SemaChecking.cpp revision f7ccbad5d9949e7ddd1cbef43d482553b811e026
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::IntegerLiteralClass: 1385 // Technically -Wformat-nonliteral does not warn about this case. 1386 // The behavior of printf and friends in this case is implementation 1387 // dependent. Ideally if the format string cannot be null then 1388 // it should have a 'nonnull' attribute in the function prototype. 1389 return true; 1390 1391 case Stmt::ImplicitCastExprClass: { 1392 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 1393 goto tryAgain; 1394 } 1395 1396 case Stmt::OpaqueValueExprClass: 1397 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 1398 E = src; 1399 goto tryAgain; 1400 } 1401 return false; 1402 1403 case Stmt::PredefinedExprClass: 1404 // While __func__, etc., are technically not string literals, they 1405 // cannot contain format specifiers and thus are not a security 1406 // liability. 1407 return true; 1408 1409 case Stmt::DeclRefExprClass: { 1410 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1411 1412 // As an exception, do not flag errors for variables binding to 1413 // const string literals. 1414 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1415 bool isConstant = false; 1416 QualType T = DR->getType(); 1417 1418 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1419 isConstant = AT->getElementType().isConstant(Context); 1420 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1421 isConstant = T.isConstant(Context) && 1422 PT->getPointeeType().isConstant(Context); 1423 } else if (T->isObjCObjectPointerType()) { 1424 // In ObjC, there is usually no "const ObjectPointer" type, 1425 // so don't check if the pointee type is constant. 1426 isConstant = T.isConstant(Context); 1427 } 1428 1429 if (isConstant) { 1430 if (const Expr *Init = VD->getAnyInitializer()) 1431 return SemaCheckStringLiteral(Init, Args, NumArgs, 1432 HasVAListArg, format_idx, firstDataArg, 1433 Type, /*inFunctionCall*/false); 1434 } 1435 1436 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1437 // special check to see if the format string is a function parameter 1438 // of the function calling the printf function. If the function 1439 // has an attribute indicating it is a printf-like function, then we 1440 // should suppress warnings concerning non-literals being used in a call 1441 // to a vprintf function. For example: 1442 // 1443 // void 1444 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1445 // va_list ap; 1446 // va_start(ap, fmt); 1447 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1448 // ... 1449 // 1450 // 1451 // FIXME: We don't have full attribute support yet, so just check to see 1452 // if the argument is a DeclRefExpr that references a parameter. We'll 1453 // add proper support for checking the attribute later. 1454 if (HasVAListArg) 1455 if (isa<ParmVarDecl>(VD)) 1456 return true; 1457 } 1458 1459 return false; 1460 } 1461 1462 case Stmt::CallExprClass: { 1463 const CallExpr *CE = cast<CallExpr>(E); 1464 if (const ImplicitCastExpr *ICE 1465 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 1466 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 1467 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 1468 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 1469 unsigned ArgIndex = FA->getFormatIdx(); 1470 const Expr *Arg = CE->getArg(ArgIndex - 1); 1471 1472 return SemaCheckStringLiteral(Arg, Args, NumArgs, HasVAListArg, 1473 format_idx, firstDataArg, Type, 1474 inFunctionCall); 1475 } 1476 } 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 if (isa<CXXMemberCallExpr>(TheCall)) { 1538 const CXXMethodDecl *method_decl = 1539 dyn_cast<CXXMethodDecl>(TheCall->getCalleeDecl()); 1540 IsCXXMember = method_decl && method_decl->isInstance(); 1541 } 1542 CheckFormatArguments(Format, TheCall->getArgs(), TheCall->getNumArgs(), 1543 IsCXXMember, TheCall->getRParenLoc(), 1544 TheCall->getCallee()->getSourceRange()); 1545} 1546 1547void Sema::CheckFormatArguments(const FormatAttr *Format, Expr **Args, 1548 unsigned NumArgs, bool IsCXXMember, 1549 SourceLocation Loc, SourceRange Range) { 1550 bool HasVAListArg = Format->getFirstArg() == 0; 1551 unsigned format_idx = Format->getFormatIdx() - 1; 1552 unsigned firstDataArg = HasVAListArg ? 0 : Format->getFirstArg() - 1; 1553 if (IsCXXMember) { 1554 if (format_idx == 0) 1555 return; 1556 --format_idx; 1557 if(firstDataArg != 0) 1558 --firstDataArg; 1559 } 1560 CheckFormatArguments(Args, NumArgs, HasVAListArg, format_idx, 1561 firstDataArg, GetFormatStringType(Format), Loc, Range); 1562} 1563 1564void Sema::CheckFormatArguments(Expr **Args, unsigned NumArgs, 1565 bool HasVAListArg, unsigned format_idx, 1566 unsigned firstDataArg, FormatStringType Type, 1567 SourceLocation Loc, SourceRange Range) { 1568 // CHECK: printf/scanf-like function is called with no format string. 1569 if (format_idx >= NumArgs) { 1570 Diag(Loc, diag::warn_missing_format_string) << Range; 1571 return; 1572 } 1573 1574 const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts(); 1575 1576 // CHECK: format string is not a string literal. 1577 // 1578 // Dynamically generated format strings are difficult to 1579 // automatically vet at compile time. Requiring that format strings 1580 // are string literals: (1) permits the checking of format strings by 1581 // the compiler and thereby (2) can practically remove the source of 1582 // many format string exploits. 1583 1584 // Format string can be either ObjC string (e.g. @"%d") or 1585 // C string (e.g. "%d") 1586 // ObjC string uses the same format specifiers as C string, so we can use 1587 // the same format string checking logic for both ObjC and C strings. 1588 if (SemaCheckStringLiteral(OrigFormatExpr, Args, NumArgs, HasVAListArg, 1589 format_idx, firstDataArg, Type)) 1590 return; // Literal format string found, check done! 1591 1592 // Do not emit diag when the string param is a macro expansion and the 1593 // format is either NSString or CFString. This is a hack to prevent 1594 // diag when using the NSLocalizedString and CFCopyLocalizedString macros 1595 // which are usually used in place of NS and CF string literals. 1596 if (Type == FST_NSString && Args[format_idx]->getLocStart().isMacroID()) 1597 return; 1598 1599 // If there are no arguments specified, warn with -Wformat-security, otherwise 1600 // warn only with -Wformat-nonliteral. 1601 if (NumArgs == format_idx+1) 1602 Diag(Args[format_idx]->getLocStart(), 1603 diag::warn_format_nonliteral_noargs) 1604 << OrigFormatExpr->getSourceRange(); 1605 else 1606 Diag(Args[format_idx]->getLocStart(), 1607 diag::warn_format_nonliteral) 1608 << OrigFormatExpr->getSourceRange(); 1609} 1610 1611namespace { 1612class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1613protected: 1614 Sema &S; 1615 const StringLiteral *FExpr; 1616 const Expr *OrigFormatExpr; 1617 const unsigned FirstDataArg; 1618 const unsigned NumDataArgs; 1619 const bool IsObjCLiteral; 1620 const char *Beg; // Start of format string. 1621 const bool HasVAListArg; 1622 const Expr * const *Args; 1623 const unsigned NumArgs; 1624 unsigned FormatIdx; 1625 llvm::BitVector CoveredArgs; 1626 bool usesPositionalArgs; 1627 bool atFirstArg; 1628 bool inFunctionCall; 1629public: 1630 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1631 const Expr *origFormatExpr, unsigned firstDataArg, 1632 unsigned numDataArgs, bool isObjCLiteral, 1633 const char *beg, bool hasVAListArg, 1634 Expr **args, unsigned numArgs, 1635 unsigned formatIdx, bool inFunctionCall) 1636 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1637 FirstDataArg(firstDataArg), 1638 NumDataArgs(numDataArgs), 1639 IsObjCLiteral(isObjCLiteral), Beg(beg), 1640 HasVAListArg(hasVAListArg), 1641 Args(args), NumArgs(numArgs), FormatIdx(formatIdx), 1642 usesPositionalArgs(false), atFirstArg(true), 1643 inFunctionCall(inFunctionCall) { 1644 CoveredArgs.resize(numDataArgs); 1645 CoveredArgs.reset(); 1646 } 1647 1648 void DoneProcessing(); 1649 1650 void HandleIncompleteSpecifier(const char *startSpecifier, 1651 unsigned specifierLen); 1652 1653 virtual void HandleInvalidPosition(const char *startSpecifier, 1654 unsigned specifierLen, 1655 analyze_format_string::PositionContext p); 1656 1657 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1658 1659 void HandleNullChar(const char *nullCharacter); 1660 1661 template <typename Range> 1662 static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall, 1663 const Expr *ArgumentExpr, 1664 PartialDiagnostic PDiag, 1665 SourceLocation StringLoc, 1666 bool IsStringLocation, Range StringRange, 1667 FixItHint Fixit = FixItHint()); 1668 1669protected: 1670 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1671 const char *startSpec, 1672 unsigned specifierLen, 1673 const char *csStart, unsigned csLen); 1674 1675 void HandlePositionalNonpositionalArgs(SourceLocation Loc, 1676 const char *startSpec, 1677 unsigned specifierLen); 1678 1679 SourceRange getFormatStringRange(); 1680 CharSourceRange getSpecifierRange(const char *startSpecifier, 1681 unsigned specifierLen); 1682 SourceLocation getLocationOfByte(const char *x); 1683 1684 const Expr *getDataArg(unsigned i) const; 1685 1686 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1687 const analyze_format_string::ConversionSpecifier &CS, 1688 const char *startSpecifier, unsigned specifierLen, 1689 unsigned argIndex); 1690 1691 template <typename Range> 1692 void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc, 1693 bool IsStringLocation, Range StringRange, 1694 FixItHint Fixit = FixItHint()); 1695 1696 void CheckPositionalAndNonpositionalArgs( 1697 const analyze_format_string::FormatSpecifier *FS); 1698}; 1699} 1700 1701SourceRange CheckFormatHandler::getFormatStringRange() { 1702 return OrigFormatExpr->getSourceRange(); 1703} 1704 1705CharSourceRange CheckFormatHandler:: 1706getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1707 SourceLocation Start = getLocationOfByte(startSpecifier); 1708 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1709 1710 // Advance the end SourceLocation by one due to half-open ranges. 1711 End = End.getLocWithOffset(1); 1712 1713 return CharSourceRange::getCharRange(Start, End); 1714} 1715 1716SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 1717 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1718} 1719 1720void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 1721 unsigned specifierLen){ 1722 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier), 1723 getLocationOfByte(startSpecifier), 1724 /*IsStringLocation*/true, 1725 getSpecifierRange(startSpecifier, specifierLen)); 1726} 1727 1728void 1729CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1730 analyze_format_string::PositionContext p) { 1731 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier) 1732 << (unsigned) p, 1733 getLocationOfByte(startPos), /*IsStringLocation*/true, 1734 getSpecifierRange(startPos, posLen)); 1735} 1736 1737void CheckFormatHandler::HandleZeroPosition(const char *startPos, 1738 unsigned posLen) { 1739 EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier), 1740 getLocationOfByte(startPos), 1741 /*IsStringLocation*/true, 1742 getSpecifierRange(startPos, posLen)); 1743} 1744 1745void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 1746 if (!IsObjCLiteral) { 1747 // The presence of a null character is likely an error. 1748 EmitFormatDiagnostic( 1749 S.PDiag(diag::warn_printf_format_string_contains_null_char), 1750 getLocationOfByte(nullCharacter), /*IsStringLocation*/true, 1751 getFormatStringRange()); 1752 } 1753} 1754 1755const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 1756 return Args[FirstDataArg + i]; 1757} 1758 1759void CheckFormatHandler::DoneProcessing() { 1760 // Does the number of data arguments exceed the number of 1761 // format conversions in the format string? 1762 if (!HasVAListArg) { 1763 // Find any arguments that weren't covered. 1764 CoveredArgs.flip(); 1765 signed notCoveredArg = CoveredArgs.find_first(); 1766 if (notCoveredArg >= 0) { 1767 assert((unsigned)notCoveredArg < NumDataArgs); 1768 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used), 1769 getDataArg((unsigned) notCoveredArg)->getLocStart(), 1770 /*IsStringLocation*/false, getFormatStringRange()); 1771 } 1772 } 1773} 1774 1775bool 1776CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 1777 SourceLocation Loc, 1778 const char *startSpec, 1779 unsigned specifierLen, 1780 const char *csStart, 1781 unsigned csLen) { 1782 1783 bool keepGoing = true; 1784 if (argIndex < NumDataArgs) { 1785 // Consider the argument coverered, even though the specifier doesn't 1786 // make sense. 1787 CoveredArgs.set(argIndex); 1788 } 1789 else { 1790 // If argIndex exceeds the number of data arguments we 1791 // don't issue a warning because that is just a cascade of warnings (and 1792 // they may have intended '%%' anyway). We don't want to continue processing 1793 // the format string after this point, however, as we will like just get 1794 // gibberish when trying to match arguments. 1795 keepGoing = false; 1796 } 1797 1798 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion) 1799 << StringRef(csStart, csLen), 1800 Loc, /*IsStringLocation*/true, 1801 getSpecifierRange(startSpec, specifierLen)); 1802 1803 return keepGoing; 1804} 1805 1806void 1807CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc, 1808 const char *startSpec, 1809 unsigned specifierLen) { 1810 EmitFormatDiagnostic( 1811 S.PDiag(diag::warn_format_mix_positional_nonpositional_args), 1812 Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen)); 1813} 1814 1815bool 1816CheckFormatHandler::CheckNumArgs( 1817 const analyze_format_string::FormatSpecifier &FS, 1818 const analyze_format_string::ConversionSpecifier &CS, 1819 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 1820 1821 if (argIndex >= NumDataArgs) { 1822 PartialDiagnostic PDiag = FS.usesPositionalArg() 1823 ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args) 1824 << (argIndex+1) << NumDataArgs) 1825 : S.PDiag(diag::warn_printf_insufficient_data_args); 1826 EmitFormatDiagnostic( 1827 PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true, 1828 getSpecifierRange(startSpecifier, specifierLen)); 1829 return false; 1830 } 1831 return true; 1832} 1833 1834template<typename Range> 1835void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag, 1836 SourceLocation Loc, 1837 bool IsStringLocation, 1838 Range StringRange, 1839 FixItHint FixIt) { 1840 EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag, 1841 Loc, IsStringLocation, StringRange, FixIt); 1842} 1843 1844/// \brief If the format string is not within the funcion call, emit a note 1845/// so that the function call and string are in diagnostic messages. 1846/// 1847/// \param inFunctionCall if true, the format string is within the function 1848/// call and only one diagnostic message will be produced. Otherwise, an 1849/// extra note will be emitted pointing to location of the format string. 1850/// 1851/// \param ArgumentExpr the expression that is passed as the format string 1852/// argument in the function call. Used for getting locations when two 1853/// diagnostics are emitted. 1854/// 1855/// \param PDiag the callee should already have provided any strings for the 1856/// diagnostic message. This function only adds locations and fixits 1857/// to diagnostics. 1858/// 1859/// \param Loc primary location for diagnostic. If two diagnostics are 1860/// required, one will be at Loc and a new SourceLocation will be created for 1861/// the other one. 1862/// 1863/// \param IsStringLocation if true, Loc points to the format string should be 1864/// used for the note. Otherwise, Loc points to the argument list and will 1865/// be used with PDiag. 1866/// 1867/// \param StringRange some or all of the string to highlight. This is 1868/// templated so it can accept either a CharSourceRange or a SourceRange. 1869/// 1870/// \param Fixit optional fix it hint for the format string. 1871template<typename Range> 1872void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall, 1873 const Expr *ArgumentExpr, 1874 PartialDiagnostic PDiag, 1875 SourceLocation Loc, 1876 bool IsStringLocation, 1877 Range StringRange, 1878 FixItHint FixIt) { 1879 if (InFunctionCall) 1880 S.Diag(Loc, PDiag) << StringRange << FixIt; 1881 else { 1882 S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag) 1883 << ArgumentExpr->getSourceRange(); 1884 S.Diag(IsStringLocation ? Loc : StringRange.getBegin(), 1885 diag::note_format_string_defined) 1886 << StringRange << FixIt; 1887 } 1888} 1889 1890//===--- CHECK: Printf format string checking ------------------------------===// 1891 1892namespace { 1893class CheckPrintfHandler : public CheckFormatHandler { 1894public: 1895 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1896 const Expr *origFormatExpr, unsigned firstDataArg, 1897 unsigned numDataArgs, bool isObjCLiteral, 1898 const char *beg, bool hasVAListArg, 1899 Expr **Args, unsigned NumArgs, 1900 unsigned formatIdx, bool inFunctionCall) 1901 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1902 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1903 Args, NumArgs, formatIdx, inFunctionCall) {} 1904 1905 1906 bool HandleInvalidPrintfConversionSpecifier( 1907 const analyze_printf::PrintfSpecifier &FS, 1908 const char *startSpecifier, 1909 unsigned specifierLen); 1910 1911 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 1912 const char *startSpecifier, 1913 unsigned specifierLen); 1914 1915 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 1916 const char *startSpecifier, unsigned specifierLen); 1917 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 1918 const analyze_printf::OptionalAmount &Amt, 1919 unsigned type, 1920 const char *startSpecifier, unsigned specifierLen); 1921 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1922 const analyze_printf::OptionalFlag &flag, 1923 const char *startSpecifier, unsigned specifierLen); 1924 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 1925 const analyze_printf::OptionalFlag &ignoredFlag, 1926 const analyze_printf::OptionalFlag &flag, 1927 const char *startSpecifier, unsigned specifierLen); 1928}; 1929} 1930 1931bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 1932 const analyze_printf::PrintfSpecifier &FS, 1933 const char *startSpecifier, 1934 unsigned specifierLen) { 1935 const analyze_printf::PrintfConversionSpecifier &CS = 1936 FS.getConversionSpecifier(); 1937 1938 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1939 getLocationOfByte(CS.getStart()), 1940 startSpecifier, specifierLen, 1941 CS.getStart(), CS.getLength()); 1942} 1943 1944bool CheckPrintfHandler::HandleAmount( 1945 const analyze_format_string::OptionalAmount &Amt, 1946 unsigned k, const char *startSpecifier, 1947 unsigned specifierLen) { 1948 1949 if (Amt.hasDataArgument()) { 1950 if (!HasVAListArg) { 1951 unsigned argIndex = Amt.getArgIndex(); 1952 if (argIndex >= NumDataArgs) { 1953 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg) 1954 << k, 1955 getLocationOfByte(Amt.getStart()), 1956 /*IsStringLocation*/true, 1957 getSpecifierRange(startSpecifier, specifierLen)); 1958 // Don't do any more checking. We will just emit 1959 // spurious errors. 1960 return false; 1961 } 1962 1963 // Type check the data argument. It should be an 'int'. 1964 // Although not in conformance with C99, we also allow the argument to be 1965 // an 'unsigned int' as that is a reasonably safe case. GCC also 1966 // doesn't emit a warning for that case. 1967 CoveredArgs.set(argIndex); 1968 const Expr *Arg = getDataArg(argIndex); 1969 QualType T = Arg->getType(); 1970 1971 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1972 assert(ATR.isValid()); 1973 1974 if (!ATR.matchesType(S.Context, T)) { 1975 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type) 1976 << k << ATR.getRepresentativeTypeName(S.Context) 1977 << T << Arg->getSourceRange(), 1978 getLocationOfByte(Amt.getStart()), 1979 /*IsStringLocation*/true, 1980 getSpecifierRange(startSpecifier, specifierLen)); 1981 // Don't do any more checking. We will just emit 1982 // spurious errors. 1983 return false; 1984 } 1985 } 1986 } 1987 return true; 1988} 1989 1990void CheckPrintfHandler::HandleInvalidAmount( 1991 const analyze_printf::PrintfSpecifier &FS, 1992 const analyze_printf::OptionalAmount &Amt, 1993 unsigned type, 1994 const char *startSpecifier, 1995 unsigned specifierLen) { 1996 const analyze_printf::PrintfConversionSpecifier &CS = 1997 FS.getConversionSpecifier(); 1998 1999 FixItHint fixit = 2000 Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant 2001 ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 2002 Amt.getConstantLength())) 2003 : FixItHint(); 2004 2005 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount) 2006 << type << CS.toString(), 2007 getLocationOfByte(Amt.getStart()), 2008 /*IsStringLocation*/true, 2009 getSpecifierRange(startSpecifier, specifierLen), 2010 fixit); 2011} 2012 2013void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2014 const analyze_printf::OptionalFlag &flag, 2015 const char *startSpecifier, 2016 unsigned specifierLen) { 2017 // Warn about pointless flag with a fixit removal. 2018 const analyze_printf::PrintfConversionSpecifier &CS = 2019 FS.getConversionSpecifier(); 2020 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag) 2021 << flag.toString() << CS.toString(), 2022 getLocationOfByte(flag.getPosition()), 2023 /*IsStringLocation*/true, 2024 getSpecifierRange(startSpecifier, specifierLen), 2025 FixItHint::CreateRemoval( 2026 getSpecifierRange(flag.getPosition(), 1))); 2027} 2028 2029void CheckPrintfHandler::HandleIgnoredFlag( 2030 const analyze_printf::PrintfSpecifier &FS, 2031 const analyze_printf::OptionalFlag &ignoredFlag, 2032 const analyze_printf::OptionalFlag &flag, 2033 const char *startSpecifier, 2034 unsigned specifierLen) { 2035 // Warn about ignored flag with a fixit removal. 2036 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag) 2037 << ignoredFlag.toString() << flag.toString(), 2038 getLocationOfByte(ignoredFlag.getPosition()), 2039 /*IsStringLocation*/true, 2040 getSpecifierRange(startSpecifier, specifierLen), 2041 FixItHint::CreateRemoval( 2042 getSpecifierRange(ignoredFlag.getPosition(), 1))); 2043} 2044 2045bool 2046CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 2047 &FS, 2048 const char *startSpecifier, 2049 unsigned specifierLen) { 2050 2051 using namespace analyze_format_string; 2052 using namespace analyze_printf; 2053 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 2054 2055 if (FS.consumesDataArgument()) { 2056 if (atFirstArg) { 2057 atFirstArg = false; 2058 usesPositionalArgs = FS.usesPositionalArg(); 2059 } 2060 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2061 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2062 startSpecifier, specifierLen); 2063 return false; 2064 } 2065 } 2066 2067 // First check if the field width, precision, and conversion specifier 2068 // have matching data arguments. 2069 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 2070 startSpecifier, specifierLen)) { 2071 return false; 2072 } 2073 2074 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 2075 startSpecifier, specifierLen)) { 2076 return false; 2077 } 2078 2079 if (!CS.consumesDataArgument()) { 2080 // FIXME: Technically specifying a precision or field width here 2081 // makes no sense. Worth issuing a warning at some point. 2082 return true; 2083 } 2084 2085 // Consume the argument. 2086 unsigned argIndex = FS.getArgIndex(); 2087 if (argIndex < NumDataArgs) { 2088 // The check to see if the argIndex is valid will come later. 2089 // We set the bit here because we may exit early from this 2090 // function if we encounter some other error. 2091 CoveredArgs.set(argIndex); 2092 } 2093 2094 // Check for using an Objective-C specific conversion specifier 2095 // in a non-ObjC literal. 2096 if (!IsObjCLiteral && CS.isObjCArg()) { 2097 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 2098 specifierLen); 2099 } 2100 2101 // Check for invalid use of field width 2102 if (!FS.hasValidFieldWidth()) { 2103 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 2104 startSpecifier, specifierLen); 2105 } 2106 2107 // Check for invalid use of precision 2108 if (!FS.hasValidPrecision()) { 2109 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 2110 startSpecifier, specifierLen); 2111 } 2112 2113 // Check each flag does not conflict with any other component. 2114 if (!FS.hasValidThousandsGroupingPrefix()) 2115 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 2116 if (!FS.hasValidLeadingZeros()) 2117 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 2118 if (!FS.hasValidPlusPrefix()) 2119 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 2120 if (!FS.hasValidSpacePrefix()) 2121 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 2122 if (!FS.hasValidAlternativeForm()) 2123 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 2124 if (!FS.hasValidLeftJustified()) 2125 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 2126 2127 // Check that flags are not ignored by another flag 2128 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 2129 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 2130 startSpecifier, specifierLen); 2131 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 2132 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 2133 startSpecifier, specifierLen); 2134 2135 // Check the length modifier is valid with the given conversion specifier. 2136 const LengthModifier &LM = FS.getLengthModifier(); 2137 if (!FS.hasValidLengthModifier()) 2138 EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length) 2139 << LM.toString() << CS.toString(), 2140 getLocationOfByte(LM.getStart()), 2141 /*IsStringLocation*/true, 2142 getSpecifierRange(startSpecifier, specifierLen), 2143 FixItHint::CreateRemoval( 2144 getSpecifierRange(LM.getStart(), 2145 LM.getLength()))); 2146 2147 // Are we using '%n'? 2148 if (CS.getKind() == ConversionSpecifier::nArg) { 2149 // Issue a warning about this being a possible security issue. 2150 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_write_back), 2151 getLocationOfByte(CS.getStart()), 2152 /*IsStringLocation*/true, 2153 getSpecifierRange(startSpecifier, specifierLen)); 2154 // Continue checking the other format specifiers. 2155 return true; 2156 } 2157 2158 // The remaining checks depend on the data arguments. 2159 if (HasVAListArg) 2160 return true; 2161 2162 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2163 return false; 2164 2165 // Now type check the data expression that matches the 2166 // format specifier. 2167 const Expr *Ex = getDataArg(argIndex); 2168 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context, 2169 IsObjCLiteral); 2170 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 2171 // Check if we didn't match because of an implicit cast from a 'char' 2172 // or 'short' to an 'int'. This is done because printf is a varargs 2173 // function. 2174 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 2175 if (ICE->getType() == S.Context.IntTy) { 2176 // All further checking is done on the subexpression. 2177 Ex = ICE->getSubExpr(); 2178 if (ATR.matchesType(S.Context, Ex->getType())) 2179 return true; 2180 } 2181 2182 // We may be able to offer a FixItHint if it is a supported type. 2183 PrintfSpecifier fixedFS = FS; 2184 bool success = fixedFS.fixType(Ex->getType(), S.getLangOptions()); 2185 2186 if (success) { 2187 // Get the fix string from the fixed format specifier 2188 SmallString<128> buf; 2189 llvm::raw_svector_ostream os(buf); 2190 fixedFS.toString(os); 2191 2192 EmitFormatDiagnostic( 2193 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2194 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2195 << Ex->getSourceRange(), 2196 getLocationOfByte(CS.getStart()), 2197 /*IsStringLocation*/true, 2198 getSpecifierRange(startSpecifier, specifierLen), 2199 FixItHint::CreateReplacement( 2200 getSpecifierRange(startSpecifier, specifierLen), 2201 os.str())); 2202 } 2203 else { 2204 EmitFormatDiagnostic( 2205 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2206 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2207 << getSpecifierRange(startSpecifier, specifierLen) 2208 << Ex->getSourceRange(), 2209 getLocationOfByte(CS.getStart()), 2210 true, 2211 getSpecifierRange(startSpecifier, specifierLen)); 2212 } 2213 } 2214 2215 return true; 2216} 2217 2218//===--- CHECK: Scanf format string checking ------------------------------===// 2219 2220namespace { 2221class CheckScanfHandler : public CheckFormatHandler { 2222public: 2223 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 2224 const Expr *origFormatExpr, unsigned firstDataArg, 2225 unsigned numDataArgs, bool isObjCLiteral, 2226 const char *beg, bool hasVAListArg, 2227 Expr **Args, unsigned NumArgs, 2228 unsigned formatIdx, bool inFunctionCall) 2229 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2230 numDataArgs, isObjCLiteral, beg, hasVAListArg, 2231 Args, NumArgs, formatIdx, inFunctionCall) {} 2232 2233 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 2234 const char *startSpecifier, 2235 unsigned specifierLen); 2236 2237 bool HandleInvalidScanfConversionSpecifier( 2238 const analyze_scanf::ScanfSpecifier &FS, 2239 const char *startSpecifier, 2240 unsigned specifierLen); 2241 2242 void HandleIncompleteScanList(const char *start, const char *end); 2243}; 2244} 2245 2246void CheckScanfHandler::HandleIncompleteScanList(const char *start, 2247 const char *end) { 2248 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 2249 getLocationOfByte(end), /*IsStringLocation*/true, 2250 getSpecifierRange(start, end - start)); 2251} 2252 2253bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 2254 const analyze_scanf::ScanfSpecifier &FS, 2255 const char *startSpecifier, 2256 unsigned specifierLen) { 2257 2258 const analyze_scanf::ScanfConversionSpecifier &CS = 2259 FS.getConversionSpecifier(); 2260 2261 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2262 getLocationOfByte(CS.getStart()), 2263 startSpecifier, specifierLen, 2264 CS.getStart(), CS.getLength()); 2265} 2266 2267bool CheckScanfHandler::HandleScanfSpecifier( 2268 const analyze_scanf::ScanfSpecifier &FS, 2269 const char *startSpecifier, 2270 unsigned specifierLen) { 2271 2272 using namespace analyze_scanf; 2273 using namespace analyze_format_string; 2274 2275 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 2276 2277 // Handle case where '%' and '*' don't consume an argument. These shouldn't 2278 // be used to decide if we are using positional arguments consistently. 2279 if (FS.consumesDataArgument()) { 2280 if (atFirstArg) { 2281 atFirstArg = false; 2282 usesPositionalArgs = FS.usesPositionalArg(); 2283 } 2284 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2285 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2286 startSpecifier, specifierLen); 2287 return false; 2288 } 2289 } 2290 2291 // Check if the field with is non-zero. 2292 const OptionalAmount &Amt = FS.getFieldWidth(); 2293 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 2294 if (Amt.getConstantAmount() == 0) { 2295 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 2296 Amt.getConstantLength()); 2297 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 2298 getLocationOfByte(Amt.getStart()), 2299 /*IsStringLocation*/true, R, 2300 FixItHint::CreateRemoval(R)); 2301 } 2302 } 2303 2304 if (!FS.consumesDataArgument()) { 2305 // FIXME: Technically specifying a precision or field width here 2306 // makes no sense. Worth issuing a warning at some point. 2307 return true; 2308 } 2309 2310 // Consume the argument. 2311 unsigned argIndex = FS.getArgIndex(); 2312 if (argIndex < NumDataArgs) { 2313 // The check to see if the argIndex is valid will come later. 2314 // We set the bit here because we may exit early from this 2315 // function if we encounter some other error. 2316 CoveredArgs.set(argIndex); 2317 } 2318 2319 // Check the length modifier is valid with the given conversion specifier. 2320 const LengthModifier &LM = FS.getLengthModifier(); 2321 if (!FS.hasValidLengthModifier()) { 2322 const CharSourceRange &R = getSpecifierRange(LM.getStart(), LM.getLength()); 2323 EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length) 2324 << LM.toString() << CS.toString() 2325 << getSpecifierRange(startSpecifier, specifierLen), 2326 getLocationOfByte(LM.getStart()), 2327 /*IsStringLocation*/true, R, 2328 FixItHint::CreateRemoval(R)); 2329 } 2330 2331 // The remaining checks depend on the data arguments. 2332 if (HasVAListArg) 2333 return true; 2334 2335 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2336 return false; 2337 2338 // Check that the argument type matches the format specifier. 2339 const Expr *Ex = getDataArg(argIndex); 2340 const analyze_scanf::ScanfArgTypeResult &ATR = FS.getArgType(S.Context); 2341 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 2342 ScanfSpecifier fixedFS = FS; 2343 bool success = fixedFS.fixType(Ex->getType(), S.getLangOptions()); 2344 2345 if (success) { 2346 // Get the fix string from the fixed format specifier. 2347 SmallString<128> buf; 2348 llvm::raw_svector_ostream os(buf); 2349 fixedFS.toString(os); 2350 2351 EmitFormatDiagnostic( 2352 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2353 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2354 << Ex->getSourceRange(), 2355 getLocationOfByte(CS.getStart()), 2356 /*IsStringLocation*/true, 2357 getSpecifierRange(startSpecifier, specifierLen), 2358 FixItHint::CreateReplacement( 2359 getSpecifierRange(startSpecifier, specifierLen), 2360 os.str())); 2361 } else { 2362 EmitFormatDiagnostic( 2363 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2364 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2365 << Ex->getSourceRange(), 2366 getLocationOfByte(CS.getStart()), 2367 /*IsStringLocation*/true, 2368 getSpecifierRange(startSpecifier, specifierLen)); 2369 } 2370 } 2371 2372 return true; 2373} 2374 2375void Sema::CheckFormatString(const StringLiteral *FExpr, 2376 const Expr *OrigFormatExpr, 2377 Expr **Args, unsigned NumArgs, 2378 bool HasVAListArg, unsigned format_idx, 2379 unsigned firstDataArg, FormatStringType Type, 2380 bool inFunctionCall) { 2381 2382 // CHECK: is the format string a wide literal? 2383 if (!FExpr->isAscii()) { 2384 CheckFormatHandler::EmitFormatDiagnostic( 2385 *this, inFunctionCall, Args[format_idx], 2386 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 2387 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2388 return; 2389 } 2390 2391 // Str - The format string. NOTE: this is NOT null-terminated! 2392 StringRef StrRef = FExpr->getString(); 2393 const char *Str = StrRef.data(); 2394 unsigned StrLen = StrRef.size(); 2395 const unsigned numDataArgs = NumArgs - firstDataArg; 2396 2397 // CHECK: empty format string? 2398 if (StrLen == 0 && numDataArgs > 0) { 2399 CheckFormatHandler::EmitFormatDiagnostic( 2400 *this, inFunctionCall, Args[format_idx], 2401 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 2402 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2403 return; 2404 } 2405 2406 if (Type == FST_Printf || Type == FST_NSString) { 2407 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2408 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2409 Str, HasVAListArg, Args, NumArgs, format_idx, 2410 inFunctionCall); 2411 2412 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen, 2413 getLangOptions())) 2414 H.DoneProcessing(); 2415 } else if (Type == FST_Scanf) { 2416 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2417 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2418 Str, HasVAListArg, Args, NumArgs, format_idx, 2419 inFunctionCall); 2420 2421 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen, 2422 getLangOptions())) 2423 H.DoneProcessing(); 2424 } // TODO: handle other formats 2425} 2426 2427//===--- CHECK: Standard memory functions ---------------------------------===// 2428 2429/// \brief Determine whether the given type is a dynamic class type (e.g., 2430/// whether it has a vtable). 2431static bool isDynamicClassType(QualType T) { 2432 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 2433 if (CXXRecordDecl *Definition = Record->getDefinition()) 2434 if (Definition->isDynamicClass()) 2435 return true; 2436 2437 return false; 2438} 2439 2440/// \brief If E is a sizeof expression, returns its argument expression, 2441/// otherwise returns NULL. 2442static const Expr *getSizeOfExprArg(const Expr* E) { 2443 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2444 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2445 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 2446 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 2447 2448 return 0; 2449} 2450 2451/// \brief If E is a sizeof expression, returns its argument type. 2452static QualType getSizeOfArgType(const Expr* E) { 2453 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2454 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2455 if (SizeOf->getKind() == clang::UETT_SizeOf) 2456 return SizeOf->getTypeOfArgument(); 2457 2458 return QualType(); 2459} 2460 2461/// \brief Check for dangerous or invalid arguments to memset(). 2462/// 2463/// This issues warnings on known problematic, dangerous or unspecified 2464/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 2465/// function calls. 2466/// 2467/// \param Call The call expression to diagnose. 2468void Sema::CheckMemaccessArguments(const CallExpr *Call, 2469 unsigned BId, 2470 IdentifierInfo *FnName) { 2471 assert(BId != 0); 2472 2473 // It is possible to have a non-standard definition of memset. Validate 2474 // we have enough arguments, and if not, abort further checking. 2475 unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3); 2476 if (Call->getNumArgs() < ExpectedNumArgs) 2477 return; 2478 2479 unsigned LastArg = (BId == Builtin::BImemset || 2480 BId == Builtin::BIstrndup ? 1 : 2); 2481 unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2); 2482 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 2483 2484 // We have special checking when the length is a sizeof expression. 2485 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 2486 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 2487 llvm::FoldingSetNodeID SizeOfArgID; 2488 2489 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 2490 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 2491 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 2492 2493 QualType DestTy = Dest->getType(); 2494 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 2495 QualType PointeeTy = DestPtrTy->getPointeeType(); 2496 2497 // Never warn about void type pointers. This can be used to suppress 2498 // false positives. 2499 if (PointeeTy->isVoidType()) 2500 continue; 2501 2502 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 2503 // actually comparing the expressions for equality. Because computing the 2504 // expression IDs can be expensive, we only do this if the diagnostic is 2505 // enabled. 2506 if (SizeOfArg && 2507 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 2508 SizeOfArg->getExprLoc())) { 2509 // We only compute IDs for expressions if the warning is enabled, and 2510 // cache the sizeof arg's ID. 2511 if (SizeOfArgID == llvm::FoldingSetNodeID()) 2512 SizeOfArg->Profile(SizeOfArgID, Context, true); 2513 llvm::FoldingSetNodeID DestID; 2514 Dest->Profile(DestID, Context, true); 2515 if (DestID == SizeOfArgID) { 2516 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 2517 // over sizeof(src) as well. 2518 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 2519 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 2520 if (UnaryOp->getOpcode() == UO_AddrOf) 2521 ActionIdx = 1; // If its an address-of operator, just remove it. 2522 if (Context.getTypeSize(PointeeTy) == Context.getCharWidth()) 2523 ActionIdx = 2; // If the pointee's size is sizeof(char), 2524 // suggest an explicit length. 2525 unsigned DestSrcSelect = 2526 (BId == Builtin::BIstrndup ? 1 : ArgIdx); 2527 DiagRuntimeBehavior(SizeOfArg->getExprLoc(), Dest, 2528 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 2529 << FnName << DestSrcSelect << ActionIdx 2530 << Dest->getSourceRange() 2531 << SizeOfArg->getSourceRange()); 2532 break; 2533 } 2534 } 2535 2536 // Also check for cases where the sizeof argument is the exact same 2537 // type as the memory argument, and where it points to a user-defined 2538 // record type. 2539 if (SizeOfArgTy != QualType()) { 2540 if (PointeeTy->isRecordType() && 2541 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 2542 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 2543 PDiag(diag::warn_sizeof_pointer_type_memaccess) 2544 << FnName << SizeOfArgTy << ArgIdx 2545 << PointeeTy << Dest->getSourceRange() 2546 << LenExpr->getSourceRange()); 2547 break; 2548 } 2549 } 2550 2551 // Always complain about dynamic classes. 2552 if (isDynamicClassType(PointeeTy)) { 2553 2554 unsigned OperationType = 0; 2555 // "overwritten" if we're warning about the destination for any call 2556 // but memcmp; otherwise a verb appropriate to the call. 2557 if (ArgIdx != 0 || BId == Builtin::BImemcmp) { 2558 if (BId == Builtin::BImemcpy) 2559 OperationType = 1; 2560 else if(BId == Builtin::BImemmove) 2561 OperationType = 2; 2562 else if (BId == Builtin::BImemcmp) 2563 OperationType = 3; 2564 } 2565 2566 DiagRuntimeBehavior( 2567 Dest->getExprLoc(), Dest, 2568 PDiag(diag::warn_dyn_class_memaccess) 2569 << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx) 2570 << FnName << PointeeTy 2571 << OperationType 2572 << Call->getCallee()->getSourceRange()); 2573 } else if (PointeeTy.hasNonTrivialObjCLifetime() && 2574 BId != Builtin::BImemset) 2575 DiagRuntimeBehavior( 2576 Dest->getExprLoc(), Dest, 2577 PDiag(diag::warn_arc_object_memaccess) 2578 << ArgIdx << FnName << PointeeTy 2579 << Call->getCallee()->getSourceRange()); 2580 else 2581 continue; 2582 2583 DiagRuntimeBehavior( 2584 Dest->getExprLoc(), Dest, 2585 PDiag(diag::note_bad_memaccess_silence) 2586 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 2587 break; 2588 } 2589 } 2590} 2591 2592// A little helper routine: ignore addition and subtraction of integer literals. 2593// This intentionally does not ignore all integer constant expressions because 2594// we don't want to remove sizeof(). 2595static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 2596 Ex = Ex->IgnoreParenCasts(); 2597 2598 for (;;) { 2599 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 2600 if (!BO || !BO->isAdditiveOp()) 2601 break; 2602 2603 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 2604 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 2605 2606 if (isa<IntegerLiteral>(RHS)) 2607 Ex = LHS; 2608 else if (isa<IntegerLiteral>(LHS)) 2609 Ex = RHS; 2610 else 2611 break; 2612 } 2613 2614 return Ex; 2615} 2616 2617// Warn if the user has made the 'size' argument to strlcpy or strlcat 2618// be the size of the source, instead of the destination. 2619void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 2620 IdentifierInfo *FnName) { 2621 2622 // Don't crash if the user has the wrong number of arguments 2623 if (Call->getNumArgs() != 3) 2624 return; 2625 2626 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 2627 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 2628 const Expr *CompareWithSrc = NULL; 2629 2630 // Look for 'strlcpy(dst, x, sizeof(x))' 2631 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 2632 CompareWithSrc = Ex; 2633 else { 2634 // Look for 'strlcpy(dst, x, strlen(x))' 2635 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 2636 if (SizeCall->isBuiltinCall() == Builtin::BIstrlen 2637 && SizeCall->getNumArgs() == 1) 2638 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 2639 } 2640 } 2641 2642 if (!CompareWithSrc) 2643 return; 2644 2645 // Determine if the argument to sizeof/strlen is equal to the source 2646 // argument. In principle there's all kinds of things you could do 2647 // here, for instance creating an == expression and evaluating it with 2648 // EvaluateAsBooleanCondition, but this uses a more direct technique: 2649 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 2650 if (!SrcArgDRE) 2651 return; 2652 2653 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 2654 if (!CompareWithSrcDRE || 2655 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 2656 return; 2657 2658 const Expr *OriginalSizeArg = Call->getArg(2); 2659 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 2660 << OriginalSizeArg->getSourceRange() << FnName; 2661 2662 // Output a FIXIT hint if the destination is an array (rather than a 2663 // pointer to an array). This could be enhanced to handle some 2664 // pointers if we know the actual size, like if DstArg is 'array+2' 2665 // we could say 'sizeof(array)-2'. 2666 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 2667 QualType DstArgTy = DstArg->getType(); 2668 2669 // Only handle constant-sized or VLAs, but not flexible members. 2670 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 2671 // Only issue the FIXIT for arrays of size > 1. 2672 if (CAT->getSize().getSExtValue() <= 1) 2673 return; 2674 } else if (!DstArgTy->isVariableArrayType()) { 2675 return; 2676 } 2677 2678 SmallString<128> sizeString; 2679 llvm::raw_svector_ostream OS(sizeString); 2680 OS << "sizeof("; 2681 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 2682 OS << ")"; 2683 2684 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 2685 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 2686 OS.str()); 2687} 2688 2689/// Check if two expressions refer to the same declaration. 2690static bool referToTheSameDecl(const Expr *E1, const Expr *E2) { 2691 if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1)) 2692 if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2)) 2693 return D1->getDecl() == D2->getDecl(); 2694 return false; 2695} 2696 2697static const Expr *getStrlenExprArg(const Expr *E) { 2698 if (const CallExpr *CE = dyn_cast<CallExpr>(E)) { 2699 const FunctionDecl *FD = CE->getDirectCallee(); 2700 if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen) 2701 return 0; 2702 return CE->getArg(0)->IgnoreParenCasts(); 2703 } 2704 return 0; 2705} 2706 2707// Warn on anti-patterns as the 'size' argument to strncat. 2708// The correct size argument should look like following: 2709// strncat(dst, src, sizeof(dst) - strlen(dest) - 1); 2710void Sema::CheckStrncatArguments(const CallExpr *CE, 2711 IdentifierInfo *FnName) { 2712 // Don't crash if the user has the wrong number of arguments. 2713 if (CE->getNumArgs() < 3) 2714 return; 2715 const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts(); 2716 const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts(); 2717 const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts(); 2718 2719 // Identify common expressions, which are wrongly used as the size argument 2720 // to strncat and may lead to buffer overflows. 2721 unsigned PatternType = 0; 2722 if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) { 2723 // - sizeof(dst) 2724 if (referToTheSameDecl(SizeOfArg, DstArg)) 2725 PatternType = 1; 2726 // - sizeof(src) 2727 else if (referToTheSameDecl(SizeOfArg, SrcArg)) 2728 PatternType = 2; 2729 } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) { 2730 if (BE->getOpcode() == BO_Sub) { 2731 const Expr *L = BE->getLHS()->IgnoreParenCasts(); 2732 const Expr *R = BE->getRHS()->IgnoreParenCasts(); 2733 // - sizeof(dst) - strlen(dst) 2734 if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) && 2735 referToTheSameDecl(DstArg, getStrlenExprArg(R))) 2736 PatternType = 1; 2737 // - sizeof(src) - (anything) 2738 else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L))) 2739 PatternType = 2; 2740 } 2741 } 2742 2743 if (PatternType == 0) 2744 return; 2745 2746 // Generate the diagnostic. 2747 SourceLocation SL = LenArg->getLocStart(); 2748 SourceRange SR = LenArg->getSourceRange(); 2749 SourceManager &SM = PP.getSourceManager(); 2750 2751 // If the function is defined as a builtin macro, do not show macro expansion. 2752 if (SM.isMacroArgExpansion(SL)) { 2753 SL = SM.getSpellingLoc(SL); 2754 SR = SourceRange(SM.getSpellingLoc(SR.getBegin()), 2755 SM.getSpellingLoc(SR.getEnd())); 2756 } 2757 2758 if (PatternType == 1) 2759 Diag(SL, diag::warn_strncat_large_size) << SR; 2760 else 2761 Diag(SL, diag::warn_strncat_src_size) << SR; 2762 2763 // Output a FIXIT hint if the destination is an array (rather than a 2764 // pointer to an array). This could be enhanced to handle some 2765 // pointers if we know the actual size, like if DstArg is 'array+2' 2766 // we could say 'sizeof(array)-2'. 2767 QualType DstArgTy = DstArg->getType(); 2768 2769 // Only handle constant-sized or VLAs, but not flexible members. 2770 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 2771 // Only issue the FIXIT for arrays of size > 1. 2772 if (CAT->getSize().getSExtValue() <= 1) 2773 return; 2774 } else if (!DstArgTy->isVariableArrayType()) { 2775 return; 2776 } 2777 2778 SmallString<128> sizeString; 2779 llvm::raw_svector_ostream OS(sizeString); 2780 OS << "sizeof("; 2781 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 2782 OS << ") - "; 2783 OS << "strlen("; 2784 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 2785 OS << ") - 1"; 2786 2787 Diag(SL, diag::note_strncat_wrong_size) 2788 << FixItHint::CreateReplacement(SR, OS.str()); 2789} 2790 2791//===--- CHECK: Return Address of Stack Variable --------------------------===// 2792 2793static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars); 2794static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars); 2795 2796/// CheckReturnStackAddr - Check if a return statement returns the address 2797/// of a stack variable. 2798void 2799Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 2800 SourceLocation ReturnLoc) { 2801 2802 Expr *stackE = 0; 2803 SmallVector<DeclRefExpr *, 8> refVars; 2804 2805 // Perform checking for returned stack addresses, local blocks, 2806 // label addresses or references to temporaries. 2807 if (lhsType->isPointerType() || 2808 (!getLangOptions().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 2809 stackE = EvalAddr(RetValExp, refVars); 2810 } else if (lhsType->isReferenceType()) { 2811 stackE = EvalVal(RetValExp, refVars); 2812 } 2813 2814 if (stackE == 0) 2815 return; // Nothing suspicious was found. 2816 2817 SourceLocation diagLoc; 2818 SourceRange diagRange; 2819 if (refVars.empty()) { 2820 diagLoc = stackE->getLocStart(); 2821 diagRange = stackE->getSourceRange(); 2822 } else { 2823 // We followed through a reference variable. 'stackE' contains the 2824 // problematic expression but we will warn at the return statement pointing 2825 // at the reference variable. We will later display the "trail" of 2826 // reference variables using notes. 2827 diagLoc = refVars[0]->getLocStart(); 2828 diagRange = refVars[0]->getSourceRange(); 2829 } 2830 2831 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 2832 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 2833 : diag::warn_ret_stack_addr) 2834 << DR->getDecl()->getDeclName() << diagRange; 2835 } else if (isa<BlockExpr>(stackE)) { // local block. 2836 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 2837 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 2838 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 2839 } else { // local temporary. 2840 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 2841 : diag::warn_ret_local_temp_addr) 2842 << diagRange; 2843 } 2844 2845 // Display the "trail" of reference variables that we followed until we 2846 // found the problematic expression using notes. 2847 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 2848 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 2849 // If this var binds to another reference var, show the range of the next 2850 // var, otherwise the var binds to the problematic expression, in which case 2851 // show the range of the expression. 2852 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 2853 : stackE->getSourceRange(); 2854 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 2855 << VD->getDeclName() << range; 2856 } 2857} 2858 2859/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 2860/// check if the expression in a return statement evaluates to an address 2861/// to a location on the stack, a local block, an address of a label, or a 2862/// reference to local temporary. The recursion is used to traverse the 2863/// AST of the return expression, with recursion backtracking when we 2864/// encounter a subexpression that (1) clearly does not lead to one of the 2865/// above problematic expressions (2) is something we cannot determine leads to 2866/// a problematic expression based on such local checking. 2867/// 2868/// Both EvalAddr and EvalVal follow through reference variables to evaluate 2869/// the expression that they point to. Such variables are added to the 2870/// 'refVars' vector so that we know what the reference variable "trail" was. 2871/// 2872/// EvalAddr processes expressions that are pointers that are used as 2873/// references (and not L-values). EvalVal handles all other values. 2874/// At the base case of the recursion is a check for the above problematic 2875/// expressions. 2876/// 2877/// This implementation handles: 2878/// 2879/// * pointer-to-pointer casts 2880/// * implicit conversions from array references to pointers 2881/// * taking the address of fields 2882/// * arbitrary interplay between "&" and "*" operators 2883/// * pointer arithmetic from an address of a stack variable 2884/// * taking the address of an array element where the array is on the stack 2885static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 2886 if (E->isTypeDependent()) 2887 return NULL; 2888 2889 // We should only be called for evaluating pointer expressions. 2890 assert((E->getType()->isAnyPointerType() || 2891 E->getType()->isBlockPointerType() || 2892 E->getType()->isObjCQualifiedIdType()) && 2893 "EvalAddr only works on pointers"); 2894 2895 E = E->IgnoreParens(); 2896 2897 // Our "symbolic interpreter" is just a dispatch off the currently 2898 // viewed AST node. We then recursively traverse the AST by calling 2899 // EvalAddr and EvalVal appropriately. 2900 switch (E->getStmtClass()) { 2901 case Stmt::DeclRefExprClass: { 2902 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2903 2904 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2905 // If this is a reference variable, follow through to the expression that 2906 // it points to. 2907 if (V->hasLocalStorage() && 2908 V->getType()->isReferenceType() && V->hasInit()) { 2909 // Add the reference variable to the "trail". 2910 refVars.push_back(DR); 2911 return EvalAddr(V->getInit(), refVars); 2912 } 2913 2914 return NULL; 2915 } 2916 2917 case Stmt::UnaryOperatorClass: { 2918 // The only unary operator that make sense to handle here 2919 // is AddrOf. All others don't make sense as pointers. 2920 UnaryOperator *U = cast<UnaryOperator>(E); 2921 2922 if (U->getOpcode() == UO_AddrOf) 2923 return EvalVal(U->getSubExpr(), refVars); 2924 else 2925 return NULL; 2926 } 2927 2928 case Stmt::BinaryOperatorClass: { 2929 // Handle pointer arithmetic. All other binary operators are not valid 2930 // in this context. 2931 BinaryOperator *B = cast<BinaryOperator>(E); 2932 BinaryOperatorKind op = B->getOpcode(); 2933 2934 if (op != BO_Add && op != BO_Sub) 2935 return NULL; 2936 2937 Expr *Base = B->getLHS(); 2938 2939 // Determine which argument is the real pointer base. It could be 2940 // the RHS argument instead of the LHS. 2941 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 2942 2943 assert (Base->getType()->isPointerType()); 2944 return EvalAddr(Base, refVars); 2945 } 2946 2947 // For conditional operators we need to see if either the LHS or RHS are 2948 // valid DeclRefExpr*s. If one of them is valid, we return it. 2949 case Stmt::ConditionalOperatorClass: { 2950 ConditionalOperator *C = cast<ConditionalOperator>(E); 2951 2952 // Handle the GNU extension for missing LHS. 2953 if (Expr *lhsExpr = C->getLHS()) { 2954 // In C++, we can have a throw-expression, which has 'void' type. 2955 if (!lhsExpr->getType()->isVoidType()) 2956 if (Expr* LHS = EvalAddr(lhsExpr, refVars)) 2957 return LHS; 2958 } 2959 2960 // In C++, we can have a throw-expression, which has 'void' type. 2961 if (C->getRHS()->getType()->isVoidType()) 2962 return NULL; 2963 2964 return EvalAddr(C->getRHS(), refVars); 2965 } 2966 2967 case Stmt::BlockExprClass: 2968 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 2969 return E; // local block. 2970 return NULL; 2971 2972 case Stmt::AddrLabelExprClass: 2973 return E; // address of label. 2974 2975 case Stmt::ExprWithCleanupsClass: 2976 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars); 2977 2978 // For casts, we need to handle conversions from arrays to 2979 // pointer values, and pointer-to-pointer conversions. 2980 case Stmt::ImplicitCastExprClass: 2981 case Stmt::CStyleCastExprClass: 2982 case Stmt::CXXFunctionalCastExprClass: 2983 case Stmt::ObjCBridgedCastExprClass: { 2984 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 2985 QualType T = SubExpr->getType(); 2986 2987 if (SubExpr->getType()->isPointerType() || 2988 SubExpr->getType()->isBlockPointerType() || 2989 SubExpr->getType()->isObjCQualifiedIdType()) 2990 return EvalAddr(SubExpr, refVars); 2991 else if (T->isArrayType()) 2992 return EvalVal(SubExpr, refVars); 2993 else 2994 return 0; 2995 } 2996 2997 // C++ casts. For dynamic casts, static casts, and const casts, we 2998 // are always converting from a pointer-to-pointer, so we just blow 2999 // through the cast. In the case the dynamic cast doesn't fail (and 3000 // return NULL), we take the conservative route and report cases 3001 // where we return the address of a stack variable. For Reinterpre 3002 // FIXME: The comment about is wrong; we're not always converting 3003 // from pointer to pointer. I'm guessing that this code should also 3004 // handle references to objects. 3005 case Stmt::CXXStaticCastExprClass: 3006 case Stmt::CXXDynamicCastExprClass: 3007 case Stmt::CXXConstCastExprClass: 3008 case Stmt::CXXReinterpretCastExprClass: { 3009 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 3010 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 3011 return EvalAddr(S, refVars); 3012 else 3013 return NULL; 3014 } 3015 3016 case Stmt::MaterializeTemporaryExprClass: 3017 if (Expr *Result = EvalAddr( 3018 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3019 refVars)) 3020 return Result; 3021 3022 return E; 3023 3024 // Everything else: we simply don't reason about them. 3025 default: 3026 return NULL; 3027 } 3028} 3029 3030 3031/// EvalVal - This function is complements EvalAddr in the mutual recursion. 3032/// See the comments for EvalAddr for more details. 3033static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 3034do { 3035 // We should only be called for evaluating non-pointer expressions, or 3036 // expressions with a pointer type that are not used as references but instead 3037 // are l-values (e.g., DeclRefExpr with a pointer type). 3038 3039 // Our "symbolic interpreter" is just a dispatch off the currently 3040 // viewed AST node. We then recursively traverse the AST by calling 3041 // EvalAddr and EvalVal appropriately. 3042 3043 E = E->IgnoreParens(); 3044 switch (E->getStmtClass()) { 3045 case Stmt::ImplicitCastExprClass: { 3046 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 3047 if (IE->getValueKind() == VK_LValue) { 3048 E = IE->getSubExpr(); 3049 continue; 3050 } 3051 return NULL; 3052 } 3053 3054 case Stmt::ExprWithCleanupsClass: 3055 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars); 3056 3057 case Stmt::DeclRefExprClass: { 3058 // When we hit a DeclRefExpr we are looking at code that refers to a 3059 // variable's name. If it's not a reference variable we check if it has 3060 // local storage within the function, and if so, return the expression. 3061 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3062 3063 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 3064 if (V->hasLocalStorage()) { 3065 if (!V->getType()->isReferenceType()) 3066 return DR; 3067 3068 // Reference variable, follow through to the expression that 3069 // it points to. 3070 if (V->hasInit()) { 3071 // Add the reference variable to the "trail". 3072 refVars.push_back(DR); 3073 return EvalVal(V->getInit(), refVars); 3074 } 3075 } 3076 3077 return NULL; 3078 } 3079 3080 case Stmt::UnaryOperatorClass: { 3081 // The only unary operator that make sense to handle here 3082 // is Deref. All others don't resolve to a "name." This includes 3083 // handling all sorts of rvalues passed to a unary operator. 3084 UnaryOperator *U = cast<UnaryOperator>(E); 3085 3086 if (U->getOpcode() == UO_Deref) 3087 return EvalAddr(U->getSubExpr(), refVars); 3088 3089 return NULL; 3090 } 3091 3092 case Stmt::ArraySubscriptExprClass: { 3093 // Array subscripts are potential references to data on the stack. We 3094 // retrieve the DeclRefExpr* for the array variable if it indeed 3095 // has local storage. 3096 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars); 3097 } 3098 3099 case Stmt::ConditionalOperatorClass: { 3100 // For conditional operators we need to see if either the LHS or RHS are 3101 // non-NULL Expr's. If one is non-NULL, we return it. 3102 ConditionalOperator *C = cast<ConditionalOperator>(E); 3103 3104 // Handle the GNU extension for missing LHS. 3105 if (Expr *lhsExpr = C->getLHS()) 3106 if (Expr *LHS = EvalVal(lhsExpr, refVars)) 3107 return LHS; 3108 3109 return EvalVal(C->getRHS(), refVars); 3110 } 3111 3112 // Accesses to members are potential references to data on the stack. 3113 case Stmt::MemberExprClass: { 3114 MemberExpr *M = cast<MemberExpr>(E); 3115 3116 // Check for indirect access. We only want direct field accesses. 3117 if (M->isArrow()) 3118 return NULL; 3119 3120 // Check whether the member type is itself a reference, in which case 3121 // we're not going to refer to the member, but to what the member refers to. 3122 if (M->getMemberDecl()->getType()->isReferenceType()) 3123 return NULL; 3124 3125 return EvalVal(M->getBase(), refVars); 3126 } 3127 3128 case Stmt::MaterializeTemporaryExprClass: 3129 if (Expr *Result = EvalVal( 3130 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3131 refVars)) 3132 return Result; 3133 3134 return E; 3135 3136 default: 3137 // Check that we don't return or take the address of a reference to a 3138 // temporary. This is only useful in C++. 3139 if (!E->isTypeDependent() && E->isRValue()) 3140 return E; 3141 3142 // Everything else: we simply don't reason about them. 3143 return NULL; 3144 } 3145} while (true); 3146} 3147 3148//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 3149 3150/// Check for comparisons of floating point operands using != and ==. 3151/// Issue a warning if these are no self-comparisons, as they are not likely 3152/// to do what the programmer intended. 3153void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 3154 bool EmitWarning = true; 3155 3156 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 3157 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 3158 3159 // Special case: check for x == x (which is OK). 3160 // Do not emit warnings for such cases. 3161 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 3162 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 3163 if (DRL->getDecl() == DRR->getDecl()) 3164 EmitWarning = false; 3165 3166 3167 // Special case: check for comparisons against literals that can be exactly 3168 // represented by APFloat. In such cases, do not emit a warning. This 3169 // is a heuristic: often comparison against such literals are used to 3170 // detect if a value in a variable has not changed. This clearly can 3171 // lead to false negatives. 3172 if (EmitWarning) { 3173 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 3174 if (FLL->isExact()) 3175 EmitWarning = false; 3176 } else 3177 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 3178 if (FLR->isExact()) 3179 EmitWarning = false; 3180 } 3181 } 3182 3183 // Check for comparisons with builtin types. 3184 if (EmitWarning) 3185 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 3186 if (CL->isBuiltinCall()) 3187 EmitWarning = false; 3188 3189 if (EmitWarning) 3190 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 3191 if (CR->isBuiltinCall()) 3192 EmitWarning = false; 3193 3194 // Emit the diagnostic. 3195 if (EmitWarning) 3196 Diag(Loc, diag::warn_floatingpoint_eq) 3197 << LHS->getSourceRange() << RHS->getSourceRange(); 3198} 3199 3200//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 3201//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 3202 3203namespace { 3204 3205/// Structure recording the 'active' range of an integer-valued 3206/// expression. 3207struct IntRange { 3208 /// The number of bits active in the int. 3209 unsigned Width; 3210 3211 /// True if the int is known not to have negative values. 3212 bool NonNegative; 3213 3214 IntRange(unsigned Width, bool NonNegative) 3215 : Width(Width), NonNegative(NonNegative) 3216 {} 3217 3218 /// Returns the range of the bool type. 3219 static IntRange forBoolType() { 3220 return IntRange(1, true); 3221 } 3222 3223 /// Returns the range of an opaque value of the given integral type. 3224 static IntRange forValueOfType(ASTContext &C, QualType T) { 3225 return forValueOfCanonicalType(C, 3226 T->getCanonicalTypeInternal().getTypePtr()); 3227 } 3228 3229 /// Returns the range of an opaque value of a canonical integral type. 3230 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 3231 assert(T->isCanonicalUnqualified()); 3232 3233 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3234 T = VT->getElementType().getTypePtr(); 3235 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3236 T = CT->getElementType().getTypePtr(); 3237 3238 // For enum types, use the known bit width of the enumerators. 3239 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 3240 EnumDecl *Enum = ET->getDecl(); 3241 if (!Enum->isCompleteDefinition()) 3242 return IntRange(C.getIntWidth(QualType(T, 0)), false); 3243 3244 unsigned NumPositive = Enum->getNumPositiveBits(); 3245 unsigned NumNegative = Enum->getNumNegativeBits(); 3246 3247 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 3248 } 3249 3250 const BuiltinType *BT = cast<BuiltinType>(T); 3251 assert(BT->isInteger()); 3252 3253 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3254 } 3255 3256 /// Returns the "target" range of a canonical integral type, i.e. 3257 /// the range of values expressible in the type. 3258 /// 3259 /// This matches forValueOfCanonicalType except that enums have the 3260 /// full range of their type, not the range of their enumerators. 3261 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 3262 assert(T->isCanonicalUnqualified()); 3263 3264 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3265 T = VT->getElementType().getTypePtr(); 3266 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3267 T = CT->getElementType().getTypePtr(); 3268 if (const EnumType *ET = dyn_cast<EnumType>(T)) 3269 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 3270 3271 const BuiltinType *BT = cast<BuiltinType>(T); 3272 assert(BT->isInteger()); 3273 3274 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3275 } 3276 3277 /// Returns the supremum of two ranges: i.e. their conservative merge. 3278 static IntRange join(IntRange L, IntRange R) { 3279 return IntRange(std::max(L.Width, R.Width), 3280 L.NonNegative && R.NonNegative); 3281 } 3282 3283 /// Returns the infinum of two ranges: i.e. their aggressive merge. 3284 static IntRange meet(IntRange L, IntRange R) { 3285 return IntRange(std::min(L.Width, R.Width), 3286 L.NonNegative || R.NonNegative); 3287 } 3288}; 3289 3290static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, 3291 unsigned MaxWidth) { 3292 if (value.isSigned() && value.isNegative()) 3293 return IntRange(value.getMinSignedBits(), false); 3294 3295 if (value.getBitWidth() > MaxWidth) 3296 value = value.trunc(MaxWidth); 3297 3298 // isNonNegative() just checks the sign bit without considering 3299 // signedness. 3300 return IntRange(value.getActiveBits(), true); 3301} 3302 3303static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 3304 unsigned MaxWidth) { 3305 if (result.isInt()) 3306 return GetValueRange(C, result.getInt(), MaxWidth); 3307 3308 if (result.isVector()) { 3309 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 3310 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 3311 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 3312 R = IntRange::join(R, El); 3313 } 3314 return R; 3315 } 3316 3317 if (result.isComplexInt()) { 3318 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 3319 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 3320 return IntRange::join(R, I); 3321 } 3322 3323 // This can happen with lossless casts to intptr_t of "based" lvalues. 3324 // Assume it might use arbitrary bits. 3325 // FIXME: The only reason we need to pass the type in here is to get 3326 // the sign right on this one case. It would be nice if APValue 3327 // preserved this. 3328 assert(result.isLValue() || result.isAddrLabelDiff()); 3329 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 3330} 3331 3332/// Pseudo-evaluate the given integer expression, estimating the 3333/// range of values it might take. 3334/// 3335/// \param MaxWidth - the width to which the value will be truncated 3336static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 3337 E = E->IgnoreParens(); 3338 3339 // Try a full evaluation first. 3340 Expr::EvalResult result; 3341 if (E->EvaluateAsRValue(result, C)) 3342 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 3343 3344 // I think we only want to look through implicit casts here; if the 3345 // user has an explicit widening cast, we should treat the value as 3346 // being of the new, wider type. 3347 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 3348 if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue) 3349 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 3350 3351 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 3352 3353 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 3354 3355 // Assume that non-integer casts can span the full range of the type. 3356 if (!isIntegerCast) 3357 return OutputTypeRange; 3358 3359 IntRange SubRange 3360 = GetExprRange(C, CE->getSubExpr(), 3361 std::min(MaxWidth, OutputTypeRange.Width)); 3362 3363 // Bail out if the subexpr's range is as wide as the cast type. 3364 if (SubRange.Width >= OutputTypeRange.Width) 3365 return OutputTypeRange; 3366 3367 // Otherwise, we take the smaller width, and we're non-negative if 3368 // either the output type or the subexpr is. 3369 return IntRange(SubRange.Width, 3370 SubRange.NonNegative || OutputTypeRange.NonNegative); 3371 } 3372 3373 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 3374 // If we can fold the condition, just take that operand. 3375 bool CondResult; 3376 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 3377 return GetExprRange(C, CondResult ? CO->getTrueExpr() 3378 : CO->getFalseExpr(), 3379 MaxWidth); 3380 3381 // Otherwise, conservatively merge. 3382 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 3383 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 3384 return IntRange::join(L, R); 3385 } 3386 3387 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 3388 switch (BO->getOpcode()) { 3389 3390 // Boolean-valued operations are single-bit and positive. 3391 case BO_LAnd: 3392 case BO_LOr: 3393 case BO_LT: 3394 case BO_GT: 3395 case BO_LE: 3396 case BO_GE: 3397 case BO_EQ: 3398 case BO_NE: 3399 return IntRange::forBoolType(); 3400 3401 // The type of the assignments is the type of the LHS, so the RHS 3402 // is not necessarily the same type. 3403 case BO_MulAssign: 3404 case BO_DivAssign: 3405 case BO_RemAssign: 3406 case BO_AddAssign: 3407 case BO_SubAssign: 3408 case BO_XorAssign: 3409 case BO_OrAssign: 3410 // TODO: bitfields? 3411 return IntRange::forValueOfType(C, E->getType()); 3412 3413 // Simple assignments just pass through the RHS, which will have 3414 // been coerced to the LHS type. 3415 case BO_Assign: 3416 // TODO: bitfields? 3417 return GetExprRange(C, BO->getRHS(), MaxWidth); 3418 3419 // Operations with opaque sources are black-listed. 3420 case BO_PtrMemD: 3421 case BO_PtrMemI: 3422 return IntRange::forValueOfType(C, E->getType()); 3423 3424 // Bitwise-and uses the *infinum* of the two source ranges. 3425 case BO_And: 3426 case BO_AndAssign: 3427 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 3428 GetExprRange(C, BO->getRHS(), MaxWidth)); 3429 3430 // Left shift gets black-listed based on a judgement call. 3431 case BO_Shl: 3432 // ...except that we want to treat '1 << (blah)' as logically 3433 // positive. It's an important idiom. 3434 if (IntegerLiteral *I 3435 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 3436 if (I->getValue() == 1) { 3437 IntRange R = IntRange::forValueOfType(C, E->getType()); 3438 return IntRange(R.Width, /*NonNegative*/ true); 3439 } 3440 } 3441 // fallthrough 3442 3443 case BO_ShlAssign: 3444 return IntRange::forValueOfType(C, E->getType()); 3445 3446 // Right shift by a constant can narrow its left argument. 3447 case BO_Shr: 3448 case BO_ShrAssign: { 3449 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3450 3451 // If the shift amount is a positive constant, drop the width by 3452 // that much. 3453 llvm::APSInt shift; 3454 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 3455 shift.isNonNegative()) { 3456 unsigned zext = shift.getZExtValue(); 3457 if (zext >= L.Width) 3458 L.Width = (L.NonNegative ? 0 : 1); 3459 else 3460 L.Width -= zext; 3461 } 3462 3463 return L; 3464 } 3465 3466 // Comma acts as its right operand. 3467 case BO_Comma: 3468 return GetExprRange(C, BO->getRHS(), MaxWidth); 3469 3470 // Black-list pointer subtractions. 3471 case BO_Sub: 3472 if (BO->getLHS()->getType()->isPointerType()) 3473 return IntRange::forValueOfType(C, E->getType()); 3474 break; 3475 3476 // The width of a division result is mostly determined by the size 3477 // of the LHS. 3478 case BO_Div: { 3479 // Don't 'pre-truncate' the operands. 3480 unsigned opWidth = C.getIntWidth(E->getType()); 3481 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3482 3483 // If the divisor is constant, use that. 3484 llvm::APSInt divisor; 3485 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 3486 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 3487 if (log2 >= L.Width) 3488 L.Width = (L.NonNegative ? 0 : 1); 3489 else 3490 L.Width = std::min(L.Width - log2, MaxWidth); 3491 return L; 3492 } 3493 3494 // Otherwise, just use the LHS's width. 3495 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3496 return IntRange(L.Width, L.NonNegative && R.NonNegative); 3497 } 3498 3499 // The result of a remainder can't be larger than the result of 3500 // either side. 3501 case BO_Rem: { 3502 // Don't 'pre-truncate' the operands. 3503 unsigned opWidth = C.getIntWidth(E->getType()); 3504 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3505 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3506 3507 IntRange meet = IntRange::meet(L, R); 3508 meet.Width = std::min(meet.Width, MaxWidth); 3509 return meet; 3510 } 3511 3512 // The default behavior is okay for these. 3513 case BO_Mul: 3514 case BO_Add: 3515 case BO_Xor: 3516 case BO_Or: 3517 break; 3518 } 3519 3520 // The default case is to treat the operation as if it were closed 3521 // on the narrowest type that encompasses both operands. 3522 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3523 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 3524 return IntRange::join(L, R); 3525 } 3526 3527 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 3528 switch (UO->getOpcode()) { 3529 // Boolean-valued operations are white-listed. 3530 case UO_LNot: 3531 return IntRange::forBoolType(); 3532 3533 // Operations with opaque sources are black-listed. 3534 case UO_Deref: 3535 case UO_AddrOf: // should be impossible 3536 return IntRange::forValueOfType(C, E->getType()); 3537 3538 default: 3539 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 3540 } 3541 } 3542 3543 if (dyn_cast<OffsetOfExpr>(E)) { 3544 IntRange::forValueOfType(C, E->getType()); 3545 } 3546 3547 if (FieldDecl *BitField = E->getBitField()) 3548 return IntRange(BitField->getBitWidthValue(C), 3549 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 3550 3551 return IntRange::forValueOfType(C, E->getType()); 3552} 3553 3554static IntRange GetExprRange(ASTContext &C, Expr *E) { 3555 return GetExprRange(C, E, C.getIntWidth(E->getType())); 3556} 3557 3558/// Checks whether the given value, which currently has the given 3559/// source semantics, has the same value when coerced through the 3560/// target semantics. 3561static bool IsSameFloatAfterCast(const llvm::APFloat &value, 3562 const llvm::fltSemantics &Src, 3563 const llvm::fltSemantics &Tgt) { 3564 llvm::APFloat truncated = value; 3565 3566 bool ignored; 3567 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 3568 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 3569 3570 return truncated.bitwiseIsEqual(value); 3571} 3572 3573/// Checks whether the given value, which currently has the given 3574/// source semantics, has the same value when coerced through the 3575/// target semantics. 3576/// 3577/// The value might be a vector of floats (or a complex number). 3578static bool IsSameFloatAfterCast(const APValue &value, 3579 const llvm::fltSemantics &Src, 3580 const llvm::fltSemantics &Tgt) { 3581 if (value.isFloat()) 3582 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 3583 3584 if (value.isVector()) { 3585 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 3586 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 3587 return false; 3588 return true; 3589 } 3590 3591 assert(value.isComplexFloat()); 3592 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 3593 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 3594} 3595 3596static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 3597 3598static bool IsZero(Sema &S, Expr *E) { 3599 // Suppress cases where we are comparing against an enum constant. 3600 if (const DeclRefExpr *DR = 3601 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 3602 if (isa<EnumConstantDecl>(DR->getDecl())) 3603 return false; 3604 3605 // Suppress cases where the '0' value is expanded from a macro. 3606 if (E->getLocStart().isMacroID()) 3607 return false; 3608 3609 llvm::APSInt Value; 3610 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 3611} 3612 3613static bool HasEnumType(Expr *E) { 3614 // Strip off implicit integral promotions. 3615 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 3616 if (ICE->getCastKind() != CK_IntegralCast && 3617 ICE->getCastKind() != CK_NoOp) 3618 break; 3619 E = ICE->getSubExpr(); 3620 } 3621 3622 return E->getType()->isEnumeralType(); 3623} 3624 3625static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 3626 BinaryOperatorKind op = E->getOpcode(); 3627 if (E->isValueDependent()) 3628 return; 3629 3630 if (op == BO_LT && IsZero(S, E->getRHS())) { 3631 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3632 << "< 0" << "false" << HasEnumType(E->getLHS()) 3633 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3634 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 3635 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3636 << ">= 0" << "true" << HasEnumType(E->getLHS()) 3637 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3638 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 3639 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3640 << "0 >" << "false" << HasEnumType(E->getRHS()) 3641 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3642 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 3643 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3644 << "0 <=" << "true" << HasEnumType(E->getRHS()) 3645 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3646 } 3647} 3648 3649/// Analyze the operands of the given comparison. Implements the 3650/// fallback case from AnalyzeComparison. 3651static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 3652 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3653 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3654} 3655 3656/// \brief Implements -Wsign-compare. 3657/// 3658/// \param E the binary operator to check for warnings 3659static void AnalyzeComparison(Sema &S, BinaryOperator *E) { 3660 // The type the comparison is being performed in. 3661 QualType T = E->getLHS()->getType(); 3662 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 3663 && "comparison with mismatched types"); 3664 3665 // We don't do anything special if this isn't an unsigned integral 3666 // comparison: we're only interested in integral comparisons, and 3667 // signed comparisons only happen in cases we don't care to warn about. 3668 // 3669 // We also don't care about value-dependent expressions or expressions 3670 // whose result is a constant. 3671 if (!T->hasUnsignedIntegerRepresentation() 3672 || E->isValueDependent() || E->isIntegerConstantExpr(S.Context)) 3673 return AnalyzeImpConvsInComparison(S, E); 3674 3675 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 3676 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 3677 3678 // Check to see if one of the (unmodified) operands is of different 3679 // signedness. 3680 Expr *signedOperand, *unsignedOperand; 3681 if (LHS->getType()->hasSignedIntegerRepresentation()) { 3682 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 3683 "unsigned comparison between two signed integer expressions?"); 3684 signedOperand = LHS; 3685 unsignedOperand = RHS; 3686 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 3687 signedOperand = RHS; 3688 unsignedOperand = LHS; 3689 } else { 3690 CheckTrivialUnsignedComparison(S, E); 3691 return AnalyzeImpConvsInComparison(S, E); 3692 } 3693 3694 // Otherwise, calculate the effective range of the signed operand. 3695 IntRange signedRange = GetExprRange(S.Context, signedOperand); 3696 3697 // Go ahead and analyze implicit conversions in the operands. Note 3698 // that we skip the implicit conversions on both sides. 3699 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 3700 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 3701 3702 // If the signed range is non-negative, -Wsign-compare won't fire, 3703 // but we should still check for comparisons which are always true 3704 // or false. 3705 if (signedRange.NonNegative) 3706 return CheckTrivialUnsignedComparison(S, E); 3707 3708 // For (in)equality comparisons, if the unsigned operand is a 3709 // constant which cannot collide with a overflowed signed operand, 3710 // then reinterpreting the signed operand as unsigned will not 3711 // change the result of the comparison. 3712 if (E->isEqualityOp()) { 3713 unsigned comparisonWidth = S.Context.getIntWidth(T); 3714 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 3715 3716 // We should never be unable to prove that the unsigned operand is 3717 // non-negative. 3718 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 3719 3720 if (unsignedRange.Width < comparisonWidth) 3721 return; 3722 } 3723 3724 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison) 3725 << LHS->getType() << RHS->getType() 3726 << LHS->getSourceRange() << RHS->getSourceRange(); 3727} 3728 3729/// Analyzes an attempt to assign the given value to a bitfield. 3730/// 3731/// Returns true if there was something fishy about the attempt. 3732static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 3733 SourceLocation InitLoc) { 3734 assert(Bitfield->isBitField()); 3735 if (Bitfield->isInvalidDecl()) 3736 return false; 3737 3738 // White-list bool bitfields. 3739 if (Bitfield->getType()->isBooleanType()) 3740 return false; 3741 3742 // Ignore value- or type-dependent expressions. 3743 if (Bitfield->getBitWidth()->isValueDependent() || 3744 Bitfield->getBitWidth()->isTypeDependent() || 3745 Init->isValueDependent() || 3746 Init->isTypeDependent()) 3747 return false; 3748 3749 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 3750 3751 llvm::APSInt Value; 3752 if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects)) 3753 return false; 3754 3755 unsigned OriginalWidth = Value.getBitWidth(); 3756 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 3757 3758 if (OriginalWidth <= FieldWidth) 3759 return false; 3760 3761 // Compute the value which the bitfield will contain. 3762 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 3763 TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType()); 3764 3765 // Check whether the stored value is equal to the original value. 3766 TruncatedValue = TruncatedValue.extend(OriginalWidth); 3767 if (Value == TruncatedValue) 3768 return false; 3769 3770 // Special-case bitfields of width 1: booleans are naturally 0/1, and 3771 // therefore don't strictly fit into a signed bitfield of width 1. 3772 if (FieldWidth == 1 && Value == 1) 3773 return false; 3774 3775 std::string PrettyValue = Value.toString(10); 3776 std::string PrettyTrunc = TruncatedValue.toString(10); 3777 3778 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 3779 << PrettyValue << PrettyTrunc << OriginalInit->getType() 3780 << Init->getSourceRange(); 3781 3782 return true; 3783} 3784 3785/// Analyze the given simple or compound assignment for warning-worthy 3786/// operations. 3787static void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 3788 // Just recurse on the LHS. 3789 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3790 3791 // We want to recurse on the RHS as normal unless we're assigning to 3792 // a bitfield. 3793 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 3794 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 3795 E->getOperatorLoc())) { 3796 // Recurse, ignoring any implicit conversions on the RHS. 3797 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 3798 E->getOperatorLoc()); 3799 } 3800 } 3801 3802 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3803} 3804 3805/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3806static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 3807 SourceLocation CContext, unsigned diag, 3808 bool pruneControlFlow = false) { 3809 if (pruneControlFlow) { 3810 S.DiagRuntimeBehavior(E->getExprLoc(), E, 3811 S.PDiag(diag) 3812 << SourceType << T << E->getSourceRange() 3813 << SourceRange(CContext)); 3814 return; 3815 } 3816 S.Diag(E->getExprLoc(), diag) 3817 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 3818} 3819 3820/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3821static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, 3822 SourceLocation CContext, unsigned diag, 3823 bool pruneControlFlow = false) { 3824 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow); 3825} 3826 3827/// Diagnose an implicit cast from a literal expression. Does not warn when the 3828/// cast wouldn't lose information. 3829void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 3830 SourceLocation CContext) { 3831 // Try to convert the literal exactly to an integer. If we can, don't warn. 3832 bool isExact = false; 3833 const llvm::APFloat &Value = FL->getValue(); 3834 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 3835 T->hasUnsignedIntegerRepresentation()); 3836 if (Value.convertToInteger(IntegerValue, 3837 llvm::APFloat::rmTowardZero, &isExact) 3838 == llvm::APFloat::opOK && isExact) 3839 return; 3840 3841 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 3842 << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext); 3843} 3844 3845std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 3846 if (!Range.Width) return "0"; 3847 3848 llvm::APSInt ValueInRange = Value; 3849 ValueInRange.setIsSigned(!Range.NonNegative); 3850 ValueInRange = ValueInRange.trunc(Range.Width); 3851 return ValueInRange.toString(10); 3852} 3853 3854void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 3855 SourceLocation CC, bool *ICContext = 0) { 3856 if (E->isTypeDependent() || E->isValueDependent()) return; 3857 3858 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 3859 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 3860 if (Source == Target) return; 3861 if (Target->isDependentType()) return; 3862 3863 // If the conversion context location is invalid don't complain. We also 3864 // don't want to emit a warning if the issue occurs from the expansion of 3865 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 3866 // delay this check as long as possible. Once we detect we are in that 3867 // scenario, we just return. 3868 if (CC.isInvalid()) 3869 return; 3870 3871 // Diagnose implicit casts to bool. 3872 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 3873 if (isa<StringLiteral>(E)) 3874 // Warn on string literal to bool. Checks for string literals in logical 3875 // expressions, for instances, assert(0 && "error here"), is prevented 3876 // by a check in AnalyzeImplicitConversions(). 3877 return DiagnoseImpCast(S, E, T, CC, 3878 diag::warn_impcast_string_literal_to_bool); 3879 if (Source->isFunctionType()) { 3880 // Warn on function to bool. Checks free functions and static member 3881 // functions. Weakly imported functions are excluded from the check, 3882 // since it's common to test their value to check whether the linker 3883 // found a definition for them. 3884 ValueDecl *D = 0; 3885 if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) { 3886 D = R->getDecl(); 3887 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 3888 D = M->getMemberDecl(); 3889 } 3890 3891 if (D && !D->isWeak()) { 3892 if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) { 3893 S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool) 3894 << F << E->getSourceRange() << SourceRange(CC); 3895 S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence) 3896 << FixItHint::CreateInsertion(E->getExprLoc(), "&"); 3897 QualType ReturnType; 3898 UnresolvedSet<4> NonTemplateOverloads; 3899 S.isExprCallable(*E, ReturnType, NonTemplateOverloads); 3900 if (!ReturnType.isNull() 3901 && ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 3902 S.Diag(E->getExprLoc(), diag::note_function_to_bool_call) 3903 << FixItHint::CreateInsertion( 3904 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()"); 3905 return; 3906 } 3907 } 3908 } 3909 return; // Other casts to bool are not checked. 3910 } 3911 3912 // Strip vector types. 3913 if (isa<VectorType>(Source)) { 3914 if (!isa<VectorType>(Target)) { 3915 if (S.SourceMgr.isInSystemMacro(CC)) 3916 return; 3917 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 3918 } 3919 3920 // If the vector cast is cast between two vectors of the same size, it is 3921 // a bitcast, not a conversion. 3922 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 3923 return; 3924 3925 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 3926 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 3927 } 3928 3929 // Strip complex types. 3930 if (isa<ComplexType>(Source)) { 3931 if (!isa<ComplexType>(Target)) { 3932 if (S.SourceMgr.isInSystemMacro(CC)) 3933 return; 3934 3935 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 3936 } 3937 3938 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 3939 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 3940 } 3941 3942 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 3943 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 3944 3945 // If the source is floating point... 3946 if (SourceBT && SourceBT->isFloatingPoint()) { 3947 // ...and the target is floating point... 3948 if (TargetBT && TargetBT->isFloatingPoint()) { 3949 // ...then warn if we're dropping FP rank. 3950 3951 // Builtin FP kinds are ordered by increasing FP rank. 3952 if (SourceBT->getKind() > TargetBT->getKind()) { 3953 // Don't warn about float constants that are precisely 3954 // representable in the target type. 3955 Expr::EvalResult result; 3956 if (E->EvaluateAsRValue(result, S.Context)) { 3957 // Value might be a float, a float vector, or a float complex. 3958 if (IsSameFloatAfterCast(result.Val, 3959 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 3960 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 3961 return; 3962 } 3963 3964 if (S.SourceMgr.isInSystemMacro(CC)) 3965 return; 3966 3967 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 3968 } 3969 return; 3970 } 3971 3972 // If the target is integral, always warn. 3973 if ((TargetBT && TargetBT->isInteger())) { 3974 if (S.SourceMgr.isInSystemMacro(CC)) 3975 return; 3976 3977 Expr *InnerE = E->IgnoreParenImpCasts(); 3978 // We also want to warn on, e.g., "int i = -1.234" 3979 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 3980 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 3981 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 3982 3983 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 3984 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 3985 } else { 3986 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 3987 } 3988 } 3989 3990 return; 3991 } 3992 3993 if (!Source->isIntegerType() || !Target->isIntegerType()) 3994 return; 3995 3996 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 3997 == Expr::NPCK_GNUNull) && Target->isIntegerType()) { 3998 S.Diag(E->getExprLoc(), diag::warn_impcast_null_pointer_to_integer) 3999 << E->getSourceRange() << clang::SourceRange(CC); 4000 return; 4001 } 4002 4003 IntRange SourceRange = GetExprRange(S.Context, E); 4004 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 4005 4006 if (SourceRange.Width > TargetRange.Width) { 4007 // If the source is a constant, use a default-on diagnostic. 4008 // TODO: this should happen for bitfield stores, too. 4009 llvm::APSInt Value(32); 4010 if (E->isIntegerConstantExpr(Value, S.Context)) { 4011 if (S.SourceMgr.isInSystemMacro(CC)) 4012 return; 4013 4014 std::string PrettySourceValue = Value.toString(10); 4015 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 4016 4017 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4018 S.PDiag(diag::warn_impcast_integer_precision_constant) 4019 << PrettySourceValue << PrettyTargetValue 4020 << E->getType() << T << E->getSourceRange() 4021 << clang::SourceRange(CC)); 4022 return; 4023 } 4024 4025 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 4026 if (S.SourceMgr.isInSystemMacro(CC)) 4027 return; 4028 4029 if (SourceRange.Width == 64 && TargetRange.Width == 32) 4030 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32, 4031 /* pruneControlFlow */ true); 4032 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 4033 } 4034 4035 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 4036 (!TargetRange.NonNegative && SourceRange.NonNegative && 4037 SourceRange.Width == TargetRange.Width)) { 4038 4039 if (S.SourceMgr.isInSystemMacro(CC)) 4040 return; 4041 4042 unsigned DiagID = diag::warn_impcast_integer_sign; 4043 4044 // Traditionally, gcc has warned about this under -Wsign-compare. 4045 // We also want to warn about it in -Wconversion. 4046 // So if -Wconversion is off, use a completely identical diagnostic 4047 // in the sign-compare group. 4048 // The conditional-checking code will 4049 if (ICContext) { 4050 DiagID = diag::warn_impcast_integer_sign_conditional; 4051 *ICContext = true; 4052 } 4053 4054 return DiagnoseImpCast(S, E, T, CC, DiagID); 4055 } 4056 4057 // Diagnose conversions between different enumeration types. 4058 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 4059 // type, to give us better diagnostics. 4060 QualType SourceType = E->getType(); 4061 if (!S.getLangOptions().CPlusPlus) { 4062 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 4063 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 4064 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 4065 SourceType = S.Context.getTypeDeclType(Enum); 4066 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 4067 } 4068 } 4069 4070 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 4071 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 4072 if ((SourceEnum->getDecl()->getIdentifier() || 4073 SourceEnum->getDecl()->getTypedefNameForAnonDecl()) && 4074 (TargetEnum->getDecl()->getIdentifier() || 4075 TargetEnum->getDecl()->getTypedefNameForAnonDecl()) && 4076 SourceEnum != TargetEnum) { 4077 if (S.SourceMgr.isInSystemMacro(CC)) 4078 return; 4079 4080 return DiagnoseImpCast(S, E, SourceType, T, CC, 4081 diag::warn_impcast_different_enum_types); 4082 } 4083 4084 return; 4085} 4086 4087void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 4088 4089void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 4090 SourceLocation CC, bool &ICContext) { 4091 E = E->IgnoreParenImpCasts(); 4092 4093 if (isa<ConditionalOperator>(E)) 4094 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 4095 4096 AnalyzeImplicitConversions(S, E, CC); 4097 if (E->getType() != T) 4098 return CheckImplicitConversion(S, E, T, CC, &ICContext); 4099 return; 4100} 4101 4102void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 4103 SourceLocation CC = E->getQuestionLoc(); 4104 4105 AnalyzeImplicitConversions(S, E->getCond(), CC); 4106 4107 bool Suspicious = false; 4108 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 4109 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 4110 4111 // If -Wconversion would have warned about either of the candidates 4112 // for a signedness conversion to the context type... 4113 if (!Suspicious) return; 4114 4115 // ...but it's currently ignored... 4116 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 4117 CC)) 4118 return; 4119 4120 // ...then check whether it would have warned about either of the 4121 // candidates for a signedness conversion to the condition type. 4122 if (E->getType() == T) return; 4123 4124 Suspicious = false; 4125 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 4126 E->getType(), CC, &Suspicious); 4127 if (!Suspicious) 4128 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 4129 E->getType(), CC, &Suspicious); 4130} 4131 4132/// AnalyzeImplicitConversions - Find and report any interesting 4133/// implicit conversions in the given expression. There are a couple 4134/// of competing diagnostics here, -Wconversion and -Wsign-compare. 4135void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 4136 QualType T = OrigE->getType(); 4137 Expr *E = OrigE->IgnoreParenImpCasts(); 4138 4139 if (E->isTypeDependent() || E->isValueDependent()) 4140 return; 4141 4142 // For conditional operators, we analyze the arguments as if they 4143 // were being fed directly into the output. 4144 if (isa<ConditionalOperator>(E)) { 4145 ConditionalOperator *CO = cast<ConditionalOperator>(E); 4146 CheckConditionalOperator(S, CO, T); 4147 return; 4148 } 4149 4150 // Go ahead and check any implicit conversions we might have skipped. 4151 // The non-canonical typecheck is just an optimization; 4152 // CheckImplicitConversion will filter out dead implicit conversions. 4153 if (E->getType() != T) 4154 CheckImplicitConversion(S, E, T, CC); 4155 4156 // Now continue drilling into this expression. 4157 4158 // Skip past explicit casts. 4159 if (isa<ExplicitCastExpr>(E)) { 4160 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 4161 return AnalyzeImplicitConversions(S, E, CC); 4162 } 4163 4164 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 4165 // Do a somewhat different check with comparison operators. 4166 if (BO->isComparisonOp()) 4167 return AnalyzeComparison(S, BO); 4168 4169 // And with simple assignments. 4170 if (BO->getOpcode() == BO_Assign) 4171 return AnalyzeAssignment(S, BO); 4172 } 4173 4174 // These break the otherwise-useful invariant below. Fortunately, 4175 // we don't really need to recurse into them, because any internal 4176 // expressions should have been analyzed already when they were 4177 // built into statements. 4178 if (isa<StmtExpr>(E)) return; 4179 4180 // Don't descend into unevaluated contexts. 4181 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 4182 4183 // Now just recurse over the expression's children. 4184 CC = E->getExprLoc(); 4185 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 4186 bool IsLogicalOperator = BO && BO->isLogicalOp(); 4187 for (Stmt::child_range I = E->children(); I; ++I) { 4188 Expr *ChildExpr = cast<Expr>(*I); 4189 if (IsLogicalOperator && 4190 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 4191 // Ignore checking string literals that are in logical operators. 4192 continue; 4193 AnalyzeImplicitConversions(S, ChildExpr, CC); 4194 } 4195} 4196 4197} // end anonymous namespace 4198 4199/// Diagnoses "dangerous" implicit conversions within the given 4200/// expression (which is a full expression). Implements -Wconversion 4201/// and -Wsign-compare. 4202/// 4203/// \param CC the "context" location of the implicit conversion, i.e. 4204/// the most location of the syntactic entity requiring the implicit 4205/// conversion 4206void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 4207 // Don't diagnose in unevaluated contexts. 4208 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 4209 return; 4210 4211 // Don't diagnose for value- or type-dependent expressions. 4212 if (E->isTypeDependent() || E->isValueDependent()) 4213 return; 4214 4215 // Check for array bounds violations in cases where the check isn't triggered 4216 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 4217 // ArraySubscriptExpr is on the RHS of a variable initialization. 4218 CheckArrayAccess(E); 4219 4220 // This is not the right CC for (e.g.) a variable initialization. 4221 AnalyzeImplicitConversions(*this, E, CC); 4222} 4223 4224void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 4225 FieldDecl *BitField, 4226 Expr *Init) { 4227 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 4228} 4229 4230/// CheckParmsForFunctionDef - Check that the parameters of the given 4231/// function are appropriate for the definition of a function. This 4232/// takes care of any checks that cannot be performed on the 4233/// declaration itself, e.g., that the types of each of the function 4234/// parameters are complete. 4235bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 4236 bool CheckParameterNames) { 4237 bool HasInvalidParm = false; 4238 for (; P != PEnd; ++P) { 4239 ParmVarDecl *Param = *P; 4240 4241 // C99 6.7.5.3p4: the parameters in a parameter type list in a 4242 // function declarator that is part of a function definition of 4243 // that function shall not have incomplete type. 4244 // 4245 // This is also C++ [dcl.fct]p6. 4246 if (!Param->isInvalidDecl() && 4247 RequireCompleteType(Param->getLocation(), Param->getType(), 4248 diag::err_typecheck_decl_incomplete_type)) { 4249 Param->setInvalidDecl(); 4250 HasInvalidParm = true; 4251 } 4252 4253 // C99 6.9.1p5: If the declarator includes a parameter type list, the 4254 // declaration of each parameter shall include an identifier. 4255 if (CheckParameterNames && 4256 Param->getIdentifier() == 0 && 4257 !Param->isImplicit() && 4258 !getLangOptions().CPlusPlus) 4259 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 4260 4261 // C99 6.7.5.3p12: 4262 // If the function declarator is not part of a definition of that 4263 // function, parameters may have incomplete type and may use the [*] 4264 // notation in their sequences of declarator specifiers to specify 4265 // variable length array types. 4266 QualType PType = Param->getOriginalType(); 4267 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 4268 if (AT->getSizeModifier() == ArrayType::Star) { 4269 // FIXME: This diagnosic should point the the '[*]' if source-location 4270 // information is added for it. 4271 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 4272 } 4273 } 4274 } 4275 4276 return HasInvalidParm; 4277} 4278 4279/// CheckCastAlign - Implements -Wcast-align, which warns when a 4280/// pointer cast increases the alignment requirements. 4281void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 4282 // This is actually a lot of work to potentially be doing on every 4283 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 4284 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 4285 TRange.getBegin()) 4286 == DiagnosticsEngine::Ignored) 4287 return; 4288 4289 // Ignore dependent types. 4290 if (T->isDependentType() || Op->getType()->isDependentType()) 4291 return; 4292 4293 // Require that the destination be a pointer type. 4294 const PointerType *DestPtr = T->getAs<PointerType>(); 4295 if (!DestPtr) return; 4296 4297 // If the destination has alignment 1, we're done. 4298 QualType DestPointee = DestPtr->getPointeeType(); 4299 if (DestPointee->isIncompleteType()) return; 4300 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 4301 if (DestAlign.isOne()) return; 4302 4303 // Require that the source be a pointer type. 4304 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 4305 if (!SrcPtr) return; 4306 QualType SrcPointee = SrcPtr->getPointeeType(); 4307 4308 // Whitelist casts from cv void*. We already implicitly 4309 // whitelisted casts to cv void*, since they have alignment 1. 4310 // Also whitelist casts involving incomplete types, which implicitly 4311 // includes 'void'. 4312 if (SrcPointee->isIncompleteType()) return; 4313 4314 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 4315 if (SrcAlign >= DestAlign) return; 4316 4317 Diag(TRange.getBegin(), diag::warn_cast_align) 4318 << Op->getType() << T 4319 << static_cast<unsigned>(SrcAlign.getQuantity()) 4320 << static_cast<unsigned>(DestAlign.getQuantity()) 4321 << TRange << Op->getSourceRange(); 4322} 4323 4324static const Type* getElementType(const Expr *BaseExpr) { 4325 const Type* EltType = BaseExpr->getType().getTypePtr(); 4326 if (EltType->isAnyPointerType()) 4327 return EltType->getPointeeType().getTypePtr(); 4328 else if (EltType->isArrayType()) 4329 return EltType->getBaseElementTypeUnsafe(); 4330 return EltType; 4331} 4332 4333/// \brief Check whether this array fits the idiom of a size-one tail padded 4334/// array member of a struct. 4335/// 4336/// We avoid emitting out-of-bounds access warnings for such arrays as they are 4337/// commonly used to emulate flexible arrays in C89 code. 4338static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 4339 const NamedDecl *ND) { 4340 if (Size != 1 || !ND) return false; 4341 4342 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 4343 if (!FD) return false; 4344 4345 // Don't consider sizes resulting from macro expansions or template argument 4346 // substitution to form C89 tail-padded arrays. 4347 ConstantArrayTypeLoc TL = 4348 cast<ConstantArrayTypeLoc>(FD->getTypeSourceInfo()->getTypeLoc()); 4349 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(TL.getSizeExpr()); 4350 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 4351 return false; 4352 4353 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 4354 if (!RD) return false; 4355 if (RD->isUnion()) return false; 4356 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 4357 if (!CRD->isStandardLayout()) return false; 4358 } 4359 4360 // See if this is the last field decl in the record. 4361 const Decl *D = FD; 4362 while ((D = D->getNextDeclInContext())) 4363 if (isa<FieldDecl>(D)) 4364 return false; 4365 return true; 4366} 4367 4368void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 4369 const ArraySubscriptExpr *ASE, 4370 bool AllowOnePastEnd, bool IndexNegated) { 4371 IndexExpr = IndexExpr->IgnoreParenCasts(); 4372 if (IndexExpr->isValueDependent()) 4373 return; 4374 4375 const Type *EffectiveType = getElementType(BaseExpr); 4376 BaseExpr = BaseExpr->IgnoreParenCasts(); 4377 const ConstantArrayType *ArrayTy = 4378 Context.getAsConstantArrayType(BaseExpr->getType()); 4379 if (!ArrayTy) 4380 return; 4381 4382 llvm::APSInt index; 4383 if (!IndexExpr->EvaluateAsInt(index, Context)) 4384 return; 4385 if (IndexNegated) 4386 index = -index; 4387 4388 const NamedDecl *ND = NULL; 4389 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4390 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 4391 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 4392 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 4393 4394 if (index.isUnsigned() || !index.isNegative()) { 4395 llvm::APInt size = ArrayTy->getSize(); 4396 if (!size.isStrictlyPositive()) 4397 return; 4398 4399 const Type* BaseType = getElementType(BaseExpr); 4400 if (BaseType != EffectiveType) { 4401 // Make sure we're comparing apples to apples when comparing index to size 4402 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 4403 uint64_t array_typesize = Context.getTypeSize(BaseType); 4404 // Handle ptrarith_typesize being zero, such as when casting to void* 4405 if (!ptrarith_typesize) ptrarith_typesize = 1; 4406 if (ptrarith_typesize != array_typesize) { 4407 // There's a cast to a different size type involved 4408 uint64_t ratio = array_typesize / ptrarith_typesize; 4409 // TODO: Be smarter about handling cases where array_typesize is not a 4410 // multiple of ptrarith_typesize 4411 if (ptrarith_typesize * ratio == array_typesize) 4412 size *= llvm::APInt(size.getBitWidth(), ratio); 4413 } 4414 } 4415 4416 if (size.getBitWidth() > index.getBitWidth()) 4417 index = index.sext(size.getBitWidth()); 4418 else if (size.getBitWidth() < index.getBitWidth()) 4419 size = size.sext(index.getBitWidth()); 4420 4421 // For array subscripting the index must be less than size, but for pointer 4422 // arithmetic also allow the index (offset) to be equal to size since 4423 // computing the next address after the end of the array is legal and 4424 // commonly done e.g. in C++ iterators and range-based for loops. 4425 if (AllowOnePastEnd ? index.sle(size) : index.slt(size)) 4426 return; 4427 4428 // Also don't warn for arrays of size 1 which are members of some 4429 // structure. These are often used to approximate flexible arrays in C89 4430 // code. 4431 if (IsTailPaddedMemberArray(*this, size, ND)) 4432 return; 4433 4434 // Suppress the warning if the subscript expression (as identified by the 4435 // ']' location) and the index expression are both from macro expansions 4436 // within a system header. 4437 if (ASE) { 4438 SourceLocation RBracketLoc = SourceMgr.getSpellingLoc( 4439 ASE->getRBracketLoc()); 4440 if (SourceMgr.isInSystemHeader(RBracketLoc)) { 4441 SourceLocation IndexLoc = SourceMgr.getSpellingLoc( 4442 IndexExpr->getLocStart()); 4443 if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc)) 4444 return; 4445 } 4446 } 4447 4448 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 4449 if (ASE) 4450 DiagID = diag::warn_array_index_exceeds_bounds; 4451 4452 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4453 PDiag(DiagID) << index.toString(10, true) 4454 << size.toString(10, true) 4455 << (unsigned)size.getLimitedValue(~0U) 4456 << IndexExpr->getSourceRange()); 4457 } else { 4458 unsigned DiagID = diag::warn_array_index_precedes_bounds; 4459 if (!ASE) { 4460 DiagID = diag::warn_ptr_arith_precedes_bounds; 4461 if (index.isNegative()) index = -index; 4462 } 4463 4464 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4465 PDiag(DiagID) << index.toString(10, true) 4466 << IndexExpr->getSourceRange()); 4467 } 4468 4469 if (!ND) { 4470 // Try harder to find a NamedDecl to point at in the note. 4471 while (const ArraySubscriptExpr *ASE = 4472 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 4473 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 4474 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4475 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 4476 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 4477 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 4478 } 4479 4480 if (ND) 4481 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 4482 PDiag(diag::note_array_index_out_of_bounds) 4483 << ND->getDeclName()); 4484} 4485 4486void Sema::CheckArrayAccess(const Expr *expr) { 4487 int AllowOnePastEnd = 0; 4488 while (expr) { 4489 expr = expr->IgnoreParenImpCasts(); 4490 switch (expr->getStmtClass()) { 4491 case Stmt::ArraySubscriptExprClass: { 4492 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 4493 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE, 4494 AllowOnePastEnd > 0); 4495 return; 4496 } 4497 case Stmt::UnaryOperatorClass: { 4498 // Only unwrap the * and & unary operators 4499 const UnaryOperator *UO = cast<UnaryOperator>(expr); 4500 expr = UO->getSubExpr(); 4501 switch (UO->getOpcode()) { 4502 case UO_AddrOf: 4503 AllowOnePastEnd++; 4504 break; 4505 case UO_Deref: 4506 AllowOnePastEnd--; 4507 break; 4508 default: 4509 return; 4510 } 4511 break; 4512 } 4513 case Stmt::ConditionalOperatorClass: { 4514 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 4515 if (const Expr *lhs = cond->getLHS()) 4516 CheckArrayAccess(lhs); 4517 if (const Expr *rhs = cond->getRHS()) 4518 CheckArrayAccess(rhs); 4519 return; 4520 } 4521 default: 4522 return; 4523 } 4524 } 4525} 4526 4527//===--- CHECK: Objective-C retain cycles ----------------------------------// 4528 4529namespace { 4530 struct RetainCycleOwner { 4531 RetainCycleOwner() : Variable(0), Indirect(false) {} 4532 VarDecl *Variable; 4533 SourceRange Range; 4534 SourceLocation Loc; 4535 bool Indirect; 4536 4537 void setLocsFrom(Expr *e) { 4538 Loc = e->getExprLoc(); 4539 Range = e->getSourceRange(); 4540 } 4541 }; 4542} 4543 4544/// Consider whether capturing the given variable can possibly lead to 4545/// a retain cycle. 4546static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 4547 // In ARC, it's captured strongly iff the variable has __strong 4548 // lifetime. In MRR, it's captured strongly if the variable is 4549 // __block and has an appropriate type. 4550 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4551 return false; 4552 4553 owner.Variable = var; 4554 owner.setLocsFrom(ref); 4555 return true; 4556} 4557 4558static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) { 4559 while (true) { 4560 e = e->IgnoreParens(); 4561 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 4562 switch (cast->getCastKind()) { 4563 case CK_BitCast: 4564 case CK_LValueBitCast: 4565 case CK_LValueToRValue: 4566 case CK_ARCReclaimReturnedObject: 4567 e = cast->getSubExpr(); 4568 continue; 4569 4570 default: 4571 return false; 4572 } 4573 } 4574 4575 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 4576 ObjCIvarDecl *ivar = ref->getDecl(); 4577 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4578 return false; 4579 4580 // Try to find a retain cycle in the base. 4581 if (!findRetainCycleOwner(S, ref->getBase(), owner)) 4582 return false; 4583 4584 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 4585 owner.Indirect = true; 4586 return true; 4587 } 4588 4589 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 4590 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 4591 if (!var) return false; 4592 return considerVariable(var, ref, owner); 4593 } 4594 4595 if (BlockDeclRefExpr *ref = dyn_cast<BlockDeclRefExpr>(e)) { 4596 owner.Variable = ref->getDecl(); 4597 owner.setLocsFrom(ref); 4598 return true; 4599 } 4600 4601 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 4602 if (member->isArrow()) return false; 4603 4604 // Don't count this as an indirect ownership. 4605 e = member->getBase(); 4606 continue; 4607 } 4608 4609 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 4610 // Only pay attention to pseudo-objects on property references. 4611 ObjCPropertyRefExpr *pre 4612 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 4613 ->IgnoreParens()); 4614 if (!pre) return false; 4615 if (pre->isImplicitProperty()) return false; 4616 ObjCPropertyDecl *property = pre->getExplicitProperty(); 4617 if (!property->isRetaining() && 4618 !(property->getPropertyIvarDecl() && 4619 property->getPropertyIvarDecl()->getType() 4620 .getObjCLifetime() == Qualifiers::OCL_Strong)) 4621 return false; 4622 4623 owner.Indirect = true; 4624 if (pre->isSuperReceiver()) { 4625 owner.Variable = S.getCurMethodDecl()->getSelfDecl(); 4626 if (!owner.Variable) 4627 return false; 4628 owner.Loc = pre->getLocation(); 4629 owner.Range = pre->getSourceRange(); 4630 return true; 4631 } 4632 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 4633 ->getSourceExpr()); 4634 continue; 4635 } 4636 4637 // Array ivars? 4638 4639 return false; 4640 } 4641} 4642 4643namespace { 4644 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 4645 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 4646 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 4647 Variable(variable), Capturer(0) {} 4648 4649 VarDecl *Variable; 4650 Expr *Capturer; 4651 4652 void VisitDeclRefExpr(DeclRefExpr *ref) { 4653 if (ref->getDecl() == Variable && !Capturer) 4654 Capturer = ref; 4655 } 4656 4657 void VisitBlockDeclRefExpr(BlockDeclRefExpr *ref) { 4658 if (ref->getDecl() == Variable && !Capturer) 4659 Capturer = ref; 4660 } 4661 4662 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 4663 if (Capturer) return; 4664 Visit(ref->getBase()); 4665 if (Capturer && ref->isFreeIvar()) 4666 Capturer = ref; 4667 } 4668 4669 void VisitBlockExpr(BlockExpr *block) { 4670 // Look inside nested blocks 4671 if (block->getBlockDecl()->capturesVariable(Variable)) 4672 Visit(block->getBlockDecl()->getBody()); 4673 } 4674 }; 4675} 4676 4677/// Check whether the given argument is a block which captures a 4678/// variable. 4679static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 4680 assert(owner.Variable && owner.Loc.isValid()); 4681 4682 e = e->IgnoreParenCasts(); 4683 BlockExpr *block = dyn_cast<BlockExpr>(e); 4684 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 4685 return 0; 4686 4687 FindCaptureVisitor visitor(S.Context, owner.Variable); 4688 visitor.Visit(block->getBlockDecl()->getBody()); 4689 return visitor.Capturer; 4690} 4691 4692static void diagnoseRetainCycle(Sema &S, Expr *capturer, 4693 RetainCycleOwner &owner) { 4694 assert(capturer); 4695 assert(owner.Variable && owner.Loc.isValid()); 4696 4697 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 4698 << owner.Variable << capturer->getSourceRange(); 4699 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 4700 << owner.Indirect << owner.Range; 4701} 4702 4703/// Check for a keyword selector that starts with the word 'add' or 4704/// 'set'. 4705static bool isSetterLikeSelector(Selector sel) { 4706 if (sel.isUnarySelector()) return false; 4707 4708 StringRef str = sel.getNameForSlot(0); 4709 while (!str.empty() && str.front() == '_') str = str.substr(1); 4710 if (str.startswith("set")) 4711 str = str.substr(3); 4712 else if (str.startswith("add")) { 4713 // Specially whitelist 'addOperationWithBlock:'. 4714 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 4715 return false; 4716 str = str.substr(3); 4717 } 4718 else 4719 return false; 4720 4721 if (str.empty()) return true; 4722 return !islower(str.front()); 4723} 4724 4725/// Check a message send to see if it's likely to cause a retain cycle. 4726void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 4727 // Only check instance methods whose selector looks like a setter. 4728 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 4729 return; 4730 4731 // Try to find a variable that the receiver is strongly owned by. 4732 RetainCycleOwner owner; 4733 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 4734 if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner)) 4735 return; 4736 } else { 4737 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 4738 owner.Variable = getCurMethodDecl()->getSelfDecl(); 4739 owner.Loc = msg->getSuperLoc(); 4740 owner.Range = msg->getSuperLoc(); 4741 } 4742 4743 // Check whether the receiver is captured by any of the arguments. 4744 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 4745 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 4746 return diagnoseRetainCycle(*this, capturer, owner); 4747} 4748 4749/// Check a property assign to see if it's likely to cause a retain cycle. 4750void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 4751 RetainCycleOwner owner; 4752 if (!findRetainCycleOwner(*this, receiver, owner)) 4753 return; 4754 4755 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 4756 diagnoseRetainCycle(*this, capturer, owner); 4757} 4758 4759bool Sema::checkUnsafeAssigns(SourceLocation Loc, 4760 QualType LHS, Expr *RHS) { 4761 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 4762 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 4763 return false; 4764 // strip off any implicit cast added to get to the one arc-specific 4765 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4766 if (cast->getCastKind() == CK_ARCConsumeObject) { 4767 Diag(Loc, diag::warn_arc_retained_assign) 4768 << (LT == Qualifiers::OCL_ExplicitNone) 4769 << RHS->getSourceRange(); 4770 return true; 4771 } 4772 RHS = cast->getSubExpr(); 4773 } 4774 return false; 4775} 4776 4777void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 4778 Expr *LHS, Expr *RHS) { 4779 QualType LHSType; 4780 // PropertyRef on LHS type need be directly obtained from 4781 // its declaration as it has a PsuedoType. 4782 ObjCPropertyRefExpr *PRE 4783 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens()); 4784 if (PRE && !PRE->isImplicitProperty()) { 4785 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 4786 if (PD) 4787 LHSType = PD->getType(); 4788 } 4789 4790 if (LHSType.isNull()) 4791 LHSType = LHS->getType(); 4792 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 4793 return; 4794 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 4795 // FIXME. Check for other life times. 4796 if (LT != Qualifiers::OCL_None) 4797 return; 4798 4799 if (PRE) { 4800 if (PRE->isImplicitProperty()) 4801 return; 4802 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 4803 if (!PD) 4804 return; 4805 4806 unsigned Attributes = PD->getPropertyAttributes(); 4807 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) { 4808 // when 'assign' attribute was not explicitly specified 4809 // by user, ignore it and rely on property type itself 4810 // for lifetime info. 4811 unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten(); 4812 if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) && 4813 LHSType->isObjCRetainableType()) 4814 return; 4815 4816 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4817 if (cast->getCastKind() == CK_ARCConsumeObject) { 4818 Diag(Loc, diag::warn_arc_retained_property_assign) 4819 << RHS->getSourceRange(); 4820 return; 4821 } 4822 RHS = cast->getSubExpr(); 4823 } 4824 } 4825 } 4826} 4827