SemaChecking.cpp revision 3b6bef9a213249c6ab6d67c07b1ac6380961be3e
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/Lookup.h" 20#include "clang/Sema/ScopeInfo.h" 21#include "clang/Analysis/Analyses/FormatString.h" 22#include "clang/AST/ASTContext.h" 23#include "clang/AST/CharUnits.h" 24#include "clang/AST/DeclCXX.h" 25#include "clang/AST/DeclObjC.h" 26#include "clang/AST/Expr.h" 27#include "clang/AST/ExprCXX.h" 28#include "clang/AST/ExprObjC.h" 29#include "clang/AST/EvaluatedExprVisitor.h" 30#include "clang/AST/DeclObjC.h" 31#include "clang/AST/StmtCXX.h" 32#include "clang/AST/StmtObjC.h" 33#include "clang/Lex/Preprocessor.h" 34#include "llvm/ADT/BitVector.h" 35#include "llvm/ADT/SmallString.h" 36#include "llvm/ADT/STLExtras.h" 37#include "llvm/Support/raw_ostream.h" 38#include "clang/Basic/TargetBuiltins.h" 39#include "clang/Basic/TargetInfo.h" 40#include "clang/Basic/ConvertUTF.h" 41#include <limits> 42using namespace clang; 43using namespace sema; 44 45SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 46 unsigned ByteNo) const { 47 return SL->getLocationOfByte(ByteNo, PP.getSourceManager(), 48 PP.getLangOpts(), PP.getTargetInfo()); 49} 50 51/// Checks that a call expression's argument count is the desired number. 52/// This is useful when doing custom type-checking. Returns true on error. 53static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) { 54 unsigned argCount = call->getNumArgs(); 55 if (argCount == desiredArgCount) return false; 56 57 if (argCount < desiredArgCount) 58 return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args) 59 << 0 /*function call*/ << desiredArgCount << argCount 60 << call->getSourceRange(); 61 62 // Highlight all the excess arguments. 63 SourceRange range(call->getArg(desiredArgCount)->getLocStart(), 64 call->getArg(argCount - 1)->getLocEnd()); 65 66 return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args) 67 << 0 /*function call*/ << desiredArgCount << argCount 68 << call->getArg(1)->getSourceRange(); 69} 70 71/// Check that the first argument to __builtin_annotation is an integer 72/// and the second argument is a non-wide string literal. 73static bool SemaBuiltinAnnotation(Sema &S, CallExpr *TheCall) { 74 if (checkArgCount(S, TheCall, 2)) 75 return true; 76 77 // First argument should be an integer. 78 Expr *ValArg = TheCall->getArg(0); 79 QualType Ty = ValArg->getType(); 80 if (!Ty->isIntegerType()) { 81 S.Diag(ValArg->getLocStart(), diag::err_builtin_annotation_first_arg) 82 << ValArg->getSourceRange(); 83 return true; 84 } 85 86 // Second argument should be a constant string. 87 Expr *StrArg = TheCall->getArg(1)->IgnoreParenCasts(); 88 StringLiteral *Literal = dyn_cast<StringLiteral>(StrArg); 89 if (!Literal || !Literal->isAscii()) { 90 S.Diag(StrArg->getLocStart(), diag::err_builtin_annotation_second_arg) 91 << StrArg->getSourceRange(); 92 return true; 93 } 94 95 TheCall->setType(Ty); 96 return false; 97} 98 99ExprResult 100Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 101 ExprResult TheCallResult(Owned(TheCall)); 102 103 // Find out if any arguments are required to be integer constant expressions. 104 unsigned ICEArguments = 0; 105 ASTContext::GetBuiltinTypeError Error; 106 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments); 107 if (Error != ASTContext::GE_None) 108 ICEArguments = 0; // Don't diagnose previously diagnosed errors. 109 110 // If any arguments are required to be ICE's, check and diagnose. 111 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) { 112 // Skip arguments not required to be ICE's. 113 if ((ICEArguments & (1 << ArgNo)) == 0) continue; 114 115 llvm::APSInt Result; 116 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 117 return true; 118 ICEArguments &= ~(1 << ArgNo); 119 } 120 121 switch (BuiltinID) { 122 case Builtin::BI__builtin___CFStringMakeConstantString: 123 assert(TheCall->getNumArgs() == 1 && 124 "Wrong # arguments to builtin CFStringMakeConstantString"); 125 if (CheckObjCString(TheCall->getArg(0))) 126 return ExprError(); 127 break; 128 case Builtin::BI__builtin_stdarg_start: 129 case Builtin::BI__builtin_va_start: 130 if (SemaBuiltinVAStart(TheCall)) 131 return ExprError(); 132 break; 133 case Builtin::BI__builtin_isgreater: 134 case Builtin::BI__builtin_isgreaterequal: 135 case Builtin::BI__builtin_isless: 136 case Builtin::BI__builtin_islessequal: 137 case Builtin::BI__builtin_islessgreater: 138 case Builtin::BI__builtin_isunordered: 139 if (SemaBuiltinUnorderedCompare(TheCall)) 140 return ExprError(); 141 break; 142 case Builtin::BI__builtin_fpclassify: 143 if (SemaBuiltinFPClassification(TheCall, 6)) 144 return ExprError(); 145 break; 146 case Builtin::BI__builtin_isfinite: 147 case Builtin::BI__builtin_isinf: 148 case Builtin::BI__builtin_isinf_sign: 149 case Builtin::BI__builtin_isnan: 150 case Builtin::BI__builtin_isnormal: 151 if (SemaBuiltinFPClassification(TheCall, 1)) 152 return ExprError(); 153 break; 154 case Builtin::BI__builtin_shufflevector: 155 return SemaBuiltinShuffleVector(TheCall); 156 // TheCall will be freed by the smart pointer here, but that's fine, since 157 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 158 case Builtin::BI__builtin_prefetch: 159 if (SemaBuiltinPrefetch(TheCall)) 160 return ExprError(); 161 break; 162 case Builtin::BI__builtin_object_size: 163 if (SemaBuiltinObjectSize(TheCall)) 164 return ExprError(); 165 break; 166 case Builtin::BI__builtin_longjmp: 167 if (SemaBuiltinLongjmp(TheCall)) 168 return ExprError(); 169 break; 170 171 case Builtin::BI__builtin_classify_type: 172 if (checkArgCount(*this, TheCall, 1)) return true; 173 TheCall->setType(Context.IntTy); 174 break; 175 case Builtin::BI__builtin_constant_p: 176 if (checkArgCount(*this, TheCall, 1)) return true; 177 TheCall->setType(Context.IntTy); 178 break; 179 case Builtin::BI__sync_fetch_and_add: 180 case Builtin::BI__sync_fetch_and_add_1: 181 case Builtin::BI__sync_fetch_and_add_2: 182 case Builtin::BI__sync_fetch_and_add_4: 183 case Builtin::BI__sync_fetch_and_add_8: 184 case Builtin::BI__sync_fetch_and_add_16: 185 case Builtin::BI__sync_fetch_and_sub: 186 case Builtin::BI__sync_fetch_and_sub_1: 187 case Builtin::BI__sync_fetch_and_sub_2: 188 case Builtin::BI__sync_fetch_and_sub_4: 189 case Builtin::BI__sync_fetch_and_sub_8: 190 case Builtin::BI__sync_fetch_and_sub_16: 191 case Builtin::BI__sync_fetch_and_or: 192 case Builtin::BI__sync_fetch_and_or_1: 193 case Builtin::BI__sync_fetch_and_or_2: 194 case Builtin::BI__sync_fetch_and_or_4: 195 case Builtin::BI__sync_fetch_and_or_8: 196 case Builtin::BI__sync_fetch_and_or_16: 197 case Builtin::BI__sync_fetch_and_and: 198 case Builtin::BI__sync_fetch_and_and_1: 199 case Builtin::BI__sync_fetch_and_and_2: 200 case Builtin::BI__sync_fetch_and_and_4: 201 case Builtin::BI__sync_fetch_and_and_8: 202 case Builtin::BI__sync_fetch_and_and_16: 203 case Builtin::BI__sync_fetch_and_xor: 204 case Builtin::BI__sync_fetch_and_xor_1: 205 case Builtin::BI__sync_fetch_and_xor_2: 206 case Builtin::BI__sync_fetch_and_xor_4: 207 case Builtin::BI__sync_fetch_and_xor_8: 208 case Builtin::BI__sync_fetch_and_xor_16: 209 case Builtin::BI__sync_add_and_fetch: 210 case Builtin::BI__sync_add_and_fetch_1: 211 case Builtin::BI__sync_add_and_fetch_2: 212 case Builtin::BI__sync_add_and_fetch_4: 213 case Builtin::BI__sync_add_and_fetch_8: 214 case Builtin::BI__sync_add_and_fetch_16: 215 case Builtin::BI__sync_sub_and_fetch: 216 case Builtin::BI__sync_sub_and_fetch_1: 217 case Builtin::BI__sync_sub_and_fetch_2: 218 case Builtin::BI__sync_sub_and_fetch_4: 219 case Builtin::BI__sync_sub_and_fetch_8: 220 case Builtin::BI__sync_sub_and_fetch_16: 221 case Builtin::BI__sync_and_and_fetch: 222 case Builtin::BI__sync_and_and_fetch_1: 223 case Builtin::BI__sync_and_and_fetch_2: 224 case Builtin::BI__sync_and_and_fetch_4: 225 case Builtin::BI__sync_and_and_fetch_8: 226 case Builtin::BI__sync_and_and_fetch_16: 227 case Builtin::BI__sync_or_and_fetch: 228 case Builtin::BI__sync_or_and_fetch_1: 229 case Builtin::BI__sync_or_and_fetch_2: 230 case Builtin::BI__sync_or_and_fetch_4: 231 case Builtin::BI__sync_or_and_fetch_8: 232 case Builtin::BI__sync_or_and_fetch_16: 233 case Builtin::BI__sync_xor_and_fetch: 234 case Builtin::BI__sync_xor_and_fetch_1: 235 case Builtin::BI__sync_xor_and_fetch_2: 236 case Builtin::BI__sync_xor_and_fetch_4: 237 case Builtin::BI__sync_xor_and_fetch_8: 238 case Builtin::BI__sync_xor_and_fetch_16: 239 case Builtin::BI__sync_val_compare_and_swap: 240 case Builtin::BI__sync_val_compare_and_swap_1: 241 case Builtin::BI__sync_val_compare_and_swap_2: 242 case Builtin::BI__sync_val_compare_and_swap_4: 243 case Builtin::BI__sync_val_compare_and_swap_8: 244 case Builtin::BI__sync_val_compare_and_swap_16: 245 case Builtin::BI__sync_bool_compare_and_swap: 246 case Builtin::BI__sync_bool_compare_and_swap_1: 247 case Builtin::BI__sync_bool_compare_and_swap_2: 248 case Builtin::BI__sync_bool_compare_and_swap_4: 249 case Builtin::BI__sync_bool_compare_and_swap_8: 250 case Builtin::BI__sync_bool_compare_and_swap_16: 251 case Builtin::BI__sync_lock_test_and_set: 252 case Builtin::BI__sync_lock_test_and_set_1: 253 case Builtin::BI__sync_lock_test_and_set_2: 254 case Builtin::BI__sync_lock_test_and_set_4: 255 case Builtin::BI__sync_lock_test_and_set_8: 256 case Builtin::BI__sync_lock_test_and_set_16: 257 case Builtin::BI__sync_lock_release: 258 case Builtin::BI__sync_lock_release_1: 259 case Builtin::BI__sync_lock_release_2: 260 case Builtin::BI__sync_lock_release_4: 261 case Builtin::BI__sync_lock_release_8: 262 case Builtin::BI__sync_lock_release_16: 263 case Builtin::BI__sync_swap: 264 case Builtin::BI__sync_swap_1: 265 case Builtin::BI__sync_swap_2: 266 case Builtin::BI__sync_swap_4: 267 case Builtin::BI__sync_swap_8: 268 case Builtin::BI__sync_swap_16: 269 return SemaBuiltinAtomicOverloaded(TheCallResult); 270#define BUILTIN(ID, TYPE, ATTRS) 271#define ATOMIC_BUILTIN(ID, TYPE, ATTRS) \ 272 case Builtin::BI##ID: \ 273 return SemaAtomicOpsOverloaded(TheCallResult, AtomicExpr::AO##ID); 274#include "clang/Basic/Builtins.def" 275 case Builtin::BI__builtin_annotation: 276 if (SemaBuiltinAnnotation(*this, TheCall)) 277 return ExprError(); 278 break; 279 } 280 281 // Since the target specific builtins for each arch overlap, only check those 282 // of the arch we are compiling for. 283 if (BuiltinID >= Builtin::FirstTSBuiltin) { 284 switch (Context.getTargetInfo().getTriple().getArch()) { 285 case llvm::Triple::arm: 286 case llvm::Triple::thumb: 287 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 288 return ExprError(); 289 break; 290 case llvm::Triple::mips: 291 case llvm::Triple::mipsel: 292 case llvm::Triple::mips64: 293 case llvm::Triple::mips64el: 294 if (CheckMipsBuiltinFunctionCall(BuiltinID, TheCall)) 295 return ExprError(); 296 break; 297 default: 298 break; 299 } 300 } 301 302 return TheCallResult; 303} 304 305// Get the valid immediate range for the specified NEON type code. 306static unsigned RFT(unsigned t, bool shift = false) { 307 NeonTypeFlags Type(t); 308 int IsQuad = Type.isQuad(); 309 switch (Type.getEltType()) { 310 case NeonTypeFlags::Int8: 311 case NeonTypeFlags::Poly8: 312 return shift ? 7 : (8 << IsQuad) - 1; 313 case NeonTypeFlags::Int16: 314 case NeonTypeFlags::Poly16: 315 return shift ? 15 : (4 << IsQuad) - 1; 316 case NeonTypeFlags::Int32: 317 return shift ? 31 : (2 << IsQuad) - 1; 318 case NeonTypeFlags::Int64: 319 return shift ? 63 : (1 << IsQuad) - 1; 320 case NeonTypeFlags::Float16: 321 assert(!shift && "cannot shift float types!"); 322 return (4 << IsQuad) - 1; 323 case NeonTypeFlags::Float32: 324 assert(!shift && "cannot shift float types!"); 325 return (2 << IsQuad) - 1; 326 } 327 llvm_unreachable("Invalid NeonTypeFlag!"); 328} 329 330/// getNeonEltType - Return the QualType corresponding to the elements of 331/// the vector type specified by the NeonTypeFlags. This is used to check 332/// the pointer arguments for Neon load/store intrinsics. 333static QualType getNeonEltType(NeonTypeFlags Flags, ASTContext &Context) { 334 switch (Flags.getEltType()) { 335 case NeonTypeFlags::Int8: 336 return Flags.isUnsigned() ? Context.UnsignedCharTy : Context.SignedCharTy; 337 case NeonTypeFlags::Int16: 338 return Flags.isUnsigned() ? Context.UnsignedShortTy : Context.ShortTy; 339 case NeonTypeFlags::Int32: 340 return Flags.isUnsigned() ? Context.UnsignedIntTy : Context.IntTy; 341 case NeonTypeFlags::Int64: 342 return Flags.isUnsigned() ? Context.UnsignedLongLongTy : Context.LongLongTy; 343 case NeonTypeFlags::Poly8: 344 return Context.SignedCharTy; 345 case NeonTypeFlags::Poly16: 346 return Context.ShortTy; 347 case NeonTypeFlags::Float16: 348 return Context.UnsignedShortTy; 349 case NeonTypeFlags::Float32: 350 return Context.FloatTy; 351 } 352 llvm_unreachable("Invalid NeonTypeFlag!"); 353} 354 355bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 356 llvm::APSInt Result; 357 358 uint64_t mask = 0; 359 unsigned TV = 0; 360 int PtrArgNum = -1; 361 bool HasConstPtr = false; 362 switch (BuiltinID) { 363#define GET_NEON_OVERLOAD_CHECK 364#include "clang/Basic/arm_neon.inc" 365#undef GET_NEON_OVERLOAD_CHECK 366 } 367 368 // For NEON intrinsics which are overloaded on vector element type, validate 369 // the immediate which specifies which variant to emit. 370 unsigned ImmArg = TheCall->getNumArgs()-1; 371 if (mask) { 372 if (SemaBuiltinConstantArg(TheCall, ImmArg, Result)) 373 return true; 374 375 TV = Result.getLimitedValue(64); 376 if ((TV > 63) || (mask & (1ULL << TV)) == 0) 377 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 378 << TheCall->getArg(ImmArg)->getSourceRange(); 379 } 380 381 if (PtrArgNum >= 0) { 382 // Check that pointer arguments have the specified type. 383 Expr *Arg = TheCall->getArg(PtrArgNum); 384 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) 385 Arg = ICE->getSubExpr(); 386 ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg); 387 QualType RHSTy = RHS.get()->getType(); 388 QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context); 389 if (HasConstPtr) 390 EltTy = EltTy.withConst(); 391 QualType LHSTy = Context.getPointerType(EltTy); 392 AssignConvertType ConvTy; 393 ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS); 394 if (RHS.isInvalid()) 395 return true; 396 if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy, 397 RHS.get(), AA_Assigning)) 398 return true; 399 } 400 401 // For NEON intrinsics which take an immediate value as part of the 402 // instruction, range check them here. 403 unsigned i = 0, l = 0, u = 0; 404 switch (BuiltinID) { 405 default: return false; 406 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 407 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 408 case ARM::BI__builtin_arm_vcvtr_f: 409 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 410#define GET_NEON_IMMEDIATE_CHECK 411#include "clang/Basic/arm_neon.inc" 412#undef GET_NEON_IMMEDIATE_CHECK 413 }; 414 415 // We can't check the value of a dependent argument. 416 if (TheCall->getArg(i)->isTypeDependent() || 417 TheCall->getArg(i)->isValueDependent()) 418 return false; 419 420 // Check that the immediate argument is actually a constant. 421 if (SemaBuiltinConstantArg(TheCall, i, Result)) 422 return true; 423 424 // Range check against the upper/lower values for this isntruction. 425 unsigned Val = Result.getZExtValue(); 426 if (Val < l || Val > (u + l)) 427 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 428 << l << u+l << TheCall->getArg(i)->getSourceRange(); 429 430 // FIXME: VFP Intrinsics should error if VFP not present. 431 return false; 432} 433 434bool Sema::CheckMipsBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 435 unsigned i = 0, l = 0, u = 0; 436 switch (BuiltinID) { 437 default: return false; 438 case Mips::BI__builtin_mips_wrdsp: i = 1; l = 0; u = 63; break; 439 case Mips::BI__builtin_mips_rddsp: i = 0; l = 0; u = 63; break; 440 }; 441 442 // We can't check the value of a dependent argument. 443 if (TheCall->getArg(i)->isTypeDependent() || 444 TheCall->getArg(i)->isValueDependent()) 445 return false; 446 447 // Check that the immediate argument is actually a constant. 448 llvm::APSInt Result; 449 if (SemaBuiltinConstantArg(TheCall, i, Result)) 450 return true; 451 452 // Range check against the upper/lower values for this instruction. 453 unsigned Val = Result.getZExtValue(); 454 if (Val < l || Val > u) 455 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 456 << l << u << TheCall->getArg(i)->getSourceRange(); 457 458 return false; 459} 460 461/// Given a FunctionDecl's FormatAttr, attempts to populate the FomatStringInfo 462/// parameter with the FormatAttr's correct format_idx and firstDataArg. 463/// Returns true when the format fits the function and the FormatStringInfo has 464/// been populated. 465bool Sema::getFormatStringInfo(const FormatAttr *Format, bool IsCXXMember, 466 FormatStringInfo *FSI) { 467 FSI->HasVAListArg = Format->getFirstArg() == 0; 468 FSI->FormatIdx = Format->getFormatIdx() - 1; 469 FSI->FirstDataArg = FSI->HasVAListArg ? 0 : Format->getFirstArg() - 1; 470 471 // The way the format attribute works in GCC, the implicit this argument 472 // of member functions is counted. However, it doesn't appear in our own 473 // lists, so decrement format_idx in that case. 474 if (IsCXXMember) { 475 if(FSI->FormatIdx == 0) 476 return false; 477 --FSI->FormatIdx; 478 if (FSI->FirstDataArg != 0) 479 --FSI->FirstDataArg; 480 } 481 return true; 482} 483 484/// Handles the checks for format strings, non-POD arguments to vararg 485/// functions, and NULL arguments passed to non-NULL parameters. 486void Sema::checkCall(NamedDecl *FDecl, Expr **Args, 487 unsigned NumArgs, 488 unsigned NumProtoArgs, 489 bool IsMemberFunction, 490 SourceLocation Loc, 491 SourceRange Range, 492 VariadicCallType CallType) { 493 // FIXME: This mechanism should be abstracted to be less fragile and 494 // more efficient. For example, just map function ids to custom 495 // handlers. 496 497 // Printf and scanf checking. 498 bool HandledFormatString = false; 499 for (specific_attr_iterator<FormatAttr> 500 I = FDecl->specific_attr_begin<FormatAttr>(), 501 E = FDecl->specific_attr_end<FormatAttr>(); I != E ; ++I) 502 if (CheckFormatArguments(*I, Args, NumArgs, IsMemberFunction, CallType, 503 Loc, Range)) 504 HandledFormatString = true; 505 506 // Refuse POD arguments that weren't caught by the format string 507 // checks above. 508 if (!HandledFormatString && CallType != VariadicDoesNotApply) 509 for (unsigned ArgIdx = NumProtoArgs; ArgIdx < NumArgs; ++ArgIdx) 510 variadicArgumentPODCheck(Args[ArgIdx], CallType); 511 512 for (specific_attr_iterator<NonNullAttr> 513 I = FDecl->specific_attr_begin<NonNullAttr>(), 514 E = FDecl->specific_attr_end<NonNullAttr>(); I != E; ++I) 515 CheckNonNullArguments(*I, Args, Loc); 516 517 // Type safety checking. 518 for (specific_attr_iterator<ArgumentWithTypeTagAttr> 519 i = FDecl->specific_attr_begin<ArgumentWithTypeTagAttr>(), 520 e = FDecl->specific_attr_end<ArgumentWithTypeTagAttr>(); i != e; ++i) { 521 CheckArgumentWithTypeTag(*i, Args); 522 } 523} 524 525/// CheckConstructorCall - Check a constructor call for correctness and safety 526/// properties not enforced by the C type system. 527void Sema::CheckConstructorCall(FunctionDecl *FDecl, Expr **Args, 528 unsigned NumArgs, 529 const FunctionProtoType *Proto, 530 SourceLocation Loc) { 531 VariadicCallType CallType = 532 Proto->isVariadic() ? VariadicConstructor : VariadicDoesNotApply; 533 checkCall(FDecl, Args, NumArgs, Proto->getNumArgs(), 534 /*IsMemberFunction=*/true, Loc, SourceRange(), CallType); 535} 536 537/// CheckFunctionCall - Check a direct function call for various correctness 538/// and safety properties not strictly enforced by the C type system. 539bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall, 540 const FunctionProtoType *Proto) { 541 bool IsMemberFunction = isa<CXXMemberCallExpr>(TheCall); 542 VariadicCallType CallType = getVariadicCallType(FDecl, Proto, 543 TheCall->getCallee()); 544 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 545 checkCall(FDecl, TheCall->getArgs(), TheCall->getNumArgs(), NumProtoArgs, 546 IsMemberFunction, TheCall->getRParenLoc(), 547 TheCall->getCallee()->getSourceRange(), CallType); 548 549 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 550 // None of the checks below are needed for functions that don't have 551 // simple names (e.g., C++ conversion functions). 552 if (!FnInfo) 553 return false; 554 555 unsigned CMId = FDecl->getMemoryFunctionKind(); 556 if (CMId == 0) 557 return false; 558 559 // Handle memory setting and copying functions. 560 if (CMId == Builtin::BIstrlcpy || CMId == Builtin::BIstrlcat) 561 CheckStrlcpycatArguments(TheCall, FnInfo); 562 else if (CMId == Builtin::BIstrncat) 563 CheckStrncatArguments(TheCall, FnInfo); 564 else 565 CheckMemaccessArguments(TheCall, CMId, FnInfo); 566 567 return false; 568} 569 570bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac, 571 Expr **Args, unsigned NumArgs) { 572 VariadicCallType CallType = 573 Method->isVariadic() ? VariadicMethod : VariadicDoesNotApply; 574 575 checkCall(Method, Args, NumArgs, Method->param_size(), 576 /*IsMemberFunction=*/false, 577 lbrac, Method->getSourceRange(), CallType); 578 579 return false; 580} 581 582bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall, 583 const FunctionProtoType *Proto) { 584 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 585 if (!V) 586 return false; 587 588 QualType Ty = V->getType(); 589 if (!Ty->isBlockPointerType()) 590 return false; 591 592 VariadicCallType CallType = 593 Proto && Proto->isVariadic() ? VariadicBlock : VariadicDoesNotApply ; 594 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 595 596 checkCall(NDecl, TheCall->getArgs(), TheCall->getNumArgs(), 597 NumProtoArgs, /*IsMemberFunction=*/false, 598 TheCall->getRParenLoc(), 599 TheCall->getCallee()->getSourceRange(), CallType); 600 601 return false; 602} 603 604ExprResult Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, 605 AtomicExpr::AtomicOp Op) { 606 CallExpr *TheCall = cast<CallExpr>(TheCallResult.get()); 607 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 608 609 // All these operations take one of the following forms: 610 enum { 611 // C __c11_atomic_init(A *, C) 612 Init, 613 // C __c11_atomic_load(A *, int) 614 Load, 615 // void __atomic_load(A *, CP, int) 616 Copy, 617 // C __c11_atomic_add(A *, M, int) 618 Arithmetic, 619 // C __atomic_exchange_n(A *, CP, int) 620 Xchg, 621 // void __atomic_exchange(A *, C *, CP, int) 622 GNUXchg, 623 // bool __c11_atomic_compare_exchange_strong(A *, C *, CP, int, int) 624 C11CmpXchg, 625 // bool __atomic_compare_exchange(A *, C *, CP, bool, int, int) 626 GNUCmpXchg 627 } Form = Init; 628 const unsigned NumArgs[] = { 2, 2, 3, 3, 3, 4, 5, 6 }; 629 const unsigned NumVals[] = { 1, 0, 1, 1, 1, 2, 2, 3 }; 630 // where: 631 // C is an appropriate type, 632 // A is volatile _Atomic(C) for __c11 builtins and is C for GNU builtins, 633 // CP is C for __c11 builtins and GNU _n builtins and is C * otherwise, 634 // M is C if C is an integer, and ptrdiff_t if C is a pointer, and 635 // the int parameters are for orderings. 636 637 assert(AtomicExpr::AO__c11_atomic_init == 0 && 638 AtomicExpr::AO__c11_atomic_fetch_xor + 1 == AtomicExpr::AO__atomic_load 639 && "need to update code for modified C11 atomics"); 640 bool IsC11 = Op >= AtomicExpr::AO__c11_atomic_init && 641 Op <= AtomicExpr::AO__c11_atomic_fetch_xor; 642 bool IsN = Op == AtomicExpr::AO__atomic_load_n || 643 Op == AtomicExpr::AO__atomic_store_n || 644 Op == AtomicExpr::AO__atomic_exchange_n || 645 Op == AtomicExpr::AO__atomic_compare_exchange_n; 646 bool IsAddSub = false; 647 648 switch (Op) { 649 case AtomicExpr::AO__c11_atomic_init: 650 Form = Init; 651 break; 652 653 case AtomicExpr::AO__c11_atomic_load: 654 case AtomicExpr::AO__atomic_load_n: 655 Form = Load; 656 break; 657 658 case AtomicExpr::AO__c11_atomic_store: 659 case AtomicExpr::AO__atomic_load: 660 case AtomicExpr::AO__atomic_store: 661 case AtomicExpr::AO__atomic_store_n: 662 Form = Copy; 663 break; 664 665 case AtomicExpr::AO__c11_atomic_fetch_add: 666 case AtomicExpr::AO__c11_atomic_fetch_sub: 667 case AtomicExpr::AO__atomic_fetch_add: 668 case AtomicExpr::AO__atomic_fetch_sub: 669 case AtomicExpr::AO__atomic_add_fetch: 670 case AtomicExpr::AO__atomic_sub_fetch: 671 IsAddSub = true; 672 // Fall through. 673 case AtomicExpr::AO__c11_atomic_fetch_and: 674 case AtomicExpr::AO__c11_atomic_fetch_or: 675 case AtomicExpr::AO__c11_atomic_fetch_xor: 676 case AtomicExpr::AO__atomic_fetch_and: 677 case AtomicExpr::AO__atomic_fetch_or: 678 case AtomicExpr::AO__atomic_fetch_xor: 679 case AtomicExpr::AO__atomic_fetch_nand: 680 case AtomicExpr::AO__atomic_and_fetch: 681 case AtomicExpr::AO__atomic_or_fetch: 682 case AtomicExpr::AO__atomic_xor_fetch: 683 case AtomicExpr::AO__atomic_nand_fetch: 684 Form = Arithmetic; 685 break; 686 687 case AtomicExpr::AO__c11_atomic_exchange: 688 case AtomicExpr::AO__atomic_exchange_n: 689 Form = Xchg; 690 break; 691 692 case AtomicExpr::AO__atomic_exchange: 693 Form = GNUXchg; 694 break; 695 696 case AtomicExpr::AO__c11_atomic_compare_exchange_strong: 697 case AtomicExpr::AO__c11_atomic_compare_exchange_weak: 698 Form = C11CmpXchg; 699 break; 700 701 case AtomicExpr::AO__atomic_compare_exchange: 702 case AtomicExpr::AO__atomic_compare_exchange_n: 703 Form = GNUCmpXchg; 704 break; 705 } 706 707 // Check we have the right number of arguments. 708 if (TheCall->getNumArgs() < NumArgs[Form]) { 709 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 710 << 0 << NumArgs[Form] << TheCall->getNumArgs() 711 << TheCall->getCallee()->getSourceRange(); 712 return ExprError(); 713 } else if (TheCall->getNumArgs() > NumArgs[Form]) { 714 Diag(TheCall->getArg(NumArgs[Form])->getLocStart(), 715 diag::err_typecheck_call_too_many_args) 716 << 0 << NumArgs[Form] << TheCall->getNumArgs() 717 << TheCall->getCallee()->getSourceRange(); 718 return ExprError(); 719 } 720 721 // Inspect the first argument of the atomic operation. 722 Expr *Ptr = TheCall->getArg(0); 723 Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get(); 724 const PointerType *pointerType = Ptr->getType()->getAs<PointerType>(); 725 if (!pointerType) { 726 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 727 << Ptr->getType() << Ptr->getSourceRange(); 728 return ExprError(); 729 } 730 731 // For a __c11 builtin, this should be a pointer to an _Atomic type. 732 QualType AtomTy = pointerType->getPointeeType(); // 'A' 733 QualType ValType = AtomTy; // 'C' 734 if (IsC11) { 735 if (!AtomTy->isAtomicType()) { 736 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 737 << Ptr->getType() << Ptr->getSourceRange(); 738 return ExprError(); 739 } 740 ValType = AtomTy->getAs<AtomicType>()->getValueType(); 741 } 742 743 // For an arithmetic operation, the implied arithmetic must be well-formed. 744 if (Form == Arithmetic) { 745 // gcc does not enforce these rules for GNU atomics, but we do so for sanity. 746 if (IsAddSub && !ValType->isIntegerType() && !ValType->isPointerType()) { 747 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 748 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 749 return ExprError(); 750 } 751 if (!IsAddSub && !ValType->isIntegerType()) { 752 Diag(DRE->getLocStart(), diag::err_atomic_op_bitwise_needs_atomic_int) 753 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 754 return ExprError(); 755 } 756 } else if (IsN && !ValType->isIntegerType() && !ValType->isPointerType()) { 757 // For __atomic_*_n operations, the value type must be a scalar integral or 758 // pointer type which is 1, 2, 4, 8 or 16 bytes in length. 759 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 760 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 761 return ExprError(); 762 } 763 764 if (!IsC11 && !AtomTy.isTriviallyCopyableType(Context)) { 765 // For GNU atomics, require a trivially-copyable type. This is not part of 766 // the GNU atomics specification, but we enforce it for sanity. 767 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_trivial_copy) 768 << Ptr->getType() << Ptr->getSourceRange(); 769 return ExprError(); 770 } 771 772 // FIXME: For any builtin other than a load, the ValType must not be 773 // const-qualified. 774 775 switch (ValType.getObjCLifetime()) { 776 case Qualifiers::OCL_None: 777 case Qualifiers::OCL_ExplicitNone: 778 // okay 779 break; 780 781 case Qualifiers::OCL_Weak: 782 case Qualifiers::OCL_Strong: 783 case Qualifiers::OCL_Autoreleasing: 784 // FIXME: Can this happen? By this point, ValType should be known 785 // to be trivially copyable. 786 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 787 << ValType << Ptr->getSourceRange(); 788 return ExprError(); 789 } 790 791 QualType ResultType = ValType; 792 if (Form == Copy || Form == GNUXchg || Form == Init) 793 ResultType = Context.VoidTy; 794 else if (Form == C11CmpXchg || Form == GNUCmpXchg) 795 ResultType = Context.BoolTy; 796 797 // The type of a parameter passed 'by value'. In the GNU atomics, such 798 // arguments are actually passed as pointers. 799 QualType ByValType = ValType; // 'CP' 800 if (!IsC11 && !IsN) 801 ByValType = Ptr->getType(); 802 803 // The first argument --- the pointer --- has a fixed type; we 804 // deduce the types of the rest of the arguments accordingly. Walk 805 // the remaining arguments, converting them to the deduced value type. 806 for (unsigned i = 1; i != NumArgs[Form]; ++i) { 807 QualType Ty; 808 if (i < NumVals[Form] + 1) { 809 switch (i) { 810 case 1: 811 // The second argument is the non-atomic operand. For arithmetic, this 812 // is always passed by value, and for a compare_exchange it is always 813 // passed by address. For the rest, GNU uses by-address and C11 uses 814 // by-value. 815 assert(Form != Load); 816 if (Form == Init || (Form == Arithmetic && ValType->isIntegerType())) 817 Ty = ValType; 818 else if (Form == Copy || Form == Xchg) 819 Ty = ByValType; 820 else if (Form == Arithmetic) 821 Ty = Context.getPointerDiffType(); 822 else 823 Ty = Context.getPointerType(ValType.getUnqualifiedType()); 824 break; 825 case 2: 826 // The third argument to compare_exchange / GNU exchange is a 827 // (pointer to a) desired value. 828 Ty = ByValType; 829 break; 830 case 3: 831 // The fourth argument to GNU compare_exchange is a 'weak' flag. 832 Ty = Context.BoolTy; 833 break; 834 } 835 } else { 836 // The order(s) are always converted to int. 837 Ty = Context.IntTy; 838 } 839 840 InitializedEntity Entity = 841 InitializedEntity::InitializeParameter(Context, Ty, false); 842 ExprResult Arg = TheCall->getArg(i); 843 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 844 if (Arg.isInvalid()) 845 return true; 846 TheCall->setArg(i, Arg.get()); 847 } 848 849 // Permute the arguments into a 'consistent' order. 850 SmallVector<Expr*, 5> SubExprs; 851 SubExprs.push_back(Ptr); 852 switch (Form) { 853 case Init: 854 // Note, AtomicExpr::getVal1() has a special case for this atomic. 855 SubExprs.push_back(TheCall->getArg(1)); // Val1 856 break; 857 case Load: 858 SubExprs.push_back(TheCall->getArg(1)); // Order 859 break; 860 case Copy: 861 case Arithmetic: 862 case Xchg: 863 SubExprs.push_back(TheCall->getArg(2)); // Order 864 SubExprs.push_back(TheCall->getArg(1)); // Val1 865 break; 866 case GNUXchg: 867 // Note, AtomicExpr::getVal2() has a special case for this atomic. 868 SubExprs.push_back(TheCall->getArg(3)); // Order 869 SubExprs.push_back(TheCall->getArg(1)); // Val1 870 SubExprs.push_back(TheCall->getArg(2)); // Val2 871 break; 872 case C11CmpXchg: 873 SubExprs.push_back(TheCall->getArg(3)); // Order 874 SubExprs.push_back(TheCall->getArg(1)); // Val1 875 SubExprs.push_back(TheCall->getArg(4)); // OrderFail 876 SubExprs.push_back(TheCall->getArg(2)); // Val2 877 break; 878 case GNUCmpXchg: 879 SubExprs.push_back(TheCall->getArg(4)); // Order 880 SubExprs.push_back(TheCall->getArg(1)); // Val1 881 SubExprs.push_back(TheCall->getArg(5)); // OrderFail 882 SubExprs.push_back(TheCall->getArg(2)); // Val2 883 SubExprs.push_back(TheCall->getArg(3)); // Weak 884 break; 885 } 886 887 return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 888 SubExprs, ResultType, Op, 889 TheCall->getRParenLoc())); 890} 891 892 893/// checkBuiltinArgument - Given a call to a builtin function, perform 894/// normal type-checking on the given argument, updating the call in 895/// place. This is useful when a builtin function requires custom 896/// type-checking for some of its arguments but not necessarily all of 897/// them. 898/// 899/// Returns true on error. 900static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 901 FunctionDecl *Fn = E->getDirectCallee(); 902 assert(Fn && "builtin call without direct callee!"); 903 904 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 905 InitializedEntity Entity = 906 InitializedEntity::InitializeParameter(S.Context, Param); 907 908 ExprResult Arg = E->getArg(0); 909 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 910 if (Arg.isInvalid()) 911 return true; 912 913 E->setArg(ArgIndex, Arg.take()); 914 return false; 915} 916 917/// SemaBuiltinAtomicOverloaded - We have a call to a function like 918/// __sync_fetch_and_add, which is an overloaded function based on the pointer 919/// type of its first argument. The main ActOnCallExpr routines have already 920/// promoted the types of arguments because all of these calls are prototyped as 921/// void(...). 922/// 923/// This function goes through and does final semantic checking for these 924/// builtins, 925ExprResult 926Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 927 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 928 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 929 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 930 931 // Ensure that we have at least one argument to do type inference from. 932 if (TheCall->getNumArgs() < 1) { 933 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 934 << 0 << 1 << TheCall->getNumArgs() 935 << TheCall->getCallee()->getSourceRange(); 936 return ExprError(); 937 } 938 939 // Inspect the first argument of the atomic builtin. This should always be 940 // a pointer type, whose element is an integral scalar or pointer type. 941 // Because it is a pointer type, we don't have to worry about any implicit 942 // casts here. 943 // FIXME: We don't allow floating point scalars as input. 944 Expr *FirstArg = TheCall->getArg(0); 945 ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg); 946 if (FirstArgResult.isInvalid()) 947 return ExprError(); 948 FirstArg = FirstArgResult.take(); 949 TheCall->setArg(0, FirstArg); 950 951 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 952 if (!pointerType) { 953 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 954 << FirstArg->getType() << FirstArg->getSourceRange(); 955 return ExprError(); 956 } 957 958 QualType ValType = pointerType->getPointeeType(); 959 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 960 !ValType->isBlockPointerType()) { 961 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 962 << FirstArg->getType() << FirstArg->getSourceRange(); 963 return ExprError(); 964 } 965 966 switch (ValType.getObjCLifetime()) { 967 case Qualifiers::OCL_None: 968 case Qualifiers::OCL_ExplicitNone: 969 // okay 970 break; 971 972 case Qualifiers::OCL_Weak: 973 case Qualifiers::OCL_Strong: 974 case Qualifiers::OCL_Autoreleasing: 975 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 976 << ValType << FirstArg->getSourceRange(); 977 return ExprError(); 978 } 979 980 // Strip any qualifiers off ValType. 981 ValType = ValType.getUnqualifiedType(); 982 983 // The majority of builtins return a value, but a few have special return 984 // types, so allow them to override appropriately below. 985 QualType ResultType = ValType; 986 987 // We need to figure out which concrete builtin this maps onto. For example, 988 // __sync_fetch_and_add with a 2 byte object turns into 989 // __sync_fetch_and_add_2. 990#define BUILTIN_ROW(x) \ 991 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 992 Builtin::BI##x##_8, Builtin::BI##x##_16 } 993 994 static const unsigned BuiltinIndices[][5] = { 995 BUILTIN_ROW(__sync_fetch_and_add), 996 BUILTIN_ROW(__sync_fetch_and_sub), 997 BUILTIN_ROW(__sync_fetch_and_or), 998 BUILTIN_ROW(__sync_fetch_and_and), 999 BUILTIN_ROW(__sync_fetch_and_xor), 1000 1001 BUILTIN_ROW(__sync_add_and_fetch), 1002 BUILTIN_ROW(__sync_sub_and_fetch), 1003 BUILTIN_ROW(__sync_and_and_fetch), 1004 BUILTIN_ROW(__sync_or_and_fetch), 1005 BUILTIN_ROW(__sync_xor_and_fetch), 1006 1007 BUILTIN_ROW(__sync_val_compare_and_swap), 1008 BUILTIN_ROW(__sync_bool_compare_and_swap), 1009 BUILTIN_ROW(__sync_lock_test_and_set), 1010 BUILTIN_ROW(__sync_lock_release), 1011 BUILTIN_ROW(__sync_swap) 1012 }; 1013#undef BUILTIN_ROW 1014 1015 // Determine the index of the size. 1016 unsigned SizeIndex; 1017 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 1018 case 1: SizeIndex = 0; break; 1019 case 2: SizeIndex = 1; break; 1020 case 4: SizeIndex = 2; break; 1021 case 8: SizeIndex = 3; break; 1022 case 16: SizeIndex = 4; break; 1023 default: 1024 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 1025 << FirstArg->getType() << FirstArg->getSourceRange(); 1026 return ExprError(); 1027 } 1028 1029 // Each of these builtins has one pointer argument, followed by some number of 1030 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 1031 // that we ignore. Find out which row of BuiltinIndices to read from as well 1032 // as the number of fixed args. 1033 unsigned BuiltinID = FDecl->getBuiltinID(); 1034 unsigned BuiltinIndex, NumFixed = 1; 1035 switch (BuiltinID) { 1036 default: llvm_unreachable("Unknown overloaded atomic builtin!"); 1037 case Builtin::BI__sync_fetch_and_add: 1038 case Builtin::BI__sync_fetch_and_add_1: 1039 case Builtin::BI__sync_fetch_and_add_2: 1040 case Builtin::BI__sync_fetch_and_add_4: 1041 case Builtin::BI__sync_fetch_and_add_8: 1042 case Builtin::BI__sync_fetch_and_add_16: 1043 BuiltinIndex = 0; 1044 break; 1045 1046 case Builtin::BI__sync_fetch_and_sub: 1047 case Builtin::BI__sync_fetch_and_sub_1: 1048 case Builtin::BI__sync_fetch_and_sub_2: 1049 case Builtin::BI__sync_fetch_and_sub_4: 1050 case Builtin::BI__sync_fetch_and_sub_8: 1051 case Builtin::BI__sync_fetch_and_sub_16: 1052 BuiltinIndex = 1; 1053 break; 1054 1055 case Builtin::BI__sync_fetch_and_or: 1056 case Builtin::BI__sync_fetch_and_or_1: 1057 case Builtin::BI__sync_fetch_and_or_2: 1058 case Builtin::BI__sync_fetch_and_or_4: 1059 case Builtin::BI__sync_fetch_and_or_8: 1060 case Builtin::BI__sync_fetch_and_or_16: 1061 BuiltinIndex = 2; 1062 break; 1063 1064 case Builtin::BI__sync_fetch_and_and: 1065 case Builtin::BI__sync_fetch_and_and_1: 1066 case Builtin::BI__sync_fetch_and_and_2: 1067 case Builtin::BI__sync_fetch_and_and_4: 1068 case Builtin::BI__sync_fetch_and_and_8: 1069 case Builtin::BI__sync_fetch_and_and_16: 1070 BuiltinIndex = 3; 1071 break; 1072 1073 case Builtin::BI__sync_fetch_and_xor: 1074 case Builtin::BI__sync_fetch_and_xor_1: 1075 case Builtin::BI__sync_fetch_and_xor_2: 1076 case Builtin::BI__sync_fetch_and_xor_4: 1077 case Builtin::BI__sync_fetch_and_xor_8: 1078 case Builtin::BI__sync_fetch_and_xor_16: 1079 BuiltinIndex = 4; 1080 break; 1081 1082 case Builtin::BI__sync_add_and_fetch: 1083 case Builtin::BI__sync_add_and_fetch_1: 1084 case Builtin::BI__sync_add_and_fetch_2: 1085 case Builtin::BI__sync_add_and_fetch_4: 1086 case Builtin::BI__sync_add_and_fetch_8: 1087 case Builtin::BI__sync_add_and_fetch_16: 1088 BuiltinIndex = 5; 1089 break; 1090 1091 case Builtin::BI__sync_sub_and_fetch: 1092 case Builtin::BI__sync_sub_and_fetch_1: 1093 case Builtin::BI__sync_sub_and_fetch_2: 1094 case Builtin::BI__sync_sub_and_fetch_4: 1095 case Builtin::BI__sync_sub_and_fetch_8: 1096 case Builtin::BI__sync_sub_and_fetch_16: 1097 BuiltinIndex = 6; 1098 break; 1099 1100 case Builtin::BI__sync_and_and_fetch: 1101 case Builtin::BI__sync_and_and_fetch_1: 1102 case Builtin::BI__sync_and_and_fetch_2: 1103 case Builtin::BI__sync_and_and_fetch_4: 1104 case Builtin::BI__sync_and_and_fetch_8: 1105 case Builtin::BI__sync_and_and_fetch_16: 1106 BuiltinIndex = 7; 1107 break; 1108 1109 case Builtin::BI__sync_or_and_fetch: 1110 case Builtin::BI__sync_or_and_fetch_1: 1111 case Builtin::BI__sync_or_and_fetch_2: 1112 case Builtin::BI__sync_or_and_fetch_4: 1113 case Builtin::BI__sync_or_and_fetch_8: 1114 case Builtin::BI__sync_or_and_fetch_16: 1115 BuiltinIndex = 8; 1116 break; 1117 1118 case Builtin::BI__sync_xor_and_fetch: 1119 case Builtin::BI__sync_xor_and_fetch_1: 1120 case Builtin::BI__sync_xor_and_fetch_2: 1121 case Builtin::BI__sync_xor_and_fetch_4: 1122 case Builtin::BI__sync_xor_and_fetch_8: 1123 case Builtin::BI__sync_xor_and_fetch_16: 1124 BuiltinIndex = 9; 1125 break; 1126 1127 case Builtin::BI__sync_val_compare_and_swap: 1128 case Builtin::BI__sync_val_compare_and_swap_1: 1129 case Builtin::BI__sync_val_compare_and_swap_2: 1130 case Builtin::BI__sync_val_compare_and_swap_4: 1131 case Builtin::BI__sync_val_compare_and_swap_8: 1132 case Builtin::BI__sync_val_compare_and_swap_16: 1133 BuiltinIndex = 10; 1134 NumFixed = 2; 1135 break; 1136 1137 case Builtin::BI__sync_bool_compare_and_swap: 1138 case Builtin::BI__sync_bool_compare_and_swap_1: 1139 case Builtin::BI__sync_bool_compare_and_swap_2: 1140 case Builtin::BI__sync_bool_compare_and_swap_4: 1141 case Builtin::BI__sync_bool_compare_and_swap_8: 1142 case Builtin::BI__sync_bool_compare_and_swap_16: 1143 BuiltinIndex = 11; 1144 NumFixed = 2; 1145 ResultType = Context.BoolTy; 1146 break; 1147 1148 case Builtin::BI__sync_lock_test_and_set: 1149 case Builtin::BI__sync_lock_test_and_set_1: 1150 case Builtin::BI__sync_lock_test_and_set_2: 1151 case Builtin::BI__sync_lock_test_and_set_4: 1152 case Builtin::BI__sync_lock_test_and_set_8: 1153 case Builtin::BI__sync_lock_test_and_set_16: 1154 BuiltinIndex = 12; 1155 break; 1156 1157 case Builtin::BI__sync_lock_release: 1158 case Builtin::BI__sync_lock_release_1: 1159 case Builtin::BI__sync_lock_release_2: 1160 case Builtin::BI__sync_lock_release_4: 1161 case Builtin::BI__sync_lock_release_8: 1162 case Builtin::BI__sync_lock_release_16: 1163 BuiltinIndex = 13; 1164 NumFixed = 0; 1165 ResultType = Context.VoidTy; 1166 break; 1167 1168 case Builtin::BI__sync_swap: 1169 case Builtin::BI__sync_swap_1: 1170 case Builtin::BI__sync_swap_2: 1171 case Builtin::BI__sync_swap_4: 1172 case Builtin::BI__sync_swap_8: 1173 case Builtin::BI__sync_swap_16: 1174 BuiltinIndex = 14; 1175 break; 1176 } 1177 1178 // Now that we know how many fixed arguments we expect, first check that we 1179 // have at least that many. 1180 if (TheCall->getNumArgs() < 1+NumFixed) { 1181 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 1182 << 0 << 1+NumFixed << TheCall->getNumArgs() 1183 << TheCall->getCallee()->getSourceRange(); 1184 return ExprError(); 1185 } 1186 1187 // Get the decl for the concrete builtin from this, we can tell what the 1188 // concrete integer type we should convert to is. 1189 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 1190 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 1191 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 1192 FunctionDecl *NewBuiltinDecl = 1193 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 1194 TUScope, false, DRE->getLocStart())); 1195 1196 // The first argument --- the pointer --- has a fixed type; we 1197 // deduce the types of the rest of the arguments accordingly. Walk 1198 // the remaining arguments, converting them to the deduced value type. 1199 for (unsigned i = 0; i != NumFixed; ++i) { 1200 ExprResult Arg = TheCall->getArg(i+1); 1201 1202 // GCC does an implicit conversion to the pointer or integer ValType. This 1203 // can fail in some cases (1i -> int**), check for this error case now. 1204 // Initialize the argument. 1205 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context, 1206 ValType, /*consume*/ false); 1207 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 1208 if (Arg.isInvalid()) 1209 return ExprError(); 1210 1211 // Okay, we have something that *can* be converted to the right type. Check 1212 // to see if there is a potentially weird extension going on here. This can 1213 // happen when you do an atomic operation on something like an char* and 1214 // pass in 42. The 42 gets converted to char. This is even more strange 1215 // for things like 45.123 -> char, etc. 1216 // FIXME: Do this check. 1217 TheCall->setArg(i+1, Arg.take()); 1218 } 1219 1220 ASTContext& Context = this->getASTContext(); 1221 1222 // Create a new DeclRefExpr to refer to the new decl. 1223 DeclRefExpr* NewDRE = DeclRefExpr::Create( 1224 Context, 1225 DRE->getQualifierLoc(), 1226 SourceLocation(), 1227 NewBuiltinDecl, 1228 /*enclosing*/ false, 1229 DRE->getLocation(), 1230 NewBuiltinDecl->getType(), 1231 DRE->getValueKind()); 1232 1233 // Set the callee in the CallExpr. 1234 // FIXME: This leaks the original parens and implicit casts. 1235 ExprResult PromotedCall = UsualUnaryConversions(NewDRE); 1236 if (PromotedCall.isInvalid()) 1237 return ExprError(); 1238 TheCall->setCallee(PromotedCall.take()); 1239 1240 // Change the result type of the call to match the original value type. This 1241 // is arbitrary, but the codegen for these builtins ins design to handle it 1242 // gracefully. 1243 TheCall->setType(ResultType); 1244 1245 return TheCallResult; 1246} 1247 1248/// CheckObjCString - Checks that the argument to the builtin 1249/// CFString constructor is correct 1250/// Note: It might also make sense to do the UTF-16 conversion here (would 1251/// simplify the backend). 1252bool Sema::CheckObjCString(Expr *Arg) { 1253 Arg = Arg->IgnoreParenCasts(); 1254 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 1255 1256 if (!Literal || !Literal->isAscii()) { 1257 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 1258 << Arg->getSourceRange(); 1259 return true; 1260 } 1261 1262 if (Literal->containsNonAsciiOrNull()) { 1263 StringRef String = Literal->getString(); 1264 unsigned NumBytes = String.size(); 1265 SmallVector<UTF16, 128> ToBuf(NumBytes); 1266 const UTF8 *FromPtr = (UTF8 *)String.data(); 1267 UTF16 *ToPtr = &ToBuf[0]; 1268 1269 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 1270 &ToPtr, ToPtr + NumBytes, 1271 strictConversion); 1272 // Check for conversion failure. 1273 if (Result != conversionOK) 1274 Diag(Arg->getLocStart(), 1275 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 1276 } 1277 return false; 1278} 1279 1280/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 1281/// Emit an error and return true on failure, return false on success. 1282bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 1283 Expr *Fn = TheCall->getCallee(); 1284 if (TheCall->getNumArgs() > 2) { 1285 Diag(TheCall->getArg(2)->getLocStart(), 1286 diag::err_typecheck_call_too_many_args) 1287 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1288 << Fn->getSourceRange() 1289 << SourceRange(TheCall->getArg(2)->getLocStart(), 1290 (*(TheCall->arg_end()-1))->getLocEnd()); 1291 return true; 1292 } 1293 1294 if (TheCall->getNumArgs() < 2) { 1295 return Diag(TheCall->getLocEnd(), 1296 diag::err_typecheck_call_too_few_args_at_least) 1297 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 1298 } 1299 1300 // Type-check the first argument normally. 1301 if (checkBuiltinArgument(*this, TheCall, 0)) 1302 return true; 1303 1304 // Determine whether the current function is variadic or not. 1305 BlockScopeInfo *CurBlock = getCurBlock(); 1306 bool isVariadic; 1307 if (CurBlock) 1308 isVariadic = CurBlock->TheDecl->isVariadic(); 1309 else if (FunctionDecl *FD = getCurFunctionDecl()) 1310 isVariadic = FD->isVariadic(); 1311 else 1312 isVariadic = getCurMethodDecl()->isVariadic(); 1313 1314 if (!isVariadic) { 1315 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 1316 return true; 1317 } 1318 1319 // Verify that the second argument to the builtin is the last argument of the 1320 // current function or method. 1321 bool SecondArgIsLastNamedArgument = false; 1322 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 1323 1324 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 1325 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 1326 // FIXME: This isn't correct for methods (results in bogus warning). 1327 // Get the last formal in the current function. 1328 const ParmVarDecl *LastArg; 1329 if (CurBlock) 1330 LastArg = *(CurBlock->TheDecl->param_end()-1); 1331 else if (FunctionDecl *FD = getCurFunctionDecl()) 1332 LastArg = *(FD->param_end()-1); 1333 else 1334 LastArg = *(getCurMethodDecl()->param_end()-1); 1335 SecondArgIsLastNamedArgument = PV == LastArg; 1336 } 1337 } 1338 1339 if (!SecondArgIsLastNamedArgument) 1340 Diag(TheCall->getArg(1)->getLocStart(), 1341 diag::warn_second_parameter_of_va_start_not_last_named_argument); 1342 return false; 1343} 1344 1345/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 1346/// friends. This is declared to take (...), so we have to check everything. 1347bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 1348 if (TheCall->getNumArgs() < 2) 1349 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1350 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 1351 if (TheCall->getNumArgs() > 2) 1352 return Diag(TheCall->getArg(2)->getLocStart(), 1353 diag::err_typecheck_call_too_many_args) 1354 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1355 << SourceRange(TheCall->getArg(2)->getLocStart(), 1356 (*(TheCall->arg_end()-1))->getLocEnd()); 1357 1358 ExprResult OrigArg0 = TheCall->getArg(0); 1359 ExprResult OrigArg1 = TheCall->getArg(1); 1360 1361 // Do standard promotions between the two arguments, returning their common 1362 // type. 1363 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 1364 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 1365 return true; 1366 1367 // Make sure any conversions are pushed back into the call; this is 1368 // type safe since unordered compare builtins are declared as "_Bool 1369 // foo(...)". 1370 TheCall->setArg(0, OrigArg0.get()); 1371 TheCall->setArg(1, OrigArg1.get()); 1372 1373 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 1374 return false; 1375 1376 // If the common type isn't a real floating type, then the arguments were 1377 // invalid for this operation. 1378 if (Res.isNull() || !Res->isRealFloatingType()) 1379 return Diag(OrigArg0.get()->getLocStart(), 1380 diag::err_typecheck_call_invalid_ordered_compare) 1381 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 1382 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 1383 1384 return false; 1385} 1386 1387/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 1388/// __builtin_isnan and friends. This is declared to take (...), so we have 1389/// to check everything. We expect the last argument to be a floating point 1390/// value. 1391bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 1392 if (TheCall->getNumArgs() < NumArgs) 1393 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1394 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 1395 if (TheCall->getNumArgs() > NumArgs) 1396 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 1397 diag::err_typecheck_call_too_many_args) 1398 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 1399 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 1400 (*(TheCall->arg_end()-1))->getLocEnd()); 1401 1402 Expr *OrigArg = TheCall->getArg(NumArgs-1); 1403 1404 if (OrigArg->isTypeDependent()) 1405 return false; 1406 1407 // This operation requires a non-_Complex floating-point number. 1408 if (!OrigArg->getType()->isRealFloatingType()) 1409 return Diag(OrigArg->getLocStart(), 1410 diag::err_typecheck_call_invalid_unary_fp) 1411 << OrigArg->getType() << OrigArg->getSourceRange(); 1412 1413 // If this is an implicit conversion from float -> double, remove it. 1414 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 1415 Expr *CastArg = Cast->getSubExpr(); 1416 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 1417 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 1418 "promotion from float to double is the only expected cast here"); 1419 Cast->setSubExpr(0); 1420 TheCall->setArg(NumArgs-1, CastArg); 1421 } 1422 } 1423 1424 return false; 1425} 1426 1427/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1428// This is declared to take (...), so we have to check everything. 1429ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1430 if (TheCall->getNumArgs() < 2) 1431 return ExprError(Diag(TheCall->getLocEnd(), 1432 diag::err_typecheck_call_too_few_args_at_least) 1433 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1434 << TheCall->getSourceRange()); 1435 1436 // Determine which of the following types of shufflevector we're checking: 1437 // 1) unary, vector mask: (lhs, mask) 1438 // 2) binary, vector mask: (lhs, rhs, mask) 1439 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1440 QualType resType = TheCall->getArg(0)->getType(); 1441 unsigned numElements = 0; 1442 1443 if (!TheCall->getArg(0)->isTypeDependent() && 1444 !TheCall->getArg(1)->isTypeDependent()) { 1445 QualType LHSType = TheCall->getArg(0)->getType(); 1446 QualType RHSType = TheCall->getArg(1)->getType(); 1447 1448 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 1449 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 1450 << SourceRange(TheCall->getArg(0)->getLocStart(), 1451 TheCall->getArg(1)->getLocEnd()); 1452 return ExprError(); 1453 } 1454 1455 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1456 unsigned numResElements = TheCall->getNumArgs() - 2; 1457 1458 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1459 // with mask. If so, verify that RHS is an integer vector type with the 1460 // same number of elts as lhs. 1461 if (TheCall->getNumArgs() == 2) { 1462 if (!RHSType->hasIntegerRepresentation() || 1463 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1464 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1465 << SourceRange(TheCall->getArg(1)->getLocStart(), 1466 TheCall->getArg(1)->getLocEnd()); 1467 numResElements = numElements; 1468 } 1469 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1470 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1471 << SourceRange(TheCall->getArg(0)->getLocStart(), 1472 TheCall->getArg(1)->getLocEnd()); 1473 return ExprError(); 1474 } else if (numElements != numResElements) { 1475 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1476 resType = Context.getVectorType(eltType, numResElements, 1477 VectorType::GenericVector); 1478 } 1479 } 1480 1481 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1482 if (TheCall->getArg(i)->isTypeDependent() || 1483 TheCall->getArg(i)->isValueDependent()) 1484 continue; 1485 1486 llvm::APSInt Result(32); 1487 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1488 return ExprError(Diag(TheCall->getLocStart(), 1489 diag::err_shufflevector_nonconstant_argument) 1490 << TheCall->getArg(i)->getSourceRange()); 1491 1492 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1493 return ExprError(Diag(TheCall->getLocStart(), 1494 diag::err_shufflevector_argument_too_large) 1495 << TheCall->getArg(i)->getSourceRange()); 1496 } 1497 1498 SmallVector<Expr*, 32> exprs; 1499 1500 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1501 exprs.push_back(TheCall->getArg(i)); 1502 TheCall->setArg(i, 0); 1503 } 1504 1505 return Owned(new (Context) ShuffleVectorExpr(Context, exprs, resType, 1506 TheCall->getCallee()->getLocStart(), 1507 TheCall->getRParenLoc())); 1508} 1509 1510/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1511// This is declared to take (const void*, ...) and can take two 1512// optional constant int args. 1513bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1514 unsigned NumArgs = TheCall->getNumArgs(); 1515 1516 if (NumArgs > 3) 1517 return Diag(TheCall->getLocEnd(), 1518 diag::err_typecheck_call_too_many_args_at_most) 1519 << 0 /*function call*/ << 3 << NumArgs 1520 << TheCall->getSourceRange(); 1521 1522 // Argument 0 is checked for us and the remaining arguments must be 1523 // constant integers. 1524 for (unsigned i = 1; i != NumArgs; ++i) { 1525 Expr *Arg = TheCall->getArg(i); 1526 1527 // We can't check the value of a dependent argument. 1528 if (Arg->isTypeDependent() || Arg->isValueDependent()) 1529 continue; 1530 1531 llvm::APSInt Result; 1532 if (SemaBuiltinConstantArg(TheCall, i, Result)) 1533 return true; 1534 1535 // FIXME: gcc issues a warning and rewrites these to 0. These 1536 // seems especially odd for the third argument since the default 1537 // is 3. 1538 if (i == 1) { 1539 if (Result.getLimitedValue() > 1) 1540 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1541 << "0" << "1" << Arg->getSourceRange(); 1542 } else { 1543 if (Result.getLimitedValue() > 3) 1544 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1545 << "0" << "3" << Arg->getSourceRange(); 1546 } 1547 } 1548 1549 return false; 1550} 1551 1552/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1553/// TheCall is a constant expression. 1554bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1555 llvm::APSInt &Result) { 1556 Expr *Arg = TheCall->getArg(ArgNum); 1557 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1558 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1559 1560 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 1561 1562 if (!Arg->isIntegerConstantExpr(Result, Context)) 1563 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 1564 << FDecl->getDeclName() << Arg->getSourceRange(); 1565 1566 return false; 1567} 1568 1569/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 1570/// int type). This simply type checks that type is one of the defined 1571/// constants (0-3). 1572// For compatibility check 0-3, llvm only handles 0 and 2. 1573bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 1574 llvm::APSInt Result; 1575 1576 // We can't check the value of a dependent argument. 1577 if (TheCall->getArg(1)->isTypeDependent() || 1578 TheCall->getArg(1)->isValueDependent()) 1579 return false; 1580 1581 // Check constant-ness first. 1582 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1583 return true; 1584 1585 Expr *Arg = TheCall->getArg(1); 1586 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 1587 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1588 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1589 } 1590 1591 return false; 1592} 1593 1594/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1595/// This checks that val is a constant 1. 1596bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1597 Expr *Arg = TheCall->getArg(1); 1598 llvm::APSInt Result; 1599 1600 // TODO: This is less than ideal. Overload this to take a value. 1601 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1602 return true; 1603 1604 if (Result != 1) 1605 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1606 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1607 1608 return false; 1609} 1610 1611// Determine if an expression is a string literal or constant string. 1612// If this function returns false on the arguments to a function expecting a 1613// format string, we will usually need to emit a warning. 1614// True string literals are then checked by CheckFormatString. 1615Sema::StringLiteralCheckType 1616Sema::checkFormatStringExpr(const Expr *E, Expr **Args, 1617 unsigned NumArgs, bool HasVAListArg, 1618 unsigned format_idx, unsigned firstDataArg, 1619 FormatStringType Type, VariadicCallType CallType, 1620 bool inFunctionCall) { 1621 tryAgain: 1622 if (E->isTypeDependent() || E->isValueDependent()) 1623 return SLCT_NotALiteral; 1624 1625 E = E->IgnoreParenCasts(); 1626 1627 if (E->isNullPointerConstant(Context, Expr::NPC_ValueDependentIsNotNull)) 1628 // Technically -Wformat-nonliteral does not warn about this case. 1629 // The behavior of printf and friends in this case is implementation 1630 // dependent. Ideally if the format string cannot be null then 1631 // it should have a 'nonnull' attribute in the function prototype. 1632 return SLCT_CheckedLiteral; 1633 1634 switch (E->getStmtClass()) { 1635 case Stmt::BinaryConditionalOperatorClass: 1636 case Stmt::ConditionalOperatorClass: { 1637 // The expression is a literal if both sub-expressions were, and it was 1638 // completely checked only if both sub-expressions were checked. 1639 const AbstractConditionalOperator *C = 1640 cast<AbstractConditionalOperator>(E); 1641 StringLiteralCheckType Left = 1642 checkFormatStringExpr(C->getTrueExpr(), Args, NumArgs, 1643 HasVAListArg, format_idx, firstDataArg, 1644 Type, CallType, inFunctionCall); 1645 if (Left == SLCT_NotALiteral) 1646 return SLCT_NotALiteral; 1647 StringLiteralCheckType Right = 1648 checkFormatStringExpr(C->getFalseExpr(), Args, NumArgs, 1649 HasVAListArg, format_idx, firstDataArg, 1650 Type, CallType, inFunctionCall); 1651 return Left < Right ? Left : Right; 1652 } 1653 1654 case Stmt::ImplicitCastExprClass: { 1655 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 1656 goto tryAgain; 1657 } 1658 1659 case Stmt::OpaqueValueExprClass: 1660 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 1661 E = src; 1662 goto tryAgain; 1663 } 1664 return SLCT_NotALiteral; 1665 1666 case Stmt::PredefinedExprClass: 1667 // While __func__, etc., are technically not string literals, they 1668 // cannot contain format specifiers and thus are not a security 1669 // liability. 1670 return SLCT_UncheckedLiteral; 1671 1672 case Stmt::DeclRefExprClass: { 1673 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1674 1675 // As an exception, do not flag errors for variables binding to 1676 // const string literals. 1677 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1678 bool isConstant = false; 1679 QualType T = DR->getType(); 1680 1681 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1682 isConstant = AT->getElementType().isConstant(Context); 1683 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1684 isConstant = T.isConstant(Context) && 1685 PT->getPointeeType().isConstant(Context); 1686 } else if (T->isObjCObjectPointerType()) { 1687 // In ObjC, there is usually no "const ObjectPointer" type, 1688 // so don't check if the pointee type is constant. 1689 isConstant = T.isConstant(Context); 1690 } 1691 1692 if (isConstant) { 1693 if (const Expr *Init = VD->getAnyInitializer()) { 1694 // Look through initializers like const char c[] = { "foo" } 1695 if (const InitListExpr *InitList = dyn_cast<InitListExpr>(Init)) { 1696 if (InitList->isStringLiteralInit()) 1697 Init = InitList->getInit(0)->IgnoreParenImpCasts(); 1698 } 1699 return checkFormatStringExpr(Init, Args, NumArgs, 1700 HasVAListArg, format_idx, 1701 firstDataArg, Type, CallType, 1702 /*inFunctionCall*/false); 1703 } 1704 } 1705 1706 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1707 // special check to see if the format string is a function parameter 1708 // of the function calling the printf function. If the function 1709 // has an attribute indicating it is a printf-like function, then we 1710 // should suppress warnings concerning non-literals being used in a call 1711 // to a vprintf function. For example: 1712 // 1713 // void 1714 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1715 // va_list ap; 1716 // va_start(ap, fmt); 1717 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1718 // ... 1719 // 1720 if (HasVAListArg) { 1721 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(VD)) { 1722 if (const NamedDecl *ND = dyn_cast<NamedDecl>(PV->getDeclContext())) { 1723 int PVIndex = PV->getFunctionScopeIndex() + 1; 1724 for (specific_attr_iterator<FormatAttr> 1725 i = ND->specific_attr_begin<FormatAttr>(), 1726 e = ND->specific_attr_end<FormatAttr>(); i != e ; ++i) { 1727 FormatAttr *PVFormat = *i; 1728 // adjust for implicit parameter 1729 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1730 if (MD->isInstance()) 1731 ++PVIndex; 1732 // We also check if the formats are compatible. 1733 // We can't pass a 'scanf' string to a 'printf' function. 1734 if (PVIndex == PVFormat->getFormatIdx() && 1735 Type == GetFormatStringType(PVFormat)) 1736 return SLCT_UncheckedLiteral; 1737 } 1738 } 1739 } 1740 } 1741 } 1742 1743 return SLCT_NotALiteral; 1744 } 1745 1746 case Stmt::CallExprClass: 1747 case Stmt::CXXMemberCallExprClass: { 1748 const CallExpr *CE = cast<CallExpr>(E); 1749 if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(CE->getCalleeDecl())) { 1750 if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) { 1751 unsigned ArgIndex = FA->getFormatIdx(); 1752 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1753 if (MD->isInstance()) 1754 --ArgIndex; 1755 const Expr *Arg = CE->getArg(ArgIndex - 1); 1756 1757 return checkFormatStringExpr(Arg, Args, NumArgs, 1758 HasVAListArg, format_idx, firstDataArg, 1759 Type, CallType, inFunctionCall); 1760 } else if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(ND)) { 1761 unsigned BuiltinID = FD->getBuiltinID(); 1762 if (BuiltinID == Builtin::BI__builtin___CFStringMakeConstantString || 1763 BuiltinID == Builtin::BI__builtin___NSStringMakeConstantString) { 1764 const Expr *Arg = CE->getArg(0); 1765 return checkFormatStringExpr(Arg, Args, NumArgs, 1766 HasVAListArg, format_idx, 1767 firstDataArg, Type, CallType, 1768 inFunctionCall); 1769 } 1770 } 1771 } 1772 1773 return SLCT_NotALiteral; 1774 } 1775 case Stmt::ObjCStringLiteralClass: 1776 case Stmt::StringLiteralClass: { 1777 const StringLiteral *StrE = NULL; 1778 1779 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1780 StrE = ObjCFExpr->getString(); 1781 else 1782 StrE = cast<StringLiteral>(E); 1783 1784 if (StrE) { 1785 CheckFormatString(StrE, E, Args, NumArgs, HasVAListArg, format_idx, 1786 firstDataArg, Type, inFunctionCall, CallType); 1787 return SLCT_CheckedLiteral; 1788 } 1789 1790 return SLCT_NotALiteral; 1791 } 1792 1793 default: 1794 return SLCT_NotALiteral; 1795 } 1796} 1797 1798void 1799Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1800 const Expr * const *ExprArgs, 1801 SourceLocation CallSiteLoc) { 1802 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1803 e = NonNull->args_end(); 1804 i != e; ++i) { 1805 const Expr *ArgExpr = ExprArgs[*i]; 1806 if (ArgExpr->isNullPointerConstant(Context, 1807 Expr::NPC_ValueDependentIsNotNull)) 1808 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 1809 } 1810} 1811 1812Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) { 1813 return llvm::StringSwitch<FormatStringType>(Format->getType()) 1814 .Case("scanf", FST_Scanf) 1815 .Cases("printf", "printf0", FST_Printf) 1816 .Cases("NSString", "CFString", FST_NSString) 1817 .Case("strftime", FST_Strftime) 1818 .Case("strfmon", FST_Strfmon) 1819 .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf) 1820 .Default(FST_Unknown); 1821} 1822 1823/// CheckFormatArguments - Check calls to printf and scanf (and similar 1824/// functions) for correct use of format strings. 1825/// Returns true if a format string has been fully checked. 1826bool Sema::CheckFormatArguments(const FormatAttr *Format, Expr **Args, 1827 unsigned NumArgs, bool IsCXXMember, 1828 VariadicCallType CallType, 1829 SourceLocation Loc, SourceRange Range) { 1830 FormatStringInfo FSI; 1831 if (getFormatStringInfo(Format, IsCXXMember, &FSI)) 1832 return CheckFormatArguments(Args, NumArgs, FSI.HasVAListArg, FSI.FormatIdx, 1833 FSI.FirstDataArg, GetFormatStringType(Format), 1834 CallType, Loc, Range); 1835 return false; 1836} 1837 1838bool Sema::CheckFormatArguments(Expr **Args, unsigned NumArgs, 1839 bool HasVAListArg, unsigned format_idx, 1840 unsigned firstDataArg, FormatStringType Type, 1841 VariadicCallType CallType, 1842 SourceLocation Loc, SourceRange Range) { 1843 // CHECK: printf/scanf-like function is called with no format string. 1844 if (format_idx >= NumArgs) { 1845 Diag(Loc, diag::warn_missing_format_string) << Range; 1846 return false; 1847 } 1848 1849 const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts(); 1850 1851 // CHECK: format string is not a string literal. 1852 // 1853 // Dynamically generated format strings are difficult to 1854 // automatically vet at compile time. Requiring that format strings 1855 // are string literals: (1) permits the checking of format strings by 1856 // the compiler and thereby (2) can practically remove the source of 1857 // many format string exploits. 1858 1859 // Format string can be either ObjC string (e.g. @"%d") or 1860 // C string (e.g. "%d") 1861 // ObjC string uses the same format specifiers as C string, so we can use 1862 // the same format string checking logic for both ObjC and C strings. 1863 StringLiteralCheckType CT = 1864 checkFormatStringExpr(OrigFormatExpr, Args, NumArgs, HasVAListArg, 1865 format_idx, firstDataArg, Type, CallType); 1866 if (CT != SLCT_NotALiteral) 1867 // Literal format string found, check done! 1868 return CT == SLCT_CheckedLiteral; 1869 1870 // Strftime is particular as it always uses a single 'time' argument, 1871 // so it is safe to pass a non-literal string. 1872 if (Type == FST_Strftime) 1873 return false; 1874 1875 // Do not emit diag when the string param is a macro expansion and the 1876 // format is either NSString or CFString. This is a hack to prevent 1877 // diag when using the NSLocalizedString and CFCopyLocalizedString macros 1878 // which are usually used in place of NS and CF string literals. 1879 if (Type == FST_NSString && 1880 SourceMgr.isInSystemMacro(Args[format_idx]->getLocStart())) 1881 return false; 1882 1883 // If there are no arguments specified, warn with -Wformat-security, otherwise 1884 // warn only with -Wformat-nonliteral. 1885 if (NumArgs == format_idx+1) 1886 Diag(Args[format_idx]->getLocStart(), 1887 diag::warn_format_nonliteral_noargs) 1888 << OrigFormatExpr->getSourceRange(); 1889 else 1890 Diag(Args[format_idx]->getLocStart(), 1891 diag::warn_format_nonliteral) 1892 << OrigFormatExpr->getSourceRange(); 1893 return false; 1894} 1895 1896namespace { 1897class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1898protected: 1899 Sema &S; 1900 const StringLiteral *FExpr; 1901 const Expr *OrigFormatExpr; 1902 const unsigned FirstDataArg; 1903 const unsigned NumDataArgs; 1904 const char *Beg; // Start of format string. 1905 const bool HasVAListArg; 1906 const Expr * const *Args; 1907 const unsigned NumArgs; 1908 unsigned FormatIdx; 1909 llvm::BitVector CoveredArgs; 1910 bool usesPositionalArgs; 1911 bool atFirstArg; 1912 bool inFunctionCall; 1913 Sema::VariadicCallType CallType; 1914public: 1915 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1916 const Expr *origFormatExpr, unsigned firstDataArg, 1917 unsigned numDataArgs, const char *beg, bool hasVAListArg, 1918 Expr **args, unsigned numArgs, 1919 unsigned formatIdx, bool inFunctionCall, 1920 Sema::VariadicCallType callType) 1921 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1922 FirstDataArg(firstDataArg), NumDataArgs(numDataArgs), 1923 Beg(beg), HasVAListArg(hasVAListArg), 1924 Args(args), NumArgs(numArgs), FormatIdx(formatIdx), 1925 usesPositionalArgs(false), atFirstArg(true), 1926 inFunctionCall(inFunctionCall), CallType(callType) { 1927 CoveredArgs.resize(numDataArgs); 1928 CoveredArgs.reset(); 1929 } 1930 1931 void DoneProcessing(); 1932 1933 void HandleIncompleteSpecifier(const char *startSpecifier, 1934 unsigned specifierLen); 1935 1936 void HandleNonStandardLengthModifier( 1937 const analyze_format_string::LengthModifier &LM, 1938 const char *startSpecifier, unsigned specifierLen); 1939 1940 void HandleNonStandardConversionSpecifier( 1941 const analyze_format_string::ConversionSpecifier &CS, 1942 const char *startSpecifier, unsigned specifierLen); 1943 1944 void HandleNonStandardConversionSpecification( 1945 const analyze_format_string::LengthModifier &LM, 1946 const analyze_format_string::ConversionSpecifier &CS, 1947 const char *startSpecifier, unsigned specifierLen); 1948 1949 virtual void HandlePosition(const char *startPos, unsigned posLen); 1950 1951 virtual void HandleInvalidPosition(const char *startSpecifier, 1952 unsigned specifierLen, 1953 analyze_format_string::PositionContext p); 1954 1955 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1956 1957 void HandleNullChar(const char *nullCharacter); 1958 1959 template <typename Range> 1960 static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall, 1961 const Expr *ArgumentExpr, 1962 PartialDiagnostic PDiag, 1963 SourceLocation StringLoc, 1964 bool IsStringLocation, Range StringRange, 1965 FixItHint Fixit = FixItHint()); 1966 1967protected: 1968 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1969 const char *startSpec, 1970 unsigned specifierLen, 1971 const char *csStart, unsigned csLen); 1972 1973 void HandlePositionalNonpositionalArgs(SourceLocation Loc, 1974 const char *startSpec, 1975 unsigned specifierLen); 1976 1977 SourceRange getFormatStringRange(); 1978 CharSourceRange getSpecifierRange(const char *startSpecifier, 1979 unsigned specifierLen); 1980 SourceLocation getLocationOfByte(const char *x); 1981 1982 const Expr *getDataArg(unsigned i) const; 1983 1984 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1985 const analyze_format_string::ConversionSpecifier &CS, 1986 const char *startSpecifier, unsigned specifierLen, 1987 unsigned argIndex); 1988 1989 template <typename Range> 1990 void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc, 1991 bool IsStringLocation, Range StringRange, 1992 FixItHint Fixit = FixItHint()); 1993 1994 void CheckPositionalAndNonpositionalArgs( 1995 const analyze_format_string::FormatSpecifier *FS); 1996}; 1997} 1998 1999SourceRange CheckFormatHandler::getFormatStringRange() { 2000 return OrigFormatExpr->getSourceRange(); 2001} 2002 2003CharSourceRange CheckFormatHandler:: 2004getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 2005 SourceLocation Start = getLocationOfByte(startSpecifier); 2006 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 2007 2008 // Advance the end SourceLocation by one due to half-open ranges. 2009 End = End.getLocWithOffset(1); 2010 2011 return CharSourceRange::getCharRange(Start, End); 2012} 2013 2014SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 2015 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 2016} 2017 2018void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 2019 unsigned specifierLen){ 2020 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier), 2021 getLocationOfByte(startSpecifier), 2022 /*IsStringLocation*/true, 2023 getSpecifierRange(startSpecifier, specifierLen)); 2024} 2025 2026void CheckFormatHandler::HandleNonStandardLengthModifier( 2027 const analyze_format_string::LengthModifier &LM, 2028 const char *startSpecifier, unsigned specifierLen) { 2029 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) << LM.toString() 2030 << 0, 2031 getLocationOfByte(LM.getStart()), 2032 /*IsStringLocation*/true, 2033 getSpecifierRange(startSpecifier, specifierLen)); 2034} 2035 2036void CheckFormatHandler::HandleNonStandardConversionSpecifier( 2037 const analyze_format_string::ConversionSpecifier &CS, 2038 const char *startSpecifier, unsigned specifierLen) { 2039 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) << CS.toString() 2040 << 1, 2041 getLocationOfByte(CS.getStart()), 2042 /*IsStringLocation*/true, 2043 getSpecifierRange(startSpecifier, specifierLen)); 2044} 2045 2046void CheckFormatHandler::HandleNonStandardConversionSpecification( 2047 const analyze_format_string::LengthModifier &LM, 2048 const analyze_format_string::ConversionSpecifier &CS, 2049 const char *startSpecifier, unsigned specifierLen) { 2050 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_conversion_spec) 2051 << LM.toString() << CS.toString(), 2052 getLocationOfByte(LM.getStart()), 2053 /*IsStringLocation*/true, 2054 getSpecifierRange(startSpecifier, specifierLen)); 2055} 2056 2057void CheckFormatHandler::HandlePosition(const char *startPos, 2058 unsigned posLen) { 2059 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_positional_arg), 2060 getLocationOfByte(startPos), 2061 /*IsStringLocation*/true, 2062 getSpecifierRange(startPos, posLen)); 2063} 2064 2065void 2066CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 2067 analyze_format_string::PositionContext p) { 2068 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier) 2069 << (unsigned) p, 2070 getLocationOfByte(startPos), /*IsStringLocation*/true, 2071 getSpecifierRange(startPos, posLen)); 2072} 2073 2074void CheckFormatHandler::HandleZeroPosition(const char *startPos, 2075 unsigned posLen) { 2076 EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier), 2077 getLocationOfByte(startPos), 2078 /*IsStringLocation*/true, 2079 getSpecifierRange(startPos, posLen)); 2080} 2081 2082void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 2083 if (!isa<ObjCStringLiteral>(OrigFormatExpr)) { 2084 // The presence of a null character is likely an error. 2085 EmitFormatDiagnostic( 2086 S.PDiag(diag::warn_printf_format_string_contains_null_char), 2087 getLocationOfByte(nullCharacter), /*IsStringLocation*/true, 2088 getFormatStringRange()); 2089 } 2090} 2091 2092// Note that this may return NULL if there was an error parsing or building 2093// one of the argument expressions. 2094const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 2095 return Args[FirstDataArg + i]; 2096} 2097 2098void CheckFormatHandler::DoneProcessing() { 2099 // Does the number of data arguments exceed the number of 2100 // format conversions in the format string? 2101 if (!HasVAListArg) { 2102 // Find any arguments that weren't covered. 2103 CoveredArgs.flip(); 2104 signed notCoveredArg = CoveredArgs.find_first(); 2105 if (notCoveredArg >= 0) { 2106 assert((unsigned)notCoveredArg < NumDataArgs); 2107 if (const Expr *E = getDataArg((unsigned) notCoveredArg)) { 2108 SourceLocation Loc = E->getLocStart(); 2109 if (!S.getSourceManager().isInSystemMacro(Loc)) { 2110 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used), 2111 Loc, /*IsStringLocation*/false, 2112 getFormatStringRange()); 2113 } 2114 } 2115 } 2116 } 2117} 2118 2119bool 2120CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 2121 SourceLocation Loc, 2122 const char *startSpec, 2123 unsigned specifierLen, 2124 const char *csStart, 2125 unsigned csLen) { 2126 2127 bool keepGoing = true; 2128 if (argIndex < NumDataArgs) { 2129 // Consider the argument coverered, even though the specifier doesn't 2130 // make sense. 2131 CoveredArgs.set(argIndex); 2132 } 2133 else { 2134 // If argIndex exceeds the number of data arguments we 2135 // don't issue a warning because that is just a cascade of warnings (and 2136 // they may have intended '%%' anyway). We don't want to continue processing 2137 // the format string after this point, however, as we will like just get 2138 // gibberish when trying to match arguments. 2139 keepGoing = false; 2140 } 2141 2142 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion) 2143 << StringRef(csStart, csLen), 2144 Loc, /*IsStringLocation*/true, 2145 getSpecifierRange(startSpec, specifierLen)); 2146 2147 return keepGoing; 2148} 2149 2150void 2151CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc, 2152 const char *startSpec, 2153 unsigned specifierLen) { 2154 EmitFormatDiagnostic( 2155 S.PDiag(diag::warn_format_mix_positional_nonpositional_args), 2156 Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen)); 2157} 2158 2159bool 2160CheckFormatHandler::CheckNumArgs( 2161 const analyze_format_string::FormatSpecifier &FS, 2162 const analyze_format_string::ConversionSpecifier &CS, 2163 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 2164 2165 if (argIndex >= NumDataArgs) { 2166 PartialDiagnostic PDiag = FS.usesPositionalArg() 2167 ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args) 2168 << (argIndex+1) << NumDataArgs) 2169 : S.PDiag(diag::warn_printf_insufficient_data_args); 2170 EmitFormatDiagnostic( 2171 PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true, 2172 getSpecifierRange(startSpecifier, specifierLen)); 2173 return false; 2174 } 2175 return true; 2176} 2177 2178template<typename Range> 2179void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag, 2180 SourceLocation Loc, 2181 bool IsStringLocation, 2182 Range StringRange, 2183 FixItHint FixIt) { 2184 EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag, 2185 Loc, IsStringLocation, StringRange, FixIt); 2186} 2187 2188/// \brief If the format string is not within the funcion call, emit a note 2189/// so that the function call and string are in diagnostic messages. 2190/// 2191/// \param InFunctionCall if true, the format string is within the function 2192/// call and only one diagnostic message will be produced. Otherwise, an 2193/// extra note will be emitted pointing to location of the format string. 2194/// 2195/// \param ArgumentExpr the expression that is passed as the format string 2196/// argument in the function call. Used for getting locations when two 2197/// diagnostics are emitted. 2198/// 2199/// \param PDiag the callee should already have provided any strings for the 2200/// diagnostic message. This function only adds locations and fixits 2201/// to diagnostics. 2202/// 2203/// \param Loc primary location for diagnostic. If two diagnostics are 2204/// required, one will be at Loc and a new SourceLocation will be created for 2205/// the other one. 2206/// 2207/// \param IsStringLocation if true, Loc points to the format string should be 2208/// used for the note. Otherwise, Loc points to the argument list and will 2209/// be used with PDiag. 2210/// 2211/// \param StringRange some or all of the string to highlight. This is 2212/// templated so it can accept either a CharSourceRange or a SourceRange. 2213/// 2214/// \param FixIt optional fix it hint for the format string. 2215template<typename Range> 2216void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall, 2217 const Expr *ArgumentExpr, 2218 PartialDiagnostic PDiag, 2219 SourceLocation Loc, 2220 bool IsStringLocation, 2221 Range StringRange, 2222 FixItHint FixIt) { 2223 if (InFunctionCall) 2224 S.Diag(Loc, PDiag) << StringRange << FixIt; 2225 else { 2226 S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag) 2227 << ArgumentExpr->getSourceRange(); 2228 S.Diag(IsStringLocation ? Loc : StringRange.getBegin(), 2229 diag::note_format_string_defined) 2230 << StringRange << FixIt; 2231 } 2232} 2233 2234//===--- CHECK: Printf format string checking ------------------------------===// 2235 2236namespace { 2237class CheckPrintfHandler : public CheckFormatHandler { 2238 bool ObjCContext; 2239public: 2240 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 2241 const Expr *origFormatExpr, unsigned firstDataArg, 2242 unsigned numDataArgs, bool isObjC, 2243 const char *beg, bool hasVAListArg, 2244 Expr **Args, unsigned NumArgs, 2245 unsigned formatIdx, bool inFunctionCall, 2246 Sema::VariadicCallType CallType) 2247 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2248 numDataArgs, beg, hasVAListArg, Args, NumArgs, 2249 formatIdx, inFunctionCall, CallType), ObjCContext(isObjC) 2250 {} 2251 2252 2253 bool HandleInvalidPrintfConversionSpecifier( 2254 const analyze_printf::PrintfSpecifier &FS, 2255 const char *startSpecifier, 2256 unsigned specifierLen); 2257 2258 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 2259 const char *startSpecifier, 2260 unsigned specifierLen); 2261 bool checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 2262 const char *StartSpecifier, 2263 unsigned SpecifierLen, 2264 const Expr *E); 2265 2266 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 2267 const char *startSpecifier, unsigned specifierLen); 2268 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 2269 const analyze_printf::OptionalAmount &Amt, 2270 unsigned type, 2271 const char *startSpecifier, unsigned specifierLen); 2272 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2273 const analyze_printf::OptionalFlag &flag, 2274 const char *startSpecifier, unsigned specifierLen); 2275 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 2276 const analyze_printf::OptionalFlag &ignoredFlag, 2277 const analyze_printf::OptionalFlag &flag, 2278 const char *startSpecifier, unsigned specifierLen); 2279 bool checkForCStrMembers(const analyze_printf::ArgType &AT, 2280 const Expr *E, const CharSourceRange &CSR); 2281 2282}; 2283} 2284 2285bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 2286 const analyze_printf::PrintfSpecifier &FS, 2287 const char *startSpecifier, 2288 unsigned specifierLen) { 2289 const analyze_printf::PrintfConversionSpecifier &CS = 2290 FS.getConversionSpecifier(); 2291 2292 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2293 getLocationOfByte(CS.getStart()), 2294 startSpecifier, specifierLen, 2295 CS.getStart(), CS.getLength()); 2296} 2297 2298bool CheckPrintfHandler::HandleAmount( 2299 const analyze_format_string::OptionalAmount &Amt, 2300 unsigned k, const char *startSpecifier, 2301 unsigned specifierLen) { 2302 2303 if (Amt.hasDataArgument()) { 2304 if (!HasVAListArg) { 2305 unsigned argIndex = Amt.getArgIndex(); 2306 if (argIndex >= NumDataArgs) { 2307 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg) 2308 << k, 2309 getLocationOfByte(Amt.getStart()), 2310 /*IsStringLocation*/true, 2311 getSpecifierRange(startSpecifier, specifierLen)); 2312 // Don't do any more checking. We will just emit 2313 // spurious errors. 2314 return false; 2315 } 2316 2317 // Type check the data argument. It should be an 'int'. 2318 // Although not in conformance with C99, we also allow the argument to be 2319 // an 'unsigned int' as that is a reasonably safe case. GCC also 2320 // doesn't emit a warning for that case. 2321 CoveredArgs.set(argIndex); 2322 const Expr *Arg = getDataArg(argIndex); 2323 if (!Arg) 2324 return false; 2325 2326 QualType T = Arg->getType(); 2327 2328 const analyze_printf::ArgType &AT = Amt.getArgType(S.Context); 2329 assert(AT.isValid()); 2330 2331 if (!AT.matchesType(S.Context, T)) { 2332 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type) 2333 << k << AT.getRepresentativeTypeName(S.Context) 2334 << T << Arg->getSourceRange(), 2335 getLocationOfByte(Amt.getStart()), 2336 /*IsStringLocation*/true, 2337 getSpecifierRange(startSpecifier, specifierLen)); 2338 // Don't do any more checking. We will just emit 2339 // spurious errors. 2340 return false; 2341 } 2342 } 2343 } 2344 return true; 2345} 2346 2347void CheckPrintfHandler::HandleInvalidAmount( 2348 const analyze_printf::PrintfSpecifier &FS, 2349 const analyze_printf::OptionalAmount &Amt, 2350 unsigned type, 2351 const char *startSpecifier, 2352 unsigned specifierLen) { 2353 const analyze_printf::PrintfConversionSpecifier &CS = 2354 FS.getConversionSpecifier(); 2355 2356 FixItHint fixit = 2357 Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant 2358 ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 2359 Amt.getConstantLength())) 2360 : FixItHint(); 2361 2362 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount) 2363 << type << CS.toString(), 2364 getLocationOfByte(Amt.getStart()), 2365 /*IsStringLocation*/true, 2366 getSpecifierRange(startSpecifier, specifierLen), 2367 fixit); 2368} 2369 2370void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2371 const analyze_printf::OptionalFlag &flag, 2372 const char *startSpecifier, 2373 unsigned specifierLen) { 2374 // Warn about pointless flag with a fixit removal. 2375 const analyze_printf::PrintfConversionSpecifier &CS = 2376 FS.getConversionSpecifier(); 2377 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag) 2378 << flag.toString() << CS.toString(), 2379 getLocationOfByte(flag.getPosition()), 2380 /*IsStringLocation*/true, 2381 getSpecifierRange(startSpecifier, specifierLen), 2382 FixItHint::CreateRemoval( 2383 getSpecifierRange(flag.getPosition(), 1))); 2384} 2385 2386void CheckPrintfHandler::HandleIgnoredFlag( 2387 const analyze_printf::PrintfSpecifier &FS, 2388 const analyze_printf::OptionalFlag &ignoredFlag, 2389 const analyze_printf::OptionalFlag &flag, 2390 const char *startSpecifier, 2391 unsigned specifierLen) { 2392 // Warn about ignored flag with a fixit removal. 2393 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag) 2394 << ignoredFlag.toString() << flag.toString(), 2395 getLocationOfByte(ignoredFlag.getPosition()), 2396 /*IsStringLocation*/true, 2397 getSpecifierRange(startSpecifier, specifierLen), 2398 FixItHint::CreateRemoval( 2399 getSpecifierRange(ignoredFlag.getPosition(), 1))); 2400} 2401 2402// Determines if the specified is a C++ class or struct containing 2403// a member with the specified name and kind (e.g. a CXXMethodDecl named 2404// "c_str()"). 2405template<typename MemberKind> 2406static llvm::SmallPtrSet<MemberKind*, 1> 2407CXXRecordMembersNamed(StringRef Name, Sema &S, QualType Ty) { 2408 const RecordType *RT = Ty->getAs<RecordType>(); 2409 llvm::SmallPtrSet<MemberKind*, 1> Results; 2410 2411 if (!RT) 2412 return Results; 2413 const CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(RT->getDecl()); 2414 if (!RD) 2415 return Results; 2416 2417 LookupResult R(S, &S.PP.getIdentifierTable().get(Name), SourceLocation(), 2418 Sema::LookupMemberName); 2419 2420 // We just need to include all members of the right kind turned up by the 2421 // filter, at this point. 2422 if (S.LookupQualifiedName(R, RT->getDecl())) 2423 for (LookupResult::iterator I = R.begin(), E = R.end(); I != E; ++I) { 2424 NamedDecl *decl = (*I)->getUnderlyingDecl(); 2425 if (MemberKind *FK = dyn_cast<MemberKind>(decl)) 2426 Results.insert(FK); 2427 } 2428 return Results; 2429} 2430 2431// Check if a (w)string was passed when a (w)char* was needed, and offer a 2432// better diagnostic if so. AT is assumed to be valid. 2433// Returns true when a c_str() conversion method is found. 2434bool CheckPrintfHandler::checkForCStrMembers( 2435 const analyze_printf::ArgType &AT, const Expr *E, 2436 const CharSourceRange &CSR) { 2437 typedef llvm::SmallPtrSet<CXXMethodDecl*, 1> MethodSet; 2438 2439 MethodSet Results = 2440 CXXRecordMembersNamed<CXXMethodDecl>("c_str", S, E->getType()); 2441 2442 for (MethodSet::iterator MI = Results.begin(), ME = Results.end(); 2443 MI != ME; ++MI) { 2444 const CXXMethodDecl *Method = *MI; 2445 if (Method->getNumParams() == 0 && 2446 AT.matchesType(S.Context, Method->getResultType())) { 2447 // FIXME: Suggest parens if the expression needs them. 2448 SourceLocation EndLoc = 2449 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()); 2450 S.Diag(E->getLocStart(), diag::note_printf_c_str) 2451 << "c_str()" 2452 << FixItHint::CreateInsertion(EndLoc, ".c_str()"); 2453 return true; 2454 } 2455 } 2456 2457 return false; 2458} 2459 2460bool 2461CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 2462 &FS, 2463 const char *startSpecifier, 2464 unsigned specifierLen) { 2465 2466 using namespace analyze_format_string; 2467 using namespace analyze_printf; 2468 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 2469 2470 if (FS.consumesDataArgument()) { 2471 if (atFirstArg) { 2472 atFirstArg = false; 2473 usesPositionalArgs = FS.usesPositionalArg(); 2474 } 2475 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2476 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2477 startSpecifier, specifierLen); 2478 return false; 2479 } 2480 } 2481 2482 // First check if the field width, precision, and conversion specifier 2483 // have matching data arguments. 2484 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 2485 startSpecifier, specifierLen)) { 2486 return false; 2487 } 2488 2489 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 2490 startSpecifier, specifierLen)) { 2491 return false; 2492 } 2493 2494 if (!CS.consumesDataArgument()) { 2495 // FIXME: Technically specifying a precision or field width here 2496 // makes no sense. Worth issuing a warning at some point. 2497 return true; 2498 } 2499 2500 // Consume the argument. 2501 unsigned argIndex = FS.getArgIndex(); 2502 if (argIndex < NumDataArgs) { 2503 // The check to see if the argIndex is valid will come later. 2504 // We set the bit here because we may exit early from this 2505 // function if we encounter some other error. 2506 CoveredArgs.set(argIndex); 2507 } 2508 2509 // Check for using an Objective-C specific conversion specifier 2510 // in a non-ObjC literal. 2511 if (!ObjCContext && CS.isObjCArg()) { 2512 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 2513 specifierLen); 2514 } 2515 2516 // Check for invalid use of field width 2517 if (!FS.hasValidFieldWidth()) { 2518 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 2519 startSpecifier, specifierLen); 2520 } 2521 2522 // Check for invalid use of precision 2523 if (!FS.hasValidPrecision()) { 2524 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 2525 startSpecifier, specifierLen); 2526 } 2527 2528 // Check each flag does not conflict with any other component. 2529 if (!FS.hasValidThousandsGroupingPrefix()) 2530 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 2531 if (!FS.hasValidLeadingZeros()) 2532 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 2533 if (!FS.hasValidPlusPrefix()) 2534 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 2535 if (!FS.hasValidSpacePrefix()) 2536 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 2537 if (!FS.hasValidAlternativeForm()) 2538 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 2539 if (!FS.hasValidLeftJustified()) 2540 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 2541 2542 // Check that flags are not ignored by another flag 2543 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 2544 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 2545 startSpecifier, specifierLen); 2546 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 2547 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 2548 startSpecifier, specifierLen); 2549 2550 // Check the length modifier is valid with the given conversion specifier. 2551 const LengthModifier &LM = FS.getLengthModifier(); 2552 if (!FS.hasValidLengthModifier()) 2553 EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length) 2554 << LM.toString() << CS.toString(), 2555 getLocationOfByte(LM.getStart()), 2556 /*IsStringLocation*/true, 2557 getSpecifierRange(startSpecifier, specifierLen), 2558 FixItHint::CreateRemoval( 2559 getSpecifierRange(LM.getStart(), 2560 LM.getLength()))); 2561 if (!FS.hasStandardLengthModifier()) 2562 HandleNonStandardLengthModifier(LM, startSpecifier, specifierLen); 2563 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 2564 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 2565 if (!FS.hasStandardLengthConversionCombination()) 2566 HandleNonStandardConversionSpecification(LM, CS, startSpecifier, 2567 specifierLen); 2568 2569 // The remaining checks depend on the data arguments. 2570 if (HasVAListArg) 2571 return true; 2572 2573 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2574 return false; 2575 2576 const Expr *Arg = getDataArg(argIndex); 2577 if (!Arg) 2578 return true; 2579 2580 return checkFormatExpr(FS, startSpecifier, specifierLen, Arg); 2581} 2582 2583bool 2584CheckPrintfHandler::checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 2585 const char *StartSpecifier, 2586 unsigned SpecifierLen, 2587 const Expr *E) { 2588 using namespace analyze_format_string; 2589 using namespace analyze_printf; 2590 // Now type check the data expression that matches the 2591 // format specifier. 2592 const analyze_printf::ArgType &AT = FS.getArgType(S.Context, 2593 ObjCContext); 2594 if (AT.isValid() && !AT.matchesType(S.Context, E->getType())) { 2595 // Look through argument promotions for our error message's reported type. 2596 // This includes the integral and floating promotions, but excludes array 2597 // and function pointer decay; seeing that an argument intended to be a 2598 // string has type 'char [6]' is probably more confusing than 'char *'. 2599 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 2600 if (ICE->getCastKind() == CK_IntegralCast || 2601 ICE->getCastKind() == CK_FloatingCast) { 2602 E = ICE->getSubExpr(); 2603 2604 // Check if we didn't match because of an implicit cast from a 'char' 2605 // or 'short' to an 'int'. This is done because printf is a varargs 2606 // function. 2607 if (ICE->getType() == S.Context.IntTy || 2608 ICE->getType() == S.Context.UnsignedIntTy) { 2609 // All further checking is done on the subexpression. 2610 if (AT.matchesType(S.Context, E->getType())) 2611 return true; 2612 } 2613 } 2614 } 2615 2616 // We may be able to offer a FixItHint if it is a supported type. 2617 PrintfSpecifier fixedFS = FS; 2618 bool success = fixedFS.fixType(E->getType(), S.getLangOpts(), 2619 S.Context, ObjCContext); 2620 2621 if (success) { 2622 // Get the fix string from the fixed format specifier 2623 SmallString<16> buf; 2624 llvm::raw_svector_ostream os(buf); 2625 fixedFS.toString(os); 2626 2627 EmitFormatDiagnostic( 2628 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2629 << AT.getRepresentativeTypeName(S.Context) << E->getType() 2630 << E->getSourceRange(), 2631 E->getLocStart(), 2632 /*IsStringLocation*/false, 2633 getSpecifierRange(StartSpecifier, SpecifierLen), 2634 FixItHint::CreateReplacement( 2635 getSpecifierRange(StartSpecifier, SpecifierLen), 2636 os.str())); 2637 } else { 2638 const CharSourceRange &CSR = getSpecifierRange(StartSpecifier, 2639 SpecifierLen); 2640 // Since the warning for passing non-POD types to variadic functions 2641 // was deferred until now, we emit a warning for non-POD 2642 // arguments here. 2643 if (S.isValidVarArgType(E->getType()) == Sema::VAK_Invalid) { 2644 unsigned DiagKind; 2645 if (E->getType()->isObjCObjectType()) 2646 DiagKind = diag::err_cannot_pass_objc_interface_to_vararg_format; 2647 else 2648 DiagKind = diag::warn_non_pod_vararg_with_format_string; 2649 2650 EmitFormatDiagnostic( 2651 S.PDiag(DiagKind) 2652 << S.getLangOpts().CPlusPlus0x 2653 << E->getType() 2654 << CallType 2655 << AT.getRepresentativeTypeName(S.Context) 2656 << CSR 2657 << E->getSourceRange(), 2658 E->getLocStart(), /*IsStringLocation*/false, CSR); 2659 2660 checkForCStrMembers(AT, E, CSR); 2661 } else 2662 EmitFormatDiagnostic( 2663 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2664 << AT.getRepresentativeTypeName(S.Context) << E->getType() 2665 << CSR 2666 << E->getSourceRange(), 2667 E->getLocStart(), /*IsStringLocation*/false, CSR); 2668 } 2669 } 2670 2671 return true; 2672} 2673 2674//===--- CHECK: Scanf format string checking ------------------------------===// 2675 2676namespace { 2677class CheckScanfHandler : public CheckFormatHandler { 2678public: 2679 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 2680 const Expr *origFormatExpr, unsigned firstDataArg, 2681 unsigned numDataArgs, const char *beg, bool hasVAListArg, 2682 Expr **Args, unsigned NumArgs, 2683 unsigned formatIdx, bool inFunctionCall, 2684 Sema::VariadicCallType CallType) 2685 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2686 numDataArgs, beg, hasVAListArg, 2687 Args, NumArgs, formatIdx, inFunctionCall, CallType) 2688 {} 2689 2690 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 2691 const char *startSpecifier, 2692 unsigned specifierLen); 2693 2694 bool HandleInvalidScanfConversionSpecifier( 2695 const analyze_scanf::ScanfSpecifier &FS, 2696 const char *startSpecifier, 2697 unsigned specifierLen); 2698 2699 void HandleIncompleteScanList(const char *start, const char *end); 2700}; 2701} 2702 2703void CheckScanfHandler::HandleIncompleteScanList(const char *start, 2704 const char *end) { 2705 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 2706 getLocationOfByte(end), /*IsStringLocation*/true, 2707 getSpecifierRange(start, end - start)); 2708} 2709 2710bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 2711 const analyze_scanf::ScanfSpecifier &FS, 2712 const char *startSpecifier, 2713 unsigned specifierLen) { 2714 2715 const analyze_scanf::ScanfConversionSpecifier &CS = 2716 FS.getConversionSpecifier(); 2717 2718 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2719 getLocationOfByte(CS.getStart()), 2720 startSpecifier, specifierLen, 2721 CS.getStart(), CS.getLength()); 2722} 2723 2724bool CheckScanfHandler::HandleScanfSpecifier( 2725 const analyze_scanf::ScanfSpecifier &FS, 2726 const char *startSpecifier, 2727 unsigned specifierLen) { 2728 2729 using namespace analyze_scanf; 2730 using namespace analyze_format_string; 2731 2732 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 2733 2734 // Handle case where '%' and '*' don't consume an argument. These shouldn't 2735 // be used to decide if we are using positional arguments consistently. 2736 if (FS.consumesDataArgument()) { 2737 if (atFirstArg) { 2738 atFirstArg = false; 2739 usesPositionalArgs = FS.usesPositionalArg(); 2740 } 2741 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2742 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2743 startSpecifier, specifierLen); 2744 return false; 2745 } 2746 } 2747 2748 // Check if the field with is non-zero. 2749 const OptionalAmount &Amt = FS.getFieldWidth(); 2750 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 2751 if (Amt.getConstantAmount() == 0) { 2752 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 2753 Amt.getConstantLength()); 2754 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 2755 getLocationOfByte(Amt.getStart()), 2756 /*IsStringLocation*/true, R, 2757 FixItHint::CreateRemoval(R)); 2758 } 2759 } 2760 2761 if (!FS.consumesDataArgument()) { 2762 // FIXME: Technically specifying a precision or field width here 2763 // makes no sense. Worth issuing a warning at some point. 2764 return true; 2765 } 2766 2767 // Consume the argument. 2768 unsigned argIndex = FS.getArgIndex(); 2769 if (argIndex < NumDataArgs) { 2770 // The check to see if the argIndex is valid will come later. 2771 // We set the bit here because we may exit early from this 2772 // function if we encounter some other error. 2773 CoveredArgs.set(argIndex); 2774 } 2775 2776 // Check the length modifier is valid with the given conversion specifier. 2777 const LengthModifier &LM = FS.getLengthModifier(); 2778 if (!FS.hasValidLengthModifier()) { 2779 const CharSourceRange &R = getSpecifierRange(LM.getStart(), LM.getLength()); 2780 EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length) 2781 << LM.toString() << CS.toString() 2782 << getSpecifierRange(startSpecifier, specifierLen), 2783 getLocationOfByte(LM.getStart()), 2784 /*IsStringLocation*/true, R, 2785 FixItHint::CreateRemoval(R)); 2786 } 2787 2788 if (!FS.hasStandardLengthModifier()) 2789 HandleNonStandardLengthModifier(LM, startSpecifier, specifierLen); 2790 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 2791 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 2792 if (!FS.hasStandardLengthConversionCombination()) 2793 HandleNonStandardConversionSpecification(LM, CS, startSpecifier, 2794 specifierLen); 2795 2796 // The remaining checks depend on the data arguments. 2797 if (HasVAListArg) 2798 return true; 2799 2800 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2801 return false; 2802 2803 // Check that the argument type matches the format specifier. 2804 const Expr *Ex = getDataArg(argIndex); 2805 if (!Ex) 2806 return true; 2807 2808 const analyze_format_string::ArgType &AT = FS.getArgType(S.Context); 2809 if (AT.isValid() && !AT.matchesType(S.Context, Ex->getType())) { 2810 ScanfSpecifier fixedFS = FS; 2811 bool success = fixedFS.fixType(Ex->getType(), S.getLangOpts(), 2812 S.Context); 2813 2814 if (success) { 2815 // Get the fix string from the fixed format specifier. 2816 SmallString<128> buf; 2817 llvm::raw_svector_ostream os(buf); 2818 fixedFS.toString(os); 2819 2820 EmitFormatDiagnostic( 2821 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2822 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() 2823 << Ex->getSourceRange(), 2824 Ex->getLocStart(), 2825 /*IsStringLocation*/false, 2826 getSpecifierRange(startSpecifier, specifierLen), 2827 FixItHint::CreateReplacement( 2828 getSpecifierRange(startSpecifier, specifierLen), 2829 os.str())); 2830 } else { 2831 EmitFormatDiagnostic( 2832 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2833 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() 2834 << Ex->getSourceRange(), 2835 Ex->getLocStart(), 2836 /*IsStringLocation*/false, 2837 getSpecifierRange(startSpecifier, specifierLen)); 2838 } 2839 } 2840 2841 return true; 2842} 2843 2844void Sema::CheckFormatString(const StringLiteral *FExpr, 2845 const Expr *OrigFormatExpr, 2846 Expr **Args, unsigned NumArgs, 2847 bool HasVAListArg, unsigned format_idx, 2848 unsigned firstDataArg, FormatStringType Type, 2849 bool inFunctionCall, VariadicCallType CallType) { 2850 2851 // CHECK: is the format string a wide literal? 2852 if (!FExpr->isAscii() && !FExpr->isUTF8()) { 2853 CheckFormatHandler::EmitFormatDiagnostic( 2854 *this, inFunctionCall, Args[format_idx], 2855 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 2856 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2857 return; 2858 } 2859 2860 // Str - The format string. NOTE: this is NOT null-terminated! 2861 StringRef StrRef = FExpr->getString(); 2862 const char *Str = StrRef.data(); 2863 unsigned StrLen = StrRef.size(); 2864 const unsigned numDataArgs = NumArgs - firstDataArg; 2865 2866 // CHECK: empty format string? 2867 if (StrLen == 0 && numDataArgs > 0) { 2868 CheckFormatHandler::EmitFormatDiagnostic( 2869 *this, inFunctionCall, Args[format_idx], 2870 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 2871 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2872 return; 2873 } 2874 2875 if (Type == FST_Printf || Type == FST_NSString) { 2876 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2877 numDataArgs, (Type == FST_NSString), 2878 Str, HasVAListArg, Args, NumArgs, format_idx, 2879 inFunctionCall, CallType); 2880 2881 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen, 2882 getLangOpts())) 2883 H.DoneProcessing(); 2884 } else if (Type == FST_Scanf) { 2885 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, numDataArgs, 2886 Str, HasVAListArg, Args, NumArgs, format_idx, 2887 inFunctionCall, CallType); 2888 2889 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen, 2890 getLangOpts())) 2891 H.DoneProcessing(); 2892 } // TODO: handle other formats 2893} 2894 2895//===--- CHECK: Standard memory functions ---------------------------------===// 2896 2897/// \brief Determine whether the given type is a dynamic class type (e.g., 2898/// whether it has a vtable). 2899static bool isDynamicClassType(QualType T) { 2900 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 2901 if (CXXRecordDecl *Definition = Record->getDefinition()) 2902 if (Definition->isDynamicClass()) 2903 return true; 2904 2905 return false; 2906} 2907 2908/// \brief If E is a sizeof expression, returns its argument expression, 2909/// otherwise returns NULL. 2910static const Expr *getSizeOfExprArg(const Expr* E) { 2911 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2912 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2913 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 2914 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 2915 2916 return 0; 2917} 2918 2919/// \brief If E is a sizeof expression, returns its argument type. 2920static QualType getSizeOfArgType(const Expr* E) { 2921 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2922 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2923 if (SizeOf->getKind() == clang::UETT_SizeOf) 2924 return SizeOf->getTypeOfArgument(); 2925 2926 return QualType(); 2927} 2928 2929/// \brief Check for dangerous or invalid arguments to memset(). 2930/// 2931/// This issues warnings on known problematic, dangerous or unspecified 2932/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 2933/// function calls. 2934/// 2935/// \param Call The call expression to diagnose. 2936void Sema::CheckMemaccessArguments(const CallExpr *Call, 2937 unsigned BId, 2938 IdentifierInfo *FnName) { 2939 assert(BId != 0); 2940 2941 // It is possible to have a non-standard definition of memset. Validate 2942 // we have enough arguments, and if not, abort further checking. 2943 unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3); 2944 if (Call->getNumArgs() < ExpectedNumArgs) 2945 return; 2946 2947 unsigned LastArg = (BId == Builtin::BImemset || 2948 BId == Builtin::BIstrndup ? 1 : 2); 2949 unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2); 2950 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 2951 2952 // We have special checking when the length is a sizeof expression. 2953 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 2954 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 2955 llvm::FoldingSetNodeID SizeOfArgID; 2956 2957 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 2958 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 2959 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 2960 2961 QualType DestTy = Dest->getType(); 2962 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 2963 QualType PointeeTy = DestPtrTy->getPointeeType(); 2964 2965 // Never warn about void type pointers. This can be used to suppress 2966 // false positives. 2967 if (PointeeTy->isVoidType()) 2968 continue; 2969 2970 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 2971 // actually comparing the expressions for equality. Because computing the 2972 // expression IDs can be expensive, we only do this if the diagnostic is 2973 // enabled. 2974 if (SizeOfArg && 2975 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 2976 SizeOfArg->getExprLoc())) { 2977 // We only compute IDs for expressions if the warning is enabled, and 2978 // cache the sizeof arg's ID. 2979 if (SizeOfArgID == llvm::FoldingSetNodeID()) 2980 SizeOfArg->Profile(SizeOfArgID, Context, true); 2981 llvm::FoldingSetNodeID DestID; 2982 Dest->Profile(DestID, Context, true); 2983 if (DestID == SizeOfArgID) { 2984 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 2985 // over sizeof(src) as well. 2986 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 2987 StringRef ReadableName = FnName->getName(); 2988 2989 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 2990 if (UnaryOp->getOpcode() == UO_AddrOf) 2991 ActionIdx = 1; // If its an address-of operator, just remove it. 2992 if (Context.getTypeSize(PointeeTy) == Context.getCharWidth()) 2993 ActionIdx = 2; // If the pointee's size is sizeof(char), 2994 // suggest an explicit length. 2995 2996 // If the function is defined as a builtin macro, do not show macro 2997 // expansion. 2998 SourceLocation SL = SizeOfArg->getExprLoc(); 2999 SourceRange DSR = Dest->getSourceRange(); 3000 SourceRange SSR = SizeOfArg->getSourceRange(); 3001 SourceManager &SM = PP.getSourceManager(); 3002 3003 if (SM.isMacroArgExpansion(SL)) { 3004 ReadableName = Lexer::getImmediateMacroName(SL, SM, LangOpts); 3005 SL = SM.getSpellingLoc(SL); 3006 DSR = SourceRange(SM.getSpellingLoc(DSR.getBegin()), 3007 SM.getSpellingLoc(DSR.getEnd())); 3008 SSR = SourceRange(SM.getSpellingLoc(SSR.getBegin()), 3009 SM.getSpellingLoc(SSR.getEnd())); 3010 } 3011 3012 DiagRuntimeBehavior(SL, SizeOfArg, 3013 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 3014 << ReadableName 3015 << PointeeTy 3016 << DestTy 3017 << DSR 3018 << SSR); 3019 DiagRuntimeBehavior(SL, SizeOfArg, 3020 PDiag(diag::warn_sizeof_pointer_expr_memaccess_note) 3021 << ActionIdx 3022 << SSR); 3023 3024 break; 3025 } 3026 } 3027 3028 // Also check for cases where the sizeof argument is the exact same 3029 // type as the memory argument, and where it points to a user-defined 3030 // record type. 3031 if (SizeOfArgTy != QualType()) { 3032 if (PointeeTy->isRecordType() && 3033 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 3034 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 3035 PDiag(diag::warn_sizeof_pointer_type_memaccess) 3036 << FnName << SizeOfArgTy << ArgIdx 3037 << PointeeTy << Dest->getSourceRange() 3038 << LenExpr->getSourceRange()); 3039 break; 3040 } 3041 } 3042 3043 // Always complain about dynamic classes. 3044 if (isDynamicClassType(PointeeTy)) { 3045 3046 unsigned OperationType = 0; 3047 // "overwritten" if we're warning about the destination for any call 3048 // but memcmp; otherwise a verb appropriate to the call. 3049 if (ArgIdx != 0 || BId == Builtin::BImemcmp) { 3050 if (BId == Builtin::BImemcpy) 3051 OperationType = 1; 3052 else if(BId == Builtin::BImemmove) 3053 OperationType = 2; 3054 else if (BId == Builtin::BImemcmp) 3055 OperationType = 3; 3056 } 3057 3058 DiagRuntimeBehavior( 3059 Dest->getExprLoc(), Dest, 3060 PDiag(diag::warn_dyn_class_memaccess) 3061 << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx) 3062 << FnName << PointeeTy 3063 << OperationType 3064 << Call->getCallee()->getSourceRange()); 3065 } else if (PointeeTy.hasNonTrivialObjCLifetime() && 3066 BId != Builtin::BImemset) 3067 DiagRuntimeBehavior( 3068 Dest->getExprLoc(), Dest, 3069 PDiag(diag::warn_arc_object_memaccess) 3070 << ArgIdx << FnName << PointeeTy 3071 << Call->getCallee()->getSourceRange()); 3072 else 3073 continue; 3074 3075 DiagRuntimeBehavior( 3076 Dest->getExprLoc(), Dest, 3077 PDiag(diag::note_bad_memaccess_silence) 3078 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 3079 break; 3080 } 3081 } 3082} 3083 3084// A little helper routine: ignore addition and subtraction of integer literals. 3085// This intentionally does not ignore all integer constant expressions because 3086// we don't want to remove sizeof(). 3087static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 3088 Ex = Ex->IgnoreParenCasts(); 3089 3090 for (;;) { 3091 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 3092 if (!BO || !BO->isAdditiveOp()) 3093 break; 3094 3095 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 3096 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 3097 3098 if (isa<IntegerLiteral>(RHS)) 3099 Ex = LHS; 3100 else if (isa<IntegerLiteral>(LHS)) 3101 Ex = RHS; 3102 else 3103 break; 3104 } 3105 3106 return Ex; 3107} 3108 3109static bool isConstantSizeArrayWithMoreThanOneElement(QualType Ty, 3110 ASTContext &Context) { 3111 // Only handle constant-sized or VLAs, but not flexible members. 3112 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(Ty)) { 3113 // Only issue the FIXIT for arrays of size > 1. 3114 if (CAT->getSize().getSExtValue() <= 1) 3115 return false; 3116 } else if (!Ty->isVariableArrayType()) { 3117 return false; 3118 } 3119 return true; 3120} 3121 3122// Warn if the user has made the 'size' argument to strlcpy or strlcat 3123// be the size of the source, instead of the destination. 3124void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 3125 IdentifierInfo *FnName) { 3126 3127 // Don't crash if the user has the wrong number of arguments 3128 if (Call->getNumArgs() != 3) 3129 return; 3130 3131 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 3132 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 3133 const Expr *CompareWithSrc = NULL; 3134 3135 // Look for 'strlcpy(dst, x, sizeof(x))' 3136 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 3137 CompareWithSrc = Ex; 3138 else { 3139 // Look for 'strlcpy(dst, x, strlen(x))' 3140 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 3141 if (SizeCall->isBuiltinCall() == Builtin::BIstrlen 3142 && SizeCall->getNumArgs() == 1) 3143 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 3144 } 3145 } 3146 3147 if (!CompareWithSrc) 3148 return; 3149 3150 // Determine if the argument to sizeof/strlen is equal to the source 3151 // argument. In principle there's all kinds of things you could do 3152 // here, for instance creating an == expression and evaluating it with 3153 // EvaluateAsBooleanCondition, but this uses a more direct technique: 3154 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 3155 if (!SrcArgDRE) 3156 return; 3157 3158 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 3159 if (!CompareWithSrcDRE || 3160 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 3161 return; 3162 3163 const Expr *OriginalSizeArg = Call->getArg(2); 3164 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 3165 << OriginalSizeArg->getSourceRange() << FnName; 3166 3167 // Output a FIXIT hint if the destination is an array (rather than a 3168 // pointer to an array). This could be enhanced to handle some 3169 // pointers if we know the actual size, like if DstArg is 'array+2' 3170 // we could say 'sizeof(array)-2'. 3171 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 3172 if (!isConstantSizeArrayWithMoreThanOneElement(DstArg->getType(), Context)) 3173 return; 3174 3175 SmallString<128> sizeString; 3176 llvm::raw_svector_ostream OS(sizeString); 3177 OS << "sizeof("; 3178 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3179 OS << ")"; 3180 3181 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 3182 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 3183 OS.str()); 3184} 3185 3186/// Check if two expressions refer to the same declaration. 3187static bool referToTheSameDecl(const Expr *E1, const Expr *E2) { 3188 if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1)) 3189 if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2)) 3190 return D1->getDecl() == D2->getDecl(); 3191 return false; 3192} 3193 3194static const Expr *getStrlenExprArg(const Expr *E) { 3195 if (const CallExpr *CE = dyn_cast<CallExpr>(E)) { 3196 const FunctionDecl *FD = CE->getDirectCallee(); 3197 if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen) 3198 return 0; 3199 return CE->getArg(0)->IgnoreParenCasts(); 3200 } 3201 return 0; 3202} 3203 3204// Warn on anti-patterns as the 'size' argument to strncat. 3205// The correct size argument should look like following: 3206// strncat(dst, src, sizeof(dst) - strlen(dest) - 1); 3207void Sema::CheckStrncatArguments(const CallExpr *CE, 3208 IdentifierInfo *FnName) { 3209 // Don't crash if the user has the wrong number of arguments. 3210 if (CE->getNumArgs() < 3) 3211 return; 3212 const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts(); 3213 const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts(); 3214 const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts(); 3215 3216 // Identify common expressions, which are wrongly used as the size argument 3217 // to strncat and may lead to buffer overflows. 3218 unsigned PatternType = 0; 3219 if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) { 3220 // - sizeof(dst) 3221 if (referToTheSameDecl(SizeOfArg, DstArg)) 3222 PatternType = 1; 3223 // - sizeof(src) 3224 else if (referToTheSameDecl(SizeOfArg, SrcArg)) 3225 PatternType = 2; 3226 } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) { 3227 if (BE->getOpcode() == BO_Sub) { 3228 const Expr *L = BE->getLHS()->IgnoreParenCasts(); 3229 const Expr *R = BE->getRHS()->IgnoreParenCasts(); 3230 // - sizeof(dst) - strlen(dst) 3231 if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) && 3232 referToTheSameDecl(DstArg, getStrlenExprArg(R))) 3233 PatternType = 1; 3234 // - sizeof(src) - (anything) 3235 else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L))) 3236 PatternType = 2; 3237 } 3238 } 3239 3240 if (PatternType == 0) 3241 return; 3242 3243 // Generate the diagnostic. 3244 SourceLocation SL = LenArg->getLocStart(); 3245 SourceRange SR = LenArg->getSourceRange(); 3246 SourceManager &SM = PP.getSourceManager(); 3247 3248 // If the function is defined as a builtin macro, do not show macro expansion. 3249 if (SM.isMacroArgExpansion(SL)) { 3250 SL = SM.getSpellingLoc(SL); 3251 SR = SourceRange(SM.getSpellingLoc(SR.getBegin()), 3252 SM.getSpellingLoc(SR.getEnd())); 3253 } 3254 3255 // Check if the destination is an array (rather than a pointer to an array). 3256 QualType DstTy = DstArg->getType(); 3257 bool isKnownSizeArray = isConstantSizeArrayWithMoreThanOneElement(DstTy, 3258 Context); 3259 if (!isKnownSizeArray) { 3260 if (PatternType == 1) 3261 Diag(SL, diag::warn_strncat_wrong_size) << SR; 3262 else 3263 Diag(SL, diag::warn_strncat_src_size) << SR; 3264 return; 3265 } 3266 3267 if (PatternType == 1) 3268 Diag(SL, diag::warn_strncat_large_size) << SR; 3269 else 3270 Diag(SL, diag::warn_strncat_src_size) << SR; 3271 3272 SmallString<128> sizeString; 3273 llvm::raw_svector_ostream OS(sizeString); 3274 OS << "sizeof("; 3275 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3276 OS << ") - "; 3277 OS << "strlen("; 3278 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3279 OS << ") - 1"; 3280 3281 Diag(SL, diag::note_strncat_wrong_size) 3282 << FixItHint::CreateReplacement(SR, OS.str()); 3283} 3284 3285//===--- CHECK: Return Address of Stack Variable --------------------------===// 3286 3287static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3288 Decl *ParentDecl); 3289static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars, 3290 Decl *ParentDecl); 3291 3292/// CheckReturnStackAddr - Check if a return statement returns the address 3293/// of a stack variable. 3294void 3295Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 3296 SourceLocation ReturnLoc) { 3297 3298 Expr *stackE = 0; 3299 SmallVector<DeclRefExpr *, 8> refVars; 3300 3301 // Perform checking for returned stack addresses, local blocks, 3302 // label addresses or references to temporaries. 3303 if (lhsType->isPointerType() || 3304 (!getLangOpts().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 3305 stackE = EvalAddr(RetValExp, refVars, /*ParentDecl=*/0); 3306 } else if (lhsType->isReferenceType()) { 3307 stackE = EvalVal(RetValExp, refVars, /*ParentDecl=*/0); 3308 } 3309 3310 if (stackE == 0) 3311 return; // Nothing suspicious was found. 3312 3313 SourceLocation diagLoc; 3314 SourceRange diagRange; 3315 if (refVars.empty()) { 3316 diagLoc = stackE->getLocStart(); 3317 diagRange = stackE->getSourceRange(); 3318 } else { 3319 // We followed through a reference variable. 'stackE' contains the 3320 // problematic expression but we will warn at the return statement pointing 3321 // at the reference variable. We will later display the "trail" of 3322 // reference variables using notes. 3323 diagLoc = refVars[0]->getLocStart(); 3324 diagRange = refVars[0]->getSourceRange(); 3325 } 3326 3327 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 3328 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 3329 : diag::warn_ret_stack_addr) 3330 << DR->getDecl()->getDeclName() << diagRange; 3331 } else if (isa<BlockExpr>(stackE)) { // local block. 3332 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 3333 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 3334 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 3335 } else { // local temporary. 3336 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 3337 : diag::warn_ret_local_temp_addr) 3338 << diagRange; 3339 } 3340 3341 // Display the "trail" of reference variables that we followed until we 3342 // found the problematic expression using notes. 3343 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 3344 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 3345 // If this var binds to another reference var, show the range of the next 3346 // var, otherwise the var binds to the problematic expression, in which case 3347 // show the range of the expression. 3348 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 3349 : stackE->getSourceRange(); 3350 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 3351 << VD->getDeclName() << range; 3352 } 3353} 3354 3355/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 3356/// check if the expression in a return statement evaluates to an address 3357/// to a location on the stack, a local block, an address of a label, or a 3358/// reference to local temporary. The recursion is used to traverse the 3359/// AST of the return expression, with recursion backtracking when we 3360/// encounter a subexpression that (1) clearly does not lead to one of the 3361/// above problematic expressions (2) is something we cannot determine leads to 3362/// a problematic expression based on such local checking. 3363/// 3364/// Both EvalAddr and EvalVal follow through reference variables to evaluate 3365/// the expression that they point to. Such variables are added to the 3366/// 'refVars' vector so that we know what the reference variable "trail" was. 3367/// 3368/// EvalAddr processes expressions that are pointers that are used as 3369/// references (and not L-values). EvalVal handles all other values. 3370/// At the base case of the recursion is a check for the above problematic 3371/// expressions. 3372/// 3373/// This implementation handles: 3374/// 3375/// * pointer-to-pointer casts 3376/// * implicit conversions from array references to pointers 3377/// * taking the address of fields 3378/// * arbitrary interplay between "&" and "*" operators 3379/// * pointer arithmetic from an address of a stack variable 3380/// * taking the address of an array element where the array is on the stack 3381static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3382 Decl *ParentDecl) { 3383 if (E->isTypeDependent()) 3384 return NULL; 3385 3386 // We should only be called for evaluating pointer expressions. 3387 assert((E->getType()->isAnyPointerType() || 3388 E->getType()->isBlockPointerType() || 3389 E->getType()->isObjCQualifiedIdType()) && 3390 "EvalAddr only works on pointers"); 3391 3392 E = E->IgnoreParens(); 3393 3394 // Our "symbolic interpreter" is just a dispatch off the currently 3395 // viewed AST node. We then recursively traverse the AST by calling 3396 // EvalAddr and EvalVal appropriately. 3397 switch (E->getStmtClass()) { 3398 case Stmt::DeclRefExprClass: { 3399 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3400 3401 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 3402 // If this is a reference variable, follow through to the expression that 3403 // it points to. 3404 if (V->hasLocalStorage() && 3405 V->getType()->isReferenceType() && V->hasInit()) { 3406 // Add the reference variable to the "trail". 3407 refVars.push_back(DR); 3408 return EvalAddr(V->getInit(), refVars, ParentDecl); 3409 } 3410 3411 return NULL; 3412 } 3413 3414 case Stmt::UnaryOperatorClass: { 3415 // The only unary operator that make sense to handle here 3416 // is AddrOf. All others don't make sense as pointers. 3417 UnaryOperator *U = cast<UnaryOperator>(E); 3418 3419 if (U->getOpcode() == UO_AddrOf) 3420 return EvalVal(U->getSubExpr(), refVars, ParentDecl); 3421 else 3422 return NULL; 3423 } 3424 3425 case Stmt::BinaryOperatorClass: { 3426 // Handle pointer arithmetic. All other binary operators are not valid 3427 // in this context. 3428 BinaryOperator *B = cast<BinaryOperator>(E); 3429 BinaryOperatorKind op = B->getOpcode(); 3430 3431 if (op != BO_Add && op != BO_Sub) 3432 return NULL; 3433 3434 Expr *Base = B->getLHS(); 3435 3436 // Determine which argument is the real pointer base. It could be 3437 // the RHS argument instead of the LHS. 3438 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 3439 3440 assert (Base->getType()->isPointerType()); 3441 return EvalAddr(Base, refVars, ParentDecl); 3442 } 3443 3444 // For conditional operators we need to see if either the LHS or RHS are 3445 // valid DeclRefExpr*s. If one of them is valid, we return it. 3446 case Stmt::ConditionalOperatorClass: { 3447 ConditionalOperator *C = cast<ConditionalOperator>(E); 3448 3449 // Handle the GNU extension for missing LHS. 3450 if (Expr *lhsExpr = C->getLHS()) { 3451 // In C++, we can have a throw-expression, which has 'void' type. 3452 if (!lhsExpr->getType()->isVoidType()) 3453 if (Expr* LHS = EvalAddr(lhsExpr, refVars, ParentDecl)) 3454 return LHS; 3455 } 3456 3457 // In C++, we can have a throw-expression, which has 'void' type. 3458 if (C->getRHS()->getType()->isVoidType()) 3459 return NULL; 3460 3461 return EvalAddr(C->getRHS(), refVars, ParentDecl); 3462 } 3463 3464 case Stmt::BlockExprClass: 3465 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 3466 return E; // local block. 3467 return NULL; 3468 3469 case Stmt::AddrLabelExprClass: 3470 return E; // address of label. 3471 3472 case Stmt::ExprWithCleanupsClass: 3473 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars, 3474 ParentDecl); 3475 3476 // For casts, we need to handle conversions from arrays to 3477 // pointer values, and pointer-to-pointer conversions. 3478 case Stmt::ImplicitCastExprClass: 3479 case Stmt::CStyleCastExprClass: 3480 case Stmt::CXXFunctionalCastExprClass: 3481 case Stmt::ObjCBridgedCastExprClass: 3482 case Stmt::CXXStaticCastExprClass: 3483 case Stmt::CXXDynamicCastExprClass: 3484 case Stmt::CXXConstCastExprClass: 3485 case Stmt::CXXReinterpretCastExprClass: { 3486 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 3487 switch (cast<CastExpr>(E)->getCastKind()) { 3488 case CK_BitCast: 3489 case CK_LValueToRValue: 3490 case CK_NoOp: 3491 case CK_BaseToDerived: 3492 case CK_DerivedToBase: 3493 case CK_UncheckedDerivedToBase: 3494 case CK_Dynamic: 3495 case CK_CPointerToObjCPointerCast: 3496 case CK_BlockPointerToObjCPointerCast: 3497 case CK_AnyPointerToBlockPointerCast: 3498 return EvalAddr(SubExpr, refVars, ParentDecl); 3499 3500 case CK_ArrayToPointerDecay: 3501 return EvalVal(SubExpr, refVars, ParentDecl); 3502 3503 default: 3504 return 0; 3505 } 3506 } 3507 3508 case Stmt::MaterializeTemporaryExprClass: 3509 if (Expr *Result = EvalAddr( 3510 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3511 refVars, ParentDecl)) 3512 return Result; 3513 3514 return E; 3515 3516 // Everything else: we simply don't reason about them. 3517 default: 3518 return NULL; 3519 } 3520} 3521 3522 3523/// EvalVal - This function is complements EvalAddr in the mutual recursion. 3524/// See the comments for EvalAddr for more details. 3525static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3526 Decl *ParentDecl) { 3527do { 3528 // We should only be called for evaluating non-pointer expressions, or 3529 // expressions with a pointer type that are not used as references but instead 3530 // are l-values (e.g., DeclRefExpr with a pointer type). 3531 3532 // Our "symbolic interpreter" is just a dispatch off the currently 3533 // viewed AST node. We then recursively traverse the AST by calling 3534 // EvalAddr and EvalVal appropriately. 3535 3536 E = E->IgnoreParens(); 3537 switch (E->getStmtClass()) { 3538 case Stmt::ImplicitCastExprClass: { 3539 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 3540 if (IE->getValueKind() == VK_LValue) { 3541 E = IE->getSubExpr(); 3542 continue; 3543 } 3544 return NULL; 3545 } 3546 3547 case Stmt::ExprWithCleanupsClass: 3548 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,ParentDecl); 3549 3550 case Stmt::DeclRefExprClass: { 3551 // When we hit a DeclRefExpr we are looking at code that refers to a 3552 // variable's name. If it's not a reference variable we check if it has 3553 // local storage within the function, and if so, return the expression. 3554 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3555 3556 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) { 3557 // Check if it refers to itself, e.g. "int& i = i;". 3558 if (V == ParentDecl) 3559 return DR; 3560 3561 if (V->hasLocalStorage()) { 3562 if (!V->getType()->isReferenceType()) 3563 return DR; 3564 3565 // Reference variable, follow through to the expression that 3566 // it points to. 3567 if (V->hasInit()) { 3568 // Add the reference variable to the "trail". 3569 refVars.push_back(DR); 3570 return EvalVal(V->getInit(), refVars, V); 3571 } 3572 } 3573 } 3574 3575 return NULL; 3576 } 3577 3578 case Stmt::UnaryOperatorClass: { 3579 // The only unary operator that make sense to handle here 3580 // is Deref. All others don't resolve to a "name." This includes 3581 // handling all sorts of rvalues passed to a unary operator. 3582 UnaryOperator *U = cast<UnaryOperator>(E); 3583 3584 if (U->getOpcode() == UO_Deref) 3585 return EvalAddr(U->getSubExpr(), refVars, ParentDecl); 3586 3587 return NULL; 3588 } 3589 3590 case Stmt::ArraySubscriptExprClass: { 3591 // Array subscripts are potential references to data on the stack. We 3592 // retrieve the DeclRefExpr* for the array variable if it indeed 3593 // has local storage. 3594 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars,ParentDecl); 3595 } 3596 3597 case Stmt::ConditionalOperatorClass: { 3598 // For conditional operators we need to see if either the LHS or RHS are 3599 // non-NULL Expr's. If one is non-NULL, we return it. 3600 ConditionalOperator *C = cast<ConditionalOperator>(E); 3601 3602 // Handle the GNU extension for missing LHS. 3603 if (Expr *lhsExpr = C->getLHS()) 3604 if (Expr *LHS = EvalVal(lhsExpr, refVars, ParentDecl)) 3605 return LHS; 3606 3607 return EvalVal(C->getRHS(), refVars, ParentDecl); 3608 } 3609 3610 // Accesses to members are potential references to data on the stack. 3611 case Stmt::MemberExprClass: { 3612 MemberExpr *M = cast<MemberExpr>(E); 3613 3614 // Check for indirect access. We only want direct field accesses. 3615 if (M->isArrow()) 3616 return NULL; 3617 3618 // Check whether the member type is itself a reference, in which case 3619 // we're not going to refer to the member, but to what the member refers to. 3620 if (M->getMemberDecl()->getType()->isReferenceType()) 3621 return NULL; 3622 3623 return EvalVal(M->getBase(), refVars, ParentDecl); 3624 } 3625 3626 case Stmt::MaterializeTemporaryExprClass: 3627 if (Expr *Result = EvalVal( 3628 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3629 refVars, ParentDecl)) 3630 return Result; 3631 3632 return E; 3633 3634 default: 3635 // Check that we don't return or take the address of a reference to a 3636 // temporary. This is only useful in C++. 3637 if (!E->isTypeDependent() && E->isRValue()) 3638 return E; 3639 3640 // Everything else: we simply don't reason about them. 3641 return NULL; 3642 } 3643} while (true); 3644} 3645 3646//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 3647 3648/// Check for comparisons of floating point operands using != and ==. 3649/// Issue a warning if these are no self-comparisons, as they are not likely 3650/// to do what the programmer intended. 3651void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 3652 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 3653 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 3654 3655 // Special case: check for x == x (which is OK). 3656 // Do not emit warnings for such cases. 3657 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 3658 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 3659 if (DRL->getDecl() == DRR->getDecl()) 3660 return; 3661 3662 3663 // Special case: check for comparisons against literals that can be exactly 3664 // represented by APFloat. In such cases, do not emit a warning. This 3665 // is a heuristic: often comparison against such literals are used to 3666 // detect if a value in a variable has not changed. This clearly can 3667 // lead to false negatives. 3668 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 3669 if (FLL->isExact()) 3670 return; 3671 } else 3672 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)) 3673 if (FLR->isExact()) 3674 return; 3675 3676 // Check for comparisons with builtin types. 3677 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 3678 if (CL->isBuiltinCall()) 3679 return; 3680 3681 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 3682 if (CR->isBuiltinCall()) 3683 return; 3684 3685 // Emit the diagnostic. 3686 Diag(Loc, diag::warn_floatingpoint_eq) 3687 << LHS->getSourceRange() << RHS->getSourceRange(); 3688} 3689 3690//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 3691//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 3692 3693namespace { 3694 3695/// Structure recording the 'active' range of an integer-valued 3696/// expression. 3697struct IntRange { 3698 /// The number of bits active in the int. 3699 unsigned Width; 3700 3701 /// True if the int is known not to have negative values. 3702 bool NonNegative; 3703 3704 IntRange(unsigned Width, bool NonNegative) 3705 : Width(Width), NonNegative(NonNegative) 3706 {} 3707 3708 /// Returns the range of the bool type. 3709 static IntRange forBoolType() { 3710 return IntRange(1, true); 3711 } 3712 3713 /// Returns the range of an opaque value of the given integral type. 3714 static IntRange forValueOfType(ASTContext &C, QualType T) { 3715 return forValueOfCanonicalType(C, 3716 T->getCanonicalTypeInternal().getTypePtr()); 3717 } 3718 3719 /// Returns the range of an opaque value of a canonical integral type. 3720 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 3721 assert(T->isCanonicalUnqualified()); 3722 3723 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3724 T = VT->getElementType().getTypePtr(); 3725 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3726 T = CT->getElementType().getTypePtr(); 3727 3728 // For enum types, use the known bit width of the enumerators. 3729 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 3730 EnumDecl *Enum = ET->getDecl(); 3731 if (!Enum->isCompleteDefinition()) 3732 return IntRange(C.getIntWidth(QualType(T, 0)), false); 3733 3734 unsigned NumPositive = Enum->getNumPositiveBits(); 3735 unsigned NumNegative = Enum->getNumNegativeBits(); 3736 3737 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 3738 } 3739 3740 const BuiltinType *BT = cast<BuiltinType>(T); 3741 assert(BT->isInteger()); 3742 3743 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3744 } 3745 3746 /// Returns the "target" range of a canonical integral type, i.e. 3747 /// the range of values expressible in the type. 3748 /// 3749 /// This matches forValueOfCanonicalType except that enums have the 3750 /// full range of their type, not the range of their enumerators. 3751 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 3752 assert(T->isCanonicalUnqualified()); 3753 3754 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3755 T = VT->getElementType().getTypePtr(); 3756 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3757 T = CT->getElementType().getTypePtr(); 3758 if (const EnumType *ET = dyn_cast<EnumType>(T)) 3759 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 3760 3761 const BuiltinType *BT = cast<BuiltinType>(T); 3762 assert(BT->isInteger()); 3763 3764 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3765 } 3766 3767 /// Returns the supremum of two ranges: i.e. their conservative merge. 3768 static IntRange join(IntRange L, IntRange R) { 3769 return IntRange(std::max(L.Width, R.Width), 3770 L.NonNegative && R.NonNegative); 3771 } 3772 3773 /// Returns the infinum of two ranges: i.e. their aggressive merge. 3774 static IntRange meet(IntRange L, IntRange R) { 3775 return IntRange(std::min(L.Width, R.Width), 3776 L.NonNegative || R.NonNegative); 3777 } 3778}; 3779 3780static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, 3781 unsigned MaxWidth) { 3782 if (value.isSigned() && value.isNegative()) 3783 return IntRange(value.getMinSignedBits(), false); 3784 3785 if (value.getBitWidth() > MaxWidth) 3786 value = value.trunc(MaxWidth); 3787 3788 // isNonNegative() just checks the sign bit without considering 3789 // signedness. 3790 return IntRange(value.getActiveBits(), true); 3791} 3792 3793static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 3794 unsigned MaxWidth) { 3795 if (result.isInt()) 3796 return GetValueRange(C, result.getInt(), MaxWidth); 3797 3798 if (result.isVector()) { 3799 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 3800 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 3801 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 3802 R = IntRange::join(R, El); 3803 } 3804 return R; 3805 } 3806 3807 if (result.isComplexInt()) { 3808 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 3809 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 3810 return IntRange::join(R, I); 3811 } 3812 3813 // This can happen with lossless casts to intptr_t of "based" lvalues. 3814 // Assume it might use arbitrary bits. 3815 // FIXME: The only reason we need to pass the type in here is to get 3816 // the sign right on this one case. It would be nice if APValue 3817 // preserved this. 3818 assert(result.isLValue() || result.isAddrLabelDiff()); 3819 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 3820} 3821 3822/// Pseudo-evaluate the given integer expression, estimating the 3823/// range of values it might take. 3824/// 3825/// \param MaxWidth - the width to which the value will be truncated 3826static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 3827 E = E->IgnoreParens(); 3828 3829 // Try a full evaluation first. 3830 Expr::EvalResult result; 3831 if (E->EvaluateAsRValue(result, C)) 3832 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 3833 3834 // I think we only want to look through implicit casts here; if the 3835 // user has an explicit widening cast, we should treat the value as 3836 // being of the new, wider type. 3837 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 3838 if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue) 3839 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 3840 3841 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 3842 3843 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 3844 3845 // Assume that non-integer casts can span the full range of the type. 3846 if (!isIntegerCast) 3847 return OutputTypeRange; 3848 3849 IntRange SubRange 3850 = GetExprRange(C, CE->getSubExpr(), 3851 std::min(MaxWidth, OutputTypeRange.Width)); 3852 3853 // Bail out if the subexpr's range is as wide as the cast type. 3854 if (SubRange.Width >= OutputTypeRange.Width) 3855 return OutputTypeRange; 3856 3857 // Otherwise, we take the smaller width, and we're non-negative if 3858 // either the output type or the subexpr is. 3859 return IntRange(SubRange.Width, 3860 SubRange.NonNegative || OutputTypeRange.NonNegative); 3861 } 3862 3863 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 3864 // If we can fold the condition, just take that operand. 3865 bool CondResult; 3866 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 3867 return GetExprRange(C, CondResult ? CO->getTrueExpr() 3868 : CO->getFalseExpr(), 3869 MaxWidth); 3870 3871 // Otherwise, conservatively merge. 3872 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 3873 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 3874 return IntRange::join(L, R); 3875 } 3876 3877 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 3878 switch (BO->getOpcode()) { 3879 3880 // Boolean-valued operations are single-bit and positive. 3881 case BO_LAnd: 3882 case BO_LOr: 3883 case BO_LT: 3884 case BO_GT: 3885 case BO_LE: 3886 case BO_GE: 3887 case BO_EQ: 3888 case BO_NE: 3889 return IntRange::forBoolType(); 3890 3891 // The type of the assignments is the type of the LHS, so the RHS 3892 // is not necessarily the same type. 3893 case BO_MulAssign: 3894 case BO_DivAssign: 3895 case BO_RemAssign: 3896 case BO_AddAssign: 3897 case BO_SubAssign: 3898 case BO_XorAssign: 3899 case BO_OrAssign: 3900 // TODO: bitfields? 3901 return IntRange::forValueOfType(C, E->getType()); 3902 3903 // Simple assignments just pass through the RHS, which will have 3904 // been coerced to the LHS type. 3905 case BO_Assign: 3906 // TODO: bitfields? 3907 return GetExprRange(C, BO->getRHS(), MaxWidth); 3908 3909 // Operations with opaque sources are black-listed. 3910 case BO_PtrMemD: 3911 case BO_PtrMemI: 3912 return IntRange::forValueOfType(C, E->getType()); 3913 3914 // Bitwise-and uses the *infinum* of the two source ranges. 3915 case BO_And: 3916 case BO_AndAssign: 3917 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 3918 GetExprRange(C, BO->getRHS(), MaxWidth)); 3919 3920 // Left shift gets black-listed based on a judgement call. 3921 case BO_Shl: 3922 // ...except that we want to treat '1 << (blah)' as logically 3923 // positive. It's an important idiom. 3924 if (IntegerLiteral *I 3925 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 3926 if (I->getValue() == 1) { 3927 IntRange R = IntRange::forValueOfType(C, E->getType()); 3928 return IntRange(R.Width, /*NonNegative*/ true); 3929 } 3930 } 3931 // fallthrough 3932 3933 case BO_ShlAssign: 3934 return IntRange::forValueOfType(C, E->getType()); 3935 3936 // Right shift by a constant can narrow its left argument. 3937 case BO_Shr: 3938 case BO_ShrAssign: { 3939 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3940 3941 // If the shift amount is a positive constant, drop the width by 3942 // that much. 3943 llvm::APSInt shift; 3944 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 3945 shift.isNonNegative()) { 3946 unsigned zext = shift.getZExtValue(); 3947 if (zext >= L.Width) 3948 L.Width = (L.NonNegative ? 0 : 1); 3949 else 3950 L.Width -= zext; 3951 } 3952 3953 return L; 3954 } 3955 3956 // Comma acts as its right operand. 3957 case BO_Comma: 3958 return GetExprRange(C, BO->getRHS(), MaxWidth); 3959 3960 // Black-list pointer subtractions. 3961 case BO_Sub: 3962 if (BO->getLHS()->getType()->isPointerType()) 3963 return IntRange::forValueOfType(C, E->getType()); 3964 break; 3965 3966 // The width of a division result is mostly determined by the size 3967 // of the LHS. 3968 case BO_Div: { 3969 // Don't 'pre-truncate' the operands. 3970 unsigned opWidth = C.getIntWidth(E->getType()); 3971 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3972 3973 // If the divisor is constant, use that. 3974 llvm::APSInt divisor; 3975 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 3976 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 3977 if (log2 >= L.Width) 3978 L.Width = (L.NonNegative ? 0 : 1); 3979 else 3980 L.Width = std::min(L.Width - log2, MaxWidth); 3981 return L; 3982 } 3983 3984 // Otherwise, just use the LHS's width. 3985 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3986 return IntRange(L.Width, L.NonNegative && R.NonNegative); 3987 } 3988 3989 // The result of a remainder can't be larger than the result of 3990 // either side. 3991 case BO_Rem: { 3992 // Don't 'pre-truncate' the operands. 3993 unsigned opWidth = C.getIntWidth(E->getType()); 3994 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3995 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3996 3997 IntRange meet = IntRange::meet(L, R); 3998 meet.Width = std::min(meet.Width, MaxWidth); 3999 return meet; 4000 } 4001 4002 // The default behavior is okay for these. 4003 case BO_Mul: 4004 case BO_Add: 4005 case BO_Xor: 4006 case BO_Or: 4007 break; 4008 } 4009 4010 // The default case is to treat the operation as if it were closed 4011 // on the narrowest type that encompasses both operands. 4012 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 4013 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 4014 return IntRange::join(L, R); 4015 } 4016 4017 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 4018 switch (UO->getOpcode()) { 4019 // Boolean-valued operations are white-listed. 4020 case UO_LNot: 4021 return IntRange::forBoolType(); 4022 4023 // Operations with opaque sources are black-listed. 4024 case UO_Deref: 4025 case UO_AddrOf: // should be impossible 4026 return IntRange::forValueOfType(C, E->getType()); 4027 4028 default: 4029 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 4030 } 4031 } 4032 4033 if (dyn_cast<OffsetOfExpr>(E)) { 4034 IntRange::forValueOfType(C, E->getType()); 4035 } 4036 4037 if (FieldDecl *BitField = E->getBitField()) 4038 return IntRange(BitField->getBitWidthValue(C), 4039 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 4040 4041 return IntRange::forValueOfType(C, E->getType()); 4042} 4043 4044static IntRange GetExprRange(ASTContext &C, Expr *E) { 4045 return GetExprRange(C, E, C.getIntWidth(E->getType())); 4046} 4047 4048/// Checks whether the given value, which currently has the given 4049/// source semantics, has the same value when coerced through the 4050/// target semantics. 4051static bool IsSameFloatAfterCast(const llvm::APFloat &value, 4052 const llvm::fltSemantics &Src, 4053 const llvm::fltSemantics &Tgt) { 4054 llvm::APFloat truncated = value; 4055 4056 bool ignored; 4057 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 4058 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 4059 4060 return truncated.bitwiseIsEqual(value); 4061} 4062 4063/// Checks whether the given value, which currently has the given 4064/// source semantics, has the same value when coerced through the 4065/// target semantics. 4066/// 4067/// The value might be a vector of floats (or a complex number). 4068static bool IsSameFloatAfterCast(const APValue &value, 4069 const llvm::fltSemantics &Src, 4070 const llvm::fltSemantics &Tgt) { 4071 if (value.isFloat()) 4072 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 4073 4074 if (value.isVector()) { 4075 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 4076 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 4077 return false; 4078 return true; 4079 } 4080 4081 assert(value.isComplexFloat()); 4082 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 4083 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 4084} 4085 4086static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 4087 4088static bool IsZero(Sema &S, Expr *E) { 4089 // Suppress cases where we are comparing against an enum constant. 4090 if (const DeclRefExpr *DR = 4091 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 4092 if (isa<EnumConstantDecl>(DR->getDecl())) 4093 return false; 4094 4095 // Suppress cases where the '0' value is expanded from a macro. 4096 if (E->getLocStart().isMacroID()) 4097 return false; 4098 4099 llvm::APSInt Value; 4100 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 4101} 4102 4103static bool HasEnumType(Expr *E) { 4104 // Strip off implicit integral promotions. 4105 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 4106 if (ICE->getCastKind() != CK_IntegralCast && 4107 ICE->getCastKind() != CK_NoOp) 4108 break; 4109 E = ICE->getSubExpr(); 4110 } 4111 4112 return E->getType()->isEnumeralType(); 4113} 4114 4115static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 4116 BinaryOperatorKind op = E->getOpcode(); 4117 if (E->isValueDependent()) 4118 return; 4119 4120 if (op == BO_LT && IsZero(S, E->getRHS())) { 4121 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4122 << "< 0" << "false" << HasEnumType(E->getLHS()) 4123 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4124 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 4125 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4126 << ">= 0" << "true" << HasEnumType(E->getLHS()) 4127 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4128 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 4129 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4130 << "0 >" << "false" << HasEnumType(E->getRHS()) 4131 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4132 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 4133 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4134 << "0 <=" << "true" << HasEnumType(E->getRHS()) 4135 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4136 } 4137} 4138 4139/// Analyze the operands of the given comparison. Implements the 4140/// fallback case from AnalyzeComparison. 4141static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 4142 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4143 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4144} 4145 4146/// \brief Implements -Wsign-compare. 4147/// 4148/// \param E the binary operator to check for warnings 4149static void AnalyzeComparison(Sema &S, BinaryOperator *E) { 4150 // The type the comparison is being performed in. 4151 QualType T = E->getLHS()->getType(); 4152 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 4153 && "comparison with mismatched types"); 4154 4155 // We don't do anything special if this isn't an unsigned integral 4156 // comparison: we're only interested in integral comparisons, and 4157 // signed comparisons only happen in cases we don't care to warn about. 4158 // 4159 // We also don't care about value-dependent expressions or expressions 4160 // whose result is a constant. 4161 if (!T->hasUnsignedIntegerRepresentation() 4162 || E->isValueDependent() || E->isIntegerConstantExpr(S.Context)) 4163 return AnalyzeImpConvsInComparison(S, E); 4164 4165 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 4166 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 4167 4168 // Check to see if one of the (unmodified) operands is of different 4169 // signedness. 4170 Expr *signedOperand, *unsignedOperand; 4171 if (LHS->getType()->hasSignedIntegerRepresentation()) { 4172 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 4173 "unsigned comparison between two signed integer expressions?"); 4174 signedOperand = LHS; 4175 unsignedOperand = RHS; 4176 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 4177 signedOperand = RHS; 4178 unsignedOperand = LHS; 4179 } else { 4180 CheckTrivialUnsignedComparison(S, E); 4181 return AnalyzeImpConvsInComparison(S, E); 4182 } 4183 4184 // Otherwise, calculate the effective range of the signed operand. 4185 IntRange signedRange = GetExprRange(S.Context, signedOperand); 4186 4187 // Go ahead and analyze implicit conversions in the operands. Note 4188 // that we skip the implicit conversions on both sides. 4189 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 4190 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 4191 4192 // If the signed range is non-negative, -Wsign-compare won't fire, 4193 // but we should still check for comparisons which are always true 4194 // or false. 4195 if (signedRange.NonNegative) 4196 return CheckTrivialUnsignedComparison(S, E); 4197 4198 // For (in)equality comparisons, if the unsigned operand is a 4199 // constant which cannot collide with a overflowed signed operand, 4200 // then reinterpreting the signed operand as unsigned will not 4201 // change the result of the comparison. 4202 if (E->isEqualityOp()) { 4203 unsigned comparisonWidth = S.Context.getIntWidth(T); 4204 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 4205 4206 // We should never be unable to prove that the unsigned operand is 4207 // non-negative. 4208 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 4209 4210 if (unsignedRange.Width < comparisonWidth) 4211 return; 4212 } 4213 4214 S.DiagRuntimeBehavior(E->getOperatorLoc(), E, 4215 S.PDiag(diag::warn_mixed_sign_comparison) 4216 << LHS->getType() << RHS->getType() 4217 << LHS->getSourceRange() << RHS->getSourceRange()); 4218} 4219 4220/// Analyzes an attempt to assign the given value to a bitfield. 4221/// 4222/// Returns true if there was something fishy about the attempt. 4223static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 4224 SourceLocation InitLoc) { 4225 assert(Bitfield->isBitField()); 4226 if (Bitfield->isInvalidDecl()) 4227 return false; 4228 4229 // White-list bool bitfields. 4230 if (Bitfield->getType()->isBooleanType()) 4231 return false; 4232 4233 // Ignore value- or type-dependent expressions. 4234 if (Bitfield->getBitWidth()->isValueDependent() || 4235 Bitfield->getBitWidth()->isTypeDependent() || 4236 Init->isValueDependent() || 4237 Init->isTypeDependent()) 4238 return false; 4239 4240 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 4241 4242 llvm::APSInt Value; 4243 if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects)) 4244 return false; 4245 4246 unsigned OriginalWidth = Value.getBitWidth(); 4247 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 4248 4249 if (OriginalWidth <= FieldWidth) 4250 return false; 4251 4252 // Compute the value which the bitfield will contain. 4253 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 4254 TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType()); 4255 4256 // Check whether the stored value is equal to the original value. 4257 TruncatedValue = TruncatedValue.extend(OriginalWidth); 4258 if (llvm::APSInt::isSameValue(Value, TruncatedValue)) 4259 return false; 4260 4261 // Special-case bitfields of width 1: booleans are naturally 0/1, and 4262 // therefore don't strictly fit into a signed bitfield of width 1. 4263 if (FieldWidth == 1 && Value == 1) 4264 return false; 4265 4266 std::string PrettyValue = Value.toString(10); 4267 std::string PrettyTrunc = TruncatedValue.toString(10); 4268 4269 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 4270 << PrettyValue << PrettyTrunc << OriginalInit->getType() 4271 << Init->getSourceRange(); 4272 4273 return true; 4274} 4275 4276/// Analyze the given simple or compound assignment for warning-worthy 4277/// operations. 4278static void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 4279 // Just recurse on the LHS. 4280 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4281 4282 // We want to recurse on the RHS as normal unless we're assigning to 4283 // a bitfield. 4284 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 4285 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 4286 E->getOperatorLoc())) { 4287 // Recurse, ignoring any implicit conversions on the RHS. 4288 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 4289 E->getOperatorLoc()); 4290 } 4291 } 4292 4293 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4294} 4295 4296/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4297static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 4298 SourceLocation CContext, unsigned diag, 4299 bool pruneControlFlow = false) { 4300 if (pruneControlFlow) { 4301 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4302 S.PDiag(diag) 4303 << SourceType << T << E->getSourceRange() 4304 << SourceRange(CContext)); 4305 return; 4306 } 4307 S.Diag(E->getExprLoc(), diag) 4308 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 4309} 4310 4311/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4312static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, 4313 SourceLocation CContext, unsigned diag, 4314 bool pruneControlFlow = false) { 4315 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow); 4316} 4317 4318/// Diagnose an implicit cast from a literal expression. Does not warn when the 4319/// cast wouldn't lose information. 4320void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 4321 SourceLocation CContext) { 4322 // Try to convert the literal exactly to an integer. If we can, don't warn. 4323 bool isExact = false; 4324 const llvm::APFloat &Value = FL->getValue(); 4325 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 4326 T->hasUnsignedIntegerRepresentation()); 4327 if (Value.convertToInteger(IntegerValue, 4328 llvm::APFloat::rmTowardZero, &isExact) 4329 == llvm::APFloat::opOK && isExact) 4330 return; 4331 4332 SmallString<16> PrettySourceValue; 4333 Value.toString(PrettySourceValue); 4334 SmallString<16> PrettyTargetValue; 4335 if (T->isSpecificBuiltinType(BuiltinType::Bool)) 4336 PrettyTargetValue = IntegerValue == 0 ? "false" : "true"; 4337 else 4338 IntegerValue.toString(PrettyTargetValue); 4339 4340 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 4341 << FL->getType() << T.getUnqualifiedType() << PrettySourceValue 4342 << PrettyTargetValue << FL->getSourceRange() << SourceRange(CContext); 4343} 4344 4345std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 4346 if (!Range.Width) return "0"; 4347 4348 llvm::APSInt ValueInRange = Value; 4349 ValueInRange.setIsSigned(!Range.NonNegative); 4350 ValueInRange = ValueInRange.trunc(Range.Width); 4351 return ValueInRange.toString(10); 4352} 4353 4354void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 4355 SourceLocation CC, bool *ICContext = 0) { 4356 if (E->isTypeDependent() || E->isValueDependent()) return; 4357 4358 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 4359 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 4360 if (Source == Target) return; 4361 if (Target->isDependentType()) return; 4362 4363 // If the conversion context location is invalid don't complain. We also 4364 // don't want to emit a warning if the issue occurs from the expansion of 4365 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 4366 // delay this check as long as possible. Once we detect we are in that 4367 // scenario, we just return. 4368 if (CC.isInvalid()) 4369 return; 4370 4371 // Diagnose implicit casts to bool. 4372 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 4373 if (isa<StringLiteral>(E)) 4374 // Warn on string literal to bool. Checks for string literals in logical 4375 // expressions, for instances, assert(0 && "error here"), is prevented 4376 // by a check in AnalyzeImplicitConversions(). 4377 return DiagnoseImpCast(S, E, T, CC, 4378 diag::warn_impcast_string_literal_to_bool); 4379 if (Source->isFunctionType()) { 4380 // Warn on function to bool. Checks free functions and static member 4381 // functions. Weakly imported functions are excluded from the check, 4382 // since it's common to test their value to check whether the linker 4383 // found a definition for them. 4384 ValueDecl *D = 0; 4385 if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) { 4386 D = R->getDecl(); 4387 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 4388 D = M->getMemberDecl(); 4389 } 4390 4391 if (D && !D->isWeak()) { 4392 if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) { 4393 S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool) 4394 << F << E->getSourceRange() << SourceRange(CC); 4395 S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence) 4396 << FixItHint::CreateInsertion(E->getExprLoc(), "&"); 4397 QualType ReturnType; 4398 UnresolvedSet<4> NonTemplateOverloads; 4399 S.isExprCallable(*E, ReturnType, NonTemplateOverloads); 4400 if (!ReturnType.isNull() 4401 && ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 4402 S.Diag(E->getExprLoc(), diag::note_function_to_bool_call) 4403 << FixItHint::CreateInsertion( 4404 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()"); 4405 return; 4406 } 4407 } 4408 } 4409 } 4410 4411 // Strip vector types. 4412 if (isa<VectorType>(Source)) { 4413 if (!isa<VectorType>(Target)) { 4414 if (S.SourceMgr.isInSystemMacro(CC)) 4415 return; 4416 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 4417 } 4418 4419 // If the vector cast is cast between two vectors of the same size, it is 4420 // a bitcast, not a conversion. 4421 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 4422 return; 4423 4424 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 4425 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 4426 } 4427 4428 // Strip complex types. 4429 if (isa<ComplexType>(Source)) { 4430 if (!isa<ComplexType>(Target)) { 4431 if (S.SourceMgr.isInSystemMacro(CC)) 4432 return; 4433 4434 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 4435 } 4436 4437 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 4438 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 4439 } 4440 4441 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 4442 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 4443 4444 // If the source is floating point... 4445 if (SourceBT && SourceBT->isFloatingPoint()) { 4446 // ...and the target is floating point... 4447 if (TargetBT && TargetBT->isFloatingPoint()) { 4448 // ...then warn if we're dropping FP rank. 4449 4450 // Builtin FP kinds are ordered by increasing FP rank. 4451 if (SourceBT->getKind() > TargetBT->getKind()) { 4452 // Don't warn about float constants that are precisely 4453 // representable in the target type. 4454 Expr::EvalResult result; 4455 if (E->EvaluateAsRValue(result, S.Context)) { 4456 // Value might be a float, a float vector, or a float complex. 4457 if (IsSameFloatAfterCast(result.Val, 4458 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 4459 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 4460 return; 4461 } 4462 4463 if (S.SourceMgr.isInSystemMacro(CC)) 4464 return; 4465 4466 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 4467 } 4468 return; 4469 } 4470 4471 // If the target is integral, always warn. 4472 if (TargetBT && TargetBT->isInteger()) { 4473 if (S.SourceMgr.isInSystemMacro(CC)) 4474 return; 4475 4476 Expr *InnerE = E->IgnoreParenImpCasts(); 4477 // We also want to warn on, e.g., "int i = -1.234" 4478 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 4479 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 4480 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 4481 4482 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 4483 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 4484 } else { 4485 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 4486 } 4487 } 4488 4489 return; 4490 } 4491 4492 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 4493 == Expr::NPCK_GNUNull) && !Target->isAnyPointerType() 4494 && !Target->isBlockPointerType() && !Target->isMemberPointerType()) { 4495 SourceLocation Loc = E->getSourceRange().getBegin(); 4496 if (Loc.isMacroID()) 4497 Loc = S.SourceMgr.getImmediateExpansionRange(Loc).first; 4498 if (!Loc.isMacroID() || CC.isMacroID()) 4499 S.Diag(Loc, diag::warn_impcast_null_pointer_to_integer) 4500 << T << clang::SourceRange(CC) 4501 << FixItHint::CreateReplacement(Loc, S.getFixItZeroLiteralForType(T)); 4502 } 4503 4504 if (!Source->isIntegerType() || !Target->isIntegerType()) 4505 return; 4506 4507 // TODO: remove this early return once the false positives for constant->bool 4508 // in templates, macros, etc, are reduced or removed. 4509 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 4510 return; 4511 4512 IntRange SourceRange = GetExprRange(S.Context, E); 4513 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 4514 4515 if (SourceRange.Width > TargetRange.Width) { 4516 // If the source is a constant, use a default-on diagnostic. 4517 // TODO: this should happen for bitfield stores, too. 4518 llvm::APSInt Value(32); 4519 if (E->isIntegerConstantExpr(Value, S.Context)) { 4520 if (S.SourceMgr.isInSystemMacro(CC)) 4521 return; 4522 4523 std::string PrettySourceValue = Value.toString(10); 4524 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 4525 4526 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4527 S.PDiag(diag::warn_impcast_integer_precision_constant) 4528 << PrettySourceValue << PrettyTargetValue 4529 << E->getType() << T << E->getSourceRange() 4530 << clang::SourceRange(CC)); 4531 return; 4532 } 4533 4534 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 4535 if (S.SourceMgr.isInSystemMacro(CC)) 4536 return; 4537 4538 if (TargetRange.Width == 32 && S.Context.getIntWidth(E->getType()) == 64) 4539 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32, 4540 /* pruneControlFlow */ true); 4541 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 4542 } 4543 4544 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 4545 (!TargetRange.NonNegative && SourceRange.NonNegative && 4546 SourceRange.Width == TargetRange.Width)) { 4547 4548 if (S.SourceMgr.isInSystemMacro(CC)) 4549 return; 4550 4551 unsigned DiagID = diag::warn_impcast_integer_sign; 4552 4553 // Traditionally, gcc has warned about this under -Wsign-compare. 4554 // We also want to warn about it in -Wconversion. 4555 // So if -Wconversion is off, use a completely identical diagnostic 4556 // in the sign-compare group. 4557 // The conditional-checking code will 4558 if (ICContext) { 4559 DiagID = diag::warn_impcast_integer_sign_conditional; 4560 *ICContext = true; 4561 } 4562 4563 return DiagnoseImpCast(S, E, T, CC, DiagID); 4564 } 4565 4566 // Diagnose conversions between different enumeration types. 4567 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 4568 // type, to give us better diagnostics. 4569 QualType SourceType = E->getType(); 4570 if (!S.getLangOpts().CPlusPlus) { 4571 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 4572 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 4573 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 4574 SourceType = S.Context.getTypeDeclType(Enum); 4575 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 4576 } 4577 } 4578 4579 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 4580 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 4581 if ((SourceEnum->getDecl()->getIdentifier() || 4582 SourceEnum->getDecl()->getTypedefNameForAnonDecl()) && 4583 (TargetEnum->getDecl()->getIdentifier() || 4584 TargetEnum->getDecl()->getTypedefNameForAnonDecl()) && 4585 SourceEnum != TargetEnum) { 4586 if (S.SourceMgr.isInSystemMacro(CC)) 4587 return; 4588 4589 return DiagnoseImpCast(S, E, SourceType, T, CC, 4590 diag::warn_impcast_different_enum_types); 4591 } 4592 4593 return; 4594} 4595 4596void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 4597 SourceLocation CC, QualType T); 4598 4599void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 4600 SourceLocation CC, bool &ICContext) { 4601 E = E->IgnoreParenImpCasts(); 4602 4603 if (isa<ConditionalOperator>(E)) 4604 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), CC, T); 4605 4606 AnalyzeImplicitConversions(S, E, CC); 4607 if (E->getType() != T) 4608 return CheckImplicitConversion(S, E, T, CC, &ICContext); 4609 return; 4610} 4611 4612void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 4613 SourceLocation CC, QualType T) { 4614 AnalyzeImplicitConversions(S, E->getCond(), CC); 4615 4616 bool Suspicious = false; 4617 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 4618 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 4619 4620 // If -Wconversion would have warned about either of the candidates 4621 // for a signedness conversion to the context type... 4622 if (!Suspicious) return; 4623 4624 // ...but it's currently ignored... 4625 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 4626 CC)) 4627 return; 4628 4629 // ...then check whether it would have warned about either of the 4630 // candidates for a signedness conversion to the condition type. 4631 if (E->getType() == T) return; 4632 4633 Suspicious = false; 4634 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 4635 E->getType(), CC, &Suspicious); 4636 if (!Suspicious) 4637 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 4638 E->getType(), CC, &Suspicious); 4639} 4640 4641/// AnalyzeImplicitConversions - Find and report any interesting 4642/// implicit conversions in the given expression. There are a couple 4643/// of competing diagnostics here, -Wconversion and -Wsign-compare. 4644void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 4645 QualType T = OrigE->getType(); 4646 Expr *E = OrigE->IgnoreParenImpCasts(); 4647 4648 if (E->isTypeDependent() || E->isValueDependent()) 4649 return; 4650 4651 // For conditional operators, we analyze the arguments as if they 4652 // were being fed directly into the output. 4653 if (isa<ConditionalOperator>(E)) { 4654 ConditionalOperator *CO = cast<ConditionalOperator>(E); 4655 CheckConditionalOperator(S, CO, CC, T); 4656 return; 4657 } 4658 4659 // Go ahead and check any implicit conversions we might have skipped. 4660 // The non-canonical typecheck is just an optimization; 4661 // CheckImplicitConversion will filter out dead implicit conversions. 4662 if (E->getType() != T) 4663 CheckImplicitConversion(S, E, T, CC); 4664 4665 // Now continue drilling into this expression. 4666 4667 // Skip past explicit casts. 4668 if (isa<ExplicitCastExpr>(E)) { 4669 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 4670 return AnalyzeImplicitConversions(S, E, CC); 4671 } 4672 4673 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 4674 // Do a somewhat different check with comparison operators. 4675 if (BO->isComparisonOp()) 4676 return AnalyzeComparison(S, BO); 4677 4678 // And with simple assignments. 4679 if (BO->getOpcode() == BO_Assign) 4680 return AnalyzeAssignment(S, BO); 4681 } 4682 4683 // These break the otherwise-useful invariant below. Fortunately, 4684 // we don't really need to recurse into them, because any internal 4685 // expressions should have been analyzed already when they were 4686 // built into statements. 4687 if (isa<StmtExpr>(E)) return; 4688 4689 // Don't descend into unevaluated contexts. 4690 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 4691 4692 // Now just recurse over the expression's children. 4693 CC = E->getExprLoc(); 4694 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 4695 bool IsLogicalOperator = BO && BO->isLogicalOp(); 4696 for (Stmt::child_range I = E->children(); I; ++I) { 4697 Expr *ChildExpr = dyn_cast_or_null<Expr>(*I); 4698 if (!ChildExpr) 4699 continue; 4700 4701 if (IsLogicalOperator && 4702 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 4703 // Ignore checking string literals that are in logical operators. 4704 continue; 4705 AnalyzeImplicitConversions(S, ChildExpr, CC); 4706 } 4707} 4708 4709} // end anonymous namespace 4710 4711/// Diagnoses "dangerous" implicit conversions within the given 4712/// expression (which is a full expression). Implements -Wconversion 4713/// and -Wsign-compare. 4714/// 4715/// \param CC the "context" location of the implicit conversion, i.e. 4716/// the most location of the syntactic entity requiring the implicit 4717/// conversion 4718void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 4719 // Don't diagnose in unevaluated contexts. 4720 if (isUnevaluatedContext()) 4721 return; 4722 4723 // Don't diagnose for value- or type-dependent expressions. 4724 if (E->isTypeDependent() || E->isValueDependent()) 4725 return; 4726 4727 // Check for array bounds violations in cases where the check isn't triggered 4728 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 4729 // ArraySubscriptExpr is on the RHS of a variable initialization. 4730 CheckArrayAccess(E); 4731 4732 // This is not the right CC for (e.g.) a variable initialization. 4733 AnalyzeImplicitConversions(*this, E, CC); 4734} 4735 4736void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 4737 FieldDecl *BitField, 4738 Expr *Init) { 4739 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 4740} 4741 4742/// CheckParmsForFunctionDef - Check that the parameters of the given 4743/// function are appropriate for the definition of a function. This 4744/// takes care of any checks that cannot be performed on the 4745/// declaration itself, e.g., that the types of each of the function 4746/// parameters are complete. 4747bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 4748 bool CheckParameterNames) { 4749 bool HasInvalidParm = false; 4750 for (; P != PEnd; ++P) { 4751 ParmVarDecl *Param = *P; 4752 4753 // C99 6.7.5.3p4: the parameters in a parameter type list in a 4754 // function declarator that is part of a function definition of 4755 // that function shall not have incomplete type. 4756 // 4757 // This is also C++ [dcl.fct]p6. 4758 if (!Param->isInvalidDecl() && 4759 RequireCompleteType(Param->getLocation(), Param->getType(), 4760 diag::err_typecheck_decl_incomplete_type)) { 4761 Param->setInvalidDecl(); 4762 HasInvalidParm = true; 4763 } 4764 4765 // C99 6.9.1p5: If the declarator includes a parameter type list, the 4766 // declaration of each parameter shall include an identifier. 4767 if (CheckParameterNames && 4768 Param->getIdentifier() == 0 && 4769 !Param->isImplicit() && 4770 !getLangOpts().CPlusPlus) 4771 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 4772 4773 // C99 6.7.5.3p12: 4774 // If the function declarator is not part of a definition of that 4775 // function, parameters may have incomplete type and may use the [*] 4776 // notation in their sequences of declarator specifiers to specify 4777 // variable length array types. 4778 QualType PType = Param->getOriginalType(); 4779 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 4780 if (AT->getSizeModifier() == ArrayType::Star) { 4781 // FIXME: This diagnosic should point the '[*]' if source-location 4782 // information is added for it. 4783 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 4784 } 4785 } 4786 } 4787 4788 return HasInvalidParm; 4789} 4790 4791/// CheckCastAlign - Implements -Wcast-align, which warns when a 4792/// pointer cast increases the alignment requirements. 4793void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 4794 // This is actually a lot of work to potentially be doing on every 4795 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 4796 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 4797 TRange.getBegin()) 4798 == DiagnosticsEngine::Ignored) 4799 return; 4800 4801 // Ignore dependent types. 4802 if (T->isDependentType() || Op->getType()->isDependentType()) 4803 return; 4804 4805 // Require that the destination be a pointer type. 4806 const PointerType *DestPtr = T->getAs<PointerType>(); 4807 if (!DestPtr) return; 4808 4809 // If the destination has alignment 1, we're done. 4810 QualType DestPointee = DestPtr->getPointeeType(); 4811 if (DestPointee->isIncompleteType()) return; 4812 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 4813 if (DestAlign.isOne()) return; 4814 4815 // Require that the source be a pointer type. 4816 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 4817 if (!SrcPtr) return; 4818 QualType SrcPointee = SrcPtr->getPointeeType(); 4819 4820 // Whitelist casts from cv void*. We already implicitly 4821 // whitelisted casts to cv void*, since they have alignment 1. 4822 // Also whitelist casts involving incomplete types, which implicitly 4823 // includes 'void'. 4824 if (SrcPointee->isIncompleteType()) return; 4825 4826 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 4827 if (SrcAlign >= DestAlign) return; 4828 4829 Diag(TRange.getBegin(), diag::warn_cast_align) 4830 << Op->getType() << T 4831 << static_cast<unsigned>(SrcAlign.getQuantity()) 4832 << static_cast<unsigned>(DestAlign.getQuantity()) 4833 << TRange << Op->getSourceRange(); 4834} 4835 4836static const Type* getElementType(const Expr *BaseExpr) { 4837 const Type* EltType = BaseExpr->getType().getTypePtr(); 4838 if (EltType->isAnyPointerType()) 4839 return EltType->getPointeeType().getTypePtr(); 4840 else if (EltType->isArrayType()) 4841 return EltType->getBaseElementTypeUnsafe(); 4842 return EltType; 4843} 4844 4845/// \brief Check whether this array fits the idiom of a size-one tail padded 4846/// array member of a struct. 4847/// 4848/// We avoid emitting out-of-bounds access warnings for such arrays as they are 4849/// commonly used to emulate flexible arrays in C89 code. 4850static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 4851 const NamedDecl *ND) { 4852 if (Size != 1 || !ND) return false; 4853 4854 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 4855 if (!FD) return false; 4856 4857 // Don't consider sizes resulting from macro expansions or template argument 4858 // substitution to form C89 tail-padded arrays. 4859 4860 TypeSourceInfo *TInfo = FD->getTypeSourceInfo(); 4861 while (TInfo) { 4862 TypeLoc TL = TInfo->getTypeLoc(); 4863 // Look through typedefs. 4864 const TypedefTypeLoc *TTL = dyn_cast<TypedefTypeLoc>(&TL); 4865 if (TTL) { 4866 const TypedefNameDecl *TDL = TTL->getTypedefNameDecl(); 4867 TInfo = TDL->getTypeSourceInfo(); 4868 continue; 4869 } 4870 ConstantArrayTypeLoc CTL = cast<ConstantArrayTypeLoc>(TL); 4871 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(CTL.getSizeExpr()); 4872 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 4873 return false; 4874 break; 4875 } 4876 4877 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 4878 if (!RD) return false; 4879 if (RD->isUnion()) return false; 4880 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 4881 if (!CRD->isStandardLayout()) return false; 4882 } 4883 4884 // See if this is the last field decl in the record. 4885 const Decl *D = FD; 4886 while ((D = D->getNextDeclInContext())) 4887 if (isa<FieldDecl>(D)) 4888 return false; 4889 return true; 4890} 4891 4892void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 4893 const ArraySubscriptExpr *ASE, 4894 bool AllowOnePastEnd, bool IndexNegated) { 4895 IndexExpr = IndexExpr->IgnoreParenImpCasts(); 4896 if (IndexExpr->isValueDependent()) 4897 return; 4898 4899 const Type *EffectiveType = getElementType(BaseExpr); 4900 BaseExpr = BaseExpr->IgnoreParenCasts(); 4901 const ConstantArrayType *ArrayTy = 4902 Context.getAsConstantArrayType(BaseExpr->getType()); 4903 if (!ArrayTy) 4904 return; 4905 4906 llvm::APSInt index; 4907 if (!IndexExpr->EvaluateAsInt(index, Context)) 4908 return; 4909 if (IndexNegated) 4910 index = -index; 4911 4912 const NamedDecl *ND = NULL; 4913 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4914 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 4915 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 4916 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 4917 4918 if (index.isUnsigned() || !index.isNegative()) { 4919 llvm::APInt size = ArrayTy->getSize(); 4920 if (!size.isStrictlyPositive()) 4921 return; 4922 4923 const Type* BaseType = getElementType(BaseExpr); 4924 if (BaseType != EffectiveType) { 4925 // Make sure we're comparing apples to apples when comparing index to size 4926 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 4927 uint64_t array_typesize = Context.getTypeSize(BaseType); 4928 // Handle ptrarith_typesize being zero, such as when casting to void* 4929 if (!ptrarith_typesize) ptrarith_typesize = 1; 4930 if (ptrarith_typesize != array_typesize) { 4931 // There's a cast to a different size type involved 4932 uint64_t ratio = array_typesize / ptrarith_typesize; 4933 // TODO: Be smarter about handling cases where array_typesize is not a 4934 // multiple of ptrarith_typesize 4935 if (ptrarith_typesize * ratio == array_typesize) 4936 size *= llvm::APInt(size.getBitWidth(), ratio); 4937 } 4938 } 4939 4940 if (size.getBitWidth() > index.getBitWidth()) 4941 index = index.zext(size.getBitWidth()); 4942 else if (size.getBitWidth() < index.getBitWidth()) 4943 size = size.zext(index.getBitWidth()); 4944 4945 // For array subscripting the index must be less than size, but for pointer 4946 // arithmetic also allow the index (offset) to be equal to size since 4947 // computing the next address after the end of the array is legal and 4948 // commonly done e.g. in C++ iterators and range-based for loops. 4949 if (AllowOnePastEnd ? index.ule(size) : index.ult(size)) 4950 return; 4951 4952 // Also don't warn for arrays of size 1 which are members of some 4953 // structure. These are often used to approximate flexible arrays in C89 4954 // code. 4955 if (IsTailPaddedMemberArray(*this, size, ND)) 4956 return; 4957 4958 // Suppress the warning if the subscript expression (as identified by the 4959 // ']' location) and the index expression are both from macro expansions 4960 // within a system header. 4961 if (ASE) { 4962 SourceLocation RBracketLoc = SourceMgr.getSpellingLoc( 4963 ASE->getRBracketLoc()); 4964 if (SourceMgr.isInSystemHeader(RBracketLoc)) { 4965 SourceLocation IndexLoc = SourceMgr.getSpellingLoc( 4966 IndexExpr->getLocStart()); 4967 if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc)) 4968 return; 4969 } 4970 } 4971 4972 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 4973 if (ASE) 4974 DiagID = diag::warn_array_index_exceeds_bounds; 4975 4976 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4977 PDiag(DiagID) << index.toString(10, true) 4978 << size.toString(10, true) 4979 << (unsigned)size.getLimitedValue(~0U) 4980 << IndexExpr->getSourceRange()); 4981 } else { 4982 unsigned DiagID = diag::warn_array_index_precedes_bounds; 4983 if (!ASE) { 4984 DiagID = diag::warn_ptr_arith_precedes_bounds; 4985 if (index.isNegative()) index = -index; 4986 } 4987 4988 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4989 PDiag(DiagID) << index.toString(10, true) 4990 << IndexExpr->getSourceRange()); 4991 } 4992 4993 if (!ND) { 4994 // Try harder to find a NamedDecl to point at in the note. 4995 while (const ArraySubscriptExpr *ASE = 4996 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 4997 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 4998 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4999 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 5000 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 5001 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 5002 } 5003 5004 if (ND) 5005 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 5006 PDiag(diag::note_array_index_out_of_bounds) 5007 << ND->getDeclName()); 5008} 5009 5010void Sema::CheckArrayAccess(const Expr *expr) { 5011 int AllowOnePastEnd = 0; 5012 while (expr) { 5013 expr = expr->IgnoreParenImpCasts(); 5014 switch (expr->getStmtClass()) { 5015 case Stmt::ArraySubscriptExprClass: { 5016 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 5017 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE, 5018 AllowOnePastEnd > 0); 5019 return; 5020 } 5021 case Stmt::UnaryOperatorClass: { 5022 // Only unwrap the * and & unary operators 5023 const UnaryOperator *UO = cast<UnaryOperator>(expr); 5024 expr = UO->getSubExpr(); 5025 switch (UO->getOpcode()) { 5026 case UO_AddrOf: 5027 AllowOnePastEnd++; 5028 break; 5029 case UO_Deref: 5030 AllowOnePastEnd--; 5031 break; 5032 default: 5033 return; 5034 } 5035 break; 5036 } 5037 case Stmt::ConditionalOperatorClass: { 5038 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 5039 if (const Expr *lhs = cond->getLHS()) 5040 CheckArrayAccess(lhs); 5041 if (const Expr *rhs = cond->getRHS()) 5042 CheckArrayAccess(rhs); 5043 return; 5044 } 5045 default: 5046 return; 5047 } 5048 } 5049} 5050 5051//===--- CHECK: Objective-C retain cycles ----------------------------------// 5052 5053namespace { 5054 struct RetainCycleOwner { 5055 RetainCycleOwner() : Variable(0), Indirect(false) {} 5056 VarDecl *Variable; 5057 SourceRange Range; 5058 SourceLocation Loc; 5059 bool Indirect; 5060 5061 void setLocsFrom(Expr *e) { 5062 Loc = e->getExprLoc(); 5063 Range = e->getSourceRange(); 5064 } 5065 }; 5066} 5067 5068/// Consider whether capturing the given variable can possibly lead to 5069/// a retain cycle. 5070static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 5071 // In ARC, it's captured strongly iff the variable has __strong 5072 // lifetime. In MRR, it's captured strongly if the variable is 5073 // __block and has an appropriate type. 5074 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 5075 return false; 5076 5077 owner.Variable = var; 5078 owner.setLocsFrom(ref); 5079 return true; 5080} 5081 5082static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) { 5083 while (true) { 5084 e = e->IgnoreParens(); 5085 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 5086 switch (cast->getCastKind()) { 5087 case CK_BitCast: 5088 case CK_LValueBitCast: 5089 case CK_LValueToRValue: 5090 case CK_ARCReclaimReturnedObject: 5091 e = cast->getSubExpr(); 5092 continue; 5093 5094 default: 5095 return false; 5096 } 5097 } 5098 5099 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 5100 ObjCIvarDecl *ivar = ref->getDecl(); 5101 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 5102 return false; 5103 5104 // Try to find a retain cycle in the base. 5105 if (!findRetainCycleOwner(S, ref->getBase(), owner)) 5106 return false; 5107 5108 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 5109 owner.Indirect = true; 5110 return true; 5111 } 5112 5113 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 5114 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 5115 if (!var) return false; 5116 return considerVariable(var, ref, owner); 5117 } 5118 5119 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 5120 if (member->isArrow()) return false; 5121 5122 // Don't count this as an indirect ownership. 5123 e = member->getBase(); 5124 continue; 5125 } 5126 5127 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 5128 // Only pay attention to pseudo-objects on property references. 5129 ObjCPropertyRefExpr *pre 5130 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 5131 ->IgnoreParens()); 5132 if (!pre) return false; 5133 if (pre->isImplicitProperty()) return false; 5134 ObjCPropertyDecl *property = pre->getExplicitProperty(); 5135 if (!property->isRetaining() && 5136 !(property->getPropertyIvarDecl() && 5137 property->getPropertyIvarDecl()->getType() 5138 .getObjCLifetime() == Qualifiers::OCL_Strong)) 5139 return false; 5140 5141 owner.Indirect = true; 5142 if (pre->isSuperReceiver()) { 5143 owner.Variable = S.getCurMethodDecl()->getSelfDecl(); 5144 if (!owner.Variable) 5145 return false; 5146 owner.Loc = pre->getLocation(); 5147 owner.Range = pre->getSourceRange(); 5148 return true; 5149 } 5150 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 5151 ->getSourceExpr()); 5152 continue; 5153 } 5154 5155 // Array ivars? 5156 5157 return false; 5158 } 5159} 5160 5161namespace { 5162 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 5163 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 5164 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 5165 Variable(variable), Capturer(0) {} 5166 5167 VarDecl *Variable; 5168 Expr *Capturer; 5169 5170 void VisitDeclRefExpr(DeclRefExpr *ref) { 5171 if (ref->getDecl() == Variable && !Capturer) 5172 Capturer = ref; 5173 } 5174 5175 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 5176 if (Capturer) return; 5177 Visit(ref->getBase()); 5178 if (Capturer && ref->isFreeIvar()) 5179 Capturer = ref; 5180 } 5181 5182 void VisitBlockExpr(BlockExpr *block) { 5183 // Look inside nested blocks 5184 if (block->getBlockDecl()->capturesVariable(Variable)) 5185 Visit(block->getBlockDecl()->getBody()); 5186 } 5187 }; 5188} 5189 5190/// Check whether the given argument is a block which captures a 5191/// variable. 5192static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 5193 assert(owner.Variable && owner.Loc.isValid()); 5194 5195 e = e->IgnoreParenCasts(); 5196 BlockExpr *block = dyn_cast<BlockExpr>(e); 5197 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 5198 return 0; 5199 5200 FindCaptureVisitor visitor(S.Context, owner.Variable); 5201 visitor.Visit(block->getBlockDecl()->getBody()); 5202 return visitor.Capturer; 5203} 5204 5205static void diagnoseRetainCycle(Sema &S, Expr *capturer, 5206 RetainCycleOwner &owner) { 5207 assert(capturer); 5208 assert(owner.Variable && owner.Loc.isValid()); 5209 5210 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 5211 << owner.Variable << capturer->getSourceRange(); 5212 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 5213 << owner.Indirect << owner.Range; 5214} 5215 5216/// Check for a keyword selector that starts with the word 'add' or 5217/// 'set'. 5218static bool isSetterLikeSelector(Selector sel) { 5219 if (sel.isUnarySelector()) return false; 5220 5221 StringRef str = sel.getNameForSlot(0); 5222 while (!str.empty() && str.front() == '_') str = str.substr(1); 5223 if (str.startswith("set")) 5224 str = str.substr(3); 5225 else if (str.startswith("add")) { 5226 // Specially whitelist 'addOperationWithBlock:'. 5227 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 5228 return false; 5229 str = str.substr(3); 5230 } 5231 else 5232 return false; 5233 5234 if (str.empty()) return true; 5235 return !islower(str.front()); 5236} 5237 5238/// Check a message send to see if it's likely to cause a retain cycle. 5239void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 5240 // Only check instance methods whose selector looks like a setter. 5241 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 5242 return; 5243 5244 // Try to find a variable that the receiver is strongly owned by. 5245 RetainCycleOwner owner; 5246 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 5247 if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner)) 5248 return; 5249 } else { 5250 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 5251 owner.Variable = getCurMethodDecl()->getSelfDecl(); 5252 owner.Loc = msg->getSuperLoc(); 5253 owner.Range = msg->getSuperLoc(); 5254 } 5255 5256 // Check whether the receiver is captured by any of the arguments. 5257 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 5258 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 5259 return diagnoseRetainCycle(*this, capturer, owner); 5260} 5261 5262/// Check a property assign to see if it's likely to cause a retain cycle. 5263void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 5264 RetainCycleOwner owner; 5265 if (!findRetainCycleOwner(*this, receiver, owner)) 5266 return; 5267 5268 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 5269 diagnoseRetainCycle(*this, capturer, owner); 5270} 5271 5272bool Sema::checkUnsafeAssigns(SourceLocation Loc, 5273 QualType LHS, Expr *RHS) { 5274 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 5275 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 5276 return false; 5277 // strip off any implicit cast added to get to the one arc-specific 5278 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 5279 if (cast->getCastKind() == CK_ARCConsumeObject) { 5280 Diag(Loc, diag::warn_arc_retained_assign) 5281 << (LT == Qualifiers::OCL_ExplicitNone) << 1 5282 << RHS->getSourceRange(); 5283 return true; 5284 } 5285 RHS = cast->getSubExpr(); 5286 } 5287 return false; 5288} 5289 5290void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 5291 Expr *LHS, Expr *RHS) { 5292 QualType LHSType; 5293 // PropertyRef on LHS type need be directly obtained from 5294 // its declaration as it has a PsuedoType. 5295 ObjCPropertyRefExpr *PRE 5296 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens()); 5297 if (PRE && !PRE->isImplicitProperty()) { 5298 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 5299 if (PD) 5300 LHSType = PD->getType(); 5301 } 5302 5303 if (LHSType.isNull()) 5304 LHSType = LHS->getType(); 5305 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 5306 return; 5307 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 5308 // FIXME. Check for other life times. 5309 if (LT != Qualifiers::OCL_None) 5310 return; 5311 5312 if (PRE) { 5313 if (PRE->isImplicitProperty()) 5314 return; 5315 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 5316 if (!PD) 5317 return; 5318 5319 unsigned Attributes = PD->getPropertyAttributes(); 5320 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) { 5321 // when 'assign' attribute was not explicitly specified 5322 // by user, ignore it and rely on property type itself 5323 // for lifetime info. 5324 unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten(); 5325 if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) && 5326 LHSType->isObjCRetainableType()) 5327 return; 5328 5329 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 5330 if (cast->getCastKind() == CK_ARCConsumeObject) { 5331 Diag(Loc, diag::warn_arc_retained_property_assign) 5332 << RHS->getSourceRange(); 5333 return; 5334 } 5335 RHS = cast->getSubExpr(); 5336 } 5337 } 5338 else if (Attributes & ObjCPropertyDecl::OBJC_PR_weak) { 5339 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 5340 if (cast->getCastKind() == CK_ARCConsumeObject) { 5341 Diag(Loc, diag::warn_arc_retained_assign) 5342 << 0 << 0<< RHS->getSourceRange(); 5343 return; 5344 } 5345 RHS = cast->getSubExpr(); 5346 } 5347 } 5348 } 5349} 5350 5351//===--- CHECK: Empty statement body (-Wempty-body) ---------------------===// 5352 5353namespace { 5354bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr, 5355 SourceLocation StmtLoc, 5356 const NullStmt *Body) { 5357 // Do not warn if the body is a macro that expands to nothing, e.g: 5358 // 5359 // #define CALL(x) 5360 // if (condition) 5361 // CALL(0); 5362 // 5363 if (Body->hasLeadingEmptyMacro()) 5364 return false; 5365 5366 // Get line numbers of statement and body. 5367 bool StmtLineInvalid; 5368 unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc, 5369 &StmtLineInvalid); 5370 if (StmtLineInvalid) 5371 return false; 5372 5373 bool BodyLineInvalid; 5374 unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(), 5375 &BodyLineInvalid); 5376 if (BodyLineInvalid) 5377 return false; 5378 5379 // Warn if null statement and body are on the same line. 5380 if (StmtLine != BodyLine) 5381 return false; 5382 5383 return true; 5384} 5385} // Unnamed namespace 5386 5387void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc, 5388 const Stmt *Body, 5389 unsigned DiagID) { 5390 // Since this is a syntactic check, don't emit diagnostic for template 5391 // instantiations, this just adds noise. 5392 if (CurrentInstantiationScope) 5393 return; 5394 5395 // The body should be a null statement. 5396 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 5397 if (!NBody) 5398 return; 5399 5400 // Do the usual checks. 5401 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 5402 return; 5403 5404 Diag(NBody->getSemiLoc(), DiagID); 5405 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 5406} 5407 5408void Sema::DiagnoseEmptyLoopBody(const Stmt *S, 5409 const Stmt *PossibleBody) { 5410 assert(!CurrentInstantiationScope); // Ensured by caller 5411 5412 SourceLocation StmtLoc; 5413 const Stmt *Body; 5414 unsigned DiagID; 5415 if (const ForStmt *FS = dyn_cast<ForStmt>(S)) { 5416 StmtLoc = FS->getRParenLoc(); 5417 Body = FS->getBody(); 5418 DiagID = diag::warn_empty_for_body; 5419 } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) { 5420 StmtLoc = WS->getCond()->getSourceRange().getEnd(); 5421 Body = WS->getBody(); 5422 DiagID = diag::warn_empty_while_body; 5423 } else 5424 return; // Neither `for' nor `while'. 5425 5426 // The body should be a null statement. 5427 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 5428 if (!NBody) 5429 return; 5430 5431 // Skip expensive checks if diagnostic is disabled. 5432 if (Diags.getDiagnosticLevel(DiagID, NBody->getSemiLoc()) == 5433 DiagnosticsEngine::Ignored) 5434 return; 5435 5436 // Do the usual checks. 5437 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 5438 return; 5439 5440 // `for(...);' and `while(...);' are popular idioms, so in order to keep 5441 // noise level low, emit diagnostics only if for/while is followed by a 5442 // CompoundStmt, e.g.: 5443 // for (int i = 0; i < n; i++); 5444 // { 5445 // a(i); 5446 // } 5447 // or if for/while is followed by a statement with more indentation 5448 // than for/while itself: 5449 // for (int i = 0; i < n; i++); 5450 // a(i); 5451 bool ProbableTypo = isa<CompoundStmt>(PossibleBody); 5452 if (!ProbableTypo) { 5453 bool BodyColInvalid; 5454 unsigned BodyCol = SourceMgr.getPresumedColumnNumber( 5455 PossibleBody->getLocStart(), 5456 &BodyColInvalid); 5457 if (BodyColInvalid) 5458 return; 5459 5460 bool StmtColInvalid; 5461 unsigned StmtCol = SourceMgr.getPresumedColumnNumber( 5462 S->getLocStart(), 5463 &StmtColInvalid); 5464 if (StmtColInvalid) 5465 return; 5466 5467 if (BodyCol > StmtCol) 5468 ProbableTypo = true; 5469 } 5470 5471 if (ProbableTypo) { 5472 Diag(NBody->getSemiLoc(), DiagID); 5473 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 5474 } 5475} 5476 5477//===--- Layout compatibility ----------------------------------------------// 5478 5479namespace { 5480 5481bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2); 5482 5483/// \brief Check if two enumeration types are layout-compatible. 5484bool isLayoutCompatible(ASTContext &C, EnumDecl *ED1, EnumDecl *ED2) { 5485 // C++11 [dcl.enum] p8: 5486 // Two enumeration types are layout-compatible if they have the same 5487 // underlying type. 5488 return ED1->isComplete() && ED2->isComplete() && 5489 C.hasSameType(ED1->getIntegerType(), ED2->getIntegerType()); 5490} 5491 5492/// \brief Check if two fields are layout-compatible. 5493bool isLayoutCompatible(ASTContext &C, FieldDecl *Field1, FieldDecl *Field2) { 5494 if (!isLayoutCompatible(C, Field1->getType(), Field2->getType())) 5495 return false; 5496 5497 if (Field1->isBitField() != Field2->isBitField()) 5498 return false; 5499 5500 if (Field1->isBitField()) { 5501 // Make sure that the bit-fields are the same length. 5502 unsigned Bits1 = Field1->getBitWidthValue(C); 5503 unsigned Bits2 = Field2->getBitWidthValue(C); 5504 5505 if (Bits1 != Bits2) 5506 return false; 5507 } 5508 5509 return true; 5510} 5511 5512/// \brief Check if two standard-layout structs are layout-compatible. 5513/// (C++11 [class.mem] p17) 5514bool isLayoutCompatibleStruct(ASTContext &C, 5515 RecordDecl *RD1, 5516 RecordDecl *RD2) { 5517 // If both records are C++ classes, check that base classes match. 5518 if (const CXXRecordDecl *D1CXX = dyn_cast<CXXRecordDecl>(RD1)) { 5519 // If one of records is a CXXRecordDecl we are in C++ mode, 5520 // thus the other one is a CXXRecordDecl, too. 5521 const CXXRecordDecl *D2CXX = cast<CXXRecordDecl>(RD2); 5522 // Check number of base classes. 5523 if (D1CXX->getNumBases() != D2CXX->getNumBases()) 5524 return false; 5525 5526 // Check the base classes. 5527 for (CXXRecordDecl::base_class_const_iterator 5528 Base1 = D1CXX->bases_begin(), 5529 BaseEnd1 = D1CXX->bases_end(), 5530 Base2 = D2CXX->bases_begin(); 5531 Base1 != BaseEnd1; 5532 ++Base1, ++Base2) { 5533 if (!isLayoutCompatible(C, Base1->getType(), Base2->getType())) 5534 return false; 5535 } 5536 } else if (const CXXRecordDecl *D2CXX = dyn_cast<CXXRecordDecl>(RD2)) { 5537 // If only RD2 is a C++ class, it should have zero base classes. 5538 if (D2CXX->getNumBases() > 0) 5539 return false; 5540 } 5541 5542 // Check the fields. 5543 RecordDecl::field_iterator Field2 = RD2->field_begin(), 5544 Field2End = RD2->field_end(), 5545 Field1 = RD1->field_begin(), 5546 Field1End = RD1->field_end(); 5547 for ( ; Field1 != Field1End && Field2 != Field2End; ++Field1, ++Field2) { 5548 if (!isLayoutCompatible(C, *Field1, *Field2)) 5549 return false; 5550 } 5551 if (Field1 != Field1End || Field2 != Field2End) 5552 return false; 5553 5554 return true; 5555} 5556 5557/// \brief Check if two standard-layout unions are layout-compatible. 5558/// (C++11 [class.mem] p18) 5559bool isLayoutCompatibleUnion(ASTContext &C, 5560 RecordDecl *RD1, 5561 RecordDecl *RD2) { 5562 llvm::SmallPtrSet<FieldDecl *, 8> UnmatchedFields; 5563 for (RecordDecl::field_iterator Field2 = RD2->field_begin(), 5564 Field2End = RD2->field_end(); 5565 Field2 != Field2End; ++Field2) { 5566 UnmatchedFields.insert(*Field2); 5567 } 5568 5569 for (RecordDecl::field_iterator Field1 = RD1->field_begin(), 5570 Field1End = RD1->field_end(); 5571 Field1 != Field1End; ++Field1) { 5572 llvm::SmallPtrSet<FieldDecl *, 8>::iterator 5573 I = UnmatchedFields.begin(), 5574 E = UnmatchedFields.end(); 5575 5576 for ( ; I != E; ++I) { 5577 if (isLayoutCompatible(C, *Field1, *I)) { 5578 bool Result = UnmatchedFields.erase(*I); 5579 (void) Result; 5580 assert(Result); 5581 break; 5582 } 5583 } 5584 if (I == E) 5585 return false; 5586 } 5587 5588 return UnmatchedFields.empty(); 5589} 5590 5591bool isLayoutCompatible(ASTContext &C, RecordDecl *RD1, RecordDecl *RD2) { 5592 if (RD1->isUnion() != RD2->isUnion()) 5593 return false; 5594 5595 if (RD1->isUnion()) 5596 return isLayoutCompatibleUnion(C, RD1, RD2); 5597 else 5598 return isLayoutCompatibleStruct(C, RD1, RD2); 5599} 5600 5601/// \brief Check if two types are layout-compatible in C++11 sense. 5602bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2) { 5603 if (T1.isNull() || T2.isNull()) 5604 return false; 5605 5606 // C++11 [basic.types] p11: 5607 // If two types T1 and T2 are the same type, then T1 and T2 are 5608 // layout-compatible types. 5609 if (C.hasSameType(T1, T2)) 5610 return true; 5611 5612 T1 = T1.getCanonicalType().getUnqualifiedType(); 5613 T2 = T2.getCanonicalType().getUnqualifiedType(); 5614 5615 const Type::TypeClass TC1 = T1->getTypeClass(); 5616 const Type::TypeClass TC2 = T2->getTypeClass(); 5617 5618 if (TC1 != TC2) 5619 return false; 5620 5621 if (TC1 == Type::Enum) { 5622 return isLayoutCompatible(C, 5623 cast<EnumType>(T1)->getDecl(), 5624 cast<EnumType>(T2)->getDecl()); 5625 } else if (TC1 == Type::Record) { 5626 if (!T1->isStandardLayoutType() || !T2->isStandardLayoutType()) 5627 return false; 5628 5629 return isLayoutCompatible(C, 5630 cast<RecordType>(T1)->getDecl(), 5631 cast<RecordType>(T2)->getDecl()); 5632 } 5633 5634 return false; 5635} 5636} 5637 5638//===--- CHECK: pointer_with_type_tag attribute: datatypes should match ----// 5639 5640namespace { 5641/// \brief Given a type tag expression find the type tag itself. 5642/// 5643/// \param TypeExpr Type tag expression, as it appears in user's code. 5644/// 5645/// \param VD Declaration of an identifier that appears in a type tag. 5646/// 5647/// \param MagicValue Type tag magic value. 5648bool FindTypeTagExpr(const Expr *TypeExpr, const ASTContext &Ctx, 5649 const ValueDecl **VD, uint64_t *MagicValue) { 5650 while(true) { 5651 if (!TypeExpr) 5652 return false; 5653 5654 TypeExpr = TypeExpr->IgnoreParenImpCasts()->IgnoreParenCasts(); 5655 5656 switch (TypeExpr->getStmtClass()) { 5657 case Stmt::UnaryOperatorClass: { 5658 const UnaryOperator *UO = cast<UnaryOperator>(TypeExpr); 5659 if (UO->getOpcode() == UO_AddrOf || UO->getOpcode() == UO_Deref) { 5660 TypeExpr = UO->getSubExpr(); 5661 continue; 5662 } 5663 return false; 5664 } 5665 5666 case Stmt::DeclRefExprClass: { 5667 const DeclRefExpr *DRE = cast<DeclRefExpr>(TypeExpr); 5668 *VD = DRE->getDecl(); 5669 return true; 5670 } 5671 5672 case Stmt::IntegerLiteralClass: { 5673 const IntegerLiteral *IL = cast<IntegerLiteral>(TypeExpr); 5674 llvm::APInt MagicValueAPInt = IL->getValue(); 5675 if (MagicValueAPInt.getActiveBits() <= 64) { 5676 *MagicValue = MagicValueAPInt.getZExtValue(); 5677 return true; 5678 } else 5679 return false; 5680 } 5681 5682 case Stmt::BinaryConditionalOperatorClass: 5683 case Stmt::ConditionalOperatorClass: { 5684 const AbstractConditionalOperator *ACO = 5685 cast<AbstractConditionalOperator>(TypeExpr); 5686 bool Result; 5687 if (ACO->getCond()->EvaluateAsBooleanCondition(Result, Ctx)) { 5688 if (Result) 5689 TypeExpr = ACO->getTrueExpr(); 5690 else 5691 TypeExpr = ACO->getFalseExpr(); 5692 continue; 5693 } 5694 return false; 5695 } 5696 5697 case Stmt::BinaryOperatorClass: { 5698 const BinaryOperator *BO = cast<BinaryOperator>(TypeExpr); 5699 if (BO->getOpcode() == BO_Comma) { 5700 TypeExpr = BO->getRHS(); 5701 continue; 5702 } 5703 return false; 5704 } 5705 5706 default: 5707 return false; 5708 } 5709 } 5710} 5711 5712/// \brief Retrieve the C type corresponding to type tag TypeExpr. 5713/// 5714/// \param TypeExpr Expression that specifies a type tag. 5715/// 5716/// \param MagicValues Registered magic values. 5717/// 5718/// \param FoundWrongKind Set to true if a type tag was found, but of a wrong 5719/// kind. 5720/// 5721/// \param TypeInfo Information about the corresponding C type. 5722/// 5723/// \returns true if the corresponding C type was found. 5724bool GetMatchingCType( 5725 const IdentifierInfo *ArgumentKind, 5726 const Expr *TypeExpr, const ASTContext &Ctx, 5727 const llvm::DenseMap<Sema::TypeTagMagicValue, 5728 Sema::TypeTagData> *MagicValues, 5729 bool &FoundWrongKind, 5730 Sema::TypeTagData &TypeInfo) { 5731 FoundWrongKind = false; 5732 5733 // Variable declaration that has type_tag_for_datatype attribute. 5734 const ValueDecl *VD = NULL; 5735 5736 uint64_t MagicValue; 5737 5738 if (!FindTypeTagExpr(TypeExpr, Ctx, &VD, &MagicValue)) 5739 return false; 5740 5741 if (VD) { 5742 for (specific_attr_iterator<TypeTagForDatatypeAttr> 5743 I = VD->specific_attr_begin<TypeTagForDatatypeAttr>(), 5744 E = VD->specific_attr_end<TypeTagForDatatypeAttr>(); 5745 I != E; ++I) { 5746 if (I->getArgumentKind() != ArgumentKind) { 5747 FoundWrongKind = true; 5748 return false; 5749 } 5750 TypeInfo.Type = I->getMatchingCType(); 5751 TypeInfo.LayoutCompatible = I->getLayoutCompatible(); 5752 TypeInfo.MustBeNull = I->getMustBeNull(); 5753 return true; 5754 } 5755 return false; 5756 } 5757 5758 if (!MagicValues) 5759 return false; 5760 5761 llvm::DenseMap<Sema::TypeTagMagicValue, 5762 Sema::TypeTagData>::const_iterator I = 5763 MagicValues->find(std::make_pair(ArgumentKind, MagicValue)); 5764 if (I == MagicValues->end()) 5765 return false; 5766 5767 TypeInfo = I->second; 5768 return true; 5769} 5770} // unnamed namespace 5771 5772void Sema::RegisterTypeTagForDatatype(const IdentifierInfo *ArgumentKind, 5773 uint64_t MagicValue, QualType Type, 5774 bool LayoutCompatible, 5775 bool MustBeNull) { 5776 if (!TypeTagForDatatypeMagicValues) 5777 TypeTagForDatatypeMagicValues.reset( 5778 new llvm::DenseMap<TypeTagMagicValue, TypeTagData>); 5779 5780 TypeTagMagicValue Magic(ArgumentKind, MagicValue); 5781 (*TypeTagForDatatypeMagicValues)[Magic] = 5782 TypeTagData(Type, LayoutCompatible, MustBeNull); 5783} 5784 5785namespace { 5786bool IsSameCharType(QualType T1, QualType T2) { 5787 const BuiltinType *BT1 = T1->getAs<BuiltinType>(); 5788 if (!BT1) 5789 return false; 5790 5791 const BuiltinType *BT2 = T2->getAs<BuiltinType>(); 5792 if (!BT2) 5793 return false; 5794 5795 BuiltinType::Kind T1Kind = BT1->getKind(); 5796 BuiltinType::Kind T2Kind = BT2->getKind(); 5797 5798 return (T1Kind == BuiltinType::SChar && T2Kind == BuiltinType::Char_S) || 5799 (T1Kind == BuiltinType::UChar && T2Kind == BuiltinType::Char_U) || 5800 (T1Kind == BuiltinType::Char_U && T2Kind == BuiltinType::UChar) || 5801 (T1Kind == BuiltinType::Char_S && T2Kind == BuiltinType::SChar); 5802} 5803} // unnamed namespace 5804 5805void Sema::CheckArgumentWithTypeTag(const ArgumentWithTypeTagAttr *Attr, 5806 const Expr * const *ExprArgs) { 5807 const IdentifierInfo *ArgumentKind = Attr->getArgumentKind(); 5808 bool IsPointerAttr = Attr->getIsPointer(); 5809 5810 const Expr *TypeTagExpr = ExprArgs[Attr->getTypeTagIdx()]; 5811 bool FoundWrongKind; 5812 TypeTagData TypeInfo; 5813 if (!GetMatchingCType(ArgumentKind, TypeTagExpr, Context, 5814 TypeTagForDatatypeMagicValues.get(), 5815 FoundWrongKind, TypeInfo)) { 5816 if (FoundWrongKind) 5817 Diag(TypeTagExpr->getExprLoc(), 5818 diag::warn_type_tag_for_datatype_wrong_kind) 5819 << TypeTagExpr->getSourceRange(); 5820 return; 5821 } 5822 5823 const Expr *ArgumentExpr = ExprArgs[Attr->getArgumentIdx()]; 5824 if (IsPointerAttr) { 5825 // Skip implicit cast of pointer to `void *' (as a function argument). 5826 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(ArgumentExpr)) 5827 if (ICE->getType()->isVoidPointerType()) 5828 ArgumentExpr = ICE->getSubExpr(); 5829 } 5830 QualType ArgumentType = ArgumentExpr->getType(); 5831 5832 // Passing a `void*' pointer shouldn't trigger a warning. 5833 if (IsPointerAttr && ArgumentType->isVoidPointerType()) 5834 return; 5835 5836 if (TypeInfo.MustBeNull) { 5837 // Type tag with matching void type requires a null pointer. 5838 if (!ArgumentExpr->isNullPointerConstant(Context, 5839 Expr::NPC_ValueDependentIsNotNull)) { 5840 Diag(ArgumentExpr->getExprLoc(), 5841 diag::warn_type_safety_null_pointer_required) 5842 << ArgumentKind->getName() 5843 << ArgumentExpr->getSourceRange() 5844 << TypeTagExpr->getSourceRange(); 5845 } 5846 return; 5847 } 5848 5849 QualType RequiredType = TypeInfo.Type; 5850 if (IsPointerAttr) 5851 RequiredType = Context.getPointerType(RequiredType); 5852 5853 bool mismatch = false; 5854 if (!TypeInfo.LayoutCompatible) { 5855 mismatch = !Context.hasSameType(ArgumentType, RequiredType); 5856 5857 // C++11 [basic.fundamental] p1: 5858 // Plain char, signed char, and unsigned char are three distinct types. 5859 // 5860 // But we treat plain `char' as equivalent to `signed char' or `unsigned 5861 // char' depending on the current char signedness mode. 5862 if (mismatch) 5863 if ((IsPointerAttr && IsSameCharType(ArgumentType->getPointeeType(), 5864 RequiredType->getPointeeType())) || 5865 (!IsPointerAttr && IsSameCharType(ArgumentType, RequiredType))) 5866 mismatch = false; 5867 } else 5868 if (IsPointerAttr) 5869 mismatch = !isLayoutCompatible(Context, 5870 ArgumentType->getPointeeType(), 5871 RequiredType->getPointeeType()); 5872 else 5873 mismatch = !isLayoutCompatible(Context, ArgumentType, RequiredType); 5874 5875 if (mismatch) 5876 Diag(ArgumentExpr->getExprLoc(), diag::warn_type_safety_type_mismatch) 5877 << ArgumentType << ArgumentKind->getName() 5878 << TypeInfo.LayoutCompatible << RequiredType 5879 << ArgumentExpr->getSourceRange() 5880 << TypeTagExpr->getSourceRange(); 5881} 5882 5883