Poly8: 339 return Context.SignedCharTy; 340 case NeonTypeFlags::Poly16: 341 return Context.ShortTy; 342 case NeonTypeFlags::Float16: 343 return Context.UnsignedShortTy; 344 case NeonTypeFlags::Float32: 345 return Context.FloatTy; 346 } 347 llvm_unreachable("Invalid NeonTypeFlag!"); 348} 349 350bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 351 llvm::APSInt Result; 352 353 unsigned mask = 0; 354 unsigned TV = 0; 355 int PtrArgNum = -1; 356 bool HasConstPtr = false; 357 switch (BuiltinID) { 358#define GET_NEON_OVERLOAD_CHECK 359#include "clang/Basic/arm_neon.inc" 360#undef GET_NEON_OVERLOAD_CHECK 361 } 362 363 // For NEON intrinsics which are overloaded on vector element type, validate 364 // the immediate which specifies which variant to emit. 365 unsigned ImmArg = TheCall->getNumArgs()-1; 366 if (mask) { 367 if (SemaBuiltinConstantArg(TheCall, ImmArg, Result)) 368 return true; 369 370 TV = Result.getLimitedValue(64); 371 if ((TV > 63) || (mask & (1 << TV)) == 0) 372 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 373 << TheCall->getArg(ImmArg)->getSourceRange(); 374 } 375 376 if (PtrArgNum >= 0) { 377 // Check that pointer arguments have the specified type. 378 Expr *Arg = TheCall->getArg(PtrArgNum); 379 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) 380 Arg = ICE->getSubExpr(); 381 ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg); 382 QualType RHSTy = RHS.get()->getType(); 383 QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context); 384 if (HasConstPtr) 385 EltTy = EltTy.withConst(); 386 QualType LHSTy = Context.getPointerType(EltTy); 387 AssignConvertType ConvTy; 388 ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS); 389 if (RHS.isInvalid()) 390 return true; 391 if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy, 392 RHS.get(), AA_Assigning)) 393 return true; 394 } 395 396 // For NEON intrinsics which take an immediate value as part of the 397 // instruction, range check them here. 398 unsigned i = 0, l = 0, u = 0; 399 switch (BuiltinID) { 400 default: return false; 401 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 402 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 403 case ARM::BI__builtin_arm_vcvtr_f: 404 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 405#define GET_NEON_IMMEDIATE_CHECK 406#include "clang/Basic/arm_neon.inc" 407#undef GET_NEON_IMMEDIATE_CHECK 408 }; 409 410 // Check that the immediate argument is actually a constant. 411 if (SemaBuiltinConstantArg(TheCall, i, Result)) 412 return true; 413 414 // Range check against the upper/lower values for this isntruction. 415 unsigned Val = Result.getZExtValue(); 416 if (Val < l || Val > (u + l)) 417 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 418 << l << u+l << TheCall->getArg(i)->getSourceRange(); 419 420 // FIXME: VFP Intrinsics should error if VFP not present. 421 return false; 422} 423 424/// CheckFunctionCall - Check a direct function call for various correctness 425/// and safety properties not strictly enforced by the C type system. 426bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 427 // Get the IdentifierInfo* for the called function. 428 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 429 430 // None of the checks below are needed for functions that don't have 431 // simple names (e.g., C++ conversion functions). 432 if (!FnInfo) 433 return false; 434 435 // FIXME: This mechanism should be abstracted to be less fragile and 436 // more efficient. For example, just map function ids to custom 437 // handlers. 438 439 // Printf and scanf checking. 440 for (specific_attr_iterator<FormatAttr> 441 i = FDecl->specific_attr_begin<FormatAttr>(), 442 e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { 443 CheckFormatArguments(*i, TheCall); 444 } 445 446 for (specific_attr_iterator<NonNullAttr> 447 i = FDecl->specific_attr_begin<NonNullAttr>(), 448 e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) { 449 CheckNonNullArguments(*i, TheCall->getArgs(), 450 TheCall->getCallee()->getLocStart()); 451 } 452 453 unsigned CMId = FDecl->getMemoryFunctionKind(); 454 if (CMId == 0) 455 return false; 456 457 // Handle memory setting and copying functions. 458 if (CMId == Builtin::BIstrlcpy || CMId == Builtin::BIstrlcat) 459 CheckStrlcpycatArguments(TheCall, FnInfo); 460 else if (CMId == Builtin::BIstrncat) 461 CheckStrncatArguments(TheCall, FnInfo); 462 else 463 CheckMemaccessArguments(TheCall, CMId, FnInfo); 464 465 return false; 466} 467 468bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac, 469 Expr **Args, unsigned NumArgs) { 470 for (specific_attr_iterator<FormatAttr> 471 i = Method->specific_attr_begin<FormatAttr>(), 472 e = Method->specific_attr_end<FormatAttr>(); i != e ; ++i) { 473 474 CheckFormatArguments(*i, Args, NumArgs, false, lbrac, 475 Method->getSourceRange()); 476 } 477 478 // diagnose nonnull arguments. 479 for (specific_attr_iterator<NonNullAttr> 480 i = Method->specific_attr_begin<NonNullAttr>(), 481 e = Method->specific_attr_end<NonNullAttr>(); i != e; ++i) { 482 CheckNonNullArguments(*i, Args, lbrac); 483 } 484 485 return false; 486} 487 488bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 489 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 490 if (!V) 491 return false; 492 493 QualType Ty = V->getType(); 494 if (!Ty->isBlockPointerType()) 495 return false; 496 497 // format string checking. 498 for (specific_attr_iterator<FormatAttr> 499 i = NDecl->specific_attr_begin<FormatAttr>(), 500 e = NDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { 501 CheckFormatArguments(*i, TheCall); 502 } 503 504 return false; 505} 506 507ExprResult 508Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, AtomicExpr::AtomicOp Op) { 509 CallExpr *TheCall = cast<CallExpr>(TheCallResult.get()); 510 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 511 512 // All these operations take one of the following four forms: 513 // T __atomic_load(_Atomic(T)*, int) (loads) 514 // T* __atomic_add(_Atomic(T*)*, ptrdiff_t, int) (pointer add/sub) 515 // int __atomic_compare_exchange_strong(_Atomic(T)*, T*, T, int, int) 516 // (cmpxchg) 517 // T __atomic_exchange(_Atomic(T)*, T, int) (everything else) 518 // where T is an appropriate type, and the int paremeterss are for orderings. 519 unsigned NumVals = 1; 520 unsigned NumOrders = 1; 521 if (Op == AtomicExpr::Load) { 522 NumVals = 0; 523 } else if (Op == AtomicExpr::CmpXchgWeak || Op == AtomicExpr::CmpXchgStrong) { 524 NumVals = 2; 525 NumOrders = 2; 526 } 527 if (Op == AtomicExpr::Init) 528 NumOrders = 0; 529 530 if (TheCall->getNumArgs() < NumVals+NumOrders+1) { 531 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 532 << 0 << NumVals+NumOrders+1 << TheCall->getNumArgs() 533 << TheCall->getCallee()->getSourceRange(); 534 return ExprError(); 535 } else if (TheCall->getNumArgs() > NumVals+NumOrders+1) { 536 Diag(TheCall->getArg(NumVals+NumOrders+1)->getLocStart(), 537 diag::err_typecheck_call_too_many_args) 538 << 0 << NumVals+NumOrders+1 << TheCall->getNumArgs() 539 << TheCall->getCallee()->getSourceRange(); 540 return ExprError(); 541 } 542 543 // Inspect the first argument of the atomic operation. This should always be 544 // a pointer to an _Atomic type. 545 Expr *Ptr = TheCall->getArg(0); 546 Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get(); 547 const PointerType *pointerType = Ptr->getType()->getAs<PointerType>(); 548 if (!pointerType) { 549 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 550 << Ptr->getType() << Ptr->getSourceRange(); 551 return ExprError(); 552 } 553 554 QualType AtomTy = pointerType->getPointeeType(); 555 if (!AtomTy->isAtomicType()) { 556 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 557 << Ptr->getType() << Ptr->getSourceRange(); 558 return ExprError(); 559 } 560 QualType ValType = AtomTy->getAs<AtomicType>()->getValueType(); 561 562 if ((Op == AtomicExpr::Add || Op == AtomicExpr::Sub) && 563 !ValType->isIntegerType() && !ValType->isPointerType()) { 564 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 565 << Ptr->getType() << Ptr->getSourceRange(); 566 return ExprError(); 567 } 568 569 if (!ValType->isIntegerType() && 570 (Op == AtomicExpr::And || Op == AtomicExpr::Or || Op == AtomicExpr::Xor)){ 571 Diag(DRE->getLocStart(), diag::err_atomic_op_logical_needs_atomic_int) 572 << Ptr->getType() << Ptr->getSourceRange(); 573 return ExprError(); 574 } 575 576 switch (ValType.getObjCLifetime()) { 577 case Qualifiers::OCL_None: 578 case Qualifiers::OCL_ExplicitNone: 579 // okay 580 break; 581 582 case Qualifiers::OCL_Weak: 583 case Qualifiers::OCL_Strong: 584 case Qualifiers::OCL_Autoreleasing: 585 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 586 << ValType << Ptr->getSourceRange(); 587 return ExprError(); 588 } 589 590 QualType ResultType = ValType; 591 if (Op == AtomicExpr::Store || Op == AtomicExpr::Init) 592 ResultType = Context.VoidTy; 593 else if (Op == AtomicExpr::CmpXchgWeak || Op == AtomicExpr::CmpXchgStrong) 594 ResultType = Context.BoolTy; 595 596 // The first argument --- the pointer --- has a fixed type; we 597 // deduce the types of the rest of the arguments accordingly. Walk 598 // the remaining arguments, converting them to the deduced value type. 599 for (unsigned i = 1; i != NumVals+NumOrders+1; ++i) { 600 ExprResult Arg = TheCall->getArg(i); 601 QualType Ty; 602 if (i < NumVals+1) { 603 // The second argument to a cmpxchg is a pointer to the data which will 604 // be exchanged. The second argument to a pointer add/subtract is the 605 // amount to add/subtract, which must be a ptrdiff_t. The third 606 // argument to a cmpxchg and the second argument in all other cases 607 // is the type of the value. 608 if (i == 1 && (Op == AtomicExpr::CmpXchgWeak || 609 Op == AtomicExpr::CmpXchgStrong)) 610 Ty = Context.getPointerType(ValType.getUnqualifiedType()); 611 else if (!ValType->isIntegerType() && 612 (Op == AtomicExpr::Add || Op == AtomicExpr::Sub)) 613 Ty = Context.getPointerDiffType(); 614 else 615 Ty = ValType; 616 } else { 617 // The order(s) are always converted to int. 618 Ty = Context.IntTy; 619 } 620 InitializedEntity Entity = 621 InitializedEntity::InitializeParameter(Context, Ty, false); 622 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 623 if (Arg.isInvalid()) 624 return true; 625 TheCall->setArg(i, Arg.get()); 626 } 627 628 SmallVector<Expr*, 5> SubExprs; 629 SubExprs.push_back(Ptr); 630 if (Op == AtomicExpr::Load) { 631 SubExprs.push_back(TheCall->getArg(1)); // Order 632 } else if (Op == AtomicExpr::Init) { 633 SubExprs.push_back(TheCall->getArg(1)); // Val1 634 } else if (Op != AtomicExpr::CmpXchgWeak && Op != AtomicExpr::CmpXchgStrong) { 635 SubExprs.push_back(TheCall->getArg(2)); // Order 636 SubExprs.push_back(TheCall->getArg(1)); // Val1 637 } else { 638 SubExprs.push_back(TheCall->getArg(3)); // Order 639 SubExprs.push_back(TheCall->getArg(1)); // Val1 640 SubExprs.push_back(TheCall->getArg(2)); // Val2 641 SubExprs.push_back(TheCall->getArg(4)); // OrderFail 642 } 643 644 return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 645 SubExprs.data(), SubExprs.size(), 646 ResultType, Op, 647 TheCall->getRParenLoc())); 648} 649 650 651/// checkBuiltinArgument - Given a call to a builtin function, perform 652/// normal type-checking on the given argument, updating the call in 653/// place. This is useful when a builtin function requires custom 654/// type-checking for some of its arguments but not necessarily all of 655/// them. 656/// 657/// Returns true on error. 658static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 659 FunctionDecl *Fn = E->getDirectCallee(); 660 assert(Fn && "builtin call without direct callee!"); 661 662 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 663 InitializedEntity Entity = 664 InitializedEntity::InitializeParameter(S.Context, Param); 665 666 ExprResult Arg = E->getArg(0); 667 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 668 if (Arg.isInvalid()) 669 return true; 670 671 E->setArg(ArgIndex, Arg.take()); 672 return false; 673} 674 675/// SemaBuiltinAtomicOverloaded - We have a call to a function like 676/// __sync_fetch_and_add, which is an overloaded function based on the pointer 677/// type of its first argument. The main ActOnCallExpr routines have already 678/// promoted the types of arguments because all of these calls are prototyped as 679/// void(...). 680/// 681/// This function goes through and does final semantic checking for these 682/// builtins, 683ExprResult 684Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 685 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 686 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 687 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 688 689 // Ensure that we have at least one argument to do type inference from. 690 if (TheCall->getNumArgs() < 1) { 691 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 692 << 0 << 1 << TheCall->getNumArgs() 693 << TheCall->getCallee()->getSourceRange(); 694 return ExprError(); 695 } 696 697 // Inspect the first argument of the atomic builtin. This should always be 698 // a pointer type, whose element is an integral scalar or pointer type. 699 // Because it is a pointer type, we don't have to worry about any implicit 700 // casts here. 701 // FIXME: We don't allow floating point scalars as input. 702 Expr *FirstArg = TheCall->getArg(0); 703 ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg); 704 if (FirstArgResult.isInvalid()) 705 return ExprError(); 706 FirstArg = FirstArgResult.take(); 707 TheCall->setArg(0, FirstArg); 708 709 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 710 if (!pointerType) { 711 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 712 << FirstArg->getType() << FirstArg->getSourceRange(); 713 return ExprError(); 714 } 715 716 QualType ValType = pointerType->getPointeeType(); 717 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 718 !ValType->isBlockPointerType()) { 719 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 720 << FirstArg->getType() << FirstArg->getSourceRange(); 721 return ExprError(); 722 } 723 724 switch (ValType.getObjCLifetime()) { 725 case Qualifiers::OCL_None: 726 case Qualifiers::OCL_ExplicitNone: 727 // okay 728 break; 729 730 case Qualifiers::OCL_Weak: 731 case Qualifiers::OCL_Strong: 732 case Qualifiers::OCL_Autoreleasing: 733 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 734 << ValType << FirstArg->getSourceRange(); 735 return ExprError(); 736 } 737 738 // Strip any qualifiers off ValType. 739 ValType = ValType.getUnqualifiedType(); 740 741 // The majority of builtins return a value, but a few have special return 742 // types, so allow them to override appropriately below. 743 QualType ResultType = ValType; 744 745 // We need to figure out which concrete builtin this maps onto. For example, 746 // __sync_fetch_and_add with a 2 byte object turns into 747 // __sync_fetch_and_add_2. 748#define BUILTIN_ROW(x) \ 749 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 750 Builtin::BI##x##_8, Builtin::BI##x##_16 } 751 752 static const unsigned BuiltinIndices[][5] = { 753 BUILTIN_ROW(__sync_fetch_and_add), 754 BUILTIN_ROW(__sync_fetch_and_sub), 755 BUILTIN_ROW(__sync_fetch_and_or), 756 BUILTIN_ROW(__sync_fetch_and_and), 757 BUILTIN_ROW(__sync_fetch_and_xor), 758 759 BUILTIN_ROW(__sync_add_and_fetch), 760 BUILTIN_ROW(__sync_sub_and_fetch), 761 BUILTIN_ROW(__sync_and_and_fetch), 762 BUILTIN_ROW(__sync_or_and_fetch), 763 BUILTIN_ROW(__sync_xor_and_fetch), 764 765 BUILTIN_ROW(__sync_val_compare_and_swap), 766 BUILTIN_ROW(__sync_bool_compare_and_swap), 767 BUILTIN_ROW(__sync_lock_test_and_set), 768 BUILTIN_ROW(__sync_lock_release), 769 BUILTIN_ROW(__sync_swap) 770 }; 771#undef BUILTIN_ROW 772 773 // Determine the index of the size. 774 unsigned SizeIndex; 775 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 776 case 1: SizeIndex = 0; break; 777 case 2: SizeIndex = 1; break; 778 case 4: SizeIndex = 2; break; 779 case 8: SizeIndex = 3; break; 780 case 16: SizeIndex = 4; break; 781 default: 782 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 783 << FirstArg->getType() << FirstArg->getSourceRange(); 784 return ExprError(); 785 } 786 787 // Each of these builtins has one pointer argument, followed by some number of 788 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 789 // that we ignore. Find out which row of BuiltinIndices to read from as well 790 // as the number of fixed args. 791 unsigned BuiltinID = FDecl->getBuiltinID(); 792 unsigned BuiltinIndex, NumFixed = 1; 793 switch (BuiltinID) { 794 default: llvm_unreachable("Unknown overloaded atomic builtin!"); 795 case Builtin::BI__sync_fetch_and_add: 796 case Builtin::BI__sync_fetch_and_add_1: 797 case Builtin::BI__sync_fetch_and_add_2: 798 case Builtin::BI__sync_fetch_and_add_4: 799 case Builtin::BI__sync_fetch_and_add_8: 800 case Builtin::BI__sync_fetch_and_add_16: 801 BuiltinIndex = 0; 802 break; 803 804 case Builtin::BI__sync_fetch_and_sub: 805 case Builtin::BI__sync_fetch_and_sub_1: 806 case Builtin::BI__sync_fetch_and_sub_2: 807 case Builtin::BI__sync_fetch_and_sub_4: 808 case Builtin::BI__sync_fetch_and_sub_8: 809 case Builtin::BI__sync_fetch_and_sub_16: 810 BuiltinIndex = 1; 811 break; 812 813 case Builtin::BI__sync_fetch_and_or: 814 case Builtin::BI__sync_fetch_and_or_1: 815 case Builtin::BI__sync_fetch_and_or_2: 816 case Builtin::BI__sync_fetch_and_or_4: 817 case Builtin::BI__sync_fetch_and_or_8: 818 case Builtin::BI__sync_fetch_and_or_16: 819 BuiltinIndex = 2; 820 break; 821 822 case Builtin::BI__sync_fetch_and_and: 823 case Builtin::BI__sync_fetch_and_and_1: 824 case Builtin::BI__sync_fetch_and_and_2: 825 case Builtin::BI__sync_fetch_and_and_4: 826 case Builtin::BI__sync_fetch_and_and_8: 827 case Builtin::BI__sync_fetch_and_and_16: 828 BuiltinIndex = 3; 829 break; 830 831 case Builtin::BI__sync_fetch_and_xor: 832 case Builtin::BI__sync_fetch_and_xor_1: 833 case Builtin::BI__sync_fetch_and_xor_2: 834 case Builtin::BI__sync_fetch_and_xor_4: 835 case Builtin::BI__sync_fetch_and_xor_8: 836 case Builtin::BI__sync_fetch_and_xor_16: 837 BuiltinIndex = 4; 838 break; 839 840 case Builtin::BI__sync_add_and_fetch: 841 case Builtin::BI__sync_add_and_fetch_1: 842 case Builtin::BI__sync_add_and_fetch_2: 843 case Builtin::BI__sync_add_and_fetch_4: 844 case Builtin::BI__sync_add_and_fetch_8: 845 case Builtin::BI__sync_add_and_fetch_16: 846 BuiltinIndex = 5; 847 break; 848 849 case Builtin::BI__sync_sub_and_fetch: 850 case Builtin::BI__sync_sub_and_fetch_1: 851 case Builtin::BI__sync_sub_and_fetch_2: 852 case Builtin::BI__sync_sub_and_fetch_4: 853 case Builtin::BI__sync_sub_and_fetch_8: 854 case Builtin::BI__sync_sub_and_fetch_16: 855 BuiltinIndex = 6; 856 break; 857 858 case Builtin::BI__sync_and_and_fetch: 859 case Builtin::BI__sync_and_and_fetch_1: 860 case Builtin::BI__sync_and_and_fetch_2: 861 case Builtin::BI__sync_and_and_fetch_4: 862 case Builtin::BI__sync_and_and_fetch_8: 863 case Builtin::BI__sync_and_and_fetch_16: 864 BuiltinIndex = 7; 865 break; 866 867 case Builtin::BI__sync_or_and_fetch: 868 case Builtin::BI__sync_or_and_fetch_1: 869 case Builtin::BI__sync_or_and_fetch_2: 870 case Builtin::BI__sync_or_and_fetch_4: 871 case Builtin::BI__sync_or_and_fetch_8: 872 case Builtin::BI__sync_or_and_fetch_16: 873 BuiltinIndex = 8; 874 break; 875 876 case Builtin::BI__sync_xor_and_fetch: 877 case Builtin::BI__sync_xor_and_fetch_1: 878 case Builtin::BI__sync_xor_and_fetch_2: 879 case Builtin::BI__sync_xor_and_fetch_4: 880 case Builtin::BI__sync_xor_and_fetch_8: 881 case Builtin::BI__sync_xor_and_fetch_16: 882 BuiltinIndex = 9; 883 break; 884 885 case Builtin::BI__sync_val_compare_and_swap: 886 case Builtin::BI__sync_val_compare_and_swap_1: 887 case Builtin::BI__sync_val_compare_and_swap_2: 888 case Builtin::BI__sync_val_compare_and_swap_4: 889 case Builtin::BI__sync_val_compare_and_swap_8: 890 case Builtin::BI__sync_val_compare_and_swap_16: 891 BuiltinIndex = 10; 892 NumFixed = 2; 893 break; 894 895 case Builtin::BI__sync_bool_compare_and_swap: 896 case Builtin::BI__sync_bool_compare_and_swap_1: 897 case Builtin::BI__sync_bool_compare_and_swap_2: 898 case Builtin::BI__sync_bool_compare_and_swap_4: 899 case Builtin::BI__sync_bool_compare_and_swap_8: 900 case Builtin::BI__sync_bool_compare_and_swap_16: 901 BuiltinIndex = 11; 902 NumFixed = 2; 903 ResultType = Context.BoolTy; 904 break; 905 906 case Builtin::BI__sync_lock_test_and_set: 907 case Builtin::BI__sync_lock_test_and_set_1: 908 case Builtin::BI__sync_lock_test_and_set_2: 909 case Builtin::BI__sync_lock_test_and_set_4: 910 case Builtin::BI__sync_lock_test_and_set_8: 911 case Builtin::BI__sync_lock_test_and_set_16: 912 BuiltinIndex = 12; 913 break; 914 915 case Builtin::BI__sync_lock_release: 916 case Builtin::BI__sync_lock_release_1: 917 case Builtin::BI__sync_lock_release_2: 918 case Builtin::BI__sync_lock_release_4: 919 case Builtin::BI__sync_lock_release_8: 920 case Builtin::BI__sync_lock_release_16: 921 BuiltinIndex = 13; 922 NumFixed = 0; 923 ResultType = Context.VoidTy; 924 break; 925 926 case Builtin::BI__sync_swap: 927 case Builtin::BI__sync_swap_1: 928 case Builtin::BI__sync_swap_2: 929 case Builtin::BI__sync_swap_4: 930 case Builtin::BI__sync_swap_8: 931 case Builtin::BI__sync_swap_16: 932 BuiltinIndex = 14; 933 break; 934 } 935 936 // Now that we know how many fixed arguments we expect, first check that we 937 // have at least that many. 938 if (TheCall->getNumArgs() < 1+NumFixed) { 939 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 940 << 0 << 1+NumFixed << TheCall->getNumArgs() 941 << TheCall->getCallee()->getSourceRange(); 942 return ExprError(); 943 } 944 945 // Get the decl for the concrete builtin from this, we can tell what the 946 // concrete integer type we should convert to is. 947 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 948 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 949 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 950 FunctionDecl *NewBuiltinDecl = 951 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 952 TUScope, false, DRE->getLocStart())); 953 954 // The first argument --- the pointer --- has a fixed type; we 955 // deduce the types of the rest of the arguments accordingly. Walk 956 // the remaining arguments, converting them to the deduced value type. 957 for (unsigned i = 0; i != NumFixed; ++i) { 958 ExprResult Arg = TheCall->getArg(i+1); 959 960 // GCC does an implicit conversion to the pointer or integer ValType. This 961 // can fail in some cases (1i -> int**), check for this error case now. 962 // Initialize the argument. 963 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context, 964 ValType, /*consume*/ false); 965 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 966 if (Arg.isInvalid()) 967 return ExprError(); 968 969 // Okay, we have something that *can* be converted to the right type. Check 970 // to see if there is a potentially weird extension going on here. This can 971 // happen when you do an atomic operation on something like an char* and 972 // pass in 42. The 42 gets converted to char. This is even more strange 973 // for things like 45.123 -> char, etc. 974 // FIXME: Do this check. 975 TheCall->setArg(i+1, Arg.take()); 976 } 977 978 ASTContext& Context = this->getASTContext(); 979 980 // Create a new DeclRefExpr to refer to the new decl. 981 DeclRefExpr* NewDRE = DeclRefExpr::Create( 982 Context, 983 DRE->getQualifierLoc(), 984 SourceLocation(), 985 NewBuiltinDecl, 986 DRE->getLocation(), 987 NewBuiltinDecl->getType(), 988 DRE->getValueKind()); 989 990 // Set the callee in the CallExpr. 991 // FIXME: This leaks the original parens and implicit casts. 992 ExprResult PromotedCall = UsualUnaryConversions(NewDRE); 993 if (PromotedCall.isInvalid()) 994 return ExprError(); 995 TheCall->setCallee(PromotedCall.take()); 996 997 // Change the result type of the call to match the original value type. This 998 // is arbitrary, but the codegen for these builtins ins design to handle it 999 // gracefully. 1000 TheCall->setType(ResultType); 1001 1002 return move(TheCallResult); 1003} 1004 1005/// CheckObjCString - Checks that the argument to the builtin 1006/// CFString constructor is correct 1007/// Note: It might also make sense to do the UTF-16 conversion here (would 1008/// simplify the backend). 1009bool Sema::CheckObjCString(Expr *Arg) { 1010 Arg = Arg->IgnoreParenCasts(); 1011 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 1012 1013 if (!Literal || !Literal->isAscii()) { 1014 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 1015 << Arg->getSourceRange(); 1016 return true; 1017 } 1018 1019 if (Literal->containsNonAsciiOrNull()) { 1020 StringRef String = Literal->getString(); 1021 unsigned NumBytes = String.size(); 1022 SmallVector<UTF16, 128> ToBuf(NumBytes); 1023 const UTF8 *FromPtr = (UTF8 *)String.data(); 1024 UTF16 *ToPtr = &ToBuf[0]; 1025 1026 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 1027 &ToPtr, ToPtr + NumBytes, 1028 strictConversion); 1029 // Check for conversion failure. 1030 if (Result != conversionOK) 1031 Diag(Arg->getLocStart(), 1032 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 1033 } 1034 return false; 1035} 1036 1037/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 1038/// Emit an error and return true on failure, return false on success. 1039bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 1040 Expr *Fn = TheCall->getCallee(); 1041 if (TheCall->getNumArgs() > 2) { 1042 Diag(TheCall->getArg(2)->getLocStart(), 1043 diag::err_typecheck_call_too_many_args) 1044 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1045 << Fn->getSourceRange() 1046 << SourceRange(TheCall->getArg(2)->getLocStart(), 1047 (*(TheCall->arg_end()-1))->getLocEnd()); 1048 return true; 1049 } 1050 1051 if (TheCall->getNumArgs() < 2) { 1052 return Diag(TheCall->getLocEnd(), 1053 diag::err_typecheck_call_too_few_args_at_least) 1054 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 1055 } 1056 1057 // Type-check the first argument normally. 1058 if (checkBuiltinArgument(*this, TheCall, 0)) 1059 return true; 1060 1061 // Determine whether the current function is variadic or not. 1062 BlockScopeInfo *CurBlock = getCurBlock(); 1063 bool isVariadic; 1064 if (CurBlock) 1065 isVariadic = CurBlock->TheDecl->isVariadic(); 1066 else if (FunctionDecl *FD = getCurFunctionDecl()) 1067 isVariadic = FD->isVariadic(); 1068 else 1069 isVariadic = getCurMethodDecl()->isVariadic(); 1070 1071 if (!isVariadic) { 1072 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 1073 return true; 1074 } 1075 1076 // Verify that the second argument to the builtin is the last argument of the 1077 // current function or method. 1078 bool SecondArgIsLastNamedArgument = false; 1079 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 1080 1081 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 1082 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 1083 // FIXME: This isn't correct for methods (results in bogus warning). 1084 // Get the last formal in the current function. 1085 const ParmVarDecl *LastArg; 1086 if (CurBlock) 1087 LastArg = *(CurBlock->TheDecl->param_end()-1); 1088 else if (FunctionDecl *FD = getCurFunctionDecl()) 1089 LastArg = *(FD->param_end()-1); 1090 else 1091 LastArg = *(getCurMethodDecl()->param_end()-1); 1092 SecondArgIsLastNamedArgument = PV == LastArg; 1093 } 1094 } 1095 1096 if (!SecondArgIsLastNamedArgument) 1097 Diag(TheCall->getArg(1)->getLocStart(), 1098 diag::warn_second_parameter_of_va_start_not_last_named_argument); 1099 return false; 1100} 1101 1102/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 1103/// friends. This is declared to take (...), so we have to check everything. 1104bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 1105 if (TheCall->getNumArgs() < 2) 1106 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1107 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 1108 if (TheCall->getNumArgs() > 2) 1109 return Diag(TheCall->getArg(2)->getLocStart(), 1110 diag::err_typecheck_call_too_many_args) 1111 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1112 << SourceRange(TheCall->getArg(2)->getLocStart(), 1113 (*(TheCall->arg_end()-1))->getLocEnd()); 1114 1115 ExprResult OrigArg0 = TheCall->getArg(0); 1116 ExprResult OrigArg1 = TheCall->getArg(1); 1117 1118 // Do standard promotions between the two arguments, returning their common 1119 // type. 1120 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 1121 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 1122 return true; 1123 1124 // Make sure any conversions are pushed back into the call; this is 1125 // type safe since unordered compare builtins are declared as "_Bool 1126 // foo(...)". 1127 TheCall->setArg(0, OrigArg0.get()); 1128 TheCall->setArg(1, OrigArg1.get()); 1129 1130 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 1131 return false; 1132 1133 // If the common type isn't a real floating type, then the arguments were 1134 // invalid for this operation. 1135 if (!Res->isRealFloatingType()) 1136 return Diag(OrigArg0.get()->getLocStart(), 1137 diag::err_typecheck_call_invalid_ordered_compare) 1138 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 1139 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 1140 1141 return false; 1142} 1143 1144/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 1145/// __builtin_isnan and friends. This is declared to take (...), so we have 1146/// to check everything. We expect the last argument to be a floating point 1147/// value. 1148bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 1149 if (TheCall->getNumArgs() < NumArgs) 1150 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1151 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 1152 if (TheCall->getNumArgs() > NumArgs) 1153 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 1154 diag::err_typecheck_call_too_many_args) 1155 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 1156 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 1157 (*(TheCall->arg_end()-1))->getLocEnd()); 1158 1159 Expr *OrigArg = TheCall->getArg(NumArgs-1); 1160 1161 if (OrigArg->isTypeDependent()) 1162 return false; 1163 1164 // This operation requires a non-_Complex floating-point number. 1165 if (!OrigArg->getType()->isRealFloatingType()) 1166 return Diag(OrigArg->getLocStart(), 1167 diag::err_typecheck_call_invalid_unary_fp) 1168 << OrigArg->getType() << OrigArg->getSourceRange(); 1169 1170 // If this is an implicit conversion from float -> double, remove it. 1171 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 1172 Expr *CastArg = Cast->getSubExpr(); 1173 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 1174 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 1175 "promotion from float to double is the only expected cast here"); 1176 Cast->setSubExpr(0); 1177 TheCall->setArg(NumArgs-1, CastArg); 1178 OrigArg = CastArg; 1179 } 1180 } 1181 1182 return false; 1183} 1184 1185/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1186// This is declared to take (...), so we have to check everything. 1187ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1188 if (TheCall->getNumArgs() < 2) 1189 return ExprError(Diag(TheCall->getLocEnd(), 1190 diag::err_typecheck_call_too_few_args_at_least) 1191 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1192 << TheCall->getSourceRange()); 1193 1194 // Determine which of the following types of shufflevector we're checking: 1195 // 1) unary, vector mask: (lhs, mask) 1196 // 2) binary, vector mask: (lhs, rhs, mask) 1197 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1198 QualType resType = TheCall->getArg(0)->getType(); 1199 unsigned numElements = 0; 1200 1201 if (!TheCall->getArg(0)->isTypeDependent() && 1202 !TheCall->getArg(1)->isTypeDependent()) { 1203 QualType LHSType = TheCall->getArg(0)->getType(); 1204 QualType RHSType = TheCall->getArg(1)->getType(); 1205 1206 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 1207 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 1208 << SourceRange(TheCall->getArg(0)->getLocStart(), 1209 TheCall->getArg(1)->getLocEnd()); 1210 return ExprError(); 1211 } 1212 1213 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1214 unsigned numResElements = TheCall->getNumArgs() - 2; 1215 1216 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1217 // with mask. If so, verify that RHS is an integer vector type with the 1218 // same number of elts as lhs. 1219 if (TheCall->getNumArgs() == 2) { 1220 if (!RHSType->hasIntegerRepresentation() || 1221 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1222 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1223 << SourceRange(TheCall->getArg(1)->getLocStart(), 1224 TheCall->getArg(1)->getLocEnd()); 1225 numResElements = numElements; 1226 } 1227 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1228 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1229 << SourceRange(TheCall->getArg(0)->getLocStart(), 1230 TheCall->getArg(1)->getLocEnd()); 1231 return ExprError(); 1232 } else if (numElements != numResElements) { 1233 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1234 resType = Context.getVectorType(eltType, numResElements, 1235 VectorType::GenericVector); 1236 } 1237 } 1238 1239 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1240 if (TheCall->getArg(i)->isTypeDependent() || 1241 TheCall->getArg(i)->isValueDependent()) 1242 continue; 1243 1244 llvm::APSInt Result(32); 1245 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1246 return ExprError(Diag(TheCall->getLocStart(), 1247 diag::err_shufflevector_nonconstant_argument) 1248 << TheCall->getArg(i)->getSourceRange()); 1249 1250 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1251 return ExprError(Diag(TheCall->getLocStart(), 1252 diag::err_shufflevector_argument_too_large) 1253 << TheCall->getArg(i)->getSourceRange()); 1254 } 1255 1256 SmallVector<Expr*, 32> exprs; 1257 1258 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1259 exprs.push_back(TheCall->getArg(i)); 1260 TheCall->setArg(i, 0); 1261 } 1262 1263 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 1264 exprs.size(), resType, 1265 TheCall->getCallee()->getLocStart(), 1266 TheCall->getRParenLoc())); 1267} 1268 1269/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1270// This is declared to take (const void*, ...) and can take two 1271// optional constant int args. 1272bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1273 unsigned NumArgs = TheCall->getNumArgs(); 1274 1275 if (NumArgs > 3) 1276 return Diag(TheCall->getLocEnd(), 1277 diag::err_typecheck_call_too_many_args_at_most) 1278 << 0 /*function call*/ << 3 << NumArgs 1279 << TheCall->getSourceRange(); 1280 1281 // Argument 0 is checked for us and the remaining arguments must be 1282 // constant integers. 1283 for (unsigned i = 1; i != NumArgs; ++i) { 1284 Expr *Arg = TheCall->getArg(i); 1285 1286 llvm::APSInt Result; 1287 if (SemaBuiltinConstantArg(TheCall, i, Result)) 1288 return true; 1289 1290 // FIXME: gcc issues a warning and rewrites these to 0. These 1291 // seems especially odd for the third argument since the default 1292 // is 3. 1293 if (i == 1) { 1294 if (Result.getLimitedValue() > 1) 1295 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1296 << "0" << "1" << Arg->getSourceRange(); 1297 } else { 1298 if (Result.getLimitedValue() > 3) 1299 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1300 << "0" << "3" << Arg->getSourceRange(); 1301 } 1302 } 1303 1304 return false; 1305} 1306 1307/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1308/// TheCall is a constant expression. 1309bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1310 llvm::APSInt &Result) { 1311 Expr *Arg = TheCall->getArg(ArgNum); 1312 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1313 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1314 1315 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 1316 1317 if (!Arg->isIntegerConstantExpr(Result, Context)) 1318 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 1319 << FDecl->getDeclName() << Arg->getSourceRange(); 1320 1321 return false; 1322} 1323 1324/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 1325/// int type). This simply type checks that type is one of the defined 1326/// constants (0-3). 1327// For compatibility check 0-3, llvm only handles 0 and 2. 1328bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 1329 llvm::APSInt Result; 1330 1331 // Check constant-ness first. 1332 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1333 return true; 1334 1335 Expr *Arg = TheCall->getArg(1); 1336 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 1337 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1338 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1339 } 1340 1341 return false; 1342} 1343 1344/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1345/// This checks that val is a constant 1. 1346bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1347 Expr *Arg = TheCall->getArg(1); 1348 llvm::APSInt Result; 1349 1350 // TODO: This is less than ideal. Overload this to take a value. 1351 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1352 return true; 1353 1354 if (Result != 1) 1355 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1356 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1357 1358 return false; 1359} 1360 1361// Handle i > 1 ? "x" : "y", recursively. 1362bool Sema::SemaCheckStringLiteral(const Expr *E, Expr **Args, 1363 unsigned NumArgs, bool HasVAListArg, 1364 unsigned format_idx, unsigned firstDataArg, 1365 FormatStringType Type, bool inFunctionCall) { 1366 tryAgain: 1367 if (E->isTypeDependent() || E->isValueDependent()) 1368 return false; 1369 1370 E = E->IgnoreParenCasts(); 1371 1372 switch (E->getStmtClass()) { 1373 case Stmt::BinaryConditionalOperatorClass: 1374 case Stmt::ConditionalOperatorClass: { 1375 const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E); 1376 return SemaCheckStringLiteral(C->getTrueExpr(), Args, NumArgs, HasVAListArg, 1377 format_idx, firstDataArg, Type, 1378 inFunctionCall) 1379 && SemaCheckStringLiteral(C->getFalseExpr(), Args, NumArgs, HasVAListArg, 1380 format_idx, firstDataArg, Type, 1381 inFunctionCall); 1382 } 1383 1384 case Stmt::GNUNullExprClass: 1385 case Stmt::IntegerLiteralClass: 1386 // Technically -Wformat-nonliteral does not warn about this case. 1387 // The behavior of printf and friends in this case is implementation 1388 // dependent. Ideally if the format string cannot be null then 1389 // it should have a 'nonnull' attribute in the function prototype. 1390 return true; 1391 1392 case Stmt::ImplicitCastExprClass: { 1393 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 1394 goto tryAgain; 1395 } 1396 1397 case Stmt::OpaqueValueExprClass: 1398 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 1399 E = src; 1400 goto tryAgain; 1401 } 1402 return false; 1403 1404 case Stmt::PredefinedExprClass: 1405 // While __func__, etc., are technically not string literals, they 1406 // cannot contain format specifiers and thus are not a security 1407 // liability. 1408 return true; 1409 1410 case Stmt::DeclRefExprClass: { 1411 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1412 1413 // As an exception, do not flag errors for variables binding to 1414 // const string literals. 1415 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1416 bool isConstant = false; 1417 QualType T = DR->getType(); 1418 1419 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1420 isConstant = AT->getElementType().isConstant(Context); 1421 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1422 isConstant = T.isConstant(Context) && 1423 PT->getPointeeType().isConstant(Context); 1424 } else if (T->isObjCObjectPointerType()) { 1425 // In ObjC, there is usually no "const ObjectPointer" type, 1426 // so don't check if the pointee type is constant. 1427 isConstant = T.isConstant(Context); 1428 } 1429 1430 if (isConstant) { 1431 if (const Expr *Init = VD->getAnyInitializer()) 1432 return SemaCheckStringLiteral(Init, Args, NumArgs, 1433 HasVAListArg, format_idx, firstDataArg, 1434 Type, /*inFunctionCall*/false); 1435 } 1436 1437 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1438 // special check to see if the format string is a function parameter 1439 // of the function calling the printf function. If the function 1440 // has an attribute indicating it is a printf-like function, then we 1441 // should suppress warnings concerning non-literals being used in a call 1442 // to a vprintf function. For example: 1443 // 1444 // void 1445 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1446 // va_list ap; 1447 // va_start(ap, fmt); 1448 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1449 // ... 1450 // 1451 // 1452 // FIXME: We don't have full attribute support yet, so just check to see 1453 // if the argument is a DeclRefExpr that references a parameter. We'll 1454 // add proper support for checking the attribute later. 1455 if (HasVAListArg) 1456 if (isa<ParmVarDecl>(VD)) 1457 return true; 1458 } 1459 1460 return false; 1461 } 1462 1463 case Stmt::CallExprClass: 1464 case Stmt::CXXMemberCallExprClass: { 1465 const CallExpr *CE = cast<CallExpr>(E); 1466 if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(CE->getCalleeDecl())) { 1467 if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) { 1468 unsigned ArgIndex = FA->getFormatIdx(); 1469 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1470 if (MD->isInstance()) 1471 --ArgIndex; 1472 const Expr *Arg = CE->getArg(ArgIndex - 1); 1473 1474 return SemaCheckStringLiteral(Arg, Args, NumArgs, HasVAListArg, 1475 format_idx, firstDataArg, Type, 1476 inFunctionCall); 1477 } 1478 } 1479 1480 return false; 1481 } 1482 case Stmt::ObjCStringLiteralClass: 1483 case Stmt::StringLiteralClass: { 1484 const StringLiteral *StrE = NULL; 1485 1486 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1487 StrE = ObjCFExpr->getString(); 1488 else 1489 StrE = cast<StringLiteral>(E); 1490 1491 if (StrE) { 1492 CheckFormatString(StrE, E, Args, NumArgs, HasVAListArg, format_idx, 1493 firstDataArg, Type, inFunctionCall); 1494 return true; 1495 } 1496 1497 return false; 1498 } 1499 1500 default: 1501 return false; 1502 } 1503} 1504 1505void 1506Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1507 const Expr * const *ExprArgs, 1508 SourceLocation CallSiteLoc) { 1509 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1510 e = NonNull->args_end(); 1511 i != e; ++i) { 1512 const Expr *ArgExpr = ExprArgs[*i]; 1513 if (ArgExpr->isNullPointerConstant(Context, 1514 Expr::NPC_ValueDependentIsNotNull)) 1515 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 1516 } 1517} 1518 1519Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) { 1520 return llvm::StringSwitch<FormatStringType>(Format->getType()) 1521 .Case("scanf", FST_Scanf) 1522 .Cases("printf", "printf0", FST_Printf) 1523 .Cases("NSString", "CFString", FST_NSString) 1524 .Case("strftime", FST_Strftime) 1525 .Case("strfmon", FST_Strfmon) 1526 .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf) 1527 .Default(FST_Unknown); 1528} 1529 1530/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar 1531/// functions) for correct use of format strings. 1532void Sema::CheckFormatArguments(const FormatAttr *Format, CallExpr *TheCall) { 1533 bool IsCXXMember = false; 1534 // The way the format attribute works in GCC, the implicit this argument 1535 // of member functions is counted. However, it doesn't appear in our own 1536 // lists, so decrement format_idx in that case. 1537 IsCXXMember = isa<CXXMemberCallExpr>(TheCall); 1538 CheckFormatArguments(Format, TheCall->getArgs(), TheCall->getNumArgs(), 1539 IsCXXMember, TheCall->getRParenLoc(), 1540 TheCall->getCallee()->getSourceRange()); 1541} 1542 1543void Sema::CheckFormatArguments(const FormatAttr *Format, Expr **Args, 1544 unsigned NumArgs, bool IsCXXMember, 1545 SourceLocation Loc, SourceRange Range) { 1546 bool HasVAListArg = Format->getFirstArg() == 0; 1547 unsigned format_idx = Format->getFormatIdx() - 1; 1548 unsigned firstDataArg = HasVAListArg ? 0 : Format->getFirstArg() - 1; 1549 if (IsCXXMember) { 1550 if (format_idx == 0) 1551 return; 1552 --format_idx; 1553 if(firstDataArg != 0) 1554 --firstDataArg; 1555 } 1556 CheckFormatArguments(Args, NumArgs, HasVAListArg, format_idx, 1557 firstDataArg, GetFormatStringType(Format), Loc, Range); 1558} 1559 1560void Sema::CheckFormatArguments(Expr **Args, unsigned NumArgs, 1561 bool HasVAListArg, unsigned format_idx, 1562 unsigned firstDataArg, FormatStringType Type, 1563 SourceLocation Loc, SourceRange Range) { 1564 // CHECK: printf/scanf-like function is called with no format string. 1565 if (format_idx >= NumArgs) { 1566 Diag(Loc, diag::warn_missing_format_string) << Range; 1567 return; 1568 } 1569 1570 const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts(); 1571 1572 // CHECK: format string is not a string literal. 1573 // 1574 // Dynamically generated format strings are difficult to 1575 // automatically vet at compile time. Requiring that format strings 1576 // are string literals: (1) permits the checking of format strings by 1577 // the compiler and thereby (2) can practically remove the source of 1578 // many format string exploits. 1579 1580 // Format string can be either ObjC string (e.g. @"%d") or 1581 // C string (e.g. "%d") 1582 // ObjC string uses the same format specifiers as C string, so we can use 1583 // the same format string checking logic for both ObjC and C strings. 1584 if (SemaCheckStringLiteral(OrigFormatExpr, Args, NumArgs, HasVAListArg, 1585 format_idx, firstDataArg, Type)) 1586 return; // Literal format string found, check done! 1587 1588 // Strftime is particular as it always uses a single 'time' argument, 1589 // so it is safe to pass a non-literal string. 1590 if (Type == FST_Strftime) 1591 return; 1592 1593 // Do not emit diag when the string param is a macro expansion and the 1594 // format is either NSString or CFString. This is a hack to prevent 1595 // diag when using the NSLocalizedString and CFCopyLocalizedString macros 1596 // which are usually used in place of NS and CF string literals. 1597 if (Type == FST_NSString && Args[format_idx]->getLocStart().isMacroID()) 1598 return; 1599 1600 // If there are no arguments specified, warn with -Wformat-security, otherwise 1601 // warn only with -Wformat-nonliteral. 1602 if (NumArgs == format_idx+1) 1603 Diag(Args[format_idx]->getLocStart(), 1604 diag::warn_format_nonliteral_noargs) 1605 << OrigFormatExpr->getSourceRange(); 1606 else 1607 Diag(Args[format_idx]->getLocStart(), 1608 diag::warn_format_nonliteral) 1609 << OrigFormatExpr->getSourceRange(); 1610} 1611 1612namespace { 1613class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1614protected: 1615 Sema &S; 1616 const StringLiteral *FExpr; 1617 const Expr *OrigFormatExpr; 1618 const unsigned FirstDataArg; 1619 const unsigned NumDataArgs; 1620 const bool IsObjCLiteral; 1621 const char *Beg; // Start of format string. 1622 const bool HasVAListArg; 1623 const Expr * const *Args; 1624 const unsigned NumArgs; 1625 unsigned FormatIdx; 1626 llvm::BitVector CoveredArgs; 1627 bool usesPositionalArgs; 1628 bool atFirstArg; 1629 bool inFunctionCall; 1630public: 1631 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1632 const Expr *origFormatExpr, unsigned firstDataArg, 1633 unsigned numDataArgs, bool isObjCLiteral, 1634 const char *beg, bool hasVAListArg, 1635 Expr **args, unsigned numArgs, 1636 unsigned formatIdx, bool inFunctionCall) 1637 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1638 FirstDataArg(firstDataArg), 1639 NumDataArgs(numDataArgs), 1640 IsObjCLiteral(isObjCLiteral), Beg(beg), 1641 HasVAListArg(hasVAListArg), 1642 Args(args), NumArgs(numArgs), FormatIdx(formatIdx), 1643 usesPositionalArgs(false), atFirstArg(true), 1644 inFunctionCall(inFunctionCall) { 1645 CoveredArgs.resize(numDataArgs); 1646 CoveredArgs.reset(); 1647 } 1648 1649 void DoneProcessing(); 1650 1651 void HandleIncompleteSpecifier(const char *startSpecifier, 1652 unsigned specifierLen); 1653 1654 virtual void HandleInvalidPosition(const char *startSpecifier, 1655 unsigned specifierLen, 1656 analyze_format_string::PositionContext p); 1657 1658 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1659 1660 void HandleNullChar(const char *nullCharacter); 1661 1662 template <typename Range> 1663 static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall, 1664 const Expr *ArgumentExpr, 1665 PartialDiagnostic PDiag, 1666 SourceLocation StringLoc, 1667 bool IsStringLocation, Range StringRange, 1668 FixItHint Fixit = FixItHint()); 1669 1670protected: 1671 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1672 const char *startSpec, 1673 unsigned specifierLen, 1674 const char *csStart, unsigned csLen); 1675 1676 void HandlePositionalNonpositionalArgs(SourceLocation Loc, 1677 const char *startSpec, 1678 unsigned specifierLen); 1679 1680 SourceRange getFormatStringRange(); 1681 CharSourceRange getSpecifierRange(const char *startSpecifier, 1682 unsigned specifierLen); 1683 SourceLocation getLocationOfByte(const char *x); 1684 1685 const Expr *getDataArg(unsigned i) const; 1686 1687 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1688 const analyze_format_string::ConversionSpecifier &CS, 1689 const char *startSpecifier, unsigned specifierLen, 1690 unsigned argIndex); 1691 1692 template <typename Range> 1693 void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc, 1694 bool IsStringLocation, Range StringRange, 1695 FixItHint Fixit = FixItHint()); 1696 1697 void CheckPositionalAndNonpositionalArgs( 1698 const analyze_format_string::FormatSpecifier *FS); 1699}; 1700} 1701 1702SourceRange CheckFormatHandler::getFormatStringRange() { 1703 return OrigFormatExpr->getSourceRange(); 1704} 1705 1706CharSourceRange CheckFormatHandler:: 1707getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1708 SourceLocation Start = getLocationOfByte(startSpecifier); 1709 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1710 1711 // Advance the end SourceLocation by one due to half-open ranges. 1712 End = End.getLocWithOffset(1); 1713 1714 return CharSourceRange::getCharRange(Start, End); 1715} 1716 1717SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 1718 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1719} 1720 1721void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 1722 unsigned specifierLen){ 1723 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier), 1724 getLocationOfByte(startSpecifier), 1725 /*IsStringLocation*/true, 1726 getSpecifierRange(startSpecifier, specifierLen)); 1727} 1728 1729void 1730CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1731 analyze_format_string::PositionContext p) { 1732 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier) 1733 << (unsigned) p, 1734 getLocationOfByte(startPos), /*IsStringLocation*/true, 1735 getSpecifierRange(startPos, posLen)); 1736} 1737 1738void CheckFormatHandler::HandleZeroPosition(const char *startPos, 1739 unsigned posLen) { 1740 EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier), 1741 getLocationOfByte(startPos), 1742 /*IsStringLocation*/true, 1743 getSpecifierRange(startPos, posLen)); 1744} 1745 1746void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 1747 if (!IsObjCLiteral) { 1748 // The presence of a null character is likely an error. 1749 EmitFormatDiagnostic( 1750 S.PDiag(diag::warn_printf_format_string_contains_null_char), 1751 getLocationOfByte(nullCharacter), /*IsStringLocation*/true, 1752 getFormatStringRange()); 1753 } 1754} 1755 1756const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 1757 return Args[FirstDataArg + i]; 1758} 1759 1760void CheckFormatHandler::DoneProcessing() { 1761 // Does the number of data arguments exceed the number of 1762 // format conversions in the format string? 1763 if (!HasVAListArg) { 1764 // Find any arguments that weren't covered. 1765 CoveredArgs.flip(); 1766 signed notCoveredArg = CoveredArgs.find_first(); 1767 if (notCoveredArg >= 0) { 1768 assert((unsigned)notCoveredArg < NumDataArgs); 1769 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used), 1770 getDataArg((unsigned) notCoveredArg)->getLocStart(), 1771 /*IsStringLocation*/false, getFormatStringRange()); 1772 } 1773 } 1774} 1775 1776bool 1777CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 1778 SourceLocation Loc, 1779 const char *startSpec, 1780 unsigned specifierLen, 1781 const char *csStart, 1782 unsigned csLen) { 1783 1784 bool keepGoing = true; 1785 if (argIndex < NumDataArgs) { 1786 // Consider the argument coverered, even though the specifier doesn't 1787 // make sense. 1788 CoveredArgs.set(argIndex); 1789 } 1790 else { 1791 // If argIndex exceeds the number of data arguments we 1792 // don't issue a warning because that is just a cascade of warnings (and 1793 // they may have intended '%%' anyway). We don't want to continue processing 1794 // the format string after this point, however, as we will like just get 1795 // gibberish when trying to match arguments. 1796 keepGoing = false; 1797 } 1798 1799 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion) 1800 << StringRef(csStart, csLen), 1801 Loc, /*IsStringLocation*/true, 1802 getSpecifierRange(startSpec, specifierLen)); 1803 1804 return keepGoing; 1805} 1806 1807void 1808CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc, 1809 const char *startSpec, 1810 unsigned specifierLen) { 1811 EmitFormatDiagnostic( 1812 S.PDiag(diag::warn_format_mix_positional_nonpositional_args), 1813 Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen)); 1814} 1815 1816bool 1817CheckFormatHandler::CheckNumArgs( 1818 const analyze_format_string::FormatSpecifier &FS, 1819 const analyze_format_string::ConversionSpecifier &CS, 1820 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 1821 1822 if (argIndex >= NumDataArgs) { 1823 PartialDiagnostic PDiag = FS.usesPositionalArg() 1824 ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args) 1825 << (argIndex+1) << NumDataArgs) 1826 : S.PDiag(diag::warn_printf_insufficient_data_args); 1827 EmitFormatDiagnostic( 1828 PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true, 1829 getSpecifierRange(startSpecifier, specifierLen)); 1830 return false; 1831 } 1832 return true; 1833} 1834 1835template<typename Range> 1836void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag, 1837 SourceLocation Loc, 1838 bool IsStringLocation, 1839 Range StringRange, 1840 FixItHint FixIt) { 1841 EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag, 1842 Loc, IsStringLocation, StringRange, FixIt); 1843} 1844 1845/// \brief If the format string is not within the funcion call, emit a note 1846/// so that the function call and string are in diagnostic messages. 1847/// 1848/// \param inFunctionCall if true, the format string is within the function 1849/// call and only one diagnostic message will be produced. Otherwise, an 1850/// extra note will be emitted pointing to location of the format string. 1851/// 1852/// \param ArgumentExpr the expression that is passed as the format string 1853/// argument in the function call. Used for getting locations when two 1854/// diagnostics are emitted. 1855/// 1856/// \param PDiag the callee should already have provided any strings for the 1857/// diagnostic message. This function only adds locations and fixits 1858/// to diagnostics. 1859/// 1860/// \param Loc primary location for diagnostic. If two diagnostics are 1861/// required, one will be at Loc and a new SourceLocation will be created for 1862/// the other one. 1863/// 1864/// \param IsStringLocation if true, Loc points to the format string should be 1865/// used for the note. Otherwise, Loc points to the argument list and will 1866/// be used with PDiag. 1867/// 1868/// \param StringRange some or all of the string to highlight. This is 1869/// templated so it can accept either a CharSourceRange or a SourceRange. 1870/// 1871/// \param Fixit optional fix it hint for the format string. 1872template<typename Range> 1873void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall, 1874 const Expr *ArgumentExpr, 1875 PartialDiagnostic PDiag, 1876 SourceLocation Loc, 1877 bool IsStringLocation, 1878 Range StringRange, 1879 FixItHint FixIt) { 1880 if (InFunctionCall) 1881 S.Diag(Loc, PDiag) << StringRange << FixIt; 1882 else { 1883 S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag) 1884 << ArgumentExpr->getSourceRange(); 1885 S.Diag(IsStringLocation ? Loc : StringRange.getBegin(), 1886 diag::note_format_string_defined) 1887 << StringRange << FixIt; 1888 } 1889} 1890 1891//===--- CHECK: Printf format string checking ------------------------------===// 1892 1893namespace { 1894class CheckPrintfHandler : public CheckFormatHandler { 1895public: 1896 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1897 const Expr *origFormatExpr, unsigned firstDataArg, 1898 unsigned numDataArgs, bool isObjCLiteral, 1899 const char *beg, bool hasVAListArg, 1900 Expr **Args, unsigned NumArgs, 1901 unsigned formatIdx, bool inFunctionCall) 1902 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1903 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1904 Args, NumArgs, formatIdx, inFunctionCall) {} 1905 1906 1907 bool HandleInvalidPrintfConversionSpecifier( 1908 const analyze_printf::PrintfSpecifier &FS, 1909 const char *startSpecifier, 1910 unsigned specifierLen); 1911 1912 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 1913 const char *startSpecifier, 1914 unsigned specifierLen); 1915 1916 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 1917 const char *startSpecifier, unsigned specifierLen); 1918 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 1919 const analyze_printf::OptionalAmount &Amt, 1920 unsigned type, 1921 const char *startSpecifier, unsigned specifierLen); 1922 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1923 const analyze_printf::OptionalFlag &flag, 1924 const char *startSpecifier, unsigned specifierLen); 1925 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 1926 const analyze_printf::OptionalFlag &ignoredFlag, 1927 const analyze_printf::OptionalFlag &flag, 1928 const char *startSpecifier, unsigned specifierLen); 1929}; 1930} 1931 1932bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 1933 const analyze_printf::PrintfSpecifier &FS, 1934 const char *startSpecifier, 1935 unsigned specifierLen) { 1936 const analyze_printf::PrintfConversionSpecifier &CS = 1937 FS.getConversionSpecifier(); 1938 1939 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1940 getLocationOfByte(CS.getStart()), 1941 startSpecifier, specifierLen, 1942 CS.getStart(), CS.getLength()); 1943} 1944 1945bool CheckPrintfHandler::HandleAmount( 1946 const analyze_format_string::OptionalAmount &Amt, 1947 unsigned k, const char *startSpecifier, 1948 unsigned specifierLen) { 1949 1950 if (Amt.hasDataArgument()) { 1951 if (!HasVAListArg) { 1952 unsigned argIndex = Amt.getArgIndex(); 1953 if (argIndex >= NumDataArgs) { 1954 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg) 1955 << k, 1956 getLocationOfByte(Amt.getStart()), 1957 /*IsStringLocation*/true, 1958 getSpecifierRange(startSpecifier, specifierLen)); 1959 // Don't do any more checking. We will just emit 1960 // spurious errors. 1961 return false; 1962 } 1963 1964 // Type check the data argument. It should be an 'int'. 1965 // Although not in conformance with C99, we also allow the argument to be 1966 // an 'unsigned int' as that is a reasonably safe case. GCC also 1967 // doesn't emit a warning for that case. 1968 CoveredArgs.set(argIndex); 1969 const Expr *Arg = getDataArg(argIndex); 1970 QualType T = Arg->getType(); 1971 1972 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1973 assert(ATR.isValid()); 1974 1975 if (!ATR.matchesType(S.Context, T)) { 1976 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type) 1977 << k << ATR.getRepresentativeTypeName(S.Context) 1978 << T << Arg->getSourceRange(), 1979 getLocationOfByte(Amt.getStart()), 1980 /*IsStringLocation*/true, 1981 getSpecifierRange(startSpecifier, specifierLen)); 1982 // Don't do any more checking. We will just emit 1983 // spurious errors. 1984 return false; 1985 } 1986 } 1987 } 1988 return true; 1989} 1990 1991void CheckPrintfHandler::HandleInvalidAmount( 1992 const analyze_printf::PrintfSpecifier &FS, 1993 const analyze_printf::OptionalAmount &Amt, 1994 unsigned type, 1995 const char *startSpecifier, 1996 unsigned specifierLen) { 1997 const analyze_printf::PrintfConversionSpecifier &CS = 1998 FS.getConversionSpecifier(); 1999 2000 FixItHint fixit = 2001 Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant 2002 ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 2003 Amt.getConstantLength())) 2004 : FixItHint(); 2005 2006 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount) 2007 << type << CS.toString(), 2008 getLocationOfByte(Amt.getStart()), 2009 /*IsStringLocation*/true, 2010 getSpecifierRange(startSpecifier, specifierLen), 2011 fixit); 2012} 2013 2014void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2015 const analyze_printf::OptionalFlag &flag, 2016 const char *startSpecifier, 2017 unsigned specifierLen) { 2018 // Warn about pointless flag with a fixit removal. 2019 const analyze_printf::PrintfConversionSpecifier &CS = 2020 FS.getConversionSpecifier(); 2021 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag) 2022 << flag.toString() << CS.toString(), 2023 getLocationOfByte(flag.getPosition()), 2024 /*IsStringLocation*/true, 2025 getSpecifierRange(startSpecifier, specifierLen), 2026 FixItHint::CreateRemoval( 2027 getSpecifierRange(flag.getPosition(), 1))); 2028} 2029 2030void CheckPrintfHandler::HandleIgnoredFlag( 2031 const analyze_printf::PrintfSpecifier &FS, 2032 const analyze_printf::OptionalFlag &ignoredFlag, 2033 const analyze_printf::OptionalFlag &flag, 2034 const char *startSpecifier, 2035 unsigned specifierLen) { 2036 // Warn about ignored flag with a fixit removal. 2037 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag) 2038 << ignoredFlag.toString() << flag.toString(), 2039 getLocationOfByte(ignoredFlag.getPosition()), 2040 /*IsStringLocation*/true, 2041 getSpecifierRange(startSpecifier, specifierLen), 2042 FixItHint::CreateRemoval( 2043 getSpecifierRange(ignoredFlag.getPosition(), 1))); 2044} 2045 2046bool 2047CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 2048 &FS, 2049 const char *startSpecifier, 2050 unsigned specifierLen) { 2051 2052 using namespace analyze_format_string; 2053 using namespace analyze_printf; 2054 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 2055 2056 if (FS.consumesDataArgument()) { 2057 if (atFirstArg) { 2058 atFirstArg = false; 2059 usesPositionalArgs = FS.usesPositionalArg(); 2060 } 2061 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2062 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2063 startSpecifier, specifierLen); 2064 return false; 2065 } 2066 } 2067 2068 // First check if the field width, precision, and conversion specifier 2069 // have matching data arguments. 2070 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 2071 startSpecifier, specifierLen)) { 2072 return false; 2073 } 2074 2075 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 2076 startSpecifier, specifierLen)) { 2077 return false; 2078 } 2079 2080 if (!CS.consumesDataArgument()) { 2081 // FIXME: Technically specifying a precision or field width here 2082 // makes no sense. Worth issuing a warning at some point. 2083 return true; 2084 } 2085 2086 // Consume the argument. 2087 unsigned argIndex = FS.getArgIndex(); 2088 if (argIndex < NumDataArgs) { 2089 // The check to see if the argIndex is valid will come later. 2090 // We set the bit here because we may exit early from this 2091 // function if we encounter some other error. 2092 CoveredArgs.set(argIndex); 2093 } 2094 2095 // Check for using an Objective-C specific conversion specifier 2096 // in a non-ObjC literal. 2097 if (!IsObjCLiteral && CS.isObjCArg()) { 2098 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 2099 specifierLen); 2100 } 2101 2102 // Check for invalid use of field width 2103 if (!FS.hasValidFieldWidth()) { 2104 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 2105 startSpecifier, specifierLen); 2106 } 2107 2108 // Check for invalid use of precision 2109 if (!FS.hasValidPrecision()) { 2110 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 2111 startSpecifier, specifierLen); 2112 } 2113 2114 // Check each flag does not conflict with any other component. 2115 if (!FS.hasValidThousandsGroupingPrefix()) 2116 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 2117 if (!FS.hasValidLeadingZeros()) 2118 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 2119 if (!FS.hasValidPlusPrefix()) 2120 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 2121 if (!FS.hasValidSpacePrefix()) 2122 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 2123 if (!FS.hasValidAlternativeForm()) 2124 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 2125 if (!FS.hasValidLeftJustified()) 2126 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 2127 2128 // Check that flags are not ignored by another flag 2129 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 2130 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 2131 startSpecifier, specifierLen); 2132 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 2133 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 2134 startSpecifier, specifierLen); 2135 2136 // Check the length modifier is valid with the given conversion specifier. 2137 const LengthModifier &LM = FS.getLengthModifier(); 2138 if (!FS.hasValidLengthModifier()) 2139 EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length) 2140 << LM.toString() << CS.toString(), 2141 getLocationOfByte(LM.getStart()), 2142 /*IsStringLocation*/true, 2143 getSpecifierRange(startSpecifier, specifierLen), 2144 FixItHint::CreateRemoval( 2145 getSpecifierRange(LM.getStart(), 2146 LM.getLength()))); 2147 2148 // Are we using '%n'? 2149 if (CS.getKind() == ConversionSpecifier::nArg) { 2150 // Issue a warning about this being a possible security issue. 2151 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_write_back), 2152 getLocationOfByte(CS.getStart()), 2153 /*IsStringLocation*/true, 2154 getSpecifierRange(startSpecifier, specifierLen)); 2155 // Continue checking the other format specifiers. 2156 return true; 2157 } 2158 2159 // The remaining checks depend on the data arguments. 2160 if (HasVAListArg) 2161 return true; 2162 2163 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2164 return false; 2165 2166 // Now type check the data expression that matches the 2167 // format specifier. 2168 const Expr *Ex = getDataArg(argIndex); 2169 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context, 2170 IsObjCLiteral); 2171 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 2172 // Check if we didn't match because of an implicit cast from a 'char' 2173 // or 'short' to an 'int'. This is done because printf is a varargs 2174 // function. 2175 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 2176 if (ICE->getType() == S.Context.IntTy) { 2177 // All further checking is done on the subexpression. 2178 Ex = ICE->getSubExpr(); 2179 if (ATR.matchesType(S.Context, Ex->getType())) 2180 return true; 2181 } 2182 2183 // We may be able to offer a FixItHint if it is a supported type. 2184 PrintfSpecifier fixedFS = FS; 2185 bool success = fixedFS.fixType(Ex->getType(), S.getLangOptions()); 2186 2187 if (success) { 2188 // Get the fix string from the fixed format specifier 2189 SmallString<128> buf; 2190 llvm::raw_svector_ostream os(buf); 2191 fixedFS.toString(os); 2192 2193 EmitFormatDiagnostic( 2194 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2195 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2196 << Ex->getSourceRange(), 2197 getLocationOfByte(CS.getStart()), 2198 /*IsStringLocation*/true, 2199 getSpecifierRange(startSpecifier, specifierLen), 2200 FixItHint::CreateReplacement( 2201 getSpecifierRange(startSpecifier, specifierLen), 2202 os.str())); 2203 } 2204 else { 2205 EmitFormatDiagnostic( 2206 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2207 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2208 << getSpecifierRange(startSpecifier, specifierLen) 2209 << Ex->getSourceRange(), 2210 getLocationOfByte(CS.getStart()), 2211 true, 2212 getSpecifierRange(startSpecifier, specifierLen)); 2213 } 2214 } 2215 2216 return true; 2217} 2218 2219//===--- CHECK: Scanf format string checking ------------------------------===// 2220 2221namespace { 2222class CheckScanfHandler : public CheckFormatHandler { 2223public: 2224 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 2225 const Expr *origFormatExpr, unsigned firstDataArg, 2226 unsigned numDataArgs, bool isObjCLiteral, 2227 const char *beg, bool hasVAListArg, 2228 Expr **Args, unsigned NumArgs, 2229 unsigned formatIdx, bool inFunctionCall) 2230 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2231 numDataArgs, isObjCLiteral, beg, hasVAListArg, 2232 Args, NumArgs, formatIdx, inFunctionCall) {} 2233 2234 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 2235 const char *startSpecifier, 2236 unsigned specifierLen); 2237 2238 bool HandleInvalidScanfConversionSpecifier( 2239 const analyze_scanf::ScanfSpecifier &FS, 2240 const char *startSpecifier, 2241 unsigned specifierLen); 2242 2243 void HandleIncompleteScanList(const char *start, const char *end); 2244}; 2245} 2246 2247void CheckScanfHandler::HandleIncompleteScanList(const char *start, 2248 const char *end) { 2249 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 2250 getLocationOfByte(end), /*IsStringLocation*/true, 2251 getSpecifierRange(start, end - start)); 2252} 2253 2254bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 2255 const analyze_scanf::ScanfSpecifier &FS, 2256 const char *startSpecifier, 2257 unsigned specifierLen) { 2258 2259 const analyze_scanf::ScanfConversionSpecifier &CS = 2260 FS.getConversionSpecifier(); 2261 2262 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2263 getLocationOfByte(CS.getStart()), 2264 startSpecifier, specifierLen, 2265 CS.getStart(), CS.getLength()); 2266} 2267 2268bool CheckScanfHandler::HandleScanfSpecifier( 2269 const analyze_scanf::ScanfSpecifier &FS, 2270 const char *startSpecifier, 2271 unsigned specifierLen) { 2272 2273 using namespace analyze_scanf; 2274 using namespace analyze_format_string; 2275 2276 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 2277 2278 // Handle case where '%' and '*' don't consume an argument. These shouldn't 2279 // be used to decide if we are using positional arguments consistently. 2280 if (FS.consumesDataArgument()) { 2281 if (atFirstArg) { 2282 atFirstArg = false; 2283 usesPositionalArgs = FS.usesPositionalArg(); 2284 } 2285 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2286 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2287 startSpecifier, specifierLen); 2288 return false; 2289 } 2290 } 2291 2292 // Check if the field with is non-zero. 2293 const OptionalAmount &Amt = FS.getFieldWidth(); 2294 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 2295 if (Amt.getConstantAmount() == 0) { 2296 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 2297 Amt.getConstantLength()); 2298 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 2299 getLocationOfByte(Amt.getStart()), 2300 /*IsStringLocation*/true, R, 2301 FixItHint::CreateRemoval(R)); 2302 } 2303 } 2304 2305 if (!FS.consumesDataArgument()) { 2306 // FIXME: Technically specifying a precision or field width here 2307 // makes no sense. Worth issuing a warning at some point. 2308 return true; 2309 } 2310 2311 // Consume the argument. 2312 unsigned argIndex = FS.getArgIndex(); 2313 if (argIndex < NumDataArgs) { 2314 // The check to see if the argIndex is valid will come later. 2315 // We set the bit here because we may exit early from this 2316 // function if we encounter some other error. 2317 CoveredArgs.set(argIndex); 2318 } 2319 2320 // Check the length modifier is valid with the given conversion specifier. 2321 const LengthModifier &LM = FS.getLengthModifier(); 2322 if (!FS.hasValidLengthModifier()) { 2323 const CharSourceRange &R = getSpecifierRange(LM.getStart(), LM.getLength()); 2324 EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length) 2325 << LM.toString() << CS.toString() 2326 << getSpecifierRange(startSpecifier, specifierLen), 2327 getLocationOfByte(LM.getStart()), 2328 /*IsStringLocation*/true, R, 2329 FixItHint::CreateRemoval(R)); 2330 } 2331 2332 // The remaining checks depend on the data arguments. 2333 if (HasVAListArg) 2334 return true; 2335 2336 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2337 return false; 2338 2339 // Check that the argument type matches the format specifier. 2340 const Expr *Ex = getDataArg(argIndex); 2341 const analyze_scanf::ScanfArgTypeResult &ATR = FS.getArgType(S.Context); 2342 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 2343 ScanfSpecifier fixedFS = FS; 2344 bool success = fixedFS.fixType(Ex->getType(), S.getLangOptions()); 2345 2346 if (success) { 2347 // Get the fix string from the fixed format specifier. 2348 SmallString<128> buf; 2349 llvm::raw_svector_ostream os(buf); 2350 fixedFS.toString(os); 2351 2352 EmitFormatDiagnostic( 2353 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2354 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2355 << Ex->getSourceRange(), 2356 getLocationOfByte(CS.getStart()), 2357 /*IsStringLocation*/true, 2358 getSpecifierRange(startSpecifier, specifierLen), 2359 FixItHint::CreateReplacement( 2360 getSpecifierRange(startSpecifier, specifierLen), 2361 os.str())); 2362 } else { 2363 EmitFormatDiagnostic( 2364 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2365 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2366 << Ex->getSourceRange(), 2367 getLocationOfByte(CS.getStart()), 2368 /*IsStringLocation*/true, 2369 getSpecifierRange(startSpecifier, specifierLen)); 2370 } 2371 } 2372 2373 return true; 2374} 2375 2376void Sema::CheckFormatString(const StringLiteral *FExpr, 2377 const Expr *OrigFormatExpr, 2378 Expr **Args, unsigned NumArgs, 2379 bool HasVAListArg, unsigned format_idx, 2380 unsigned firstDataArg, FormatStringType Type, 2381 bool inFunctionCall) { 2382 2383 // CHECK: is the format string a wide literal? 2384 if (!FExpr->isAscii()) { 2385 CheckFormatHandler::EmitFormatDiagnostic( 2386 *this, inFunctionCall, Args[format_idx], 2387 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 2388 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2389 return; 2390 } 2391 2392 // Str - The format string. NOTE: this is NOT null-terminated! 2393 StringRef StrRef = FExpr->getString(); 2394 const char *Str = StrRef.data(); 2395 unsigned StrLen = StrRef.size(); 2396 const unsigned numDataArgs = NumArgs - firstDataArg; 2397 2398 // CHECK: empty format string? 2399 if (StrLen == 0 && numDataArgs > 0) { 2400 CheckFormatHandler::EmitFormatDiagnostic( 2401 *this, inFunctionCall, Args[format_idx], 2402 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 2403 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2404 return; 2405 } 2406 2407 if (Type == FST_Printf || Type == FST_NSString) { 2408 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2409 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2410 Str, HasVAListArg, Args, NumArgs, format_idx, 2411 inFunctionCall); 2412 2413 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen, 2414 getLangOptions())) 2415 H.DoneProcessing(); 2416 } else if (Type == FST_Scanf) { 2417 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2418 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2419 Str, HasVAListArg, Args, NumArgs, format_idx, 2420 inFunctionCall); 2421 2422 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen, 2423 getLangOptions())) 2424 H.DoneProcessing(); 2425 } // TODO: handle other formats 2426} 2427 2428//===--- CHECK: Standard memory functions ---------------------------------===// 2429 2430/// \brief Determine whether the given type is a dynamic class type (e.g., 2431/// whether it has a vtable). 2432static bool isDynamicClassType(QualType T) { 2433 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 2434 if (CXXRecordDecl *Definition = Record->getDefinition()) 2435 if (Definition->isDynamicClass()) 2436 return true; 2437 2438 return false; 2439} 2440 2441/// \brief If E is a sizeof expression, returns its argument expression, 2442/// otherwise returns NULL. 2443static const Expr *getSizeOfExprArg(const Expr* E) { 2444 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2445 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2446 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 2447 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 2448 2449 return 0; 2450} 2451 2452/// \brief If E is a sizeof expression, returns its argument type. 2453static QualType getSizeOfArgType(const Expr* E) { 2454 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2455 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2456 if (SizeOf->getKind() == clang::UETT_SizeOf) 2457 return SizeOf->getTypeOfArgument(); 2458 2459 return QualType(); 2460} 2461 2462/// \brief Check for dangerous or invalid arguments to memset(). 2463/// 2464/// This issues warnings on known problematic, dangerous or unspecified 2465/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 2466/// function calls. 2467/// 2468/// \param Call The call expression to diagnose. 2469void Sema::CheckMemaccessArguments(const CallExpr *Call, 2470 unsigned BId, 2471 IdentifierInfo *FnName) { 2472 assert(BId != 0); 2473 2474 // It is possible to have a non-standard definition of memset. Validate 2475 // we have enough arguments, and if not, abort further checking. 2476 unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3); 2477 if (Call->getNumArgs() < ExpectedNumArgs) 2478 return; 2479 2480 unsigned LastArg = (BId == Builtin::BImemset || 2481 BId == Builtin::BIstrndup ? 1 : 2); 2482 unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2); 2483 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 2484 2485 // We have special checking when the length is a sizeof expression. 2486 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 2487 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 2488 llvm::FoldingSetNodeID SizeOfArgID; 2489 2490 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 2491 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 2492 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 2493 2494 QualType DestTy = Dest->getType(); 2495 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 2496 QualType PointeeTy = DestPtrTy->getPointeeType(); 2497 2498 // Never warn about void type pointers. This can be used to suppress 2499 // false positives. 2500 if (PointeeTy->isVoidType()) 2501 continue; 2502 2503 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 2504 // actually comparing the expressions for equality. Because computing the 2505 // expression IDs can be expensive, we only do this if the diagnostic is 2506 // enabled. 2507 if (SizeOfArg && 2508 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 2509 SizeOfArg->getExprLoc())) { 2510 // We only compute IDs for expressions if the warning is enabled, and 2511 // cache the sizeof arg's ID. 2512 if (SizeOfArgID == llvm::FoldingSetNodeID()) 2513 SizeOfArg->Profile(SizeOfArgID, Context, true); 2514 llvm::FoldingSetNodeID DestID; 2515 Dest->Profile(DestID, Context, true); 2516 if (DestID == SizeOfArgID) { 2517 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 2518 // over sizeof(src) as well. 2519 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 2520 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 2521 if (UnaryOp->getOpcode() == UO_AddrOf) 2522 ActionIdx = 1; // If its an address-of operator, just remove it. 2523 if (Context.getTypeSize(PointeeTy) == Context.getCharWidth()) 2524 ActionIdx = 2; // If the pointee's size is sizeof(char), 2525 // suggest an explicit length. 2526 unsigned DestSrcSelect = 2527 (BId == Builtin::BIstrndup ? 1 : ArgIdx); 2528 DiagRuntimeBehavior(SizeOfArg->getExprLoc(), Dest, 2529 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 2530 << FnName << DestSrcSelect << ActionIdx 2531 << Dest->getSourceRange() 2532 << SizeOfArg->getSourceRange()); 2533 break; 2534 } 2535 } 2536 2537 // Also check for cases where the sizeof argument is the exact same 2538 // type as the memory argument, and where it points to a user-defined 2539 // record type. 2540 if (SizeOfArgTy != QualType()) { 2541 if (PointeeTy->isRecordType() && 2542 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 2543 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 2544 PDiag(diag::warn_sizeof_pointer_type_memaccess) 2545 << FnName << SizeOfArgTy << ArgIdx 2546 << PointeeTy << Dest->getSourceRange() 2547 << LenExpr->getSourceRange()); 2548 break; 2549 } 2550 } 2551 2552 // Always complain about dynamic classes. 2553 if (isDynamicClassType(PointeeTy)) { 2554 2555 unsigned OperationType = 0; 2556 // "overwritten" if we're warning about the destination for any call 2557 // but memcmp; otherwise a verb appropriate to the call. 2558 if (ArgIdx != 0 || BId == Builtin::BImemcmp) { 2559 if (BId == Builtin::BImemcpy) 2560 OperationType = 1; 2561 else if(BId == Builtin::BImemmove) 2562 OperationType = 2; 2563 else if (BId == Builtin::BImemcmp) 2564 OperationType = 3; 2565 } 2566 2567 DiagRuntimeBehavior( 2568 Dest->getExprLoc(), Dest, 2569 PDiag(diag::warn_dyn_class_memaccess) 2570 << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx) 2571 << FnName << PointeeTy 2572 << OperationType 2573 << Call->getCallee()->getSourceRange()); 2574 } else if (PointeeTy.hasNonTrivialObjCLifetime() && 2575 BId != Builtin::BImemset) 2576 DiagRuntimeBehavior( 2577 Dest->getExprLoc(), Dest, 2578 PDiag(diag::warn_arc_object_memaccess) 2579 << ArgIdx << FnName << PointeeTy 2580 << Call->getCallee()->getSourceRange()); 2581 else 2582 continue; 2583 2584 DiagRuntimeBehavior( 2585 Dest->getExprLoc(), Dest, 2586 PDiag(diag::note_bad_memaccess_silence) 2587 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 2588 break; 2589 } 2590 } 2591} 2592 2593// A little helper routine: ignore addition and subtraction of integer literals. 2594// This intentionally does not ignore all integer constant expressions because 2595// we don't want to remove sizeof(). 2596static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 2597 Ex = Ex->IgnoreParenCasts(); 2598 2599 for (;;) { 2600 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 2601 if (!BO || !BO->isAdditiveOp()) 2602 break; 2603 2604 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 2605 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 2606 2607 if (isa<IntegerLiteral>(RHS)) 2608 Ex = LHS; 2609 else if (isa<IntegerLiteral>(LHS)) 2610 Ex = RHS; 2611 else 2612 break; 2613 } 2614 2615 return Ex; 2616} 2617 2618// Warn if the user has made the 'size' argument to strlcpy or strlcat 2619// be the size of the source, instead of the destination. 2620void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 2621 IdentifierInfo *FnName) { 2622 2623 // Don't crash if the user has the wrong number of arguments 2624 if (Call->getNumArgs() != 3) 2625 return; 2626 2627 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 2628 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 2629 const Expr *CompareWithSrc = NULL; 2630 2631 // Look for 'strlcpy(dst, x, sizeof(x))' 2632 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 2633 CompareWithSrc = Ex; 2634 else { 2635 // Look for 'strlcpy(dst, x, strlen(x))' 2636 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 2637 if (SizeCall->isBuiltinCall() == Builtin::BIstrlen 2638 && SizeCall->getNumArgs() == 1) 2639 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 2640 } 2641 } 2642 2643 if (!CompareWithSrc) 2644 return; 2645 2646 // Determine if the argument to sizeof/strlen is equal to the source 2647 // argument. In principle there's all kinds of things you could do 2648 // here, for instance creating an == expression and evaluating it with 2649 // EvaluateAsBooleanCondition, but this uses a more direct technique: 2650 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 2651 if (!SrcArgDRE) 2652 return; 2653 2654 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 2655 if (!CompareWithSrcDRE || 2656 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 2657 return; 2658 2659 const Expr *OriginalSizeArg = Call->getArg(2); 2660 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 2661 << OriginalSizeArg->getSourceRange() << FnName; 2662 2663 // Output a FIXIT hint if the destination is an array (rather than a 2664 // pointer to an array). This could be enhanced to handle some 2665 // pointers if we know the actual size, like if DstArg is 'array+2' 2666 // we could say 'sizeof(array)-2'. 2667 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 2668 QualType DstArgTy = DstArg->getType(); 2669 2670 // Only handle constant-sized or VLAs, but not flexible members. 2671 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 2672 // Only issue the FIXIT for arrays of size > 1. 2673 if (CAT->getSize().getSExtValue() <= 1) 2674 return; 2675 } else if (!DstArgTy->isVariableArrayType()) { 2676 return; 2677 } 2678 2679 SmallString<128> sizeString; 2680 llvm::raw_svector_ostream OS(sizeString); 2681 OS << "sizeof("; 2682 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 2683 OS << ")"; 2684 2685 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 2686 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 2687 OS.str()); 2688} 2689 2690/// Check if two expressions refer to the same declaration. 2691static bool referToTheSameDecl(const Expr *E1, const Expr *E2) { 2692 if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1)) 2693 if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2)) 2694 return D1->getDecl() == D2->getDecl(); 2695 return false; 2696} 2697 2698static const Expr *getStrlenExprArg(const Expr *E) { 2699 if (const CallExpr *CE = dyn_cast<CallExpr>(E)) { 2700 const FunctionDecl *FD = CE->getDirectCallee(); 2701 if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen) 2702 return 0; 2703 return CE->getArg(0)->IgnoreParenCasts(); 2704 } 2705 return 0; 2706} 2707 2708// Warn on anti-patterns as the 'size' argument to strncat. 2709// The correct size argument should look like following: 2710// strncat(dst, src, sizeof(dst) - strlen(dest) - 1); 2711void Sema::CheckStrncatArguments(const CallExpr *CE, 2712 IdentifierInfo *FnName) { 2713 // Don't crash if the user has the wrong number of arguments. 2714 if (CE->getNumArgs() < 3) 2715 return; 2716 const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts(); 2717 const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts(); 2718 const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts(); 2719 2720 // Identify common expressions, which are wrongly used as the size argument 2721 // to strncat and may lead to buffer overflows. 2722 unsigned PatternType = 0; 2723 if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) { 2724 // - sizeof(dst) 2725 if (referToTheSameDecl(SizeOfArg, DstArg)) 2726 PatternType = 1; 2727 // - sizeof(src) 2728 else if (referToTheSameDecl(SizeOfArg, SrcArg)) 2729 PatternType = 2; 2730 } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) { 2731 if (BE->getOpcode() == BO_Sub) { 2732 const Expr *L = BE->getLHS()->IgnoreParenCasts(); 2733 const Expr *R = BE->getRHS()->IgnoreParenCasts(); 2734 // - sizeof(dst) - strlen(dst) 2735 if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) && 2736 referToTheSameDecl(DstArg, getStrlenExprArg(R))) 2737 PatternType = 1; 2738 // - sizeof(src) - (anything) 2739 else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L))) 2740 PatternType = 2; 2741 } 2742 } 2743 2744 if (PatternType == 0) 2745 return; 2746 2747 // Generate the diagnostic. 2748 SourceLocation SL = LenArg->getLocStart(); 2749 SourceRange SR = LenArg->getSourceRange(); 2750 SourceManager &SM = PP.getSourceManager(); 2751 2752 // If the function is defined as a builtin macro, do not show macro expansion. 2753 if (SM.isMacroArgExpansion(SL)) { 2754 SL = SM.getSpellingLoc(SL); 2755 SR = SourceRange(SM.getSpellingLoc(SR.getBegin()), 2756 SM.getSpellingLoc(SR.getEnd())); 2757 } 2758 2759 if (PatternType == 1) 2760 Diag(SL, diag::warn_strncat_large_size) << SR; 2761 else 2762 Diag(SL, diag::warn_strncat_src_size) << SR; 2763 2764 // Output a FIXIT hint if the destination is an array (rather than a 2765 // pointer to an array). This could be enhanced to handle some 2766 // pointers if we know the actual size, like if DstArg is 'array+2' 2767 // we could say 'sizeof(array)-2'. 2768 QualType DstArgTy = DstArg->getType(); 2769 2770 // Only handle constant-sized or VLAs, but not flexible members. 2771 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 2772 // Only issue the FIXIT for arrays of size > 1. 2773 if (CAT->getSize().getSExtValue() <= 1) 2774 return; 2775 } else if (!DstArgTy->isVariableArrayType()) { 2776 return; 2777 } 2778 2779 SmallString<128> sizeString; 2780 llvm::raw_svector_ostream OS(sizeString); 2781 OS << "sizeof("; 2782 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 2783 OS << ") - "; 2784 OS << "strlen("; 2785 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 2786 OS << ") - 1"; 2787 2788 Diag(SL, diag::note_strncat_wrong_size) 2789 << FixItHint::CreateReplacement(SR, OS.str()); 2790} 2791 2792//===--- CHECK: Return Address of Stack Variable --------------------------===// 2793 2794static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars); 2795static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars); 2796 2797/// CheckReturnStackAddr - Check if a return statement returns the address 2798/// of a stack variable. 2799void 2800Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 2801 SourceLocation ReturnLoc) { 2802 2803 Expr *stackE = 0; 2804 SmallVector<DeclRefExpr *, 8> refVars; 2805 2806 // Perform checking for returned stack addresses, local blocks, 2807 // label addresses or references to temporaries. 2808 if (lhsType->isPointerType() || 2809 (!getLangOptions().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 2810 stackE = EvalAddr(RetValExp, refVars); 2811 } else if (lhsType->isReferenceType()) { 2812 stackE = EvalVal(RetValExp, refVars); 2813 } 2814 2815 if (stackE == 0) 2816 return; // Nothing suspicious was found. 2817 2818 SourceLocation diagLoc; 2819 SourceRange diagRange; 2820 if (refVars.empty()) { 2821 diagLoc = stackE->getLocStart(); 2822 diagRange = stackE->getSourceRange(); 2823 } else { 2824 // We followed through a reference variable. 'stackE' contains the 2825 // problematic expression but we will warn at the return statement pointing 2826 // at the reference variable. We will later display the "trail" of 2827 // reference variables using notes. 2828 diagLoc = refVars[0]->getLocStart(); 2829 diagRange = refVars[0]->getSourceRange(); 2830 } 2831 2832 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 2833 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 2834 : diag::warn_ret_stack_addr) 2835 << DR->getDecl()->getDeclName() << diagRange; 2836 } else if (isa<BlockExpr>(stackE)) { // local block. 2837 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 2838 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 2839 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 2840 } else { // local temporary. 2841 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 2842 : diag::warn_ret_local_temp_addr) 2843 << diagRange; 2844 } 2845 2846 // Display the "trail" of reference variables that we followed until we 2847 // found the problematic expression using notes. 2848 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 2849 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 2850 // If this var binds to another reference var, show the range of the next 2851 // var, otherwise the var binds to the problematic expression, in which case 2852 // show the range of the expression. 2853 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 2854 : stackE->getSourceRange(); 2855 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 2856 << VD->getDeclName() << range; 2857 } 2858} 2859 2860/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 2861/// check if the expression in a return statement evaluates to an address 2862/// to a location on the stack, a local block, an address of a label, or a 2863/// reference to local temporary. The recursion is used to traverse the 2864/// AST of the return expression, with recursion backtracking when we 2865/// encounter a subexpression that (1) clearly does not lead to one of the 2866/// above problematic expressions (2) is something we cannot determine leads to 2867/// a problematic expression based on such local checking. 2868/// 2869/// Both EvalAddr and EvalVal follow through reference variables to evaluate 2870/// the expression that they point to. Such variables are added to the 2871/// 'refVars' vector so that we know what the reference variable "trail" was. 2872/// 2873/// EvalAddr processes expressions that are pointers that are used as 2874/// references (and not L-values). EvalVal handles all other values. 2875/// At the base case of the recursion is a check for the above problematic 2876/// expressions. 2877/// 2878/// This implementation handles: 2879/// 2880/// * pointer-to-pointer casts 2881/// * implicit conversions from array references to pointers 2882/// * taking the address of fields 2883/// * arbitrary interplay between "&" and "*" operators 2884/// * pointer arithmetic from an address of a stack variable 2885/// * taking the address of an array element where the array is on the stack 2886static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 2887 if (E->isTypeDependent()) 2888 return NULL; 2889 2890 // We should only be called for evaluating pointer expressions. 2891 assert((E->getType()->isAnyPointerType() || 2892 E->getType()->isBlockPointerType() || 2893 E->getType()->isObjCQualifiedIdType()) && 2894 "EvalAddr only works on pointers"); 2895 2896 E = E->IgnoreParens(); 2897 2898 // Our "symbolic interpreter" is just a dispatch off the currently 2899 // viewed AST node. We then recursively traverse the AST by calling 2900 // EvalAddr and EvalVal appropriately. 2901 switch (E->getStmtClass()) { 2902 case Stmt::DeclRefExprClass: { 2903 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2904 2905 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2906 // If this is a reference variable, follow through to the expression that 2907 // it points to. 2908 if (V->hasLocalStorage() && 2909 V->getType()->isReferenceType() && V->hasInit()) { 2910 // Add the reference variable to the "trail". 2911 refVars.push_back(DR); 2912 return EvalAddr(V->getInit(), refVars); 2913 } 2914 2915 return NULL; 2916 } 2917 2918 case Stmt::UnaryOperatorClass: { 2919 // The only unary operator that make sense to handle here 2920 // is AddrOf. All others don't make sense as pointers. 2921 UnaryOperator *U = cast<UnaryOperator>(E); 2922 2923 if (U->getOpcode() == UO_AddrOf) 2924 return EvalVal(U->getSubExpr(), refVars); 2925 else 2926 return NULL; 2927 } 2928 2929 case Stmt::BinaryOperatorClass: { 2930 // Handle pointer arithmetic. All other binary operators are not valid 2931 // in this context. 2932 BinaryOperator *B = cast<BinaryOperator>(E); 2933 BinaryOperatorKind op = B->getOpcode(); 2934 2935 if (op != BO_Add && op != BO_Sub) 2936 return NULL; 2937 2938 Expr *Base = B->getLHS(); 2939 2940 // Determine which argument is the real pointer base. It could be 2941 // the RHS argument instead of the LHS. 2942 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 2943 2944 assert (Base->getType()->isPointerType()); 2945 return EvalAddr(Base, refVars); 2946 } 2947 2948 // For conditional operators we need to see if either the LHS or RHS are 2949 // valid DeclRefExpr*s. If one of them is valid, we return it. 2950 case Stmt::ConditionalOperatorClass: { 2951 ConditionalOperator *C = cast<ConditionalOperator>(E); 2952 2953 // Handle the GNU extension for missing LHS. 2954 if (Expr *lhsExpr = C->getLHS()) { 2955 // In C++, we can have a throw-expression, which has 'void' type. 2956 if (!lhsExpr->getType()->isVoidType()) 2957 if (Expr* LHS = EvalAddr(lhsExpr, refVars)) 2958 return LHS; 2959 } 2960 2961 // In C++, we can have a throw-expression, which has 'void' type. 2962 if (C->getRHS()->getType()->isVoidType()) 2963 return NULL; 2964 2965 return EvalAddr(C->getRHS(), refVars); 2966 } 2967 2968 case Stmt::BlockExprClass: 2969 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 2970 return E; // local block. 2971 return NULL; 2972 2973 case Stmt::AddrLabelExprClass: 2974 return E; // address of label. 2975 2976 case Stmt::ExprWithCleanupsClass: 2977 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars); 2978 2979 // For casts, we need to handle conversions from arrays to 2980 // pointer values, and pointer-to-pointer conversions. 2981 case Stmt::ImplicitCastExprClass: 2982 case Stmt::CStyleCastExprClass: 2983 case Stmt::CXXFunctionalCastExprClass: 2984 case Stmt::ObjCBridgedCastExprClass: { 2985 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 2986 QualType T = SubExpr->getType(); 2987 2988 if (SubExpr->getType()->isPointerType() || 2989 SubExpr->getType()->isBlockPointerType() || 2990 SubExpr->getType()->isObjCQualifiedIdType()) 2991 return EvalAddr(SubExpr, refVars); 2992 else if (T->isArrayType()) 2993 return EvalVal(SubExpr, refVars); 2994 else 2995 return 0; 2996 } 2997 2998 // C++ casts. For dynamic casts, static casts, and const casts, we 2999 // are always converting from a pointer-to-pointer, so we just blow 3000 // through the cast. In the case the dynamic cast doesn't fail (and 3001 // return NULL), we take the conservative route and report cases 3002 // where we return the address of a stack variable. For Reinterpre 3003 // FIXME: The comment about is wrong; we're not always converting 3004 // from pointer to pointer. I'm guessing that this code should also 3005 // handle references to objects. 3006 case Stmt::CXXStaticCastExprClass: 3007 case Stmt::CXXDynamicCastExprClass: 3008 case Stmt::CXXConstCastExprClass: 3009 case Stmt::CXXReinterpretCastExprClass: { 3010 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 3011 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 3012 return EvalAddr(S, refVars); 3013 else 3014 return NULL; 3015 } 3016 3017 case Stmt::MaterializeTemporaryExprClass: 3018 if (Expr *Result = EvalAddr( 3019 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3020 refVars)) 3021 return Result; 3022 3023 return E; 3024 3025 // Everything else: we simply don't reason about them. 3026 default: 3027 return NULL; 3028 } 3029} 3030 3031 3032/// EvalVal - This function is complements EvalAddr in the mutual recursion. 3033/// See the comments for EvalAddr for more details. 3034static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 3035do { 3036 // We should only be called for evaluating non-pointer expressions, or 3037 // expressions with a pointer type that are not used as references but instead 3038 // are l-values (e.g., DeclRefExpr with a pointer type). 3039 3040 // Our "symbolic interpreter" is just a dispatch off the currently 3041 // viewed AST node. We then recursively traverse the AST by calling 3042 // EvalAddr and EvalVal appropriately. 3043 3044 E = E->IgnoreParens(); 3045 switch (E->getStmtClass()) { 3046 case Stmt::ImplicitCastExprClass: { 3047 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 3048 if (IE->getValueKind() == VK_LValue) { 3049 E = IE->getSubExpr(); 3050 continue; 3051 } 3052 return NULL; 3053 } 3054 3055 case Stmt::ExprWithCleanupsClass: 3056 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars); 3057 3058 case Stmt::DeclRefExprClass: { 3059 // When we hit a DeclRefExpr we are looking at code that refers to a 3060 // variable's name. If it's not a reference variable we check if it has 3061 // local storage within the function, and if so, return the expression. 3062 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3063 3064 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 3065 if (V->hasLocalStorage()) { 3066 if (!V->getType()->isReferenceType()) 3067 return DR; 3068 3069 // Reference variable, follow through to the expression that 3070 // it points to. 3071 if (V->hasInit()) { 3072 // Add the reference variable to the "trail". 3073 refVars.push_back(DR); 3074 return EvalVal(V->getInit(), refVars); 3075 } 3076 } 3077 3078 return NULL; 3079 } 3080 3081 case Stmt::UnaryOperatorClass: { 3082 // The only unary operator that make sense to handle here 3083 // is Deref. All others don't resolve to a "name." This includes 3084 // handling all sorts of rvalues passed to a unary operator. 3085 UnaryOperator *U = cast<UnaryOperator>(E); 3086 3087 if (U->getOpcode() == UO_Deref) 3088 return EvalAddr(U->getSubExpr(), refVars); 3089 3090 return NULL; 3091 } 3092 3093 case Stmt::ArraySubscriptExprClass: { 3094 // Array subscripts are potential references to data on the stack. We 3095 // retrieve the DeclRefExpr* for the array variable if it indeed 3096 // has local storage. 3097 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars); 3098 } 3099 3100 case Stmt::ConditionalOperatorClass: { 3101 // For conditional operators we need to see if either the LHS or RHS are 3102 // non-NULL Expr's. If one is non-NULL, we return it. 3103 ConditionalOperator *C = cast<ConditionalOperator>(E); 3104 3105 // Handle the GNU extension for missing LHS. 3106 if (Expr *lhsExpr = C->getLHS()) 3107 if (Expr *LHS = EvalVal(lhsExpr, refVars)) 3108 return LHS; 3109 3110 return EvalVal(C->getRHS(), refVars); 3111 } 3112 3113 // Accesses to members are potential references to data on the stack. 3114 case Stmt::MemberExprClass: { 3115 MemberExpr *M = cast<MemberExpr>(E); 3116 3117 // Check for indirect access. We only want direct field accesses. 3118 if (M->isArrow()) 3119 return NULL; 3120 3121 // Check whether the member type is itself a reference, in which case 3122 // we're not going to refer to the member, but to what the member refers to. 3123 if (M->getMemberDecl()->getType()->isReferenceType()) 3124 return NULL; 3125 3126 return EvalVal(M->getBase(), refVars); 3127 } 3128 3129 case Stmt::MaterializeTemporaryExprClass: 3130 if (Expr *Result = EvalVal( 3131 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3132 refVars)) 3133 return Result; 3134 3135 return E; 3136 3137 default: 3138 // Check that we don't return or take the address of a reference to a 3139 // temporary. This is only useful in C++. 3140 if (!E->isTypeDependent() && E->isRValue()) 3141 return E; 3142 3143 // Everything else: we simply don't reason about them. 3144 return NULL; 3145 } 3146} while (true); 3147} 3148 3149//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 3150 3151/// Check for comparisons of floating point operands using != and ==. 3152/// Issue a warning if these are no self-comparisons, as they are not likely 3153/// to do what the programmer intended. 3154void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 3155 bool EmitWarning = true; 3156 3157 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 3158 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 3159 3160 // Special case: check for x == x (which is OK). 3161 // Do not emit warnings for such cases. 3162 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 3163 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 3164 if (DRL->getDecl() == DRR->getDecl()) 3165 EmitWarning = false; 3166 3167 3168 // Special case: check for comparisons against literals that can be exactly 3169 // represented by APFloat. In such cases, do not emit a warning. This 3170 // is a heuristic: often comparison against such literals are used to 3171 // detect if a value in a variable has not changed. This clearly can 3172 // lead to false negatives. 3173 if (EmitWarning) { 3174 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 3175 if (FLL->isExact()) 3176 EmitWarning = false; 3177 } else 3178 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 3179 if (FLR->isExact()) 3180 EmitWarning = false; 3181 } 3182 } 3183 3184 // Check for comparisons with builtin types. 3185 if (EmitWarning) 3186 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 3187 if (CL->isBuiltinCall()) 3188 EmitWarning = false; 3189 3190 if (EmitWarning) 3191 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 3192 if (CR->isBuiltinCall()) 3193 EmitWarning = false; 3194 3195 // Emit the diagnostic. 3196 if (EmitWarning) 3197 Diag(Loc, diag::warn_floatingpoint_eq) 3198 << LHS->getSourceRange() << RHS->getSourceRange(); 3199} 3200 3201//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 3202//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 3203 3204namespace { 3205 3206/// Structure recording the 'active' range of an integer-valued 3207/// expression. 3208struct IntRange { 3209 /// The number of bits active in the int. 3210 unsigned Width; 3211 3212 /// True if the int is known not to have negative values. 3213 bool NonNegative; 3214 3215 IntRange(unsigned Width, bool NonNegative) 3216 : Width(Width), NonNegative(NonNegative) 3217 {} 3218 3219 /// Returns the range of the bool type. 3220 static IntRange forBoolType() { 3221 return IntRange(1, true); 3222 } 3223 3224 /// Returns the range of an opaque value of the given integral type. 3225 static IntRange forValueOfType(ASTContext &C, QualType T) { 3226 return forValueOfCanonicalType(C, 3227 T->getCanonicalTypeInternal().getTypePtr()); 3228 } 3229 3230 /// Returns the range of an opaque value of a canonical integral type. 3231 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 3232 assert(T->isCanonicalUnqualified()); 3233 3234 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3235 T = VT->getElementType().getTypePtr(); 3236 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3237 T = CT->getElementType().getTypePtr(); 3238 3239 // For enum types, use the known bit width of the enumerators. 3240 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 3241 EnumDecl *Enum = ET->getDecl(); 3242 if (!Enum->isCompleteDefinition()) 3243 return IntRange(C.getIntWidth(QualType(T, 0)), false); 3244 3245 unsigned NumPositive = Enum->getNumPositiveBits(); 3246 unsigned NumNegative = Enum->getNumNegativeBits(); 3247 3248 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 3249 } 3250 3251 const BuiltinType *BT = cast<BuiltinType>(T); 3252 assert(BT->isInteger()); 3253 3254 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3255 } 3256 3257 /// Returns the "target" range of a canonical integral type, i.e. 3258 /// the range of values expressible in the type. 3259 /// 3260 /// This matches forValueOfCanonicalType except that enums have the 3261 /// full range of their type, not the range of their enumerators. 3262 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 3263 assert(T->isCanonicalUnqualified()); 3264 3265 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3266 T = VT->getElementType().getTypePtr(); 3267 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3268 T = CT->getElementType().getTypePtr(); 3269 if (const EnumType *ET = dyn_cast<EnumType>(T)) 3270 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 3271 3272 const BuiltinType *BT = cast<BuiltinType>(T); 3273 assert(BT->isInteger()); 3274 3275 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3276 } 3277 3278 /// Returns the supremum of two ranges: i.e. their conservative merge. 3279 static IntRange join(IntRange L, IntRange R) { 3280 return IntRange(std::max(L.Width, R.Width), 3281 L.NonNegative && R.NonNegative); 3282 } 3283 3284 /// Returns the infinum of two ranges: i.e. their aggressive merge. 3285 static IntRange meet(IntRange L, IntRange R) { 3286 return IntRange(std::min(L.Width, R.Width), 3287 L.NonNegative || R.NonNegative); 3288 } 3289}; 3290 3291static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, 3292 unsigned MaxWidth) { 3293 if (value.isSigned() && value.isNegative()) 3294 return IntRange(value.getMinSignedBits(), false); 3295 3296 if (value.getBitWidth() > MaxWidth) 3297 value = value.trunc(MaxWidth); 3298 3299 // isNonNegative() just checks the sign bit without considering 3300 // signedness. 3301 return IntRange(value.getActiveBits(), true); 3302} 3303 3304static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 3305 unsigned MaxWidth) { 3306 if (result.isInt()) 3307 return GetValueRange(C, result.getInt(), MaxWidth); 3308 3309 if (result.isVector()) { 3310 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 3311 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 3312 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 3313 R = IntRange::join(R, El); 3314 } 3315 return R; 3316 } 3317 3318 if (result.isComplexInt()) { 3319 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 3320 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 3321 return IntRange::join(R, I); 3322 } 3323 3324 // This can happen with lossless casts to intptr_t of "based" lvalues. 3325 // Assume it might use arbitrary bits. 3326 // FIXME: The only reason we need to pass the type in here is to get 3327 // the sign right on this one case. It would be nice if APValue 3328 // preserved this. 3329 assert(result.isLValue() || result.isAddrLabelDiff()); 3330 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 3331} 3332 3333/// Pseudo-evaluate the given integer expression, estimating the 3334/// range of values it might take. 3335/// 3336/// \param MaxWidth - the width to which the value will be truncated 3337static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 3338 E = E->IgnoreParens(); 3339 3340 // Try a full evaluation first. 3341 Expr::EvalResult result; 3342 if (E->EvaluateAsRValue(result, C)) 3343 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 3344 3345 // I think we only want to look through implicit casts here; if the 3346 // user has an explicit widening cast, we should treat the value as 3347 // being of the new, wider type. 3348 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 3349 if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue) 3350 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 3351 3352 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 3353 3354 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 3355 3356 // Assume that non-integer casts can span the full range of the type. 3357 if (!isIntegerCast) 3358 return OutputTypeRange; 3359 3360 IntRange SubRange 3361 = GetExprRange(C, CE->getSubExpr(), 3362 std::min(MaxWidth, OutputTypeRange.Width)); 3363 3364 // Bail out if the subexpr's range is as wide as the cast type. 3365 if (SubRange.Width >= OutputTypeRange.Width) 3366 return OutputTypeRange; 3367 3368 // Otherwise, we take the smaller width, and we're non-negative if 3369 // either the output type or the subexpr is. 3370 return IntRange(SubRange.Width, 3371 SubRange.NonNegative || OutputTypeRange.NonNegative); 3372 } 3373 3374 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 3375 // If we can fold the condition, just take that operand. 3376 bool CondResult; 3377 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 3378 return GetExprRange(C, CondResult ? CO->getTrueExpr() 3379 : CO->getFalseExpr(), 3380 MaxWidth); 3381 3382 // Otherwise, conservatively merge. 3383 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 3384 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 3385 return IntRange::join(L, R); 3386 } 3387 3388 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 3389 switch (BO->getOpcode()) { 3390 3391 // Boolean-valued operations are single-bit and positive. 3392 case BO_LAnd: 3393 case BO_LOr: 3394 case BO_LT: 3395 case BO_GT: 3396 case BO_LE: 3397 case BO_GE: 3398 case BO_EQ: 3399 case BO_NE: 3400 return IntRange::forBoolType(); 3401 3402 // The type of the assignments is the type of the LHS, so the RHS 3403 // is not necessarily the same type. 3404 case BO_MulAssign: 3405 case BO_DivAssign: 3406 case BO_RemAssign: 3407 case BO_AddAssign: 3408 case BO_SubAssign: 3409 case BO_XorAssign: 3410 case BO_OrAssign: 3411 // TODO: bitfields? 3412 return IntRange::forValueOfType(C, E->getType()); 3413 3414 // Simple assignments just pass through the RHS, which will have 3415 // been coerced to the LHS type. 3416 case BO_Assign: 3417 // TODO: bitfields? 3418 return GetExprRange(C, BO->getRHS(), MaxWidth); 3419 3420 // Operations with opaque sources are black-listed. 3421 case BO_PtrMemD: 3422 case BO_PtrMemI: 3423 return IntRange::forValueOfType(C, E->getType()); 3424 3425 // Bitwise-and uses the *infinum* of the two source ranges. 3426 case BO_And: 3427 case BO_AndAssign: 3428 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 3429 GetExprRange(C, BO->getRHS(), MaxWidth)); 3430 3431 // Left shift gets black-listed based on a judgement call. 3432 case BO_Shl: 3433 // ...except that we want to treat '1 << (blah)' as logically 3434 // positive. It's an important idiom. 3435 if (IntegerLiteral *I 3436 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 3437 if (I->getValue() == 1) { 3438 IntRange R = IntRange::forValueOfType(C, E->getType()); 3439 return IntRange(R.Width, /*NonNegative*/ true); 3440 } 3441 } 3442 // fallthrough 3443 3444 case BO_ShlAssign: 3445 return IntRange::forValueOfType(C, E->getType()); 3446 3447 // Right shift by a constant can narrow its left argument. 3448 case BO_Shr: 3449 case BO_ShrAssign: { 3450 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3451 3452 // If the shift amount is a positive constant, drop the width by 3453 // that much. 3454 llvm::APSInt shift; 3455 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 3456 shift.isNonNegative()) { 3457 unsigned zext = shift.getZExtValue(); 3458 if (zext >= L.Width) 3459 L.Width = (L.NonNegative ? 0 : 1); 3460 else 3461 L.Width -= zext; 3462 } 3463 3464 return L; 3465 } 3466 3467 // Comma acts as its right operand. 3468 case BO_Comma: 3469 return GetExprRange(C, BO->getRHS(), MaxWidth); 3470 3471 // Black-list pointer subtractions. 3472 case BO_Sub: 3473 if (BO->getLHS()->getType()->isPointerType()) 3474 return IntRange::forValueOfType(C, E->getType()); 3475 break; 3476 3477 // The width of a division result is mostly determined by the size 3478 // of the LHS. 3479 case BO_Div: { 3480 // Don't 'pre-truncate' the operands. 3481 unsigned opWidth = C.getIntWidth(E->getType()); 3482 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3483 3484 // If the divisor is constant, use that. 3485 llvm::APSInt divisor; 3486 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 3487 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 3488 if (log2 >= L.Width) 3489 L.Width = (L.NonNegative ? 0 : 1); 3490 else 3491 L.Width = std::min(L.Width - log2, MaxWidth); 3492 return L; 3493 } 3494 3495 // Otherwise, just use the LHS's width. 3496 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3497 return IntRange(L.Width, L.NonNegative && R.NonNegative); 3498 } 3499 3500 // The result of a remainder can't be larger than the result of 3501 // either side. 3502 case BO_Rem: { 3503 // Don't 'pre-truncate' the operands. 3504 unsigned opWidth = C.getIntWidth(E->getType()); 3505 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3506 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3507 3508 IntRange meet = IntRange::meet(L, R); 3509 meet.Width = std::min(meet.Width, MaxWidth); 3510 return meet; 3511 } 3512 3513 // The default behavior is okay for these. 3514 case BO_Mul: 3515 case BO_Add: 3516 case BO_Xor: 3517 case BO_Or: 3518 break; 3519 } 3520 3521 // The default case is to treat the operation as if it were closed 3522 // on the narrowest type that encompasses both operands. 3523 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3524 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 3525 return IntRange::join(L, R); 3526 } 3527 3528 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 3529 switch (UO->getOpcode()) { 3530 // Boolean-valued operations are white-listed. 3531 case UO_LNot: 3532 return IntRange::forBoolType(); 3533 3534 // Operations with opaque sources are black-listed. 3535 case UO_Deref: 3536 case UO_AddrOf: // should be impossible 3537 return IntRange::forValueOfType(C, E->getType()); 3538 3539 default: 3540 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 3541 } 3542 } 3543 3544 if (dyn_cast<OffsetOfExpr>(E)) { 3545 IntRange::forValueOfType(C, E->getType()); 3546 } 3547 3548 if (FieldDecl *BitField = E->getBitField()) 3549 return IntRange(BitField->getBitWidthValue(C), 3550 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 3551 3552 return IntRange::forValueOfType(C, E->getType()); 3553} 3554 3555static IntRange GetExprRange(ASTContext &C, Expr *E) { 3556 return GetExprRange(C, E, C.getIntWidth(E->getType())); 3557} 3558 3559/// Checks whether the given value, which currently has the given 3560/// source semantics, has the same value when coerced through the 3561/// target semantics. 3562static bool IsSameFloatAfterCast(const llvm::APFloat &value, 3563 const llvm::fltSemantics &Src, 3564 const llvm::fltSemantics &Tgt) { 3565 llvm::APFloat truncated = value; 3566 3567 bool ignored; 3568 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 3569 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 3570 3571 return truncated.bitwiseIsEqual(value); 3572} 3573 3574/// Checks whether the given value, which currently has the given 3575/// source semantics, has the same value when coerced through the 3576/// target semantics. 3577/// 3578/// The value might be a vector of floats (or a complex number). 3579static bool IsSameFloatAfterCast(const APValue &value, 3580 const llvm::fltSemantics &Src, 3581 const llvm::fltSemantics &Tgt) { 3582 if (value.isFloat()) 3583 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 3584 3585 if (value.isVector()) { 3586 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 3587 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 3588 return false; 3589 return true; 3590 } 3591 3592 assert(value.isComplexFloat()); 3593 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 3594 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 3595} 3596 3597static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 3598 3599static bool IsZero(Sema &S, Expr *E) { 3600 // Suppress cases where we are comparing against an enum constant. 3601 if (const DeclRefExpr *DR = 3602 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 3603 if (isa<EnumConstantDecl>(DR->getDecl())) 3604 return false; 3605 3606 // Suppress cases where the '0' value is expanded from a macro. 3607 if (E->getLocStart().isMacroID()) 3608 return false; 3609 3610 llvm::APSInt Value; 3611 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 3612} 3613 3614static bool HasEnumType(Expr *E) { 3615 // Strip off implicit integral promotions. 3616 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 3617 if (ICE->getCastKind() != CK_IntegralCast && 3618 ICE->getCastKind() != CK_NoOp) 3619 break; 3620 E = ICE->getSubExpr(); 3621 } 3622 3623 return E->getType()->isEnumeralType(); 3624} 3625 3626static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 3627 BinaryOperatorKind op = E->getOpcode(); 3628 if (E->isValueDependent()) 3629 return; 3630 3631 if (op == BO_LT && IsZero(S, E->getRHS())) { 3632 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3633 << "< 0" << "false" << HasEnumType(E->getLHS()) 3634 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3635 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 3636 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3637 << ">= 0" << "true" << HasEnumType(E->getLHS()) 3638 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3639 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 3640 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3641 << "0 >" << "false" << HasEnumType(E->getRHS()) 3642 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3643 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 3644 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3645 << "0 <=" << "true" << HasEnumType(E->getRHS()) 3646 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3647 } 3648} 3649 3650/// Analyze the operands of the given comparison. Implements the 3651/// fallback case from AnalyzeComparison. 3652static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 3653 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3654 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3655} 3656 3657/// \brief Implements -Wsign-compare. 3658/// 3659/// \param E the binary operator to check for warnings 3660static void AnalyzeComparison(Sema &S, BinaryOperator *E) { 3661 // The type the comparison is being performed in. 3662 QualType T = E->getLHS()->getType(); 3663 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 3664 && "comparison with mismatched types"); 3665 3666 // We don't do anything special if this isn't an unsigned integral 3667 // comparison: we're only interested in integral comparisons, and 3668 // signed comparisons only happen in cases we don't care to warn about. 3669 // 3670 // We also don't care about value-dependent expressions or expressions 3671 // whose result is a constant. 3672 if (!T->hasUnsignedIntegerRepresentation() 3673 || E->isValueDependent() || E->isIntegerConstantExpr(S.Context)) 3674 return AnalyzeImpConvsInComparison(S, E); 3675 3676 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 3677 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 3678 3679 // Check to see if one of the (unmodified) operands is of different 3680 // signedness. 3681 Expr *signedOperand, *unsignedOperand; 3682 if (LHS->getType()->hasSignedIntegerRepresentation()) { 3683 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 3684 "unsigned comparison between two signed integer expressions?"); 3685 signedOperand = LHS; 3686 unsignedOperand = RHS; 3687 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 3688 signedOperand = RHS; 3689 unsignedOperand = LHS; 3690 } else { 3691 CheckTrivialUnsignedComparison(S, E); 3692 return AnalyzeImpConvsInComparison(S, E); 3693 } 3694 3695 // Otherwise, calculate the effective range of the signed operand. 3696 IntRange signedRange = GetExprRange(S.Context, signedOperand); 3697 3698 // Go ahead and analyze implicit conversions in the operands. Note 3699 // that we skip the implicit conversions on both sides. 3700 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 3701 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 3702 3703 // If the signed range is non-negative, -Wsign-compare won't fire, 3704 // but we should still check for comparisons which are always true 3705 // or false. 3706 if (signedRange.NonNegative) 3707 return CheckTrivialUnsignedComparison(S, E); 3708 3709 // For (in)equality comparisons, if the unsigned operand is a 3710 // constant which cannot collide with a overflowed signed operand, 3711 // then reinterpreting the signed operand as unsigned will not 3712 // change the result of the comparison. 3713 if (E->isEqualityOp()) { 3714 unsigned comparisonWidth = S.Context.getIntWidth(T); 3715 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 3716 3717 // We should never be unable to prove that the unsigned operand is 3718 // non-negative. 3719 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 3720 3721 if (unsignedRange.Width < comparisonWidth) 3722 return; 3723 } 3724 3725 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison) 3726 << LHS->getType() << RHS->getType() 3727 << LHS->getSourceRange() << RHS->getSourceRange(); 3728} 3729 3730/// Analyzes an attempt to assign the given value to a bitfield. 3731/// 3732/// Returns true if there was something fishy about the attempt. 3733static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 3734 SourceLocation InitLoc) { 3735 assert(Bitfield->isBitField()); 3736 if (Bitfield->isInvalidDecl()) 3737 return false; 3738 3739 // White-list bool bitfields. 3740 if (Bitfield->getType()->isBooleanType()) 3741 return false; 3742 3743 // Ignore value- or type-dependent expressions. 3744 if (Bitfield->getBitWidth()->isValueDependent() || 3745 Bitfield->getBitWidth()->isTypeDependent() || 3746 Init->isValueDependent() || 3747 Init->isTypeDependent()) 3748 return false; 3749 3750 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 3751 3752 llvm::APSInt Value; 3753 if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects)) 3754 return false; 3755 3756 unsigned OriginalWidth = Value.getBitWidth(); 3757 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 3758 3759 if (OriginalWidth <= FieldWidth) 3760 return false; 3761 3762 // Compute the value which the bitfield will contain. 3763 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 3764 TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType()); 3765 3766 // Check whether the stored value is equal to the original value. 3767 TruncatedValue = TruncatedValue.extend(OriginalWidth); 3768 if (Value == TruncatedValue) 3769 return false; 3770 3771 // Special-case bitfields of width 1: booleans are naturally 0/1, and 3772 // therefore don't strictly fit into a signed bitfield of width 1. 3773 if (FieldWidth == 1 && Value == 1) 3774 return false; 3775 3776 std::string PrettyValue = Value.toString(10); 3777 std::string PrettyTrunc = TruncatedValue.toString(10); 3778 3779 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 3780 << PrettyValue << PrettyTrunc << OriginalInit->getType() 3781 << Init->getSourceRange(); 3782 3783 return true; 3784} 3785 3786/// Analyze the given simple or compound assignment for warning-worthy 3787/// operations. 3788static void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 3789 // Just recurse on the LHS. 3790 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3791 3792 // We want to recurse on the RHS as normal unless we're assigning to 3793 // a bitfield. 3794 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 3795 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 3796 E->getOperatorLoc())) { 3797 // Recurse, ignoring any implicit conversions on the RHS. 3798 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 3799 E->getOperatorLoc()); 3800 } 3801 } 3802 3803 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3804} 3805 3806/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3807static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 3808 SourceLocation CContext, unsigned diag, 3809 bool pruneControlFlow = false) { 3810 if (pruneControlFlow) { 3811 S.DiagRuntimeBehavior(E->getExprLoc(), E, 3812 S.PDiag(diag) 3813 << SourceType << T << E->getSourceRange() 3814 << SourceRange(CContext)); 3815 return; 3816 } 3817 S.Diag(E->getExprLoc(), diag) 3818 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 3819} 3820 3821/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3822static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, 3823 SourceLocation CContext, unsigned diag, 3824 bool pruneControlFlow = false) { 3825 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow); 3826} 3827 3828/// Diagnose an implicit cast from a literal expression. Does not warn when the 3829/// cast wouldn't lose information. 3830void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 3831 SourceLocation CContext) { 3832 // Try to convert the literal exactly to an integer. If we can, don't warn. 3833 bool isExact = false; 3834 const llvm::APFloat &Value = FL->getValue(); 3835 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 3836 T->hasUnsignedIntegerRepresentation()); 3837 if (Value.convertToInteger(IntegerValue, 3838 llvm::APFloat::rmTowardZero, &isExact) 3839 == llvm::APFloat::opOK && isExact) 3840 return; 3841 3842 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 3843 << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext); 3844} 3845 3846std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 3847 if (!Range.Width) return "0"; 3848 3849 llvm::APSInt ValueInRange = Value; 3850 ValueInRange.setIsSigned(!Range.NonNegative); 3851 ValueInRange = ValueInRange.trunc(Range.Width); 3852 return ValueInRange.toString(10); 3853} 3854 3855void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 3856 SourceLocation CC, bool *ICContext = 0) { 3857 if (E->isTypeDependent() || E->isValueDependent()) return; 3858 3859 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 3860 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 3861 if (Source == Target) return; 3862 if (Target->isDependentType()) return; 3863 3864 // If the conversion context location is invalid don't complain. We also 3865 // don't want to emit a warning if the issue occurs from the expansion of 3866 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 3867 // delay this check as long as possible. Once we detect we are in that 3868 // scenario, we just return. 3869 if (CC.isInvalid()) 3870 return; 3871 3872 // Diagnose implicit casts to bool. 3873 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 3874 if (isa<StringLiteral>(E)) 3875 // Warn on string literal to bool. Checks for string literals in logical 3876 // expressions, for instances, assert(0 && "error here"), is prevented 3877 // by a check in AnalyzeImplicitConversions(). 3878 return DiagnoseImpCast(S, E, T, CC, 3879 diag::warn_impcast_string_literal_to_bool); 3880 if (Source->isFunctionType()) { 3881 // Warn on function to bool. Checks free functions and static member 3882 // functions. Weakly imported functions are excluded from the check, 3883 // since it's common to test their value to check whether the linker 3884 // found a definition for them. 3885 ValueDecl *D = 0; 3886 if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) { 3887 D = R->getDecl(); 3888 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 3889 D = M->getMemberDecl(); 3890 } 3891 3892 if (D && !D->isWeak()) { 3893 if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) { 3894 S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool) 3895 << F << E->getSourceRange() << SourceRange(CC); 3896 S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence) 3897 << FixItHint::CreateInsertion(E->getExprLoc(), "&"); 3898 QualType ReturnType; 3899 UnresolvedSet<4> NonTemplateOverloads; 3900 S.isExprCallable(*E, ReturnType, NonTemplateOverloads); 3901 if (!ReturnType.isNull() 3902 && ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 3903 S.Diag(E->getExprLoc(), diag::note_function_to_bool_call) 3904 << FixItHint::CreateInsertion( 3905 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()"); 3906 return; 3907 } 3908 } 3909 } 3910 return; // Other casts to bool are not checked. 3911 } 3912 3913 // Strip vector types. 3914 if (isa<VectorType>(Source)) { 3915 if (!isa<VectorType>(Target)) { 3916 if (S.SourceMgr.isInSystemMacro(CC)) 3917 return; 3918 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 3919 } 3920 3921 // If the vector cast is cast between two vectors of the same size, it is 3922 // a bitcast, not a conversion. 3923 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 3924 return; 3925 3926 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 3927 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 3928 } 3929 3930 // Strip complex types. 3931 if (isa<ComplexType>(Source)) { 3932 if (!isa<ComplexType>(Target)) { 3933 if (S.SourceMgr.isInSystemMacro(CC)) 3934 return; 3935 3936 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 3937 } 3938 3939 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 3940 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 3941 } 3942 3943 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 3944 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 3945 3946 // If the source is floating point... 3947 if (SourceBT && SourceBT->isFloatingPoint()) { 3948 // ...and the target is floating point... 3949 if (TargetBT && TargetBT->isFloatingPoint()) { 3950 // ...then warn if we're dropping FP rank. 3951 3952 // Builtin FP kinds are ordered by increasing FP rank. 3953 if (SourceBT->getKind() > TargetBT->getKind()) { 3954 // Don't warn about float constants that are precisely 3955 // representable in the target type. 3956 Expr::EvalResult result; 3957 if (E->EvaluateAsRValue(result, S.Context)) { 3958 // Value might be a float, a float vector, or a float complex. 3959 if (IsSameFloatAfterCast(result.Val, 3960 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 3961 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 3962 return; 3963 } 3964 3965 if (S.SourceMgr.isInSystemMacro(CC)) 3966 return; 3967 3968 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 3969 } 3970 return; 3971 } 3972 3973 // If the target is integral, always warn. 3974 if ((TargetBT && TargetBT->isInteger())) { 3975 if (S.SourceMgr.isInSystemMacro(CC)) 3976 return; 3977 3978 Expr *InnerE = E->IgnoreParenImpCasts(); 3979 // We also want to warn on, e.g., "int i = -1.234" 3980 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 3981 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 3982 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 3983 3984 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 3985 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 3986 } else { 3987 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 3988 } 3989 } 3990 3991 return; 3992 } 3993 3994 if (!Source->isIntegerType() || !Target->isIntegerType()) 3995 return; 3996 3997 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 3998 == Expr::NPCK_GNUNull) && Target->isIntegerType()) { 3999 S.Diag(E->getExprLoc(), diag::warn_impcast_null_pointer_to_integer) 4000 << E->getSourceRange() << clang::SourceRange(CC); 4001 return; 4002 } 4003 4004 IntRange SourceRange = GetExprRange(S.Context, E); 4005 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 4006 4007 if (SourceRange.Width > TargetRange.Width) { 4008 // If the source is a constant, use a default-on diagnostic. 4009 // TODO: this should happen for bitfield stores, too. 4010 llvm::APSInt Value(32); 4011 if (E->isIntegerConstantExpr(Value, S.Context)) { 4012 if (S.SourceMgr.isInSystemMacro(CC)) 4013 return; 4014 4015 std::string PrettySourceValue = Value.toString(10); 4016 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 4017 4018 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4019 S.PDiag(diag::warn_impcast_integer_precision_constant) 4020 << PrettySourceValue << PrettyTargetValue 4021 << E->getType() << T << E->getSourceRange() 4022 << clang::SourceRange(CC)); 4023 return; 4024 } 4025 4026 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 4027 if (S.SourceMgr.isInSystemMacro(CC)) 4028 return; 4029 4030 if (SourceRange.Width == 64 && TargetRange.Width == 32) 4031 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32, 4032 /* pruneControlFlow */ true); 4033 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 4034 } 4035 4036 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 4037 (!TargetRange.NonNegative && SourceRange.NonNegative && 4038 SourceRange.Width == TargetRange.Width)) { 4039 4040 if (S.SourceMgr.isInSystemMacro(CC)) 4041 return; 4042 4043 unsigned DiagID = diag::warn_impcast_integer_sign; 4044 4045 // Traditionally, gcc has warned about this under -Wsign-compare. 4046 // We also want to warn about it in -Wconversion. 4047 // So if -Wconversion is off, use a completely identical diagnostic 4048 // in the sign-compare group. 4049 // The conditional-checking code will 4050 if (ICContext) { 4051 DiagID = diag::warn_impcast_integer_sign_conditional; 4052 *ICContext = true; 4053 } 4054 4055 return DiagnoseImpCast(S, E, T, CC, DiagID); 4056 } 4057 4058 // Diagnose conversions between different enumeration types. 4059 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 4060 // type, to give us better diagnostics. 4061 QualType SourceType = E->getType(); 4062 if (!S.getLangOptions().CPlusPlus) { 4063 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 4064 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 4065 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 4066 SourceType = S.Context.getTypeDeclType(Enum); 4067 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 4068 } 4069 } 4070 4071 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 4072 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 4073 if ((SourceEnum->getDecl()->getIdentifier() || 4074 SourceEnum->getDecl()->getTypedefNameForAnonDecl()) && 4075 (TargetEnum->getDecl()->getIdentifier() || 4076 TargetEnum->getDecl()->getTypedefNameForAnonDecl()) && 4077 SourceEnum != TargetEnum) { 4078 if (S.SourceMgr.isInSystemMacro(CC)) 4079 return; 4080 4081 return DiagnoseImpCast(S, E, SourceType, T, CC, 4082 diag::warn_impcast_different_enum_types); 4083 } 4084 4085 return; 4086} 4087 4088void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 4089 4090void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 4091 SourceLocation CC, bool &ICContext) { 4092 E = E->IgnoreParenImpCasts(); 4093 4094 if (isa<ConditionalOperator>(E)) 4095 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 4096 4097 AnalyzeImplicitConversions(S, E, CC); 4098 if (E->getType() != T) 4099 return CheckImplicitConversion(S, E, T, CC, &ICContext); 4100 return; 4101} 4102 4103void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 4104 SourceLocation CC = E->getQuestionLoc(); 4105 4106 AnalyzeImplicitConversions(S, E->getCond(), CC); 4107 4108 bool Suspicious = false; 4109 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 4110 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 4111 4112 // If -Wconversion would have warned about either of the candidates 4113 // for a signedness conversion to the context type... 4114 if (!Suspicious) return; 4115 4116 // ...but it's currently ignored... 4117 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 4118 CC)) 4119 return; 4120 4121 // ...then check whether it would have warned about either of the 4122 // candidates for a signedness conversion to the condition type. 4123 if (E->getType() == T) return; 4124 4125 Suspicious = false; 4126 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 4127 E->getType(), CC, &Suspicious); 4128 if (!Suspicious) 4129 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 4130 E->getType(), CC, &Suspicious); 4131} 4132 4133/// AnalyzeImplicitConversions - Find and report any interesting 4134/// implicit conversions in the given expression. There are a couple 4135/// of competing diagnostics here, -Wconversion and -Wsign-compare. 4136void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 4137 QualType T = OrigE->getType(); 4138 Expr *E = OrigE->IgnoreParenImpCasts(); 4139 4140 if (E->isTypeDependent() || E->isValueDependent()) 4141 return; 4142 4143 // For conditional operators, we analyze the arguments as if they 4144 // were being fed directly into the output. 4145 if (isa<ConditionalOperator>(E)) { 4146 ConditionalOperator *CO = cast<ConditionalOperator>(E); 4147 CheckConditionalOperator(S, CO, T); 4148 return; 4149 } 4150 4151 // Go ahead and check any implicit conversions we might have skipped. 4152 // The non-canonical typecheck is just an optimization; 4153 // CheckImplicitConversion will filter out dead implicit conversions. 4154 if (E->getType() != T) 4155 CheckImplicitConversion(S, E, T, CC); 4156 4157 // Now continue drilling into this expression. 4158 4159 // Skip past explicit casts. 4160 if (isa<ExplicitCastExpr>(E)) { 4161 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 4162 return AnalyzeImplicitConversions(S, E, CC); 4163 } 4164 4165 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 4166 // Do a somewhat different check with comparison operators. 4167 if (BO->isComparisonOp()) 4168 return AnalyzeComparison(S, BO); 4169 4170 // And with simple assignments. 4171 if (BO->getOpcode() == BO_Assign) 4172 return AnalyzeAssignment(S, BO); 4173 } 4174 4175 // These break the otherwise-useful invariant below. Fortunately, 4176 // we don't really need to recurse into them, because any internal 4177 // expressions should have been analyzed already when they were 4178 // built into statements. 4179 if (isa<StmtExpr>(E)) return; 4180 4181 // Don't descend into unevaluated contexts. 4182 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 4183 4184 // Now just recurse over the expression's children. 4185 CC = E->getExprLoc(); 4186 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 4187 bool IsLogicalOperator = BO && BO->isLogicalOp(); 4188 for (Stmt::child_range I = E->children(); I; ++I) { 4189 Expr *ChildExpr = dyn_cast_or_null<Expr>(*I); 4190 if (!ChildExpr) 4191 continue; 4192 4193 if (IsLogicalOperator && 4194 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 4195 // Ignore checking string literals that are in logical operators. 4196 continue; 4197 AnalyzeImplicitConversions(S, ChildExpr, CC); 4198 } 4199} 4200 4201} // end anonymous namespace 4202 4203/// Diagnoses "dangerous" implicit conversions within the given 4204/// expression (which is a full expression). Implements -Wconversion 4205/// and -Wsign-compare. 4206/// 4207/// \param CC the "context" location of the implicit conversion, i.e. 4208/// the most location of the syntactic entity requiring the implicit 4209/// conversion 4210void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 4211 // Don't diagnose in unevaluated contexts. 4212 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 4213 return; 4214 4215 // Don't diagnose for value- or type-dependent expressions. 4216 if (E->isTypeDependent() || E->isValueDependent()) 4217 return; 4218 4219 // Check for array bounds violations in cases where the check isn't triggered 4220 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 4221 // ArraySubscriptExpr is on the RHS of a variable initialization. 4222 CheckArrayAccess(E); 4223 4224 // This is not the right CC for (e.g.) a variable initialization. 4225 AnalyzeImplicitConversions(*this, E, CC); 4226} 4227 4228void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 4229 FieldDecl *BitField, 4230 Expr *Init) { 4231 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 4232} 4233 4234/// CheckParmsForFunctionDef - Check that the parameters of the given 4235/// function are appropriate for the definition of a function. This 4236/// takes care of any checks that cannot be performed on the 4237/// declaration itself, e.g., that the types of each of the function 4238/// parameters are complete. 4239bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 4240 bool CheckParameterNames) { 4241 bool HasInvalidParm = false; 4242 for (; P != PEnd; ++P) { 4243 ParmVarDecl *Param = *P; 4244 4245 // C99 6.7.5.3p4: the parameters in a parameter type list in a 4246 // function declarator that is part of a function definition of 4247 // that function shall not have incomplete type. 4248 // 4249 // This is also C++ [dcl.fct]p6. 4250 if (!Param->isInvalidDecl() && 4251 RequireCompleteType(Param->getLocation(), Param->getType(), 4252 diag::err_typecheck_decl_incomplete_type)) { 4253 Param->setInvalidDecl(); 4254 HasInvalidParm = true; 4255 } 4256 4257 // C99 6.9.1p5: If the declarator includes a parameter type list, the 4258 // declaration of each parameter shall include an identifier. 4259 if (CheckParameterNames && 4260 Param->getIdentifier() == 0 && 4261 !Param->isImplicit() && 4262 !getLangOptions().CPlusPlus) 4263 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 4264 4265 // C99 6.7.5.3p12: 4266 // If the function declarator is not part of a definition of that 4267 // function, parameters may have incomplete type and may use the [*] 4268 // notation in their sequences of declarator specifiers to specify 4269 // variable length array types. 4270 QualType PType = Param->getOriginalType(); 4271 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 4272 if (AT->getSizeModifier() == ArrayType::Star) { 4273 // FIXME: This diagnosic should point the the '[*]' if source-location 4274 // information is added for it. 4275 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 4276 } 4277 } 4278 } 4279 4280 return HasInvalidParm; 4281} 4282 4283/// CheckCastAlign - Implements -Wcast-align, which warns when a 4284/// pointer cast increases the alignment requirements. 4285void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 4286 // This is actually a lot of work to potentially be doing on every 4287 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 4288 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 4289 TRange.getBegin()) 4290 == DiagnosticsEngine::Ignored) 4291 return; 4292 4293 // Ignore dependent types. 4294 if (T->isDependentType() || Op->getType()->isDependentType()) 4295 return; 4296 4297 // Require that the destination be a pointer type. 4298 const PointerType *DestPtr = T->getAs<PointerType>(); 4299 if (!DestPtr) return; 4300 4301 // If the destination has alignment 1, we're done. 4302 QualType DestPointee = DestPtr->getPointeeType(); 4303 if (DestPointee->isIncompleteType()) return; 4304 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 4305 if (DestAlign.isOne()) return; 4306 4307 // Require that the source be a pointer type. 4308 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 4309 if (!SrcPtr) return; 4310 QualType SrcPointee = SrcPtr->getPointeeType(); 4311 4312 // Whitelist casts from cv void*. We already implicitly 4313 // whitelisted casts to cv void*, since they have alignment 1. 4314 // Also whitelist casts involving incomplete types, which implicitly 4315 // includes 'void'. 4316 if (SrcPointee->isIncompleteType()) return; 4317 4318 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 4319 if (SrcAlign >= DestAlign) return; 4320 4321 Diag(TRange.getBegin(), diag::warn_cast_align) 4322 << Op->getType() << T 4323 << static_cast<unsigned>(SrcAlign.getQuantity()) 4324 << static_cast<unsigned>(DestAlign.getQuantity()) 4325 << TRange << Op->getSourceRange(); 4326} 4327 4328static const Type* getElementType(const Expr *BaseExpr) { 4329 const Type* EltType = BaseExpr->getType().getTypePtr(); 4330 if (EltType->isAnyPointerType()) 4331 return EltType->getPointeeType().getTypePtr(); 4332 else if (EltType->isArrayType()) 4333 return EltType->getBaseElementTypeUnsafe(); 4334 return EltType; 4335} 4336 4337/// \brief Check whether this array fits the idiom of a size-one tail padded 4338/// array member of a struct. 4339/// 4340/// We avoid emitting out-of-bounds access warnings for such arrays as they are 4341/// commonly used to emulate flexible arrays in C89 code. 4342static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 4343 const NamedDecl *ND) { 4344 if (Size != 1 || !ND) return false; 4345 4346 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 4347 if (!FD) return false; 4348 4349 // Don't consider sizes resulting from macro expansions or template argument 4350 // substitution to form C89 tail-padded arrays. 4351 ConstantArrayTypeLoc TL = 4352 cast<ConstantArrayTypeLoc>(FD->getTypeSourceInfo()->getTypeLoc()); 4353 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(TL.getSizeExpr()); 4354 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 4355 return false; 4356 4357 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 4358 if (!RD) return false; 4359 if (RD->isUnion()) return false; 4360 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 4361 if (!CRD->isStandardLayout()) return false; 4362 } 4363 4364 // See if this is the last field decl in the record. 4365 const Decl *D = FD; 4366 while ((D = D->getNextDeclInContext())) 4367 if (isa<FieldDecl>(D)) 4368 return false; 4369 return true; 4370} 4371 4372void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 4373 const ArraySubscriptExpr *ASE, 4374 bool AllowOnePastEnd, bool IndexNegated) { 4375 IndexExpr = IndexExpr->IgnoreParenCasts(); 4376 if (IndexExpr->isValueDependent()) 4377 return; 4378 4379 const Type *EffectiveType = getElementType(BaseExpr); 4380 BaseExpr = BaseExpr->IgnoreParenCasts(); 4381 const ConstantArrayType *ArrayTy = 4382 Context.getAsConstantArrayType(BaseExpr->getType()); 4383 if (!ArrayTy) 4384 return; 4385 4386 llvm::APSInt index; 4387 if (!IndexExpr->EvaluateAsInt(index, Context)) 4388 return; 4389 if (IndexNegated) 4390 index = -index; 4391 4392 const NamedDecl *ND = NULL; 4393 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4394 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 4395 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 4396 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 4397 4398 if (index.isUnsigned() || !index.isNegative()) { 4399 llvm::APInt size = ArrayTy->getSize(); 4400 if (!size.isStrictlyPositive()) 4401 return; 4402 4403 const Type* BaseType = getElementType(BaseExpr); 4404 if (BaseType != EffectiveType) { 4405 // Make sure we're comparing apples to apples when comparing index to size 4406 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 4407 uint64_t array_typesize = Context.getTypeSize(BaseType); 4408 // Handle ptrarith_typesize being zero, such as when casting to void* 4409 if (!ptrarith_typesize) ptrarith_typesize = 1; 4410 if (ptrarith_typesize != array_typesize) { 4411 // There's a cast to a different size type involved 4412 uint64_t ratio = array_typesize / ptrarith_typesize; 4413 // TODO: Be smarter about handling cases where array_typesize is not a 4414 // multiple of ptrarith_typesize 4415 if (ptrarith_typesize * ratio == array_typesize) 4416 size *= llvm::APInt(size.getBitWidth(), ratio); 4417 } 4418 } 4419 4420 if (size.getBitWidth() > index.getBitWidth()) 4421 index = index.sext(size.getBitWidth()); 4422 else if (size.getBitWidth() < index.getBitWidth()) 4423 size = size.sext(index.getBitWidth()); 4424 4425 // For array subscripting the index must be less than size, but for pointer 4426 // arithmetic also allow the index (offset) to be equal to size since 4427 // computing the next address after the end of the array is legal and 4428 // commonly done e.g. in C++ iterators and range-based for loops. 4429 if (AllowOnePastEnd ? index.sle(size) : index.slt(size)) 4430 return; 4431 4432 // Also don't warn for arrays of size 1 which are members of some 4433 // structure. These are often used to approximate flexible arrays in C89 4434 // code. 4435 if (IsTailPaddedMemberArray(*this, size, ND)) 4436 return; 4437 4438 // Suppress the warning if the subscript expression (as identified by the 4439 // ']' location) and the index expression are both from macro expansions 4440 // within a system header. 4441 if (ASE) { 4442 SourceLocation RBracketLoc = SourceMgr.getSpellingLoc( 4443 ASE->getRBracketLoc()); 4444 if (SourceMgr.isInSystemHeader(RBracketLoc)) { 4445 SourceLocation IndexLoc = SourceMgr.getSpellingLoc( 4446 IndexExpr->getLocStart()); 4447 if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc)) 4448 return; 4449 } 4450 } 4451 4452 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 4453 if (ASE) 4454 DiagID = diag::warn_array_index_exceeds_bounds; 4455 4456 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4457 PDiag(DiagID) << index.toString(10, true) 4458 << size.toString(10, true) 4459 << (unsigned)size.getLimitedValue(~0U) 4460 << IndexExpr->getSourceRange()); 4461 } else { 4462 unsigned DiagID = diag::warn_array_index_precedes_bounds; 4463 if (!ASE) { 4464 DiagID = diag::warn_ptr_arith_precedes_bounds; 4465 if (index.isNegative()) index = -index; 4466 } 4467 4468 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4469 PDiag(DiagID) << index.toString(10, true) 4470 << IndexExpr->getSourceRange()); 4471 } 4472 4473 if (!ND) { 4474 // Try harder to find a NamedDecl to point at in the note. 4475 while (const ArraySubscriptExpr *ASE = 4476 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 4477 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 4478 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4479 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 4480 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 4481 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 4482 } 4483 4484 if (ND) 4485 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 4486 PDiag(diag::note_array_index_out_of_bounds) 4487 << ND->getDeclName()); 4488} 4489 4490void Sema::CheckArrayAccess(const Expr *expr) { 4491 int AllowOnePastEnd = 0; 4492 while (expr) { 4493 expr = expr->IgnoreParenImpCasts(); 4494 switch (expr->getStmtClass()) { 4495 case Stmt::ArraySubscriptExprClass: { 4496 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 4497 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE, 4498 AllowOnePastEnd > 0); 4499 return; 4500 } 4501 case Stmt::UnaryOperatorClass: { 4502 // Only unwrap the * and & unary operators 4503 const UnaryOperator *UO = cast<UnaryOperator>(expr); 4504 expr = UO->getSubExpr(); 4505 switch (UO->getOpcode()) { 4506 case UO_AddrOf: 4507 AllowOnePastEnd++; 4508 break; 4509 case UO_Deref: 4510 AllowOnePastEnd--; 4511 break; 4512 default: 4513 return; 4514 } 4515 break; 4516 } 4517 case Stmt::ConditionalOperatorClass: { 4518 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 4519 if (const Expr *lhs = cond->getLHS()) 4520 CheckArrayAccess(lhs); 4521 if (const Expr *rhs = cond->getRHS()) 4522 CheckArrayAccess(rhs); 4523 return; 4524 } 4525 default: 4526 return; 4527 } 4528 } 4529} 4530 4531//===--- CHECK: Objective-C retain cycles ----------------------------------// 4532 4533namespace { 4534 struct RetainCycleOwner { 4535 RetainCycleOwner() : Variable(0), Indirect(false) {} 4536 VarDecl *Variable; 4537 SourceRange Range; 4538 SourceLocation Loc; 4539 bool Indirect; 4540 4541 void setLocsFrom(Expr *e) { 4542 Loc = e->getExprLoc(); 4543 Range = e->getSourceRange(); 4544 } 4545 }; 4546} 4547 4548/// Consider whether capturing the given variable can possibly lead to 4549/// a retain cycle. 4550static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 4551 // In ARC, it's captured strongly iff the variable has __strong 4552 // lifetime. In MRR, it's captured strongly if the variable is 4553 // __block and has an appropriate type. 4554 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4555 return false; 4556 4557 owner.Variable = var; 4558 owner.setLocsFrom(ref); 4559 return true; 4560} 4561 4562static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) { 4563 while (true) { 4564 e = e->IgnoreParens(); 4565 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 4566 switch (cast->getCastKind()) { 4567 case CK_BitCast: 4568 case CK_LValueBitCast: 4569 case CK_LValueToRValue: 4570 case CK_ARCReclaimReturnedObject: 4571 e = cast->getSubExpr(); 4572 continue; 4573 4574 default: 4575 return false; 4576 } 4577 } 4578 4579 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 4580 ObjCIvarDecl *ivar = ref->getDecl(); 4581 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4582 return false; 4583 4584 // Try to find a retain cycle in the base. 4585 if (!findRetainCycleOwner(S, ref->getBase(), owner)) 4586 return false; 4587 4588 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 4589 owner.Indirect = true; 4590 return true; 4591 } 4592 4593 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 4594 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 4595 if (!var) return false; 4596 return considerVariable(var, ref, owner); 4597 } 4598 4599 if (BlockDeclRefExpr *ref = dyn_cast<BlockDeclRefExpr>(e)) { 4600 owner.Variable = ref->getDecl(); 4601 owner.setLocsFrom(ref); 4602 return true; 4603 } 4604 4605 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 4606 if (member->isArrow()) return false; 4607 4608 // Don't count this as an indirect ownership. 4609 e = member->getBase(); 4610 continue; 4611 } 4612 4613 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 4614 // Only pay attention to pseudo-objects on property references. 4615 ObjCPropertyRefExpr *pre 4616 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 4617 ->IgnoreParens()); 4618 if (!pre) return false; 4619 if (pre->isImplicitProperty()) return false; 4620 ObjCPropertyDecl *property = pre->getExplicitProperty(); 4621 if (!property->isRetaining() && 4622 !(property->getPropertyIvarDecl() && 4623 property->getPropertyIvarDecl()->getType() 4624 .getObjCLifetime() == Qualifiers::OCL_Strong)) 4625 return false; 4626 4627 owner.Indirect = true; 4628 if (pre->isSuperReceiver()) { 4629 owner.Variable = S.getCurMethodDecl()->getSelfDecl(); 4630 if (!owner.Variable) 4631 return false; 4632 owner.Loc = pre->getLocation(); 4633 owner.Range = pre->getSourceRange(); 4634 return true; 4635 } 4636 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 4637 ->getSourceExpr()); 4638 continue; 4639 } 4640 4641 // Array ivars? 4642 4643 return false; 4644 } 4645} 4646 4647namespace { 4648 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 4649 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 4650 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 4651 Variable(variable), Capturer(0) {} 4652 4653 VarDecl *Variable; 4654 Expr *Capturer; 4655 4656 void VisitDeclRefExpr(DeclRefExpr *ref) { 4657 if (ref->getDecl() == Variable && !Capturer) 4658 Capturer = ref; 4659 } 4660 4661 void VisitBlockDeclRefExpr(BlockDeclRefExpr *ref) { 4662 if (ref->getDecl() == Variable && !Capturer) 4663 Capturer = ref; 4664 } 4665 4666 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 4667 if (Capturer) return; 4668 Visit(ref->getBase()); 4669 if (Capturer && ref->isFreeIvar()) 4670 Capturer = ref; 4671 } 4672 4673 void VisitBlockExpr(BlockExpr *block) { 4674 // Look inside nested blocks 4675 if (block->getBlockDecl()->capturesVariable(Variable)) 4676 Visit(block->getBlockDecl()->getBody()); 4677 } 4678 }; 4679} 4680 4681/// Check whether the given argument is a block which captures a 4682/// variable. 4683static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 4684 assert(owner.Variable && owner.Loc.isValid()); 4685 4686 e = e->IgnoreParenCasts(); 4687 BlockExpr *block = dyn_cast<BlockExpr>(e); 4688 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 4689 return 0; 4690 4691 FindCaptureVisitor visitor(S.Context, owner.Variable); 4692 visitor.Visit(block->getBlockDecl()->getBody()); 4693 return visitor.Capturer; 4694} 4695 4696static void diagnoseRetainCycle(Sema &S, Expr *capturer, 4697 RetainCycleOwner &owner) { 4698 assert(capturer); 4699 assert(owner.Variable && owner.Loc.isValid()); 4700 4701 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 4702 << owner.Variable << capturer->getSourceRange(); 4703 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 4704 << owner.Indirect << owner.Range; 4705} 4706 4707/// Check for a keyword selector that starts with the word 'add' or 4708/// 'set'. 4709static bool isSetterLikeSelector(Selector sel) { 4710 if (sel.isUnarySelector()) return false; 4711 4712 StringRef str = sel.getNameForSlot(0); 4713 while (!str.empty() && str.front() == '_') str = str.substr(1); 4714 if (str.startswith("set")) 4715 str = str.substr(3); 4716 else if (str.startswith("add")) { 4717 // Specially whitelist 'addOperationWithBlock:'. 4718 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 4719 return false; 4720 str = str.substr(3); 4721 } 4722 else 4723 return false; 4724 4725 if (str.empty()) return true; 4726 return !islower(str.front()); 4727} 4728 4729/// Check a message send to see if it's likely to cause a retain cycle. 4730void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 4731 // Only check instance methods whose selector looks like a setter. 4732 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 4733 return; 4734 4735 // Try to find a variable that the receiver is strongly owned by. 4736 RetainCycleOwner owner; 4737 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 4738 if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner)) 4739 return; 4740 } else { 4741 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 4742 owner.Variable = getCurMethodDecl()->getSelfDecl(); 4743 owner.Loc = msg->getSuperLoc(); 4744 owner.Range = msg->getSuperLoc(); 4745 } 4746 4747 // Check whether the receiver is captured by any of the arguments. 4748 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 4749 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 4750 return diagnoseRetainCycle(*this, capturer, owner); 4751} 4752 4753/// Check a property assign to see if it's likely to cause a retain cycle. 4754void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 4755 RetainCycleOwner owner; 4756 if (!findRetainCycleOwner(*this, receiver, owner)) 4757 return; 4758 4759 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 4760 diagnoseRetainCycle(*this, capturer, owner); 4761} 4762 4763bool Sema::checkUnsafeAssigns(SourceLocation Loc, 4764 QualType LHS, Expr *RHS) { 4765 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 4766 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 4767 return false; 4768 // strip off any implicit cast added to get to the one arc-specific 4769 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4770 if (cast->getCastKind() == CK_ARCConsumeObject) { 4771 Diag(Loc, diag::warn_arc_retained_assign) 4772 << (LT == Qualifiers::OCL_ExplicitNone) 4773 << RHS->getSourceRange(); 4774 return true; 4775 } 4776 RHS = cast->getSubExpr(); 4777 } 4778 return false; 4779} 4780 4781void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 4782 Expr *LHS, Expr *RHS) { 4783 QualType LHSType; 4784 // PropertyRef on LHS type need be directly obtained from 4785 // its declaration as it has a PsuedoType. 4786 ObjCPropertyRefExpr *PRE 4787 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens()); 4788 if (PRE && !PRE->isImplicitProperty()) { 4789 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 4790 if (PD) 4791 LHSType = PD->getType(); 4792 } 4793 4794 if (LHSType.isNull()) 4795 LHSType = LHS->getType(); 4796 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 4797 return; 4798 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 4799 // FIXME. Check for other life times. 4800 if (LT != Qualifiers::OCL_None) 4801 return; 4802 4803 if (PRE) { 4804 if (PRE->isImplicitProperty()) 4805 return; 4806 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 4807 if (!PD) 4808 return; 4809 4810 unsigned Attributes = PD->getPropertyAttributes(); 4811 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) { 4812 // when 'assign' attribute was not explicitly specified 4813 // by user, ignore it and rely on property type itself 4814 // for lifetime info. 4815 unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten(); 4816 if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) && 4817 LHSType->isObjCRetainableType()) 4818 return; 4819 4820 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4821 if (cast->getCastKind() == CK_ARCConsumeObject) { 4822 Diag(Loc, diag::warn_arc_retained_property_assign) 4823 << RHS->getSourceRange(); 4824 return; 4825 } 4826 RHS = cast->getSubExpr(); 4827 } 4828 } 4829 } 4830} 4831