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> 39 using namespace clang; 40 using namespace sema; 41 42 SourceLocation 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? 51 bool 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. 74 static 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. 94 static 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 } 104 105 ExprResult 106 Sema::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_sub: 187 case Builtin::BI__sync_fetch_and_or: 188 case Builtin::BI__sync_fetch_and_and: 189 case Builtin::BI__sync_fetch_and_xor: 190 case Builtin::BI__sync_add_and_fetch: 191 case Builtin::BI__sync_sub_and_fetch: 192 case Builtin::BI__sync_and_and_fetch: 193 case Builtin::BI__sync_or_and_fetch: 194 case Builtin::BI__sync_xor_and_fetch: 195 case Builtin::BI__sync_val_compare_and_swap: 196 case Builtin::BI__sync_bool_compare_and_swap: 197 case Builtin::BI__sync_lock_test_and_set: 198 case Builtin::BI__sync_lock_release: 199 case Builtin::BI__sync_swap: 200 return SemaBuiltinAtomicOverloaded(move(TheCallResult)); 201 case Builtin::BI__atomic_load: 202 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Load); 203 case Builtin::BI__atomic_store: 204 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Store); 205 case Builtin::BI__atomic_exchange: 206 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Xchg); 207 case Builtin::BI__atomic_compare_exchange_strong: 208 return SemaAtomicOpsOverloaded(move(TheCallResult), 209 AtomicExpr::CmpXchgStrong); 210 case Builtin::BI__atomic_compare_exchange_weak: 211 return SemaAtomicOpsOverloaded(move(TheCallResult), 212 AtomicExpr::CmpXchgWeak); 213 case Builtin::BI__atomic_fetch_add: 214 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Add); 215 case Builtin::BI__atomic_fetch_sub: 216 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Sub); 217 case Builtin::BI__atomic_fetch_and: 218 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::And); 219 case Builtin::BI__atomic_fetch_or: 220 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Or); 221 case Builtin::BI__atomic_fetch_xor: 222 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Xor); 223 case Builtin::BI__builtin_annotation: 224 if (CheckBuiltinAnnotationString(*this, TheCall->getArg(1))) 225 return ExprError(); 226 break; 227 } 228 229 // Since the target specific builtins for each arch overlap, only check those 230 // of the arch we are compiling for. 231 if (BuiltinID >= Builtin::FirstTSBuiltin) { 232 switch (Context.getTargetInfo().getTriple().getArch()) { 233 case llvm::Triple::arm: 234 case llvm::Triple::thumb: 235 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 236 return ExprError(); 237 break; 238 default: 239 break; 240 } 241 } 242 243 return move(TheCallResult); 244 } 245 246 // Get the valid immediate range for the specified NEON type code. 247 static unsigned RFT(unsigned t, bool shift = false) { 248 bool quad = t & 0x10; 249 250 switch (t & 0x7) { 251 case 0: // i8 252 return shift ? 7 : (8 << (int)quad) - 1; 253 case 1: // i16 254 return shift ? 15 : (4 << (int)quad) - 1; 255 case 2: // i32 256 return shift ? 31 : (2 << (int)quad) - 1; 257 case 3: // i64 258 return shift ? 63 : (1 << (int)quad) - 1; 259 case 4: // f32 260 assert(!shift && "cannot shift float types!"); 261 return (2 << (int)quad) - 1; 262 case 5: // poly8 263 return shift ? 7 : (8 << (int)quad) - 1; 264 case 6: // poly16 265 return shift ? 15 : (4 << (int)quad) - 1; 266 case 7: // float16 267 assert(!shift && "cannot shift float types!"); 268 return (4 << (int)quad) - 1; 269 } 270 return 0; 271 } 272 273 bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 274 llvm::APSInt Result; 275 276 unsigned mask = 0; 277 unsigned TV = 0; 278 switch (BuiltinID) { 279 #define GET_NEON_OVERLOAD_CHECK 280 #include "clang/Basic/arm_neon.inc" 281 #undef GET_NEON_OVERLOAD_CHECK 282 } 283 284 // For NEON intrinsics which are overloaded on vector element type, validate 285 // the immediate which specifies which variant to emit. 286 if (mask) { 287 unsigned ArgNo = TheCall->getNumArgs()-1; 288 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 289 return true; 290 291 TV = Result.getLimitedValue(32); 292 if ((TV > 31) || (mask & (1 << TV)) == 0) 293 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 294 << TheCall->getArg(ArgNo)->getSourceRange(); 295 } 296 297 // For NEON intrinsics which take an immediate value as part of the 298 // instruction, range check them here. 299 unsigned i = 0, l = 0, u = 0; 300 switch (BuiltinID) { 301 default: return false; 302 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 303 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 304 case ARM::BI__builtin_arm_vcvtr_f: 305 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 306 #define GET_NEON_IMMEDIATE_CHECK 307 #include "clang/Basic/arm_neon.inc" 308 #undef GET_NEON_IMMEDIATE_CHECK 309 }; 310 311 // Check that the immediate argument is actually a constant. 312 if (SemaBuiltinConstantArg(TheCall, i, Result)) 313 return true; 314 315 // Range check against the upper/lower values for this isntruction. 316 unsigned Val = Result.getZExtValue(); 317 if (Val < l || Val > (u + l)) 318 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 319 << l << u+l << TheCall->getArg(i)->getSourceRange(); 320 321 // FIXME: VFP Intrinsics should error if VFP not present. 322 return false; 323 } 324 325 /// CheckFunctionCall - Check a direct function call for various correctness 326 /// and safety properties not strictly enforced by the C type system. 327 bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 328 // Get the IdentifierInfo* for the called function. 329 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 330 331 // None of the checks below are needed for functions that don't have 332 // simple names (e.g., C++ conversion functions). 333 if (!FnInfo) 334 return false; 335 336 // FIXME: This mechanism should be abstracted to be less fragile and 337 // more efficient. For example, just map function ids to custom 338 // handlers. 339 340 // Printf and scanf checking. 341 for (specific_attr_iterator<FormatAttr> 342 i = FDecl->specific_attr_begin<FormatAttr>(), 343 e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { 344 345 const FormatAttr *Format = *i; 346 const bool b = Format->getType() == "scanf"; 347 if (b || CheckablePrintfAttr(Format, TheCall)) { 348 bool HasVAListArg = Format->getFirstArg() == 0; 349 CheckPrintfScanfArguments(TheCall, HasVAListArg, 350 Format->getFormatIdx() - 1, 351 HasVAListArg ? 0 : Format->getFirstArg() - 1, 352 !b); 353 } 354 } 355 356 for (specific_attr_iterator<NonNullAttr> 357 i = FDecl->specific_attr_begin<NonNullAttr>(), 358 e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) { 359 CheckNonNullArguments(*i, TheCall->getArgs(), 360 TheCall->getCallee()->getLocStart()); 361 } 362 363 // Builtin handling 364 int CMF = -1; 365 switch (FDecl->getBuiltinID()) { 366 case Builtin::BI__builtin_memset: 367 case Builtin::BI__builtin___memset_chk: 368 case Builtin::BImemset: 369 CMF = CMF_Memset; 370 break; 371 372 case Builtin::BI__builtin_memcpy: 373 case Builtin::BI__builtin___memcpy_chk: 374 case Builtin::BImemcpy: 375 CMF = CMF_Memcpy; 376 break; 377 378 case Builtin::BI__builtin_memmove: 379 case Builtin::BI__builtin___memmove_chk: 380 case Builtin::BImemmove: 381 CMF = CMF_Memmove; 382 break; 383 384 case Builtin::BIstrlcpy: 385 case Builtin::BIstrlcat: 386 CheckStrlcpycatArguments(TheCall, FnInfo); 387 break; 388 389 case Builtin::BI__builtin_memcmp: 390 CMF = CMF_Memcmp; 391 break; 392 393 case Builtin::BI__builtin_strncpy: 394 case Builtin::BI__builtin___strncpy_chk: 395 case Builtin::BIstrncpy: 396 CMF = CMF_Strncpy; 397 break; 398 399 case Builtin::BI__builtin_strncmp: 400 CMF = CMF_Strncmp; 401 break; 402 403 case Builtin::BI__builtin_strncasecmp: 404 CMF = CMF_Strncasecmp; 405 break; 406 407 case Builtin::BI__builtin_strncat: 408 case Builtin::BIstrncat: 409 CMF = CMF_Strncat; 410 break; 411 412 case Builtin::BI__builtin_strndup: 413 case Builtin::BIstrndup: 414 CMF = CMF_Strndup; 415 break; 416 417 default: 418 if (FDecl->getLinkage() == ExternalLinkage && 419 (!getLangOptions().CPlusPlus || FDecl->isExternC())) { 420 if (FnInfo->isStr("memset")) 421 CMF = CMF_Memset; 422 else if (FnInfo->isStr("memcpy")) 423 CMF = CMF_Memcpy; 424 else if (FnInfo->isStr("memmove")) 425 CMF = CMF_Memmove; 426 else if (FnInfo->isStr("memcmp")) 427 CMF = CMF_Memcmp; 428 else if (FnInfo->isStr("strncpy")) 429 CMF = CMF_Strncpy; 430 else if (FnInfo->isStr("strncmp")) 431 CMF = CMF_Strncmp; 432 else if (FnInfo->isStr("strncasecmp")) 433 CMF = CMF_Strncasecmp; 434 else if (FnInfo->isStr("strncat")) 435 CMF = CMF_Strncat; 436 else if (FnInfo->isStr("strndup")) 437 CMF = CMF_Strndup; 438 } 439 break; 440 } 441 442 // Memset/memcpy/memmove handling 443 if (CMF != -1) 444 CheckMemaccessArguments(TheCall, CheckedMemoryFunction(CMF), FnInfo); 445 446 return false; 447 } 448 449 bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 450 // Printf checking. 451 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 452 if (!Format) 453 return false; 454 455 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 456 if (!V) 457 return false; 458 459 QualType Ty = V->getType(); 460 if (!Ty->isBlockPointerType()) 461 return false; 462 463 const bool b = Format->getType() == "scanf"; 464 if (!b && !CheckablePrintfAttr(Format, TheCall)) 465 return false; 466 467 bool HasVAListArg = Format->getFirstArg() == 0; 468 CheckPrintfScanfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 469 HasVAListArg ? 0 : Format->getFirstArg() - 1, !b); 470 471 return false; 472 } 473 474 ExprResult 475 Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, AtomicExpr::AtomicOp Op) { 476 CallExpr *TheCall = cast<CallExpr>(TheCallResult.get()); 477 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 478 479 // All these operations take one of the following four forms: 480 // T __atomic_load(_Atomic(T)*, int) (loads) 481 // T* __atomic_add(_Atomic(T*)*, ptrdiff_t, int) (pointer add/sub) 482 // int __atomic_compare_exchange_strong(_Atomic(T)*, T*, T, int, int) 483 // (cmpxchg) 484 // T __atomic_exchange(_Atomic(T)*, T, int) (everything else) 485 // where T is an appropriate type, and the int paremeterss are for orderings. 486 unsigned NumVals = 1; 487 unsigned NumOrders = 1; 488 if (Op == AtomicExpr::Load) { 489 NumVals = 0; 490 } else if (Op == AtomicExpr::CmpXchgWeak || Op == AtomicExpr::CmpXchgStrong) { 491 NumVals = 2; 492 NumOrders = 2; 493 } 494 495 if (TheCall->getNumArgs() < NumVals+NumOrders+1) { 496 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 497 << 0 << NumVals+NumOrders+1 << TheCall->getNumArgs() 498 << TheCall->getCallee()->getSourceRange(); 499 return ExprError(); 500 } else if (TheCall->getNumArgs() > NumVals+NumOrders+1) { 501 Diag(TheCall->getArg(NumVals+NumOrders+1)->getLocStart(), 502 diag::err_typecheck_call_too_many_args) 503 << 0 << NumVals+NumOrders+1 << TheCall->getNumArgs() 504 << TheCall->getCallee()->getSourceRange(); 505 return ExprError(); 506 } 507 508 // Inspect the first argument of the atomic operation. This should always be 509 // a pointer to an _Atomic type. 510 Expr *Ptr = TheCall->getArg(0); 511 Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get(); 512 const PointerType *pointerType = Ptr->getType()->getAs<PointerType>(); 513 if (!pointerType) { 514 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 515 << Ptr->getType() << Ptr->getSourceRange(); 516 return ExprError(); 517 } 518 519 QualType AtomTy = pointerType->getPointeeType(); 520 if (!AtomTy->isAtomicType()) { 521 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 522 << Ptr->getType() << Ptr->getSourceRange(); 523 return ExprError(); 524 } 525 QualType ValType = AtomTy->getAs<AtomicType>()->getValueType(); 526 527 if ((Op == AtomicExpr::Add || Op == AtomicExpr::Sub) && 528 !ValType->isIntegerType() && !ValType->isPointerType()) { 529 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 530 << Ptr->getType() << Ptr->getSourceRange(); 531 return ExprError(); 532 } 533 534 if (!ValType->isIntegerType() && 535 (Op == AtomicExpr::And || Op == AtomicExpr::Or || Op == AtomicExpr::Xor)){ 536 Diag(DRE->getLocStart(), diag::err_atomic_op_logical_needs_atomic_int) 537 << Ptr->getType() << Ptr->getSourceRange(); 538 return ExprError(); 539 } 540 541 switch (ValType.getObjCLifetime()) { 542 case Qualifiers::OCL_None: 543 case Qualifiers::OCL_ExplicitNone: 544 // okay 545 break; 546 547 case Qualifiers::OCL_Weak: 548 case Qualifiers::OCL_Strong: 549 case Qualifiers::OCL_Autoreleasing: 550 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 551 << ValType << Ptr->getSourceRange(); 552 return ExprError(); 553 } 554 555 QualType ResultType = ValType; 556 if (Op == AtomicExpr::Store) 557 ResultType = Context.VoidTy; 558 else if (Op == AtomicExpr::CmpXchgWeak || Op == AtomicExpr::CmpXchgStrong) 559 ResultType = Context.BoolTy; 560 561 // The first argument --- the pointer --- has a fixed type; we 562 // deduce the types of the rest of the arguments accordingly. Walk 563 // the remaining arguments, converting them to the deduced value type. 564 for (unsigned i = 1; i != NumVals+NumOrders+1; ++i) { 565 ExprResult Arg = TheCall->getArg(i); 566 QualType Ty; 567 if (i < NumVals+1) { 568 // The second argument to a cmpxchg is a pointer to the data which will 569 // be exchanged. The second argument to a pointer add/subtract is the 570 // amount to add/subtract, which must be a ptrdiff_t. The third 571 // argument to a cmpxchg and the second argument in all other cases 572 // is the type of the value. 573 if (i == 1 && (Op == AtomicExpr::CmpXchgWeak || 574 Op == AtomicExpr::CmpXchgStrong)) 575 Ty = Context.getPointerType(ValType.getUnqualifiedType()); 576 else if (!ValType->isIntegerType() && 577 (Op == AtomicExpr::Add || Op == AtomicExpr::Sub)) 578 Ty = Context.getPointerDiffType(); 579 else 580 Ty = ValType; 581 } else { 582 // The order(s) are always converted to int. 583 Ty = Context.IntTy; 584 } 585 InitializedEntity Entity = 586 InitializedEntity::InitializeParameter(Context, Ty, false); 587 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 588 if (Arg.isInvalid()) 589 return true; 590 TheCall->setArg(i, Arg.get()); 591 } 592 593 SmallVector<Expr*, 5> SubExprs; 594 SubExprs.push_back(Ptr); 595 if (Op == AtomicExpr::Load) { 596 SubExprs.push_back(TheCall->getArg(1)); // Order 597 } else if (Op != AtomicExpr::CmpXchgWeak && Op != AtomicExpr::CmpXchgStrong) { 598 SubExprs.push_back(TheCall->getArg(2)); // Order 599 SubExprs.push_back(TheCall->getArg(1)); // Val1 600 } else { 601 SubExprs.push_back(TheCall->getArg(3)); // Order 602 SubExprs.push_back(TheCall->getArg(1)); // Val1 603 SubExprs.push_back(TheCall->getArg(2)); // Val2 604 SubExprs.push_back(TheCall->getArg(4)); // OrderFail 605 } 606 607 return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 608 SubExprs.data(), SubExprs.size(), 609 ResultType, Op, 610 TheCall->getRParenLoc())); 611 } 612 613 614 /// checkBuiltinArgument - Given a call to a builtin function, perform 615 /// normal type-checking on the given argument, updating the call in 616 /// place. This is useful when a builtin function requires custom 617 /// type-checking for some of its arguments but not necessarily all of 618 /// them. 619 /// 620 /// Returns true on error. 621 static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 622 FunctionDecl *Fn = E->getDirectCallee(); 623 assert(Fn && "builtin call without direct callee!"); 624 625 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 626 InitializedEntity Entity = 627 InitializedEntity::InitializeParameter(S.Context, Param); 628 629 ExprResult Arg = E->getArg(0); 630 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 631 if (Arg.isInvalid()) 632 return true; 633 634 E->setArg(ArgIndex, Arg.take()); 635 return false; 636 } 637 638 /// SemaBuiltinAtomicOverloaded - We have a call to a function like 639 /// __sync_fetch_and_add, which is an overloaded function based on the pointer 640 /// type of its first argument. The main ActOnCallExpr routines have already 641 /// promoted the types of arguments because all of these calls are prototyped as 642 /// void(...). 643 /// 644 /// This function goes through and does final semantic checking for these 645 /// builtins, 646 ExprResult 647 Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 648 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 649 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 650 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 651 652 // Ensure that we have at least one argument to do type inference from. 653 if (TheCall->getNumArgs() < 1) { 654 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 655 << 0 << 1 << TheCall->getNumArgs() 656 << TheCall->getCallee()->getSourceRange(); 657 return ExprError(); 658 } 659 660 // Inspect the first argument of the atomic builtin. This should always be 661 // a pointer type, whose element is an integral scalar or pointer type. 662 // Because it is a pointer type, we don't have to worry about any implicit 663 // casts here. 664 // FIXME: We don't allow floating point scalars as input. 665 Expr *FirstArg = TheCall->getArg(0); 666 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 667 if (!pointerType) { 668 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 669 << FirstArg->getType() << FirstArg->getSourceRange(); 670 return ExprError(); 671 } 672 673 QualType ValType = pointerType->getPointeeType(); 674 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 675 !ValType->isBlockPointerType()) { 676 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 677 << FirstArg->getType() << FirstArg->getSourceRange(); 678 return ExprError(); 679 } 680 681 switch (ValType.getObjCLifetime()) { 682 case Qualifiers::OCL_None: 683 case Qualifiers::OCL_ExplicitNone: 684 // okay 685 break; 686 687 case Qualifiers::OCL_Weak: 688 case Qualifiers::OCL_Strong: 689 case Qualifiers::OCL_Autoreleasing: 690 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 691 << ValType << FirstArg->getSourceRange(); 692 return ExprError(); 693 } 694 695 // Strip any qualifiers off ValType. 696 ValType = ValType.getUnqualifiedType(); 697 698 // The majority of builtins return a value, but a few have special return 699 // types, so allow them to override appropriately below. 700 QualType ResultType = ValType; 701 702 // We need to figure out which concrete builtin this maps onto. For example, 703 // __sync_fetch_and_add with a 2 byte object turns into 704 // __sync_fetch_and_add_2. 705 #define BUILTIN_ROW(x) \ 706 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 707 Builtin::BI##x##_8, Builtin::BI##x##_16 } 708 709 static const unsigned BuiltinIndices[][5] = { 710 BUILTIN_ROW(__sync_fetch_and_add), 711 BUILTIN_ROW(__sync_fetch_and_sub), 712 BUILTIN_ROW(__sync_fetch_and_or), 713 BUILTIN_ROW(__sync_fetch_and_and), 714 BUILTIN_ROW(__sync_fetch_and_xor), 715 716 BUILTIN_ROW(__sync_add_and_fetch), 717 BUILTIN_ROW(__sync_sub_and_fetch), 718 BUILTIN_ROW(__sync_and_and_fetch), 719 BUILTIN_ROW(__sync_or_and_fetch), 720 BUILTIN_ROW(__sync_xor_and_fetch), 721 722 BUILTIN_ROW(__sync_val_compare_and_swap), 723 BUILTIN_ROW(__sync_bool_compare_and_swap), 724 BUILTIN_ROW(__sync_lock_test_and_set), 725 BUILTIN_ROW(__sync_lock_release), 726 BUILTIN_ROW(__sync_swap) 727 }; 728 #undef BUILTIN_ROW 729 730 // Determine the index of the size. 731 unsigned SizeIndex; 732 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 733 case 1: SizeIndex = 0; break; 734 case 2: SizeIndex = 1; break; 735 case 4: SizeIndex = 2; break; 736 case 8: SizeIndex = 3; break; 737 case 16: SizeIndex = 4; break; 738 default: 739 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 740 << FirstArg->getType() << FirstArg->getSourceRange(); 741 return ExprError(); 742 } 743 744 // Each of these builtins has one pointer argument, followed by some number of 745 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 746 // that we ignore. Find out which row of BuiltinIndices to read from as well 747 // as the number of fixed args. 748 unsigned BuiltinID = FDecl->getBuiltinID(); 749 unsigned BuiltinIndex, NumFixed = 1; 750 switch (BuiltinID) { 751 default: llvm_unreachable("Unknown overloaded atomic builtin!"); 752 case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break; 753 case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break; 754 case Builtin::BI__sync_fetch_and_or: BuiltinIndex = 2; break; 755 case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break; 756 case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break; 757 758 case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break; 759 case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break; 760 case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break; 761 case Builtin::BI__sync_or_and_fetch: BuiltinIndex = 8; break; 762 case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break; 763 764 case Builtin::BI__sync_val_compare_and_swap: 765 BuiltinIndex = 10; 766 NumFixed = 2; 767 break; 768 case Builtin::BI__sync_bool_compare_and_swap: 769 BuiltinIndex = 11; 770 NumFixed = 2; 771 ResultType = Context.BoolTy; 772 break; 773 case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break; 774 case Builtin::BI__sync_lock_release: 775 BuiltinIndex = 13; 776 NumFixed = 0; 777 ResultType = Context.VoidTy; 778 break; 779 case Builtin::BI__sync_swap: BuiltinIndex = 14; break; 780 } 781 782 // Now that we know how many fixed arguments we expect, first check that we 783 // have at least that many. 784 if (TheCall->getNumArgs() < 1+NumFixed) { 785 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 786 << 0 << 1+NumFixed << TheCall->getNumArgs() 787 << TheCall->getCallee()->getSourceRange(); 788 return ExprError(); 789 } 790 791 // Get the decl for the concrete builtin from this, we can tell what the 792 // concrete integer type we should convert to is. 793 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 794 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 795 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 796 FunctionDecl *NewBuiltinDecl = 797 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 798 TUScope, false, DRE->getLocStart())); 799 800 // The first argument --- the pointer --- has a fixed type; we 801 // deduce the types of the rest of the arguments accordingly. Walk 802 // the remaining arguments, converting them to the deduced value type. 803 for (unsigned i = 0; i != NumFixed; ++i) { 804 ExprResult Arg = TheCall->getArg(i+1); 805 806 // If the argument is an implicit cast, then there was a promotion due to 807 // "...", just remove it now. 808 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg.get())) { 809 Arg = ICE->getSubExpr(); 810 ICE->setSubExpr(0); 811 TheCall->setArg(i+1, Arg.get()); 812 } 813 814 // GCC does an implicit conversion to the pointer or integer ValType. This 815 // can fail in some cases (1i -> int**), check for this error case now. 816 // Initialize the argument. 817 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context, 818 ValType, /*consume*/ false); 819 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 820 if (Arg.isInvalid()) 821 return ExprError(); 822 823 // Okay, we have something that *can* be converted to the right type. Check 824 // to see if there is a potentially weird extension going on here. This can 825 // happen when you do an atomic operation on something like an char* and 826 // pass in 42. The 42 gets converted to char. This is even more strange 827 // for things like 45.123 -> char, etc. 828 // FIXME: Do this check. 829 TheCall->setArg(i+1, Arg.take()); 830 } 831 832 ASTContext& Context = this->getASTContext(); 833 834 // Create a new DeclRefExpr to refer to the new decl. 835 DeclRefExpr* NewDRE = DeclRefExpr::Create( 836 Context, 837 DRE->getQualifierLoc(), 838 NewBuiltinDecl, 839 DRE->getLocation(), 840 NewBuiltinDecl->getType(), 841 DRE->getValueKind()); 842 843 // Set the callee in the CallExpr. 844 // FIXME: This leaks the original parens and implicit casts. 845 ExprResult PromotedCall = UsualUnaryConversions(NewDRE); 846 if (PromotedCall.isInvalid()) 847 return ExprError(); 848 TheCall->setCallee(PromotedCall.take()); 849 850 // Change the result type of the call to match the original value type. This 851 // is arbitrary, but the codegen for these builtins ins design to handle it 852 // gracefully. 853 TheCall->setType(ResultType); 854 855 return move(TheCallResult); 856 } 857 858 /// CheckObjCString - Checks that the argument to the builtin 859 /// CFString constructor is correct 860 /// Note: It might also make sense to do the UTF-16 conversion here (would 861 /// simplify the backend). 862 bool Sema::CheckObjCString(Expr *Arg) { 863 Arg = Arg->IgnoreParenCasts(); 864 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 865 866 if (!Literal || !Literal->isAscii()) { 867 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 868 << Arg->getSourceRange(); 869 return true; 870 } 871 872 if (Literal->containsNonAsciiOrNull()) { 873 StringRef String = Literal->getString(); 874 unsigned NumBytes = String.size(); 875 SmallVector<UTF16, 128> ToBuf(NumBytes); 876 const UTF8 *FromPtr = (UTF8 *)String.data(); 877 UTF16 *ToPtr = &ToBuf[0]; 878 879 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 880 &ToPtr, ToPtr + NumBytes, 881 strictConversion); 882 // Check for conversion failure. 883 if (Result != conversionOK) 884 Diag(Arg->getLocStart(), 885 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 886 } 887 return false; 888 } 889 890 /// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 891 /// Emit an error and return true on failure, return false on success. 892 bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 893 Expr *Fn = TheCall->getCallee(); 894 if (TheCall->getNumArgs() > 2) { 895 Diag(TheCall->getArg(2)->getLocStart(), 896 diag::err_typecheck_call_too_many_args) 897 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 898 << Fn->getSourceRange() 899 << SourceRange(TheCall->getArg(2)->getLocStart(), 900 (*(TheCall->arg_end()-1))->getLocEnd()); 901 return true; 902 } 903 904 if (TheCall->getNumArgs() < 2) { 905 return Diag(TheCall->getLocEnd(), 906 diag::err_typecheck_call_too_few_args_at_least) 907 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 908 } 909 910 // Type-check the first argument normally. 911 if (checkBuiltinArgument(*this, TheCall, 0)) 912 return true; 913 914 // Determine whether the current function is variadic or not. 915 BlockScopeInfo *CurBlock = getCurBlock(); 916 bool isVariadic; 917 if (CurBlock) 918 isVariadic = CurBlock->TheDecl->isVariadic(); 919 else if (FunctionDecl *FD = getCurFunctionDecl()) 920 isVariadic = FD->isVariadic(); 921 else 922 isVariadic = getCurMethodDecl()->isVariadic(); 923 924 if (!isVariadic) { 925 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 926 return true; 927 } 928 929 // Verify that the second argument to the builtin is the last argument of the 930 // current function or method. 931 bool SecondArgIsLastNamedArgument = false; 932 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 933 934 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 935 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 936 // FIXME: This isn't correct for methods (results in bogus warning). 937 // Get the last formal in the current function. 938 const ParmVarDecl *LastArg; 939 if (CurBlock) 940 LastArg = *(CurBlock->TheDecl->param_end()-1); 941 else if (FunctionDecl *FD = getCurFunctionDecl()) 942 LastArg = *(FD->param_end()-1); 943 else 944 LastArg = *(getCurMethodDecl()->param_end()-1); 945 SecondArgIsLastNamedArgument = PV == LastArg; 946 } 947 } 948 949 if (!SecondArgIsLastNamedArgument) 950 Diag(TheCall->getArg(1)->getLocStart(), 951 diag::warn_second_parameter_of_va_start_not_last_named_argument); 952 return false; 953 } 954 955 /// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 956 /// friends. This is declared to take (...), so we have to check everything. 957 bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 958 if (TheCall->getNumArgs() < 2) 959 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 960 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 961 if (TheCall->getNumArgs() > 2) 962 return Diag(TheCall->getArg(2)->getLocStart(), 963 diag::err_typecheck_call_too_many_args) 964 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 965 << SourceRange(TheCall->getArg(2)->getLocStart(), 966 (*(TheCall->arg_end()-1))->getLocEnd()); 967 968 ExprResult OrigArg0 = TheCall->getArg(0); 969 ExprResult OrigArg1 = TheCall->getArg(1); 970 971 // Do standard promotions between the two arguments, returning their common 972 // type. 973 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 974 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 975 return true; 976 977 // Make sure any conversions are pushed back into the call; this is 978 // type safe since unordered compare builtins are declared as "_Bool 979 // foo(...)". 980 TheCall->setArg(0, OrigArg0.get()); 981 TheCall->setArg(1, OrigArg1.get()); 982 983 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 984 return false; 985 986 // If the common type isn't a real floating type, then the arguments were 987 // invalid for this operation. 988 if (!Res->isRealFloatingType()) 989 return Diag(OrigArg0.get()->getLocStart(), 990 diag::err_typecheck_call_invalid_ordered_compare) 991 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 992 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 993 994 return false; 995 } 996 997 /// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 998 /// __builtin_isnan and friends. This is declared to take (...), so we have 999 /// to check everything. We expect the last argument to be a floating point 1000 /// value. 1001 bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 1002 if (TheCall->getNumArgs() < NumArgs) 1003 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1004 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 1005 if (TheCall->getNumArgs() > NumArgs) 1006 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 1007 diag::err_typecheck_call_too_many_args) 1008 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 1009 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 1010 (*(TheCall->arg_end()-1))->getLocEnd()); 1011 1012 Expr *OrigArg = TheCall->getArg(NumArgs-1); 1013 1014 if (OrigArg->isTypeDependent()) 1015 return false; 1016 1017 // This operation requires a non-_Complex floating-point number. 1018 if (!OrigArg->getType()->isRealFloatingType()) 1019 return Diag(OrigArg->getLocStart(), 1020 diag::err_typecheck_call_invalid_unary_fp) 1021 << OrigArg->getType() << OrigArg->getSourceRange(); 1022 1023 // If this is an implicit conversion from float -> double, remove it. 1024 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 1025 Expr *CastArg = Cast->getSubExpr(); 1026 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 1027 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 1028 "promotion from float to double is the only expected cast here"); 1029 Cast->setSubExpr(0); 1030 TheCall->setArg(NumArgs-1, CastArg); 1031 OrigArg = CastArg; 1032 } 1033 } 1034 1035 return false; 1036 } 1037 1038 /// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1039 // This is declared to take (...), so we have to check everything. 1040 ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1041 if (TheCall->getNumArgs() < 2) 1042 return ExprError(Diag(TheCall->getLocEnd(), 1043 diag::err_typecheck_call_too_few_args_at_least) 1044 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1045 << TheCall->getSourceRange()); 1046 1047 // Determine which of the following types of shufflevector we're checking: 1048 // 1) unary, vector mask: (lhs, mask) 1049 // 2) binary, vector mask: (lhs, rhs, mask) 1050 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1051 QualType resType = TheCall->getArg(0)->getType(); 1052 unsigned numElements = 0; 1053 1054 if (!TheCall->getArg(0)->isTypeDependent() && 1055 !TheCall->getArg(1)->isTypeDependent()) { 1056 QualType LHSType = TheCall->getArg(0)->getType(); 1057 QualType RHSType = TheCall->getArg(1)->getType(); 1058 1059 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 1060 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 1061 << SourceRange(TheCall->getArg(0)->getLocStart(), 1062 TheCall->getArg(1)->getLocEnd()); 1063 return ExprError(); 1064 } 1065 1066 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1067 unsigned numResElements = TheCall->getNumArgs() - 2; 1068 1069 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1070 // with mask. If so, verify that RHS is an integer vector type with the 1071 // same number of elts as lhs. 1072 if (TheCall->getNumArgs() == 2) { 1073 if (!RHSType->hasIntegerRepresentation() || 1074 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1075 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1076 << SourceRange(TheCall->getArg(1)->getLocStart(), 1077 TheCall->getArg(1)->getLocEnd()); 1078 numResElements = numElements; 1079 } 1080 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1081 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1082 << SourceRange(TheCall->getArg(0)->getLocStart(), 1083 TheCall->getArg(1)->getLocEnd()); 1084 return ExprError(); 1085 } else if (numElements != numResElements) { 1086 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1087 resType = Context.getVectorType(eltType, numResElements, 1088 VectorType::GenericVector); 1089 } 1090 } 1091 1092 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1093 if (TheCall->getArg(i)->isTypeDependent() || 1094 TheCall->getArg(i)->isValueDependent()) 1095 continue; 1096 1097 llvm::APSInt Result(32); 1098 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1099 return ExprError(Diag(TheCall->getLocStart(), 1100 diag::err_shufflevector_nonconstant_argument) 1101 << TheCall->getArg(i)->getSourceRange()); 1102 1103 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1104 return ExprError(Diag(TheCall->getLocStart(), 1105 diag::err_shufflevector_argument_too_large) 1106 << TheCall->getArg(i)->getSourceRange()); 1107 } 1108 1109 SmallVector<Expr*, 32> exprs; 1110 1111 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1112 exprs.push_back(TheCall->getArg(i)); 1113 TheCall->setArg(i, 0); 1114 } 1115 1116 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 1117 exprs.size(), resType, 1118 TheCall->getCallee()->getLocStart(), 1119 TheCall->getRParenLoc())); 1120 } 1121 1122 /// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1123 // This is declared to take (const void*, ...) and can take two 1124 // optional constant int args. 1125 bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1126 unsigned NumArgs = TheCall->getNumArgs(); 1127 1128 if (NumArgs > 3) 1129 return Diag(TheCall->getLocEnd(), 1130 diag::err_typecheck_call_too_many_args_at_most) 1131 << 0 /*function call*/ << 3 << NumArgs 1132 << TheCall->getSourceRange(); 1133 1134 // Argument 0 is checked for us and the remaining arguments must be 1135 // constant integers. 1136 for (unsigned i = 1; i != NumArgs; ++i) { 1137 Expr *Arg = TheCall->getArg(i); 1138 1139 llvm::APSInt Result; 1140 if (SemaBuiltinConstantArg(TheCall, i, Result)) 1141 return true; 1142 1143 // FIXME: gcc issues a warning and rewrites these to 0. These 1144 // seems especially odd for the third argument since the default 1145 // is 3. 1146 if (i == 1) { 1147 if (Result.getLimitedValue() > 1) 1148 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1149 << "0" << "1" << Arg->getSourceRange(); 1150 } else { 1151 if (Result.getLimitedValue() > 3) 1152 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1153 << "0" << "3" << Arg->getSourceRange(); 1154 } 1155 } 1156 1157 return false; 1158 } 1159 1160 /// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1161 /// TheCall is a constant expression. 1162 bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1163 llvm::APSInt &Result) { 1164 Expr *Arg = TheCall->getArg(ArgNum); 1165 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1166 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1167 1168 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 1169 1170 if (!Arg->isIntegerConstantExpr(Result, Context)) 1171 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 1172 << FDecl->getDeclName() << Arg->getSourceRange(); 1173 1174 return false; 1175 } 1176 1177 /// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 1178 /// int type). This simply type checks that type is one of the defined 1179 /// constants (0-3). 1180 // For compatibility check 0-3, llvm only handles 0 and 2. 1181 bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 1182 llvm::APSInt Result; 1183 1184 // Check constant-ness first. 1185 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1186 return true; 1187 1188 Expr *Arg = TheCall->getArg(1); 1189 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 1190 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1191 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1192 } 1193 1194 return false; 1195 } 1196 1197 /// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1198 /// This checks that val is a constant 1. 1199 bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1200 Expr *Arg = TheCall->getArg(1); 1201 llvm::APSInt Result; 1202 1203 // TODO: This is less than ideal. Overload this to take a value. 1204 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1205 return true; 1206 1207 if (Result != 1) 1208 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1209 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1210 1211 return false; 1212 } 1213 1214 // Handle i > 1 ? "x" : "y", recursively. 1215 bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 1216 bool HasVAListArg, 1217 unsigned format_idx, unsigned firstDataArg, 1218 bool isPrintf) { 1219 tryAgain: 1220 if (E->isTypeDependent() || E->isValueDependent()) 1221 return false; 1222 1223 E = E->IgnoreParens(); 1224 1225 switch (E->getStmtClass()) { 1226 case Stmt::BinaryConditionalOperatorClass: 1227 case Stmt::ConditionalOperatorClass: { 1228 const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E); 1229 return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, HasVAListArg, 1230 format_idx, firstDataArg, isPrintf) 1231 && SemaCheckStringLiteral(C->getFalseExpr(), TheCall, HasVAListArg, 1232 format_idx, firstDataArg, isPrintf); 1233 } 1234 1235 case Stmt::IntegerLiteralClass: 1236 // Technically -Wformat-nonliteral does not warn about this case. 1237 // The behavior of printf and friends in this case is implementation 1238 // dependent. Ideally if the format string cannot be null then 1239 // it should have a 'nonnull' attribute in the function prototype. 1240 return true; 1241 1242 case Stmt::ImplicitCastExprClass: { 1243 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 1244 goto tryAgain; 1245 } 1246 1247 case Stmt::OpaqueValueExprClass: 1248 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 1249 E = src; 1250 goto tryAgain; 1251 } 1252 return false; 1253 1254 case Stmt::PredefinedExprClass: 1255 // While __func__, etc., are technically not string literals, they 1256 // cannot contain format specifiers and thus are not a security 1257 // liability. 1258 return true; 1259 1260 case Stmt::DeclRefExprClass: { 1261 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1262 1263 // As an exception, do not flag errors for variables binding to 1264 // const string literals. 1265 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1266 bool isConstant = false; 1267 QualType T = DR->getType(); 1268 1269 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1270 isConstant = AT->getElementType().isConstant(Context); 1271 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1272 isConstant = T.isConstant(Context) && 1273 PT->getPointeeType().isConstant(Context); 1274 } 1275 1276 if (isConstant) { 1277 if (const Expr *Init = VD->getAnyInitializer()) 1278 return SemaCheckStringLiteral(Init, TheCall, 1279 HasVAListArg, format_idx, firstDataArg, 1280 isPrintf); 1281 } 1282 1283 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1284 // special check to see if the format string is a function parameter 1285 // of the function calling the printf function. If the function 1286 // has an attribute indicating it is a printf-like function, then we 1287 // should suppress warnings concerning non-literals being used in a call 1288 // to a vprintf function. For example: 1289 // 1290 // void 1291 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1292 // va_list ap; 1293 // va_start(ap, fmt); 1294 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1295 // ... 1296 // 1297 // 1298 // FIXME: We don't have full attribute support yet, so just check to see 1299 // if the argument is a DeclRefExpr that references a parameter. We'll 1300 // add proper support for checking the attribute later. 1301 if (HasVAListArg) 1302 if (isa<ParmVarDecl>(VD)) 1303 return true; 1304 } 1305 1306 return false; 1307 } 1308 1309 case Stmt::CallExprClass: { 1310 const CallExpr *CE = cast<CallExpr>(E); 1311 if (const ImplicitCastExpr *ICE 1312 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 1313 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 1314 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 1315 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 1316 unsigned ArgIndex = FA->getFormatIdx(); 1317 const Expr *Arg = CE->getArg(ArgIndex - 1); 1318 1319 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 1320 format_idx, firstDataArg, isPrintf); 1321 } 1322 } 1323 } 1324 } 1325 1326 return false; 1327 } 1328 case Stmt::ObjCStringLiteralClass: 1329 case Stmt::StringLiteralClass: { 1330 const StringLiteral *StrE = NULL; 1331 1332 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1333 StrE = ObjCFExpr->getString(); 1334 else 1335 StrE = cast<StringLiteral>(E); 1336 1337 if (StrE) { 1338 CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx, 1339 firstDataArg, isPrintf); 1340 return true; 1341 } 1342 1343 return false; 1344 } 1345 1346 default: 1347 return false; 1348 } 1349 } 1350 1351 void 1352 Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1353 const Expr * const *ExprArgs, 1354 SourceLocation CallSiteLoc) { 1355 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1356 e = NonNull->args_end(); 1357 i != e; ++i) { 1358 const Expr *ArgExpr = ExprArgs[*i]; 1359 if (ArgExpr->isNullPointerConstant(Context, 1360 Expr::NPC_ValueDependentIsNotNull)) 1361 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 1362 } 1363 } 1364 1365 /// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar 1366 /// functions) for correct use of format strings. 1367 void 1368 Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg, 1369 unsigned format_idx, unsigned firstDataArg, 1370 bool isPrintf) { 1371 1372 const Expr *Fn = TheCall->getCallee(); 1373 1374 // The way the format attribute works in GCC, the implicit this argument 1375 // of member functions is counted. However, it doesn't appear in our own 1376 // lists, so decrement format_idx in that case. 1377 if (isa<CXXMemberCallExpr>(TheCall)) { 1378 const CXXMethodDecl *method_decl = 1379 dyn_cast<CXXMethodDecl>(TheCall->getCalleeDecl()); 1380 if (method_decl && method_decl->isInstance()) { 1381 // Catch a format attribute mistakenly referring to the object argument. 1382 if (format_idx == 0) 1383 return; 1384 --format_idx; 1385 if(firstDataArg != 0) 1386 --firstDataArg; 1387 } 1388 } 1389 1390 // CHECK: printf/scanf-like function is called with no format string. 1391 if (format_idx >= TheCall->getNumArgs()) { 1392 Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string) 1393 << Fn->getSourceRange(); 1394 return; 1395 } 1396 1397 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 1398 1399 // CHECK: format string is not a string literal. 1400 // 1401 // Dynamically generated format strings are difficult to 1402 // automatically vet at compile time. Requiring that format strings 1403 // are string literals: (1) permits the checking of format strings by 1404 // the compiler and thereby (2) can practically remove the source of 1405 // many format string exploits. 1406 1407 // Format string can be either ObjC string (e.g. @"%d") or 1408 // C string (e.g. "%d") 1409 // ObjC string uses the same format specifiers as C string, so we can use 1410 // the same format string checking logic for both ObjC and C strings. 1411 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 1412 firstDataArg, isPrintf)) 1413 return; // Literal format string found, check done! 1414 1415 // If there are no arguments specified, warn with -Wformat-security, otherwise 1416 // warn only with -Wformat-nonliteral. 1417 if (TheCall->getNumArgs() == format_idx+1) 1418 Diag(TheCall->getArg(format_idx)->getLocStart(), 1419 diag::warn_format_nonliteral_noargs) 1420 << OrigFormatExpr->getSourceRange(); 1421 else 1422 Diag(TheCall->getArg(format_idx)->getLocStart(), 1423 diag::warn_format_nonliteral) 1424 << OrigFormatExpr->getSourceRange(); 1425 } 1426 1427 namespace { 1428 class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1429 protected: 1430 Sema &S; 1431 const StringLiteral *FExpr; 1432 const Expr *OrigFormatExpr; 1433 const unsigned FirstDataArg; 1434 const unsigned NumDataArgs; 1435 const bool IsObjCLiteral; 1436 const char *Beg; // Start of format string. 1437 const bool HasVAListArg; 1438 const CallExpr *TheCall; 1439 unsigned FormatIdx; 1440 llvm::BitVector CoveredArgs; 1441 bool usesPositionalArgs; 1442 bool atFirstArg; 1443 public: 1444 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1445 const Expr *origFormatExpr, unsigned firstDataArg, 1446 unsigned numDataArgs, bool isObjCLiteral, 1447 const char *beg, bool hasVAListArg, 1448 const CallExpr *theCall, unsigned formatIdx) 1449 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1450 FirstDataArg(firstDataArg), 1451 NumDataArgs(numDataArgs), 1452 IsObjCLiteral(isObjCLiteral), Beg(beg), 1453 HasVAListArg(hasVAListArg), 1454 TheCall(theCall), FormatIdx(formatIdx), 1455 usesPositionalArgs(false), atFirstArg(true) { 1456 CoveredArgs.resize(numDataArgs); 1457 CoveredArgs.reset(); 1458 } 1459 1460 void DoneProcessing(); 1461 1462 void HandleIncompleteSpecifier(const char *startSpecifier, 1463 unsigned specifierLen); 1464 1465 virtual void HandleInvalidPosition(const char *startSpecifier, 1466 unsigned specifierLen, 1467 analyze_format_string::PositionContext p); 1468 1469 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1470 1471 void HandleNullChar(const char *nullCharacter); 1472 1473 protected: 1474 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1475 const char *startSpec, 1476 unsigned specifierLen, 1477 const char *csStart, unsigned csLen); 1478 1479 SourceRange getFormatStringRange(); 1480 CharSourceRange getSpecifierRange(const char *startSpecifier, 1481 unsigned specifierLen); 1482 SourceLocation getLocationOfByte(const char *x); 1483 1484 const Expr *getDataArg(unsigned i) const; 1485 1486 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1487 const analyze_format_string::ConversionSpecifier &CS, 1488 const char *startSpecifier, unsigned specifierLen, 1489 unsigned argIndex); 1490 }; 1491 } 1492 1493 SourceRange CheckFormatHandler::getFormatStringRange() { 1494 return OrigFormatExpr->getSourceRange(); 1495 } 1496 1497 CharSourceRange CheckFormatHandler:: 1498 getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1499 SourceLocation Start = getLocationOfByte(startSpecifier); 1500 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1501 1502 // Advance the end SourceLocation by one due to half-open ranges. 1503 End = End.getLocWithOffset(1); 1504 1505 return CharSourceRange::getCharRange(Start, End); 1506 } 1507 1508 SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 1509 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1510 } 1511 1512 void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 1513 unsigned specifierLen){ 1514 SourceLocation Loc = getLocationOfByte(startSpecifier); 1515 S.Diag(Loc, diag::warn_printf_incomplete_specifier) 1516 << getSpecifierRange(startSpecifier, specifierLen); 1517 } 1518 1519 void 1520 CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1521 analyze_format_string::PositionContext p) { 1522 SourceLocation Loc = getLocationOfByte(startPos); 1523 S.Diag(Loc, diag::warn_format_invalid_positional_specifier) 1524 << (unsigned) p << getSpecifierRange(startPos, posLen); 1525 } 1526 1527 void CheckFormatHandler::HandleZeroPosition(const char *startPos, 1528 unsigned posLen) { 1529 SourceLocation Loc = getLocationOfByte(startPos); 1530 S.Diag(Loc, diag::warn_format_zero_positional_specifier) 1531 << getSpecifierRange(startPos, posLen); 1532 } 1533 1534 void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 1535 if (!IsObjCLiteral) { 1536 // The presence of a null character is likely an error. 1537 S.Diag(getLocationOfByte(nullCharacter), 1538 diag::warn_printf_format_string_contains_null_char) 1539 << getFormatStringRange(); 1540 } 1541 } 1542 1543 const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 1544 return TheCall->getArg(FirstDataArg + i); 1545 } 1546 1547 void CheckFormatHandler::DoneProcessing() { 1548 // Does the number of data arguments exceed the number of 1549 // format conversions in the format string? 1550 if (!HasVAListArg) { 1551 // Find any arguments that weren't covered. 1552 CoveredArgs.flip(); 1553 signed notCoveredArg = CoveredArgs.find_first(); 1554 if (notCoveredArg >= 0) { 1555 assert((unsigned)notCoveredArg < NumDataArgs); 1556 S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(), 1557 diag::warn_printf_data_arg_not_used) 1558 << getFormatStringRange(); 1559 } 1560 } 1561 } 1562 1563 bool 1564 CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 1565 SourceLocation Loc, 1566 const char *startSpec, 1567 unsigned specifierLen, 1568 const char *csStart, 1569 unsigned csLen) { 1570 1571 bool keepGoing = true; 1572 if (argIndex < NumDataArgs) { 1573 // Consider the argument coverered, even though the specifier doesn't 1574 // make sense. 1575 CoveredArgs.set(argIndex); 1576 } 1577 else { 1578 // If argIndex exceeds the number of data arguments we 1579 // don't issue a warning because that is just a cascade of warnings (and 1580 // they may have intended '%%' anyway). We don't want to continue processing 1581 // the format string after this point, however, as we will like just get 1582 // gibberish when trying to match arguments. 1583 keepGoing = false; 1584 } 1585 1586 S.Diag(Loc, diag::warn_format_invalid_conversion) 1587 << StringRef(csStart, csLen) 1588 << getSpecifierRange(startSpec, specifierLen); 1589 1590 return keepGoing; 1591 } 1592 1593 bool 1594 CheckFormatHandler::CheckNumArgs( 1595 const analyze_format_string::FormatSpecifier &FS, 1596 const analyze_format_string::ConversionSpecifier &CS, 1597 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 1598 1599 if (argIndex >= NumDataArgs) { 1600 if (FS.usesPositionalArg()) { 1601 S.Diag(getLocationOfByte(CS.getStart()), 1602 diag::warn_printf_positional_arg_exceeds_data_args) 1603 << (argIndex+1) << NumDataArgs 1604 << getSpecifierRange(startSpecifier, specifierLen); 1605 } 1606 else { 1607 S.Diag(getLocationOfByte(CS.getStart()), 1608 diag::warn_printf_insufficient_data_args) 1609 << getSpecifierRange(startSpecifier, specifierLen); 1610 } 1611 1612 return false; 1613 } 1614 return true; 1615 } 1616 1617 //===--- CHECK: Printf format string checking ------------------------------===// 1618 1619 namespace { 1620 class CheckPrintfHandler : public CheckFormatHandler { 1621 public: 1622 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1623 const Expr *origFormatExpr, unsigned firstDataArg, 1624 unsigned numDataArgs, bool isObjCLiteral, 1625 const char *beg, bool hasVAListArg, 1626 const CallExpr *theCall, unsigned formatIdx) 1627 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1628 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1629 theCall, formatIdx) {} 1630 1631 1632 bool HandleInvalidPrintfConversionSpecifier( 1633 const analyze_printf::PrintfSpecifier &FS, 1634 const char *startSpecifier, 1635 unsigned specifierLen); 1636 1637 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 1638 const char *startSpecifier, 1639 unsigned specifierLen); 1640 1641 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 1642 const char *startSpecifier, unsigned specifierLen); 1643 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 1644 const analyze_printf::OptionalAmount &Amt, 1645 unsigned type, 1646 const char *startSpecifier, unsigned specifierLen); 1647 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1648 const analyze_printf::OptionalFlag &flag, 1649 const char *startSpecifier, unsigned specifierLen); 1650 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 1651 const analyze_printf::OptionalFlag &ignoredFlag, 1652 const analyze_printf::OptionalFlag &flag, 1653 const char *startSpecifier, unsigned specifierLen); 1654 }; 1655 } 1656 1657 bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 1658 const analyze_printf::PrintfSpecifier &FS, 1659 const char *startSpecifier, 1660 unsigned specifierLen) { 1661 const analyze_printf::PrintfConversionSpecifier &CS = 1662 FS.getConversionSpecifier(); 1663 1664 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1665 getLocationOfByte(CS.getStart()), 1666 startSpecifier, specifierLen, 1667 CS.getStart(), CS.getLength()); 1668 } 1669 1670 bool CheckPrintfHandler::HandleAmount( 1671 const analyze_format_string::OptionalAmount &Amt, 1672 unsigned k, const char *startSpecifier, 1673 unsigned specifierLen) { 1674 1675 if (Amt.hasDataArgument()) { 1676 if (!HasVAListArg) { 1677 unsigned argIndex = Amt.getArgIndex(); 1678 if (argIndex >= NumDataArgs) { 1679 S.Diag(getLocationOfByte(Amt.getStart()), 1680 diag::warn_printf_asterisk_missing_arg) 1681 << k << getSpecifierRange(startSpecifier, specifierLen); 1682 // Don't do any more checking. We will just emit 1683 // spurious errors. 1684 return false; 1685 } 1686 1687 // Type check the data argument. It should be an 'int'. 1688 // Although not in conformance with C99, we also allow the argument to be 1689 // an 'unsigned int' as that is a reasonably safe case. GCC also 1690 // doesn't emit a warning for that case. 1691 CoveredArgs.set(argIndex); 1692 const Expr *Arg = getDataArg(argIndex); 1693 QualType T = Arg->getType(); 1694 1695 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1696 assert(ATR.isValid()); 1697 1698 if (!ATR.matchesType(S.Context, T)) { 1699 S.Diag(getLocationOfByte(Amt.getStart()), 1700 diag::warn_printf_asterisk_wrong_type) 1701 << k 1702 << ATR.getRepresentativeType(S.Context) << T 1703 << getSpecifierRange(startSpecifier, specifierLen) 1704 << Arg->getSourceRange(); 1705 // Don't do any more checking. We will just emit 1706 // spurious errors. 1707 return false; 1708 } 1709 } 1710 } 1711 return true; 1712 } 1713 1714 void CheckPrintfHandler::HandleInvalidAmount( 1715 const analyze_printf::PrintfSpecifier &FS, 1716 const analyze_printf::OptionalAmount &Amt, 1717 unsigned type, 1718 const char *startSpecifier, 1719 unsigned specifierLen) { 1720 const analyze_printf::PrintfConversionSpecifier &CS = 1721 FS.getConversionSpecifier(); 1722 switch (Amt.getHowSpecified()) { 1723 case analyze_printf::OptionalAmount::Constant: 1724 S.Diag(getLocationOfByte(Amt.getStart()), 1725 diag::warn_printf_nonsensical_optional_amount) 1726 << type 1727 << CS.toString() 1728 << getSpecifierRange(startSpecifier, specifierLen) 1729 << FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 1730 Amt.getConstantLength())); 1731 break; 1732 1733 default: 1734 S.Diag(getLocationOfByte(Amt.getStart()), 1735 diag::warn_printf_nonsensical_optional_amount) 1736 << type 1737 << CS.toString() 1738 << getSpecifierRange(startSpecifier, specifierLen); 1739 break; 1740 } 1741 } 1742 1743 void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1744 const analyze_printf::OptionalFlag &flag, 1745 const char *startSpecifier, 1746 unsigned specifierLen) { 1747 // Warn about pointless flag with a fixit removal. 1748 const analyze_printf::PrintfConversionSpecifier &CS = 1749 FS.getConversionSpecifier(); 1750 S.Diag(getLocationOfByte(flag.getPosition()), 1751 diag::warn_printf_nonsensical_flag) 1752 << flag.toString() << CS.toString() 1753 << getSpecifierRange(startSpecifier, specifierLen) 1754 << FixItHint::CreateRemoval(getSpecifierRange(flag.getPosition(), 1)); 1755 } 1756 1757 void CheckPrintfHandler::HandleIgnoredFlag( 1758 const analyze_printf::PrintfSpecifier &FS, 1759 const analyze_printf::OptionalFlag &ignoredFlag, 1760 const analyze_printf::OptionalFlag &flag, 1761 const char *startSpecifier, 1762 unsigned specifierLen) { 1763 // Warn about ignored flag with a fixit removal. 1764 S.Diag(getLocationOfByte(ignoredFlag.getPosition()), 1765 diag::warn_printf_ignored_flag) 1766 << ignoredFlag.toString() << flag.toString() 1767 << getSpecifierRange(startSpecifier, specifierLen) 1768 << FixItHint::CreateRemoval(getSpecifierRange( 1769 ignoredFlag.getPosition(), 1)); 1770 } 1771 1772 bool 1773 CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 1774 &FS, 1775 const char *startSpecifier, 1776 unsigned specifierLen) { 1777 1778 using namespace analyze_format_string; 1779 using namespace analyze_printf; 1780 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 1781 1782 if (FS.consumesDataArgument()) { 1783 if (atFirstArg) { 1784 atFirstArg = false; 1785 usesPositionalArgs = FS.usesPositionalArg(); 1786 } 1787 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1788 // Cannot mix-and-match positional and non-positional arguments. 1789 S.Diag(getLocationOfByte(CS.getStart()), 1790 diag::warn_format_mix_positional_nonpositional_args) 1791 << getSpecifierRange(startSpecifier, specifierLen); 1792 return false; 1793 } 1794 } 1795 1796 // First check if the field width, precision, and conversion specifier 1797 // have matching data arguments. 1798 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 1799 startSpecifier, specifierLen)) { 1800 return false; 1801 } 1802 1803 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 1804 startSpecifier, specifierLen)) { 1805 return false; 1806 } 1807 1808 if (!CS.consumesDataArgument()) { 1809 // FIXME: Technically specifying a precision or field width here 1810 // makes no sense. Worth issuing a warning at some point. 1811 return true; 1812 } 1813 1814 // Consume the argument. 1815 unsigned argIndex = FS.getArgIndex(); 1816 if (argIndex < NumDataArgs) { 1817 // The check to see if the argIndex is valid will come later. 1818 // We set the bit here because we may exit early from this 1819 // function if we encounter some other error. 1820 CoveredArgs.set(argIndex); 1821 } 1822 1823 // Check for using an Objective-C specific conversion specifier 1824 // in a non-ObjC literal. 1825 if (!IsObjCLiteral && CS.isObjCArg()) { 1826 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 1827 specifierLen); 1828 } 1829 1830 // Check for invalid use of field width 1831 if (!FS.hasValidFieldWidth()) { 1832 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 1833 startSpecifier, specifierLen); 1834 } 1835 1836 // Check for invalid use of precision 1837 if (!FS.hasValidPrecision()) { 1838 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 1839 startSpecifier, specifierLen); 1840 } 1841 1842 // Check each flag does not conflict with any other component. 1843 if (!FS.hasValidThousandsGroupingPrefix()) 1844 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 1845 if (!FS.hasValidLeadingZeros()) 1846 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 1847 if (!FS.hasValidPlusPrefix()) 1848 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 1849 if (!FS.hasValidSpacePrefix()) 1850 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 1851 if (!FS.hasValidAlternativeForm()) 1852 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 1853 if (!FS.hasValidLeftJustified()) 1854 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 1855 1856 // Check that flags are not ignored by another flag 1857 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 1858 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 1859 startSpecifier, specifierLen); 1860 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 1861 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 1862 startSpecifier, specifierLen); 1863 1864 // Check the length modifier is valid with the given conversion specifier. 1865 const LengthModifier &LM = FS.getLengthModifier(); 1866 if (!FS.hasValidLengthModifier()) 1867 S.Diag(getLocationOfByte(LM.getStart()), 1868 diag::warn_format_nonsensical_length) 1869 << LM.toString() << CS.toString() 1870 << getSpecifierRange(startSpecifier, specifierLen) 1871 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 1872 LM.getLength())); 1873 1874 // Are we using '%n'? 1875 if (CS.getKind() == ConversionSpecifier::nArg) { 1876 // Issue a warning about this being a possible security issue. 1877 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back) 1878 << getSpecifierRange(startSpecifier, specifierLen); 1879 // Continue checking the other format specifiers. 1880 return true; 1881 } 1882 1883 // The remaining checks depend on the data arguments. 1884 if (HasVAListArg) 1885 return true; 1886 1887 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 1888 return false; 1889 1890 // Now type check the data expression that matches the 1891 // format specifier. 1892 const Expr *Ex = getDataArg(argIndex); 1893 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context); 1894 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 1895 // Check if we didn't match because of an implicit cast from a 'char' 1896 // or 'short' to an 'int'. This is done because printf is a varargs 1897 // function. 1898 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 1899 if (ICE->getType() == S.Context.IntTy) { 1900 // All further checking is done on the subexpression. 1901 Ex = ICE->getSubExpr(); 1902 if (ATR.matchesType(S.Context, Ex->getType())) 1903 return true; 1904 } 1905 1906 // We may be able to offer a FixItHint if it is a supported type. 1907 PrintfSpecifier fixedFS = FS; 1908 bool success = fixedFS.fixType(Ex->getType(), S.getLangOptions()); 1909 1910 if (success) { 1911 // Get the fix string from the fixed format specifier 1912 llvm::SmallString<128> buf; 1913 llvm::raw_svector_ostream os(buf); 1914 fixedFS.toString(os); 1915 1916 // FIXME: getRepresentativeType() perhaps should return a string 1917 // instead of a QualType to better handle when the representative 1918 // type is 'wint_t' (which is defined in the system headers). 1919 S.Diag(getLocationOfByte(CS.getStart()), 1920 diag::warn_printf_conversion_argument_type_mismatch) 1921 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1922 << getSpecifierRange(startSpecifier, specifierLen) 1923 << Ex->getSourceRange() 1924 << FixItHint::CreateReplacement( 1925 getSpecifierRange(startSpecifier, specifierLen), 1926 os.str()); 1927 } 1928 else { 1929 S.Diag(getLocationOfByte(CS.getStart()), 1930 diag::warn_printf_conversion_argument_type_mismatch) 1931 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1932 << getSpecifierRange(startSpecifier, specifierLen) 1933 << Ex->getSourceRange(); 1934 } 1935 } 1936 1937 return true; 1938 } 1939 1940 //===--- CHECK: Scanf format string checking ------------------------------===// 1941 1942 namespace { 1943 class CheckScanfHandler : public CheckFormatHandler { 1944 public: 1945 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 1946 const Expr *origFormatExpr, unsigned firstDataArg, 1947 unsigned numDataArgs, bool isObjCLiteral, 1948 const char *beg, bool hasVAListArg, 1949 const CallExpr *theCall, unsigned formatIdx) 1950 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1951 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1952 theCall, formatIdx) {} 1953 1954 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 1955 const char *startSpecifier, 1956 unsigned specifierLen); 1957 1958 bool HandleInvalidScanfConversionSpecifier( 1959 const analyze_scanf::ScanfSpecifier &FS, 1960 const char *startSpecifier, 1961 unsigned specifierLen); 1962 1963 void HandleIncompleteScanList(const char *start, const char *end); 1964 }; 1965 } 1966 1967 void CheckScanfHandler::HandleIncompleteScanList(const char *start, 1968 const char *end) { 1969 S.Diag(getLocationOfByte(end), diag::warn_scanf_scanlist_incomplete) 1970 << getSpecifierRange(start, end - start); 1971 } 1972 1973 bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 1974 const analyze_scanf::ScanfSpecifier &FS, 1975 const char *startSpecifier, 1976 unsigned specifierLen) { 1977 1978 const analyze_scanf::ScanfConversionSpecifier &CS = 1979 FS.getConversionSpecifier(); 1980 1981 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1982 getLocationOfByte(CS.getStart()), 1983 startSpecifier, specifierLen, 1984 CS.getStart(), CS.getLength()); 1985 } 1986 1987 bool CheckScanfHandler::HandleScanfSpecifier( 1988 const analyze_scanf::ScanfSpecifier &FS, 1989 const char *startSpecifier, 1990 unsigned specifierLen) { 1991 1992 using namespace analyze_scanf; 1993 using namespace analyze_format_string; 1994 1995 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 1996 1997 // Handle case where '%' and '*' don't consume an argument. These shouldn't 1998 // be used to decide if we are using positional arguments consistently. 1999 if (FS.consumesDataArgument()) { 2000 if (atFirstArg) { 2001 atFirstArg = false; 2002 usesPositionalArgs = FS.usesPositionalArg(); 2003 } 2004 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2005 // Cannot mix-and-match positional and non-positional arguments. 2006 S.Diag(getLocationOfByte(CS.getStart()), 2007 diag::warn_format_mix_positional_nonpositional_args) 2008 << getSpecifierRange(startSpecifier, specifierLen); 2009 return false; 2010 } 2011 } 2012 2013 // Check if the field with is non-zero. 2014 const OptionalAmount &Amt = FS.getFieldWidth(); 2015 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 2016 if (Amt.getConstantAmount() == 0) { 2017 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 2018 Amt.getConstantLength()); 2019 S.Diag(getLocationOfByte(Amt.getStart()), 2020 diag::warn_scanf_nonzero_width) 2021 << R << FixItHint::CreateRemoval(R); 2022 } 2023 } 2024 2025 if (!FS.consumesDataArgument()) { 2026 // FIXME: Technically specifying a precision or field width here 2027 // makes no sense. Worth issuing a warning at some point. 2028 return true; 2029 } 2030 2031 // Consume the argument. 2032 unsigned argIndex = FS.getArgIndex(); 2033 if (argIndex < NumDataArgs) { 2034 // The check to see if the argIndex is valid will come later. 2035 // We set the bit here because we may exit early from this 2036 // function if we encounter some other error. 2037 CoveredArgs.set(argIndex); 2038 } 2039 2040 // Check the length modifier is valid with the given conversion specifier. 2041 const LengthModifier &LM = FS.getLengthModifier(); 2042 if (!FS.hasValidLengthModifier()) { 2043 S.Diag(getLocationOfByte(LM.getStart()), 2044 diag::warn_format_nonsensical_length) 2045 << LM.toString() << CS.toString() 2046 << getSpecifierRange(startSpecifier, specifierLen) 2047 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 2048 LM.getLength())); 2049 } 2050 2051 // The remaining checks depend on the data arguments. 2052 if (HasVAListArg) 2053 return true; 2054 2055 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2056 return false; 2057 2058 // FIXME: Check that the argument type matches the format specifier. 2059 2060 return true; 2061 } 2062 2063 void Sema::CheckFormatString(const StringLiteral *FExpr, 2064 const Expr *OrigFormatExpr, 2065 const CallExpr *TheCall, bool HasVAListArg, 2066 unsigned format_idx, unsigned firstDataArg, 2067 bool isPrintf) { 2068 2069 // CHECK: is the format string a wide literal? 2070 if (!FExpr->isAscii()) { 2071 Diag(FExpr->getLocStart(), 2072 diag::warn_format_string_is_wide_literal) 2073 << OrigFormatExpr->getSourceRange(); 2074 return; 2075 } 2076 2077 // Str - The format string. NOTE: this is NOT null-terminated! 2078 StringRef StrRef = FExpr->getString(); 2079 const char *Str = StrRef.data(); 2080 unsigned StrLen = StrRef.size(); 2081 const unsigned numDataArgs = TheCall->getNumArgs() - firstDataArg; 2082 2083 // CHECK: empty format string? 2084 if (StrLen == 0 && numDataArgs > 0) { 2085 Diag(FExpr->getLocStart(), diag::warn_empty_format_string) 2086 << OrigFormatExpr->getSourceRange(); 2087 return; 2088 } 2089 2090 if (isPrintf) { 2091 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2092 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2093 Str, HasVAListArg, TheCall, format_idx); 2094 2095 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen)) 2096 H.DoneProcessing(); 2097 } 2098 else { 2099 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2100 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2101 Str, HasVAListArg, TheCall, format_idx); 2102 2103 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen)) 2104 H.DoneProcessing(); 2105 } 2106 } 2107 2108 //===--- CHECK: Standard memory functions ---------------------------------===// 2109 2110 /// \brief Determine whether the given type is a dynamic class type (e.g., 2111 /// whether it has a vtable). 2112 static bool isDynamicClassType(QualType T) { 2113 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 2114 if (CXXRecordDecl *Definition = Record->getDefinition()) 2115 if (Definition->isDynamicClass()) 2116 return true; 2117 2118 return false; 2119 } 2120 2121 /// \brief If E is a sizeof expression, returns its argument expression, 2122 /// otherwise returns NULL. 2123 static const Expr *getSizeOfExprArg(const Expr* E) { 2124 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2125 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2126 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 2127 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 2128 2129 return 0; 2130 } 2131 2132 /// \brief If E is a sizeof expression, returns its argument type. 2133 static QualType getSizeOfArgType(const Expr* E) { 2134 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2135 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2136 if (SizeOf->getKind() == clang::UETT_SizeOf) 2137 return SizeOf->getTypeOfArgument(); 2138 2139 return QualType(); 2140 } 2141 2142 /// \brief Check for dangerous or invalid arguments to memset(). 2143 /// 2144 /// This issues warnings on known problematic, dangerous or unspecified 2145 /// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 2146 /// function calls. 2147 /// 2148 /// \param Call The call expression to diagnose. 2149 void Sema::CheckMemaccessArguments(const CallExpr *Call, 2150 CheckedMemoryFunction CMF, 2151 IdentifierInfo *FnName) { 2152 // It is possible to have a non-standard definition of memset. Validate 2153 // we have enough arguments, and if not, abort further checking. 2154 unsigned ExpectedNumArgs = (CMF == CMF_Strndup ? 2 : 3); 2155 if (Call->getNumArgs() < ExpectedNumArgs) 2156 return; 2157 2158 unsigned LastArg = (CMF == CMF_Memset || CMF == CMF_Strndup ? 1 : 2); 2159 unsigned LenArg = (CMF == CMF_Strndup ? 1 : 2); 2160 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 2161 2162 // We have special checking when the length is a sizeof expression. 2163 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 2164 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 2165 llvm::FoldingSetNodeID SizeOfArgID; 2166 2167 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 2168 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 2169 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 2170 2171 QualType DestTy = Dest->getType(); 2172 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 2173 QualType PointeeTy = DestPtrTy->getPointeeType(); 2174 2175 // Never warn about void type pointers. This can be used to suppress 2176 // false positives. 2177 if (PointeeTy->isVoidType()) 2178 continue; 2179 2180 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 2181 // actually comparing the expressions for equality. Because computing the 2182 // expression IDs can be expensive, we only do this if the diagnostic is 2183 // enabled. 2184 if (SizeOfArg && 2185 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 2186 SizeOfArg->getExprLoc())) { 2187 // We only compute IDs for expressions if the warning is enabled, and 2188 // cache the sizeof arg's ID. 2189 if (SizeOfArgID == llvm::FoldingSetNodeID()) 2190 SizeOfArg->Profile(SizeOfArgID, Context, true); 2191 llvm::FoldingSetNodeID DestID; 2192 Dest->Profile(DestID, Context, true); 2193 if (DestID == SizeOfArgID) { 2194 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 2195 // over sizeof(src) as well. 2196 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 2197 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 2198 if (UnaryOp->getOpcode() == UO_AddrOf) 2199 ActionIdx = 1; // If its an address-of operator, just remove it. 2200 if (Context.getTypeSize(PointeeTy) == Context.getCharWidth()) 2201 ActionIdx = 2; // If the pointee's size is sizeof(char), 2202 // suggest an explicit length. 2203 unsigned DestSrcSelect = (CMF == CMF_Strndup ? 1 : ArgIdx); 2204 DiagRuntimeBehavior(SizeOfArg->getExprLoc(), Dest, 2205 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 2206 << FnName << DestSrcSelect << ActionIdx 2207 << Dest->getSourceRange() 2208 << SizeOfArg->getSourceRange()); 2209 break; 2210 } 2211 } 2212 2213 // Also check for cases where the sizeof argument is the exact same 2214 // type as the memory argument, and where it points to a user-defined 2215 // record type. 2216 if (SizeOfArgTy != QualType()) { 2217 if (PointeeTy->isRecordType() && 2218 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 2219 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 2220 PDiag(diag::warn_sizeof_pointer_type_memaccess) 2221 << FnName << SizeOfArgTy << ArgIdx 2222 << PointeeTy << Dest->getSourceRange() 2223 << LenExpr->getSourceRange()); 2224 break; 2225 } 2226 } 2227 2228 // Always complain about dynamic classes. 2229 if (isDynamicClassType(PointeeTy)) 2230 DiagRuntimeBehavior( 2231 Dest->getExprLoc(), Dest, 2232 PDiag(diag::warn_dyn_class_memaccess) 2233 << (CMF == CMF_Memcmp ? ArgIdx + 2 : ArgIdx) << FnName << PointeeTy 2234 // "overwritten" if we're warning about the destination for any call 2235 // but memcmp; otherwise a verb appropriate to the call. 2236 << (ArgIdx == 0 && CMF != CMF_Memcmp ? 0 : (unsigned)CMF) 2237 << Call->getCallee()->getSourceRange()); 2238 else if (PointeeTy.hasNonTrivialObjCLifetime() && CMF != CMF_Memset) 2239 DiagRuntimeBehavior( 2240 Dest->getExprLoc(), Dest, 2241 PDiag(diag::warn_arc_object_memaccess) 2242 << ArgIdx << FnName << PointeeTy 2243 << Call->getCallee()->getSourceRange()); 2244 else 2245 continue; 2246 2247 DiagRuntimeBehavior( 2248 Dest->getExprLoc(), Dest, 2249 PDiag(diag::note_bad_memaccess_silence) 2250 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 2251 break; 2252 } 2253 } 2254 } 2255 2256 // A little helper routine: ignore addition and subtraction of integer literals. 2257 // This intentionally does not ignore all integer constant expressions because 2258 // we don't want to remove sizeof(). 2259 static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 2260 Ex = Ex->IgnoreParenCasts(); 2261 2262 for (;;) { 2263 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 2264 if (!BO || !BO->isAdditiveOp()) 2265 break; 2266 2267 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 2268 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 2269 2270 if (isa<IntegerLiteral>(RHS)) 2271 Ex = LHS; 2272 else if (isa<IntegerLiteral>(LHS)) 2273 Ex = RHS; 2274 else 2275 break; 2276 } 2277 2278 return Ex; 2279 } 2280 2281 // Warn if the user has made the 'size' argument to strlcpy or strlcat 2282 // be the size of the source, instead of the destination. 2283 void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 2284 IdentifierInfo *FnName) { 2285 2286 // Don't crash if the user has the wrong number of arguments 2287 if (Call->getNumArgs() != 3) 2288 return; 2289 2290 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 2291 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 2292 const Expr *CompareWithSrc = NULL; 2293 2294 // Look for 'strlcpy(dst, x, sizeof(x))' 2295 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 2296 CompareWithSrc = Ex; 2297 else { 2298 // Look for 'strlcpy(dst, x, strlen(x))' 2299 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 2300 if (SizeCall->isBuiltinCall(Context) == Builtin::BIstrlen 2301 && SizeCall->getNumArgs() == 1) 2302 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 2303 } 2304 } 2305 2306 if (!CompareWithSrc) 2307 return; 2308 2309 // Determine if the argument to sizeof/strlen is equal to the source 2310 // argument. In principle there's all kinds of things you could do 2311 // here, for instance creating an == expression and evaluating it with 2312 // EvaluateAsBooleanCondition, but this uses a more direct technique: 2313 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 2314 if (!SrcArgDRE) 2315 return; 2316 2317 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 2318 if (!CompareWithSrcDRE || 2319 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 2320 return; 2321 2322 const Expr *OriginalSizeArg = Call->getArg(2); 2323 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 2324 << OriginalSizeArg->getSourceRange() << FnName; 2325 2326 // Output a FIXIT hint if the destination is an array (rather than a 2327 // pointer to an array). This could be enhanced to handle some 2328 // pointers if we know the actual size, like if DstArg is 'array+2' 2329 // we could say 'sizeof(array)-2'. 2330 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 2331 QualType DstArgTy = DstArg->getType(); 2332 2333 // Only handle constant-sized or VLAs, but not flexible members. 2334 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 2335 // Only issue the FIXIT for arrays of size > 1. 2336 if (CAT->getSize().getSExtValue() <= 1) 2337 return; 2338 } else if (!DstArgTy->isVariableArrayType()) { 2339 return; 2340 } 2341 2342 llvm::SmallString<128> sizeString; 2343 llvm::raw_svector_ostream OS(sizeString); 2344 OS << "sizeof("; 2345 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 2346 OS << ")"; 2347 2348 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 2349 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 2350 OS.str()); 2351 } 2352 2353 //===--- CHECK: Return Address of Stack Variable --------------------------===// 2354 2355 static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars); 2356 static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars); 2357 2358 /// CheckReturnStackAddr - Check if a return statement returns the address 2359 /// of a stack variable. 2360 void 2361 Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 2362 SourceLocation ReturnLoc) { 2363 2364 Expr *stackE = 0; 2365 SmallVector<DeclRefExpr *, 8> refVars; 2366 2367 // Perform checking for returned stack addresses, local blocks, 2368 // label addresses or references to temporaries. 2369 if (lhsType->isPointerType() || 2370 (!getLangOptions().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 2371 stackE = EvalAddr(RetValExp, refVars); 2372 } else if (lhsType->isReferenceType()) { 2373 stackE = EvalVal(RetValExp, refVars); 2374 } 2375 2376 if (stackE == 0) 2377 return; // Nothing suspicious was found. 2378 2379 SourceLocation diagLoc; 2380 SourceRange diagRange; 2381 if (refVars.empty()) { 2382 diagLoc = stackE->getLocStart(); 2383 diagRange = stackE->getSourceRange(); 2384 } else { 2385 // We followed through a reference variable. 'stackE' contains the 2386 // problematic expression but we will warn at the return statement pointing 2387 // at the reference variable. We will later display the "trail" of 2388 // reference variables using notes. 2389 diagLoc = refVars[0]->getLocStart(); 2390 diagRange = refVars[0]->getSourceRange(); 2391 } 2392 2393 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 2394 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 2395 : diag::warn_ret_stack_addr) 2396 << DR->getDecl()->getDeclName() << diagRange; 2397 } else if (isa<BlockExpr>(stackE)) { // local block. 2398 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 2399 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 2400 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 2401 } else { // local temporary. 2402 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 2403 : diag::warn_ret_local_temp_addr) 2404 << diagRange; 2405 } 2406 2407 // Display the "trail" of reference variables that we followed until we 2408 // found the problematic expression using notes. 2409 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 2410 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 2411 // If this var binds to another reference var, show the range of the next 2412 // var, otherwise the var binds to the problematic expression, in which case 2413 // show the range of the expression. 2414 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 2415 : stackE->getSourceRange(); 2416 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 2417 << VD->getDeclName() << range; 2418 } 2419 } 2420 2421 /// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 2422 /// check if the expression in a return statement evaluates to an address 2423 /// to a location on the stack, a local block, an address of a label, or a 2424 /// reference to local temporary. The recursion is used to traverse the 2425 /// AST of the return expression, with recursion backtracking when we 2426 /// encounter a subexpression that (1) clearly does not lead to one of the 2427 /// above problematic expressions (2) is something we cannot determine leads to 2428 /// a problematic expression based on such local checking. 2429 /// 2430 /// Both EvalAddr and EvalVal follow through reference variables to evaluate 2431 /// the expression that they point to. Such variables are added to the 2432 /// 'refVars' vector so that we know what the reference variable "trail" was. 2433 /// 2434 /// EvalAddr processes expressions that are pointers that are used as 2435 /// references (and not L-values). EvalVal handles all other values. 2436 /// At the base case of the recursion is a check for the above problematic 2437 /// expressions. 2438 /// 2439 /// This implementation handles: 2440 /// 2441 /// * pointer-to-pointer casts 2442 /// * implicit conversions from array references to pointers 2443 /// * taking the address of fields 2444 /// * arbitrary interplay between "&" and "*" operators 2445 /// * pointer arithmetic from an address of a stack variable 2446 /// * taking the address of an array element where the array is on the stack 2447 static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 2448 if (E->isTypeDependent()) 2449 return NULL; 2450 2451 // We should only be called for evaluating pointer expressions. 2452 assert((E->getType()->isAnyPointerType() || 2453 E->getType()->isBlockPointerType() || 2454 E->getType()->isObjCQualifiedIdType()) && 2455 "EvalAddr only works on pointers"); 2456 2457 E = E->IgnoreParens(); 2458 2459 // Our "symbolic interpreter" is just a dispatch off the currently 2460 // viewed AST node. We then recursively traverse the AST by calling 2461 // EvalAddr and EvalVal appropriately. 2462 switch (E->getStmtClass()) { 2463 case Stmt::DeclRefExprClass: { 2464 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2465 2466 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2467 // If this is a reference variable, follow through to the expression that 2468 // it points to. 2469 if (V->hasLocalStorage() && 2470 V->getType()->isReferenceType() && V->hasInit()) { 2471 // Add the reference variable to the "trail". 2472 refVars.push_back(DR); 2473 return EvalAddr(V->getInit(), refVars); 2474 } 2475 2476 return NULL; 2477 } 2478 2479 case Stmt::UnaryOperatorClass: { 2480 // The only unary operator that make sense to handle here 2481 // is AddrOf. All others don't make sense as pointers. 2482 UnaryOperator *U = cast<UnaryOperator>(E); 2483 2484 if (U->getOpcode() == UO_AddrOf) 2485 return EvalVal(U->getSubExpr(), refVars); 2486 else 2487 return NULL; 2488 } 2489 2490 case Stmt::BinaryOperatorClass: { 2491 // Handle pointer arithmetic. All other binary operators are not valid 2492 // in this context. 2493 BinaryOperator *B = cast<BinaryOperator>(E); 2494 BinaryOperatorKind op = B->getOpcode(); 2495 2496 if (op != BO_Add && op != BO_Sub) 2497 return NULL; 2498 2499 Expr *Base = B->getLHS(); 2500 2501 // Determine which argument is the real pointer base. It could be 2502 // the RHS argument instead of the LHS. 2503 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 2504 2505 assert (Base->getType()->isPointerType()); 2506 return EvalAddr(Base, refVars); 2507 } 2508 2509 // For conditional operators we need to see if either the LHS or RHS are 2510 // valid DeclRefExpr*s. If one of them is valid, we return it. 2511 case Stmt::ConditionalOperatorClass: { 2512 ConditionalOperator *C = cast<ConditionalOperator>(E); 2513 2514 // Handle the GNU extension for missing LHS. 2515 if (Expr *lhsExpr = C->getLHS()) { 2516 // In C++, we can have a throw-expression, which has 'void' type. 2517 if (!lhsExpr->getType()->isVoidType()) 2518 if (Expr* LHS = EvalAddr(lhsExpr, refVars)) 2519 return LHS; 2520 } 2521 2522 // In C++, we can have a throw-expression, which has 'void' type. 2523 if (C->getRHS()->getType()->isVoidType()) 2524 return NULL; 2525 2526 return EvalAddr(C->getRHS(), refVars); 2527 } 2528 2529 case Stmt::BlockExprClass: 2530 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 2531 return E; // local block. 2532 return NULL; 2533 2534 case Stmt::AddrLabelExprClass: 2535 return E; // address of label. 2536 2537 // For casts, we need to handle conversions from arrays to 2538 // pointer values, and pointer-to-pointer conversions. 2539 case Stmt::ImplicitCastExprClass: 2540 case Stmt::CStyleCastExprClass: 2541 case Stmt::CXXFunctionalCastExprClass: 2542 case Stmt::ObjCBridgedCastExprClass: { 2543 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 2544 QualType T = SubExpr->getType(); 2545 2546 if (SubExpr->getType()->isPointerType() || 2547 SubExpr->getType()->isBlockPointerType() || 2548 SubExpr->getType()->isObjCQualifiedIdType()) 2549 return EvalAddr(SubExpr, refVars); 2550 else if (T->isArrayType()) 2551 return EvalVal(SubExpr, refVars); 2552 else 2553 return 0; 2554 } 2555 2556 // C++ casts. For dynamic casts, static casts, and const casts, we 2557 // are always converting from a pointer-to-pointer, so we just blow 2558 // through the cast. In the case the dynamic cast doesn't fail (and 2559 // return NULL), we take the conservative route and report cases 2560 // where we return the address of a stack variable. For Reinterpre 2561 // FIXME: The comment about is wrong; we're not always converting 2562 // from pointer to pointer. I'm guessing that this code should also 2563 // handle references to objects. 2564 case Stmt::CXXStaticCastExprClass: 2565 case Stmt::CXXDynamicCastExprClass: 2566 case Stmt::CXXConstCastExprClass: 2567 case Stmt::CXXReinterpretCastExprClass: { 2568 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 2569 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 2570 return EvalAddr(S, refVars); 2571 else 2572 return NULL; 2573 } 2574 2575 case Stmt::MaterializeTemporaryExprClass: 2576 if (Expr *Result = EvalAddr( 2577 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 2578 refVars)) 2579 return Result; 2580 2581 return E; 2582 2583 // Everything else: we simply don't reason about them. 2584 default: 2585 return NULL; 2586 } 2587 } 2588 2589 2590 /// EvalVal - This function is complements EvalAddr in the mutual recursion. 2591 /// See the comments for EvalAddr for more details. 2592 static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 2593 do { 2594 // We should only be called for evaluating non-pointer expressions, or 2595 // expressions with a pointer type that are not used as references but instead 2596 // are l-values (e.g., DeclRefExpr with a pointer type). 2597 2598 // Our "symbolic interpreter" is just a dispatch off the currently 2599 // viewed AST node. We then recursively traverse the AST by calling 2600 // EvalAddr and EvalVal appropriately. 2601 2602 E = E->IgnoreParens(); 2603 switch (E->getStmtClass()) { 2604 case Stmt::ImplicitCastExprClass: { 2605 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 2606 if (IE->getValueKind() == VK_LValue) { 2607 E = IE->getSubExpr(); 2608 continue; 2609 } 2610 return NULL; 2611 } 2612 2613 case Stmt::DeclRefExprClass: { 2614 // When we hit a DeclRefExpr we are looking at code that refers to a 2615 // variable's name. If it's not a reference variable we check if it has 2616 // local storage within the function, and if so, return the expression. 2617 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2618 2619 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2620 if (V->hasLocalStorage()) { 2621 if (!V->getType()->isReferenceType()) 2622 return DR; 2623 2624 // Reference variable, follow through to the expression that 2625 // it points to. 2626 if (V->hasInit()) { 2627 // Add the reference variable to the "trail". 2628 refVars.push_back(DR); 2629 return EvalVal(V->getInit(), refVars); 2630 } 2631 } 2632 2633 return NULL; 2634 } 2635 2636 case Stmt::UnaryOperatorClass: { 2637 // The only unary operator that make sense to handle here 2638 // is Deref. All others don't resolve to a "name." This includes 2639 // handling all sorts of rvalues passed to a unary operator. 2640 UnaryOperator *U = cast<UnaryOperator>(E); 2641 2642 if (U->getOpcode() == UO_Deref) 2643 return EvalAddr(U->getSubExpr(), refVars); 2644 2645 return NULL; 2646 } 2647 2648 case Stmt::ArraySubscriptExprClass: { 2649 // Array subscripts are potential references to data on the stack. We 2650 // retrieve the DeclRefExpr* for the array variable if it indeed 2651 // has local storage. 2652 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars); 2653 } 2654 2655 case Stmt::ConditionalOperatorClass: { 2656 // For conditional operators we need to see if either the LHS or RHS are 2657 // non-NULL Expr's. If one is non-NULL, we return it. 2658 ConditionalOperator *C = cast<ConditionalOperator>(E); 2659 2660 // Handle the GNU extension for missing LHS. 2661 if (Expr *lhsExpr = C->getLHS()) 2662 if (Expr *LHS = EvalVal(lhsExpr, refVars)) 2663 return LHS; 2664 2665 return EvalVal(C->getRHS(), refVars); 2666 } 2667 2668 // Accesses to members are potential references to data on the stack. 2669 case Stmt::MemberExprClass: { 2670 MemberExpr *M = cast<MemberExpr>(E); 2671 2672 // Check for indirect access. We only want direct field accesses. 2673 if (M->isArrow()) 2674 return NULL; 2675 2676 // Check whether the member type is itself a reference, in which case 2677 // we're not going to refer to the member, but to what the member refers to. 2678 if (M->getMemberDecl()->getType()->isReferenceType()) 2679 return NULL; 2680 2681 return EvalVal(M->getBase(), refVars); 2682 } 2683 2684 case Stmt::MaterializeTemporaryExprClass: 2685 if (Expr *Result = EvalVal( 2686 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 2687 refVars)) 2688 return Result; 2689 2690 return E; 2691 2692 default: 2693 // Check that we don't return or take the address of a reference to a 2694 // temporary. This is only useful in C++. 2695 if (!E->isTypeDependent() && E->isRValue()) 2696 return E; 2697 2698 // Everything else: we simply don't reason about them. 2699 return NULL; 2700 } 2701 } while (true); 2702 } 2703 2704 //===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 2705 2706 /// Check for comparisons of floating point operands using != and ==. 2707 /// Issue a warning if these are no self-comparisons, as they are not likely 2708 /// to do what the programmer intended. 2709 void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 2710 bool EmitWarning = true; 2711 2712 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 2713 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 2714 2715 // Special case: check for x == x (which is OK). 2716 // Do not emit warnings for such cases. 2717 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 2718 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 2719 if (DRL->getDecl() == DRR->getDecl()) 2720 EmitWarning = false; 2721 2722 2723 // Special case: check for comparisons against literals that can be exactly 2724 // represented by APFloat. In such cases, do not emit a warning. This 2725 // is a heuristic: often comparison against such literals are used to 2726 // detect if a value in a variable has not changed. This clearly can 2727 // lead to false negatives. 2728 if (EmitWarning) { 2729 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 2730 if (FLL->isExact()) 2731 EmitWarning = false; 2732 } else 2733 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 2734 if (FLR->isExact()) 2735 EmitWarning = false; 2736 } 2737 } 2738 2739 // Check for comparisons with builtin types. 2740 if (EmitWarning) 2741 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 2742 if (CL->isBuiltinCall(Context)) 2743 EmitWarning = false; 2744 2745 if (EmitWarning) 2746 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 2747 if (CR->isBuiltinCall(Context)) 2748 EmitWarning = false; 2749 2750 // Emit the diagnostic. 2751 if (EmitWarning) 2752 Diag(Loc, diag::warn_floatingpoint_eq) 2753 << LHS->getSourceRange() << RHS->getSourceRange(); 2754 } 2755 2756 //===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 2757 //===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 2758 2759 namespace { 2760 2761 /// Structure recording the 'active' range of an integer-valued 2762 /// expression. 2763 struct IntRange { 2764 /// The number of bits active in the int. 2765 unsigned Width; 2766 2767 /// True if the int is known not to have negative values. 2768 bool NonNegative; 2769 2770 IntRange(unsigned Width, bool NonNegative) 2771 : Width(Width), NonNegative(NonNegative) 2772 {} 2773 2774 /// Returns the range of the bool type. 2775 static IntRange forBoolType() { 2776 return IntRange(1, true); 2777 } 2778 2779 /// Returns the range of an opaque value of the given integral type. 2780 static IntRange forValueOfType(ASTContext &C, QualType T) { 2781 return forValueOfCanonicalType(C, 2782 T->getCanonicalTypeInternal().getTypePtr()); 2783 } 2784 2785 /// Returns the range of an opaque value of a canonical integral type. 2786 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 2787 assert(T->isCanonicalUnqualified()); 2788 2789 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2790 T = VT->getElementType().getTypePtr(); 2791 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2792 T = CT->getElementType().getTypePtr(); 2793 2794 // For enum types, use the known bit width of the enumerators. 2795 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 2796 EnumDecl *Enum = ET->getDecl(); 2797 if (!Enum->isCompleteDefinition()) 2798 return IntRange(C.getIntWidth(QualType(T, 0)), false); 2799 2800 unsigned NumPositive = Enum->getNumPositiveBits(); 2801 unsigned NumNegative = Enum->getNumNegativeBits(); 2802 2803 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 2804 } 2805 2806 const BuiltinType *BT = cast<BuiltinType>(T); 2807 assert(BT->isInteger()); 2808 2809 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2810 } 2811 2812 /// Returns the "target" range of a canonical integral type, i.e. 2813 /// the range of values expressible in the type. 2814 /// 2815 /// This matches forValueOfCanonicalType except that enums have the 2816 /// full range of their type, not the range of their enumerators. 2817 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 2818 assert(T->isCanonicalUnqualified()); 2819 2820 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2821 T = VT->getElementType().getTypePtr(); 2822 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2823 T = CT->getElementType().getTypePtr(); 2824 if (const EnumType *ET = dyn_cast<EnumType>(T)) 2825 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 2826 2827 const BuiltinType *BT = cast<BuiltinType>(T); 2828 assert(BT->isInteger()); 2829 2830 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2831 } 2832 2833 /// Returns the supremum of two ranges: i.e. their conservative merge. 2834 static IntRange join(IntRange L, IntRange R) { 2835 return IntRange(std::max(L.Width, R.Width), 2836 L.NonNegative && R.NonNegative); 2837 } 2838 2839 /// Returns the infinum of two ranges: i.e. their aggressive merge. 2840 static IntRange meet(IntRange L, IntRange R) { 2841 return IntRange(std::min(L.Width, R.Width), 2842 L.NonNegative || R.NonNegative); 2843 } 2844 }; 2845 2846 IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 2847 if (value.isSigned() && value.isNegative()) 2848 return IntRange(value.getMinSignedBits(), false); 2849 2850 if (value.getBitWidth() > MaxWidth) 2851 value = value.trunc(MaxWidth); 2852 2853 // isNonNegative() just checks the sign bit without considering 2854 // signedness. 2855 return IntRange(value.getActiveBits(), true); 2856 } 2857 2858 IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 2859 unsigned MaxWidth) { 2860 if (result.isInt()) 2861 return GetValueRange(C, result.getInt(), MaxWidth); 2862 2863 if (result.isVector()) { 2864 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 2865 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 2866 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 2867 R = IntRange::join(R, El); 2868 } 2869 return R; 2870 } 2871 2872 if (result.isComplexInt()) { 2873 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 2874 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 2875 return IntRange::join(R, I); 2876 } 2877 2878 // This can happen with lossless casts to intptr_t of "based" lvalues. 2879 // Assume it might use arbitrary bits. 2880 // FIXME: The only reason we need to pass the type in here is to get 2881 // the sign right on this one case. It would be nice if APValue 2882 // preserved this. 2883 assert(result.isLValue()); 2884 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 2885 } 2886 2887 /// Pseudo-evaluate the given integer expression, estimating the 2888 /// range of values it might take. 2889 /// 2890 /// \param MaxWidth - the width to which the value will be truncated 2891 IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 2892 E = E->IgnoreParens(); 2893 2894 // Try a full evaluation first. 2895 Expr::EvalResult result; 2896 if (E->Evaluate(result, C)) 2897 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 2898 2899 // I think we only want to look through implicit casts here; if the 2900 // user has an explicit widening cast, we should treat the value as 2901 // being of the new, wider type. 2902 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 2903 if (CE->getCastKind() == CK_NoOp) 2904 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 2905 2906 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 2907 2908 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 2909 2910 // Assume that non-integer casts can span the full range of the type. 2911 if (!isIntegerCast) 2912 return OutputTypeRange; 2913 2914 IntRange SubRange 2915 = GetExprRange(C, CE->getSubExpr(), 2916 std::min(MaxWidth, OutputTypeRange.Width)); 2917 2918 // Bail out if the subexpr's range is as wide as the cast type. 2919 if (SubRange.Width >= OutputTypeRange.Width) 2920 return OutputTypeRange; 2921 2922 // Otherwise, we take the smaller width, and we're non-negative if 2923 // either the output type or the subexpr is. 2924 return IntRange(SubRange.Width, 2925 SubRange.NonNegative || OutputTypeRange.NonNegative); 2926 } 2927 2928 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 2929 // If we can fold the condition, just take that operand. 2930 bool CondResult; 2931 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 2932 return GetExprRange(C, CondResult ? CO->getTrueExpr() 2933 : CO->getFalseExpr(), 2934 MaxWidth); 2935 2936 // Otherwise, conservatively merge. 2937 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 2938 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 2939 return IntRange::join(L, R); 2940 } 2941 2942 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 2943 switch (BO->getOpcode()) { 2944 2945 // Boolean-valued operations are single-bit and positive. 2946 case BO_LAnd: 2947 case BO_LOr: 2948 case BO_LT: 2949 case BO_GT: 2950 case BO_LE: 2951 case BO_GE: 2952 case BO_EQ: 2953 case BO_NE: 2954 return IntRange::forBoolType(); 2955 2956 // The type of the assignments is the type of the LHS, so the RHS 2957 // is not necessarily the same type. 2958 case BO_MulAssign: 2959 case BO_DivAssign: 2960 case BO_RemAssign: 2961 case BO_AddAssign: 2962 case BO_SubAssign: 2963 case BO_XorAssign: 2964 case BO_OrAssign: 2965 // TODO: bitfields? 2966 return IntRange::forValueOfType(C, E->getType()); 2967 2968 // Simple assignments just pass through the RHS, which will have 2969 // been coerced to the LHS type. 2970 case BO_Assign: 2971 // TODO: bitfields? 2972 return GetExprRange(C, BO->getRHS(), MaxWidth); 2973 2974 // Operations with opaque sources are black-listed. 2975 case BO_PtrMemD: 2976 case BO_PtrMemI: 2977 return IntRange::forValueOfType(C, E->getType()); 2978 2979 // Bitwise-and uses the *infinum* of the two source ranges. 2980 case BO_And: 2981 case BO_AndAssign: 2982 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 2983 GetExprRange(C, BO->getRHS(), MaxWidth)); 2984 2985 // Left shift gets black-listed based on a judgement call. 2986 case BO_Shl: 2987 // ...except that we want to treat '1 << (blah)' as logically 2988 // positive. It's an important idiom. 2989 if (IntegerLiteral *I 2990 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 2991 if (I->getValue() == 1) { 2992 IntRange R = IntRange::forValueOfType(C, E->getType()); 2993 return IntRange(R.Width, /*NonNegative*/ true); 2994 } 2995 } 2996 // fallthrough 2997 2998 case BO_ShlAssign: 2999 return IntRange::forValueOfType(C, E->getType()); 3000 3001 // Right shift by a constant can narrow its left argument. 3002 case BO_Shr: 3003 case BO_ShrAssign: { 3004 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3005 3006 // If the shift amount is a positive constant, drop the width by 3007 // that much. 3008 llvm::APSInt shift; 3009 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 3010 shift.isNonNegative()) { 3011 unsigned zext = shift.getZExtValue(); 3012 if (zext >= L.Width) 3013 L.Width = (L.NonNegative ? 0 : 1); 3014 else 3015 L.Width -= zext; 3016 } 3017 3018 return L; 3019 } 3020 3021 // Comma acts as its right operand. 3022 case BO_Comma: 3023 return GetExprRange(C, BO->getRHS(), MaxWidth); 3024 3025 // Black-list pointer subtractions. 3026 case BO_Sub: 3027 if (BO->getLHS()->getType()->isPointerType()) 3028 return IntRange::forValueOfType(C, E->getType()); 3029 break; 3030 3031 // The width of a division result is mostly determined by the size 3032 // of the LHS. 3033 case BO_Div: { 3034 // Don't 'pre-truncate' the operands. 3035 unsigned opWidth = C.getIntWidth(E->getType()); 3036 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3037 3038 // If the divisor is constant, use that. 3039 llvm::APSInt divisor; 3040 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 3041 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 3042 if (log2 >= L.Width) 3043 L.Width = (L.NonNegative ? 0 : 1); 3044 else 3045 L.Width = std::min(L.Width - log2, MaxWidth); 3046 return L; 3047 } 3048 3049 // Otherwise, just use the LHS's width. 3050 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3051 return IntRange(L.Width, L.NonNegative && R.NonNegative); 3052 } 3053 3054 // The result of a remainder can't be larger than the result of 3055 // either side. 3056 case BO_Rem: { 3057 // Don't 'pre-truncate' the operands. 3058 unsigned opWidth = C.getIntWidth(E->getType()); 3059 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3060 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3061 3062 IntRange meet = IntRange::meet(L, R); 3063 meet.Width = std::min(meet.Width, MaxWidth); 3064 return meet; 3065 } 3066 3067 // The default behavior is okay for these. 3068 case BO_Mul: 3069 case BO_Add: 3070 case BO_Xor: 3071 case BO_Or: 3072 break; 3073 } 3074 3075 // The default case is to treat the operation as if it were closed 3076 // on the narrowest type that encompasses both operands. 3077 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3078 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 3079 return IntRange::join(L, R); 3080 } 3081 3082 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 3083 switch (UO->getOpcode()) { 3084 // Boolean-valued operations are white-listed. 3085 case UO_LNot: 3086 return IntRange::forBoolType(); 3087 3088 // Operations with opaque sources are black-listed. 3089 case UO_Deref: 3090 case UO_AddrOf: // should be impossible 3091 return IntRange::forValueOfType(C, E->getType()); 3092 3093 default: 3094 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 3095 } 3096 } 3097 3098 if (dyn_cast<OffsetOfExpr>(E)) { 3099 IntRange::forValueOfType(C, E->getType()); 3100 } 3101 3102 if (FieldDecl *BitField = E->getBitField()) 3103 return IntRange(BitField->getBitWidthValue(C), 3104 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 3105 3106 return IntRange::forValueOfType(C, E->getType()); 3107 } 3108 3109 IntRange GetExprRange(ASTContext &C, Expr *E) { 3110 return GetExprRange(C, E, C.getIntWidth(E->getType())); 3111 } 3112 3113 /// Checks whether the given value, which currently has the given 3114 /// source semantics, has the same value when coerced through the 3115 /// target semantics. 3116 bool IsSameFloatAfterCast(const llvm::APFloat &value, 3117 const llvm::fltSemantics &Src, 3118 const llvm::fltSemantics &Tgt) { 3119 llvm::APFloat truncated = value; 3120 3121 bool ignored; 3122 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 3123 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 3124 3125 return truncated.bitwiseIsEqual(value); 3126 } 3127 3128 /// Checks whether the given value, which currently has the given 3129 /// source semantics, has the same value when coerced through the 3130 /// target semantics. 3131 /// 3132 /// The value might be a vector of floats (or a complex number). 3133 bool IsSameFloatAfterCast(const APValue &value, 3134 const llvm::fltSemantics &Src, 3135 const llvm::fltSemantics &Tgt) { 3136 if (value.isFloat()) 3137 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 3138 3139 if (value.isVector()) { 3140 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 3141 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 3142 return false; 3143 return true; 3144 } 3145 3146 assert(value.isComplexFloat()); 3147 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 3148 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 3149 } 3150 3151 void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 3152 3153 static bool IsZero(Sema &S, Expr *E) { 3154 // Suppress cases where we are comparing against an enum constant. 3155 if (const DeclRefExpr *DR = 3156 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 3157 if (isa<EnumConstantDecl>(DR->getDecl())) 3158 return false; 3159 3160 // Suppress cases where the '0' value is expanded from a macro. 3161 if (E->getLocStart().isMacroID()) 3162 return false; 3163 3164 llvm::APSInt Value; 3165 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 3166 } 3167 3168 static bool HasEnumType(Expr *E) { 3169 // Strip off implicit integral promotions. 3170 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 3171 if (ICE->getCastKind() != CK_IntegralCast && 3172 ICE->getCastKind() != CK_NoOp) 3173 break; 3174 E = ICE->getSubExpr(); 3175 } 3176 3177 return E->getType()->isEnumeralType(); 3178 } 3179 3180 void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 3181 BinaryOperatorKind op = E->getOpcode(); 3182 if (E->isValueDependent()) 3183 return; 3184 3185 if (op == BO_LT && IsZero(S, E->getRHS())) { 3186 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3187 << "< 0" << "false" << HasEnumType(E->getLHS()) 3188 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3189 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 3190 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3191 << ">= 0" << "true" << HasEnumType(E->getLHS()) 3192 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3193 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 3194 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3195 << "0 >" << "false" << HasEnumType(E->getRHS()) 3196 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3197 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 3198 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3199 << "0 <=" << "true" << HasEnumType(E->getRHS()) 3200 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3201 } 3202 } 3203 3204 /// Analyze the operands of the given comparison. Implements the 3205 /// fallback case from AnalyzeComparison. 3206 void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 3207 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3208 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3209 } 3210 3211 /// \brief Implements -Wsign-compare. 3212 /// 3213 /// \param E the binary operator to check for warnings 3214 void AnalyzeComparison(Sema &S, BinaryOperator *E) { 3215 // The type the comparison is being performed in. 3216 QualType T = E->getLHS()->getType(); 3217 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 3218 && "comparison with mismatched types"); 3219 3220 // We don't do anything special if this isn't an unsigned integral 3221 // comparison: we're only interested in integral comparisons, and 3222 // signed comparisons only happen in cases we don't care to warn about. 3223 // 3224 // We also don't care about value-dependent expressions or expressions 3225 // whose result is a constant. 3226 if (!T->hasUnsignedIntegerRepresentation() 3227 || E->isValueDependent() || E->isIntegerConstantExpr(S.Context)) 3228 return AnalyzeImpConvsInComparison(S, E); 3229 3230 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 3231 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 3232 3233 // Check to see if one of the (unmodified) operands is of different 3234 // signedness. 3235 Expr *signedOperand, *unsignedOperand; 3236 if (LHS->getType()->hasSignedIntegerRepresentation()) { 3237 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 3238 "unsigned comparison between two signed integer expressions?"); 3239 signedOperand = LHS; 3240 unsignedOperand = RHS; 3241 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 3242 signedOperand = RHS; 3243 unsignedOperand = LHS; 3244 } else { 3245 CheckTrivialUnsignedComparison(S, E); 3246 return AnalyzeImpConvsInComparison(S, E); 3247 } 3248 3249 // Otherwise, calculate the effective range of the signed operand. 3250 IntRange signedRange = GetExprRange(S.Context, signedOperand); 3251 3252 // Go ahead and analyze implicit conversions in the operands. Note 3253 // that we skip the implicit conversions on both sides. 3254 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 3255 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 3256 3257 // If the signed range is non-negative, -Wsign-compare won't fire, 3258 // but we should still check for comparisons which are always true 3259 // or false. 3260 if (signedRange.NonNegative) 3261 return CheckTrivialUnsignedComparison(S, E); 3262 3263 // For (in)equality comparisons, if the unsigned operand is a 3264 // constant which cannot collide with a overflowed signed operand, 3265 // then reinterpreting the signed operand as unsigned will not 3266 // change the result of the comparison. 3267 if (E->isEqualityOp()) { 3268 unsigned comparisonWidth = S.Context.getIntWidth(T); 3269 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 3270 3271 // We should never be unable to prove that the unsigned operand is 3272 // non-negative. 3273 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 3274 3275 if (unsignedRange.Width < comparisonWidth) 3276 return; 3277 } 3278 3279 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison) 3280 << LHS->getType() << RHS->getType() 3281 << LHS->getSourceRange() << RHS->getSourceRange(); 3282 } 3283 3284 /// Analyzes an attempt to assign the given value to a bitfield. 3285 /// 3286 /// Returns true if there was something fishy about the attempt. 3287 bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 3288 SourceLocation InitLoc) { 3289 assert(Bitfield->isBitField()); 3290 if (Bitfield->isInvalidDecl()) 3291 return false; 3292 3293 // White-list bool bitfields. 3294 if (Bitfield->getType()->isBooleanType()) 3295 return false; 3296 3297 // Ignore value- or type-dependent expressions. 3298 if (Bitfield->getBitWidth()->isValueDependent() || 3299 Bitfield->getBitWidth()->isTypeDependent() || 3300 Init->isValueDependent() || 3301 Init->isTypeDependent()) 3302 return false; 3303 3304 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 3305 3306 Expr::EvalResult InitValue; 3307 if (!OriginalInit->Evaluate(InitValue, S.Context) || 3308 !InitValue.Val.isInt()) 3309 return false; 3310 3311 const llvm::APSInt &Value = InitValue.Val.getInt(); 3312 unsigned OriginalWidth = Value.getBitWidth(); 3313 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 3314 3315 if (OriginalWidth <= FieldWidth) 3316 return false; 3317 3318 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 3319 3320 // It's fairly common to write values into signed bitfields 3321 // that, if sign-extended, would end up becoming a different 3322 // value. We don't want to warn about that. 3323 if (Value.isSigned() && Value.isNegative()) 3324 TruncatedValue = TruncatedValue.sext(OriginalWidth); 3325 else 3326 TruncatedValue = TruncatedValue.zext(OriginalWidth); 3327 3328 if (Value == TruncatedValue) 3329 return false; 3330 3331 std::string PrettyValue = Value.toString(10); 3332 std::string PrettyTrunc = TruncatedValue.toString(10); 3333 3334 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 3335 << PrettyValue << PrettyTrunc << OriginalInit->getType() 3336 << Init->getSourceRange(); 3337 3338 return true; 3339 } 3340 3341 /// Analyze the given simple or compound assignment for warning-worthy 3342 /// operations. 3343 void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 3344 // Just recurse on the LHS. 3345 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3346 3347 // We want to recurse on the RHS as normal unless we're assigning to 3348 // a bitfield. 3349 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 3350 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 3351 E->getOperatorLoc())) { 3352 // Recurse, ignoring any implicit conversions on the RHS. 3353 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 3354 E->getOperatorLoc()); 3355 } 3356 } 3357 3358 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3359 } 3360 3361 /// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3362 void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 3363 SourceLocation CContext, unsigned diag) { 3364 S.Diag(E->getExprLoc(), diag) 3365 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 3366 } 3367 3368 /// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3369 void DiagnoseImpCast(Sema &S, Expr *E, QualType T, SourceLocation CContext, 3370 unsigned diag) { 3371 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag); 3372 } 3373 3374 /// Diagnose an implicit cast from a literal expression. Does not warn when the 3375 /// cast wouldn't lose information. 3376 void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 3377 SourceLocation CContext) { 3378 // Try to convert the literal exactly to an integer. If we can, don't warn. 3379 bool isExact = false; 3380 const llvm::APFloat &Value = FL->getValue(); 3381 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 3382 T->hasUnsignedIntegerRepresentation()); 3383 if (Value.convertToInteger(IntegerValue, 3384 llvm::APFloat::rmTowardZero, &isExact) 3385 == llvm::APFloat::opOK && isExact) 3386 return; 3387 3388 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 3389 << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext); 3390 } 3391 3392 std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 3393 if (!Range.Width) return "0"; 3394 3395 llvm::APSInt ValueInRange = Value; 3396 ValueInRange.setIsSigned(!Range.NonNegative); 3397 ValueInRange = ValueInRange.trunc(Range.Width); 3398 return ValueInRange.toString(10); 3399 } 3400 3401 static bool isFromSystemMacro(Sema &S, SourceLocation loc) { 3402 SourceManager &smgr = S.Context.getSourceManager(); 3403 return loc.isMacroID() && smgr.isInSystemHeader(smgr.getSpellingLoc(loc)); 3404 } 3405 3406 void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 3407 SourceLocation CC, bool *ICContext = 0) { 3408 if (E->isTypeDependent() || E->isValueDependent()) return; 3409 3410 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 3411 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 3412 if (Source == Target) return; 3413 if (Target->isDependentType()) return; 3414 3415 // If the conversion context location is invalid don't complain. We also 3416 // don't want to emit a warning if the issue occurs from the expansion of 3417 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 3418 // delay this check as long as possible. Once we detect we are in that 3419 // scenario, we just return. 3420 if (CC.isInvalid()) 3421 return; 3422 3423 // Diagnose implicit casts to bool. 3424 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 3425 if (isa<StringLiteral>(E)) 3426 // Warn on string literal to bool. Checks for string literals in logical 3427 // expressions, for instances, assert(0 && "error here"), is prevented 3428 // by a check in AnalyzeImplicitConversions(). 3429 return DiagnoseImpCast(S, E, T, CC, 3430 diag::warn_impcast_string_literal_to_bool); 3431 return; // Other casts to bool are not checked. 3432 } 3433 3434 // Strip vector types. 3435 if (isa<VectorType>(Source)) { 3436 if (!isa<VectorType>(Target)) { 3437 if (isFromSystemMacro(S, CC)) 3438 return; 3439 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 3440 } 3441 3442 // If the vector cast is cast between two vectors of the same size, it is 3443 // a bitcast, not a conversion. 3444 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 3445 return; 3446 3447 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 3448 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 3449 } 3450 3451 // Strip complex types. 3452 if (isa<ComplexType>(Source)) { 3453 if (!isa<ComplexType>(Target)) { 3454 if (isFromSystemMacro(S, CC)) 3455 return; 3456 3457 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 3458 } 3459 3460 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 3461 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 3462 } 3463 3464 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 3465 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 3466 3467 // If the source is floating point... 3468 if (SourceBT && SourceBT->isFloatingPoint()) { 3469 // ...and the target is floating point... 3470 if (TargetBT && TargetBT->isFloatingPoint()) { 3471 // ...then warn if we're dropping FP rank. 3472 3473 // Builtin FP kinds are ordered by increasing FP rank. 3474 if (SourceBT->getKind() > TargetBT->getKind()) { 3475 // Don't warn about float constants that are precisely 3476 // representable in the target type. 3477 Expr::EvalResult result; 3478 if (E->Evaluate(result, S.Context)) { 3479 // Value might be a float, a float vector, or a float complex. 3480 if (IsSameFloatAfterCast(result.Val, 3481 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 3482 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 3483 return; 3484 } 3485 3486 if (isFromSystemMacro(S, CC)) 3487 return; 3488 3489 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 3490 } 3491 return; 3492 } 3493 3494 // If the target is integral, always warn. 3495 if ((TargetBT && TargetBT->isInteger())) { 3496 if (isFromSystemMacro(S, CC)) 3497 return; 3498 3499 Expr *InnerE = E->IgnoreParenImpCasts(); 3500 // We also want to warn on, e.g., "int i = -1.234" 3501 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 3502 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 3503 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 3504 3505 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 3506 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 3507 } else { 3508 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 3509 } 3510 } 3511 3512 return; 3513 } 3514 3515 if (!Source->isIntegerType() || !Target->isIntegerType()) 3516 return; 3517 3518 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 3519 == Expr::NPCK_GNUNull) && Target->isIntegerType()) { 3520 S.Diag(E->getExprLoc(), diag::warn_impcast_null_pointer_to_integer) 3521 << E->getSourceRange() << clang::SourceRange(CC); 3522 return; 3523 } 3524 3525 IntRange SourceRange = GetExprRange(S.Context, E); 3526 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 3527 3528 if (SourceRange.Width > TargetRange.Width) { 3529 // If the source is a constant, use a default-on diagnostic. 3530 // TODO: this should happen for bitfield stores, too. 3531 llvm::APSInt Value(32); 3532 if (E->isIntegerConstantExpr(Value, S.Context)) { 3533 if (isFromSystemMacro(S, CC)) 3534 return; 3535 3536 std::string PrettySourceValue = Value.toString(10); 3537 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 3538 3539 S.Diag(E->getExprLoc(), diag::warn_impcast_integer_precision_constant) 3540 << PrettySourceValue << PrettyTargetValue 3541 << E->getType() << T << E->getSourceRange() << clang::SourceRange(CC); 3542 return; 3543 } 3544 3545 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 3546 if (isFromSystemMacro(S, CC)) 3547 return; 3548 3549 if (SourceRange.Width == 64 && TargetRange.Width == 32) 3550 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32); 3551 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 3552 } 3553 3554 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 3555 (!TargetRange.NonNegative && SourceRange.NonNegative && 3556 SourceRange.Width == TargetRange.Width)) { 3557 3558 if (isFromSystemMacro(S, CC)) 3559 return; 3560 3561 unsigned DiagID = diag::warn_impcast_integer_sign; 3562 3563 // Traditionally, gcc has warned about this under -Wsign-compare. 3564 // We also want to warn about it in -Wconversion. 3565 // So if -Wconversion is off, use a completely identical diagnostic 3566 // in the sign-compare group. 3567 // The conditional-checking code will 3568 if (ICContext) { 3569 DiagID = diag::warn_impcast_integer_sign_conditional; 3570 *ICContext = true; 3571 } 3572 3573 return DiagnoseImpCast(S, E, T, CC, DiagID); 3574 } 3575 3576 // Diagnose conversions between different enumeration types. 3577 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 3578 // type, to give us better diagnostics. 3579 QualType SourceType = E->getType(); 3580 if (!S.getLangOptions().CPlusPlus) { 3581 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 3582 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 3583 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 3584 SourceType = S.Context.getTypeDeclType(Enum); 3585 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 3586 } 3587 } 3588 3589 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 3590 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 3591 if ((SourceEnum->getDecl()->getIdentifier() || 3592 SourceEnum->getDecl()->getTypedefNameForAnonDecl()) && 3593 (TargetEnum->getDecl()->getIdentifier() || 3594 TargetEnum->getDecl()->getTypedefNameForAnonDecl()) && 3595 SourceEnum != TargetEnum) { 3596 if (isFromSystemMacro(S, CC)) 3597 return; 3598 3599 return DiagnoseImpCast(S, E, SourceType, T, CC, 3600 diag::warn_impcast_different_enum_types); 3601 } 3602 3603 return; 3604 } 3605 3606 void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 3607 3608 void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 3609 SourceLocation CC, bool &ICContext) { 3610 E = E->IgnoreParenImpCasts(); 3611 3612 if (isa<ConditionalOperator>(E)) 3613 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 3614 3615 AnalyzeImplicitConversions(S, E, CC); 3616 if (E->getType() != T) 3617 return CheckImplicitConversion(S, E, T, CC, &ICContext); 3618 return; 3619 } 3620 3621 void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 3622 SourceLocation CC = E->getQuestionLoc(); 3623 3624 AnalyzeImplicitConversions(S, E->getCond(), CC); 3625 3626 bool Suspicious = false; 3627 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 3628 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 3629 3630 // If -Wconversion would have warned about either of the candidates 3631 // for a signedness conversion to the context type... 3632 if (!Suspicious) return; 3633 3634 // ...but it's currently ignored... 3635 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 3636 CC)) 3637 return; 3638 3639 // ...then check whether it would have warned about either of the 3640 // candidates for a signedness conversion to the condition type. 3641 if (E->getType() == T) return; 3642 3643 Suspicious = false; 3644 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 3645 E->getType(), CC, &Suspicious); 3646 if (!Suspicious) 3647 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 3648 E->getType(), CC, &Suspicious); 3649 } 3650 3651 /// AnalyzeImplicitConversions - Find and report any interesting 3652 /// implicit conversions in the given expression. There are a couple 3653 /// of competing diagnostics here, -Wconversion and -Wsign-compare. 3654 void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 3655 QualType T = OrigE->getType(); 3656 Expr *E = OrigE->IgnoreParenImpCasts(); 3657 3658 if (E->isTypeDependent() || E->isValueDependent()) 3659 return; 3660 3661 // For conditional operators, we analyze the arguments as if they 3662 // were being fed directly into the output. 3663 if (isa<ConditionalOperator>(E)) { 3664 ConditionalOperator *CO = cast<ConditionalOperator>(E); 3665 CheckConditionalOperator(S, CO, T); 3666 return; 3667 } 3668 3669 // Go ahead and check any implicit conversions we might have skipped. 3670 // The non-canonical typecheck is just an optimization; 3671 // CheckImplicitConversion will filter out dead implicit conversions. 3672 if (E->getType() != T) 3673 CheckImplicitConversion(S, E, T, CC); 3674 3675 // Now continue drilling into this expression. 3676 3677 // Skip past explicit casts. 3678 if (isa<ExplicitCastExpr>(E)) { 3679 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 3680 return AnalyzeImplicitConversions(S, E, CC); 3681 } 3682 3683 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 3684 // Do a somewhat different check with comparison operators. 3685 if (BO->isComparisonOp()) 3686 return AnalyzeComparison(S, BO); 3687 3688 // And with assignments and compound assignments. 3689 if (BO->isAssignmentOp()) 3690 return AnalyzeAssignment(S, BO); 3691 } 3692 3693 // These break the otherwise-useful invariant below. Fortunately, 3694 // we don't really need to recurse into them, because any internal 3695 // expressions should have been analyzed already when they were 3696 // built into statements. 3697 if (isa<StmtExpr>(E)) return; 3698 3699 // Don't descend into unevaluated contexts. 3700 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 3701 3702 // Now just recurse over the expression's children. 3703 CC = E->getExprLoc(); 3704 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 3705 bool IsLogicalOperator = BO && BO->isLogicalOp(); 3706 for (Stmt::child_range I = E->children(); I; ++I) { 3707 Expr *ChildExpr = cast<Expr>(*I); 3708 if (IsLogicalOperator && 3709 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 3710 // Ignore checking string literals that are in logical operators. 3711 continue; 3712 AnalyzeImplicitConversions(S, ChildExpr, CC); 3713 } 3714 } 3715 3716 } // end anonymous namespace 3717 3718 /// Diagnoses "dangerous" implicit conversions within the given 3719 /// expression (which is a full expression). Implements -Wconversion 3720 /// and -Wsign-compare. 3721 /// 3722 /// \param CC the "context" location of the implicit conversion, i.e. 3723 /// the most location of the syntactic entity requiring the implicit 3724 /// conversion 3725 void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 3726 // Don't diagnose in unevaluated contexts. 3727 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 3728 return; 3729 3730 // Don't diagnose for value- or type-dependent expressions. 3731 if (E->isTypeDependent() || E->isValueDependent()) 3732 return; 3733 3734 // Check for array bounds violations in cases where the check isn't triggered 3735 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 3736 // ArraySubscriptExpr is on the RHS of a variable initialization. 3737 CheckArrayAccess(E); 3738 3739 // This is not the right CC for (e.g.) a variable initialization. 3740 AnalyzeImplicitConversions(*this, E, CC); 3741 } 3742 3743 void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 3744 FieldDecl *BitField, 3745 Expr *Init) { 3746 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 3747 } 3748 3749 /// CheckParmsForFunctionDef - Check that the parameters of the given 3750 /// function are appropriate for the definition of a function. This 3751 /// takes care of any checks that cannot be performed on the 3752 /// declaration itself, e.g., that the types of each of the function 3753 /// parameters are complete. 3754 bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 3755 bool CheckParameterNames) { 3756 bool HasInvalidParm = false; 3757 for (; P != PEnd; ++P) { 3758 ParmVarDecl *Param = *P; 3759 3760 // C99 6.7.5.3p4: the parameters in a parameter type list in a 3761 // function declarator that is part of a function definition of 3762 // that function shall not have incomplete type. 3763 // 3764 // This is also C++ [dcl.fct]p6. 3765 if (!Param->isInvalidDecl() && 3766 RequireCompleteType(Param->getLocation(), Param->getType(), 3767 diag::err_typecheck_decl_incomplete_type)) { 3768 Param->setInvalidDecl(); 3769 HasInvalidParm = true; 3770 } 3771 3772 // C99 6.9.1p5: If the declarator includes a parameter type list, the 3773 // declaration of each parameter shall include an identifier. 3774 if (CheckParameterNames && 3775 Param->getIdentifier() == 0 && 3776 !Param->isImplicit() && 3777 !getLangOptions().CPlusPlus) 3778 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 3779 3780 // C99 6.7.5.3p12: 3781 // If the function declarator is not part of a definition of that 3782 // function, parameters may have incomplete type and may use the [*] 3783 // notation in their sequences of declarator specifiers to specify 3784 // variable length array types. 3785 QualType PType = Param->getOriginalType(); 3786 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 3787 if (AT->getSizeModifier() == ArrayType::Star) { 3788 // FIXME: This diagnosic should point the the '[*]' if source-location 3789 // information is added for it. 3790 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 3791 } 3792 } 3793 } 3794 3795 return HasInvalidParm; 3796 } 3797 3798 /// CheckCastAlign - Implements -Wcast-align, which warns when a 3799 /// pointer cast increases the alignment requirements. 3800 void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 3801 // This is actually a lot of work to potentially be doing on every 3802 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 3803 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 3804 TRange.getBegin()) 3805 == DiagnosticsEngine::Ignored) 3806 return; 3807 3808 // Ignore dependent types. 3809 if (T->isDependentType() || Op->getType()->isDependentType()) 3810 return; 3811 3812 // Require that the destination be a pointer type. 3813 const PointerType *DestPtr = T->getAs<PointerType>(); 3814 if (!DestPtr) return; 3815 3816 // If the destination has alignment 1, we're done. 3817 QualType DestPointee = DestPtr->getPointeeType(); 3818 if (DestPointee->isIncompleteType()) return; 3819 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 3820 if (DestAlign.isOne()) return; 3821 3822 // Require that the source be a pointer type. 3823 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 3824 if (!SrcPtr) return; 3825 QualType SrcPointee = SrcPtr->getPointeeType(); 3826 3827 // Whitelist casts from cv void*. We already implicitly 3828 // whitelisted casts to cv void*, since they have alignment 1. 3829 // Also whitelist casts involving incomplete types, which implicitly 3830 // includes 'void'. 3831 if (SrcPointee->isIncompleteType()) return; 3832 3833 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 3834 if (SrcAlign >= DestAlign) return; 3835 3836 Diag(TRange.getBegin(), diag::warn_cast_align) 3837 << Op->getType() << T 3838 << static_cast<unsigned>(SrcAlign.getQuantity()) 3839 << static_cast<unsigned>(DestAlign.getQuantity()) 3840 << TRange << Op->getSourceRange(); 3841 } 3842 3843 static const Type* getElementType(const Expr *BaseExpr) { 3844 const Type* EltType = BaseExpr->getType().getTypePtr(); 3845 if (EltType->isAnyPointerType()) 3846 return EltType->getPointeeType().getTypePtr(); 3847 else if (EltType->isArrayType()) 3848 return EltType->getBaseElementTypeUnsafe(); 3849 return EltType; 3850 } 3851 3852 /// \brief Check whether this array fits the idiom of a size-one tail padded 3853 /// array member of a struct. 3854 /// 3855 /// We avoid emitting out-of-bounds access warnings for such arrays as they are 3856 /// commonly used to emulate flexible arrays in C89 code. 3857 static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 3858 const NamedDecl *ND) { 3859 if (Size != 1 || !ND) return false; 3860 3861 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 3862 if (!FD) return false; 3863 3864 // Don't consider sizes resulting from macro expansions or template argument 3865 // substitution to form C89 tail-padded arrays. 3866 ConstantArrayTypeLoc TL = 3867 cast<ConstantArrayTypeLoc>(FD->getTypeSourceInfo()->getTypeLoc()); 3868 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(TL.getSizeExpr()); 3869 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 3870 return false; 3871 3872 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 3873 if (!RD || !RD->isStruct()) 3874 return false; 3875 3876 // See if this is the last field decl in the record. 3877 const Decl *D = FD; 3878 while ((D = D->getNextDeclInContext())) 3879 if (isa<FieldDecl>(D)) 3880 return false; 3881 return true; 3882 } 3883 3884 void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 3885 bool isSubscript, bool AllowOnePastEnd) { 3886 const Type* EffectiveType = getElementType(BaseExpr); 3887 BaseExpr = BaseExpr->IgnoreParenCasts(); 3888 IndexExpr = IndexExpr->IgnoreParenCasts(); 3889 3890 const ConstantArrayType *ArrayTy = 3891 Context.getAsConstantArrayType(BaseExpr->getType()); 3892 if (!ArrayTy) 3893 return; 3894 3895 if (IndexExpr->isValueDependent()) 3896 return; 3897 llvm::APSInt index; 3898 if (!IndexExpr->isIntegerConstantExpr(index, Context)) 3899 return; 3900 3901 const NamedDecl *ND = NULL; 3902 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 3903 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 3904 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 3905 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 3906 3907 if (index.isUnsigned() || !index.isNegative()) { 3908 llvm::APInt size = ArrayTy->getSize(); 3909 if (!size.isStrictlyPositive()) 3910 return; 3911 3912 const Type* BaseType = getElementType(BaseExpr); 3913 if (BaseType != EffectiveType) { 3914 // Make sure we're comparing apples to apples when comparing index to size 3915 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 3916 uint64_t array_typesize = Context.getTypeSize(BaseType); 3917 // Handle ptrarith_typesize being zero, such as when casting to void* 3918 if (!ptrarith_typesize) ptrarith_typesize = 1; 3919 if (ptrarith_typesize != array_typesize) { 3920 // There's a cast to a different size type involved 3921 uint64_t ratio = array_typesize / ptrarith_typesize; 3922 // TODO: Be smarter about handling cases where array_typesize is not a 3923 // multiple of ptrarith_typesize 3924 if (ptrarith_typesize * ratio == array_typesize) 3925 size *= llvm::APInt(size.getBitWidth(), ratio); 3926 } 3927 } 3928 3929 if (size.getBitWidth() > index.getBitWidth()) 3930 index = index.sext(size.getBitWidth()); 3931 else if (size.getBitWidth() < index.getBitWidth()) 3932 size = size.sext(index.getBitWidth()); 3933 3934 // For array subscripting the index must be less than size, but for pointer 3935 // arithmetic also allow the index (offset) to be equal to size since 3936 // computing the next address after the end of the array is legal and 3937 // commonly done e.g. in C++ iterators and range-based for loops. 3938 if (AllowOnePastEnd ? index.sle(size) : index.slt(size)) 3939 return; 3940 3941 // Also don't warn for arrays of size 1 which are members of some 3942 // structure. These are often used to approximate flexible arrays in C89 3943 // code. 3944 if (IsTailPaddedMemberArray(*this, size, ND)) 3945 return; 3946 3947 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 3948 if (isSubscript) 3949 DiagID = diag::warn_array_index_exceeds_bounds; 3950 3951 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 3952 PDiag(DiagID) << index.toString(10, true) 3953 << size.toString(10, true) 3954 << (unsigned)size.getLimitedValue(~0U) 3955 << IndexExpr->getSourceRange()); 3956 } else { 3957 unsigned DiagID = diag::warn_array_index_precedes_bounds; 3958 if (!isSubscript) { 3959 DiagID = diag::warn_ptr_arith_precedes_bounds; 3960 if (index.isNegative()) index = -index; 3961 } 3962 3963 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 3964 PDiag(DiagID) << index.toString(10, true) 3965 << IndexExpr->getSourceRange()); 3966 } 3967 3968 if (ND) 3969 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 3970 PDiag(diag::note_array_index_out_of_bounds) 3971 << ND->getDeclName()); 3972 } 3973 3974 void Sema::CheckArrayAccess(const Expr *expr) { 3975 int AllowOnePastEnd = 0; 3976 while (expr) { 3977 expr = expr->IgnoreParenImpCasts(); 3978 switch (expr->getStmtClass()) { 3979 case Stmt::ArraySubscriptExprClass: { 3980 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 3981 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), true, 3982 AllowOnePastEnd > 0); 3983 return; 3984 } 3985 case Stmt::UnaryOperatorClass: { 3986 // Only unwrap the * and & unary operators 3987 const UnaryOperator *UO = cast<UnaryOperator>(expr); 3988 expr = UO->getSubExpr(); 3989 switch (UO->getOpcode()) { 3990 case UO_AddrOf: 3991 AllowOnePastEnd++; 3992 break; 3993 case UO_Deref: 3994 AllowOnePastEnd--; 3995 break; 3996 default: 3997 return; 3998 } 3999 break; 4000 } 4001 case Stmt::ConditionalOperatorClass: { 4002 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 4003 if (const Expr *lhs = cond->getLHS()) 4004 CheckArrayAccess(lhs); 4005 if (const Expr *rhs = cond->getRHS()) 4006 CheckArrayAccess(rhs); 4007 return; 4008 } 4009 default: 4010 return; 4011 } 4012 } 4013 } 4014 4015 //===--- CHECK: Objective-C retain cycles ----------------------------------// 4016 4017 namespace { 4018 struct RetainCycleOwner { 4019 RetainCycleOwner() : Variable(0), Indirect(false) {} 4020 VarDecl *Variable; 4021 SourceRange Range; 4022 SourceLocation Loc; 4023 bool Indirect; 4024 4025 void setLocsFrom(Expr *e) { 4026 Loc = e->getExprLoc(); 4027 Range = e->getSourceRange(); 4028 } 4029 }; 4030 } 4031 4032 /// Consider whether capturing the given variable can possibly lead to 4033 /// a retain cycle. 4034 static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 4035 // In ARC, it's captured strongly iff the variable has __strong 4036 // lifetime. In MRR, it's captured strongly if the variable is 4037 // __block and has an appropriate type. 4038 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4039 return false; 4040 4041 owner.Variable = var; 4042 owner.setLocsFrom(ref); 4043 return true; 4044 } 4045 4046 static bool findRetainCycleOwner(Expr *e, RetainCycleOwner &owner) { 4047 while (true) { 4048 e = e->IgnoreParens(); 4049 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 4050 switch (cast->getCastKind()) { 4051 case CK_BitCast: 4052 case CK_LValueBitCast: 4053 case CK_LValueToRValue: 4054 case CK_ARCReclaimReturnedObject: 4055 e = cast->getSubExpr(); 4056 continue; 4057 4058 case CK_GetObjCProperty: { 4059 // Bail out if this isn't a strong explicit property. 4060 const ObjCPropertyRefExpr *pre = cast->getSubExpr()->getObjCProperty(); 4061 if (pre->isImplicitProperty()) return false; 4062 ObjCPropertyDecl *property = pre->getExplicitProperty(); 4063 if (!property->isRetaining() && 4064 !(property->getPropertyIvarDecl() && 4065 property->getPropertyIvarDecl()->getType() 4066 .getObjCLifetime() == Qualifiers::OCL_Strong)) 4067 return false; 4068 4069 owner.Indirect = true; 4070 e = const_cast<Expr*>(pre->getBase()); 4071 continue; 4072 } 4073 4074 default: 4075 return false; 4076 } 4077 } 4078 4079 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 4080 ObjCIvarDecl *ivar = ref->getDecl(); 4081 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4082 return false; 4083 4084 // Try to find a retain cycle in the base. 4085 if (!findRetainCycleOwner(ref->getBase(), owner)) 4086 return false; 4087 4088 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 4089 owner.Indirect = true; 4090 return true; 4091 } 4092 4093 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 4094 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 4095 if (!var) return false; 4096 return considerVariable(var, ref, owner); 4097 } 4098 4099 if (BlockDeclRefExpr *ref = dyn_cast<BlockDeclRefExpr>(e)) { 4100 owner.Variable = ref->getDecl(); 4101 owner.setLocsFrom(ref); 4102 return true; 4103 } 4104 4105 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 4106 if (member->isArrow()) return false; 4107 4108 // Don't count this as an indirect ownership. 4109 e = member->getBase(); 4110 continue; 4111 } 4112 4113 // Array ivars? 4114 4115 return false; 4116 } 4117 } 4118 4119 namespace { 4120 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 4121 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 4122 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 4123 Variable(variable), Capturer(0) {} 4124 4125 VarDecl *Variable; 4126 Expr *Capturer; 4127 4128 void VisitDeclRefExpr(DeclRefExpr *ref) { 4129 if (ref->getDecl() == Variable && !Capturer) 4130 Capturer = ref; 4131 } 4132 4133 void VisitBlockDeclRefExpr(BlockDeclRefExpr *ref) { 4134 if (ref->getDecl() == Variable && !Capturer) 4135 Capturer = ref; 4136 } 4137 4138 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 4139 if (Capturer) return; 4140 Visit(ref->getBase()); 4141 if (Capturer && ref->isFreeIvar()) 4142 Capturer = ref; 4143 } 4144 4145 void VisitBlockExpr(BlockExpr *block) { 4146 // Look inside nested blocks 4147 if (block->getBlockDecl()->capturesVariable(Variable)) 4148 Visit(block->getBlockDecl()->getBody()); 4149 } 4150 }; 4151 } 4152 4153 /// Check whether the given argument is a block which captures a 4154 /// variable. 4155 static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 4156 assert(owner.Variable && owner.Loc.isValid()); 4157 4158 e = e->IgnoreParenCasts(); 4159 BlockExpr *block = dyn_cast<BlockExpr>(e); 4160 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 4161 return 0; 4162 4163 FindCaptureVisitor visitor(S.Context, owner.Variable); 4164 visitor.Visit(block->getBlockDecl()->getBody()); 4165 return visitor.Capturer; 4166 } 4167 4168 static void diagnoseRetainCycle(Sema &S, Expr *capturer, 4169 RetainCycleOwner &owner) { 4170 assert(capturer); 4171 assert(owner.Variable && owner.Loc.isValid()); 4172 4173 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 4174 << owner.Variable << capturer->getSourceRange(); 4175 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 4176 << owner.Indirect << owner.Range; 4177 } 4178 4179 /// Check for a keyword selector that starts with the word 'add' or 4180 /// 'set'. 4181 static bool isSetterLikeSelector(Selector sel) { 4182 if (sel.isUnarySelector()) return false; 4183 4184 StringRef str = sel.getNameForSlot(0); 4185 while (!str.empty() && str.front() == '_') str = str.substr(1); 4186 if (str.startswith("set") || str.startswith("add")) 4187 str = str.substr(3); 4188 else 4189 return false; 4190 4191 if (str.empty()) return true; 4192 return !islower(str.front()); 4193 } 4194 4195 /// Check a message send to see if it's likely to cause a retain cycle. 4196 void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 4197 // Only check instance methods whose selector looks like a setter. 4198 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 4199 return; 4200 4201 // Try to find a variable that the receiver is strongly owned by. 4202 RetainCycleOwner owner; 4203 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 4204 if (!findRetainCycleOwner(msg->getInstanceReceiver(), owner)) 4205 return; 4206 } else { 4207 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 4208 owner.Variable = getCurMethodDecl()->getSelfDecl(); 4209 owner.Loc = msg->getSuperLoc(); 4210 owner.Range = msg->getSuperLoc(); 4211 } 4212 4213 // Check whether the receiver is captured by any of the arguments. 4214 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 4215 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 4216 return diagnoseRetainCycle(*this, capturer, owner); 4217 } 4218 4219 /// Check a property assign to see if it's likely to cause a retain cycle. 4220 void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 4221 RetainCycleOwner owner; 4222 if (!findRetainCycleOwner(receiver, owner)) 4223 return; 4224 4225 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 4226 diagnoseRetainCycle(*this, capturer, owner); 4227 } 4228 4229 bool Sema::checkUnsafeAssigns(SourceLocation Loc, 4230 QualType LHS, Expr *RHS) { 4231 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 4232 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 4233 return false; 4234 // strip off any implicit cast added to get to the one arc-specific 4235 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4236 if (cast->getCastKind() == CK_ARCConsumeObject) { 4237 Diag(Loc, diag::warn_arc_retained_assign) 4238 << (LT == Qualifiers::OCL_ExplicitNone) 4239 << RHS->getSourceRange(); 4240 return true; 4241 } 4242 RHS = cast->getSubExpr(); 4243 } 4244 return false; 4245 } 4246 4247 void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 4248 Expr *LHS, Expr *RHS) { 4249 QualType LHSType = LHS->getType(); 4250 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 4251 return; 4252 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 4253 // FIXME. Check for other life times. 4254 if (LT != Qualifiers::OCL_None) 4255 return; 4256 4257 if (ObjCPropertyRefExpr *PRE = dyn_cast<ObjCPropertyRefExpr>(LHS)) { 4258 if (PRE->isImplicitProperty()) 4259 return; 4260 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 4261 if (!PD) 4262 return; 4263 4264 unsigned Attributes = PD->getPropertyAttributes(); 4265 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) 4266 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4267 if (cast->getCastKind() == CK_ARCConsumeObject) { 4268 Diag(Loc, diag::warn_arc_retained_property_assign) 4269 << RHS->getSourceRange(); 4270 return; 4271 } 4272 RHS = cast->getSubExpr(); 4273 } 4274 } 4275 } 4276