SemaChecking.cpp revision cec9ce49dcf4b4b768043f96c8ef8c1d4cdbb6b9
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 } 2572 2573 // The remaining checks depend on the data arguments. 2574 if (HasVAListArg) 2575 return true; 2576 2577 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2578 return false; 2579 2580 const Expr *Arg = getDataArg(argIndex); 2581 if (!Arg) 2582 return true; 2583 2584 return checkFormatExpr(FS, startSpecifier, specifierLen, Arg); 2585} 2586 2587bool 2588CheckPrintfHandler::checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 2589 const char *StartSpecifier, 2590 unsigned SpecifierLen, 2591 const Expr *E) { 2592 using namespace analyze_format_string; 2593 using namespace analyze_printf; 2594 // Now type check the data expression that matches the 2595 // format specifier. 2596 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context, 2597 ObjCContext); 2598 if (ATR.isValid() && !ATR.matchesType(S.Context, E->getType())) { 2599 // Look through argument promotions for our error message's reported type. 2600 // This includes the integral and floating promotions, but excludes array 2601 // and function pointer decay; seeing that an argument intended to be a 2602 // string has type 'char [6]' is probably more confusing than 'char *'. 2603 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 2604 if (ICE->getCastKind() == CK_IntegralCast || 2605 ICE->getCastKind() == CK_FloatingCast) { 2606 E = ICE->getSubExpr(); 2607 2608 // Check if we didn't match because of an implicit cast from a 'char' 2609 // or 'short' to an 'int'. This is done because printf is a varargs 2610 // function. 2611 if (ICE->getType() == S.Context.IntTy || 2612 ICE->getType() == S.Context.UnsignedIntTy) { 2613 // All further checking is done on the subexpression. 2614 if (ATR.matchesType(S.Context, E->getType())) 2615 return true; 2616 } 2617 } 2618 } 2619 2620 // We may be able to offer a FixItHint if it is a supported type. 2621 PrintfSpecifier fixedFS = FS; 2622 bool success = fixedFS.fixType(E->getType(), S.getLangOpts(), 2623 S.Context, ObjCContext); 2624 2625 if (success) { 2626 // Get the fix string from the fixed format specifier 2627 SmallString<16> buf; 2628 llvm::raw_svector_ostream os(buf); 2629 fixedFS.toString(os); 2630 2631 EmitFormatDiagnostic( 2632 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2633 << ATR.getRepresentativeTypeName(S.Context) << E->getType() 2634 << E->getSourceRange(), 2635 E->getLocStart(), 2636 /*IsStringLocation*/false, 2637 getSpecifierRange(StartSpecifier, SpecifierLen), 2638 FixItHint::CreateReplacement( 2639 getSpecifierRange(StartSpecifier, SpecifierLen), 2640 os.str())); 2641 } else { 2642 const CharSourceRange &CSR = getSpecifierRange(StartSpecifier, 2643 SpecifierLen); 2644 // Since the warning for passing non-POD types to variadic functions 2645 // was deferred until now, we emit a warning for non-POD 2646 // arguments here. 2647 if (S.isValidVarArgType(E->getType()) == Sema::VAK_Invalid) { 2648 unsigned DiagKind; 2649 if (E->getType()->isObjCObjectType()) 2650 DiagKind = diag::err_cannot_pass_objc_interface_to_vararg_format; 2651 else 2652 DiagKind = diag::warn_non_pod_vararg_with_format_string; 2653 2654 EmitFormatDiagnostic( 2655 S.PDiag(DiagKind) 2656 << S.getLangOpts().CPlusPlus0x 2657 << E->getType() 2658 << CallType 2659 << ATR.getRepresentativeTypeName(S.Context) 2660 << CSR 2661 << E->getSourceRange(), 2662 E->getLocStart(), /*IsStringLocation*/false, CSR); 2663 2664 checkForCStrMembers(ATR, E, CSR); 2665 } else 2666 EmitFormatDiagnostic( 2667 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2668 << ATR.getRepresentativeTypeName(S.Context) << E->getType() 2669 << CSR 2670 << E->getSourceRange(), 2671 E->getLocStart(), /*IsStringLocation*/false, CSR); 2672 } 2673 } 2674 2675 return true; 2676} 2677 2678//===--- CHECK: Scanf format string checking ------------------------------===// 2679 2680namespace { 2681class CheckScanfHandler : public CheckFormatHandler { 2682public: 2683 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 2684 const Expr *origFormatExpr, unsigned firstDataArg, 2685 unsigned numDataArgs, const char *beg, bool hasVAListArg, 2686 Expr **Args, unsigned NumArgs, 2687 unsigned formatIdx, bool inFunctionCall, 2688 Sema::VariadicCallType CallType) 2689 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2690 numDataArgs, beg, hasVAListArg, 2691 Args, NumArgs, formatIdx, inFunctionCall, CallType) 2692 {} 2693 2694 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 2695 const char *startSpecifier, 2696 unsigned specifierLen); 2697 2698 bool HandleInvalidScanfConversionSpecifier( 2699 const analyze_scanf::ScanfSpecifier &FS, 2700 const char *startSpecifier, 2701 unsigned specifierLen); 2702 2703 void HandleIncompleteScanList(const char *start, const char *end); 2704}; 2705} 2706 2707void CheckScanfHandler::HandleIncompleteScanList(const char *start, 2708 const char *end) { 2709 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 2710 getLocationOfByte(end), /*IsStringLocation*/true, 2711 getSpecifierRange(start, end - start)); 2712} 2713 2714bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 2715 const analyze_scanf::ScanfSpecifier &FS, 2716 const char *startSpecifier, 2717 unsigned specifierLen) { 2718 2719 const analyze_scanf::ScanfConversionSpecifier &CS = 2720 FS.getConversionSpecifier(); 2721 2722 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2723 getLocationOfByte(CS.getStart()), 2724 startSpecifier, specifierLen, 2725 CS.getStart(), CS.getLength()); 2726} 2727 2728bool CheckScanfHandler::HandleScanfSpecifier( 2729 const analyze_scanf::ScanfSpecifier &FS, 2730 const char *startSpecifier, 2731 unsigned specifierLen) { 2732 2733 using namespace analyze_scanf; 2734 using namespace analyze_format_string; 2735 2736 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 2737 2738 // Handle case where '%' and '*' don't consume an argument. These shouldn't 2739 // be used to decide if we are using positional arguments consistently. 2740 if (FS.consumesDataArgument()) { 2741 if (atFirstArg) { 2742 atFirstArg = false; 2743 usesPositionalArgs = FS.usesPositionalArg(); 2744 } 2745 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2746 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2747 startSpecifier, specifierLen); 2748 return false; 2749 } 2750 } 2751 2752 // Check if the field with is non-zero. 2753 const OptionalAmount &Amt = FS.getFieldWidth(); 2754 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 2755 if (Amt.getConstantAmount() == 0) { 2756 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 2757 Amt.getConstantLength()); 2758 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 2759 getLocationOfByte(Amt.getStart()), 2760 /*IsStringLocation*/true, R, 2761 FixItHint::CreateRemoval(R)); 2762 } 2763 } 2764 2765 if (!FS.consumesDataArgument()) { 2766 // FIXME: Technically specifying a precision or field width here 2767 // makes no sense. Worth issuing a warning at some point. 2768 return true; 2769 } 2770 2771 // Consume the argument. 2772 unsigned argIndex = FS.getArgIndex(); 2773 if (argIndex < NumDataArgs) { 2774 // The check to see if the argIndex is valid will come later. 2775 // We set the bit here because we may exit early from this 2776 // function if we encounter some other error. 2777 CoveredArgs.set(argIndex); 2778 } 2779 2780 // Check the length modifier is valid with the given conversion specifier. 2781 const LengthModifier &LM = FS.getLengthModifier(); 2782 if (!FS.hasValidLengthModifier()) { 2783 const CharSourceRange &R = getSpecifierRange(LM.getStart(), LM.getLength()); 2784 EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length) 2785 << LM.toString() << CS.toString() 2786 << getSpecifierRange(startSpecifier, specifierLen), 2787 getLocationOfByte(LM.getStart()), 2788 /*IsStringLocation*/true, R, 2789 FixItHint::CreateRemoval(R)); 2790 } 2791 2792 if (!FS.hasStandardLengthModifier()) 2793 HandleNonStandardLengthModifier(LM, startSpecifier, specifierLen); 2794 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 2795 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 2796 if (!FS.hasStandardLengthConversionCombination()) 2797 HandleNonStandardConversionSpecification(LM, CS, startSpecifier, 2798 specifierLen); 2799 2800 // The remaining checks depend on the data arguments. 2801 if (HasVAListArg) 2802 return true; 2803 2804 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2805 return false; 2806 2807 // Check that the argument type matches the format specifier. 2808 const Expr *Ex = getDataArg(argIndex); 2809 if (!Ex) 2810 return true; 2811 2812 const analyze_scanf::ScanfArgTypeResult &ATR = FS.getArgType(S.Context); 2813 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 2814 ScanfSpecifier fixedFS = FS; 2815 bool success = fixedFS.fixType(Ex->getType(), S.getLangOpts(), 2816 S.Context); 2817 2818 if (success) { 2819 // Get the fix string from the fixed format specifier. 2820 SmallString<128> buf; 2821 llvm::raw_svector_ostream os(buf); 2822 fixedFS.toString(os); 2823 2824 EmitFormatDiagnostic( 2825 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2826 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2827 << Ex->getSourceRange(), 2828 Ex->getLocStart(), 2829 /*IsStringLocation*/false, 2830 getSpecifierRange(startSpecifier, specifierLen), 2831 FixItHint::CreateReplacement( 2832 getSpecifierRange(startSpecifier, specifierLen), 2833 os.str())); 2834 } else { 2835 EmitFormatDiagnostic( 2836 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2837 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2838 << Ex->getSourceRange(), 2839 Ex->getLocStart(), 2840 /*IsStringLocation*/false, 2841 getSpecifierRange(startSpecifier, specifierLen)); 2842 } 2843 } 2844 2845 return true; 2846} 2847 2848void Sema::CheckFormatString(const StringLiteral *FExpr, 2849 const Expr *OrigFormatExpr, 2850 Expr **Args, unsigned NumArgs, 2851 bool HasVAListArg, unsigned format_idx, 2852 unsigned firstDataArg, FormatStringType Type, 2853 bool inFunctionCall, VariadicCallType CallType) { 2854 2855 // CHECK: is the format string a wide literal? 2856 if (!FExpr->isAscii() && !FExpr->isUTF8()) { 2857 CheckFormatHandler::EmitFormatDiagnostic( 2858 *this, inFunctionCall, Args[format_idx], 2859 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 2860 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2861 return; 2862 } 2863 2864 // Str - The format string. NOTE: this is NOT null-terminated! 2865 StringRef StrRef = FExpr->getString(); 2866 const char *Str = StrRef.data(); 2867 unsigned StrLen = StrRef.size(); 2868 const unsigned numDataArgs = NumArgs - firstDataArg; 2869 2870 // CHECK: empty format string? 2871 if (StrLen == 0 && numDataArgs > 0) { 2872 CheckFormatHandler::EmitFormatDiagnostic( 2873 *this, inFunctionCall, Args[format_idx], 2874 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 2875 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2876 return; 2877 } 2878 2879 if (Type == FST_Printf || Type == FST_NSString) { 2880 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2881 numDataArgs, (Type == FST_NSString), 2882 Str, HasVAListArg, Args, NumArgs, format_idx, 2883 inFunctionCall, CallType); 2884 2885 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen, 2886 getLangOpts())) 2887 H.DoneProcessing(); 2888 } else if (Type == FST_Scanf) { 2889 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, numDataArgs, 2890 Str, HasVAListArg, Args, NumArgs, format_idx, 2891 inFunctionCall, CallType); 2892 2893 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen, 2894 getLangOpts())) 2895 H.DoneProcessing(); 2896 } // TODO: handle other formats 2897} 2898 2899//===--- CHECK: Standard memory functions ---------------------------------===// 2900 2901/// \brief Determine whether the given type is a dynamic class type (e.g., 2902/// whether it has a vtable). 2903static bool isDynamicClassType(QualType T) { 2904 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 2905 if (CXXRecordDecl *Definition = Record->getDefinition()) 2906 if (Definition->isDynamicClass()) 2907 return true; 2908 2909 return false; 2910} 2911 2912/// \brief If E is a sizeof expression, returns its argument expression, 2913/// otherwise returns NULL. 2914static const Expr *getSizeOfExprArg(const Expr* E) { 2915 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2916 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2917 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 2918 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 2919 2920 return 0; 2921} 2922 2923/// \brief If E is a sizeof expression, returns its argument type. 2924static QualType getSizeOfArgType(const Expr* E) { 2925 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2926 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2927 if (SizeOf->getKind() == clang::UETT_SizeOf) 2928 return SizeOf->getTypeOfArgument(); 2929 2930 return QualType(); 2931} 2932 2933/// \brief Check for dangerous or invalid arguments to memset(). 2934/// 2935/// This issues warnings on known problematic, dangerous or unspecified 2936/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 2937/// function calls. 2938/// 2939/// \param Call The call expression to diagnose. 2940void Sema::CheckMemaccessArguments(const CallExpr *Call, 2941 unsigned BId, 2942 IdentifierInfo *FnName) { 2943 assert(BId != 0); 2944 2945 // It is possible to have a non-standard definition of memset. Validate 2946 // we have enough arguments, and if not, abort further checking. 2947 unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3); 2948 if (Call->getNumArgs() < ExpectedNumArgs) 2949 return; 2950 2951 unsigned LastArg = (BId == Builtin::BImemset || 2952 BId == Builtin::BIstrndup ? 1 : 2); 2953 unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2); 2954 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 2955 2956 // We have special checking when the length is a sizeof expression. 2957 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 2958 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 2959 llvm::FoldingSetNodeID SizeOfArgID; 2960 2961 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 2962 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 2963 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 2964 2965 QualType DestTy = Dest->getType(); 2966 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 2967 QualType PointeeTy = DestPtrTy->getPointeeType(); 2968 2969 // Never warn about void type pointers. This can be used to suppress 2970 // false positives. 2971 if (PointeeTy->isVoidType()) 2972 continue; 2973 2974 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 2975 // actually comparing the expressions for equality. Because computing the 2976 // expression IDs can be expensive, we only do this if the diagnostic is 2977 // enabled. 2978 if (SizeOfArg && 2979 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 2980 SizeOfArg->getExprLoc())) { 2981 // We only compute IDs for expressions if the warning is enabled, and 2982 // cache the sizeof arg's ID. 2983 if (SizeOfArgID == llvm::FoldingSetNodeID()) 2984 SizeOfArg->Profile(SizeOfArgID, Context, true); 2985 llvm::FoldingSetNodeID DestID; 2986 Dest->Profile(DestID, Context, true); 2987 if (DestID == SizeOfArgID) { 2988 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 2989 // over sizeof(src) as well. 2990 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 2991 StringRef ReadableName = FnName->getName(); 2992 2993 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 2994 if (UnaryOp->getOpcode() == UO_AddrOf) 2995 ActionIdx = 1; // If its an address-of operator, just remove it. 2996 if (Context.getTypeSize(PointeeTy) == Context.getCharWidth()) 2997 ActionIdx = 2; // If the pointee's size is sizeof(char), 2998 // suggest an explicit length. 2999 3000 // If the function is defined as a builtin macro, do not show macro 3001 // expansion. 3002 SourceLocation SL = SizeOfArg->getExprLoc(); 3003 SourceRange DSR = Dest->getSourceRange(); 3004 SourceRange SSR = SizeOfArg->getSourceRange(); 3005 SourceManager &SM = PP.getSourceManager(); 3006 3007 if (SM.isMacroArgExpansion(SL)) { 3008 ReadableName = Lexer::getImmediateMacroName(SL, SM, LangOpts); 3009 SL = SM.getSpellingLoc(SL); 3010 DSR = SourceRange(SM.getSpellingLoc(DSR.getBegin()), 3011 SM.getSpellingLoc(DSR.getEnd())); 3012 SSR = SourceRange(SM.getSpellingLoc(SSR.getBegin()), 3013 SM.getSpellingLoc(SSR.getEnd())); 3014 } 3015 3016 DiagRuntimeBehavior(SL, SizeOfArg, 3017 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 3018 << ReadableName 3019 << PointeeTy 3020 << DestTy 3021 << DSR 3022 << SSR); 3023 DiagRuntimeBehavior(SL, SizeOfArg, 3024 PDiag(diag::warn_sizeof_pointer_expr_memaccess_note) 3025 << ActionIdx 3026 << SSR); 3027 3028 break; 3029 } 3030 } 3031 3032 // Also check for cases where the sizeof argument is the exact same 3033 // type as the memory argument, and where it points to a user-defined 3034 // record type. 3035 if (SizeOfArgTy != QualType()) { 3036 if (PointeeTy->isRecordType() && 3037 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 3038 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 3039 PDiag(diag::warn_sizeof_pointer_type_memaccess) 3040 << FnName << SizeOfArgTy << ArgIdx 3041 << PointeeTy << Dest->getSourceRange() 3042 << LenExpr->getSourceRange()); 3043 break; 3044 } 3045 } 3046 3047 // Always complain about dynamic classes. 3048 if (isDynamicClassType(PointeeTy)) { 3049 3050 unsigned OperationType = 0; 3051 // "overwritten" if we're warning about the destination for any call 3052 // but memcmp; otherwise a verb appropriate to the call. 3053 if (ArgIdx != 0 || BId == Builtin::BImemcmp) { 3054 if (BId == Builtin::BImemcpy) 3055 OperationType = 1; 3056 else if(BId == Builtin::BImemmove) 3057 OperationType = 2; 3058 else if (BId == Builtin::BImemcmp) 3059 OperationType = 3; 3060 } 3061 3062 DiagRuntimeBehavior( 3063 Dest->getExprLoc(), Dest, 3064 PDiag(diag::warn_dyn_class_memaccess) 3065 << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx) 3066 << FnName << PointeeTy 3067 << OperationType 3068 << Call->getCallee()->getSourceRange()); 3069 } else if (PointeeTy.hasNonTrivialObjCLifetime() && 3070 BId != Builtin::BImemset) 3071 DiagRuntimeBehavior( 3072 Dest->getExprLoc(), Dest, 3073 PDiag(diag::warn_arc_object_memaccess) 3074 << ArgIdx << FnName << PointeeTy 3075 << Call->getCallee()->getSourceRange()); 3076 else 3077 continue; 3078 3079 DiagRuntimeBehavior( 3080 Dest->getExprLoc(), Dest, 3081 PDiag(diag::note_bad_memaccess_silence) 3082 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 3083 break; 3084 } 3085 } 3086} 3087 3088// A little helper routine: ignore addition and subtraction of integer literals. 3089// This intentionally does not ignore all integer constant expressions because 3090// we don't want to remove sizeof(). 3091static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 3092 Ex = Ex->IgnoreParenCasts(); 3093 3094 for (;;) { 3095 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 3096 if (!BO || !BO->isAdditiveOp()) 3097 break; 3098 3099 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 3100 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 3101 3102 if (isa<IntegerLiteral>(RHS)) 3103 Ex = LHS; 3104 else if (isa<IntegerLiteral>(LHS)) 3105 Ex = RHS; 3106 else 3107 break; 3108 } 3109 3110 return Ex; 3111} 3112 3113// Warn if the user has made the 'size' argument to strlcpy or strlcat 3114// be the size of the source, instead of the destination. 3115void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 3116 IdentifierInfo *FnName) { 3117 3118 // Don't crash if the user has the wrong number of arguments 3119 if (Call->getNumArgs() != 3) 3120 return; 3121 3122 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 3123 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 3124 const Expr *CompareWithSrc = NULL; 3125 3126 // Look for 'strlcpy(dst, x, sizeof(x))' 3127 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 3128 CompareWithSrc = Ex; 3129 else { 3130 // Look for 'strlcpy(dst, x, strlen(x))' 3131 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 3132 if (SizeCall->isBuiltinCall() == Builtin::BIstrlen 3133 && SizeCall->getNumArgs() == 1) 3134 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 3135 } 3136 } 3137 3138 if (!CompareWithSrc) 3139 return; 3140 3141 // Determine if the argument to sizeof/strlen is equal to the source 3142 // argument. In principle there's all kinds of things you could do 3143 // here, for instance creating an == expression and evaluating it with 3144 // EvaluateAsBooleanCondition, but this uses a more direct technique: 3145 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 3146 if (!SrcArgDRE) 3147 return; 3148 3149 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 3150 if (!CompareWithSrcDRE || 3151 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 3152 return; 3153 3154 const Expr *OriginalSizeArg = Call->getArg(2); 3155 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 3156 << OriginalSizeArg->getSourceRange() << FnName; 3157 3158 // Output a FIXIT hint if the destination is an array (rather than a 3159 // pointer to an array). This could be enhanced to handle some 3160 // pointers if we know the actual size, like if DstArg is 'array+2' 3161 // we could say 'sizeof(array)-2'. 3162 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 3163 QualType DstArgTy = DstArg->getType(); 3164 3165 // Only handle constant-sized or VLAs, but not flexible members. 3166 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 3167 // Only issue the FIXIT for arrays of size > 1. 3168 if (CAT->getSize().getSExtValue() <= 1) 3169 return; 3170 } else if (!DstArgTy->isVariableArrayType()) { 3171 return; 3172 } 3173 3174 SmallString<128> sizeString; 3175 llvm::raw_svector_ostream OS(sizeString); 3176 OS << "sizeof("; 3177 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 3178 OS << ")"; 3179 3180 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 3181 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 3182 OS.str()); 3183} 3184 3185/// Check if two expressions refer to the same declaration. 3186static bool referToTheSameDecl(const Expr *E1, const Expr *E2) { 3187 if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1)) 3188 if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2)) 3189 return D1->getDecl() == D2->getDecl(); 3190 return false; 3191} 3192 3193static const Expr *getStrlenExprArg(const Expr *E) { 3194 if (const CallExpr *CE = dyn_cast<CallExpr>(E)) { 3195 const FunctionDecl *FD = CE->getDirectCallee(); 3196 if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen) 3197 return 0; 3198 return CE->getArg(0)->IgnoreParenCasts(); 3199 } 3200 return 0; 3201} 3202 3203// Warn on anti-patterns as the 'size' argument to strncat. 3204// The correct size argument should look like following: 3205// strncat(dst, src, sizeof(dst) - strlen(dest) - 1); 3206void Sema::CheckStrncatArguments(const CallExpr *CE, 3207 IdentifierInfo *FnName) { 3208 // Don't crash if the user has the wrong number of arguments. 3209 if (CE->getNumArgs() < 3) 3210 return; 3211 const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts(); 3212 const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts(); 3213 const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts(); 3214 3215 // Identify common expressions, which are wrongly used as the size argument 3216 // to strncat and may lead to buffer overflows. 3217 unsigned PatternType = 0; 3218 if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) { 3219 // - sizeof(dst) 3220 if (referToTheSameDecl(SizeOfArg, DstArg)) 3221 PatternType = 1; 3222 // - sizeof(src) 3223 else if (referToTheSameDecl(SizeOfArg, SrcArg)) 3224 PatternType = 2; 3225 } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) { 3226 if (BE->getOpcode() == BO_Sub) { 3227 const Expr *L = BE->getLHS()->IgnoreParenCasts(); 3228 const Expr *R = BE->getRHS()->IgnoreParenCasts(); 3229 // - sizeof(dst) - strlen(dst) 3230 if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) && 3231 referToTheSameDecl(DstArg, getStrlenExprArg(R))) 3232 PatternType = 1; 3233 // - sizeof(src) - (anything) 3234 else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L))) 3235 PatternType = 2; 3236 } 3237 } 3238 3239 if (PatternType == 0) 3240 return; 3241 3242 // Generate the diagnostic. 3243 SourceLocation SL = LenArg->getLocStart(); 3244 SourceRange SR = LenArg->getSourceRange(); 3245 SourceManager &SM = PP.getSourceManager(); 3246 3247 // If the function is defined as a builtin macro, do not show macro expansion. 3248 if (SM.isMacroArgExpansion(SL)) { 3249 SL = SM.getSpellingLoc(SL); 3250 SR = SourceRange(SM.getSpellingLoc(SR.getBegin()), 3251 SM.getSpellingLoc(SR.getEnd())); 3252 } 3253 3254 if (PatternType == 1) 3255 Diag(SL, diag::warn_strncat_large_size) << SR; 3256 else 3257 Diag(SL, diag::warn_strncat_src_size) << SR; 3258 3259 // Output a FIXIT hint if the destination is an array (rather than a 3260 // pointer to an array). This could be enhanced to handle some 3261 // pointers if we know the actual size, like if DstArg is 'array+2' 3262 // we could say 'sizeof(array)-2'. 3263 QualType DstArgTy = DstArg->getType(); 3264 3265 // Only handle constant-sized or VLAs, but not flexible members. 3266 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 3267 // Only issue the FIXIT for arrays of size > 1. 3268 if (CAT->getSize().getSExtValue() <= 1) 3269 return; 3270 } else if (!DstArgTy->isVariableArrayType()) { 3271 return; 3272 } 3273 3274 SmallString<128> sizeString; 3275 llvm::raw_svector_ostream OS(sizeString); 3276 OS << "sizeof("; 3277 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 3278 OS << ") - "; 3279 OS << "strlen("; 3280 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 3281 OS << ") - 1"; 3282 3283 Diag(SL, diag::note_strncat_wrong_size) 3284 << FixItHint::CreateReplacement(SR, OS.str()); 3285} 3286 3287//===--- CHECK: Return Address of Stack Variable --------------------------===// 3288 3289static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3290 Decl *ParentDecl); 3291static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars, 3292 Decl *ParentDecl); 3293 3294/// CheckReturnStackAddr - Check if a return statement returns the address 3295/// of a stack variable. 3296void 3297Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 3298 SourceLocation ReturnLoc) { 3299 3300 Expr *stackE = 0; 3301 SmallVector<DeclRefExpr *, 8> refVars; 3302 3303 // Perform checking for returned stack addresses, local blocks, 3304 // label addresses or references to temporaries. 3305 if (lhsType->isPointerType() || 3306 (!getLangOpts().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 3307 stackE = EvalAddr(RetValExp, refVars, /*ParentDecl=*/0); 3308 } else if (lhsType->isReferenceType()) { 3309 stackE = EvalVal(RetValExp, refVars, /*ParentDecl=*/0); 3310 } 3311 3312 if (stackE == 0) 3313 return; // Nothing suspicious was found. 3314 3315 SourceLocation diagLoc; 3316 SourceRange diagRange; 3317 if (refVars.empty()) { 3318 diagLoc = stackE->getLocStart(); 3319 diagRange = stackE->getSourceRange(); 3320 } else { 3321 // We followed through a reference variable. 'stackE' contains the 3322 // problematic expression but we will warn at the return statement pointing 3323 // at the reference variable. We will later display the "trail" of 3324 // reference variables using notes. 3325 diagLoc = refVars[0]->getLocStart(); 3326 diagRange = refVars[0]->getSourceRange(); 3327 } 3328 3329 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 3330 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 3331 : diag::warn_ret_stack_addr) 3332 << DR->getDecl()->getDeclName() << diagRange; 3333 } else if (isa<BlockExpr>(stackE)) { // local block. 3334 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 3335 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 3336 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 3337 } else { // local temporary. 3338 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 3339 : diag::warn_ret_local_temp_addr) 3340 << diagRange; 3341 } 3342 3343 // Display the "trail" of reference variables that we followed until we 3344 // found the problematic expression using notes. 3345 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 3346 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 3347 // If this var binds to another reference var, show the range of the next 3348 // var, otherwise the var binds to the problematic expression, in which case 3349 // show the range of the expression. 3350 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 3351 : stackE->getSourceRange(); 3352 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 3353 << VD->getDeclName() << range; 3354 } 3355} 3356 3357/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 3358/// check if the expression in a return statement evaluates to an address 3359/// to a location on the stack, a local block, an address of a label, or a 3360/// reference to local temporary. The recursion is used to traverse the 3361/// AST of the return expression, with recursion backtracking when we 3362/// encounter a subexpression that (1) clearly does not lead to one of the 3363/// above problematic expressions (2) is something we cannot determine leads to 3364/// a problematic expression based on such local checking. 3365/// 3366/// Both EvalAddr and EvalVal follow through reference variables to evaluate 3367/// the expression that they point to. Such variables are added to the 3368/// 'refVars' vector so that we know what the reference variable "trail" was. 3369/// 3370/// EvalAddr processes expressions that are pointers that are used as 3371/// references (and not L-values). EvalVal handles all other values. 3372/// At the base case of the recursion is a check for the above problematic 3373/// expressions. 3374/// 3375/// This implementation handles: 3376/// 3377/// * pointer-to-pointer casts 3378/// * implicit conversions from array references to pointers 3379/// * taking the address of fields 3380/// * arbitrary interplay between "&" and "*" operators 3381/// * pointer arithmetic from an address of a stack variable 3382/// * taking the address of an array element where the array is on the stack 3383static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3384 Decl *ParentDecl) { 3385 if (E->isTypeDependent()) 3386 return NULL; 3387 3388 // We should only be called for evaluating pointer expressions. 3389 assert((E->getType()->isAnyPointerType() || 3390 E->getType()->isBlockPointerType() || 3391 E->getType()->isObjCQualifiedIdType()) && 3392 "EvalAddr only works on pointers"); 3393 3394 E = E->IgnoreParens(); 3395 3396 // Our "symbolic interpreter" is just a dispatch off the currently 3397 // viewed AST node. We then recursively traverse the AST by calling 3398 // EvalAddr and EvalVal appropriately. 3399 switch (E->getStmtClass()) { 3400 case Stmt::DeclRefExprClass: { 3401 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3402 3403 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 3404 // If this is a reference variable, follow through to the expression that 3405 // it points to. 3406 if (V->hasLocalStorage() && 3407 V->getType()->isReferenceType() && V->hasInit()) { 3408 // Add the reference variable to the "trail". 3409 refVars.push_back(DR); 3410 return EvalAddr(V->getInit(), refVars, ParentDecl); 3411 } 3412 3413 return NULL; 3414 } 3415 3416 case Stmt::UnaryOperatorClass: { 3417 // The only unary operator that make sense to handle here 3418 // is AddrOf. All others don't make sense as pointers. 3419 UnaryOperator *U = cast<UnaryOperator>(E); 3420 3421 if (U->getOpcode() == UO_AddrOf) 3422 return EvalVal(U->getSubExpr(), refVars, ParentDecl); 3423 else 3424 return NULL; 3425 } 3426 3427 case Stmt::BinaryOperatorClass: { 3428 // Handle pointer arithmetic. All other binary operators are not valid 3429 // in this context. 3430 BinaryOperator *B = cast<BinaryOperator>(E); 3431 BinaryOperatorKind op = B->getOpcode(); 3432 3433 if (op != BO_Add && op != BO_Sub) 3434 return NULL; 3435 3436 Expr *Base = B->getLHS(); 3437 3438 // Determine which argument is the real pointer base. It could be 3439 // the RHS argument instead of the LHS. 3440 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 3441 3442 assert (Base->getType()->isPointerType()); 3443 return EvalAddr(Base, refVars, ParentDecl); 3444 } 3445 3446 // For conditional operators we need to see if either the LHS or RHS are 3447 // valid DeclRefExpr*s. If one of them is valid, we return it. 3448 case Stmt::ConditionalOperatorClass: { 3449 ConditionalOperator *C = cast<ConditionalOperator>(E); 3450 3451 // Handle the GNU extension for missing LHS. 3452 if (Expr *lhsExpr = C->getLHS()) { 3453 // In C++, we can have a throw-expression, which has 'void' type. 3454 if (!lhsExpr->getType()->isVoidType()) 3455 if (Expr* LHS = EvalAddr(lhsExpr, refVars, ParentDecl)) 3456 return LHS; 3457 } 3458 3459 // In C++, we can have a throw-expression, which has 'void' type. 3460 if (C->getRHS()->getType()->isVoidType()) 3461 return NULL; 3462 3463 return EvalAddr(C->getRHS(), refVars, ParentDecl); 3464 } 3465 3466 case Stmt::BlockExprClass: 3467 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 3468 return E; // local block. 3469 return NULL; 3470 3471 case Stmt::AddrLabelExprClass: 3472 return E; // address of label. 3473 3474 case Stmt::ExprWithCleanupsClass: 3475 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars, 3476 ParentDecl); 3477 3478 // For casts, we need to handle conversions from arrays to 3479 // pointer values, and pointer-to-pointer conversions. 3480 case Stmt::ImplicitCastExprClass: 3481 case Stmt::CStyleCastExprClass: 3482 case Stmt::CXXFunctionalCastExprClass: 3483 case Stmt::ObjCBridgedCastExprClass: 3484 case Stmt::CXXStaticCastExprClass: 3485 case Stmt::CXXDynamicCastExprClass: 3486 case Stmt::CXXConstCastExprClass: 3487 case Stmt::CXXReinterpretCastExprClass: { 3488 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 3489 switch (cast<CastExpr>(E)->getCastKind()) { 3490 case CK_BitCast: 3491 case CK_LValueToRValue: 3492 case CK_NoOp: 3493 case CK_BaseToDerived: 3494 case CK_DerivedToBase: 3495 case CK_UncheckedDerivedToBase: 3496 case CK_Dynamic: 3497 case CK_CPointerToObjCPointerCast: 3498 case CK_BlockPointerToObjCPointerCast: 3499 case CK_AnyPointerToBlockPointerCast: 3500 return EvalAddr(SubExpr, refVars, ParentDecl); 3501 3502 case CK_ArrayToPointerDecay: 3503 return EvalVal(SubExpr, refVars, ParentDecl); 3504 3505 default: 3506 return 0; 3507 } 3508 } 3509 3510 case Stmt::MaterializeTemporaryExprClass: 3511 if (Expr *Result = EvalAddr( 3512 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3513 refVars, ParentDecl)) 3514 return Result; 3515 3516 return E; 3517 3518 // Everything else: we simply don't reason about them. 3519 default: 3520 return NULL; 3521 } 3522} 3523 3524 3525/// EvalVal - This function is complements EvalAddr in the mutual recursion. 3526/// See the comments for EvalAddr for more details. 3527static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3528 Decl *ParentDecl) { 3529do { 3530 // We should only be called for evaluating non-pointer expressions, or 3531 // expressions with a pointer type that are not used as references but instead 3532 // are l-values (e.g., DeclRefExpr with a pointer type). 3533 3534 // Our "symbolic interpreter" is just a dispatch off the currently 3535 // viewed AST node. We then recursively traverse the AST by calling 3536 // EvalAddr and EvalVal appropriately. 3537 3538 E = E->IgnoreParens(); 3539 switch (E->getStmtClass()) { 3540 case Stmt::ImplicitCastExprClass: { 3541 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 3542 if (IE->getValueKind() == VK_LValue) { 3543 E = IE->getSubExpr(); 3544 continue; 3545 } 3546 return NULL; 3547 } 3548 3549 case Stmt::ExprWithCleanupsClass: 3550 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,ParentDecl); 3551 3552 case Stmt::DeclRefExprClass: { 3553 // When we hit a DeclRefExpr we are looking at code that refers to a 3554 // variable's name. If it's not a reference variable we check if it has 3555 // local storage within the function, and if so, return the expression. 3556 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3557 3558 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) { 3559 // Check if it refers to itself, e.g. "int& i = i;". 3560 if (V == ParentDecl) 3561 return DR; 3562 3563 if (V->hasLocalStorage()) { 3564 if (!V->getType()->isReferenceType()) 3565 return DR; 3566 3567 // Reference variable, follow through to the expression that 3568 // it points to. 3569 if (V->hasInit()) { 3570 // Add the reference variable to the "trail". 3571 refVars.push_back(DR); 3572 return EvalVal(V->getInit(), refVars, V); 3573 } 3574 } 3575 } 3576 3577 return NULL; 3578 } 3579 3580 case Stmt::UnaryOperatorClass: { 3581 // The only unary operator that make sense to handle here 3582 // is Deref. All others don't resolve to a "name." This includes 3583 // handling all sorts of rvalues passed to a unary operator. 3584 UnaryOperator *U = cast<UnaryOperator>(E); 3585 3586 if (U->getOpcode() == UO_Deref) 3587 return EvalAddr(U->getSubExpr(), refVars, ParentDecl); 3588 3589 return NULL; 3590 } 3591 3592 case Stmt::ArraySubscriptExprClass: { 3593 // Array subscripts are potential references to data on the stack. We 3594 // retrieve the DeclRefExpr* for the array variable if it indeed 3595 // has local storage. 3596 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars,ParentDecl); 3597 } 3598 3599 case Stmt::ConditionalOperatorClass: { 3600 // For conditional operators we need to see if either the LHS or RHS are 3601 // non-NULL Expr's. If one is non-NULL, we return it. 3602 ConditionalOperator *C = cast<ConditionalOperator>(E); 3603 3604 // Handle the GNU extension for missing LHS. 3605 if (Expr *lhsExpr = C->getLHS()) 3606 if (Expr *LHS = EvalVal(lhsExpr, refVars, ParentDecl)) 3607 return LHS; 3608 3609 return EvalVal(C->getRHS(), refVars, ParentDecl); 3610 } 3611 3612 // Accesses to members are potential references to data on the stack. 3613 case Stmt::MemberExprClass: { 3614 MemberExpr *M = cast<MemberExpr>(E); 3615 3616 // Check for indirect access. We only want direct field accesses. 3617 if (M->isArrow()) 3618 return NULL; 3619 3620 // Check whether the member type is itself a reference, in which case 3621 // we're not going to refer to the member, but to what the member refers to. 3622 if (M->getMemberDecl()->getType()->isReferenceType()) 3623 return NULL; 3624 3625 return EvalVal(M->getBase(), refVars, ParentDecl); 3626 } 3627 3628 case Stmt::MaterializeTemporaryExprClass: 3629 if (Expr *Result = EvalVal( 3630 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3631 refVars, ParentDecl)) 3632 return Result; 3633 3634 return E; 3635 3636 default: 3637 // Check that we don't return or take the address of a reference to a 3638 // temporary. This is only useful in C++. 3639 if (!E->isTypeDependent() && E->isRValue()) 3640 return E; 3641 3642 // Everything else: we simply don't reason about them. 3643 return NULL; 3644 } 3645} while (true); 3646} 3647 3648//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 3649 3650/// Check for comparisons of floating point operands using != and ==. 3651/// Issue a warning if these are no self-comparisons, as they are not likely 3652/// to do what the programmer intended. 3653void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 3654 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 3655 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 3656 3657 // Special case: check for x == x (which is OK). 3658 // Do not emit warnings for such cases. 3659 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 3660 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 3661 if (DRL->getDecl() == DRR->getDecl()) 3662 return; 3663 3664 3665 // Special case: check for comparisons against literals that can be exactly 3666 // represented by APFloat. In such cases, do not emit a warning. This 3667 // is a heuristic: often comparison against such literals are used to 3668 // detect if a value in a variable has not changed. This clearly can 3669 // lead to false negatives. 3670 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 3671 if (FLL->isExact()) 3672 return; 3673 } else 3674 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)) 3675 if (FLR->isExact()) 3676 return; 3677 3678 // Check for comparisons with builtin types. 3679 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 3680 if (CL->isBuiltinCall()) 3681 return; 3682 3683 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 3684 if (CR->isBuiltinCall()) 3685 return; 3686 3687 // Emit the diagnostic. 3688 Diag(Loc, diag::warn_floatingpoint_eq) 3689 << LHS->getSourceRange() << RHS->getSourceRange(); 3690} 3691 3692//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 3693//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 3694 3695namespace { 3696 3697/// Structure recording the 'active' range of an integer-valued 3698/// expression. 3699struct IntRange { 3700 /// The number of bits active in the int. 3701 unsigned Width; 3702 3703 /// True if the int is known not to have negative values. 3704 bool NonNegative; 3705 3706 IntRange(unsigned Width, bool NonNegative) 3707 : Width(Width), NonNegative(NonNegative) 3708 {} 3709 3710 /// Returns the range of the bool type. 3711 static IntRange forBoolType() { 3712 return IntRange(1, true); 3713 } 3714 3715 /// Returns the range of an opaque value of the given integral type. 3716 static IntRange forValueOfType(ASTContext &C, QualType T) { 3717 return forValueOfCanonicalType(C, 3718 T->getCanonicalTypeInternal().getTypePtr()); 3719 } 3720 3721 /// Returns the range of an opaque value of a canonical integral type. 3722 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 3723 assert(T->isCanonicalUnqualified()); 3724 3725 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3726 T = VT->getElementType().getTypePtr(); 3727 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3728 T = CT->getElementType().getTypePtr(); 3729 3730 // For enum types, use the known bit width of the enumerators. 3731 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 3732 EnumDecl *Enum = ET->getDecl(); 3733 if (!Enum->isCompleteDefinition()) 3734 return IntRange(C.getIntWidth(QualType(T, 0)), false); 3735 3736 unsigned NumPositive = Enum->getNumPositiveBits(); 3737 unsigned NumNegative = Enum->getNumNegativeBits(); 3738 3739 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 3740 } 3741 3742 const BuiltinType *BT = cast<BuiltinType>(T); 3743 assert(BT->isInteger()); 3744 3745 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3746 } 3747 3748 /// Returns the "target" range of a canonical integral type, i.e. 3749 /// the range of values expressible in the type. 3750 /// 3751 /// This matches forValueOfCanonicalType except that enums have the 3752 /// full range of their type, not the range of their enumerators. 3753 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 3754 assert(T->isCanonicalUnqualified()); 3755 3756 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3757 T = VT->getElementType().getTypePtr(); 3758 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3759 T = CT->getElementType().getTypePtr(); 3760 if (const EnumType *ET = dyn_cast<EnumType>(T)) 3761 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 3762 3763 const BuiltinType *BT = cast<BuiltinType>(T); 3764 assert(BT->isInteger()); 3765 3766 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3767 } 3768 3769 /// Returns the supremum of two ranges: i.e. their conservative merge. 3770 static IntRange join(IntRange L, IntRange R) { 3771 return IntRange(std::max(L.Width, R.Width), 3772 L.NonNegative && R.NonNegative); 3773 } 3774 3775 /// Returns the infinum of two ranges: i.e. their aggressive merge. 3776 static IntRange meet(IntRange L, IntRange R) { 3777 return IntRange(std::min(L.Width, R.Width), 3778 L.NonNegative || R.NonNegative); 3779 } 3780}; 3781 3782static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, 3783 unsigned MaxWidth) { 3784 if (value.isSigned() && value.isNegative()) 3785 return IntRange(value.getMinSignedBits(), false); 3786 3787 if (value.getBitWidth() > MaxWidth) 3788 value = value.trunc(MaxWidth); 3789 3790 // isNonNegative() just checks the sign bit without considering 3791 // signedness. 3792 return IntRange(value.getActiveBits(), true); 3793} 3794 3795static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 3796 unsigned MaxWidth) { 3797 if (result.isInt()) 3798 return GetValueRange(C, result.getInt(), MaxWidth); 3799 3800 if (result.isVector()) { 3801 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 3802 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 3803 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 3804 R = IntRange::join(R, El); 3805 } 3806 return R; 3807 } 3808 3809 if (result.isComplexInt()) { 3810 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 3811 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 3812 return IntRange::join(R, I); 3813 } 3814 3815 // This can happen with lossless casts to intptr_t of "based" lvalues. 3816 // Assume it might use arbitrary bits. 3817 // FIXME: The only reason we need to pass the type in here is to get 3818 // the sign right on this one case. It would be nice if APValue 3819 // preserved this. 3820 assert(result.isLValue() || result.isAddrLabelDiff()); 3821 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 3822} 3823 3824/// Pseudo-evaluate the given integer expression, estimating the 3825/// range of values it might take. 3826/// 3827/// \param MaxWidth - the width to which the value will be truncated 3828static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 3829 E = E->IgnoreParens(); 3830 3831 // Try a full evaluation first. 3832 Expr::EvalResult result; 3833 if (E->EvaluateAsRValue(result, C)) 3834 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 3835 3836 // I think we only want to look through implicit casts here; if the 3837 // user has an explicit widening cast, we should treat the value as 3838 // being of the new, wider type. 3839 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 3840 if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue) 3841 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 3842 3843 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 3844 3845 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 3846 3847 // Assume that non-integer casts can span the full range of the type. 3848 if (!isIntegerCast) 3849 return OutputTypeRange; 3850 3851 IntRange SubRange 3852 = GetExprRange(C, CE->getSubExpr(), 3853 std::min(MaxWidth, OutputTypeRange.Width)); 3854 3855 // Bail out if the subexpr's range is as wide as the cast type. 3856 if (SubRange.Width >= OutputTypeRange.Width) 3857 return OutputTypeRange; 3858 3859 // Otherwise, we take the smaller width, and we're non-negative if 3860 // either the output type or the subexpr is. 3861 return IntRange(SubRange.Width, 3862 SubRange.NonNegative || OutputTypeRange.NonNegative); 3863 } 3864 3865 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 3866 // If we can fold the condition, just take that operand. 3867 bool CondResult; 3868 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 3869 return GetExprRange(C, CondResult ? CO->getTrueExpr() 3870 : CO->getFalseExpr(), 3871 MaxWidth); 3872 3873 // Otherwise, conservatively merge. 3874 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 3875 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 3876 return IntRange::join(L, R); 3877 } 3878 3879 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 3880 switch (BO->getOpcode()) { 3881 3882 // Boolean-valued operations are single-bit and positive. 3883 case BO_LAnd: 3884 case BO_LOr: 3885 case BO_LT: 3886 case BO_GT: 3887 case BO_LE: 3888 case BO_GE: 3889 case BO_EQ: 3890 case BO_NE: 3891 return IntRange::forBoolType(); 3892 3893 // The type of the assignments is the type of the LHS, so the RHS 3894 // is not necessarily the same type. 3895 case BO_MulAssign: 3896 case BO_DivAssign: 3897 case BO_RemAssign: 3898 case BO_AddAssign: 3899 case BO_SubAssign: 3900 case BO_XorAssign: 3901 case BO_OrAssign: 3902 // TODO: bitfields? 3903 return IntRange::forValueOfType(C, E->getType()); 3904 3905 // Simple assignments just pass through the RHS, which will have 3906 // been coerced to the LHS type. 3907 case BO_Assign: 3908 // TODO: bitfields? 3909 return GetExprRange(C, BO->getRHS(), MaxWidth); 3910 3911 // Operations with opaque sources are black-listed. 3912 case BO_PtrMemD: 3913 case BO_PtrMemI: 3914 return IntRange::forValueOfType(C, E->getType()); 3915 3916 // Bitwise-and uses the *infinum* of the two source ranges. 3917 case BO_And: 3918 case BO_AndAssign: 3919 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 3920 GetExprRange(C, BO->getRHS(), MaxWidth)); 3921 3922 // Left shift gets black-listed based on a judgement call. 3923 case BO_Shl: 3924 // ...except that we want to treat '1 << (blah)' as logically 3925 // positive. It's an important idiom. 3926 if (IntegerLiteral *I 3927 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 3928 if (I->getValue() == 1) { 3929 IntRange R = IntRange::forValueOfType(C, E->getType()); 3930 return IntRange(R.Width, /*NonNegative*/ true); 3931 } 3932 } 3933 // fallthrough 3934 3935 case BO_ShlAssign: 3936 return IntRange::forValueOfType(C, E->getType()); 3937 3938 // Right shift by a constant can narrow its left argument. 3939 case BO_Shr: 3940 case BO_ShrAssign: { 3941 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3942 3943 // If the shift amount is a positive constant, drop the width by 3944 // that much. 3945 llvm::APSInt shift; 3946 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 3947 shift.isNonNegative()) { 3948 unsigned zext = shift.getZExtValue(); 3949 if (zext >= L.Width) 3950 L.Width = (L.NonNegative ? 0 : 1); 3951 else 3952 L.Width -= zext; 3953 } 3954 3955 return L; 3956 } 3957 3958 // Comma acts as its right operand. 3959 case BO_Comma: 3960 return GetExprRange(C, BO->getRHS(), MaxWidth); 3961 3962 // Black-list pointer subtractions. 3963 case BO_Sub: 3964 if (BO->getLHS()->getType()->isPointerType()) 3965 return IntRange::forValueOfType(C, E->getType()); 3966 break; 3967 3968 // The width of a division result is mostly determined by the size 3969 // of the LHS. 3970 case BO_Div: { 3971 // Don't 'pre-truncate' the operands. 3972 unsigned opWidth = C.getIntWidth(E->getType()); 3973 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3974 3975 // If the divisor is constant, use that. 3976 llvm::APSInt divisor; 3977 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 3978 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 3979 if (log2 >= L.Width) 3980 L.Width = (L.NonNegative ? 0 : 1); 3981 else 3982 L.Width = std::min(L.Width - log2, MaxWidth); 3983 return L; 3984 } 3985 3986 // Otherwise, just use the LHS's width. 3987 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3988 return IntRange(L.Width, L.NonNegative && R.NonNegative); 3989 } 3990 3991 // The result of a remainder can't be larger than the result of 3992 // either side. 3993 case BO_Rem: { 3994 // Don't 'pre-truncate' the operands. 3995 unsigned opWidth = C.getIntWidth(E->getType()); 3996 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3997 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3998 3999 IntRange meet = IntRange::meet(L, R); 4000 meet.Width = std::min(meet.Width, MaxWidth); 4001 return meet; 4002 } 4003 4004 // The default behavior is okay for these. 4005 case BO_Mul: 4006 case BO_Add: 4007 case BO_Xor: 4008 case BO_Or: 4009 break; 4010 } 4011 4012 // The default case is to treat the operation as if it were closed 4013 // on the narrowest type that encompasses both operands. 4014 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 4015 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 4016 return IntRange::join(L, R); 4017 } 4018 4019 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 4020 switch (UO->getOpcode()) { 4021 // Boolean-valued operations are white-listed. 4022 case UO_LNot: 4023 return IntRange::forBoolType(); 4024 4025 // Operations with opaque sources are black-listed. 4026 case UO_Deref: 4027 case UO_AddrOf: // should be impossible 4028 return IntRange::forValueOfType(C, E->getType()); 4029 4030 default: 4031 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 4032 } 4033 } 4034 4035 if (dyn_cast<OffsetOfExpr>(E)) { 4036 IntRange::forValueOfType(C, E->getType()); 4037 } 4038 4039 if (FieldDecl *BitField = E->getBitField()) 4040 return IntRange(BitField->getBitWidthValue(C), 4041 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 4042 4043 return IntRange::forValueOfType(C, E->getType()); 4044} 4045 4046static IntRange GetExprRange(ASTContext &C, Expr *E) { 4047 return GetExprRange(C, E, C.getIntWidth(E->getType())); 4048} 4049 4050/// Checks whether the given value, which currently has the given 4051/// source semantics, has the same value when coerced through the 4052/// target semantics. 4053static bool IsSameFloatAfterCast(const llvm::APFloat &value, 4054 const llvm::fltSemantics &Src, 4055 const llvm::fltSemantics &Tgt) { 4056 llvm::APFloat truncated = value; 4057 4058 bool ignored; 4059 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 4060 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 4061 4062 return truncated.bitwiseIsEqual(value); 4063} 4064 4065/// Checks whether the given value, which currently has the given 4066/// source semantics, has the same value when coerced through the 4067/// target semantics. 4068/// 4069/// The value might be a vector of floats (or a complex number). 4070static bool IsSameFloatAfterCast(const APValue &value, 4071 const llvm::fltSemantics &Src, 4072 const llvm::fltSemantics &Tgt) { 4073 if (value.isFloat()) 4074 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 4075 4076 if (value.isVector()) { 4077 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 4078 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 4079 return false; 4080 return true; 4081 } 4082 4083 assert(value.isComplexFloat()); 4084 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 4085 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 4086} 4087 4088static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 4089 4090static bool IsZero(Sema &S, Expr *E) { 4091 // Suppress cases where we are comparing against an enum constant. 4092 if (const DeclRefExpr *DR = 4093 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 4094 if (isa<EnumConstantDecl>(DR->getDecl())) 4095 return false; 4096 4097 // Suppress cases where the '0' value is expanded from a macro. 4098 if (E->getLocStart().isMacroID()) 4099 return false; 4100 4101 llvm::APSInt Value; 4102 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 4103} 4104 4105static bool HasEnumType(Expr *E) { 4106 // Strip off implicit integral promotions. 4107 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 4108 if (ICE->getCastKind() != CK_IntegralCast && 4109 ICE->getCastKind() != CK_NoOp) 4110 break; 4111 E = ICE->getSubExpr(); 4112 } 4113 4114 return E->getType()->isEnumeralType(); 4115} 4116 4117static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 4118 BinaryOperatorKind op = E->getOpcode(); 4119 if (E->isValueDependent()) 4120 return; 4121 4122 if (op == BO_LT && IsZero(S, E->getRHS())) { 4123 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4124 << "< 0" << "false" << HasEnumType(E->getLHS()) 4125 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4126 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 4127 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4128 << ">= 0" << "true" << HasEnumType(E->getLHS()) 4129 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4130 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 4131 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4132 << "0 >" << "false" << HasEnumType(E->getRHS()) 4133 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4134 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 4135 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4136 << "0 <=" << "true" << HasEnumType(E->getRHS()) 4137 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4138 } 4139} 4140 4141/// Analyze the operands of the given comparison. Implements the 4142/// fallback case from AnalyzeComparison. 4143static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 4144 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4145 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4146} 4147 4148/// \brief Implements -Wsign-compare. 4149/// 4150/// \param E the binary operator to check for warnings 4151static void AnalyzeComparison(Sema &S, BinaryOperator *E) { 4152 // The type the comparison is being performed in. 4153 QualType T = E->getLHS()->getType(); 4154 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 4155 && "comparison with mismatched types"); 4156 4157 // We don't do anything special if this isn't an unsigned integral 4158 // comparison: we're only interested in integral comparisons, and 4159 // signed comparisons only happen in cases we don't care to warn about. 4160 // 4161 // We also don't care about value-dependent expressions or expressions 4162 // whose result is a constant. 4163 if (!T->hasUnsignedIntegerRepresentation() 4164 || E->isValueDependent() || E->isIntegerConstantExpr(S.Context)) 4165 return AnalyzeImpConvsInComparison(S, E); 4166 4167 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 4168 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 4169 4170 // Check to see if one of the (unmodified) operands is of different 4171 // signedness. 4172 Expr *signedOperand, *unsignedOperand; 4173 if (LHS->getType()->hasSignedIntegerRepresentation()) { 4174 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 4175 "unsigned comparison between two signed integer expressions?"); 4176 signedOperand = LHS; 4177 unsignedOperand = RHS; 4178 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 4179 signedOperand = RHS; 4180 unsignedOperand = LHS; 4181 } else { 4182 CheckTrivialUnsignedComparison(S, E); 4183 return AnalyzeImpConvsInComparison(S, E); 4184 } 4185 4186 // Otherwise, calculate the effective range of the signed operand. 4187 IntRange signedRange = GetExprRange(S.Context, signedOperand); 4188 4189 // Go ahead and analyze implicit conversions in the operands. Note 4190 // that we skip the implicit conversions on both sides. 4191 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 4192 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 4193 4194 // If the signed range is non-negative, -Wsign-compare won't fire, 4195 // but we should still check for comparisons which are always true 4196 // or false. 4197 if (signedRange.NonNegative) 4198 return CheckTrivialUnsignedComparison(S, E); 4199 4200 // For (in)equality comparisons, if the unsigned operand is a 4201 // constant which cannot collide with a overflowed signed operand, 4202 // then reinterpreting the signed operand as unsigned will not 4203 // change the result of the comparison. 4204 if (E->isEqualityOp()) { 4205 unsigned comparisonWidth = S.Context.getIntWidth(T); 4206 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 4207 4208 // We should never be unable to prove that the unsigned operand is 4209 // non-negative. 4210 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 4211 4212 if (unsignedRange.Width < comparisonWidth) 4213 return; 4214 } 4215 4216 S.DiagRuntimeBehavior(E->getOperatorLoc(), E, 4217 S.PDiag(diag::warn_mixed_sign_comparison) 4218 << LHS->getType() << RHS->getType() 4219 << LHS->getSourceRange() << RHS->getSourceRange()); 4220} 4221 4222/// Analyzes an attempt to assign the given value to a bitfield. 4223/// 4224/// Returns true if there was something fishy about the attempt. 4225static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 4226 SourceLocation InitLoc) { 4227 assert(Bitfield->isBitField()); 4228 if (Bitfield->isInvalidDecl()) 4229 return false; 4230 4231 // White-list bool bitfields. 4232 if (Bitfield->getType()->isBooleanType()) 4233 return false; 4234 4235 // Ignore value- or type-dependent expressions. 4236 if (Bitfield->getBitWidth()->isValueDependent() || 4237 Bitfield->getBitWidth()->isTypeDependent() || 4238 Init->isValueDependent() || 4239 Init->isTypeDependent()) 4240 return false; 4241 4242 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 4243 4244 llvm::APSInt Value; 4245 if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects)) 4246 return false; 4247 4248 unsigned OriginalWidth = Value.getBitWidth(); 4249 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 4250 4251 if (OriginalWidth <= FieldWidth) 4252 return false; 4253 4254 // Compute the value which the bitfield will contain. 4255 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 4256 TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType()); 4257 4258 // Check whether the stored value is equal to the original value. 4259 TruncatedValue = TruncatedValue.extend(OriginalWidth); 4260 if (llvm::APSInt::isSameValue(Value, TruncatedValue)) 4261 return false; 4262 4263 // Special-case bitfields of width 1: booleans are naturally 0/1, and 4264 // therefore don't strictly fit into a signed bitfield of width 1. 4265 if (FieldWidth == 1 && Value == 1) 4266 return false; 4267 4268 std::string PrettyValue = Value.toString(10); 4269 std::string PrettyTrunc = TruncatedValue.toString(10); 4270 4271 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 4272 << PrettyValue << PrettyTrunc << OriginalInit->getType() 4273 << Init->getSourceRange(); 4274 4275 return true; 4276} 4277 4278/// Analyze the given simple or compound assignment for warning-worthy 4279/// operations. 4280static void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 4281 // Just recurse on the LHS. 4282 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4283 4284 // We want to recurse on the RHS as normal unless we're assigning to 4285 // a bitfield. 4286 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 4287 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 4288 E->getOperatorLoc())) { 4289 // Recurse, ignoring any implicit conversions on the RHS. 4290 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 4291 E->getOperatorLoc()); 4292 } 4293 } 4294 4295 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4296} 4297 4298/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4299static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 4300 SourceLocation CContext, unsigned diag, 4301 bool pruneControlFlow = false) { 4302 if (pruneControlFlow) { 4303 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4304 S.PDiag(diag) 4305 << SourceType << T << E->getSourceRange() 4306 << SourceRange(CContext)); 4307 return; 4308 } 4309 S.Diag(E->getExprLoc(), diag) 4310 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 4311} 4312 4313/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4314static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, 4315 SourceLocation CContext, unsigned diag, 4316 bool pruneControlFlow = false) { 4317 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow); 4318} 4319 4320/// Diagnose an implicit cast from a literal expression. Does not warn when the 4321/// cast wouldn't lose information. 4322void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 4323 SourceLocation CContext) { 4324 // Try to convert the literal exactly to an integer. If we can, don't warn. 4325 bool isExact = false; 4326 const llvm::APFloat &Value = FL->getValue(); 4327 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 4328 T->hasUnsignedIntegerRepresentation()); 4329 if (Value.convertToInteger(IntegerValue, 4330 llvm::APFloat::rmTowardZero, &isExact) 4331 == llvm::APFloat::opOK && isExact) 4332 return; 4333 4334 SmallString<16> PrettySourceValue; 4335 Value.toString(PrettySourceValue); 4336 SmallString<16> PrettyTargetValue; 4337 if (T->isSpecificBuiltinType(BuiltinType::Bool)) 4338 PrettyTargetValue = IntegerValue == 0 ? "false" : "true"; 4339 else 4340 IntegerValue.toString(PrettyTargetValue); 4341 4342 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 4343 << FL->getType() << T.getUnqualifiedType() << PrettySourceValue 4344 << PrettyTargetValue << FL->getSourceRange() << SourceRange(CContext); 4345} 4346 4347std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 4348 if (!Range.Width) return "0"; 4349 4350 llvm::APSInt ValueInRange = Value; 4351 ValueInRange.setIsSigned(!Range.NonNegative); 4352 ValueInRange = ValueInRange.trunc(Range.Width); 4353 return ValueInRange.toString(10); 4354} 4355 4356void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 4357 SourceLocation CC, bool *ICContext = 0) { 4358 if (E->isTypeDependent() || E->isValueDependent()) return; 4359 4360 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 4361 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 4362 if (Source == Target) return; 4363 if (Target->isDependentType()) return; 4364 4365 // If the conversion context location is invalid don't complain. We also 4366 // don't want to emit a warning if the issue occurs from the expansion of 4367 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 4368 // delay this check as long as possible. Once we detect we are in that 4369 // scenario, we just return. 4370 if (CC.isInvalid()) 4371 return; 4372 4373 // Diagnose implicit casts to bool. 4374 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 4375 if (isa<StringLiteral>(E)) 4376 // Warn on string literal to bool. Checks for string literals in logical 4377 // expressions, for instances, assert(0 && "error here"), is prevented 4378 // by a check in AnalyzeImplicitConversions(). 4379 return DiagnoseImpCast(S, E, T, CC, 4380 diag::warn_impcast_string_literal_to_bool); 4381 if (Source->isFunctionType()) { 4382 // Warn on function to bool. Checks free functions and static member 4383 // functions. Weakly imported functions are excluded from the check, 4384 // since it's common to test their value to check whether the linker 4385 // found a definition for them. 4386 ValueDecl *D = 0; 4387 if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) { 4388 D = R->getDecl(); 4389 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 4390 D = M->getMemberDecl(); 4391 } 4392 4393 if (D && !D->isWeak()) { 4394 if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) { 4395 S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool) 4396 << F << E->getSourceRange() << SourceRange(CC); 4397 S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence) 4398 << FixItHint::CreateInsertion(E->getExprLoc(), "&"); 4399 QualType ReturnType; 4400 UnresolvedSet<4> NonTemplateOverloads; 4401 S.isExprCallable(*E, ReturnType, NonTemplateOverloads); 4402 if (!ReturnType.isNull() 4403 && ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 4404 S.Diag(E->getExprLoc(), diag::note_function_to_bool_call) 4405 << FixItHint::CreateInsertion( 4406 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()"); 4407 return; 4408 } 4409 } 4410 } 4411 } 4412 4413 // Strip vector types. 4414 if (isa<VectorType>(Source)) { 4415 if (!isa<VectorType>(Target)) { 4416 if (S.SourceMgr.isInSystemMacro(CC)) 4417 return; 4418 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 4419 } 4420 4421 // If the vector cast is cast between two vectors of the same size, it is 4422 // a bitcast, not a conversion. 4423 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 4424 return; 4425 4426 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 4427 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 4428 } 4429 4430 // Strip complex types. 4431 if (isa<ComplexType>(Source)) { 4432 if (!isa<ComplexType>(Target)) { 4433 if (S.SourceMgr.isInSystemMacro(CC)) 4434 return; 4435 4436 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 4437 } 4438 4439 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 4440 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 4441 } 4442 4443 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 4444 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 4445 4446 // If the source is floating point... 4447 if (SourceBT && SourceBT->isFloatingPoint()) { 4448 // ...and the target is floating point... 4449 if (TargetBT && TargetBT->isFloatingPoint()) { 4450 // ...then warn if we're dropping FP rank. 4451 4452 // Builtin FP kinds are ordered by increasing FP rank. 4453 if (SourceBT->getKind() > TargetBT->getKind()) { 4454 // Don't warn about float constants that are precisely 4455 // representable in the target type. 4456 Expr::EvalResult result; 4457 if (E->EvaluateAsRValue(result, S.Context)) { 4458 // Value might be a float, a float vector, or a float complex. 4459 if (IsSameFloatAfterCast(result.Val, 4460 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 4461 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 4462 return; 4463 } 4464 4465 if (S.SourceMgr.isInSystemMacro(CC)) 4466 return; 4467 4468 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 4469 } 4470 return; 4471 } 4472 4473 // If the target is integral, always warn. 4474 if (TargetBT && TargetBT->isInteger()) { 4475 if (S.SourceMgr.isInSystemMacro(CC)) 4476 return; 4477 4478 Expr *InnerE = E->IgnoreParenImpCasts(); 4479 // We also want to warn on, e.g., "int i = -1.234" 4480 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 4481 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 4482 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 4483 4484 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 4485 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 4486 } else { 4487 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 4488 } 4489 } 4490 4491 return; 4492 } 4493 4494 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 4495 == Expr::NPCK_GNUNull) && !Target->isAnyPointerType() 4496 && !Target->isBlockPointerType() && !Target->isMemberPointerType()) { 4497 SourceLocation Loc = E->getSourceRange().getBegin(); 4498 if (Loc.isMacroID()) 4499 Loc = S.SourceMgr.getImmediateExpansionRange(Loc).first; 4500 if (!Loc.isMacroID() || CC.isMacroID()) 4501 S.Diag(Loc, diag::warn_impcast_null_pointer_to_integer) 4502 << T << clang::SourceRange(CC) 4503 << FixItHint::CreateReplacement(Loc, S.getFixItZeroLiteralForType(T)); 4504 } 4505 4506 if (!Source->isIntegerType() || !Target->isIntegerType()) 4507 return; 4508 4509 // TODO: remove this early return once the false positives for constant->bool 4510 // in templates, macros, etc, are reduced or removed. 4511 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 4512 return; 4513 4514 IntRange SourceRange = GetExprRange(S.Context, E); 4515 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 4516 4517 if (SourceRange.Width > TargetRange.Width) { 4518 // If the source is a constant, use a default-on diagnostic. 4519 // TODO: this should happen for bitfield stores, too. 4520 llvm::APSInt Value(32); 4521 if (E->isIntegerConstantExpr(Value, S.Context)) { 4522 if (S.SourceMgr.isInSystemMacro(CC)) 4523 return; 4524 4525 std::string PrettySourceValue = Value.toString(10); 4526 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 4527 4528 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4529 S.PDiag(diag::warn_impcast_integer_precision_constant) 4530 << PrettySourceValue << PrettyTargetValue 4531 << E->getType() << T << E->getSourceRange() 4532 << clang::SourceRange(CC)); 4533 return; 4534 } 4535 4536 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 4537 if (S.SourceMgr.isInSystemMacro(CC)) 4538 return; 4539 4540 if (TargetRange.Width == 32 && S.Context.getIntWidth(E->getType()) == 64) 4541 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32, 4542 /* pruneControlFlow */ true); 4543 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 4544 } 4545 4546 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 4547 (!TargetRange.NonNegative && SourceRange.NonNegative && 4548 SourceRange.Width == TargetRange.Width)) { 4549 4550 if (S.SourceMgr.isInSystemMacro(CC)) 4551 return; 4552 4553 unsigned DiagID = diag::warn_impcast_integer_sign; 4554 4555 // Traditionally, gcc has warned about this under -Wsign-compare. 4556 // We also want to warn about it in -Wconversion. 4557 // So if -Wconversion is off, use a completely identical diagnostic 4558 // in the sign-compare group. 4559 // The conditional-checking code will 4560 if (ICContext) { 4561 DiagID = diag::warn_impcast_integer_sign_conditional; 4562 *ICContext = true; 4563 } 4564 4565 return DiagnoseImpCast(S, E, T, CC, DiagID); 4566 } 4567 4568 // Diagnose conversions between different enumeration types. 4569 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 4570 // type, to give us better diagnostics. 4571 QualType SourceType = E->getType(); 4572 if (!S.getLangOpts().CPlusPlus) { 4573 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 4574 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 4575 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 4576 SourceType = S.Context.getTypeDeclType(Enum); 4577 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 4578 } 4579 } 4580 4581 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 4582 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 4583 if ((SourceEnum->getDecl()->getIdentifier() || 4584 SourceEnum->getDecl()->getTypedefNameForAnonDecl()) && 4585 (TargetEnum->getDecl()->getIdentifier() || 4586 TargetEnum->getDecl()->getTypedefNameForAnonDecl()) && 4587 SourceEnum != TargetEnum) { 4588 if (S.SourceMgr.isInSystemMacro(CC)) 4589 return; 4590 4591 return DiagnoseImpCast(S, E, SourceType, T, CC, 4592 diag::warn_impcast_different_enum_types); 4593 } 4594 4595 return; 4596} 4597 4598void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 4599 SourceLocation CC, QualType T); 4600 4601void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 4602 SourceLocation CC, bool &ICContext) { 4603 E = E->IgnoreParenImpCasts(); 4604 4605 if (isa<ConditionalOperator>(E)) 4606 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), CC, T); 4607 4608 AnalyzeImplicitConversions(S, E, CC); 4609 if (E->getType() != T) 4610 return CheckImplicitConversion(S, E, T, CC, &ICContext); 4611 return; 4612} 4613 4614void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 4615 SourceLocation CC, QualType T) { 4616 AnalyzeImplicitConversions(S, E->getCond(), CC); 4617 4618 bool Suspicious = false; 4619 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 4620 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 4621 4622 // If -Wconversion would have warned about either of the candidates 4623 // for a signedness conversion to the context type... 4624 if (!Suspicious) return; 4625 4626 // ...but it's currently ignored... 4627 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 4628 CC)) 4629 return; 4630 4631 // ...then check whether it would have warned about either of the 4632 // candidates for a signedness conversion to the condition type. 4633 if (E->getType() == T) return; 4634 4635 Suspicious = false; 4636 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 4637 E->getType(), CC, &Suspicious); 4638 if (!Suspicious) 4639 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 4640 E->getType(), CC, &Suspicious); 4641} 4642 4643/// AnalyzeImplicitConversions - Find and report any interesting 4644/// implicit conversions in the given expression. There are a couple 4645/// of competing diagnostics here, -Wconversion and -Wsign-compare. 4646void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 4647 QualType T = OrigE->getType(); 4648 Expr *E = OrigE->IgnoreParenImpCasts(); 4649 4650 if (E->isTypeDependent() || E->isValueDependent()) 4651 return; 4652 4653 // For conditional operators, we analyze the arguments as if they 4654 // were being fed directly into the output. 4655 if (isa<ConditionalOperator>(E)) { 4656 ConditionalOperator *CO = cast<ConditionalOperator>(E); 4657 CheckConditionalOperator(S, CO, CC, T); 4658 return; 4659 } 4660 4661 // Go ahead and check any implicit conversions we might have skipped. 4662 // The non-canonical typecheck is just an optimization; 4663 // CheckImplicitConversion will filter out dead implicit conversions. 4664 if (E->getType() != T) 4665 CheckImplicitConversion(S, E, T, CC); 4666 4667 // Now continue drilling into this expression. 4668 4669 // Skip past explicit casts. 4670 if (isa<ExplicitCastExpr>(E)) { 4671 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 4672 return AnalyzeImplicitConversions(S, E, CC); 4673 } 4674 4675 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 4676 // Do a somewhat different check with comparison operators. 4677 if (BO->isComparisonOp()) 4678 return AnalyzeComparison(S, BO); 4679 4680 // And with simple assignments. 4681 if (BO->getOpcode() == BO_Assign) 4682 return AnalyzeAssignment(S, BO); 4683 } 4684 4685 // These break the otherwise-useful invariant below. Fortunately, 4686 // we don't really need to recurse into them, because any internal 4687 // expressions should have been analyzed already when they were 4688 // built into statements. 4689 if (isa<StmtExpr>(E)) return; 4690 4691 // Don't descend into unevaluated contexts. 4692 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 4693 4694 // Now just recurse over the expression's children. 4695 CC = E->getExprLoc(); 4696 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 4697 bool IsLogicalOperator = BO && BO->isLogicalOp(); 4698 for (Stmt::child_range I = E->children(); I; ++I) { 4699 Expr *ChildExpr = dyn_cast_or_null<Expr>(*I); 4700 if (!ChildExpr) 4701 continue; 4702 4703 if (IsLogicalOperator && 4704 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 4705 // Ignore checking string literals that are in logical operators. 4706 continue; 4707 AnalyzeImplicitConversions(S, ChildExpr, CC); 4708 } 4709} 4710 4711} // end anonymous namespace 4712 4713/// Diagnoses "dangerous" implicit conversions within the given 4714/// expression (which is a full expression). Implements -Wconversion 4715/// and -Wsign-compare. 4716/// 4717/// \param CC the "context" location of the implicit conversion, i.e. 4718/// the most location of the syntactic entity requiring the implicit 4719/// conversion 4720void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 4721 // Don't diagnose in unevaluated contexts. 4722 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 4723 return; 4724 4725 // Don't diagnose for value- or type-dependent expressions. 4726 if (E->isTypeDependent() || E->isValueDependent()) 4727 return; 4728 4729 // Check for array bounds violations in cases where the check isn't triggered 4730 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 4731 // ArraySubscriptExpr is on the RHS of a variable initialization. 4732 CheckArrayAccess(E); 4733 4734 // This is not the right CC for (e.g.) a variable initialization. 4735 AnalyzeImplicitConversions(*this, E, CC); 4736} 4737 4738void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 4739 FieldDecl *BitField, 4740 Expr *Init) { 4741 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 4742} 4743 4744/// CheckParmsForFunctionDef - Check that the parameters of the given 4745/// function are appropriate for the definition of a function. This 4746/// takes care of any checks that cannot be performed on the 4747/// declaration itself, e.g., that the types of each of the function 4748/// parameters are complete. 4749bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 4750 bool CheckParameterNames) { 4751 bool HasInvalidParm = false; 4752 for (; P != PEnd; ++P) { 4753 ParmVarDecl *Param = *P; 4754 4755 // C99 6.7.5.3p4: the parameters in a parameter type list in a 4756 // function declarator that is part of a function definition of 4757 // that function shall not have incomplete type. 4758 // 4759 // This is also C++ [dcl.fct]p6. 4760 if (!Param->isInvalidDecl() && 4761 RequireCompleteType(Param->getLocation(), Param->getType(), 4762 diag::err_typecheck_decl_incomplete_type)) { 4763 Param->setInvalidDecl(); 4764 HasInvalidParm = true; 4765 } 4766 4767 // C99 6.9.1p5: If the declarator includes a parameter type list, the 4768 // declaration of each parameter shall include an identifier. 4769 if (CheckParameterNames && 4770 Param->getIdentifier() == 0 && 4771 !Param->isImplicit() && 4772 !getLangOpts().CPlusPlus) 4773 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 4774 4775 // C99 6.7.5.3p12: 4776 // If the function declarator is not part of a definition of that 4777 // function, parameters may have incomplete type and may use the [*] 4778 // notation in their sequences of declarator specifiers to specify 4779 // variable length array types. 4780 QualType PType = Param->getOriginalType(); 4781 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 4782 if (AT->getSizeModifier() == ArrayType::Star) { 4783 // FIXME: This diagnosic should point the '[*]' if source-location 4784 // information is added for it. 4785 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 4786 } 4787 } 4788 } 4789 4790 return HasInvalidParm; 4791} 4792 4793/// CheckCastAlign - Implements -Wcast-align, which warns when a 4794/// pointer cast increases the alignment requirements. 4795void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 4796 // This is actually a lot of work to potentially be doing on every 4797 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 4798 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 4799 TRange.getBegin()) 4800 == DiagnosticsEngine::Ignored) 4801 return; 4802 4803 // Ignore dependent types. 4804 if (T->isDependentType() || Op->getType()->isDependentType()) 4805 return; 4806 4807 // Require that the destination be a pointer type. 4808 const PointerType *DestPtr = T->getAs<PointerType>(); 4809 if (!DestPtr) return; 4810 4811 // If the destination has alignment 1, we're done. 4812 QualType DestPointee = DestPtr->getPointeeType(); 4813 if (DestPointee->isIncompleteType()) return; 4814 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 4815 if (DestAlign.isOne()) return; 4816 4817 // Require that the source be a pointer type. 4818 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 4819 if (!SrcPtr) return; 4820 QualType SrcPointee = SrcPtr->getPointeeType(); 4821 4822 // Whitelist casts from cv void*. We already implicitly 4823 // whitelisted casts to cv void*, since they have alignment 1. 4824 // Also whitelist casts involving incomplete types, which implicitly 4825 // includes 'void'. 4826 if (SrcPointee->isIncompleteType()) return; 4827 4828 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 4829 if (SrcAlign >= DestAlign) return; 4830 4831 Diag(TRange.getBegin(), diag::warn_cast_align) 4832 << Op->getType() << T 4833 << static_cast<unsigned>(SrcAlign.getQuantity()) 4834 << static_cast<unsigned>(DestAlign.getQuantity()) 4835 << TRange << Op->getSourceRange(); 4836} 4837 4838static const Type* getElementType(const Expr *BaseExpr) { 4839 const Type* EltType = BaseExpr->getType().getTypePtr(); 4840 if (EltType->isAnyPointerType()) 4841 return EltType->getPointeeType().getTypePtr(); 4842 else if (EltType->isArrayType()) 4843 return EltType->getBaseElementTypeUnsafe(); 4844 return EltType; 4845} 4846 4847/// \brief Check whether this array fits the idiom of a size-one tail padded 4848/// array member of a struct. 4849/// 4850/// We avoid emitting out-of-bounds access warnings for such arrays as they are 4851/// commonly used to emulate flexible arrays in C89 code. 4852static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 4853 const NamedDecl *ND) { 4854 if (Size != 1 || !ND) return false; 4855 4856 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 4857 if (!FD) return false; 4858 4859 // Don't consider sizes resulting from macro expansions or template argument 4860 // substitution to form C89 tail-padded arrays. 4861 4862 TypeSourceInfo *TInfo = FD->getTypeSourceInfo(); 4863 while (TInfo) { 4864 TypeLoc TL = TInfo->getTypeLoc(); 4865 // Look through typedefs. 4866 const TypedefTypeLoc *TTL = dyn_cast<TypedefTypeLoc>(&TL); 4867 if (TTL) { 4868 const TypedefNameDecl *TDL = TTL->getTypedefNameDecl(); 4869 TInfo = TDL->getTypeSourceInfo(); 4870 continue; 4871 } 4872 ConstantArrayTypeLoc CTL = cast<ConstantArrayTypeLoc>(TL); 4873 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(CTL.getSizeExpr()); 4874 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 4875 return false; 4876 break; 4877 } 4878 4879 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 4880 if (!RD) return false; 4881 if (RD->isUnion()) return false; 4882 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 4883 if (!CRD->isStandardLayout()) return false; 4884 } 4885 4886 // See if this is the last field decl in the record. 4887 const Decl *D = FD; 4888 while ((D = D->getNextDeclInContext())) 4889 if (isa<FieldDecl>(D)) 4890 return false; 4891 return true; 4892} 4893 4894void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 4895 const ArraySubscriptExpr *ASE, 4896 bool AllowOnePastEnd, bool IndexNegated) { 4897 IndexExpr = IndexExpr->IgnoreParenImpCasts(); 4898 if (IndexExpr->isValueDependent()) 4899 return; 4900 4901 const Type *EffectiveType = getElementType(BaseExpr); 4902 BaseExpr = BaseExpr->IgnoreParenCasts(); 4903 const ConstantArrayType *ArrayTy = 4904 Context.getAsConstantArrayType(BaseExpr->getType()); 4905 if (!ArrayTy) 4906 return; 4907 4908 llvm::APSInt index; 4909 if (!IndexExpr->EvaluateAsInt(index, Context)) 4910 return; 4911 if (IndexNegated) 4912 index = -index; 4913 4914 const NamedDecl *ND = NULL; 4915 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4916 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 4917 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 4918 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 4919 4920 if (index.isUnsigned() || !index.isNegative()) { 4921 llvm::APInt size = ArrayTy->getSize(); 4922 if (!size.isStrictlyPositive()) 4923 return; 4924 4925 const Type* BaseType = getElementType(BaseExpr); 4926 if (BaseType != EffectiveType) { 4927 // Make sure we're comparing apples to apples when comparing index to size 4928 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 4929 uint64_t array_typesize = Context.getTypeSize(BaseType); 4930 // Handle ptrarith_typesize being zero, such as when casting to void* 4931 if (!ptrarith_typesize) ptrarith_typesize = 1; 4932 if (ptrarith_typesize != array_typesize) { 4933 // There's a cast to a different size type involved 4934 uint64_t ratio = array_typesize / ptrarith_typesize; 4935 // TODO: Be smarter about handling cases where array_typesize is not a 4936 // multiple of ptrarith_typesize 4937 if (ptrarith_typesize * ratio == array_typesize) 4938 size *= llvm::APInt(size.getBitWidth(), ratio); 4939 } 4940 } 4941 4942 if (size.getBitWidth() > index.getBitWidth()) 4943 index = index.zext(size.getBitWidth()); 4944 else if (size.getBitWidth() < index.getBitWidth()) 4945 size = size.zext(index.getBitWidth()); 4946 4947 // For array subscripting the index must be less than size, but for pointer 4948 // arithmetic also allow the index (offset) to be equal to size since 4949 // computing the next address after the end of the array is legal and 4950 // commonly done e.g. in C++ iterators and range-based for loops. 4951 if (AllowOnePastEnd ? index.ule(size) : index.ult(size)) 4952 return; 4953 4954 // Also don't warn for arrays of size 1 which are members of some 4955 // structure. These are often used to approximate flexible arrays in C89 4956 // code. 4957 if (IsTailPaddedMemberArray(*this, size, ND)) 4958 return; 4959 4960 // Suppress the warning if the subscript expression (as identified by the 4961 // ']' location) and the index expression are both from macro expansions 4962 // within a system header. 4963 if (ASE) { 4964 SourceLocation RBracketLoc = SourceMgr.getSpellingLoc( 4965 ASE->getRBracketLoc()); 4966 if (SourceMgr.isInSystemHeader(RBracketLoc)) { 4967 SourceLocation IndexLoc = SourceMgr.getSpellingLoc( 4968 IndexExpr->getLocStart()); 4969 if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc)) 4970 return; 4971 } 4972 } 4973 4974 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 4975 if (ASE) 4976 DiagID = diag::warn_array_index_exceeds_bounds; 4977 4978 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4979 PDiag(DiagID) << index.toString(10, true) 4980 << size.toString(10, true) 4981 << (unsigned)size.getLimitedValue(~0U) 4982 << IndexExpr->getSourceRange()); 4983 } else { 4984 unsigned DiagID = diag::warn_array_index_precedes_bounds; 4985 if (!ASE) { 4986 DiagID = diag::warn_ptr_arith_precedes_bounds; 4987 if (index.isNegative()) index = -index; 4988 } 4989 4990 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4991 PDiag(DiagID) << index.toString(10, true) 4992 << IndexExpr->getSourceRange()); 4993 } 4994 4995 if (!ND) { 4996 // Try harder to find a NamedDecl to point at in the note. 4997 while (const ArraySubscriptExpr *ASE = 4998 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 4999 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 5000 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 5001 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 5002 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 5003 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 5004 } 5005 5006 if (ND) 5007 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 5008 PDiag(diag::note_array_index_out_of_bounds) 5009 << ND->getDeclName()); 5010} 5011 5012void Sema::CheckArrayAccess(const Expr *expr) { 5013 int AllowOnePastEnd = 0; 5014 while (expr) { 5015 expr = expr->IgnoreParenImpCasts(); 5016 switch (expr->getStmtClass()) { 5017 case Stmt::ArraySubscriptExprClass: { 5018 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 5019 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE, 5020 AllowOnePastEnd > 0); 5021 return; 5022 } 5023 case Stmt::UnaryOperatorClass: { 5024 // Only unwrap the * and & unary operators 5025 const UnaryOperator *UO = cast<UnaryOperator>(expr); 5026 expr = UO->getSubExpr(); 5027 switch (UO->getOpcode()) { 5028 case UO_AddrOf: 5029 AllowOnePastEnd++; 5030 break; 5031 case UO_Deref: 5032 AllowOnePastEnd--; 5033 break; 5034 default: 5035 return; 5036 } 5037 break; 5038 } 5039 case Stmt::ConditionalOperatorClass: { 5040 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 5041 if (const Expr *lhs = cond->getLHS()) 5042 CheckArrayAccess(lhs); 5043 if (const Expr *rhs = cond->getRHS()) 5044 CheckArrayAccess(rhs); 5045 return; 5046 } 5047 default: 5048 return; 5049 } 5050 } 5051} 5052 5053//===--- CHECK: Objective-C retain cycles ----------------------------------// 5054 5055namespace { 5056 struct RetainCycleOwner { 5057 RetainCycleOwner() : Variable(0), Indirect(false) {} 5058 VarDecl *Variable; 5059 SourceRange Range; 5060 SourceLocation Loc; 5061 bool Indirect; 5062 5063 void setLocsFrom(Expr *e) { 5064 Loc = e->getExprLoc(); 5065 Range = e->getSourceRange(); 5066 } 5067 }; 5068} 5069 5070/// Consider whether capturing the given variable can possibly lead to 5071/// a retain cycle. 5072static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 5073 // In ARC, it's captured strongly iff the variable has __strong 5074 // lifetime. In MRR, it's captured strongly if the variable is 5075 // __block and has an appropriate type. 5076 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 5077 return false; 5078 5079 owner.Variable = var; 5080 owner.setLocsFrom(ref); 5081 return true; 5082} 5083 5084static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) { 5085 while (true) { 5086 e = e->IgnoreParens(); 5087 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 5088 switch (cast->getCastKind()) { 5089 case CK_BitCast: 5090 case CK_LValueBitCast: 5091 case CK_LValueToRValue: 5092 case CK_ARCReclaimReturnedObject: 5093 e = cast->getSubExpr(); 5094 continue; 5095 5096 default: 5097 return false; 5098 } 5099 } 5100 5101 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 5102 ObjCIvarDecl *ivar = ref->getDecl(); 5103 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 5104 return false; 5105 5106 // Try to find a retain cycle in the base. 5107 if (!findRetainCycleOwner(S, ref->getBase(), owner)) 5108 return false; 5109 5110 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 5111 owner.Indirect = true; 5112 return true; 5113 } 5114 5115 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 5116 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 5117 if (!var) return false; 5118 return considerVariable(var, ref, owner); 5119 } 5120 5121 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 5122 if (member->isArrow()) return false; 5123 5124 // Don't count this as an indirect ownership. 5125 e = member->getBase(); 5126 continue; 5127 } 5128 5129 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 5130 // Only pay attention to pseudo-objects on property references. 5131 ObjCPropertyRefExpr *pre 5132 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 5133 ->IgnoreParens()); 5134 if (!pre) return false; 5135 if (pre->isImplicitProperty()) return false; 5136 ObjCPropertyDecl *property = pre->getExplicitProperty(); 5137 if (!property->isRetaining() && 5138 !(property->getPropertyIvarDecl() && 5139 property->getPropertyIvarDecl()->getType() 5140 .getObjCLifetime() == Qualifiers::OCL_Strong)) 5141 return false; 5142 5143 owner.Indirect = true; 5144 if (pre->isSuperReceiver()) { 5145 owner.Variable = S.getCurMethodDecl()->getSelfDecl(); 5146 if (!owner.Variable) 5147 return false; 5148 owner.Loc = pre->getLocation(); 5149 owner.Range = pre->getSourceRange(); 5150 return true; 5151 } 5152 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 5153 ->getSourceExpr()); 5154 continue; 5155 } 5156 5157 // Array ivars? 5158 5159 return false; 5160 } 5161} 5162 5163namespace { 5164 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 5165 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 5166 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 5167 Variable(variable), Capturer(0) {} 5168 5169 VarDecl *Variable; 5170 Expr *Capturer; 5171 5172 void VisitDeclRefExpr(DeclRefExpr *ref) { 5173 if (ref->getDecl() == Variable && !Capturer) 5174 Capturer = ref; 5175 } 5176 5177 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 5178 if (Capturer) return; 5179 Visit(ref->getBase()); 5180 if (Capturer && ref->isFreeIvar()) 5181 Capturer = ref; 5182 } 5183 5184 void VisitBlockExpr(BlockExpr *block) { 5185 // Look inside nested blocks 5186 if (block->getBlockDecl()->capturesVariable(Variable)) 5187 Visit(block->getBlockDecl()->getBody()); 5188 } 5189 }; 5190} 5191 5192/// Check whether the given argument is a block which captures a 5193/// variable. 5194static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 5195 assert(owner.Variable && owner.Loc.isValid()); 5196 5197 e = e->IgnoreParenCasts(); 5198 BlockExpr *block = dyn_cast<BlockExpr>(e); 5199 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 5200 return 0; 5201 5202 FindCaptureVisitor visitor(S.Context, owner.Variable); 5203 visitor.Visit(block->getBlockDecl()->getBody()); 5204 return visitor.Capturer; 5205} 5206 5207static void diagnoseRetainCycle(Sema &S, Expr *capturer, 5208 RetainCycleOwner &owner) { 5209 assert(capturer); 5210 assert(owner.Variable && owner.Loc.isValid()); 5211 5212 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 5213 << owner.Variable << capturer->getSourceRange(); 5214 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 5215 << owner.Indirect << owner.Range; 5216} 5217 5218/// Check for a keyword selector that starts with the word 'add' or 5219/// 'set'. 5220static bool isSetterLikeSelector(Selector sel) { 5221 if (sel.isUnarySelector()) return false; 5222 5223 StringRef str = sel.getNameForSlot(0); 5224 while (!str.empty() && str.front() == '_') str = str.substr(1); 5225 if (str.startswith("set")) 5226 str = str.substr(3); 5227 else if (str.startswith("add")) { 5228 // Specially whitelist 'addOperationWithBlock:'. 5229 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 5230 return false; 5231 str = str.substr(3); 5232 } 5233 else 5234 return false; 5235 5236 if (str.empty()) return true; 5237 return !islower(str.front()); 5238} 5239 5240/// Check a message send to see if it's likely to cause a retain cycle. 5241void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 5242 // Only check instance methods whose selector looks like a setter. 5243 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 5244 return; 5245 5246 // Try to find a variable that the receiver is strongly owned by. 5247 RetainCycleOwner owner; 5248 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 5249 if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner)) 5250 return; 5251 } else { 5252 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 5253 owner.Variable = getCurMethodDecl()->getSelfDecl(); 5254 owner.Loc = msg->getSuperLoc(); 5255 owner.Range = msg->getSuperLoc(); 5256 } 5257 5258 // Check whether the receiver is captured by any of the arguments. 5259 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 5260 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 5261 return diagnoseRetainCycle(*this, capturer, owner); 5262} 5263 5264/// Check a property assign to see if it's likely to cause a retain cycle. 5265void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 5266 RetainCycleOwner owner; 5267 if (!findRetainCycleOwner(*this, receiver, owner)) 5268 return; 5269 5270 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 5271 diagnoseRetainCycle(*this, capturer, owner); 5272} 5273 5274bool Sema::checkUnsafeAssigns(SourceLocation Loc, 5275 QualType LHS, Expr *RHS) { 5276 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 5277 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 5278 return false; 5279 // strip off any implicit cast added to get to the one arc-specific 5280 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 5281 if (cast->getCastKind() == CK_ARCConsumeObject) { 5282 Diag(Loc, diag::warn_arc_retained_assign) 5283 << (LT == Qualifiers::OCL_ExplicitNone) << 1 5284 << RHS->getSourceRange(); 5285 return true; 5286 } 5287 RHS = cast->getSubExpr(); 5288 } 5289 return false; 5290} 5291 5292void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 5293 Expr *LHS, Expr *RHS) { 5294 QualType LHSType; 5295 // PropertyRef on LHS type need be directly obtained from 5296 // its declaration as it has a PsuedoType. 5297 ObjCPropertyRefExpr *PRE 5298 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens()); 5299 if (PRE && !PRE->isImplicitProperty()) { 5300 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 5301 if (PD) 5302 LHSType = PD->getType(); 5303 } 5304 5305 if (LHSType.isNull()) 5306 LHSType = LHS->getType(); 5307 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 5308 return; 5309 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 5310 // FIXME. Check for other life times. 5311 if (LT != Qualifiers::OCL_None) 5312 return; 5313 5314 if (PRE) { 5315 if (PRE->isImplicitProperty()) 5316 return; 5317 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 5318 if (!PD) 5319 return; 5320 5321 unsigned Attributes = PD->getPropertyAttributes(); 5322 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) { 5323 // when 'assign' attribute was not explicitly specified 5324 // by user, ignore it and rely on property type itself 5325 // for lifetime info. 5326 unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten(); 5327 if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) && 5328 LHSType->isObjCRetainableType()) 5329 return; 5330 5331 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 5332 if (cast->getCastKind() == CK_ARCConsumeObject) { 5333 Diag(Loc, diag::warn_arc_retained_property_assign) 5334 << RHS->getSourceRange(); 5335 return; 5336 } 5337 RHS = cast->getSubExpr(); 5338 } 5339 } 5340 else if (Attributes & ObjCPropertyDecl::OBJC_PR_weak) { 5341 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 5342 if (cast->getCastKind() == CK_ARCConsumeObject) { 5343 Diag(Loc, diag::warn_arc_retained_assign) 5344 << 0 << 0<< RHS->getSourceRange(); 5345 return; 5346 } 5347 RHS = cast->getSubExpr(); 5348 } 5349 } 5350 } 5351} 5352 5353//===--- CHECK: Empty statement body (-Wempty-body) ---------------------===// 5354 5355namespace { 5356bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr, 5357 SourceLocation StmtLoc, 5358 const NullStmt *Body) { 5359 // Do not warn if the body is a macro that expands to nothing, e.g: 5360 // 5361 // #define CALL(x) 5362 // if (condition) 5363 // CALL(0); 5364 // 5365 if (Body->hasLeadingEmptyMacro()) 5366 return false; 5367 5368 // Get line numbers of statement and body. 5369 bool StmtLineInvalid; 5370 unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc, 5371 &StmtLineInvalid); 5372 if (StmtLineInvalid) 5373 return false; 5374 5375 bool BodyLineInvalid; 5376 unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(), 5377 &BodyLineInvalid); 5378 if (BodyLineInvalid) 5379 return false; 5380 5381 // Warn if null statement and body are on the same line. 5382 if (StmtLine != BodyLine) 5383 return false; 5384 5385 return true; 5386} 5387} // Unnamed namespace 5388 5389void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc, 5390 const Stmt *Body, 5391 unsigned DiagID) { 5392 // Since this is a syntactic check, don't emit diagnostic for template 5393 // instantiations, this just adds noise. 5394 if (CurrentInstantiationScope) 5395 return; 5396 5397 // The body should be a null statement. 5398 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 5399 if (!NBody) 5400 return; 5401 5402 // Do the usual checks. 5403 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 5404 return; 5405 5406 Diag(NBody->getSemiLoc(), DiagID); 5407 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 5408} 5409 5410void Sema::DiagnoseEmptyLoopBody(const Stmt *S, 5411 const Stmt *PossibleBody) { 5412 assert(!CurrentInstantiationScope); // Ensured by caller 5413 5414 SourceLocation StmtLoc; 5415 const Stmt *Body; 5416 unsigned DiagID; 5417 if (const ForStmt *FS = dyn_cast<ForStmt>(S)) { 5418 StmtLoc = FS->getRParenLoc(); 5419 Body = FS->getBody(); 5420 DiagID = diag::warn_empty_for_body; 5421 } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) { 5422 StmtLoc = WS->getCond()->getSourceRange().getEnd(); 5423 Body = WS->getBody(); 5424 DiagID = diag::warn_empty_while_body; 5425 } else 5426 return; // Neither `for' nor `while'. 5427 5428 // The body should be a null statement. 5429 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 5430 if (!NBody) 5431 return; 5432 5433 // Skip expensive checks if diagnostic is disabled. 5434 if (Diags.getDiagnosticLevel(DiagID, NBody->getSemiLoc()) == 5435 DiagnosticsEngine::Ignored) 5436 return; 5437 5438 // Do the usual checks. 5439 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 5440 return; 5441 5442 // `for(...);' and `while(...);' are popular idioms, so in order to keep 5443 // noise level low, emit diagnostics only if for/while is followed by a 5444 // CompoundStmt, e.g.: 5445 // for (int i = 0; i < n; i++); 5446 // { 5447 // a(i); 5448 // } 5449 // or if for/while is followed by a statement with more indentation 5450 // than for/while itself: 5451 // for (int i = 0; i < n; i++); 5452 // a(i); 5453 bool ProbableTypo = isa<CompoundStmt>(PossibleBody); 5454 if (!ProbableTypo) { 5455 bool BodyColInvalid; 5456 unsigned BodyCol = SourceMgr.getPresumedColumnNumber( 5457 PossibleBody->getLocStart(), 5458 &BodyColInvalid); 5459 if (BodyColInvalid) 5460 return; 5461 5462 bool StmtColInvalid; 5463 unsigned StmtCol = SourceMgr.getPresumedColumnNumber( 5464 S->getLocStart(), 5465 &StmtColInvalid); 5466 if (StmtColInvalid) 5467 return; 5468 5469 if (BodyCol > StmtCol) 5470 ProbableTypo = true; 5471 } 5472 5473 if (ProbableTypo) { 5474 Diag(NBody->getSemiLoc(), DiagID); 5475 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 5476 } 5477} 5478