SemaChecking.cpp revision bed28ac1d1463adca3ecf24fca5c30646fa9dbb2
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(move(TheCallResult)); 270#define BUILTIN(ID, TYPE, ATTRS) 271#define ATOMIC_BUILTIN(ID, TYPE, ATTRS) \ 272 case Builtin::BI##ID: \ 273 return SemaAtomicOpsOverloaded(move(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 move(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 unsigned 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 & (1 << TV)) == 0) 377 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 378 << TheCall->getArg(ImmArg)->getSourceRange(); 379 } 380 381 if (PtrArgNum >= 0) { 382 // Check that pointer arguments have the specified type. 383 Expr *Arg = TheCall->getArg(PtrArgNum); 384 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) 385 Arg = ICE->getSubExpr(); 386 ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg); 387 QualType RHSTy = RHS.get()->getType(); 388 QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context); 389 if (HasConstPtr) 390 EltTy = EltTy.withConst(); 391 QualType LHSTy = Context.getPointerType(EltTy); 392 AssignConvertType ConvTy; 393 ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS); 394 if (RHS.isInvalid()) 395 return true; 396 if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy, 397 RHS.get(), AA_Assigning)) 398 return true; 399 } 400 401 // For NEON intrinsics which take an immediate value as part of the 402 // instruction, range check them here. 403 unsigned i = 0, l = 0, u = 0; 404 switch (BuiltinID) { 405 default: return false; 406 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 407 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 408 case ARM::BI__builtin_arm_vcvtr_f: 409 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 410#define GET_NEON_IMMEDIATE_CHECK 411#include "clang/Basic/arm_neon.inc" 412#undef GET_NEON_IMMEDIATE_CHECK 413 }; 414 415 // We can't check the value of a dependent argument. 416 if (TheCall->getArg(i)->isTypeDependent() || 417 TheCall->getArg(i)->isValueDependent()) 418 return false; 419 420 // Check that the immediate argument is actually a constant. 421 if (SemaBuiltinConstantArg(TheCall, i, Result)) 422 return true; 423 424 // Range check against the upper/lower values for this isntruction. 425 unsigned Val = Result.getZExtValue(); 426 if (Val < l || Val > (u + l)) 427 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 428 << l << u+l << TheCall->getArg(i)->getSourceRange(); 429 430 // FIXME: VFP Intrinsics should error if VFP not present. 431 return false; 432} 433 434bool Sema::CheckMipsBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 435 unsigned i = 0, l = 0, u = 0; 436 switch (BuiltinID) { 437 default: return false; 438 case Mips::BI__builtin_mips_wrdsp: i = 1; l = 0; u = 63; break; 439 case Mips::BI__builtin_mips_rddsp: i = 0; l = 0; u = 63; break; 440 }; 441 442 // We can't check the value of a dependent argument. 443 if (TheCall->getArg(i)->isTypeDependent() || 444 TheCall->getArg(i)->isValueDependent()) 445 return false; 446 447 // Check that the immediate argument is actually a constant. 448 llvm::APSInt Result; 449 if (SemaBuiltinConstantArg(TheCall, i, Result)) 450 return true; 451 452 // Range check against the upper/lower values for this instruction. 453 unsigned Val = Result.getZExtValue(); 454 if (Val < l || Val > u) 455 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 456 << l << u << TheCall->getArg(i)->getSourceRange(); 457 458 return false; 459} 460 461/// Given a FunctionDecl's FormatAttr, attempts to populate the FomatStringInfo 462/// parameter with the FormatAttr's correct format_idx and firstDataArg. 463/// Returns true when the format fits the function and the FormatStringInfo has 464/// been populated. 465bool Sema::getFormatStringInfo(const FormatAttr *Format, bool IsCXXMember, 466 FormatStringInfo *FSI) { 467 FSI->HasVAListArg = Format->getFirstArg() == 0; 468 FSI->FormatIdx = Format->getFormatIdx() - 1; 469 FSI->FirstDataArg = FSI->HasVAListArg ? 0 : Format->getFirstArg() - 1; 470 471 // The way the format attribute works in GCC, the implicit this argument 472 // of member functions is counted. However, it doesn't appear in our own 473 // lists, so decrement format_idx in that case. 474 if (IsCXXMember) { 475 if(FSI->FormatIdx == 0) 476 return false; 477 --FSI->FormatIdx; 478 if (FSI->FirstDataArg != 0) 479 --FSI->FirstDataArg; 480 } 481 return true; 482} 483 484/// Handles the checks for format strings, non-POD arguments to vararg 485/// functions, and NULL arguments passed to non-NULL parameters. 486void Sema::checkCall(NamedDecl *FDecl, Expr **Args, 487 unsigned NumArgs, 488 unsigned NumProtoArgs, 489 bool IsMemberFunction, 490 SourceLocation Loc, 491 SourceRange Range, 492 VariadicCallType CallType) { 493 // FIXME: This mechanism should be abstracted to be less fragile and 494 // more efficient. For example, just map function ids to custom 495 // handlers. 496 497 // Printf and scanf checking. 498 bool HandledFormatString = false; 499 for (specific_attr_iterator<FormatAttr> 500 I = FDecl->specific_attr_begin<FormatAttr>(), 501 E = FDecl->specific_attr_end<FormatAttr>(); I != E ; ++I) 502 if (CheckFormatArguments(*I, Args, NumArgs, IsMemberFunction, CallType, 503 Loc, Range)) 504 HandledFormatString = true; 505 506 // Refuse POD arguments that weren't caught by the format string 507 // checks above. 508 if (!HandledFormatString && CallType != VariadicDoesNotApply) 509 for (unsigned ArgIdx = NumProtoArgs; ArgIdx < NumArgs; ++ArgIdx) 510 variadicArgumentPODCheck(Args[ArgIdx], CallType); 511 512 for (specific_attr_iterator<NonNullAttr> 513 I = FDecl->specific_attr_begin<NonNullAttr>(), 514 E = FDecl->specific_attr_end<NonNullAttr>(); I != E; ++I) 515 CheckNonNullArguments(*I, Args, Loc); 516} 517 518/// CheckConstructorCall - Check a constructor call for correctness and safety 519/// properties not enforced by the C type system. 520void Sema::CheckConstructorCall(FunctionDecl *FDecl, Expr **Args, 521 unsigned NumArgs, 522 const FunctionProtoType *Proto, 523 SourceLocation Loc) { 524 VariadicCallType CallType = 525 Proto->isVariadic() ? VariadicConstructor : VariadicDoesNotApply; 526 checkCall(FDecl, Args, NumArgs, Proto->getNumArgs(), 527 /*IsMemberFunction=*/true, Loc, SourceRange(), CallType); 528} 529 530/// CheckFunctionCall - Check a direct function call for various correctness 531/// and safety properties not strictly enforced by the C type system. 532bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall, 533 const FunctionProtoType *Proto) { 534 bool IsMemberFunction = isa<CXXMemberCallExpr>(TheCall); 535 VariadicCallType CallType = getVariadicCallType(FDecl, Proto, 536 TheCall->getCallee()); 537 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 538 checkCall(FDecl, TheCall->getArgs(), TheCall->getNumArgs(), NumProtoArgs, 539 IsMemberFunction, TheCall->getRParenLoc(), 540 TheCall->getCallee()->getSourceRange(), CallType); 541 542 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 543 // None of the checks below are needed for functions that don't have 544 // simple names (e.g., C++ conversion functions). 545 if (!FnInfo) 546 return false; 547 548 unsigned CMId = FDecl->getMemoryFunctionKind(); 549 if (CMId == 0) 550 return false; 551 552 // Handle memory setting and copying functions. 553 if (CMId == Builtin::BIstrlcpy || CMId == Builtin::BIstrlcat) 554 CheckStrlcpycatArguments(TheCall, FnInfo); 555 else if (CMId == Builtin::BIstrncat) 556 CheckStrncatArguments(TheCall, FnInfo); 557 else 558 CheckMemaccessArguments(TheCall, CMId, FnInfo); 559 560 return false; 561} 562 563bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac, 564 Expr **Args, unsigned NumArgs) { 565 VariadicCallType CallType = 566 Method->isVariadic() ? VariadicMethod : VariadicDoesNotApply; 567 568 checkCall(Method, Args, NumArgs, Method->param_size(), 569 /*IsMemberFunction=*/false, 570 lbrac, Method->getSourceRange(), CallType); 571 572 return false; 573} 574 575bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall, 576 const FunctionProtoType *Proto) { 577 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 578 if (!V) 579 return false; 580 581 QualType Ty = V->getType(); 582 if (!Ty->isBlockPointerType()) 583 return false; 584 585 VariadicCallType CallType = 586 Proto && Proto->isVariadic() ? VariadicBlock : VariadicDoesNotApply ; 587 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 588 589 checkCall(NDecl, TheCall->getArgs(), TheCall->getNumArgs(), 590 NumProtoArgs, /*IsMemberFunction=*/false, 591 TheCall->getRParenLoc(), 592 TheCall->getCallee()->getSourceRange(), CallType); 593 594 return false; 595} 596 597ExprResult Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, 598 AtomicExpr::AtomicOp Op) { 599 CallExpr *TheCall = cast<CallExpr>(TheCallResult.get()); 600 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 601 602 // All these operations take one of the following forms: 603 enum { 604 // C __c11_atomic_init(A *, C) 605 Init, 606 // C __c11_atomic_load(A *, int) 607 Load, 608 // void __atomic_load(A *, CP, int) 609 Copy, 610 // C __c11_atomic_add(A *, M, int) 611 Arithmetic, 612 // C __atomic_exchange_n(A *, CP, int) 613 Xchg, 614 // void __atomic_exchange(A *, C *, CP, int) 615 GNUXchg, 616 // bool __c11_atomic_compare_exchange_strong(A *, C *, CP, int, int) 617 C11CmpXchg, 618 // bool __atomic_compare_exchange(A *, C *, CP, bool, int, int) 619 GNUCmpXchg 620 } Form = Init; 621 const unsigned NumArgs[] = { 2, 2, 3, 3, 3, 4, 5, 6 }; 622 const unsigned NumVals[] = { 1, 0, 1, 1, 1, 2, 2, 3 }; 623 // where: 624 // C is an appropriate type, 625 // A is volatile _Atomic(C) for __c11 builtins and is C for GNU builtins, 626 // CP is C for __c11 builtins and GNU _n builtins and is C * otherwise, 627 // M is C if C is an integer, and ptrdiff_t if C is a pointer, and 628 // the int parameters are for orderings. 629 630 assert(AtomicExpr::AO__c11_atomic_init == 0 && 631 AtomicExpr::AO__c11_atomic_fetch_xor + 1 == AtomicExpr::AO__atomic_load 632 && "need to update code for modified C11 atomics"); 633 bool IsC11 = Op >= AtomicExpr::AO__c11_atomic_init && 634 Op <= AtomicExpr::AO__c11_atomic_fetch_xor; 635 bool IsN = Op == AtomicExpr::AO__atomic_load_n || 636 Op == AtomicExpr::AO__atomic_store_n || 637 Op == AtomicExpr::AO__atomic_exchange_n || 638 Op == AtomicExpr::AO__atomic_compare_exchange_n; 639 bool IsAddSub = false; 640 641 switch (Op) { 642 case AtomicExpr::AO__c11_atomic_init: 643 Form = Init; 644 break; 645 646 case AtomicExpr::AO__c11_atomic_load: 647 case AtomicExpr::AO__atomic_load_n: 648 Form = Load; 649 break; 650 651 case AtomicExpr::AO__c11_atomic_store: 652 case AtomicExpr::AO__atomic_load: 653 case AtomicExpr::AO__atomic_store: 654 case AtomicExpr::AO__atomic_store_n: 655 Form = Copy; 656 break; 657 658 case AtomicExpr::AO__c11_atomic_fetch_add: 659 case AtomicExpr::AO__c11_atomic_fetch_sub: 660 case AtomicExpr::AO__atomic_fetch_add: 661 case AtomicExpr::AO__atomic_fetch_sub: 662 case AtomicExpr::AO__atomic_add_fetch: 663 case AtomicExpr::AO__atomic_sub_fetch: 664 IsAddSub = true; 665 // Fall through. 666 case AtomicExpr::AO__c11_atomic_fetch_and: 667 case AtomicExpr::AO__c11_atomic_fetch_or: 668 case AtomicExpr::AO__c11_atomic_fetch_xor: 669 case AtomicExpr::AO__atomic_fetch_and: 670 case AtomicExpr::AO__atomic_fetch_or: 671 case AtomicExpr::AO__atomic_fetch_xor: 672 case AtomicExpr::AO__atomic_fetch_nand: 673 case AtomicExpr::AO__atomic_and_fetch: 674 case AtomicExpr::AO__atomic_or_fetch: 675 case AtomicExpr::AO__atomic_xor_fetch: 676 case AtomicExpr::AO__atomic_nand_fetch: 677 Form = Arithmetic; 678 break; 679 680 case AtomicExpr::AO__c11_atomic_exchange: 681 case AtomicExpr::AO__atomic_exchange_n: 682 Form = Xchg; 683 break; 684 685 case AtomicExpr::AO__atomic_exchange: 686 Form = GNUXchg; 687 break; 688 689 case AtomicExpr::AO__c11_atomic_compare_exchange_strong: 690 case AtomicExpr::AO__c11_atomic_compare_exchange_weak: 691 Form = C11CmpXchg; 692 break; 693 694 case AtomicExpr::AO__atomic_compare_exchange: 695 case AtomicExpr::AO__atomic_compare_exchange_n: 696 Form = GNUCmpXchg; 697 break; 698 } 699 700 // Check we have the right number of arguments. 701 if (TheCall->getNumArgs() < NumArgs[Form]) { 702 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 703 << 0 << NumArgs[Form] << TheCall->getNumArgs() 704 << TheCall->getCallee()->getSourceRange(); 705 return ExprError(); 706 } else if (TheCall->getNumArgs() > NumArgs[Form]) { 707 Diag(TheCall->getArg(NumArgs[Form])->getLocStart(), 708 diag::err_typecheck_call_too_many_args) 709 << 0 << NumArgs[Form] << TheCall->getNumArgs() 710 << TheCall->getCallee()->getSourceRange(); 711 return ExprError(); 712 } 713 714 // Inspect the first argument of the atomic operation. 715 Expr *Ptr = TheCall->getArg(0); 716 Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get(); 717 const PointerType *pointerType = Ptr->getType()->getAs<PointerType>(); 718 if (!pointerType) { 719 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 720 << Ptr->getType() << Ptr->getSourceRange(); 721 return ExprError(); 722 } 723 724 // For a __c11 builtin, this should be a pointer to an _Atomic type. 725 QualType AtomTy = pointerType->getPointeeType(); // 'A' 726 QualType ValType = AtomTy; // 'C' 727 if (IsC11) { 728 if (!AtomTy->isAtomicType()) { 729 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 730 << Ptr->getType() << Ptr->getSourceRange(); 731 return ExprError(); 732 } 733 ValType = AtomTy->getAs<AtomicType>()->getValueType(); 734 } 735 736 // For an arithmetic operation, the implied arithmetic must be well-formed. 737 if (Form == Arithmetic) { 738 // gcc does not enforce these rules for GNU atomics, but we do so for sanity. 739 if (IsAddSub && !ValType->isIntegerType() && !ValType->isPointerType()) { 740 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 741 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 742 return ExprError(); 743 } 744 if (!IsAddSub && !ValType->isIntegerType()) { 745 Diag(DRE->getLocStart(), diag::err_atomic_op_bitwise_needs_atomic_int) 746 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 747 return ExprError(); 748 } 749 } else if (IsN && !ValType->isIntegerType() && !ValType->isPointerType()) { 750 // For __atomic_*_n operations, the value type must be a scalar integral or 751 // pointer type which is 1, 2, 4, 8 or 16 bytes in length. 752 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 753 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 754 return ExprError(); 755 } 756 757 if (!IsC11 && !AtomTy.isTriviallyCopyableType(Context)) { 758 // For GNU atomics, require a trivially-copyable type. This is not part of 759 // the GNU atomics specification, but we enforce it for sanity. 760 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_trivial_copy) 761 << Ptr->getType() << Ptr->getSourceRange(); 762 return ExprError(); 763 } 764 765 // FIXME: For any builtin other than a load, the ValType must not be 766 // const-qualified. 767 768 switch (ValType.getObjCLifetime()) { 769 case Qualifiers::OCL_None: 770 case Qualifiers::OCL_ExplicitNone: 771 // okay 772 break; 773 774 case Qualifiers::OCL_Weak: 775 case Qualifiers::OCL_Strong: 776 case Qualifiers::OCL_Autoreleasing: 777 // FIXME: Can this happen? By this point, ValType should be known 778 // to be trivially copyable. 779 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 780 << ValType << Ptr->getSourceRange(); 781 return ExprError(); 782 } 783 784 QualType ResultType = ValType; 785 if (Form == Copy || Form == GNUXchg || Form == Init) 786 ResultType = Context.VoidTy; 787 else if (Form == C11CmpXchg || Form == GNUCmpXchg) 788 ResultType = Context.BoolTy; 789 790 // The type of a parameter passed 'by value'. In the GNU atomics, such 791 // arguments are actually passed as pointers. 792 QualType ByValType = ValType; // 'CP' 793 if (!IsC11 && !IsN) 794 ByValType = Ptr->getType(); 795 796 // The first argument --- the pointer --- has a fixed type; we 797 // deduce the types of the rest of the arguments accordingly. Walk 798 // the remaining arguments, converting them to the deduced value type. 799 for (unsigned i = 1; i != NumArgs[Form]; ++i) { 800 QualType Ty; 801 if (i < NumVals[Form] + 1) { 802 switch (i) { 803 case 1: 804 // The second argument is the non-atomic operand. For arithmetic, this 805 // is always passed by value, and for a compare_exchange it is always 806 // passed by address. For the rest, GNU uses by-address and C11 uses 807 // by-value. 808 assert(Form != Load); 809 if (Form == Init || (Form == Arithmetic && ValType->isIntegerType())) 810 Ty = ValType; 811 else if (Form == Copy || Form == Xchg) 812 Ty = ByValType; 813 else if (Form == Arithmetic) 814 Ty = Context.getPointerDiffType(); 815 else 816 Ty = Context.getPointerType(ValType.getUnqualifiedType()); 817 break; 818 case 2: 819 // The third argument to compare_exchange / GNU exchange is a 820 // (pointer to a) desired value. 821 Ty = ByValType; 822 break; 823 case 3: 824 // The fourth argument to GNU compare_exchange is a 'weak' flag. 825 Ty = Context.BoolTy; 826 break; 827 } 828 } else { 829 // The order(s) are always converted to int. 830 Ty = Context.IntTy; 831 } 832 833 InitializedEntity Entity = 834 InitializedEntity::InitializeParameter(Context, Ty, false); 835 ExprResult Arg = TheCall->getArg(i); 836 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 837 if (Arg.isInvalid()) 838 return true; 839 TheCall->setArg(i, Arg.get()); 840 } 841 842 // Permute the arguments into a 'consistent' order. 843 SmallVector<Expr*, 5> SubExprs; 844 SubExprs.push_back(Ptr); 845 switch (Form) { 846 case Init: 847 // Note, AtomicExpr::getVal1() has a special case for this atomic. 848 SubExprs.push_back(TheCall->getArg(1)); // Val1 849 break; 850 case Load: 851 SubExprs.push_back(TheCall->getArg(1)); // Order 852 break; 853 case Copy: 854 case Arithmetic: 855 case Xchg: 856 SubExprs.push_back(TheCall->getArg(2)); // Order 857 SubExprs.push_back(TheCall->getArg(1)); // Val1 858 break; 859 case GNUXchg: 860 // Note, AtomicExpr::getVal2() has a special case for this atomic. 861 SubExprs.push_back(TheCall->getArg(3)); // Order 862 SubExprs.push_back(TheCall->getArg(1)); // Val1 863 SubExprs.push_back(TheCall->getArg(2)); // Val2 864 break; 865 case C11CmpXchg: 866 SubExprs.push_back(TheCall->getArg(3)); // Order 867 SubExprs.push_back(TheCall->getArg(1)); // Val1 868 SubExprs.push_back(TheCall->getArg(4)); // OrderFail 869 SubExprs.push_back(TheCall->getArg(2)); // Val2 870 break; 871 case GNUCmpXchg: 872 SubExprs.push_back(TheCall->getArg(4)); // Order 873 SubExprs.push_back(TheCall->getArg(1)); // Val1 874 SubExprs.push_back(TheCall->getArg(5)); // OrderFail 875 SubExprs.push_back(TheCall->getArg(2)); // Val2 876 SubExprs.push_back(TheCall->getArg(3)); // Weak 877 break; 878 } 879 880 return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 881 SubExprs.data(), SubExprs.size(), 882 ResultType, Op, 883 TheCall->getRParenLoc())); 884} 885 886 887/// checkBuiltinArgument - Given a call to a builtin function, perform 888/// normal type-checking on the given argument, updating the call in 889/// place. This is useful when a builtin function requires custom 890/// type-checking for some of its arguments but not necessarily all of 891/// them. 892/// 893/// Returns true on error. 894static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 895 FunctionDecl *Fn = E->getDirectCallee(); 896 assert(Fn && "builtin call without direct callee!"); 897 898 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 899 InitializedEntity Entity = 900 InitializedEntity::InitializeParameter(S.Context, Param); 901 902 ExprResult Arg = E->getArg(0); 903 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 904 if (Arg.isInvalid()) 905 return true; 906 907 E->setArg(ArgIndex, Arg.take()); 908 return false; 909} 910 911/// SemaBuiltinAtomicOverloaded - We have a call to a function like 912/// __sync_fetch_and_add, which is an overloaded function based on the pointer 913/// type of its first argument. The main ActOnCallExpr routines have already 914/// promoted the types of arguments because all of these calls are prototyped as 915/// void(...). 916/// 917/// This function goes through and does final semantic checking for these 918/// builtins, 919ExprResult 920Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 921 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 922 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 923 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 924 925 // Ensure that we have at least one argument to do type inference from. 926 if (TheCall->getNumArgs() < 1) { 927 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 928 << 0 << 1 << TheCall->getNumArgs() 929 << TheCall->getCallee()->getSourceRange(); 930 return ExprError(); 931 } 932 933 // Inspect the first argument of the atomic builtin. This should always be 934 // a pointer type, whose element is an integral scalar or pointer type. 935 // Because it is a pointer type, we don't have to worry about any implicit 936 // casts here. 937 // FIXME: We don't allow floating point scalars as input. 938 Expr *FirstArg = TheCall->getArg(0); 939 ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg); 940 if (FirstArgResult.isInvalid()) 941 return ExprError(); 942 FirstArg = FirstArgResult.take(); 943 TheCall->setArg(0, FirstArg); 944 945 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 946 if (!pointerType) { 947 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 948 << FirstArg->getType() << FirstArg->getSourceRange(); 949 return ExprError(); 950 } 951 952 QualType ValType = pointerType->getPointeeType(); 953 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 954 !ValType->isBlockPointerType()) { 955 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 956 << FirstArg->getType() << FirstArg->getSourceRange(); 957 return ExprError(); 958 } 959 960 switch (ValType.getObjCLifetime()) { 961 case Qualifiers::OCL_None: 962 case Qualifiers::OCL_ExplicitNone: 963 // okay 964 break; 965 966 case Qualifiers::OCL_Weak: 967 case Qualifiers::OCL_Strong: 968 case Qualifiers::OCL_Autoreleasing: 969 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 970 << ValType << FirstArg->getSourceRange(); 971 return ExprError(); 972 } 973 974 // Strip any qualifiers off ValType. 975 ValType = ValType.getUnqualifiedType(); 976 977 // The majority of builtins return a value, but a few have special return 978 // types, so allow them to override appropriately below. 979 QualType ResultType = ValType; 980 981 // We need to figure out which concrete builtin this maps onto. For example, 982 // __sync_fetch_and_add with a 2 byte object turns into 983 // __sync_fetch_and_add_2. 984#define BUILTIN_ROW(x) \ 985 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 986 Builtin::BI##x##_8, Builtin::BI##x##_16 } 987 988 static const unsigned BuiltinIndices[][5] = { 989 BUILTIN_ROW(__sync_fetch_and_add), 990 BUILTIN_ROW(__sync_fetch_and_sub), 991 BUILTIN_ROW(__sync_fetch_and_or), 992 BUILTIN_ROW(__sync_fetch_and_and), 993 BUILTIN_ROW(__sync_fetch_and_xor), 994 995 BUILTIN_ROW(__sync_add_and_fetch), 996 BUILTIN_ROW(__sync_sub_and_fetch), 997 BUILTIN_ROW(__sync_and_and_fetch), 998 BUILTIN_ROW(__sync_or_and_fetch), 999 BUILTIN_ROW(__sync_xor_and_fetch), 1000 1001 BUILTIN_ROW(__sync_val_compare_and_swap), 1002 BUILTIN_ROW(__sync_bool_compare_and_swap), 1003 BUILTIN_ROW(__sync_lock_test_and_set), 1004 BUILTIN_ROW(__sync_lock_release), 1005 BUILTIN_ROW(__sync_swap) 1006 }; 1007#undef BUILTIN_ROW 1008 1009 // Determine the index of the size. 1010 unsigned SizeIndex; 1011 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 1012 case 1: SizeIndex = 0; break; 1013 case 2: SizeIndex = 1; break; 1014 case 4: SizeIndex = 2; break; 1015 case 8: SizeIndex = 3; break; 1016 case 16: SizeIndex = 4; break; 1017 default: 1018 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 1019 << FirstArg->getType() << FirstArg->getSourceRange(); 1020 return ExprError(); 1021 } 1022 1023 // Each of these builtins has one pointer argument, followed by some number of 1024 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 1025 // that we ignore. Find out which row of BuiltinIndices to read from as well 1026 // as the number of fixed args. 1027 unsigned BuiltinID = FDecl->getBuiltinID(); 1028 unsigned BuiltinIndex, NumFixed = 1; 1029 switch (BuiltinID) { 1030 default: llvm_unreachable("Unknown overloaded atomic builtin!"); 1031 case Builtin::BI__sync_fetch_and_add: 1032 case Builtin::BI__sync_fetch_and_add_1: 1033 case Builtin::BI__sync_fetch_and_add_2: 1034 case Builtin::BI__sync_fetch_and_add_4: 1035 case Builtin::BI__sync_fetch_and_add_8: 1036 case Builtin::BI__sync_fetch_and_add_16: 1037 BuiltinIndex = 0; 1038 break; 1039 1040 case Builtin::BI__sync_fetch_and_sub: 1041 case Builtin::BI__sync_fetch_and_sub_1: 1042 case Builtin::BI__sync_fetch_and_sub_2: 1043 case Builtin::BI__sync_fetch_and_sub_4: 1044 case Builtin::BI__sync_fetch_and_sub_8: 1045 case Builtin::BI__sync_fetch_and_sub_16: 1046 BuiltinIndex = 1; 1047 break; 1048 1049 case Builtin::BI__sync_fetch_and_or: 1050 case Builtin::BI__sync_fetch_and_or_1: 1051 case Builtin::BI__sync_fetch_and_or_2: 1052 case Builtin::BI__sync_fetch_and_or_4: 1053 case Builtin::BI__sync_fetch_and_or_8: 1054 case Builtin::BI__sync_fetch_and_or_16: 1055 BuiltinIndex = 2; 1056 break; 1057 1058 case Builtin::BI__sync_fetch_and_and: 1059 case Builtin::BI__sync_fetch_and_and_1: 1060 case Builtin::BI__sync_fetch_and_and_2: 1061 case Builtin::BI__sync_fetch_and_and_4: 1062 case Builtin::BI__sync_fetch_and_and_8: 1063 case Builtin::BI__sync_fetch_and_and_16: 1064 BuiltinIndex = 3; 1065 break; 1066 1067 case Builtin::BI__sync_fetch_and_xor: 1068 case Builtin::BI__sync_fetch_and_xor_1: 1069 case Builtin::BI__sync_fetch_and_xor_2: 1070 case Builtin::BI__sync_fetch_and_xor_4: 1071 case Builtin::BI__sync_fetch_and_xor_8: 1072 case Builtin::BI__sync_fetch_and_xor_16: 1073 BuiltinIndex = 4; 1074 break; 1075 1076 case Builtin::BI__sync_add_and_fetch: 1077 case Builtin::BI__sync_add_and_fetch_1: 1078 case Builtin::BI__sync_add_and_fetch_2: 1079 case Builtin::BI__sync_add_and_fetch_4: 1080 case Builtin::BI__sync_add_and_fetch_8: 1081 case Builtin::BI__sync_add_and_fetch_16: 1082 BuiltinIndex = 5; 1083 break; 1084 1085 case Builtin::BI__sync_sub_and_fetch: 1086 case Builtin::BI__sync_sub_and_fetch_1: 1087 case Builtin::BI__sync_sub_and_fetch_2: 1088 case Builtin::BI__sync_sub_and_fetch_4: 1089 case Builtin::BI__sync_sub_and_fetch_8: 1090 case Builtin::BI__sync_sub_and_fetch_16: 1091 BuiltinIndex = 6; 1092 break; 1093 1094 case Builtin::BI__sync_and_and_fetch: 1095 case Builtin::BI__sync_and_and_fetch_1: 1096 case Builtin::BI__sync_and_and_fetch_2: 1097 case Builtin::BI__sync_and_and_fetch_4: 1098 case Builtin::BI__sync_and_and_fetch_8: 1099 case Builtin::BI__sync_and_and_fetch_16: 1100 BuiltinIndex = 7; 1101 break; 1102 1103 case Builtin::BI__sync_or_and_fetch: 1104 case Builtin::BI__sync_or_and_fetch_1: 1105 case Builtin::BI__sync_or_and_fetch_2: 1106 case Builtin::BI__sync_or_and_fetch_4: 1107 case Builtin::BI__sync_or_and_fetch_8: 1108 case Builtin::BI__sync_or_and_fetch_16: 1109 BuiltinIndex = 8; 1110 break; 1111 1112 case Builtin::BI__sync_xor_and_fetch: 1113 case Builtin::BI__sync_xor_and_fetch_1: 1114 case Builtin::BI__sync_xor_and_fetch_2: 1115 case Builtin::BI__sync_xor_and_fetch_4: 1116 case Builtin::BI__sync_xor_and_fetch_8: 1117 case Builtin::BI__sync_xor_and_fetch_16: 1118 BuiltinIndex = 9; 1119 break; 1120 1121 case Builtin::BI__sync_val_compare_and_swap: 1122 case Builtin::BI__sync_val_compare_and_swap_1: 1123 case Builtin::BI__sync_val_compare_and_swap_2: 1124 case Builtin::BI__sync_val_compare_and_swap_4: 1125 case Builtin::BI__sync_val_compare_and_swap_8: 1126 case Builtin::BI__sync_val_compare_and_swap_16: 1127 BuiltinIndex = 10; 1128 NumFixed = 2; 1129 break; 1130 1131 case Builtin::BI__sync_bool_compare_and_swap: 1132 case Builtin::BI__sync_bool_compare_and_swap_1: 1133 case Builtin::BI__sync_bool_compare_and_swap_2: 1134 case Builtin::BI__sync_bool_compare_and_swap_4: 1135 case Builtin::BI__sync_bool_compare_and_swap_8: 1136 case Builtin::BI__sync_bool_compare_and_swap_16: 1137 BuiltinIndex = 11; 1138 NumFixed = 2; 1139 ResultType = Context.BoolTy; 1140 break; 1141 1142 case Builtin::BI__sync_lock_test_and_set: 1143 case Builtin::BI__sync_lock_test_and_set_1: 1144 case Builtin::BI__sync_lock_test_and_set_2: 1145 case Builtin::BI__sync_lock_test_and_set_4: 1146 case Builtin::BI__sync_lock_test_and_set_8: 1147 case Builtin::BI__sync_lock_test_and_set_16: 1148 BuiltinIndex = 12; 1149 break; 1150 1151 case Builtin::BI__sync_lock_release: 1152 case Builtin::BI__sync_lock_release_1: 1153 case Builtin::BI__sync_lock_release_2: 1154 case Builtin::BI__sync_lock_release_4: 1155 case Builtin::BI__sync_lock_release_8: 1156 case Builtin::BI__sync_lock_release_16: 1157 BuiltinIndex = 13; 1158 NumFixed = 0; 1159 ResultType = Context.VoidTy; 1160 break; 1161 1162 case Builtin::BI__sync_swap: 1163 case Builtin::BI__sync_swap_1: 1164 case Builtin::BI__sync_swap_2: 1165 case Builtin::BI__sync_swap_4: 1166 case Builtin::BI__sync_swap_8: 1167 case Builtin::BI__sync_swap_16: 1168 BuiltinIndex = 14; 1169 break; 1170 } 1171 1172 // Now that we know how many fixed arguments we expect, first check that we 1173 // have at least that many. 1174 if (TheCall->getNumArgs() < 1+NumFixed) { 1175 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 1176 << 0 << 1+NumFixed << TheCall->getNumArgs() 1177 << TheCall->getCallee()->getSourceRange(); 1178 return ExprError(); 1179 } 1180 1181 // Get the decl for the concrete builtin from this, we can tell what the 1182 // concrete integer type we should convert to is. 1183 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 1184 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 1185 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 1186 FunctionDecl *NewBuiltinDecl = 1187 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 1188 TUScope, false, DRE->getLocStart())); 1189 1190 // The first argument --- the pointer --- has a fixed type; we 1191 // deduce the types of the rest of the arguments accordingly. Walk 1192 // the remaining arguments, converting them to the deduced value type. 1193 for (unsigned i = 0; i != NumFixed; ++i) { 1194 ExprResult Arg = TheCall->getArg(i+1); 1195 1196 // GCC does an implicit conversion to the pointer or integer ValType. This 1197 // can fail in some cases (1i -> int**), check for this error case now. 1198 // Initialize the argument. 1199 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context, 1200 ValType, /*consume*/ false); 1201 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 1202 if (Arg.isInvalid()) 1203 return ExprError(); 1204 1205 // Okay, we have something that *can* be converted to the right type. Check 1206 // to see if there is a potentially weird extension going on here. This can 1207 // happen when you do an atomic operation on something like an char* and 1208 // pass in 42. The 42 gets converted to char. This is even more strange 1209 // for things like 45.123 -> char, etc. 1210 // FIXME: Do this check. 1211 TheCall->setArg(i+1, Arg.take()); 1212 } 1213 1214 ASTContext& Context = this->getASTContext(); 1215 1216 // Create a new DeclRefExpr to refer to the new decl. 1217 DeclRefExpr* NewDRE = DeclRefExpr::Create( 1218 Context, 1219 DRE->getQualifierLoc(), 1220 SourceLocation(), 1221 NewBuiltinDecl, 1222 /*enclosing*/ false, 1223 DRE->getLocation(), 1224 NewBuiltinDecl->getType(), 1225 DRE->getValueKind()); 1226 1227 // Set the callee in the CallExpr. 1228 // FIXME: This leaks the original parens and implicit casts. 1229 ExprResult PromotedCall = UsualUnaryConversions(NewDRE); 1230 if (PromotedCall.isInvalid()) 1231 return ExprError(); 1232 TheCall->setCallee(PromotedCall.take()); 1233 1234 // Change the result type of the call to match the original value type. This 1235 // is arbitrary, but the codegen for these builtins ins design to handle it 1236 // gracefully. 1237 TheCall->setType(ResultType); 1238 1239 return move(TheCallResult); 1240} 1241 1242/// CheckObjCString - Checks that the argument to the builtin 1243/// CFString constructor is correct 1244/// Note: It might also make sense to do the UTF-16 conversion here (would 1245/// simplify the backend). 1246bool Sema::CheckObjCString(Expr *Arg) { 1247 Arg = Arg->IgnoreParenCasts(); 1248 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 1249 1250 if (!Literal || !Literal->isAscii()) { 1251 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 1252 << Arg->getSourceRange(); 1253 return true; 1254 } 1255 1256 if (Literal->containsNonAsciiOrNull()) { 1257 StringRef String = Literal->getString(); 1258 unsigned NumBytes = String.size(); 1259 SmallVector<UTF16, 128> ToBuf(NumBytes); 1260 const UTF8 *FromPtr = (UTF8 *)String.data(); 1261 UTF16 *ToPtr = &ToBuf[0]; 1262 1263 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 1264 &ToPtr, ToPtr + NumBytes, 1265 strictConversion); 1266 // Check for conversion failure. 1267 if (Result != conversionOK) 1268 Diag(Arg->getLocStart(), 1269 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 1270 } 1271 return false; 1272} 1273 1274/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 1275/// Emit an error and return true on failure, return false on success. 1276bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 1277 Expr *Fn = TheCall->getCallee(); 1278 if (TheCall->getNumArgs() > 2) { 1279 Diag(TheCall->getArg(2)->getLocStart(), 1280 diag::err_typecheck_call_too_many_args) 1281 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1282 << Fn->getSourceRange() 1283 << SourceRange(TheCall->getArg(2)->getLocStart(), 1284 (*(TheCall->arg_end()-1))->getLocEnd()); 1285 return true; 1286 } 1287 1288 if (TheCall->getNumArgs() < 2) { 1289 return Diag(TheCall->getLocEnd(), 1290 diag::err_typecheck_call_too_few_args_at_least) 1291 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 1292 } 1293 1294 // Type-check the first argument normally. 1295 if (checkBuiltinArgument(*this, TheCall, 0)) 1296 return true; 1297 1298 // Determine whether the current function is variadic or not. 1299 BlockScopeInfo *CurBlock = getCurBlock(); 1300 bool isVariadic; 1301 if (CurBlock) 1302 isVariadic = CurBlock->TheDecl->isVariadic(); 1303 else if (FunctionDecl *FD = getCurFunctionDecl()) 1304 isVariadic = FD->isVariadic(); 1305 else 1306 isVariadic = getCurMethodDecl()->isVariadic(); 1307 1308 if (!isVariadic) { 1309 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 1310 return true; 1311 } 1312 1313 // Verify that the second argument to the builtin is the last argument of the 1314 // current function or method. 1315 bool SecondArgIsLastNamedArgument = false; 1316 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 1317 1318 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 1319 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 1320 // FIXME: This isn't correct for methods (results in bogus warning). 1321 // Get the last formal in the current function. 1322 const ParmVarDecl *LastArg; 1323 if (CurBlock) 1324 LastArg = *(CurBlock->TheDecl->param_end()-1); 1325 else if (FunctionDecl *FD = getCurFunctionDecl()) 1326 LastArg = *(FD->param_end()-1); 1327 else 1328 LastArg = *(getCurMethodDecl()->param_end()-1); 1329 SecondArgIsLastNamedArgument = PV == LastArg; 1330 } 1331 } 1332 1333 if (!SecondArgIsLastNamedArgument) 1334 Diag(TheCall->getArg(1)->getLocStart(), 1335 diag::warn_second_parameter_of_va_start_not_last_named_argument); 1336 return false; 1337} 1338 1339/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 1340/// friends. This is declared to take (...), so we have to check everything. 1341bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 1342 if (TheCall->getNumArgs() < 2) 1343 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1344 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 1345 if (TheCall->getNumArgs() > 2) 1346 return Diag(TheCall->getArg(2)->getLocStart(), 1347 diag::err_typecheck_call_too_many_args) 1348 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1349 << SourceRange(TheCall->getArg(2)->getLocStart(), 1350 (*(TheCall->arg_end()-1))->getLocEnd()); 1351 1352 ExprResult OrigArg0 = TheCall->getArg(0); 1353 ExprResult OrigArg1 = TheCall->getArg(1); 1354 1355 // Do standard promotions between the two arguments, returning their common 1356 // type. 1357 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 1358 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 1359 return true; 1360 1361 // Make sure any conversions are pushed back into the call; this is 1362 // type safe since unordered compare builtins are declared as "_Bool 1363 // foo(...)". 1364 TheCall->setArg(0, OrigArg0.get()); 1365 TheCall->setArg(1, OrigArg1.get()); 1366 1367 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 1368 return false; 1369 1370 // If the common type isn't a real floating type, then the arguments were 1371 // invalid for this operation. 1372 if (Res.isNull() || !Res->isRealFloatingType()) 1373 return Diag(OrigArg0.get()->getLocStart(), 1374 diag::err_typecheck_call_invalid_ordered_compare) 1375 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 1376 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 1377 1378 return false; 1379} 1380 1381/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 1382/// __builtin_isnan and friends. This is declared to take (...), so we have 1383/// to check everything. We expect the last argument to be a floating point 1384/// value. 1385bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 1386 if (TheCall->getNumArgs() < NumArgs) 1387 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1388 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 1389 if (TheCall->getNumArgs() > NumArgs) 1390 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 1391 diag::err_typecheck_call_too_many_args) 1392 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 1393 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 1394 (*(TheCall->arg_end()-1))->getLocEnd()); 1395 1396 Expr *OrigArg = TheCall->getArg(NumArgs-1); 1397 1398 if (OrigArg->isTypeDependent()) 1399 return false; 1400 1401 // This operation requires a non-_Complex floating-point number. 1402 if (!OrigArg->getType()->isRealFloatingType()) 1403 return Diag(OrigArg->getLocStart(), 1404 diag::err_typecheck_call_invalid_unary_fp) 1405 << OrigArg->getType() << OrigArg->getSourceRange(); 1406 1407 // If this is an implicit conversion from float -> double, remove it. 1408 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 1409 Expr *CastArg = Cast->getSubExpr(); 1410 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 1411 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 1412 "promotion from float to double is the only expected cast here"); 1413 Cast->setSubExpr(0); 1414 TheCall->setArg(NumArgs-1, CastArg); 1415 } 1416 } 1417 1418 return false; 1419} 1420 1421/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1422// This is declared to take (...), so we have to check everything. 1423ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1424 if (TheCall->getNumArgs() < 2) 1425 return ExprError(Diag(TheCall->getLocEnd(), 1426 diag::err_typecheck_call_too_few_args_at_least) 1427 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1428 << TheCall->getSourceRange()); 1429 1430 // Determine which of the following types of shufflevector we're checking: 1431 // 1) unary, vector mask: (lhs, mask) 1432 // 2) binary, vector mask: (lhs, rhs, mask) 1433 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1434 QualType resType = TheCall->getArg(0)->getType(); 1435 unsigned numElements = 0; 1436 1437 if (!TheCall->getArg(0)->isTypeDependent() && 1438 !TheCall->getArg(1)->isTypeDependent()) { 1439 QualType LHSType = TheCall->getArg(0)->getType(); 1440 QualType RHSType = TheCall->getArg(1)->getType(); 1441 1442 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 1443 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 1444 << SourceRange(TheCall->getArg(0)->getLocStart(), 1445 TheCall->getArg(1)->getLocEnd()); 1446 return ExprError(); 1447 } 1448 1449 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1450 unsigned numResElements = TheCall->getNumArgs() - 2; 1451 1452 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1453 // with mask. If so, verify that RHS is an integer vector type with the 1454 // same number of elts as lhs. 1455 if (TheCall->getNumArgs() == 2) { 1456 if (!RHSType->hasIntegerRepresentation() || 1457 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1458 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1459 << SourceRange(TheCall->getArg(1)->getLocStart(), 1460 TheCall->getArg(1)->getLocEnd()); 1461 numResElements = numElements; 1462 } 1463 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1464 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1465 << SourceRange(TheCall->getArg(0)->getLocStart(), 1466 TheCall->getArg(1)->getLocEnd()); 1467 return ExprError(); 1468 } else if (numElements != numResElements) { 1469 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1470 resType = Context.getVectorType(eltType, numResElements, 1471 VectorType::GenericVector); 1472 } 1473 } 1474 1475 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1476 if (TheCall->getArg(i)->isTypeDependent() || 1477 TheCall->getArg(i)->isValueDependent()) 1478 continue; 1479 1480 llvm::APSInt Result(32); 1481 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1482 return ExprError(Diag(TheCall->getLocStart(), 1483 diag::err_shufflevector_nonconstant_argument) 1484 << TheCall->getArg(i)->getSourceRange()); 1485 1486 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1487 return ExprError(Diag(TheCall->getLocStart(), 1488 diag::err_shufflevector_argument_too_large) 1489 << TheCall->getArg(i)->getSourceRange()); 1490 } 1491 1492 SmallVector<Expr*, 32> exprs; 1493 1494 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1495 exprs.push_back(TheCall->getArg(i)); 1496 TheCall->setArg(i, 0); 1497 } 1498 1499 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 1500 exprs.size(), resType, 1501 TheCall->getCallee()->getLocStart(), 1502 TheCall->getRParenLoc())); 1503} 1504 1505/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1506// This is declared to take (const void*, ...) and can take two 1507// optional constant int args. 1508bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1509 unsigned NumArgs = TheCall->getNumArgs(); 1510 1511 if (NumArgs > 3) 1512 return Diag(TheCall->getLocEnd(), 1513 diag::err_typecheck_call_too_many_args_at_most) 1514 << 0 /*function call*/ << 3 << NumArgs 1515 << TheCall->getSourceRange(); 1516 1517 // Argument 0 is checked for us and the remaining arguments must be 1518 // constant integers. 1519 for (unsigned i = 1; i != NumArgs; ++i) { 1520 Expr *Arg = TheCall->getArg(i); 1521 1522 // We can't check the value of a dependent argument. 1523 if (Arg->isTypeDependent() || Arg->isValueDependent()) 1524 continue; 1525 1526 llvm::APSInt Result; 1527 if (SemaBuiltinConstantArg(TheCall, i, Result)) 1528 return true; 1529 1530 // FIXME: gcc issues a warning and rewrites these to 0. These 1531 // seems especially odd for the third argument since the default 1532 // is 3. 1533 if (i == 1) { 1534 if (Result.getLimitedValue() > 1) 1535 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1536 << "0" << "1" << Arg->getSourceRange(); 1537 } else { 1538 if (Result.getLimitedValue() > 3) 1539 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1540 << "0" << "3" << Arg->getSourceRange(); 1541 } 1542 } 1543 1544 return false; 1545} 1546 1547/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1548/// TheCall is a constant expression. 1549bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1550 llvm::APSInt &Result) { 1551 Expr *Arg = TheCall->getArg(ArgNum); 1552 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1553 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1554 1555 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 1556 1557 if (!Arg->isIntegerConstantExpr(Result, Context)) 1558 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 1559 << FDecl->getDeclName() << Arg->getSourceRange(); 1560 1561 return false; 1562} 1563 1564/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 1565/// int type). This simply type checks that type is one of the defined 1566/// constants (0-3). 1567// For compatibility check 0-3, llvm only handles 0 and 2. 1568bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 1569 llvm::APSInt Result; 1570 1571 // We can't check the value of a dependent argument. 1572 if (TheCall->getArg(1)->isTypeDependent() || 1573 TheCall->getArg(1)->isValueDependent()) 1574 return false; 1575 1576 // Check constant-ness first. 1577 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1578 return true; 1579 1580 Expr *Arg = TheCall->getArg(1); 1581 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 1582 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1583 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1584 } 1585 1586 return false; 1587} 1588 1589/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1590/// This checks that val is a constant 1. 1591bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1592 Expr *Arg = TheCall->getArg(1); 1593 llvm::APSInt Result; 1594 1595 // TODO: This is less than ideal. Overload this to take a value. 1596 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1597 return true; 1598 1599 if (Result != 1) 1600 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1601 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1602 1603 return false; 1604} 1605 1606// Determine if an expression is a string literal or constant string. 1607// If this function returns false on the arguments to a function expecting a 1608// format string, we will usually need to emit a warning. 1609// True string literals are then checked by CheckFormatString. 1610Sema::StringLiteralCheckType 1611Sema::checkFormatStringExpr(const Expr *E, Expr **Args, 1612 unsigned NumArgs, bool HasVAListArg, 1613 unsigned format_idx, unsigned firstDataArg, 1614 FormatStringType Type, VariadicCallType CallType, 1615 bool inFunctionCall) { 1616 tryAgain: 1617 if (E->isTypeDependent() || E->isValueDependent()) 1618 return SLCT_NotALiteral; 1619 1620 E = E->IgnoreParenCasts(); 1621 1622 if (E->isNullPointerConstant(Context, Expr::NPC_ValueDependentIsNotNull)) 1623 // Technically -Wformat-nonliteral does not warn about this case. 1624 // The behavior of printf and friends in this case is implementation 1625 // dependent. Ideally if the format string cannot be null then 1626 // it should have a 'nonnull' attribute in the function prototype. 1627 return SLCT_CheckedLiteral; 1628 1629 switch (E->getStmtClass()) { 1630 case Stmt::BinaryConditionalOperatorClass: 1631 case Stmt::ConditionalOperatorClass: { 1632 // The expression is a literal if both sub-expressions were, and it was 1633 // completely checked only if both sub-expressions were checked. 1634 const AbstractConditionalOperator *C = 1635 cast<AbstractConditionalOperator>(E); 1636 StringLiteralCheckType Left = 1637 checkFormatStringExpr(C->getTrueExpr(), Args, NumArgs, 1638 HasVAListArg, format_idx, firstDataArg, 1639 Type, CallType, inFunctionCall); 1640 if (Left == SLCT_NotALiteral) 1641 return SLCT_NotALiteral; 1642 StringLiteralCheckType Right = 1643 checkFormatStringExpr(C->getFalseExpr(), Args, NumArgs, 1644 HasVAListArg, format_idx, firstDataArg, 1645 Type, CallType, inFunctionCall); 1646 return Left < Right ? Left : Right; 1647 } 1648 1649 case Stmt::ImplicitCastExprClass: { 1650 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 1651 goto tryAgain; 1652 } 1653 1654 case Stmt::OpaqueValueExprClass: 1655 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 1656 E = src; 1657 goto tryAgain; 1658 } 1659 return SLCT_NotALiteral; 1660 1661 case Stmt::PredefinedExprClass: 1662 // While __func__, etc., are technically not string literals, they 1663 // cannot contain format specifiers and thus are not a security 1664 // liability. 1665 return SLCT_UncheckedLiteral; 1666 1667 case Stmt::DeclRefExprClass: { 1668 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1669 1670 // As an exception, do not flag errors for variables binding to 1671 // const string literals. 1672 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1673 bool isConstant = false; 1674 QualType T = DR->getType(); 1675 1676 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1677 isConstant = AT->getElementType().isConstant(Context); 1678 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1679 isConstant = T.isConstant(Context) && 1680 PT->getPointeeType().isConstant(Context); 1681 } else if (T->isObjCObjectPointerType()) { 1682 // In ObjC, there is usually no "const ObjectPointer" type, 1683 // so don't check if the pointee type is constant. 1684 isConstant = T.isConstant(Context); 1685 } 1686 1687 if (isConstant) { 1688 if (const Expr *Init = VD->getAnyInitializer()) { 1689 // Look through initializers like const char c[] = { "foo" } 1690 if (const InitListExpr *InitList = dyn_cast<InitListExpr>(Init)) { 1691 if (InitList->isStringLiteralInit()) 1692 Init = InitList->getInit(0)->IgnoreParenImpCasts(); 1693 } 1694 return checkFormatStringExpr(Init, Args, NumArgs, 1695 HasVAListArg, format_idx, 1696 firstDataArg, Type, CallType, 1697 /*inFunctionCall*/false); 1698 } 1699 } 1700 1701 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1702 // special check to see if the format string is a function parameter 1703 // of the function calling the printf function. If the function 1704 // has an attribute indicating it is a printf-like function, then we 1705 // should suppress warnings concerning non-literals being used in a call 1706 // to a vprintf function. For example: 1707 // 1708 // void 1709 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1710 // va_list ap; 1711 // va_start(ap, fmt); 1712 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1713 // ... 1714 // 1715 if (HasVAListArg) { 1716 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(VD)) { 1717 if (const NamedDecl *ND = dyn_cast<NamedDecl>(PV->getDeclContext())) { 1718 int PVIndex = PV->getFunctionScopeIndex() + 1; 1719 for (specific_attr_iterator<FormatAttr> 1720 i = ND->specific_attr_begin<FormatAttr>(), 1721 e = ND->specific_attr_end<FormatAttr>(); i != e ; ++i) { 1722 FormatAttr *PVFormat = *i; 1723 // adjust for implicit parameter 1724 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1725 if (MD->isInstance()) 1726 ++PVIndex; 1727 // We also check if the formats are compatible. 1728 // We can't pass a 'scanf' string to a 'printf' function. 1729 if (PVIndex == PVFormat->getFormatIdx() && 1730 Type == GetFormatStringType(PVFormat)) 1731 return SLCT_UncheckedLiteral; 1732 } 1733 } 1734 } 1735 } 1736 } 1737 1738 return SLCT_NotALiteral; 1739 } 1740 1741 case Stmt::CallExprClass: 1742 case Stmt::CXXMemberCallExprClass: { 1743 const CallExpr *CE = cast<CallExpr>(E); 1744 if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(CE->getCalleeDecl())) { 1745 if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) { 1746 unsigned ArgIndex = FA->getFormatIdx(); 1747 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1748 if (MD->isInstance()) 1749 --ArgIndex; 1750 const Expr *Arg = CE->getArg(ArgIndex - 1); 1751 1752 return checkFormatStringExpr(Arg, Args, NumArgs, 1753 HasVAListArg, format_idx, firstDataArg, 1754 Type, CallType, inFunctionCall); 1755 } else if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(ND)) { 1756 unsigned BuiltinID = FD->getBuiltinID(); 1757 if (BuiltinID == Builtin::BI__builtin___CFStringMakeConstantString || 1758 BuiltinID == Builtin::BI__builtin___NSStringMakeConstantString) { 1759 const Expr *Arg = CE->getArg(0); 1760 return checkFormatStringExpr(Arg, Args, NumArgs, 1761 HasVAListArg, format_idx, 1762 firstDataArg, Type, CallType, 1763 inFunctionCall); 1764 } 1765 } 1766 } 1767 1768 return SLCT_NotALiteral; 1769 } 1770 case Stmt::ObjCStringLiteralClass: 1771 case Stmt::StringLiteralClass: { 1772 const StringLiteral *StrE = NULL; 1773 1774 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1775 StrE = ObjCFExpr->getString(); 1776 else 1777 StrE = cast<StringLiteral>(E); 1778 1779 if (StrE) { 1780 CheckFormatString(StrE, E, Args, NumArgs, HasVAListArg, format_idx, 1781 firstDataArg, Type, inFunctionCall, CallType); 1782 return SLCT_CheckedLiteral; 1783 } 1784 1785 return SLCT_NotALiteral; 1786 } 1787 1788 default: 1789 return SLCT_NotALiteral; 1790 } 1791} 1792 1793void 1794Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1795 const Expr * const *ExprArgs, 1796 SourceLocation CallSiteLoc) { 1797 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1798 e = NonNull->args_end(); 1799 i != e; ++i) { 1800 const Expr *ArgExpr = ExprArgs[*i]; 1801 if (ArgExpr->isNullPointerConstant(Context, 1802 Expr::NPC_ValueDependentIsNotNull)) 1803 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 1804 } 1805} 1806 1807Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) { 1808 return llvm::StringSwitch<FormatStringType>(Format->getType()) 1809 .Case("scanf", FST_Scanf) 1810 .Cases("printf", "printf0", FST_Printf) 1811 .Cases("NSString", "CFString", FST_NSString) 1812 .Case("strftime", FST_Strftime) 1813 .Case("strfmon", FST_Strfmon) 1814 .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf) 1815 .Default(FST_Unknown); 1816} 1817 1818/// CheckFormatArguments - Check calls to printf and scanf (and similar 1819/// functions) for correct use of format strings. 1820/// Returns true if a format string has been fully checked. 1821bool Sema::CheckFormatArguments(const FormatAttr *Format, Expr **Args, 1822 unsigned NumArgs, bool IsCXXMember, 1823 VariadicCallType CallType, 1824 SourceLocation Loc, SourceRange Range) { 1825 FormatStringInfo FSI; 1826 if (getFormatStringInfo(Format, IsCXXMember, &FSI)) 1827 return CheckFormatArguments(Args, NumArgs, FSI.HasVAListArg, FSI.FormatIdx, 1828 FSI.FirstDataArg, GetFormatStringType(Format), 1829 CallType, Loc, Range); 1830 return false; 1831} 1832 1833bool Sema::CheckFormatArguments(Expr **Args, unsigned NumArgs, 1834 bool HasVAListArg, unsigned format_idx, 1835 unsigned firstDataArg, FormatStringType Type, 1836 VariadicCallType CallType, 1837 SourceLocation Loc, SourceRange Range) { 1838 // CHECK: printf/scanf-like function is called with no format string. 1839 if (format_idx >= NumArgs) { 1840 Diag(Loc, diag::warn_missing_format_string) << Range; 1841 return false; 1842 } 1843 1844 const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts(); 1845 1846 // CHECK: format string is not a string literal. 1847 // 1848 // Dynamically generated format strings are difficult to 1849 // automatically vet at compile time. Requiring that format strings 1850 // are string literals: (1) permits the checking of format strings by 1851 // the compiler and thereby (2) can practically remove the source of 1852 // many format string exploits. 1853 1854 // Format string can be either ObjC string (e.g. @"%d") or 1855 // C string (e.g. "%d") 1856 // ObjC string uses the same format specifiers as C string, so we can use 1857 // the same format string checking logic for both ObjC and C strings. 1858 StringLiteralCheckType CT = 1859 checkFormatStringExpr(OrigFormatExpr, Args, NumArgs, HasVAListArg, 1860 format_idx, firstDataArg, Type, CallType); 1861 if (CT != SLCT_NotALiteral) 1862 // Literal format string found, check done! 1863 return CT == SLCT_CheckedLiteral; 1864 1865 // Strftime is particular as it always uses a single 'time' argument, 1866 // so it is safe to pass a non-literal string. 1867 if (Type == FST_Strftime) 1868 return false; 1869 1870 // Do not emit diag when the string param is a macro expansion and the 1871 // format is either NSString or CFString. This is a hack to prevent 1872 // diag when using the NSLocalizedString and CFCopyLocalizedString macros 1873 // which are usually used in place of NS and CF string literals. 1874 if (Type == FST_NSString && 1875 SourceMgr.isInSystemMacro(Args[format_idx]->getLocStart())) 1876 return false; 1877 1878 // If there are no arguments specified, warn with -Wformat-security, otherwise 1879 // warn only with -Wformat-nonliteral. 1880 if (NumArgs == format_idx+1) 1881 Diag(Args[format_idx]->getLocStart(), 1882 diag::warn_format_nonliteral_noargs) 1883 << OrigFormatExpr->getSourceRange(); 1884 else 1885 Diag(Args[format_idx]->getLocStart(), 1886 diag::warn_format_nonliteral) 1887 << OrigFormatExpr->getSourceRange(); 1888 return false; 1889} 1890 1891namespace { 1892class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1893protected: 1894 Sema &S; 1895 const StringLiteral *FExpr; 1896 const Expr *OrigFormatExpr; 1897 const unsigned FirstDataArg; 1898 const unsigned NumDataArgs; 1899 const char *Beg; // Start of format string. 1900 const bool HasVAListArg; 1901 const Expr * const *Args; 1902 const unsigned NumArgs; 1903 unsigned FormatIdx; 1904 llvm::BitVector CoveredArgs; 1905 bool usesPositionalArgs; 1906 bool atFirstArg; 1907 bool inFunctionCall; 1908 Sema::VariadicCallType CallType; 1909public: 1910 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1911 const Expr *origFormatExpr, unsigned firstDataArg, 1912 unsigned numDataArgs, const char *beg, bool hasVAListArg, 1913 Expr **args, unsigned numArgs, 1914 unsigned formatIdx, bool inFunctionCall, 1915 Sema::VariadicCallType callType) 1916 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1917 FirstDataArg(firstDataArg), NumDataArgs(numDataArgs), 1918 Beg(beg), HasVAListArg(hasVAListArg), 1919 Args(args), NumArgs(numArgs), FormatIdx(formatIdx), 1920 usesPositionalArgs(false), atFirstArg(true), 1921 inFunctionCall(inFunctionCall), CallType(callType) { 1922 CoveredArgs.resize(numDataArgs); 1923 CoveredArgs.reset(); 1924 } 1925 1926 void DoneProcessing(); 1927 1928 void HandleIncompleteSpecifier(const char *startSpecifier, 1929 unsigned specifierLen); 1930 1931 void HandleNonStandardLengthModifier( 1932 const analyze_format_string::LengthModifier &LM, 1933 const char *startSpecifier, unsigned specifierLen); 1934 1935 void HandleNonStandardConversionSpecifier( 1936 const analyze_format_string::ConversionSpecifier &CS, 1937 const char *startSpecifier, unsigned specifierLen); 1938 1939 void HandleNonStandardConversionSpecification( 1940 const analyze_format_string::LengthModifier &LM, 1941 const analyze_format_string::ConversionSpecifier &CS, 1942 const char *startSpecifier, unsigned specifierLen); 1943 1944 virtual void HandlePosition(const char *startPos, unsigned posLen); 1945 1946 virtual void HandleInvalidPosition(const char *startSpecifier, 1947 unsigned specifierLen, 1948 analyze_format_string::PositionContext p); 1949 1950 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1951 1952 void HandleNullChar(const char *nullCharacter); 1953 1954 template <typename Range> 1955 static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall, 1956 const Expr *ArgumentExpr, 1957 PartialDiagnostic PDiag, 1958 SourceLocation StringLoc, 1959 bool IsStringLocation, Range StringRange, 1960 FixItHint Fixit = FixItHint()); 1961 1962protected: 1963 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1964 const char *startSpec, 1965 unsigned specifierLen, 1966 const char *csStart, unsigned csLen); 1967 1968 void HandlePositionalNonpositionalArgs(SourceLocation Loc, 1969 const char *startSpec, 1970 unsigned specifierLen); 1971 1972 SourceRange getFormatStringRange(); 1973 CharSourceRange getSpecifierRange(const char *startSpecifier, 1974 unsigned specifierLen); 1975 SourceLocation getLocationOfByte(const char *x); 1976 1977 const Expr *getDataArg(unsigned i) const; 1978 1979 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1980 const analyze_format_string::ConversionSpecifier &CS, 1981 const char *startSpecifier, unsigned specifierLen, 1982 unsigned argIndex); 1983 1984 template <typename Range> 1985 void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc, 1986 bool IsStringLocation, Range StringRange, 1987 FixItHint Fixit = FixItHint()); 1988 1989 void CheckPositionalAndNonpositionalArgs( 1990 const analyze_format_string::FormatSpecifier *FS); 1991}; 1992} 1993 1994SourceRange CheckFormatHandler::getFormatStringRange() { 1995 return OrigFormatExpr->getSourceRange(); 1996} 1997 1998CharSourceRange CheckFormatHandler:: 1999getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 2000 SourceLocation Start = getLocationOfByte(startSpecifier); 2001 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 2002 2003 // Advance the end SourceLocation by one due to half-open ranges. 2004 End = End.getLocWithOffset(1); 2005 2006 return CharSourceRange::getCharRange(Start, End); 2007} 2008 2009SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 2010 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 2011} 2012 2013void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 2014 unsigned specifierLen){ 2015 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier), 2016 getLocationOfByte(startSpecifier), 2017 /*IsStringLocation*/true, 2018 getSpecifierRange(startSpecifier, specifierLen)); 2019} 2020 2021void CheckFormatHandler::HandleNonStandardLengthModifier( 2022 const analyze_format_string::LengthModifier &LM, 2023 const char *startSpecifier, unsigned specifierLen) { 2024 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) << LM.toString() 2025 << 0, 2026 getLocationOfByte(LM.getStart()), 2027 /*IsStringLocation*/true, 2028 getSpecifierRange(startSpecifier, specifierLen)); 2029} 2030 2031void CheckFormatHandler::HandleNonStandardConversionSpecifier( 2032 const analyze_format_string::ConversionSpecifier &CS, 2033 const char *startSpecifier, unsigned specifierLen) { 2034 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) << CS.toString() 2035 << 1, 2036 getLocationOfByte(CS.getStart()), 2037 /*IsStringLocation*/true, 2038 getSpecifierRange(startSpecifier, specifierLen)); 2039} 2040 2041void CheckFormatHandler::HandleNonStandardConversionSpecification( 2042 const analyze_format_string::LengthModifier &LM, 2043 const analyze_format_string::ConversionSpecifier &CS, 2044 const char *startSpecifier, unsigned specifierLen) { 2045 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_conversion_spec) 2046 << LM.toString() << CS.toString(), 2047 getLocationOfByte(LM.getStart()), 2048 /*IsStringLocation*/true, 2049 getSpecifierRange(startSpecifier, specifierLen)); 2050} 2051 2052void CheckFormatHandler::HandlePosition(const char *startPos, 2053 unsigned posLen) { 2054 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_positional_arg), 2055 getLocationOfByte(startPos), 2056 /*IsStringLocation*/true, 2057 getSpecifierRange(startPos, posLen)); 2058} 2059 2060void 2061CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 2062 analyze_format_string::PositionContext p) { 2063 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier) 2064 << (unsigned) p, 2065 getLocationOfByte(startPos), /*IsStringLocation*/true, 2066 getSpecifierRange(startPos, posLen)); 2067} 2068 2069void CheckFormatHandler::HandleZeroPosition(const char *startPos, 2070 unsigned posLen) { 2071 EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier), 2072 getLocationOfByte(startPos), 2073 /*IsStringLocation*/true, 2074 getSpecifierRange(startPos, posLen)); 2075} 2076 2077void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 2078 if (!isa<ObjCStringLiteral>(OrigFormatExpr)) { 2079 // The presence of a null character is likely an error. 2080 EmitFormatDiagnostic( 2081 S.PDiag(diag::warn_printf_format_string_contains_null_char), 2082 getLocationOfByte(nullCharacter), /*IsStringLocation*/true, 2083 getFormatStringRange()); 2084 } 2085} 2086 2087// Note that this may return NULL if there was an error parsing or building 2088// one of the argument expressions. 2089const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 2090 return Args[FirstDataArg + i]; 2091} 2092 2093void CheckFormatHandler::DoneProcessing() { 2094 // Does the number of data arguments exceed the number of 2095 // format conversions in the format string? 2096 if (!HasVAListArg) { 2097 // Find any arguments that weren't covered. 2098 CoveredArgs.flip(); 2099 signed notCoveredArg = CoveredArgs.find_first(); 2100 if (notCoveredArg >= 0) { 2101 assert((unsigned)notCoveredArg < NumDataArgs); 2102 if (const Expr *E = getDataArg((unsigned) notCoveredArg)) { 2103 SourceLocation Loc = E->getLocStart(); 2104 if (!S.getSourceManager().isInSystemMacro(Loc)) { 2105 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used), 2106 Loc, /*IsStringLocation*/false, 2107 getFormatStringRange()); 2108 } 2109 } 2110 } 2111 } 2112} 2113 2114bool 2115CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 2116 SourceLocation Loc, 2117 const char *startSpec, 2118 unsigned specifierLen, 2119 const char *csStart, 2120 unsigned csLen) { 2121 2122 bool keepGoing = true; 2123 if (argIndex < NumDataArgs) { 2124 // Consider the argument coverered, even though the specifier doesn't 2125 // make sense. 2126 CoveredArgs.set(argIndex); 2127 } 2128 else { 2129 // If argIndex exceeds the number of data arguments we 2130 // don't issue a warning because that is just a cascade of warnings (and 2131 // they may have intended '%%' anyway). We don't want to continue processing 2132 // the format string after this point, however, as we will like just get 2133 // gibberish when trying to match arguments. 2134 keepGoing = false; 2135 } 2136 2137 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion) 2138 << StringRef(csStart, csLen), 2139 Loc, /*IsStringLocation*/true, 2140 getSpecifierRange(startSpec, specifierLen)); 2141 2142 return keepGoing; 2143} 2144 2145void 2146CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc, 2147 const char *startSpec, 2148 unsigned specifierLen) { 2149 EmitFormatDiagnostic( 2150 S.PDiag(diag::warn_format_mix_positional_nonpositional_args), 2151 Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen)); 2152} 2153 2154bool 2155CheckFormatHandler::CheckNumArgs( 2156 const analyze_format_string::FormatSpecifier &FS, 2157 const analyze_format_string::ConversionSpecifier &CS, 2158 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 2159 2160 if (argIndex >= NumDataArgs) { 2161 PartialDiagnostic PDiag = FS.usesPositionalArg() 2162 ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args) 2163 << (argIndex+1) << NumDataArgs) 2164 : S.PDiag(diag::warn_printf_insufficient_data_args); 2165 EmitFormatDiagnostic( 2166 PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true, 2167 getSpecifierRange(startSpecifier, specifierLen)); 2168 return false; 2169 } 2170 return true; 2171} 2172 2173template<typename Range> 2174void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag, 2175 SourceLocation Loc, 2176 bool IsStringLocation, 2177 Range StringRange, 2178 FixItHint FixIt) { 2179 EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag, 2180 Loc, IsStringLocation, StringRange, FixIt); 2181} 2182 2183/// \brief If the format string is not within the funcion call, emit a note 2184/// so that the function call and string are in diagnostic messages. 2185/// 2186/// \param inFunctionCall if true, the format string is within the function 2187/// call and only one diagnostic message will be produced. Otherwise, an 2188/// extra note will be emitted pointing to location of the format string. 2189/// 2190/// \param ArgumentExpr the expression that is passed as the format string 2191/// argument in the function call. Used for getting locations when two 2192/// diagnostics are emitted. 2193/// 2194/// \param PDiag the callee should already have provided any strings for the 2195/// diagnostic message. This function only adds locations and fixits 2196/// to diagnostics. 2197/// 2198/// \param Loc primary location for diagnostic. If two diagnostics are 2199/// required, one will be at Loc and a new SourceLocation will be created for 2200/// the other one. 2201/// 2202/// \param IsStringLocation if true, Loc points to the format string should be 2203/// used for the note. Otherwise, Loc points to the argument list and will 2204/// be used with PDiag. 2205/// 2206/// \param StringRange some or all of the string to highlight. This is 2207/// templated so it can accept either a CharSourceRange or a SourceRange. 2208/// 2209/// \param Fixit optional fix it hint for the format string. 2210template<typename Range> 2211void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall, 2212 const Expr *ArgumentExpr, 2213 PartialDiagnostic PDiag, 2214 SourceLocation Loc, 2215 bool IsStringLocation, 2216 Range StringRange, 2217 FixItHint FixIt) { 2218 if (InFunctionCall) 2219 S.Diag(Loc, PDiag) << StringRange << FixIt; 2220 else { 2221 S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag) 2222 << ArgumentExpr->getSourceRange(); 2223 S.Diag(IsStringLocation ? Loc : StringRange.getBegin(), 2224 diag::note_format_string_defined) 2225 << StringRange << FixIt; 2226 } 2227} 2228 2229//===--- CHECK: Printf format string checking ------------------------------===// 2230 2231namespace { 2232class CheckPrintfHandler : public CheckFormatHandler { 2233 bool ObjCContext; 2234public: 2235 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 2236 const Expr *origFormatExpr, unsigned firstDataArg, 2237 unsigned numDataArgs, bool isObjC, 2238 const char *beg, bool hasVAListArg, 2239 Expr **Args, unsigned NumArgs, 2240 unsigned formatIdx, bool inFunctionCall, 2241 Sema::VariadicCallType CallType) 2242 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2243 numDataArgs, beg, hasVAListArg, Args, NumArgs, 2244 formatIdx, inFunctionCall, CallType), ObjCContext(isObjC) 2245 {} 2246 2247 2248 bool HandleInvalidPrintfConversionSpecifier( 2249 const analyze_printf::PrintfSpecifier &FS, 2250 const char *startSpecifier, 2251 unsigned specifierLen); 2252 2253 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 2254 const char *startSpecifier, 2255 unsigned specifierLen); 2256 bool checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 2257 const char *StartSpecifier, 2258 unsigned SpecifierLen, 2259 const Expr *E); 2260 2261 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 2262 const char *startSpecifier, unsigned specifierLen); 2263 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 2264 const analyze_printf::OptionalAmount &Amt, 2265 unsigned type, 2266 const char *startSpecifier, unsigned specifierLen); 2267 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2268 const analyze_printf::OptionalFlag &flag, 2269 const char *startSpecifier, unsigned specifierLen); 2270 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 2271 const analyze_printf::OptionalFlag &ignoredFlag, 2272 const analyze_printf::OptionalFlag &flag, 2273 const char *startSpecifier, unsigned specifierLen); 2274 bool checkForCStrMembers(const analyze_printf::ArgTypeResult &ATR, 2275 const Expr *E, const CharSourceRange &CSR); 2276 2277}; 2278} 2279 2280bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 2281 const analyze_printf::PrintfSpecifier &FS, 2282 const char *startSpecifier, 2283 unsigned specifierLen) { 2284 const analyze_printf::PrintfConversionSpecifier &CS = 2285 FS.getConversionSpecifier(); 2286 2287 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2288 getLocationOfByte(CS.getStart()), 2289 startSpecifier, specifierLen, 2290 CS.getStart(), CS.getLength()); 2291} 2292 2293bool CheckPrintfHandler::HandleAmount( 2294 const analyze_format_string::OptionalAmount &Amt, 2295 unsigned k, const char *startSpecifier, 2296 unsigned specifierLen) { 2297 2298 if (Amt.hasDataArgument()) { 2299 if (!HasVAListArg) { 2300 unsigned argIndex = Amt.getArgIndex(); 2301 if (argIndex >= NumDataArgs) { 2302 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg) 2303 << k, 2304 getLocationOfByte(Amt.getStart()), 2305 /*IsStringLocation*/true, 2306 getSpecifierRange(startSpecifier, specifierLen)); 2307 // Don't do any more checking. We will just emit 2308 // spurious errors. 2309 return false; 2310 } 2311 2312 // Type check the data argument. It should be an 'int'. 2313 // Although not in conformance with C99, we also allow the argument to be 2314 // an 'unsigned int' as that is a reasonably safe case. GCC also 2315 // doesn't emit a warning for that case. 2316 CoveredArgs.set(argIndex); 2317 const Expr *Arg = getDataArg(argIndex); 2318 if (!Arg) 2319 return false; 2320 2321 QualType T = Arg->getType(); 2322 2323 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 2324 assert(ATR.isValid()); 2325 2326 if (!ATR.matchesType(S.Context, T)) { 2327 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type) 2328 << k << ATR.getRepresentativeTypeName(S.Context) 2329 << T << Arg->getSourceRange(), 2330 getLocationOfByte(Amt.getStart()), 2331 /*IsStringLocation*/true, 2332 getSpecifierRange(startSpecifier, specifierLen)); 2333 // Don't do any more checking. We will just emit 2334 // spurious errors. 2335 return false; 2336 } 2337 } 2338 } 2339 return true; 2340} 2341 2342void CheckPrintfHandler::HandleInvalidAmount( 2343 const analyze_printf::PrintfSpecifier &FS, 2344 const analyze_printf::OptionalAmount &Amt, 2345 unsigned type, 2346 const char *startSpecifier, 2347 unsigned specifierLen) { 2348 const analyze_printf::PrintfConversionSpecifier &CS = 2349 FS.getConversionSpecifier(); 2350 2351 FixItHint fixit = 2352 Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant 2353 ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 2354 Amt.getConstantLength())) 2355 : FixItHint(); 2356 2357 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount) 2358 << type << CS.toString(), 2359 getLocationOfByte(Amt.getStart()), 2360 /*IsStringLocation*/true, 2361 getSpecifierRange(startSpecifier, specifierLen), 2362 fixit); 2363} 2364 2365void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2366 const analyze_printf::OptionalFlag &flag, 2367 const char *startSpecifier, 2368 unsigned specifierLen) { 2369 // Warn about pointless flag with a fixit removal. 2370 const analyze_printf::PrintfConversionSpecifier &CS = 2371 FS.getConversionSpecifier(); 2372 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag) 2373 << flag.toString() << CS.toString(), 2374 getLocationOfByte(flag.getPosition()), 2375 /*IsStringLocation*/true, 2376 getSpecifierRange(startSpecifier, specifierLen), 2377 FixItHint::CreateRemoval( 2378 getSpecifierRange(flag.getPosition(), 1))); 2379} 2380 2381void CheckPrintfHandler::HandleIgnoredFlag( 2382 const analyze_printf::PrintfSpecifier &FS, 2383 const analyze_printf::OptionalFlag &ignoredFlag, 2384 const analyze_printf::OptionalFlag &flag, 2385 const char *startSpecifier, 2386 unsigned specifierLen) { 2387 // Warn about ignored flag with a fixit removal. 2388 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag) 2389 << ignoredFlag.toString() << flag.toString(), 2390 getLocationOfByte(ignoredFlag.getPosition()), 2391 /*IsStringLocation*/true, 2392 getSpecifierRange(startSpecifier, specifierLen), 2393 FixItHint::CreateRemoval( 2394 getSpecifierRange(ignoredFlag.getPosition(), 1))); 2395} 2396 2397// Determines if the specified is a C++ class or struct containing 2398// a member with the specified name and kind (e.g. a CXXMethodDecl named 2399// "c_str()"). 2400template<typename MemberKind> 2401static llvm::SmallPtrSet<MemberKind*, 1> 2402CXXRecordMembersNamed(StringRef Name, Sema &S, QualType Ty) { 2403 const RecordType *RT = Ty->getAs<RecordType>(); 2404 llvm::SmallPtrSet<MemberKind*, 1> Results; 2405 2406 if (!RT) 2407 return Results; 2408 const CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(RT->getDecl()); 2409 if (!RD) 2410 return Results; 2411 2412 LookupResult R(S, &S.PP.getIdentifierTable().get(Name), SourceLocation(), 2413 Sema::LookupMemberName); 2414 2415 // We just need to include all members of the right kind turned up by the 2416 // filter, at this point. 2417 if (S.LookupQualifiedName(R, RT->getDecl())) 2418 for (LookupResult::iterator I = R.begin(), E = R.end(); I != E; ++I) { 2419 NamedDecl *decl = (*I)->getUnderlyingDecl(); 2420 if (MemberKind *FK = dyn_cast<MemberKind>(decl)) 2421 Results.insert(FK); 2422 } 2423 return Results; 2424} 2425 2426// Check if a (w)string was passed when a (w)char* was needed, and offer a 2427// better diagnostic if so. ATR is assumed to be valid. 2428// Returns true when a c_str() conversion method is found. 2429bool CheckPrintfHandler::checkForCStrMembers( 2430 const analyze_printf::ArgTypeResult &ATR, const Expr *E, 2431 const CharSourceRange &CSR) { 2432 typedef llvm::SmallPtrSet<CXXMethodDecl*, 1> MethodSet; 2433 2434 MethodSet Results = 2435 CXXRecordMembersNamed<CXXMethodDecl>("c_str", S, E->getType()); 2436 2437 for (MethodSet::iterator MI = Results.begin(), ME = Results.end(); 2438 MI != ME; ++MI) { 2439 const CXXMethodDecl *Method = *MI; 2440 if (Method->getNumParams() == 0 && 2441 ATR.matchesType(S.Context, Method->getResultType())) { 2442 // FIXME: Suggest parens if the expression needs them. 2443 SourceLocation EndLoc = 2444 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()); 2445 S.Diag(E->getLocStart(), diag::note_printf_c_str) 2446 << "c_str()" 2447 << FixItHint::CreateInsertion(EndLoc, ".c_str()"); 2448 return true; 2449 } 2450 } 2451 2452 return false; 2453} 2454 2455bool 2456CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 2457 &FS, 2458 const char *startSpecifier, 2459 unsigned specifierLen) { 2460 2461 using namespace analyze_format_string; 2462 using namespace analyze_printf; 2463 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 2464 2465 if (FS.consumesDataArgument()) { 2466 if (atFirstArg) { 2467 atFirstArg = false; 2468 usesPositionalArgs = FS.usesPositionalArg(); 2469 } 2470 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2471 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2472 startSpecifier, specifierLen); 2473 return false; 2474 } 2475 } 2476 2477 // First check if the field width, precision, and conversion specifier 2478 // have matching data arguments. 2479 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 2480 startSpecifier, specifierLen)) { 2481 return false; 2482 } 2483 2484 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 2485 startSpecifier, specifierLen)) { 2486 return false; 2487 } 2488 2489 if (!CS.consumesDataArgument()) { 2490 // FIXME: Technically specifying a precision or field width here 2491 // makes no sense. Worth issuing a warning at some point. 2492 return true; 2493 } 2494 2495 // Consume the argument. 2496 unsigned argIndex = FS.getArgIndex(); 2497 if (argIndex < NumDataArgs) { 2498 // The check to see if the argIndex is valid will come later. 2499 // We set the bit here because we may exit early from this 2500 // function if we encounter some other error. 2501 CoveredArgs.set(argIndex); 2502 } 2503 2504 // Check for using an Objective-C specific conversion specifier 2505 // in a non-ObjC literal. 2506 if (!ObjCContext && CS.isObjCArg()) { 2507 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 2508 specifierLen); 2509 } 2510 2511 // Check for invalid use of field width 2512 if (!FS.hasValidFieldWidth()) { 2513 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 2514 startSpecifier, specifierLen); 2515 } 2516 2517 // Check for invalid use of precision 2518 if (!FS.hasValidPrecision()) { 2519 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 2520 startSpecifier, specifierLen); 2521 } 2522 2523 // Check each flag does not conflict with any other component. 2524 if (!FS.hasValidThousandsGroupingPrefix()) 2525 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 2526 if (!FS.hasValidLeadingZeros()) 2527 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 2528 if (!FS.hasValidPlusPrefix()) 2529 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 2530 if (!FS.hasValidSpacePrefix()) 2531 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 2532 if (!FS.hasValidAlternativeForm()) 2533 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 2534 if (!FS.hasValidLeftJustified()) 2535 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 2536 2537 // Check that flags are not ignored by another flag 2538 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 2539 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 2540 startSpecifier, specifierLen); 2541 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 2542 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 2543 startSpecifier, specifierLen); 2544 2545 // Check the length modifier is valid with the given conversion specifier. 2546 const LengthModifier &LM = FS.getLengthModifier(); 2547 if (!FS.hasValidLengthModifier()) 2548 EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length) 2549 << LM.toString() << CS.toString(), 2550 getLocationOfByte(LM.getStart()), 2551 /*IsStringLocation*/true, 2552 getSpecifierRange(startSpecifier, specifierLen), 2553 FixItHint::CreateRemoval( 2554 getSpecifierRange(LM.getStart(), 2555 LM.getLength()))); 2556 if (!FS.hasStandardLengthModifier()) 2557 HandleNonStandardLengthModifier(LM, startSpecifier, specifierLen); 2558 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 2559 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 2560 if (!FS.hasStandardLengthConversionCombination()) 2561 HandleNonStandardConversionSpecification(LM, CS, startSpecifier, 2562 specifierLen); 2563 2564 // Are we using '%n'? 2565 if (CS.getKind() == ConversionSpecifier::nArg) { 2566 // Issue a warning about this being a possible security issue. 2567 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_write_back), 2568 getLocationOfByte(CS.getStart()), 2569 /*IsStringLocation*/true, 2570 getSpecifierRange(startSpecifier, specifierLen)); 2571 // Continue checking the other format specifiers. 2572 return true; 2573 } 2574 2575 // The remaining checks depend on the data arguments. 2576 if (HasVAListArg) 2577 return true; 2578 2579 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2580 return false; 2581 2582 const Expr *Arg = getDataArg(argIndex); 2583 if (!Arg) 2584 return true; 2585 2586 return checkFormatExpr(FS, startSpecifier, specifierLen, Arg); 2587} 2588 2589bool 2590CheckPrintfHandler::checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 2591 const char *StartSpecifier, 2592 unsigned SpecifierLen, 2593 const Expr *E) { 2594 using namespace analyze_format_string; 2595 using namespace analyze_printf; 2596 // Now type check the data expression that matches the 2597 // format specifier. 2598 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context, 2599 ObjCContext); 2600 if (ATR.isValid() && !ATR.matchesType(S.Context, E->getType())) { 2601 // Look through argument promotions for our error message's reported type. 2602 // This includes the integral and floating promotions, but excludes array 2603 // and function pointer decay; seeing that an argument intended to be a 2604 // string has type 'char [6]' is probably more confusing than 'char *'. 2605 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 2606 if (ICE->getCastKind() == CK_IntegralCast || 2607 ICE->getCastKind() == CK_FloatingCast) { 2608 E = ICE->getSubExpr(); 2609 2610 // Check if we didn't match because of an implicit cast from a 'char' 2611 // or 'short' to an 'int'. This is done because printf is a varargs 2612 // function. 2613 if (ICE->getType() == S.Context.IntTy || 2614 ICE->getType() == S.Context.UnsignedIntTy) { 2615 // All further checking is done on the subexpression. 2616 if (ATR.matchesType(S.Context, E->getType())) 2617 return true; 2618 } 2619 } 2620 } 2621 2622 // We may be able to offer a FixItHint if it is a supported type. 2623 PrintfSpecifier fixedFS = FS; 2624 bool success = fixedFS.fixType(E->getType(), S.getLangOpts(), 2625 S.Context, ObjCContext); 2626 2627 if (success) { 2628 // Get the fix string from the fixed format specifier 2629 SmallString<16> buf; 2630 llvm::raw_svector_ostream os(buf); 2631 fixedFS.toString(os); 2632 2633 EmitFormatDiagnostic( 2634 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2635 << ATR.getRepresentativeTypeName(S.Context) << E->getType() 2636 << E->getSourceRange(), 2637 E->getLocStart(), 2638 /*IsStringLocation*/false, 2639 getSpecifierRange(StartSpecifier, SpecifierLen), 2640 FixItHint::CreateReplacement( 2641 getSpecifierRange(StartSpecifier, SpecifierLen), 2642 os.str())); 2643 } else { 2644 const CharSourceRange &CSR = getSpecifierRange(StartSpecifier, 2645 SpecifierLen); 2646 // Since the warning for passing non-POD types to variadic functions 2647 // was deferred until now, we emit a warning for non-POD 2648 // arguments here. 2649 if (S.isValidVarArgType(E->getType()) == Sema::VAK_Invalid) { 2650 unsigned DiagKind; 2651 if (E->getType()->isObjCObjectType()) 2652 DiagKind = diag::err_cannot_pass_objc_interface_to_vararg_format; 2653 else 2654 DiagKind = diag::warn_non_pod_vararg_with_format_string; 2655 2656 EmitFormatDiagnostic( 2657 S.PDiag(DiagKind) 2658 << S.getLangOpts().CPlusPlus0x 2659 << E->getType() 2660 << CallType 2661 << ATR.getRepresentativeTypeName(S.Context) 2662 << CSR 2663 << E->getSourceRange(), 2664 E->getLocStart(), /*IsStringLocation*/false, CSR); 2665 2666 checkForCStrMembers(ATR, E, CSR); 2667 } else 2668 EmitFormatDiagnostic( 2669 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2670 << ATR.getRepresentativeTypeName(S.Context) << E->getType() 2671 << CSR 2672 << E->getSourceRange(), 2673 E->getLocStart(), /*IsStringLocation*/false, CSR); 2674 } 2675 } 2676 2677 return true; 2678} 2679 2680//===--- CHECK: Scanf format string checking ------------------------------===// 2681 2682namespace { 2683class CheckScanfHandler : public CheckFormatHandler { 2684public: 2685 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 2686 const Expr *origFormatExpr, unsigned firstDataArg, 2687 unsigned numDataArgs, const char *beg, bool hasVAListArg, 2688 Expr **Args, unsigned NumArgs, 2689 unsigned formatIdx, bool inFunctionCall, 2690 Sema::VariadicCallType CallType) 2691 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2692 numDataArgs, beg, hasVAListArg, 2693 Args, NumArgs, formatIdx, inFunctionCall, CallType) 2694 {} 2695 2696 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 2697 const char *startSpecifier, 2698 unsigned specifierLen); 2699 2700 bool HandleInvalidScanfConversionSpecifier( 2701 const analyze_scanf::ScanfSpecifier &FS, 2702 const char *startSpecifier, 2703 unsigned specifierLen); 2704 2705 void HandleIncompleteScanList(const char *start, const char *end); 2706}; 2707} 2708 2709void CheckScanfHandler::HandleIncompleteScanList(const char *start, 2710 const char *end) { 2711 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 2712 getLocationOfByte(end), /*IsStringLocation*/true, 2713 getSpecifierRange(start, end - start)); 2714} 2715 2716bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 2717 const analyze_scanf::ScanfSpecifier &FS, 2718 const char *startSpecifier, 2719 unsigned specifierLen) { 2720 2721 const analyze_scanf::ScanfConversionSpecifier &CS = 2722 FS.getConversionSpecifier(); 2723 2724 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2725 getLocationOfByte(CS.getStart()), 2726 startSpecifier, specifierLen, 2727 CS.getStart(), CS.getLength()); 2728} 2729 2730bool CheckScanfHandler::HandleScanfSpecifier( 2731 const analyze_scanf::ScanfSpecifier &FS, 2732 const char *startSpecifier, 2733 unsigned specifierLen) { 2734 2735 using namespace analyze_scanf; 2736 using namespace analyze_format_string; 2737 2738 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 2739 2740 // Handle case where '%' and '*' don't consume an argument. These shouldn't 2741 // be used to decide if we are using positional arguments consistently. 2742 if (FS.consumesDataArgument()) { 2743 if (atFirstArg) { 2744 atFirstArg = false; 2745 usesPositionalArgs = FS.usesPositionalArg(); 2746 } 2747 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2748 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2749 startSpecifier, specifierLen); 2750 return false; 2751 } 2752 } 2753 2754 // Check if the field with is non-zero. 2755 const OptionalAmount &Amt = FS.getFieldWidth(); 2756 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 2757 if (Amt.getConstantAmount() == 0) { 2758 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 2759 Amt.getConstantLength()); 2760 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 2761 getLocationOfByte(Amt.getStart()), 2762 /*IsStringLocation*/true, R, 2763 FixItHint::CreateRemoval(R)); 2764 } 2765 } 2766 2767 if (!FS.consumesDataArgument()) { 2768 // FIXME: Technically specifying a precision or field width here 2769 // makes no sense. Worth issuing a warning at some point. 2770 return true; 2771 } 2772 2773 // Consume the argument. 2774 unsigned argIndex = FS.getArgIndex(); 2775 if (argIndex < NumDataArgs) { 2776 // The check to see if the argIndex is valid will come later. 2777 // We set the bit here because we may exit early from this 2778 // function if we encounter some other error. 2779 CoveredArgs.set(argIndex); 2780 } 2781 2782 // Check the length modifier is valid with the given conversion specifier. 2783 const LengthModifier &LM = FS.getLengthModifier(); 2784 if (!FS.hasValidLengthModifier()) { 2785 const CharSourceRange &R = getSpecifierRange(LM.getStart(), LM.getLength()); 2786 EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length) 2787 << LM.toString() << CS.toString() 2788 << getSpecifierRange(startSpecifier, specifierLen), 2789 getLocationOfByte(LM.getStart()), 2790 /*IsStringLocation*/true, R, 2791 FixItHint::CreateRemoval(R)); 2792 } 2793 2794 if (!FS.hasStandardLengthModifier()) 2795 HandleNonStandardLengthModifier(LM, startSpecifier, specifierLen); 2796 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 2797 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 2798 if (!FS.hasStandardLengthConversionCombination()) 2799 HandleNonStandardConversionSpecification(LM, CS, startSpecifier, 2800 specifierLen); 2801 2802 // The remaining checks depend on the data arguments. 2803 if (HasVAListArg) 2804 return true; 2805 2806 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2807 return false; 2808 2809 // Check that the argument type matches the format specifier. 2810 const Expr *Ex = getDataArg(argIndex); 2811 if (!Ex) 2812 return true; 2813 2814 const analyze_scanf::ScanfArgTypeResult &ATR = FS.getArgType(S.Context); 2815 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 2816 ScanfSpecifier fixedFS = FS; 2817 bool success = fixedFS.fixType(Ex->getType(), S.getLangOpts(), 2818 S.Context); 2819 2820 if (success) { 2821 // Get the fix string from the fixed format specifier. 2822 SmallString<128> buf; 2823 llvm::raw_svector_ostream os(buf); 2824 fixedFS.toString(os); 2825 2826 EmitFormatDiagnostic( 2827 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2828 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2829 << Ex->getSourceRange(), 2830 Ex->getLocStart(), 2831 /*IsStringLocation*/false, 2832 getSpecifierRange(startSpecifier, specifierLen), 2833 FixItHint::CreateReplacement( 2834 getSpecifierRange(startSpecifier, specifierLen), 2835 os.str())); 2836 } else { 2837 EmitFormatDiagnostic( 2838 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2839 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2840 << Ex->getSourceRange(), 2841 Ex->getLocStart(), 2842 /*IsStringLocation*/false, 2843 getSpecifierRange(startSpecifier, specifierLen)); 2844 } 2845 } 2846 2847 return true; 2848} 2849 2850void Sema::CheckFormatString(const StringLiteral *FExpr, 2851 const Expr *OrigFormatExpr, 2852 Expr **Args, unsigned NumArgs, 2853 bool HasVAListArg, unsigned format_idx, 2854 unsigned firstDataArg, FormatStringType Type, 2855 bool inFunctionCall, VariadicCallType CallType) { 2856 2857 // CHECK: is the format string a wide literal? 2858 if (!FExpr->isAscii() && !FExpr->isUTF8()) { 2859 CheckFormatHandler::EmitFormatDiagnostic( 2860 *this, inFunctionCall, Args[format_idx], 2861 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 2862 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2863 return; 2864 } 2865 2866 // Str - The format string. NOTE: this is NOT null-terminated! 2867 StringRef StrRef = FExpr->getString(); 2868 const char *Str = StrRef.data(); 2869 unsigned StrLen = StrRef.size(); 2870 const unsigned numDataArgs = NumArgs - firstDataArg; 2871 2872 // CHECK: empty format string? 2873 if (StrLen == 0 && numDataArgs > 0) { 2874 CheckFormatHandler::EmitFormatDiagnostic( 2875 *this, inFunctionCall, Args[format_idx], 2876 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 2877 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2878 return; 2879 } 2880 2881 if (Type == FST_Printf || Type == FST_NSString) { 2882 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2883 numDataArgs, (Type == FST_NSString), 2884 Str, HasVAListArg, Args, NumArgs, format_idx, 2885 inFunctionCall, CallType); 2886 2887 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen, 2888 getLangOpts())) 2889 H.DoneProcessing(); 2890 } else if (Type == FST_Scanf) { 2891 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, numDataArgs, 2892 Str, HasVAListArg, Args, NumArgs, format_idx, 2893 inFunctionCall, CallType); 2894 2895 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen, 2896 getLangOpts())) 2897 H.DoneProcessing(); 2898 } // TODO: handle other formats 2899} 2900 2901//===--- CHECK: Standard memory functions ---------------------------------===// 2902 2903/// \brief Determine whether the given type is a dynamic class type (e.g., 2904/// whether it has a vtable). 2905static bool isDynamicClassType(QualType T) { 2906 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 2907 if (CXXRecordDecl *Definition = Record->getDefinition()) 2908 if (Definition->isDynamicClass()) 2909 return true; 2910 2911 return false; 2912} 2913 2914/// \brief If E is a sizeof expression, returns its argument expression, 2915/// otherwise returns NULL. 2916static const Expr *getSizeOfExprArg(const Expr* E) { 2917 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2918 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2919 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 2920 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 2921 2922 return 0; 2923} 2924 2925/// \brief If E is a sizeof expression, returns its argument type. 2926static QualType getSizeOfArgType(const Expr* E) { 2927 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2928 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2929 if (SizeOf->getKind() == clang::UETT_SizeOf) 2930 return SizeOf->getTypeOfArgument(); 2931 2932 return QualType(); 2933} 2934 2935/// \brief Check for dangerous or invalid arguments to memset(). 2936/// 2937/// This issues warnings on known problematic, dangerous or unspecified 2938/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 2939/// function calls. 2940/// 2941/// \param Call The call expression to diagnose. 2942void Sema::CheckMemaccessArguments(const CallExpr *Call, 2943 unsigned BId, 2944 IdentifierInfo *FnName) { 2945 assert(BId != 0); 2946 2947 // It is possible to have a non-standard definition of memset. Validate 2948 // we have enough arguments, and if not, abort further checking. 2949 unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3); 2950 if (Call->getNumArgs() < ExpectedNumArgs) 2951 return; 2952 2953 unsigned LastArg = (BId == Builtin::BImemset || 2954 BId == Builtin::BIstrndup ? 1 : 2); 2955 unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2); 2956 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 2957 2958 // We have special checking when the length is a sizeof expression. 2959 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 2960 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 2961 llvm::FoldingSetNodeID SizeOfArgID; 2962 2963 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 2964 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 2965 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 2966 2967 QualType DestTy = Dest->getType(); 2968 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 2969 QualType PointeeTy = DestPtrTy->getPointeeType(); 2970 2971 // Never warn about void type pointers. This can be used to suppress 2972 // false positives. 2973 if (PointeeTy->isVoidType()) 2974 continue; 2975 2976 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 2977 // actually comparing the expressions for equality. Because computing the 2978 // expression IDs can be expensive, we only do this if the diagnostic is 2979 // enabled. 2980 if (SizeOfArg && 2981 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 2982 SizeOfArg->getExprLoc())) { 2983 // We only compute IDs for expressions if the warning is enabled, and 2984 // cache the sizeof arg's ID. 2985 if (SizeOfArgID == llvm::FoldingSetNodeID()) 2986 SizeOfArg->Profile(SizeOfArgID, Context, true); 2987 llvm::FoldingSetNodeID DestID; 2988 Dest->Profile(DestID, Context, true); 2989 if (DestID == SizeOfArgID) { 2990 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 2991 // over sizeof(src) as well. 2992 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 2993 StringRef ReadableName = FnName->getName(); 2994 2995 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 2996 if (UnaryOp->getOpcode() == UO_AddrOf) 2997 ActionIdx = 1; // If its an address-of operator, just remove it. 2998 if (Context.getTypeSize(PointeeTy) == Context.getCharWidth()) 2999 ActionIdx = 2; // If the pointee's size is sizeof(char), 3000 // suggest an explicit length. 3001 3002 // If the function is defined as a builtin macro, do not show macro 3003 // expansion. 3004 SourceLocation SL = SizeOfArg->getExprLoc(); 3005 SourceRange DSR = Dest->getSourceRange(); 3006 SourceRange SSR = SizeOfArg->getSourceRange(); 3007 SourceManager &SM = PP.getSourceManager(); 3008 3009 if (SM.isMacroArgExpansion(SL)) { 3010 ReadableName = Lexer::getImmediateMacroName(SL, SM, LangOpts); 3011 SL = SM.getSpellingLoc(SL); 3012 DSR = SourceRange(SM.getSpellingLoc(DSR.getBegin()), 3013 SM.getSpellingLoc(DSR.getEnd())); 3014 SSR = SourceRange(SM.getSpellingLoc(SSR.getBegin()), 3015 SM.getSpellingLoc(SSR.getEnd())); 3016 } 3017 3018 DiagRuntimeBehavior(SL, SizeOfArg, 3019 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 3020 << ReadableName 3021 << PointeeTy 3022 << DestTy 3023 << DSR 3024 << SSR); 3025 DiagRuntimeBehavior(SL, SizeOfArg, 3026 PDiag(diag::warn_sizeof_pointer_expr_memaccess_note) 3027 << ActionIdx 3028 << SSR); 3029 3030 break; 3031 } 3032 } 3033 3034 // Also check for cases where the sizeof argument is the exact same 3035 // type as the memory argument, and where it points to a user-defined 3036 // record type. 3037 if (SizeOfArgTy != QualType()) { 3038 if (PointeeTy->isRecordType() && 3039 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 3040 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 3041 PDiag(diag::warn_sizeof_pointer_type_memaccess) 3042 << FnName << SizeOfArgTy << ArgIdx 3043 << PointeeTy << Dest->getSourceRange() 3044 << LenExpr->getSourceRange()); 3045 break; 3046 } 3047 } 3048 3049 // Always complain about dynamic classes. 3050 if (isDynamicClassType(PointeeTy)) { 3051 3052 unsigned OperationType = 0; 3053 // "overwritten" if we're warning about the destination for any call 3054 // but memcmp; otherwise a verb appropriate to the call. 3055 if (ArgIdx != 0 || BId == Builtin::BImemcmp) { 3056 if (BId == Builtin::BImemcpy) 3057 OperationType = 1; 3058 else if(BId == Builtin::BImemmove) 3059 OperationType = 2; 3060 else if (BId == Builtin::BImemcmp) 3061 OperationType = 3; 3062 } 3063 3064 DiagRuntimeBehavior( 3065 Dest->getExprLoc(), Dest, 3066 PDiag(diag::warn_dyn_class_memaccess) 3067 << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx) 3068 << FnName << PointeeTy 3069 << OperationType 3070 << Call->getCallee()->getSourceRange()); 3071 } else if (PointeeTy.hasNonTrivialObjCLifetime() && 3072 BId != Builtin::BImemset) 3073 DiagRuntimeBehavior( 3074 Dest->getExprLoc(), Dest, 3075 PDiag(diag::warn_arc_object_memaccess) 3076 << ArgIdx << FnName << PointeeTy 3077 << Call->getCallee()->getSourceRange()); 3078 else 3079 continue; 3080 3081 DiagRuntimeBehavior( 3082 Dest->getExprLoc(), Dest, 3083 PDiag(diag::note_bad_memaccess_silence) 3084 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 3085 break; 3086 } 3087 } 3088} 3089 3090// A little helper routine: ignore addition and subtraction of integer literals. 3091// This intentionally does not ignore all integer constant expressions because 3092// we don't want to remove sizeof(). 3093static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 3094 Ex = Ex->IgnoreParenCasts(); 3095 3096 for (;;) { 3097 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 3098 if (!BO || !BO->isAdditiveOp()) 3099 break; 3100 3101 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 3102 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 3103 3104 if (isa<IntegerLiteral>(RHS)) 3105 Ex = LHS; 3106 else if (isa<IntegerLiteral>(LHS)) 3107 Ex = RHS; 3108 else 3109 break; 3110 } 3111 3112 return Ex; 3113} 3114 3115// Warn if the user has made the 'size' argument to strlcpy or strlcat 3116// be the size of the source, instead of the destination. 3117void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 3118 IdentifierInfo *FnName) { 3119 3120 // Don't crash if the user has the wrong number of arguments 3121 if (Call->getNumArgs() != 3) 3122 return; 3123 3124 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 3125 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 3126 const Expr *CompareWithSrc = NULL; 3127 3128 // Look for 'strlcpy(dst, x, sizeof(x))' 3129 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 3130 CompareWithSrc = Ex; 3131 else { 3132 // Look for 'strlcpy(dst, x, strlen(x))' 3133 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 3134 if (SizeCall->isBuiltinCall() == Builtin::BIstrlen 3135 && SizeCall->getNumArgs() == 1) 3136 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 3137 } 3138 } 3139 3140 if (!CompareWithSrc) 3141 return; 3142 3143 // Determine if the argument to sizeof/strlen is equal to the source 3144 // argument. In principle there's all kinds of things you could do 3145 // here, for instance creating an == expression and evaluating it with 3146 // EvaluateAsBooleanCondition, but this uses a more direct technique: 3147 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 3148 if (!SrcArgDRE) 3149 return; 3150 3151 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 3152 if (!CompareWithSrcDRE || 3153 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 3154 return; 3155 3156 const Expr *OriginalSizeArg = Call->getArg(2); 3157 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 3158 << OriginalSizeArg->getSourceRange() << FnName; 3159 3160 // Output a FIXIT hint if the destination is an array (rather than a 3161 // pointer to an array). This could be enhanced to handle some 3162 // pointers if we know the actual size, like if DstArg is 'array+2' 3163 // we could say 'sizeof(array)-2'. 3164 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 3165 QualType DstArgTy = DstArg->getType(); 3166 3167 // Only handle constant-sized or VLAs, but not flexible members. 3168 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 3169 // Only issue the FIXIT for arrays of size > 1. 3170 if (CAT->getSize().getSExtValue() <= 1) 3171 return; 3172 } else if (!DstArgTy->isVariableArrayType()) { 3173 return; 3174 } 3175 3176 SmallString<128> sizeString; 3177 llvm::raw_svector_ostream OS(sizeString); 3178 OS << "sizeof("; 3179 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 3180 OS << ")"; 3181 3182 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 3183 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 3184 OS.str()); 3185} 3186 3187/// Check if two expressions refer to the same declaration. 3188static bool referToTheSameDecl(const Expr *E1, const Expr *E2) { 3189 if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1)) 3190 if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2)) 3191 return D1->getDecl() == D2->getDecl(); 3192 return false; 3193} 3194 3195static const Expr *getStrlenExprArg(const Expr *E) { 3196 if (const CallExpr *CE = dyn_cast<CallExpr>(E)) { 3197 const FunctionDecl *FD = CE->getDirectCallee(); 3198 if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen) 3199 return 0; 3200 return CE->getArg(0)->IgnoreParenCasts(); 3201 } 3202 return 0; 3203} 3204 3205// Warn on anti-patterns as the 'size' argument to strncat. 3206// The correct size argument should look like following: 3207// strncat(dst, src, sizeof(dst) - strlen(dest) - 1); 3208void Sema::CheckStrncatArguments(const CallExpr *CE, 3209 IdentifierInfo *FnName) { 3210 // Don't crash if the user has the wrong number of arguments. 3211 if (CE->getNumArgs() < 3) 3212 return; 3213 const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts(); 3214 const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts(); 3215 const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts(); 3216 3217 // Identify common expressions, which are wrongly used as the size argument 3218 // to strncat and may lead to buffer overflows. 3219 unsigned PatternType = 0; 3220 if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) { 3221 // - sizeof(dst) 3222 if (referToTheSameDecl(SizeOfArg, DstArg)) 3223 PatternType = 1; 3224 // - sizeof(src) 3225 else if (referToTheSameDecl(SizeOfArg, SrcArg)) 3226 PatternType = 2; 3227 } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) { 3228 if (BE->getOpcode() == BO_Sub) { 3229 const Expr *L = BE->getLHS()->IgnoreParenCasts(); 3230 const Expr *R = BE->getRHS()->IgnoreParenCasts(); 3231 // - sizeof(dst) - strlen(dst) 3232 if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) && 3233 referToTheSameDecl(DstArg, getStrlenExprArg(R))) 3234 PatternType = 1; 3235 // - sizeof(src) - (anything) 3236 else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L))) 3237 PatternType = 2; 3238 } 3239 } 3240 3241 if (PatternType == 0) 3242 return; 3243 3244 // Generate the diagnostic. 3245 SourceLocation SL = LenArg->getLocStart(); 3246 SourceRange SR = LenArg->getSourceRange(); 3247 SourceManager &SM = PP.getSourceManager(); 3248 3249 // If the function is defined as a builtin macro, do not show macro expansion. 3250 if (SM.isMacroArgExpansion(SL)) { 3251 SL = SM.getSpellingLoc(SL); 3252 SR = SourceRange(SM.getSpellingLoc(SR.getBegin()), 3253 SM.getSpellingLoc(SR.getEnd())); 3254 } 3255 3256 if (PatternType == 1) 3257 Diag(SL, diag::warn_strncat_large_size) << SR; 3258 else 3259 Diag(SL, diag::warn_strncat_src_size) << SR; 3260 3261 // Output a FIXIT hint if the destination is an array (rather than a 3262 // pointer to an array). This could be enhanced to handle some 3263 // pointers if we know the actual size, like if DstArg is 'array+2' 3264 // we could say 'sizeof(array)-2'. 3265 QualType DstArgTy = DstArg->getType(); 3266 3267 // Only handle constant-sized or VLAs, but not flexible members. 3268 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 3269 // Only issue the FIXIT for arrays of size > 1. 3270 if (CAT->getSize().getSExtValue() <= 1) 3271 return; 3272 } else if (!DstArgTy->isVariableArrayType()) { 3273 return; 3274 } 3275 3276 SmallString<128> sizeString; 3277 llvm::raw_svector_ostream OS(sizeString); 3278 OS << "sizeof("; 3279 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 3280 OS << ") - "; 3281 OS << "strlen("; 3282 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 3283 OS << ") - 1"; 3284 3285 Diag(SL, diag::note_strncat_wrong_size) 3286 << FixItHint::CreateReplacement(SR, OS.str()); 3287} 3288 3289//===--- CHECK: Return Address of Stack Variable --------------------------===// 3290 3291static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3292 Decl *ParentDecl); 3293static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars, 3294 Decl *ParentDecl); 3295 3296/// CheckReturnStackAddr - Check if a return statement returns the address 3297/// of a stack variable. 3298void 3299Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 3300 SourceLocation ReturnLoc) { 3301 3302 Expr *stackE = 0; 3303 SmallVector<DeclRefExpr *, 8> refVars; 3304 3305 // Perform checking for returned stack addresses, local blocks, 3306 // label addresses or references to temporaries. 3307 if (lhsType->isPointerType() || 3308 (!getLangOpts().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 3309 stackE = EvalAddr(RetValExp, refVars, /*ParentDecl=*/0); 3310 } else if (lhsType->isReferenceType()) { 3311 stackE = EvalVal(RetValExp, refVars, /*ParentDecl=*/0); 3312 } 3313 3314 if (stackE == 0) 3315 return; // Nothing suspicious was found. 3316 3317 SourceLocation diagLoc; 3318 SourceRange diagRange; 3319 if (refVars.empty()) { 3320 diagLoc = stackE->getLocStart(); 3321 diagRange = stackE->getSourceRange(); 3322 } else { 3323 // We followed through a reference variable. 'stackE' contains the 3324 // problematic expression but we will warn at the return statement pointing 3325 // at the reference variable. We will later display the "trail" of 3326 // reference variables using notes. 3327 diagLoc = refVars[0]->getLocStart(); 3328 diagRange = refVars[0]->getSourceRange(); 3329 } 3330 3331 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 3332 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 3333 : diag::warn_ret_stack_addr) 3334 << DR->getDecl()->getDeclName() << diagRange; 3335 } else if (isa<BlockExpr>(stackE)) { // local block. 3336 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 3337 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 3338 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 3339 } else { // local temporary. 3340 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 3341 : diag::warn_ret_local_temp_addr) 3342 << diagRange; 3343 } 3344 3345 // Display the "trail" of reference variables that we followed until we 3346 // found the problematic expression using notes. 3347 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 3348 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 3349 // If this var binds to another reference var, show the range of the next 3350 // var, otherwise the var binds to the problematic expression, in which case 3351 // show the range of the expression. 3352 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 3353 : stackE->getSourceRange(); 3354 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 3355 << VD->getDeclName() << range; 3356 } 3357} 3358 3359/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 3360/// check if the expression in a return statement evaluates to an address 3361/// to a location on the stack, a local block, an address of a label, or a 3362/// reference to local temporary. The recursion is used to traverse the 3363/// AST of the return expression, with recursion backtracking when we 3364/// encounter a subexpression that (1) clearly does not lead to one of the 3365/// above problematic expressions (2) is something we cannot determine leads to 3366/// a problematic expression based on such local checking. 3367/// 3368/// Both EvalAddr and EvalVal follow through reference variables to evaluate 3369/// the expression that they point to. Such variables are added to the 3370/// 'refVars' vector so that we know what the reference variable "trail" was. 3371/// 3372/// EvalAddr processes expressions that are pointers that are used as 3373/// references (and not L-values). EvalVal handles all other values. 3374/// At the base case of the recursion is a check for the above problematic 3375/// expressions. 3376/// 3377/// This implementation handles: 3378/// 3379/// * pointer-to-pointer casts 3380/// * implicit conversions from array references to pointers 3381/// * taking the address of fields 3382/// * arbitrary interplay between "&" and "*" operators 3383/// * pointer arithmetic from an address of a stack variable 3384/// * taking the address of an array element where the array is on the stack 3385static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3386 Decl *ParentDecl) { 3387 if (E->isTypeDependent()) 3388 return NULL; 3389 3390 // We should only be called for evaluating pointer expressions. 3391 assert((E->getType()->isAnyPointerType() || 3392 E->getType()->isBlockPointerType() || 3393 E->getType()->isObjCQualifiedIdType()) && 3394 "EvalAddr only works on pointers"); 3395 3396 E = E->IgnoreParens(); 3397 3398 // Our "symbolic interpreter" is just a dispatch off the currently 3399 // viewed AST node. We then recursively traverse the AST by calling 3400 // EvalAddr and EvalVal appropriately. 3401 switch (E->getStmtClass()) { 3402 case Stmt::DeclRefExprClass: { 3403 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3404 3405 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 3406 // If this is a reference variable, follow through to the expression that 3407 // it points to. 3408 if (V->hasLocalStorage() && 3409 V->getType()->isReferenceType() && V->hasInit()) { 3410 // Add the reference variable to the "trail". 3411 refVars.push_back(DR); 3412 return EvalAddr(V->getInit(), refVars, ParentDecl); 3413 } 3414 3415 return NULL; 3416 } 3417 3418 case Stmt::UnaryOperatorClass: { 3419 // The only unary operator that make sense to handle here 3420 // is AddrOf. All others don't make sense as pointers. 3421 UnaryOperator *U = cast<UnaryOperator>(E); 3422 3423 if (U->getOpcode() == UO_AddrOf) 3424 return EvalVal(U->getSubExpr(), refVars, ParentDecl); 3425 else 3426 return NULL; 3427 } 3428 3429 case Stmt::BinaryOperatorClass: { 3430 // Handle pointer arithmetic. All other binary operators are not valid 3431 // in this context. 3432 BinaryOperator *B = cast<BinaryOperator>(E); 3433 BinaryOperatorKind op = B->getOpcode(); 3434 3435 if (op != BO_Add && op != BO_Sub) 3436 return NULL; 3437 3438 Expr *Base = B->getLHS(); 3439 3440 // Determine which argument is the real pointer base. It could be 3441 // the RHS argument instead of the LHS. 3442 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 3443 3444 assert (Base->getType()->isPointerType()); 3445 return EvalAddr(Base, refVars, ParentDecl); 3446 } 3447 3448 // For conditional operators we need to see if either the LHS or RHS are 3449 // valid DeclRefExpr*s. If one of them is valid, we return it. 3450 case Stmt::ConditionalOperatorClass: { 3451 ConditionalOperator *C = cast<ConditionalOperator>(E); 3452 3453 // Handle the GNU extension for missing LHS. 3454 if (Expr *lhsExpr = C->getLHS()) { 3455 // In C++, we can have a throw-expression, which has 'void' type. 3456 if (!lhsExpr->getType()->isVoidType()) 3457 if (Expr* LHS = EvalAddr(lhsExpr, refVars, ParentDecl)) 3458 return LHS; 3459 } 3460 3461 // In C++, we can have a throw-expression, which has 'void' type. 3462 if (C->getRHS()->getType()->isVoidType()) 3463 return NULL; 3464 3465 return EvalAddr(C->getRHS(), refVars, ParentDecl); 3466 } 3467 3468 case Stmt::BlockExprClass: 3469 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 3470 return E; // local block. 3471 return NULL; 3472 3473 case Stmt::AddrLabelExprClass: 3474 return E; // address of label. 3475 3476 case Stmt::ExprWithCleanupsClass: 3477 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars, 3478 ParentDecl); 3479 3480 // For casts, we need to handle conversions from arrays to 3481 // pointer values, and pointer-to-pointer conversions. 3482 case Stmt::ImplicitCastExprClass: 3483 case Stmt::CStyleCastExprClass: 3484 case Stmt::CXXFunctionalCastExprClass: 3485 case Stmt::ObjCBridgedCastExprClass: 3486 case Stmt::CXXStaticCastExprClass: 3487 case Stmt::CXXDynamicCastExprClass: 3488 case Stmt::CXXConstCastExprClass: 3489 case Stmt::CXXReinterpretCastExprClass: { 3490 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 3491 switch (cast<CastExpr>(E)->getCastKind()) { 3492 case CK_BitCast: 3493 case CK_LValueToRValue: 3494 case CK_NoOp: 3495 case CK_BaseToDerived: 3496 case CK_DerivedToBase: 3497 case CK_UncheckedDerivedToBase: 3498 case CK_Dynamic: 3499 case CK_CPointerToObjCPointerCast: 3500 case CK_BlockPointerToObjCPointerCast: 3501 case CK_AnyPointerToBlockPointerCast: 3502 return EvalAddr(SubExpr, refVars, ParentDecl); 3503 3504 case CK_ArrayToPointerDecay: 3505 return EvalVal(SubExpr, refVars, ParentDecl); 3506 3507 default: 3508 return 0; 3509 } 3510 } 3511 3512 case Stmt::MaterializeTemporaryExprClass: 3513 if (Expr *Result = EvalAddr( 3514 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3515 refVars, ParentDecl)) 3516 return Result; 3517 3518 return E; 3519 3520 // Everything else: we simply don't reason about them. 3521 default: 3522 return NULL; 3523 } 3524} 3525 3526 3527/// EvalVal - This function is complements EvalAddr in the mutual recursion. 3528/// See the comments for EvalAddr for more details. 3529static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3530 Decl *ParentDecl) { 3531do { 3532 // We should only be called for evaluating non-pointer expressions, or 3533 // expressions with a pointer type that are not used as references but instead 3534 // are l-values (e.g., DeclRefExpr with a pointer type). 3535 3536 // Our "symbolic interpreter" is just a dispatch off the currently 3537 // viewed AST node. We then recursively traverse the AST by calling 3538 // EvalAddr and EvalVal appropriately. 3539 3540 E = E->IgnoreParens(); 3541 switch (E->getStmtClass()) { 3542 case Stmt::ImplicitCastExprClass: { 3543 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 3544 if (IE->getValueKind() == VK_LValue) { 3545 E = IE->getSubExpr(); 3546 continue; 3547 } 3548 return NULL; 3549 } 3550 3551 case Stmt::ExprWithCleanupsClass: 3552 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,ParentDecl); 3553 3554 case Stmt::DeclRefExprClass: { 3555 // When we hit a DeclRefExpr we are looking at code that refers to a 3556 // variable's name. If it's not a reference variable we check if it has 3557 // local storage within the function, and if so, return the expression. 3558 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3559 3560 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) { 3561 // Check if it refers to itself, e.g. "int& i = i;". 3562 if (V == ParentDecl) 3563 return DR; 3564 3565 if (V->hasLocalStorage()) { 3566 if (!V->getType()->isReferenceType()) 3567 return DR; 3568 3569 // Reference variable, follow through to the expression that 3570 // it points to. 3571 if (V->hasInit()) { 3572 // Add the reference variable to the "trail". 3573 refVars.push_back(DR); 3574 return EvalVal(V->getInit(), refVars, V); 3575 } 3576 } 3577 } 3578 3579 return NULL; 3580 } 3581 3582 case Stmt::UnaryOperatorClass: { 3583 // The only unary operator that make sense to handle here 3584 // is Deref. All others don't resolve to a "name." This includes 3585 // handling all sorts of rvalues passed to a unary operator. 3586 UnaryOperator *U = cast<UnaryOperator>(E); 3587 3588 if (U->getOpcode() == UO_Deref) 3589 return EvalAddr(U->getSubExpr(), refVars, ParentDecl); 3590 3591 return NULL; 3592 } 3593 3594 case Stmt::ArraySubscriptExprClass: { 3595 // Array subscripts are potential references to data on the stack. We 3596 // retrieve the DeclRefExpr* for the array variable if it indeed 3597 // has local storage. 3598 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars,ParentDecl); 3599 } 3600 3601 case Stmt::ConditionalOperatorClass: { 3602 // For conditional operators we need to see if either the LHS or RHS are 3603 // non-NULL Expr's. If one is non-NULL, we return it. 3604 ConditionalOperator *C = cast<ConditionalOperator>(E); 3605 3606 // Handle the GNU extension for missing LHS. 3607 if (Expr *lhsExpr = C->getLHS()) 3608 if (Expr *LHS = EvalVal(lhsExpr, refVars, ParentDecl)) 3609 return LHS; 3610 3611 return EvalVal(C->getRHS(), refVars, ParentDecl); 3612 } 3613 3614 // Accesses to members are potential references to data on the stack. 3615 case Stmt::MemberExprClass: { 3616 MemberExpr *M = cast<MemberExpr>(E); 3617 3618 // Check for indirect access. We only want direct field accesses. 3619 if (M->isArrow()) 3620 return NULL; 3621 3622 // Check whether the member type is itself a reference, in which case 3623 // we're not going to refer to the member, but to what the member refers to. 3624 if (M->getMemberDecl()->getType()->isReferenceType()) 3625 return NULL; 3626 3627 return EvalVal(M->getBase(), refVars, ParentDecl); 3628 } 3629 3630 case Stmt::MaterializeTemporaryExprClass: 3631 if (Expr *Result = EvalVal( 3632 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3633 refVars, ParentDecl)) 3634 return Result; 3635 3636 return E; 3637 3638 default: 3639 // Check that we don't return or take the address of a reference to a 3640 // temporary. This is only useful in C++. 3641 if (!E->isTypeDependent() && E->isRValue()) 3642 return E; 3643 3644 // Everything else: we simply don't reason about them. 3645 return NULL; 3646 } 3647} while (true); 3648} 3649 3650//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 3651 3652/// Check for comparisons of floating point operands using != and ==. 3653/// Issue a warning if these are no self-comparisons, as they are not likely 3654/// to do what the programmer intended. 3655void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 3656 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 3657 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 3658 3659 // Special case: check for x == x (which is OK). 3660 // Do not emit warnings for such cases. 3661 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 3662 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 3663 if (DRL->getDecl() == DRR->getDecl()) 3664 return; 3665 3666 3667 // Special case: check for comparisons against literals that can be exactly 3668 // represented by APFloat. In such cases, do not emit a warning. This 3669 // is a heuristic: often comparison against such literals are used to 3670 // detect if a value in a variable has not changed. This clearly can 3671 // lead to false negatives. 3672 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 3673 if (FLL->isExact()) 3674 return; 3675 } else 3676 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)) 3677 if (FLR->isExact()) 3678 return; 3679 3680 // Check for comparisons with builtin types. 3681 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 3682 if (CL->isBuiltinCall()) 3683 return; 3684 3685 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 3686 if (CR->isBuiltinCall()) 3687 return; 3688 3689 // Emit the diagnostic. 3690 Diag(Loc, diag::warn_floatingpoint_eq) 3691 << LHS->getSourceRange() << RHS->getSourceRange(); 3692} 3693 3694//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 3695//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 3696 3697namespace { 3698 3699/// Structure recording the 'active' range of an integer-valued 3700/// expression. 3701struct IntRange { 3702 /// The number of bits active in the int. 3703 unsigned Width; 3704 3705 /// True if the int is known not to have negative values. 3706 bool NonNegative; 3707 3708 IntRange(unsigned Width, bool NonNegative) 3709 : Width(Width), NonNegative(NonNegative) 3710 {} 3711 3712 /// Returns the range of the bool type. 3713 static IntRange forBoolType() { 3714 return IntRange(1, true); 3715 } 3716 3717 /// Returns the range of an opaque value of the given integral type. 3718 static IntRange forValueOfType(ASTContext &C, QualType T) { 3719 return forValueOfCanonicalType(C, 3720 T->getCanonicalTypeInternal().getTypePtr()); 3721 } 3722 3723 /// Returns the range of an opaque value of a canonical integral type. 3724 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 3725 assert(T->isCanonicalUnqualified()); 3726 3727 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3728 T = VT->getElementType().getTypePtr(); 3729 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3730 T = CT->getElementType().getTypePtr(); 3731 3732 // For enum types, use the known bit width of the enumerators. 3733 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 3734 EnumDecl *Enum = ET->getDecl(); 3735 if (!Enum->isCompleteDefinition()) 3736 return IntRange(C.getIntWidth(QualType(T, 0)), false); 3737 3738 unsigned NumPositive = Enum->getNumPositiveBits(); 3739 unsigned NumNegative = Enum->getNumNegativeBits(); 3740 3741 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 3742 } 3743 3744 const BuiltinType *BT = cast<BuiltinType>(T); 3745 assert(BT->isInteger()); 3746 3747 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3748 } 3749 3750 /// Returns the "target" range of a canonical integral type, i.e. 3751 /// the range of values expressible in the type. 3752 /// 3753 /// This matches forValueOfCanonicalType except that enums have the 3754 /// full range of their type, not the range of their enumerators. 3755 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 3756 assert(T->isCanonicalUnqualified()); 3757 3758 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3759 T = VT->getElementType().getTypePtr(); 3760 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3761 T = CT->getElementType().getTypePtr(); 3762 if (const EnumType *ET = dyn_cast<EnumType>(T)) 3763 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 3764 3765 const BuiltinType *BT = cast<BuiltinType>(T); 3766 assert(BT->isInteger()); 3767 3768 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3769 } 3770 3771 /// Returns the supremum of two ranges: i.e. their conservative merge. 3772 static IntRange join(IntRange L, IntRange R) { 3773 return IntRange(std::max(L.Width, R.Width), 3774 L.NonNegative && R.NonNegative); 3775 } 3776 3777 /// Returns the infinum of two ranges: i.e. their aggressive merge. 3778 static IntRange meet(IntRange L, IntRange R) { 3779 return IntRange(std::min(L.Width, R.Width), 3780 L.NonNegative || R.NonNegative); 3781 } 3782}; 3783 3784static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, 3785 unsigned MaxWidth) { 3786 if (value.isSigned() && value.isNegative()) 3787 return IntRange(value.getMinSignedBits(), false); 3788 3789 if (value.getBitWidth() > MaxWidth) 3790 value = value.trunc(MaxWidth); 3791 3792 // isNonNegative() just checks the sign bit without considering 3793 // signedness. 3794 return IntRange(value.getActiveBits(), true); 3795} 3796 3797static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 3798 unsigned MaxWidth) { 3799 if (result.isInt()) 3800 return GetValueRange(C, result.getInt(), MaxWidth); 3801 3802 if (result.isVector()) { 3803 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 3804 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 3805 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 3806 R = IntRange::join(R, El); 3807 } 3808 return R; 3809 } 3810 3811 if (result.isComplexInt()) { 3812 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 3813 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 3814 return IntRange::join(R, I); 3815 } 3816 3817 // This can happen with lossless casts to intptr_t of "based" lvalues. 3818 // Assume it might use arbitrary bits. 3819 // FIXME: The only reason we need to pass the type in here is to get 3820 // the sign right on this one case. It would be nice if APValue 3821 // preserved this. 3822 assert(result.isLValue() || result.isAddrLabelDiff()); 3823 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 3824} 3825 3826/// Pseudo-evaluate the given integer expression, estimating the 3827/// range of values it might take. 3828/// 3829/// \param MaxWidth - the width to which the value will be truncated 3830static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 3831 E = E->IgnoreParens(); 3832 3833 // Try a full evaluation first. 3834 Expr::EvalResult result; 3835 if (E->EvaluateAsRValue(result, C)) 3836 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 3837 3838 // I think we only want to look through implicit casts here; if the 3839 // user has an explicit widening cast, we should treat the value as 3840 // being of the new, wider type. 3841 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 3842 if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue) 3843 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 3844 3845 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 3846 3847 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 3848 3849 // Assume that non-integer casts can span the full range of the type. 3850 if (!isIntegerCast) 3851 return OutputTypeRange; 3852 3853 IntRange SubRange 3854 = GetExprRange(C, CE->getSubExpr(), 3855 std::min(MaxWidth, OutputTypeRange.Width)); 3856 3857 // Bail out if the subexpr's range is as wide as the cast type. 3858 if (SubRange.Width >= OutputTypeRange.Width) 3859 return OutputTypeRange; 3860 3861 // Otherwise, we take the smaller width, and we're non-negative if 3862 // either the output type or the subexpr is. 3863 return IntRange(SubRange.Width, 3864 SubRange.NonNegative || OutputTypeRange.NonNegative); 3865 } 3866 3867 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 3868 // If we can fold the condition, just take that operand. 3869 bool CondResult; 3870 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 3871 return GetExprRange(C, CondResult ? CO->getTrueExpr() 3872 : CO->getFalseExpr(), 3873 MaxWidth); 3874 3875 // Otherwise, conservatively merge. 3876 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 3877 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 3878 return IntRange::join(L, R); 3879 } 3880 3881 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 3882 switch (BO->getOpcode()) { 3883 3884 // Boolean-valued operations are single-bit and positive. 3885 case BO_LAnd: 3886 case BO_LOr: 3887 case BO_LT: 3888 case BO_GT: 3889 case BO_LE: 3890 case BO_GE: 3891 case BO_EQ: 3892 case BO_NE: 3893 return IntRange::forBoolType(); 3894 3895 // The type of the assignments is the type of the LHS, so the RHS 3896 // is not necessarily the same type. 3897 case BO_MulAssign: 3898 case BO_DivAssign: 3899 case BO_RemAssign: 3900 case BO_AddAssign: 3901 case BO_SubAssign: 3902 case BO_XorAssign: 3903 case BO_OrAssign: 3904 // TODO: bitfields? 3905 return IntRange::forValueOfType(C, E->getType()); 3906 3907 // Simple assignments just pass through the RHS, which will have 3908 // been coerced to the LHS type. 3909 case BO_Assign: 3910 // TODO: bitfields? 3911 return GetExprRange(C, BO->getRHS(), MaxWidth); 3912 3913 // Operations with opaque sources are black-listed. 3914 case BO_PtrMemD: 3915 case BO_PtrMemI: 3916 return IntRange::forValueOfType(C, E->getType()); 3917 3918 // Bitwise-and uses the *infinum* of the two source ranges. 3919 case BO_And: 3920 case BO_AndAssign: 3921 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 3922 GetExprRange(C, BO->getRHS(), MaxWidth)); 3923 3924 // Left shift gets black-listed based on a judgement call. 3925 case BO_Shl: 3926 // ...except that we want to treat '1 << (blah)' as logically 3927 // positive. It's an important idiom. 3928 if (IntegerLiteral *I 3929 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 3930 if (I->getValue() == 1) { 3931 IntRange R = IntRange::forValueOfType(C, E->getType()); 3932 return IntRange(R.Width, /*NonNegative*/ true); 3933 } 3934 } 3935 // fallthrough 3936 3937 case BO_ShlAssign: 3938 return IntRange::forValueOfType(C, E->getType()); 3939 3940 // Right shift by a constant can narrow its left argument. 3941 case BO_Shr: 3942 case BO_ShrAssign: { 3943 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3944 3945 // If the shift amount is a positive constant, drop the width by 3946 // that much. 3947 llvm::APSInt shift; 3948 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 3949 shift.isNonNegative()) { 3950 unsigned zext = shift.getZExtValue(); 3951 if (zext >= L.Width) 3952 L.Width = (L.NonNegative ? 0 : 1); 3953 else 3954 L.Width -= zext; 3955 } 3956 3957 return L; 3958 } 3959 3960 // Comma acts as its right operand. 3961 case BO_Comma: 3962 return GetExprRange(C, BO->getRHS(), MaxWidth); 3963 3964 // Black-list pointer subtractions. 3965 case BO_Sub: 3966 if (BO->getLHS()->getType()->isPointerType()) 3967 return IntRange::forValueOfType(C, E->getType()); 3968 break; 3969 3970 // The width of a division result is mostly determined by the size 3971 // of the LHS. 3972 case BO_Div: { 3973 // Don't 'pre-truncate' the operands. 3974 unsigned opWidth = C.getIntWidth(E->getType()); 3975 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3976 3977 // If the divisor is constant, use that. 3978 llvm::APSInt divisor; 3979 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 3980 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 3981 if (log2 >= L.Width) 3982 L.Width = (L.NonNegative ? 0 : 1); 3983 else 3984 L.Width = std::min(L.Width - log2, MaxWidth); 3985 return L; 3986 } 3987 3988 // Otherwise, just use the LHS's width. 3989 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3990 return IntRange(L.Width, L.NonNegative && R.NonNegative); 3991 } 3992 3993 // The result of a remainder can't be larger than the result of 3994 // either side. 3995 case BO_Rem: { 3996 // Don't 'pre-truncate' the operands. 3997 unsigned opWidth = C.getIntWidth(E->getType()); 3998 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3999 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 4000 4001 IntRange meet = IntRange::meet(L, R); 4002 meet.Width = std::min(meet.Width, MaxWidth); 4003 return meet; 4004 } 4005 4006 // The default behavior is okay for these. 4007 case BO_Mul: 4008 case BO_Add: 4009 case BO_Xor: 4010 case BO_Or: 4011 break; 4012 } 4013 4014 // The default case is to treat the operation as if it were closed 4015 // on the narrowest type that encompasses both operands. 4016 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 4017 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 4018 return IntRange::join(L, R); 4019 } 4020 4021 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 4022 switch (UO->getOpcode()) { 4023 // Boolean-valued operations are white-listed. 4024 case UO_LNot: 4025 return IntRange::forBoolType(); 4026 4027 // Operations with opaque sources are black-listed. 4028 case UO_Deref: 4029 case UO_AddrOf: // should be impossible 4030 return IntRange::forValueOfType(C, E->getType()); 4031 4032 default: 4033 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 4034 } 4035 } 4036 4037 if (dyn_cast<OffsetOfExpr>(E)) { 4038 IntRange::forValueOfType(C, E->getType()); 4039 } 4040 4041 if (FieldDecl *BitField = E->getBitField()) 4042 return IntRange(BitField->getBitWidthValue(C), 4043 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 4044 4045 return IntRange::forValueOfType(C, E->getType()); 4046} 4047 4048static IntRange GetExprRange(ASTContext &C, Expr *E) { 4049 return GetExprRange(C, E, C.getIntWidth(E->getType())); 4050} 4051 4052/// Checks whether the given value, which currently has the given 4053/// source semantics, has the same value when coerced through the 4054/// target semantics. 4055static bool IsSameFloatAfterCast(const llvm::APFloat &value, 4056 const llvm::fltSemantics &Src, 4057 const llvm::fltSemantics &Tgt) { 4058 llvm::APFloat truncated = value; 4059 4060 bool ignored; 4061 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 4062 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 4063 4064 return truncated.bitwiseIsEqual(value); 4065} 4066 4067/// Checks whether the given value, which currently has the given 4068/// source semantics, has the same value when coerced through the 4069/// target semantics. 4070/// 4071/// The value might be a vector of floats (or a complex number). 4072static bool IsSameFloatAfterCast(const APValue &value, 4073 const llvm::fltSemantics &Src, 4074 const llvm::fltSemantics &Tgt) { 4075 if (value.isFloat()) 4076 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 4077 4078 if (value.isVector()) { 4079 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 4080 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 4081 return false; 4082 return true; 4083 } 4084 4085 assert(value.isComplexFloat()); 4086 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 4087 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 4088} 4089 4090static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 4091 4092static bool IsZero(Sema &S, Expr *E) { 4093 // Suppress cases where we are comparing against an enum constant. 4094 if (const DeclRefExpr *DR = 4095 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 4096 if (isa<EnumConstantDecl>(DR->getDecl())) 4097 return false; 4098 4099 // Suppress cases where the '0' value is expanded from a macro. 4100 if (E->getLocStart().isMacroID()) 4101 return false; 4102 4103 llvm::APSInt Value; 4104 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 4105} 4106 4107static bool HasEnumType(Expr *E) { 4108 // Strip off implicit integral promotions. 4109 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 4110 if (ICE->getCastKind() != CK_IntegralCast && 4111 ICE->getCastKind() != CK_NoOp) 4112 break; 4113 E = ICE->getSubExpr(); 4114 } 4115 4116 return E->getType()->isEnumeralType(); 4117} 4118 4119static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 4120 BinaryOperatorKind op = E->getOpcode(); 4121 if (E->isValueDependent()) 4122 return; 4123 4124 if (op == BO_LT && IsZero(S, E->getRHS())) { 4125 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4126 << "< 0" << "false" << HasEnumType(E->getLHS()) 4127 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4128 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 4129 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4130 << ">= 0" << "true" << HasEnumType(E->getLHS()) 4131 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4132 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 4133 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4134 << "0 >" << "false" << HasEnumType(E->getRHS()) 4135 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4136 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 4137 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4138 << "0 <=" << "true" << HasEnumType(E->getRHS()) 4139 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4140 } 4141} 4142 4143/// Analyze the operands of the given comparison. Implements the 4144/// fallback case from AnalyzeComparison. 4145static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 4146 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4147 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4148} 4149 4150/// \brief Implements -Wsign-compare. 4151/// 4152/// \param E the binary operator to check for warnings 4153static void AnalyzeComparison(Sema &S, BinaryOperator *E) { 4154 // The type the comparison is being performed in. 4155 QualType T = E->getLHS()->getType(); 4156 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 4157 && "comparison with mismatched types"); 4158 4159 // We don't do anything special if this isn't an unsigned integral 4160 // comparison: we're only interested in integral comparisons, and 4161 // signed comparisons only happen in cases we don't care to warn about. 4162 // 4163 // We also don't care about value-dependent expressions or expressions 4164 // whose result is a constant. 4165 if (!T->hasUnsignedIntegerRepresentation() 4166 || E->isValueDependent() || E->isIntegerConstantExpr(S.Context)) 4167 return AnalyzeImpConvsInComparison(S, E); 4168 4169 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 4170 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 4171 4172 // Check to see if one of the (unmodified) operands is of different 4173 // signedness. 4174 Expr *signedOperand, *unsignedOperand; 4175 if (LHS->getType()->hasSignedIntegerRepresentation()) { 4176 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 4177 "unsigned comparison between two signed integer expressions?"); 4178 signedOperand = LHS; 4179 unsignedOperand = RHS; 4180 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 4181 signedOperand = RHS; 4182 unsignedOperand = LHS; 4183 } else { 4184 CheckTrivialUnsignedComparison(S, E); 4185 return AnalyzeImpConvsInComparison(S, E); 4186 } 4187 4188 // Otherwise, calculate the effective range of the signed operand. 4189 IntRange signedRange = GetExprRange(S.Context, signedOperand); 4190 4191 // Go ahead and analyze implicit conversions in the operands. Note 4192 // that we skip the implicit conversions on both sides. 4193 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 4194 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 4195 4196 // If the signed range is non-negative, -Wsign-compare won't fire, 4197 // but we should still check for comparisons which are always true 4198 // or false. 4199 if (signedRange.NonNegative) 4200 return CheckTrivialUnsignedComparison(S, E); 4201 4202 // For (in)equality comparisons, if the unsigned operand is a 4203 // constant which cannot collide with a overflowed signed operand, 4204 // then reinterpreting the signed operand as unsigned will not 4205 // change the result of the comparison. 4206 if (E->isEqualityOp()) { 4207 unsigned comparisonWidth = S.Context.getIntWidth(T); 4208 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 4209 4210 // We should never be unable to prove that the unsigned operand is 4211 // non-negative. 4212 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 4213 4214 if (unsignedRange.Width < comparisonWidth) 4215 return; 4216 } 4217 4218 S.DiagRuntimeBehavior(E->getOperatorLoc(), E, 4219 S.PDiag(diag::warn_mixed_sign_comparison) 4220 << LHS->getType() << RHS->getType() 4221 << LHS->getSourceRange() << RHS->getSourceRange()); 4222} 4223 4224/// Analyzes an attempt to assign the given value to a bitfield. 4225/// 4226/// Returns true if there was something fishy about the attempt. 4227static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 4228 SourceLocation InitLoc) { 4229 assert(Bitfield->isBitField()); 4230 if (Bitfield->isInvalidDecl()) 4231 return false; 4232 4233 // White-list bool bitfields. 4234 if (Bitfield->getType()->isBooleanType()) 4235 return false; 4236 4237 // Ignore value- or type-dependent expressions. 4238 if (Bitfield->getBitWidth()->isValueDependent() || 4239 Bitfield->getBitWidth()->isTypeDependent() || 4240 Init->isValueDependent() || 4241 Init->isTypeDependent()) 4242 return false; 4243 4244 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 4245 4246 llvm::APSInt Value; 4247 if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects)) 4248 return false; 4249 4250 unsigned OriginalWidth = Value.getBitWidth(); 4251 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 4252 4253 if (OriginalWidth <= FieldWidth) 4254 return false; 4255 4256 // Compute the value which the bitfield will contain. 4257 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 4258 TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType()); 4259 4260 // Check whether the stored value is equal to the original value. 4261 TruncatedValue = TruncatedValue.extend(OriginalWidth); 4262 if (Value == TruncatedValue) 4263 return false; 4264 4265 // Special-case bitfields of width 1: booleans are naturally 0/1, and 4266 // therefore don't strictly fit into a signed bitfield of width 1. 4267 if (FieldWidth == 1 && Value == 1) 4268 return false; 4269 4270 std::string PrettyValue = Value.toString(10); 4271 std::string PrettyTrunc = TruncatedValue.toString(10); 4272 4273 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 4274 << PrettyValue << PrettyTrunc << OriginalInit->getType() 4275 << Init->getSourceRange(); 4276 4277 return true; 4278} 4279 4280/// Analyze the given simple or compound assignment for warning-worthy 4281/// operations. 4282static void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 4283 // Just recurse on the LHS. 4284 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4285 4286 // We want to recurse on the RHS as normal unless we're assigning to 4287 // a bitfield. 4288 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 4289 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 4290 E->getOperatorLoc())) { 4291 // Recurse, ignoring any implicit conversions on the RHS. 4292 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 4293 E->getOperatorLoc()); 4294 } 4295 } 4296 4297 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4298} 4299 4300/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4301static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 4302 SourceLocation CContext, unsigned diag, 4303 bool pruneControlFlow = false) { 4304 if (pruneControlFlow) { 4305 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4306 S.PDiag(diag) 4307 << SourceType << T << E->getSourceRange() 4308 << SourceRange(CContext)); 4309 return; 4310 } 4311 S.Diag(E->getExprLoc(), diag) 4312 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 4313} 4314 4315/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4316static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, 4317 SourceLocation CContext, unsigned diag, 4318 bool pruneControlFlow = false) { 4319 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow); 4320} 4321 4322/// Diagnose an implicit cast from a literal expression. Does not warn when the 4323/// cast wouldn't lose information. 4324void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 4325 SourceLocation CContext) { 4326 // Try to convert the literal exactly to an integer. If we can, don't warn. 4327 bool isExact = false; 4328 const llvm::APFloat &Value = FL->getValue(); 4329 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 4330 T->hasUnsignedIntegerRepresentation()); 4331 if (Value.convertToInteger(IntegerValue, 4332 llvm::APFloat::rmTowardZero, &isExact) 4333 == llvm::APFloat::opOK && isExact) 4334 return; 4335 4336 SmallString<16> PrettySourceValue; 4337 Value.toString(PrettySourceValue); 4338 SmallString<16> PrettyTargetValue; 4339 if (T->isSpecificBuiltinType(BuiltinType::Bool)) 4340 PrettyTargetValue = IntegerValue == 0 ? "false" : "true"; 4341 else 4342 IntegerValue.toString(PrettyTargetValue); 4343 4344 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 4345 << FL->getType() << T.getUnqualifiedType() << PrettySourceValue 4346 << PrettyTargetValue << FL->getSourceRange() << SourceRange(CContext); 4347} 4348 4349std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 4350 if (!Range.Width) return "0"; 4351 4352 llvm::APSInt ValueInRange = Value; 4353 ValueInRange.setIsSigned(!Range.NonNegative); 4354 ValueInRange = ValueInRange.trunc(Range.Width); 4355 return ValueInRange.toString(10); 4356} 4357 4358void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 4359 SourceLocation CC, bool *ICContext = 0) { 4360 if (E->isTypeDependent() || E->isValueDependent()) return; 4361 4362 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 4363 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 4364 if (Source == Target) return; 4365 if (Target->isDependentType()) return; 4366 4367 // If the conversion context location is invalid don't complain. We also 4368 // don't want to emit a warning if the issue occurs from the expansion of 4369 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 4370 // delay this check as long as possible. Once we detect we are in that 4371 // scenario, we just return. 4372 if (CC.isInvalid()) 4373 return; 4374 4375 // Diagnose implicit casts to bool. 4376 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 4377 if (isa<StringLiteral>(E)) 4378 // Warn on string literal to bool. Checks for string literals in logical 4379 // expressions, for instances, assert(0 && "error here"), is prevented 4380 // by a check in AnalyzeImplicitConversions(). 4381 return DiagnoseImpCast(S, E, T, CC, 4382 diag::warn_impcast_string_literal_to_bool); 4383 if (Source->isFunctionType()) { 4384 // Warn on function to bool. Checks free functions and static member 4385 // functions. Weakly imported functions are excluded from the check, 4386 // since it's common to test their value to check whether the linker 4387 // found a definition for them. 4388 ValueDecl *D = 0; 4389 if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) { 4390 D = R->getDecl(); 4391 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 4392 D = M->getMemberDecl(); 4393 } 4394 4395 if (D && !D->isWeak()) { 4396 if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) { 4397 S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool) 4398 << F << E->getSourceRange() << SourceRange(CC); 4399 S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence) 4400 << FixItHint::CreateInsertion(E->getExprLoc(), "&"); 4401 QualType ReturnType; 4402 UnresolvedSet<4> NonTemplateOverloads; 4403 S.isExprCallable(*E, ReturnType, NonTemplateOverloads); 4404 if (!ReturnType.isNull() 4405 && ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 4406 S.Diag(E->getExprLoc(), diag::note_function_to_bool_call) 4407 << FixItHint::CreateInsertion( 4408 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()"); 4409 return; 4410 } 4411 } 4412 } 4413 } 4414 4415 // Strip vector types. 4416 if (isa<VectorType>(Source)) { 4417 if (!isa<VectorType>(Target)) { 4418 if (S.SourceMgr.isInSystemMacro(CC)) 4419 return; 4420 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 4421 } 4422 4423 // If the vector cast is cast between two vectors of the same size, it is 4424 // a bitcast, not a conversion. 4425 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 4426 return; 4427 4428 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 4429 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 4430 } 4431 4432 // Strip complex types. 4433 if (isa<ComplexType>(Source)) { 4434 if (!isa<ComplexType>(Target)) { 4435 if (S.SourceMgr.isInSystemMacro(CC)) 4436 return; 4437 4438 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 4439 } 4440 4441 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 4442 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 4443 } 4444 4445 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 4446 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 4447 4448 // If the source is floating point... 4449 if (SourceBT && SourceBT->isFloatingPoint()) { 4450 // ...and the target is floating point... 4451 if (TargetBT && TargetBT->isFloatingPoint()) { 4452 // ...then warn if we're dropping FP rank. 4453 4454 // Builtin FP kinds are ordered by increasing FP rank. 4455 if (SourceBT->getKind() > TargetBT->getKind()) { 4456 // Don't warn about float constants that are precisely 4457 // representable in the target type. 4458 Expr::EvalResult result; 4459 if (E->EvaluateAsRValue(result, S.Context)) { 4460 // Value might be a float, a float vector, or a float complex. 4461 if (IsSameFloatAfterCast(result.Val, 4462 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 4463 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 4464 return; 4465 } 4466 4467 if (S.SourceMgr.isInSystemMacro(CC)) 4468 return; 4469 4470 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 4471 } 4472 return; 4473 } 4474 4475 // If the target is integral, always warn. 4476 if (TargetBT && TargetBT->isInteger()) { 4477 if (S.SourceMgr.isInSystemMacro(CC)) 4478 return; 4479 4480 Expr *InnerE = E->IgnoreParenImpCasts(); 4481 // We also want to warn on, e.g., "int i = -1.234" 4482 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 4483 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 4484 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 4485 4486 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 4487 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 4488 } else { 4489 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 4490 } 4491 } 4492 4493 return; 4494 } 4495 4496 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 4497 == Expr::NPCK_GNUNull) && !Target->isAnyPointerType() 4498 && !Target->isBlockPointerType() && !Target->isMemberPointerType()) { 4499 SourceLocation Loc = E->getSourceRange().getBegin(); 4500 if (Loc.isMacroID()) 4501 Loc = S.SourceMgr.getImmediateExpansionRange(Loc).first; 4502 if (!Loc.isMacroID() || CC.isMacroID()) 4503 S.Diag(Loc, diag::warn_impcast_null_pointer_to_integer) 4504 << T << clang::SourceRange(CC) 4505 << FixItHint::CreateReplacement(Loc, S.getFixItZeroLiteralForType(T)); 4506 } 4507 4508 if (!Source->isIntegerType() || !Target->isIntegerType()) 4509 return; 4510 4511 // TODO: remove this early return once the false positives for constant->bool 4512 // in templates, macros, etc, are reduced or removed. 4513 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 4514 return; 4515 4516 IntRange SourceRange = GetExprRange(S.Context, E); 4517 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 4518 4519 if (SourceRange.Width > TargetRange.Width) { 4520 // If the source is a constant, use a default-on diagnostic. 4521 // TODO: this should happen for bitfield stores, too. 4522 llvm::APSInt Value(32); 4523 if (E->isIntegerConstantExpr(Value, S.Context)) { 4524 if (S.SourceMgr.isInSystemMacro(CC)) 4525 return; 4526 4527 std::string PrettySourceValue = Value.toString(10); 4528 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 4529 4530 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4531 S.PDiag(diag::warn_impcast_integer_precision_constant) 4532 << PrettySourceValue << PrettyTargetValue 4533 << E->getType() << T << E->getSourceRange() 4534 << clang::SourceRange(CC)); 4535 return; 4536 } 4537 4538 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 4539 if (S.SourceMgr.isInSystemMacro(CC)) 4540 return; 4541 4542 if (TargetRange.Width == 32 && S.Context.getIntWidth(E->getType()) == 64) 4543 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32, 4544 /* pruneControlFlow */ true); 4545 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 4546 } 4547 4548 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 4549 (!TargetRange.NonNegative && SourceRange.NonNegative && 4550 SourceRange.Width == TargetRange.Width)) { 4551 4552 if (S.SourceMgr.isInSystemMacro(CC)) 4553 return; 4554 4555 unsigned DiagID = diag::warn_impcast_integer_sign; 4556 4557 // Traditionally, gcc has warned about this under -Wsign-compare. 4558 // We also want to warn about it in -Wconversion. 4559 // So if -Wconversion is off, use a completely identical diagnostic 4560 // in the sign-compare group. 4561 // The conditional-checking code will 4562 if (ICContext) { 4563 DiagID = diag::warn_impcast_integer_sign_conditional; 4564 *ICContext = true; 4565 } 4566 4567 return DiagnoseImpCast(S, E, T, CC, DiagID); 4568 } 4569 4570 // Diagnose conversions between different enumeration types. 4571 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 4572 // type, to give us better diagnostics. 4573 QualType SourceType = E->getType(); 4574 if (!S.getLangOpts().CPlusPlus) { 4575 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 4576 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 4577 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 4578 SourceType = S.Context.getTypeDeclType(Enum); 4579 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 4580 } 4581 } 4582 4583 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 4584 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 4585 if ((SourceEnum->getDecl()->getIdentifier() || 4586 SourceEnum->getDecl()->getTypedefNameForAnonDecl()) && 4587 (TargetEnum->getDecl()->getIdentifier() || 4588 TargetEnum->getDecl()->getTypedefNameForAnonDecl()) && 4589 SourceEnum != TargetEnum) { 4590 if (S.SourceMgr.isInSystemMacro(CC)) 4591 return; 4592 4593 return DiagnoseImpCast(S, E, SourceType, T, CC, 4594 diag::warn_impcast_different_enum_types); 4595 } 4596 4597 return; 4598} 4599 4600void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 4601 SourceLocation CC, QualType T); 4602 4603void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 4604 SourceLocation CC, bool &ICContext) { 4605 E = E->IgnoreParenImpCasts(); 4606 4607 if (isa<ConditionalOperator>(E)) 4608 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), CC, T); 4609 4610 AnalyzeImplicitConversions(S, E, CC); 4611 if (E->getType() != T) 4612 return CheckImplicitConversion(S, E, T, CC, &ICContext); 4613 return; 4614} 4615 4616void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 4617 SourceLocation CC, QualType T) { 4618 AnalyzeImplicitConversions(S, E->getCond(), CC); 4619 4620 bool Suspicious = false; 4621 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 4622 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 4623 4624 // If -Wconversion would have warned about either of the candidates 4625 // for a signedness conversion to the context type... 4626 if (!Suspicious) return; 4627 4628 // ...but it's currently ignored... 4629 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 4630 CC)) 4631 return; 4632 4633 // ...then check whether it would have warned about either of the 4634 // candidates for a signedness conversion to the condition type. 4635 if (E->getType() == T) return; 4636 4637 Suspicious = false; 4638 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 4639 E->getType(), CC, &Suspicious); 4640 if (!Suspicious) 4641 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 4642 E->getType(), CC, &Suspicious); 4643} 4644 4645/// AnalyzeImplicitConversions - Find and report any interesting 4646/// implicit conversions in the given expression. There are a couple 4647/// of competing diagnostics here, -Wconversion and -Wsign-compare. 4648void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 4649 QualType T = OrigE->getType(); 4650 Expr *E = OrigE->IgnoreParenImpCasts(); 4651 4652 if (E->isTypeDependent() || E->isValueDependent()) 4653 return; 4654 4655 // For conditional operators, we analyze the arguments as if they 4656 // were being fed directly into the output. 4657 if (isa<ConditionalOperator>(E)) { 4658 ConditionalOperator *CO = cast<ConditionalOperator>(E); 4659 CheckConditionalOperator(S, CO, CC, T); 4660 return; 4661 } 4662 4663 // Go ahead and check any implicit conversions we might have skipped. 4664 // The non-canonical typecheck is just an optimization; 4665 // CheckImplicitConversion will filter out dead implicit conversions. 4666 if (E->getType() != T) 4667 CheckImplicitConversion(S, E, T, CC); 4668 4669 // Now continue drilling into this expression. 4670 4671 // Skip past explicit casts. 4672 if (isa<ExplicitCastExpr>(E)) { 4673 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 4674 return AnalyzeImplicitConversions(S, E, CC); 4675 } 4676 4677 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 4678 // Do a somewhat different check with comparison operators. 4679 if (BO->isComparisonOp()) 4680 return AnalyzeComparison(S, BO); 4681 4682 // And with simple assignments. 4683 if (BO->getOpcode() == BO_Assign) 4684 return AnalyzeAssignment(S, BO); 4685 } 4686 4687 // These break the otherwise-useful invariant below. Fortunately, 4688 // we don't really need to recurse into them, because any internal 4689 // expressions should have been analyzed already when they were 4690 // built into statements. 4691 if (isa<StmtExpr>(E)) return; 4692 4693 // Don't descend into unevaluated contexts. 4694 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 4695 4696 // Now just recurse over the expression's children. 4697 CC = E->getExprLoc(); 4698 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 4699 bool IsLogicalOperator = BO && BO->isLogicalOp(); 4700 for (Stmt::child_range I = E->children(); I; ++I) { 4701 Expr *ChildExpr = dyn_cast_or_null<Expr>(*I); 4702 if (!ChildExpr) 4703 continue; 4704 4705 if (IsLogicalOperator && 4706 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 4707 // Ignore checking string literals that are in logical operators. 4708 continue; 4709 AnalyzeImplicitConversions(S, ChildExpr, CC); 4710 } 4711} 4712 4713} // end anonymous namespace 4714 4715/// Diagnoses "dangerous" implicit conversions within the given 4716/// expression (which is a full expression). Implements -Wconversion 4717/// and -Wsign-compare. 4718/// 4719/// \param CC the "context" location of the implicit conversion, i.e. 4720/// the most location of the syntactic entity requiring the implicit 4721/// conversion 4722void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 4723 // Don't diagnose in unevaluated contexts. 4724 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 4725 return; 4726 4727 // Don't diagnose for value- or type-dependent expressions. 4728 if (E->isTypeDependent() || E->isValueDependent()) 4729 return; 4730 4731 // Check for array bounds violations in cases where the check isn't triggered 4732 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 4733 // ArraySubscriptExpr is on the RHS of a variable initialization. 4734 CheckArrayAccess(E); 4735 4736 // This is not the right CC for (e.g.) a variable initialization. 4737 AnalyzeImplicitConversions(*this, E, CC); 4738} 4739 4740void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 4741 FieldDecl *BitField, 4742 Expr *Init) { 4743 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 4744} 4745 4746/// CheckParmsForFunctionDef - Check that the parameters of the given 4747/// function are appropriate for the definition of a function. This 4748/// takes care of any checks that cannot be performed on the 4749/// declaration itself, e.g., that the types of each of the function 4750/// parameters are complete. 4751bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 4752 bool CheckParameterNames) { 4753 bool HasInvalidParm = false; 4754 for (; P != PEnd; ++P) { 4755 ParmVarDecl *Param = *P; 4756 4757 // C99 6.7.5.3p4: the parameters in a parameter type list in a 4758 // function declarator that is part of a function definition of 4759 // that function shall not have incomplete type. 4760 // 4761 // This is also C++ [dcl.fct]p6. 4762 if (!Param->isInvalidDecl() && 4763 RequireCompleteType(Param->getLocation(), Param->getType(), 4764 diag::err_typecheck_decl_incomplete_type)) { 4765 Param->setInvalidDecl(); 4766 HasInvalidParm = true; 4767 } 4768 4769 // C99 6.9.1p5: If the declarator includes a parameter type list, the 4770 // declaration of each parameter shall include an identifier. 4771 if (CheckParameterNames && 4772 Param->getIdentifier() == 0 && 4773 !Param->isImplicit() && 4774 !getLangOpts().CPlusPlus) 4775 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 4776 4777 // C99 6.7.5.3p12: 4778 // If the function declarator is not part of a definition of that 4779 // function, parameters may have incomplete type and may use the [*] 4780 // notation in their sequences of declarator specifiers to specify 4781 // variable length array types. 4782 QualType PType = Param->getOriginalType(); 4783 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 4784 if (AT->getSizeModifier() == ArrayType::Star) { 4785 // FIXME: This diagnosic should point the '[*]' if source-location 4786 // information is added for it. 4787 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 4788 } 4789 } 4790 } 4791 4792 return HasInvalidParm; 4793} 4794 4795/// CheckCastAlign - Implements -Wcast-align, which warns when a 4796/// pointer cast increases the alignment requirements. 4797void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 4798 // This is actually a lot of work to potentially be doing on every 4799 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 4800 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 4801 TRange.getBegin()) 4802 == DiagnosticsEngine::Ignored) 4803 return; 4804 4805 // Ignore dependent types. 4806 if (T->isDependentType() || Op->getType()->isDependentType()) 4807 return; 4808 4809 // Require that the destination be a pointer type. 4810 const PointerType *DestPtr = T->getAs<PointerType>(); 4811 if (!DestPtr) return; 4812 4813 // If the destination has alignment 1, we're done. 4814 QualType DestPointee = DestPtr->getPointeeType(); 4815 if (DestPointee->isIncompleteType()) return; 4816 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 4817 if (DestAlign.isOne()) return; 4818 4819 // Require that the source be a pointer type. 4820 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 4821 if (!SrcPtr) return; 4822 QualType SrcPointee = SrcPtr->getPointeeType(); 4823 4824 // Whitelist casts from cv void*. We already implicitly 4825 // whitelisted casts to cv void*, since they have alignment 1. 4826 // Also whitelist casts involving incomplete types, which implicitly 4827 // includes 'void'. 4828 if (SrcPointee->isIncompleteType()) return; 4829 4830 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 4831 if (SrcAlign >= DestAlign) return; 4832 4833 Diag(TRange.getBegin(), diag::warn_cast_align) 4834 << Op->getType() << T 4835 << static_cast<unsigned>(SrcAlign.getQuantity()) 4836 << static_cast<unsigned>(DestAlign.getQuantity()) 4837 << TRange << Op->getSourceRange(); 4838} 4839 4840static const Type* getElementType(const Expr *BaseExpr) { 4841 const Type* EltType = BaseExpr->getType().getTypePtr(); 4842 if (EltType->isAnyPointerType()) 4843 return EltType->getPointeeType().getTypePtr(); 4844 else if (EltType->isArrayType()) 4845 return EltType->getBaseElementTypeUnsafe(); 4846 return EltType; 4847} 4848 4849/// \brief Check whether this array fits the idiom of a size-one tail padded 4850/// array member of a struct. 4851/// 4852/// We avoid emitting out-of-bounds access warnings for such arrays as they are 4853/// commonly used to emulate flexible arrays in C89 code. 4854static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 4855 const NamedDecl *ND) { 4856 if (Size != 1 || !ND) return false; 4857 4858 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 4859 if (!FD) return false; 4860 4861 // Don't consider sizes resulting from macro expansions or template argument 4862 // substitution to form C89 tail-padded arrays. 4863 4864 TypeSourceInfo *TInfo = FD->getTypeSourceInfo(); 4865 while (TInfo) { 4866 TypeLoc TL = TInfo->getTypeLoc(); 4867 // Look through typedefs. 4868 const TypedefTypeLoc *TTL = dyn_cast<TypedefTypeLoc>(&TL); 4869 if (TTL) { 4870 const TypedefNameDecl *TDL = TTL->getTypedefNameDecl(); 4871 TInfo = TDL->getTypeSourceInfo(); 4872 continue; 4873 } 4874 ConstantArrayTypeLoc CTL = cast<ConstantArrayTypeLoc>(TL); 4875 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(CTL.getSizeExpr()); 4876 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 4877 return false; 4878 break; 4879 } 4880 4881 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 4882 if (!RD) return false; 4883 if (RD->isUnion()) return false; 4884 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 4885 if (!CRD->isStandardLayout()) return false; 4886 } 4887 4888 // See if this is the last field decl in the record. 4889 const Decl *D = FD; 4890 while ((D = D->getNextDeclInContext())) 4891 if (isa<FieldDecl>(D)) 4892 return false; 4893 return true; 4894} 4895 4896void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 4897 const ArraySubscriptExpr *ASE, 4898 bool AllowOnePastEnd, bool IndexNegated) { 4899 IndexExpr = IndexExpr->IgnoreParenImpCasts(); 4900 if (IndexExpr->isValueDependent()) 4901 return; 4902 4903 const Type *EffectiveType = getElementType(BaseExpr); 4904 BaseExpr = BaseExpr->IgnoreParenCasts(); 4905 const ConstantArrayType *ArrayTy = 4906 Context.getAsConstantArrayType(BaseExpr->getType()); 4907 if (!ArrayTy) 4908 return; 4909 4910 llvm::APSInt index; 4911 if (!IndexExpr->EvaluateAsInt(index, Context)) 4912 return; 4913 if (IndexNegated) 4914 index = -index; 4915 4916 const NamedDecl *ND = NULL; 4917 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4918 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 4919 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 4920 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 4921 4922 if (index.isUnsigned() || !index.isNegative()) { 4923 llvm::APInt size = ArrayTy->getSize(); 4924 if (!size.isStrictlyPositive()) 4925 return; 4926 4927 const Type* BaseType = getElementType(BaseExpr); 4928 if (BaseType != EffectiveType) { 4929 // Make sure we're comparing apples to apples when comparing index to size 4930 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 4931 uint64_t array_typesize = Context.getTypeSize(BaseType); 4932 // Handle ptrarith_typesize being zero, such as when casting to void* 4933 if (!ptrarith_typesize) ptrarith_typesize = 1; 4934 if (ptrarith_typesize != array_typesize) { 4935 // There's a cast to a different size type involved 4936 uint64_t ratio = array_typesize / ptrarith_typesize; 4937 // TODO: Be smarter about handling cases where array_typesize is not a 4938 // multiple of ptrarith_typesize 4939 if (ptrarith_typesize * ratio == array_typesize) 4940 size *= llvm::APInt(size.getBitWidth(), ratio); 4941 } 4942 } 4943 4944 if (size.getBitWidth() > index.getBitWidth()) 4945 index = index.zext(size.getBitWidth()); 4946 else if (size.getBitWidth() < index.getBitWidth()) 4947 size = size.zext(index.getBitWidth()); 4948 4949 // For array subscripting the index must be less than size, but for pointer 4950 // arithmetic also allow the index (offset) to be equal to size since 4951 // computing the next address after the end of the array is legal and 4952 // commonly done e.g. in C++ iterators and range-based for loops. 4953 if (AllowOnePastEnd ? index.ule(size) : index.ult(size)) 4954 return; 4955 4956 // Also don't warn for arrays of size 1 which are members of some 4957 // structure. These are often used to approximate flexible arrays in C89 4958 // code. 4959 if (IsTailPaddedMemberArray(*this, size, ND)) 4960 return; 4961 4962 // Suppress the warning if the subscript expression (as identified by the 4963 // ']' location) and the index expression are both from macro expansions 4964 // within a system header. 4965 if (ASE) { 4966 SourceLocation RBracketLoc = SourceMgr.getSpellingLoc( 4967 ASE->getRBracketLoc()); 4968 if (SourceMgr.isInSystemHeader(RBracketLoc)) { 4969 SourceLocation IndexLoc = SourceMgr.getSpellingLoc( 4970 IndexExpr->getLocStart()); 4971 if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc)) 4972 return; 4973 } 4974 } 4975 4976 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 4977 if (ASE) 4978 DiagID = diag::warn_array_index_exceeds_bounds; 4979 4980 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4981 PDiag(DiagID) << index.toString(10, true) 4982 << size.toString(10, true) 4983 << (unsigned)size.getLimitedValue(~0U) 4984 << IndexExpr->getSourceRange()); 4985 } else { 4986 unsigned DiagID = diag::warn_array_index_precedes_bounds; 4987 if (!ASE) { 4988 DiagID = diag::warn_ptr_arith_precedes_bounds; 4989 if (index.isNegative()) index = -index; 4990 } 4991 4992 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4993 PDiag(DiagID) << index.toString(10, true) 4994 << IndexExpr->getSourceRange()); 4995 } 4996 4997 if (!ND) { 4998 // Try harder to find a NamedDecl to point at in the note. 4999 while (const ArraySubscriptExpr *ASE = 5000 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 5001 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 5002 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 5003 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 5004 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 5005 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 5006 } 5007 5008 if (ND) 5009 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 5010 PDiag(diag::note_array_index_out_of_bounds) 5011 << ND->getDeclName()); 5012} 5013 5014void Sema::CheckArrayAccess(const Expr *expr) { 5015 int AllowOnePastEnd = 0; 5016 while (expr) { 5017 expr = expr->IgnoreParenImpCasts(); 5018 switch (expr->getStmtClass()) { 5019 case Stmt::ArraySubscriptExprClass: { 5020 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 5021 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE, 5022 AllowOnePastEnd > 0); 5023 return; 5024 } 5025 case Stmt::UnaryOperatorClass: { 5026 // Only unwrap the * and & unary operators 5027 const UnaryOperator *UO = cast<UnaryOperator>(expr); 5028 expr = UO->getSubExpr(); 5029 switch (UO->getOpcode()) { 5030 case UO_AddrOf: 5031 AllowOnePastEnd++; 5032 break; 5033 case UO_Deref: 5034 AllowOnePastEnd--; 5035 break; 5036 default: 5037 return; 5038 } 5039 break; 5040 } 5041 case Stmt::ConditionalOperatorClass: { 5042 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 5043 if (const Expr *lhs = cond->getLHS()) 5044 CheckArrayAccess(lhs); 5045 if (const Expr *rhs = cond->getRHS()) 5046 CheckArrayAccess(rhs); 5047 return; 5048 } 5049 default: 5050 return; 5051 } 5052 } 5053} 5054 5055//===--- CHECK: Objective-C retain cycles ----------------------------------// 5056 5057namespace { 5058 struct RetainCycleOwner { 5059 RetainCycleOwner() : Variable(0), Indirect(false) {} 5060 VarDecl *Variable; 5061 SourceRange Range; 5062 SourceLocation Loc; 5063 bool Indirect; 5064 5065 void setLocsFrom(Expr *e) { 5066 Loc = e->getExprLoc(); 5067 Range = e->getSourceRange(); 5068 } 5069 }; 5070} 5071 5072/// Consider whether capturing the given variable can possibly lead to 5073/// a retain cycle. 5074static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 5075 // In ARC, it's captured strongly iff the variable has __strong 5076 // lifetime. In MRR, it's captured strongly if the variable is 5077 // __block and has an appropriate type. 5078 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 5079 return false; 5080 5081 owner.Variable = var; 5082 owner.setLocsFrom(ref); 5083 return true; 5084} 5085 5086static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) { 5087 while (true) { 5088 e = e->IgnoreParens(); 5089 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 5090 switch (cast->getCastKind()) { 5091 case CK_BitCast: 5092 case CK_LValueBitCast: 5093 case CK_LValueToRValue: 5094 case CK_ARCReclaimReturnedObject: 5095 e = cast->getSubExpr(); 5096 continue; 5097 5098 default: 5099 return false; 5100 } 5101 } 5102 5103 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 5104 ObjCIvarDecl *ivar = ref->getDecl(); 5105 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 5106 return false; 5107 5108 // Try to find a retain cycle in the base. 5109 if (!findRetainCycleOwner(S, ref->getBase(), owner)) 5110 return false; 5111 5112 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 5113 owner.Indirect = true; 5114 return true; 5115 } 5116 5117 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 5118 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 5119 if (!var) return false; 5120 return considerVariable(var, ref, owner); 5121 } 5122 5123 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 5124 if (member->isArrow()) return false; 5125 5126 // Don't count this as an indirect ownership. 5127 e = member->getBase(); 5128 continue; 5129 } 5130 5131 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 5132 // Only pay attention to pseudo-objects on property references. 5133 ObjCPropertyRefExpr *pre 5134 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 5135 ->IgnoreParens()); 5136 if (!pre) return false; 5137 if (pre->isImplicitProperty()) return false; 5138 ObjCPropertyDecl *property = pre->getExplicitProperty(); 5139 if (!property->isRetaining() && 5140 !(property->getPropertyIvarDecl() && 5141 property->getPropertyIvarDecl()->getType() 5142 .getObjCLifetime() == Qualifiers::OCL_Strong)) 5143 return false; 5144 5145 owner.Indirect = true; 5146 if (pre->isSuperReceiver()) { 5147 owner.Variable = S.getCurMethodDecl()->getSelfDecl(); 5148 if (!owner.Variable) 5149 return false; 5150 owner.Loc = pre->getLocation(); 5151 owner.Range = pre->getSourceRange(); 5152 return true; 5153 } 5154 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 5155 ->getSourceExpr()); 5156 continue; 5157 } 5158 5159 // Array ivars? 5160 5161 return false; 5162 } 5163} 5164 5165namespace { 5166 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 5167 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 5168 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 5169 Variable(variable), Capturer(0) {} 5170 5171 VarDecl *Variable; 5172 Expr *Capturer; 5173 5174 void VisitDeclRefExpr(DeclRefExpr *ref) { 5175 if (ref->getDecl() == Variable && !Capturer) 5176 Capturer = ref; 5177 } 5178 5179 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 5180 if (Capturer) return; 5181 Visit(ref->getBase()); 5182 if (Capturer && ref->isFreeIvar()) 5183 Capturer = ref; 5184 } 5185 5186 void VisitBlockExpr(BlockExpr *block) { 5187 // Look inside nested blocks 5188 if (block->getBlockDecl()->capturesVariable(Variable)) 5189 Visit(block->getBlockDecl()->getBody()); 5190 } 5191 }; 5192} 5193 5194/// Check whether the given argument is a block which captures a 5195/// variable. 5196static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 5197 assert(owner.Variable && owner.Loc.isValid()); 5198 5199 e = e->IgnoreParenCasts(); 5200 BlockExpr *block = dyn_cast<BlockExpr>(e); 5201 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 5202 return 0; 5203 5204 FindCaptureVisitor visitor(S.Context, owner.Variable); 5205 visitor.Visit(block->getBlockDecl()->getBody()); 5206 return visitor.Capturer; 5207} 5208 5209static void diagnoseRetainCycle(Sema &S, Expr *capturer, 5210 RetainCycleOwner &owner) { 5211 assert(capturer); 5212 assert(owner.Variable && owner.Loc.isValid()); 5213 5214 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 5215 << owner.Variable << capturer->getSourceRange(); 5216 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 5217 << owner.Indirect << owner.Range; 5218} 5219 5220/// Check for a keyword selector that starts with the word 'add' or 5221/// 'set'. 5222static bool isSetterLikeSelector(Selector sel) { 5223 if (sel.isUnarySelector()) return false; 5224 5225 StringRef str = sel.getNameForSlot(0); 5226 while (!str.empty() && str.front() == '_') str = str.substr(1); 5227 if (str.startswith("set")) 5228 str = str.substr(3); 5229 else if (str.startswith("add")) { 5230 // Specially whitelist 'addOperationWithBlock:'. 5231 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 5232 return false; 5233 str = str.substr(3); 5234 } 5235 else 5236 return false; 5237 5238 if (str.empty()) return true; 5239 return !islower(str.front()); 5240} 5241 5242/// Check a message send to see if it's likely to cause a retain cycle. 5243void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 5244 // Only check instance methods whose selector looks like a setter. 5245 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 5246 return; 5247 5248 // Try to find a variable that the receiver is strongly owned by. 5249 RetainCycleOwner owner; 5250 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 5251 if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner)) 5252 return; 5253 } else { 5254 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 5255 owner.Variable = getCurMethodDecl()->getSelfDecl(); 5256 owner.Loc = msg->getSuperLoc(); 5257 owner.Range = msg->getSuperLoc(); 5258 } 5259 5260 // Check whether the receiver is captured by any of the arguments. 5261 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 5262 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 5263 return diagnoseRetainCycle(*this, capturer, owner); 5264} 5265 5266/// Check a property assign to see if it's likely to cause a retain cycle. 5267void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 5268 RetainCycleOwner owner; 5269 if (!findRetainCycleOwner(*this, receiver, owner)) 5270 return; 5271 5272 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 5273 diagnoseRetainCycle(*this, capturer, owner); 5274} 5275 5276bool Sema::checkUnsafeAssigns(SourceLocation Loc, 5277 QualType LHS, Expr *RHS) { 5278 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 5279 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 5280 return false; 5281 // strip off any implicit cast added to get to the one arc-specific 5282 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 5283 if (cast->getCastKind() == CK_ARCConsumeObject) { 5284 Diag(Loc, diag::warn_arc_retained_assign) 5285 << (LT == Qualifiers::OCL_ExplicitNone) << 1 5286 << RHS->getSourceRange(); 5287 return true; 5288 } 5289 RHS = cast->getSubExpr(); 5290 } 5291 return false; 5292} 5293 5294void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 5295 Expr *LHS, Expr *RHS) { 5296 QualType LHSType; 5297 // PropertyRef on LHS type need be directly obtained from 5298 // its declaration as it has a PsuedoType. 5299 ObjCPropertyRefExpr *PRE 5300 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens()); 5301 if (PRE && !PRE->isImplicitProperty()) { 5302 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 5303 if (PD) 5304 LHSType = PD->getType(); 5305 } 5306 5307 if (LHSType.isNull()) 5308 LHSType = LHS->getType(); 5309 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 5310 return; 5311 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 5312 // FIXME. Check for other life times. 5313 if (LT != Qualifiers::OCL_None) 5314 return; 5315 5316 if (PRE) { 5317 if (PRE->isImplicitProperty()) 5318 return; 5319 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 5320 if (!PD) 5321 return; 5322 5323 unsigned Attributes = PD->getPropertyAttributes(); 5324 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) { 5325 // when 'assign' attribute was not explicitly specified 5326 // by user, ignore it and rely on property type itself 5327 // for lifetime info. 5328 unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten(); 5329 if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) && 5330 LHSType->isObjCRetainableType()) 5331 return; 5332 5333 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 5334 if (cast->getCastKind() == CK_ARCConsumeObject) { 5335 Diag(Loc, diag::warn_arc_retained_property_assign) 5336 << RHS->getSourceRange(); 5337 return; 5338 } 5339 RHS = cast->getSubExpr(); 5340 } 5341 } 5342 else if (Attributes & ObjCPropertyDecl::OBJC_PR_weak) { 5343 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 5344 if (cast->getCastKind() == CK_ARCConsumeObject) { 5345 Diag(Loc, diag::warn_arc_retained_assign) 5346 << 0 << 0<< RHS->getSourceRange(); 5347 return; 5348 } 5349 RHS = cast->getSubExpr(); 5350 } 5351 } 5352 } 5353} 5354 5355//===--- CHECK: Empty statement body (-Wempty-body) ---------------------===// 5356 5357namespace { 5358bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr, 5359 SourceLocation StmtLoc, 5360 const NullStmt *Body) { 5361 // Do not warn if the body is a macro that expands to nothing, e.g: 5362 // 5363 // #define CALL(x) 5364 // if (condition) 5365 // CALL(0); 5366 // 5367 if (Body->hasLeadingEmptyMacro()) 5368 return false; 5369 5370 // Get line numbers of statement and body. 5371 bool StmtLineInvalid; 5372 unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc, 5373 &StmtLineInvalid); 5374 if (StmtLineInvalid) 5375 return false; 5376 5377 bool BodyLineInvalid; 5378 unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(), 5379 &BodyLineInvalid); 5380 if (BodyLineInvalid) 5381 return false; 5382 5383 // Warn if null statement and body are on the same line. 5384 if (StmtLine != BodyLine) 5385 return false; 5386 5387 return true; 5388} 5389} // Unnamed namespace 5390 5391void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc, 5392 const Stmt *Body, 5393 unsigned DiagID) { 5394 // Since this is a syntactic check, don't emit diagnostic for template 5395 // instantiations, this just adds noise. 5396 if (CurrentInstantiationScope) 5397 return; 5398 5399 // The body should be a null statement. 5400 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 5401 if (!NBody) 5402 return; 5403 5404 // Do the usual checks. 5405 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 5406 return; 5407 5408 Diag(NBody->getSemiLoc(), DiagID); 5409 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 5410} 5411 5412void Sema::DiagnoseEmptyLoopBody(const Stmt *S, 5413 const Stmt *PossibleBody) { 5414 assert(!CurrentInstantiationScope); // Ensured by caller 5415 5416 SourceLocation StmtLoc; 5417 const Stmt *Body; 5418 unsigned DiagID; 5419 if (const ForStmt *FS = dyn_cast<ForStmt>(S)) { 5420 StmtLoc = FS->getRParenLoc(); 5421 Body = FS->getBody(); 5422 DiagID = diag::warn_empty_for_body; 5423 } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) { 5424 StmtLoc = WS->getCond()->getSourceRange().getEnd(); 5425 Body = WS->getBody(); 5426 DiagID = diag::warn_empty_while_body; 5427 } else 5428 return; // Neither `for' nor `while'. 5429 5430 // The body should be a null statement. 5431 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 5432 if (!NBody) 5433 return; 5434 5435 // Skip expensive checks if diagnostic is disabled. 5436 if (Diags.getDiagnosticLevel(DiagID, NBody->getSemiLoc()) == 5437 DiagnosticsEngine::Ignored) 5438 return; 5439 5440 // Do the usual checks. 5441 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 5442 return; 5443 5444 // `for(...);' and `while(...);' are popular idioms, so in order to keep 5445 // noise level low, emit diagnostics only if for/while is followed by a 5446 // CompoundStmt, e.g.: 5447 // for (int i = 0; i < n; i++); 5448 // { 5449 // a(i); 5450 // } 5451 // or if for/while is followed by a statement with more indentation 5452 // than for/while itself: 5453 // for (int i = 0; i < n; i++); 5454 // a(i); 5455 bool ProbableTypo = isa<CompoundStmt>(PossibleBody); 5456 if (!ProbableTypo) { 5457 bool BodyColInvalid; 5458 unsigned BodyCol = SourceMgr.getPresumedColumnNumber( 5459 PossibleBody->getLocStart(), 5460 &BodyColInvalid); 5461 if (BodyColInvalid) 5462 return; 5463 5464 bool StmtColInvalid; 5465 unsigned StmtCol = SourceMgr.getPresumedColumnNumber( 5466 S->getLocStart(), 5467 &StmtColInvalid); 5468 if (StmtColInvalid) 5469 return; 5470 5471 if (BodyCol > StmtCol) 5472 ProbableTypo = true; 5473 } 5474 5475 if (ProbableTypo) { 5476 Diag(NBody->getSemiLoc(), DiagID); 5477 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 5478 } 5479} 5480