SemaChecking.cpp revision d9b859a74ecaede23a78d37f364498102ef418c9
1//===--- SemaChecking.cpp - Extra Semantic Checking -----------------------===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// This file implements extra semantic analysis beyond what is enforced 11// by the C type system. 12// 13//===----------------------------------------------------------------------===// 14 15#include "clang/Sema/Initialization.h" 16#include "clang/Sema/Sema.h" 17#include "clang/Sema/SemaInternal.h" 18#include "clang/Sema/Initialization.h" 19#include "clang/Sema/ScopeInfo.h" 20#include "clang/Analysis/Analyses/FormatString.h" 21#include "clang/AST/ASTContext.h" 22#include "clang/AST/CharUnits.h" 23#include "clang/AST/DeclCXX.h" 24#include "clang/AST/DeclObjC.h" 25#include "clang/AST/ExprCXX.h" 26#include "clang/AST/ExprObjC.h" 27#include "clang/AST/EvaluatedExprVisitor.h" 28#include "clang/AST/DeclObjC.h" 29#include "clang/AST/StmtCXX.h" 30#include "clang/AST/StmtObjC.h" 31#include "clang/Lex/Preprocessor.h" 32#include "llvm/ADT/BitVector.h" 33#include "llvm/ADT/STLExtras.h" 34#include "llvm/Support/raw_ostream.h" 35#include "clang/Basic/TargetBuiltins.h" 36#include "clang/Basic/TargetInfo.h" 37#include "clang/Basic/ConvertUTF.h" 38#include <limits> 39using namespace clang; 40using namespace sema; 41 42SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 43 unsigned ByteNo) const { 44 return SL->getLocationOfByte(ByteNo, PP.getSourceManager(), 45 PP.getLangOptions(), PP.getTargetInfo()); 46} 47 48 49/// CheckablePrintfAttr - does a function call have a "printf" attribute 50/// and arguments that merit checking? 51bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) { 52 if (Format->getType() == "printf") return true; 53 if (Format->getType() == "printf0") { 54 // printf0 allows null "format" string; if so don't check format/args 55 unsigned format_idx = Format->getFormatIdx() - 1; 56 // Does the index refer to the implicit object argument? 57 if (isa<CXXMemberCallExpr>(TheCall)) { 58 if (format_idx == 0) 59 return false; 60 --format_idx; 61 } 62 if (format_idx < TheCall->getNumArgs()) { 63 Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts(); 64 if (!Format->isNullPointerConstant(Context, 65 Expr::NPC_ValueDependentIsNull)) 66 return true; 67 } 68 } 69 return false; 70} 71 72/// Checks that a call expression's argument count is the desired number. 73/// This is useful when doing custom type-checking. Returns true on error. 74static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) { 75 unsigned argCount = call->getNumArgs(); 76 if (argCount == desiredArgCount) return false; 77 78 if (argCount < desiredArgCount) 79 return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args) 80 << 0 /*function call*/ << desiredArgCount << argCount 81 << call->getSourceRange(); 82 83 // Highlight all the excess arguments. 84 SourceRange range(call->getArg(desiredArgCount)->getLocStart(), 85 call->getArg(argCount - 1)->getLocEnd()); 86 87 return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args) 88 << 0 /*function call*/ << desiredArgCount << argCount 89 << call->getArg(1)->getSourceRange(); 90} 91 92/// CheckBuiltinAnnotationString - Checks that string argument to the builtin 93/// annotation is a non wide string literal. 94static bool CheckBuiltinAnnotationString(Sema &S, Expr *Arg) { 95 Arg = Arg->IgnoreParenCasts(); 96 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 97 if (!Literal || !Literal->isAscii()) { 98 S.Diag(Arg->getLocStart(), diag::err_builtin_annotation_not_string_constant) 99 << Arg->getSourceRange(); 100 return true; 101 } 102 return false; 103} 104 105ExprResult 106Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 107 ExprResult TheCallResult(Owned(TheCall)); 108 109 // Find out if any arguments are required to be integer constant expressions. 110 unsigned ICEArguments = 0; 111 ASTContext::GetBuiltinTypeError Error; 112 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments); 113 if (Error != ASTContext::GE_None) 114 ICEArguments = 0; // Don't diagnose previously diagnosed errors. 115 116 // If any arguments are required to be ICE's, check and diagnose. 117 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) { 118 // Skip arguments not required to be ICE's. 119 if ((ICEArguments & (1 << ArgNo)) == 0) continue; 120 121 llvm::APSInt Result; 122 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 123 return true; 124 ICEArguments &= ~(1 << ArgNo); 125 } 126 127 switch (BuiltinID) { 128 case Builtin::BI__builtin___CFStringMakeConstantString: 129 assert(TheCall->getNumArgs() == 1 && 130 "Wrong # arguments to builtin CFStringMakeConstantString"); 131 if (CheckObjCString(TheCall->getArg(0))) 132 return ExprError(); 133 break; 134 case Builtin::BI__builtin_stdarg_start: 135 case Builtin::BI__builtin_va_start: 136 if (SemaBuiltinVAStart(TheCall)) 137 return ExprError(); 138 break; 139 case Builtin::BI__builtin_isgreater: 140 case Builtin::BI__builtin_isgreaterequal: 141 case Builtin::BI__builtin_isless: 142 case Builtin::BI__builtin_islessequal: 143 case Builtin::BI__builtin_islessgreater: 144 case Builtin::BI__builtin_isunordered: 145 if (SemaBuiltinUnorderedCompare(TheCall)) 146 return ExprError(); 147 break; 148 case Builtin::BI__builtin_fpclassify: 149 if (SemaBuiltinFPClassification(TheCall, 6)) 150 return ExprError(); 151 break; 152 case Builtin::BI__builtin_isfinite: 153 case Builtin::BI__builtin_isinf: 154 case Builtin::BI__builtin_isinf_sign: 155 case Builtin::BI__builtin_isnan: 156 case Builtin::BI__builtin_isnormal: 157 if (SemaBuiltinFPClassification(TheCall, 1)) 158 return ExprError(); 159 break; 160 case Builtin::BI__builtin_shufflevector: 161 return SemaBuiltinShuffleVector(TheCall); 162 // TheCall will be freed by the smart pointer here, but that's fine, since 163 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 164 case Builtin::BI__builtin_prefetch: 165 if (SemaBuiltinPrefetch(TheCall)) 166 return ExprError(); 167 break; 168 case Builtin::BI__builtin_object_size: 169 if (SemaBuiltinObjectSize(TheCall)) 170 return ExprError(); 171 break; 172 case Builtin::BI__builtin_longjmp: 173 if (SemaBuiltinLongjmp(TheCall)) 174 return ExprError(); 175 break; 176 177 case Builtin::BI__builtin_classify_type: 178 if (checkArgCount(*this, TheCall, 1)) return true; 179 TheCall->setType(Context.IntTy); 180 break; 181 case Builtin::BI__builtin_constant_p: 182 if (checkArgCount(*this, TheCall, 1)) return true; 183 TheCall->setType(Context.IntTy); 184 break; 185 case Builtin::BI__sync_fetch_and_add: 186 case Builtin::BI__sync_fetch_and_add_1: 187 case Builtin::BI__sync_fetch_and_add_2: 188 case Builtin::BI__sync_fetch_and_add_4: 189 case Builtin::BI__sync_fetch_and_add_8: 190 case Builtin::BI__sync_fetch_and_add_16: 191 case Builtin::BI__sync_fetch_and_sub: 192 case Builtin::BI__sync_fetch_and_sub_1: 193 case Builtin::BI__sync_fetch_and_sub_2: 194 case Builtin::BI__sync_fetch_and_sub_4: 195 case Builtin::BI__sync_fetch_and_sub_8: 196 case Builtin::BI__sync_fetch_and_sub_16: 197 case Builtin::BI__sync_fetch_and_or: 198 case Builtin::BI__sync_fetch_and_or_1: 199 case Builtin::BI__sync_fetch_and_or_2: 200 case Builtin::BI__sync_fetch_and_or_4: 201 case Builtin::BI__sync_fetch_and_or_8: 202 case Builtin::BI__sync_fetch_and_or_16: 203 case Builtin::BI__sync_fetch_and_and: 204 case Builtin::BI__sync_fetch_and_and_1: 205 case Builtin::BI__sync_fetch_and_and_2: 206 case Builtin::BI__sync_fetch_and_and_4: 207 case Builtin::BI__sync_fetch_and_and_8: 208 case Builtin::BI__sync_fetch_and_and_16: 209 case Builtin::BI__sync_fetch_and_xor: 210 case Builtin::BI__sync_fetch_and_xor_1: 211 case Builtin::BI__sync_fetch_and_xor_2: 212 case Builtin::BI__sync_fetch_and_xor_4: 213 case Builtin::BI__sync_fetch_and_xor_8: 214 case Builtin::BI__sync_fetch_and_xor_16: 215 case Builtin::BI__sync_add_and_fetch: 216 case Builtin::BI__sync_add_and_fetch_1: 217 case Builtin::BI__sync_add_and_fetch_2: 218 case Builtin::BI__sync_add_and_fetch_4: 219 case Builtin::BI__sync_add_and_fetch_8: 220 case Builtin::BI__sync_add_and_fetch_16: 221 case Builtin::BI__sync_sub_and_fetch: 222 case Builtin::BI__sync_sub_and_fetch_1: 223 case Builtin::BI__sync_sub_and_fetch_2: 224 case Builtin::BI__sync_sub_and_fetch_4: 225 case Builtin::BI__sync_sub_and_fetch_8: 226 case Builtin::BI__sync_sub_and_fetch_16: 227 case Builtin::BI__sync_and_and_fetch: 228 case Builtin::BI__sync_and_and_fetch_1: 229 case Builtin::BI__sync_and_and_fetch_2: 230 case Builtin::BI__sync_and_and_fetch_4: 231 case Builtin::BI__sync_and_and_fetch_8: 232 case Builtin::BI__sync_and_and_fetch_16: 233 case Builtin::BI__sync_or_and_fetch: 234 case Builtin::BI__sync_or_and_fetch_1: 235 case Builtin::BI__sync_or_and_fetch_2: 236 case Builtin::BI__sync_or_and_fetch_4: 237 case Builtin::BI__sync_or_and_fetch_8: 238 case Builtin::BI__sync_or_and_fetch_16: 239 case Builtin::BI__sync_xor_and_fetch: 240 case Builtin::BI__sync_xor_and_fetch_1: 241 case Builtin::BI__sync_xor_and_fetch_2: 242 case Builtin::BI__sync_xor_and_fetch_4: 243 case Builtin::BI__sync_xor_and_fetch_8: 244 case Builtin::BI__sync_xor_and_fetch_16: 245 case Builtin::BI__sync_val_compare_and_swap: 246 case Builtin::BI__sync_val_compare_and_swap_1: 247 case Builtin::BI__sync_val_compare_and_swap_2: 248 case Builtin::BI__sync_val_compare_and_swap_4: 249 case Builtin::BI__sync_val_compare_and_swap_8: 250 case Builtin::BI__sync_val_compare_and_swap_16: 251 case Builtin::BI__sync_bool_compare_and_swap: 252 case Builtin::BI__sync_bool_compare_and_swap_1: 253 case Builtin::BI__sync_bool_compare_and_swap_2: 254 case Builtin::BI__sync_bool_compare_and_swap_4: 255 case Builtin::BI__sync_bool_compare_and_swap_8: 256 case Builtin::BI__sync_bool_compare_and_swap_16: 257 case Builtin::BI__sync_lock_test_and_set: 258 case Builtin::BI__sync_lock_test_and_set_1: 259 case Builtin::BI__sync_lock_test_and_set_2: 260 case Builtin::BI__sync_lock_test_and_set_4: 261 case Builtin::BI__sync_lock_test_and_set_8: 262 case Builtin::BI__sync_lock_test_and_set_16: 263 case Builtin::BI__sync_lock_release: 264 case Builtin::BI__sync_lock_release_1: 265 case Builtin::BI__sync_lock_release_2: 266 case Builtin::BI__sync_lock_release_4: 267 case Builtin::BI__sync_lock_release_8: 268 case Builtin::BI__sync_lock_release_16: 269 case Builtin::BI__sync_swap: 270 case Builtin::BI__sync_swap_1: 271 case Builtin::BI__sync_swap_2: 272 case Builtin::BI__sync_swap_4: 273 case Builtin::BI__sync_swap_8: 274 case Builtin::BI__sync_swap_16: 275 return SemaBuiltinAtomicOverloaded(move(TheCallResult)); 276 case Builtin::BI__atomic_load: 277 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Load); 278 case Builtin::BI__atomic_store: 279 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Store); 280 case Builtin::BI__atomic_exchange: 281 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Xchg); 282 case Builtin::BI__atomic_compare_exchange_strong: 283 return SemaAtomicOpsOverloaded(move(TheCallResult), 284 AtomicExpr::CmpXchgStrong); 285 case Builtin::BI__atomic_compare_exchange_weak: 286 return SemaAtomicOpsOverloaded(move(TheCallResult), 287 AtomicExpr::CmpXchgWeak); 288 case Builtin::BI__atomic_fetch_add: 289 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Add); 290 case Builtin::BI__atomic_fetch_sub: 291 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Sub); 292 case Builtin::BI__atomic_fetch_and: 293 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::And); 294 case Builtin::BI__atomic_fetch_or: 295 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Or); 296 case Builtin::BI__atomic_fetch_xor: 297 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Xor); 298 case Builtin::BI__builtin_annotation: 299 if (CheckBuiltinAnnotationString(*this, TheCall->getArg(1))) 300 return ExprError(); 301 break; 302 } 303 304 // Since the target specific builtins for each arch overlap, only check those 305 // of the arch we are compiling for. 306 if (BuiltinID >= Builtin::FirstTSBuiltin) { 307 switch (Context.getTargetInfo().getTriple().getArch()) { 308 case llvm::Triple::arm: 309 case llvm::Triple::thumb: 310 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 311 return ExprError(); 312 break; 313 default: 314 break; 315 } 316 } 317 318 return move(TheCallResult); 319} 320 321// Get the valid immediate range for the specified NEON type code. 322static unsigned RFT(unsigned t, bool shift = false) { 323 NeonTypeFlags Type(t); 324 int IsQuad = Type.isQuad(); 325 switch (Type.getEltType()) { 326 case NeonTypeFlags::Int8: 327 case NeonTypeFlags::Poly8: 328 return shift ? 7 : (8 << IsQuad) - 1; 329 case NeonTypeFlags::Int16: 330 case NeonTypeFlags::Poly16: 331 return shift ? 15 : (4 << IsQuad) - 1; 332 case NeonTypeFlags::Int32: 333 return shift ? 31 : (2 << IsQuad) - 1; 334 case NeonTypeFlags::Int64: 335 return shift ? 63 : (1 << IsQuad) - 1; 336 case NeonTypeFlags::Float16: 337 assert(!shift && "cannot shift float types!"); 338 return (4 << IsQuad) - 1; 339 case NeonTypeFlags::Float32: 340 assert(!shift && "cannot shift float types!"); 341 return (2 << IsQuad) - 1; 342 } 343 return 0; 344} 345 346/// getNeonEltType - Return the QualType corresponding to the elements of 347/// the vector type specified by the NeonTypeFlags. This is used to check 348/// the pointer arguments for Neon load/store intrinsics. 349static QualType getNeonEltType(NeonTypeFlags Flags, ASTContext &Context) { 350 switch (Flags.getEltType()) { 351 case NeonTypeFlags::Int8: 352 return Flags.isUnsigned() ? Context.UnsignedCharTy : Context.SignedCharTy; 353 case NeonTypeFlags::Int16: 354 return Flags.isUnsigned() ? Context.UnsignedShortTy : Context.ShortTy; 355 case NeonTypeFlags::Int32: 356 return Flags.isUnsigned() ? Context.UnsignedIntTy : Context.IntTy; 357 case NeonTypeFlags::Int64: 358 return Flags.isUnsigned() ? Context.UnsignedLongLongTy : Context.LongLongTy; 359 case NeonTypeFlags::Poly8: 360 return Context.SignedCharTy; 361 case NeonTypeFlags::Poly16: 362 return Context.ShortTy; 363 case NeonTypeFlags::Float16: 364 return Context.UnsignedShortTy; 365 case NeonTypeFlags::Float32: 366 return Context.FloatTy; 367 } 368 return QualType(); 369} 370 371bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 372 llvm::APSInt Result; 373 374 unsigned mask = 0; 375 unsigned TV = 0; 376 int PtrArgNum = -1; 377 bool HasConstPtr = false; 378 switch (BuiltinID) { 379#define GET_NEON_OVERLOAD_CHECK 380#include "clang/Basic/arm_neon.inc" 381#undef GET_NEON_OVERLOAD_CHECK 382 } 383 384 // For NEON intrinsics which are overloaded on vector element type, validate 385 // the immediate which specifies which variant to emit. 386 unsigned ImmArg = TheCall->getNumArgs()-1; 387 if (mask) { 388 if (SemaBuiltinConstantArg(TheCall, ImmArg, Result)) 389 return true; 390 391 TV = Result.getLimitedValue(64); 392 if ((TV > 63) || (mask & (1 << TV)) == 0) 393 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 394 << TheCall->getArg(ImmArg)->getSourceRange(); 395 } 396 397 if (PtrArgNum >= 0) { 398 // Check that pointer arguments have the specified type. 399 Expr *Arg = TheCall->getArg(PtrArgNum); 400 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) 401 Arg = ICE->getSubExpr(); 402 ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg); 403 QualType RHSTy = RHS.get()->getType(); 404 QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context); 405 if (HasConstPtr) 406 EltTy = EltTy.withConst(); 407 QualType LHSTy = Context.getPointerType(EltTy); 408 AssignConvertType ConvTy; 409 ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS); 410 if (RHS.isInvalid()) 411 return true; 412 if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy, 413 RHS.get(), AA_Assigning)) 414 return true; 415 } 416 417 // For NEON intrinsics which take an immediate value as part of the 418 // instruction, range check them here. 419 unsigned i = 0, l = 0, u = 0; 420 switch (BuiltinID) { 421 default: return false; 422 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 423 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 424 case ARM::BI__builtin_arm_vcvtr_f: 425 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 426#define GET_NEON_IMMEDIATE_CHECK 427#include "clang/Basic/arm_neon.inc" 428#undef GET_NEON_IMMEDIATE_CHECK 429 }; 430 431 // Check that the immediate argument is actually a constant. 432 if (SemaBuiltinConstantArg(TheCall, i, Result)) 433 return true; 434 435 // Range check against the upper/lower values for this isntruction. 436 unsigned Val = Result.getZExtValue(); 437 if (Val < l || Val > (u + l)) 438 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 439 << l << u+l << TheCall->getArg(i)->getSourceRange(); 440 441 // FIXME: VFP Intrinsics should error if VFP not present. 442 return false; 443} 444 445/// CheckFunctionCall - Check a direct function call for various correctness 446/// and safety properties not strictly enforced by the C type system. 447bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 448 // Get the IdentifierInfo* for the called function. 449 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 450 451 // None of the checks below are needed for functions that don't have 452 // simple names (e.g., C++ conversion functions). 453 if (!FnInfo) 454 return false; 455 456 // FIXME: This mechanism should be abstracted to be less fragile and 457 // more efficient. For example, just map function ids to custom 458 // handlers. 459 460 // Printf and scanf checking. 461 for (specific_attr_iterator<FormatAttr> 462 i = FDecl->specific_attr_begin<FormatAttr>(), 463 e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { 464 465 const FormatAttr *Format = *i; 466 const bool b = Format->getType() == "scanf"; 467 if (b || CheckablePrintfAttr(Format, TheCall)) { 468 bool HasVAListArg = Format->getFirstArg() == 0; 469 CheckPrintfScanfArguments(TheCall, HasVAListArg, 470 Format->getFormatIdx() - 1, 471 HasVAListArg ? 0 : Format->getFirstArg() - 1, 472 !b); 473 } 474 } 475 476 for (specific_attr_iterator<NonNullAttr> 477 i = FDecl->specific_attr_begin<NonNullAttr>(), 478 e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) { 479 CheckNonNullArguments(*i, TheCall->getArgs(), 480 TheCall->getCallee()->getLocStart()); 481 } 482 483 FunctionDecl::MemoryFunctionKind CMF = FDecl->getMemoryFunctionKind(); 484 if (CMF == FunctionDecl::MFK_Invalid) 485 return false; 486 487 // Handle memory setting and copying functions. 488 if (CMF == FunctionDecl::MFK_Strlcpy || CMF == FunctionDecl::MFK_Strlcat) 489 CheckStrlcpycatArguments(TheCall, FnInfo); 490 else 491 CheckMemaccessArguments(TheCall, CMF, FnInfo); 492 493 return false; 494} 495 496bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 497 // Printf checking. 498 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 499 if (!Format) 500 return false; 501 502 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 503 if (!V) 504 return false; 505 506 QualType Ty = V->getType(); 507 if (!Ty->isBlockPointerType()) 508 return false; 509 510 const bool b = Format->getType() == "scanf"; 511 if (!b && !CheckablePrintfAttr(Format, TheCall)) 512 return false; 513 514 bool HasVAListArg = Format->getFirstArg() == 0; 515 CheckPrintfScanfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 516 HasVAListArg ? 0 : Format->getFirstArg() - 1, !b); 517 518 return false; 519} 520 521ExprResult 522Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, AtomicExpr::AtomicOp Op) { 523 CallExpr *TheCall = cast<CallExpr>(TheCallResult.get()); 524 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 525 526 // All these operations take one of the following four forms: 527 // T __atomic_load(_Atomic(T)*, int) (loads) 528 // T* __atomic_add(_Atomic(T*)*, ptrdiff_t, int) (pointer add/sub) 529 // int __atomic_compare_exchange_strong(_Atomic(T)*, T*, T, int, int) 530 // (cmpxchg) 531 // T __atomic_exchange(_Atomic(T)*, T, int) (everything else) 532 // where T is an appropriate type, and the int paremeterss are for orderings. 533 unsigned NumVals = 1; 534 unsigned NumOrders = 1; 535 if (Op == AtomicExpr::Load) { 536 NumVals = 0; 537 } else if (Op == AtomicExpr::CmpXchgWeak || Op == AtomicExpr::CmpXchgStrong) { 538 NumVals = 2; 539 NumOrders = 2; 540 } 541 542 if (TheCall->getNumArgs() < NumVals+NumOrders+1) { 543 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 544 << 0 << NumVals+NumOrders+1 << TheCall->getNumArgs() 545 << TheCall->getCallee()->getSourceRange(); 546 return ExprError(); 547 } else if (TheCall->getNumArgs() > NumVals+NumOrders+1) { 548 Diag(TheCall->getArg(NumVals+NumOrders+1)->getLocStart(), 549 diag::err_typecheck_call_too_many_args) 550 << 0 << NumVals+NumOrders+1 << TheCall->getNumArgs() 551 << TheCall->getCallee()->getSourceRange(); 552 return ExprError(); 553 } 554 555 // Inspect the first argument of the atomic operation. This should always be 556 // a pointer to an _Atomic type. 557 Expr *Ptr = TheCall->getArg(0); 558 Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get(); 559 const PointerType *pointerType = Ptr->getType()->getAs<PointerType>(); 560 if (!pointerType) { 561 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 562 << Ptr->getType() << Ptr->getSourceRange(); 563 return ExprError(); 564 } 565 566 QualType AtomTy = pointerType->getPointeeType(); 567 if (!AtomTy->isAtomicType()) { 568 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 569 << Ptr->getType() << Ptr->getSourceRange(); 570 return ExprError(); 571 } 572 QualType ValType = AtomTy->getAs<AtomicType>()->getValueType(); 573 574 if ((Op == AtomicExpr::Add || Op == AtomicExpr::Sub) && 575 !ValType->isIntegerType() && !ValType->isPointerType()) { 576 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 577 << Ptr->getType() << Ptr->getSourceRange(); 578 return ExprError(); 579 } 580 581 if (!ValType->isIntegerType() && 582 (Op == AtomicExpr::And || Op == AtomicExpr::Or || Op == AtomicExpr::Xor)){ 583 Diag(DRE->getLocStart(), diag::err_atomic_op_logical_needs_atomic_int) 584 << Ptr->getType() << Ptr->getSourceRange(); 585 return ExprError(); 586 } 587 588 switch (ValType.getObjCLifetime()) { 589 case Qualifiers::OCL_None: 590 case Qualifiers::OCL_ExplicitNone: 591 // okay 592 break; 593 594 case Qualifiers::OCL_Weak: 595 case Qualifiers::OCL_Strong: 596 case Qualifiers::OCL_Autoreleasing: 597 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 598 << ValType << Ptr->getSourceRange(); 599 return ExprError(); 600 } 601 602 QualType ResultType = ValType; 603 if (Op == AtomicExpr::Store) 604 ResultType = Context.VoidTy; 605 else if (Op == AtomicExpr::CmpXchgWeak || Op == AtomicExpr::CmpXchgStrong) 606 ResultType = Context.BoolTy; 607 608 // The first argument --- the pointer --- has a fixed type; we 609 // deduce the types of the rest of the arguments accordingly. Walk 610 // the remaining arguments, converting them to the deduced value type. 611 for (unsigned i = 1; i != NumVals+NumOrders+1; ++i) { 612 ExprResult Arg = TheCall->getArg(i); 613 QualType Ty; 614 if (i < NumVals+1) { 615 // The second argument to a cmpxchg is a pointer to the data which will 616 // be exchanged. The second argument to a pointer add/subtract is the 617 // amount to add/subtract, which must be a ptrdiff_t. The third 618 // argument to a cmpxchg and the second argument in all other cases 619 // is the type of the value. 620 if (i == 1 && (Op == AtomicExpr::CmpXchgWeak || 621 Op == AtomicExpr::CmpXchgStrong)) 622 Ty = Context.getPointerType(ValType.getUnqualifiedType()); 623 else if (!ValType->isIntegerType() && 624 (Op == AtomicExpr::Add || Op == AtomicExpr::Sub)) 625 Ty = Context.getPointerDiffType(); 626 else 627 Ty = ValType; 628 } else { 629 // The order(s) are always converted to int. 630 Ty = Context.IntTy; 631 } 632 InitializedEntity Entity = 633 InitializedEntity::InitializeParameter(Context, Ty, false); 634 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 635 if (Arg.isInvalid()) 636 return true; 637 TheCall->setArg(i, Arg.get()); 638 } 639 640 SmallVector<Expr*, 5> SubExprs; 641 SubExprs.push_back(Ptr); 642 if (Op == AtomicExpr::Load) { 643 SubExprs.push_back(TheCall->getArg(1)); // Order 644 } else if (Op != AtomicExpr::CmpXchgWeak && Op != AtomicExpr::CmpXchgStrong) { 645 SubExprs.push_back(TheCall->getArg(2)); // Order 646 SubExprs.push_back(TheCall->getArg(1)); // Val1 647 } else { 648 SubExprs.push_back(TheCall->getArg(3)); // Order 649 SubExprs.push_back(TheCall->getArg(1)); // Val1 650 SubExprs.push_back(TheCall->getArg(2)); // Val2 651 SubExprs.push_back(TheCall->getArg(4)); // OrderFail 652 } 653 654 return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 655 SubExprs.data(), SubExprs.size(), 656 ResultType, Op, 657 TheCall->getRParenLoc())); 658} 659 660 661/// checkBuiltinArgument - Given a call to a builtin function, perform 662/// normal type-checking on the given argument, updating the call in 663/// place. This is useful when a builtin function requires custom 664/// type-checking for some of its arguments but not necessarily all of 665/// them. 666/// 667/// Returns true on error. 668static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 669 FunctionDecl *Fn = E->getDirectCallee(); 670 assert(Fn && "builtin call without direct callee!"); 671 672 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 673 InitializedEntity Entity = 674 InitializedEntity::InitializeParameter(S.Context, Param); 675 676 ExprResult Arg = E->getArg(0); 677 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 678 if (Arg.isInvalid()) 679 return true; 680 681 E->setArg(ArgIndex, Arg.take()); 682 return false; 683} 684 685/// SemaBuiltinAtomicOverloaded - We have a call to a function like 686/// __sync_fetch_and_add, which is an overloaded function based on the pointer 687/// type of its first argument. The main ActOnCallExpr routines have already 688/// promoted the types of arguments because all of these calls are prototyped as 689/// void(...). 690/// 691/// This function goes through and does final semantic checking for these 692/// builtins, 693ExprResult 694Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 695 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 696 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 697 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 698 699 // Ensure that we have at least one argument to do type inference from. 700 if (TheCall->getNumArgs() < 1) { 701 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 702 << 0 << 1 << TheCall->getNumArgs() 703 << TheCall->getCallee()->getSourceRange(); 704 return ExprError(); 705 } 706 707 // Inspect the first argument of the atomic builtin. This should always be 708 // a pointer type, whose element is an integral scalar or pointer type. 709 // Because it is a pointer type, we don't have to worry about any implicit 710 // casts here. 711 // FIXME: We don't allow floating point scalars as input. 712 Expr *FirstArg = TheCall->getArg(0); 713 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 714 if (!pointerType) { 715 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 716 << FirstArg->getType() << FirstArg->getSourceRange(); 717 return ExprError(); 718 } 719 720 QualType ValType = pointerType->getPointeeType(); 721 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 722 !ValType->isBlockPointerType()) { 723 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 724 << FirstArg->getType() << FirstArg->getSourceRange(); 725 return ExprError(); 726 } 727 728 switch (ValType.getObjCLifetime()) { 729 case Qualifiers::OCL_None: 730 case Qualifiers::OCL_ExplicitNone: 731 // okay 732 break; 733 734 case Qualifiers::OCL_Weak: 735 case Qualifiers::OCL_Strong: 736 case Qualifiers::OCL_Autoreleasing: 737 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 738 << ValType << FirstArg->getSourceRange(); 739 return ExprError(); 740 } 741 742 // Strip any qualifiers off ValType. 743 ValType = ValType.getUnqualifiedType(); 744 745 // The majority of builtins return a value, but a few have special return 746 // types, so allow them to override appropriately below. 747 QualType ResultType = ValType; 748 749 // We need to figure out which concrete builtin this maps onto. For example, 750 // __sync_fetch_and_add with a 2 byte object turns into 751 // __sync_fetch_and_add_2. 752#define BUILTIN_ROW(x) \ 753 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 754 Builtin::BI##x##_8, Builtin::BI##x##_16 } 755 756 static const unsigned BuiltinIndices[][5] = { 757 BUILTIN_ROW(__sync_fetch_and_add), 758 BUILTIN_ROW(__sync_fetch_and_sub), 759 BUILTIN_ROW(__sync_fetch_and_or), 760 BUILTIN_ROW(__sync_fetch_and_and), 761 BUILTIN_ROW(__sync_fetch_and_xor), 762 763 BUILTIN_ROW(__sync_add_and_fetch), 764 BUILTIN_ROW(__sync_sub_and_fetch), 765 BUILTIN_ROW(__sync_and_and_fetch), 766 BUILTIN_ROW(__sync_or_and_fetch), 767 BUILTIN_ROW(__sync_xor_and_fetch), 768 769 BUILTIN_ROW(__sync_val_compare_and_swap), 770 BUILTIN_ROW(__sync_bool_compare_and_swap), 771 BUILTIN_ROW(__sync_lock_test_and_set), 772 BUILTIN_ROW(__sync_lock_release), 773 BUILTIN_ROW(__sync_swap) 774 }; 775#undef BUILTIN_ROW 776 777 // Determine the index of the size. 778 unsigned SizeIndex; 779 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 780 case 1: SizeIndex = 0; break; 781 case 2: SizeIndex = 1; break; 782 case 4: SizeIndex = 2; break; 783 case 8: SizeIndex = 3; break; 784 case 16: SizeIndex = 4; break; 785 default: 786 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 787 << FirstArg->getType() << FirstArg->getSourceRange(); 788 return ExprError(); 789 } 790 791 // Each of these builtins has one pointer argument, followed by some number of 792 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 793 // that we ignore. Find out which row of BuiltinIndices to read from as well 794 // as the number of fixed args. 795 unsigned BuiltinID = FDecl->getBuiltinID(); 796 unsigned BuiltinIndex, NumFixed = 1; 797 switch (BuiltinID) { 798 default: llvm_unreachable("Unknown overloaded atomic builtin!"); 799 case Builtin::BI__sync_fetch_and_add: 800 case Builtin::BI__sync_fetch_and_add_1: 801 case Builtin::BI__sync_fetch_and_add_2: 802 case Builtin::BI__sync_fetch_and_add_4: 803 case Builtin::BI__sync_fetch_and_add_8: 804 case Builtin::BI__sync_fetch_and_add_16: 805 BuiltinIndex = 0; 806 break; 807 808 case Builtin::BI__sync_fetch_and_sub: 809 case Builtin::BI__sync_fetch_and_sub_1: 810 case Builtin::BI__sync_fetch_and_sub_2: 811 case Builtin::BI__sync_fetch_and_sub_4: 812 case Builtin::BI__sync_fetch_and_sub_8: 813 case Builtin::BI__sync_fetch_and_sub_16: 814 BuiltinIndex = 1; 815 break; 816 817 case Builtin::BI__sync_fetch_and_or: 818 case Builtin::BI__sync_fetch_and_or_1: 819 case Builtin::BI__sync_fetch_and_or_2: 820 case Builtin::BI__sync_fetch_and_or_4: 821 case Builtin::BI__sync_fetch_and_or_8: 822 case Builtin::BI__sync_fetch_and_or_16: 823 BuiltinIndex = 2; 824 break; 825 826 case Builtin::BI__sync_fetch_and_and: 827 case Builtin::BI__sync_fetch_and_and_1: 828 case Builtin::BI__sync_fetch_and_and_2: 829 case Builtin::BI__sync_fetch_and_and_4: 830 case Builtin::BI__sync_fetch_and_and_8: 831 case Builtin::BI__sync_fetch_and_and_16: 832 BuiltinIndex = 3; 833 break; 834 835 case Builtin::BI__sync_fetch_and_xor: 836 case Builtin::BI__sync_fetch_and_xor_1: 837 case Builtin::BI__sync_fetch_and_xor_2: 838 case Builtin::BI__sync_fetch_and_xor_4: 839 case Builtin::BI__sync_fetch_and_xor_8: 840 case Builtin::BI__sync_fetch_and_xor_16: 841 BuiltinIndex = 4; 842 break; 843 844 case Builtin::BI__sync_add_and_fetch: 845 case Builtin::BI__sync_add_and_fetch_1: 846 case Builtin::BI__sync_add_and_fetch_2: 847 case Builtin::BI__sync_add_and_fetch_4: 848 case Builtin::BI__sync_add_and_fetch_8: 849 case Builtin::BI__sync_add_and_fetch_16: 850 BuiltinIndex = 5; 851 break; 852 853 case Builtin::BI__sync_sub_and_fetch: 854 case Builtin::BI__sync_sub_and_fetch_1: 855 case Builtin::BI__sync_sub_and_fetch_2: 856 case Builtin::BI__sync_sub_and_fetch_4: 857 case Builtin::BI__sync_sub_and_fetch_8: 858 case Builtin::BI__sync_sub_and_fetch_16: 859 BuiltinIndex = 6; 860 break; 861 862 case Builtin::BI__sync_and_and_fetch: 863 case Builtin::BI__sync_and_and_fetch_1: 864 case Builtin::BI__sync_and_and_fetch_2: 865 case Builtin::BI__sync_and_and_fetch_4: 866 case Builtin::BI__sync_and_and_fetch_8: 867 case Builtin::BI__sync_and_and_fetch_16: 868 BuiltinIndex = 7; 869 break; 870 871 case Builtin::BI__sync_or_and_fetch: 872 case Builtin::BI__sync_or_and_fetch_1: 873 case Builtin::BI__sync_or_and_fetch_2: 874 case Builtin::BI__sync_or_and_fetch_4: 875 case Builtin::BI__sync_or_and_fetch_8: 876 case Builtin::BI__sync_or_and_fetch_16: 877 BuiltinIndex = 8; 878 break; 879 880 case Builtin::BI__sync_xor_and_fetch: 881 case Builtin::BI__sync_xor_and_fetch_1: 882 case Builtin::BI__sync_xor_and_fetch_2: 883 case Builtin::BI__sync_xor_and_fetch_4: 884 case Builtin::BI__sync_xor_and_fetch_8: 885 case Builtin::BI__sync_xor_and_fetch_16: 886 BuiltinIndex = 9; 887 break; 888 889 case Builtin::BI__sync_val_compare_and_swap: 890 case Builtin::BI__sync_val_compare_and_swap_1: 891 case Builtin::BI__sync_val_compare_and_swap_2: 892 case Builtin::BI__sync_val_compare_and_swap_4: 893 case Builtin::BI__sync_val_compare_and_swap_8: 894 case Builtin::BI__sync_val_compare_and_swap_16: 895 BuiltinIndex = 10; 896 NumFixed = 2; 897 break; 898 899 case Builtin::BI__sync_bool_compare_and_swap: 900 case Builtin::BI__sync_bool_compare_and_swap_1: 901 case Builtin::BI__sync_bool_compare_and_swap_2: 902 case Builtin::BI__sync_bool_compare_and_swap_4: 903 case Builtin::BI__sync_bool_compare_and_swap_8: 904 case Builtin::BI__sync_bool_compare_and_swap_16: 905 BuiltinIndex = 11; 906 NumFixed = 2; 907 ResultType = Context.BoolTy; 908 break; 909 910 case Builtin::BI__sync_lock_test_and_set: 911 case Builtin::BI__sync_lock_test_and_set_1: 912 case Builtin::BI__sync_lock_test_and_set_2: 913 case Builtin::BI__sync_lock_test_and_set_4: 914 case Builtin::BI__sync_lock_test_and_set_8: 915 case Builtin::BI__sync_lock_test_and_set_16: 916 BuiltinIndex = 12; 917 break; 918 919 case Builtin::BI__sync_lock_release: 920 case Builtin::BI__sync_lock_release_1: 921 case Builtin::BI__sync_lock_release_2: 922 case Builtin::BI__sync_lock_release_4: 923 case Builtin::BI__sync_lock_release_8: 924 case Builtin::BI__sync_lock_release_16: 925 BuiltinIndex = 13; 926 NumFixed = 0; 927 ResultType = Context.VoidTy; 928 break; 929 930 case Builtin::BI__sync_swap: 931 case Builtin::BI__sync_swap_1: 932 case Builtin::BI__sync_swap_2: 933 case Builtin::BI__sync_swap_4: 934 case Builtin::BI__sync_swap_8: 935 case Builtin::BI__sync_swap_16: 936 BuiltinIndex = 14; 937 break; 938 } 939 940 // Now that we know how many fixed arguments we expect, first check that we 941 // have at least that many. 942 if (TheCall->getNumArgs() < 1+NumFixed) { 943 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 944 << 0 << 1+NumFixed << TheCall->getNumArgs() 945 << TheCall->getCallee()->getSourceRange(); 946 return ExprError(); 947 } 948 949 // Get the decl for the concrete builtin from this, we can tell what the 950 // concrete integer type we should convert to is. 951 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 952 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 953 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 954 FunctionDecl *NewBuiltinDecl = 955 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 956 TUScope, false, DRE->getLocStart())); 957 958 // The first argument --- the pointer --- has a fixed type; we 959 // deduce the types of the rest of the arguments accordingly. Walk 960 // the remaining arguments, converting them to the deduced value type. 961 for (unsigned i = 0; i != NumFixed; ++i) { 962 ExprResult Arg = TheCall->getArg(i+1); 963 964 // GCC does an implicit conversion to the pointer or integer ValType. This 965 // can fail in some cases (1i -> int**), check for this error case now. 966 // Initialize the argument. 967 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context, 968 ValType, /*consume*/ false); 969 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 970 if (Arg.isInvalid()) 971 return ExprError(); 972 973 // Okay, we have something that *can* be converted to the right type. Check 974 // to see if there is a potentially weird extension going on here. This can 975 // happen when you do an atomic operation on something like an char* and 976 // pass in 42. The 42 gets converted to char. This is even more strange 977 // for things like 45.123 -> char, etc. 978 // FIXME: Do this check. 979 TheCall->setArg(i+1, Arg.take()); 980 } 981 982 ASTContext& Context = this->getASTContext(); 983 984 // Create a new DeclRefExpr to refer to the new decl. 985 DeclRefExpr* NewDRE = DeclRefExpr::Create( 986 Context, 987 DRE->getQualifierLoc(), 988 NewBuiltinDecl, 989 DRE->getLocation(), 990 NewBuiltinDecl->getType(), 991 DRE->getValueKind()); 992 993 // Set the callee in the CallExpr. 994 // FIXME: This leaks the original parens and implicit casts. 995 ExprResult PromotedCall = UsualUnaryConversions(NewDRE); 996 if (PromotedCall.isInvalid()) 997 return ExprError(); 998 TheCall->setCallee(PromotedCall.take()); 999 1000 // Change the result type of the call to match the original value type. This 1001 // is arbitrary, but the codegen for these builtins ins design to handle it 1002 // gracefully. 1003 TheCall->setType(ResultType); 1004 1005 return move(TheCallResult); 1006} 1007 1008/// CheckObjCString - Checks that the argument to the builtin 1009/// CFString constructor is correct 1010/// Note: It might also make sense to do the UTF-16 conversion here (would 1011/// simplify the backend). 1012bool Sema::CheckObjCString(Expr *Arg) { 1013 Arg = Arg->IgnoreParenCasts(); 1014 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 1015 1016 if (!Literal || !Literal->isAscii()) { 1017 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 1018 << Arg->getSourceRange(); 1019 return true; 1020 } 1021 1022 if (Literal->containsNonAsciiOrNull()) { 1023 StringRef String = Literal->getString(); 1024 unsigned NumBytes = String.size(); 1025 SmallVector<UTF16, 128> ToBuf(NumBytes); 1026 const UTF8 *FromPtr = (UTF8 *)String.data(); 1027 UTF16 *ToPtr = &ToBuf[0]; 1028 1029 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 1030 &ToPtr, ToPtr + NumBytes, 1031 strictConversion); 1032 // Check for conversion failure. 1033 if (Result != conversionOK) 1034 Diag(Arg->getLocStart(), 1035 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 1036 } 1037 return false; 1038} 1039 1040/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 1041/// Emit an error and return true on failure, return false on success. 1042bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 1043 Expr *Fn = TheCall->getCallee(); 1044 if (TheCall->getNumArgs() > 2) { 1045 Diag(TheCall->getArg(2)->getLocStart(), 1046 diag::err_typecheck_call_too_many_args) 1047 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1048 << Fn->getSourceRange() 1049 << SourceRange(TheCall->getArg(2)->getLocStart(), 1050 (*(TheCall->arg_end()-1))->getLocEnd()); 1051 return true; 1052 } 1053 1054 if (TheCall->getNumArgs() < 2) { 1055 return Diag(TheCall->getLocEnd(), 1056 diag::err_typecheck_call_too_few_args_at_least) 1057 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 1058 } 1059 1060 // Type-check the first argument normally. 1061 if (checkBuiltinArgument(*this, TheCall, 0)) 1062 return true; 1063 1064 // Determine whether the current function is variadic or not. 1065 BlockScopeInfo *CurBlock = getCurBlock(); 1066 bool isVariadic; 1067 if (CurBlock) 1068 isVariadic = CurBlock->TheDecl->isVariadic(); 1069 else if (FunctionDecl *FD = getCurFunctionDecl()) 1070 isVariadic = FD->isVariadic(); 1071 else 1072 isVariadic = getCurMethodDecl()->isVariadic(); 1073 1074 if (!isVariadic) { 1075 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 1076 return true; 1077 } 1078 1079 // Verify that the second argument to the builtin is the last argument of the 1080 // current function or method. 1081 bool SecondArgIsLastNamedArgument = false; 1082 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 1083 1084 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 1085 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 1086 // FIXME: This isn't correct for methods (results in bogus warning). 1087 // Get the last formal in the current function. 1088 const ParmVarDecl *LastArg; 1089 if (CurBlock) 1090 LastArg = *(CurBlock->TheDecl->param_end()-1); 1091 else if (FunctionDecl *FD = getCurFunctionDecl()) 1092 LastArg = *(FD->param_end()-1); 1093 else 1094 LastArg = *(getCurMethodDecl()->param_end()-1); 1095 SecondArgIsLastNamedArgument = PV == LastArg; 1096 } 1097 } 1098 1099 if (!SecondArgIsLastNamedArgument) 1100 Diag(TheCall->getArg(1)->getLocStart(), 1101 diag::warn_second_parameter_of_va_start_not_last_named_argument); 1102 return false; 1103} 1104 1105/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 1106/// friends. This is declared to take (...), so we have to check everything. 1107bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 1108 if (TheCall->getNumArgs() < 2) 1109 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1110 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 1111 if (TheCall->getNumArgs() > 2) 1112 return Diag(TheCall->getArg(2)->getLocStart(), 1113 diag::err_typecheck_call_too_many_args) 1114 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1115 << SourceRange(TheCall->getArg(2)->getLocStart(), 1116 (*(TheCall->arg_end()-1))->getLocEnd()); 1117 1118 ExprResult OrigArg0 = TheCall->getArg(0); 1119 ExprResult OrigArg1 = TheCall->getArg(1); 1120 1121 // Do standard promotions between the two arguments, returning their common 1122 // type. 1123 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 1124 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 1125 return true; 1126 1127 // Make sure any conversions are pushed back into the call; this is 1128 // type safe since unordered compare builtins are declared as "_Bool 1129 // foo(...)". 1130 TheCall->setArg(0, OrigArg0.get()); 1131 TheCall->setArg(1, OrigArg1.get()); 1132 1133 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 1134 return false; 1135 1136 // If the common type isn't a real floating type, then the arguments were 1137 // invalid for this operation. 1138 if (!Res->isRealFloatingType()) 1139 return Diag(OrigArg0.get()->getLocStart(), 1140 diag::err_typecheck_call_invalid_ordered_compare) 1141 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 1142 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 1143 1144 return false; 1145} 1146 1147/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 1148/// __builtin_isnan and friends. This is declared to take (...), so we have 1149/// to check everything. We expect the last argument to be a floating point 1150/// value. 1151bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 1152 if (TheCall->getNumArgs() < NumArgs) 1153 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1154 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 1155 if (TheCall->getNumArgs() > NumArgs) 1156 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 1157 diag::err_typecheck_call_too_many_args) 1158 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 1159 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 1160 (*(TheCall->arg_end()-1))->getLocEnd()); 1161 1162 Expr *OrigArg = TheCall->getArg(NumArgs-1); 1163 1164 if (OrigArg->isTypeDependent()) 1165 return false; 1166 1167 // This operation requires a non-_Complex floating-point number. 1168 if (!OrigArg->getType()->isRealFloatingType()) 1169 return Diag(OrigArg->getLocStart(), 1170 diag::err_typecheck_call_invalid_unary_fp) 1171 << OrigArg->getType() << OrigArg->getSourceRange(); 1172 1173 // If this is an implicit conversion from float -> double, remove it. 1174 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 1175 Expr *CastArg = Cast->getSubExpr(); 1176 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 1177 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 1178 "promotion from float to double is the only expected cast here"); 1179 Cast->setSubExpr(0); 1180 TheCall->setArg(NumArgs-1, CastArg); 1181 OrigArg = CastArg; 1182 } 1183 } 1184 1185 return false; 1186} 1187 1188/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1189// This is declared to take (...), so we have to check everything. 1190ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1191 if (TheCall->getNumArgs() < 2) 1192 return ExprError(Diag(TheCall->getLocEnd(), 1193 diag::err_typecheck_call_too_few_args_at_least) 1194 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1195 << TheCall->getSourceRange()); 1196 1197 // Determine which of the following types of shufflevector we're checking: 1198 // 1) unary, vector mask: (lhs, mask) 1199 // 2) binary, vector mask: (lhs, rhs, mask) 1200 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1201 QualType resType = TheCall->getArg(0)->getType(); 1202 unsigned numElements = 0; 1203 1204 if (!TheCall->getArg(0)->isTypeDependent() && 1205 !TheCall->getArg(1)->isTypeDependent()) { 1206 QualType LHSType = TheCall->getArg(0)->getType(); 1207 QualType RHSType = TheCall->getArg(1)->getType(); 1208 1209 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 1210 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 1211 << SourceRange(TheCall->getArg(0)->getLocStart(), 1212 TheCall->getArg(1)->getLocEnd()); 1213 return ExprError(); 1214 } 1215 1216 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1217 unsigned numResElements = TheCall->getNumArgs() - 2; 1218 1219 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1220 // with mask. If so, verify that RHS is an integer vector type with the 1221 // same number of elts as lhs. 1222 if (TheCall->getNumArgs() == 2) { 1223 if (!RHSType->hasIntegerRepresentation() || 1224 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1225 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1226 << SourceRange(TheCall->getArg(1)->getLocStart(), 1227 TheCall->getArg(1)->getLocEnd()); 1228 numResElements = numElements; 1229 } 1230 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1231 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1232 << SourceRange(TheCall->getArg(0)->getLocStart(), 1233 TheCall->getArg(1)->getLocEnd()); 1234 return ExprError(); 1235 } else if (numElements != numResElements) { 1236 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1237 resType = Context.getVectorType(eltType, numResElements, 1238 VectorType::GenericVector); 1239 } 1240 } 1241 1242 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1243 if (TheCall->getArg(i)->isTypeDependent() || 1244 TheCall->getArg(i)->isValueDependent()) 1245 continue; 1246 1247 llvm::APSInt Result(32); 1248 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1249 return ExprError(Diag(TheCall->getLocStart(), 1250 diag::err_shufflevector_nonconstant_argument) 1251 << TheCall->getArg(i)->getSourceRange()); 1252 1253 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1254 return ExprError(Diag(TheCall->getLocStart(), 1255 diag::err_shufflevector_argument_too_large) 1256 << TheCall->getArg(i)->getSourceRange()); 1257 } 1258 1259 SmallVector<Expr*, 32> exprs; 1260 1261 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1262 exprs.push_back(TheCall->getArg(i)); 1263 TheCall->setArg(i, 0); 1264 } 1265 1266 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 1267 exprs.size(), resType, 1268 TheCall->getCallee()->getLocStart(), 1269 TheCall->getRParenLoc())); 1270} 1271 1272/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1273// This is declared to take (const void*, ...) and can take two 1274// optional constant int args. 1275bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1276 unsigned NumArgs = TheCall->getNumArgs(); 1277 1278 if (NumArgs > 3) 1279 return Diag(TheCall->getLocEnd(), 1280 diag::err_typecheck_call_too_many_args_at_most) 1281 << 0 /*function call*/ << 3 << NumArgs 1282 << TheCall->getSourceRange(); 1283 1284 // Argument 0 is checked for us and the remaining arguments must be 1285 // constant integers. 1286 for (unsigned i = 1; i != NumArgs; ++i) { 1287 Expr *Arg = TheCall->getArg(i); 1288 1289 llvm::APSInt Result; 1290 if (SemaBuiltinConstantArg(TheCall, i, Result)) 1291 return true; 1292 1293 // FIXME: gcc issues a warning and rewrites these to 0. These 1294 // seems especially odd for the third argument since the default 1295 // is 3. 1296 if (i == 1) { 1297 if (Result.getLimitedValue() > 1) 1298 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1299 << "0" << "1" << Arg->getSourceRange(); 1300 } else { 1301 if (Result.getLimitedValue() > 3) 1302 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1303 << "0" << "3" << Arg->getSourceRange(); 1304 } 1305 } 1306 1307 return false; 1308} 1309 1310/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1311/// TheCall is a constant expression. 1312bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1313 llvm::APSInt &Result) { 1314 Expr *Arg = TheCall->getArg(ArgNum); 1315 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1316 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1317 1318 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 1319 1320 if (!Arg->isIntegerConstantExpr(Result, Context)) 1321 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 1322 << FDecl->getDeclName() << Arg->getSourceRange(); 1323 1324 return false; 1325} 1326 1327/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 1328/// int type). This simply type checks that type is one of the defined 1329/// constants (0-3). 1330// For compatibility check 0-3, llvm only handles 0 and 2. 1331bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 1332 llvm::APSInt Result; 1333 1334 // Check constant-ness first. 1335 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1336 return true; 1337 1338 Expr *Arg = TheCall->getArg(1); 1339 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 1340 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1341 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1342 } 1343 1344 return false; 1345} 1346 1347/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1348/// This checks that val is a constant 1. 1349bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1350 Expr *Arg = TheCall->getArg(1); 1351 llvm::APSInt Result; 1352 1353 // TODO: This is less than ideal. Overload this to take a value. 1354 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1355 return true; 1356 1357 if (Result != 1) 1358 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1359 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1360 1361 return false; 1362} 1363 1364// Handle i > 1 ? "x" : "y", recursively. 1365bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 1366 bool HasVAListArg, 1367 unsigned format_idx, unsigned firstDataArg, 1368 bool isPrintf, bool inFunctionCall) { 1369 tryAgain: 1370 if (E->isTypeDependent() || E->isValueDependent()) 1371 return false; 1372 1373 E = E->IgnoreParens(); 1374 1375 switch (E->getStmtClass()) { 1376 case Stmt::BinaryConditionalOperatorClass: 1377 case Stmt::ConditionalOperatorClass: { 1378 const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E); 1379 return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, HasVAListArg, 1380 format_idx, firstDataArg, isPrintf, 1381 inFunctionCall) 1382 && SemaCheckStringLiteral(C->getFalseExpr(), TheCall, HasVAListArg, 1383 format_idx, firstDataArg, isPrintf, 1384 inFunctionCall); 1385 } 1386 1387 case Stmt::IntegerLiteralClass: 1388 // Technically -Wformat-nonliteral does not warn about this case. 1389 // The behavior of printf and friends in this case is implementation 1390 // dependent. Ideally if the format string cannot be null then 1391 // it should have a 'nonnull' attribute in the function prototype. 1392 return true; 1393 1394 case Stmt::ImplicitCastExprClass: { 1395 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 1396 goto tryAgain; 1397 } 1398 1399 case Stmt::OpaqueValueExprClass: 1400 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 1401 E = src; 1402 goto tryAgain; 1403 } 1404 return false; 1405 1406 case Stmt::PredefinedExprClass: 1407 // While __func__, etc., are technically not string literals, they 1408 // cannot contain format specifiers and thus are not a security 1409 // liability. 1410 return true; 1411 1412 case Stmt::DeclRefExprClass: { 1413 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1414 1415 // As an exception, do not flag errors for variables binding to 1416 // const string literals. 1417 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1418 bool isConstant = false; 1419 QualType T = DR->getType(); 1420 1421 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1422 isConstant = AT->getElementType().isConstant(Context); 1423 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1424 isConstant = T.isConstant(Context) && 1425 PT->getPointeeType().isConstant(Context); 1426 } 1427 1428 if (isConstant) { 1429 if (const Expr *Init = VD->getAnyInitializer()) 1430 return SemaCheckStringLiteral(Init, TheCall, 1431 HasVAListArg, format_idx, firstDataArg, 1432 isPrintf, /*inFunctionCall*/false); 1433 } 1434 1435 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1436 // special check to see if the format string is a function parameter 1437 // of the function calling the printf function. If the function 1438 // has an attribute indicating it is a printf-like function, then we 1439 // should suppress warnings concerning non-literals being used in a call 1440 // to a vprintf function. For example: 1441 // 1442 // void 1443 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1444 // va_list ap; 1445 // va_start(ap, fmt); 1446 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1447 // ... 1448 // 1449 // 1450 // FIXME: We don't have full attribute support yet, so just check to see 1451 // if the argument is a DeclRefExpr that references a parameter. We'll 1452 // add proper support for checking the attribute later. 1453 if (HasVAListArg) 1454 if (isa<ParmVarDecl>(VD)) 1455 return true; 1456 } 1457 1458 return false; 1459 } 1460 1461 case Stmt::CallExprClass: { 1462 const CallExpr *CE = cast<CallExpr>(E); 1463 if (const ImplicitCastExpr *ICE 1464 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 1465 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 1466 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 1467 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 1468 unsigned ArgIndex = FA->getFormatIdx(); 1469 const Expr *Arg = CE->getArg(ArgIndex - 1); 1470 1471 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 1472 format_idx, firstDataArg, isPrintf, 1473 inFunctionCall); 1474 } 1475 } 1476 } 1477 } 1478 1479 return false; 1480 } 1481 case Stmt::ObjCStringLiteralClass: 1482 case Stmt::StringLiteralClass: { 1483 const StringLiteral *StrE = NULL; 1484 1485 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1486 StrE = ObjCFExpr->getString(); 1487 else 1488 StrE = cast<StringLiteral>(E); 1489 1490 if (StrE) { 1491 CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx, 1492 firstDataArg, isPrintf, inFunctionCall); 1493 return true; 1494 } 1495 1496 return false; 1497 } 1498 1499 default: 1500 return false; 1501 } 1502} 1503 1504void 1505Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1506 const Expr * const *ExprArgs, 1507 SourceLocation CallSiteLoc) { 1508 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1509 e = NonNull->args_end(); 1510 i != e; ++i) { 1511 const Expr *ArgExpr = ExprArgs[*i]; 1512 if (ArgExpr->isNullPointerConstant(Context, 1513 Expr::NPC_ValueDependentIsNotNull)) 1514 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 1515 } 1516} 1517 1518/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar 1519/// functions) for correct use of format strings. 1520void 1521Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg, 1522 unsigned format_idx, unsigned firstDataArg, 1523 bool isPrintf) { 1524 1525 const Expr *Fn = TheCall->getCallee(); 1526 1527 // The way the format attribute works in GCC, the implicit this argument 1528 // of member functions is counted. However, it doesn't appear in our own 1529 // lists, so decrement format_idx in that case. 1530 if (isa<CXXMemberCallExpr>(TheCall)) { 1531 const CXXMethodDecl *method_decl = 1532 dyn_cast<CXXMethodDecl>(TheCall->getCalleeDecl()); 1533 if (method_decl && method_decl->isInstance()) { 1534 // Catch a format attribute mistakenly referring to the object argument. 1535 if (format_idx == 0) 1536 return; 1537 --format_idx; 1538 if(firstDataArg != 0) 1539 --firstDataArg; 1540 } 1541 } 1542 1543 // CHECK: printf/scanf-like function is called with no format string. 1544 if (format_idx >= TheCall->getNumArgs()) { 1545 Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string) 1546 << Fn->getSourceRange(); 1547 return; 1548 } 1549 1550 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 1551 1552 // CHECK: format string is not a string literal. 1553 // 1554 // Dynamically generated format strings are difficult to 1555 // automatically vet at compile time. Requiring that format strings 1556 // are string literals: (1) permits the checking of format strings by 1557 // the compiler and thereby (2) can practically remove the source of 1558 // many format string exploits. 1559 1560 // Format string can be either ObjC string (e.g. @"%d") or 1561 // C string (e.g. "%d") 1562 // ObjC string uses the same format specifiers as C string, so we can use 1563 // the same format string checking logic for both ObjC and C strings. 1564 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 1565 firstDataArg, isPrintf)) 1566 return; // Literal format string found, check done! 1567 1568 // If there are no arguments specified, warn with -Wformat-security, otherwise 1569 // warn only with -Wformat-nonliteral. 1570 if (TheCall->getNumArgs() == format_idx+1) 1571 Diag(TheCall->getArg(format_idx)->getLocStart(), 1572 diag::warn_format_nonliteral_noargs) 1573 << OrigFormatExpr->getSourceRange(); 1574 else 1575 Diag(TheCall->getArg(format_idx)->getLocStart(), 1576 diag::warn_format_nonliteral) 1577 << OrigFormatExpr->getSourceRange(); 1578} 1579 1580namespace { 1581class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1582protected: 1583 Sema &S; 1584 const StringLiteral *FExpr; 1585 const Expr *OrigFormatExpr; 1586 const unsigned FirstDataArg; 1587 const unsigned NumDataArgs; 1588 const bool IsObjCLiteral; 1589 const char *Beg; // Start of format string. 1590 const bool HasVAListArg; 1591 const CallExpr *TheCall; 1592 unsigned FormatIdx; 1593 llvm::BitVector CoveredArgs; 1594 bool usesPositionalArgs; 1595 bool atFirstArg; 1596 bool inFunctionCall; 1597public: 1598 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1599 const Expr *origFormatExpr, unsigned firstDataArg, 1600 unsigned numDataArgs, bool isObjCLiteral, 1601 const char *beg, bool hasVAListArg, 1602 const CallExpr *theCall, unsigned formatIdx, 1603 bool inFunctionCall) 1604 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1605 FirstDataArg(firstDataArg), 1606 NumDataArgs(numDataArgs), 1607 IsObjCLiteral(isObjCLiteral), Beg(beg), 1608 HasVAListArg(hasVAListArg), 1609 TheCall(theCall), FormatIdx(formatIdx), 1610 usesPositionalArgs(false), atFirstArg(true), 1611 inFunctionCall(inFunctionCall) { 1612 CoveredArgs.resize(numDataArgs); 1613 CoveredArgs.reset(); 1614 } 1615 1616 void DoneProcessing(); 1617 1618 void HandleIncompleteSpecifier(const char *startSpecifier, 1619 unsigned specifierLen); 1620 1621 virtual void HandleInvalidPosition(const char *startSpecifier, 1622 unsigned specifierLen, 1623 analyze_format_string::PositionContext p); 1624 1625 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1626 1627 void HandleNullChar(const char *nullCharacter); 1628 1629 template <typename Range> 1630 static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall, 1631 const Expr *ArgumentExpr, 1632 PartialDiagnostic PDiag, 1633 SourceLocation StringLoc, 1634 bool IsStringLocation, Range StringRange, 1635 FixItHint Fixit = FixItHint()); 1636 1637protected: 1638 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1639 const char *startSpec, 1640 unsigned specifierLen, 1641 const char *csStart, unsigned csLen); 1642 1643 void HandlePositionalNonpositionalArgs(SourceLocation Loc, 1644 const char *startSpec, 1645 unsigned specifierLen); 1646 1647 SourceRange getFormatStringRange(); 1648 CharSourceRange getSpecifierRange(const char *startSpecifier, 1649 unsigned specifierLen); 1650 SourceLocation getLocationOfByte(const char *x); 1651 1652 const Expr *getDataArg(unsigned i) const; 1653 1654 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1655 const analyze_format_string::ConversionSpecifier &CS, 1656 const char *startSpecifier, unsigned specifierLen, 1657 unsigned argIndex); 1658 1659 template <typename Range> 1660 void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc, 1661 bool IsStringLocation, Range StringRange, 1662 FixItHint Fixit = FixItHint()); 1663 1664 void CheckPositionalAndNonpositionalArgs( 1665 const analyze_format_string::FormatSpecifier *FS); 1666}; 1667} 1668 1669SourceRange CheckFormatHandler::getFormatStringRange() { 1670 return OrigFormatExpr->getSourceRange(); 1671} 1672 1673CharSourceRange CheckFormatHandler:: 1674getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1675 SourceLocation Start = getLocationOfByte(startSpecifier); 1676 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1677 1678 // Advance the end SourceLocation by one due to half-open ranges. 1679 End = End.getLocWithOffset(1); 1680 1681 return CharSourceRange::getCharRange(Start, End); 1682} 1683 1684SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 1685 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1686} 1687 1688void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 1689 unsigned specifierLen){ 1690 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier), 1691 getLocationOfByte(startSpecifier), 1692 /*IsStringLocation*/true, 1693 getSpecifierRange(startSpecifier, specifierLen)); 1694} 1695 1696void 1697CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1698 analyze_format_string::PositionContext p) { 1699 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier) 1700 << (unsigned) p, 1701 getLocationOfByte(startPos), /*IsStringLocation*/true, 1702 getSpecifierRange(startPos, posLen)); 1703} 1704 1705void CheckFormatHandler::HandleZeroPosition(const char *startPos, 1706 unsigned posLen) { 1707 EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier), 1708 getLocationOfByte(startPos), 1709 /*IsStringLocation*/true, 1710 getSpecifierRange(startPos, posLen)); 1711} 1712 1713void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 1714 if (!IsObjCLiteral) { 1715 // The presence of a null character is likely an error. 1716 EmitFormatDiagnostic( 1717 S.PDiag(diag::warn_printf_format_string_contains_null_char), 1718 getLocationOfByte(nullCharacter), /*IsStringLocation*/true, 1719 getFormatStringRange()); 1720 } 1721} 1722 1723const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 1724 return TheCall->getArg(FirstDataArg + i); 1725} 1726 1727void CheckFormatHandler::DoneProcessing() { 1728 // Does the number of data arguments exceed the number of 1729 // format conversions in the format string? 1730 if (!HasVAListArg) { 1731 // Find any arguments that weren't covered. 1732 CoveredArgs.flip(); 1733 signed notCoveredArg = CoveredArgs.find_first(); 1734 if (notCoveredArg >= 0) { 1735 assert((unsigned)notCoveredArg < NumDataArgs); 1736 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used), 1737 getDataArg((unsigned) notCoveredArg)->getLocStart(), 1738 /*IsStringLocation*/false, getFormatStringRange()); 1739 } 1740 } 1741} 1742 1743bool 1744CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 1745 SourceLocation Loc, 1746 const char *startSpec, 1747 unsigned specifierLen, 1748 const char *csStart, 1749 unsigned csLen) { 1750 1751 bool keepGoing = true; 1752 if (argIndex < NumDataArgs) { 1753 // Consider the argument coverered, even though the specifier doesn't 1754 // make sense. 1755 CoveredArgs.set(argIndex); 1756 } 1757 else { 1758 // If argIndex exceeds the number of data arguments we 1759 // don't issue a warning because that is just a cascade of warnings (and 1760 // they may have intended '%%' anyway). We don't want to continue processing 1761 // the format string after this point, however, as we will like just get 1762 // gibberish when trying to match arguments. 1763 keepGoing = false; 1764 } 1765 1766 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion) 1767 << StringRef(csStart, csLen), 1768 Loc, /*IsStringLocation*/true, 1769 getSpecifierRange(startSpec, specifierLen)); 1770 1771 return keepGoing; 1772} 1773 1774void 1775CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc, 1776 const char *startSpec, 1777 unsigned specifierLen) { 1778 EmitFormatDiagnostic( 1779 S.PDiag(diag::warn_format_mix_positional_nonpositional_args), 1780 Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen)); 1781} 1782 1783bool 1784CheckFormatHandler::CheckNumArgs( 1785 const analyze_format_string::FormatSpecifier &FS, 1786 const analyze_format_string::ConversionSpecifier &CS, 1787 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 1788 1789 if (argIndex >= NumDataArgs) { 1790 PartialDiagnostic PDiag = FS.usesPositionalArg() 1791 ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args) 1792 << (argIndex+1) << NumDataArgs) 1793 : S.PDiag(diag::warn_printf_insufficient_data_args); 1794 EmitFormatDiagnostic( 1795 PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true, 1796 getSpecifierRange(startSpecifier, specifierLen)); 1797 return false; 1798 } 1799 return true; 1800} 1801 1802template<typename Range> 1803void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag, 1804 SourceLocation Loc, 1805 bool IsStringLocation, 1806 Range StringRange, 1807 FixItHint FixIt) { 1808 EmitFormatDiagnostic(S, inFunctionCall, TheCall->getArg(FormatIdx), PDiag, 1809 Loc, IsStringLocation, StringRange, FixIt); 1810} 1811 1812/// \brief If the format string is not within the funcion call, emit a note 1813/// so that the function call and string are in diagnostic messages. 1814/// 1815/// \param inFunctionCall if true, the format string is within the function 1816/// call and only one diagnostic message will be produced. Otherwise, an 1817/// extra note will be emitted pointing to location of the format string. 1818/// 1819/// \param ArgumentExpr the expression that is passed as the format string 1820/// argument in the function call. Used for getting locations when two 1821/// diagnostics are emitted. 1822/// 1823/// \param PDiag the callee should already have provided any strings for the 1824/// diagnostic message. This function only adds locations and fixits 1825/// to diagnostics. 1826/// 1827/// \param Loc primary location for diagnostic. If two diagnostics are 1828/// required, one will be at Loc and a new SourceLocation will be created for 1829/// the other one. 1830/// 1831/// \param IsStringLocation if true, Loc points to the format string should be 1832/// used for the note. Otherwise, Loc points to the argument list and will 1833/// be used with PDiag. 1834/// 1835/// \param StringRange some or all of the string to highlight. This is 1836/// templated so it can accept either a CharSourceRange or a SourceRange. 1837/// 1838/// \param Fixit optional fix it hint for the format string. 1839template<typename Range> 1840void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall, 1841 const Expr *ArgumentExpr, 1842 PartialDiagnostic PDiag, 1843 SourceLocation Loc, 1844 bool IsStringLocation, 1845 Range StringRange, 1846 FixItHint FixIt) { 1847 if (InFunctionCall) 1848 S.Diag(Loc, PDiag) << StringRange << FixIt; 1849 else { 1850 S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag) 1851 << ArgumentExpr->getSourceRange(); 1852 S.Diag(IsStringLocation ? Loc : StringRange.getBegin(), 1853 diag::note_format_string_defined) 1854 << StringRange << FixIt; 1855 } 1856} 1857 1858//===--- CHECK: Printf format string checking ------------------------------===// 1859 1860namespace { 1861class CheckPrintfHandler : public CheckFormatHandler { 1862public: 1863 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1864 const Expr *origFormatExpr, unsigned firstDataArg, 1865 unsigned numDataArgs, bool isObjCLiteral, 1866 const char *beg, bool hasVAListArg, 1867 const CallExpr *theCall, unsigned formatIdx, 1868 bool inFunctionCall) 1869 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1870 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1871 theCall, formatIdx, inFunctionCall) {} 1872 1873 1874 bool HandleInvalidPrintfConversionSpecifier( 1875 const analyze_printf::PrintfSpecifier &FS, 1876 const char *startSpecifier, 1877 unsigned specifierLen); 1878 1879 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 1880 const char *startSpecifier, 1881 unsigned specifierLen); 1882 1883 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 1884 const char *startSpecifier, unsigned specifierLen); 1885 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 1886 const analyze_printf::OptionalAmount &Amt, 1887 unsigned type, 1888 const char *startSpecifier, unsigned specifierLen); 1889 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1890 const analyze_printf::OptionalFlag &flag, 1891 const char *startSpecifier, unsigned specifierLen); 1892 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 1893 const analyze_printf::OptionalFlag &ignoredFlag, 1894 const analyze_printf::OptionalFlag &flag, 1895 const char *startSpecifier, unsigned specifierLen); 1896}; 1897} 1898 1899bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 1900 const analyze_printf::PrintfSpecifier &FS, 1901 const char *startSpecifier, 1902 unsigned specifierLen) { 1903 const analyze_printf::PrintfConversionSpecifier &CS = 1904 FS.getConversionSpecifier(); 1905 1906 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1907 getLocationOfByte(CS.getStart()), 1908 startSpecifier, specifierLen, 1909 CS.getStart(), CS.getLength()); 1910} 1911 1912bool CheckPrintfHandler::HandleAmount( 1913 const analyze_format_string::OptionalAmount &Amt, 1914 unsigned k, const char *startSpecifier, 1915 unsigned specifierLen) { 1916 1917 if (Amt.hasDataArgument()) { 1918 if (!HasVAListArg) { 1919 unsigned argIndex = Amt.getArgIndex(); 1920 if (argIndex >= NumDataArgs) { 1921 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg) 1922 << k, 1923 getLocationOfByte(Amt.getStart()), 1924 /*IsStringLocation*/true, 1925 getSpecifierRange(startSpecifier, specifierLen)); 1926 // Don't do any more checking. We will just emit 1927 // spurious errors. 1928 return false; 1929 } 1930 1931 // Type check the data argument. It should be an 'int'. 1932 // Although not in conformance with C99, we also allow the argument to be 1933 // an 'unsigned int' as that is a reasonably safe case. GCC also 1934 // doesn't emit a warning for that case. 1935 CoveredArgs.set(argIndex); 1936 const Expr *Arg = getDataArg(argIndex); 1937 QualType T = Arg->getType(); 1938 1939 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1940 assert(ATR.isValid()); 1941 1942 if (!ATR.matchesType(S.Context, T)) { 1943 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type) 1944 << k << ATR.getRepresentativeTypeName(S.Context) 1945 << T << Arg->getSourceRange(), 1946 getLocationOfByte(Amt.getStart()), 1947 /*IsStringLocation*/true, 1948 getSpecifierRange(startSpecifier, specifierLen)); 1949 // Don't do any more checking. We will just emit 1950 // spurious errors. 1951 return false; 1952 } 1953 } 1954 } 1955 return true; 1956} 1957 1958void CheckPrintfHandler::HandleInvalidAmount( 1959 const analyze_printf::PrintfSpecifier &FS, 1960 const analyze_printf::OptionalAmount &Amt, 1961 unsigned type, 1962 const char *startSpecifier, 1963 unsigned specifierLen) { 1964 const analyze_printf::PrintfConversionSpecifier &CS = 1965 FS.getConversionSpecifier(); 1966 1967 FixItHint fixit = 1968 Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant 1969 ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 1970 Amt.getConstantLength())) 1971 : FixItHint(); 1972 1973 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount) 1974 << type << CS.toString(), 1975 getLocationOfByte(Amt.getStart()), 1976 /*IsStringLocation*/true, 1977 getSpecifierRange(startSpecifier, specifierLen), 1978 fixit); 1979} 1980 1981void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1982 const analyze_printf::OptionalFlag &flag, 1983 const char *startSpecifier, 1984 unsigned specifierLen) { 1985 // Warn about pointless flag with a fixit removal. 1986 const analyze_printf::PrintfConversionSpecifier &CS = 1987 FS.getConversionSpecifier(); 1988 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag) 1989 << flag.toString() << CS.toString(), 1990 getLocationOfByte(flag.getPosition()), 1991 /*IsStringLocation*/true, 1992 getSpecifierRange(startSpecifier, specifierLen), 1993 FixItHint::CreateRemoval( 1994 getSpecifierRange(flag.getPosition(), 1))); 1995} 1996 1997void CheckPrintfHandler::HandleIgnoredFlag( 1998 const analyze_printf::PrintfSpecifier &FS, 1999 const analyze_printf::OptionalFlag &ignoredFlag, 2000 const analyze_printf::OptionalFlag &flag, 2001 const char *startSpecifier, 2002 unsigned specifierLen) { 2003 // Warn about ignored flag with a fixit removal. 2004 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag) 2005 << ignoredFlag.toString() << flag.toString(), 2006 getLocationOfByte(ignoredFlag.getPosition()), 2007 /*IsStringLocation*/true, 2008 getSpecifierRange(startSpecifier, specifierLen), 2009 FixItHint::CreateRemoval( 2010 getSpecifierRange(ignoredFlag.getPosition(), 1))); 2011} 2012 2013bool 2014CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 2015 &FS, 2016 const char *startSpecifier, 2017 unsigned specifierLen) { 2018 2019 using namespace analyze_format_string; 2020 using namespace analyze_printf; 2021 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 2022 2023 if (FS.consumesDataArgument()) { 2024 if (atFirstArg) { 2025 atFirstArg = false; 2026 usesPositionalArgs = FS.usesPositionalArg(); 2027 } 2028 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2029 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2030 startSpecifier, specifierLen); 2031 return false; 2032 } 2033 } 2034 2035 // First check if the field width, precision, and conversion specifier 2036 // have matching data arguments. 2037 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 2038 startSpecifier, specifierLen)) { 2039 return false; 2040 } 2041 2042 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 2043 startSpecifier, specifierLen)) { 2044 return false; 2045 } 2046 2047 if (!CS.consumesDataArgument()) { 2048 // FIXME: Technically specifying a precision or field width here 2049 // makes no sense. Worth issuing a warning at some point. 2050 return true; 2051 } 2052 2053 // Consume the argument. 2054 unsigned argIndex = FS.getArgIndex(); 2055 if (argIndex < NumDataArgs) { 2056 // The check to see if the argIndex is valid will come later. 2057 // We set the bit here because we may exit early from this 2058 // function if we encounter some other error. 2059 CoveredArgs.set(argIndex); 2060 } 2061 2062 // Check for using an Objective-C specific conversion specifier 2063 // in a non-ObjC literal. 2064 if (!IsObjCLiteral && CS.isObjCArg()) { 2065 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 2066 specifierLen); 2067 } 2068 2069 // Check for invalid use of field width 2070 if (!FS.hasValidFieldWidth()) { 2071 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 2072 startSpecifier, specifierLen); 2073 } 2074 2075 // Check for invalid use of precision 2076 if (!FS.hasValidPrecision()) { 2077 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 2078 startSpecifier, specifierLen); 2079 } 2080 2081 // Check each flag does not conflict with any other component. 2082 if (!FS.hasValidThousandsGroupingPrefix()) 2083 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 2084 if (!FS.hasValidLeadingZeros()) 2085 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 2086 if (!FS.hasValidPlusPrefix()) 2087 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 2088 if (!FS.hasValidSpacePrefix()) 2089 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 2090 if (!FS.hasValidAlternativeForm()) 2091 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 2092 if (!FS.hasValidLeftJustified()) 2093 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 2094 2095 // Check that flags are not ignored by another flag 2096 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 2097 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 2098 startSpecifier, specifierLen); 2099 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 2100 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 2101 startSpecifier, specifierLen); 2102 2103 // Check the length modifier is valid with the given conversion specifier. 2104 const LengthModifier &LM = FS.getLengthModifier(); 2105 if (!FS.hasValidLengthModifier()) 2106 EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length) 2107 << LM.toString() << CS.toString(), 2108 getLocationOfByte(LM.getStart()), 2109 /*IsStringLocation*/true, 2110 getSpecifierRange(startSpecifier, specifierLen), 2111 FixItHint::CreateRemoval( 2112 getSpecifierRange(LM.getStart(), 2113 LM.getLength()))); 2114 2115 // Are we using '%n'? 2116 if (CS.getKind() == ConversionSpecifier::nArg) { 2117 // Issue a warning about this being a possible security issue. 2118 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_write_back), 2119 getLocationOfByte(CS.getStart()), 2120 /*IsStringLocation*/true, 2121 getSpecifierRange(startSpecifier, specifierLen)); 2122 // Continue checking the other format specifiers. 2123 return true; 2124 } 2125 2126 // The remaining checks depend on the data arguments. 2127 if (HasVAListArg) 2128 return true; 2129 2130 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2131 return false; 2132 2133 // Now type check the data expression that matches the 2134 // format specifier. 2135 const Expr *Ex = getDataArg(argIndex); 2136 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context); 2137 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 2138 // Check if we didn't match because of an implicit cast from a 'char' 2139 // or 'short' to an 'int'. This is done because printf is a varargs 2140 // function. 2141 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 2142 if (ICE->getType() == S.Context.IntTy) { 2143 // All further checking is done on the subexpression. 2144 Ex = ICE->getSubExpr(); 2145 if (ATR.matchesType(S.Context, Ex->getType())) 2146 return true; 2147 } 2148 2149 // We may be able to offer a FixItHint if it is a supported type. 2150 PrintfSpecifier fixedFS = FS; 2151 bool success = fixedFS.fixType(Ex->getType(), S.getLangOptions()); 2152 2153 if (success) { 2154 // Get the fix string from the fixed format specifier 2155 llvm::SmallString<128> buf; 2156 llvm::raw_svector_ostream os(buf); 2157 fixedFS.toString(os); 2158 2159 EmitFormatDiagnostic( 2160 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2161 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2162 << Ex->getSourceRange(), 2163 getLocationOfByte(CS.getStart()), 2164 /*IsStringLocation*/true, 2165 getSpecifierRange(startSpecifier, specifierLen), 2166 FixItHint::CreateReplacement( 2167 getSpecifierRange(startSpecifier, specifierLen), 2168 os.str())); 2169 } 2170 else { 2171 S.Diag(getLocationOfByte(CS.getStart()), 2172 diag::warn_printf_conversion_argument_type_mismatch) 2173 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2174 << getSpecifierRange(startSpecifier, specifierLen) 2175 << Ex->getSourceRange(); 2176 } 2177 } 2178 2179 return true; 2180} 2181 2182//===--- CHECK: Scanf format string checking ------------------------------===// 2183 2184namespace { 2185class CheckScanfHandler : public CheckFormatHandler { 2186public: 2187 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 2188 const Expr *origFormatExpr, unsigned firstDataArg, 2189 unsigned numDataArgs, bool isObjCLiteral, 2190 const char *beg, bool hasVAListArg, 2191 const CallExpr *theCall, unsigned formatIdx, 2192 bool inFunctionCall) 2193 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2194 numDataArgs, isObjCLiteral, beg, hasVAListArg, 2195 theCall, formatIdx, inFunctionCall) {} 2196 2197 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 2198 const char *startSpecifier, 2199 unsigned specifierLen); 2200 2201 bool HandleInvalidScanfConversionSpecifier( 2202 const analyze_scanf::ScanfSpecifier &FS, 2203 const char *startSpecifier, 2204 unsigned specifierLen); 2205 2206 void HandleIncompleteScanList(const char *start, const char *end); 2207}; 2208} 2209 2210void CheckScanfHandler::HandleIncompleteScanList(const char *start, 2211 const char *end) { 2212 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 2213 getLocationOfByte(end), /*IsStringLocation*/true, 2214 getSpecifierRange(start, end - start)); 2215} 2216 2217bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 2218 const analyze_scanf::ScanfSpecifier &FS, 2219 const char *startSpecifier, 2220 unsigned specifierLen) { 2221 2222 const analyze_scanf::ScanfConversionSpecifier &CS = 2223 FS.getConversionSpecifier(); 2224 2225 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2226 getLocationOfByte(CS.getStart()), 2227 startSpecifier, specifierLen, 2228 CS.getStart(), CS.getLength()); 2229} 2230 2231bool CheckScanfHandler::HandleScanfSpecifier( 2232 const analyze_scanf::ScanfSpecifier &FS, 2233 const char *startSpecifier, 2234 unsigned specifierLen) { 2235 2236 using namespace analyze_scanf; 2237 using namespace analyze_format_string; 2238 2239 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 2240 2241 // Handle case where '%' and '*' don't consume an argument. These shouldn't 2242 // be used to decide if we are using positional arguments consistently. 2243 if (FS.consumesDataArgument()) { 2244 if (atFirstArg) { 2245 atFirstArg = false; 2246 usesPositionalArgs = FS.usesPositionalArg(); 2247 } 2248 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2249 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2250 startSpecifier, specifierLen); 2251 return false; 2252 } 2253 } 2254 2255 // Check if the field with is non-zero. 2256 const OptionalAmount &Amt = FS.getFieldWidth(); 2257 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 2258 if (Amt.getConstantAmount() == 0) { 2259 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 2260 Amt.getConstantLength()); 2261 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 2262 getLocationOfByte(Amt.getStart()), 2263 /*IsStringLocation*/true, R, 2264 FixItHint::CreateRemoval(R)); 2265 } 2266 } 2267 2268 if (!FS.consumesDataArgument()) { 2269 // FIXME: Technically specifying a precision or field width here 2270 // makes no sense. Worth issuing a warning at some point. 2271 return true; 2272 } 2273 2274 // Consume the argument. 2275 unsigned argIndex = FS.getArgIndex(); 2276 if (argIndex < NumDataArgs) { 2277 // The check to see if the argIndex is valid will come later. 2278 // We set the bit here because we may exit early from this 2279 // function if we encounter some other error. 2280 CoveredArgs.set(argIndex); 2281 } 2282 2283 // Check the length modifier is valid with the given conversion specifier. 2284 const LengthModifier &LM = FS.getLengthModifier(); 2285 if (!FS.hasValidLengthModifier()) { 2286 S.Diag(getLocationOfByte(LM.getStart()), 2287 diag::warn_format_nonsensical_length) 2288 << LM.toString() << CS.toString() 2289 << getSpecifierRange(startSpecifier, specifierLen) 2290 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 2291 LM.getLength())); 2292 } 2293 2294 // The remaining checks depend on the data arguments. 2295 if (HasVAListArg) 2296 return true; 2297 2298 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2299 return false; 2300 2301 // Check that the argument type matches the format specifier. 2302 const Expr *Ex = getDataArg(argIndex); 2303 const analyze_scanf::ScanfArgTypeResult &ATR = FS.getArgType(S.Context); 2304 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 2305 ScanfSpecifier fixedFS = FS; 2306 bool success = fixedFS.fixType(Ex->getType(), S.getLangOptions()); 2307 2308 if (success) { 2309 // Get the fix string from the fixed format specifier. 2310 llvm::SmallString<128> buf; 2311 llvm::raw_svector_ostream os(buf); 2312 fixedFS.toString(os); 2313 2314 EmitFormatDiagnostic( 2315 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2316 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2317 << Ex->getSourceRange(), 2318 getLocationOfByte(CS.getStart()), 2319 /*IsStringLocation*/true, 2320 getSpecifierRange(startSpecifier, specifierLen), 2321 FixItHint::CreateReplacement( 2322 getSpecifierRange(startSpecifier, specifierLen), 2323 os.str())); 2324 } else { 2325 S.Diag(getLocationOfByte(CS.getStart()), 2326 diag::warn_printf_conversion_argument_type_mismatch) 2327 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2328 << getSpecifierRange(startSpecifier, specifierLen) 2329 << Ex->getSourceRange(); 2330 } 2331 } 2332 2333 return true; 2334} 2335 2336void Sema::CheckFormatString(const StringLiteral *FExpr, 2337 const Expr *OrigFormatExpr, 2338 const CallExpr *TheCall, bool HasVAListArg, 2339 unsigned format_idx, unsigned firstDataArg, 2340 bool isPrintf, bool inFunctionCall) { 2341 2342 // CHECK: is the format string a wide literal? 2343 if (!FExpr->isAscii()) { 2344 CheckFormatHandler::EmitFormatDiagnostic( 2345 *this, inFunctionCall, TheCall->getArg(format_idx), 2346 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 2347 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2348 return; 2349 } 2350 2351 // Str - The format string. NOTE: this is NOT null-terminated! 2352 StringRef StrRef = FExpr->getString(); 2353 const char *Str = StrRef.data(); 2354 unsigned StrLen = StrRef.size(); 2355 const unsigned numDataArgs = TheCall->getNumArgs() - firstDataArg; 2356 2357 // CHECK: empty format string? 2358 if (StrLen == 0 && numDataArgs > 0) { 2359 CheckFormatHandler::EmitFormatDiagnostic( 2360 *this, inFunctionCall, TheCall->getArg(format_idx), 2361 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 2362 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2363 return; 2364 } 2365 2366 if (isPrintf) { 2367 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2368 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2369 Str, HasVAListArg, TheCall, format_idx, 2370 inFunctionCall); 2371 2372 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen, 2373 getLangOptions())) 2374 H.DoneProcessing(); 2375 } 2376 else { 2377 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2378 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2379 Str, HasVAListArg, TheCall, format_idx, 2380 inFunctionCall); 2381 2382 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen, 2383 getLangOptions())) 2384 H.DoneProcessing(); 2385 } 2386} 2387 2388//===--- CHECK: Standard memory functions ---------------------------------===// 2389 2390/// \brief Determine whether the given type is a dynamic class type (e.g., 2391/// whether it has a vtable). 2392static bool isDynamicClassType(QualType T) { 2393 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 2394 if (CXXRecordDecl *Definition = Record->getDefinition()) 2395 if (Definition->isDynamicClass()) 2396 return true; 2397 2398 return false; 2399} 2400 2401/// \brief If E is a sizeof expression, returns its argument expression, 2402/// otherwise returns NULL. 2403static const Expr *getSizeOfExprArg(const Expr* E) { 2404 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2405 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2406 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 2407 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 2408 2409 return 0; 2410} 2411 2412/// \brief If E is a sizeof expression, returns its argument type. 2413static QualType getSizeOfArgType(const Expr* E) { 2414 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2415 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2416 if (SizeOf->getKind() == clang::UETT_SizeOf) 2417 return SizeOf->getTypeOfArgument(); 2418 2419 return QualType(); 2420} 2421 2422/// \brief Check for dangerous or invalid arguments to memset(). 2423/// 2424/// This issues warnings on known problematic, dangerous or unspecified 2425/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 2426/// function calls. 2427/// 2428/// \param Call The call expression to diagnose. 2429void Sema::CheckMemaccessArguments(const CallExpr *Call, 2430 FunctionDecl::MemoryFunctionKind CMF, 2431 IdentifierInfo *FnName) { 2432 // It is possible to have a non-standard definition of memset. Validate 2433 // we have enough arguments, and if not, abort further checking. 2434 unsigned ExpectedNumArgs = (CMF == FunctionDecl::MFK_Strndup ? 2 : 3); 2435 if (Call->getNumArgs() < ExpectedNumArgs) 2436 return; 2437 2438 unsigned LastArg = (CMF == FunctionDecl::MFK_Memset || 2439 CMF == FunctionDecl::MFK_Strndup ? 1 : 2); 2440 unsigned LenArg = (CMF == FunctionDecl::MFK_Strndup ? 1 : 2); 2441 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 2442 2443 // We have special checking when the length is a sizeof expression. 2444 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 2445 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 2446 llvm::FoldingSetNodeID SizeOfArgID; 2447 2448 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 2449 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 2450 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 2451 2452 QualType DestTy = Dest->getType(); 2453 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 2454 QualType PointeeTy = DestPtrTy->getPointeeType(); 2455 2456 // Never warn about void type pointers. This can be used to suppress 2457 // false positives. 2458 if (PointeeTy->isVoidType()) 2459 continue; 2460 2461 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 2462 // actually comparing the expressions for equality. Because computing the 2463 // expression IDs can be expensive, we only do this if the diagnostic is 2464 // enabled. 2465 if (SizeOfArg && 2466 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 2467 SizeOfArg->getExprLoc())) { 2468 // We only compute IDs for expressions if the warning is enabled, and 2469 // cache the sizeof arg's ID. 2470 if (SizeOfArgID == llvm::FoldingSetNodeID()) 2471 SizeOfArg->Profile(SizeOfArgID, Context, true); 2472 llvm::FoldingSetNodeID DestID; 2473 Dest->Profile(DestID, Context, true); 2474 if (DestID == SizeOfArgID) { 2475 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 2476 // over sizeof(src) as well. 2477 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 2478 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 2479 if (UnaryOp->getOpcode() == UO_AddrOf) 2480 ActionIdx = 1; // If its an address-of operator, just remove it. 2481 if (Context.getTypeSize(PointeeTy) == Context.getCharWidth()) 2482 ActionIdx = 2; // If the pointee's size is sizeof(char), 2483 // suggest an explicit length. 2484 unsigned DestSrcSelect = 2485 (CMF == FunctionDecl::MFK_Strndup ? 1 : ArgIdx); 2486 DiagRuntimeBehavior(SizeOfArg->getExprLoc(), Dest, 2487 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 2488 << FnName << DestSrcSelect << ActionIdx 2489 << Dest->getSourceRange() 2490 << SizeOfArg->getSourceRange()); 2491 break; 2492 } 2493 } 2494 2495 // Also check for cases where the sizeof argument is the exact same 2496 // type as the memory argument, and where it points to a user-defined 2497 // record type. 2498 if (SizeOfArgTy != QualType()) { 2499 if (PointeeTy->isRecordType() && 2500 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 2501 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 2502 PDiag(diag::warn_sizeof_pointer_type_memaccess) 2503 << FnName << SizeOfArgTy << ArgIdx 2504 << PointeeTy << Dest->getSourceRange() 2505 << LenExpr->getSourceRange()); 2506 break; 2507 } 2508 } 2509 2510 // Always complain about dynamic classes. 2511 if (isDynamicClassType(PointeeTy)) 2512 DiagRuntimeBehavior( 2513 Dest->getExprLoc(), Dest, 2514 PDiag(diag::warn_dyn_class_memaccess) 2515 << (CMF == FunctionDecl::MFK_Memcmp ? ArgIdx + 2 : ArgIdx) 2516 << FnName << PointeeTy 2517 // "overwritten" if we're warning about the destination for any call 2518 // but memcmp; otherwise a verb appropriate to the call. 2519 << (ArgIdx == 0 && 2520 CMF != FunctionDecl::MFK_Memcmp ? 0 : (unsigned)CMF) 2521 << Call->getCallee()->getSourceRange()); 2522 else if (PointeeTy.hasNonTrivialObjCLifetime() && 2523 CMF != FunctionDecl::MFK_Memset) 2524 DiagRuntimeBehavior( 2525 Dest->getExprLoc(), Dest, 2526 PDiag(diag::warn_arc_object_memaccess) 2527 << ArgIdx << FnName << PointeeTy 2528 << Call->getCallee()->getSourceRange()); 2529 else 2530 continue; 2531 2532 DiagRuntimeBehavior( 2533 Dest->getExprLoc(), Dest, 2534 PDiag(diag::note_bad_memaccess_silence) 2535 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 2536 break; 2537 } 2538 } 2539} 2540 2541// A little helper routine: ignore addition and subtraction of integer literals. 2542// This intentionally does not ignore all integer constant expressions because 2543// we don't want to remove sizeof(). 2544static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 2545 Ex = Ex->IgnoreParenCasts(); 2546 2547 for (;;) { 2548 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 2549 if (!BO || !BO->isAdditiveOp()) 2550 break; 2551 2552 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 2553 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 2554 2555 if (isa<IntegerLiteral>(RHS)) 2556 Ex = LHS; 2557 else if (isa<IntegerLiteral>(LHS)) 2558 Ex = RHS; 2559 else 2560 break; 2561 } 2562 2563 return Ex; 2564} 2565 2566// Warn if the user has made the 'size' argument to strlcpy or strlcat 2567// be the size of the source, instead of the destination. 2568void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 2569 IdentifierInfo *FnName) { 2570 2571 // Don't crash if the user has the wrong number of arguments 2572 if (Call->getNumArgs() != 3) 2573 return; 2574 2575 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 2576 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 2577 const Expr *CompareWithSrc = NULL; 2578 2579 // Look for 'strlcpy(dst, x, sizeof(x))' 2580 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 2581 CompareWithSrc = Ex; 2582 else { 2583 // Look for 'strlcpy(dst, x, strlen(x))' 2584 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 2585 if (SizeCall->isBuiltinCall() == Builtin::BIstrlen 2586 && SizeCall->getNumArgs() == 1) 2587 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 2588 } 2589 } 2590 2591 if (!CompareWithSrc) 2592 return; 2593 2594 // Determine if the argument to sizeof/strlen is equal to the source 2595 // argument. In principle there's all kinds of things you could do 2596 // here, for instance creating an == expression and evaluating it with 2597 // EvaluateAsBooleanCondition, but this uses a more direct technique: 2598 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 2599 if (!SrcArgDRE) 2600 return; 2601 2602 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 2603 if (!CompareWithSrcDRE || 2604 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 2605 return; 2606 2607 const Expr *OriginalSizeArg = Call->getArg(2); 2608 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 2609 << OriginalSizeArg->getSourceRange() << FnName; 2610 2611 // Output a FIXIT hint if the destination is an array (rather than a 2612 // pointer to an array). This could be enhanced to handle some 2613 // pointers if we know the actual size, like if DstArg is 'array+2' 2614 // we could say 'sizeof(array)-2'. 2615 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 2616 QualType DstArgTy = DstArg->getType(); 2617 2618 // Only handle constant-sized or VLAs, but not flexible members. 2619 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 2620 // Only issue the FIXIT for arrays of size > 1. 2621 if (CAT->getSize().getSExtValue() <= 1) 2622 return; 2623 } else if (!DstArgTy->isVariableArrayType()) { 2624 return; 2625 } 2626 2627 llvm::SmallString<128> sizeString; 2628 llvm::raw_svector_ostream OS(sizeString); 2629 OS << "sizeof("; 2630 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 2631 OS << ")"; 2632 2633 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 2634 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 2635 OS.str()); 2636} 2637 2638//===--- CHECK: Return Address of Stack Variable --------------------------===// 2639 2640static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars); 2641static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars); 2642 2643/// CheckReturnStackAddr - Check if a return statement returns the address 2644/// of a stack variable. 2645void 2646Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 2647 SourceLocation ReturnLoc) { 2648 2649 Expr *stackE = 0; 2650 SmallVector<DeclRefExpr *, 8> refVars; 2651 2652 // Perform checking for returned stack addresses, local blocks, 2653 // label addresses or references to temporaries. 2654 if (lhsType->isPointerType() || 2655 (!getLangOptions().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 2656 stackE = EvalAddr(RetValExp, refVars); 2657 } else if (lhsType->isReferenceType()) { 2658 stackE = EvalVal(RetValExp, refVars); 2659 } 2660 2661 if (stackE == 0) 2662 return; // Nothing suspicious was found. 2663 2664 SourceLocation diagLoc; 2665 SourceRange diagRange; 2666 if (refVars.empty()) { 2667 diagLoc = stackE->getLocStart(); 2668 diagRange = stackE->getSourceRange(); 2669 } else { 2670 // We followed through a reference variable. 'stackE' contains the 2671 // problematic expression but we will warn at the return statement pointing 2672 // at the reference variable. We will later display the "trail" of 2673 // reference variables using notes. 2674 diagLoc = refVars[0]->getLocStart(); 2675 diagRange = refVars[0]->getSourceRange(); 2676 } 2677 2678 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 2679 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 2680 : diag::warn_ret_stack_addr) 2681 << DR->getDecl()->getDeclName() << diagRange; 2682 } else if (isa<BlockExpr>(stackE)) { // local block. 2683 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 2684 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 2685 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 2686 } else { // local temporary. 2687 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 2688 : diag::warn_ret_local_temp_addr) 2689 << diagRange; 2690 } 2691 2692 // Display the "trail" of reference variables that we followed until we 2693 // found the problematic expression using notes. 2694 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 2695 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 2696 // If this var binds to another reference var, show the range of the next 2697 // var, otherwise the var binds to the problematic expression, in which case 2698 // show the range of the expression. 2699 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 2700 : stackE->getSourceRange(); 2701 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 2702 << VD->getDeclName() << range; 2703 } 2704} 2705 2706/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 2707/// check if the expression in a return statement evaluates to an address 2708/// to a location on the stack, a local block, an address of a label, or a 2709/// reference to local temporary. The recursion is used to traverse the 2710/// AST of the return expression, with recursion backtracking when we 2711/// encounter a subexpression that (1) clearly does not lead to one of the 2712/// above problematic expressions (2) is something we cannot determine leads to 2713/// a problematic expression based on such local checking. 2714/// 2715/// Both EvalAddr and EvalVal follow through reference variables to evaluate 2716/// the expression that they point to. Such variables are added to the 2717/// 'refVars' vector so that we know what the reference variable "trail" was. 2718/// 2719/// EvalAddr processes expressions that are pointers that are used as 2720/// references (and not L-values). EvalVal handles all other values. 2721/// At the base case of the recursion is a check for the above problematic 2722/// expressions. 2723/// 2724/// This implementation handles: 2725/// 2726/// * pointer-to-pointer casts 2727/// * implicit conversions from array references to pointers 2728/// * taking the address of fields 2729/// * arbitrary interplay between "&" and "*" operators 2730/// * pointer arithmetic from an address of a stack variable 2731/// * taking the address of an array element where the array is on the stack 2732static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 2733 if (E->isTypeDependent()) 2734 return NULL; 2735 2736 // We should only be called for evaluating pointer expressions. 2737 assert((E->getType()->isAnyPointerType() || 2738 E->getType()->isBlockPointerType() || 2739 E->getType()->isObjCQualifiedIdType()) && 2740 "EvalAddr only works on pointers"); 2741 2742 E = E->IgnoreParens(); 2743 2744 // Our "symbolic interpreter" is just a dispatch off the currently 2745 // viewed AST node. We then recursively traverse the AST by calling 2746 // EvalAddr and EvalVal appropriately. 2747 switch (E->getStmtClass()) { 2748 case Stmt::DeclRefExprClass: { 2749 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2750 2751 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2752 // If this is a reference variable, follow through to the expression that 2753 // it points to. 2754 if (V->hasLocalStorage() && 2755 V->getType()->isReferenceType() && V->hasInit()) { 2756 // Add the reference variable to the "trail". 2757 refVars.push_back(DR); 2758 return EvalAddr(V->getInit(), refVars); 2759 } 2760 2761 return NULL; 2762 } 2763 2764 case Stmt::UnaryOperatorClass: { 2765 // The only unary operator that make sense to handle here 2766 // is AddrOf. All others don't make sense as pointers. 2767 UnaryOperator *U = cast<UnaryOperator>(E); 2768 2769 if (U->getOpcode() == UO_AddrOf) 2770 return EvalVal(U->getSubExpr(), refVars); 2771 else 2772 return NULL; 2773 } 2774 2775 case Stmt::BinaryOperatorClass: { 2776 // Handle pointer arithmetic. All other binary operators are not valid 2777 // in this context. 2778 BinaryOperator *B = cast<BinaryOperator>(E); 2779 BinaryOperatorKind op = B->getOpcode(); 2780 2781 if (op != BO_Add && op != BO_Sub) 2782 return NULL; 2783 2784 Expr *Base = B->getLHS(); 2785 2786 // Determine which argument is the real pointer base. It could be 2787 // the RHS argument instead of the LHS. 2788 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 2789 2790 assert (Base->getType()->isPointerType()); 2791 return EvalAddr(Base, refVars); 2792 } 2793 2794 // For conditional operators we need to see if either the LHS or RHS are 2795 // valid DeclRefExpr*s. If one of them is valid, we return it. 2796 case Stmt::ConditionalOperatorClass: { 2797 ConditionalOperator *C = cast<ConditionalOperator>(E); 2798 2799 // Handle the GNU extension for missing LHS. 2800 if (Expr *lhsExpr = C->getLHS()) { 2801 // In C++, we can have a throw-expression, which has 'void' type. 2802 if (!lhsExpr->getType()->isVoidType()) 2803 if (Expr* LHS = EvalAddr(lhsExpr, refVars)) 2804 return LHS; 2805 } 2806 2807 // In C++, we can have a throw-expression, which has 'void' type. 2808 if (C->getRHS()->getType()->isVoidType()) 2809 return NULL; 2810 2811 return EvalAddr(C->getRHS(), refVars); 2812 } 2813 2814 case Stmt::BlockExprClass: 2815 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 2816 return E; // local block. 2817 return NULL; 2818 2819 case Stmt::AddrLabelExprClass: 2820 return E; // address of label. 2821 2822 case Stmt::ExprWithCleanupsClass: 2823 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars); 2824 2825 // For casts, we need to handle conversions from arrays to 2826 // pointer values, and pointer-to-pointer conversions. 2827 case Stmt::ImplicitCastExprClass: 2828 case Stmt::CStyleCastExprClass: 2829 case Stmt::CXXFunctionalCastExprClass: 2830 case Stmt::ObjCBridgedCastExprClass: { 2831 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 2832 QualType T = SubExpr->getType(); 2833 2834 if (SubExpr->getType()->isPointerType() || 2835 SubExpr->getType()->isBlockPointerType() || 2836 SubExpr->getType()->isObjCQualifiedIdType()) 2837 return EvalAddr(SubExpr, refVars); 2838 else if (T->isArrayType()) 2839 return EvalVal(SubExpr, refVars); 2840 else 2841 return 0; 2842 } 2843 2844 // C++ casts. For dynamic casts, static casts, and const casts, we 2845 // are always converting from a pointer-to-pointer, so we just blow 2846 // through the cast. In the case the dynamic cast doesn't fail (and 2847 // return NULL), we take the conservative route and report cases 2848 // where we return the address of a stack variable. For Reinterpre 2849 // FIXME: The comment about is wrong; we're not always converting 2850 // from pointer to pointer. I'm guessing that this code should also 2851 // handle references to objects. 2852 case Stmt::CXXStaticCastExprClass: 2853 case Stmt::CXXDynamicCastExprClass: 2854 case Stmt::CXXConstCastExprClass: 2855 case Stmt::CXXReinterpretCastExprClass: { 2856 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 2857 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 2858 return EvalAddr(S, refVars); 2859 else 2860 return NULL; 2861 } 2862 2863 case Stmt::MaterializeTemporaryExprClass: 2864 if (Expr *Result = EvalAddr( 2865 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 2866 refVars)) 2867 return Result; 2868 2869 return E; 2870 2871 // Everything else: we simply don't reason about them. 2872 default: 2873 return NULL; 2874 } 2875} 2876 2877 2878/// EvalVal - This function is complements EvalAddr in the mutual recursion. 2879/// See the comments for EvalAddr for more details. 2880static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 2881do { 2882 // We should only be called for evaluating non-pointer expressions, or 2883 // expressions with a pointer type that are not used as references but instead 2884 // are l-values (e.g., DeclRefExpr with a pointer type). 2885 2886 // Our "symbolic interpreter" is just a dispatch off the currently 2887 // viewed AST node. We then recursively traverse the AST by calling 2888 // EvalAddr and EvalVal appropriately. 2889 2890 E = E->IgnoreParens(); 2891 switch (E->getStmtClass()) { 2892 case Stmt::ImplicitCastExprClass: { 2893 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 2894 if (IE->getValueKind() == VK_LValue) { 2895 E = IE->getSubExpr(); 2896 continue; 2897 } 2898 return NULL; 2899 } 2900 2901 case Stmt::ExprWithCleanupsClass: 2902 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars); 2903 2904 case Stmt::DeclRefExprClass: { 2905 // When we hit a DeclRefExpr we are looking at code that refers to a 2906 // variable's name. If it's not a reference variable we check if it has 2907 // local storage within the function, and if so, return the expression. 2908 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2909 2910 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2911 if (V->hasLocalStorage()) { 2912 if (!V->getType()->isReferenceType()) 2913 return DR; 2914 2915 // Reference variable, follow through to the expression that 2916 // it points to. 2917 if (V->hasInit()) { 2918 // Add the reference variable to the "trail". 2919 refVars.push_back(DR); 2920 return EvalVal(V->getInit(), refVars); 2921 } 2922 } 2923 2924 return NULL; 2925 } 2926 2927 case Stmt::UnaryOperatorClass: { 2928 // The only unary operator that make sense to handle here 2929 // is Deref. All others don't resolve to a "name." This includes 2930 // handling all sorts of rvalues passed to a unary operator. 2931 UnaryOperator *U = cast<UnaryOperator>(E); 2932 2933 if (U->getOpcode() == UO_Deref) 2934 return EvalAddr(U->getSubExpr(), refVars); 2935 2936 return NULL; 2937 } 2938 2939 case Stmt::ArraySubscriptExprClass: { 2940 // Array subscripts are potential references to data on the stack. We 2941 // retrieve the DeclRefExpr* for the array variable if it indeed 2942 // has local storage. 2943 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars); 2944 } 2945 2946 case Stmt::ConditionalOperatorClass: { 2947 // For conditional operators we need to see if either the LHS or RHS are 2948 // non-NULL Expr's. If one is non-NULL, we return it. 2949 ConditionalOperator *C = cast<ConditionalOperator>(E); 2950 2951 // Handle the GNU extension for missing LHS. 2952 if (Expr *lhsExpr = C->getLHS()) 2953 if (Expr *LHS = EvalVal(lhsExpr, refVars)) 2954 return LHS; 2955 2956 return EvalVal(C->getRHS(), refVars); 2957 } 2958 2959 // Accesses to members are potential references to data on the stack. 2960 case Stmt::MemberExprClass: { 2961 MemberExpr *M = cast<MemberExpr>(E); 2962 2963 // Check for indirect access. We only want direct field accesses. 2964 if (M->isArrow()) 2965 return NULL; 2966 2967 // Check whether the member type is itself a reference, in which case 2968 // we're not going to refer to the member, but to what the member refers to. 2969 if (M->getMemberDecl()->getType()->isReferenceType()) 2970 return NULL; 2971 2972 return EvalVal(M->getBase(), refVars); 2973 } 2974 2975 case Stmt::MaterializeTemporaryExprClass: 2976 if (Expr *Result = EvalVal( 2977 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 2978 refVars)) 2979 return Result; 2980 2981 return E; 2982 2983 default: 2984 // Check that we don't return or take the address of a reference to a 2985 // temporary. This is only useful in C++. 2986 if (!E->isTypeDependent() && E->isRValue()) 2987 return E; 2988 2989 // Everything else: we simply don't reason about them. 2990 return NULL; 2991 } 2992} while (true); 2993} 2994 2995//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 2996 2997/// Check for comparisons of floating point operands using != and ==. 2998/// Issue a warning if these are no self-comparisons, as they are not likely 2999/// to do what the programmer intended. 3000void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 3001 bool EmitWarning = true; 3002 3003 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 3004 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 3005 3006 // Special case: check for x == x (which is OK). 3007 // Do not emit warnings for such cases. 3008 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 3009 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 3010 if (DRL->getDecl() == DRR->getDecl()) 3011 EmitWarning = false; 3012 3013 3014 // Special case: check for comparisons against literals that can be exactly 3015 // represented by APFloat. In such cases, do not emit a warning. This 3016 // is a heuristic: often comparison against such literals are used to 3017 // detect if a value in a variable has not changed. This clearly can 3018 // lead to false negatives. 3019 if (EmitWarning) { 3020 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 3021 if (FLL->isExact()) 3022 EmitWarning = false; 3023 } else 3024 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 3025 if (FLR->isExact()) 3026 EmitWarning = false; 3027 } 3028 } 3029 3030 // Check for comparisons with builtin types. 3031 if (EmitWarning) 3032 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 3033 if (CL->isBuiltinCall()) 3034 EmitWarning = false; 3035 3036 if (EmitWarning) 3037 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 3038 if (CR->isBuiltinCall()) 3039 EmitWarning = false; 3040 3041 // Emit the diagnostic. 3042 if (EmitWarning) 3043 Diag(Loc, diag::warn_floatingpoint_eq) 3044 << LHS->getSourceRange() << RHS->getSourceRange(); 3045} 3046 3047//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 3048//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 3049 3050namespace { 3051 3052/// Structure recording the 'active' range of an integer-valued 3053/// expression. 3054struct IntRange { 3055 /// The number of bits active in the int. 3056 unsigned Width; 3057 3058 /// True if the int is known not to have negative values. 3059 bool NonNegative; 3060 3061 IntRange(unsigned Width, bool NonNegative) 3062 : Width(Width), NonNegative(NonNegative) 3063 {} 3064 3065 /// Returns the range of the bool type. 3066 static IntRange forBoolType() { 3067 return IntRange(1, true); 3068 } 3069 3070 /// Returns the range of an opaque value of the given integral type. 3071 static IntRange forValueOfType(ASTContext &C, QualType T) { 3072 return forValueOfCanonicalType(C, 3073 T->getCanonicalTypeInternal().getTypePtr()); 3074 } 3075 3076 /// Returns the range of an opaque value of a canonical integral type. 3077 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 3078 assert(T->isCanonicalUnqualified()); 3079 3080 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3081 T = VT->getElementType().getTypePtr(); 3082 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3083 T = CT->getElementType().getTypePtr(); 3084 3085 // For enum types, use the known bit width of the enumerators. 3086 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 3087 EnumDecl *Enum = ET->getDecl(); 3088 if (!Enum->isCompleteDefinition()) 3089 return IntRange(C.getIntWidth(QualType(T, 0)), false); 3090 3091 unsigned NumPositive = Enum->getNumPositiveBits(); 3092 unsigned NumNegative = Enum->getNumNegativeBits(); 3093 3094 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 3095 } 3096 3097 const BuiltinType *BT = cast<BuiltinType>(T); 3098 assert(BT->isInteger()); 3099 3100 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3101 } 3102 3103 /// Returns the "target" range of a canonical integral type, i.e. 3104 /// the range of values expressible in the type. 3105 /// 3106 /// This matches forValueOfCanonicalType except that enums have the 3107 /// full range of their type, not the range of their enumerators. 3108 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 3109 assert(T->isCanonicalUnqualified()); 3110 3111 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3112 T = VT->getElementType().getTypePtr(); 3113 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3114 T = CT->getElementType().getTypePtr(); 3115 if (const EnumType *ET = dyn_cast<EnumType>(T)) 3116 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 3117 3118 const BuiltinType *BT = cast<BuiltinType>(T); 3119 assert(BT->isInteger()); 3120 3121 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3122 } 3123 3124 /// Returns the supremum of two ranges: i.e. their conservative merge. 3125 static IntRange join(IntRange L, IntRange R) { 3126 return IntRange(std::max(L.Width, R.Width), 3127 L.NonNegative && R.NonNegative); 3128 } 3129 3130 /// Returns the infinum of two ranges: i.e. their aggressive merge. 3131 static IntRange meet(IntRange L, IntRange R) { 3132 return IntRange(std::min(L.Width, R.Width), 3133 L.NonNegative || R.NonNegative); 3134 } 3135}; 3136 3137IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 3138 if (value.isSigned() && value.isNegative()) 3139 return IntRange(value.getMinSignedBits(), false); 3140 3141 if (value.getBitWidth() > MaxWidth) 3142 value = value.trunc(MaxWidth); 3143 3144 // isNonNegative() just checks the sign bit without considering 3145 // signedness. 3146 return IntRange(value.getActiveBits(), true); 3147} 3148 3149IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 3150 unsigned MaxWidth) { 3151 if (result.isInt()) 3152 return GetValueRange(C, result.getInt(), MaxWidth); 3153 3154 if (result.isVector()) { 3155 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 3156 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 3157 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 3158 R = IntRange::join(R, El); 3159 } 3160 return R; 3161 } 3162 3163 if (result.isComplexInt()) { 3164 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 3165 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 3166 return IntRange::join(R, I); 3167 } 3168 3169 // This can happen with lossless casts to intptr_t of "based" lvalues. 3170 // Assume it might use arbitrary bits. 3171 // FIXME: The only reason we need to pass the type in here is to get 3172 // the sign right on this one case. It would be nice if APValue 3173 // preserved this. 3174 assert(result.isLValue() || result.isAddrLabelDiff()); 3175 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 3176} 3177 3178/// Pseudo-evaluate the given integer expression, estimating the 3179/// range of values it might take. 3180/// 3181/// \param MaxWidth - the width to which the value will be truncated 3182IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 3183 E = E->IgnoreParens(); 3184 3185 // Try a full evaluation first. 3186 Expr::EvalResult result; 3187 if (E->EvaluateAsRValue(result, C)) 3188 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 3189 3190 // I think we only want to look through implicit casts here; if the 3191 // user has an explicit widening cast, we should treat the value as 3192 // being of the new, wider type. 3193 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 3194 if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue) 3195 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 3196 3197 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 3198 3199 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 3200 3201 // Assume that non-integer casts can span the full range of the type. 3202 if (!isIntegerCast) 3203 return OutputTypeRange; 3204 3205 IntRange SubRange 3206 = GetExprRange(C, CE->getSubExpr(), 3207 std::min(MaxWidth, OutputTypeRange.Width)); 3208 3209 // Bail out if the subexpr's range is as wide as the cast type. 3210 if (SubRange.Width >= OutputTypeRange.Width) 3211 return OutputTypeRange; 3212 3213 // Otherwise, we take the smaller width, and we're non-negative if 3214 // either the output type or the subexpr is. 3215 return IntRange(SubRange.Width, 3216 SubRange.NonNegative || OutputTypeRange.NonNegative); 3217 } 3218 3219 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 3220 // If we can fold the condition, just take that operand. 3221 bool CondResult; 3222 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 3223 return GetExprRange(C, CondResult ? CO->getTrueExpr() 3224 : CO->getFalseExpr(), 3225 MaxWidth); 3226 3227 // Otherwise, conservatively merge. 3228 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 3229 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 3230 return IntRange::join(L, R); 3231 } 3232 3233 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 3234 switch (BO->getOpcode()) { 3235 3236 // Boolean-valued operations are single-bit and positive. 3237 case BO_LAnd: 3238 case BO_LOr: 3239 case BO_LT: 3240 case BO_GT: 3241 case BO_LE: 3242 case BO_GE: 3243 case BO_EQ: 3244 case BO_NE: 3245 return IntRange::forBoolType(); 3246 3247 // The type of the assignments is the type of the LHS, so the RHS 3248 // is not necessarily the same type. 3249 case BO_MulAssign: 3250 case BO_DivAssign: 3251 case BO_RemAssign: 3252 case BO_AddAssign: 3253 case BO_SubAssign: 3254 case BO_XorAssign: 3255 case BO_OrAssign: 3256 // TODO: bitfields? 3257 return IntRange::forValueOfType(C, E->getType()); 3258 3259 // Simple assignments just pass through the RHS, which will have 3260 // been coerced to the LHS type. 3261 case BO_Assign: 3262 // TODO: bitfields? 3263 return GetExprRange(C, BO->getRHS(), MaxWidth); 3264 3265 // Operations with opaque sources are black-listed. 3266 case BO_PtrMemD: 3267 case BO_PtrMemI: 3268 return IntRange::forValueOfType(C, E->getType()); 3269 3270 // Bitwise-and uses the *infinum* of the two source ranges. 3271 case BO_And: 3272 case BO_AndAssign: 3273 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 3274 GetExprRange(C, BO->getRHS(), MaxWidth)); 3275 3276 // Left shift gets black-listed based on a judgement call. 3277 case BO_Shl: 3278 // ...except that we want to treat '1 << (blah)' as logically 3279 // positive. It's an important idiom. 3280 if (IntegerLiteral *I 3281 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 3282 if (I->getValue() == 1) { 3283 IntRange R = IntRange::forValueOfType(C, E->getType()); 3284 return IntRange(R.Width, /*NonNegative*/ true); 3285 } 3286 } 3287 // fallthrough 3288 3289 case BO_ShlAssign: 3290 return IntRange::forValueOfType(C, E->getType()); 3291 3292 // Right shift by a constant can narrow its left argument. 3293 case BO_Shr: 3294 case BO_ShrAssign: { 3295 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3296 3297 // If the shift amount is a positive constant, drop the width by 3298 // that much. 3299 llvm::APSInt shift; 3300 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 3301 shift.isNonNegative()) { 3302 unsigned zext = shift.getZExtValue(); 3303 if (zext >= L.Width) 3304 L.Width = (L.NonNegative ? 0 : 1); 3305 else 3306 L.Width -= zext; 3307 } 3308 3309 return L; 3310 } 3311 3312 // Comma acts as its right operand. 3313 case BO_Comma: 3314 return GetExprRange(C, BO->getRHS(), MaxWidth); 3315 3316 // Black-list pointer subtractions. 3317 case BO_Sub: 3318 if (BO->getLHS()->getType()->isPointerType()) 3319 return IntRange::forValueOfType(C, E->getType()); 3320 break; 3321 3322 // The width of a division result is mostly determined by the size 3323 // of the LHS. 3324 case BO_Div: { 3325 // Don't 'pre-truncate' the operands. 3326 unsigned opWidth = C.getIntWidth(E->getType()); 3327 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3328 3329 // If the divisor is constant, use that. 3330 llvm::APSInt divisor; 3331 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 3332 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 3333 if (log2 >= L.Width) 3334 L.Width = (L.NonNegative ? 0 : 1); 3335 else 3336 L.Width = std::min(L.Width - log2, MaxWidth); 3337 return L; 3338 } 3339 3340 // Otherwise, just use the LHS's width. 3341 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3342 return IntRange(L.Width, L.NonNegative && R.NonNegative); 3343 } 3344 3345 // The result of a remainder can't be larger than the result of 3346 // either side. 3347 case BO_Rem: { 3348 // Don't 'pre-truncate' the operands. 3349 unsigned opWidth = C.getIntWidth(E->getType()); 3350 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3351 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3352 3353 IntRange meet = IntRange::meet(L, R); 3354 meet.Width = std::min(meet.Width, MaxWidth); 3355 return meet; 3356 } 3357 3358 // The default behavior is okay for these. 3359 case BO_Mul: 3360 case BO_Add: 3361 case BO_Xor: 3362 case BO_Or: 3363 break; 3364 } 3365 3366 // The default case is to treat the operation as if it were closed 3367 // on the narrowest type that encompasses both operands. 3368 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3369 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 3370 return IntRange::join(L, R); 3371 } 3372 3373 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 3374 switch (UO->getOpcode()) { 3375 // Boolean-valued operations are white-listed. 3376 case UO_LNot: 3377 return IntRange::forBoolType(); 3378 3379 // Operations with opaque sources are black-listed. 3380 case UO_Deref: 3381 case UO_AddrOf: // should be impossible 3382 return IntRange::forValueOfType(C, E->getType()); 3383 3384 default: 3385 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 3386 } 3387 } 3388 3389 if (dyn_cast<OffsetOfExpr>(E)) { 3390 IntRange::forValueOfType(C, E->getType()); 3391 } 3392 3393 if (FieldDecl *BitField = E->getBitField()) 3394 return IntRange(BitField->getBitWidthValue(C), 3395 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 3396 3397 return IntRange::forValueOfType(C, E->getType()); 3398} 3399 3400IntRange GetExprRange(ASTContext &C, Expr *E) { 3401 return GetExprRange(C, E, C.getIntWidth(E->getType())); 3402} 3403 3404/// Checks whether the given value, which currently has the given 3405/// source semantics, has the same value when coerced through the 3406/// target semantics. 3407bool IsSameFloatAfterCast(const llvm::APFloat &value, 3408 const llvm::fltSemantics &Src, 3409 const llvm::fltSemantics &Tgt) { 3410 llvm::APFloat truncated = value; 3411 3412 bool ignored; 3413 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 3414 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 3415 3416 return truncated.bitwiseIsEqual(value); 3417} 3418 3419/// Checks whether the given value, which currently has the given 3420/// source semantics, has the same value when coerced through the 3421/// target semantics. 3422/// 3423/// The value might be a vector of floats (or a complex number). 3424bool IsSameFloatAfterCast(const APValue &value, 3425 const llvm::fltSemantics &Src, 3426 const llvm::fltSemantics &Tgt) { 3427 if (value.isFloat()) 3428 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 3429 3430 if (value.isVector()) { 3431 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 3432 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 3433 return false; 3434 return true; 3435 } 3436 3437 assert(value.isComplexFloat()); 3438 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 3439 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 3440} 3441 3442void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 3443 3444static bool IsZero(Sema &S, Expr *E) { 3445 // Suppress cases where we are comparing against an enum constant. 3446 if (const DeclRefExpr *DR = 3447 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 3448 if (isa<EnumConstantDecl>(DR->getDecl())) 3449 return false; 3450 3451 // Suppress cases where the '0' value is expanded from a macro. 3452 if (E->getLocStart().isMacroID()) 3453 return false; 3454 3455 llvm::APSInt Value; 3456 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 3457} 3458 3459static bool HasEnumType(Expr *E) { 3460 // Strip off implicit integral promotions. 3461 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 3462 if (ICE->getCastKind() != CK_IntegralCast && 3463 ICE->getCastKind() != CK_NoOp) 3464 break; 3465 E = ICE->getSubExpr(); 3466 } 3467 3468 return E->getType()->isEnumeralType(); 3469} 3470 3471void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 3472 BinaryOperatorKind op = E->getOpcode(); 3473 if (E->isValueDependent()) 3474 return; 3475 3476 if (op == BO_LT && IsZero(S, E->getRHS())) { 3477 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3478 << "< 0" << "false" << HasEnumType(E->getLHS()) 3479 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3480 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 3481 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3482 << ">= 0" << "true" << HasEnumType(E->getLHS()) 3483 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3484 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 3485 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3486 << "0 >" << "false" << HasEnumType(E->getRHS()) 3487 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3488 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 3489 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3490 << "0 <=" << "true" << HasEnumType(E->getRHS()) 3491 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3492 } 3493} 3494 3495/// Analyze the operands of the given comparison. Implements the 3496/// fallback case from AnalyzeComparison. 3497void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 3498 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3499 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3500} 3501 3502/// \brief Implements -Wsign-compare. 3503/// 3504/// \param E the binary operator to check for warnings 3505void AnalyzeComparison(Sema &S, BinaryOperator *E) { 3506 // The type the comparison is being performed in. 3507 QualType T = E->getLHS()->getType(); 3508 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 3509 && "comparison with mismatched types"); 3510 3511 // We don't do anything special if this isn't an unsigned integral 3512 // comparison: we're only interested in integral comparisons, and 3513 // signed comparisons only happen in cases we don't care to warn about. 3514 // 3515 // We also don't care about value-dependent expressions or expressions 3516 // whose result is a constant. 3517 if (!T->hasUnsignedIntegerRepresentation() 3518 || E->isValueDependent() || E->isIntegerConstantExpr(S.Context)) 3519 return AnalyzeImpConvsInComparison(S, E); 3520 3521 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 3522 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 3523 3524 // Check to see if one of the (unmodified) operands is of different 3525 // signedness. 3526 Expr *signedOperand, *unsignedOperand; 3527 if (LHS->getType()->hasSignedIntegerRepresentation()) { 3528 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 3529 "unsigned comparison between two signed integer expressions?"); 3530 signedOperand = LHS; 3531 unsignedOperand = RHS; 3532 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 3533 signedOperand = RHS; 3534 unsignedOperand = LHS; 3535 } else { 3536 CheckTrivialUnsignedComparison(S, E); 3537 return AnalyzeImpConvsInComparison(S, E); 3538 } 3539 3540 // Otherwise, calculate the effective range of the signed operand. 3541 IntRange signedRange = GetExprRange(S.Context, signedOperand); 3542 3543 // Go ahead and analyze implicit conversions in the operands. Note 3544 // that we skip the implicit conversions on both sides. 3545 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 3546 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 3547 3548 // If the signed range is non-negative, -Wsign-compare won't fire, 3549 // but we should still check for comparisons which are always true 3550 // or false. 3551 if (signedRange.NonNegative) 3552 return CheckTrivialUnsignedComparison(S, E); 3553 3554 // For (in)equality comparisons, if the unsigned operand is a 3555 // constant which cannot collide with a overflowed signed operand, 3556 // then reinterpreting the signed operand as unsigned will not 3557 // change the result of the comparison. 3558 if (E->isEqualityOp()) { 3559 unsigned comparisonWidth = S.Context.getIntWidth(T); 3560 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 3561 3562 // We should never be unable to prove that the unsigned operand is 3563 // non-negative. 3564 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 3565 3566 if (unsignedRange.Width < comparisonWidth) 3567 return; 3568 } 3569 3570 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison) 3571 << LHS->getType() << RHS->getType() 3572 << LHS->getSourceRange() << RHS->getSourceRange(); 3573} 3574 3575/// Analyzes an attempt to assign the given value to a bitfield. 3576/// 3577/// Returns true if there was something fishy about the attempt. 3578bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 3579 SourceLocation InitLoc) { 3580 assert(Bitfield->isBitField()); 3581 if (Bitfield->isInvalidDecl()) 3582 return false; 3583 3584 // White-list bool bitfields. 3585 if (Bitfield->getType()->isBooleanType()) 3586 return false; 3587 3588 // Ignore value- or type-dependent expressions. 3589 if (Bitfield->getBitWidth()->isValueDependent() || 3590 Bitfield->getBitWidth()->isTypeDependent() || 3591 Init->isValueDependent() || 3592 Init->isTypeDependent()) 3593 return false; 3594 3595 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 3596 3597 llvm::APSInt Value; 3598 if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects)) 3599 return false; 3600 3601 unsigned OriginalWidth = Value.getBitWidth(); 3602 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 3603 3604 if (OriginalWidth <= FieldWidth) 3605 return false; 3606 3607 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 3608 3609 // It's fairly common to write values into signed bitfields 3610 // that, if sign-extended, would end up becoming a different 3611 // value. We don't want to warn about that. 3612 if (Value.isSigned() && Value.isNegative()) 3613 TruncatedValue = TruncatedValue.sext(OriginalWidth); 3614 else 3615 TruncatedValue = TruncatedValue.zext(OriginalWidth); 3616 3617 if (Value == TruncatedValue) 3618 return false; 3619 3620 std::string PrettyValue = Value.toString(10); 3621 std::string PrettyTrunc = TruncatedValue.toString(10); 3622 3623 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 3624 << PrettyValue << PrettyTrunc << OriginalInit->getType() 3625 << Init->getSourceRange(); 3626 3627 return true; 3628} 3629 3630/// Analyze the given simple or compound assignment for warning-worthy 3631/// operations. 3632void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 3633 // Just recurse on the LHS. 3634 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3635 3636 // We want to recurse on the RHS as normal unless we're assigning to 3637 // a bitfield. 3638 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 3639 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 3640 E->getOperatorLoc())) { 3641 // Recurse, ignoring any implicit conversions on the RHS. 3642 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 3643 E->getOperatorLoc()); 3644 } 3645 } 3646 3647 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3648} 3649 3650/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3651void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 3652 SourceLocation CContext, unsigned diag) { 3653 S.Diag(E->getExprLoc(), diag) 3654 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 3655} 3656 3657/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3658void DiagnoseImpCast(Sema &S, Expr *E, QualType T, SourceLocation CContext, 3659 unsigned diag) { 3660 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag); 3661} 3662 3663/// Diagnose an implicit cast from a literal expression. Does not warn when the 3664/// cast wouldn't lose information. 3665void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 3666 SourceLocation CContext) { 3667 // Try to convert the literal exactly to an integer. If we can, don't warn. 3668 bool isExact = false; 3669 const llvm::APFloat &Value = FL->getValue(); 3670 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 3671 T->hasUnsignedIntegerRepresentation()); 3672 if (Value.convertToInteger(IntegerValue, 3673 llvm::APFloat::rmTowardZero, &isExact) 3674 == llvm::APFloat::opOK && isExact) 3675 return; 3676 3677 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 3678 << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext); 3679} 3680 3681std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 3682 if (!Range.Width) return "0"; 3683 3684 llvm::APSInt ValueInRange = Value; 3685 ValueInRange.setIsSigned(!Range.NonNegative); 3686 ValueInRange = ValueInRange.trunc(Range.Width); 3687 return ValueInRange.toString(10); 3688} 3689 3690void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 3691 SourceLocation CC, bool *ICContext = 0) { 3692 if (E->isTypeDependent() || E->isValueDependent()) return; 3693 3694 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 3695 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 3696 if (Source == Target) return; 3697 if (Target->isDependentType()) return; 3698 3699 // If the conversion context location is invalid don't complain. We also 3700 // don't want to emit a warning if the issue occurs from the expansion of 3701 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 3702 // delay this check as long as possible. Once we detect we are in that 3703 // scenario, we just return. 3704 if (CC.isInvalid()) 3705 return; 3706 3707 // Diagnose implicit casts to bool. 3708 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 3709 if (isa<StringLiteral>(E)) 3710 // Warn on string literal to bool. Checks for string literals in logical 3711 // expressions, for instances, assert(0 && "error here"), is prevented 3712 // by a check in AnalyzeImplicitConversions(). 3713 return DiagnoseImpCast(S, E, T, CC, 3714 diag::warn_impcast_string_literal_to_bool); 3715 if (Source->isFunctionType()) { 3716 // Warn on function to bool. Checks free functions and static member 3717 // functions. Weakly imported functions are excluded from the check, 3718 // since it's common to test their value to check whether the linker 3719 // found a definition for them. 3720 ValueDecl *D = 0; 3721 if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) { 3722 D = R->getDecl(); 3723 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 3724 D = M->getMemberDecl(); 3725 } 3726 3727 if (D && !D->isWeak()) { 3728 if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) { 3729 S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool) 3730 << F << E->getSourceRange() << SourceRange(CC); 3731 S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence) 3732 << FixItHint::CreateInsertion(E->getExprLoc(), "&"); 3733 QualType ReturnType; 3734 UnresolvedSet<4> NonTemplateOverloads; 3735 S.isExprCallable(*E, ReturnType, NonTemplateOverloads); 3736 if (!ReturnType.isNull() 3737 && ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 3738 S.Diag(E->getExprLoc(), diag::note_function_to_bool_call) 3739 << FixItHint::CreateInsertion( 3740 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()"); 3741 return; 3742 } 3743 } 3744 } 3745 return; // Other casts to bool are not checked. 3746 } 3747 3748 // Strip vector types. 3749 if (isa<VectorType>(Source)) { 3750 if (!isa<VectorType>(Target)) { 3751 if (S.SourceMgr.isInSystemMacro(CC)) 3752 return; 3753 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 3754 } 3755 3756 // If the vector cast is cast between two vectors of the same size, it is 3757 // a bitcast, not a conversion. 3758 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 3759 return; 3760 3761 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 3762 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 3763 } 3764 3765 // Strip complex types. 3766 if (isa<ComplexType>(Source)) { 3767 if (!isa<ComplexType>(Target)) { 3768 if (S.SourceMgr.isInSystemMacro(CC)) 3769 return; 3770 3771 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 3772 } 3773 3774 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 3775 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 3776 } 3777 3778 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 3779 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 3780 3781 // If the source is floating point... 3782 if (SourceBT && SourceBT->isFloatingPoint()) { 3783 // ...and the target is floating point... 3784 if (TargetBT && TargetBT->isFloatingPoint()) { 3785 // ...then warn if we're dropping FP rank. 3786 3787 // Builtin FP kinds are ordered by increasing FP rank. 3788 if (SourceBT->getKind() > TargetBT->getKind()) { 3789 // Don't warn about float constants that are precisely 3790 // representable in the target type. 3791 Expr::EvalResult result; 3792 if (E->EvaluateAsRValue(result, S.Context)) { 3793 // Value might be a float, a float vector, or a float complex. 3794 if (IsSameFloatAfterCast(result.Val, 3795 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 3796 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 3797 return; 3798 } 3799 3800 if (S.SourceMgr.isInSystemMacro(CC)) 3801 return; 3802 3803 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 3804 } 3805 return; 3806 } 3807 3808 // If the target is integral, always warn. 3809 if ((TargetBT && TargetBT->isInteger())) { 3810 if (S.SourceMgr.isInSystemMacro(CC)) 3811 return; 3812 3813 Expr *InnerE = E->IgnoreParenImpCasts(); 3814 // We also want to warn on, e.g., "int i = -1.234" 3815 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 3816 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 3817 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 3818 3819 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 3820 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 3821 } else { 3822 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 3823 } 3824 } 3825 3826 return; 3827 } 3828 3829 if (!Source->isIntegerType() || !Target->isIntegerType()) 3830 return; 3831 3832 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 3833 == Expr::NPCK_GNUNull) && Target->isIntegerType()) { 3834 S.Diag(E->getExprLoc(), diag::warn_impcast_null_pointer_to_integer) 3835 << E->getSourceRange() << clang::SourceRange(CC); 3836 return; 3837 } 3838 3839 IntRange SourceRange = GetExprRange(S.Context, E); 3840 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 3841 3842 if (SourceRange.Width > TargetRange.Width) { 3843 // If the source is a constant, use a default-on diagnostic. 3844 // TODO: this should happen for bitfield stores, too. 3845 llvm::APSInt Value(32); 3846 if (E->isIntegerConstantExpr(Value, S.Context)) { 3847 if (S.SourceMgr.isInSystemMacro(CC)) 3848 return; 3849 3850 std::string PrettySourceValue = Value.toString(10); 3851 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 3852 3853 S.DiagRuntimeBehavior(E->getExprLoc(), E, 3854 S.PDiag(diag::warn_impcast_integer_precision_constant) 3855 << PrettySourceValue << PrettyTargetValue 3856 << E->getType() << T << E->getSourceRange() 3857 << clang::SourceRange(CC)); 3858 return; 3859 } 3860 3861 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 3862 if (S.SourceMgr.isInSystemMacro(CC)) 3863 return; 3864 3865 if (SourceRange.Width == 64 && TargetRange.Width == 32) 3866 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32); 3867 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 3868 } 3869 3870 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 3871 (!TargetRange.NonNegative && SourceRange.NonNegative && 3872 SourceRange.Width == TargetRange.Width)) { 3873 3874 if (S.SourceMgr.isInSystemMacro(CC)) 3875 return; 3876 3877 unsigned DiagID = diag::warn_impcast_integer_sign; 3878 3879 // Traditionally, gcc has warned about this under -Wsign-compare. 3880 // We also want to warn about it in -Wconversion. 3881 // So if -Wconversion is off, use a completely identical diagnostic 3882 // in the sign-compare group. 3883 // The conditional-checking code will 3884 if (ICContext) { 3885 DiagID = diag::warn_impcast_integer_sign_conditional; 3886 *ICContext = true; 3887 } 3888 3889 return DiagnoseImpCast(S, E, T, CC, DiagID); 3890 } 3891 3892 // Diagnose conversions between different enumeration types. 3893 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 3894 // type, to give us better diagnostics. 3895 QualType SourceType = E->getType(); 3896 if (!S.getLangOptions().CPlusPlus) { 3897 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 3898 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 3899 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 3900 SourceType = S.Context.getTypeDeclType(Enum); 3901 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 3902 } 3903 } 3904 3905 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 3906 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 3907 if ((SourceEnum->getDecl()->getIdentifier() || 3908 SourceEnum->getDecl()->getTypedefNameForAnonDecl()) && 3909 (TargetEnum->getDecl()->getIdentifier() || 3910 TargetEnum->getDecl()->getTypedefNameForAnonDecl()) && 3911 SourceEnum != TargetEnum) { 3912 if (S.SourceMgr.isInSystemMacro(CC)) 3913 return; 3914 3915 return DiagnoseImpCast(S, E, SourceType, T, CC, 3916 diag::warn_impcast_different_enum_types); 3917 } 3918 3919 return; 3920} 3921 3922void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 3923 3924void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 3925 SourceLocation CC, bool &ICContext) { 3926 E = E->IgnoreParenImpCasts(); 3927 3928 if (isa<ConditionalOperator>(E)) 3929 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 3930 3931 AnalyzeImplicitConversions(S, E, CC); 3932 if (E->getType() != T) 3933 return CheckImplicitConversion(S, E, T, CC, &ICContext); 3934 return; 3935} 3936 3937void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 3938 SourceLocation CC = E->getQuestionLoc(); 3939 3940 AnalyzeImplicitConversions(S, E->getCond(), CC); 3941 3942 bool Suspicious = false; 3943 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 3944 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 3945 3946 // If -Wconversion would have warned about either of the candidates 3947 // for a signedness conversion to the context type... 3948 if (!Suspicious) return; 3949 3950 // ...but it's currently ignored... 3951 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 3952 CC)) 3953 return; 3954 3955 // ...then check whether it would have warned about either of the 3956 // candidates for a signedness conversion to the condition type. 3957 if (E->getType() == T) return; 3958 3959 Suspicious = false; 3960 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 3961 E->getType(), CC, &Suspicious); 3962 if (!Suspicious) 3963 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 3964 E->getType(), CC, &Suspicious); 3965} 3966 3967/// AnalyzeImplicitConversions - Find and report any interesting 3968/// implicit conversions in the given expression. There are a couple 3969/// of competing diagnostics here, -Wconversion and -Wsign-compare. 3970void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 3971 QualType T = OrigE->getType(); 3972 Expr *E = OrigE->IgnoreParenImpCasts(); 3973 3974 if (E->isTypeDependent() || E->isValueDependent()) 3975 return; 3976 3977 // For conditional operators, we analyze the arguments as if they 3978 // were being fed directly into the output. 3979 if (isa<ConditionalOperator>(E)) { 3980 ConditionalOperator *CO = cast<ConditionalOperator>(E); 3981 CheckConditionalOperator(S, CO, T); 3982 return; 3983 } 3984 3985 // Go ahead and check any implicit conversions we might have skipped. 3986 // The non-canonical typecheck is just an optimization; 3987 // CheckImplicitConversion will filter out dead implicit conversions. 3988 if (E->getType() != T) 3989 CheckImplicitConversion(S, E, T, CC); 3990 3991 // Now continue drilling into this expression. 3992 3993 // Skip past explicit casts. 3994 if (isa<ExplicitCastExpr>(E)) { 3995 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 3996 return AnalyzeImplicitConversions(S, E, CC); 3997 } 3998 3999 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 4000 // Do a somewhat different check with comparison operators. 4001 if (BO->isComparisonOp()) 4002 return AnalyzeComparison(S, BO); 4003 4004 // And with assignments and compound assignments. 4005 if (BO->isAssignmentOp()) 4006 return AnalyzeAssignment(S, BO); 4007 } 4008 4009 // These break the otherwise-useful invariant below. Fortunately, 4010 // we don't really need to recurse into them, because any internal 4011 // expressions should have been analyzed already when they were 4012 // built into statements. 4013 if (isa<StmtExpr>(E)) return; 4014 4015 // Don't descend into unevaluated contexts. 4016 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 4017 4018 // Now just recurse over the expression's children. 4019 CC = E->getExprLoc(); 4020 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 4021 bool IsLogicalOperator = BO && BO->isLogicalOp(); 4022 for (Stmt::child_range I = E->children(); I; ++I) { 4023 Expr *ChildExpr = cast<Expr>(*I); 4024 if (IsLogicalOperator && 4025 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 4026 // Ignore checking string literals that are in logical operators. 4027 continue; 4028 AnalyzeImplicitConversions(S, ChildExpr, CC); 4029 } 4030} 4031 4032} // end anonymous namespace 4033 4034/// Diagnoses "dangerous" implicit conversions within the given 4035/// expression (which is a full expression). Implements -Wconversion 4036/// and -Wsign-compare. 4037/// 4038/// \param CC the "context" location of the implicit conversion, i.e. 4039/// the most location of the syntactic entity requiring the implicit 4040/// conversion 4041void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 4042 // Don't diagnose in unevaluated contexts. 4043 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 4044 return; 4045 4046 // Don't diagnose for value- or type-dependent expressions. 4047 if (E->isTypeDependent() || E->isValueDependent()) 4048 return; 4049 4050 // Check for array bounds violations in cases where the check isn't triggered 4051 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 4052 // ArraySubscriptExpr is on the RHS of a variable initialization. 4053 CheckArrayAccess(E); 4054 4055 // This is not the right CC for (e.g.) a variable initialization. 4056 AnalyzeImplicitConversions(*this, E, CC); 4057} 4058 4059void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 4060 FieldDecl *BitField, 4061 Expr *Init) { 4062 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 4063} 4064 4065/// CheckParmsForFunctionDef - Check that the parameters of the given 4066/// function are appropriate for the definition of a function. This 4067/// takes care of any checks that cannot be performed on the 4068/// declaration itself, e.g., that the types of each of the function 4069/// parameters are complete. 4070bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 4071 bool CheckParameterNames) { 4072 bool HasInvalidParm = false; 4073 for (; P != PEnd; ++P) { 4074 ParmVarDecl *Param = *P; 4075 4076 // C99 6.7.5.3p4: the parameters in a parameter type list in a 4077 // function declarator that is part of a function definition of 4078 // that function shall not have incomplete type. 4079 // 4080 // This is also C++ [dcl.fct]p6. 4081 if (!Param->isInvalidDecl() && 4082 RequireCompleteType(Param->getLocation(), Param->getType(), 4083 diag::err_typecheck_decl_incomplete_type)) { 4084 Param->setInvalidDecl(); 4085 HasInvalidParm = true; 4086 } 4087 4088 // C99 6.9.1p5: If the declarator includes a parameter type list, the 4089 // declaration of each parameter shall include an identifier. 4090 if (CheckParameterNames && 4091 Param->getIdentifier() == 0 && 4092 !Param->isImplicit() && 4093 !getLangOptions().CPlusPlus) 4094 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 4095 4096 // C99 6.7.5.3p12: 4097 // If the function declarator is not part of a definition of that 4098 // function, parameters may have incomplete type and may use the [*] 4099 // notation in their sequences of declarator specifiers to specify 4100 // variable length array types. 4101 QualType PType = Param->getOriginalType(); 4102 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 4103 if (AT->getSizeModifier() == ArrayType::Star) { 4104 // FIXME: This diagnosic should point the the '[*]' if source-location 4105 // information is added for it. 4106 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 4107 } 4108 } 4109 } 4110 4111 return HasInvalidParm; 4112} 4113 4114/// CheckCastAlign - Implements -Wcast-align, which warns when a 4115/// pointer cast increases the alignment requirements. 4116void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 4117 // This is actually a lot of work to potentially be doing on every 4118 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 4119 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 4120 TRange.getBegin()) 4121 == DiagnosticsEngine::Ignored) 4122 return; 4123 4124 // Ignore dependent types. 4125 if (T->isDependentType() || Op->getType()->isDependentType()) 4126 return; 4127 4128 // Require that the destination be a pointer type. 4129 const PointerType *DestPtr = T->getAs<PointerType>(); 4130 if (!DestPtr) return; 4131 4132 // If the destination has alignment 1, we're done. 4133 QualType DestPointee = DestPtr->getPointeeType(); 4134 if (DestPointee->isIncompleteType()) return; 4135 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 4136 if (DestAlign.isOne()) return; 4137 4138 // Require that the source be a pointer type. 4139 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 4140 if (!SrcPtr) return; 4141 QualType SrcPointee = SrcPtr->getPointeeType(); 4142 4143 // Whitelist casts from cv void*. We already implicitly 4144 // whitelisted casts to cv void*, since they have alignment 1. 4145 // Also whitelist casts involving incomplete types, which implicitly 4146 // includes 'void'. 4147 if (SrcPointee->isIncompleteType()) return; 4148 4149 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 4150 if (SrcAlign >= DestAlign) return; 4151 4152 Diag(TRange.getBegin(), diag::warn_cast_align) 4153 << Op->getType() << T 4154 << static_cast<unsigned>(SrcAlign.getQuantity()) 4155 << static_cast<unsigned>(DestAlign.getQuantity()) 4156 << TRange << Op->getSourceRange(); 4157} 4158 4159static const Type* getElementType(const Expr *BaseExpr) { 4160 const Type* EltType = BaseExpr->getType().getTypePtr(); 4161 if (EltType->isAnyPointerType()) 4162 return EltType->getPointeeType().getTypePtr(); 4163 else if (EltType->isArrayType()) 4164 return EltType->getBaseElementTypeUnsafe(); 4165 return EltType; 4166} 4167 4168/// \brief Check whether this array fits the idiom of a size-one tail padded 4169/// array member of a struct. 4170/// 4171/// We avoid emitting out-of-bounds access warnings for such arrays as they are 4172/// commonly used to emulate flexible arrays in C89 code. 4173static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 4174 const NamedDecl *ND) { 4175 if (Size != 1 || !ND) return false; 4176 4177 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 4178 if (!FD) return false; 4179 4180 // Don't consider sizes resulting from macro expansions or template argument 4181 // substitution to form C89 tail-padded arrays. 4182 ConstantArrayTypeLoc TL = 4183 cast<ConstantArrayTypeLoc>(FD->getTypeSourceInfo()->getTypeLoc()); 4184 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(TL.getSizeExpr()); 4185 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 4186 return false; 4187 4188 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 4189 if (!RD) return false; 4190 if (RD->isUnion()) return false; 4191 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 4192 if (!CRD->isStandardLayout()) return false; 4193 } 4194 4195 // See if this is the last field decl in the record. 4196 const Decl *D = FD; 4197 while ((D = D->getNextDeclInContext())) 4198 if (isa<FieldDecl>(D)) 4199 return false; 4200 return true; 4201} 4202 4203void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 4204 const ArraySubscriptExpr *ASE, 4205 bool AllowOnePastEnd, bool IndexNegated) { 4206 IndexExpr = IndexExpr->IgnoreParenCasts(); 4207 if (IndexExpr->isValueDependent()) 4208 return; 4209 4210 const Type *EffectiveType = getElementType(BaseExpr); 4211 BaseExpr = BaseExpr->IgnoreParenCasts(); 4212 const ConstantArrayType *ArrayTy = 4213 Context.getAsConstantArrayType(BaseExpr->getType()); 4214 if (!ArrayTy) 4215 return; 4216 4217 llvm::APSInt index; 4218 if (!IndexExpr->EvaluateAsInt(index, Context)) 4219 return; 4220 if (IndexNegated) 4221 index = -index; 4222 4223 const NamedDecl *ND = NULL; 4224 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4225 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 4226 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 4227 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 4228 4229 if (index.isUnsigned() || !index.isNegative()) { 4230 llvm::APInt size = ArrayTy->getSize(); 4231 if (!size.isStrictlyPositive()) 4232 return; 4233 4234 const Type* BaseType = getElementType(BaseExpr); 4235 if (BaseType != EffectiveType) { 4236 // Make sure we're comparing apples to apples when comparing index to size 4237 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 4238 uint64_t array_typesize = Context.getTypeSize(BaseType); 4239 // Handle ptrarith_typesize being zero, such as when casting to void* 4240 if (!ptrarith_typesize) ptrarith_typesize = 1; 4241 if (ptrarith_typesize != array_typesize) { 4242 // There's a cast to a different size type involved 4243 uint64_t ratio = array_typesize / ptrarith_typesize; 4244 // TODO: Be smarter about handling cases where array_typesize is not a 4245 // multiple of ptrarith_typesize 4246 if (ptrarith_typesize * ratio == array_typesize) 4247 size *= llvm::APInt(size.getBitWidth(), ratio); 4248 } 4249 } 4250 4251 if (size.getBitWidth() > index.getBitWidth()) 4252 index = index.sext(size.getBitWidth()); 4253 else if (size.getBitWidth() < index.getBitWidth()) 4254 size = size.sext(index.getBitWidth()); 4255 4256 // For array subscripting the index must be less than size, but for pointer 4257 // arithmetic also allow the index (offset) to be equal to size since 4258 // computing the next address after the end of the array is legal and 4259 // commonly done e.g. in C++ iterators and range-based for loops. 4260 if (AllowOnePastEnd ? index.sle(size) : index.slt(size)) 4261 return; 4262 4263 // Also don't warn for arrays of size 1 which are members of some 4264 // structure. These are often used to approximate flexible arrays in C89 4265 // code. 4266 if (IsTailPaddedMemberArray(*this, size, ND)) 4267 return; 4268 4269 // Suppress the warning if the subscript expression (as identified by the 4270 // ']' location) and the index expression are both from macro expansions 4271 // within a system header. 4272 if (ASE) { 4273 SourceLocation RBracketLoc = SourceMgr.getSpellingLoc( 4274 ASE->getRBracketLoc()); 4275 if (SourceMgr.isInSystemHeader(RBracketLoc)) { 4276 SourceLocation IndexLoc = SourceMgr.getSpellingLoc( 4277 IndexExpr->getLocStart()); 4278 if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc)) 4279 return; 4280 } 4281 } 4282 4283 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 4284 if (ASE) 4285 DiagID = diag::warn_array_index_exceeds_bounds; 4286 4287 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4288 PDiag(DiagID) << index.toString(10, true) 4289 << size.toString(10, true) 4290 << (unsigned)size.getLimitedValue(~0U) 4291 << IndexExpr->getSourceRange()); 4292 } else { 4293 unsigned DiagID = diag::warn_array_index_precedes_bounds; 4294 if (!ASE) { 4295 DiagID = diag::warn_ptr_arith_precedes_bounds; 4296 if (index.isNegative()) index = -index; 4297 } 4298 4299 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4300 PDiag(DiagID) << index.toString(10, true) 4301 << IndexExpr->getSourceRange()); 4302 } 4303 4304 if (!ND) { 4305 // Try harder to find a NamedDecl to point at in the note. 4306 while (const ArraySubscriptExpr *ASE = 4307 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 4308 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 4309 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4310 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 4311 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 4312 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 4313 } 4314 4315 if (ND) 4316 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 4317 PDiag(diag::note_array_index_out_of_bounds) 4318 << ND->getDeclName()); 4319} 4320 4321void Sema::CheckArrayAccess(const Expr *expr) { 4322 int AllowOnePastEnd = 0; 4323 while (expr) { 4324 expr = expr->IgnoreParenImpCasts(); 4325 switch (expr->getStmtClass()) { 4326 case Stmt::ArraySubscriptExprClass: { 4327 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 4328 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE, 4329 AllowOnePastEnd > 0); 4330 return; 4331 } 4332 case Stmt::UnaryOperatorClass: { 4333 // Only unwrap the * and & unary operators 4334 const UnaryOperator *UO = cast<UnaryOperator>(expr); 4335 expr = UO->getSubExpr(); 4336 switch (UO->getOpcode()) { 4337 case UO_AddrOf: 4338 AllowOnePastEnd++; 4339 break; 4340 case UO_Deref: 4341 AllowOnePastEnd--; 4342 break; 4343 default: 4344 return; 4345 } 4346 break; 4347 } 4348 case Stmt::ConditionalOperatorClass: { 4349 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 4350 if (const Expr *lhs = cond->getLHS()) 4351 CheckArrayAccess(lhs); 4352 if (const Expr *rhs = cond->getRHS()) 4353 CheckArrayAccess(rhs); 4354 return; 4355 } 4356 default: 4357 return; 4358 } 4359 } 4360} 4361 4362//===--- CHECK: Objective-C retain cycles ----------------------------------// 4363 4364namespace { 4365 struct RetainCycleOwner { 4366 RetainCycleOwner() : Variable(0), Indirect(false) {} 4367 VarDecl *Variable; 4368 SourceRange Range; 4369 SourceLocation Loc; 4370 bool Indirect; 4371 4372 void setLocsFrom(Expr *e) { 4373 Loc = e->getExprLoc(); 4374 Range = e->getSourceRange(); 4375 } 4376 }; 4377} 4378 4379/// Consider whether capturing the given variable can possibly lead to 4380/// a retain cycle. 4381static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 4382 // In ARC, it's captured strongly iff the variable has __strong 4383 // lifetime. In MRR, it's captured strongly if the variable is 4384 // __block and has an appropriate type. 4385 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4386 return false; 4387 4388 owner.Variable = var; 4389 owner.setLocsFrom(ref); 4390 return true; 4391} 4392 4393static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) { 4394 while (true) { 4395 e = e->IgnoreParens(); 4396 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 4397 switch (cast->getCastKind()) { 4398 case CK_BitCast: 4399 case CK_LValueBitCast: 4400 case CK_LValueToRValue: 4401 case CK_ARCReclaimReturnedObject: 4402 e = cast->getSubExpr(); 4403 continue; 4404 4405 default: 4406 return false; 4407 } 4408 } 4409 4410 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 4411 ObjCIvarDecl *ivar = ref->getDecl(); 4412 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4413 return false; 4414 4415 // Try to find a retain cycle in the base. 4416 if (!findRetainCycleOwner(S, ref->getBase(), owner)) 4417 return false; 4418 4419 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 4420 owner.Indirect = true; 4421 return true; 4422 } 4423 4424 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 4425 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 4426 if (!var) return false; 4427 return considerVariable(var, ref, owner); 4428 } 4429 4430 if (BlockDeclRefExpr *ref = dyn_cast<BlockDeclRefExpr>(e)) { 4431 owner.Variable = ref->getDecl(); 4432 owner.setLocsFrom(ref); 4433 return true; 4434 } 4435 4436 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 4437 if (member->isArrow()) return false; 4438 4439 // Don't count this as an indirect ownership. 4440 e = member->getBase(); 4441 continue; 4442 } 4443 4444 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 4445 // Only pay attention to pseudo-objects on property references. 4446 ObjCPropertyRefExpr *pre 4447 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 4448 ->IgnoreParens()); 4449 if (!pre) return false; 4450 if (pre->isImplicitProperty()) return false; 4451 ObjCPropertyDecl *property = pre->getExplicitProperty(); 4452 if (!property->isRetaining() && 4453 !(property->getPropertyIvarDecl() && 4454 property->getPropertyIvarDecl()->getType() 4455 .getObjCLifetime() == Qualifiers::OCL_Strong)) 4456 return false; 4457 4458 owner.Indirect = true; 4459 if (pre->isSuperReceiver()) { 4460 owner.Variable = S.getCurMethodDecl()->getSelfDecl(); 4461 if (!owner.Variable) 4462 return false; 4463 owner.Loc = pre->getLocation(); 4464 owner.Range = pre->getSourceRange(); 4465 return true; 4466 } 4467 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 4468 ->getSourceExpr()); 4469 continue; 4470 } 4471 4472 // Array ivars? 4473 4474 return false; 4475 } 4476} 4477 4478namespace { 4479 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 4480 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 4481 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 4482 Variable(variable), Capturer(0) {} 4483 4484 VarDecl *Variable; 4485 Expr *Capturer; 4486 4487 void VisitDeclRefExpr(DeclRefExpr *ref) { 4488 if (ref->getDecl() == Variable && !Capturer) 4489 Capturer = ref; 4490 } 4491 4492 void VisitBlockDeclRefExpr(BlockDeclRefExpr *ref) { 4493 if (ref->getDecl() == Variable && !Capturer) 4494 Capturer = ref; 4495 } 4496 4497 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 4498 if (Capturer) return; 4499 Visit(ref->getBase()); 4500 if (Capturer && ref->isFreeIvar()) 4501 Capturer = ref; 4502 } 4503 4504 void VisitBlockExpr(BlockExpr *block) { 4505 // Look inside nested blocks 4506 if (block->getBlockDecl()->capturesVariable(Variable)) 4507 Visit(block->getBlockDecl()->getBody()); 4508 } 4509 }; 4510} 4511 4512/// Check whether the given argument is a block which captures a 4513/// variable. 4514static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 4515 assert(owner.Variable && owner.Loc.isValid()); 4516 4517 e = e->IgnoreParenCasts(); 4518 BlockExpr *block = dyn_cast<BlockExpr>(e); 4519 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 4520 return 0; 4521 4522 FindCaptureVisitor visitor(S.Context, owner.Variable); 4523 visitor.Visit(block->getBlockDecl()->getBody()); 4524 return visitor.Capturer; 4525} 4526 4527static void diagnoseRetainCycle(Sema &S, Expr *capturer, 4528 RetainCycleOwner &owner) { 4529 assert(capturer); 4530 assert(owner.Variable && owner.Loc.isValid()); 4531 4532 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 4533 << owner.Variable << capturer->getSourceRange(); 4534 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 4535 << owner.Indirect << owner.Range; 4536} 4537 4538/// Check for a keyword selector that starts with the word 'add' or 4539/// 'set'. 4540static bool isSetterLikeSelector(Selector sel) { 4541 if (sel.isUnarySelector()) return false; 4542 4543 StringRef str = sel.getNameForSlot(0); 4544 while (!str.empty() && str.front() == '_') str = str.substr(1); 4545 if (str.startswith("set")) 4546 str = str.substr(3); 4547 else if (str.startswith("add")) { 4548 // Specially whitelist 'addOperationWithBlock:'. 4549 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 4550 return false; 4551 str = str.substr(3); 4552 } 4553 else 4554 return false; 4555 4556 if (str.empty()) return true; 4557 return !islower(str.front()); 4558} 4559 4560/// Check a message send to see if it's likely to cause a retain cycle. 4561void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 4562 // Only check instance methods whose selector looks like a setter. 4563 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 4564 return; 4565 4566 // Try to find a variable that the receiver is strongly owned by. 4567 RetainCycleOwner owner; 4568 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 4569 if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner)) 4570 return; 4571 } else { 4572 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 4573 owner.Variable = getCurMethodDecl()->getSelfDecl(); 4574 owner.Loc = msg->getSuperLoc(); 4575 owner.Range = msg->getSuperLoc(); 4576 } 4577 4578 // Check whether the receiver is captured by any of the arguments. 4579 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 4580 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 4581 return diagnoseRetainCycle(*this, capturer, owner); 4582} 4583 4584/// Check a property assign to see if it's likely to cause a retain cycle. 4585void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 4586 RetainCycleOwner owner; 4587 if (!findRetainCycleOwner(*this, receiver, owner)) 4588 return; 4589 4590 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 4591 diagnoseRetainCycle(*this, capturer, owner); 4592} 4593 4594bool Sema::checkUnsafeAssigns(SourceLocation Loc, 4595 QualType LHS, Expr *RHS) { 4596 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 4597 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 4598 return false; 4599 // strip off any implicit cast added to get to the one arc-specific 4600 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4601 if (cast->getCastKind() == CK_ARCConsumeObject) { 4602 Diag(Loc, diag::warn_arc_retained_assign) 4603 << (LT == Qualifiers::OCL_ExplicitNone) 4604 << RHS->getSourceRange(); 4605 return true; 4606 } 4607 RHS = cast->getSubExpr(); 4608 } 4609 return false; 4610} 4611 4612void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 4613 Expr *LHS, Expr *RHS) { 4614 QualType LHSType = LHS->getType(); 4615 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 4616 return; 4617 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 4618 // FIXME. Check for other life times. 4619 if (LT != Qualifiers::OCL_None) 4620 return; 4621 4622 if (ObjCPropertyRefExpr *PRE 4623 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens())) { 4624 if (PRE->isImplicitProperty()) 4625 return; 4626 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 4627 if (!PD) 4628 return; 4629 4630 unsigned Attributes = PD->getPropertyAttributes(); 4631 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) 4632 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4633 if (cast->getCastKind() == CK_ARCConsumeObject) { 4634 Diag(Loc, diag::warn_arc_retained_property_assign) 4635 << RHS->getSourceRange(); 4636 return; 4637 } 4638 RHS = cast->getSubExpr(); 4639 } 4640 } 4641} 4642