SemaChecking.cpp revision 8c382060c9e6668a94f1485dd16f012cda526c5f
1//===--- SemaChecking.cpp - Extra Semantic Checking -----------------------===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// This file implements extra semantic analysis beyond what is enforced 11// by the C type system. 12// 13//===----------------------------------------------------------------------===// 14 15#include "clang/Sema/Initialization.h" 16#include "clang/Sema/Sema.h" 17#include "clang/Sema/SemaInternal.h" 18#include "clang/Sema/Initialization.h" 19#include "clang/Sema/ScopeInfo.h" 20#include "clang/Analysis/Analyses/FormatString.h" 21#include "clang/AST/ASTContext.h" 22#include "clang/AST/CharUnits.h" 23#include "clang/AST/DeclCXX.h" 24#include "clang/AST/DeclObjC.h" 25#include "clang/AST/ExprCXX.h" 26#include "clang/AST/ExprObjC.h" 27#include "clang/AST/EvaluatedExprVisitor.h" 28#include "clang/AST/DeclObjC.h" 29#include "clang/AST/StmtCXX.h" 30#include "clang/AST/StmtObjC.h" 31#include "clang/Lex/Preprocessor.h" 32#include "llvm/ADT/BitVector.h" 33#include "llvm/ADT/STLExtras.h" 34#include "llvm/Support/raw_ostream.h" 35#include "clang/Basic/TargetBuiltins.h" 36#include "clang/Basic/TargetInfo.h" 37#include "clang/Basic/ConvertUTF.h" 38#include <limits> 39using namespace clang; 40using namespace sema; 41 42SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 43 unsigned ByteNo) const { 44 return SL->getLocationOfByte(ByteNo, PP.getSourceManager(), 45 PP.getLangOptions(), PP.getTargetInfo()); 46} 47 48bool Sema::CheckablePrintfAttr(const FormatAttr *Format, Expr **Args, 49 unsigned NumArgs, bool IsCXXMemberCall) { 50 StringRef Type = Format->getType(); 51 // FIXME: add support for "CFString" Type. They are not string literal though, 52 // so they need special handling. 53 if (Type == "printf" || Type == "NSString") return true; 54 if (Type == "printf0") { 55 // printf0 allows null "format" string; if so don't check format/args 56 unsigned format_idx = Format->getFormatIdx() - 1; 57 // Does the index refer to the implicit object argument? 58 if (IsCXXMemberCall) { 59 if (format_idx == 0) 60 return false; 61 --format_idx; 62 } 63 if (format_idx < NumArgs) { 64 Expr *Format = Args[format_idx]->IgnoreParenCasts(); 65 if (!Format->isNullPointerConstant(Context, 66 Expr::NPC_ValueDependentIsNull)) 67 return true; 68 } 69 } 70 return false; 71} 72 73/// Checks that a call expression's argument count is the desired number. 74/// This is useful when doing custom type-checking. Returns true on error. 75static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) { 76 unsigned argCount = call->getNumArgs(); 77 if (argCount == desiredArgCount) return false; 78 79 if (argCount < desiredArgCount) 80 return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args) 81 << 0 /*function call*/ << desiredArgCount << argCount 82 << call->getSourceRange(); 83 84 // Highlight all the excess arguments. 85 SourceRange range(call->getArg(desiredArgCount)->getLocStart(), 86 call->getArg(argCount - 1)->getLocEnd()); 87 88 return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args) 89 << 0 /*function call*/ << desiredArgCount << argCount 90 << call->getArg(1)->getSourceRange(); 91} 92 93/// CheckBuiltinAnnotationString - Checks that string argument to the builtin 94/// annotation is a non wide string literal. 95static bool CheckBuiltinAnnotationString(Sema &S, Expr *Arg) { 96 Arg = Arg->IgnoreParenCasts(); 97 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 98 if (!Literal || !Literal->isAscii()) { 99 S.Diag(Arg->getLocStart(), diag::err_builtin_annotation_not_string_constant) 100 << Arg->getSourceRange(); 101 return true; 102 } 103 return false; 104} 105 106ExprResult 107Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 108 ExprResult TheCallResult(Owned(TheCall)); 109 110 // Find out if any arguments are required to be integer constant expressions. 111 unsigned ICEArguments = 0; 112 ASTContext::GetBuiltinTypeError Error; 113 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments); 114 if (Error != ASTContext::GE_None) 115 ICEArguments = 0; // Don't diagnose previously diagnosed errors. 116 117 // If any arguments are required to be ICE's, check and diagnose. 118 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) { 119 // Skip arguments not required to be ICE's. 120 if ((ICEArguments & (1 << ArgNo)) == 0) continue; 121 122 llvm::APSInt Result; 123 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 124 return true; 125 ICEArguments &= ~(1 << ArgNo); 126 } 127 128 switch (BuiltinID) { 129 case Builtin::BI__builtin___CFStringMakeConstantString: 130 assert(TheCall->getNumArgs() == 1 && 131 "Wrong # arguments to builtin CFStringMakeConstantString"); 132 if (CheckObjCString(TheCall->getArg(0))) 133 return ExprError(); 134 break; 135 case Builtin::BI__builtin_stdarg_start: 136 case Builtin::BI__builtin_va_start: 137 if (SemaBuiltinVAStart(TheCall)) 138 return ExprError(); 139 break; 140 case Builtin::BI__builtin_isgreater: 141 case Builtin::BI__builtin_isgreaterequal: 142 case Builtin::BI__builtin_isless: 143 case Builtin::BI__builtin_islessequal: 144 case Builtin::BI__builtin_islessgreater: 145 case Builtin::BI__builtin_isunordered: 146 if (SemaBuiltinUnorderedCompare(TheCall)) 147 return ExprError(); 148 break; 149 case Builtin::BI__builtin_fpclassify: 150 if (SemaBuiltinFPClassification(TheCall, 6)) 151 return ExprError(); 152 break; 153 case Builtin::BI__builtin_isfinite: 154 case Builtin::BI__builtin_isinf: 155 case Builtin::BI__builtin_isinf_sign: 156 case Builtin::BI__builtin_isnan: 157 case Builtin::BI__builtin_isnormal: 158 if (SemaBuiltinFPClassification(TheCall, 1)) 159 return ExprError(); 160 break; 161 case Builtin::BI__builtin_shufflevector: 162 return SemaBuiltinShuffleVector(TheCall); 163 // TheCall will be freed by the smart pointer here, but that's fine, since 164 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 165 case Builtin::BI__builtin_prefetch: 166 if (SemaBuiltinPrefetch(TheCall)) 167 return ExprError(); 168 break; 169 case Builtin::BI__builtin_object_size: 170 if (SemaBuiltinObjectSize(TheCall)) 171 return ExprError(); 172 break; 173 case Builtin::BI__builtin_longjmp: 174 if (SemaBuiltinLongjmp(TheCall)) 175 return ExprError(); 176 break; 177 178 case Builtin::BI__builtin_classify_type: 179 if (checkArgCount(*this, TheCall, 1)) return true; 180 TheCall->setType(Context.IntTy); 181 break; 182 case Builtin::BI__builtin_constant_p: 183 if (checkArgCount(*this, TheCall, 1)) return true; 184 TheCall->setType(Context.IntTy); 185 break; 186 case Builtin::BI__sync_fetch_and_add: 187 case Builtin::BI__sync_fetch_and_add_1: 188 case Builtin::BI__sync_fetch_and_add_2: 189 case Builtin::BI__sync_fetch_and_add_4: 190 case Builtin::BI__sync_fetch_and_add_8: 191 case Builtin::BI__sync_fetch_and_add_16: 192 case Builtin::BI__sync_fetch_and_sub: 193 case Builtin::BI__sync_fetch_and_sub_1: 194 case Builtin::BI__sync_fetch_and_sub_2: 195 case Builtin::BI__sync_fetch_and_sub_4: 196 case Builtin::BI__sync_fetch_and_sub_8: 197 case Builtin::BI__sync_fetch_and_sub_16: 198 case Builtin::BI__sync_fetch_and_or: 199 case Builtin::BI__sync_fetch_and_or_1: 200 case Builtin::BI__sync_fetch_and_or_2: 201 case Builtin::BI__sync_fetch_and_or_4: 202 case Builtin::BI__sync_fetch_and_or_8: 203 case Builtin::BI__sync_fetch_and_or_16: 204 case Builtin::BI__sync_fetch_and_and: 205 case Builtin::BI__sync_fetch_and_and_1: 206 case Builtin::BI__sync_fetch_and_and_2: 207 case Builtin::BI__sync_fetch_and_and_4: 208 case Builtin::BI__sync_fetch_and_and_8: 209 case Builtin::BI__sync_fetch_and_and_16: 210 case Builtin::BI__sync_fetch_and_xor: 211 case Builtin::BI__sync_fetch_and_xor_1: 212 case Builtin::BI__sync_fetch_and_xor_2: 213 case Builtin::BI__sync_fetch_and_xor_4: 214 case Builtin::BI__sync_fetch_and_xor_8: 215 case Builtin::BI__sync_fetch_and_xor_16: 216 case Builtin::BI__sync_add_and_fetch: 217 case Builtin::BI__sync_add_and_fetch_1: 218 case Builtin::BI__sync_add_and_fetch_2: 219 case Builtin::BI__sync_add_and_fetch_4: 220 case Builtin::BI__sync_add_and_fetch_8: 221 case Builtin::BI__sync_add_and_fetch_16: 222 case Builtin::BI__sync_sub_and_fetch: 223 case Builtin::BI__sync_sub_and_fetch_1: 224 case Builtin::BI__sync_sub_and_fetch_2: 225 case Builtin::BI__sync_sub_and_fetch_4: 226 case Builtin::BI__sync_sub_and_fetch_8: 227 case Builtin::BI__sync_sub_and_fetch_16: 228 case Builtin::BI__sync_and_and_fetch: 229 case Builtin::BI__sync_and_and_fetch_1: 230 case Builtin::BI__sync_and_and_fetch_2: 231 case Builtin::BI__sync_and_and_fetch_4: 232 case Builtin::BI__sync_and_and_fetch_8: 233 case Builtin::BI__sync_and_and_fetch_16: 234 case Builtin::BI__sync_or_and_fetch: 235 case Builtin::BI__sync_or_and_fetch_1: 236 case Builtin::BI__sync_or_and_fetch_2: 237 case Builtin::BI__sync_or_and_fetch_4: 238 case Builtin::BI__sync_or_and_fetch_8: 239 case Builtin::BI__sync_or_and_fetch_16: 240 case Builtin::BI__sync_xor_and_fetch: 241 case Builtin::BI__sync_xor_and_fetch_1: 242 case Builtin::BI__sync_xor_and_fetch_2: 243 case Builtin::BI__sync_xor_and_fetch_4: 244 case Builtin::BI__sync_xor_and_fetch_8: 245 case Builtin::BI__sync_xor_and_fetch_16: 246 case Builtin::BI__sync_val_compare_and_swap: 247 case Builtin::BI__sync_val_compare_and_swap_1: 248 case Builtin::BI__sync_val_compare_and_swap_2: 249 case Builtin::BI__sync_val_compare_and_swap_4: 250 case Builtin::BI__sync_val_compare_and_swap_8: 251 case Builtin::BI__sync_val_compare_and_swap_16: 252 case Builtin::BI__sync_bool_compare_and_swap: 253 case Builtin::BI__sync_bool_compare_and_swap_1: 254 case Builtin::BI__sync_bool_compare_and_swap_2: 255 case Builtin::BI__sync_bool_compare_and_swap_4: 256 case Builtin::BI__sync_bool_compare_and_swap_8: 257 case Builtin::BI__sync_bool_compare_and_swap_16: 258 case Builtin::BI__sync_lock_test_and_set: 259 case Builtin::BI__sync_lock_test_and_set_1: 260 case Builtin::BI__sync_lock_test_and_set_2: 261 case Builtin::BI__sync_lock_test_and_set_4: 262 case Builtin::BI__sync_lock_test_and_set_8: 263 case Builtin::BI__sync_lock_test_and_set_16: 264 case Builtin::BI__sync_lock_release: 265 case Builtin::BI__sync_lock_release_1: 266 case Builtin::BI__sync_lock_release_2: 267 case Builtin::BI__sync_lock_release_4: 268 case Builtin::BI__sync_lock_release_8: 269 case Builtin::BI__sync_lock_release_16: 270 case Builtin::BI__sync_swap: 271 case Builtin::BI__sync_swap_1: 272 case Builtin::BI__sync_swap_2: 273 case Builtin::BI__sync_swap_4: 274 case Builtin::BI__sync_swap_8: 275 case Builtin::BI__sync_swap_16: 276 return SemaBuiltinAtomicOverloaded(move(TheCallResult)); 277 case Builtin::BI__atomic_load: 278 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Load); 279 case Builtin::BI__atomic_store: 280 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Store); 281 case Builtin::BI__atomic_init: 282 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Init); 283 case Builtin::BI__atomic_exchange: 284 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Xchg); 285 case Builtin::BI__atomic_compare_exchange_strong: 286 return SemaAtomicOpsOverloaded(move(TheCallResult), 287 AtomicExpr::CmpXchgStrong); 288 case Builtin::BI__atomic_compare_exchange_weak: 289 return SemaAtomicOpsOverloaded(move(TheCallResult), 290 AtomicExpr::CmpXchgWeak); 291 case Builtin::BI__atomic_fetch_add: 292 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Add); 293 case Builtin::BI__atomic_fetch_sub: 294 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Sub); 295 case Builtin::BI__atomic_fetch_and: 296 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::And); 297 case Builtin::BI__atomic_fetch_or: 298 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Or); 299 case Builtin::BI__atomic_fetch_xor: 300 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Xor); 301 case Builtin::BI__builtin_annotation: 302 if (CheckBuiltinAnnotationString(*this, TheCall->getArg(1))) 303 return ExprError(); 304 break; 305 } 306 307 // Since the target specific builtins for each arch overlap, only check those 308 // of the arch we are compiling for. 309 if (BuiltinID >= Builtin::FirstTSBuiltin) { 310 switch (Context.getTargetInfo().getTriple().getArch()) { 311 case llvm::Triple::arm: 312 case llvm::Triple::thumb: 313 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 314 return ExprError(); 315 break; 316 default: 317 break; 318 } 319 } 320 321 return move(TheCallResult); 322} 323 324// Get the valid immediate range for the specified NEON type code. 325static unsigned RFT(unsigned t, bool shift = false) { 326 NeonTypeFlags Type(t); 327 int IsQuad = Type.isQuad(); 328 switch (Type.getEltType()) { 329 case NeonTypeFlags::Int8: 330 case NeonTypeFlags::Poly8: 331 return shift ? 7 : (8 << IsQuad) - 1; 332 case NeonTypeFlags::Int16: 333 case NeonTypeFlags::Poly16: 334 return shift ? 15 : (4 << IsQuad) - 1; 335 case NeonTypeFlags::Int32: 336 return shift ? 31 : (2 << IsQuad) - 1; 337 case NeonTypeFlags::Int64: 338 return shift ? 63 : (1 << IsQuad) - 1; 339 case NeonTypeFlags::Float16: 340 assert(!shift && "cannot shift float types!"); 341 return (4 << IsQuad) - 1; 342 case NeonTypeFlags::Float32: 343 assert(!shift && "cannot shift float types!"); 344 return (2 << IsQuad) - 1; 345 } 346 llvm_unreachable("Invalid NeonTypeFlag!"); 347} 348 349/// getNeonEltType - Return the QualType corresponding to the elements of 350/// the vector type specified by the NeonTypeFlags. This is used to check 351/// the pointer arguments for Neon load/store intrinsics. 352static QualType getNeonEltType(NeonTypeFlags Flags, ASTContext &Context) { 353 switch (Flags.getEltType()) { 354 case NeonTypeFlags::Int8: 355 return Flags.isUnsigned() ? Context.UnsignedCharTy : Context.SignedCharTy; 356 case NeonTypeFlags::Int16: 357 return Flags.isUnsigned() ? Context.UnsignedShortTy : Context.ShortTy; 358 case NeonTypeFlags::Int32: 359 return Flags.isUnsigned() ? Context.UnsignedIntTy : Context.IntTy; 360 case NeonTypeFlags::Int64: 361 return Flags.isUnsigned() ? Context.UnsignedLongLongTy : Context.LongLongTy; 362 case NeonTypeFlags::Poly8: 363 return Context.SignedCharTy; 364 case NeonTypeFlags::Poly16: 365 return Context.ShortTy; 366 case NeonTypeFlags::Float16: 367 return Context.UnsignedShortTy; 368 case NeonTypeFlags::Float32: 369 return Context.FloatTy; 370 } 371 llvm_unreachable("Invalid NeonTypeFlag!"); 372} 373 374bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 375 llvm::APSInt Result; 376 377 unsigned mask = 0; 378 unsigned TV = 0; 379 int PtrArgNum = -1; 380 bool HasConstPtr = false; 381 switch (BuiltinID) { 382#define GET_NEON_OVERLOAD_CHECK 383#include "clang/Basic/arm_neon.inc" 384#undef GET_NEON_OVERLOAD_CHECK 385 } 386 387 // For NEON intrinsics which are overloaded on vector element type, validate 388 // the immediate which specifies which variant to emit. 389 unsigned ImmArg = TheCall->getNumArgs()-1; 390 if (mask) { 391 if (SemaBuiltinConstantArg(TheCall, ImmArg, Result)) 392 return true; 393 394 TV = Result.getLimitedValue(64); 395 if ((TV > 63) || (mask & (1 << TV)) == 0) 396 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 397 << TheCall->getArg(ImmArg)->getSourceRange(); 398 } 399 400 if (PtrArgNum >= 0) { 401 // Check that pointer arguments have the specified type. 402 Expr *Arg = TheCall->getArg(PtrArgNum); 403 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) 404 Arg = ICE->getSubExpr(); 405 ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg); 406 QualType RHSTy = RHS.get()->getType(); 407 QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context); 408 if (HasConstPtr) 409 EltTy = EltTy.withConst(); 410 QualType LHSTy = Context.getPointerType(EltTy); 411 AssignConvertType ConvTy; 412 ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS); 413 if (RHS.isInvalid()) 414 return true; 415 if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy, 416 RHS.get(), AA_Assigning)) 417 return true; 418 } 419 420 // For NEON intrinsics which take an immediate value as part of the 421 // instruction, range check them here. 422 unsigned i = 0, l = 0, u = 0; 423 switch (BuiltinID) { 424 default: return false; 425 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 426 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 427 case ARM::BI__builtin_arm_vcvtr_f: 428 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 429#define GET_NEON_IMMEDIATE_CHECK 430#include "clang/Basic/arm_neon.inc" 431#undef GET_NEON_IMMEDIATE_CHECK 432 }; 433 434 // Check that the immediate argument is actually a constant. 435 if (SemaBuiltinConstantArg(TheCall, i, Result)) 436 return true; 437 438 // Range check against the upper/lower values for this isntruction. 439 unsigned Val = Result.getZExtValue(); 440 if (Val < l || Val > (u + l)) 441 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 442 << l << u+l << TheCall->getArg(i)->getSourceRange(); 443 444 // FIXME: VFP Intrinsics should error if VFP not present. 445 return false; 446} 447 448/// CheckFunctionCall - Check a direct function call for various correctness 449/// and safety properties not strictly enforced by the C type system. 450bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 451 // Get the IdentifierInfo* for the called function. 452 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 453 454 // None of the checks below are needed for functions that don't have 455 // simple names (e.g., C++ conversion functions). 456 if (!FnInfo) 457 return false; 458 459 // FIXME: This mechanism should be abstracted to be less fragile and 460 // more efficient. For example, just map function ids to custom 461 // handlers. 462 463 // Printf and scanf checking. 464 for (specific_attr_iterator<FormatAttr> 465 i = FDecl->specific_attr_begin<FormatAttr>(), 466 e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { 467 CheckFormatArguments(*i, TheCall); 468 } 469 470 for (specific_attr_iterator<NonNullAttr> 471 i = FDecl->specific_attr_begin<NonNullAttr>(), 472 e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) { 473 CheckNonNullArguments(*i, TheCall->getArgs(), 474 TheCall->getCallee()->getLocStart()); 475 } 476 477 unsigned CMId = FDecl->getMemoryFunctionKind(); 478 if (CMId == 0) 479 return false; 480 481 // Handle memory setting and copying functions. 482 if (CMId == Builtin::BIstrlcpy || CMId == Builtin::BIstrlcat) 483 CheckStrlcpycatArguments(TheCall, FnInfo); 484 else 485 CheckMemaccessArguments(TheCall, CMId, FnInfo); 486 487 return false; 488} 489 490bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac, 491 Expr **Args, unsigned NumArgs) { 492 for (specific_attr_iterator<FormatAttr> 493 i = Method->specific_attr_begin<FormatAttr>(), 494 e = Method->specific_attr_end<FormatAttr>(); i != e ; ++i) { 495 496 CheckFormatArguments(*i, Args, NumArgs, false, lbrac, 497 Method->getSourceRange()); 498 } 499 500 // diagnose nonnull arguments. 501 for (specific_attr_iterator<NonNullAttr> 502 i = Method->specific_attr_begin<NonNullAttr>(), 503 e = Method->specific_attr_end<NonNullAttr>(); i != e; ++i) { 504 CheckNonNullArguments(*i, Args, lbrac); 505 } 506 507 return false; 508} 509 510bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 511 // Printf checking. 512 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 513 if (!Format) 514 return false; 515 516 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 517 if (!V) 518 return false; 519 520 QualType Ty = V->getType(); 521 if (!Ty->isBlockPointerType()) 522 return false; 523 524 CheckFormatArguments(Format, TheCall); 525 526 return false; 527} 528 529ExprResult 530Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, AtomicExpr::AtomicOp Op) { 531 CallExpr *TheCall = cast<CallExpr>(TheCallResult.get()); 532 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 533 534 // All these operations take one of the following four forms: 535 // T __atomic_load(_Atomic(T)*, int) (loads) 536 // T* __atomic_add(_Atomic(T*)*, ptrdiff_t, int) (pointer add/sub) 537 // int __atomic_compare_exchange_strong(_Atomic(T)*, T*, T, int, int) 538 // (cmpxchg) 539 // T __atomic_exchange(_Atomic(T)*, T, int) (everything else) 540 // where T is an appropriate type, and the int paremeterss are for orderings. 541 unsigned NumVals = 1; 542 unsigned NumOrders = 1; 543 if (Op == AtomicExpr::Load) { 544 NumVals = 0; 545 } else if (Op == AtomicExpr::CmpXchgWeak || Op == AtomicExpr::CmpXchgStrong) { 546 NumVals = 2; 547 NumOrders = 2; 548 } 549 if (Op == AtomicExpr::Init) 550 NumOrders = 0; 551 552 if (TheCall->getNumArgs() < NumVals+NumOrders+1) { 553 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 554 << 0 << NumVals+NumOrders+1 << TheCall->getNumArgs() 555 << TheCall->getCallee()->getSourceRange(); 556 return ExprError(); 557 } else if (TheCall->getNumArgs() > NumVals+NumOrders+1) { 558 Diag(TheCall->getArg(NumVals+NumOrders+1)->getLocStart(), 559 diag::err_typecheck_call_too_many_args) 560 << 0 << NumVals+NumOrders+1 << TheCall->getNumArgs() 561 << TheCall->getCallee()->getSourceRange(); 562 return ExprError(); 563 } 564 565 // Inspect the first argument of the atomic operation. This should always be 566 // a pointer to an _Atomic type. 567 Expr *Ptr = TheCall->getArg(0); 568 Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get(); 569 const PointerType *pointerType = Ptr->getType()->getAs<PointerType>(); 570 if (!pointerType) { 571 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 572 << Ptr->getType() << Ptr->getSourceRange(); 573 return ExprError(); 574 } 575 576 QualType AtomTy = pointerType->getPointeeType(); 577 if (!AtomTy->isAtomicType()) { 578 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 579 << Ptr->getType() << Ptr->getSourceRange(); 580 return ExprError(); 581 } 582 QualType ValType = AtomTy->getAs<AtomicType>()->getValueType(); 583 584 if ((Op == AtomicExpr::Add || Op == AtomicExpr::Sub) && 585 !ValType->isIntegerType() && !ValType->isPointerType()) { 586 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 587 << Ptr->getType() << Ptr->getSourceRange(); 588 return ExprError(); 589 } 590 591 if (!ValType->isIntegerType() && 592 (Op == AtomicExpr::And || Op == AtomicExpr::Or || Op == AtomicExpr::Xor)){ 593 Diag(DRE->getLocStart(), diag::err_atomic_op_logical_needs_atomic_int) 594 << Ptr->getType() << Ptr->getSourceRange(); 595 return ExprError(); 596 } 597 598 switch (ValType.getObjCLifetime()) { 599 case Qualifiers::OCL_None: 600 case Qualifiers::OCL_ExplicitNone: 601 // okay 602 break; 603 604 case Qualifiers::OCL_Weak: 605 case Qualifiers::OCL_Strong: 606 case Qualifiers::OCL_Autoreleasing: 607 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 608 << ValType << Ptr->getSourceRange(); 609 return ExprError(); 610 } 611 612 QualType ResultType = ValType; 613 if (Op == AtomicExpr::Store || Op == AtomicExpr::Init) 614 ResultType = Context.VoidTy; 615 else if (Op == AtomicExpr::CmpXchgWeak || Op == AtomicExpr::CmpXchgStrong) 616 ResultType = Context.BoolTy; 617 618 // The first argument --- the pointer --- has a fixed type; we 619 // deduce the types of the rest of the arguments accordingly. Walk 620 // the remaining arguments, converting them to the deduced value type. 621 for (unsigned i = 1; i != NumVals+NumOrders+1; ++i) { 622 ExprResult Arg = TheCall->getArg(i); 623 QualType Ty; 624 if (i < NumVals+1) { 625 // The second argument to a cmpxchg is a pointer to the data which will 626 // be exchanged. The second argument to a pointer add/subtract is the 627 // amount to add/subtract, which must be a ptrdiff_t. The third 628 // argument to a cmpxchg and the second argument in all other cases 629 // is the type of the value. 630 if (i == 1 && (Op == AtomicExpr::CmpXchgWeak || 631 Op == AtomicExpr::CmpXchgStrong)) 632 Ty = Context.getPointerType(ValType.getUnqualifiedType()); 633 else if (!ValType->isIntegerType() && 634 (Op == AtomicExpr::Add || Op == AtomicExpr::Sub)) 635 Ty = Context.getPointerDiffType(); 636 else 637 Ty = ValType; 638 } else { 639 // The order(s) are always converted to int. 640 Ty = Context.IntTy; 641 } 642 InitializedEntity Entity = 643 InitializedEntity::InitializeParameter(Context, Ty, false); 644 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 645 if (Arg.isInvalid()) 646 return true; 647 TheCall->setArg(i, Arg.get()); 648 } 649 650 SmallVector<Expr*, 5> SubExprs; 651 SubExprs.push_back(Ptr); 652 if (Op == AtomicExpr::Load) { 653 SubExprs.push_back(TheCall->getArg(1)); // Order 654 } else if (Op == AtomicExpr::Init) { 655 SubExprs.push_back(TheCall->getArg(1)); // Val1 656 } else if (Op != AtomicExpr::CmpXchgWeak && Op != AtomicExpr::CmpXchgStrong) { 657 SubExprs.push_back(TheCall->getArg(2)); // Order 658 SubExprs.push_back(TheCall->getArg(1)); // Val1 659 } else { 660 SubExprs.push_back(TheCall->getArg(3)); // Order 661 SubExprs.push_back(TheCall->getArg(1)); // Val1 662 SubExprs.push_back(TheCall->getArg(2)); // Val2 663 SubExprs.push_back(TheCall->getArg(4)); // OrderFail 664 } 665 666 return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 667 SubExprs.data(), SubExprs.size(), 668 ResultType, Op, 669 TheCall->getRParenLoc())); 670} 671 672 673/// checkBuiltinArgument - Given a call to a builtin function, perform 674/// normal type-checking on the given argument, updating the call in 675/// place. This is useful when a builtin function requires custom 676/// type-checking for some of its arguments but not necessarily all of 677/// them. 678/// 679/// Returns true on error. 680static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 681 FunctionDecl *Fn = E->getDirectCallee(); 682 assert(Fn && "builtin call without direct callee!"); 683 684 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 685 InitializedEntity Entity = 686 InitializedEntity::InitializeParameter(S.Context, Param); 687 688 ExprResult Arg = E->getArg(0); 689 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 690 if (Arg.isInvalid()) 691 return true; 692 693 E->setArg(ArgIndex, Arg.take()); 694 return false; 695} 696 697/// SemaBuiltinAtomicOverloaded - We have a call to a function like 698/// __sync_fetch_and_add, which is an overloaded function based on the pointer 699/// type of its first argument. The main ActOnCallExpr routines have already 700/// promoted the types of arguments because all of these calls are prototyped as 701/// void(...). 702/// 703/// This function goes through and does final semantic checking for these 704/// builtins, 705ExprResult 706Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 707 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 708 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 709 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 710 711 // Ensure that we have at least one argument to do type inference from. 712 if (TheCall->getNumArgs() < 1) { 713 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 714 << 0 << 1 << TheCall->getNumArgs() 715 << TheCall->getCallee()->getSourceRange(); 716 return ExprError(); 717 } 718 719 // Inspect the first argument of the atomic builtin. This should always be 720 // a pointer type, whose element is an integral scalar or pointer type. 721 // Because it is a pointer type, we don't have to worry about any implicit 722 // casts here. 723 // FIXME: We don't allow floating point scalars as input. 724 Expr *FirstArg = TheCall->getArg(0); 725 ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg); 726 if (FirstArgResult.isInvalid()) 727 return ExprError(); 728 FirstArg = FirstArgResult.take(); 729 TheCall->setArg(0, FirstArg); 730 731 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 732 if (!pointerType) { 733 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 734 << FirstArg->getType() << FirstArg->getSourceRange(); 735 return ExprError(); 736 } 737 738 QualType ValType = pointerType->getPointeeType(); 739 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 740 !ValType->isBlockPointerType()) { 741 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 742 << FirstArg->getType() << FirstArg->getSourceRange(); 743 return ExprError(); 744 } 745 746 switch (ValType.getObjCLifetime()) { 747 case Qualifiers::OCL_None: 748 case Qualifiers::OCL_ExplicitNone: 749 // okay 750 break; 751 752 case Qualifiers::OCL_Weak: 753 case Qualifiers::OCL_Strong: 754 case Qualifiers::OCL_Autoreleasing: 755 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 756 << ValType << FirstArg->getSourceRange(); 757 return ExprError(); 758 } 759 760 // Strip any qualifiers off ValType. 761 ValType = ValType.getUnqualifiedType(); 762 763 // The majority of builtins return a value, but a few have special return 764 // types, so allow them to override appropriately below. 765 QualType ResultType = ValType; 766 767 // We need to figure out which concrete builtin this maps onto. For example, 768 // __sync_fetch_and_add with a 2 byte object turns into 769 // __sync_fetch_and_add_2. 770#define BUILTIN_ROW(x) \ 771 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 772 Builtin::BI##x##_8, Builtin::BI##x##_16 } 773 774 static const unsigned BuiltinIndices[][5] = { 775 BUILTIN_ROW(__sync_fetch_and_add), 776 BUILTIN_ROW(__sync_fetch_and_sub), 777 BUILTIN_ROW(__sync_fetch_and_or), 778 BUILTIN_ROW(__sync_fetch_and_and), 779 BUILTIN_ROW(__sync_fetch_and_xor), 780 781 BUILTIN_ROW(__sync_add_and_fetch), 782 BUILTIN_ROW(__sync_sub_and_fetch), 783 BUILTIN_ROW(__sync_and_and_fetch), 784 BUILTIN_ROW(__sync_or_and_fetch), 785 BUILTIN_ROW(__sync_xor_and_fetch), 786 787 BUILTIN_ROW(__sync_val_compare_and_swap), 788 BUILTIN_ROW(__sync_bool_compare_and_swap), 789 BUILTIN_ROW(__sync_lock_test_and_set), 790 BUILTIN_ROW(__sync_lock_release), 791 BUILTIN_ROW(__sync_swap) 792 }; 793#undef BUILTIN_ROW 794 795 // Determine the index of the size. 796 unsigned SizeIndex; 797 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 798 case 1: SizeIndex = 0; break; 799 case 2: SizeIndex = 1; break; 800 case 4: SizeIndex = 2; break; 801 case 8: SizeIndex = 3; break; 802 case 16: SizeIndex = 4; break; 803 default: 804 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 805 << FirstArg->getType() << FirstArg->getSourceRange(); 806 return ExprError(); 807 } 808 809 // Each of these builtins has one pointer argument, followed by some number of 810 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 811 // that we ignore. Find out which row of BuiltinIndices to read from as well 812 // as the number of fixed args. 813 unsigned BuiltinID = FDecl->getBuiltinID(); 814 unsigned BuiltinIndex, NumFixed = 1; 815 switch (BuiltinID) { 816 default: llvm_unreachable("Unknown overloaded atomic builtin!"); 817 case Builtin::BI__sync_fetch_and_add: 818 case Builtin::BI__sync_fetch_and_add_1: 819 case Builtin::BI__sync_fetch_and_add_2: 820 case Builtin::BI__sync_fetch_and_add_4: 821 case Builtin::BI__sync_fetch_and_add_8: 822 case Builtin::BI__sync_fetch_and_add_16: 823 BuiltinIndex = 0; 824 break; 825 826 case Builtin::BI__sync_fetch_and_sub: 827 case Builtin::BI__sync_fetch_and_sub_1: 828 case Builtin::BI__sync_fetch_and_sub_2: 829 case Builtin::BI__sync_fetch_and_sub_4: 830 case Builtin::BI__sync_fetch_and_sub_8: 831 case Builtin::BI__sync_fetch_and_sub_16: 832 BuiltinIndex = 1; 833 break; 834 835 case Builtin::BI__sync_fetch_and_or: 836 case Builtin::BI__sync_fetch_and_or_1: 837 case Builtin::BI__sync_fetch_and_or_2: 838 case Builtin::BI__sync_fetch_and_or_4: 839 case Builtin::BI__sync_fetch_and_or_8: 840 case Builtin::BI__sync_fetch_and_or_16: 841 BuiltinIndex = 2; 842 break; 843 844 case Builtin::BI__sync_fetch_and_and: 845 case Builtin::BI__sync_fetch_and_and_1: 846 case Builtin::BI__sync_fetch_and_and_2: 847 case Builtin::BI__sync_fetch_and_and_4: 848 case Builtin::BI__sync_fetch_and_and_8: 849 case Builtin::BI__sync_fetch_and_and_16: 850 BuiltinIndex = 3; 851 break; 852 853 case Builtin::BI__sync_fetch_and_xor: 854 case Builtin::BI__sync_fetch_and_xor_1: 855 case Builtin::BI__sync_fetch_and_xor_2: 856 case Builtin::BI__sync_fetch_and_xor_4: 857 case Builtin::BI__sync_fetch_and_xor_8: 858 case Builtin::BI__sync_fetch_and_xor_16: 859 BuiltinIndex = 4; 860 break; 861 862 case Builtin::BI__sync_add_and_fetch: 863 case Builtin::BI__sync_add_and_fetch_1: 864 case Builtin::BI__sync_add_and_fetch_2: 865 case Builtin::BI__sync_add_and_fetch_4: 866 case Builtin::BI__sync_add_and_fetch_8: 867 case Builtin::BI__sync_add_and_fetch_16: 868 BuiltinIndex = 5; 869 break; 870 871 case Builtin::BI__sync_sub_and_fetch: 872 case Builtin::BI__sync_sub_and_fetch_1: 873 case Builtin::BI__sync_sub_and_fetch_2: 874 case Builtin::BI__sync_sub_and_fetch_4: 875 case Builtin::BI__sync_sub_and_fetch_8: 876 case Builtin::BI__sync_sub_and_fetch_16: 877 BuiltinIndex = 6; 878 break; 879 880 case Builtin::BI__sync_and_and_fetch: 881 case Builtin::BI__sync_and_and_fetch_1: 882 case Builtin::BI__sync_and_and_fetch_2: 883 case Builtin::BI__sync_and_and_fetch_4: 884 case Builtin::BI__sync_and_and_fetch_8: 885 case Builtin::BI__sync_and_and_fetch_16: 886 BuiltinIndex = 7; 887 break; 888 889 case Builtin::BI__sync_or_and_fetch: 890 case Builtin::BI__sync_or_and_fetch_1: 891 case Builtin::BI__sync_or_and_fetch_2: 892 case Builtin::BI__sync_or_and_fetch_4: 893 case Builtin::BI__sync_or_and_fetch_8: 894 case Builtin::BI__sync_or_and_fetch_16: 895 BuiltinIndex = 8; 896 break; 897 898 case Builtin::BI__sync_xor_and_fetch: 899 case Builtin::BI__sync_xor_and_fetch_1: 900 case Builtin::BI__sync_xor_and_fetch_2: 901 case Builtin::BI__sync_xor_and_fetch_4: 902 case Builtin::BI__sync_xor_and_fetch_8: 903 case Builtin::BI__sync_xor_and_fetch_16: 904 BuiltinIndex = 9; 905 break; 906 907 case Builtin::BI__sync_val_compare_and_swap: 908 case Builtin::BI__sync_val_compare_and_swap_1: 909 case Builtin::BI__sync_val_compare_and_swap_2: 910 case Builtin::BI__sync_val_compare_and_swap_4: 911 case Builtin::BI__sync_val_compare_and_swap_8: 912 case Builtin::BI__sync_val_compare_and_swap_16: 913 BuiltinIndex = 10; 914 NumFixed = 2; 915 break; 916 917 case Builtin::BI__sync_bool_compare_and_swap: 918 case Builtin::BI__sync_bool_compare_and_swap_1: 919 case Builtin::BI__sync_bool_compare_and_swap_2: 920 case Builtin::BI__sync_bool_compare_and_swap_4: 921 case Builtin::BI__sync_bool_compare_and_swap_8: 922 case Builtin::BI__sync_bool_compare_and_swap_16: 923 BuiltinIndex = 11; 924 NumFixed = 2; 925 ResultType = Context.BoolTy; 926 break; 927 928 case Builtin::BI__sync_lock_test_and_set: 929 case Builtin::BI__sync_lock_test_and_set_1: 930 case Builtin::BI__sync_lock_test_and_set_2: 931 case Builtin::BI__sync_lock_test_and_set_4: 932 case Builtin::BI__sync_lock_test_and_set_8: 933 case Builtin::BI__sync_lock_test_and_set_16: 934 BuiltinIndex = 12; 935 break; 936 937 case Builtin::BI__sync_lock_release: 938 case Builtin::BI__sync_lock_release_1: 939 case Builtin::BI__sync_lock_release_2: 940 case Builtin::BI__sync_lock_release_4: 941 case Builtin::BI__sync_lock_release_8: 942 case Builtin::BI__sync_lock_release_16: 943 BuiltinIndex = 13; 944 NumFixed = 0; 945 ResultType = Context.VoidTy; 946 break; 947 948 case Builtin::BI__sync_swap: 949 case Builtin::BI__sync_swap_1: 950 case Builtin::BI__sync_swap_2: 951 case Builtin::BI__sync_swap_4: 952 case Builtin::BI__sync_swap_8: 953 case Builtin::BI__sync_swap_16: 954 BuiltinIndex = 14; 955 break; 956 } 957 958 // Now that we know how many fixed arguments we expect, first check that we 959 // have at least that many. 960 if (TheCall->getNumArgs() < 1+NumFixed) { 961 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 962 << 0 << 1+NumFixed << TheCall->getNumArgs() 963 << TheCall->getCallee()->getSourceRange(); 964 return ExprError(); 965 } 966 967 // Get the decl for the concrete builtin from this, we can tell what the 968 // concrete integer type we should convert to is. 969 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 970 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 971 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 972 FunctionDecl *NewBuiltinDecl = 973 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 974 TUScope, false, DRE->getLocStart())); 975 976 // The first argument --- the pointer --- has a fixed type; we 977 // deduce the types of the rest of the arguments accordingly. Walk 978 // the remaining arguments, converting them to the deduced value type. 979 for (unsigned i = 0; i != NumFixed; ++i) { 980 ExprResult Arg = TheCall->getArg(i+1); 981 982 // GCC does an implicit conversion to the pointer or integer ValType. This 983 // can fail in some cases (1i -> int**), check for this error case now. 984 // Initialize the argument. 985 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context, 986 ValType, /*consume*/ false); 987 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 988 if (Arg.isInvalid()) 989 return ExprError(); 990 991 // Okay, we have something that *can* be converted to the right type. Check 992 // to see if there is a potentially weird extension going on here. This can 993 // happen when you do an atomic operation on something like an char* and 994 // pass in 42. The 42 gets converted to char. This is even more strange 995 // for things like 45.123 -> char, etc. 996 // FIXME: Do this check. 997 TheCall->setArg(i+1, Arg.take()); 998 } 999 1000 ASTContext& Context = this->getASTContext(); 1001 1002 // Create a new DeclRefExpr to refer to the new decl. 1003 DeclRefExpr* NewDRE = DeclRefExpr::Create( 1004 Context, 1005 DRE->getQualifierLoc(), 1006 NewBuiltinDecl, 1007 DRE->getLocation(), 1008 NewBuiltinDecl->getType(), 1009 DRE->getValueKind()); 1010 1011 // Set the callee in the CallExpr. 1012 // FIXME: This leaks the original parens and implicit casts. 1013 ExprResult PromotedCall = UsualUnaryConversions(NewDRE); 1014 if (PromotedCall.isInvalid()) 1015 return ExprError(); 1016 TheCall->setCallee(PromotedCall.take()); 1017 1018 // Change the result type of the call to match the original value type. This 1019 // is arbitrary, but the codegen for these builtins ins design to handle it 1020 // gracefully. 1021 TheCall->setType(ResultType); 1022 1023 return move(TheCallResult); 1024} 1025 1026/// CheckObjCString - Checks that the argument to the builtin 1027/// CFString constructor is correct 1028/// Note: It might also make sense to do the UTF-16 conversion here (would 1029/// simplify the backend). 1030bool Sema::CheckObjCString(Expr *Arg) { 1031 Arg = Arg->IgnoreParenCasts(); 1032 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 1033 1034 if (!Literal || !Literal->isAscii()) { 1035 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 1036 << Arg->getSourceRange(); 1037 return true; 1038 } 1039 1040 if (Literal->containsNonAsciiOrNull()) { 1041 StringRef String = Literal->getString(); 1042 unsigned NumBytes = String.size(); 1043 SmallVector<UTF16, 128> ToBuf(NumBytes); 1044 const UTF8 *FromPtr = (UTF8 *)String.data(); 1045 UTF16 *ToPtr = &ToBuf[0]; 1046 1047 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 1048 &ToPtr, ToPtr + NumBytes, 1049 strictConversion); 1050 // Check for conversion failure. 1051 if (Result != conversionOK) 1052 Diag(Arg->getLocStart(), 1053 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 1054 } 1055 return false; 1056} 1057 1058/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 1059/// Emit an error and return true on failure, return false on success. 1060bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 1061 Expr *Fn = TheCall->getCallee(); 1062 if (TheCall->getNumArgs() > 2) { 1063 Diag(TheCall->getArg(2)->getLocStart(), 1064 diag::err_typecheck_call_too_many_args) 1065 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1066 << Fn->getSourceRange() 1067 << SourceRange(TheCall->getArg(2)->getLocStart(), 1068 (*(TheCall->arg_end()-1))->getLocEnd()); 1069 return true; 1070 } 1071 1072 if (TheCall->getNumArgs() < 2) { 1073 return Diag(TheCall->getLocEnd(), 1074 diag::err_typecheck_call_too_few_args_at_least) 1075 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 1076 } 1077 1078 // Type-check the first argument normally. 1079 if (checkBuiltinArgument(*this, TheCall, 0)) 1080 return true; 1081 1082 // Determine whether the current function is variadic or not. 1083 BlockScopeInfo *CurBlock = getCurBlock(); 1084 bool isVariadic; 1085 if (CurBlock) 1086 isVariadic = CurBlock->TheDecl->isVariadic(); 1087 else if (FunctionDecl *FD = getCurFunctionDecl()) 1088 isVariadic = FD->isVariadic(); 1089 else 1090 isVariadic = getCurMethodDecl()->isVariadic(); 1091 1092 if (!isVariadic) { 1093 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 1094 return true; 1095 } 1096 1097 // Verify that the second argument to the builtin is the last argument of the 1098 // current function or method. 1099 bool SecondArgIsLastNamedArgument = false; 1100 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 1101 1102 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 1103 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 1104 // FIXME: This isn't correct for methods (results in bogus warning). 1105 // Get the last formal in the current function. 1106 const ParmVarDecl *LastArg; 1107 if (CurBlock) 1108 LastArg = *(CurBlock->TheDecl->param_end()-1); 1109 else if (FunctionDecl *FD = getCurFunctionDecl()) 1110 LastArg = *(FD->param_end()-1); 1111 else 1112 LastArg = *(getCurMethodDecl()->param_end()-1); 1113 SecondArgIsLastNamedArgument = PV == LastArg; 1114 } 1115 } 1116 1117 if (!SecondArgIsLastNamedArgument) 1118 Diag(TheCall->getArg(1)->getLocStart(), 1119 diag::warn_second_parameter_of_va_start_not_last_named_argument); 1120 return false; 1121} 1122 1123/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 1124/// friends. This is declared to take (...), so we have to check everything. 1125bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 1126 if (TheCall->getNumArgs() < 2) 1127 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1128 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 1129 if (TheCall->getNumArgs() > 2) 1130 return Diag(TheCall->getArg(2)->getLocStart(), 1131 diag::err_typecheck_call_too_many_args) 1132 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1133 << SourceRange(TheCall->getArg(2)->getLocStart(), 1134 (*(TheCall->arg_end()-1))->getLocEnd()); 1135 1136 ExprResult OrigArg0 = TheCall->getArg(0); 1137 ExprResult OrigArg1 = TheCall->getArg(1); 1138 1139 // Do standard promotions between the two arguments, returning their common 1140 // type. 1141 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 1142 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 1143 return true; 1144 1145 // Make sure any conversions are pushed back into the call; this is 1146 // type safe since unordered compare builtins are declared as "_Bool 1147 // foo(...)". 1148 TheCall->setArg(0, OrigArg0.get()); 1149 TheCall->setArg(1, OrigArg1.get()); 1150 1151 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 1152 return false; 1153 1154 // If the common type isn't a real floating type, then the arguments were 1155 // invalid for this operation. 1156 if (!Res->isRealFloatingType()) 1157 return Diag(OrigArg0.get()->getLocStart(), 1158 diag::err_typecheck_call_invalid_ordered_compare) 1159 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 1160 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 1161 1162 return false; 1163} 1164 1165/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 1166/// __builtin_isnan and friends. This is declared to take (...), so we have 1167/// to check everything. We expect the last argument to be a floating point 1168/// value. 1169bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 1170 if (TheCall->getNumArgs() < NumArgs) 1171 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1172 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 1173 if (TheCall->getNumArgs() > NumArgs) 1174 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 1175 diag::err_typecheck_call_too_many_args) 1176 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 1177 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 1178 (*(TheCall->arg_end()-1))->getLocEnd()); 1179 1180 Expr *OrigArg = TheCall->getArg(NumArgs-1); 1181 1182 if (OrigArg->isTypeDependent()) 1183 return false; 1184 1185 // This operation requires a non-_Complex floating-point number. 1186 if (!OrigArg->getType()->isRealFloatingType()) 1187 return Diag(OrigArg->getLocStart(), 1188 diag::err_typecheck_call_invalid_unary_fp) 1189 << OrigArg->getType() << OrigArg->getSourceRange(); 1190 1191 // If this is an implicit conversion from float -> double, remove it. 1192 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 1193 Expr *CastArg = Cast->getSubExpr(); 1194 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 1195 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 1196 "promotion from float to double is the only expected cast here"); 1197 Cast->setSubExpr(0); 1198 TheCall->setArg(NumArgs-1, CastArg); 1199 OrigArg = CastArg; 1200 } 1201 } 1202 1203 return false; 1204} 1205 1206/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1207// This is declared to take (...), so we have to check everything. 1208ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1209 if (TheCall->getNumArgs() < 2) 1210 return ExprError(Diag(TheCall->getLocEnd(), 1211 diag::err_typecheck_call_too_few_args_at_least) 1212 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1213 << TheCall->getSourceRange()); 1214 1215 // Determine which of the following types of shufflevector we're checking: 1216 // 1) unary, vector mask: (lhs, mask) 1217 // 2) binary, vector mask: (lhs, rhs, mask) 1218 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1219 QualType resType = TheCall->getArg(0)->getType(); 1220 unsigned numElements = 0; 1221 1222 if (!TheCall->getArg(0)->isTypeDependent() && 1223 !TheCall->getArg(1)->isTypeDependent()) { 1224 QualType LHSType = TheCall->getArg(0)->getType(); 1225 QualType RHSType = TheCall->getArg(1)->getType(); 1226 1227 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 1228 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 1229 << SourceRange(TheCall->getArg(0)->getLocStart(), 1230 TheCall->getArg(1)->getLocEnd()); 1231 return ExprError(); 1232 } 1233 1234 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1235 unsigned numResElements = TheCall->getNumArgs() - 2; 1236 1237 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1238 // with mask. If so, verify that RHS is an integer vector type with the 1239 // same number of elts as lhs. 1240 if (TheCall->getNumArgs() == 2) { 1241 if (!RHSType->hasIntegerRepresentation() || 1242 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1243 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1244 << SourceRange(TheCall->getArg(1)->getLocStart(), 1245 TheCall->getArg(1)->getLocEnd()); 1246 numResElements = numElements; 1247 } 1248 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1249 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1250 << SourceRange(TheCall->getArg(0)->getLocStart(), 1251 TheCall->getArg(1)->getLocEnd()); 1252 return ExprError(); 1253 } else if (numElements != numResElements) { 1254 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1255 resType = Context.getVectorType(eltType, numResElements, 1256 VectorType::GenericVector); 1257 } 1258 } 1259 1260 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1261 if (TheCall->getArg(i)->isTypeDependent() || 1262 TheCall->getArg(i)->isValueDependent()) 1263 continue; 1264 1265 llvm::APSInt Result(32); 1266 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1267 return ExprError(Diag(TheCall->getLocStart(), 1268 diag::err_shufflevector_nonconstant_argument) 1269 << TheCall->getArg(i)->getSourceRange()); 1270 1271 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1272 return ExprError(Diag(TheCall->getLocStart(), 1273 diag::err_shufflevector_argument_too_large) 1274 << TheCall->getArg(i)->getSourceRange()); 1275 } 1276 1277 SmallVector<Expr*, 32> exprs; 1278 1279 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1280 exprs.push_back(TheCall->getArg(i)); 1281 TheCall->setArg(i, 0); 1282 } 1283 1284 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 1285 exprs.size(), resType, 1286 TheCall->getCallee()->getLocStart(), 1287 TheCall->getRParenLoc())); 1288} 1289 1290/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1291// This is declared to take (const void*, ...) and can take two 1292// optional constant int args. 1293bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1294 unsigned NumArgs = TheCall->getNumArgs(); 1295 1296 if (NumArgs > 3) 1297 return Diag(TheCall->getLocEnd(), 1298 diag::err_typecheck_call_too_many_args_at_most) 1299 << 0 /*function call*/ << 3 << NumArgs 1300 << TheCall->getSourceRange(); 1301 1302 // Argument 0 is checked for us and the remaining arguments must be 1303 // constant integers. 1304 for (unsigned i = 1; i != NumArgs; ++i) { 1305 Expr *Arg = TheCall->getArg(i); 1306 1307 llvm::APSInt Result; 1308 if (SemaBuiltinConstantArg(TheCall, i, Result)) 1309 return true; 1310 1311 // FIXME: gcc issues a warning and rewrites these to 0. These 1312 // seems especially odd for the third argument since the default 1313 // is 3. 1314 if (i == 1) { 1315 if (Result.getLimitedValue() > 1) 1316 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1317 << "0" << "1" << Arg->getSourceRange(); 1318 } else { 1319 if (Result.getLimitedValue() > 3) 1320 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1321 << "0" << "3" << Arg->getSourceRange(); 1322 } 1323 } 1324 1325 return false; 1326} 1327 1328/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1329/// TheCall is a constant expression. 1330bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1331 llvm::APSInt &Result) { 1332 Expr *Arg = TheCall->getArg(ArgNum); 1333 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1334 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1335 1336 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 1337 1338 if (!Arg->isIntegerConstantExpr(Result, Context)) 1339 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 1340 << FDecl->getDeclName() << Arg->getSourceRange(); 1341 1342 return false; 1343} 1344 1345/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 1346/// int type). This simply type checks that type is one of the defined 1347/// constants (0-3). 1348// For compatibility check 0-3, llvm only handles 0 and 2. 1349bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 1350 llvm::APSInt Result; 1351 1352 // Check constant-ness first. 1353 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1354 return true; 1355 1356 Expr *Arg = TheCall->getArg(1); 1357 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 1358 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1359 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1360 } 1361 1362 return false; 1363} 1364 1365/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1366/// This checks that val is a constant 1. 1367bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1368 Expr *Arg = TheCall->getArg(1); 1369 llvm::APSInt Result; 1370 1371 // TODO: This is less than ideal. Overload this to take a value. 1372 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1373 return true; 1374 1375 if (Result != 1) 1376 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1377 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1378 1379 return false; 1380} 1381 1382// Handle i > 1 ? "x" : "y", recursively. 1383bool Sema::SemaCheckStringLiteral(const Expr *E, Expr **Args, 1384 unsigned NumArgs, bool HasVAListArg, 1385 unsigned format_idx, unsigned firstDataArg, 1386 bool isPrintf, bool inFunctionCall) { 1387 tryAgain: 1388 if (E->isTypeDependent() || E->isValueDependent()) 1389 return false; 1390 1391 E = E->IgnoreParens(); 1392 1393 switch (E->getStmtClass()) { 1394 case Stmt::BinaryConditionalOperatorClass: 1395 case Stmt::ConditionalOperatorClass: { 1396 const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E); 1397 return SemaCheckStringLiteral(C->getTrueExpr(), Args, NumArgs, HasVAListArg, 1398 format_idx, firstDataArg, isPrintf, 1399 inFunctionCall) 1400 && SemaCheckStringLiteral(C->getFalseExpr(), Args, NumArgs, HasVAListArg, 1401 format_idx, firstDataArg, isPrintf, 1402 inFunctionCall); 1403 } 1404 1405 case Stmt::IntegerLiteralClass: 1406 // Technically -Wformat-nonliteral does not warn about this case. 1407 // The behavior of printf and friends in this case is implementation 1408 // dependent. Ideally if the format string cannot be null then 1409 // it should have a 'nonnull' attribute in the function prototype. 1410 return true; 1411 1412 case Stmt::ImplicitCastExprClass: { 1413 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 1414 goto tryAgain; 1415 } 1416 1417 case Stmt::OpaqueValueExprClass: 1418 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 1419 E = src; 1420 goto tryAgain; 1421 } 1422 return false; 1423 1424 case Stmt::PredefinedExprClass: 1425 // While __func__, etc., are technically not string literals, they 1426 // cannot contain format specifiers and thus are not a security 1427 // liability. 1428 return true; 1429 1430 case Stmt::DeclRefExprClass: { 1431 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1432 1433 // As an exception, do not flag errors for variables binding to 1434 // const string literals. 1435 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1436 bool isConstant = false; 1437 QualType T = DR->getType(); 1438 1439 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1440 isConstant = AT->getElementType().isConstant(Context); 1441 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1442 isConstant = T.isConstant(Context) && 1443 PT->getPointeeType().isConstant(Context); 1444 } 1445 1446 if (isConstant) { 1447 if (const Expr *Init = VD->getAnyInitializer()) 1448 return SemaCheckStringLiteral(Init, Args, NumArgs, 1449 HasVAListArg, format_idx, firstDataArg, 1450 isPrintf, /*inFunctionCall*/false); 1451 } 1452 1453 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1454 // special check to see if the format string is a function parameter 1455 // of the function calling the printf function. If the function 1456 // has an attribute indicating it is a printf-like function, then we 1457 // should suppress warnings concerning non-literals being used in a call 1458 // to a vprintf function. For example: 1459 // 1460 // void 1461 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1462 // va_list ap; 1463 // va_start(ap, fmt); 1464 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1465 // ... 1466 // 1467 // 1468 // FIXME: We don't have full attribute support yet, so just check to see 1469 // if the argument is a DeclRefExpr that references a parameter. We'll 1470 // add proper support for checking the attribute later. 1471 if (HasVAListArg) 1472 if (isa<ParmVarDecl>(VD)) 1473 return true; 1474 } 1475 1476 return false; 1477 } 1478 1479 case Stmt::CallExprClass: { 1480 const CallExpr *CE = cast<CallExpr>(E); 1481 if (const ImplicitCastExpr *ICE 1482 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 1483 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 1484 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 1485 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 1486 unsigned ArgIndex = FA->getFormatIdx(); 1487 const Expr *Arg = CE->getArg(ArgIndex - 1); 1488 1489 return SemaCheckStringLiteral(Arg, Args, NumArgs, HasVAListArg, 1490 format_idx, firstDataArg, isPrintf, 1491 inFunctionCall); 1492 } 1493 } 1494 } 1495 } 1496 1497 return false; 1498 } 1499 case Stmt::ObjCStringLiteralClass: 1500 case Stmt::StringLiteralClass: { 1501 const StringLiteral *StrE = NULL; 1502 1503 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1504 StrE = ObjCFExpr->getString(); 1505 else 1506 StrE = cast<StringLiteral>(E); 1507 1508 if (StrE) { 1509 CheckFormatString(StrE, E, Args, NumArgs, HasVAListArg, format_idx, 1510 firstDataArg, isPrintf, inFunctionCall); 1511 return true; 1512 } 1513 1514 return false; 1515 } 1516 1517 default: 1518 return false; 1519 } 1520} 1521 1522void 1523Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1524 const Expr * const *ExprArgs, 1525 SourceLocation CallSiteLoc) { 1526 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1527 e = NonNull->args_end(); 1528 i != e; ++i) { 1529 const Expr *ArgExpr = ExprArgs[*i]; 1530 if (ArgExpr->isNullPointerConstant(Context, 1531 Expr::NPC_ValueDependentIsNotNull)) 1532 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 1533 } 1534} 1535 1536/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar 1537/// functions) for correct use of format strings. 1538void Sema::CheckFormatArguments(const FormatAttr *Format, CallExpr *TheCall) { 1539 bool IsCXXMember = false; 1540 // The way the format attribute works in GCC, the implicit this argument 1541 // of member functions is counted. However, it doesn't appear in our own 1542 // lists, so decrement format_idx in that case. 1543 if (isa<CXXMemberCallExpr>(TheCall)) { 1544 const CXXMethodDecl *method_decl = 1545 dyn_cast<CXXMethodDecl>(TheCall->getCalleeDecl()); 1546 IsCXXMember = method_decl && method_decl->isInstance(); 1547 } 1548 CheckFormatArguments(Format, TheCall->getArgs(), TheCall->getNumArgs(), 1549 IsCXXMember, TheCall->getRParenLoc(), 1550 TheCall->getCallee()->getSourceRange()); 1551} 1552 1553void Sema::CheckFormatArguments(const FormatAttr *Format, Expr **Args, 1554 unsigned NumArgs, bool IsCXXMember, 1555 SourceLocation Loc, SourceRange Range) { 1556 const bool b = Format->getType() == "scanf"; 1557 if (b || CheckablePrintfAttr(Format, Args, NumArgs, IsCXXMember)) { 1558 bool HasVAListArg = Format->getFirstArg() == 0; 1559 unsigned format_idx = Format->getFormatIdx() - 1; 1560 unsigned firstDataArg = HasVAListArg ? 0 : Format->getFirstArg() - 1; 1561 if (IsCXXMember) { 1562 if (format_idx == 0) 1563 return; 1564 --format_idx; 1565 if(firstDataArg != 0) 1566 --firstDataArg; 1567 } 1568 CheckPrintfScanfArguments(Args, NumArgs, HasVAListArg, format_idx, 1569 firstDataArg, !b, Loc, Range); 1570 } 1571} 1572 1573void Sema::CheckPrintfScanfArguments(Expr **Args, unsigned NumArgs, 1574 bool HasVAListArg, unsigned format_idx, 1575 unsigned firstDataArg, bool isPrintf, 1576 SourceLocation Loc, SourceRange Range) { 1577 // CHECK: printf/scanf-like function is called with no format string. 1578 if (format_idx >= NumArgs) { 1579 Diag(Loc, diag::warn_missing_format_string) << Range; 1580 return; 1581 } 1582 1583 const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts(); 1584 1585 // CHECK: format string is not a string literal. 1586 // 1587 // Dynamically generated format strings are difficult to 1588 // automatically vet at compile time. Requiring that format strings 1589 // are string literals: (1) permits the checking of format strings by 1590 // the compiler and thereby (2) can practically remove the source of 1591 // many format string exploits. 1592 1593 // Format string can be either ObjC string (e.g. @"%d") or 1594 // C string (e.g. "%d") 1595 // ObjC string uses the same format specifiers as C string, so we can use 1596 // the same format string checking logic for both ObjC and C strings. 1597 if (SemaCheckStringLiteral(OrigFormatExpr, Args, NumArgs, HasVAListArg, 1598 format_idx, firstDataArg, isPrintf)) 1599 return; // Literal format string found, check done! 1600 1601 // If there are no arguments specified, warn with -Wformat-security, otherwise 1602 // warn only with -Wformat-nonliteral. 1603 if (NumArgs == format_idx+1) 1604 Diag(Args[format_idx]->getLocStart(), 1605 diag::warn_format_nonliteral_noargs) 1606 << OrigFormatExpr->getSourceRange(); 1607 else 1608 Diag(Args[format_idx]->getLocStart(), 1609 diag::warn_format_nonliteral) 1610 << OrigFormatExpr->getSourceRange(); 1611} 1612 1613namespace { 1614class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1615protected: 1616 Sema &S; 1617 const StringLiteral *FExpr; 1618 const Expr *OrigFormatExpr; 1619 const unsigned FirstDataArg; 1620 const unsigned NumDataArgs; 1621 const bool IsObjCLiteral; 1622 const char *Beg; // Start of format string. 1623 const bool HasVAListArg; 1624 const Expr * const *Args; 1625 const unsigned NumArgs; 1626 unsigned FormatIdx; 1627 llvm::BitVector CoveredArgs; 1628 bool usesPositionalArgs; 1629 bool atFirstArg; 1630 bool inFunctionCall; 1631public: 1632 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1633 const Expr *origFormatExpr, unsigned firstDataArg, 1634 unsigned numDataArgs, bool isObjCLiteral, 1635 const char *beg, bool hasVAListArg, 1636 Expr **args, unsigned numArgs, 1637 unsigned formatIdx, bool inFunctionCall) 1638 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1639 FirstDataArg(firstDataArg), 1640 NumDataArgs(numDataArgs), 1641 IsObjCLiteral(isObjCLiteral), Beg(beg), 1642 HasVAListArg(hasVAListArg), 1643 Args(args), NumArgs(numArgs), FormatIdx(formatIdx), 1644 usesPositionalArgs(false), atFirstArg(true), 1645 inFunctionCall(inFunctionCall) { 1646 CoveredArgs.resize(numDataArgs); 1647 CoveredArgs.reset(); 1648 } 1649 1650 void DoneProcessing(); 1651 1652 void HandleIncompleteSpecifier(const char *startSpecifier, 1653 unsigned specifierLen); 1654 1655 virtual void HandleInvalidPosition(const char *startSpecifier, 1656 unsigned specifierLen, 1657 analyze_format_string::PositionContext p); 1658 1659 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1660 1661 void HandleNullChar(const char *nullCharacter); 1662 1663 template <typename Range> 1664 static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall, 1665 const Expr *ArgumentExpr, 1666 PartialDiagnostic PDiag, 1667 SourceLocation StringLoc, 1668 bool IsStringLocation, Range StringRange, 1669 FixItHint Fixit = FixItHint()); 1670 1671protected: 1672 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1673 const char *startSpec, 1674 unsigned specifierLen, 1675 const char *csStart, unsigned csLen); 1676 1677 void HandlePositionalNonpositionalArgs(SourceLocation Loc, 1678 const char *startSpec, 1679 unsigned specifierLen); 1680 1681 SourceRange getFormatStringRange(); 1682 CharSourceRange getSpecifierRange(const char *startSpecifier, 1683 unsigned specifierLen); 1684 SourceLocation getLocationOfByte(const char *x); 1685 1686 const Expr *getDataArg(unsigned i) const; 1687 1688 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1689 const analyze_format_string::ConversionSpecifier &CS, 1690 const char *startSpecifier, unsigned specifierLen, 1691 unsigned argIndex); 1692 1693 template <typename Range> 1694 void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc, 1695 bool IsStringLocation, Range StringRange, 1696 FixItHint Fixit = FixItHint()); 1697 1698 void CheckPositionalAndNonpositionalArgs( 1699 const analyze_format_string::FormatSpecifier *FS); 1700}; 1701} 1702 1703SourceRange CheckFormatHandler::getFormatStringRange() { 1704 return OrigFormatExpr->getSourceRange(); 1705} 1706 1707CharSourceRange CheckFormatHandler:: 1708getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1709 SourceLocation Start = getLocationOfByte(startSpecifier); 1710 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1711 1712 // Advance the end SourceLocation by one due to half-open ranges. 1713 End = End.getLocWithOffset(1); 1714 1715 return CharSourceRange::getCharRange(Start, End); 1716} 1717 1718SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 1719 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1720} 1721 1722void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 1723 unsigned specifierLen){ 1724 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier), 1725 getLocationOfByte(startSpecifier), 1726 /*IsStringLocation*/true, 1727 getSpecifierRange(startSpecifier, specifierLen)); 1728} 1729 1730void 1731CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1732 analyze_format_string::PositionContext p) { 1733 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier) 1734 << (unsigned) p, 1735 getLocationOfByte(startPos), /*IsStringLocation*/true, 1736 getSpecifierRange(startPos, posLen)); 1737} 1738 1739void CheckFormatHandler::HandleZeroPosition(const char *startPos, 1740 unsigned posLen) { 1741 EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier), 1742 getLocationOfByte(startPos), 1743 /*IsStringLocation*/true, 1744 getSpecifierRange(startPos, posLen)); 1745} 1746 1747void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 1748 if (!IsObjCLiteral) { 1749 // The presence of a null character is likely an error. 1750 EmitFormatDiagnostic( 1751 S.PDiag(diag::warn_printf_format_string_contains_null_char), 1752 getLocationOfByte(nullCharacter), /*IsStringLocation*/true, 1753 getFormatStringRange()); 1754 } 1755} 1756 1757const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 1758 return Args[FirstDataArg + i]; 1759} 1760 1761void CheckFormatHandler::DoneProcessing() { 1762 // Does the number of data arguments exceed the number of 1763 // format conversions in the format string? 1764 if (!HasVAListArg) { 1765 // Find any arguments that weren't covered. 1766 CoveredArgs.flip(); 1767 signed notCoveredArg = CoveredArgs.find_first(); 1768 if (notCoveredArg >= 0) { 1769 assert((unsigned)notCoveredArg < NumDataArgs); 1770 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used), 1771 getDataArg((unsigned) notCoveredArg)->getLocStart(), 1772 /*IsStringLocation*/false, getFormatStringRange()); 1773 } 1774 } 1775} 1776 1777bool 1778CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 1779 SourceLocation Loc, 1780 const char *startSpec, 1781 unsigned specifierLen, 1782 const char *csStart, 1783 unsigned csLen) { 1784 1785 bool keepGoing = true; 1786 if (argIndex < NumDataArgs) { 1787 // Consider the argument coverered, even though the specifier doesn't 1788 // make sense. 1789 CoveredArgs.set(argIndex); 1790 } 1791 else { 1792 // If argIndex exceeds the number of data arguments we 1793 // don't issue a warning because that is just a cascade of warnings (and 1794 // they may have intended '%%' anyway). We don't want to continue processing 1795 // the format string after this point, however, as we will like just get 1796 // gibberish when trying to match arguments. 1797 keepGoing = false; 1798 } 1799 1800 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion) 1801 << StringRef(csStart, csLen), 1802 Loc, /*IsStringLocation*/true, 1803 getSpecifierRange(startSpec, specifierLen)); 1804 1805 return keepGoing; 1806} 1807 1808void 1809CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc, 1810 const char *startSpec, 1811 unsigned specifierLen) { 1812 EmitFormatDiagnostic( 1813 S.PDiag(diag::warn_format_mix_positional_nonpositional_args), 1814 Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen)); 1815} 1816 1817bool 1818CheckFormatHandler::CheckNumArgs( 1819 const analyze_format_string::FormatSpecifier &FS, 1820 const analyze_format_string::ConversionSpecifier &CS, 1821 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 1822 1823 if (argIndex >= NumDataArgs) { 1824 PartialDiagnostic PDiag = FS.usesPositionalArg() 1825 ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args) 1826 << (argIndex+1) << NumDataArgs) 1827 : S.PDiag(diag::warn_printf_insufficient_data_args); 1828 EmitFormatDiagnostic( 1829 PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true, 1830 getSpecifierRange(startSpecifier, specifierLen)); 1831 return false; 1832 } 1833 return true; 1834} 1835 1836template<typename Range> 1837void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag, 1838 SourceLocation Loc, 1839 bool IsStringLocation, 1840 Range StringRange, 1841 FixItHint FixIt) { 1842 EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag, 1843 Loc, IsStringLocation, StringRange, FixIt); 1844} 1845 1846/// \brief If the format string is not within the funcion call, emit a note 1847/// so that the function call and string are in diagnostic messages. 1848/// 1849/// \param inFunctionCall if true, the format string is within the function 1850/// call and only one diagnostic message will be produced. Otherwise, an 1851/// extra note will be emitted pointing to location of the format string. 1852/// 1853/// \param ArgumentExpr the expression that is passed as the format string 1854/// argument in the function call. Used for getting locations when two 1855/// diagnostics are emitted. 1856/// 1857/// \param PDiag the callee should already have provided any strings for the 1858/// diagnostic message. This function only adds locations and fixits 1859/// to diagnostics. 1860/// 1861/// \param Loc primary location for diagnostic. If two diagnostics are 1862/// required, one will be at Loc and a new SourceLocation will be created for 1863/// the other one. 1864/// 1865/// \param IsStringLocation if true, Loc points to the format string should be 1866/// used for the note. Otherwise, Loc points to the argument list and will 1867/// be used with PDiag. 1868/// 1869/// \param StringRange some or all of the string to highlight. This is 1870/// templated so it can accept either a CharSourceRange or a SourceRange. 1871/// 1872/// \param Fixit optional fix it hint for the format string. 1873template<typename Range> 1874void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall, 1875 const Expr *ArgumentExpr, 1876 PartialDiagnostic PDiag, 1877 SourceLocation Loc, 1878 bool IsStringLocation, 1879 Range StringRange, 1880 FixItHint FixIt) { 1881 if (InFunctionCall) 1882 S.Diag(Loc, PDiag) << StringRange << FixIt; 1883 else { 1884 S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag) 1885 << ArgumentExpr->getSourceRange(); 1886 S.Diag(IsStringLocation ? Loc : StringRange.getBegin(), 1887 diag::note_format_string_defined) 1888 << StringRange << FixIt; 1889 } 1890} 1891 1892//===--- CHECK: Printf format string checking ------------------------------===// 1893 1894namespace { 1895class CheckPrintfHandler : public CheckFormatHandler { 1896public: 1897 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1898 const Expr *origFormatExpr, unsigned firstDataArg, 1899 unsigned numDataArgs, bool isObjCLiteral, 1900 const char *beg, bool hasVAListArg, 1901 Expr **Args, unsigned NumArgs, 1902 unsigned formatIdx, bool inFunctionCall) 1903 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1904 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1905 Args, NumArgs, formatIdx, inFunctionCall) {} 1906 1907 1908 bool HandleInvalidPrintfConversionSpecifier( 1909 const analyze_printf::PrintfSpecifier &FS, 1910 const char *startSpecifier, 1911 unsigned specifierLen); 1912 1913 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 1914 const char *startSpecifier, 1915 unsigned specifierLen); 1916 1917 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 1918 const char *startSpecifier, unsigned specifierLen); 1919 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 1920 const analyze_printf::OptionalAmount &Amt, 1921 unsigned type, 1922 const char *startSpecifier, unsigned specifierLen); 1923 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1924 const analyze_printf::OptionalFlag &flag, 1925 const char *startSpecifier, unsigned specifierLen); 1926 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 1927 const analyze_printf::OptionalFlag &ignoredFlag, 1928 const analyze_printf::OptionalFlag &flag, 1929 const char *startSpecifier, unsigned specifierLen); 1930}; 1931} 1932 1933bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 1934 const analyze_printf::PrintfSpecifier &FS, 1935 const char *startSpecifier, 1936 unsigned specifierLen) { 1937 const analyze_printf::PrintfConversionSpecifier &CS = 1938 FS.getConversionSpecifier(); 1939 1940 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1941 getLocationOfByte(CS.getStart()), 1942 startSpecifier, specifierLen, 1943 CS.getStart(), CS.getLength()); 1944} 1945 1946bool CheckPrintfHandler::HandleAmount( 1947 const analyze_format_string::OptionalAmount &Amt, 1948 unsigned k, const char *startSpecifier, 1949 unsigned specifierLen) { 1950 1951 if (Amt.hasDataArgument()) { 1952 if (!HasVAListArg) { 1953 unsigned argIndex = Amt.getArgIndex(); 1954 if (argIndex >= NumDataArgs) { 1955 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg) 1956 << k, 1957 getLocationOfByte(Amt.getStart()), 1958 /*IsStringLocation*/true, 1959 getSpecifierRange(startSpecifier, specifierLen)); 1960 // Don't do any more checking. We will just emit 1961 // spurious errors. 1962 return false; 1963 } 1964 1965 // Type check the data argument. It should be an 'int'. 1966 // Although not in conformance with C99, we also allow the argument to be 1967 // an 'unsigned int' as that is a reasonably safe case. GCC also 1968 // doesn't emit a warning for that case. 1969 CoveredArgs.set(argIndex); 1970 const Expr *Arg = getDataArg(argIndex); 1971 QualType T = Arg->getType(); 1972 1973 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1974 assert(ATR.isValid()); 1975 1976 if (!ATR.matchesType(S.Context, T)) { 1977 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type) 1978 << k << ATR.getRepresentativeTypeName(S.Context) 1979 << T << Arg->getSourceRange(), 1980 getLocationOfByte(Amt.getStart()), 1981 /*IsStringLocation*/true, 1982 getSpecifierRange(startSpecifier, specifierLen)); 1983 // Don't do any more checking. We will just emit 1984 // spurious errors. 1985 return false; 1986 } 1987 } 1988 } 1989 return true; 1990} 1991 1992void CheckPrintfHandler::HandleInvalidAmount( 1993 const analyze_printf::PrintfSpecifier &FS, 1994 const analyze_printf::OptionalAmount &Amt, 1995 unsigned type, 1996 const char *startSpecifier, 1997 unsigned specifierLen) { 1998 const analyze_printf::PrintfConversionSpecifier &CS = 1999 FS.getConversionSpecifier(); 2000 2001 FixItHint fixit = 2002 Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant 2003 ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 2004 Amt.getConstantLength())) 2005 : FixItHint(); 2006 2007 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount) 2008 << type << CS.toString(), 2009 getLocationOfByte(Amt.getStart()), 2010 /*IsStringLocation*/true, 2011 getSpecifierRange(startSpecifier, specifierLen), 2012 fixit); 2013} 2014 2015void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2016 const analyze_printf::OptionalFlag &flag, 2017 const char *startSpecifier, 2018 unsigned specifierLen) { 2019 // Warn about pointless flag with a fixit removal. 2020 const analyze_printf::PrintfConversionSpecifier &CS = 2021 FS.getConversionSpecifier(); 2022 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag) 2023 << flag.toString() << CS.toString(), 2024 getLocationOfByte(flag.getPosition()), 2025 /*IsStringLocation*/true, 2026 getSpecifierRange(startSpecifier, specifierLen), 2027 FixItHint::CreateRemoval( 2028 getSpecifierRange(flag.getPosition(), 1))); 2029} 2030 2031void CheckPrintfHandler::HandleIgnoredFlag( 2032 const analyze_printf::PrintfSpecifier &FS, 2033 const analyze_printf::OptionalFlag &ignoredFlag, 2034 const analyze_printf::OptionalFlag &flag, 2035 const char *startSpecifier, 2036 unsigned specifierLen) { 2037 // Warn about ignored flag with a fixit removal. 2038 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag) 2039 << ignoredFlag.toString() << flag.toString(), 2040 getLocationOfByte(ignoredFlag.getPosition()), 2041 /*IsStringLocation*/true, 2042 getSpecifierRange(startSpecifier, specifierLen), 2043 FixItHint::CreateRemoval( 2044 getSpecifierRange(ignoredFlag.getPosition(), 1))); 2045} 2046 2047bool 2048CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 2049 &FS, 2050 const char *startSpecifier, 2051 unsigned specifierLen) { 2052 2053 using namespace analyze_format_string; 2054 using namespace analyze_printf; 2055 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 2056 2057 if (FS.consumesDataArgument()) { 2058 if (atFirstArg) { 2059 atFirstArg = false; 2060 usesPositionalArgs = FS.usesPositionalArg(); 2061 } 2062 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2063 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2064 startSpecifier, specifierLen); 2065 return false; 2066 } 2067 } 2068 2069 // First check if the field width, precision, and conversion specifier 2070 // have matching data arguments. 2071 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 2072 startSpecifier, specifierLen)) { 2073 return false; 2074 } 2075 2076 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 2077 startSpecifier, specifierLen)) { 2078 return false; 2079 } 2080 2081 if (!CS.consumesDataArgument()) { 2082 // FIXME: Technically specifying a precision or field width here 2083 // makes no sense. Worth issuing a warning at some point. 2084 return true; 2085 } 2086 2087 // Consume the argument. 2088 unsigned argIndex = FS.getArgIndex(); 2089 if (argIndex < NumDataArgs) { 2090 // The check to see if the argIndex is valid will come later. 2091 // We set the bit here because we may exit early from this 2092 // function if we encounter some other error. 2093 CoveredArgs.set(argIndex); 2094 } 2095 2096 // Check for using an Objective-C specific conversion specifier 2097 // in a non-ObjC literal. 2098 if (!IsObjCLiteral && CS.isObjCArg()) { 2099 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 2100 specifierLen); 2101 } 2102 2103 // Check for invalid use of field width 2104 if (!FS.hasValidFieldWidth()) { 2105 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 2106 startSpecifier, specifierLen); 2107 } 2108 2109 // Check for invalid use of precision 2110 if (!FS.hasValidPrecision()) { 2111 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 2112 startSpecifier, specifierLen); 2113 } 2114 2115 // Check each flag does not conflict with any other component. 2116 if (!FS.hasValidThousandsGroupingPrefix()) 2117 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 2118 if (!FS.hasValidLeadingZeros()) 2119 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 2120 if (!FS.hasValidPlusPrefix()) 2121 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 2122 if (!FS.hasValidSpacePrefix()) 2123 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 2124 if (!FS.hasValidAlternativeForm()) 2125 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 2126 if (!FS.hasValidLeftJustified()) 2127 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 2128 2129 // Check that flags are not ignored by another flag 2130 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 2131 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 2132 startSpecifier, specifierLen); 2133 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 2134 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 2135 startSpecifier, specifierLen); 2136 2137 // Check the length modifier is valid with the given conversion specifier. 2138 const LengthModifier &LM = FS.getLengthModifier(); 2139 if (!FS.hasValidLengthModifier()) 2140 EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length) 2141 << LM.toString() << CS.toString(), 2142 getLocationOfByte(LM.getStart()), 2143 /*IsStringLocation*/true, 2144 getSpecifierRange(startSpecifier, specifierLen), 2145 FixItHint::CreateRemoval( 2146 getSpecifierRange(LM.getStart(), 2147 LM.getLength()))); 2148 2149 // Are we using '%n'? 2150 if (CS.getKind() == ConversionSpecifier::nArg) { 2151 // Issue a warning about this being a possible security issue. 2152 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_write_back), 2153 getLocationOfByte(CS.getStart()), 2154 /*IsStringLocation*/true, 2155 getSpecifierRange(startSpecifier, specifierLen)); 2156 // Continue checking the other format specifiers. 2157 return true; 2158 } 2159 2160 // The remaining checks depend on the data arguments. 2161 if (HasVAListArg) 2162 return true; 2163 2164 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2165 return false; 2166 2167 // Now type check the data expression that matches the 2168 // format specifier. 2169 const Expr *Ex = getDataArg(argIndex); 2170 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context); 2171 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 2172 // Check if we didn't match because of an implicit cast from a 'char' 2173 // or 'short' to an 'int'. This is done because printf is a varargs 2174 // function. 2175 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 2176 if (ICE->getType() == S.Context.IntTy) { 2177 // All further checking is done on the subexpression. 2178 Ex = ICE->getSubExpr(); 2179 if (ATR.matchesType(S.Context, Ex->getType())) 2180 return true; 2181 } 2182 2183 // We may be able to offer a FixItHint if it is a supported type. 2184 PrintfSpecifier fixedFS = FS; 2185 bool success = fixedFS.fixType(Ex->getType(), S.getLangOptions()); 2186 2187 if (success) { 2188 // Get the fix string from the fixed format specifier 2189 llvm::SmallString<128> buf; 2190 llvm::raw_svector_ostream os(buf); 2191 fixedFS.toString(os); 2192 2193 EmitFormatDiagnostic( 2194 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2195 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2196 << Ex->getSourceRange(), 2197 getLocationOfByte(CS.getStart()), 2198 /*IsStringLocation*/true, 2199 getSpecifierRange(startSpecifier, specifierLen), 2200 FixItHint::CreateReplacement( 2201 getSpecifierRange(startSpecifier, specifierLen), 2202 os.str())); 2203 } 2204 else { 2205 S.Diag(getLocationOfByte(CS.getStart()), 2206 diag::warn_printf_conversion_argument_type_mismatch) 2207 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2208 << getSpecifierRange(startSpecifier, specifierLen) 2209 << Ex->getSourceRange(); 2210 } 2211 } 2212 2213 return true; 2214} 2215 2216//===--- CHECK: Scanf format string checking ------------------------------===// 2217 2218namespace { 2219class CheckScanfHandler : public CheckFormatHandler { 2220public: 2221 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 2222 const Expr *origFormatExpr, unsigned firstDataArg, 2223 unsigned numDataArgs, bool isObjCLiteral, 2224 const char *beg, bool hasVAListArg, 2225 Expr **Args, unsigned NumArgs, 2226 unsigned formatIdx, bool inFunctionCall) 2227 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2228 numDataArgs, isObjCLiteral, beg, hasVAListArg, 2229 Args, NumArgs, formatIdx, inFunctionCall) {} 2230 2231 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 2232 const char *startSpecifier, 2233 unsigned specifierLen); 2234 2235 bool HandleInvalidScanfConversionSpecifier( 2236 const analyze_scanf::ScanfSpecifier &FS, 2237 const char *startSpecifier, 2238 unsigned specifierLen); 2239 2240 void HandleIncompleteScanList(const char *start, const char *end); 2241}; 2242} 2243 2244void CheckScanfHandler::HandleIncompleteScanList(const char *start, 2245 const char *end) { 2246 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 2247 getLocationOfByte(end), /*IsStringLocation*/true, 2248 getSpecifierRange(start, end - start)); 2249} 2250 2251bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 2252 const analyze_scanf::ScanfSpecifier &FS, 2253 const char *startSpecifier, 2254 unsigned specifierLen) { 2255 2256 const analyze_scanf::ScanfConversionSpecifier &CS = 2257 FS.getConversionSpecifier(); 2258 2259 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2260 getLocationOfByte(CS.getStart()), 2261 startSpecifier, specifierLen, 2262 CS.getStart(), CS.getLength()); 2263} 2264 2265bool CheckScanfHandler::HandleScanfSpecifier( 2266 const analyze_scanf::ScanfSpecifier &FS, 2267 const char *startSpecifier, 2268 unsigned specifierLen) { 2269 2270 using namespace analyze_scanf; 2271 using namespace analyze_format_string; 2272 2273 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 2274 2275 // Handle case where '%' and '*' don't consume an argument. These shouldn't 2276 // be used to decide if we are using positional arguments consistently. 2277 if (FS.consumesDataArgument()) { 2278 if (atFirstArg) { 2279 atFirstArg = false; 2280 usesPositionalArgs = FS.usesPositionalArg(); 2281 } 2282 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2283 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2284 startSpecifier, specifierLen); 2285 return false; 2286 } 2287 } 2288 2289 // Check if the field with is non-zero. 2290 const OptionalAmount &Amt = FS.getFieldWidth(); 2291 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 2292 if (Amt.getConstantAmount() == 0) { 2293 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 2294 Amt.getConstantLength()); 2295 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 2296 getLocationOfByte(Amt.getStart()), 2297 /*IsStringLocation*/true, R, 2298 FixItHint::CreateRemoval(R)); 2299 } 2300 } 2301 2302 if (!FS.consumesDataArgument()) { 2303 // FIXME: Technically specifying a precision or field width here 2304 // makes no sense. Worth issuing a warning at some point. 2305 return true; 2306 } 2307 2308 // Consume the argument. 2309 unsigned argIndex = FS.getArgIndex(); 2310 if (argIndex < NumDataArgs) { 2311 // The check to see if the argIndex is valid will come later. 2312 // We set the bit here because we may exit early from this 2313 // function if we encounter some other error. 2314 CoveredArgs.set(argIndex); 2315 } 2316 2317 // Check the length modifier is valid with the given conversion specifier. 2318 const LengthModifier &LM = FS.getLengthModifier(); 2319 if (!FS.hasValidLengthModifier()) { 2320 S.Diag(getLocationOfByte(LM.getStart()), 2321 diag::warn_format_nonsensical_length) 2322 << LM.toString() << CS.toString() 2323 << getSpecifierRange(startSpecifier, specifierLen) 2324 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 2325 LM.getLength())); 2326 } 2327 2328 // The remaining checks depend on the data arguments. 2329 if (HasVAListArg) 2330 return true; 2331 2332 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2333 return false; 2334 2335 // Check that the argument type matches the format specifier. 2336 const Expr *Ex = getDataArg(argIndex); 2337 const analyze_scanf::ScanfArgTypeResult &ATR = FS.getArgType(S.Context); 2338 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 2339 ScanfSpecifier fixedFS = FS; 2340 bool success = fixedFS.fixType(Ex->getType(), S.getLangOptions()); 2341 2342 if (success) { 2343 // Get the fix string from the fixed format specifier. 2344 llvm::SmallString<128> buf; 2345 llvm::raw_svector_ostream os(buf); 2346 fixedFS.toString(os); 2347 2348 EmitFormatDiagnostic( 2349 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2350 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2351 << Ex->getSourceRange(), 2352 getLocationOfByte(CS.getStart()), 2353 /*IsStringLocation*/true, 2354 getSpecifierRange(startSpecifier, specifierLen), 2355 FixItHint::CreateReplacement( 2356 getSpecifierRange(startSpecifier, specifierLen), 2357 os.str())); 2358 } else { 2359 S.Diag(getLocationOfByte(CS.getStart()), 2360 diag::warn_printf_conversion_argument_type_mismatch) 2361 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2362 << getSpecifierRange(startSpecifier, specifierLen) 2363 << Ex->getSourceRange(); 2364 } 2365 } 2366 2367 return true; 2368} 2369 2370void Sema::CheckFormatString(const StringLiteral *FExpr, 2371 const Expr *OrigFormatExpr, 2372 Expr **Args, unsigned NumArgs, 2373 bool HasVAListArg, unsigned format_idx, 2374 unsigned firstDataArg, bool isPrintf, 2375 bool inFunctionCall) { 2376 2377 // CHECK: is the format string a wide literal? 2378 if (!FExpr->isAscii()) { 2379 CheckFormatHandler::EmitFormatDiagnostic( 2380 *this, inFunctionCall, Args[format_idx], 2381 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 2382 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2383 return; 2384 } 2385 2386 // Str - The format string. NOTE: this is NOT null-terminated! 2387 StringRef StrRef = FExpr->getString(); 2388 const char *Str = StrRef.data(); 2389 unsigned StrLen = StrRef.size(); 2390 const unsigned numDataArgs = NumArgs - firstDataArg; 2391 2392 // CHECK: empty format string? 2393 if (StrLen == 0 && numDataArgs > 0) { 2394 CheckFormatHandler::EmitFormatDiagnostic( 2395 *this, inFunctionCall, Args[format_idx], 2396 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 2397 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2398 return; 2399 } 2400 2401 if (isPrintf) { 2402 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2403 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2404 Str, HasVAListArg, Args, NumArgs, format_idx, 2405 inFunctionCall); 2406 2407 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen, 2408 getLangOptions())) 2409 H.DoneProcessing(); 2410 } 2411 else { 2412 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2413 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2414 Str, HasVAListArg, Args, NumArgs, format_idx, 2415 inFunctionCall); 2416 2417 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen, 2418 getLangOptions())) 2419 H.DoneProcessing(); 2420 } 2421} 2422 2423//===--- CHECK: Standard memory functions ---------------------------------===// 2424 2425/// \brief Determine whether the given type is a dynamic class type (e.g., 2426/// whether it has a vtable). 2427static bool isDynamicClassType(QualType T) { 2428 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 2429 if (CXXRecordDecl *Definition = Record->getDefinition()) 2430 if (Definition->isDynamicClass()) 2431 return true; 2432 2433 return false; 2434} 2435 2436/// \brief If E is a sizeof expression, returns its argument expression, 2437/// otherwise returns NULL. 2438static const Expr *getSizeOfExprArg(const Expr* E) { 2439 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2440 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2441 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 2442 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 2443 2444 return 0; 2445} 2446 2447/// \brief If E is a sizeof expression, returns its argument type. 2448static QualType getSizeOfArgType(const Expr* E) { 2449 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2450 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2451 if (SizeOf->getKind() == clang::UETT_SizeOf) 2452 return SizeOf->getTypeOfArgument(); 2453 2454 return QualType(); 2455} 2456 2457/// \brief Check for dangerous or invalid arguments to memset(). 2458/// 2459/// This issues warnings on known problematic, dangerous or unspecified 2460/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 2461/// function calls. 2462/// 2463/// \param Call The call expression to diagnose. 2464void Sema::CheckMemaccessArguments(const CallExpr *Call, 2465 unsigned BId, 2466 IdentifierInfo *FnName) { 2467 assert(BId != 0); 2468 2469 // It is possible to have a non-standard definition of memset. Validate 2470 // we have enough arguments, and if not, abort further checking. 2471 unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3); 2472 if (Call->getNumArgs() < ExpectedNumArgs) 2473 return; 2474 2475 unsigned LastArg = (BId == Builtin::BImemset || 2476 BId == Builtin::BIstrndup ? 1 : 2); 2477 unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2); 2478 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 2479 2480 // We have special checking when the length is a sizeof expression. 2481 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 2482 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 2483 llvm::FoldingSetNodeID SizeOfArgID; 2484 2485 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 2486 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 2487 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 2488 2489 QualType DestTy = Dest->getType(); 2490 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 2491 QualType PointeeTy = DestPtrTy->getPointeeType(); 2492 2493 // Never warn about void type pointers. This can be used to suppress 2494 // false positives. 2495 if (PointeeTy->isVoidType()) 2496 continue; 2497 2498 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 2499 // actually comparing the expressions for equality. Because computing the 2500 // expression IDs can be expensive, we only do this if the diagnostic is 2501 // enabled. 2502 if (SizeOfArg && 2503 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 2504 SizeOfArg->getExprLoc())) { 2505 // We only compute IDs for expressions if the warning is enabled, and 2506 // cache the sizeof arg's ID. 2507 if (SizeOfArgID == llvm::FoldingSetNodeID()) 2508 SizeOfArg->Profile(SizeOfArgID, Context, true); 2509 llvm::FoldingSetNodeID DestID; 2510 Dest->Profile(DestID, Context, true); 2511 if (DestID == SizeOfArgID) { 2512 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 2513 // over sizeof(src) as well. 2514 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 2515 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 2516 if (UnaryOp->getOpcode() == UO_AddrOf) 2517 ActionIdx = 1; // If its an address-of operator, just remove it. 2518 if (Context.getTypeSize(PointeeTy) == Context.getCharWidth()) 2519 ActionIdx = 2; // If the pointee's size is sizeof(char), 2520 // suggest an explicit length. 2521 unsigned DestSrcSelect = 2522 (BId == Builtin::BIstrndup ? 1 : ArgIdx); 2523 DiagRuntimeBehavior(SizeOfArg->getExprLoc(), Dest, 2524 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 2525 << FnName << DestSrcSelect << ActionIdx 2526 << Dest->getSourceRange() 2527 << SizeOfArg->getSourceRange()); 2528 break; 2529 } 2530 } 2531 2532 // Also check for cases where the sizeof argument is the exact same 2533 // type as the memory argument, and where it points to a user-defined 2534 // record type. 2535 if (SizeOfArgTy != QualType()) { 2536 if (PointeeTy->isRecordType() && 2537 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 2538 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 2539 PDiag(diag::warn_sizeof_pointer_type_memaccess) 2540 << FnName << SizeOfArgTy << ArgIdx 2541 << PointeeTy << Dest->getSourceRange() 2542 << LenExpr->getSourceRange()); 2543 break; 2544 } 2545 } 2546 2547 // Always complain about dynamic classes. 2548 if (isDynamicClassType(PointeeTy)) { 2549 2550 unsigned OperationType = 0; 2551 // "overwritten" if we're warning about the destination for any call 2552 // but memcmp; otherwise a verb appropriate to the call. 2553 if (ArgIdx != 0 || BId == Builtin::BImemcmp) { 2554 if (BId == Builtin::BImemcpy) 2555 OperationType = 1; 2556 else if(BId == Builtin::BImemmove) 2557 OperationType = 2; 2558 else if (BId == Builtin::BImemcmp) 2559 OperationType = 3; 2560 } 2561 2562 DiagRuntimeBehavior( 2563 Dest->getExprLoc(), Dest, 2564 PDiag(diag::warn_dyn_class_memaccess) 2565 << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx) 2566 << FnName << PointeeTy 2567 << OperationType 2568 << Call->getCallee()->getSourceRange()); 2569 } else if (PointeeTy.hasNonTrivialObjCLifetime() && 2570 BId != Builtin::BImemset) 2571 DiagRuntimeBehavior( 2572 Dest->getExprLoc(), Dest, 2573 PDiag(diag::warn_arc_object_memaccess) 2574 << ArgIdx << FnName << PointeeTy 2575 << Call->getCallee()->getSourceRange()); 2576 else 2577 continue; 2578 2579 DiagRuntimeBehavior( 2580 Dest->getExprLoc(), Dest, 2581 PDiag(diag::note_bad_memaccess_silence) 2582 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 2583 break; 2584 } 2585 } 2586} 2587 2588// A little helper routine: ignore addition and subtraction of integer literals. 2589// This intentionally does not ignore all integer constant expressions because 2590// we don't want to remove sizeof(). 2591static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 2592 Ex = Ex->IgnoreParenCasts(); 2593 2594 for (;;) { 2595 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 2596 if (!BO || !BO->isAdditiveOp()) 2597 break; 2598 2599 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 2600 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 2601 2602 if (isa<IntegerLiteral>(RHS)) 2603 Ex = LHS; 2604 else if (isa<IntegerLiteral>(LHS)) 2605 Ex = RHS; 2606 else 2607 break; 2608 } 2609 2610 return Ex; 2611} 2612 2613// Warn if the user has made the 'size' argument to strlcpy or strlcat 2614// be the size of the source, instead of the destination. 2615void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 2616 IdentifierInfo *FnName) { 2617 2618 // Don't crash if the user has the wrong number of arguments 2619 if (Call->getNumArgs() != 3) 2620 return; 2621 2622 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 2623 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 2624 const Expr *CompareWithSrc = NULL; 2625 2626 // Look for 'strlcpy(dst, x, sizeof(x))' 2627 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 2628 CompareWithSrc = Ex; 2629 else { 2630 // Look for 'strlcpy(dst, x, strlen(x))' 2631 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 2632 if (SizeCall->isBuiltinCall() == Builtin::BIstrlen 2633 && SizeCall->getNumArgs() == 1) 2634 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 2635 } 2636 } 2637 2638 if (!CompareWithSrc) 2639 return; 2640 2641 // Determine if the argument to sizeof/strlen is equal to the source 2642 // argument. In principle there's all kinds of things you could do 2643 // here, for instance creating an == expression and evaluating it with 2644 // EvaluateAsBooleanCondition, but this uses a more direct technique: 2645 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 2646 if (!SrcArgDRE) 2647 return; 2648 2649 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 2650 if (!CompareWithSrcDRE || 2651 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 2652 return; 2653 2654 const Expr *OriginalSizeArg = Call->getArg(2); 2655 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 2656 << OriginalSizeArg->getSourceRange() << FnName; 2657 2658 // Output a FIXIT hint if the destination is an array (rather than a 2659 // pointer to an array). This could be enhanced to handle some 2660 // pointers if we know the actual size, like if DstArg is 'array+2' 2661 // we could say 'sizeof(array)-2'. 2662 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 2663 QualType DstArgTy = DstArg->getType(); 2664 2665 // Only handle constant-sized or VLAs, but not flexible members. 2666 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 2667 // Only issue the FIXIT for arrays of size > 1. 2668 if (CAT->getSize().getSExtValue() <= 1) 2669 return; 2670 } else if (!DstArgTy->isVariableArrayType()) { 2671 return; 2672 } 2673 2674 llvm::SmallString<128> sizeString; 2675 llvm::raw_svector_ostream OS(sizeString); 2676 OS << "sizeof("; 2677 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 2678 OS << ")"; 2679 2680 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 2681 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 2682 OS.str()); 2683} 2684 2685//===--- CHECK: Return Address of Stack Variable --------------------------===// 2686 2687static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars); 2688static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars); 2689 2690/// CheckReturnStackAddr - Check if a return statement returns the address 2691/// of a stack variable. 2692void 2693Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 2694 SourceLocation ReturnLoc) { 2695 2696 Expr *stackE = 0; 2697 SmallVector<DeclRefExpr *, 8> refVars; 2698 2699 // Perform checking for returned stack addresses, local blocks, 2700 // label addresses or references to temporaries. 2701 if (lhsType->isPointerType() || 2702 (!getLangOptions().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 2703 stackE = EvalAddr(RetValExp, refVars); 2704 } else if (lhsType->isReferenceType()) { 2705 stackE = EvalVal(RetValExp, refVars); 2706 } 2707 2708 if (stackE == 0) 2709 return; // Nothing suspicious was found. 2710 2711 SourceLocation diagLoc; 2712 SourceRange diagRange; 2713 if (refVars.empty()) { 2714 diagLoc = stackE->getLocStart(); 2715 diagRange = stackE->getSourceRange(); 2716 } else { 2717 // We followed through a reference variable. 'stackE' contains the 2718 // problematic expression but we will warn at the return statement pointing 2719 // at the reference variable. We will later display the "trail" of 2720 // reference variables using notes. 2721 diagLoc = refVars[0]->getLocStart(); 2722 diagRange = refVars[0]->getSourceRange(); 2723 } 2724 2725 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 2726 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 2727 : diag::warn_ret_stack_addr) 2728 << DR->getDecl()->getDeclName() << diagRange; 2729 } else if (isa<BlockExpr>(stackE)) { // local block. 2730 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 2731 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 2732 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 2733 } else { // local temporary. 2734 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 2735 : diag::warn_ret_local_temp_addr) 2736 << diagRange; 2737 } 2738 2739 // Display the "trail" of reference variables that we followed until we 2740 // found the problematic expression using notes. 2741 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 2742 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 2743 // If this var binds to another reference var, show the range of the next 2744 // var, otherwise the var binds to the problematic expression, in which case 2745 // show the range of the expression. 2746 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 2747 : stackE->getSourceRange(); 2748 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 2749 << VD->getDeclName() << range; 2750 } 2751} 2752 2753/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 2754/// check if the expression in a return statement evaluates to an address 2755/// to a location on the stack, a local block, an address of a label, or a 2756/// reference to local temporary. The recursion is used to traverse the 2757/// AST of the return expression, with recursion backtracking when we 2758/// encounter a subexpression that (1) clearly does not lead to one of the 2759/// above problematic expressions (2) is something we cannot determine leads to 2760/// a problematic expression based on such local checking. 2761/// 2762/// Both EvalAddr and EvalVal follow through reference variables to evaluate 2763/// the expression that they point to. Such variables are added to the 2764/// 'refVars' vector so that we know what the reference variable "trail" was. 2765/// 2766/// EvalAddr processes expressions that are pointers that are used as 2767/// references (and not L-values). EvalVal handles all other values. 2768/// At the base case of the recursion is a check for the above problematic 2769/// expressions. 2770/// 2771/// This implementation handles: 2772/// 2773/// * pointer-to-pointer casts 2774/// * implicit conversions from array references to pointers 2775/// * taking the address of fields 2776/// * arbitrary interplay between "&" and "*" operators 2777/// * pointer arithmetic from an address of a stack variable 2778/// * taking the address of an array element where the array is on the stack 2779static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 2780 if (E->isTypeDependent()) 2781 return NULL; 2782 2783 // We should only be called for evaluating pointer expressions. 2784 assert((E->getType()->isAnyPointerType() || 2785 E->getType()->isBlockPointerType() || 2786 E->getType()->isObjCQualifiedIdType()) && 2787 "EvalAddr only works on pointers"); 2788 2789 E = E->IgnoreParens(); 2790 2791 // Our "symbolic interpreter" is just a dispatch off the currently 2792 // viewed AST node. We then recursively traverse the AST by calling 2793 // EvalAddr and EvalVal appropriately. 2794 switch (E->getStmtClass()) { 2795 case Stmt::DeclRefExprClass: { 2796 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2797 2798 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2799 // If this is a reference variable, follow through to the expression that 2800 // it points to. 2801 if (V->hasLocalStorage() && 2802 V->getType()->isReferenceType() && V->hasInit()) { 2803 // Add the reference variable to the "trail". 2804 refVars.push_back(DR); 2805 return EvalAddr(V->getInit(), refVars); 2806 } 2807 2808 return NULL; 2809 } 2810 2811 case Stmt::UnaryOperatorClass: { 2812 // The only unary operator that make sense to handle here 2813 // is AddrOf. All others don't make sense as pointers. 2814 UnaryOperator *U = cast<UnaryOperator>(E); 2815 2816 if (U->getOpcode() == UO_AddrOf) 2817 return EvalVal(U->getSubExpr(), refVars); 2818 else 2819 return NULL; 2820 } 2821 2822 case Stmt::BinaryOperatorClass: { 2823 // Handle pointer arithmetic. All other binary operators are not valid 2824 // in this context. 2825 BinaryOperator *B = cast<BinaryOperator>(E); 2826 BinaryOperatorKind op = B->getOpcode(); 2827 2828 if (op != BO_Add && op != BO_Sub) 2829 return NULL; 2830 2831 Expr *Base = B->getLHS(); 2832 2833 // Determine which argument is the real pointer base. It could be 2834 // the RHS argument instead of the LHS. 2835 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 2836 2837 assert (Base->getType()->isPointerType()); 2838 return EvalAddr(Base, refVars); 2839 } 2840 2841 // For conditional operators we need to see if either the LHS or RHS are 2842 // valid DeclRefExpr*s. If one of them is valid, we return it. 2843 case Stmt::ConditionalOperatorClass: { 2844 ConditionalOperator *C = cast<ConditionalOperator>(E); 2845 2846 // Handle the GNU extension for missing LHS. 2847 if (Expr *lhsExpr = C->getLHS()) { 2848 // In C++, we can have a throw-expression, which has 'void' type. 2849 if (!lhsExpr->getType()->isVoidType()) 2850 if (Expr* LHS = EvalAddr(lhsExpr, refVars)) 2851 return LHS; 2852 } 2853 2854 // In C++, we can have a throw-expression, which has 'void' type. 2855 if (C->getRHS()->getType()->isVoidType()) 2856 return NULL; 2857 2858 return EvalAddr(C->getRHS(), refVars); 2859 } 2860 2861 case Stmt::BlockExprClass: 2862 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 2863 return E; // local block. 2864 return NULL; 2865 2866 case Stmt::AddrLabelExprClass: 2867 return E; // address of label. 2868 2869 case Stmt::ExprWithCleanupsClass: 2870 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars); 2871 2872 // For casts, we need to handle conversions from arrays to 2873 // pointer values, and pointer-to-pointer conversions. 2874 case Stmt::ImplicitCastExprClass: 2875 case Stmt::CStyleCastExprClass: 2876 case Stmt::CXXFunctionalCastExprClass: 2877 case Stmt::ObjCBridgedCastExprClass: { 2878 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 2879 QualType T = SubExpr->getType(); 2880 2881 if (SubExpr->getType()->isPointerType() || 2882 SubExpr->getType()->isBlockPointerType() || 2883 SubExpr->getType()->isObjCQualifiedIdType()) 2884 return EvalAddr(SubExpr, refVars); 2885 else if (T->isArrayType()) 2886 return EvalVal(SubExpr, refVars); 2887 else 2888 return 0; 2889 } 2890 2891 // C++ casts. For dynamic casts, static casts, and const casts, we 2892 // are always converting from a pointer-to-pointer, so we just blow 2893 // through the cast. In the case the dynamic cast doesn't fail (and 2894 // return NULL), we take the conservative route and report cases 2895 // where we return the address of a stack variable. For Reinterpre 2896 // FIXME: The comment about is wrong; we're not always converting 2897 // from pointer to pointer. I'm guessing that this code should also 2898 // handle references to objects. 2899 case Stmt::CXXStaticCastExprClass: 2900 case Stmt::CXXDynamicCastExprClass: 2901 case Stmt::CXXConstCastExprClass: 2902 case Stmt::CXXReinterpretCastExprClass: { 2903 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 2904 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 2905 return EvalAddr(S, refVars); 2906 else 2907 return NULL; 2908 } 2909 2910 case Stmt::MaterializeTemporaryExprClass: 2911 if (Expr *Result = EvalAddr( 2912 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 2913 refVars)) 2914 return Result; 2915 2916 return E; 2917 2918 // Everything else: we simply don't reason about them. 2919 default: 2920 return NULL; 2921 } 2922} 2923 2924 2925/// EvalVal - This function is complements EvalAddr in the mutual recursion. 2926/// See the comments for EvalAddr for more details. 2927static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 2928do { 2929 // We should only be called for evaluating non-pointer expressions, or 2930 // expressions with a pointer type that are not used as references but instead 2931 // are l-values (e.g., DeclRefExpr with a pointer type). 2932 2933 // Our "symbolic interpreter" is just a dispatch off the currently 2934 // viewed AST node. We then recursively traverse the AST by calling 2935 // EvalAddr and EvalVal appropriately. 2936 2937 E = E->IgnoreParens(); 2938 switch (E->getStmtClass()) { 2939 case Stmt::ImplicitCastExprClass: { 2940 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 2941 if (IE->getValueKind() == VK_LValue) { 2942 E = IE->getSubExpr(); 2943 continue; 2944 } 2945 return NULL; 2946 } 2947 2948 case Stmt::ExprWithCleanupsClass: 2949 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars); 2950 2951 case Stmt::DeclRefExprClass: { 2952 // When we hit a DeclRefExpr we are looking at code that refers to a 2953 // variable's name. If it's not a reference variable we check if it has 2954 // local storage within the function, and if so, return the expression. 2955 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2956 2957 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2958 if (V->hasLocalStorage()) { 2959 if (!V->getType()->isReferenceType()) 2960 return DR; 2961 2962 // Reference variable, follow through to the expression that 2963 // it points to. 2964 if (V->hasInit()) { 2965 // Add the reference variable to the "trail". 2966 refVars.push_back(DR); 2967 return EvalVal(V->getInit(), refVars); 2968 } 2969 } 2970 2971 return NULL; 2972 } 2973 2974 case Stmt::UnaryOperatorClass: { 2975 // The only unary operator that make sense to handle here 2976 // is Deref. All others don't resolve to a "name." This includes 2977 // handling all sorts of rvalues passed to a unary operator. 2978 UnaryOperator *U = cast<UnaryOperator>(E); 2979 2980 if (U->getOpcode() == UO_Deref) 2981 return EvalAddr(U->getSubExpr(), refVars); 2982 2983 return NULL; 2984 } 2985 2986 case Stmt::ArraySubscriptExprClass: { 2987 // Array subscripts are potential references to data on the stack. We 2988 // retrieve the DeclRefExpr* for the array variable if it indeed 2989 // has local storage. 2990 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars); 2991 } 2992 2993 case Stmt::ConditionalOperatorClass: { 2994 // For conditional operators we need to see if either the LHS or RHS are 2995 // non-NULL Expr's. If one is non-NULL, we return it. 2996 ConditionalOperator *C = cast<ConditionalOperator>(E); 2997 2998 // Handle the GNU extension for missing LHS. 2999 if (Expr *lhsExpr = C->getLHS()) 3000 if (Expr *LHS = EvalVal(lhsExpr, refVars)) 3001 return LHS; 3002 3003 return EvalVal(C->getRHS(), refVars); 3004 } 3005 3006 // Accesses to members are potential references to data on the stack. 3007 case Stmt::MemberExprClass: { 3008 MemberExpr *M = cast<MemberExpr>(E); 3009 3010 // Check for indirect access. We only want direct field accesses. 3011 if (M->isArrow()) 3012 return NULL; 3013 3014 // Check whether the member type is itself a reference, in which case 3015 // we're not going to refer to the member, but to what the member refers to. 3016 if (M->getMemberDecl()->getType()->isReferenceType()) 3017 return NULL; 3018 3019 return EvalVal(M->getBase(), refVars); 3020 } 3021 3022 case Stmt::MaterializeTemporaryExprClass: 3023 if (Expr *Result = EvalVal( 3024 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3025 refVars)) 3026 return Result; 3027 3028 return E; 3029 3030 default: 3031 // Check that we don't return or take the address of a reference to a 3032 // temporary. This is only useful in C++. 3033 if (!E->isTypeDependent() && E->isRValue()) 3034 return E; 3035 3036 // Everything else: we simply don't reason about them. 3037 return NULL; 3038 } 3039} while (true); 3040} 3041 3042//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 3043 3044/// Check for comparisons of floating point operands using != and ==. 3045/// Issue a warning if these are no self-comparisons, as they are not likely 3046/// to do what the programmer intended. 3047void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 3048 bool EmitWarning = true; 3049 3050 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 3051 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 3052 3053 // Special case: check for x == x (which is OK). 3054 // Do not emit warnings for such cases. 3055 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 3056 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 3057 if (DRL->getDecl() == DRR->getDecl()) 3058 EmitWarning = false; 3059 3060 3061 // Special case: check for comparisons against literals that can be exactly 3062 // represented by APFloat. In such cases, do not emit a warning. This 3063 // is a heuristic: often comparison against such literals are used to 3064 // detect if a value in a variable has not changed. This clearly can 3065 // lead to false negatives. 3066 if (EmitWarning) { 3067 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 3068 if (FLL->isExact()) 3069 EmitWarning = false; 3070 } else 3071 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 3072 if (FLR->isExact()) 3073 EmitWarning = false; 3074 } 3075 } 3076 3077 // Check for comparisons with builtin types. 3078 if (EmitWarning) 3079 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 3080 if (CL->isBuiltinCall()) 3081 EmitWarning = false; 3082 3083 if (EmitWarning) 3084 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 3085 if (CR->isBuiltinCall()) 3086 EmitWarning = false; 3087 3088 // Emit the diagnostic. 3089 if (EmitWarning) 3090 Diag(Loc, diag::warn_floatingpoint_eq) 3091 << LHS->getSourceRange() << RHS->getSourceRange(); 3092} 3093 3094//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 3095//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 3096 3097namespace { 3098 3099/// Structure recording the 'active' range of an integer-valued 3100/// expression. 3101struct IntRange { 3102 /// The number of bits active in the int. 3103 unsigned Width; 3104 3105 /// True if the int is known not to have negative values. 3106 bool NonNegative; 3107 3108 IntRange(unsigned Width, bool NonNegative) 3109 : Width(Width), NonNegative(NonNegative) 3110 {} 3111 3112 /// Returns the range of the bool type. 3113 static IntRange forBoolType() { 3114 return IntRange(1, true); 3115 } 3116 3117 /// Returns the range of an opaque value of the given integral type. 3118 static IntRange forValueOfType(ASTContext &C, QualType T) { 3119 return forValueOfCanonicalType(C, 3120 T->getCanonicalTypeInternal().getTypePtr()); 3121 } 3122 3123 /// Returns the range of an opaque value of a canonical integral type. 3124 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 3125 assert(T->isCanonicalUnqualified()); 3126 3127 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3128 T = VT->getElementType().getTypePtr(); 3129 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3130 T = CT->getElementType().getTypePtr(); 3131 3132 // For enum types, use the known bit width of the enumerators. 3133 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 3134 EnumDecl *Enum = ET->getDecl(); 3135 if (!Enum->isCompleteDefinition()) 3136 return IntRange(C.getIntWidth(QualType(T, 0)), false); 3137 3138 unsigned NumPositive = Enum->getNumPositiveBits(); 3139 unsigned NumNegative = Enum->getNumNegativeBits(); 3140 3141 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 3142 } 3143 3144 const BuiltinType *BT = cast<BuiltinType>(T); 3145 assert(BT->isInteger()); 3146 3147 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3148 } 3149 3150 /// Returns the "target" range of a canonical integral type, i.e. 3151 /// the range of values expressible in the type. 3152 /// 3153 /// This matches forValueOfCanonicalType except that enums have the 3154 /// full range of their type, not the range of their enumerators. 3155 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 3156 assert(T->isCanonicalUnqualified()); 3157 3158 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3159 T = VT->getElementType().getTypePtr(); 3160 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3161 T = CT->getElementType().getTypePtr(); 3162 if (const EnumType *ET = dyn_cast<EnumType>(T)) 3163 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 3164 3165 const BuiltinType *BT = cast<BuiltinType>(T); 3166 assert(BT->isInteger()); 3167 3168 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3169 } 3170 3171 /// Returns the supremum of two ranges: i.e. their conservative merge. 3172 static IntRange join(IntRange L, IntRange R) { 3173 return IntRange(std::max(L.Width, R.Width), 3174 L.NonNegative && R.NonNegative); 3175 } 3176 3177 /// Returns the infinum of two ranges: i.e. their aggressive merge. 3178 static IntRange meet(IntRange L, IntRange R) { 3179 return IntRange(std::min(L.Width, R.Width), 3180 L.NonNegative || R.NonNegative); 3181 } 3182}; 3183 3184IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 3185 if (value.isSigned() && value.isNegative()) 3186 return IntRange(value.getMinSignedBits(), false); 3187 3188 if (value.getBitWidth() > MaxWidth) 3189 value = value.trunc(MaxWidth); 3190 3191 // isNonNegative() just checks the sign bit without considering 3192 // signedness. 3193 return IntRange(value.getActiveBits(), true); 3194} 3195 3196IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 3197 unsigned MaxWidth) { 3198 if (result.isInt()) 3199 return GetValueRange(C, result.getInt(), MaxWidth); 3200 3201 if (result.isVector()) { 3202 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 3203 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 3204 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 3205 R = IntRange::join(R, El); 3206 } 3207 return R; 3208 } 3209 3210 if (result.isComplexInt()) { 3211 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 3212 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 3213 return IntRange::join(R, I); 3214 } 3215 3216 // This can happen with lossless casts to intptr_t of "based" lvalues. 3217 // Assume it might use arbitrary bits. 3218 // FIXME: The only reason we need to pass the type in here is to get 3219 // the sign right on this one case. It would be nice if APValue 3220 // preserved this. 3221 assert(result.isLValue() || result.isAddrLabelDiff()); 3222 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 3223} 3224 3225/// Pseudo-evaluate the given integer expression, estimating the 3226/// range of values it might take. 3227/// 3228/// \param MaxWidth - the width to which the value will be truncated 3229IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 3230 E = E->IgnoreParens(); 3231 3232 // Try a full evaluation first. 3233 Expr::EvalResult result; 3234 if (E->EvaluateAsRValue(result, C)) 3235 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 3236 3237 // I think we only want to look through implicit casts here; if the 3238 // user has an explicit widening cast, we should treat the value as 3239 // being of the new, wider type. 3240 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 3241 if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue) 3242 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 3243 3244 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 3245 3246 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 3247 3248 // Assume that non-integer casts can span the full range of the type. 3249 if (!isIntegerCast) 3250 return OutputTypeRange; 3251 3252 IntRange SubRange 3253 = GetExprRange(C, CE->getSubExpr(), 3254 std::min(MaxWidth, OutputTypeRange.Width)); 3255 3256 // Bail out if the subexpr's range is as wide as the cast type. 3257 if (SubRange.Width >= OutputTypeRange.Width) 3258 return OutputTypeRange; 3259 3260 // Otherwise, we take the smaller width, and we're non-negative if 3261 // either the output type or the subexpr is. 3262 return IntRange(SubRange.Width, 3263 SubRange.NonNegative || OutputTypeRange.NonNegative); 3264 } 3265 3266 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 3267 // If we can fold the condition, just take that operand. 3268 bool CondResult; 3269 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 3270 return GetExprRange(C, CondResult ? CO->getTrueExpr() 3271 : CO->getFalseExpr(), 3272 MaxWidth); 3273 3274 // Otherwise, conservatively merge. 3275 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 3276 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 3277 return IntRange::join(L, R); 3278 } 3279 3280 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 3281 switch (BO->getOpcode()) { 3282 3283 // Boolean-valued operations are single-bit and positive. 3284 case BO_LAnd: 3285 case BO_LOr: 3286 case BO_LT: 3287 case BO_GT: 3288 case BO_LE: 3289 case BO_GE: 3290 case BO_EQ: 3291 case BO_NE: 3292 return IntRange::forBoolType(); 3293 3294 // The type of the assignments is the type of the LHS, so the RHS 3295 // is not necessarily the same type. 3296 case BO_MulAssign: 3297 case BO_DivAssign: 3298 case BO_RemAssign: 3299 case BO_AddAssign: 3300 case BO_SubAssign: 3301 case BO_XorAssign: 3302 case BO_OrAssign: 3303 // TODO: bitfields? 3304 return IntRange::forValueOfType(C, E->getType()); 3305 3306 // Simple assignments just pass through the RHS, which will have 3307 // been coerced to the LHS type. 3308 case BO_Assign: 3309 // TODO: bitfields? 3310 return GetExprRange(C, BO->getRHS(), MaxWidth); 3311 3312 // Operations with opaque sources are black-listed. 3313 case BO_PtrMemD: 3314 case BO_PtrMemI: 3315 return IntRange::forValueOfType(C, E->getType()); 3316 3317 // Bitwise-and uses the *infinum* of the two source ranges. 3318 case BO_And: 3319 case BO_AndAssign: 3320 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 3321 GetExprRange(C, BO->getRHS(), MaxWidth)); 3322 3323 // Left shift gets black-listed based on a judgement call. 3324 case BO_Shl: 3325 // ...except that we want to treat '1 << (blah)' as logically 3326 // positive. It's an important idiom. 3327 if (IntegerLiteral *I 3328 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 3329 if (I->getValue() == 1) { 3330 IntRange R = IntRange::forValueOfType(C, E->getType()); 3331 return IntRange(R.Width, /*NonNegative*/ true); 3332 } 3333 } 3334 // fallthrough 3335 3336 case BO_ShlAssign: 3337 return IntRange::forValueOfType(C, E->getType()); 3338 3339 // Right shift by a constant can narrow its left argument. 3340 case BO_Shr: 3341 case BO_ShrAssign: { 3342 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3343 3344 // If the shift amount is a positive constant, drop the width by 3345 // that much. 3346 llvm::APSInt shift; 3347 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 3348 shift.isNonNegative()) { 3349 unsigned zext = shift.getZExtValue(); 3350 if (zext >= L.Width) 3351 L.Width = (L.NonNegative ? 0 : 1); 3352 else 3353 L.Width -= zext; 3354 } 3355 3356 return L; 3357 } 3358 3359 // Comma acts as its right operand. 3360 case BO_Comma: 3361 return GetExprRange(C, BO->getRHS(), MaxWidth); 3362 3363 // Black-list pointer subtractions. 3364 case BO_Sub: 3365 if (BO->getLHS()->getType()->isPointerType()) 3366 return IntRange::forValueOfType(C, E->getType()); 3367 break; 3368 3369 // The width of a division result is mostly determined by the size 3370 // of the LHS. 3371 case BO_Div: { 3372 // Don't 'pre-truncate' the operands. 3373 unsigned opWidth = C.getIntWidth(E->getType()); 3374 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3375 3376 // If the divisor is constant, use that. 3377 llvm::APSInt divisor; 3378 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 3379 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 3380 if (log2 >= L.Width) 3381 L.Width = (L.NonNegative ? 0 : 1); 3382 else 3383 L.Width = std::min(L.Width - log2, MaxWidth); 3384 return L; 3385 } 3386 3387 // Otherwise, just use the LHS's width. 3388 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3389 return IntRange(L.Width, L.NonNegative && R.NonNegative); 3390 } 3391 3392 // The result of a remainder can't be larger than the result of 3393 // either side. 3394 case BO_Rem: { 3395 // Don't 'pre-truncate' the operands. 3396 unsigned opWidth = C.getIntWidth(E->getType()); 3397 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3398 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3399 3400 IntRange meet = IntRange::meet(L, R); 3401 meet.Width = std::min(meet.Width, MaxWidth); 3402 return meet; 3403 } 3404 3405 // The default behavior is okay for these. 3406 case BO_Mul: 3407 case BO_Add: 3408 case BO_Xor: 3409 case BO_Or: 3410 break; 3411 } 3412 3413 // The default case is to treat the operation as if it were closed 3414 // on the narrowest type that encompasses both operands. 3415 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3416 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 3417 return IntRange::join(L, R); 3418 } 3419 3420 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 3421 switch (UO->getOpcode()) { 3422 // Boolean-valued operations are white-listed. 3423 case UO_LNot: 3424 return IntRange::forBoolType(); 3425 3426 // Operations with opaque sources are black-listed. 3427 case UO_Deref: 3428 case UO_AddrOf: // should be impossible 3429 return IntRange::forValueOfType(C, E->getType()); 3430 3431 default: 3432 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 3433 } 3434 } 3435 3436 if (dyn_cast<OffsetOfExpr>(E)) { 3437 IntRange::forValueOfType(C, E->getType()); 3438 } 3439 3440 if (FieldDecl *BitField = E->getBitField()) 3441 return IntRange(BitField->getBitWidthValue(C), 3442 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 3443 3444 return IntRange::forValueOfType(C, E->getType()); 3445} 3446 3447IntRange GetExprRange(ASTContext &C, Expr *E) { 3448 return GetExprRange(C, E, C.getIntWidth(E->getType())); 3449} 3450 3451/// Checks whether the given value, which currently has the given 3452/// source semantics, has the same value when coerced through the 3453/// target semantics. 3454bool IsSameFloatAfterCast(const llvm::APFloat &value, 3455 const llvm::fltSemantics &Src, 3456 const llvm::fltSemantics &Tgt) { 3457 llvm::APFloat truncated = value; 3458 3459 bool ignored; 3460 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 3461 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 3462 3463 return truncated.bitwiseIsEqual(value); 3464} 3465 3466/// Checks whether the given value, which currently has the given 3467/// source semantics, has the same value when coerced through the 3468/// target semantics. 3469/// 3470/// The value might be a vector of floats (or a complex number). 3471bool IsSameFloatAfterCast(const APValue &value, 3472 const llvm::fltSemantics &Src, 3473 const llvm::fltSemantics &Tgt) { 3474 if (value.isFloat()) 3475 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 3476 3477 if (value.isVector()) { 3478 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 3479 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 3480 return false; 3481 return true; 3482 } 3483 3484 assert(value.isComplexFloat()); 3485 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 3486 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 3487} 3488 3489void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 3490 3491static bool IsZero(Sema &S, Expr *E) { 3492 // Suppress cases where we are comparing against an enum constant. 3493 if (const DeclRefExpr *DR = 3494 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 3495 if (isa<EnumConstantDecl>(DR->getDecl())) 3496 return false; 3497 3498 // Suppress cases where the '0' value is expanded from a macro. 3499 if (E->getLocStart().isMacroID()) 3500 return false; 3501 3502 llvm::APSInt Value; 3503 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 3504} 3505 3506static bool HasEnumType(Expr *E) { 3507 // Strip off implicit integral promotions. 3508 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 3509 if (ICE->getCastKind() != CK_IntegralCast && 3510 ICE->getCastKind() != CK_NoOp) 3511 break; 3512 E = ICE->getSubExpr(); 3513 } 3514 3515 return E->getType()->isEnumeralType(); 3516} 3517 3518void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 3519 BinaryOperatorKind op = E->getOpcode(); 3520 if (E->isValueDependent()) 3521 return; 3522 3523 if (op == BO_LT && IsZero(S, E->getRHS())) { 3524 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3525 << "< 0" << "false" << HasEnumType(E->getLHS()) 3526 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3527 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 3528 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3529 << ">= 0" << "true" << HasEnumType(E->getLHS()) 3530 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3531 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 3532 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3533 << "0 >" << "false" << HasEnumType(E->getRHS()) 3534 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3535 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 3536 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3537 << "0 <=" << "true" << HasEnumType(E->getRHS()) 3538 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3539 } 3540} 3541 3542/// Analyze the operands of the given comparison. Implements the 3543/// fallback case from AnalyzeComparison. 3544void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 3545 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3546 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3547} 3548 3549/// \brief Implements -Wsign-compare. 3550/// 3551/// \param E the binary operator to check for warnings 3552void AnalyzeComparison(Sema &S, BinaryOperator *E) { 3553 // The type the comparison is being performed in. 3554 QualType T = E->getLHS()->getType(); 3555 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 3556 && "comparison with mismatched types"); 3557 3558 // We don't do anything special if this isn't an unsigned integral 3559 // comparison: we're only interested in integral comparisons, and 3560 // signed comparisons only happen in cases we don't care to warn about. 3561 // 3562 // We also don't care about value-dependent expressions or expressions 3563 // whose result is a constant. 3564 if (!T->hasUnsignedIntegerRepresentation() 3565 || E->isValueDependent() || E->isIntegerConstantExpr(S.Context)) 3566 return AnalyzeImpConvsInComparison(S, E); 3567 3568 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 3569 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 3570 3571 // Check to see if one of the (unmodified) operands is of different 3572 // signedness. 3573 Expr *signedOperand, *unsignedOperand; 3574 if (LHS->getType()->hasSignedIntegerRepresentation()) { 3575 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 3576 "unsigned comparison between two signed integer expressions?"); 3577 signedOperand = LHS; 3578 unsignedOperand = RHS; 3579 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 3580 signedOperand = RHS; 3581 unsignedOperand = LHS; 3582 } else { 3583 CheckTrivialUnsignedComparison(S, E); 3584 return AnalyzeImpConvsInComparison(S, E); 3585 } 3586 3587 // Otherwise, calculate the effective range of the signed operand. 3588 IntRange signedRange = GetExprRange(S.Context, signedOperand); 3589 3590 // Go ahead and analyze implicit conversions in the operands. Note 3591 // that we skip the implicit conversions on both sides. 3592 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 3593 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 3594 3595 // If the signed range is non-negative, -Wsign-compare won't fire, 3596 // but we should still check for comparisons which are always true 3597 // or false. 3598 if (signedRange.NonNegative) 3599 return CheckTrivialUnsignedComparison(S, E); 3600 3601 // For (in)equality comparisons, if the unsigned operand is a 3602 // constant which cannot collide with a overflowed signed operand, 3603 // then reinterpreting the signed operand as unsigned will not 3604 // change the result of the comparison. 3605 if (E->isEqualityOp()) { 3606 unsigned comparisonWidth = S.Context.getIntWidth(T); 3607 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 3608 3609 // We should never be unable to prove that the unsigned operand is 3610 // non-negative. 3611 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 3612 3613 if (unsignedRange.Width < comparisonWidth) 3614 return; 3615 } 3616 3617 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison) 3618 << LHS->getType() << RHS->getType() 3619 << LHS->getSourceRange() << RHS->getSourceRange(); 3620} 3621 3622/// Analyzes an attempt to assign the given value to a bitfield. 3623/// 3624/// Returns true if there was something fishy about the attempt. 3625bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 3626 SourceLocation InitLoc) { 3627 assert(Bitfield->isBitField()); 3628 if (Bitfield->isInvalidDecl()) 3629 return false; 3630 3631 // White-list bool bitfields. 3632 if (Bitfield->getType()->isBooleanType()) 3633 return false; 3634 3635 // Ignore value- or type-dependent expressions. 3636 if (Bitfield->getBitWidth()->isValueDependent() || 3637 Bitfield->getBitWidth()->isTypeDependent() || 3638 Init->isValueDependent() || 3639 Init->isTypeDependent()) 3640 return false; 3641 3642 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 3643 3644 llvm::APSInt Value; 3645 if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects)) 3646 return false; 3647 3648 unsigned OriginalWidth = Value.getBitWidth(); 3649 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 3650 3651 if (OriginalWidth <= FieldWidth) 3652 return false; 3653 3654 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 3655 3656 // It's fairly common to write values into signed bitfields 3657 // that, if sign-extended, would end up becoming a different 3658 // value. We don't want to warn about that. 3659 if (Value.isSigned() && Value.isNegative()) 3660 TruncatedValue = TruncatedValue.sext(OriginalWidth); 3661 else 3662 TruncatedValue = TruncatedValue.zext(OriginalWidth); 3663 3664 if (Value == TruncatedValue) 3665 return false; 3666 3667 std::string PrettyValue = Value.toString(10); 3668 std::string PrettyTrunc = TruncatedValue.toString(10); 3669 3670 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 3671 << PrettyValue << PrettyTrunc << OriginalInit->getType() 3672 << Init->getSourceRange(); 3673 3674 return true; 3675} 3676 3677/// Analyze the given simple or compound assignment for warning-worthy 3678/// operations. 3679void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 3680 // Just recurse on the LHS. 3681 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3682 3683 // We want to recurse on the RHS as normal unless we're assigning to 3684 // a bitfield. 3685 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 3686 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 3687 E->getOperatorLoc())) { 3688 // Recurse, ignoring any implicit conversions on the RHS. 3689 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 3690 E->getOperatorLoc()); 3691 } 3692 } 3693 3694 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3695} 3696 3697/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3698void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 3699 SourceLocation CContext, unsigned diag) { 3700 S.Diag(E->getExprLoc(), diag) 3701 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 3702} 3703 3704/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3705void DiagnoseImpCast(Sema &S, Expr *E, QualType T, SourceLocation CContext, 3706 unsigned diag) { 3707 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag); 3708} 3709 3710/// Diagnose an implicit cast from a literal expression. Does not warn when the 3711/// cast wouldn't lose information. 3712void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 3713 SourceLocation CContext) { 3714 // Try to convert the literal exactly to an integer. If we can, don't warn. 3715 bool isExact = false; 3716 const llvm::APFloat &Value = FL->getValue(); 3717 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 3718 T->hasUnsignedIntegerRepresentation()); 3719 if (Value.convertToInteger(IntegerValue, 3720 llvm::APFloat::rmTowardZero, &isExact) 3721 == llvm::APFloat::opOK && isExact) 3722 return; 3723 3724 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 3725 << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext); 3726} 3727 3728std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 3729 if (!Range.Width) return "0"; 3730 3731 llvm::APSInt ValueInRange = Value; 3732 ValueInRange.setIsSigned(!Range.NonNegative); 3733 ValueInRange = ValueInRange.trunc(Range.Width); 3734 return ValueInRange.toString(10); 3735} 3736 3737void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 3738 SourceLocation CC, bool *ICContext = 0) { 3739 if (E->isTypeDependent() || E->isValueDependent()) return; 3740 3741 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 3742 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 3743 if (Source == Target) return; 3744 if (Target->isDependentType()) return; 3745 3746 // If the conversion context location is invalid don't complain. We also 3747 // don't want to emit a warning if the issue occurs from the expansion of 3748 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 3749 // delay this check as long as possible. Once we detect we are in that 3750 // scenario, we just return. 3751 if (CC.isInvalid()) 3752 return; 3753 3754 // Diagnose implicit casts to bool. 3755 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 3756 if (isa<StringLiteral>(E)) 3757 // Warn on string literal to bool. Checks for string literals in logical 3758 // expressions, for instances, assert(0 && "error here"), is prevented 3759 // by a check in AnalyzeImplicitConversions(). 3760 return DiagnoseImpCast(S, E, T, CC, 3761 diag::warn_impcast_string_literal_to_bool); 3762 if (Source->isFunctionType()) { 3763 // Warn on function to bool. Checks free functions and static member 3764 // functions. Weakly imported functions are excluded from the check, 3765 // since it's common to test their value to check whether the linker 3766 // found a definition for them. 3767 ValueDecl *D = 0; 3768 if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) { 3769 D = R->getDecl(); 3770 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 3771 D = M->getMemberDecl(); 3772 } 3773 3774 if (D && !D->isWeak()) { 3775 if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) { 3776 S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool) 3777 << F << E->getSourceRange() << SourceRange(CC); 3778 S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence) 3779 << FixItHint::CreateInsertion(E->getExprLoc(), "&"); 3780 QualType ReturnType; 3781 UnresolvedSet<4> NonTemplateOverloads; 3782 S.isExprCallable(*E, ReturnType, NonTemplateOverloads); 3783 if (!ReturnType.isNull() 3784 && ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 3785 S.Diag(E->getExprLoc(), diag::note_function_to_bool_call) 3786 << FixItHint::CreateInsertion( 3787 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()"); 3788 return; 3789 } 3790 } 3791 } 3792 return; // Other casts to bool are not checked. 3793 } 3794 3795 // Strip vector types. 3796 if (isa<VectorType>(Source)) { 3797 if (!isa<VectorType>(Target)) { 3798 if (S.SourceMgr.isInSystemMacro(CC)) 3799 return; 3800 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 3801 } 3802 3803 // If the vector cast is cast between two vectors of the same size, it is 3804 // a bitcast, not a conversion. 3805 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 3806 return; 3807 3808 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 3809 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 3810 } 3811 3812 // Strip complex types. 3813 if (isa<ComplexType>(Source)) { 3814 if (!isa<ComplexType>(Target)) { 3815 if (S.SourceMgr.isInSystemMacro(CC)) 3816 return; 3817 3818 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 3819 } 3820 3821 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 3822 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 3823 } 3824 3825 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 3826 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 3827 3828 // If the source is floating point... 3829 if (SourceBT && SourceBT->isFloatingPoint()) { 3830 // ...and the target is floating point... 3831 if (TargetBT && TargetBT->isFloatingPoint()) { 3832 // ...then warn if we're dropping FP rank. 3833 3834 // Builtin FP kinds are ordered by increasing FP rank. 3835 if (SourceBT->getKind() > TargetBT->getKind()) { 3836 // Don't warn about float constants that are precisely 3837 // representable in the target type. 3838 Expr::EvalResult result; 3839 if (E->EvaluateAsRValue(result, S.Context)) { 3840 // Value might be a float, a float vector, or a float complex. 3841 if (IsSameFloatAfterCast(result.Val, 3842 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 3843 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 3844 return; 3845 } 3846 3847 if (S.SourceMgr.isInSystemMacro(CC)) 3848 return; 3849 3850 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 3851 } 3852 return; 3853 } 3854 3855 // If the target is integral, always warn. 3856 if ((TargetBT && TargetBT->isInteger())) { 3857 if (S.SourceMgr.isInSystemMacro(CC)) 3858 return; 3859 3860 Expr *InnerE = E->IgnoreParenImpCasts(); 3861 // We also want to warn on, e.g., "int i = -1.234" 3862 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 3863 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 3864 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 3865 3866 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 3867 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 3868 } else { 3869 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 3870 } 3871 } 3872 3873 return; 3874 } 3875 3876 if (!Source->isIntegerType() || !Target->isIntegerType()) 3877 return; 3878 3879 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 3880 == Expr::NPCK_GNUNull) && Target->isIntegerType()) { 3881 S.Diag(E->getExprLoc(), diag::warn_impcast_null_pointer_to_integer) 3882 << E->getSourceRange() << clang::SourceRange(CC); 3883 return; 3884 } 3885 3886 IntRange SourceRange = GetExprRange(S.Context, E); 3887 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 3888 3889 if (SourceRange.Width > TargetRange.Width) { 3890 // If the source is a constant, use a default-on diagnostic. 3891 // TODO: this should happen for bitfield stores, too. 3892 llvm::APSInt Value(32); 3893 if (E->isIntegerConstantExpr(Value, S.Context)) { 3894 if (S.SourceMgr.isInSystemMacro(CC)) 3895 return; 3896 3897 std::string PrettySourceValue = Value.toString(10); 3898 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 3899 3900 S.DiagRuntimeBehavior(E->getExprLoc(), E, 3901 S.PDiag(diag::warn_impcast_integer_precision_constant) 3902 << PrettySourceValue << PrettyTargetValue 3903 << E->getType() << T << E->getSourceRange() 3904 << clang::SourceRange(CC)); 3905 return; 3906 } 3907 3908 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 3909 if (S.SourceMgr.isInSystemMacro(CC)) 3910 return; 3911 3912 if (SourceRange.Width == 64 && TargetRange.Width == 32) 3913 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32); 3914 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 3915 } 3916 3917 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 3918 (!TargetRange.NonNegative && SourceRange.NonNegative && 3919 SourceRange.Width == TargetRange.Width)) { 3920 3921 if (S.SourceMgr.isInSystemMacro(CC)) 3922 return; 3923 3924 unsigned DiagID = diag::warn_impcast_integer_sign; 3925 3926 // Traditionally, gcc has warned about this under -Wsign-compare. 3927 // We also want to warn about it in -Wconversion. 3928 // So if -Wconversion is off, use a completely identical diagnostic 3929 // in the sign-compare group. 3930 // The conditional-checking code will 3931 if (ICContext) { 3932 DiagID = diag::warn_impcast_integer_sign_conditional; 3933 *ICContext = true; 3934 } 3935 3936 return DiagnoseImpCast(S, E, T, CC, DiagID); 3937 } 3938 3939 // Diagnose conversions between different enumeration types. 3940 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 3941 // type, to give us better diagnostics. 3942 QualType SourceType = E->getType(); 3943 if (!S.getLangOptions().CPlusPlus) { 3944 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 3945 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 3946 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 3947 SourceType = S.Context.getTypeDeclType(Enum); 3948 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 3949 } 3950 } 3951 3952 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 3953 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 3954 if ((SourceEnum->getDecl()->getIdentifier() || 3955 SourceEnum->getDecl()->getTypedefNameForAnonDecl()) && 3956 (TargetEnum->getDecl()->getIdentifier() || 3957 TargetEnum->getDecl()->getTypedefNameForAnonDecl()) && 3958 SourceEnum != TargetEnum) { 3959 if (S.SourceMgr.isInSystemMacro(CC)) 3960 return; 3961 3962 return DiagnoseImpCast(S, E, SourceType, T, CC, 3963 diag::warn_impcast_different_enum_types); 3964 } 3965 3966 return; 3967} 3968 3969void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 3970 3971void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 3972 SourceLocation CC, bool &ICContext) { 3973 E = E->IgnoreParenImpCasts(); 3974 3975 if (isa<ConditionalOperator>(E)) 3976 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 3977 3978 AnalyzeImplicitConversions(S, E, CC); 3979 if (E->getType() != T) 3980 return CheckImplicitConversion(S, E, T, CC, &ICContext); 3981 return; 3982} 3983 3984void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 3985 SourceLocation CC = E->getQuestionLoc(); 3986 3987 AnalyzeImplicitConversions(S, E->getCond(), CC); 3988 3989 bool Suspicious = false; 3990 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 3991 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 3992 3993 // If -Wconversion would have warned about either of the candidates 3994 // for a signedness conversion to the context type... 3995 if (!Suspicious) return; 3996 3997 // ...but it's currently ignored... 3998 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 3999 CC)) 4000 return; 4001 4002 // ...then check whether it would have warned about either of the 4003 // candidates for a signedness conversion to the condition type. 4004 if (E->getType() == T) return; 4005 4006 Suspicious = false; 4007 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 4008 E->getType(), CC, &Suspicious); 4009 if (!Suspicious) 4010 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 4011 E->getType(), CC, &Suspicious); 4012} 4013 4014/// AnalyzeImplicitConversions - Find and report any interesting 4015/// implicit conversions in the given expression. There are a couple 4016/// of competing diagnostics here, -Wconversion and -Wsign-compare. 4017void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 4018 QualType T = OrigE->getType(); 4019 Expr *E = OrigE->IgnoreParenImpCasts(); 4020 4021 if (E->isTypeDependent() || E->isValueDependent()) 4022 return; 4023 4024 // For conditional operators, we analyze the arguments as if they 4025 // were being fed directly into the output. 4026 if (isa<ConditionalOperator>(E)) { 4027 ConditionalOperator *CO = cast<ConditionalOperator>(E); 4028 CheckConditionalOperator(S, CO, T); 4029 return; 4030 } 4031 4032 // Go ahead and check any implicit conversions we might have skipped. 4033 // The non-canonical typecheck is just an optimization; 4034 // CheckImplicitConversion will filter out dead implicit conversions. 4035 if (E->getType() != T) 4036 CheckImplicitConversion(S, E, T, CC); 4037 4038 // Now continue drilling into this expression. 4039 4040 // Skip past explicit casts. 4041 if (isa<ExplicitCastExpr>(E)) { 4042 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 4043 return AnalyzeImplicitConversions(S, E, CC); 4044 } 4045 4046 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 4047 // Do a somewhat different check with comparison operators. 4048 if (BO->isComparisonOp()) 4049 return AnalyzeComparison(S, BO); 4050 4051 // And with assignments and compound assignments. 4052 if (BO->isAssignmentOp()) 4053 return AnalyzeAssignment(S, BO); 4054 } 4055 4056 // These break the otherwise-useful invariant below. Fortunately, 4057 // we don't really need to recurse into them, because any internal 4058 // expressions should have been analyzed already when they were 4059 // built into statements. 4060 if (isa<StmtExpr>(E)) return; 4061 4062 // Don't descend into unevaluated contexts. 4063 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 4064 4065 // Now just recurse over the expression's children. 4066 CC = E->getExprLoc(); 4067 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 4068 bool IsLogicalOperator = BO && BO->isLogicalOp(); 4069 for (Stmt::child_range I = E->children(); I; ++I) { 4070 Expr *ChildExpr = cast<Expr>(*I); 4071 if (IsLogicalOperator && 4072 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 4073 // Ignore checking string literals that are in logical operators. 4074 continue; 4075 AnalyzeImplicitConversions(S, ChildExpr, CC); 4076 } 4077} 4078 4079} // end anonymous namespace 4080 4081/// Diagnoses "dangerous" implicit conversions within the given 4082/// expression (which is a full expression). Implements -Wconversion 4083/// and -Wsign-compare. 4084/// 4085/// \param CC the "context" location of the implicit conversion, i.e. 4086/// the most location of the syntactic entity requiring the implicit 4087/// conversion 4088void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 4089 // Don't diagnose in unevaluated contexts. 4090 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 4091 return; 4092 4093 // Don't diagnose for value- or type-dependent expressions. 4094 if (E->isTypeDependent() || E->isValueDependent()) 4095 return; 4096 4097 // Check for array bounds violations in cases where the check isn't triggered 4098 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 4099 // ArraySubscriptExpr is on the RHS of a variable initialization. 4100 CheckArrayAccess(E); 4101 4102 // This is not the right CC for (e.g.) a variable initialization. 4103 AnalyzeImplicitConversions(*this, E, CC); 4104} 4105 4106void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 4107 FieldDecl *BitField, 4108 Expr *Init) { 4109 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 4110} 4111 4112/// CheckParmsForFunctionDef - Check that the parameters of the given 4113/// function are appropriate for the definition of a function. This 4114/// takes care of any checks that cannot be performed on the 4115/// declaration itself, e.g., that the types of each of the function 4116/// parameters are complete. 4117bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 4118 bool CheckParameterNames) { 4119 bool HasInvalidParm = false; 4120 for (; P != PEnd; ++P) { 4121 ParmVarDecl *Param = *P; 4122 4123 // C99 6.7.5.3p4: the parameters in a parameter type list in a 4124 // function declarator that is part of a function definition of 4125 // that function shall not have incomplete type. 4126 // 4127 // This is also C++ [dcl.fct]p6. 4128 if (!Param->isInvalidDecl() && 4129 RequireCompleteType(Param->getLocation(), Param->getType(), 4130 diag::err_typecheck_decl_incomplete_type)) { 4131 Param->setInvalidDecl(); 4132 HasInvalidParm = true; 4133 } 4134 4135 // C99 6.9.1p5: If the declarator includes a parameter type list, the 4136 // declaration of each parameter shall include an identifier. 4137 if (CheckParameterNames && 4138 Param->getIdentifier() == 0 && 4139 !Param->isImplicit() && 4140 !getLangOptions().CPlusPlus) 4141 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 4142 4143 // C99 6.7.5.3p12: 4144 // If the function declarator is not part of a definition of that 4145 // function, parameters may have incomplete type and may use the [*] 4146 // notation in their sequences of declarator specifiers to specify 4147 // variable length array types. 4148 QualType PType = Param->getOriginalType(); 4149 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 4150 if (AT->getSizeModifier() == ArrayType::Star) { 4151 // FIXME: This diagnosic should point the the '[*]' if source-location 4152 // information is added for it. 4153 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 4154 } 4155 } 4156 } 4157 4158 return HasInvalidParm; 4159} 4160 4161/// CheckCastAlign - Implements -Wcast-align, which warns when a 4162/// pointer cast increases the alignment requirements. 4163void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 4164 // This is actually a lot of work to potentially be doing on every 4165 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 4166 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 4167 TRange.getBegin()) 4168 == DiagnosticsEngine::Ignored) 4169 return; 4170 4171 // Ignore dependent types. 4172 if (T->isDependentType() || Op->getType()->isDependentType()) 4173 return; 4174 4175 // Require that the destination be a pointer type. 4176 const PointerType *DestPtr = T->getAs<PointerType>(); 4177 if (!DestPtr) return; 4178 4179 // If the destination has alignment 1, we're done. 4180 QualType DestPointee = DestPtr->getPointeeType(); 4181 if (DestPointee->isIncompleteType()) return; 4182 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 4183 if (DestAlign.isOne()) return; 4184 4185 // Require that the source be a pointer type. 4186 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 4187 if (!SrcPtr) return; 4188 QualType SrcPointee = SrcPtr->getPointeeType(); 4189 4190 // Whitelist casts from cv void*. We already implicitly 4191 // whitelisted casts to cv void*, since they have alignment 1. 4192 // Also whitelist casts involving incomplete types, which implicitly 4193 // includes 'void'. 4194 if (SrcPointee->isIncompleteType()) return; 4195 4196 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 4197 if (SrcAlign >= DestAlign) return; 4198 4199 Diag(TRange.getBegin(), diag::warn_cast_align) 4200 << Op->getType() << T 4201 << static_cast<unsigned>(SrcAlign.getQuantity()) 4202 << static_cast<unsigned>(DestAlign.getQuantity()) 4203 << TRange << Op->getSourceRange(); 4204} 4205 4206static const Type* getElementType(const Expr *BaseExpr) { 4207 const Type* EltType = BaseExpr->getType().getTypePtr(); 4208 if (EltType->isAnyPointerType()) 4209 return EltType->getPointeeType().getTypePtr(); 4210 else if (EltType->isArrayType()) 4211 return EltType->getBaseElementTypeUnsafe(); 4212 return EltType; 4213} 4214 4215/// \brief Check whether this array fits the idiom of a size-one tail padded 4216/// array member of a struct. 4217/// 4218/// We avoid emitting out-of-bounds access warnings for such arrays as they are 4219/// commonly used to emulate flexible arrays in C89 code. 4220static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 4221 const NamedDecl *ND) { 4222 if (Size != 1 || !ND) return false; 4223 4224 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 4225 if (!FD) return false; 4226 4227 // Don't consider sizes resulting from macro expansions or template argument 4228 // substitution to form C89 tail-padded arrays. 4229 ConstantArrayTypeLoc TL = 4230 cast<ConstantArrayTypeLoc>(FD->getTypeSourceInfo()->getTypeLoc()); 4231 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(TL.getSizeExpr()); 4232 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 4233 return false; 4234 4235 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 4236 if (!RD) return false; 4237 if (RD->isUnion()) return false; 4238 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 4239 if (!CRD->isStandardLayout()) return false; 4240 } 4241 4242 // See if this is the last field decl in the record. 4243 const Decl *D = FD; 4244 while ((D = D->getNextDeclInContext())) 4245 if (isa<FieldDecl>(D)) 4246 return false; 4247 return true; 4248} 4249 4250void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 4251 const ArraySubscriptExpr *ASE, 4252 bool AllowOnePastEnd, bool IndexNegated) { 4253 IndexExpr = IndexExpr->IgnoreParenCasts(); 4254 if (IndexExpr->isValueDependent()) 4255 return; 4256 4257 const Type *EffectiveType = getElementType(BaseExpr); 4258 BaseExpr = BaseExpr->IgnoreParenCasts(); 4259 const ConstantArrayType *ArrayTy = 4260 Context.getAsConstantArrayType(BaseExpr->getType()); 4261 if (!ArrayTy) 4262 return; 4263 4264 llvm::APSInt index; 4265 if (!IndexExpr->EvaluateAsInt(index, Context)) 4266 return; 4267 if (IndexNegated) 4268 index = -index; 4269 4270 const NamedDecl *ND = NULL; 4271 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4272 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 4273 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 4274 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 4275 4276 if (index.isUnsigned() || !index.isNegative()) { 4277 llvm::APInt size = ArrayTy->getSize(); 4278 if (!size.isStrictlyPositive()) 4279 return; 4280 4281 const Type* BaseType = getElementType(BaseExpr); 4282 if (BaseType != EffectiveType) { 4283 // Make sure we're comparing apples to apples when comparing index to size 4284 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 4285 uint64_t array_typesize = Context.getTypeSize(BaseType); 4286 // Handle ptrarith_typesize being zero, such as when casting to void* 4287 if (!ptrarith_typesize) ptrarith_typesize = 1; 4288 if (ptrarith_typesize != array_typesize) { 4289 // There's a cast to a different size type involved 4290 uint64_t ratio = array_typesize / ptrarith_typesize; 4291 // TODO: Be smarter about handling cases where array_typesize is not a 4292 // multiple of ptrarith_typesize 4293 if (ptrarith_typesize * ratio == array_typesize) 4294 size *= llvm::APInt(size.getBitWidth(), ratio); 4295 } 4296 } 4297 4298 if (size.getBitWidth() > index.getBitWidth()) 4299 index = index.sext(size.getBitWidth()); 4300 else if (size.getBitWidth() < index.getBitWidth()) 4301 size = size.sext(index.getBitWidth()); 4302 4303 // For array subscripting the index must be less than size, but for pointer 4304 // arithmetic also allow the index (offset) to be equal to size since 4305 // computing the next address after the end of the array is legal and 4306 // commonly done e.g. in C++ iterators and range-based for loops. 4307 if (AllowOnePastEnd ? index.sle(size) : index.slt(size)) 4308 return; 4309 4310 // Also don't warn for arrays of size 1 which are members of some 4311 // structure. These are often used to approximate flexible arrays in C89 4312 // code. 4313 if (IsTailPaddedMemberArray(*this, size, ND)) 4314 return; 4315 4316 // Suppress the warning if the subscript expression (as identified by the 4317 // ']' location) and the index expression are both from macro expansions 4318 // within a system header. 4319 if (ASE) { 4320 SourceLocation RBracketLoc = SourceMgr.getSpellingLoc( 4321 ASE->getRBracketLoc()); 4322 if (SourceMgr.isInSystemHeader(RBracketLoc)) { 4323 SourceLocation IndexLoc = SourceMgr.getSpellingLoc( 4324 IndexExpr->getLocStart()); 4325 if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc)) 4326 return; 4327 } 4328 } 4329 4330 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 4331 if (ASE) 4332 DiagID = diag::warn_array_index_exceeds_bounds; 4333 4334 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4335 PDiag(DiagID) << index.toString(10, true) 4336 << size.toString(10, true) 4337 << (unsigned)size.getLimitedValue(~0U) 4338 << IndexExpr->getSourceRange()); 4339 } else { 4340 unsigned DiagID = diag::warn_array_index_precedes_bounds; 4341 if (!ASE) { 4342 DiagID = diag::warn_ptr_arith_precedes_bounds; 4343 if (index.isNegative()) index = -index; 4344 } 4345 4346 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4347 PDiag(DiagID) << index.toString(10, true) 4348 << IndexExpr->getSourceRange()); 4349 } 4350 4351 if (!ND) { 4352 // Try harder to find a NamedDecl to point at in the note. 4353 while (const ArraySubscriptExpr *ASE = 4354 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 4355 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 4356 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4357 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 4358 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 4359 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 4360 } 4361 4362 if (ND) 4363 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 4364 PDiag(diag::note_array_index_out_of_bounds) 4365 << ND->getDeclName()); 4366} 4367 4368void Sema::CheckArrayAccess(const Expr *expr) { 4369 int AllowOnePastEnd = 0; 4370 while (expr) { 4371 expr = expr->IgnoreParenImpCasts(); 4372 switch (expr->getStmtClass()) { 4373 case Stmt::ArraySubscriptExprClass: { 4374 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 4375 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE, 4376 AllowOnePastEnd > 0); 4377 return; 4378 } 4379 case Stmt::UnaryOperatorClass: { 4380 // Only unwrap the * and & unary operators 4381 const UnaryOperator *UO = cast<UnaryOperator>(expr); 4382 expr = UO->getSubExpr(); 4383 switch (UO->getOpcode()) { 4384 case UO_AddrOf: 4385 AllowOnePastEnd++; 4386 break; 4387 case UO_Deref: 4388 AllowOnePastEnd--; 4389 break; 4390 default: 4391 return; 4392 } 4393 break; 4394 } 4395 case Stmt::ConditionalOperatorClass: { 4396 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 4397 if (const Expr *lhs = cond->getLHS()) 4398 CheckArrayAccess(lhs); 4399 if (const Expr *rhs = cond->getRHS()) 4400 CheckArrayAccess(rhs); 4401 return; 4402 } 4403 default: 4404 return; 4405 } 4406 } 4407} 4408 4409//===--- CHECK: Objective-C retain cycles ----------------------------------// 4410 4411namespace { 4412 struct RetainCycleOwner { 4413 RetainCycleOwner() : Variable(0), Indirect(false) {} 4414 VarDecl *Variable; 4415 SourceRange Range; 4416 SourceLocation Loc; 4417 bool Indirect; 4418 4419 void setLocsFrom(Expr *e) { 4420 Loc = e->getExprLoc(); 4421 Range = e->getSourceRange(); 4422 } 4423 }; 4424} 4425 4426/// Consider whether capturing the given variable can possibly lead to 4427/// a retain cycle. 4428static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 4429 // In ARC, it's captured strongly iff the variable has __strong 4430 // lifetime. In MRR, it's captured strongly if the variable is 4431 // __block and has an appropriate type. 4432 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4433 return false; 4434 4435 owner.Variable = var; 4436 owner.setLocsFrom(ref); 4437 return true; 4438} 4439 4440static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) { 4441 while (true) { 4442 e = e->IgnoreParens(); 4443 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 4444 switch (cast->getCastKind()) { 4445 case CK_BitCast: 4446 case CK_LValueBitCast: 4447 case CK_LValueToRValue: 4448 case CK_ARCReclaimReturnedObject: 4449 e = cast->getSubExpr(); 4450 continue; 4451 4452 default: 4453 return false; 4454 } 4455 } 4456 4457 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 4458 ObjCIvarDecl *ivar = ref->getDecl(); 4459 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4460 return false; 4461 4462 // Try to find a retain cycle in the base. 4463 if (!findRetainCycleOwner(S, ref->getBase(), owner)) 4464 return false; 4465 4466 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 4467 owner.Indirect = true; 4468 return true; 4469 } 4470 4471 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 4472 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 4473 if (!var) return false; 4474 return considerVariable(var, ref, owner); 4475 } 4476 4477 if (BlockDeclRefExpr *ref = dyn_cast<BlockDeclRefExpr>(e)) { 4478 owner.Variable = ref->getDecl(); 4479 owner.setLocsFrom(ref); 4480 return true; 4481 } 4482 4483 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 4484 if (member->isArrow()) return false; 4485 4486 // Don't count this as an indirect ownership. 4487 e = member->getBase(); 4488 continue; 4489 } 4490 4491 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 4492 // Only pay attention to pseudo-objects on property references. 4493 ObjCPropertyRefExpr *pre 4494 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 4495 ->IgnoreParens()); 4496 if (!pre) return false; 4497 if (pre->isImplicitProperty()) return false; 4498 ObjCPropertyDecl *property = pre->getExplicitProperty(); 4499 if (!property->isRetaining() && 4500 !(property->getPropertyIvarDecl() && 4501 property->getPropertyIvarDecl()->getType() 4502 .getObjCLifetime() == Qualifiers::OCL_Strong)) 4503 return false; 4504 4505 owner.Indirect = true; 4506 if (pre->isSuperReceiver()) { 4507 owner.Variable = S.getCurMethodDecl()->getSelfDecl(); 4508 if (!owner.Variable) 4509 return false; 4510 owner.Loc = pre->getLocation(); 4511 owner.Range = pre->getSourceRange(); 4512 return true; 4513 } 4514 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 4515 ->getSourceExpr()); 4516 continue; 4517 } 4518 4519 // Array ivars? 4520 4521 return false; 4522 } 4523} 4524 4525namespace { 4526 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 4527 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 4528 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 4529 Variable(variable), Capturer(0) {} 4530 4531 VarDecl *Variable; 4532 Expr *Capturer; 4533 4534 void VisitDeclRefExpr(DeclRefExpr *ref) { 4535 if (ref->getDecl() == Variable && !Capturer) 4536 Capturer = ref; 4537 } 4538 4539 void VisitBlockDeclRefExpr(BlockDeclRefExpr *ref) { 4540 if (ref->getDecl() == Variable && !Capturer) 4541 Capturer = ref; 4542 } 4543 4544 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 4545 if (Capturer) return; 4546 Visit(ref->getBase()); 4547 if (Capturer && ref->isFreeIvar()) 4548 Capturer = ref; 4549 } 4550 4551 void VisitBlockExpr(BlockExpr *block) { 4552 // Look inside nested blocks 4553 if (block->getBlockDecl()->capturesVariable(Variable)) 4554 Visit(block->getBlockDecl()->getBody()); 4555 } 4556 }; 4557} 4558 4559/// Check whether the given argument is a block which captures a 4560/// variable. 4561static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 4562 assert(owner.Variable && owner.Loc.isValid()); 4563 4564 e = e->IgnoreParenCasts(); 4565 BlockExpr *block = dyn_cast<BlockExpr>(e); 4566 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 4567 return 0; 4568 4569 FindCaptureVisitor visitor(S.Context, owner.Variable); 4570 visitor.Visit(block->getBlockDecl()->getBody()); 4571 return visitor.Capturer; 4572} 4573 4574static void diagnoseRetainCycle(Sema &S, Expr *capturer, 4575 RetainCycleOwner &owner) { 4576 assert(capturer); 4577 assert(owner.Variable && owner.Loc.isValid()); 4578 4579 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 4580 << owner.Variable << capturer->getSourceRange(); 4581 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 4582 << owner.Indirect << owner.Range; 4583} 4584 4585/// Check for a keyword selector that starts with the word 'add' or 4586/// 'set'. 4587static bool isSetterLikeSelector(Selector sel) { 4588 if (sel.isUnarySelector()) return false; 4589 4590 StringRef str = sel.getNameForSlot(0); 4591 while (!str.empty() && str.front() == '_') str = str.substr(1); 4592 if (str.startswith("set")) 4593 str = str.substr(3); 4594 else if (str.startswith("add")) { 4595 // Specially whitelist 'addOperationWithBlock:'. 4596 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 4597 return false; 4598 str = str.substr(3); 4599 } 4600 else 4601 return false; 4602 4603 if (str.empty()) return true; 4604 return !islower(str.front()); 4605} 4606 4607/// Check a message send to see if it's likely to cause a retain cycle. 4608void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 4609 // Only check instance methods whose selector looks like a setter. 4610 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 4611 return; 4612 4613 // Try to find a variable that the receiver is strongly owned by. 4614 RetainCycleOwner owner; 4615 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 4616 if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner)) 4617 return; 4618 } else { 4619 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 4620 owner.Variable = getCurMethodDecl()->getSelfDecl(); 4621 owner.Loc = msg->getSuperLoc(); 4622 owner.Range = msg->getSuperLoc(); 4623 } 4624 4625 // Check whether the receiver is captured by any of the arguments. 4626 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 4627 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 4628 return diagnoseRetainCycle(*this, capturer, owner); 4629} 4630 4631/// Check a property assign to see if it's likely to cause a retain cycle. 4632void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 4633 RetainCycleOwner owner; 4634 if (!findRetainCycleOwner(*this, receiver, owner)) 4635 return; 4636 4637 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 4638 diagnoseRetainCycle(*this, capturer, owner); 4639} 4640 4641bool Sema::checkUnsafeAssigns(SourceLocation Loc, 4642 QualType LHS, Expr *RHS) { 4643 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 4644 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 4645 return false; 4646 // strip off any implicit cast added to get to the one arc-specific 4647 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4648 if (cast->getCastKind() == CK_ARCConsumeObject) { 4649 Diag(Loc, diag::warn_arc_retained_assign) 4650 << (LT == Qualifiers::OCL_ExplicitNone) 4651 << RHS->getSourceRange(); 4652 return true; 4653 } 4654 RHS = cast->getSubExpr(); 4655 } 4656 return false; 4657} 4658 4659void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 4660 Expr *LHS, Expr *RHS) { 4661 QualType LHSType; 4662 // PropertyRef on LHS type need be directly obtained from 4663 // its declaration as it has a PsuedoType. 4664 ObjCPropertyRefExpr *PRE 4665 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens()); 4666 if (PRE && !PRE->isImplicitProperty()) { 4667 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 4668 if (PD) 4669 LHSType = PD->getType(); 4670 } 4671 4672 if (LHSType.isNull()) 4673 LHSType = LHS->getType(); 4674 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 4675 return; 4676 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 4677 // FIXME. Check for other life times. 4678 if (LT != Qualifiers::OCL_None) 4679 return; 4680 4681 if (PRE) { 4682 if (PRE->isImplicitProperty()) 4683 return; 4684 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 4685 if (!PD) 4686 return; 4687 4688 unsigned Attributes = PD->getPropertyAttributes(); 4689 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) { 4690 // when 'assign' attribute was not explicitly specified 4691 // by user, ignore it and rely on property type itself 4692 // for lifetime info. 4693 unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten(); 4694 if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) && 4695 LHSType->isObjCRetainableType()) 4696 return; 4697 4698 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4699 if (cast->getCastKind() == CK_ARCConsumeObject) { 4700 Diag(Loc, diag::warn_arc_retained_property_assign) 4701 << RHS->getSourceRange(); 4702 return; 4703 } 4704 RHS = cast->getSubExpr(); 4705 } 4706 } 4707 } 4708} 4709