SemaChecking.cpp revision 243830
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 case Mips::BI__builtin_mips_append: i = 2; l = 0; u = 31; break; 441 case Mips::BI__builtin_mips_balign: i = 2; l = 0; u = 3; break; 442 case Mips::BI__builtin_mips_precr_sra_ph_w: i = 2; l = 0; u = 31; break; 443 case Mips::BI__builtin_mips_precr_sra_r_ph_w: i = 2; l = 0; u = 31; break; 444 case Mips::BI__builtin_mips_prepend: i = 2; l = 0; u = 31; break; 445 }; 446 447 // We can't check the value of a dependent argument. 448 if (TheCall->getArg(i)->isTypeDependent() || 449 TheCall->getArg(i)->isValueDependent()) 450 return false; 451 452 // Check that the immediate argument is actually a constant. 453 llvm::APSInt Result; 454 if (SemaBuiltinConstantArg(TheCall, i, Result)) 455 return true; 456 457 // Range check against the upper/lower values for this instruction. 458 unsigned Val = Result.getZExtValue(); 459 if (Val < l || Val > u) 460 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 461 << l << u << TheCall->getArg(i)->getSourceRange(); 462 463 return false; 464} 465 466/// Given a FunctionDecl's FormatAttr, attempts to populate the FomatStringInfo 467/// parameter with the FormatAttr's correct format_idx and firstDataArg. 468/// Returns true when the format fits the function and the FormatStringInfo has 469/// been populated. 470bool Sema::getFormatStringInfo(const FormatAttr *Format, bool IsCXXMember, 471 FormatStringInfo *FSI) { 472 FSI->HasVAListArg = Format->getFirstArg() == 0; 473 FSI->FormatIdx = Format->getFormatIdx() - 1; 474 FSI->FirstDataArg = FSI->HasVAListArg ? 0 : Format->getFirstArg() - 1; 475 476 // The way the format attribute works in GCC, the implicit this argument 477 // of member functions is counted. However, it doesn't appear in our own 478 // lists, so decrement format_idx in that case. 479 if (IsCXXMember) { 480 if(FSI->FormatIdx == 0) 481 return false; 482 --FSI->FormatIdx; 483 if (FSI->FirstDataArg != 0) 484 --FSI->FirstDataArg; 485 } 486 return true; 487} 488 489/// Handles the checks for format strings, non-POD arguments to vararg 490/// functions, and NULL arguments passed to non-NULL parameters. 491void Sema::checkCall(NamedDecl *FDecl, Expr **Args, 492 unsigned NumArgs, 493 unsigned NumProtoArgs, 494 bool IsMemberFunction, 495 SourceLocation Loc, 496 SourceRange Range, 497 VariadicCallType CallType) { 498 if (CurContext->isDependentContext()) 499 return; 500 501 // Printf and scanf checking. 502 bool HandledFormatString = false; 503 for (specific_attr_iterator<FormatAttr> 504 I = FDecl->specific_attr_begin<FormatAttr>(), 505 E = FDecl->specific_attr_end<FormatAttr>(); I != E ; ++I) 506 if (CheckFormatArguments(*I, Args, NumArgs, IsMemberFunction, CallType, 507 Loc, Range)) 508 HandledFormatString = true; 509 510 // Refuse POD arguments that weren't caught by the format string 511 // checks above. 512 if (!HandledFormatString && CallType != VariadicDoesNotApply) 513 for (unsigned ArgIdx = NumProtoArgs; ArgIdx < NumArgs; ++ArgIdx) { 514 // Args[ArgIdx] can be null in malformed code. 515 if (Expr *Arg = Args[ArgIdx]) 516 variadicArgumentPODCheck(Arg, CallType); 517 } 518 519 for (specific_attr_iterator<NonNullAttr> 520 I = FDecl->specific_attr_begin<NonNullAttr>(), 521 E = FDecl->specific_attr_end<NonNullAttr>(); I != E; ++I) 522 CheckNonNullArguments(*I, Args, Loc); 523 524 // Type safety checking. 525 for (specific_attr_iterator<ArgumentWithTypeTagAttr> 526 i = FDecl->specific_attr_begin<ArgumentWithTypeTagAttr>(), 527 e = FDecl->specific_attr_end<ArgumentWithTypeTagAttr>(); i != e; ++i) { 528 CheckArgumentWithTypeTag(*i, Args); 529 } 530} 531 532/// CheckConstructorCall - Check a constructor call for correctness and safety 533/// properties not enforced by the C type system. 534void Sema::CheckConstructorCall(FunctionDecl *FDecl, Expr **Args, 535 unsigned NumArgs, 536 const FunctionProtoType *Proto, 537 SourceLocation Loc) { 538 VariadicCallType CallType = 539 Proto->isVariadic() ? VariadicConstructor : VariadicDoesNotApply; 540 checkCall(FDecl, Args, NumArgs, Proto->getNumArgs(), 541 /*IsMemberFunction=*/true, Loc, SourceRange(), CallType); 542} 543 544/// CheckFunctionCall - Check a direct function call for various correctness 545/// and safety properties not strictly enforced by the C type system. 546bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall, 547 const FunctionProtoType *Proto) { 548 bool IsMemberOperatorCall = isa<CXXOperatorCallExpr>(TheCall) && 549 isa<CXXMethodDecl>(FDecl); 550 bool IsMemberFunction = isa<CXXMemberCallExpr>(TheCall) || 551 IsMemberOperatorCall; 552 VariadicCallType CallType = getVariadicCallType(FDecl, Proto, 553 TheCall->getCallee()); 554 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 555 Expr** Args = TheCall->getArgs(); 556 unsigned NumArgs = TheCall->getNumArgs(); 557 if (IsMemberOperatorCall) { 558 // If this is a call to a member operator, hide the first argument 559 // from checkCall. 560 // FIXME: Our choice of AST representation here is less than ideal. 561 ++Args; 562 --NumArgs; 563 } 564 checkCall(FDecl, Args, NumArgs, NumProtoArgs, 565 IsMemberFunction, TheCall->getRParenLoc(), 566 TheCall->getCallee()->getSourceRange(), CallType); 567 568 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 569 // None of the checks below are needed for functions that don't have 570 // simple names (e.g., C++ conversion functions). 571 if (!FnInfo) 572 return false; 573 574 unsigned CMId = FDecl->getMemoryFunctionKind(); 575 if (CMId == 0) 576 return false; 577 578 // Handle memory setting and copying functions. 579 if (CMId == Builtin::BIstrlcpy || CMId == Builtin::BIstrlcat) 580 CheckStrlcpycatArguments(TheCall, FnInfo); 581 else if (CMId == Builtin::BIstrncat) 582 CheckStrncatArguments(TheCall, FnInfo); 583 else 584 CheckMemaccessArguments(TheCall, CMId, FnInfo); 585 586 return false; 587} 588 589bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac, 590 Expr **Args, unsigned NumArgs) { 591 VariadicCallType CallType = 592 Method->isVariadic() ? VariadicMethod : VariadicDoesNotApply; 593 594 checkCall(Method, Args, NumArgs, Method->param_size(), 595 /*IsMemberFunction=*/false, 596 lbrac, Method->getSourceRange(), CallType); 597 598 return false; 599} 600 601bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall, 602 const FunctionProtoType *Proto) { 603 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 604 if (!V) 605 return false; 606 607 QualType Ty = V->getType(); 608 if (!Ty->isBlockPointerType()) 609 return false; 610 611 VariadicCallType CallType = 612 Proto && Proto->isVariadic() ? VariadicBlock : VariadicDoesNotApply ; 613 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 614 615 checkCall(NDecl, TheCall->getArgs(), TheCall->getNumArgs(), 616 NumProtoArgs, /*IsMemberFunction=*/false, 617 TheCall->getRParenLoc(), 618 TheCall->getCallee()->getSourceRange(), CallType); 619 620 return false; 621} 622 623ExprResult Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, 624 AtomicExpr::AtomicOp Op) { 625 CallExpr *TheCall = cast<CallExpr>(TheCallResult.get()); 626 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 627 628 // All these operations take one of the following forms: 629 enum { 630 // C __c11_atomic_init(A *, C) 631 Init, 632 // C __c11_atomic_load(A *, int) 633 Load, 634 // void __atomic_load(A *, CP, int) 635 Copy, 636 // C __c11_atomic_add(A *, M, int) 637 Arithmetic, 638 // C __atomic_exchange_n(A *, CP, int) 639 Xchg, 640 // void __atomic_exchange(A *, C *, CP, int) 641 GNUXchg, 642 // bool __c11_atomic_compare_exchange_strong(A *, C *, CP, int, int) 643 C11CmpXchg, 644 // bool __atomic_compare_exchange(A *, C *, CP, bool, int, int) 645 GNUCmpXchg 646 } Form = Init; 647 const unsigned NumArgs[] = { 2, 2, 3, 3, 3, 4, 5, 6 }; 648 const unsigned NumVals[] = { 1, 0, 1, 1, 1, 2, 2, 3 }; 649 // where: 650 // C is an appropriate type, 651 // A is volatile _Atomic(C) for __c11 builtins and is C for GNU builtins, 652 // CP is C for __c11 builtins and GNU _n builtins and is C * otherwise, 653 // M is C if C is an integer, and ptrdiff_t if C is a pointer, and 654 // the int parameters are for orderings. 655 656 assert(AtomicExpr::AO__c11_atomic_init == 0 && 657 AtomicExpr::AO__c11_atomic_fetch_xor + 1 == AtomicExpr::AO__atomic_load 658 && "need to update code for modified C11 atomics"); 659 bool IsC11 = Op >= AtomicExpr::AO__c11_atomic_init && 660 Op <= AtomicExpr::AO__c11_atomic_fetch_xor; 661 bool IsN = Op == AtomicExpr::AO__atomic_load_n || 662 Op == AtomicExpr::AO__atomic_store_n || 663 Op == AtomicExpr::AO__atomic_exchange_n || 664 Op == AtomicExpr::AO__atomic_compare_exchange_n; 665 bool IsAddSub = false; 666 667 switch (Op) { 668 case AtomicExpr::AO__c11_atomic_init: 669 Form = Init; 670 break; 671 672 case AtomicExpr::AO__c11_atomic_load: 673 case AtomicExpr::AO__atomic_load_n: 674 Form = Load; 675 break; 676 677 case AtomicExpr::AO__c11_atomic_store: 678 case AtomicExpr::AO__atomic_load: 679 case AtomicExpr::AO__atomic_store: 680 case AtomicExpr::AO__atomic_store_n: 681 Form = Copy; 682 break; 683 684 case AtomicExpr::AO__c11_atomic_fetch_add: 685 case AtomicExpr::AO__c11_atomic_fetch_sub: 686 case AtomicExpr::AO__atomic_fetch_add: 687 case AtomicExpr::AO__atomic_fetch_sub: 688 case AtomicExpr::AO__atomic_add_fetch: 689 case AtomicExpr::AO__atomic_sub_fetch: 690 IsAddSub = true; 691 // Fall through. 692 case AtomicExpr::AO__c11_atomic_fetch_and: 693 case AtomicExpr::AO__c11_atomic_fetch_or: 694 case AtomicExpr::AO__c11_atomic_fetch_xor: 695 case AtomicExpr::AO__atomic_fetch_and: 696 case AtomicExpr::AO__atomic_fetch_or: 697 case AtomicExpr::AO__atomic_fetch_xor: 698 case AtomicExpr::AO__atomic_fetch_nand: 699 case AtomicExpr::AO__atomic_and_fetch: 700 case AtomicExpr::AO__atomic_or_fetch: 701 case AtomicExpr::AO__atomic_xor_fetch: 702 case AtomicExpr::AO__atomic_nand_fetch: 703 Form = Arithmetic; 704 break; 705 706 case AtomicExpr::AO__c11_atomic_exchange: 707 case AtomicExpr::AO__atomic_exchange_n: 708 Form = Xchg; 709 break; 710 711 case AtomicExpr::AO__atomic_exchange: 712 Form = GNUXchg; 713 break; 714 715 case AtomicExpr::AO__c11_atomic_compare_exchange_strong: 716 case AtomicExpr::AO__c11_atomic_compare_exchange_weak: 717 Form = C11CmpXchg; 718 break; 719 720 case AtomicExpr::AO__atomic_compare_exchange: 721 case AtomicExpr::AO__atomic_compare_exchange_n: 722 Form = GNUCmpXchg; 723 break; 724 } 725 726 // Check we have the right number of arguments. 727 if (TheCall->getNumArgs() < NumArgs[Form]) { 728 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 729 << 0 << NumArgs[Form] << TheCall->getNumArgs() 730 << TheCall->getCallee()->getSourceRange(); 731 return ExprError(); 732 } else if (TheCall->getNumArgs() > NumArgs[Form]) { 733 Diag(TheCall->getArg(NumArgs[Form])->getLocStart(), 734 diag::err_typecheck_call_too_many_args) 735 << 0 << NumArgs[Form] << TheCall->getNumArgs() 736 << TheCall->getCallee()->getSourceRange(); 737 return ExprError(); 738 } 739 740 // Inspect the first argument of the atomic operation. 741 Expr *Ptr = TheCall->getArg(0); 742 Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get(); 743 const PointerType *pointerType = Ptr->getType()->getAs<PointerType>(); 744 if (!pointerType) { 745 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 746 << Ptr->getType() << Ptr->getSourceRange(); 747 return ExprError(); 748 } 749 750 // For a __c11 builtin, this should be a pointer to an _Atomic type. 751 QualType AtomTy = pointerType->getPointeeType(); // 'A' 752 QualType ValType = AtomTy; // 'C' 753 if (IsC11) { 754 if (!AtomTy->isAtomicType()) { 755 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 756 << Ptr->getType() << Ptr->getSourceRange(); 757 return ExprError(); 758 } 759 if (AtomTy.isConstQualified()) { 760 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_non_const_atomic) 761 << Ptr->getType() << Ptr->getSourceRange(); 762 return ExprError(); 763 } 764 ValType = AtomTy->getAs<AtomicType>()->getValueType(); 765 } 766 767 // For an arithmetic operation, the implied arithmetic must be well-formed. 768 if (Form == Arithmetic) { 769 // gcc does not enforce these rules for GNU atomics, but we do so for sanity. 770 if (IsAddSub && !ValType->isIntegerType() && !ValType->isPointerType()) { 771 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 772 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 773 return ExprError(); 774 } 775 if (!IsAddSub && !ValType->isIntegerType()) { 776 Diag(DRE->getLocStart(), diag::err_atomic_op_bitwise_needs_atomic_int) 777 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 778 return ExprError(); 779 } 780 } else if (IsN && !ValType->isIntegerType() && !ValType->isPointerType()) { 781 // For __atomic_*_n operations, the value type must be a scalar integral or 782 // pointer type which is 1, 2, 4, 8 or 16 bytes in length. 783 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 784 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 785 return ExprError(); 786 } 787 788 if (!IsC11 && !AtomTy.isTriviallyCopyableType(Context)) { 789 // For GNU atomics, require a trivially-copyable type. This is not part of 790 // the GNU atomics specification, but we enforce it for sanity. 791 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_trivial_copy) 792 << Ptr->getType() << Ptr->getSourceRange(); 793 return ExprError(); 794 } 795 796 // FIXME: For any builtin other than a load, the ValType must not be 797 // const-qualified. 798 799 switch (ValType.getObjCLifetime()) { 800 case Qualifiers::OCL_None: 801 case Qualifiers::OCL_ExplicitNone: 802 // okay 803 break; 804 805 case Qualifiers::OCL_Weak: 806 case Qualifiers::OCL_Strong: 807 case Qualifiers::OCL_Autoreleasing: 808 // FIXME: Can this happen? By this point, ValType should be known 809 // to be trivially copyable. 810 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 811 << ValType << Ptr->getSourceRange(); 812 return ExprError(); 813 } 814 815 QualType ResultType = ValType; 816 if (Form == Copy || Form == GNUXchg || Form == Init) 817 ResultType = Context.VoidTy; 818 else if (Form == C11CmpXchg || Form == GNUCmpXchg) 819 ResultType = Context.BoolTy; 820 821 // The type of a parameter passed 'by value'. In the GNU atomics, such 822 // arguments are actually passed as pointers. 823 QualType ByValType = ValType; // 'CP' 824 if (!IsC11 && !IsN) 825 ByValType = Ptr->getType(); 826 827 // The first argument --- the pointer --- has a fixed type; we 828 // deduce the types of the rest of the arguments accordingly. Walk 829 // the remaining arguments, converting them to the deduced value type. 830 for (unsigned i = 1; i != NumArgs[Form]; ++i) { 831 QualType Ty; 832 if (i < NumVals[Form] + 1) { 833 switch (i) { 834 case 1: 835 // The second argument is the non-atomic operand. For arithmetic, this 836 // is always passed by value, and for a compare_exchange it is always 837 // passed by address. For the rest, GNU uses by-address and C11 uses 838 // by-value. 839 assert(Form != Load); 840 if (Form == Init || (Form == Arithmetic && ValType->isIntegerType())) 841 Ty = ValType; 842 else if (Form == Copy || Form == Xchg) 843 Ty = ByValType; 844 else if (Form == Arithmetic) 845 Ty = Context.getPointerDiffType(); 846 else 847 Ty = Context.getPointerType(ValType.getUnqualifiedType()); 848 break; 849 case 2: 850 // The third argument to compare_exchange / GNU exchange is a 851 // (pointer to a) desired value. 852 Ty = ByValType; 853 break; 854 case 3: 855 // The fourth argument to GNU compare_exchange is a 'weak' flag. 856 Ty = Context.BoolTy; 857 break; 858 } 859 } else { 860 // The order(s) are always converted to int. 861 Ty = Context.IntTy; 862 } 863 864 InitializedEntity Entity = 865 InitializedEntity::InitializeParameter(Context, Ty, false); 866 ExprResult Arg = TheCall->getArg(i); 867 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 868 if (Arg.isInvalid()) 869 return true; 870 TheCall->setArg(i, Arg.get()); 871 } 872 873 // Permute the arguments into a 'consistent' order. 874 SmallVector<Expr*, 5> SubExprs; 875 SubExprs.push_back(Ptr); 876 switch (Form) { 877 case Init: 878 // Note, AtomicExpr::getVal1() has a special case for this atomic. 879 SubExprs.push_back(TheCall->getArg(1)); // Val1 880 break; 881 case Load: 882 SubExprs.push_back(TheCall->getArg(1)); // Order 883 break; 884 case Copy: 885 case Arithmetic: 886 case Xchg: 887 SubExprs.push_back(TheCall->getArg(2)); // Order 888 SubExprs.push_back(TheCall->getArg(1)); // Val1 889 break; 890 case GNUXchg: 891 // Note, AtomicExpr::getVal2() has a special case for this atomic. 892 SubExprs.push_back(TheCall->getArg(3)); // Order 893 SubExprs.push_back(TheCall->getArg(1)); // Val1 894 SubExprs.push_back(TheCall->getArg(2)); // Val2 895 break; 896 case C11CmpXchg: 897 SubExprs.push_back(TheCall->getArg(3)); // Order 898 SubExprs.push_back(TheCall->getArg(1)); // Val1 899 SubExprs.push_back(TheCall->getArg(4)); // OrderFail 900 SubExprs.push_back(TheCall->getArg(2)); // Val2 901 break; 902 case GNUCmpXchg: 903 SubExprs.push_back(TheCall->getArg(4)); // Order 904 SubExprs.push_back(TheCall->getArg(1)); // Val1 905 SubExprs.push_back(TheCall->getArg(5)); // OrderFail 906 SubExprs.push_back(TheCall->getArg(2)); // Val2 907 SubExprs.push_back(TheCall->getArg(3)); // Weak 908 break; 909 } 910 911 return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 912 SubExprs, ResultType, Op, 913 TheCall->getRParenLoc())); 914} 915 916 917/// checkBuiltinArgument - Given a call to a builtin function, perform 918/// normal type-checking on the given argument, updating the call in 919/// place. This is useful when a builtin function requires custom 920/// type-checking for some of its arguments but not necessarily all of 921/// them. 922/// 923/// Returns true on error. 924static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 925 FunctionDecl *Fn = E->getDirectCallee(); 926 assert(Fn && "builtin call without direct callee!"); 927 928 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 929 InitializedEntity Entity = 930 InitializedEntity::InitializeParameter(S.Context, Param); 931 932 ExprResult Arg = E->getArg(0); 933 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 934 if (Arg.isInvalid()) 935 return true; 936 937 E->setArg(ArgIndex, Arg.take()); 938 return false; 939} 940 941/// SemaBuiltinAtomicOverloaded - We have a call to a function like 942/// __sync_fetch_and_add, which is an overloaded function based on the pointer 943/// type of its first argument. The main ActOnCallExpr routines have already 944/// promoted the types of arguments because all of these calls are prototyped as 945/// void(...). 946/// 947/// This function goes through and does final semantic checking for these 948/// builtins, 949ExprResult 950Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 951 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 952 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 953 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 954 955 // Ensure that we have at least one argument to do type inference from. 956 if (TheCall->getNumArgs() < 1) { 957 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 958 << 0 << 1 << TheCall->getNumArgs() 959 << TheCall->getCallee()->getSourceRange(); 960 return ExprError(); 961 } 962 963 // Inspect the first argument of the atomic builtin. This should always be 964 // a pointer type, whose element is an integral scalar or pointer type. 965 // Because it is a pointer type, we don't have to worry about any implicit 966 // casts here. 967 // FIXME: We don't allow floating point scalars as input. 968 Expr *FirstArg = TheCall->getArg(0); 969 ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg); 970 if (FirstArgResult.isInvalid()) 971 return ExprError(); 972 FirstArg = FirstArgResult.take(); 973 TheCall->setArg(0, FirstArg); 974 975 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 976 if (!pointerType) { 977 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 978 << FirstArg->getType() << FirstArg->getSourceRange(); 979 return ExprError(); 980 } 981 982 QualType ValType = pointerType->getPointeeType(); 983 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 984 !ValType->isBlockPointerType()) { 985 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 986 << FirstArg->getType() << FirstArg->getSourceRange(); 987 return ExprError(); 988 } 989 990 switch (ValType.getObjCLifetime()) { 991 case Qualifiers::OCL_None: 992 case Qualifiers::OCL_ExplicitNone: 993 // okay 994 break; 995 996 case Qualifiers::OCL_Weak: 997 case Qualifiers::OCL_Strong: 998 case Qualifiers::OCL_Autoreleasing: 999 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 1000 << ValType << FirstArg->getSourceRange(); 1001 return ExprError(); 1002 } 1003 1004 // Strip any qualifiers off ValType. 1005 ValType = ValType.getUnqualifiedType(); 1006 1007 // The majority of builtins return a value, but a few have special return 1008 // types, so allow them to override appropriately below. 1009 QualType ResultType = ValType; 1010 1011 // We need to figure out which concrete builtin this maps onto. For example, 1012 // __sync_fetch_and_add with a 2 byte object turns into 1013 // __sync_fetch_and_add_2. 1014#define BUILTIN_ROW(x) \ 1015 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 1016 Builtin::BI##x##_8, Builtin::BI##x##_16 } 1017 1018 static const unsigned BuiltinIndices[][5] = { 1019 BUILTIN_ROW(__sync_fetch_and_add), 1020 BUILTIN_ROW(__sync_fetch_and_sub), 1021 BUILTIN_ROW(__sync_fetch_and_or), 1022 BUILTIN_ROW(__sync_fetch_and_and), 1023 BUILTIN_ROW(__sync_fetch_and_xor), 1024 1025 BUILTIN_ROW(__sync_add_and_fetch), 1026 BUILTIN_ROW(__sync_sub_and_fetch), 1027 BUILTIN_ROW(__sync_and_and_fetch), 1028 BUILTIN_ROW(__sync_or_and_fetch), 1029 BUILTIN_ROW(__sync_xor_and_fetch), 1030 1031 BUILTIN_ROW(__sync_val_compare_and_swap), 1032 BUILTIN_ROW(__sync_bool_compare_and_swap), 1033 BUILTIN_ROW(__sync_lock_test_and_set), 1034 BUILTIN_ROW(__sync_lock_release), 1035 BUILTIN_ROW(__sync_swap) 1036 }; 1037#undef BUILTIN_ROW 1038 1039 // Determine the index of the size. 1040 unsigned SizeIndex; 1041 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 1042 case 1: SizeIndex = 0; break; 1043 case 2: SizeIndex = 1; break; 1044 case 4: SizeIndex = 2; break; 1045 case 8: SizeIndex = 3; break; 1046 case 16: SizeIndex = 4; break; 1047 default: 1048 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 1049 << FirstArg->getType() << FirstArg->getSourceRange(); 1050 return ExprError(); 1051 } 1052 1053 // Each of these builtins has one pointer argument, followed by some number of 1054 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 1055 // that we ignore. Find out which row of BuiltinIndices to read from as well 1056 // as the number of fixed args. 1057 unsigned BuiltinID = FDecl->getBuiltinID(); 1058 unsigned BuiltinIndex, NumFixed = 1; 1059 switch (BuiltinID) { 1060 default: llvm_unreachable("Unknown overloaded atomic builtin!"); 1061 case Builtin::BI__sync_fetch_and_add: 1062 case Builtin::BI__sync_fetch_and_add_1: 1063 case Builtin::BI__sync_fetch_and_add_2: 1064 case Builtin::BI__sync_fetch_and_add_4: 1065 case Builtin::BI__sync_fetch_and_add_8: 1066 case Builtin::BI__sync_fetch_and_add_16: 1067 BuiltinIndex = 0; 1068 break; 1069 1070 case Builtin::BI__sync_fetch_and_sub: 1071 case Builtin::BI__sync_fetch_and_sub_1: 1072 case Builtin::BI__sync_fetch_and_sub_2: 1073 case Builtin::BI__sync_fetch_and_sub_4: 1074 case Builtin::BI__sync_fetch_and_sub_8: 1075 case Builtin::BI__sync_fetch_and_sub_16: 1076 BuiltinIndex = 1; 1077 break; 1078 1079 case Builtin::BI__sync_fetch_and_or: 1080 case Builtin::BI__sync_fetch_and_or_1: 1081 case Builtin::BI__sync_fetch_and_or_2: 1082 case Builtin::BI__sync_fetch_and_or_4: 1083 case Builtin::BI__sync_fetch_and_or_8: 1084 case Builtin::BI__sync_fetch_and_or_16: 1085 BuiltinIndex = 2; 1086 break; 1087 1088 case Builtin::BI__sync_fetch_and_and: 1089 case Builtin::BI__sync_fetch_and_and_1: 1090 case Builtin::BI__sync_fetch_and_and_2: 1091 case Builtin::BI__sync_fetch_and_and_4: 1092 case Builtin::BI__sync_fetch_and_and_8: 1093 case Builtin::BI__sync_fetch_and_and_16: 1094 BuiltinIndex = 3; 1095 break; 1096 1097 case Builtin::BI__sync_fetch_and_xor: 1098 case Builtin::BI__sync_fetch_and_xor_1: 1099 case Builtin::BI__sync_fetch_and_xor_2: 1100 case Builtin::BI__sync_fetch_and_xor_4: 1101 case Builtin::BI__sync_fetch_and_xor_8: 1102 case Builtin::BI__sync_fetch_and_xor_16: 1103 BuiltinIndex = 4; 1104 break; 1105 1106 case Builtin::BI__sync_add_and_fetch: 1107 case Builtin::BI__sync_add_and_fetch_1: 1108 case Builtin::BI__sync_add_and_fetch_2: 1109 case Builtin::BI__sync_add_and_fetch_4: 1110 case Builtin::BI__sync_add_and_fetch_8: 1111 case Builtin::BI__sync_add_and_fetch_16: 1112 BuiltinIndex = 5; 1113 break; 1114 1115 case Builtin::BI__sync_sub_and_fetch: 1116 case Builtin::BI__sync_sub_and_fetch_1: 1117 case Builtin::BI__sync_sub_and_fetch_2: 1118 case Builtin::BI__sync_sub_and_fetch_4: 1119 case Builtin::BI__sync_sub_and_fetch_8: 1120 case Builtin::BI__sync_sub_and_fetch_16: 1121 BuiltinIndex = 6; 1122 break; 1123 1124 case Builtin::BI__sync_and_and_fetch: 1125 case Builtin::BI__sync_and_and_fetch_1: 1126 case Builtin::BI__sync_and_and_fetch_2: 1127 case Builtin::BI__sync_and_and_fetch_4: 1128 case Builtin::BI__sync_and_and_fetch_8: 1129 case Builtin::BI__sync_and_and_fetch_16: 1130 BuiltinIndex = 7; 1131 break; 1132 1133 case Builtin::BI__sync_or_and_fetch: 1134 case Builtin::BI__sync_or_and_fetch_1: 1135 case Builtin::BI__sync_or_and_fetch_2: 1136 case Builtin::BI__sync_or_and_fetch_4: 1137 case Builtin::BI__sync_or_and_fetch_8: 1138 case Builtin::BI__sync_or_and_fetch_16: 1139 BuiltinIndex = 8; 1140 break; 1141 1142 case Builtin::BI__sync_xor_and_fetch: 1143 case Builtin::BI__sync_xor_and_fetch_1: 1144 case Builtin::BI__sync_xor_and_fetch_2: 1145 case Builtin::BI__sync_xor_and_fetch_4: 1146 case Builtin::BI__sync_xor_and_fetch_8: 1147 case Builtin::BI__sync_xor_and_fetch_16: 1148 BuiltinIndex = 9; 1149 break; 1150 1151 case Builtin::BI__sync_val_compare_and_swap: 1152 case Builtin::BI__sync_val_compare_and_swap_1: 1153 case Builtin::BI__sync_val_compare_and_swap_2: 1154 case Builtin::BI__sync_val_compare_and_swap_4: 1155 case Builtin::BI__sync_val_compare_and_swap_8: 1156 case Builtin::BI__sync_val_compare_and_swap_16: 1157 BuiltinIndex = 10; 1158 NumFixed = 2; 1159 break; 1160 1161 case Builtin::BI__sync_bool_compare_and_swap: 1162 case Builtin::BI__sync_bool_compare_and_swap_1: 1163 case Builtin::BI__sync_bool_compare_and_swap_2: 1164 case Builtin::BI__sync_bool_compare_and_swap_4: 1165 case Builtin::BI__sync_bool_compare_and_swap_8: 1166 case Builtin::BI__sync_bool_compare_and_swap_16: 1167 BuiltinIndex = 11; 1168 NumFixed = 2; 1169 ResultType = Context.BoolTy; 1170 break; 1171 1172 case Builtin::BI__sync_lock_test_and_set: 1173 case Builtin::BI__sync_lock_test_and_set_1: 1174 case Builtin::BI__sync_lock_test_and_set_2: 1175 case Builtin::BI__sync_lock_test_and_set_4: 1176 case Builtin::BI__sync_lock_test_and_set_8: 1177 case Builtin::BI__sync_lock_test_and_set_16: 1178 BuiltinIndex = 12; 1179 break; 1180 1181 case Builtin::BI__sync_lock_release: 1182 case Builtin::BI__sync_lock_release_1: 1183 case Builtin::BI__sync_lock_release_2: 1184 case Builtin::BI__sync_lock_release_4: 1185 case Builtin::BI__sync_lock_release_8: 1186 case Builtin::BI__sync_lock_release_16: 1187 BuiltinIndex = 13; 1188 NumFixed = 0; 1189 ResultType = Context.VoidTy; 1190 break; 1191 1192 case Builtin::BI__sync_swap: 1193 case Builtin::BI__sync_swap_1: 1194 case Builtin::BI__sync_swap_2: 1195 case Builtin::BI__sync_swap_4: 1196 case Builtin::BI__sync_swap_8: 1197 case Builtin::BI__sync_swap_16: 1198 BuiltinIndex = 14; 1199 break; 1200 } 1201 1202 // Now that we know how many fixed arguments we expect, first check that we 1203 // have at least that many. 1204 if (TheCall->getNumArgs() < 1+NumFixed) { 1205 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 1206 << 0 << 1+NumFixed << TheCall->getNumArgs() 1207 << TheCall->getCallee()->getSourceRange(); 1208 return ExprError(); 1209 } 1210 1211 // Get the decl for the concrete builtin from this, we can tell what the 1212 // concrete integer type we should convert to is. 1213 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 1214 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 1215 FunctionDecl *NewBuiltinDecl; 1216 if (NewBuiltinID == BuiltinID) 1217 NewBuiltinDecl = FDecl; 1218 else { 1219 // Perform builtin lookup to avoid redeclaring it. 1220 DeclarationName DN(&Context.Idents.get(NewBuiltinName)); 1221 LookupResult Res(*this, DN, DRE->getLocStart(), LookupOrdinaryName); 1222 LookupName(Res, TUScope, /*AllowBuiltinCreation=*/true); 1223 assert(Res.getFoundDecl()); 1224 NewBuiltinDecl = dyn_cast<FunctionDecl>(Res.getFoundDecl()); 1225 if (NewBuiltinDecl == 0) 1226 return ExprError(); 1227 } 1228 1229 // The first argument --- the pointer --- has a fixed type; we 1230 // deduce the types of the rest of the arguments accordingly. Walk 1231 // the remaining arguments, converting them to the deduced value type. 1232 for (unsigned i = 0; i != NumFixed; ++i) { 1233 ExprResult Arg = TheCall->getArg(i+1); 1234 1235 // GCC does an implicit conversion to the pointer or integer ValType. This 1236 // can fail in some cases (1i -> int**), check for this error case now. 1237 // Initialize the argument. 1238 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context, 1239 ValType, /*consume*/ false); 1240 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 1241 if (Arg.isInvalid()) 1242 return ExprError(); 1243 1244 // Okay, we have something that *can* be converted to the right type. Check 1245 // to see if there is a potentially weird extension going on here. This can 1246 // happen when you do an atomic operation on something like an char* and 1247 // pass in 42. The 42 gets converted to char. This is even more strange 1248 // for things like 45.123 -> char, etc. 1249 // FIXME: Do this check. 1250 TheCall->setArg(i+1, Arg.take()); 1251 } 1252 1253 ASTContext& Context = this->getASTContext(); 1254 1255 // Create a new DeclRefExpr to refer to the new decl. 1256 DeclRefExpr* NewDRE = DeclRefExpr::Create( 1257 Context, 1258 DRE->getQualifierLoc(), 1259 SourceLocation(), 1260 NewBuiltinDecl, 1261 /*enclosing*/ false, 1262 DRE->getLocation(), 1263 Context.BuiltinFnTy, 1264 DRE->getValueKind()); 1265 1266 // Set the callee in the CallExpr. 1267 // FIXME: This loses syntactic information. 1268 QualType CalleePtrTy = Context.getPointerType(NewBuiltinDecl->getType()); 1269 ExprResult PromotedCall = ImpCastExprToType(NewDRE, CalleePtrTy, 1270 CK_BuiltinFnToFnPtr); 1271 TheCall->setCallee(PromotedCall.take()); 1272 1273 // Change the result type of the call to match the original value type. This 1274 // is arbitrary, but the codegen for these builtins ins design to handle it 1275 // gracefully. 1276 TheCall->setType(ResultType); 1277 1278 return TheCallResult; 1279} 1280 1281/// CheckObjCString - Checks that the argument to the builtin 1282/// CFString constructor is correct 1283/// Note: It might also make sense to do the UTF-16 conversion here (would 1284/// simplify the backend). 1285bool Sema::CheckObjCString(Expr *Arg) { 1286 Arg = Arg->IgnoreParenCasts(); 1287 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 1288 1289 if (!Literal || !Literal->isAscii()) { 1290 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 1291 << Arg->getSourceRange(); 1292 return true; 1293 } 1294 1295 if (Literal->containsNonAsciiOrNull()) { 1296 StringRef String = Literal->getString(); 1297 unsigned NumBytes = String.size(); 1298 SmallVector<UTF16, 128> ToBuf(NumBytes); 1299 const UTF8 *FromPtr = (const UTF8 *)String.data(); 1300 UTF16 *ToPtr = &ToBuf[0]; 1301 1302 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 1303 &ToPtr, ToPtr + NumBytes, 1304 strictConversion); 1305 // Check for conversion failure. 1306 if (Result != conversionOK) 1307 Diag(Arg->getLocStart(), 1308 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 1309 } 1310 return false; 1311} 1312 1313/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 1314/// Emit an error and return true on failure, return false on success. 1315bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 1316 Expr *Fn = TheCall->getCallee(); 1317 if (TheCall->getNumArgs() > 2) { 1318 Diag(TheCall->getArg(2)->getLocStart(), 1319 diag::err_typecheck_call_too_many_args) 1320 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1321 << Fn->getSourceRange() 1322 << SourceRange(TheCall->getArg(2)->getLocStart(), 1323 (*(TheCall->arg_end()-1))->getLocEnd()); 1324 return true; 1325 } 1326 1327 if (TheCall->getNumArgs() < 2) { 1328 return Diag(TheCall->getLocEnd(), 1329 diag::err_typecheck_call_too_few_args_at_least) 1330 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 1331 } 1332 1333 // Type-check the first argument normally. 1334 if (checkBuiltinArgument(*this, TheCall, 0)) 1335 return true; 1336 1337 // Determine whether the current function is variadic or not. 1338 BlockScopeInfo *CurBlock = getCurBlock(); 1339 bool isVariadic; 1340 if (CurBlock) 1341 isVariadic = CurBlock->TheDecl->isVariadic(); 1342 else if (FunctionDecl *FD = getCurFunctionDecl()) 1343 isVariadic = FD->isVariadic(); 1344 else 1345 isVariadic = getCurMethodDecl()->isVariadic(); 1346 1347 if (!isVariadic) { 1348 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 1349 return true; 1350 } 1351 1352 // Verify that the second argument to the builtin is the last argument of the 1353 // current function or method. 1354 bool SecondArgIsLastNamedArgument = false; 1355 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 1356 1357 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 1358 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 1359 // FIXME: This isn't correct for methods (results in bogus warning). 1360 // Get the last formal in the current function. 1361 const ParmVarDecl *LastArg; 1362 if (CurBlock) 1363 LastArg = *(CurBlock->TheDecl->param_end()-1); 1364 else if (FunctionDecl *FD = getCurFunctionDecl()) 1365 LastArg = *(FD->param_end()-1); 1366 else 1367 LastArg = *(getCurMethodDecl()->param_end()-1); 1368 SecondArgIsLastNamedArgument = PV == LastArg; 1369 } 1370 } 1371 1372 if (!SecondArgIsLastNamedArgument) 1373 Diag(TheCall->getArg(1)->getLocStart(), 1374 diag::warn_second_parameter_of_va_start_not_last_named_argument); 1375 return false; 1376} 1377 1378/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 1379/// friends. This is declared to take (...), so we have to check everything. 1380bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 1381 if (TheCall->getNumArgs() < 2) 1382 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1383 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 1384 if (TheCall->getNumArgs() > 2) 1385 return Diag(TheCall->getArg(2)->getLocStart(), 1386 diag::err_typecheck_call_too_many_args) 1387 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1388 << SourceRange(TheCall->getArg(2)->getLocStart(), 1389 (*(TheCall->arg_end()-1))->getLocEnd()); 1390 1391 ExprResult OrigArg0 = TheCall->getArg(0); 1392 ExprResult OrigArg1 = TheCall->getArg(1); 1393 1394 // Do standard promotions between the two arguments, returning their common 1395 // type. 1396 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 1397 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 1398 return true; 1399 1400 // Make sure any conversions are pushed back into the call; this is 1401 // type safe since unordered compare builtins are declared as "_Bool 1402 // foo(...)". 1403 TheCall->setArg(0, OrigArg0.get()); 1404 TheCall->setArg(1, OrigArg1.get()); 1405 1406 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 1407 return false; 1408 1409 // If the common type isn't a real floating type, then the arguments were 1410 // invalid for this operation. 1411 if (Res.isNull() || !Res->isRealFloatingType()) 1412 return Diag(OrigArg0.get()->getLocStart(), 1413 diag::err_typecheck_call_invalid_ordered_compare) 1414 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 1415 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 1416 1417 return false; 1418} 1419 1420/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 1421/// __builtin_isnan and friends. This is declared to take (...), so we have 1422/// to check everything. We expect the last argument to be a floating point 1423/// value. 1424bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 1425 if (TheCall->getNumArgs() < NumArgs) 1426 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1427 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 1428 if (TheCall->getNumArgs() > NumArgs) 1429 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 1430 diag::err_typecheck_call_too_many_args) 1431 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 1432 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 1433 (*(TheCall->arg_end()-1))->getLocEnd()); 1434 1435 Expr *OrigArg = TheCall->getArg(NumArgs-1); 1436 1437 if (OrigArg->isTypeDependent()) 1438 return false; 1439 1440 // This operation requires a non-_Complex floating-point number. 1441 if (!OrigArg->getType()->isRealFloatingType()) 1442 return Diag(OrigArg->getLocStart(), 1443 diag::err_typecheck_call_invalid_unary_fp) 1444 << OrigArg->getType() << OrigArg->getSourceRange(); 1445 1446 // If this is an implicit conversion from float -> double, remove it. 1447 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 1448 Expr *CastArg = Cast->getSubExpr(); 1449 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 1450 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 1451 "promotion from float to double is the only expected cast here"); 1452 Cast->setSubExpr(0); 1453 TheCall->setArg(NumArgs-1, CastArg); 1454 } 1455 } 1456 1457 return false; 1458} 1459 1460/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1461// This is declared to take (...), so we have to check everything. 1462ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1463 if (TheCall->getNumArgs() < 2) 1464 return ExprError(Diag(TheCall->getLocEnd(), 1465 diag::err_typecheck_call_too_few_args_at_least) 1466 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1467 << TheCall->getSourceRange()); 1468 1469 // Determine which of the following types of shufflevector we're checking: 1470 // 1) unary, vector mask: (lhs, mask) 1471 // 2) binary, vector mask: (lhs, rhs, mask) 1472 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1473 QualType resType = TheCall->getArg(0)->getType(); 1474 unsigned numElements = 0; 1475 1476 if (!TheCall->getArg(0)->isTypeDependent() && 1477 !TheCall->getArg(1)->isTypeDependent()) { 1478 QualType LHSType = TheCall->getArg(0)->getType(); 1479 QualType RHSType = TheCall->getArg(1)->getType(); 1480 1481 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 1482 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 1483 << SourceRange(TheCall->getArg(0)->getLocStart(), 1484 TheCall->getArg(1)->getLocEnd()); 1485 return ExprError(); 1486 } 1487 1488 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1489 unsigned numResElements = TheCall->getNumArgs() - 2; 1490 1491 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1492 // with mask. If so, verify that RHS is an integer vector type with the 1493 // same number of elts as lhs. 1494 if (TheCall->getNumArgs() == 2) { 1495 if (!RHSType->hasIntegerRepresentation() || 1496 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1497 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1498 << SourceRange(TheCall->getArg(1)->getLocStart(), 1499 TheCall->getArg(1)->getLocEnd()); 1500 numResElements = numElements; 1501 } 1502 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1503 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1504 << SourceRange(TheCall->getArg(0)->getLocStart(), 1505 TheCall->getArg(1)->getLocEnd()); 1506 return ExprError(); 1507 } else if (numElements != numResElements) { 1508 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1509 resType = Context.getVectorType(eltType, numResElements, 1510 VectorType::GenericVector); 1511 } 1512 } 1513 1514 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1515 if (TheCall->getArg(i)->isTypeDependent() || 1516 TheCall->getArg(i)->isValueDependent()) 1517 continue; 1518 1519 llvm::APSInt Result(32); 1520 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1521 return ExprError(Diag(TheCall->getLocStart(), 1522 diag::err_shufflevector_nonconstant_argument) 1523 << TheCall->getArg(i)->getSourceRange()); 1524 1525 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1526 return ExprError(Diag(TheCall->getLocStart(), 1527 diag::err_shufflevector_argument_too_large) 1528 << TheCall->getArg(i)->getSourceRange()); 1529 } 1530 1531 SmallVector<Expr*, 32> exprs; 1532 1533 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1534 exprs.push_back(TheCall->getArg(i)); 1535 TheCall->setArg(i, 0); 1536 } 1537 1538 return Owned(new (Context) ShuffleVectorExpr(Context, exprs, resType, 1539 TheCall->getCallee()->getLocStart(), 1540 TheCall->getRParenLoc())); 1541} 1542 1543/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1544// This is declared to take (const void*, ...) and can take two 1545// optional constant int args. 1546bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1547 unsigned NumArgs = TheCall->getNumArgs(); 1548 1549 if (NumArgs > 3) 1550 return Diag(TheCall->getLocEnd(), 1551 diag::err_typecheck_call_too_many_args_at_most) 1552 << 0 /*function call*/ << 3 << NumArgs 1553 << TheCall->getSourceRange(); 1554 1555 // Argument 0 is checked for us and the remaining arguments must be 1556 // constant integers. 1557 for (unsigned i = 1; i != NumArgs; ++i) { 1558 Expr *Arg = TheCall->getArg(i); 1559 1560 // We can't check the value of a dependent argument. 1561 if (Arg->isTypeDependent() || Arg->isValueDependent()) 1562 continue; 1563 1564 llvm::APSInt Result; 1565 if (SemaBuiltinConstantArg(TheCall, i, Result)) 1566 return true; 1567 1568 // FIXME: gcc issues a warning and rewrites these to 0. These 1569 // seems especially odd for the third argument since the default 1570 // is 3. 1571 if (i == 1) { 1572 if (Result.getLimitedValue() > 1) 1573 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1574 << "0" << "1" << Arg->getSourceRange(); 1575 } else { 1576 if (Result.getLimitedValue() > 3) 1577 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1578 << "0" << "3" << Arg->getSourceRange(); 1579 } 1580 } 1581 1582 return false; 1583} 1584 1585/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1586/// TheCall is a constant expression. 1587bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1588 llvm::APSInt &Result) { 1589 Expr *Arg = TheCall->getArg(ArgNum); 1590 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1591 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1592 1593 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 1594 1595 if (!Arg->isIntegerConstantExpr(Result, Context)) 1596 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 1597 << FDecl->getDeclName() << Arg->getSourceRange(); 1598 1599 return false; 1600} 1601 1602/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 1603/// int type). This simply type checks that type is one of the defined 1604/// constants (0-3). 1605// For compatibility check 0-3, llvm only handles 0 and 2. 1606bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 1607 llvm::APSInt Result; 1608 1609 // We can't check the value of a dependent argument. 1610 if (TheCall->getArg(1)->isTypeDependent() || 1611 TheCall->getArg(1)->isValueDependent()) 1612 return false; 1613 1614 // Check constant-ness first. 1615 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1616 return true; 1617 1618 Expr *Arg = TheCall->getArg(1); 1619 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 1620 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1621 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1622 } 1623 1624 return false; 1625} 1626 1627/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1628/// This checks that val is a constant 1. 1629bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1630 Expr *Arg = TheCall->getArg(1); 1631 llvm::APSInt Result; 1632 1633 // TODO: This is less than ideal. Overload this to take a value. 1634 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1635 return true; 1636 1637 if (Result != 1) 1638 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1639 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1640 1641 return false; 1642} 1643 1644// Determine if an expression is a string literal or constant string. 1645// If this function returns false on the arguments to a function expecting a 1646// format string, we will usually need to emit a warning. 1647// True string literals are then checked by CheckFormatString. 1648Sema::StringLiteralCheckType 1649Sema::checkFormatStringExpr(const Expr *E, Expr **Args, 1650 unsigned NumArgs, bool HasVAListArg, 1651 unsigned format_idx, unsigned firstDataArg, 1652 FormatStringType Type, VariadicCallType CallType, 1653 bool inFunctionCall) { 1654 tryAgain: 1655 if (E->isTypeDependent() || E->isValueDependent()) 1656 return SLCT_NotALiteral; 1657 1658 E = E->IgnoreParenCasts(); 1659 1660 if (E->isNullPointerConstant(Context, Expr::NPC_ValueDependentIsNotNull)) 1661 // Technically -Wformat-nonliteral does not warn about this case. 1662 // The behavior of printf and friends in this case is implementation 1663 // dependent. Ideally if the format string cannot be null then 1664 // it should have a 'nonnull' attribute in the function prototype. 1665 return SLCT_CheckedLiteral; 1666 1667 switch (E->getStmtClass()) { 1668 case Stmt::BinaryConditionalOperatorClass: 1669 case Stmt::ConditionalOperatorClass: { 1670 // The expression is a literal if both sub-expressions were, and it was 1671 // completely checked only if both sub-expressions were checked. 1672 const AbstractConditionalOperator *C = 1673 cast<AbstractConditionalOperator>(E); 1674 StringLiteralCheckType Left = 1675 checkFormatStringExpr(C->getTrueExpr(), Args, NumArgs, 1676 HasVAListArg, format_idx, firstDataArg, 1677 Type, CallType, inFunctionCall); 1678 if (Left == SLCT_NotALiteral) 1679 return SLCT_NotALiteral; 1680 StringLiteralCheckType Right = 1681 checkFormatStringExpr(C->getFalseExpr(), Args, NumArgs, 1682 HasVAListArg, format_idx, firstDataArg, 1683 Type, CallType, inFunctionCall); 1684 return Left < Right ? Left : Right; 1685 } 1686 1687 case Stmt::ImplicitCastExprClass: { 1688 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 1689 goto tryAgain; 1690 } 1691 1692 case Stmt::OpaqueValueExprClass: 1693 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 1694 E = src; 1695 goto tryAgain; 1696 } 1697 return SLCT_NotALiteral; 1698 1699 case Stmt::PredefinedExprClass: 1700 // While __func__, etc., are technically not string literals, they 1701 // cannot contain format specifiers and thus are not a security 1702 // liability. 1703 return SLCT_UncheckedLiteral; 1704 1705 case Stmt::DeclRefExprClass: { 1706 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1707 1708 // As an exception, do not flag errors for variables binding to 1709 // const string literals. 1710 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1711 bool isConstant = false; 1712 QualType T = DR->getType(); 1713 1714 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1715 isConstant = AT->getElementType().isConstant(Context); 1716 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1717 isConstant = T.isConstant(Context) && 1718 PT->getPointeeType().isConstant(Context); 1719 } else if (T->isObjCObjectPointerType()) { 1720 // In ObjC, there is usually no "const ObjectPointer" type, 1721 // so don't check if the pointee type is constant. 1722 isConstant = T.isConstant(Context); 1723 } 1724 1725 if (isConstant) { 1726 if (const Expr *Init = VD->getAnyInitializer()) { 1727 // Look through initializers like const char c[] = { "foo" } 1728 if (const InitListExpr *InitList = dyn_cast<InitListExpr>(Init)) { 1729 if (InitList->isStringLiteralInit()) 1730 Init = InitList->getInit(0)->IgnoreParenImpCasts(); 1731 } 1732 return checkFormatStringExpr(Init, Args, NumArgs, 1733 HasVAListArg, format_idx, 1734 firstDataArg, Type, CallType, 1735 /*inFunctionCall*/false); 1736 } 1737 } 1738 1739 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1740 // special check to see if the format string is a function parameter 1741 // of the function calling the printf function. If the function 1742 // has an attribute indicating it is a printf-like function, then we 1743 // should suppress warnings concerning non-literals being used in a call 1744 // to a vprintf function. For example: 1745 // 1746 // void 1747 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1748 // va_list ap; 1749 // va_start(ap, fmt); 1750 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1751 // ... 1752 // 1753 if (HasVAListArg) { 1754 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(VD)) { 1755 if (const NamedDecl *ND = dyn_cast<NamedDecl>(PV->getDeclContext())) { 1756 int PVIndex = PV->getFunctionScopeIndex() + 1; 1757 for (specific_attr_iterator<FormatAttr> 1758 i = ND->specific_attr_begin<FormatAttr>(), 1759 e = ND->specific_attr_end<FormatAttr>(); i != e ; ++i) { 1760 FormatAttr *PVFormat = *i; 1761 // adjust for implicit parameter 1762 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1763 if (MD->isInstance()) 1764 ++PVIndex; 1765 // We also check if the formats are compatible. 1766 // We can't pass a 'scanf' string to a 'printf' function. 1767 if (PVIndex == PVFormat->getFormatIdx() && 1768 Type == GetFormatStringType(PVFormat)) 1769 return SLCT_UncheckedLiteral; 1770 } 1771 } 1772 } 1773 } 1774 } 1775 1776 return SLCT_NotALiteral; 1777 } 1778 1779 case Stmt::CallExprClass: 1780 case Stmt::CXXMemberCallExprClass: { 1781 const CallExpr *CE = cast<CallExpr>(E); 1782 if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(CE->getCalleeDecl())) { 1783 if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) { 1784 unsigned ArgIndex = FA->getFormatIdx(); 1785 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1786 if (MD->isInstance()) 1787 --ArgIndex; 1788 const Expr *Arg = CE->getArg(ArgIndex - 1); 1789 1790 return checkFormatStringExpr(Arg, Args, NumArgs, 1791 HasVAListArg, format_idx, firstDataArg, 1792 Type, CallType, inFunctionCall); 1793 } else if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(ND)) { 1794 unsigned BuiltinID = FD->getBuiltinID(); 1795 if (BuiltinID == Builtin::BI__builtin___CFStringMakeConstantString || 1796 BuiltinID == Builtin::BI__builtin___NSStringMakeConstantString) { 1797 const Expr *Arg = CE->getArg(0); 1798 return checkFormatStringExpr(Arg, Args, NumArgs, 1799 HasVAListArg, format_idx, 1800 firstDataArg, Type, CallType, 1801 inFunctionCall); 1802 } 1803 } 1804 } 1805 1806 return SLCT_NotALiteral; 1807 } 1808 case Stmt::ObjCStringLiteralClass: 1809 case Stmt::StringLiteralClass: { 1810 const StringLiteral *StrE = NULL; 1811 1812 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1813 StrE = ObjCFExpr->getString(); 1814 else 1815 StrE = cast<StringLiteral>(E); 1816 1817 if (StrE) { 1818 CheckFormatString(StrE, E, Args, NumArgs, HasVAListArg, format_idx, 1819 firstDataArg, Type, inFunctionCall, CallType); 1820 return SLCT_CheckedLiteral; 1821 } 1822 1823 return SLCT_NotALiteral; 1824 } 1825 1826 default: 1827 return SLCT_NotALiteral; 1828 } 1829} 1830 1831void 1832Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1833 const Expr * const *ExprArgs, 1834 SourceLocation CallSiteLoc) { 1835 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1836 e = NonNull->args_end(); 1837 i != e; ++i) { 1838 const Expr *ArgExpr = ExprArgs[*i]; 1839 if (ArgExpr->isNullPointerConstant(Context, 1840 Expr::NPC_ValueDependentIsNotNull)) 1841 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 1842 } 1843} 1844 1845Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) { 1846 return llvm::StringSwitch<FormatStringType>(Format->getType()) 1847 .Case("scanf", FST_Scanf) 1848 .Cases("printf", "printf0", FST_Printf) 1849 .Cases("NSString", "CFString", FST_NSString) 1850 .Case("strftime", FST_Strftime) 1851 .Case("strfmon", FST_Strfmon) 1852 .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf) 1853 .Default(FST_Unknown); 1854} 1855 1856/// CheckFormatArguments - Check calls to printf and scanf (and similar 1857/// functions) for correct use of format strings. 1858/// Returns true if a format string has been fully checked. 1859bool Sema::CheckFormatArguments(const FormatAttr *Format, Expr **Args, 1860 unsigned NumArgs, bool IsCXXMember, 1861 VariadicCallType CallType, 1862 SourceLocation Loc, SourceRange Range) { 1863 FormatStringInfo FSI; 1864 if (getFormatStringInfo(Format, IsCXXMember, &FSI)) 1865 return CheckFormatArguments(Args, NumArgs, FSI.HasVAListArg, FSI.FormatIdx, 1866 FSI.FirstDataArg, GetFormatStringType(Format), 1867 CallType, Loc, Range); 1868 return false; 1869} 1870 1871bool Sema::CheckFormatArguments(Expr **Args, unsigned NumArgs, 1872 bool HasVAListArg, unsigned format_idx, 1873 unsigned firstDataArg, FormatStringType Type, 1874 VariadicCallType CallType, 1875 SourceLocation Loc, SourceRange Range) { 1876 // CHECK: printf/scanf-like function is called with no format string. 1877 if (format_idx >= NumArgs) { 1878 Diag(Loc, diag::warn_missing_format_string) << Range; 1879 return false; 1880 } 1881 1882 const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts(); 1883 1884 // CHECK: format string is not a string literal. 1885 // 1886 // Dynamically generated format strings are difficult to 1887 // automatically vet at compile time. Requiring that format strings 1888 // are string literals: (1) permits the checking of format strings by 1889 // the compiler and thereby (2) can practically remove the source of 1890 // many format string exploits. 1891 1892 // Format string can be either ObjC string (e.g. @"%d") or 1893 // C string (e.g. "%d") 1894 // ObjC string uses the same format specifiers as C string, so we can use 1895 // the same format string checking logic for both ObjC and C strings. 1896 StringLiteralCheckType CT = 1897 checkFormatStringExpr(OrigFormatExpr, Args, NumArgs, HasVAListArg, 1898 format_idx, firstDataArg, Type, CallType); 1899 if (CT != SLCT_NotALiteral) 1900 // Literal format string found, check done! 1901 return CT == SLCT_CheckedLiteral; 1902 1903 // Strftime is particular as it always uses a single 'time' argument, 1904 // so it is safe to pass a non-literal string. 1905 if (Type == FST_Strftime) 1906 return false; 1907 1908 // Do not emit diag when the string param is a macro expansion and the 1909 // format is either NSString or CFString. This is a hack to prevent 1910 // diag when using the NSLocalizedString and CFCopyLocalizedString macros 1911 // which are usually used in place of NS and CF string literals. 1912 if (Type == FST_NSString && 1913 SourceMgr.isInSystemMacro(Args[format_idx]->getLocStart())) 1914 return false; 1915 1916 // If there are no arguments specified, warn with -Wformat-security, otherwise 1917 // warn only with -Wformat-nonliteral. 1918 if (NumArgs == format_idx+1) 1919 Diag(Args[format_idx]->getLocStart(), 1920 diag::warn_format_nonliteral_noargs) 1921 << OrigFormatExpr->getSourceRange(); 1922 else 1923 Diag(Args[format_idx]->getLocStart(), 1924 diag::warn_format_nonliteral) 1925 << OrigFormatExpr->getSourceRange(); 1926 return false; 1927} 1928 1929namespace { 1930class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1931protected: 1932 Sema &S; 1933 const StringLiteral *FExpr; 1934 const Expr *OrigFormatExpr; 1935 const unsigned FirstDataArg; 1936 const unsigned NumDataArgs; 1937 const char *Beg; // Start of format string. 1938 const bool HasVAListArg; 1939 const Expr * const *Args; 1940 const unsigned NumArgs; 1941 unsigned FormatIdx; 1942 llvm::BitVector CoveredArgs; 1943 bool usesPositionalArgs; 1944 bool atFirstArg; 1945 bool inFunctionCall; 1946 Sema::VariadicCallType CallType; 1947public: 1948 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1949 const Expr *origFormatExpr, unsigned firstDataArg, 1950 unsigned numDataArgs, const char *beg, bool hasVAListArg, 1951 Expr **args, unsigned numArgs, 1952 unsigned formatIdx, bool inFunctionCall, 1953 Sema::VariadicCallType callType) 1954 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1955 FirstDataArg(firstDataArg), NumDataArgs(numDataArgs), 1956 Beg(beg), HasVAListArg(hasVAListArg), 1957 Args(args), NumArgs(numArgs), FormatIdx(formatIdx), 1958 usesPositionalArgs(false), atFirstArg(true), 1959 inFunctionCall(inFunctionCall), CallType(callType) { 1960 CoveredArgs.resize(numDataArgs); 1961 CoveredArgs.reset(); 1962 } 1963 1964 void DoneProcessing(); 1965 1966 void HandleIncompleteSpecifier(const char *startSpecifier, 1967 unsigned specifierLen); 1968 1969 void HandleInvalidLengthModifier( 1970 const analyze_format_string::FormatSpecifier &FS, 1971 const analyze_format_string::ConversionSpecifier &CS, 1972 const char *startSpecifier, unsigned specifierLen, unsigned DiagID); 1973 1974 void HandleNonStandardLengthModifier( 1975 const analyze_format_string::FormatSpecifier &FS, 1976 const char *startSpecifier, unsigned specifierLen); 1977 1978 void HandleNonStandardConversionSpecifier( 1979 const analyze_format_string::ConversionSpecifier &CS, 1980 const char *startSpecifier, unsigned specifierLen); 1981 1982 virtual void HandlePosition(const char *startPos, unsigned posLen); 1983 1984 virtual void HandleInvalidPosition(const char *startSpecifier, 1985 unsigned specifierLen, 1986 analyze_format_string::PositionContext p); 1987 1988 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1989 1990 void HandleNullChar(const char *nullCharacter); 1991 1992 template <typename Range> 1993 static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall, 1994 const Expr *ArgumentExpr, 1995 PartialDiagnostic PDiag, 1996 SourceLocation StringLoc, 1997 bool IsStringLocation, Range StringRange, 1998 ArrayRef<FixItHint> Fixit = ArrayRef<FixItHint>()); 1999 2000protected: 2001 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 2002 const char *startSpec, 2003 unsigned specifierLen, 2004 const char *csStart, unsigned csLen); 2005 2006 void HandlePositionalNonpositionalArgs(SourceLocation Loc, 2007 const char *startSpec, 2008 unsigned specifierLen); 2009 2010 SourceRange getFormatStringRange(); 2011 CharSourceRange getSpecifierRange(const char *startSpecifier, 2012 unsigned specifierLen); 2013 SourceLocation getLocationOfByte(const char *x); 2014 2015 const Expr *getDataArg(unsigned i) const; 2016 2017 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 2018 const analyze_format_string::ConversionSpecifier &CS, 2019 const char *startSpecifier, unsigned specifierLen, 2020 unsigned argIndex); 2021 2022 template <typename Range> 2023 void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc, 2024 bool IsStringLocation, Range StringRange, 2025 ArrayRef<FixItHint> Fixit = ArrayRef<FixItHint>()); 2026 2027 void CheckPositionalAndNonpositionalArgs( 2028 const analyze_format_string::FormatSpecifier *FS); 2029}; 2030} 2031 2032SourceRange CheckFormatHandler::getFormatStringRange() { 2033 return OrigFormatExpr->getSourceRange(); 2034} 2035 2036CharSourceRange CheckFormatHandler:: 2037getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 2038 SourceLocation Start = getLocationOfByte(startSpecifier); 2039 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 2040 2041 // Advance the end SourceLocation by one due to half-open ranges. 2042 End = End.getLocWithOffset(1); 2043 2044 return CharSourceRange::getCharRange(Start, End); 2045} 2046 2047SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 2048 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 2049} 2050 2051void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 2052 unsigned specifierLen){ 2053 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier), 2054 getLocationOfByte(startSpecifier), 2055 /*IsStringLocation*/true, 2056 getSpecifierRange(startSpecifier, specifierLen)); 2057} 2058 2059void CheckFormatHandler::HandleInvalidLengthModifier( 2060 const analyze_format_string::FormatSpecifier &FS, 2061 const analyze_format_string::ConversionSpecifier &CS, 2062 const char *startSpecifier, unsigned specifierLen, unsigned DiagID) { 2063 using namespace analyze_format_string; 2064 2065 const LengthModifier &LM = FS.getLengthModifier(); 2066 CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength()); 2067 2068 // See if we know how to fix this length modifier. 2069 llvm::Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier(); 2070 if (FixedLM) { 2071 EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(), 2072 getLocationOfByte(LM.getStart()), 2073 /*IsStringLocation*/true, 2074 getSpecifierRange(startSpecifier, specifierLen)); 2075 2076 S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier) 2077 << FixedLM->toString() 2078 << FixItHint::CreateReplacement(LMRange, FixedLM->toString()); 2079 2080 } else { 2081 FixItHint Hint; 2082 if (DiagID == diag::warn_format_nonsensical_length) 2083 Hint = FixItHint::CreateRemoval(LMRange); 2084 2085 EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(), 2086 getLocationOfByte(LM.getStart()), 2087 /*IsStringLocation*/true, 2088 getSpecifierRange(startSpecifier, specifierLen), 2089 Hint); 2090 } 2091} 2092 2093void CheckFormatHandler::HandleNonStandardLengthModifier( 2094 const analyze_format_string::FormatSpecifier &FS, 2095 const char *startSpecifier, unsigned specifierLen) { 2096 using namespace analyze_format_string; 2097 2098 const LengthModifier &LM = FS.getLengthModifier(); 2099 CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength()); 2100 2101 // See if we know how to fix this length modifier. 2102 llvm::Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier(); 2103 if (FixedLM) { 2104 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2105 << LM.toString() << 0, 2106 getLocationOfByte(LM.getStart()), 2107 /*IsStringLocation*/true, 2108 getSpecifierRange(startSpecifier, specifierLen)); 2109 2110 S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier) 2111 << FixedLM->toString() 2112 << FixItHint::CreateReplacement(LMRange, FixedLM->toString()); 2113 2114 } else { 2115 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2116 << LM.toString() << 0, 2117 getLocationOfByte(LM.getStart()), 2118 /*IsStringLocation*/true, 2119 getSpecifierRange(startSpecifier, specifierLen)); 2120 } 2121} 2122 2123void CheckFormatHandler::HandleNonStandardConversionSpecifier( 2124 const analyze_format_string::ConversionSpecifier &CS, 2125 const char *startSpecifier, unsigned specifierLen) { 2126 using namespace analyze_format_string; 2127 2128 // See if we know how to fix this conversion specifier. 2129 llvm::Optional<ConversionSpecifier> FixedCS = CS.getStandardSpecifier(); 2130 if (FixedCS) { 2131 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2132 << CS.toString() << /*conversion specifier*/1, 2133 getLocationOfByte(CS.getStart()), 2134 /*IsStringLocation*/true, 2135 getSpecifierRange(startSpecifier, specifierLen)); 2136 2137 CharSourceRange CSRange = getSpecifierRange(CS.getStart(), CS.getLength()); 2138 S.Diag(getLocationOfByte(CS.getStart()), diag::note_format_fix_specifier) 2139 << FixedCS->toString() 2140 << FixItHint::CreateReplacement(CSRange, FixedCS->toString()); 2141 } else { 2142 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2143 << CS.toString() << /*conversion specifier*/1, 2144 getLocationOfByte(CS.getStart()), 2145 /*IsStringLocation*/true, 2146 getSpecifierRange(startSpecifier, specifierLen)); 2147 } 2148} 2149 2150void CheckFormatHandler::HandlePosition(const char *startPos, 2151 unsigned posLen) { 2152 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_positional_arg), 2153 getLocationOfByte(startPos), 2154 /*IsStringLocation*/true, 2155 getSpecifierRange(startPos, posLen)); 2156} 2157 2158void 2159CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 2160 analyze_format_string::PositionContext p) { 2161 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier) 2162 << (unsigned) p, 2163 getLocationOfByte(startPos), /*IsStringLocation*/true, 2164 getSpecifierRange(startPos, posLen)); 2165} 2166 2167void CheckFormatHandler::HandleZeroPosition(const char *startPos, 2168 unsigned posLen) { 2169 EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier), 2170 getLocationOfByte(startPos), 2171 /*IsStringLocation*/true, 2172 getSpecifierRange(startPos, posLen)); 2173} 2174 2175void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 2176 if (!isa<ObjCStringLiteral>(OrigFormatExpr)) { 2177 // The presence of a null character is likely an error. 2178 EmitFormatDiagnostic( 2179 S.PDiag(diag::warn_printf_format_string_contains_null_char), 2180 getLocationOfByte(nullCharacter), /*IsStringLocation*/true, 2181 getFormatStringRange()); 2182 } 2183} 2184 2185// Note that this may return NULL if there was an error parsing or building 2186// one of the argument expressions. 2187const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 2188 return Args[FirstDataArg + i]; 2189} 2190 2191void CheckFormatHandler::DoneProcessing() { 2192 // Does the number of data arguments exceed the number of 2193 // format conversions in the format string? 2194 if (!HasVAListArg) { 2195 // Find any arguments that weren't covered. 2196 CoveredArgs.flip(); 2197 signed notCoveredArg = CoveredArgs.find_first(); 2198 if (notCoveredArg >= 0) { 2199 assert((unsigned)notCoveredArg < NumDataArgs); 2200 if (const Expr *E = getDataArg((unsigned) notCoveredArg)) { 2201 SourceLocation Loc = E->getLocStart(); 2202 if (!S.getSourceManager().isInSystemMacro(Loc)) { 2203 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used), 2204 Loc, /*IsStringLocation*/false, 2205 getFormatStringRange()); 2206 } 2207 } 2208 } 2209 } 2210} 2211 2212bool 2213CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 2214 SourceLocation Loc, 2215 const char *startSpec, 2216 unsigned specifierLen, 2217 const char *csStart, 2218 unsigned csLen) { 2219 2220 bool keepGoing = true; 2221 if (argIndex < NumDataArgs) { 2222 // Consider the argument coverered, even though the specifier doesn't 2223 // make sense. 2224 CoveredArgs.set(argIndex); 2225 } 2226 else { 2227 // If argIndex exceeds the number of data arguments we 2228 // don't issue a warning because that is just a cascade of warnings (and 2229 // they may have intended '%%' anyway). We don't want to continue processing 2230 // the format string after this point, however, as we will like just get 2231 // gibberish when trying to match arguments. 2232 keepGoing = false; 2233 } 2234 2235 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion) 2236 << StringRef(csStart, csLen), 2237 Loc, /*IsStringLocation*/true, 2238 getSpecifierRange(startSpec, specifierLen)); 2239 2240 return keepGoing; 2241} 2242 2243void 2244CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc, 2245 const char *startSpec, 2246 unsigned specifierLen) { 2247 EmitFormatDiagnostic( 2248 S.PDiag(diag::warn_format_mix_positional_nonpositional_args), 2249 Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen)); 2250} 2251 2252bool 2253CheckFormatHandler::CheckNumArgs( 2254 const analyze_format_string::FormatSpecifier &FS, 2255 const analyze_format_string::ConversionSpecifier &CS, 2256 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 2257 2258 if (argIndex >= NumDataArgs) { 2259 PartialDiagnostic PDiag = FS.usesPositionalArg() 2260 ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args) 2261 << (argIndex+1) << NumDataArgs) 2262 : S.PDiag(diag::warn_printf_insufficient_data_args); 2263 EmitFormatDiagnostic( 2264 PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true, 2265 getSpecifierRange(startSpecifier, specifierLen)); 2266 return false; 2267 } 2268 return true; 2269} 2270 2271template<typename Range> 2272void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag, 2273 SourceLocation Loc, 2274 bool IsStringLocation, 2275 Range StringRange, 2276 ArrayRef<FixItHint> FixIt) { 2277 EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag, 2278 Loc, IsStringLocation, StringRange, FixIt); 2279} 2280 2281/// \brief If the format string is not within the funcion call, emit a note 2282/// so that the function call and string are in diagnostic messages. 2283/// 2284/// \param InFunctionCall if true, the format string is within the function 2285/// call and only one diagnostic message will be produced. Otherwise, an 2286/// extra note will be emitted pointing to location of the format string. 2287/// 2288/// \param ArgumentExpr the expression that is passed as the format string 2289/// argument in the function call. Used for getting locations when two 2290/// diagnostics are emitted. 2291/// 2292/// \param PDiag the callee should already have provided any strings for the 2293/// diagnostic message. This function only adds locations and fixits 2294/// to diagnostics. 2295/// 2296/// \param Loc primary location for diagnostic. If two diagnostics are 2297/// required, one will be at Loc and a new SourceLocation will be created for 2298/// the other one. 2299/// 2300/// \param IsStringLocation if true, Loc points to the format string should be 2301/// used for the note. Otherwise, Loc points to the argument list and will 2302/// be used with PDiag. 2303/// 2304/// \param StringRange some or all of the string to highlight. This is 2305/// templated so it can accept either a CharSourceRange or a SourceRange. 2306/// 2307/// \param FixIt optional fix it hint for the format string. 2308template<typename Range> 2309void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall, 2310 const Expr *ArgumentExpr, 2311 PartialDiagnostic PDiag, 2312 SourceLocation Loc, 2313 bool IsStringLocation, 2314 Range StringRange, 2315 ArrayRef<FixItHint> FixIt) { 2316 if (InFunctionCall) { 2317 const Sema::SemaDiagnosticBuilder &D = S.Diag(Loc, PDiag); 2318 D << StringRange; 2319 for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end(); 2320 I != E; ++I) { 2321 D << *I; 2322 } 2323 } else { 2324 S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag) 2325 << ArgumentExpr->getSourceRange(); 2326 2327 const Sema::SemaDiagnosticBuilder &Note = 2328 S.Diag(IsStringLocation ? Loc : StringRange.getBegin(), 2329 diag::note_format_string_defined); 2330 2331 Note << StringRange; 2332 for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end(); 2333 I != E; ++I) { 2334 Note << *I; 2335 } 2336 } 2337} 2338 2339//===--- CHECK: Printf format string checking ------------------------------===// 2340 2341namespace { 2342class CheckPrintfHandler : public CheckFormatHandler { 2343 bool ObjCContext; 2344public: 2345 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 2346 const Expr *origFormatExpr, unsigned firstDataArg, 2347 unsigned numDataArgs, bool isObjC, 2348 const char *beg, bool hasVAListArg, 2349 Expr **Args, unsigned NumArgs, 2350 unsigned formatIdx, bool inFunctionCall, 2351 Sema::VariadicCallType CallType) 2352 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2353 numDataArgs, beg, hasVAListArg, Args, NumArgs, 2354 formatIdx, inFunctionCall, CallType), ObjCContext(isObjC) 2355 {} 2356 2357 2358 bool HandleInvalidPrintfConversionSpecifier( 2359 const analyze_printf::PrintfSpecifier &FS, 2360 const char *startSpecifier, 2361 unsigned specifierLen); 2362 2363 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 2364 const char *startSpecifier, 2365 unsigned specifierLen); 2366 bool checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 2367 const char *StartSpecifier, 2368 unsigned SpecifierLen, 2369 const Expr *E); 2370 2371 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 2372 const char *startSpecifier, unsigned specifierLen); 2373 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 2374 const analyze_printf::OptionalAmount &Amt, 2375 unsigned type, 2376 const char *startSpecifier, unsigned specifierLen); 2377 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2378 const analyze_printf::OptionalFlag &flag, 2379 const char *startSpecifier, unsigned specifierLen); 2380 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 2381 const analyze_printf::OptionalFlag &ignoredFlag, 2382 const analyze_printf::OptionalFlag &flag, 2383 const char *startSpecifier, unsigned specifierLen); 2384 bool checkForCStrMembers(const analyze_printf::ArgType &AT, 2385 const Expr *E, const CharSourceRange &CSR); 2386 2387}; 2388} 2389 2390bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 2391 const analyze_printf::PrintfSpecifier &FS, 2392 const char *startSpecifier, 2393 unsigned specifierLen) { 2394 const analyze_printf::PrintfConversionSpecifier &CS = 2395 FS.getConversionSpecifier(); 2396 2397 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2398 getLocationOfByte(CS.getStart()), 2399 startSpecifier, specifierLen, 2400 CS.getStart(), CS.getLength()); 2401} 2402 2403bool CheckPrintfHandler::HandleAmount( 2404 const analyze_format_string::OptionalAmount &Amt, 2405 unsigned k, const char *startSpecifier, 2406 unsigned specifierLen) { 2407 2408 if (Amt.hasDataArgument()) { 2409 if (!HasVAListArg) { 2410 unsigned argIndex = Amt.getArgIndex(); 2411 if (argIndex >= NumDataArgs) { 2412 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg) 2413 << k, 2414 getLocationOfByte(Amt.getStart()), 2415 /*IsStringLocation*/true, 2416 getSpecifierRange(startSpecifier, specifierLen)); 2417 // Don't do any more checking. We will just emit 2418 // spurious errors. 2419 return false; 2420 } 2421 2422 // Type check the data argument. It should be an 'int'. 2423 // Although not in conformance with C99, we also allow the argument to be 2424 // an 'unsigned int' as that is a reasonably safe case. GCC also 2425 // doesn't emit a warning for that case. 2426 CoveredArgs.set(argIndex); 2427 const Expr *Arg = getDataArg(argIndex); 2428 if (!Arg) 2429 return false; 2430 2431 QualType T = Arg->getType(); 2432 2433 const analyze_printf::ArgType &AT = Amt.getArgType(S.Context); 2434 assert(AT.isValid()); 2435 2436 if (!AT.matchesType(S.Context, T)) { 2437 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type) 2438 << k << AT.getRepresentativeTypeName(S.Context) 2439 << T << Arg->getSourceRange(), 2440 getLocationOfByte(Amt.getStart()), 2441 /*IsStringLocation*/true, 2442 getSpecifierRange(startSpecifier, specifierLen)); 2443 // Don't do any more checking. We will just emit 2444 // spurious errors. 2445 return false; 2446 } 2447 } 2448 } 2449 return true; 2450} 2451 2452void CheckPrintfHandler::HandleInvalidAmount( 2453 const analyze_printf::PrintfSpecifier &FS, 2454 const analyze_printf::OptionalAmount &Amt, 2455 unsigned type, 2456 const char *startSpecifier, 2457 unsigned specifierLen) { 2458 const analyze_printf::PrintfConversionSpecifier &CS = 2459 FS.getConversionSpecifier(); 2460 2461 FixItHint fixit = 2462 Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant 2463 ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 2464 Amt.getConstantLength())) 2465 : FixItHint(); 2466 2467 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount) 2468 << type << CS.toString(), 2469 getLocationOfByte(Amt.getStart()), 2470 /*IsStringLocation*/true, 2471 getSpecifierRange(startSpecifier, specifierLen), 2472 fixit); 2473} 2474 2475void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2476 const analyze_printf::OptionalFlag &flag, 2477 const char *startSpecifier, 2478 unsigned specifierLen) { 2479 // Warn about pointless flag with a fixit removal. 2480 const analyze_printf::PrintfConversionSpecifier &CS = 2481 FS.getConversionSpecifier(); 2482 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag) 2483 << flag.toString() << CS.toString(), 2484 getLocationOfByte(flag.getPosition()), 2485 /*IsStringLocation*/true, 2486 getSpecifierRange(startSpecifier, specifierLen), 2487 FixItHint::CreateRemoval( 2488 getSpecifierRange(flag.getPosition(), 1))); 2489} 2490 2491void CheckPrintfHandler::HandleIgnoredFlag( 2492 const analyze_printf::PrintfSpecifier &FS, 2493 const analyze_printf::OptionalFlag &ignoredFlag, 2494 const analyze_printf::OptionalFlag &flag, 2495 const char *startSpecifier, 2496 unsigned specifierLen) { 2497 // Warn about ignored flag with a fixit removal. 2498 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag) 2499 << ignoredFlag.toString() << flag.toString(), 2500 getLocationOfByte(ignoredFlag.getPosition()), 2501 /*IsStringLocation*/true, 2502 getSpecifierRange(startSpecifier, specifierLen), 2503 FixItHint::CreateRemoval( 2504 getSpecifierRange(ignoredFlag.getPosition(), 1))); 2505} 2506 2507// Determines if the specified is a C++ class or struct containing 2508// a member with the specified name and kind (e.g. a CXXMethodDecl named 2509// "c_str()"). 2510template<typename MemberKind> 2511static llvm::SmallPtrSet<MemberKind*, 1> 2512CXXRecordMembersNamed(StringRef Name, Sema &S, QualType Ty) { 2513 const RecordType *RT = Ty->getAs<RecordType>(); 2514 llvm::SmallPtrSet<MemberKind*, 1> Results; 2515 2516 if (!RT) 2517 return Results; 2518 const CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(RT->getDecl()); 2519 if (!RD) 2520 return Results; 2521 2522 LookupResult R(S, &S.PP.getIdentifierTable().get(Name), SourceLocation(), 2523 Sema::LookupMemberName); 2524 2525 // We just need to include all members of the right kind turned up by the 2526 // filter, at this point. 2527 if (S.LookupQualifiedName(R, RT->getDecl())) 2528 for (LookupResult::iterator I = R.begin(), E = R.end(); I != E; ++I) { 2529 NamedDecl *decl = (*I)->getUnderlyingDecl(); 2530 if (MemberKind *FK = dyn_cast<MemberKind>(decl)) 2531 Results.insert(FK); 2532 } 2533 return Results; 2534} 2535 2536// Check if a (w)string was passed when a (w)char* was needed, and offer a 2537// better diagnostic if so. AT is assumed to be valid. 2538// Returns true when a c_str() conversion method is found. 2539bool CheckPrintfHandler::checkForCStrMembers( 2540 const analyze_printf::ArgType &AT, const Expr *E, 2541 const CharSourceRange &CSR) { 2542 typedef llvm::SmallPtrSet<CXXMethodDecl*, 1> MethodSet; 2543 2544 MethodSet Results = 2545 CXXRecordMembersNamed<CXXMethodDecl>("c_str", S, E->getType()); 2546 2547 for (MethodSet::iterator MI = Results.begin(), ME = Results.end(); 2548 MI != ME; ++MI) { 2549 const CXXMethodDecl *Method = *MI; 2550 if (Method->getNumParams() == 0 && 2551 AT.matchesType(S.Context, Method->getResultType())) { 2552 // FIXME: Suggest parens if the expression needs them. 2553 SourceLocation EndLoc = 2554 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()); 2555 S.Diag(E->getLocStart(), diag::note_printf_c_str) 2556 << "c_str()" 2557 << FixItHint::CreateInsertion(EndLoc, ".c_str()"); 2558 return true; 2559 } 2560 } 2561 2562 return false; 2563} 2564 2565bool 2566CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 2567 &FS, 2568 const char *startSpecifier, 2569 unsigned specifierLen) { 2570 2571 using namespace analyze_format_string; 2572 using namespace analyze_printf; 2573 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 2574 2575 if (FS.consumesDataArgument()) { 2576 if (atFirstArg) { 2577 atFirstArg = false; 2578 usesPositionalArgs = FS.usesPositionalArg(); 2579 } 2580 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2581 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2582 startSpecifier, specifierLen); 2583 return false; 2584 } 2585 } 2586 2587 // First check if the field width, precision, and conversion specifier 2588 // have matching data arguments. 2589 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 2590 startSpecifier, specifierLen)) { 2591 return false; 2592 } 2593 2594 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 2595 startSpecifier, specifierLen)) { 2596 return false; 2597 } 2598 2599 if (!CS.consumesDataArgument()) { 2600 // FIXME: Technically specifying a precision or field width here 2601 // makes no sense. Worth issuing a warning at some point. 2602 return true; 2603 } 2604 2605 // Consume the argument. 2606 unsigned argIndex = FS.getArgIndex(); 2607 if (argIndex < NumDataArgs) { 2608 // The check to see if the argIndex is valid will come later. 2609 // We set the bit here because we may exit early from this 2610 // function if we encounter some other error. 2611 CoveredArgs.set(argIndex); 2612 } 2613 2614 // FreeBSD extensions 2615 if (CS.getKind() == ConversionSpecifier::FreeBSDbArg || 2616 CS.getKind() == ConversionSpecifier::FreeBSDDArg) { 2617 // claim the second argument 2618 CoveredArgs.set(argIndex + 1); 2619 2620 // Now type check the data expression that matches the 2621 // format specifier. 2622 const Expr *Ex = getDataArg(argIndex); 2623 const analyze_printf::ArgType &AT = 2624 (CS.getKind() == ConversionSpecifier::FreeBSDbArg) ? 2625 ArgType(S.Context.IntTy) : ArgType::CStrTy; 2626 if (AT.isValid() && !AT.matchesType(S.Context, Ex->getType())) 2627 S.Diag(getLocationOfByte(CS.getStart()), 2628 diag::warn_printf_conversion_argument_type_mismatch) 2629 << AT.getRepresentativeType(S.Context) << Ex->getType() 2630 << getSpecifierRange(startSpecifier, specifierLen) 2631 << Ex->getSourceRange(); 2632 2633 // Now type check the data expression that matches the 2634 // format specifier. 2635 Ex = getDataArg(argIndex + 1); 2636 const analyze_printf::ArgType &AT2 = ArgType::CStrTy; 2637 if (AT2.isValid() && !AT2.matchesType(S.Context, Ex->getType())) 2638 S.Diag(getLocationOfByte(CS.getStart()), 2639 diag::warn_printf_conversion_argument_type_mismatch) 2640 << AT2.getRepresentativeType(S.Context) << Ex->getType() 2641 << getSpecifierRange(startSpecifier, specifierLen) 2642 << Ex->getSourceRange(); 2643 2644 return true; 2645 } 2646 // END OF FREEBSD EXTENSIONS 2647 2648 // Check for using an Objective-C specific conversion specifier 2649 // in a non-ObjC literal. 2650 if (!ObjCContext && CS.isObjCArg()) { 2651 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 2652 specifierLen); 2653 } 2654 2655 // Check for invalid use of field width 2656 if (!FS.hasValidFieldWidth()) { 2657 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 2658 startSpecifier, specifierLen); 2659 } 2660 2661 // Check for invalid use of precision 2662 if (!FS.hasValidPrecision()) { 2663 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 2664 startSpecifier, specifierLen); 2665 } 2666 2667 // Check each flag does not conflict with any other component. 2668 if (!FS.hasValidThousandsGroupingPrefix()) 2669 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 2670 if (!FS.hasValidLeadingZeros()) 2671 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 2672 if (!FS.hasValidPlusPrefix()) 2673 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 2674 if (!FS.hasValidSpacePrefix()) 2675 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 2676 if (!FS.hasValidAlternativeForm()) 2677 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 2678 if (!FS.hasValidLeftJustified()) 2679 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 2680 2681 // Check that flags are not ignored by another flag 2682 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 2683 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 2684 startSpecifier, specifierLen); 2685 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 2686 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 2687 startSpecifier, specifierLen); 2688 2689 // Check the length modifier is valid with the given conversion specifier. 2690 if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo())) 2691 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 2692 diag::warn_format_nonsensical_length); 2693 else if (!FS.hasStandardLengthModifier()) 2694 HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen); 2695 else if (!FS.hasStandardLengthConversionCombination()) 2696 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 2697 diag::warn_format_non_standard_conversion_spec); 2698 2699 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 2700 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 2701 2702 // The remaining checks depend on the data arguments. 2703 if (HasVAListArg) 2704 return true; 2705 2706 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2707 return false; 2708 2709 const Expr *Arg = getDataArg(argIndex); 2710 if (!Arg) 2711 return true; 2712 2713 return checkFormatExpr(FS, startSpecifier, specifierLen, Arg); 2714} 2715 2716static bool requiresParensToAddCast(const Expr *E) { 2717 // FIXME: We should have a general way to reason about operator 2718 // precedence and whether parens are actually needed here. 2719 // Take care of a few common cases where they aren't. 2720 const Expr *Inside = E->IgnoreImpCasts(); 2721 if (const PseudoObjectExpr *POE = dyn_cast<PseudoObjectExpr>(Inside)) 2722 Inside = POE->getSyntacticForm()->IgnoreImpCasts(); 2723 2724 switch (Inside->getStmtClass()) { 2725 case Stmt::ArraySubscriptExprClass: 2726 case Stmt::CallExprClass: 2727 case Stmt::DeclRefExprClass: 2728 case Stmt::MemberExprClass: 2729 case Stmt::ObjCIvarRefExprClass: 2730 case Stmt::ObjCMessageExprClass: 2731 case Stmt::ObjCPropertyRefExprClass: 2732 case Stmt::ParenExprClass: 2733 case Stmt::UnaryOperatorClass: 2734 return false; 2735 default: 2736 return true; 2737 } 2738} 2739 2740bool 2741CheckPrintfHandler::checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 2742 const char *StartSpecifier, 2743 unsigned SpecifierLen, 2744 const Expr *E) { 2745 using namespace analyze_format_string; 2746 using namespace analyze_printf; 2747 // Now type check the data expression that matches the 2748 // format specifier. 2749 const analyze_printf::ArgType &AT = FS.getArgType(S.Context, 2750 ObjCContext); 2751 if (!AT.isValid()) 2752 return true; 2753 2754 QualType IntendedTy = E->getType(); 2755 if (AT.matchesType(S.Context, IntendedTy)) 2756 return true; 2757 2758 // Look through argument promotions for our error message's reported type. 2759 // This includes the integral and floating promotions, but excludes array 2760 // and function pointer decay; seeing that an argument intended to be a 2761 // string has type 'char [6]' is probably more confusing than 'char *'. 2762 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 2763 if (ICE->getCastKind() == CK_IntegralCast || 2764 ICE->getCastKind() == CK_FloatingCast) { 2765 E = ICE->getSubExpr(); 2766 IntendedTy = E->getType(); 2767 2768 // Check if we didn't match because of an implicit cast from a 'char' 2769 // or 'short' to an 'int'. This is done because printf is a varargs 2770 // function. 2771 if (ICE->getType() == S.Context.IntTy || 2772 ICE->getType() == S.Context.UnsignedIntTy) { 2773 // All further checking is done on the subexpression. 2774 if (AT.matchesType(S.Context, IntendedTy)) 2775 return true; 2776 } 2777 } 2778 } 2779 2780 if (S.Context.getTargetInfo().getTriple().isOSDarwin()) { 2781 // Special-case some of Darwin's platform-independence types. 2782 if (const TypedefType *UserTy = IntendedTy->getAs<TypedefType>()) { 2783 StringRef Name = UserTy->getDecl()->getName(); 2784 IntendedTy = llvm::StringSwitch<QualType>(Name) 2785 .Case("NSInteger", S.Context.LongTy) 2786 .Case("NSUInteger", S.Context.UnsignedLongTy) 2787 .Case("SInt32", S.Context.IntTy) 2788 .Case("UInt32", S.Context.UnsignedIntTy) 2789 .Default(IntendedTy); 2790 } 2791 } 2792 2793 // We may be able to offer a FixItHint if it is a supported type. 2794 PrintfSpecifier fixedFS = FS; 2795 bool success = fixedFS.fixType(IntendedTy, S.getLangOpts(), 2796 S.Context, ObjCContext); 2797 2798 if (success) { 2799 // Get the fix string from the fixed format specifier 2800 SmallString<16> buf; 2801 llvm::raw_svector_ostream os(buf); 2802 fixedFS.toString(os); 2803 2804 CharSourceRange SpecRange = getSpecifierRange(StartSpecifier, SpecifierLen); 2805 2806 if (IntendedTy != E->getType()) { 2807 // The canonical type for formatting this value is different from the 2808 // actual type of the expression. (This occurs, for example, with Darwin's 2809 // NSInteger on 32-bit platforms, where it is typedef'd as 'int', but 2810 // should be printed as 'long' for 64-bit compatibility.) 2811 // Rather than emitting a normal format/argument mismatch, we want to 2812 // add a cast to the recommended type (and correct the format string 2813 // if necessary). 2814 SmallString<16> CastBuf; 2815 llvm::raw_svector_ostream CastFix(CastBuf); 2816 CastFix << "("; 2817 IntendedTy.print(CastFix, S.Context.getPrintingPolicy()); 2818 CastFix << ")"; 2819 2820 SmallVector<FixItHint,4> Hints; 2821 if (!AT.matchesType(S.Context, IntendedTy)) 2822 Hints.push_back(FixItHint::CreateReplacement(SpecRange, os.str())); 2823 2824 if (const CStyleCastExpr *CCast = dyn_cast<CStyleCastExpr>(E)) { 2825 // If there's already a cast present, just replace it. 2826 SourceRange CastRange(CCast->getLParenLoc(), CCast->getRParenLoc()); 2827 Hints.push_back(FixItHint::CreateReplacement(CastRange, CastFix.str())); 2828 2829 } else if (!requiresParensToAddCast(E)) { 2830 // If the expression has high enough precedence, 2831 // just write the C-style cast. 2832 Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(), 2833 CastFix.str())); 2834 } else { 2835 // Otherwise, add parens around the expression as well as the cast. 2836 CastFix << "("; 2837 Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(), 2838 CastFix.str())); 2839 2840 SourceLocation After = S.PP.getLocForEndOfToken(E->getLocEnd()); 2841 Hints.push_back(FixItHint::CreateInsertion(After, ")")); 2842 } 2843 2844 // We extract the name from the typedef because we don't want to show 2845 // the underlying type in the diagnostic. 2846 const TypedefType *UserTy = cast<TypedefType>(E->getType()); 2847 StringRef Name = UserTy->getDecl()->getName(); 2848 2849 // Finally, emit the diagnostic. 2850 EmitFormatDiagnostic(S.PDiag(diag::warn_format_argument_needs_cast) 2851 << Name << IntendedTy 2852 << E->getSourceRange(), 2853 E->getLocStart(), /*IsStringLocation=*/false, 2854 SpecRange, Hints); 2855 } else { 2856 EmitFormatDiagnostic( 2857 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2858 << AT.getRepresentativeTypeName(S.Context) << IntendedTy 2859 << E->getSourceRange(), 2860 E->getLocStart(), 2861 /*IsStringLocation*/false, 2862 SpecRange, 2863 FixItHint::CreateReplacement(SpecRange, os.str())); 2864 } 2865 } else { 2866 const CharSourceRange &CSR = getSpecifierRange(StartSpecifier, 2867 SpecifierLen); 2868 // Since the warning for passing non-POD types to variadic functions 2869 // was deferred until now, we emit a warning for non-POD 2870 // arguments here. 2871 if (S.isValidVarArgType(E->getType()) == Sema::VAK_Invalid) { 2872 unsigned DiagKind; 2873 if (E->getType()->isObjCObjectType()) 2874 DiagKind = diag::err_cannot_pass_objc_interface_to_vararg_format; 2875 else 2876 DiagKind = diag::warn_non_pod_vararg_with_format_string; 2877 2878 EmitFormatDiagnostic( 2879 S.PDiag(DiagKind) 2880 << S.getLangOpts().CPlusPlus0x 2881 << E->getType() 2882 << CallType 2883 << AT.getRepresentativeTypeName(S.Context) 2884 << CSR 2885 << E->getSourceRange(), 2886 E->getLocStart(), /*IsStringLocation*/false, CSR); 2887 2888 checkForCStrMembers(AT, E, CSR); 2889 } else 2890 EmitFormatDiagnostic( 2891 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2892 << AT.getRepresentativeTypeName(S.Context) << E->getType() 2893 << CSR 2894 << E->getSourceRange(), 2895 E->getLocStart(), /*IsStringLocation*/false, CSR); 2896 } 2897 2898 return true; 2899} 2900 2901//===--- CHECK: Scanf format string checking ------------------------------===// 2902 2903namespace { 2904class CheckScanfHandler : public CheckFormatHandler { 2905public: 2906 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 2907 const Expr *origFormatExpr, unsigned firstDataArg, 2908 unsigned numDataArgs, const char *beg, bool hasVAListArg, 2909 Expr **Args, unsigned NumArgs, 2910 unsigned formatIdx, bool inFunctionCall, 2911 Sema::VariadicCallType CallType) 2912 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2913 numDataArgs, beg, hasVAListArg, 2914 Args, NumArgs, formatIdx, inFunctionCall, CallType) 2915 {} 2916 2917 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 2918 const char *startSpecifier, 2919 unsigned specifierLen); 2920 2921 bool HandleInvalidScanfConversionSpecifier( 2922 const analyze_scanf::ScanfSpecifier &FS, 2923 const char *startSpecifier, 2924 unsigned specifierLen); 2925 2926 void HandleIncompleteScanList(const char *start, const char *end); 2927}; 2928} 2929 2930void CheckScanfHandler::HandleIncompleteScanList(const char *start, 2931 const char *end) { 2932 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 2933 getLocationOfByte(end), /*IsStringLocation*/true, 2934 getSpecifierRange(start, end - start)); 2935} 2936 2937bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 2938 const analyze_scanf::ScanfSpecifier &FS, 2939 const char *startSpecifier, 2940 unsigned specifierLen) { 2941 2942 const analyze_scanf::ScanfConversionSpecifier &CS = 2943 FS.getConversionSpecifier(); 2944 2945 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2946 getLocationOfByte(CS.getStart()), 2947 startSpecifier, specifierLen, 2948 CS.getStart(), CS.getLength()); 2949} 2950 2951bool CheckScanfHandler::HandleScanfSpecifier( 2952 const analyze_scanf::ScanfSpecifier &FS, 2953 const char *startSpecifier, 2954 unsigned specifierLen) { 2955 2956 using namespace analyze_scanf; 2957 using namespace analyze_format_string; 2958 2959 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 2960 2961 // Handle case where '%' and '*' don't consume an argument. These shouldn't 2962 // be used to decide if we are using positional arguments consistently. 2963 if (FS.consumesDataArgument()) { 2964 if (atFirstArg) { 2965 atFirstArg = false; 2966 usesPositionalArgs = FS.usesPositionalArg(); 2967 } 2968 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2969 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2970 startSpecifier, specifierLen); 2971 return false; 2972 } 2973 } 2974 2975 // Check if the field with is non-zero. 2976 const OptionalAmount &Amt = FS.getFieldWidth(); 2977 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 2978 if (Amt.getConstantAmount() == 0) { 2979 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 2980 Amt.getConstantLength()); 2981 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 2982 getLocationOfByte(Amt.getStart()), 2983 /*IsStringLocation*/true, R, 2984 FixItHint::CreateRemoval(R)); 2985 } 2986 } 2987 2988 if (!FS.consumesDataArgument()) { 2989 // FIXME: Technically specifying a precision or field width here 2990 // makes no sense. Worth issuing a warning at some point. 2991 return true; 2992 } 2993 2994 // Consume the argument. 2995 unsigned argIndex = FS.getArgIndex(); 2996 if (argIndex < NumDataArgs) { 2997 // The check to see if the argIndex is valid will come later. 2998 // We set the bit here because we may exit early from this 2999 // function if we encounter some other error. 3000 CoveredArgs.set(argIndex); 3001 } 3002 3003 // Check the length modifier is valid with the given conversion specifier. 3004 if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo())) 3005 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3006 diag::warn_format_nonsensical_length); 3007 else if (!FS.hasStandardLengthModifier()) 3008 HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen); 3009 else if (!FS.hasStandardLengthConversionCombination()) 3010 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3011 diag::warn_format_non_standard_conversion_spec); 3012 3013 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 3014 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 3015 3016 // The remaining checks depend on the data arguments. 3017 if (HasVAListArg) 3018 return true; 3019 3020 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 3021 return false; 3022 3023 // Check that the argument type matches the format specifier. 3024 const Expr *Ex = getDataArg(argIndex); 3025 if (!Ex) 3026 return true; 3027 3028 const analyze_format_string::ArgType &AT = FS.getArgType(S.Context); 3029 if (AT.isValid() && !AT.matchesType(S.Context, Ex->getType())) { 3030 ScanfSpecifier fixedFS = FS; 3031 bool success = fixedFS.fixType(Ex->getType(), S.getLangOpts(), 3032 S.Context); 3033 3034 if (success) { 3035 // Get the fix string from the fixed format specifier. 3036 SmallString<128> buf; 3037 llvm::raw_svector_ostream os(buf); 3038 fixedFS.toString(os); 3039 3040 EmitFormatDiagnostic( 3041 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3042 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() 3043 << Ex->getSourceRange(), 3044 Ex->getLocStart(), 3045 /*IsStringLocation*/false, 3046 getSpecifierRange(startSpecifier, specifierLen), 3047 FixItHint::CreateReplacement( 3048 getSpecifierRange(startSpecifier, specifierLen), 3049 os.str())); 3050 } else { 3051 EmitFormatDiagnostic( 3052 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3053 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() 3054 << Ex->getSourceRange(), 3055 Ex->getLocStart(), 3056 /*IsStringLocation*/false, 3057 getSpecifierRange(startSpecifier, specifierLen)); 3058 } 3059 } 3060 3061 return true; 3062} 3063 3064void Sema::CheckFormatString(const StringLiteral *FExpr, 3065 const Expr *OrigFormatExpr, 3066 Expr **Args, unsigned NumArgs, 3067 bool HasVAListArg, unsigned format_idx, 3068 unsigned firstDataArg, FormatStringType Type, 3069 bool inFunctionCall, VariadicCallType CallType) { 3070 3071 // CHECK: is the format string a wide literal? 3072 if (!FExpr->isAscii() && !FExpr->isUTF8()) { 3073 CheckFormatHandler::EmitFormatDiagnostic( 3074 *this, inFunctionCall, Args[format_idx], 3075 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 3076 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 3077 return; 3078 } 3079 3080 // Str - The format string. NOTE: this is NOT null-terminated! 3081 StringRef StrRef = FExpr->getString(); 3082 const char *Str = StrRef.data(); 3083 unsigned StrLen = StrRef.size(); 3084 const unsigned numDataArgs = NumArgs - firstDataArg; 3085 3086 // CHECK: empty format string? 3087 if (StrLen == 0 && numDataArgs > 0) { 3088 CheckFormatHandler::EmitFormatDiagnostic( 3089 *this, inFunctionCall, Args[format_idx], 3090 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 3091 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 3092 return; 3093 } 3094 3095 if (Type == FST_Printf || Type == FST_NSString) { 3096 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 3097 numDataArgs, (Type == FST_NSString), 3098 Str, HasVAListArg, Args, NumArgs, format_idx, 3099 inFunctionCall, CallType); 3100 3101 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen, 3102 getLangOpts(), 3103 Context.getTargetInfo())) 3104 H.DoneProcessing(); 3105 } else if (Type == FST_Scanf) { 3106 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, numDataArgs, 3107 Str, HasVAListArg, Args, NumArgs, format_idx, 3108 inFunctionCall, CallType); 3109 3110 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen, 3111 getLangOpts(), 3112 Context.getTargetInfo())) 3113 H.DoneProcessing(); 3114 } // TODO: handle other formats 3115} 3116 3117//===--- CHECK: Standard memory functions ---------------------------------===// 3118 3119/// \brief Determine whether the given type is a dynamic class type (e.g., 3120/// whether it has a vtable). 3121static bool isDynamicClassType(QualType T) { 3122 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 3123 if (CXXRecordDecl *Definition = Record->getDefinition()) 3124 if (Definition->isDynamicClass()) 3125 return true; 3126 3127 return false; 3128} 3129 3130/// \brief If E is a sizeof expression, returns its argument expression, 3131/// otherwise returns NULL. 3132static const Expr *getSizeOfExprArg(const Expr* E) { 3133 if (const UnaryExprOrTypeTraitExpr *SizeOf = 3134 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 3135 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 3136 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 3137 3138 return 0; 3139} 3140 3141/// \brief If E is a sizeof expression, returns its argument type. 3142static QualType getSizeOfArgType(const Expr* E) { 3143 if (const UnaryExprOrTypeTraitExpr *SizeOf = 3144 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 3145 if (SizeOf->getKind() == clang::UETT_SizeOf) 3146 return SizeOf->getTypeOfArgument(); 3147 3148 return QualType(); 3149} 3150 3151/// \brief Check for dangerous or invalid arguments to memset(). 3152/// 3153/// This issues warnings on known problematic, dangerous or unspecified 3154/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 3155/// function calls. 3156/// 3157/// \param Call The call expression to diagnose. 3158void Sema::CheckMemaccessArguments(const CallExpr *Call, 3159 unsigned BId, 3160 IdentifierInfo *FnName) { 3161 assert(BId != 0); 3162 3163 // It is possible to have a non-standard definition of memset. Validate 3164 // we have enough arguments, and if not, abort further checking. 3165 unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3); 3166 if (Call->getNumArgs() < ExpectedNumArgs) 3167 return; 3168 3169 unsigned LastArg = (BId == Builtin::BImemset || 3170 BId == Builtin::BIstrndup ? 1 : 2); 3171 unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2); 3172 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 3173 3174 // We have special checking when the length is a sizeof expression. 3175 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 3176 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 3177 llvm::FoldingSetNodeID SizeOfArgID; 3178 3179 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 3180 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 3181 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 3182 3183 QualType DestTy = Dest->getType(); 3184 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 3185 QualType PointeeTy = DestPtrTy->getPointeeType(); 3186 3187 // Never warn about void type pointers. This can be used to suppress 3188 // false positives. 3189 if (PointeeTy->isVoidType()) 3190 continue; 3191 3192 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 3193 // actually comparing the expressions for equality. Because computing the 3194 // expression IDs can be expensive, we only do this if the diagnostic is 3195 // enabled. 3196 if (SizeOfArg && 3197 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 3198 SizeOfArg->getExprLoc())) { 3199 // We only compute IDs for expressions if the warning is enabled, and 3200 // cache the sizeof arg's ID. 3201 if (SizeOfArgID == llvm::FoldingSetNodeID()) 3202 SizeOfArg->Profile(SizeOfArgID, Context, true); 3203 llvm::FoldingSetNodeID DestID; 3204 Dest->Profile(DestID, Context, true); 3205 if (DestID == SizeOfArgID) { 3206 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 3207 // over sizeof(src) as well. 3208 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 3209 StringRef ReadableName = FnName->getName(); 3210 3211 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 3212 if (UnaryOp->getOpcode() == UO_AddrOf) 3213 ActionIdx = 1; // If its an address-of operator, just remove it. 3214 if (Context.getTypeSize(PointeeTy) == Context.getCharWidth()) 3215 ActionIdx = 2; // If the pointee's size is sizeof(char), 3216 // suggest an explicit length. 3217 3218 // If the function is defined as a builtin macro, do not show macro 3219 // expansion. 3220 SourceLocation SL = SizeOfArg->getExprLoc(); 3221 SourceRange DSR = Dest->getSourceRange(); 3222 SourceRange SSR = SizeOfArg->getSourceRange(); 3223 SourceManager &SM = PP.getSourceManager(); 3224 3225 if (SM.isMacroArgExpansion(SL)) { 3226 ReadableName = Lexer::getImmediateMacroName(SL, SM, LangOpts); 3227 SL = SM.getSpellingLoc(SL); 3228 DSR = SourceRange(SM.getSpellingLoc(DSR.getBegin()), 3229 SM.getSpellingLoc(DSR.getEnd())); 3230 SSR = SourceRange(SM.getSpellingLoc(SSR.getBegin()), 3231 SM.getSpellingLoc(SSR.getEnd())); 3232 } 3233 3234 DiagRuntimeBehavior(SL, SizeOfArg, 3235 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 3236 << ReadableName 3237 << PointeeTy 3238 << DestTy 3239 << DSR 3240 << SSR); 3241 DiagRuntimeBehavior(SL, SizeOfArg, 3242 PDiag(diag::warn_sizeof_pointer_expr_memaccess_note) 3243 << ActionIdx 3244 << SSR); 3245 3246 break; 3247 } 3248 } 3249 3250 // Also check for cases where the sizeof argument is the exact same 3251 // type as the memory argument, and where it points to a user-defined 3252 // record type. 3253 if (SizeOfArgTy != QualType()) { 3254 if (PointeeTy->isRecordType() && 3255 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 3256 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 3257 PDiag(diag::warn_sizeof_pointer_type_memaccess) 3258 << FnName << SizeOfArgTy << ArgIdx 3259 << PointeeTy << Dest->getSourceRange() 3260 << LenExpr->getSourceRange()); 3261 break; 3262 } 3263 } 3264 3265 // Always complain about dynamic classes. 3266 if (isDynamicClassType(PointeeTy)) { 3267 3268 unsigned OperationType = 0; 3269 // "overwritten" if we're warning about the destination for any call 3270 // but memcmp; otherwise a verb appropriate to the call. 3271 if (ArgIdx != 0 || BId == Builtin::BImemcmp) { 3272 if (BId == Builtin::BImemcpy) 3273 OperationType = 1; 3274 else if(BId == Builtin::BImemmove) 3275 OperationType = 2; 3276 else if (BId == Builtin::BImemcmp) 3277 OperationType = 3; 3278 } 3279 3280 DiagRuntimeBehavior( 3281 Dest->getExprLoc(), Dest, 3282 PDiag(diag::warn_dyn_class_memaccess) 3283 << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx) 3284 << FnName << PointeeTy 3285 << OperationType 3286 << Call->getCallee()->getSourceRange()); 3287 } else if (PointeeTy.hasNonTrivialObjCLifetime() && 3288 BId != Builtin::BImemset) 3289 DiagRuntimeBehavior( 3290 Dest->getExprLoc(), Dest, 3291 PDiag(diag::warn_arc_object_memaccess) 3292 << ArgIdx << FnName << PointeeTy 3293 << Call->getCallee()->getSourceRange()); 3294 else 3295 continue; 3296 3297 DiagRuntimeBehavior( 3298 Dest->getExprLoc(), Dest, 3299 PDiag(diag::note_bad_memaccess_silence) 3300 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 3301 break; 3302 } 3303 } 3304} 3305 3306// A little helper routine: ignore addition and subtraction of integer literals. 3307// This intentionally does not ignore all integer constant expressions because 3308// we don't want to remove sizeof(). 3309static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 3310 Ex = Ex->IgnoreParenCasts(); 3311 3312 for (;;) { 3313 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 3314 if (!BO || !BO->isAdditiveOp()) 3315 break; 3316 3317 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 3318 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 3319 3320 if (isa<IntegerLiteral>(RHS)) 3321 Ex = LHS; 3322 else if (isa<IntegerLiteral>(LHS)) 3323 Ex = RHS; 3324 else 3325 break; 3326 } 3327 3328 return Ex; 3329} 3330 3331static bool isConstantSizeArrayWithMoreThanOneElement(QualType Ty, 3332 ASTContext &Context) { 3333 // Only handle constant-sized or VLAs, but not flexible members. 3334 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(Ty)) { 3335 // Only issue the FIXIT for arrays of size > 1. 3336 if (CAT->getSize().getSExtValue() <= 1) 3337 return false; 3338 } else if (!Ty->isVariableArrayType()) { 3339 return false; 3340 } 3341 return true; 3342} 3343 3344// Warn if the user has made the 'size' argument to strlcpy or strlcat 3345// be the size of the source, instead of the destination. 3346void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 3347 IdentifierInfo *FnName) { 3348 3349 // Don't crash if the user has the wrong number of arguments 3350 if (Call->getNumArgs() != 3) 3351 return; 3352 3353 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 3354 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 3355 const Expr *CompareWithSrc = NULL; 3356 3357 // Look for 'strlcpy(dst, x, sizeof(x))' 3358 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 3359 CompareWithSrc = Ex; 3360 else { 3361 // Look for 'strlcpy(dst, x, strlen(x))' 3362 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 3363 if (SizeCall->isBuiltinCall() == Builtin::BIstrlen 3364 && SizeCall->getNumArgs() == 1) 3365 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 3366 } 3367 } 3368 3369 if (!CompareWithSrc) 3370 return; 3371 3372 // Determine if the argument to sizeof/strlen is equal to the source 3373 // argument. In principle there's all kinds of things you could do 3374 // here, for instance creating an == expression and evaluating it with 3375 // EvaluateAsBooleanCondition, but this uses a more direct technique: 3376 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 3377 if (!SrcArgDRE) 3378 return; 3379 3380 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 3381 if (!CompareWithSrcDRE || 3382 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 3383 return; 3384 3385 const Expr *OriginalSizeArg = Call->getArg(2); 3386 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 3387 << OriginalSizeArg->getSourceRange() << FnName; 3388 3389 // Output a FIXIT hint if the destination is an array (rather than a 3390 // pointer to an array). This could be enhanced to handle some 3391 // pointers if we know the actual size, like if DstArg is 'array+2' 3392 // we could say 'sizeof(array)-2'. 3393 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 3394 if (!isConstantSizeArrayWithMoreThanOneElement(DstArg->getType(), Context)) 3395 return; 3396 3397 SmallString<128> sizeString; 3398 llvm::raw_svector_ostream OS(sizeString); 3399 OS << "sizeof("; 3400 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3401 OS << ")"; 3402 3403 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 3404 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 3405 OS.str()); 3406} 3407 3408/// Check if two expressions refer to the same declaration. 3409static bool referToTheSameDecl(const Expr *E1, const Expr *E2) { 3410 if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1)) 3411 if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2)) 3412 return D1->getDecl() == D2->getDecl(); 3413 return false; 3414} 3415 3416static const Expr *getStrlenExprArg(const Expr *E) { 3417 if (const CallExpr *CE = dyn_cast<CallExpr>(E)) { 3418 const FunctionDecl *FD = CE->getDirectCallee(); 3419 if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen) 3420 return 0; 3421 return CE->getArg(0)->IgnoreParenCasts(); 3422 } 3423 return 0; 3424} 3425 3426// Warn on anti-patterns as the 'size' argument to strncat. 3427// The correct size argument should look like following: 3428// strncat(dst, src, sizeof(dst) - strlen(dest) - 1); 3429void Sema::CheckStrncatArguments(const CallExpr *CE, 3430 IdentifierInfo *FnName) { 3431 // Don't crash if the user has the wrong number of arguments. 3432 if (CE->getNumArgs() < 3) 3433 return; 3434 const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts(); 3435 const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts(); 3436 const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts(); 3437 3438 // Identify common expressions, which are wrongly used as the size argument 3439 // to strncat and may lead to buffer overflows. 3440 unsigned PatternType = 0; 3441 if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) { 3442 // - sizeof(dst) 3443 if (referToTheSameDecl(SizeOfArg, DstArg)) 3444 PatternType = 1; 3445 // - sizeof(src) 3446 else if (referToTheSameDecl(SizeOfArg, SrcArg)) 3447 PatternType = 2; 3448 } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) { 3449 if (BE->getOpcode() == BO_Sub) { 3450 const Expr *L = BE->getLHS()->IgnoreParenCasts(); 3451 const Expr *R = BE->getRHS()->IgnoreParenCasts(); 3452 // - sizeof(dst) - strlen(dst) 3453 if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) && 3454 referToTheSameDecl(DstArg, getStrlenExprArg(R))) 3455 PatternType = 1; 3456 // - sizeof(src) - (anything) 3457 else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L))) 3458 PatternType = 2; 3459 } 3460 } 3461 3462 if (PatternType == 0) 3463 return; 3464 3465 // Generate the diagnostic. 3466 SourceLocation SL = LenArg->getLocStart(); 3467 SourceRange SR = LenArg->getSourceRange(); 3468 SourceManager &SM = PP.getSourceManager(); 3469 3470 // If the function is defined as a builtin macro, do not show macro expansion. 3471 if (SM.isMacroArgExpansion(SL)) { 3472 SL = SM.getSpellingLoc(SL); 3473 SR = SourceRange(SM.getSpellingLoc(SR.getBegin()), 3474 SM.getSpellingLoc(SR.getEnd())); 3475 } 3476 3477 // Check if the destination is an array (rather than a pointer to an array). 3478 QualType DstTy = DstArg->getType(); 3479 bool isKnownSizeArray = isConstantSizeArrayWithMoreThanOneElement(DstTy, 3480 Context); 3481 if (!isKnownSizeArray) { 3482 if (PatternType == 1) 3483 Diag(SL, diag::warn_strncat_wrong_size) << SR; 3484 else 3485 Diag(SL, diag::warn_strncat_src_size) << SR; 3486 return; 3487 } 3488 3489 if (PatternType == 1) 3490 Diag(SL, diag::warn_strncat_large_size) << SR; 3491 else 3492 Diag(SL, diag::warn_strncat_src_size) << SR; 3493 3494 SmallString<128> sizeString; 3495 llvm::raw_svector_ostream OS(sizeString); 3496 OS << "sizeof("; 3497 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3498 OS << ") - "; 3499 OS << "strlen("; 3500 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3501 OS << ") - 1"; 3502 3503 Diag(SL, diag::note_strncat_wrong_size) 3504 << FixItHint::CreateReplacement(SR, OS.str()); 3505} 3506 3507//===--- CHECK: Return Address of Stack Variable --------------------------===// 3508 3509static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3510 Decl *ParentDecl); 3511static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars, 3512 Decl *ParentDecl); 3513 3514/// CheckReturnStackAddr - Check if a return statement returns the address 3515/// of a stack variable. 3516void 3517Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 3518 SourceLocation ReturnLoc) { 3519 3520 Expr *stackE = 0; 3521 SmallVector<DeclRefExpr *, 8> refVars; 3522 3523 // Perform checking for returned stack addresses, local blocks, 3524 // label addresses or references to temporaries. 3525 if (lhsType->isPointerType() || 3526 (!getLangOpts().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 3527 stackE = EvalAddr(RetValExp, refVars, /*ParentDecl=*/0); 3528 } else if (lhsType->isReferenceType()) { 3529 stackE = EvalVal(RetValExp, refVars, /*ParentDecl=*/0); 3530 } 3531 3532 if (stackE == 0) 3533 return; // Nothing suspicious was found. 3534 3535 SourceLocation diagLoc; 3536 SourceRange diagRange; 3537 if (refVars.empty()) { 3538 diagLoc = stackE->getLocStart(); 3539 diagRange = stackE->getSourceRange(); 3540 } else { 3541 // We followed through a reference variable. 'stackE' contains the 3542 // problematic expression but we will warn at the return statement pointing 3543 // at the reference variable. We will later display the "trail" of 3544 // reference variables using notes. 3545 diagLoc = refVars[0]->getLocStart(); 3546 diagRange = refVars[0]->getSourceRange(); 3547 } 3548 3549 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 3550 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 3551 : diag::warn_ret_stack_addr) 3552 << DR->getDecl()->getDeclName() << diagRange; 3553 } else if (isa<BlockExpr>(stackE)) { // local block. 3554 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 3555 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 3556 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 3557 } else { // local temporary. 3558 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 3559 : diag::warn_ret_local_temp_addr) 3560 << diagRange; 3561 } 3562 3563 // Display the "trail" of reference variables that we followed until we 3564 // found the problematic expression using notes. 3565 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 3566 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 3567 // If this var binds to another reference var, show the range of the next 3568 // var, otherwise the var binds to the problematic expression, in which case 3569 // show the range of the expression. 3570 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 3571 : stackE->getSourceRange(); 3572 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 3573 << VD->getDeclName() << range; 3574 } 3575} 3576 3577/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 3578/// check if the expression in a return statement evaluates to an address 3579/// to a location on the stack, a local block, an address of a label, or a 3580/// reference to local temporary. The recursion is used to traverse the 3581/// AST of the return expression, with recursion backtracking when we 3582/// encounter a subexpression that (1) clearly does not lead to one of the 3583/// above problematic expressions (2) is something we cannot determine leads to 3584/// a problematic expression based on such local checking. 3585/// 3586/// Both EvalAddr and EvalVal follow through reference variables to evaluate 3587/// the expression that they point to. Such variables are added to the 3588/// 'refVars' vector so that we know what the reference variable "trail" was. 3589/// 3590/// EvalAddr processes expressions that are pointers that are used as 3591/// references (and not L-values). EvalVal handles all other values. 3592/// At the base case of the recursion is a check for the above problematic 3593/// expressions. 3594/// 3595/// This implementation handles: 3596/// 3597/// * pointer-to-pointer casts 3598/// * implicit conversions from array references to pointers 3599/// * taking the address of fields 3600/// * arbitrary interplay between "&" and "*" operators 3601/// * pointer arithmetic from an address of a stack variable 3602/// * taking the address of an array element where the array is on the stack 3603static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3604 Decl *ParentDecl) { 3605 if (E->isTypeDependent()) 3606 return NULL; 3607 3608 // We should only be called for evaluating pointer expressions. 3609 assert((E->getType()->isAnyPointerType() || 3610 E->getType()->isBlockPointerType() || 3611 E->getType()->isObjCQualifiedIdType()) && 3612 "EvalAddr only works on pointers"); 3613 3614 E = E->IgnoreParens(); 3615 3616 // Our "symbolic interpreter" is just a dispatch off the currently 3617 // viewed AST node. We then recursively traverse the AST by calling 3618 // EvalAddr and EvalVal appropriately. 3619 switch (E->getStmtClass()) { 3620 case Stmt::DeclRefExprClass: { 3621 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3622 3623 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 3624 // If this is a reference variable, follow through to the expression that 3625 // it points to. 3626 if (V->hasLocalStorage() && 3627 V->getType()->isReferenceType() && V->hasInit()) { 3628 // Add the reference variable to the "trail". 3629 refVars.push_back(DR); 3630 return EvalAddr(V->getInit(), refVars, ParentDecl); 3631 } 3632 3633 return NULL; 3634 } 3635 3636 case Stmt::UnaryOperatorClass: { 3637 // The only unary operator that make sense to handle here 3638 // is AddrOf. All others don't make sense as pointers. 3639 UnaryOperator *U = cast<UnaryOperator>(E); 3640 3641 if (U->getOpcode() == UO_AddrOf) 3642 return EvalVal(U->getSubExpr(), refVars, ParentDecl); 3643 else 3644 return NULL; 3645 } 3646 3647 case Stmt::BinaryOperatorClass: { 3648 // Handle pointer arithmetic. All other binary operators are not valid 3649 // in this context. 3650 BinaryOperator *B = cast<BinaryOperator>(E); 3651 BinaryOperatorKind op = B->getOpcode(); 3652 3653 if (op != BO_Add && op != BO_Sub) 3654 return NULL; 3655 3656 Expr *Base = B->getLHS(); 3657 3658 // Determine which argument is the real pointer base. It could be 3659 // the RHS argument instead of the LHS. 3660 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 3661 3662 assert (Base->getType()->isPointerType()); 3663 return EvalAddr(Base, refVars, ParentDecl); 3664 } 3665 3666 // For conditional operators we need to see if either the LHS or RHS are 3667 // valid DeclRefExpr*s. If one of them is valid, we return it. 3668 case Stmt::ConditionalOperatorClass: { 3669 ConditionalOperator *C = cast<ConditionalOperator>(E); 3670 3671 // Handle the GNU extension for missing LHS. 3672 if (Expr *lhsExpr = C->getLHS()) { 3673 // In C++, we can have a throw-expression, which has 'void' type. 3674 if (!lhsExpr->getType()->isVoidType()) 3675 if (Expr* LHS = EvalAddr(lhsExpr, refVars, ParentDecl)) 3676 return LHS; 3677 } 3678 3679 // In C++, we can have a throw-expression, which has 'void' type. 3680 if (C->getRHS()->getType()->isVoidType()) 3681 return NULL; 3682 3683 return EvalAddr(C->getRHS(), refVars, ParentDecl); 3684 } 3685 3686 case Stmt::BlockExprClass: 3687 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 3688 return E; // local block. 3689 return NULL; 3690 3691 case Stmt::AddrLabelExprClass: 3692 return E; // address of label. 3693 3694 case Stmt::ExprWithCleanupsClass: 3695 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars, 3696 ParentDecl); 3697 3698 // For casts, we need to handle conversions from arrays to 3699 // pointer values, and pointer-to-pointer conversions. 3700 case Stmt::ImplicitCastExprClass: 3701 case Stmt::CStyleCastExprClass: 3702 case Stmt::CXXFunctionalCastExprClass: 3703 case Stmt::ObjCBridgedCastExprClass: 3704 case Stmt::CXXStaticCastExprClass: 3705 case Stmt::CXXDynamicCastExprClass: 3706 case Stmt::CXXConstCastExprClass: 3707 case Stmt::CXXReinterpretCastExprClass: { 3708 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 3709 switch (cast<CastExpr>(E)->getCastKind()) { 3710 case CK_BitCast: 3711 case CK_LValueToRValue: 3712 case CK_NoOp: 3713 case CK_BaseToDerived: 3714 case CK_DerivedToBase: 3715 case CK_UncheckedDerivedToBase: 3716 case CK_Dynamic: 3717 case CK_CPointerToObjCPointerCast: 3718 case CK_BlockPointerToObjCPointerCast: 3719 case CK_AnyPointerToBlockPointerCast: 3720 return EvalAddr(SubExpr, refVars, ParentDecl); 3721 3722 case CK_ArrayToPointerDecay: 3723 return EvalVal(SubExpr, refVars, ParentDecl); 3724 3725 default: 3726 return 0; 3727 } 3728 } 3729 3730 case Stmt::MaterializeTemporaryExprClass: 3731 if (Expr *Result = EvalAddr( 3732 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3733 refVars, ParentDecl)) 3734 return Result; 3735 3736 return E; 3737 3738 // Everything else: we simply don't reason about them. 3739 default: 3740 return NULL; 3741 } 3742} 3743 3744 3745/// EvalVal - This function is complements EvalAddr in the mutual recursion. 3746/// See the comments for EvalAddr for more details. 3747static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3748 Decl *ParentDecl) { 3749do { 3750 // We should only be called for evaluating non-pointer expressions, or 3751 // expressions with a pointer type that are not used as references but instead 3752 // are l-values (e.g., DeclRefExpr with a pointer type). 3753 3754 // Our "symbolic interpreter" is just a dispatch off the currently 3755 // viewed AST node. We then recursively traverse the AST by calling 3756 // EvalAddr and EvalVal appropriately. 3757 3758 E = E->IgnoreParens(); 3759 switch (E->getStmtClass()) { 3760 case Stmt::ImplicitCastExprClass: { 3761 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 3762 if (IE->getValueKind() == VK_LValue) { 3763 E = IE->getSubExpr(); 3764 continue; 3765 } 3766 return NULL; 3767 } 3768 3769 case Stmt::ExprWithCleanupsClass: 3770 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,ParentDecl); 3771 3772 case Stmt::DeclRefExprClass: { 3773 // When we hit a DeclRefExpr we are looking at code that refers to a 3774 // variable's name. If it's not a reference variable we check if it has 3775 // local storage within the function, and if so, return the expression. 3776 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3777 3778 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) { 3779 // Check if it refers to itself, e.g. "int& i = i;". 3780 if (V == ParentDecl) 3781 return DR; 3782 3783 if (V->hasLocalStorage()) { 3784 if (!V->getType()->isReferenceType()) 3785 return DR; 3786 3787 // Reference variable, follow through to the expression that 3788 // it points to. 3789 if (V->hasInit()) { 3790 // Add the reference variable to the "trail". 3791 refVars.push_back(DR); 3792 return EvalVal(V->getInit(), refVars, V); 3793 } 3794 } 3795 } 3796 3797 return NULL; 3798 } 3799 3800 case Stmt::UnaryOperatorClass: { 3801 // The only unary operator that make sense to handle here 3802 // is Deref. All others don't resolve to a "name." This includes 3803 // handling all sorts of rvalues passed to a unary operator. 3804 UnaryOperator *U = cast<UnaryOperator>(E); 3805 3806 if (U->getOpcode() == UO_Deref) 3807 return EvalAddr(U->getSubExpr(), refVars, ParentDecl); 3808 3809 return NULL; 3810 } 3811 3812 case Stmt::ArraySubscriptExprClass: { 3813 // Array subscripts are potential references to data on the stack. We 3814 // retrieve the DeclRefExpr* for the array variable if it indeed 3815 // has local storage. 3816 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars,ParentDecl); 3817 } 3818 3819 case Stmt::ConditionalOperatorClass: { 3820 // For conditional operators we need to see if either the LHS or RHS are 3821 // non-NULL Expr's. If one is non-NULL, we return it. 3822 ConditionalOperator *C = cast<ConditionalOperator>(E); 3823 3824 // Handle the GNU extension for missing LHS. 3825 if (Expr *lhsExpr = C->getLHS()) 3826 if (Expr *LHS = EvalVal(lhsExpr, refVars, ParentDecl)) 3827 return LHS; 3828 3829 return EvalVal(C->getRHS(), refVars, ParentDecl); 3830 } 3831 3832 // Accesses to members are potential references to data on the stack. 3833 case Stmt::MemberExprClass: { 3834 MemberExpr *M = cast<MemberExpr>(E); 3835 3836 // Check for indirect access. We only want direct field accesses. 3837 if (M->isArrow()) 3838 return NULL; 3839 3840 // Check whether the member type is itself a reference, in which case 3841 // we're not going to refer to the member, but to what the member refers to. 3842 if (M->getMemberDecl()->getType()->isReferenceType()) 3843 return NULL; 3844 3845 return EvalVal(M->getBase(), refVars, ParentDecl); 3846 } 3847 3848 case Stmt::MaterializeTemporaryExprClass: 3849 if (Expr *Result = EvalVal( 3850 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3851 refVars, ParentDecl)) 3852 return Result; 3853 3854 return E; 3855 3856 default: 3857 // Check that we don't return or take the address of a reference to a 3858 // temporary. This is only useful in C++. 3859 if (!E->isTypeDependent() && E->isRValue()) 3860 return E; 3861 3862 // Everything else: we simply don't reason about them. 3863 return NULL; 3864 } 3865} while (true); 3866} 3867 3868//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 3869 3870/// Check for comparisons of floating point operands using != and ==. 3871/// Issue a warning if these are no self-comparisons, as they are not likely 3872/// to do what the programmer intended. 3873void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 3874 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 3875 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 3876 3877 // Special case: check for x == x (which is OK). 3878 // Do not emit warnings for such cases. 3879 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 3880 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 3881 if (DRL->getDecl() == DRR->getDecl()) 3882 return; 3883 3884 3885 // Special case: check for comparisons against literals that can be exactly 3886 // represented by APFloat. In such cases, do not emit a warning. This 3887 // is a heuristic: often comparison against such literals are used to 3888 // detect if a value in a variable has not changed. This clearly can 3889 // lead to false negatives. 3890 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 3891 if (FLL->isExact()) 3892 return; 3893 } else 3894 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)) 3895 if (FLR->isExact()) 3896 return; 3897 3898 // Check for comparisons with builtin types. 3899 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 3900 if (CL->isBuiltinCall()) 3901 return; 3902 3903 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 3904 if (CR->isBuiltinCall()) 3905 return; 3906 3907 // Emit the diagnostic. 3908 Diag(Loc, diag::warn_floatingpoint_eq) 3909 << LHS->getSourceRange() << RHS->getSourceRange(); 3910} 3911 3912//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 3913//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 3914 3915namespace { 3916 3917/// Structure recording the 'active' range of an integer-valued 3918/// expression. 3919struct IntRange { 3920 /// The number of bits active in the int. 3921 unsigned Width; 3922 3923 /// True if the int is known not to have negative values. 3924 bool NonNegative; 3925 3926 IntRange(unsigned Width, bool NonNegative) 3927 : Width(Width), NonNegative(NonNegative) 3928 {} 3929 3930 /// Returns the range of the bool type. 3931 static IntRange forBoolType() { 3932 return IntRange(1, true); 3933 } 3934 3935 /// Returns the range of an opaque value of the given integral type. 3936 static IntRange forValueOfType(ASTContext &C, QualType T) { 3937 return forValueOfCanonicalType(C, 3938 T->getCanonicalTypeInternal().getTypePtr()); 3939 } 3940 3941 /// Returns the range of an opaque value of a canonical integral type. 3942 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 3943 assert(T->isCanonicalUnqualified()); 3944 3945 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3946 T = VT->getElementType().getTypePtr(); 3947 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3948 T = CT->getElementType().getTypePtr(); 3949 3950 // For enum types, use the known bit width of the enumerators. 3951 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 3952 EnumDecl *Enum = ET->getDecl(); 3953 if (!Enum->isCompleteDefinition()) 3954 return IntRange(C.getIntWidth(QualType(T, 0)), false); 3955 3956 unsigned NumPositive = Enum->getNumPositiveBits(); 3957 unsigned NumNegative = Enum->getNumNegativeBits(); 3958 3959 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 3960 } 3961 3962 const BuiltinType *BT = cast<BuiltinType>(T); 3963 assert(BT->isInteger()); 3964 3965 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3966 } 3967 3968 /// Returns the "target" range of a canonical integral type, i.e. 3969 /// the range of values expressible in the type. 3970 /// 3971 /// This matches forValueOfCanonicalType except that enums have the 3972 /// full range of their type, not the range of their enumerators. 3973 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 3974 assert(T->isCanonicalUnqualified()); 3975 3976 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3977 T = VT->getElementType().getTypePtr(); 3978 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3979 T = CT->getElementType().getTypePtr(); 3980 if (const EnumType *ET = dyn_cast<EnumType>(T)) 3981 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 3982 3983 const BuiltinType *BT = cast<BuiltinType>(T); 3984 assert(BT->isInteger()); 3985 3986 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3987 } 3988 3989 /// Returns the supremum of two ranges: i.e. their conservative merge. 3990 static IntRange join(IntRange L, IntRange R) { 3991 return IntRange(std::max(L.Width, R.Width), 3992 L.NonNegative && R.NonNegative); 3993 } 3994 3995 /// Returns the infinum of two ranges: i.e. their aggressive merge. 3996 static IntRange meet(IntRange L, IntRange R) { 3997 return IntRange(std::min(L.Width, R.Width), 3998 L.NonNegative || R.NonNegative); 3999 } 4000}; 4001 4002static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, 4003 unsigned MaxWidth) { 4004 if (value.isSigned() && value.isNegative()) 4005 return IntRange(value.getMinSignedBits(), false); 4006 4007 if (value.getBitWidth() > MaxWidth) 4008 value = value.trunc(MaxWidth); 4009 4010 // isNonNegative() just checks the sign bit without considering 4011 // signedness. 4012 return IntRange(value.getActiveBits(), true); 4013} 4014 4015static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 4016 unsigned MaxWidth) { 4017 if (result.isInt()) 4018 return GetValueRange(C, result.getInt(), MaxWidth); 4019 4020 if (result.isVector()) { 4021 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 4022 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 4023 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 4024 R = IntRange::join(R, El); 4025 } 4026 return R; 4027 } 4028 4029 if (result.isComplexInt()) { 4030 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 4031 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 4032 return IntRange::join(R, I); 4033 } 4034 4035 // This can happen with lossless casts to intptr_t of "based" lvalues. 4036 // Assume it might use arbitrary bits. 4037 // FIXME: The only reason we need to pass the type in here is to get 4038 // the sign right on this one case. It would be nice if APValue 4039 // preserved this. 4040 assert(result.isLValue() || result.isAddrLabelDiff()); 4041 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 4042} 4043 4044/// Pseudo-evaluate the given integer expression, estimating the 4045/// range of values it might take. 4046/// 4047/// \param MaxWidth - the width to which the value will be truncated 4048static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 4049 E = E->IgnoreParens(); 4050 4051 // Try a full evaluation first. 4052 Expr::EvalResult result; 4053 if (E->EvaluateAsRValue(result, C)) 4054 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 4055 4056 // I think we only want to look through implicit casts here; if the 4057 // user has an explicit widening cast, we should treat the value as 4058 // being of the new, wider type. 4059 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 4060 if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue) 4061 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 4062 4063 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 4064 4065 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 4066 4067 // Assume that non-integer casts can span the full range of the type. 4068 if (!isIntegerCast) 4069 return OutputTypeRange; 4070 4071 IntRange SubRange 4072 = GetExprRange(C, CE->getSubExpr(), 4073 std::min(MaxWidth, OutputTypeRange.Width)); 4074 4075 // Bail out if the subexpr's range is as wide as the cast type. 4076 if (SubRange.Width >= OutputTypeRange.Width) 4077 return OutputTypeRange; 4078 4079 // Otherwise, we take the smaller width, and we're non-negative if 4080 // either the output type or the subexpr is. 4081 return IntRange(SubRange.Width, 4082 SubRange.NonNegative || OutputTypeRange.NonNegative); 4083 } 4084 4085 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 4086 // If we can fold the condition, just take that operand. 4087 bool CondResult; 4088 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 4089 return GetExprRange(C, CondResult ? CO->getTrueExpr() 4090 : CO->getFalseExpr(), 4091 MaxWidth); 4092 4093 // Otherwise, conservatively merge. 4094 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 4095 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 4096 return IntRange::join(L, R); 4097 } 4098 4099 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 4100 switch (BO->getOpcode()) { 4101 4102 // Boolean-valued operations are single-bit and positive. 4103 case BO_LAnd: 4104 case BO_LOr: 4105 case BO_LT: 4106 case BO_GT: 4107 case BO_LE: 4108 case BO_GE: 4109 case BO_EQ: 4110 case BO_NE: 4111 return IntRange::forBoolType(); 4112 4113 // The type of the assignments is the type of the LHS, so the RHS 4114 // is not necessarily the same type. 4115 case BO_MulAssign: 4116 case BO_DivAssign: 4117 case BO_RemAssign: 4118 case BO_AddAssign: 4119 case BO_SubAssign: 4120 case BO_XorAssign: 4121 case BO_OrAssign: 4122 // TODO: bitfields? 4123 return IntRange::forValueOfType(C, E->getType()); 4124 4125 // Simple assignments just pass through the RHS, which will have 4126 // been coerced to the LHS type. 4127 case BO_Assign: 4128 // TODO: bitfields? 4129 return GetExprRange(C, BO->getRHS(), MaxWidth); 4130 4131 // Operations with opaque sources are black-listed. 4132 case BO_PtrMemD: 4133 case BO_PtrMemI: 4134 return IntRange::forValueOfType(C, E->getType()); 4135 4136 // Bitwise-and uses the *infinum* of the two source ranges. 4137 case BO_And: 4138 case BO_AndAssign: 4139 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 4140 GetExprRange(C, BO->getRHS(), MaxWidth)); 4141 4142 // Left shift gets black-listed based on a judgement call. 4143 case BO_Shl: 4144 // ...except that we want to treat '1 << (blah)' as logically 4145 // positive. It's an important idiom. 4146 if (IntegerLiteral *I 4147 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 4148 if (I->getValue() == 1) { 4149 IntRange R = IntRange::forValueOfType(C, E->getType()); 4150 return IntRange(R.Width, /*NonNegative*/ true); 4151 } 4152 } 4153 // fallthrough 4154 4155 case BO_ShlAssign: 4156 return IntRange::forValueOfType(C, E->getType()); 4157 4158 // Right shift by a constant can narrow its left argument. 4159 case BO_Shr: 4160 case BO_ShrAssign: { 4161 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 4162 4163 // If the shift amount is a positive constant, drop the width by 4164 // that much. 4165 llvm::APSInt shift; 4166 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 4167 shift.isNonNegative()) { 4168 unsigned zext = shift.getZExtValue(); 4169 if (zext >= L.Width) 4170 L.Width = (L.NonNegative ? 0 : 1); 4171 else 4172 L.Width -= zext; 4173 } 4174 4175 return L; 4176 } 4177 4178 // Comma acts as its right operand. 4179 case BO_Comma: 4180 return GetExprRange(C, BO->getRHS(), MaxWidth); 4181 4182 // Black-list pointer subtractions. 4183 case BO_Sub: 4184 if (BO->getLHS()->getType()->isPointerType()) 4185 return IntRange::forValueOfType(C, E->getType()); 4186 break; 4187 4188 // The width of a division result is mostly determined by the size 4189 // of the LHS. 4190 case BO_Div: { 4191 // Don't 'pre-truncate' the operands. 4192 unsigned opWidth = C.getIntWidth(E->getType()); 4193 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 4194 4195 // If the divisor is constant, use that. 4196 llvm::APSInt divisor; 4197 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 4198 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 4199 if (log2 >= L.Width) 4200 L.Width = (L.NonNegative ? 0 : 1); 4201 else 4202 L.Width = std::min(L.Width - log2, MaxWidth); 4203 return L; 4204 } 4205 4206 // Otherwise, just use the LHS's width. 4207 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 4208 return IntRange(L.Width, L.NonNegative && R.NonNegative); 4209 } 4210 4211 // The result of a remainder can't be larger than the result of 4212 // either side. 4213 case BO_Rem: { 4214 // Don't 'pre-truncate' the operands. 4215 unsigned opWidth = C.getIntWidth(E->getType()); 4216 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 4217 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 4218 4219 IntRange meet = IntRange::meet(L, R); 4220 meet.Width = std::min(meet.Width, MaxWidth); 4221 return meet; 4222 } 4223 4224 // The default behavior is okay for these. 4225 case BO_Mul: 4226 case BO_Add: 4227 case BO_Xor: 4228 case BO_Or: 4229 break; 4230 } 4231 4232 // The default case is to treat the operation as if it were closed 4233 // on the narrowest type that encompasses both operands. 4234 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 4235 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 4236 return IntRange::join(L, R); 4237 } 4238 4239 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 4240 switch (UO->getOpcode()) { 4241 // Boolean-valued operations are white-listed. 4242 case UO_LNot: 4243 return IntRange::forBoolType(); 4244 4245 // Operations with opaque sources are black-listed. 4246 case UO_Deref: 4247 case UO_AddrOf: // should be impossible 4248 return IntRange::forValueOfType(C, E->getType()); 4249 4250 default: 4251 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 4252 } 4253 } 4254 4255 if (dyn_cast<OffsetOfExpr>(E)) { 4256 IntRange::forValueOfType(C, E->getType()); 4257 } 4258 4259 if (FieldDecl *BitField = E->getBitField()) 4260 return IntRange(BitField->getBitWidthValue(C), 4261 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 4262 4263 return IntRange::forValueOfType(C, E->getType()); 4264} 4265 4266static IntRange GetExprRange(ASTContext &C, Expr *E) { 4267 return GetExprRange(C, E, C.getIntWidth(E->getType())); 4268} 4269 4270/// Checks whether the given value, which currently has the given 4271/// source semantics, has the same value when coerced through the 4272/// target semantics. 4273static bool IsSameFloatAfterCast(const llvm::APFloat &value, 4274 const llvm::fltSemantics &Src, 4275 const llvm::fltSemantics &Tgt) { 4276 llvm::APFloat truncated = value; 4277 4278 bool ignored; 4279 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 4280 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 4281 4282 return truncated.bitwiseIsEqual(value); 4283} 4284 4285/// Checks whether the given value, which currently has the given 4286/// source semantics, has the same value when coerced through the 4287/// target semantics. 4288/// 4289/// The value might be a vector of floats (or a complex number). 4290static bool IsSameFloatAfterCast(const APValue &value, 4291 const llvm::fltSemantics &Src, 4292 const llvm::fltSemantics &Tgt) { 4293 if (value.isFloat()) 4294 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 4295 4296 if (value.isVector()) { 4297 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 4298 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 4299 return false; 4300 return true; 4301 } 4302 4303 assert(value.isComplexFloat()); 4304 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 4305 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 4306} 4307 4308static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 4309 4310static bool IsZero(Sema &S, Expr *E) { 4311 // Suppress cases where we are comparing against an enum constant. 4312 if (const DeclRefExpr *DR = 4313 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 4314 if (isa<EnumConstantDecl>(DR->getDecl())) 4315 return false; 4316 4317 // Suppress cases where the '0' value is expanded from a macro. 4318 if (E->getLocStart().isMacroID()) 4319 return false; 4320 4321 llvm::APSInt Value; 4322 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 4323} 4324 4325static bool HasEnumType(Expr *E) { 4326 // Strip off implicit integral promotions. 4327 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 4328 if (ICE->getCastKind() != CK_IntegralCast && 4329 ICE->getCastKind() != CK_NoOp) 4330 break; 4331 E = ICE->getSubExpr(); 4332 } 4333 4334 return E->getType()->isEnumeralType(); 4335} 4336 4337static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 4338 BinaryOperatorKind op = E->getOpcode(); 4339 if (E->isValueDependent()) 4340 return; 4341 4342 if (op == BO_LT && IsZero(S, E->getRHS())) { 4343 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4344 << "< 0" << "false" << HasEnumType(E->getLHS()) 4345 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4346 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 4347 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4348 << ">= 0" << "true" << HasEnumType(E->getLHS()) 4349 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4350 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 4351 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4352 << "0 >" << "false" << HasEnumType(E->getRHS()) 4353 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4354 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 4355 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4356 << "0 <=" << "true" << HasEnumType(E->getRHS()) 4357 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4358 } 4359} 4360 4361static void DiagnoseOutOfRangeComparison(Sema &S, BinaryOperator *E, 4362 Expr *Constant, Expr *Other, 4363 llvm::APSInt Value, 4364 bool RhsConstant) { 4365 BinaryOperatorKind op = E->getOpcode(); 4366 QualType OtherT = Other->getType(); 4367 QualType ConstantT = Constant->getType(); 4368 if (S.Context.hasSameUnqualifiedType(OtherT, ConstantT)) 4369 return; 4370 assert((OtherT->isIntegerType() && ConstantT->isIntegerType()) 4371 && "comparison with non-integer type"); 4372 // FIXME. handle cases for signedness to catch (signed char)N == 200 4373 IntRange OtherRange = IntRange::forValueOfType(S.Context, OtherT); 4374 IntRange LitRange = GetValueRange(S.Context, Value, Value.getBitWidth()); 4375 if (OtherRange.Width >= LitRange.Width) 4376 return; 4377 bool IsTrue = true; 4378 if (op == BO_EQ) 4379 IsTrue = false; 4380 else if (op == BO_NE) 4381 IsTrue = true; 4382 else if (RhsConstant) { 4383 if (op == BO_GT || op == BO_GE) 4384 IsTrue = !LitRange.NonNegative; 4385 else // op == BO_LT || op == BO_LE 4386 IsTrue = LitRange.NonNegative; 4387 } else { 4388 if (op == BO_LT || op == BO_LE) 4389 IsTrue = !LitRange.NonNegative; 4390 else // op == BO_GT || op == BO_GE 4391 IsTrue = LitRange.NonNegative; 4392 } 4393 SmallString<16> PrettySourceValue(Value.toString(10)); 4394 S.Diag(E->getOperatorLoc(), diag::warn_out_of_range_compare) 4395 << PrettySourceValue << OtherT << IsTrue 4396 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4397} 4398 4399/// Analyze the operands of the given comparison. Implements the 4400/// fallback case from AnalyzeComparison. 4401static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 4402 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4403 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4404} 4405 4406/// \brief Implements -Wsign-compare. 4407/// 4408/// \param E the binary operator to check for warnings 4409static void AnalyzeComparison(Sema &S, BinaryOperator *E) { 4410 // The type the comparison is being performed in. 4411 QualType T = E->getLHS()->getType(); 4412 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 4413 && "comparison with mismatched types"); 4414 if (E->isValueDependent()) 4415 return AnalyzeImpConvsInComparison(S, E); 4416 4417 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 4418 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 4419 4420 bool IsComparisonConstant = false; 4421 4422 // Check whether an integer constant comparison results in a value 4423 // of 'true' or 'false'. 4424 if (T->isIntegralType(S.Context)) { 4425 llvm::APSInt RHSValue; 4426 bool IsRHSIntegralLiteral = 4427 RHS->isIntegerConstantExpr(RHSValue, S.Context); 4428 llvm::APSInt LHSValue; 4429 bool IsLHSIntegralLiteral = 4430 LHS->isIntegerConstantExpr(LHSValue, S.Context); 4431 if (IsRHSIntegralLiteral && !IsLHSIntegralLiteral) 4432 DiagnoseOutOfRangeComparison(S, E, RHS, LHS, RHSValue, true); 4433 else if (!IsRHSIntegralLiteral && IsLHSIntegralLiteral) 4434 DiagnoseOutOfRangeComparison(S, E, LHS, RHS, LHSValue, false); 4435 else 4436 IsComparisonConstant = 4437 (IsRHSIntegralLiteral && IsLHSIntegralLiteral); 4438 } else if (!T->hasUnsignedIntegerRepresentation()) 4439 IsComparisonConstant = E->isIntegerConstantExpr(S.Context); 4440 4441 // We don't do anything special if this isn't an unsigned integral 4442 // comparison: we're only interested in integral comparisons, and 4443 // signed comparisons only happen in cases we don't care to warn about. 4444 // 4445 // We also don't care about value-dependent expressions or expressions 4446 // whose result is a constant. 4447 if (!T->hasUnsignedIntegerRepresentation() || IsComparisonConstant) 4448 return AnalyzeImpConvsInComparison(S, E); 4449 4450 // Check to see if one of the (unmodified) operands is of different 4451 // signedness. 4452 Expr *signedOperand, *unsignedOperand; 4453 if (LHS->getType()->hasSignedIntegerRepresentation()) { 4454 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 4455 "unsigned comparison between two signed integer expressions?"); 4456 signedOperand = LHS; 4457 unsignedOperand = RHS; 4458 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 4459 signedOperand = RHS; 4460 unsignedOperand = LHS; 4461 } else { 4462 CheckTrivialUnsignedComparison(S, E); 4463 return AnalyzeImpConvsInComparison(S, E); 4464 } 4465 4466 // Otherwise, calculate the effective range of the signed operand. 4467 IntRange signedRange = GetExprRange(S.Context, signedOperand); 4468 4469 // Go ahead and analyze implicit conversions in the operands. Note 4470 // that we skip the implicit conversions on both sides. 4471 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 4472 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 4473 4474 // If the signed range is non-negative, -Wsign-compare won't fire, 4475 // but we should still check for comparisons which are always true 4476 // or false. 4477 if (signedRange.NonNegative) 4478 return CheckTrivialUnsignedComparison(S, E); 4479 4480 // For (in)equality comparisons, if the unsigned operand is a 4481 // constant which cannot collide with a overflowed signed operand, 4482 // then reinterpreting the signed operand as unsigned will not 4483 // change the result of the comparison. 4484 if (E->isEqualityOp()) { 4485 unsigned comparisonWidth = S.Context.getIntWidth(T); 4486 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 4487 4488 // We should never be unable to prove that the unsigned operand is 4489 // non-negative. 4490 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 4491 4492 if (unsignedRange.Width < comparisonWidth) 4493 return; 4494 } 4495 4496 S.DiagRuntimeBehavior(E->getOperatorLoc(), E, 4497 S.PDiag(diag::warn_mixed_sign_comparison) 4498 << LHS->getType() << RHS->getType() 4499 << LHS->getSourceRange() << RHS->getSourceRange()); 4500} 4501 4502/// Analyzes an attempt to assign the given value to a bitfield. 4503/// 4504/// Returns true if there was something fishy about the attempt. 4505static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 4506 SourceLocation InitLoc) { 4507 assert(Bitfield->isBitField()); 4508 if (Bitfield->isInvalidDecl()) 4509 return false; 4510 4511 // White-list bool bitfields. 4512 if (Bitfield->getType()->isBooleanType()) 4513 return false; 4514 4515 // Ignore value- or type-dependent expressions. 4516 if (Bitfield->getBitWidth()->isValueDependent() || 4517 Bitfield->getBitWidth()->isTypeDependent() || 4518 Init->isValueDependent() || 4519 Init->isTypeDependent()) 4520 return false; 4521 4522 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 4523 4524 llvm::APSInt Value; 4525 if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects)) 4526 return false; 4527 4528 unsigned OriginalWidth = Value.getBitWidth(); 4529 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 4530 4531 if (OriginalWidth <= FieldWidth) 4532 return false; 4533 4534 // Compute the value which the bitfield will contain. 4535 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 4536 TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType()); 4537 4538 // Check whether the stored value is equal to the original value. 4539 TruncatedValue = TruncatedValue.extend(OriginalWidth); 4540 if (llvm::APSInt::isSameValue(Value, TruncatedValue)) 4541 return false; 4542 4543 // Special-case bitfields of width 1: booleans are naturally 0/1, and 4544 // therefore don't strictly fit into a signed bitfield of width 1. 4545 if (FieldWidth == 1 && Value == 1) 4546 return false; 4547 4548 std::string PrettyValue = Value.toString(10); 4549 std::string PrettyTrunc = TruncatedValue.toString(10); 4550 4551 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 4552 << PrettyValue << PrettyTrunc << OriginalInit->getType() 4553 << Init->getSourceRange(); 4554 4555 return true; 4556} 4557 4558/// Analyze the given simple or compound assignment for warning-worthy 4559/// operations. 4560static void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 4561 // Just recurse on the LHS. 4562 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4563 4564 // We want to recurse on the RHS as normal unless we're assigning to 4565 // a bitfield. 4566 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 4567 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 4568 E->getOperatorLoc())) { 4569 // Recurse, ignoring any implicit conversions on the RHS. 4570 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 4571 E->getOperatorLoc()); 4572 } 4573 } 4574 4575 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4576} 4577 4578/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4579static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 4580 SourceLocation CContext, unsigned diag, 4581 bool pruneControlFlow = false) { 4582 if (pruneControlFlow) { 4583 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4584 S.PDiag(diag) 4585 << SourceType << T << E->getSourceRange() 4586 << SourceRange(CContext)); 4587 return; 4588 } 4589 S.Diag(E->getExprLoc(), diag) 4590 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 4591} 4592 4593/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4594static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, 4595 SourceLocation CContext, unsigned diag, 4596 bool pruneControlFlow = false) { 4597 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow); 4598} 4599 4600/// Diagnose an implicit cast from a literal expression. Does not warn when the 4601/// cast wouldn't lose information. 4602void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 4603 SourceLocation CContext) { 4604 // Try to convert the literal exactly to an integer. If we can, don't warn. 4605 bool isExact = false; 4606 const llvm::APFloat &Value = FL->getValue(); 4607 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 4608 T->hasUnsignedIntegerRepresentation()); 4609 if (Value.convertToInteger(IntegerValue, 4610 llvm::APFloat::rmTowardZero, &isExact) 4611 == llvm::APFloat::opOK && isExact) 4612 return; 4613 4614 SmallString<16> PrettySourceValue; 4615 Value.toString(PrettySourceValue); 4616 SmallString<16> PrettyTargetValue; 4617 if (T->isSpecificBuiltinType(BuiltinType::Bool)) 4618 PrettyTargetValue = IntegerValue == 0 ? "false" : "true"; 4619 else 4620 IntegerValue.toString(PrettyTargetValue); 4621 4622 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 4623 << FL->getType() << T.getUnqualifiedType() << PrettySourceValue 4624 << PrettyTargetValue << FL->getSourceRange() << SourceRange(CContext); 4625} 4626 4627std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 4628 if (!Range.Width) return "0"; 4629 4630 llvm::APSInt ValueInRange = Value; 4631 ValueInRange.setIsSigned(!Range.NonNegative); 4632 ValueInRange = ValueInRange.trunc(Range.Width); 4633 return ValueInRange.toString(10); 4634} 4635 4636static bool IsImplicitBoolFloatConversion(Sema &S, Expr *Ex, bool ToBool) { 4637 if (!isa<ImplicitCastExpr>(Ex)) 4638 return false; 4639 4640 Expr *InnerE = Ex->IgnoreParenImpCasts(); 4641 const Type *Target = S.Context.getCanonicalType(Ex->getType()).getTypePtr(); 4642 const Type *Source = 4643 S.Context.getCanonicalType(InnerE->getType()).getTypePtr(); 4644 if (Target->isDependentType()) 4645 return false; 4646 4647 const BuiltinType *FloatCandidateBT = 4648 dyn_cast<BuiltinType>(ToBool ? Source : Target); 4649 const Type *BoolCandidateType = ToBool ? Target : Source; 4650 4651 return (BoolCandidateType->isSpecificBuiltinType(BuiltinType::Bool) && 4652 FloatCandidateBT && (FloatCandidateBT->isFloatingPoint())); 4653} 4654 4655void CheckImplicitArgumentConversions(Sema &S, CallExpr *TheCall, 4656 SourceLocation CC) { 4657 unsigned NumArgs = TheCall->getNumArgs(); 4658 for (unsigned i = 0; i < NumArgs; ++i) { 4659 Expr *CurrA = TheCall->getArg(i); 4660 if (!IsImplicitBoolFloatConversion(S, CurrA, true)) 4661 continue; 4662 4663 bool IsSwapped = ((i > 0) && 4664 IsImplicitBoolFloatConversion(S, TheCall->getArg(i - 1), false)); 4665 IsSwapped |= ((i < (NumArgs - 1)) && 4666 IsImplicitBoolFloatConversion(S, TheCall->getArg(i + 1), false)); 4667 if (IsSwapped) { 4668 // Warn on this floating-point to bool conversion. 4669 DiagnoseImpCast(S, CurrA->IgnoreParenImpCasts(), 4670 CurrA->getType(), CC, 4671 diag::warn_impcast_floating_point_to_bool); 4672 } 4673 } 4674} 4675 4676void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 4677 SourceLocation CC, bool *ICContext = 0) { 4678 if (E->isTypeDependent() || E->isValueDependent()) return; 4679 4680 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 4681 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 4682 if (Source == Target) return; 4683 if (Target->isDependentType()) return; 4684 4685 // If the conversion context location is invalid don't complain. We also 4686 // don't want to emit a warning if the issue occurs from the expansion of 4687 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 4688 // delay this check as long as possible. Once we detect we are in that 4689 // scenario, we just return. 4690 if (CC.isInvalid()) 4691 return; 4692 4693 // Diagnose implicit casts to bool. 4694 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 4695 if (isa<StringLiteral>(E)) 4696 // Warn on string literal to bool. Checks for string literals in logical 4697 // expressions, for instances, assert(0 && "error here"), is prevented 4698 // by a check in AnalyzeImplicitConversions(). 4699 return DiagnoseImpCast(S, E, T, CC, 4700 diag::warn_impcast_string_literal_to_bool); 4701 if (Source->isFunctionType()) { 4702 // Warn on function to bool. Checks free functions and static member 4703 // functions. Weakly imported functions are excluded from the check, 4704 // since it's common to test their value to check whether the linker 4705 // found a definition for them. 4706 ValueDecl *D = 0; 4707 if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) { 4708 D = R->getDecl(); 4709 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 4710 D = M->getMemberDecl(); 4711 } 4712 4713 if (D && !D->isWeak()) { 4714 if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) { 4715 S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool) 4716 << F << E->getSourceRange() << SourceRange(CC); 4717 S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence) 4718 << FixItHint::CreateInsertion(E->getExprLoc(), "&"); 4719 QualType ReturnType; 4720 UnresolvedSet<4> NonTemplateOverloads; 4721 S.isExprCallable(*E, ReturnType, NonTemplateOverloads); 4722 if (!ReturnType.isNull() 4723 && ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 4724 S.Diag(E->getExprLoc(), diag::note_function_to_bool_call) 4725 << FixItHint::CreateInsertion( 4726 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()"); 4727 return; 4728 } 4729 } 4730 } 4731 } 4732 4733 // Strip vector types. 4734 if (isa<VectorType>(Source)) { 4735 if (!isa<VectorType>(Target)) { 4736 if (S.SourceMgr.isInSystemMacro(CC)) 4737 return; 4738 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 4739 } 4740 4741 // If the vector cast is cast between two vectors of the same size, it is 4742 // a bitcast, not a conversion. 4743 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 4744 return; 4745 4746 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 4747 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 4748 } 4749 4750 // Strip complex types. 4751 if (isa<ComplexType>(Source)) { 4752 if (!isa<ComplexType>(Target)) { 4753 if (S.SourceMgr.isInSystemMacro(CC)) 4754 return; 4755 4756 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 4757 } 4758 4759 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 4760 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 4761 } 4762 4763 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 4764 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 4765 4766 // If the source is floating point... 4767 if (SourceBT && SourceBT->isFloatingPoint()) { 4768 // ...and the target is floating point... 4769 if (TargetBT && TargetBT->isFloatingPoint()) { 4770 // ...then warn if we're dropping FP rank. 4771 4772 // Builtin FP kinds are ordered by increasing FP rank. 4773 if (SourceBT->getKind() > TargetBT->getKind()) { 4774 // Don't warn about float constants that are precisely 4775 // representable in the target type. 4776 Expr::EvalResult result; 4777 if (E->EvaluateAsRValue(result, S.Context)) { 4778 // Value might be a float, a float vector, or a float complex. 4779 if (IsSameFloatAfterCast(result.Val, 4780 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 4781 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 4782 return; 4783 } 4784 4785 if (S.SourceMgr.isInSystemMacro(CC)) 4786 return; 4787 4788 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 4789 } 4790 return; 4791 } 4792 4793 // If the target is integral, always warn. 4794 if (TargetBT && TargetBT->isInteger()) { 4795 if (S.SourceMgr.isInSystemMacro(CC)) 4796 return; 4797 4798 Expr *InnerE = E->IgnoreParenImpCasts(); 4799 // We also want to warn on, e.g., "int i = -1.234" 4800 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 4801 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 4802 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 4803 4804 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 4805 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 4806 } else { 4807 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 4808 } 4809 } 4810 4811 // If the target is bool, warn if expr is a function or method call. 4812 if (Target->isSpecificBuiltinType(BuiltinType::Bool) && 4813 isa<CallExpr>(E)) { 4814 // Check last argument of function call to see if it is an 4815 // implicit cast from a type matching the type the result 4816 // is being cast to. 4817 CallExpr *CEx = cast<CallExpr>(E); 4818 unsigned NumArgs = CEx->getNumArgs(); 4819 if (NumArgs > 0) { 4820 Expr *LastA = CEx->getArg(NumArgs - 1); 4821 Expr *InnerE = LastA->IgnoreParenImpCasts(); 4822 const Type *InnerType = 4823 S.Context.getCanonicalType(InnerE->getType()).getTypePtr(); 4824 if (isa<ImplicitCastExpr>(LastA) && (InnerType == Target)) { 4825 // Warn on this floating-point to bool conversion 4826 DiagnoseImpCast(S, E, T, CC, 4827 diag::warn_impcast_floating_point_to_bool); 4828 } 4829 } 4830 } 4831 return; 4832 } 4833 4834 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 4835 == Expr::NPCK_GNUNull) && !Target->isAnyPointerType() 4836 && !Target->isBlockPointerType() && !Target->isMemberPointerType() 4837 && Target->isScalarType()) { 4838 SourceLocation Loc = E->getSourceRange().getBegin(); 4839 if (Loc.isMacroID()) 4840 Loc = S.SourceMgr.getImmediateExpansionRange(Loc).first; 4841 if (!Loc.isMacroID() || CC.isMacroID()) 4842 S.Diag(Loc, diag::warn_impcast_null_pointer_to_integer) 4843 << T << clang::SourceRange(CC) 4844 << FixItHint::CreateReplacement(Loc, S.getFixItZeroLiteralForType(T)); 4845 } 4846 4847 if (!Source->isIntegerType() || !Target->isIntegerType()) 4848 return; 4849 4850 // TODO: remove this early return once the false positives for constant->bool 4851 // in templates, macros, etc, are reduced or removed. 4852 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 4853 return; 4854 4855 IntRange SourceRange = GetExprRange(S.Context, E); 4856 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 4857 4858 if (SourceRange.Width > TargetRange.Width) { 4859 // If the source is a constant, use a default-on diagnostic. 4860 // TODO: this should happen for bitfield stores, too. 4861 llvm::APSInt Value(32); 4862 if (E->isIntegerConstantExpr(Value, S.Context)) { 4863 if (S.SourceMgr.isInSystemMacro(CC)) 4864 return; 4865 4866 std::string PrettySourceValue = Value.toString(10); 4867 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 4868 4869 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4870 S.PDiag(diag::warn_impcast_integer_precision_constant) 4871 << PrettySourceValue << PrettyTargetValue 4872 << E->getType() << T << E->getSourceRange() 4873 << clang::SourceRange(CC)); 4874 return; 4875 } 4876 4877 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 4878 if (S.SourceMgr.isInSystemMacro(CC)) 4879 return; 4880 4881 if (TargetRange.Width == 32 && S.Context.getIntWidth(E->getType()) == 64) 4882 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32, 4883 /* pruneControlFlow */ true); 4884 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 4885 } 4886 4887 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 4888 (!TargetRange.NonNegative && SourceRange.NonNegative && 4889 SourceRange.Width == TargetRange.Width)) { 4890 4891 if (S.SourceMgr.isInSystemMacro(CC)) 4892 return; 4893 4894 unsigned DiagID = diag::warn_impcast_integer_sign; 4895 4896 // Traditionally, gcc has warned about this under -Wsign-compare. 4897 // We also want to warn about it in -Wconversion. 4898 // So if -Wconversion is off, use a completely identical diagnostic 4899 // in the sign-compare group. 4900 // The conditional-checking code will 4901 if (ICContext) { 4902 DiagID = diag::warn_impcast_integer_sign_conditional; 4903 *ICContext = true; 4904 } 4905 4906 return DiagnoseImpCast(S, E, T, CC, DiagID); 4907 } 4908 4909 // Diagnose conversions between different enumeration types. 4910 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 4911 // type, to give us better diagnostics. 4912 QualType SourceType = E->getType(); 4913 if (!S.getLangOpts().CPlusPlus) { 4914 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 4915 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 4916 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 4917 SourceType = S.Context.getTypeDeclType(Enum); 4918 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 4919 } 4920 } 4921 4922 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 4923 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 4924 if ((SourceEnum->getDecl()->getIdentifier() || 4925 SourceEnum->getDecl()->getTypedefNameForAnonDecl()) && 4926 (TargetEnum->getDecl()->getIdentifier() || 4927 TargetEnum->getDecl()->getTypedefNameForAnonDecl()) && 4928 SourceEnum != TargetEnum) { 4929 if (S.SourceMgr.isInSystemMacro(CC)) 4930 return; 4931 4932 return DiagnoseImpCast(S, E, SourceType, T, CC, 4933 diag::warn_impcast_different_enum_types); 4934 } 4935 4936 return; 4937} 4938 4939void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 4940 SourceLocation CC, QualType T); 4941 4942void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 4943 SourceLocation CC, bool &ICContext) { 4944 E = E->IgnoreParenImpCasts(); 4945 4946 if (isa<ConditionalOperator>(E)) 4947 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), CC, T); 4948 4949 AnalyzeImplicitConversions(S, E, CC); 4950 if (E->getType() != T) 4951 return CheckImplicitConversion(S, E, T, CC, &ICContext); 4952 return; 4953} 4954 4955void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 4956 SourceLocation CC, QualType T) { 4957 AnalyzeImplicitConversions(S, E->getCond(), CC); 4958 4959 bool Suspicious = false; 4960 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 4961 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 4962 4963 // If -Wconversion would have warned about either of the candidates 4964 // for a signedness conversion to the context type... 4965 if (!Suspicious) return; 4966 4967 // ...but it's currently ignored... 4968 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 4969 CC)) 4970 return; 4971 4972 // ...then check whether it would have warned about either of the 4973 // candidates for a signedness conversion to the condition type. 4974 if (E->getType() == T) return; 4975 4976 Suspicious = false; 4977 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 4978 E->getType(), CC, &Suspicious); 4979 if (!Suspicious) 4980 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 4981 E->getType(), CC, &Suspicious); 4982} 4983 4984/// AnalyzeImplicitConversions - Find and report any interesting 4985/// implicit conversions in the given expression. There are a couple 4986/// of competing diagnostics here, -Wconversion and -Wsign-compare. 4987void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 4988 QualType T = OrigE->getType(); 4989 Expr *E = OrigE->IgnoreParenImpCasts(); 4990 4991 if (E->isTypeDependent() || E->isValueDependent()) 4992 return; 4993 4994 // For conditional operators, we analyze the arguments as if they 4995 // were being fed directly into the output. 4996 if (isa<ConditionalOperator>(E)) { 4997 ConditionalOperator *CO = cast<ConditionalOperator>(E); 4998 CheckConditionalOperator(S, CO, CC, T); 4999 return; 5000 } 5001 5002 // Check implicit argument conversions for function calls. 5003 if (CallExpr *Call = dyn_cast<CallExpr>(E)) 5004 CheckImplicitArgumentConversions(S, Call, CC); 5005 5006 // Go ahead and check any implicit conversions we might have skipped. 5007 // The non-canonical typecheck is just an optimization; 5008 // CheckImplicitConversion will filter out dead implicit conversions. 5009 if (E->getType() != T) 5010 CheckImplicitConversion(S, E, T, CC); 5011 5012 // Now continue drilling into this expression. 5013 5014 // Skip past explicit casts. 5015 if (isa<ExplicitCastExpr>(E)) { 5016 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 5017 return AnalyzeImplicitConversions(S, E, CC); 5018 } 5019 5020 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 5021 // Do a somewhat different check with comparison operators. 5022 if (BO->isComparisonOp()) 5023 return AnalyzeComparison(S, BO); 5024 5025 // And with simple assignments. 5026 if (BO->getOpcode() == BO_Assign) 5027 return AnalyzeAssignment(S, BO); 5028 } 5029 5030 // These break the otherwise-useful invariant below. Fortunately, 5031 // we don't really need to recurse into them, because any internal 5032 // expressions should have been analyzed already when they were 5033 // built into statements. 5034 if (isa<StmtExpr>(E)) return; 5035 5036 // Don't descend into unevaluated contexts. 5037 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 5038 5039 // Now just recurse over the expression's children. 5040 CC = E->getExprLoc(); 5041 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 5042 bool IsLogicalOperator = BO && BO->isLogicalOp(); 5043 for (Stmt::child_range I = E->children(); I; ++I) { 5044 Expr *ChildExpr = dyn_cast_or_null<Expr>(*I); 5045 if (!ChildExpr) 5046 continue; 5047 5048 if (IsLogicalOperator && 5049 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 5050 // Ignore checking string literals that are in logical operators. 5051 continue; 5052 AnalyzeImplicitConversions(S, ChildExpr, CC); 5053 } 5054} 5055 5056} // end anonymous namespace 5057 5058/// Diagnoses "dangerous" implicit conversions within the given 5059/// expression (which is a full expression). Implements -Wconversion 5060/// and -Wsign-compare. 5061/// 5062/// \param CC the "context" location of the implicit conversion, i.e. 5063/// the most location of the syntactic entity requiring the implicit 5064/// conversion 5065void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 5066 // Don't diagnose in unevaluated contexts. 5067 if (isUnevaluatedContext()) 5068 return; 5069 5070 // Don't diagnose for value- or type-dependent expressions. 5071 if (E->isTypeDependent() || E->isValueDependent()) 5072 return; 5073 5074 // Check for array bounds violations in cases where the check isn't triggered 5075 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 5076 // ArraySubscriptExpr is on the RHS of a variable initialization. 5077 CheckArrayAccess(E); 5078 5079 // This is not the right CC for (e.g.) a variable initialization. 5080 AnalyzeImplicitConversions(*this, E, CC); 5081} 5082 5083void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 5084 FieldDecl *BitField, 5085 Expr *Init) { 5086 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 5087} 5088 5089/// CheckParmsForFunctionDef - Check that the parameters of the given 5090/// function are appropriate for the definition of a function. This 5091/// takes care of any checks that cannot be performed on the 5092/// declaration itself, e.g., that the types of each of the function 5093/// parameters are complete. 5094bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 5095 bool CheckParameterNames) { 5096 bool HasInvalidParm = false; 5097 for (; P != PEnd; ++P) { 5098 ParmVarDecl *Param = *P; 5099 5100 // C99 6.7.5.3p4: the parameters in a parameter type list in a 5101 // function declarator that is part of a function definition of 5102 // that function shall not have incomplete type. 5103 // 5104 // This is also C++ [dcl.fct]p6. 5105 if (!Param->isInvalidDecl() && 5106 RequireCompleteType(Param->getLocation(), Param->getType(), 5107 diag::err_typecheck_decl_incomplete_type)) { 5108 Param->setInvalidDecl(); 5109 HasInvalidParm = true; 5110 } 5111 5112 // C99 6.9.1p5: If the declarator includes a parameter type list, the 5113 // declaration of each parameter shall include an identifier. 5114 if (CheckParameterNames && 5115 Param->getIdentifier() == 0 && 5116 !Param->isImplicit() && 5117 !getLangOpts().CPlusPlus) 5118 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 5119 5120 // C99 6.7.5.3p12: 5121 // If the function declarator is not part of a definition of that 5122 // function, parameters may have incomplete type and may use the [*] 5123 // notation in their sequences of declarator specifiers to specify 5124 // variable length array types. 5125 QualType PType = Param->getOriginalType(); 5126 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 5127 if (AT->getSizeModifier() == ArrayType::Star) { 5128 // FIXME: This diagnosic should point the '[*]' if source-location 5129 // information is added for it. 5130 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 5131 } 5132 } 5133 } 5134 5135 return HasInvalidParm; 5136} 5137 5138/// CheckCastAlign - Implements -Wcast-align, which warns when a 5139/// pointer cast increases the alignment requirements. 5140void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 5141 // This is actually a lot of work to potentially be doing on every 5142 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 5143 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 5144 TRange.getBegin()) 5145 == DiagnosticsEngine::Ignored) 5146 return; 5147 5148 // Ignore dependent types. 5149 if (T->isDependentType() || Op->getType()->isDependentType()) 5150 return; 5151 5152 // Require that the destination be a pointer type. 5153 const PointerType *DestPtr = T->getAs<PointerType>(); 5154 if (!DestPtr) return; 5155 5156 // If the destination has alignment 1, we're done. 5157 QualType DestPointee = DestPtr->getPointeeType(); 5158 if (DestPointee->isIncompleteType()) return; 5159 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 5160 if (DestAlign.isOne()) return; 5161 5162 // Require that the source be a pointer type. 5163 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 5164 if (!SrcPtr) return; 5165 QualType SrcPointee = SrcPtr->getPointeeType(); 5166 5167 // Whitelist casts from cv void*. We already implicitly 5168 // whitelisted casts to cv void*, since they have alignment 1. 5169 // Also whitelist casts involving incomplete types, which implicitly 5170 // includes 'void'. 5171 if (SrcPointee->isIncompleteType()) return; 5172 5173 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 5174 if (SrcAlign >= DestAlign) return; 5175 5176 Diag(TRange.getBegin(), diag::warn_cast_align) 5177 << Op->getType() << T 5178 << static_cast<unsigned>(SrcAlign.getQuantity()) 5179 << static_cast<unsigned>(DestAlign.getQuantity()) 5180 << TRange << Op->getSourceRange(); 5181} 5182 5183static const Type* getElementType(const Expr *BaseExpr) { 5184 const Type* EltType = BaseExpr->getType().getTypePtr(); 5185 if (EltType->isAnyPointerType()) 5186 return EltType->getPointeeType().getTypePtr(); 5187 else if (EltType->isArrayType()) 5188 return EltType->getBaseElementTypeUnsafe(); 5189 return EltType; 5190} 5191 5192/// \brief Check whether this array fits the idiom of a size-one tail padded 5193/// array member of a struct. 5194/// 5195/// We avoid emitting out-of-bounds access warnings for such arrays as they are 5196/// commonly used to emulate flexible arrays in C89 code. 5197static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 5198 const NamedDecl *ND) { 5199 if (Size != 1 || !ND) return false; 5200 5201 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 5202 if (!FD) return false; 5203 5204 // Don't consider sizes resulting from macro expansions or template argument 5205 // substitution to form C89 tail-padded arrays. 5206 5207 TypeSourceInfo *TInfo = FD->getTypeSourceInfo(); 5208 while (TInfo) { 5209 TypeLoc TL = TInfo->getTypeLoc(); 5210 // Look through typedefs. 5211 const TypedefTypeLoc *TTL = dyn_cast<TypedefTypeLoc>(&TL); 5212 if (TTL) { 5213 const TypedefNameDecl *TDL = TTL->getTypedefNameDecl(); 5214 TInfo = TDL->getTypeSourceInfo(); 5215 continue; 5216 } 5217 ConstantArrayTypeLoc CTL = cast<ConstantArrayTypeLoc>(TL); 5218 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(CTL.getSizeExpr()); 5219 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 5220 return false; 5221 break; 5222 } 5223 5224 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 5225 if (!RD) return false; 5226 if (RD->isUnion()) return false; 5227 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 5228 if (!CRD->isStandardLayout()) return false; 5229 } 5230 5231 // See if this is the last field decl in the record. 5232 const Decl *D = FD; 5233 while ((D = D->getNextDeclInContext())) 5234 if (isa<FieldDecl>(D)) 5235 return false; 5236 return true; 5237} 5238 5239void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 5240 const ArraySubscriptExpr *ASE, 5241 bool AllowOnePastEnd, bool IndexNegated) { 5242 IndexExpr = IndexExpr->IgnoreParenImpCasts(); 5243 if (IndexExpr->isValueDependent()) 5244 return; 5245 5246 const Type *EffectiveType = getElementType(BaseExpr); 5247 BaseExpr = BaseExpr->IgnoreParenCasts(); 5248 const ConstantArrayType *ArrayTy = 5249 Context.getAsConstantArrayType(BaseExpr->getType()); 5250 if (!ArrayTy) 5251 return; 5252 5253 llvm::APSInt index; 5254 if (!IndexExpr->EvaluateAsInt(index, Context)) 5255 return; 5256 if (IndexNegated) 5257 index = -index; 5258 5259 const NamedDecl *ND = NULL; 5260 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 5261 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 5262 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 5263 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 5264 5265 if (index.isUnsigned() || !index.isNegative()) { 5266 llvm::APInt size = ArrayTy->getSize(); 5267 if (!size.isStrictlyPositive()) 5268 return; 5269 5270 const Type* BaseType = getElementType(BaseExpr); 5271 if (BaseType != EffectiveType) { 5272 // Make sure we're comparing apples to apples when comparing index to size 5273 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 5274 uint64_t array_typesize = Context.getTypeSize(BaseType); 5275 // Handle ptrarith_typesize being zero, such as when casting to void* 5276 if (!ptrarith_typesize) ptrarith_typesize = 1; 5277 if (ptrarith_typesize != array_typesize) { 5278 // There's a cast to a different size type involved 5279 uint64_t ratio = array_typesize / ptrarith_typesize; 5280 // TODO: Be smarter about handling cases where array_typesize is not a 5281 // multiple of ptrarith_typesize 5282 if (ptrarith_typesize * ratio == array_typesize) 5283 size *= llvm::APInt(size.getBitWidth(), ratio); 5284 } 5285 } 5286 5287 if (size.getBitWidth() > index.getBitWidth()) 5288 index = index.zext(size.getBitWidth()); 5289 else if (size.getBitWidth() < index.getBitWidth()) 5290 size = size.zext(index.getBitWidth()); 5291 5292 // For array subscripting the index must be less than size, but for pointer 5293 // arithmetic also allow the index (offset) to be equal to size since 5294 // computing the next address after the end of the array is legal and 5295 // commonly done e.g. in C++ iterators and range-based for loops. 5296 if (AllowOnePastEnd ? index.ule(size) : index.ult(size)) 5297 return; 5298 5299 // Also don't warn for arrays of size 1 which are members of some 5300 // structure. These are often used to approximate flexible arrays in C89 5301 // code. 5302 if (IsTailPaddedMemberArray(*this, size, ND)) 5303 return; 5304 5305 // Suppress the warning if the subscript expression (as identified by the 5306 // ']' location) and the index expression are both from macro expansions 5307 // within a system header. 5308 if (ASE) { 5309 SourceLocation RBracketLoc = SourceMgr.getSpellingLoc( 5310 ASE->getRBracketLoc()); 5311 if (SourceMgr.isInSystemHeader(RBracketLoc)) { 5312 SourceLocation IndexLoc = SourceMgr.getSpellingLoc( 5313 IndexExpr->getLocStart()); 5314 if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc)) 5315 return; 5316 } 5317 } 5318 5319 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 5320 if (ASE) 5321 DiagID = diag::warn_array_index_exceeds_bounds; 5322 5323 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 5324 PDiag(DiagID) << index.toString(10, true) 5325 << size.toString(10, true) 5326 << (unsigned)size.getLimitedValue(~0U) 5327 << IndexExpr->getSourceRange()); 5328 } else { 5329 unsigned DiagID = diag::warn_array_index_precedes_bounds; 5330 if (!ASE) { 5331 DiagID = diag::warn_ptr_arith_precedes_bounds; 5332 if (index.isNegative()) index = -index; 5333 } 5334 5335 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 5336 PDiag(DiagID) << index.toString(10, true) 5337 << IndexExpr->getSourceRange()); 5338 } 5339 5340 if (!ND) { 5341 // Try harder to find a NamedDecl to point at in the note. 5342 while (const ArraySubscriptExpr *ASE = 5343 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 5344 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 5345 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 5346 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 5347 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 5348 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 5349 } 5350 5351 if (ND) 5352 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 5353 PDiag(diag::note_array_index_out_of_bounds) 5354 << ND->getDeclName()); 5355} 5356 5357void Sema::CheckArrayAccess(const Expr *expr) { 5358 int AllowOnePastEnd = 0; 5359 while (expr) { 5360 expr = expr->IgnoreParenImpCasts(); 5361 switch (expr->getStmtClass()) { 5362 case Stmt::ArraySubscriptExprClass: { 5363 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 5364 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE, 5365 AllowOnePastEnd > 0); 5366 return; 5367 } 5368 case Stmt::UnaryOperatorClass: { 5369 // Only unwrap the * and & unary operators 5370 const UnaryOperator *UO = cast<UnaryOperator>(expr); 5371 expr = UO->getSubExpr(); 5372 switch (UO->getOpcode()) { 5373 case UO_AddrOf: 5374 AllowOnePastEnd++; 5375 break; 5376 case UO_Deref: 5377 AllowOnePastEnd--; 5378 break; 5379 default: 5380 return; 5381 } 5382 break; 5383 } 5384 case Stmt::ConditionalOperatorClass: { 5385 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 5386 if (const Expr *lhs = cond->getLHS()) 5387 CheckArrayAccess(lhs); 5388 if (const Expr *rhs = cond->getRHS()) 5389 CheckArrayAccess(rhs); 5390 return; 5391 } 5392 default: 5393 return; 5394 } 5395 } 5396} 5397 5398//===--- CHECK: Objective-C retain cycles ----------------------------------// 5399 5400namespace { 5401 struct RetainCycleOwner { 5402 RetainCycleOwner() : Variable(0), Indirect(false) {} 5403 VarDecl *Variable; 5404 SourceRange Range; 5405 SourceLocation Loc; 5406 bool Indirect; 5407 5408 void setLocsFrom(Expr *e) { 5409 Loc = e->getExprLoc(); 5410 Range = e->getSourceRange(); 5411 } 5412 }; 5413} 5414 5415/// Consider whether capturing the given variable can possibly lead to 5416/// a retain cycle. 5417static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 5418 // In ARC, it's captured strongly iff the variable has __strong 5419 // lifetime. In MRR, it's captured strongly if the variable is 5420 // __block and has an appropriate type. 5421 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 5422 return false; 5423 5424 owner.Variable = var; 5425 if (ref) 5426 owner.setLocsFrom(ref); 5427 return true; 5428} 5429 5430static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) { 5431 while (true) { 5432 e = e->IgnoreParens(); 5433 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 5434 switch (cast->getCastKind()) { 5435 case CK_BitCast: 5436 case CK_LValueBitCast: 5437 case CK_LValueToRValue: 5438 case CK_ARCReclaimReturnedObject: 5439 e = cast->getSubExpr(); 5440 continue; 5441 5442 default: 5443 return false; 5444 } 5445 } 5446 5447 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 5448 ObjCIvarDecl *ivar = ref->getDecl(); 5449 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 5450 return false; 5451 5452 // Try to find a retain cycle in the base. 5453 if (!findRetainCycleOwner(S, ref->getBase(), owner)) 5454 return false; 5455 5456 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 5457 owner.Indirect = true; 5458 return true; 5459 } 5460 5461 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 5462 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 5463 if (!var) return false; 5464 return considerVariable(var, ref, owner); 5465 } 5466 5467 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 5468 if (member->isArrow()) return false; 5469 5470 // Don't count this as an indirect ownership. 5471 e = member->getBase(); 5472 continue; 5473 } 5474 5475 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 5476 // Only pay attention to pseudo-objects on property references. 5477 ObjCPropertyRefExpr *pre 5478 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 5479 ->IgnoreParens()); 5480 if (!pre) return false; 5481 if (pre->isImplicitProperty()) return false; 5482 ObjCPropertyDecl *property = pre->getExplicitProperty(); 5483 if (!property->isRetaining() && 5484 !(property->getPropertyIvarDecl() && 5485 property->getPropertyIvarDecl()->getType() 5486 .getObjCLifetime() == Qualifiers::OCL_Strong)) 5487 return false; 5488 5489 owner.Indirect = true; 5490 if (pre->isSuperReceiver()) { 5491 owner.Variable = S.getCurMethodDecl()->getSelfDecl(); 5492 if (!owner.Variable) 5493 return false; 5494 owner.Loc = pre->getLocation(); 5495 owner.Range = pre->getSourceRange(); 5496 return true; 5497 } 5498 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 5499 ->getSourceExpr()); 5500 continue; 5501 } 5502 5503 // Array ivars? 5504 5505 return false; 5506 } 5507} 5508 5509namespace { 5510 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 5511 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 5512 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 5513 Variable(variable), Capturer(0) {} 5514 5515 VarDecl *Variable; 5516 Expr *Capturer; 5517 5518 void VisitDeclRefExpr(DeclRefExpr *ref) { 5519 if (ref->getDecl() == Variable && !Capturer) 5520 Capturer = ref; 5521 } 5522 5523 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 5524 if (Capturer) return; 5525 Visit(ref->getBase()); 5526 if (Capturer && ref->isFreeIvar()) 5527 Capturer = ref; 5528 } 5529 5530 void VisitBlockExpr(BlockExpr *block) { 5531 // Look inside nested blocks 5532 if (block->getBlockDecl()->capturesVariable(Variable)) 5533 Visit(block->getBlockDecl()->getBody()); 5534 } 5535 5536 void VisitOpaqueValueExpr(OpaqueValueExpr *OVE) { 5537 if (Capturer) return; 5538 if (OVE->getSourceExpr()) 5539 Visit(OVE->getSourceExpr()); 5540 } 5541 }; 5542} 5543 5544/// Check whether the given argument is a block which captures a 5545/// variable. 5546static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 5547 assert(owner.Variable && owner.Loc.isValid()); 5548 5549 e = e->IgnoreParenCasts(); 5550 5551 // Look through [^{...} copy] and Block_copy(^{...}). 5552 if (ObjCMessageExpr *ME = dyn_cast<ObjCMessageExpr>(e)) { 5553 Selector Cmd = ME->getSelector(); 5554 if (Cmd.isUnarySelector() && Cmd.getNameForSlot(0) == "copy") { 5555 e = ME->getInstanceReceiver(); 5556 if (!e) 5557 return 0; 5558 e = e->IgnoreParenCasts(); 5559 } 5560 } else if (CallExpr *CE = dyn_cast<CallExpr>(e)) { 5561 if (CE->getNumArgs() == 1) { 5562 FunctionDecl *Fn = dyn_cast_or_null<FunctionDecl>(CE->getCalleeDecl()); 5563 if (Fn) { 5564 const IdentifierInfo *FnI = Fn->getIdentifier(); 5565 if (FnI && FnI->isStr("_Block_copy")) { 5566 e = CE->getArg(0)->IgnoreParenCasts(); 5567 } 5568 } 5569 } 5570 } 5571 5572 BlockExpr *block = dyn_cast<BlockExpr>(e); 5573 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 5574 return 0; 5575 5576 FindCaptureVisitor visitor(S.Context, owner.Variable); 5577 visitor.Visit(block->getBlockDecl()->getBody()); 5578 return visitor.Capturer; 5579} 5580 5581static void diagnoseRetainCycle(Sema &S, Expr *capturer, 5582 RetainCycleOwner &owner) { 5583 assert(capturer); 5584 assert(owner.Variable && owner.Loc.isValid()); 5585 5586 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 5587 << owner.Variable << capturer->getSourceRange(); 5588 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 5589 << owner.Indirect << owner.Range; 5590} 5591 5592/// Check for a keyword selector that starts with the word 'add' or 5593/// 'set'. 5594static bool isSetterLikeSelector(Selector sel) { 5595 if (sel.isUnarySelector()) return false; 5596 5597 StringRef str = sel.getNameForSlot(0); 5598 while (!str.empty() && str.front() == '_') str = str.substr(1); 5599 if (str.startswith("set")) 5600 str = str.substr(3); 5601 else if (str.startswith("add")) { 5602 // Specially whitelist 'addOperationWithBlock:'. 5603 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 5604 return false; 5605 str = str.substr(3); 5606 } 5607 else 5608 return false; 5609 5610 if (str.empty()) return true; 5611 return !islower(str.front()); 5612} 5613 5614/// Check a message send to see if it's likely to cause a retain cycle. 5615void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 5616 // Only check instance methods whose selector looks like a setter. 5617 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 5618 return; 5619 5620 // Try to find a variable that the receiver is strongly owned by. 5621 RetainCycleOwner owner; 5622 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 5623 if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner)) 5624 return; 5625 } else { 5626 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 5627 owner.Variable = getCurMethodDecl()->getSelfDecl(); 5628 owner.Loc = msg->getSuperLoc(); 5629 owner.Range = msg->getSuperLoc(); 5630 } 5631 5632 // Check whether the receiver is captured by any of the arguments. 5633 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 5634 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 5635 return diagnoseRetainCycle(*this, capturer, owner); 5636} 5637 5638/// Check a property assign to see if it's likely to cause a retain cycle. 5639void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 5640 RetainCycleOwner owner; 5641 if (!findRetainCycleOwner(*this, receiver, owner)) 5642 return; 5643 5644 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 5645 diagnoseRetainCycle(*this, capturer, owner); 5646} 5647 5648void Sema::checkRetainCycles(VarDecl *Var, Expr *Init) { 5649 RetainCycleOwner Owner; 5650 if (!considerVariable(Var, /*DeclRefExpr=*/0, Owner)) 5651 return; 5652 5653 // Because we don't have an expression for the variable, we have to set the 5654 // location explicitly here. 5655 Owner.Loc = Var->getLocation(); 5656 Owner.Range = Var->getSourceRange(); 5657 5658 if (Expr *Capturer = findCapturingExpr(*this, Init, Owner)) 5659 diagnoseRetainCycle(*this, Capturer, Owner); 5660} 5661 5662bool Sema::checkUnsafeAssigns(SourceLocation Loc, 5663 QualType LHS, Expr *RHS) { 5664 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 5665 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 5666 return false; 5667 // strip off any implicit cast added to get to the one arc-specific 5668 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 5669 if (cast->getCastKind() == CK_ARCConsumeObject) { 5670 Diag(Loc, diag::warn_arc_retained_assign) 5671 << (LT == Qualifiers::OCL_ExplicitNone) << 1 5672 << RHS->getSourceRange(); 5673 return true; 5674 } 5675 RHS = cast->getSubExpr(); 5676 } 5677 return false; 5678} 5679 5680void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 5681 Expr *LHS, Expr *RHS) { 5682 QualType LHSType; 5683 // PropertyRef on LHS type need be directly obtained from 5684 // its declaration as it has a PsuedoType. 5685 ObjCPropertyRefExpr *PRE 5686 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens()); 5687 if (PRE && !PRE->isImplicitProperty()) { 5688 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 5689 if (PD) 5690 LHSType = PD->getType(); 5691 } 5692 5693 if (LHSType.isNull()) 5694 LHSType = LHS->getType(); 5695 5696 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 5697 5698 if (LT == Qualifiers::OCL_Weak) { 5699 DiagnosticsEngine::Level Level = 5700 Diags.getDiagnosticLevel(diag::warn_arc_repeated_use_of_weak, Loc); 5701 if (Level != DiagnosticsEngine::Ignored) 5702 getCurFunction()->markSafeWeakUse(LHS); 5703 } 5704 5705 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 5706 return; 5707 5708 // FIXME. Check for other life times. 5709 if (LT != Qualifiers::OCL_None) 5710 return; 5711 5712 if (PRE) { 5713 if (PRE->isImplicitProperty()) 5714 return; 5715 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 5716 if (!PD) 5717 return; 5718 5719 unsigned Attributes = PD->getPropertyAttributes(); 5720 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) { 5721 // when 'assign' attribute was not explicitly specified 5722 // by user, ignore it and rely on property type itself 5723 // for lifetime info. 5724 unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten(); 5725 if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) && 5726 LHSType->isObjCRetainableType()) 5727 return; 5728 5729 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 5730 if (cast->getCastKind() == CK_ARCConsumeObject) { 5731 Diag(Loc, diag::warn_arc_retained_property_assign) 5732 << RHS->getSourceRange(); 5733 return; 5734 } 5735 RHS = cast->getSubExpr(); 5736 } 5737 } 5738 else if (Attributes & ObjCPropertyDecl::OBJC_PR_weak) { 5739 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 5740 if (cast->getCastKind() == CK_ARCConsumeObject) { 5741 Diag(Loc, diag::warn_arc_retained_assign) 5742 << 0 << 0<< RHS->getSourceRange(); 5743 return; 5744 } 5745 RHS = cast->getSubExpr(); 5746 } 5747 } 5748 } 5749} 5750 5751//===--- CHECK: Empty statement body (-Wempty-body) ---------------------===// 5752 5753namespace { 5754bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr, 5755 SourceLocation StmtLoc, 5756 const NullStmt *Body) { 5757 // Do not warn if the body is a macro that expands to nothing, e.g: 5758 // 5759 // #define CALL(x) 5760 // if (condition) 5761 // CALL(0); 5762 // 5763 if (Body->hasLeadingEmptyMacro()) 5764 return false; 5765 5766 // Get line numbers of statement and body. 5767 bool StmtLineInvalid; 5768 unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc, 5769 &StmtLineInvalid); 5770 if (StmtLineInvalid) 5771 return false; 5772 5773 bool BodyLineInvalid; 5774 unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(), 5775 &BodyLineInvalid); 5776 if (BodyLineInvalid) 5777 return false; 5778 5779 // Warn if null statement and body are on the same line. 5780 if (StmtLine != BodyLine) 5781 return false; 5782 5783 return true; 5784} 5785} // Unnamed namespace 5786 5787void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc, 5788 const Stmt *Body, 5789 unsigned DiagID) { 5790 // Since this is a syntactic check, don't emit diagnostic for template 5791 // instantiations, this just adds noise. 5792 if (CurrentInstantiationScope) 5793 return; 5794 5795 // The body should be a null statement. 5796 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 5797 if (!NBody) 5798 return; 5799 5800 // Do the usual checks. 5801 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 5802 return; 5803 5804 Diag(NBody->getSemiLoc(), DiagID); 5805 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 5806} 5807 5808void Sema::DiagnoseEmptyLoopBody(const Stmt *S, 5809 const Stmt *PossibleBody) { 5810 assert(!CurrentInstantiationScope); // Ensured by caller 5811 5812 SourceLocation StmtLoc; 5813 const Stmt *Body; 5814 unsigned DiagID; 5815 if (const ForStmt *FS = dyn_cast<ForStmt>(S)) { 5816 StmtLoc = FS->getRParenLoc(); 5817 Body = FS->getBody(); 5818 DiagID = diag::warn_empty_for_body; 5819 } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) { 5820 StmtLoc = WS->getCond()->getSourceRange().getEnd(); 5821 Body = WS->getBody(); 5822 DiagID = diag::warn_empty_while_body; 5823 } else 5824 return; // Neither `for' nor `while'. 5825 5826 // The body should be a null statement. 5827 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 5828 if (!NBody) 5829 return; 5830 5831 // Skip expensive checks if diagnostic is disabled. 5832 if (Diags.getDiagnosticLevel(DiagID, NBody->getSemiLoc()) == 5833 DiagnosticsEngine::Ignored) 5834 return; 5835 5836 // Do the usual checks. 5837 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 5838 return; 5839 5840 // `for(...);' and `while(...);' are popular idioms, so in order to keep 5841 // noise level low, emit diagnostics only if for/while is followed by a 5842 // CompoundStmt, e.g.: 5843 // for (int i = 0; i < n; i++); 5844 // { 5845 // a(i); 5846 // } 5847 // or if for/while is followed by a statement with more indentation 5848 // than for/while itself: 5849 // for (int i = 0; i < n; i++); 5850 // a(i); 5851 bool ProbableTypo = isa<CompoundStmt>(PossibleBody); 5852 if (!ProbableTypo) { 5853 bool BodyColInvalid; 5854 unsigned BodyCol = SourceMgr.getPresumedColumnNumber( 5855 PossibleBody->getLocStart(), 5856 &BodyColInvalid); 5857 if (BodyColInvalid) 5858 return; 5859 5860 bool StmtColInvalid; 5861 unsigned StmtCol = SourceMgr.getPresumedColumnNumber( 5862 S->getLocStart(), 5863 &StmtColInvalid); 5864 if (StmtColInvalid) 5865 return; 5866 5867 if (BodyCol > StmtCol) 5868 ProbableTypo = true; 5869 } 5870 5871 if (ProbableTypo) { 5872 Diag(NBody->getSemiLoc(), DiagID); 5873 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 5874 } 5875} 5876 5877//===--- Layout compatibility ----------------------------------------------// 5878 5879namespace { 5880 5881bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2); 5882 5883/// \brief Check if two enumeration types are layout-compatible. 5884bool isLayoutCompatible(ASTContext &C, EnumDecl *ED1, EnumDecl *ED2) { 5885 // C++11 [dcl.enum] p8: 5886 // Two enumeration types are layout-compatible if they have the same 5887 // underlying type. 5888 return ED1->isComplete() && ED2->isComplete() && 5889 C.hasSameType(ED1->getIntegerType(), ED2->getIntegerType()); 5890} 5891 5892/// \brief Check if two fields are layout-compatible. 5893bool isLayoutCompatible(ASTContext &C, FieldDecl *Field1, FieldDecl *Field2) { 5894 if (!isLayoutCompatible(C, Field1->getType(), Field2->getType())) 5895 return false; 5896 5897 if (Field1->isBitField() != Field2->isBitField()) 5898 return false; 5899 5900 if (Field1->isBitField()) { 5901 // Make sure that the bit-fields are the same length. 5902 unsigned Bits1 = Field1->getBitWidthValue(C); 5903 unsigned Bits2 = Field2->getBitWidthValue(C); 5904 5905 if (Bits1 != Bits2) 5906 return false; 5907 } 5908 5909 return true; 5910} 5911 5912/// \brief Check if two standard-layout structs are layout-compatible. 5913/// (C++11 [class.mem] p17) 5914bool isLayoutCompatibleStruct(ASTContext &C, 5915 RecordDecl *RD1, 5916 RecordDecl *RD2) { 5917 // If both records are C++ classes, check that base classes match. 5918 if (const CXXRecordDecl *D1CXX = dyn_cast<CXXRecordDecl>(RD1)) { 5919 // If one of records is a CXXRecordDecl we are in C++ mode, 5920 // thus the other one is a CXXRecordDecl, too. 5921 const CXXRecordDecl *D2CXX = cast<CXXRecordDecl>(RD2); 5922 // Check number of base classes. 5923 if (D1CXX->getNumBases() != D2CXX->getNumBases()) 5924 return false; 5925 5926 // Check the base classes. 5927 for (CXXRecordDecl::base_class_const_iterator 5928 Base1 = D1CXX->bases_begin(), 5929 BaseEnd1 = D1CXX->bases_end(), 5930 Base2 = D2CXX->bases_begin(); 5931 Base1 != BaseEnd1; 5932 ++Base1, ++Base2) { 5933 if (!isLayoutCompatible(C, Base1->getType(), Base2->getType())) 5934 return false; 5935 } 5936 } else if (const CXXRecordDecl *D2CXX = dyn_cast<CXXRecordDecl>(RD2)) { 5937 // If only RD2 is a C++ class, it should have zero base classes. 5938 if (D2CXX->getNumBases() > 0) 5939 return false; 5940 } 5941 5942 // Check the fields. 5943 RecordDecl::field_iterator Field2 = RD2->field_begin(), 5944 Field2End = RD2->field_end(), 5945 Field1 = RD1->field_begin(), 5946 Field1End = RD1->field_end(); 5947 for ( ; Field1 != Field1End && Field2 != Field2End; ++Field1, ++Field2) { 5948 if (!isLayoutCompatible(C, *Field1, *Field2)) 5949 return false; 5950 } 5951 if (Field1 != Field1End || Field2 != Field2End) 5952 return false; 5953 5954 return true; 5955} 5956 5957/// \brief Check if two standard-layout unions are layout-compatible. 5958/// (C++11 [class.mem] p18) 5959bool isLayoutCompatibleUnion(ASTContext &C, 5960 RecordDecl *RD1, 5961 RecordDecl *RD2) { 5962 llvm::SmallPtrSet<FieldDecl *, 8> UnmatchedFields; 5963 for (RecordDecl::field_iterator Field2 = RD2->field_begin(), 5964 Field2End = RD2->field_end(); 5965 Field2 != Field2End; ++Field2) { 5966 UnmatchedFields.insert(*Field2); 5967 } 5968 5969 for (RecordDecl::field_iterator Field1 = RD1->field_begin(), 5970 Field1End = RD1->field_end(); 5971 Field1 != Field1End; ++Field1) { 5972 llvm::SmallPtrSet<FieldDecl *, 8>::iterator 5973 I = UnmatchedFields.begin(), 5974 E = UnmatchedFields.end(); 5975 5976 for ( ; I != E; ++I) { 5977 if (isLayoutCompatible(C, *Field1, *I)) { 5978 bool Result = UnmatchedFields.erase(*I); 5979 (void) Result; 5980 assert(Result); 5981 break; 5982 } 5983 } 5984 if (I == E) 5985 return false; 5986 } 5987 5988 return UnmatchedFields.empty(); 5989} 5990 5991bool isLayoutCompatible(ASTContext &C, RecordDecl *RD1, RecordDecl *RD2) { 5992 if (RD1->isUnion() != RD2->isUnion()) 5993 return false; 5994 5995 if (RD1->isUnion()) 5996 return isLayoutCompatibleUnion(C, RD1, RD2); 5997 else 5998 return isLayoutCompatibleStruct(C, RD1, RD2); 5999} 6000 6001/// \brief Check if two types are layout-compatible in C++11 sense. 6002bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2) { 6003 if (T1.isNull() || T2.isNull()) 6004 return false; 6005 6006 // C++11 [basic.types] p11: 6007 // If two types T1 and T2 are the same type, then T1 and T2 are 6008 // layout-compatible types. 6009 if (C.hasSameType(T1, T2)) 6010 return true; 6011 6012 T1 = T1.getCanonicalType().getUnqualifiedType(); 6013 T2 = T2.getCanonicalType().getUnqualifiedType(); 6014 6015 const Type::TypeClass TC1 = T1->getTypeClass(); 6016 const Type::TypeClass TC2 = T2->getTypeClass(); 6017 6018 if (TC1 != TC2) 6019 return false; 6020 6021 if (TC1 == Type::Enum) { 6022 return isLayoutCompatible(C, 6023 cast<EnumType>(T1)->getDecl(), 6024 cast<EnumType>(T2)->getDecl()); 6025 } else if (TC1 == Type::Record) { 6026 if (!T1->isStandardLayoutType() || !T2->isStandardLayoutType()) 6027 return false; 6028 6029 return isLayoutCompatible(C, 6030 cast<RecordType>(T1)->getDecl(), 6031 cast<RecordType>(T2)->getDecl()); 6032 } 6033 6034 return false; 6035} 6036} 6037 6038//===--- CHECK: pointer_with_type_tag attribute: datatypes should match ----// 6039 6040namespace { 6041/// \brief Given a type tag expression find the type tag itself. 6042/// 6043/// \param TypeExpr Type tag expression, as it appears in user's code. 6044/// 6045/// \param VD Declaration of an identifier that appears in a type tag. 6046/// 6047/// \param MagicValue Type tag magic value. 6048bool FindTypeTagExpr(const Expr *TypeExpr, const ASTContext &Ctx, 6049 const ValueDecl **VD, uint64_t *MagicValue) { 6050 while(true) { 6051 if (!TypeExpr) 6052 return false; 6053 6054 TypeExpr = TypeExpr->IgnoreParenImpCasts()->IgnoreParenCasts(); 6055 6056 switch (TypeExpr->getStmtClass()) { 6057 case Stmt::UnaryOperatorClass: { 6058 const UnaryOperator *UO = cast<UnaryOperator>(TypeExpr); 6059 if (UO->getOpcode() == UO_AddrOf || UO->getOpcode() == UO_Deref) { 6060 TypeExpr = UO->getSubExpr(); 6061 continue; 6062 } 6063 return false; 6064 } 6065 6066 case Stmt::DeclRefExprClass: { 6067 const DeclRefExpr *DRE = cast<DeclRefExpr>(TypeExpr); 6068 *VD = DRE->getDecl(); 6069 return true; 6070 } 6071 6072 case Stmt::IntegerLiteralClass: { 6073 const IntegerLiteral *IL = cast<IntegerLiteral>(TypeExpr); 6074 llvm::APInt MagicValueAPInt = IL->getValue(); 6075 if (MagicValueAPInt.getActiveBits() <= 64) { 6076 *MagicValue = MagicValueAPInt.getZExtValue(); 6077 return true; 6078 } else 6079 return false; 6080 } 6081 6082 case Stmt::BinaryConditionalOperatorClass: 6083 case Stmt::ConditionalOperatorClass: { 6084 const AbstractConditionalOperator *ACO = 6085 cast<AbstractConditionalOperator>(TypeExpr); 6086 bool Result; 6087 if (ACO->getCond()->EvaluateAsBooleanCondition(Result, Ctx)) { 6088 if (Result) 6089 TypeExpr = ACO->getTrueExpr(); 6090 else 6091 TypeExpr = ACO->getFalseExpr(); 6092 continue; 6093 } 6094 return false; 6095 } 6096 6097 case Stmt::BinaryOperatorClass: { 6098 const BinaryOperator *BO = cast<BinaryOperator>(TypeExpr); 6099 if (BO->getOpcode() == BO_Comma) { 6100 TypeExpr = BO->getRHS(); 6101 continue; 6102 } 6103 return false; 6104 } 6105 6106 default: 6107 return false; 6108 } 6109 } 6110} 6111 6112/// \brief Retrieve the C type corresponding to type tag TypeExpr. 6113/// 6114/// \param TypeExpr Expression that specifies a type tag. 6115/// 6116/// \param MagicValues Registered magic values. 6117/// 6118/// \param FoundWrongKind Set to true if a type tag was found, but of a wrong 6119/// kind. 6120/// 6121/// \param TypeInfo Information about the corresponding C type. 6122/// 6123/// \returns true if the corresponding C type was found. 6124bool GetMatchingCType( 6125 const IdentifierInfo *ArgumentKind, 6126 const Expr *TypeExpr, const ASTContext &Ctx, 6127 const llvm::DenseMap<Sema::TypeTagMagicValue, 6128 Sema::TypeTagData> *MagicValues, 6129 bool &FoundWrongKind, 6130 Sema::TypeTagData &TypeInfo) { 6131 FoundWrongKind = false; 6132 6133 // Variable declaration that has type_tag_for_datatype attribute. 6134 const ValueDecl *VD = NULL; 6135 6136 uint64_t MagicValue; 6137 6138 if (!FindTypeTagExpr(TypeExpr, Ctx, &VD, &MagicValue)) 6139 return false; 6140 6141 if (VD) { 6142 for (specific_attr_iterator<TypeTagForDatatypeAttr> 6143 I = VD->specific_attr_begin<TypeTagForDatatypeAttr>(), 6144 E = VD->specific_attr_end<TypeTagForDatatypeAttr>(); 6145 I != E; ++I) { 6146 if (I->getArgumentKind() != ArgumentKind) { 6147 FoundWrongKind = true; 6148 return false; 6149 } 6150 TypeInfo.Type = I->getMatchingCType(); 6151 TypeInfo.LayoutCompatible = I->getLayoutCompatible(); 6152 TypeInfo.MustBeNull = I->getMustBeNull(); 6153 return true; 6154 } 6155 return false; 6156 } 6157 6158 if (!MagicValues) 6159 return false; 6160 6161 llvm::DenseMap<Sema::TypeTagMagicValue, 6162 Sema::TypeTagData>::const_iterator I = 6163 MagicValues->find(std::make_pair(ArgumentKind, MagicValue)); 6164 if (I == MagicValues->end()) 6165 return false; 6166 6167 TypeInfo = I->second; 6168 return true; 6169} 6170} // unnamed namespace 6171 6172void Sema::RegisterTypeTagForDatatype(const IdentifierInfo *ArgumentKind, 6173 uint64_t MagicValue, QualType Type, 6174 bool LayoutCompatible, 6175 bool MustBeNull) { 6176 if (!TypeTagForDatatypeMagicValues) 6177 TypeTagForDatatypeMagicValues.reset( 6178 new llvm::DenseMap<TypeTagMagicValue, TypeTagData>); 6179 6180 TypeTagMagicValue Magic(ArgumentKind, MagicValue); 6181 (*TypeTagForDatatypeMagicValues)[Magic] = 6182 TypeTagData(Type, LayoutCompatible, MustBeNull); 6183} 6184 6185namespace { 6186bool IsSameCharType(QualType T1, QualType T2) { 6187 const BuiltinType *BT1 = T1->getAs<BuiltinType>(); 6188 if (!BT1) 6189 return false; 6190 6191 const BuiltinType *BT2 = T2->getAs<BuiltinType>(); 6192 if (!BT2) 6193 return false; 6194 6195 BuiltinType::Kind T1Kind = BT1->getKind(); 6196 BuiltinType::Kind T2Kind = BT2->getKind(); 6197 6198 return (T1Kind == BuiltinType::SChar && T2Kind == BuiltinType::Char_S) || 6199 (T1Kind == BuiltinType::UChar && T2Kind == BuiltinType::Char_U) || 6200 (T1Kind == BuiltinType::Char_U && T2Kind == BuiltinType::UChar) || 6201 (T1Kind == BuiltinType::Char_S && T2Kind == BuiltinType::SChar); 6202} 6203} // unnamed namespace 6204 6205void Sema::CheckArgumentWithTypeTag(const ArgumentWithTypeTagAttr *Attr, 6206 const Expr * const *ExprArgs) { 6207 const IdentifierInfo *ArgumentKind = Attr->getArgumentKind(); 6208 bool IsPointerAttr = Attr->getIsPointer(); 6209 6210 const Expr *TypeTagExpr = ExprArgs[Attr->getTypeTagIdx()]; 6211 bool FoundWrongKind; 6212 TypeTagData TypeInfo; 6213 if (!GetMatchingCType(ArgumentKind, TypeTagExpr, Context, 6214 TypeTagForDatatypeMagicValues.get(), 6215 FoundWrongKind, TypeInfo)) { 6216 if (FoundWrongKind) 6217 Diag(TypeTagExpr->getExprLoc(), 6218 diag::warn_type_tag_for_datatype_wrong_kind) 6219 << TypeTagExpr->getSourceRange(); 6220 return; 6221 } 6222 6223 const Expr *ArgumentExpr = ExprArgs[Attr->getArgumentIdx()]; 6224 if (IsPointerAttr) { 6225 // Skip implicit cast of pointer to `void *' (as a function argument). 6226 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(ArgumentExpr)) 6227 if (ICE->getType()->isVoidPointerType() && 6228 ICE->getCastKind() == CK_BitCast) 6229 ArgumentExpr = ICE->getSubExpr(); 6230 } 6231 QualType ArgumentType = ArgumentExpr->getType(); 6232 6233 // Passing a `void*' pointer shouldn't trigger a warning. 6234 if (IsPointerAttr && ArgumentType->isVoidPointerType()) 6235 return; 6236 6237 if (TypeInfo.MustBeNull) { 6238 // Type tag with matching void type requires a null pointer. 6239 if (!ArgumentExpr->isNullPointerConstant(Context, 6240 Expr::NPC_ValueDependentIsNotNull)) { 6241 Diag(ArgumentExpr->getExprLoc(), 6242 diag::warn_type_safety_null_pointer_required) 6243 << ArgumentKind->getName() 6244 << ArgumentExpr->getSourceRange() 6245 << TypeTagExpr->getSourceRange(); 6246 } 6247 return; 6248 } 6249 6250 QualType RequiredType = TypeInfo.Type; 6251 if (IsPointerAttr) 6252 RequiredType = Context.getPointerType(RequiredType); 6253 6254 bool mismatch = false; 6255 if (!TypeInfo.LayoutCompatible) { 6256 mismatch = !Context.hasSameType(ArgumentType, RequiredType); 6257 6258 // C++11 [basic.fundamental] p1: 6259 // Plain char, signed char, and unsigned char are three distinct types. 6260 // 6261 // But we treat plain `char' as equivalent to `signed char' or `unsigned 6262 // char' depending on the current char signedness mode. 6263 if (mismatch) 6264 if ((IsPointerAttr && IsSameCharType(ArgumentType->getPointeeType(), 6265 RequiredType->getPointeeType())) || 6266 (!IsPointerAttr && IsSameCharType(ArgumentType, RequiredType))) 6267 mismatch = false; 6268 } else 6269 if (IsPointerAttr) 6270 mismatch = !isLayoutCompatible(Context, 6271 ArgumentType->getPointeeType(), 6272 RequiredType->getPointeeType()); 6273 else 6274 mismatch = !isLayoutCompatible(Context, ArgumentType, RequiredType); 6275 6276 if (mismatch) 6277 Diag(ArgumentExpr->getExprLoc(), diag::warn_type_safety_type_mismatch) 6278 << ArgumentType << ArgumentKind->getName() 6279 << TypeInfo.LayoutCompatible << RequiredType 6280 << ArgumentExpr->getSourceRange() 6281 << TypeTagExpr->getSourceRange(); 6282} 6283